Springer Series in

optical sciences 90/2

Founded by H.K.V. Lotsch

Editor-in-Chief: W. T. Rhodes, Atlanta

Editorial Board: T. Asakura, Sapporo

K.-H. Brenner, Mannheim
T. W. Hänsch, Garching
T. Kamiya, Tokyo
F. Krausz, Vienna and Garching
B. Monemar, Linköping
H. Venghaus, Berlin
H. Weber, Berlin
H. Weinfurter, Munich

Springer
New York
Berlin
Heidelberg
Hong Kong
London
Milan
Paris
Tokyo
Springer Series in
optical sciences
The Springer Series in Optical Sciences, under the leadership of Editor-in-Chief William T. Rhodes,
Georgia Institute of Technology, USA, and Georgia Tech Lorraine, France, provides an expanding selec-
tion of research monographs in all major areas of optics: lasers and quantum optics, ultrafast phenomena,
optical spectroscopy techniques, optoelectronics, quantum information, information optics, applied laser
technology, industrial applications, and other topics of contemporary interest.
With this broad coverage of topics, the series is of use to all research scientists and engineers who need
up-to-date reference books.
The editors encourage prospective authors to correspond with them in advance of submitting a manu-
script. Submission of manuscripts should be made to the Editor-in-Chief or one of the Editors. See also
http://www.springer.de/phys/books/optical_science/

Editor-in-Chief
William T. Rhodes Ferenc Krausz
Georgia Institute of Technology Vienna University of Technology
School of Electrical and Computer Engineering Photonics Institute
Atlanta, GA 30332-0250, USA Gusshausstrasse 27/387
E-mail: [email protected] 1040 Wien, Austria
E-mail: [email protected]
and
Max-Planck-Institut für Quantenoptik
Hans-Kopfermann-Strasse 1
Editorial Board
85748 Garching, Germany
Toshimitsu Asakura
Hokkai-Gakuen University Bo Monemar
Faculty of Engineering Department of Physics
1-1, Minami-26, Nishi 11, Chuo-ku and Measurement Technology
Sapporo, Hokkaido 064-0926, Japan Materials Science Division
E-mail: [email protected] Linköping University
58183 Linköping, Sweden
E-mail: [email protected]
Karl-Heinz Brenner
Chair of Optoelectronics
University of Mannheim
Herbert Venghaus
Institute of Computer Engineering Heinrich-Hertz-Institut
B6, 26 für Nachrichtentechnik Berlin GmbH
68131 Mannheim, Germany Einsteinufer 37
E-mail: [email protected] 10587 Berlin, Germany
E-mail: [email protected]

Theodor W. Hänsch Horst Weber

Max-Planck-Institut für Quantenoptik Technische Universität Berlin
Hans-Kopfermann-Strasse 1 Optisches Institut
85748 Garching, Germany Strasse des 17. Juni 135
E-mail: [email protected] 10623 Berlin, Germany
E-mail: [email protected]
Takeshi Kamiya
Ministry of Education, Culture, Sports Harald Weinfurter
Science and Technology Ludwig-Maximilians-Universität München
National Institution for Academic Degrees Sektion Physik
3-29-1 Otsuka, Bunkyo-ku Schellingstrasse 4/III
Tokyo 112-0012, Japan 80799 München, Germany
E-mail:[email protected] E-mail: [email protected]
Mohammed N. Islam (Ed.)

Raman Amplifiers
for Telecommunications 2
Sub-Systems and Systems
Foreword by Robert W. Lucky

With 286 Figures

Mohammed N. Islam
Department of Electrical Engineering and Computer Science
University of Michigan at Ann Arbor
1110 EECS Building
1301 Beal Avenue
Ann Arbor, MI 48109-2122
[email protected]
and
Xtera Communications, Inc.
500 West Bethany Drive, Suite 100
Allen, TX 75013
USA
[email protected]

Library of Congress Cataloging-in-Publication Data

Raman amplifiers for telecommunications 2: sub-systems and systems / editor, Mohammed N. Islam.
p. cm. – (Springer series in optical sciences ; v. 90/2)
Includes bibliographical references and index.
ISBN 0-387-40656-5 (hc. : alk. paper)
1. Fiber optics. 2. Optical communications. 3. Raman effect. 4. Optical amplifiers.
I. Islam, Mohammed N. II. Series.
TL5103.592.F52R35 2003
621.382
75–dc21 2003044945

ISBN 0-387-40656-5 ISSN 0342-4111 Printed on acid-free paper.

© 2004 Springer-Verlag New York, Inc.

All rights reserved. This work may not be translated or copied in whole or in part without the written
permission of the publisher (Springer-Verlag New York, Inc., 175 Fifth Avenue, New York, NY 10010,
USA), except for brief excerpts in connection with reviews or scholarly analysis. Use in connection with
any form of information storage and retrieval, electronic adaptation, computer software, or by similar or
dissimilar methodology now known or hereafter developed is forbidden.
The use in this publication of trade names, trademarks, service marks, and similar terms, even if they are
not identified as such, is not to be taken as an expression of opinion as to whether or not they are subject
to proprietary rights.

Printed in the United States of America.

9 8 7 6 5 4 3 2 1 SPIN 10948880 90/2

www.springer-ny.com

Springer-Verlag New York Berlin Heidelberg

A member of BertelsmannSpringer Science+Business Media GmbH
To my loving wife,
Nasreen
This page intentionally left blank
Foreword

I remember vividly the first time that I heard about the fiber amplifier. At that time,
of course, it was the erbium-doped fiber amplifier, the predecessor of the Raman
amplifier that is the subject of this book.
It was an early morning in a forgotten year in Murray Hill, New Jersey at one of
our Bell Labs monthly research staff meetings. About twenty directors and executive
directors of research organizations clustered around a long table in the imposing
executive conference room. Arno Penzias, the vice president of research, presided at
the foot of the table.
Everyone who participated in those research staff meetings will long remember
their culture and atmosphere. Arno would pick an arbitrary starting point somewhere
around the table, and the designated person would head to the front of the table to give
a short talk on “something new” in his or her research area. This first speaker would
invariably fiddle helplessly with the controls embedded in the podium that controlled
the viewgraph projector, but eventually we would hear machinery grinding in the
back room as a large hidden mirror moved into place. We would all wait quietly,
arranging and choosing our own viewgraphs from the piles that lay on the table in
front of every participant.
The rules for the staff meeting were that each speaker was allowed seven minutes
and three viewgraphs. However, in spite of Arno’s best efforts to enforce this regimen,
everyone took too long and used too many viewgraphs. Various attempts at using loud
timers and other incentives all failed. No one could give a respectable talk on a research
topic for which they had passionate feelings in seven minutes.
Another rule was that anyone could forfeit his talk by simply saying, “I pass.”
This forfeiture was always accepted without comment, but new directors always
asked their friends about whether this would constitute a black mark against their
performance. No one knew for sure, but rumor had it that it was unwise to pass unless
you were truly destitute of material. After all, the implication would be that there was
nothing new in your research organization for the last month—not a good indication
of your management skills.
With no one passing, and everyone speaking too long, these staff meetings some-
times seemed endless. Computer scientists would talk about new constructs in soft-
viii Foreword

ware, systems people would talk about new techniques for speech recognition, physi-
cists would talk about some new laser, chemists would show diagrams of new organic
materials, and so forth. It didn’t take long for each talk to exceed the understanding
of most listeners of whatever specialty was being discussed. I always left with a pro-
found sense of the limitations of my own knowledge, but with an exhilarating inkling
into the unfolding of science. It was, perhaps, the best of the old research model in
Bell Labs, and in retrospect I can say that in this new competitive world I miss those
old scientific-style management meetings.
It was in such a meeting that I first heard about the fiber amplifier. I don’t know
whether I had been paying attention, but I was immediately galvanized by the impli-
cations of this new discovery. One word came to me and blazed across my mind: that
word was “transparency.”
Surprisingly, in my experience I am not always immediately enthusiastic about
a new technology upon initial exposure. One might think that the potentials of great
breakthroughs are self-evident, but that does not seem to be the case. When I first
heard about the invention of the laser, I had no premonition that lasers would become
the primary instrument of the world’s telecommunications traffic. When one of the
inventors of public key cryptography told me his idea for having two keys, I scoffed
at the naiveté of his concept. I remember thinking on first hearing about what is now
the principal algorithm for data compression that I thought it was only a theoretical
exercise. So my track record for such insights is not altogether good.
However, with the fiber amplifier I went to the other extreme. I foresaw a dramatic
revolution in communications. I spoke up at the staff meeting that morning to say
that this invention would transform the architecture of communications networks.
This would lead to transparent networks, I said, and that this would not necessarily
be good for AT&T. I got carried away with this vision, and went on to say that pri-
vate networks could have their own wavelengths traveling transparently through the
network, untouched by the common carrier in the middle. One private network might
have “blue” light (figuratively speaking, of course, because we’re not talking about
visible wavelengths) whereas another would have “green.” I foresaw a plug on the
wall that passed only the chosen wavelength, which would be owned exclusively by
that particular customer’s network. AT&T would thus be deprived of the opportunity
to process signals for value-added services. AT&T, in fact, wouldn’t have any idea
what was packed into those wavelengths.
Well, that hasn’t exactly happened, but today’s optical networks are moving to-
wards increased transparency, and Raman amplifiers will accelerate this trend. The
advantages of transparency are compelling. A great many constituent signals can be
amplified cheaply in one fell swoop. More importantly, this amplification is indepen-
dent of the bit rates, protocols, waveforms, multiplexing, or any other particulars of
the transmission format. The design isn’t “locked in” to any specific format, and as
these details change, the amplification remains as effective as ever. In the case of the
Raman amplifier, the bandwidth is so enormous that adjectives seem inadequate to
describe its potential for bulk amplification.
Transparency in the network is so attractive that probably the only reason it isn’t
done is that it is so difficult to achieve. One reason is, of course, the necessity for
periodically unbundling the signal to add or drop subcomponents. In the digital world
 Foreword ix

this has usually meant a complete demultiplexing and remultiplexing of the overall
signal, an expensive operation. The optical world opens up the possibility of selective
transparency for certain wavelengths whereas others are unpacked to do add-drop
multiplexing.
So network topology sets limits on transparency. But the other reason transparency
is hard to achieve is the implicit accumulation of impairments as a signal incurs succes-
sive amplifications. It is ironic that the telephone network was essentially transparent
for the first half-century of its existence. Until 1960 the long-haul transmission sys-
tems used analogue amplification to boost levels as the signal traversed the nation.
The invention of the triode vacuum tube enabled the first transcontinental transmis-
sion system to be deployed in the 1920s. It was a marvelous feat to be able to send a
band of signals 3000 miles across the country, passing through many amplifiers, ac-
cumulating noise and distortion along the way, but still providing intelligible speech
at the other end. Some older readers will remember when long distance phone calls
sounded crackly and “distant.” Now, of course, it is impossible to tell how far away
a connection is. They all sound local, because of digital transmission.
Digital transmission was the triumph of the 1960s. Though now it seems obvious,
engineers found the philosophy of digitization hard to grasp for several decades after
the invention of pulse code modulation by Reeves in 1939. There is a trade-off here:
bandwidth against perfectibility. A 3 kHz voice signal, for example, is transformed by
an analogue-to-digital converter into a 64,000 kbps stream of bits, greatly expanding
the necessary transmission bandwidth. However, this digital signal can be regener-
ated perfectly, removing noise and distortion periodically as necessary. A miracle is
achieved as the bits arrive across the country in the same pristine form as when they
left.
So it was that all long distance transmission was converted to digital format.
The introduction of the first lightwave transmission systems hurried this change,
inasmuch as lightwave systems were deemed to be “intrinsically digital” because of
their nonlinearities and the lack of amplifiers. No one cared much at the time—the
early 1980s—but the entire design of the network was predicated on the transmission
of 64 kbps voice channels. The multiplexing hierarchy, the electronic switching, the
synchronization and timing, and the transmission format assumed that everything was
packaged into neat little voice channels. That, of course, was before the rise of the
Internet.
Now optical amplification has reversed this trend of the last half-century towards
digitization based upon a hierarchy of voice signals. It isn’t just that optical amplifiers
have an enormous bandwidth. They do something those old triode vacuum tubes could
never do: they amplify without substantially increasing the noise and distortion of the
signal. Raman amplifiers are particularly good in this way. Moreover, because Raman
amplification is distributed across the whole span of the fiber, the signal level never
drops as low as it does when discrete amplifiers are employed. In a system using
discrete amplifiers the signal level is at its lowest and most vulnerable right before
the point of amplification.
Back at that research staff meeting I was concerned about the implications of
transparency to the architecture of the network.Atransparent network is, by definition,
a “dumb” network. It doesn’t do anything to the signal; it can’t, because it doesn’t
x Foreword

know what the signal is. As an AT&T employee, that sounded threatening. As an
Internet user, that sounded empowering. The Internet, after all, was designed around
the so-called end-to-end principle. In the architecture of the Internet, intelligence is at
the periphery of the network, and the network is as minimally intrusive as is necessary
to achieve interconnection. It is an extremely important philosophical principle that
was just beginning to be understood in the 1980s. Since then the argument has raged,
and the concept of a “stupid” network has been put forward by a number of Internet
designers as the ultimate desired objective. If that is so, then the optical amplifier has
made possible the ultimate stupid network.
I can’t leave this foreword without mentioning another observation on perhaps
a more personal level. Raman amplifiers epitomize for me the transformation of
communications from a world of electrical circuits to one of quantum mechanical
phenomena. Of course you could argue that transistors themselves depend on quan-
tum mechanical principles, and surely the laser does, and so forth. But for many
practical and design purposes these devices could be modeled with traditional circuit
equivalents. Since then, however, photonics has increasingly become a showpiece of
modern physics. The erbium-doped fiber amplifier had to be understood as a quantum
interaction of light with the erbium atom. Raman amplifiers, by contrast, involve the
interaction of light with a material structure. We descend ever more into the realm of
quantum phenomena, into a world of small and impressive miracles.
A number of my friends and associates at Bell Labs have contributed to this
technology and even to this particular book. I’m very proud of the work that they and
their peers in academia and other industries have done in the creation of photonics
technology. I’ve seen it grow around me and have taken vicarious pride in their
accomplishments. Sometimes I tell people that, yes, I know the inventors of this or
that great technology, even though I may not have realized at the time the significance
of the invention. In the case of Raman amplifiers I remember learning about Raman
effects as one of the impairments to be overcome in optical transmission. Researchers
in my organization were even then experimenting with Raman amplification, and
although there was a glimmer of potential, I can’t say that I was aware of what their
future might bring. Perhaps now its day has come, and that’s what this book is all
about.
Robert W. Lucky
Fair Haven, New Jersey
March 2003
Preface

Technologies for fiber-optic telecommunications went through a major growth

period—some might even say a revolution—roughly during the years 1994 to 2000.
This growth came about due to the convergence of several market drivers and tech-
nologies. First were data traffic and the Internet, the key drivers of the demand for
bandwidth. Prior to the explosion of data traffic and the Internet, voice traffic only
grew at an average of 4% a year. The Internet, on the other hand, grew 100% a year
or more starting in 1992 and sustained this phenomenal growth rate at least through
about 2001. The second was the advent of the optical amplifier, which served the
role in optical networks that the transistor had played in the electronics revolution.
The optical amplifier was key because it allowed the simultaneous amplification of
a number of channels, as opposed to electronic regenerators that operated channel
by channel. The third technology was wavelength-division-multiplexing (WDM),
which made a single strand of fiber act as many virtual fibers. WDM has allowed
the capacity of fibers to be increased by more than two orders of magnitude over the
past few years, providing plenty of bandwidth to fuel the growth of data traffic and
the Internet. WDM served the role in optical networks that integrated circuits had
played in the electronics revolution. Just as the transistor permitted the revolution
associated with integrated circuits in electronics, the optical amplifier permitted the
revolution associated with WDM in optical networks. Because a number of channels
could be simultaneously amplified, the cost of deploying more wavelengths in WDM
was gated by the terminal end costs rather than the regenerator costs. Hence far more
cost-effective networks became available with the combination of optical amplifiers
and WDM.
Raman amplification has been one of the optical amplifier technologies that had
a slow start, but then experienced a wide deployment with increasing performance
needs of optical networks. It would be reasonable to assume that almost every new
or upgraded long-haul (∼300 to 600 km between regenerators) and ultra-long-haul
(>600 km between regenerators) will eventually deploy some form of Raman ampli-
fication technology. Any deployment concerns about discrete or distributed Raman
amplification have been outweighed by the performance improvements permitted
with Raman amplification. For example, distributed Raman amplification improves
xii Preface

noise performance and decreases nonlinear penalties in WDM networks, thereby al-
leviating the two main constraints in dispersion-compensated, optically amplified
systems. The improved noise performance can be used to travel longer distances be-
tween repeaters or to introduce lossy switching elements such as optical add/drop
multiplexers or optical cross-connects. Discrete and distributed Raman amplifiers
are wavelength agnostic, with the gainband being determined by the pump distribu-
tion. Also, discrete Raman amplification can efficiently be integrated with dispersion
compensation. Hence, Raman amplification permits wide bandwidth and long reach
simultaneously. For instance, commercial systems in 2002 provide 240 channels at 10
Gb/s over 100 nm bandwidth (capacity of 2.4 Tb/s) over 1500 km with static optical
add/drop multiplexers at every inline amplifier site (roughly every 80 km). Of course,
if less bandwidth is required, then the unrepeated distance can be even longer.
Although stimulated Raman scattering (SRS) dates back to 1928 [7], it was first
studied in optical fibers by Roger Stolen and coworkers in 1972 [10, 9]. Much of
the physics of Raman amplification was explored through the 1970s and early 1980s.
Then, in 1984 Linn Mollenauer, Jim Gordon, and I suggested the use of Raman
amplification in WDM soliton systems [5, 6]. We demonstrated the concept using
color center lasers as the pump lasers, and there was a flurry of research on Raman
amplification in fiber systems from about 1984 to 1988. However, by 1988 it became
clear that erbium-doped fiber amplifiers (EDFAs) were closer to practical deployment,
and the Raman work was mostly dropped in favor of rare earth-doped amplifiers.
Admittedly, it was a bit difficult to imagine how to put large tabletop lasers such as
color center lasers in a central office, in a hut, or under the sea.
Although it seemed that EDFAs would never be displaced as optical amplifiers in
fiber-optic systems, by 1997 the scene began to change. Work particularly at Lucent
Technologies by Steve Grubb, Per Hansen, Andy Stentz, and others began to show the
promise of Raman oscillators pumped by cladding-pumped fiber lasers [1, 2, 4, 3, 8].
I realized the big payoff would be not in oscillators but in Raman amplifiers, and I
spunoff from the University of Michigan a startup company called Xtera Commu-
nications. Xtera began by trying to develop S-band subsystems (Chapter 10), and
later we redirected the business plan to a wideband, long-reach, all-Raman system
(Chapter 14).
The key thing to understand is that stimulated Raman amplification had not
changed. It was the technology required to make Raman amplifiers that had changed.
The most fundamental change was the development and commercialization of prac-
tical, high-powered, laser diodes. Although we believed that commercial Raman am-
plifiers would require cladding-pumped fiber lasers, by 1999 it became clear that laser
diodes with sufficient power would be available. This was an extremely important
development because it would reduce the cost and size of Raman amplifiers while
increasing their reliability. Second, dispersion-compensating fibers (DCF) were being
commercialized for use with 10 Gb/s systems. It turns out that the DCF is an excellent
gain medium for discrete Raman amplifiers, permitting the integration of dispersion
compensation with optical amplification. Finally, all of the required passive compo-
nents became available at least with fiber pigtails and with the ability to handle high
pump powers.
 Preface xiii

By 2000, Raman amplified systems were becoming commercially available. Sev-

eral startup companies (e.g., Corvis, Qtera, Xtera) were using Raman amplification
as their differentiator. Even the stalwarts of the industry had to take notice of this im-
portant technological development. For example, Nortel Networks acquired Qtera,
and Lucent Technologies began to develop their all-Raman system, which became
commercially available in 2002. It finally looked as if Raman amplification had made
inroads in long-haul and ultra-long-haul systems. The noise figure improvement of
up to 7 dB was simply too large a system margin to ignore!
With heavy research and development of Raman amplified systems between 1997
and 2002, it would be fair to say that the physics and applications of Raman amplifiers
were pretty well understood, at least in the arena of telecommunications. The Raman
“buzz” was out there, and telecommunications engineers were constantly asking for
a “good reference” that they could read to understand Raman amplification. Raman
amplifiers were about to leave the eclectic world of research laboratories and PhDs
and perhaps enter the commercial Main Street, and a book that summarized the key
physical principles and applications was needed. After all, it would be difficult to
deploy and maintain that which was unknown. Therefore, at OFC 2002 I finally
agreed to put together this volume, Raman Amplifiers for Telecommunications.
For me the assembling of this book is another important step on a long journey.
As a graduate student and when I first joined AT&T Bell Labs between 1983 and
1987, we were convinced that Raman amplification was going to be important. At
MIT I worked with Linn Mollenauer and Jim Gordon on WDM soliton systems using
Raman amplification, and then when I first joined Bell Labs I worked on Raman
oscillators and amplifiers. Almost a decade later, I spent five years from 1997 to
2002 transferring to commercialization an all-optical, all-Raman amplified system
through Xtera Communications. Now that Raman amplification is finally prime-time
for systems, it is necessary to organize, articulate, and share the know-how so that
telecommunications and systems engineers can deploy and exploit the technology.

Acknowledgments
This book was written and assembled while I was at Xtera Communications, on a
leave of absence from the University of Michigan in Ann Arbor, as well as the first
year after I returned to the University. Thanks are due to Professors Richard Brown
and Duncan Steel at the University for permitting me to complete this book. Also, I
am particularly indebted to Dr. Jon Bayless and Carl DeWilde for encouraging me to
put this book together and having the foresight to understand the broader impact that
a startup could have by allowing this endeavor.
Many at Xtera Communications worked on Raman amplification and helped in
composing this volume. In particular, Amos Kuditcher helped significantly on Chap-
ters 10 and 14. Special thanks are due to Monica Villalobos, without whose help
this book could never have been written. Monica kept the process moving forward
throughout the year with her usual methodical and professional style. She kept contact
with all of the authors, collected all of the chapters, and helped in the hand-off to the
xiv Preface

publishers. I think that all of my coauthors would agree that Monica was a pleasure
to work with throughout the process.
Finally, I am deeply appreciative of the love, support, and encouragement from
my wife, Nasreen, and sons, Sabir and Shawn. The only regret I have in putting this
book together is the time it took away from my family.

Mohammed N. Islam
Ann Arbor, Michigan
January 2003

References
[1] S. Grubb, T. Erdogan, V. Mizrahi, T. Strasser, W.Y. Cheung, W.A. Reed, P.J. Lemaire, A.E.
Miller, S.G. Kosinski, G. Nykolak, and P.C. Becker, 1.3 µm cascaded Raman amplifier
in germanosilicate fibers. In Proceedings of Optical Amplifiers and Their Applications,
PD3-1, 187, 1994.
[2] S.G. Grubb, T. Strasser, W.Y. Cheung, W.A. Reed, V. Mizrachi, T. Erdogan, P.J. Lemaire,
A.M. Vengsarkar, D.J. DiGiovanni, D.W. Peckham, and B.H. Rockney, High-power 1.48
µm cascaded Raman laser in germanosilicate fibers, In Proceedings of Optical Amplifiers
and Their Applications, SA4, 197–199, 1995.
[3] P.B. Hansen, L. Eskildsen, S.G. Grubb, A.J. Stentz, T.A. Strasser, J. Judkins, J.J. DeMarco,
R. Pedrazzani, D.J. DiGiovanni, Capacity upgrades of transmission systems by Raman
amplification, IEEE Photon. Technol. Lett., 9:2 (Feb.), 1997.
[4] P.B. Hansen, L. Eskildsen, S.G. Grubb, A.M. Vengsarkar, S.K. Korotky, T.A. Strasser,
J.E.J. Alphonsus, J.J. Veselka, D.J. DiGiovanni, D.W. Peckham, D. Truxal, W.Y. Cheung,
S.G. Kosinski, and P.F. Wysocki, 10 Gb/s, 411 km repeaterless transmission experiment
employing dispersion compensation and remote post- and pre-amplifiers. In Proceedings
of the 21st European Conference on Optical Communications (Gent, Belgium), 1995.
[5] L.F. Mollenauer, J.P. Gordon, and M.N. Islam, Soliton propagation in long fibers with
periodically compensated loss, IEEE J. Quantum Electron. QE-22:157–173, 1986.
[6] L.F. Mollenauer, R.H. Stolen, and M.N. Islam, Experimental demonstration of soliton
propagation in long fibers: Loss compensated by Raman gain, Opt. Lett. 10:229–231,
1985.
[7] C.V. Raman and K.S. Krishnan, A new type of secondary radiation, Nature 121:3048,
501, 1928.
[8] A.J. Stentz, S.G. Grubb, C.E. Headley III, J. R. Simpson, T. Strasser, and N. Park, Raman
amplifier with improved system performance, OFC ’96 Technical Digest, TuD3, 1996.
[9] R.H.Stolen and E.P. Ippen, Raman gain in glass optical waveguides, Appl. Phys. Lett.,
22:276, 1973.
[10] R.H. Stolen, E.P. Ippen, and A.R. Tynes, Raman oscillation in glass optical waveguide,
Appl. Phys. Lett., 20:62, 1972.
Quick Summary of Book

The book is organized into two volumes with three sections. Volume 1 begins with
a chapter entitled “Overview of Raman Technologies for Telecommunications.” The
first major section (Volume 1, Section A), Raman Physics, contains eight chapters
(Chapters 2–9). The second section (Volume 2, Section B) on Subsystems and Mod-
ules contains five chapters (Chapters 10–14). Finally, the third section (Volume 2,
Section C), Systems Design and Experiments, contains an additional five chapters
(Chapters 15–19). Almost half of the book is dedicated to Raman physics, because
these are the principles that will remain time invariant. This is covered completely in
Volume 1. The second section, Subsystems and Modules, describes applications of
Raman technology that will be fairly time invariant, although the details and data of
the applications will continuously evolve. Finally, the last section, Systems Design
and Experiments, represents a snapshot of the state-of-the-art system demonstrations
as of early 2003. This is the section that must necessarily change with time, but at least
it can provide a good basis for comparison or updating from 2003. It is important to
go all the way from basic physics to systems because they are intimately linked. The
basic physics determines what can or cannot be done, and it points to the differential
advantages that Raman amplification provides. On the other hand, the systems design
and experiments ultimately define what is worth doing and where performance should
be optimized. Fortunately, Raman amplification is very rich with physical principles
as well as being one of the key enabling technologies for long-haul and ultra-long-haul
submarine, terrestrial, soliton, and high-speed systems.
In selecting the topics to be covered in this book as well as the authors to invite, a
broad, diverse, and insightful view was sought. As an example, the authors were cho-
sen from industrial labs as well as universities. The industrial laboratories represented
include Corning, Furukawa, Lucent Technologies, Nippon Telephone and Telegraph,
Nortel Networks, OFS Fitel, Siemens, Tyco Telecommunications, and Xtera Commu-
nications. Also, the authors represent the international nature of Raman technology,
with contributions from the United States, Europe, Japan, and Russia. Furthermore,
young rising stars were invited to contribute chapters as well as the “giants in the
field,” starting with Roger Stolen, Linn Mollenauer, and Evgeny Dianov.
xvi Quick Summary of Book

It is an honor that so many key researchers in Raman technology accepted the

invitation to contribute to this book. The invitation was extended to researchers who
have made significant contributions to the technology and whose work has consis-
tently represented the highest quality and deepest insight. Obviously there are many
other excellent researchers in the field, but the intent was to cover the main issues in
Raman physics, subsystems and modules and systems design and experiments within
the limited space of two volumes.
The book begins with “Overview of Raman Technologies for Telecommunica-
tions,” which I authored (Chapter 1). Then, the physics section opens with a chapter
by Stolen, “Fundamentals of Raman Amplification in Fibers” (Chapter 2), which is
fitting since he did much of the original groundbreaking work on Raman amplifica-
tion in fibers. Noise is a very important aspect of any optical amplifier, and Fludger
contributes two chapters on the topic: “Linear Noise Characteristics” (Chapter 4) and
“Noise due to Fast Gain Dynamics” (Chapter 8). The most significant technological
development for commercial Raman amplifiers is the increase in laser diode power,
and Namiki et al., in Chapter 5 describe “Pump Laser Diodes and WDM Pump-
ing.” The other major technological development is the availability of new fibers
with efficient Raman gain, and two chapters are dedicated to this topic: in Chapter
6 Grüner-Nielsen and Qian describe dispersion compensating fibers for Raman ap-
plications, and in Chapter 7 Dianov describes more forward-looking work on new
Raman fibers. The simplest Raman amplifier uses CW pumps and a counterpropa-
gating geometry (i.e., where the pump and signal propagate in opposite directions).
However, the performance of this basic Raman amplifier can be improved by a num-
ber of emerging techniques. In Chapter 3, Grant and Mollenauer describe the use of
time-division multiplexing of pump wavelengths. Then, in Chapter 9 Radic discusses
and compares forward, bidirectional, and higher-order Raman amplification.
In the second section, “Subsystems and Modules,” four types of Raman devices
are covered: discrete (or lumped) amplifiers, distributed amplifiers, lasers, and a com-
bination of discrete and distributed amplifiers. In Chapter 10 I review work on discrete
or lumped Raman amplifiers to open up new wavelength windows, particularly in the
short wavelength S-band. Then, Headley et al. review in Chapter 11 work on Ra-
man fiber lasers or oscillators. Next, in Chapter 12, Evans et al. discuss distributed
Raman transmission, applications, and fiber issues. One of the most important appli-
cations of combined discrete and distributed amplifiers is to broadband transmission
systems. One way to achieve the broadband amplifier is to combine erbium-doped
fiber amplifiers with Raman amplifiers, and in Chapter 13 Masuda describes hybrid
EDFA/Raman amplifiers. Another route to a broadband system is to use all-Raman
discrete and distributed amplifiers, and in Chapter 14 on wideband Raman amplifiers
I along with coworkers at Xtera illustrate this approach.
The third section of the book focuses on system design and experiments. Some of
the challenges of the Raman effect are covered in the first two chapters, and system
deployments of Raman amplifiers are discussed in the following three chapters. In
Chapter 15 Bromage et al. detail multiple path interference and its impact on system
design. Then, in Chapter 16 Krummrich discusses Raman impairments in WDM sys-
tems. As examples of areas where Raman amplifiers are a key enabling technology,
three system experiments are included. First, in Chapter 17 Kidorf et al. describe
 Quick Summary of Book xvii

the use of Raman amplifiers in ultra-long-haul submarine and terrestrial applica-

tions. Then, in Chapter 18 Mollenauer discusses ultra-long-haul, dense WDM using
dispersion-managed dolitons in an all-Raman system. Finally, in Chapter 19 Nelson
and Zhu illustrate 40 Gb/s Raman-amplified transmission.

Survey of Chapters

VOLUME 90/1

Overview of Raman Amplification in Telecommunications (Chapter 1)

As an overview for the book, this chapter surveys Raman amplification for telecom-
munications. It starts with a brief review of the physics of Raman amplification
in optical fibers, along with the advantages and challenges of Raman amplifiers. It
also discusses some of the recent technological advances that have caused a revived
interest in Raman amplifiers. Then, distributed and discrete Raman amplifiers are
described. Distributed Raman amplifiers improve the noise figure and reduce the
nonlinear penalty of the amplifier, allowing for longer amplifier spans, higher bit
rates, closer channel spacings, and operation near the zero dispersion wavelength.
Discrete Raman amplifiers are primarily used to increase the capacity of fiber-optic
networks. Examples of discrete amplifiers are provided in the 1310 nm band, the 1400
nm band, and the short-wavelength S-band.

Section A. Raman Physics

Fundamentals of Raman Amplification in Fibers (Chapter 2)

Raman scattering was first published by C.V. Raman in 1928, and he was awarded
the 1930 Nobel Prize for the discovery. In 1972 stimulated Raman scattering was first
observed in single-mode fibers, and the Raman gain coefficient was also measured
that same year. The chapter focuses on various treatments of the Raman interaction,
which can appear to be quite different. The quantum approach treats the problem as a
transition rate involving photon number. In the classical approach, the Raman effect
is a parametric amplifier with an interaction between signal, pump, and vibrational
wave. Finally, the Raman interaction itself can be traced to a small time delay in
the nonlinear refractive index. This chapter compares and contrasts these various
treatments of the Raman effect in optical fibers. Also, a fundamental treatment of
noise in fiber Raman amplifiers is included.

Time-Division Multiplexing of Pump Wavelengths (Chapter 3)

This chapter describes an approach to Raman pumping that uses time-division mul-
tiplexing of the pump wavelengths. TDM pumping has several advantages over CW
xviii Quick Summary of Book

pumping such as efficient gain leveling with a “smart” pump, the elimination of
four-wave mixing between pumps, and the reduction of pump-to-pump Raman inter-
actions. This technique only works with backward Raman pumping, where the pump
and signal are counterpropagating. The rate of TDM pumping needs only to exceed
1 MHz, so electronic components for these speeds are widely available and very in-
expensive. However, TDM Raman pumping does introduce a new set of problems.
The higher gain for signal propagating in the backward direction leads to a larger
backward spontaneous Raman noise level. Consequently, Rayleigh scattering of the
backward propagating noise can significantly increase the forward noise level under
high gain conditions.

Linear Noise Characteristics (Chapter 4)

Spontaneous emission is the inevitable consequence of gain in an optical amplifier. In

this chapter, the definition of noise figure is shown to be useful only in characterizing
shot noise and signal-spontaneous beat noise. The noise characteristics of both discrete
and distributed Raman amplifiers are presented. Also, a general model that accurately
predicts both signal propagation and the buildup of amplified spontaneous emission
is discussed and compared to measurements. Further measurements and analysis of
broadband Raman amplifiers show a clear dependence on temperature, which places a
fundamental limit on their performance. Interactions between the pump wavelengths
are also shown to play an important role, giving better system performance to longer
signal wavelengths at the expense of the shorter wavelengths. Finally, an analysis of
the relative linear noise performance of different transmission fibers is presented.

Pump Laser Diodes and WDM Pumping (Chapter 5)

This chapter discusses issues surrounding the pump laser diodes for broadband Ra-
man amplifiers, which range from fundamentals to industry practices of Raman pump
sources based on so-called 14XX nm pump laser diodes. The chapter also describes
the design and issues of wavelength-division-multiplexed pumping for realizing a
broad and flat Raman gain spectrum over the signal band. In addition, practical Ra-
man pump units are illustrated, and the chapter also provides insights into ongoing
issues on copumped Raman amplifiers and their pumping sources. The pump laser
diodes discussed are InGaAsP/InP GRIN-SCH strained layer MQW structure with
BH structure, which are the most widely used in the industry.

Dispersion-Compensating Fibers for Raman Applications (Chapter 6)

Dispersion-compensating fibers are the most widely used technology for dispersion
compensation. Also, DCF is a good Raman gain medium, due to a relatively high
germanium doping level and a small effective area. Dispersion-compensating Raman
amplifiers integrate two key functions, dispersion compensation and discrete Raman
amplification, into a single component. Use of DCF for broadband Raman amplifiers
 Quick Summary of Book xix

raises new requirements for the properties of the DCF including requirements for
gain, double Rayleigh scattering, and broadband dispersion compensation. Dispersion
slope compensation is now possible for all types of transmission fibers, and the next
challenge for broadband dispersion compensation is dispersion curvature. Dispersion-
compensating Raman amplifiers have been realized with high-gain, low-noise figure
and low multipath interference arising from double Rayleigh back scattering.

New Raman Fibers (Chapter 7)

Standard transmission fibers with silica core doped with a small concentration of
GeO2 have a low value of the Raman gain and a peak Raman gain at a frequency shift
of about 440 cm−1 . However, for a number of applications such as discrete Raman
amplifiers and Raman fiber lasers, special fibers with much higher Raman gain and/or
various Raman frequency shifts are often required. Early experiments show that low-
loss, high GeO2 - and P2 O5 -doped silica fibers could be promising fiber gain Raman
fibers. For example, phosphor-silicate glass has two Raman scattering bands shifted
by 650 cm−1 and 1300 cm−1 , and the cross-section for these bands is 5.7 and 3.5 times
higher compared to silica. However, these fibers have met with serious challenges
during fabrication by well-developed techniques. Nonetheless, at present germano-
silicate and phosphor-silicate Raman fibers are being widely used for constructing
CW Raman fiber lasers, which can cover the whole spectral range of 1.2 to 1.75
microns. These CW medium power (1 to 10 W) lasers are a convenient laser source
for pumping optical fiber amplifiers and some lasers.

Noise due to Fast Gain Dynamics (Chapter 8)

The time response of the Raman effect is associated with the vibrations of the
molecules in the gain medium and is on the order of several hundred femtoseconds.
Compared to current data rates, this energy transfer is practically instantaneous, re-
sulting in very fast gain dynamics. In a copumped Raman amplifier, the gain dynamics
are averaged due to chromatic dispersion between pump and signal wavelengths. This
lessens the impact of the fast physical process and results in a more improved system
performance than would otherwise be expected. In a counterpumped Raman ampli-
fier, the different propagation directions of pump and signals average the gain over
the cavity length. This much stronger averaging greatly reduces system penalties in
counterpumped amplifiers. In this chapter models are developed for co- and counter-
pumped Raman amplifiers that quantify both the transfer of relative intensity noise
from the pump to the signal and also the signal-to-signal crosstalk, mediated by the
pump (crossgain modulation). In addition, the system impact in terms of Q penalty is
determined, as well as determining the actual energy transfer from pumps to signals
and from crossgain modulation.

Forward, Bidirectional, and Higher-Order Raman Amplification (Chapter 9)

Distributed Raman amplification can be achieved by optical pumping at either end of
the fiber. In a unidirectional transmission line, all signals travel in the same direction.
xx Quick Summary of Book

In contrast, bidirectional transmission can be used to realize two-way traffic within

a single fiber line: counterpropagating signal traffic is launched and received at the
opposite ends of the optical link. A bidirectionally pumped fiber span can support both
uni- and bidirectional signal transmission. A unidirectionally pumped span, however,
almost exclusively supports unidirectional signal traffic. This chapter explores and
compares forward, bidirectional, and higher-order Raman amplification. Higher-order
pumping refers to the introduction of shorter-wavelength pumps that are used to
pump the pump; that is, the higher-order pump amplifies the first-order pump, which
in turn pumps the signal band. Different pumping schemes provide different levels
of performance, but each scheme has a trade-off of performance versus pump laser
restrictions.

VOLUME 90/2

Section B. Subsystems and Modules

S-Band Raman Amplifiers (Chapter 10)

The design, implementation, and issues associated with S-band amplification are dis-
cussed in this chapter, with a special emphasis on lumped Raman amplifiers (LRAs).
LRAs can be used in a split-band augmentation strategy with new or already deployed
C- and/or L-band systems, which are usually amplified with EDFAs. To open up the
S-band, the key enabling technology is the appropriate optical amplifier. Raman am-
plifiers appear to be a practical solution to the S-band amplifier, and they are a mature
technology ready for deployment. Utilizing silica fiber as the gain medium, Raman
amplifiers can be readily fusion spliced with the fiber used in the transport infrastruc-
ture. LRAs have also been demonstrated with performance on a par with commercial
C-band EDFAs in terms of gain, noise figure, and bandwidth. In addition, LRAs can
be implemented efficiently using DCF, which means that the lumped amplifier can be
integrated with the dispersion compensation. The major challenge of Raman ampli-
fiers has been their lower efficiency than EDFAs, but this discrepancy is narrowing
through better gain fibers, higher laser diode pump powers, and the inherent better
slope efficiency for Raman amplifiers at higher channel count. The bulk of this chapter
focuses on the issues and experimental demonstration of S-band LRAs in fiber-optic
transmission systems.

Raman Fiber Lasers (Chapter 11)

This chapter focuses on cascaded Raman fiber lasers (RFL), which use the stimulated
Raman scattering in optical fibers to shift the wavelength of light from an input pump
laser to another desired wavelength. Devices at almost any wavelength can be made
by proper choice of a pump wavelength, and by cascading the pump through several
Raman shifts. Although RFL had been demonstrated since the 1970s, the advent of
 Quick Summary of Book xxi

fiber Bragg gratings made the devices practical. A broadband flat Raman gain profile
can be obtained using multiple pump wavelengths, and it is advantageous to have all
the required wavelengths emitted from one source. This motivated the development
of a multiple wavelength RFL. Single cavities simultaneously lasing from two to
six wavelengths have been demonstrated. Finally, distributed Raman amplification
techniques have become more sophisticated with the proposed use of higher-order
pumping schemes. The use of a RFL is especially suited to this application, inasmuch
as large amounts of powers are required at the shortest pump wavelength.

Distributed Raman Transmission: Applications and Fiber Issues (Chapter 12)

The persistent demand for higher performance (capacity, system reach, data rate,
etc.) has turned system designers to distributed Raman for its lower noise figure.
Today’s data-dominated traffic patterns require reach beyond 1000 km, and Raman
amplification is one vital tool in pushing out the system reach. This chapter reviews
the two most important properties of an optical amplifier—pump efficiency and noise
figure—and compares Raman to erbium amplification. The concept of effective noise
figure is covered, which leads to a generic system scaling relationship that aids in the
prediction of Raman-assisted, system performance improvements. Raman transmis-
sion experiments at 10 Gb/s and 40 Gb/s are summarized, and design issues specific to
these systems are covered. In addition, dispersion-managed fiber consisting of optical
fiber spans that can be optimized for Raman transmission is introduced.

Hybrid EDFA/Raman Amplifiers (Chapter 13)

This chapter describes the technologies needed for cascading an EDFA and a fiber
Raman amplifier to create a hybrid amplifier, the EDFA/Raman hybrid amplifier. Two
kinds of hybrid amplifiers are defined in this chapter: the “narrowband hybrid ampli-
fier,” and the “seamless and wideband hybrid amplifier.” The narrowband amplifier
employs distributed Raman amplification in the transmission fiber together with an
EDFA; this provides low-noise transmission in the C- or L-band. The noise figure
of the transmission line is lower than it would be if only an EDFA were used. The
wideband amplifier, on the other hand, employs distributed or discrete Raman am-
plification together with an EDFA. The wideband amplifier provides a low-noise and
wideband transmission line or a low-noise and wideband discrete amplifier for the C-
and L-bands. The typical gain bandwidth of the narrowband amplifier is ∼30 to 40
nm, whereas that of the wideband amplifier is ∼70 to 80 nm.

Wideband Raman Amplifiers (Chapter 14)

This chapter describes the design and implementation of wideband Raman ampli-
fiers. All-Raman amplification enables the lowest cost and smallest footprint system,
and Raman amplification provides a simple single platform for long-haul and ultra-
long-haul fiber-optic transmission systems. Despite a significant list of advantages, a
xxii Quick Summary of Book

number of challenges exist for Raman amplification, including: pump–pump interac-

tions, interband and intraband Raman gain tilt, noise arising from thermally induced
phonons near the pump wavelengths, multipath interference from double Rayleigh
scattering, coupling of pump fluctuations to the signal, and pump-mediated signal
crosstalk. Fortunately, design techniques exist for overcoming all of these physical
limitations, thus allowing for the relatively simple implementation of 100 nm Raman
amplifiers. Although commercially available wideband Raman amplifiers have been
limited to a bandwidth of 100 nm to date, laboratory experiments have shown am-
plifiers with much larger bandwidths. Bandwidths greater than 100 nm are usually
achieved with such special techniques as new glass compositions or wavelength guard
bands around the pump wavelengths. Finally, the application of wideband Raman am-
plification in high-performance transmission systems is reviewed. For example, an
all-Raman amplifier structure with discrete and distributed amplification can give
significant advantages of reach and capacity. Such a design has been implemented,
and the transmission feasibility of 240 OC-192 channels over 1565 km standard
single-mode fiber has been demonstrated

Section C. Systems Design and Experiments

Multiple Path Interference and Its Impact on System Design (Chapter 15)

Up to the end of the 1990s, the main causes of signal degradation in transmission were
fiber nonlinearity and amplified spontaneous emission from optical amplifiers. With
the advent of Raman amplification in fiber-optic communications systems, another
source of signal degradation has become increasingly relevant: so-called multiple
path interference or MPI. This chapter focuses on MPI and its impact on receiver and
system design. Optical amplification can exacerbate MPI by providing gain for paths
that would otherwise have too much attenuation to be significant. Sources of MPI
include discrete reflections within or surrounding optical amplifiers, double Rayleigh
scattering in optical amplifiers or in the transmission span, and unwanted transverse
mode mixing in higher-order mode dispersion compensators. The properties of MPI
and Rayleigh scattering are reviewed, and the techniques for measuring MPI level
are then described. The impact of MPI on beat-noise limited receivers is discussed,
along with techniques for system design optimization.

Raman Impairments in WDM Systems (Chapter 16)

In most chapters of this book, stimulated Raman scattering is invoked intentionally.

Pump radiation is coupled into the fiber carrying the signal radiation to generate Ra-
man gain. However, SRS also occurs unintentionally in WDM transmission systems.
Due to the large number of channels inside the Raman gain bandwidth, total power
can add up to levels where considerable amounts of SRS are generated, with the signal
channels acting as pumps. In contrast to the beneficial effects of intentional Raman
pumping, the unintended generation of SRS usually degrades system performance.
This chapter addresses effects resulting from the unintended invocation of SRS and
 Quick Summary of Book xxiii

their impact on WDM signal transmission. A number of system impairments result

from the interaction between signal channels due to SRS. Effects with time scales
well below the bit period affect the mean values of the individual channel powers. On
the other hand, fast interactions between individual bits change the variances of the
respective channel powers and can be considered as noise. In addition, some selection
criteria for transmission fibers with respect to Raman efficiency are provided.

Ultra-Long-Haul Submarine and Terrestrial Applications (Chapter 17)

Ultra-long-haul (ULH) optically amplified transmission systems (defined in this chap-
ter as those spanning from 1500 to 12,000 km) are some of the most technically
challenging systems designed today. Undersea cable systems require ULH, inasmuch
as the distance across the Atlantic Ocean is approximately 6000 km and the distance
across the Pacific Ocean is approximately 9000 km. For terrestrial networks, the ULH
networks are needed because of the change in the nature of the traffic. Until a few years
ago, voice traffic dominated the network, and a span distance of 600 km satisfied more
than 60% of the connections for voice traffic. However, with the Internet dominating
the traffic now, a span distance of 3000 km is required to satisfy 60% of the connec-
tions for Internet traffic. In terrestrial systems, the marriage of Raman amplification
technology and EDFAs has demonstrated great benefit by expanding the bandwidth of
amplifiers, extending the distance between amplifiers, and allowing longer distances
to be spanned. For submarine systems where the systems are designed to achieve a
desired capacity over often the longest transmission distances, shorter span length
(than for terrestrial systems) often has to be chosen. For such shorter spans (∼50
km), the benefits of Raman amplification are not nearly as substantial. Presently, the
most promising candidate use of Raman amplification in submarine systems is the
wideband hybrid Raman–EDFA. For systems that require a very wide bandwidth this
seems like an attractive way to more than double the transmission bandwidth without
doubling the component count.

Ultra-Long-Haul, Dense WDM Using Dispersion-Managed Solitons in an

All-Raman System (Chapter 18)
In an all-Raman-amplified system, dispersion-managed solitons can provide for dense
WDM, uniquely compatible with all-optical terrestrial networking, robust and error-
free over many thousands of kilometers. This chapter discusses various aspects of
system design, including optimal dispersion maps, nonlinear and noise penalties, and
typical dense WDM system performance. For example, dispersion-managed solitons
are described as well as their special, periodic pulse behavior, their advantages over
other transmission modes, and the conditions required to create and to maintain them.
Also studied is one serious nonlinear penalty they suffer, viz. the timing jitter from
collisions with solitons of neighboring channels. Dispersion-managed solitons, in
an all-Raman, dense WDM system at 10 Gb/s per channel, makes a natural and
comfortable fit with existing terrestrial fiber spans and can provide for transmission
that is robust and error-free out to distances of 7000 km or more. In addition, such
transmission is uniquely suited to provide the backbone of an all-optical network.
xxiv Quick Summary of Book

40 Gb/s Raman-Amplified Transmission (Chapter 19)

High-capacity 40 Gb/s transmission systems offer scalable solutions for future traf-
fic growth in the core network. The challenges of 40 Gb/s systems include optical
signal-to-noise ratio, fine-tuned dispersion compensation, and polarization mode dis-
persion. Raman amplification is likely to be a key driver to ease the noise performance
and increase the available bandwidth for 40 Gb/s DWDM systems. New fiber tech-
nologies provide high system performance and enable a simple and cost-effective
dispersion-compensation scheme. More system margin can also be expected from
high-coding-gain forward error correction. Optimized modulation formats and high-
speed optoelectronics will make practical deployment of 40 Gb/s DWDM systems
possible, facilitating multiple terabit transmission over Mm distance at low cost-per-
bit-per-kilometer. The challenges of DWDM transmission at 40 Gb/s are addressed
in this chapter, along with the technologies enabling 40 Gb/s terrestrial transmission.
Also described are advanced experiments and demonstrations at 40 Gb/s using Raman
amplification.
Contents

Foreword . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii

Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi

Quick Summary of Book . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv

Contributors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxvii

VOLUME 90/1

1. Overview of Raman Amplification in Telecommunications

Mohammed N. Islam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

Section A. Raman Physics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

2. Fundamentals of Raman Amplification in Fibers

R.H. Stolen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

3. Time-Division-Multiplexing of Pump Wavelengths

A.R. Grant and L.F. Mollenauer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

4. Linear Noise Characteristics

C.R.S. Fludger . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

5. Pump Laser Diodes and WDM Pumping

Shu Namiki, Naoki Tsukiji, and Yoshihiro Emori . . . . . . . . . . . . . . . . . . . . . . 121

6. Dispersion-Compensating Fibers for Raman Applications

L. Grüner-Nielsen and Y. Qian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
7. New Raman Fibers
Evgeny M. Dianov . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
xxvi Contents

8. Noise due to Fast-Gain Dynamics

C.R.S. Fludger . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213

9. Forward, Bidirectional, and Higher-Order Raman Amplification

Stojan Radic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291

VOLUME 90/2

Section B. Subsystems and Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301

10. S-Band Raman Amplifiers

Mohammed N. Islam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
11. Raman Fiber Lasers
C. Headley, M. Mermelstein, and J.C. Bouteiller . . . . . . . . . . . . . . . . . . . . 353

12. Distributed Raman Transmission: Applications and Fiber Issues

Alan Evan, Andrey Kobyakov, and Michael Vasilyev . . . . . . . . . . . . . . . . . 383

13. Hybrid EDFA/Raman Amplifiers

Hiroji Masuda . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413

14. Wideband Raman Amplifiers

Mohammed N. Islam, Carl DeWilde, and Amos Kuditcher . . . . . . . . . . . . . 445

Section C. Systems Design and Experiments . . . . . . . . . . . . . . . . . . . . . . . . 491

15. Multiple Path Interference and Its Impact on System Design

J. Bromage, P.J. Winzer, and R.-J. Essiambre . . . . . . . . . . . . . . . . . . . . . . . 491

16. Raman Impairments in WDM Systems

P. M. Krummrich . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 569

17. Ultra-Long-Haul Submarine and Terrestrial Applications

Howard Kidorf, Morten Nissov, and Dmitri Foursa . . . . . . . . . . . . . . . . . . 595
18. Ultra-Long-Haul, Dense WDM Using Dispersion-Managed Solitons
in an All-Raman System
Linn F. Mollenauer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 627

19. 40 Gb/s Raman-Amplified Transmission

L. Nelson and B. Zhu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 673

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 723
Contributors

Jean-Christophe Bouteiller Yoshihito Emori

OFS Laboratories Fitel Photonica Laboratory
25 Schoolhouse Road Optical Subsystem Department
Somerset, NJ 08873 6 Yawata Kaigan Dori
USA Ichihara 290-8555
Email: [email protected] Japan
Email: [email protected]
Jake Bromage
OFS René-Jean Essiambre
Crawford Hill Laboratories Lucent Technologies
791 Holmdel-Keyport Road Bell Laboratories
Holmdel, NJ 07733 Crawford Hill Laboratory
USA Room HOH L-129
Email: [email protected] 791 Holmdel-Keyport Road
Holmdel, NJ 07733
Carl DeWilde USA
Xtera Communications, Inc. Email: [email protected]
500 West Bethany Drive
Alan F. Evans
Suite 100
Optical Physics
Allen, TX 75013
Science and Technology
USA
Corning Incorporated
Email: [email protected]
SP-AR-02-4
Evgeny M. Dianov Corning, NY 14831
A.M. Prokhorov General USA
Physics Institute Email: [email protected]
Russian Academy of Sciences Chris R.S. Fludger
38 Vavilov Str. Nortel Networks, UK
119991 Moscow London Road
Russia Harlow, Essex CM17 9NA
Email: [email protected] UK
Email : [email protected]
xxviii Contributors

Dmitri Foursa Howard Kidorf

Tyco Telecommunications 82 Tower Hill Drive
250 Industrial Way West Red Bank, NJ 07701
Eatontown, NJ 07724 USA
USA Email: [email protected]
Email: [email protected]
Andrey Kobyakov
Andrew R. Grant Science and Technology
Lightwave Systems Research Corning Incorporated
Bell Laboratories SP-DV-02-08
101 Crawfords Corner Road Corning, NY 14831
Room 4C-316 USA
Holmdel, NJ 07733-3030 Email: [email protected]
USA Peter Krummrich
Email: [email protected] Siemens AG, Information and
Communication Networks
Lars Grüner-Nielsen
Hofmannstrasse 51
Specialty Photonics Division
81359 Munich
OFS Denmark
Germany
Priorparken 680
Email: [email protected]
DK-2605 Brøndby
Denmark Amos Kuditcher
Email: [email protected] Xtera Communications, Inc.
500 West Bethany Drive
Clifford E. Headley Suite 100
OFS Laboratories Allen, TX 75013
25 Schoolhouse Road, Room A-16 USA
Somerset, NJ 08873 Email: [email protected]
USA
Email: [email protected] Hiroji Masuda
NTT Network Innovation Laboratories
Mohammed N. Islam 1-1 Hikarino-oka, Yokosuka
Department of Electrical Engineering Kanagawa 239-0847
and Computer Science Japan
University of Michigan at Ann Arbor E-mail: [email protected]
1110 EECS Bldg., 1301 Beal Ave.
Marc Mermelstein
Ann Arbor, MI 48109-2122
OFS Laboratories
USA
25 Schoolhouse Road, Room A-05
Email: [email protected]
Somerset, NJ 08873
and
USA
Xtera Communications, Inc.
Email: [email protected]
500 West Bethany Drive
Suite 100
Allen, TX 75013
USA
Email: [email protected]
 Contributors xxix

Linn F. Mollenauer Stojan Radic

Lucent Technologies Bell Laboratories
Bell Laboratories Lightwave Systems Research
101 Crawfords Corner Road Crawford Hill Laboratory R-231
Room 4C-306 791 Holmdel Keyport Road
Holmdel, NJ 07733 Holmdel, NJ 07733-0400
USA USA
Email: [email protected]
Shu Namiki
Manager Roger H. Stolen
Furukawa Electric Company, Ltd. Department of Electrical
Fitel Photonics Lab,Optical and Computer Engineering
Subsystems Department Virginia Polytechnic Institute
6 Yawata Kaigan Dori and State University
Ichihara 290-8555 Blacksburg, VA 24061
Japan USA
E-mail: [email protected] Email: [email protected]
Naoki Tsukiji
Lynn E. Nelson
Furukawa Electric Company, Ltd.
OFS
Optical Components Department
791 Holmdel-Keyport Road
2-4-3 Okano, Nishi-ku
Room L-137
Yokohama 220-6374
Holmdel, NJ 07733
Japan
USA
Email: [email protected]
Email: [email protected]
Michael Vasilyev
Morten Nissov Corning Incorporated
Transmission Research Photonic Research and Test Center
Tyco Telecommunications 2200 Cottontail Lane
250 Industrial Way West Somerset, NJ 08873
Room 1A225 USA
Eatontown, NJ 07724
USA Pete J. Winzer
Email: [email protected] Lucent Technologies
Thurn & Taxis Str. 10
Yujun Qian Nuremberg 90411
Specialty Photonics Division Germany
OFS Fitel Denmark I/S Email: [email protected]
Priorparken 680
Benyuan Zhu
DK-2605 Brøndby
OFS
Denmark
791 Holmdel-Keyport Road
Email: [email protected]
Room L-111
Holmdel, NJ 07733
USA
Email: [email protected]
This page intentionally left blank
Chapter 10
S-Band Raman Amplifiers

Mohammed N. Islam

10.1. Introduction
In this chapter we focus on the use of discrete or lumped Raman amplifiers in the
short-wavelength S-band. Recent advances in data communications have led to re-
quirements for higher throughput of wavelength-division-multiplexed (WDM) trans-
mission systems. Two approaches have been pursued to address the increasing
throughput requirements, namely, increasing the spectral efficiency of WDM sys-
tems within existing transmission bands, and increasing optical bandwidth to utilize
much more of the low-loss window in silica fiber than presently supported on WDM
systems. The first approach has been pursued through coding and modulation tech-
niques. The second approach to increasing throughput involves making available
new transmission bands beyond those supported by conventional erbium-doped fiber
amplifiers (EDFAs), which presently define the optical bandwidth of most WDM
systems.
The low-loss window of single-mode silica fiber is shown in Fig. 10.1. Erbium-
doped fiber amplifiers provide amplification in the C-band extending from 1530 to
1565 nm and the L-band extending from 1565 to as high as 1625 nm. On the short
wavelength side of the C-band are the S-band stretching from 1480 to 1530 nm and
the S+-band from 1450 to 1480 nm. The longer wavelength band stretching from
1625 to 1675 nm has also been suggested for WDM transmission.
Once the C- and L-bands are exhausted, the most likely band to be exploited is the
S-band, with the potential for increasing the bandwidth of most transmission systems
by 30 to 50 nm. As shown in Fig. 10.1, attenuation in the S-band is comparable to
or better than attenuation in the L-band, and sensitivity to micro- and macrobending
losses in the S-band is far less pronounced. In addition, dispersion of standard single-
mode silica fiber in the S-band is lower than dispersion in the longer wavelength
bands. Figure 10.2 shows dispersion in standard single-mode fiber. S-band dispersion
is approximately 30% less than L-band dispersion in standard single-mode fiber.
Although dispersion in the S-band is generally low, reducing dispersion compensation
requirements, WDM transmission in that band may not be attractive on some fiber
types for which the zero dispersion wavelength falls within the S-band. Because
302 M.N. Islam

Microbending

Macrobending
0.35
Loss (dB/km)

0.25

S+ S C L
0.15
1300 1400 1500 1600 1700
Wavelength (nm)

Fig. 10.1. Loss spectrum of standard single-mode fiber. The gray curves show micro- and
macrobending loss.

15
Dispersion (ps/nm km)

0
C-Band
S-Band

L-Band

–15
1200 1300 1400 1500 1600
Wavelength (nm)

Fig. 10.2. Dispersion of standard single-mode fiber. Dispersion in the S-band is lower than in
the C- and L-bands.
 10. S-Band Raman Amplifiers 303

Applications

Turn-Key Systems

Control Software Monitor Software

Mux /Demux’s ADM’s Switches

Amplifiers Transmitters Filters WDM’s

Fig. 10.3. Components of an optical network. The key technology for unlocking the S-band in
fibers is the optical amplifier.

phase matching of parametric interactions between signals readily occurs near the zero
dispersion wavelength, four-wave mixing may pose a challenge to WDM transmission
on those fiber types. Figure 1.28 in Chapter 1 shows, however, that the zero dispersion
wavelength falls within the S-band only for a few fiber types.
As illustrated in Fig. 10.3, the key technology for unlocking the S-band in fibers
is the optical amplifier. All of the other building blocks are engineering changes on C-
or L-band technologies. A number of S-band amplifiers are being studied, including
rare earth-doped fiber amplifiers, semiconductor optical amplifiers, optical paramet-
ric amplifiers, and Raman amplifiers. Raman amplification has been demonstrated
to be a viable approach to bandwidth extension. Because the spectral region where
significant Raman gain occurs depends primarily on pump wavelength and power,
amplification can be obtained practically anywhere within the low-loss window of
silica fiber. In addition, transmission fiber and dispersion-compensating fiber are both
good Raman gain media. Thus Raman amplification provides a means not only for
compensating losses within the transmission fiber, but also for integrating amplifica-
tion and dispersion compensation within a single module. The challenges that plagued
deployment of Raman amplification and led to its abandonment in favor of EDFAs in
the late 1980s have been resolved, and commercial products have already appeared.
In this chapter the S-band Raman amplified system is studied in detail. Char-
acteristics of alternative amplifier technologies for the S-band are briefly reviewed
in Section 10.2, and the advantages and disadvantages of Raman amplification are
summarized in Section 10.3. The focus of this chapter is on the Raman amplification
approach, and an experimental example of an S-band Raman amplified, long-haul
system is reported in Section 10.4. For the system described, the next few sections
study in greater detail some of the key issues in practical deployment of the S-band in
augmentation of C-band and/or L-band amplifiers. First, in Section 10.5 the Raman
304 M.N. Islam

amplifier characterization and control are illustrated, which are considerably differ-
ent than in EDFAs because the physics of Raman amplifiers and rare earth-doped
amplifiers are significantly different. A major concern of adding the S-band to an
existing C-band system has been the interband interaction, and Section 10.6 stud-
ies the impact on the C-band from the introduction of the S-band. Third, because
dispersion-compensating fiber (DCF) is used as the Raman gain medium because of
the high nonlinear coefficient in DCF, another concern arises from nonlinear penalties
on the signal channels. The dominant nonlinear effect arises from cross-phase modu-
lation (XPM), and Section 10.7 quantifies the XPM penalty on the channels passing
through the S-band amplifier. In addition, because the Raman amplifier is signifi-
cantly longer than EDFA’s, another concern arises from new noise sources such as
double Rayleigh backscattering (DRBS). In Section 10.8 the contribution of DRBS
to multipath interference (MPI) in lumped Raman amplifiers is explained. Finally,
some other S-band system experiments are described in Section 10.9, followed by a
summary and conclusion in Section 10.10.

10.2. Alternative S-Band Amplifier Technologies

Three technologies have been studied extensively in hopes of bringing to the lower
wavelengths of the S-band the same cost-effective means of optical amplification
that EDFAs provide for the C-band [1]. Semiconductor optical amplifiers (SOAs),
thulium-doped fiber amplifiers (TDFAs) utilizing either fluoride or multicomponent
silicates (MCS) as the host fiber, along with lumped, or discrete, Raman amplifiers
(LRAs) have all been proposed as enablers for opening up transmission of wave-
lengths shorter than the traditional 1530 to 1560 nm range of EDFAs.
To ascertain which of these three optical amplification approaches holds the most
promise for enabling S-band transmission, the principles of operation, implementation
challenges, and possible applications for each approach are examined. All of these
technologies have benefits to offer, however, only one provides cost-effective long-
haul transmission of additional wavelength bands. In addition, recent research has
shown the possibility of using EDFAs or optical parametric amplifiers to provide
gain in the S-band. For example, optical parametric amplifiers have been considered
for S-band amplification [2, 3]. EDFAs for use in the S-band are also briefly reviewed.

10.2.1. Semiconductor Optical Amplifiers (SOAs)

The principles of operation of the SOA differ from those of the other amplifier
approaches in that an optical pumping source is not used. Optical amplification is
achieved in an SOA through the application of an electric field to the input optical
transmission signal traveling through an SOA waveguide structure. Within the SOA
and electric field, an absorption or stimulated emission of a photon can occur. The ab-
sorption of a photon results in the generation of an electron-hole pair, and stimulated
emission is brought about by a photon initiating the recombination of an electron-
hole pair. To obtain amplification of an optical signal, the stimulated emission must
be greater than the absorption of photons.
 10. S-Band Raman Amplifiers 305

Output
Fiber

Signal Gain (a.u.)

–V n p +V

Lenses
1440 1460 1480 1500 1520
Input Wavelength (nm)
Fiber

Fig. 10.4. Schematic structure and typical gain curve of a semiconductor optical amplifier.

The characteristics of the gain, the center wavelength, and the gain bandwidth
offered by an SOA are dependent upon the type of semiconductor material used and
the structure of the device. For optical transmission applications, current emphasis
has focused on SOAs fabricated from indium phosphide (InP), indium gallium ar-
senide (InGaAs), and indium gallium arsenide phosphide (InGaAsP). The 3 dB gain
bandwidth of a single device is typically 20 to 30 nm, although operation has been
reported up to 40 to 60 nm away from the gain peak wavelength (Fig. 10.4). With
current technology, maximum gains made available through the use of SOAs have
been limited to 15 dB. The physical properties of the semiconductor material permit
the construction of SOAs that can operate across a wide range of wavelengths from
1300 up to 1600 nm.
The very aspects that make SOAs so attractive as an optical amplifier also pose the
greatest challenges to its eventual implementation for long-haul WDM applications.
This is borne out by the nonlinear nature of the amplification mechanisms involved
in an SOA being applied to a linear application of cascading optical amplifiers along
a number of consecutive fiber spans. Challenges exist because SOAs suffer from
low-gain, high-noise figures along with high bit rate and multiplex signal handling
sensitivities. For brevity, only the noise and bit rate sensitivity issues are discussed.
The rapid response time of the SOA brings about a number of difficulties when
attempting to achieve amplification of high bit rate WDM signals within a fiber. With
recovery times on the same order as some of today’s time-division-multiplexed bit
rates, current SOAs only effectively handle signals below 10 Gbps before saturation,
although 20 Gbps operation using lower powers to remain below saturation has been
reported [4]. However, these lower power levels will not operate over the typical
“standard” span length of 80 to 100 km.
With the fast gain response affecting the entire length of the signal within the
waveguide of an SOA, changes in the magnitudes of the electric field or input signal(s)
impart instantaneous changes (fluctuations) to the resulting signal(s). This leads to
interchannel and intersymbol interference along with high noise figures, typically
around 8 to 10 dB, prevalent in today’s SOAs. This high noise figure (NF) is further
306 M.N. Islam

exacerbated by the high coupling losses that are incurred when connecting an SOA
to the transmission fiber.
The diagram in Fig. 10.4 shows the simplified structure of an SOA, highlighting
the input and output fibers together with the internal waveguide structure. Here the
challenge exists in bringing together the different modes of propagation brought about
by the difference in the geometries between the fibers and SOA waveguide. Together
with the problem of the mechanical alignment, these mismatches lead to coupling
losses for SOAs on the order of 3 dB for each end (input and output fiber), further
affecting the already high noise figure.
The nonlinear operation of SOAs is becoming more understood with more re-
search. As such, the SOA will likely find applications in the world of photonics, not
as a linear amplifier for long-haul WDM, but as a device that truly exploits the very
nature of its operation. Current research has focused on the application of SOAs as
modulators, optical switches, and wavelength converters. The size of the device to-
gether with the high level of integration makes it attractive for applications that are
highly cost sensitive. Thus much of the recent work has also been for applications
within the metropolitan optical network that require a lower level of amplification to
overcome losses from multiple encounters with optical add/drop nodes and, in the
future, optical switches.

10.2.2. Thulium-Doped Fiber Amplifiers (TDFAs)

The principle of operation of a TDFA is similar to that of the EDFA in that optical
amplification of a signal is accomplished through the energy conversion from pump
signals through a length of specially doped fiber. The differences lie in the type of gain
fiber used, dopants employed, and pump configurations. An energy level schematic,
shown in Fig. 10.5(a), depicts the optical amplification process in a TDFA, utilizing
either a fluoride or MCS fiber gain medium. Two pumps, of either different [5] or the
same wavelengths [6], are utilized to achieve optical amplification. An upconversion
pump is used to overpopulate the quasistable level depicted as E2 , where the pump is
then used to populate the other quasistable E3 level, whereupon the stimulated release
of photons then occurs.
A gain response such as that shown in Fig. 10.5 centered around 1460 nm is
achieved by utilizing the pumping scheme discussed together with either a fluoride,
fluoro-zirconate (ZBLAN), or a MCS such as antimony silicate fiber doped with high
levels of thulium. Gain bandwidths of +20 dB have approached 35 nm in width, and
maximum gains of 31 dB have been reported. Unlike the SOA, though, operation is
limited over about a 65 nm range.
The fibers hosting the dopants pose the main challenge to the wide acceptance
and commercial deployment of TDFAs. These fibers are more brittle than silica-based
ones and, thus, greater care must be exercised throughout the manufacturing, deploy-
ment, and maintenance processes. Maintaining the same level of product reliability as
expected with current silica-based technologies (e.g., EDFAs) will pose a significant
challenge, as current acceptable practices must be changed.
 10. S-Band Raman Amplifiers 307

Tm3+ Doped E3
(ZBLAN or MCS)
Gain Medium hpump hsignal
Signal (in) Signal (out)

E2

hupconversion pump
Upconversion
Pump Unit
Pump Unit E1

(a)
Signal Gain (a.u.)

1440 1460 1480 1500 1520

Wavelength (nm)
(b)

Fig. 10.5. (a) Schematic energy level diagram and pump configuration for a thulium-doped
fiber amplifier; (b) typical gain spectrum of a thulium-doped fiber amplifier.

Being of different material composition than fiber deployed into the network’s
transport infrastructure, connecting TDFAs into the network poses another challenge.
The less reliable and lossier mechanical splicing, or epoxy pigtailing must be em-
ployed, as fusion splicing is not an available option. This results in a difference of
>0.3 dB per mechanical splice versus a <0.05 dB per fusion splice. The exotic
compositions of these fibers will also reveal themselves in higher component costs.
Thus there are challenges in reliability and cost that may prevent the widespread
deployment of TDFAs.

10.2.3. Erbium-Doped Fiber Amplifiers

Although the peak gain of EDFAs occurs at wavelengths near 1530 nm, attempts have
been made to extend operation into the S-band. Gain in EDFAs arises from inversion
of the 4 I13/2 level of Er3+ with respect to the 4 I15/2 level [7]. The inversion is achieved
by excitation of the 4 I11/2 upon pump absorption at 980 nm followed by relaxation to
the 4 I13/2 level, or by excitation of the 4 I13/2 level via pump absorption at 1480 nm.
The subsequent emission from the 4 I13/2 level extends into the S-band at sufficiently
high inversion rates.
308 M.N. Islam

2.0

1.5 Inversion:
1.0
1.0 0.8
Gain coefficient (m –1)
0.6
0.5 0.4
0.2
0.0 0.0
–0.2
–0.5 –0.4
–0.6
–1.0 –0.8
–1.0
–1.5

–2.0
1400 1450 1500 1550 1600 1650
Wavelength (nm)

Fig. 10.6. Representative gain spectra of erbium-doped silica fiber at various inversion levels.

Figure 10.6 shows representative emission spectra of erbium-doped silica fiber

at various inversion rates [8]. Gain in the S-band is achievable at inversion rates
exceeding 0.7; however, signal amplification in this region of the erbium-doped fiber
emission spectrum competes with amplification of spontaneous emission in the longer
wavelength bands where gain is higher. Amplified spontaneous emission (ASE) at
the peak emission wavelength, 1530 nm, is particularly strong and tends to saturate
the amplifier, severely limiting achievable S-band gain.
Adequate gain in the S-band may be achieved by suppressing the ASE. Because
the ASE is generated over the length of the amplifier, ASE suppression would only be
effective if applied over a substantial fraction of the erbium-doped fiber. One method
for increasing S-band gain by ASE suppression utilizes short-pass filters inserted
between sections of gain of gain fiber. This configuration has been demonstrated
using five sections of erbium-doped fiber and bidirectional pumping at 980 nm [9].
An average gain of 21 dB was achieved for 40 channels (−20 dBm per channel) in
the wavelength range 1489 to 1519 nm.
A second method for suppressing ASE in the long-wavelength bands utilizes a
host fiber with a long-wavelength cutoff located near the peak wavelength of the
ASE [10]. With a cutoff wavelength near 1525 nm, the core of the host fiber does
not guide ASE at 1530 nm and longer wavelengths. Saturation of the amplifier by
ASE at peak emission wavelengths is effectively eliminated, shifting the gain peak
to shorter wavelengths. This approach effectively distributes ASE filtering along the
entire length of the gain medium and high-gain S-band amplifiers.
Except for a few components, S-band EDFAs are based on standard EDFA tech-
nology developed for the C-band and share the same pump characteristics and gain
 10. S-Band Raman Amplifiers 309

dynamics. Compared to conventional EDFAs, these amplifiers require longer lengths

of gain fiber and have a higher noise figure. Dividing the gain fiber into sections and
inserting filters between them introduces lumped loss along the length of the ampli-
fier and increases the amplifier noise figure. For example, in the previous example
utilizing five fiber sections and short-pass filters, a noise figure of 7 dB was reported
for an aggregate output power of 17 dBm.

10.2.4. Lumped or Discrete Raman Amplifiers

Because it utilizes silica fiber as the gain medium, Raman amplification does not
suffer from the same implementation or deployment challenges as SOAs and TDFAs.
Amplifiers can be fusion spliced with the same type of fibers utilized within the fiber-
optic transport infrastructure. Discrete Raman amplifiers have also been demonstrated
to have performance on a par with EDFAs, in terms of gain, NF, and bandwidth.
Raman amplification arises from the transfer of power from a pump beam to
a signal beam of different frequency. The frequency difference between the pump
and signal beams corresponds to the energy of a vibrational mode of the medium in
which the pump–signal interaction occurs. Gain characteristics of Raman amplifiers
depend on the medium and the pump wavelength and power. In silica fiber, the gain
bandwidth is approximately 40 THz with maximum gain occurring for signals with
frequency offset from the pump by 13.2 THz. Gain shape depends on the number of
pump wavelengths utilized and their relative power.
Raman amplification has emerged as a viable technology for enabling WDM trans-
mission in the S-band. The following section examines the advantages and challenges
of Raman amplification. Then an S-band system using lumped Raman amplification is
illustrated in Section 10.4. Thereafter, detailed analysis is provided of various issues
that require resolution before practical deployment of the S-band in an augmentation
strategy with the C- and/or L-band systems.

10.3. Advantages and Challenges of S-Band Raman LRA

Because silica fiber is a good gain medium, Raman amplification benefits from
an established technology base with an extensive research and deployment history.
Dispersion-compensating fiber is a particularly convenient gain medium. Implement-
ing the LRA in the DCF permits the deployment of a module that simultaneously pro-
vides gain to overcome the fiber and component losses in addition to having integrated
dispersion compensation.
In addition, gain is achievable over a broad wavelength range simply by pump
wavelength and power selection. The gain flexibility of Raman amplification has led
to discrete amplifiers for many different wavelength ranges and demonstrations of flat
gain over wide gain bandwidths. Recent progress in diode laser technology has made
available high-power pump modules at various wavelengths, enabling construction
of Raman amplifiers with gain at practically any wavelength within the S as well as
S+ bands with performance comparable to C-band EDFAs.
310 M.N. Islam

10.3.1. Challenges of Raman Approach

Perhaps the most significant challenge that Raman amplifiers have had to overcome
is the relatively poor efficiency compared to EDFAs. However, recent advances in
the understanding of the Raman gain efficiency of optical fibers and the development
of DCF has led to a natural solution to the gain fiber issue. Increases of over tenfold
in Raman gain efficiencies have been reported for commercially available DCFs. In
addition, semiconductor laser diode powers have risen steadily over the past few
years. Newer generations of high-power laser diodes are today being demonstrated
to have greater than 1 W of output power [11]. New diode-array-cladding-pumped
lasers are reaching output power levels on the order of 10 W and above. The challenge
imposed by the poor efficiency of Raman amplification is quickly fading with these
increases in fiber Raman gain efficiency and laser pump output powers. Moreover, as
discussed in Chapter 1, at high channel count or high signal/pump powers, the slope
and overall efficiency of LRAs can actually exceed even 1480 nm pumped EDFAs.
Noise issues within Raman amplifiers brought about by double Rayleigh backscat-
tering (DRS) and the extremely fast response of the Raman effect itself are also effec-
tively being addressed. DRS can be controlled through the use of improved isolation
between the multiple stages of a discrete Raman amplifier. The extremely fast re-
sponse of the amplification process can be averaged out through the utilization of
counterpropagating pumping. Thus noise figures are decreasing towards 5 dB, on a
par with commercial EDFAs.
Packaging challenges for Raman amplifiers still exist as the lengths required to
achieve appreciable gain can easily reach tens of kilometers. However, as the fiber’s
Raman gain efficiencies increase, these lengths decrease. In addition, for systems
with bit rates of 10 Gb/s or higher, dispersion compensation is required, and DCF is
the most commonly deployed form of compensation. Because the LRA gain can be
achieved in the same DCF, the long fiber length limitation is muted by using a LRA
with integrated dispersion compensation.

10.4. S-Band Long-Haul Transmission Using LRAs

As an example of the application of an S-band lumped Raman amplifier (SLRA),
consider the following experiment of the first S-band long-haul WDM transmis-
sion using a cascade of dispersion-compensating LRAs. Twenty non-return-to-zero
(NRZ) channels, spanning the entire S-band, were transmitted over 10 spans of stan-
dard single-mode fiber (SSMF), each achieving bit error rate (BER) <10−12 without
forward error correction.
In particular, Puc and co-workers [12] demonstrate the cascade of 11 rack-mounted
S-band dispersion-compensating LRAs (SLRAs) to transmit 20 S-band channels
modulated at 10.67 Gb/s over 867 km of standard single-mode fiber. The margins
accumulated in this demonstration show the capability for such a system to achieve
80-channel transmission over 10 × 25 dB SSMF with standard out-of-band forward
error correction (OOB-FEC) and presently available SLRAs.
 10. S-Band Raman Amplifiers 311

OSA

APM

BCM

SLRA

Fig. 10.7. Photograph of six rack-mounted S-band lumped Raman amplifiers (SLRAs). In addi-
tion to the SLRAs, the rack also holds band couplers and optical supervisory units. OSA: optical
spectrum analyzer; APM: administrative processor module; BCM: band coupling module.

Figure 10.7 shows a photograph of a standard seven-foot rack containing six

SLRAs. Each SLRA is a two-stage amplifier containing a gain-flattening filter (GFF)
with a midstage access (Fig. 10.8(a)). The pump module corresponds to four laser
diodes wavelength and polarization multiplexed together (Fig. 10.8(b)). The pump
wavelengths are selected to achieve sufficiently flat gain across the S-band.
Figure 10.9 shows the functional block diagram of the amplifier and its typical
gain and noise performance at a total input power of −14 dBm. It can be seen that high
values of gain (up to 28 dB) can be achieved with a very small gain ripple (<1 dB),
in addition to the noise figure values on the order of 5.5 dB across the S-band (1493
to 1523 nm). The SLRA gain fiber has a high negative dispersion in the S-band, as
well as a negative dispersion slope, providing coarse dispersion and dispersion slope
compensation throughout the band. Each SLRA compensates for the dispersion of
about 75 km of SSMF.
A block diagram of the system experimental setup is shown in Fig. 10.10. Eleven
dispersion-compensating SLRAs are used to transmit 20 channels in the S-band (be-
312 M.N. Islam

1st Stage 2nd Stage

Gain Fiber Gain Fiber

2
WDM 2
WDM
Raman Pump Raman Pump

(a)

1 Polarization
Beam

1 Combiner

WDM

2 Polarization
Beam

2 Combiner

(b)

Fig. 10.8. (a) Configuration of a two-stage S-band lumped Raman amplifier. Provision is made
for midstage access; (b) schematic diagram of pump module. Two pairs of laser diodes each
polarization multiplexed are combined with a WDM coupler.

–14dBm Total Input Power

Gain / Noise Figure (dB)

30

GFF
GFF 25
20
18 dB
Pump Pump 15 21 dB
24 dB
Module Module 27 dB
10
5
0
1490 1495 1500 1505 1510 1515 1520 1525
Wavelength (nm)

Fig. 10.9. Functional block diagram and typical gain and noise figure spectra of the S-band
Raman amplifier. Total input power is −14 dBm.
 10. S-Band Raman Amplifiers 313

Mux
Tx# 1
1493.36nm • LiNbO3 SLRA 98 km SLRA 89 km SLRA SLRA
MZM #1 #2 #3 90 km #4 91 km
200GHz
•
spacing
•
Tx#20 SLRA SLRA SLRA SLRA
1521.77nm 80 km #8 80 km #7 80 km #6 90 km #5

SLRA SLRA SLRA Demux

#9 89 km #10 80 km #11
•
10.7G
• BERT
Rx
•
10.7Gb/s NRZ
PRBS 2 31– 1

attenuator

Fig. 10.10. Block diagram of S-band transmission experiment. Output power is 14 dBm for all
amplifiers (1 dBm/channel) and average span loss is 21 dB. SLRA: S-band Raman amplifier.

tween 1493.36 and 1521.77 nm) nominally spaced by 200 GHz and modulated at
10.67 Gb/s over 10 spans of SSMF, for a total length of 867 km. Each span contains
on average six connectorized joints and additional loss elements to bring the average
span loss to 21 dB.
The average amplifier output power is only +14 dBm, resulting in a launched
power per channel of +1 dBm, and each SLRA is capable of at least +19 dBm
output power. All the channels are launched in parallel polarization and modulated
with a 231 − 1 pseudorandom bit sequence NRZ pattern at 10.67 Gb/s through an
external LiNbO3 modulator. The receiver has an adjustable decision threshold and a
clock extractor, allowing all the BER readings to be made at a maximum likelihood
setting.
The dispersion map is sketched in Fig. 10.11 for the system configuration of
Fig. 10.10. The line consists of sections of SSMF, with a dispersion of about
15ps/nm.km at 1510 nm. A section of dispersion-compensating fiber is inserted be-
tween the two stages of the fifth SLRA. The zero dispersion wavelength of the disper-
sion map is at 1508 nm. The crosses and the legend in Fig. 10.11 show the measured
residual dispersion at the center and two extreme wavelengths.
The received spectrum is shown in Fig. 10.12. The average received optical signal-
to-noise ratio OSNR (0.1 nm bandwidth) is about 20.7 dB. At this nominal level, all
the channels operated at BER <10−12 without any error correction at 10.67 Gb/s line
rate. In addition, Puc et al. measure BER versus OSNR curves by raising the noise
floor of the spectrum while maintaining the peak signal power constant. The results
are summarized in Fig. 10.13. The most dispersive channel is #20, and its received
314 M.N. Islam

Distance (km)
0 100 200 300 400 500 600 700 800 900
1500

Dispersion (ps/nm) 1000

500
159 ps/nm
0
–218 ps/nm
–500

–1000 1523 nm ⇒ 159ps/nm

1510 nm ⇒ 36ps /nm
–1500 1493 nm ⇒ –218ps /nm

Fig. 10.11. Dispersion map of the S-band transmission experiment. Crosses and legend show
residual dispersion at the center and extreme wavelengths.

3.0dB/D RES: 0.5nm SENS: NORM HLD AVG: 5 SMPL:AUTO

–10.9 L1 : 1518.440nm 1

L3 : –13.62dBm 3
L4 : –27.37dBm
L3–L4 : –13.74dB
–16.9 REF
dBm

–22.9

–28.9

–34.9
1489.00nm 1509.00nm 4.00nm/D 1529.00nm
SRC AUT WL AUT VAC RES
1–2 ANA SHF OFS WL COR

Fig. 10.12. Spectrum after 867 km in the S-band transmission experiment. Resolution band-
width is 0.5 nm.

optical eye diagram is shown as an inset in Fig. 10.13. Moreover, almost no dispersion
penalty is observed.
Similarly, the channels did not suffer from the four-wave mixing (FWM) effect.
Even when five channels are spaced by 50 GHz, no FWM intermodulation product
is observed down to 28 dB below the channel level (Fig. 10.14). At the nominal
launched channel power level (+1 dBm), the self-phase modulation (SPM) appears to
 10. S-Band Raman Amplifiers 315

–3
Received Channel 20 Baseline
Channel 1
C h annel 4
Channel 8
Channel 10
–5 Channel 16
Channel 17
Channel 20
log10(BER)

Other channels
–7

–9

–11

–13

–15
14 15 16 17 18 19 20 21 22 23
OSNR (dB)

Fig. 10.13. Variation of BER with OSNR after 10 spans for 20 channels. Inset shows the eye
diagram of the most dispersive channel.

be the largest impairment aside from amplified spontaneous emission. The estimated
SPM penalty is less than 1 dB for any of the transmitted channels. No cross-phase
modulation (XPM) penalty is apparently observed.
For an 80-channel system using the full power level of the SLRAs (+19 dBm),
the power per channel would be 0 dBm. Furthermore, if a standard OOB-FEC such
as Reed–Solomon 255/239 is employed, one can expect that a reduction of the OSNR
by 5 dB would result in a similar BER performance as if no FEC were used. Therefore
the present demonstration translates into a capability of transmitting 80 channels at
10.67 Gb/s with OOB-FEC (OC-192 line rate) over 10 spans of 25 dB.
To summarize this section, a novel, S-band long-haul WDM transmission scheme
using dispersion-compensating lumped Raman amplifiers has been successfully
demonstrated. This 10-span transmission demonstration confirms the viability of
SLRAs as a key enabling technology for a cost-effective and reliable expansion of
optical networks into the S/S+ -band regions and potentially to other wavelength
windows.
316 M.N. Islam

4.0dB/D RES:0.01nm SENS:NORM HLD AVG: 5 SMPL:AUTO

–12.0 2 1
L1 : 200.2919THz
L2 : 200.3422THz
L3 : –19.53dBm
L4 : –43.16dBm
L2–L1 : 0.0503THz
L4–L3 : –23.63dB
3
–20.0 REF
dBm

–28.0

–36.0

4
–44.0
1494.41nm 1496.41nm 0.40nm/D 1498.41nm
SRC AUT WL AUT VAC RES
1–2 ANA SHF OFS WL COR

4.0dB/D RES:0.01nm SENS:NORM HLD AVG: 5 SMPL:AUTO

–12.0 2 1
L1 : 200.2919THz
L2 : 200.3422THz
L3 : –19.53dBm
L4 : –43.16dBm
L2–L1 : 0.0503THz
L4–L3 : –23.63dB
–20.0 REF 3
dBm

–28.0

–36.0

4
–44.0
1494.41nm 1496.41nm 0.40nm/D 1498.41nm
SRC AUT WL AUT VAC RES
1–2 ANA SHF OFS WL COR

Fig. 10.14. Four-wave mixing test over 10 spans at 50 GHz spacing. Channel bit rate is 10
Gb/s.
 10. S-Band Raman Amplifiers 317

10.5. Amplifier Characterization and Control

Having demonstrated the S-band system experiment in the last section, it is worth
looking now in more detail at the performance and control of the Raman ampli-
fiers [13]. The gain fiber, which is a dispersion-compensating fiber, is first charac-
terized. The gain fiber has high negative dispersion in the S-band as well as a nega-
tive dispersion slope, which provides for coarse dispersion and slope compensation
throughout the band. Figure 10.15 shows the unsaturated gain coefficient in the DCF
as compared with the gain in SiO2 glass. The peak gain coefficient is more than an
order of magnitude higher than in fused silica, and the full-width at half maximum of
the gain curve is about 37 nm wide at 1500 nm. Figure 10.16 examines the attenuation
in the gain fiber. The figure of merit for the Raman amplifier fiber is gR /αP Aeff , so
that low loss is as important as high gain coefficient. The loss as a function of wave-
length is shown in Fig. 10.16. It is important to note that a low OH− attenuation at
∼1380 nm is critical for S-band efficiency, because this is the loss seen by some of
the pump wavelengths.
Moreover, Fig. 10.17 shows the splice loss spectral variation between the gain
fiber and the transmission fiber for two different splices. Knowledge of the splice
loss variation is important for accurate prediction of amplifier performance. As an
example, Fig. 10.18 shows the calculated (lines) and measured (squares) net gain
versus wavelength. The top dotted curve is the modeling results without including the
spectrally dependent loss of the splice, and the lower solid curve includes the spectral
dependence. As can be seen, reliable prediction of experimental performance requires
accurate knowledge of the fiber parameters as well as their spectral dependence.
Beyond the loss and gain, another important aspect of the amplifier modeling is the
gain saturation performance. For instance, Fig. 10.19 shows the simulated (lines) and
experimental data (points) for signal input powers ranging from −18 dBm to 0 dBm.

0.008

DCF
0.006
g R /A eff(m –1W –1)

0.004 5THz = 37nm

@1500 nm

SiO2

0.002

0.000
0 5 10 15 20 25
Raman Shift (THz)

Fig. 10.15. Raman gain spectra in dispersion-compensating fiber (DCF) and silica.
318 M.N. Islam

1.5

1.0
Loss (dB/km)

0.5

0.0
1380 1420 1460 1500 1540 1580 1620 1660
Wavelength (nm)

Fig. 10.16. Attenuation in gain fiber used in S-band Raman amplifier. Low absorption at 1380
nm (hydroxyl absorption peak in silica) is critical for high-power conversion efficiency.

0.7

0.6
Splice 1
Splice 2
0.5

0.4
Loss (dB)

0.3

0.2

0.1

0.0
1300 1350 1400 1450 1500 1550 1600 1650
Wavelength (nm)

Fig. 10.17. Loss spectra for fusion splices of gain fiber to single-mode fiber.
 10. S-Band Raman Amplifiers 319

21

20

19
Net Gain (dB)

18

17
Experimental Data
Simulation w/ flat splice loss
16 Simulation w/
-dependent loss

15
1490 1495 1500 1505 1510 1515 1520 1525
Wavelength (nm)

Fig. 10.18. Calculated (lines) and measured (squares) gain spectra. The dotted line is for flat
splice loss spectrum and the solid line is for wavelength-dependent loss. Knowledge of splice
loss spectral variation is important for performance prediction.

20
Input Power: –18 dBm

19
–12 dBm

18
Gain (dB)

–6 dBm
17

16
{
0 dBm

15

14
1490 1495 1500 1505 1510 1515 1520 1525
Wavelength (nm)

Fig. 10.19. Measured and simulated gain spectra in SLRA for input powers of −18, −12, −6,
and 0 dBm. Solid lines: simulation results; points: experimental data.
320 M.N. Islam

40
Input Power
–18 dBm
35

–12 dBm
30
–6 dBm

25
Gain, ONF (dB)

0 dBm
20

15

NF
10

0
1490 1495 1500 1505 1510 1515 1520 1525 1530
Wavelength (nm)

Fig. 10.20. Gain spectra without gain-flattening filter at various input powers. The peak gain
is as high as +39 dB, and the NF remains virtually constant as the gain is varied.

The discrepancy between theory and experiment increases with increasing signal
input power, which may result from the nonlinearity of some of the fiber parameters.
Nonetheless, the agreement is fairly reasonable over a wide signal power range.
The saturation behavior of the two-stage Raman amplifier is detailed more care-
fully in Figs. 10.20 and 10.21. In particular, Fig. 10.20 shows the gain and noise figure
versus wavelength for different signal input powers. The data here are for open-loop
operation of the amplifier with the pumps at a nominal setting and with no gain-
flattening filter. The maximum small-signal gain is 39 dB, and the noise figure remains
virtually constant as the signal power is varied. The saturation power is more clearly
shown in Fig. 10.21, which plots the total output signal power versus input signal
power for the dual-stage amplifier with 21 input channels. The maximum output power
in a highly saturated operation reaches +24 dBm at an input of 0 dBm. The peak NF in
this condition is relatively low at 6.7 dB. Thus the major spectral effect of saturation
is a small negative gain tilt of 0.9 dB as the power changes from −18 dBm to 0 dBm.
It is also interesting to note the gain difference for the different input powers
versus wavelength. For example, Fig. 10.22 plots the gain difference for different input
powers using the −18 dBm input power as the reference curve. As the input power
increases, some inhomogeneous gain saturation behavior is observed. In general,
Raman amplifiers act homogeneous-like due to pump depletion of the common pump.
However, the situation becomes more complicated with multiple-wavelength pumps,
which themselves interact through the Raman effect. Hence, some inhomogeneity of
the gain curve is observed.
 10. S-Band Raman Amplifiers 321

27.0

25.0

23.0
Output Power (dBm)

21.0

19.0

17.0

15.0
–20 –15 –10 –5 0
Input Power (dBm)

Fig. 10.21. Variation of total output signal power with total input signal power for 21 input
channels.

–18 dBm
–15 dBm
Gain difference (dB)

–12 dBm
–5
–9 dBm
–6 dBm
–3 dBm
0 dBm
–10

–15
1490 1495 1500 1505 1510 1515 1520 1525
Wavelength (nm)

Fig. 10.22. Gain difference spectra for different input powers. Reference is gain at −18 dBm
input power.
322 M.N. Islam

Stage 1 Gain Fiber Stage 2 Gain Fiber

Output
Input

Stage 1 Signal Stage 2

pumps monitoring pumps

Feedback Control

Midstage access

Fig. 10.23. Schematic diagram of a two-stage amplifier with signal monitoring and closed loop
control.

10.5.1. Feedback Control of the SLRA

To control the performance of the two-stage Raman amplifier, the amplifier unit is
equipped with a closed loop control system that can be configured via software. As
shown in Fig. 10.23, the feedback control circuit has as an input signal monitoring at
the midstage, and the output controls the pump laser powers in the two stages. The
closed loop control acting on the pumps can be set for various conditions such as
constant gain, constant output power, and provisional gain or power tilt.
Using the closed loop control set for constant gain, flat gain and noise figure per-
formance can be achieved by adding a gain-flattening filter. For example, in Fig. 10.24
are shown the typical gain and noise performance of the amplifier over a wide range
of gain settings, at an aggregate input power of −8 dBm. A maximum output power of
+19 dBm is achieved, corresponding to transmission of 80 channels at 0 dBm/channel.
Over a setpoint range of 9 dB, the gain is substantially flat and does not exhibit tilt.
This control range is achieved solely through pump power control, without the need
for a midstage variable attenuator. The maximum optical noise figure is below 6.3
dB and does not change with gain setting (e.g., the NF variation is <0.15 dB). The
NF decreases at reduced input power and is less than 6.0 dB for input powers below
−10 dBm.
A more detailed look at the gain profiles shows the performance of the amplifier
under closed loop constant gain control. For example, Fig. 10.25 illustrates the gain
as a function of wavelength with a constant gain setting of 27 dB and different net
input signal powers. Over the input power range of −8 dBm to −14 dBm, the gain
variation is less than 0.25 dB. An alternate view is provided by looking at the ripple
variation at different input powers for different gain levels. Figure 10.26 shows the
ripple at constant signal input power for different gain levels. Again, the closed loop
gain control is found to be effective.
Another aspect of the closed loop operation is to use the control circuit to provide
provisional gain tilt. As an illustration, Fig. 10.27 shows that controlling the relative
 10. S-Band Raman Amplifiers 323

35

30 Gain:

27 dB
25
24 dB

21 dB
20
Gain (dB)

18 dB

15

10

NF
5

0
1490 1495 1500 1505 1510 1515 1520 1525 1530
Wavelength (nm)

Fig. 10.24. Gain spectra under constant gain closed loop control for different gain levels.

28.5

28.0 –8 dBm
–11 dBm
27.5 –14 dBm
Gain (dB)

27.0

26.5

26.0

25.5
1490 1495 1500 1505 1510 1515 1520 1525
Wavelength (nm)

Fig. 10.25. Gain spectra under constant gain closed loop control for different input signal
powers. Gain setting of the loop is 27 dB.
324 M.N. Islam

0.5

0
–8 dBm, 18 dBG
–8 dBm, 21 dBG
–0.5 –8 dBm, 24 dBG
–8 dBm, 27 dBG
–1
1490 1495 1500 1505 1510 1515 1520 1525 1530
Wavelength

0.5

0
–14 dBm, 18 dBG
–0.5 –14 dBm, 21 dBG
–14 dBm, 24 dBG
–14 dBm, 27 dBG
–1
1490 1495 1500 1505 1510 1515 1520 1525 1530
Wavelength

Fig. 10.26. Gain ripple variation under constant gain closed loop control for −8 dBm input
power (upper graph) and −14 dBm input power (lower graph).

pump powers can lead to a positive or negative gain tilt ranging from no tilt to up to
3 dB of tilt over the gain band. Such an operation might be important if the lumped
Raman amplifier were used with another amplifier that had a complementary gain
slope, of if the gain slope were used to compensate for other wavelength-dependent
components in the system. Alternately, the gain tilt might be useful to compensate for
interchannel or interband signal interaction through the Raman effect itself.
Finally, another setting for the control loop is the constant output power mode.
Figure 10.28 shows the output power and noise figure versus attenuation in the mid-
stage access point for the two-stage lumped Raman amplifier for different output
power settings of the control loop. The input power is set at −12.5 dBm, and the
gain tilt remains unchanged as the midstage attenuation is increased. As the output
power setpoint is decreased, the range of allowable attenuation in the midstage in-
creases while still maintaining constant output power. This results from the fact that
the amplifier can only put some maximum output power based on the pump power
limits, and for lower output power settings there is more headroom for the feedback
circuit to compensate for midstage loss. Also shown in Fig. 10.28 is the noise figure
for different midstage attenuation and different settings of the output power.

10.5.2. Polarization-Dependent Gain

Polarization-dependent gain (PDG) in Raman amplifiers is primarily dependent on
the polarization state of the pump. In the lumped Raman amplifier, the pump modules
employ polarization multiplexing to realize low PDG. Typical values of PDG are
 10. S-Band Raman Amplifiers 325

31
30.5
30 No tilt
29.5
29
Gain (dB)
–0.5dB
28.5
28 –1dB
27.5
27 –1.5dB
26.5
26 –3dB
25.5
1490 1495 1500 1505 1510 1515 1520 1525 1530
Wavelength (nm)
31
30.5
No tilt
30
29.5
29 +0.5dB
Gain (dB)

28.5
28 +1.5dB
27.5
27
26.5 +2.5dB
26
25.5
1490 1495 1500 1505 1510 1515 1520 1525 1530
Wavelength (nm)

Fig. 10.27. Gain spectra at various negative (upper plot) and positive (lower plot) gain slopes.
The different slopes are obtained through pump power control.

<0.05 dB. Figure 10.29 shows the increase in PDG when the ratio X/Y of x-polarized
to y-polarized components of the pump is deliberately varied from 0 to 100%. It is
seen that a low value of PDG of <0.05 dB is obtained for pump polarizations <20%,
and the PDG remains below 0.1 dB for the ratio X/Y > 50%. In particular, the data
represent the PDG for a single-stage, backward pumped Raman amplifier with a signal
gain of 15 dB.

10.5.3. Stimulated Brillouin Scattering in Amplifier

One of the concerns raised about Raman amplification is the penalty from stimu-
lated Brillouin scattering. Stimulated Brillouin scattering is related to interaction with
acoustic phonons (whereas Raman amplification is related to the optical phonons),
and the phase matching is in the backward direction. Therefore, to study the impact
of Brillouin scattering, the backscattered light is measured as a function of input
power. Figure 10.30 shows the reflected power as a function of input pump power
for a single-stage, unpumped gain fiber. The threshold for Brillouin scattering is of-
ten given as P T H = 21(Aeff /Leff gB ). For this gain fiber, the stimulated Brillouin
scattering threshold is ∼4 mW for an unmodulated distributed-feedback laser source
with a linewidth of 3 MHz. In addition, Fig. 10.31 shows the back-reflected power
326 M.N. Islam

20.00
75
S Pout=19dBm
18.00 70 Pout = 19dBm
S Pout=17dBm
Pout = 17dBm
S Pout=15dBm
Output [dBm]

16.00 65 Pout = 15dBm

F [dB]
S Pout=13dBm
Pout = 13dBm

14.00 60

55
12.00

50
10.00 0.0 2.5 5.0 7.5 10.0 12.5 15.0
0.00 2.50 5.00 7.50 10.00 12.00 15.00 17.00 Attenuation [dB]
Attenuation [dB]

Pump mp
Pump
Modu
Module Modu
Module

Fig. 10.28. Output power (left plot) and noise figure (right plot) versus midstage attenuation in
the two-stage lumped Raman amplifier for various output power settings of the control loop.
The lower diagram shows the configuration of the amplifier and the location of the variable
attenuator.

0.35

0.3

0.25

0.2
PDG (dB)

0.15

0.1

0.05

0
0 0.2 0.4 0.6 0.8 1
X/Y Pump Ratio

Fig. 10.29. Polarization-dependent gain (PDG) versus ratio X/Y of x-polarized to y-polarized
components of the pump.
 10. S-Band Raman Amplifiers 327

–10
Reflected Power (dBm)
–20

–30

–40 P th

–50

–60
–25 –20 –15 –10 –5 0 5 10 15
Incident Power (dBm)

Fig. 10.30. Reflected power versus incident power for a single-stage unpumped gain fiber.

10

5
m=2.6% G=15dB
Back Reflection Power (dBm)

m=1.0% G=15dB
0 m=1.0% G=10dB
m=1.0% G=7dB
–5

–10

–15

Operating Input
–20 Power range

–25
–25 –20 –15 –10 –5 0 5 10
Input Power (dBm)

Fig. 10.31. Back reflection power in the output stage versus input signal power for the two-stage
amplifier at various gain levels.
328 M.N. Islam

versus input signal power measured in the output stage of the two-stage amplifier
with a pumped gain fiber. The different curves correspond to gain levels between 7
and 15 dB, and the signal input power for the threshold appears above 0 dBm input
power. Thus the stimulated Brillouin scattering threshold is more than 10 dB above
the typically launched signal channel powers, and so the Brillouin effect should not
be a problem in typical operation of the lumped Raman amplifiers.

10.6. S- and C-Band Interaction

The work on the S-band is intended to extend the capacity of existing and new systems.
To deploy the S-band amplifiers, it is expected that the S-band will augment existing
bands in the C-band and possibly the L-band. In particular, it is expected that a split-
band architecture will be used, where the S-band amplifier sits in parallel with the C-
and L-band amplifiers. As the S-band is introduced, one of the concerns is the effect
on the existing C-band that may already be deployed. To understand the effects, the
impact of S-band channels on C-band channels is experimentally investigated in a
three-span dual-band transmission [14].
The existing C-band is affected by the addition of the S-band through the interband
stimulated Raman scattering. The S-band composite signal acts on the C-band as a
broadband copropagating Raman pump. This interaction manifests itself in several
different ways: it increases the optical signal-to-noise ratio throughout the C-band,
and also tilts the power distribution in the C-band and introduces signal fluctuations.
These effects have been quantified experimentally, and it is confirmed that the addition
of S-band channels improves transmission performance in the C-band.
A schematic diagram of the experimental setup is shown in Fig. 10.32. Sixteen
channels in the S-band (between 1492 and 1522 nm), are transmitted over three

# 1 1493.36nm
SLRA #1 SLRA #2 SLRA #3
•
16 : 1

90 km 90 km
200 GHz spacing •
BCM 4
BCM 2

BCM 5

•
BCM 1

BCM 3

#20 1521.77nm DCU DCU

DCU
# 1 1532.32nm EDFA #1 EDFA #2 EDFA #3
4:1

# 2 1541.28nm
SLRA #4
# 3 1553.52nm 10.7Gb/s NRZ • 90 km
31 •
AWG

# 4 1563.52nm PRBS 2 –1
•
BCM 6

BERT BERT Rx
LEGEND: TX RX Rx
DCU
= attenuator
= Dispersion OSA
DCU
Compensation DSO EDFA #4
xy km
Unit for xy SMF
= polarization Rx
rotator

Fig. 10.32. Experimental setup for investigating S- and C-band interaction.

 10. S-Band Raman Amplifiers 329

800

Dispersion (ps/nm)
0

–800
C-BAND
S-BAND
–1600
0 50 100 150 200 250 300
Length [km]

Fig. 10.33. Dispersion map for C- and S-bands.

spans of SSMF, for a total length of 280 km, using four dispersion-compensating
S-band lumped Raman amplifiers. The S-band WDM signals are multiplexed and
demultiplexed with the C-band WDM signals at each repeater. The C-band erbium-
doped fiber amplifiers are operated in constant gain mode to minimize the tilt induced
by the EDFAs. The launch power in the C-band is ∼3 dBm/channel, and the average
span loss is ∼21 dB. The S-band amplifiers are operated in constant output power
mode and provide a negative gain tilt (vs. wavelength) to compensate for the SRS-
induced positive tilt. For test purposes, the C-band WDM signal consists of only four
equally spaced, dispersion-compensated channels between 1532 and 1564 nm. Both
S- and C-band signals are modulated at 10.67 Gb/s with a pseudorandom bit sequence
(PRBS) of length 231 − 1. In a separate experiment, the S-band is modulated with
a 215 − 1 PRBS and the ripple induced in the C-band is monitored on a sampling
oscilloscope.
The inline amplifiers have the following characteristics. For the S-band lumped
Raman amplifier, the output power is up to 19 dBm, the noise figure is <6 dB, the
control circuit is set to adjust the gain tilt, and there is both dispersion and dispersion
slope compensation. For the C-band amplifier, a commercial EDFA is used that main-
tains the gain at 23 dB. The EDFA has an output power up to 17 dBm and a noise figure
<6 dB. With the EDFA an external dispersion and dispersion slope compensation is
required. The corresponding dispersion maps for the C- and S-bands are illustrated
in Fig. 10.33.
The spectrum at the receiver for the S-band and the C-band is shown in Fig. 10.34
as measured by an optical spectrum analyzer. The BER is measured by <10−12 for
all of the channels at a bit rate of 10.67 Gb/s. Also shown by the arrows are the C-
band channel levels at different S-band launch powers. As the S-band channel power
increases, the C-band power level increases due to the stimulated Raman scattering
effect.
Figure 10.35 depicts the measured C-band power tilt versus S-band launch power
with an 11 dBm C-band launch power level. The dashed curves in the same figure
present numerically simulated data with C-band launch power of 11, 9, and 7 dBm.
When prorated, the results show about same amount of tilt as in [15], but somewhat less
tilt than in [16]. The difference is explained by a complete depolarization between
330 M.N. Islam

3.0dB/D RES:0.5nm SENS:NORM HLD AVG: 10 SMPL:AUTO

–11.0 Signal level at S-band Plaunch=18 dBm
Signal level at S-band Plaunch=15 dBm
Signal level without S-band

–17.0 REF
dBm

–23.0

–29.0

–35.0
1490.00nm 1530.00nm 8.00nm/D 1570.00nm
AUT WL AUT VAC RES
ANA SHF OFS WL COR

Fig. 10.34. Spectra at the receiver for the S-band and C-band for S-band launch powers of 18
and 15 dBm and for no S-band signals.

0.05

C-BAND LAUNCH POWER:

0.04 11 dBm, measured
11 dBm, simulation
9 dBm, simulation
Tilt [dB/nm]

0.03
7 dBm, simulation

0.02

0.01

0
10 12 14 16 18
S-band launch power [dBm]

Fig. 10.35. Measured and simulated C-band power tilt versus S-band launch power at various
composite C-band launch powers. Composite C-band power for the measurement is 11 dBm
(solid curve). The dashed curves show simulation results for C-band launch powers of 11, 9,
and 7 dBm.
 10. S-Band Raman Amplifiers 331

1.4

1.2
1564 nm

1 1548 nm

OSNR [dB] 0.8 1532 nm

0.6

0.4

0.2

0
10 12 14 16 18
S-band Pout [dBm]

Fig. 10.36. C-band OSNR improvement versus S-band output power at various wavelengths
for C-band launch power of 11 dBm.

S- and C-band signals. Also, a number of consecutive measurements disclosed no

difference in the tilt magnitude between modulated and unmodulated data, which
is in agreement with [17]. Moreover, the results show a good agreement with the
numerical results. The compensation of this tilt could be achieved by either or any
combination of a band-flattening filter in the band-combiner module, preemphasis of
the signals, or negative gain tilt in the inline amplifiers.
OSNR improvement due to the presence of the C-band signals is shown in
Fig. 10.36. The measurements are taken for various S-band launch power levels
with a C-band launch power of 11 dBm. The C-band OSNR improves proportionally
with wavelength and S-band signal power at the expense of S-band signal power. For
a fully loaded S-band (launch power >18 dBm), the C-band OSNR improvement
would be more than 1 dB. Also, the improvement is roughly proportional (in dB) to
the number of spans.
Due to SRS dynamic crosstalk, ripples appear on the C-band channels induced
by the S-band channels. Measurement results of peak-to-peak ripple that appeared on
C-band signals when all S-band channels are modulated with the same PRBS (worst-
case scenario) are summarized in Fig. 10.37. For example, Fig. 10.38 illustrates a
typical ripple pattern. The baseband ripple, which is proportional to the bandwidth
times the modulation depth, remains approximately constant. Also, the 3 dB ripple
bandwidth is less than 5 MHz.
The BER penalty is determined by comparing BER versus OSNR measurements
in C-band channels with and without S-band transmission. Figure 10.39 summarizes
the results in terms of S-band-induced receiver sensitivity penalty at BER = 10−10
for an S-band launch power of 18 dBm. For comparison, the corresponding OSNR
improvement and the resulting BER improvement are plotted in the same diagram. The
332 M.N. Islam

C-band ripple [%] 6

4
1564 nm

1554 nm
2
1541 nm

1532 nm
0
10 dBm 15 dBm 18 dBm
S-band launch power [dBm]

Fig. 10.37. S-band induced C-band peak-to-peak ripple.

= 1563 nm:
Intensity Ripple [W]

= 1550 nm:

= 1533 nm:

T [200ns / div]

Fig. 10.38. Typical C-band ripple pattern.

 10. S-Band Raman Amplifiers 333

1.2

Improvement & Penalty [dB]

Gross OSNR
0.8 improvement
Net
0.6 improvement
Penalty (due to
0.4 "SRS ripple")

0.2

0
1530 1540 1550 1560 1570
C-band channel wavelength [nm]

Fig. 10.39. Net transmission improvement in C-band due to addition of S-band.

4

OSNR [dB]

0
0 3 6 9 12 15
Number of 90 km spans

Fig. 10.40. OSNR improvement versus number of spans. S-band launch power is 18 dBm.

curve labeled net OSNR improvement corresponds to the gross OSNR improvement
minus the penalty due to SRS dynamic crosstalk. It is interesting to observe that the
BER penalty is relatively flat across the C-band. A roughly constant product of the
ripple magnitude and spectral width can explain this result.
Finally, simulations are used to study the net improvement of transmission per-
formance for an increasing number of 90 km spans. For instance, Fig. 10.40 plots
the OSNR improvement in the C-band versus the number of 90 km spans when
334 M.N. Islam

S-band signals with 18 dBm launch power are added. The net OSNR improvement
is proportional to the number of spans, and is roughly 0.3 dB per span.
Because of a large signal walk-off between the channels in both bands, the effect of
SRS dynamic crosstalk between the S- and C-bands is small compared to the increase
in the C-band OSNR caused by SRS energy transfer from S-band to C-band. The
net result is an average improvement of roughly 0.7 dB in transmission performance
in the C-band over three spans. The improvement should grow proportionally with
the number of spans. Conversely, it can be noted that C-band performance would
deteriorate with the addition of L-band signals by about the same amount it would be
improved by the addition of identical signals in the S-band.

10.7. XPM in the S-Band Raman Amplifier

Because a small-core DCF is used as the gain fiber in lumped Raman amplifiers,
another concern arises about XPM in the gain medium. Therefore, experiments were
performed to assess the impact of XPM on the performance of WDM transmission
using a cascade of lumped Raman amplifiers. In particular, XPM measurements were
conducted on 10.67 Gb/s WDM signals over a 500 km SSMF link implemented with
six LRAs [18]. The results, in good agreement with a theoretical analysis, demonstrate
small XPM transmission impairments in this type of links. In particular, systematic
analysis shows that XPM presents a negligible additional penalty in LRA systems
with 50 GHz channels spacing and operating at realistic launch power levels.
Between the two cross-channel nonlinear phenomena, the four-wave mixing and
cross-phase modulation, the latter is the predominant impairment. XPM measure-
ments in WDM systems are partially masked by linear crosstalk, FWM, and ASE
noise. In order to avoid these undesirable side effects, a novel measuring method is
introduced, based on incoherent homodyne detection of the test channel and RF spec-
tral analysis in the signal baseband. The method allows the taking of reliable XPM
measurements with channel spacings as small as 25 GHz and at low channel launch
power levels. Errors due to linear crosstalk and FWM are practically nonexistent, as
long as the channel launch power is less than 7 dBm/channel. When comparing the
homodyne method with more conventional methods [19, 20], a good agreement is
measured for higher launch channel power level (>4 dBm) and wider channel spacing
(≥50 GHz).
The set-up block diagram is shown in Fig. 10.41. Fourteen DFB lasers with
wavelengths spanning the entire S-band (between 1493.36 and 1521.77 nm) and
nominally spaced by 100 GHz are combined using a 4:1 coupler and a 100 GHz
arrayed waveguide grating (AWG) multiplexer. The nominal test channel (probe) is
surrounded with four “interfering” channels with adjustable wavelength capability in
order to adjust for different channel spacing. Six dispersion-compensating LRAs are
used to transmit WDM channels in the S-band over 5 × 100 km spans of SSMF, for a
total length of 500 km. The probe channel launch power is maintained at 0 dBm, and
the launch power of the rest of channels varies from 0 dBm to 6 dBm, depending on
test scenarios.
 10. S-Band Raman Amplifiers 335

SLRA #1 SLRA #2 SLRA #3 SLRA #4

Wavelength
adjustable
sources

44: :1
1

DELAY
probe

REGENERATOR
1495.6nm SLRA #6 SLRA #5

AWG
100 GHz
spacing

AWG
OSA

BERT
TX
DSO RF spectrum
10.7Gb/s NRZ BERT analyzer
31
PRBS 2 –1 RX
Square law
O/E detector Low Pass F.

Fig. 10.41. Block diagram of experimental setup for XPM measurements.

1507nm
1200 1497nm
Dispersion [ps/nm]

1519nm

–1200
–20 30 80 130 180 230 280 330 380 430 480
Length [km]

Fig. 10.42. Testbed dispersion map. The transmission line consists of five spans of standard
single-mode fiber.

All of the channels are launched in parallel polarization and are modulated with
a 231 − 1 PRBS NRZ pattern at 10.67 Gb/s through external LiNbO3 modulators.
Although the probe channel would not be normally modulated, the setup allows the
modulation of the probe, as well as the direct measurement of the bit error rate
penalties. The probe and the interfering channels are modulated by two different
modulators, driven by the same pattern, but separated by an adjustable delay to test
for the worst-case scenario [19]. All laser sources are dithered by a small internal
modulation at 10 kHz in order to prevent a possible stimulated Brillouin scattering
contamination at higher signal power levels.
The dispersion map is sketched in Fig. 10.42 for all three test wavelengths, which
are at 1498, 1507, and 1519 nm. The line consists of five spans of SSMF, with a
dispersion of about 15 ps/nm.km at 1510 nm.
336 M.N. Islam

1.E–04

1 channel (no XPM)

8.E–05
2 channels

Voltage [V] 6.E–05 3 channels

4.E–05

2.E–05

0.E+00
0.E+00 1.E+09 2.E+09 3.E+09
Frequency [Hz]

Fig. 10.43. Baseband RF spectrum of unmodulated probe for 50 GHz channel spacing and 4
dBm/channel launch power.

The homodyne receiver consists of the probe signal and the “local oscillator”
combiner, O/E converter, a low-pass filter (8 GHz bandwidth), and an AC-coupled RF
spectrum analyzer. By switching the interfering channels on and off, it is determined
that more than 99.5% of the induced XPM appears within the frequency range from 30
MHz to 3 GHz. The baseband spectrogram of unmodulated probe in Fig. 10.43 shows
three spectral density lines: the baseline with no interfering channels, a spectrum with
one interfering channel, and a spectrum with two interfering channels. The numerical
integration of the spectral density lines within the entire baseband frequency range
yields the rms “noise” power with and without XPM. The signal OSNR and average
power are recorded for each measuring point in order to determine the normalized
XPM “noise” variance.
For analysis of the experimental results, the approach explained in [20] is fol-
lowed. In the absence of FWM and for a CW probe channel, the following simple
relations hold with a good accuracy.
σ 2 = σ12 + σXP
2
M
(σ )2  
XP M ∼ σXP M 2 1 Be
2
σXN = = × ×
S σ1 OSN R Bo
  ( )
Q
Q[dB] = 20 log = 10 log σXN
2
× Q2 + 1 .
QXP M
Here the total noise variance σ 2 is a linear sum of noise variances σ12 (signal–
spontaneous noise is the predominant one) and signal fluctuations due to XPM,
denoted as σXP M . S corresponds to the signal baseband power. The XPM “noise”
variance σXN , normalized with respect to the baseband signal power, can be calcu-
 10. S-Band Raman Amplifiers 337

lated either numerically or analytically [21]. Bo and Be denote the receiver’s optical
and electrical bandwidths, respectively. Q is the system penalty, defined as the ratio
of Q-values without and with XPM; Q and QXP M are system Q-values without and
with XPM.
Commercial system design software is used to simulate a number of test scenarios,
and to numerically compute the XPM-induced signal fluctuations. The simulation
assumes linearly polarized and polarization-aligned channels.
The measurements are taken at the beginning, at the middle, and at the end of the
30 nm wide band. The amount of XPM crosstalk remains unchanged, to within the
accuracy of measurements, across the band. Hence, for simplicity, only the results
measured at the beginning of the band are presented.
Figure 10.44 depicts the results for 50 GHz channel spacing and for different signal
power values. The square marks in the diagram present the measurements taken by the
direct BER measurement method. The solid line curves are obtained by numerical
simulation. The good agreement between the numerical and experimental results
indicates that the channels suffered little relative polarization rotation throughout the
500 km long testbed, an indication of a true worst-case scenario test.
Figure 10.45 summarizes the experimental data for different channel spacings.
There is good agreement between numerical analysis of the test scenarios and the
experiment. Based on this agreement, the XPM simulations are extended to a 15 span
LRA link. The results show that for the nominal channel launch power of 0 dBm, the
XPM Q penalty should be less than 0.5 dB.

–18 2 ch. sim.

3 ch. sim.
–20 5 ch. sim
2 ch. meas.
–22 3 ch. meas.
Normalized XPM [dB]

4 ch. meas.
–24 BER test, 3ch.

–26

–28

–30

–32
0 1 2 3 4 5 6
Launch Power/Channel [dBm]

Fig. 10.44. Measured and simulated XPM noise variance over 500 km for 50 GHz channel
spacing.
338 M.N. Islam

–20

–22

–24
Normalized XPM [dB]

–26

0 dBm
–28
2 dBm
–30
4 dBm

–32 6 dBm

–34
100 GHz 50 GHz 25 GHz
Channel spacing

Fig. 10.45. XPM total system penalty versus channel spacing for various channel powers.

From Figs. 10.44 and 10.45, it can be seen that for channel spacings down to 50
GHz and realistic system conditions with launch channel power levels of <2 dBm,
the worst-case XPM penalty is less than 1 dB (a small impact when compared to
SNR degradation due to amplified spontaneous noise). Therefore it is concluded that
XPM is not a limiting factor in the design of long terrestrial links with lumped Raman
amplifiers.

10.8. MPI Penalties in SLRAs

Because of the long fiber length typical of Raman amplifiers, multipath interference
in distributed Raman amplification and in lumped Raman amplifiers can create an
additional system penalty [22]. MPI is forward scattered power arising from such
sources as double Rayleigh backscattering, splice/connector reflections followed by
single Rayleigh scattering, and double reflections. DRBS corresponds to two scat-
tering events in which subwavelength density fluctuations scatter signal waves. Be-
cause Rayleigh loss constitutes the main loss mechanism in optical fiber at short
wavelengths, DRBS can significantly reduce system margins in Raman amplified
links. Indeed, DRBS is the dominant contributor to MPI for most transmission link
configurations, and DRBS increases with amplifier gain and fiber length [23].
Double Rayleigh backscatter crosstalk is defined as the ratio of scattered power
to signal power at the output of the amplifier and may be written as
L L
PDRBS −2
DRBS crosstalk = =k 2
dxG (x) dyG2 (y). (10.1)
Ps 0 x
 10. S-Band Raman Amplifiers 339

The backscatter coefficient k is the product of the capture coefficient S and the
Rayleigh loss coefficient αr at the signal wavelength. In Eq. (10.1), PDRBS is the
average DRBS output power, Ps the average signal output power, and G(z) the net
gain of the amplifier as a function of position. For the case of a single backward
propagating undepleted Raman pump, the net gain G(z) is
 )
g0 ( αp (z−L) −αp L
G(z) = exp −αs z + e −e
αp
ln(G(L)eαs L ) 1 − e−αp L
g0 = , Leff = , (10.2)
Leff αp

where G(L) is the net gain at the signal output end of the fiber. Using Eq. (10.2), the
double integral in Eq. (10.1) can be evaluated after making an additional approxima-
tion of constant gain to obtain
 2  
kL
DRBS crosstalk = G(L)2 − 1 − 2 ln(G(L)) . (10.3)
2 ln(G(L))

Measurements of DRBS crosstalk have been made and the results are compared
to predictions from Eq. (10.3) [23]. A calibrated electrical spectrum analyzer is used
to make MPI spectral measurements, as described by Fludger and Mears [24]. Back-
ground noise due to thermal noise, amplified spontaneous emission, and shot noise
is subtracted from the measured spectra to obtain the MPI spectra, as shown in
Fig. 10.46. The spikes evident in the spectra at low MPI levels arise from electri-
cal noise on the current driver for the distributed feedback (DFB) laser diode used in
the measurement. Calibration of the electrical spectrum analyzer (i.e., conversion of
the results from the electrical to optical domain) is accomplished by using a filtered
ASE source with known relative intensity noise (RIN) spectrum. After conversion
to the optical domain, the result is integrated over the frequency range from 0.5 to
100 MHz to obtain the MPI noise variance. A continuous-wave DFB laser source
at 1510.4 nm with a RIN less than −150 dB/Hz at frequencies greater than 0.5
MHz is used. A delayed self-homodyne interferometer is used to calculate a trans-
fer function relating MPI crosstalk values to MPI integrated noise values shown in
Fig. 10.46.
Figure 10.47 compares measured and calculated DRBS crosstalk for a range of
amplifier net gain values. Experimental MPI data are measured for a single-stage
S-band Raman amplifier, a two-stage S-band Raman amplifier, and an optical link
of five 80 km spans of standard single-mode fiber (25 dB loss in each span) with
six cascaded two-stage S-band Raman amplifiers. Figure 10.47 also shows calculated
MPI data and demonstrates good agreement between calculated and measured results.
Data in Fig. 10.47 show that for a stage-gain of up to 14 dB, MPI crosstalk in all
configurations is less than −30 dB.
System performance is represented by the quality factor Q = (I1 − I0 )/(σ1 + σ0 ),
where I1 and I0 are the mean detector currents for marks and spaces in the data stream;
σ1 and σ0 represent standard deviations of the mark and space current distributions,
340 M.N. Islam

–65
MPI (–15.6 dB)

Electrical Power (dBm)

MPI (–17.6 dB)
–75
MPI (–19.6 dB)
MPI (–21.6 dB)
–85 MPI (–25.6 dB)
MPI (–30.6 dB)
–95 MPI (–35.6 dB)
MPI (–40.6 dB)
MPI (–45.6 dB)
–105 MPI (–50.6 dB)

–115
0.E+00 2.E+07 4.E+07 6.E+07 8.E+07 1.E+08
Frequency (Hz)

MPI Calibration
–15

–20
Integrated MPI Noise (dB)

–25

–30

–35

–40
MPI Calibration Data
–45
Linear Fit
–50
–50 –45 –40 –35 –30 –25 –20 –15
MPI (dB)

Fig. 10.46. Characteristic MPI power spectra (left plot) for different values of MPI crosstalk
and MPI crosstalk versus integrated MPI noise (right plot).

respectively. MPI noise is added to other noise sources such as signal–spontaneous

beat noise and thermal noise so that

σ12 = σs−sp
2
+ σth
2
+ σmpi
2
,

2 is proportional to the total MPI crosstalk:

where σmpi

2
σmpi = 0.5 ∗ d ∗ Xtalk ∗ I12 .

Here, Xtalk stands for DRBS crosstalk and the factor of 0.5 accounts for the fact that
for unpolarized MPI noise, only half of the MPI power beats with the signal in the
receiver. The duty cycle d is equal to 1 (0.5) for the NRZ (RZ) data format. If, for
 10. S-Band Raman Amplifiers 341

–30

–33 MPI/Amp Theory

SFA 1 Amp
DRBS crosstalk (dB) –36
SFA 2 Amps
SFA 12 Amps (5 spans)

–39

–42

–45

–48
0 5 10 15
SFA Net Gain per Amplifier (dB)

Fig. 10.47. DRBS crosstalk as a function of net gain in an S-band amplifier. Solid line: calculated
DRBS crosstalk for a single-stage amplifier. Points: measured DRBS crosstalk for a single-
stage amplifier (triangles), a two-stage amplifier (squares), and a five-span (25 dB) link with
six double-stage amplifiers with about 12.5 dB gain/amplifier. The Rayleigh scatter parameter
for dispersion-compensation fiber is chosen to be 7.1 × 10−4 km−1 and fiber loss at the pump
wavelength is 0.65 dB/km.For the two-stage and link cases, the ordinate represents the MPI
crosstalk per stage.

simplicity, I0 and σ0 are neglected, the MPI penalty for an NRZ system becomes:
 
Q(Xtalk)
MPI Penalty = 20 log10
Qbase
I1 I1
Q(Xtalk) = * , Qbase = ,
σ12 + 0.5∗ Xtalk ∗ I12 σ1

where Qbase denotes the “baseline” quality factor, that is, the quality factor with-
out MPI.
Figure 10.48 shows variation of the system penalty with MPI crosstalk for three
different values of Qbase . The MPI penalty for a 10-span system with lumped Raman
amplifiers and average span loss of 23 dB is about −34.6 dB per stage. MPI crosstalk
for multispan systems accumulates linearly with the number of amplifiers. Thus for a
10-span system with 11 two-stage amplifiers the total MPI crosstalk is about −21.2
dB. Accounting for the finite bandwidth of MPI noise due to signal modulation (10
Gb/s) and the electrical bandwidth of the system (7 GHz) decreases the MPI crosstalk
values by about 0.5 dB. Thus MPI penalty for a 10-span system is only 0.5 dB20
according to the curve of Qbase = 15.5 dB20 (corresponding to a Q of 6 in linear
units and a bit-error rate of 10−9 ) in Fig. 10.48. This result is consistent with the
findings of Puc et al. [12], where the worst-case total system penalty resulting from
342 M.N. Islam

–0

–0.5

–1
System Penalty dB20Q

–1.5

–2

–2.5 Q base = 18 dB
Q base = 15.5 dB
–3 Q base = 13 dB

–3.5
–40 –35 –30 –25 –20 –15
X talk (dB)

Fig. 10.48. Variation of system penalty with MPI crosstalk for three different values of baseline
quality factor Qbase .

WDM transmission over 867 km standard single-mode fiber using 11 cascaded S-band
lumped Raman amplifiers was measured and found to be lower than 1 dB.

10.9. Other S-Band Experiments

Raman amplification has been used to extend the bandwidth of TDFAs. Kani and
Jinno [26] demonstrated a hybrid TDFA/Raman amplifier with a 2 dB gain bandwidth
of 50 nm. The amplifier comprised a TDFA followed by a two-stage Raman amplifier
pumped at 1415 nm (Fig. 10.49(a)). Gain and noise figure spectra are shown in
Fig. 10.49(b). It can be seen from the figure that gain greater than 25 dB and a noise
figure less than 6 dB could be achieved from 1460 to 1510 nm. These results were
obtained with a single-channel measurement using −30 dBm input power.
In another experiment on hybrid TDFA/Raman amplifiers, Masum-Thomas et
al. [25] demonstrated a bandwidth of 70 nm. The amplifier comprised a TDFA in series
with a Raman amplifier pumped at 1413 nm, the Raman amplifier providing gain at
longer wavelengths than the TDFA gain band. Gain and noise figure spectra are shown
in Fig. 10.50 for two configurations, one in which the TDFA precedes the Raman
amplifier and another in which the Raman amplifier precedes the TDFA. In both
cases, gain greater than 15 dB is achieved over the wavelength range spanning 1445
to 1520 nm. As indicated by the figure, the noise figure in this hybrid configuration
using a single Raman pump wavelength depends on whether the Raman amplifier
precedes or is preceded by the doped-fiber amplifier. Where the TDFA precedes the
Raman amplifier, the high noise figure of the TDFA dominates the overall noise figure,
which is higher than the noise figure in the configuration where the Raman amplifier
 10. S-Band Raman Amplifiers 343

TDFA FRA
pump pump pump
laser laser laser
HiNAF HiNAF
TDF (i)

in out
(ii) (iii)

(a)

35 35

30 30

25 25

noise figure, dB
gain, dB

20 20

15 15

10 10

5 5

0 0
1450 1460 1470 1480 1490 1500 1510 1520

, nm
(b)

Fig. 10.49. Configuration (a) and gain and noise figure spectra (b) of a hybrid TDFA/Raman
amplifier reported by [26]. Source: J. Kani and M. Jinno “Wideband and Flat-Gain Optical
Amplification from 1460 to 1510 nm by Serial Combination of a Thulium-Doped Fluoride
Fiber Amplifier and Fiber Raman Amplifier” IEEE Electronics Letters, Vol 35, Issue 12 (©
1999 IEEE)

is the first stage. When the Raman amplifier is in the first stage, net noise figures as
low as 5.5 dB could be obtained at long wavelengths. Also shown in the figure are
results for co- and counterpumping of the Raman stage in the case where the Raman
amplifier is the first stage. Copumping yields channel gains ranging from 14 to 39 dB
over the 1445 to 1520 nm wavelength band, and counterpumping yields a lower gain
variation over the band, from 18 to 34 dB.
An experiment employing only Raman amplification in the S band was reported
by Kani et al. [27]. Raman amplification in the wavelength range spanning 1490 and
1530 nm is demonstrated using an amplifier configuration comprising two stages of
germania-doped fiber of total length 10 km, a pump laser operating at 1420 nm, and
an interstage isolator to minimize DRBS (Fig. 10.51). Figure 10.52 shows the gain
and noise figure spectra for −30 dBm input signal power and 750 mW pump power.
344 M.N. Islam

40 40

35 35

30 30

25 25
RA Counter - Tm
Gain (dB)

NF (dB)
20 Tm - RA Counter 20
RA Counter - Tm NF
15 Tm - RA Counter NF 15

10 10

5 5

0 0
1445 1455 1465 1475 1485 1495 1505 1515 1520
Wavelength (nm)
(a)
40 40

35 35

30 30

25 25
Gain (dB)

NF (dB)
RA Counter - Tm
20 20
RA Co - Tm
RA Counter - Tm NF
15 RA Co - Tm NF 15

10 10

5 5

0 0
1445 1455 1465 1475 1485 1495 1505 1515 1520
Wavelength (nm)
(b)

Fig. 10.50. Gain (upper curves) and noise figure (lower curves) spectra for hybrid TDFA/Raman
amplifiers reported in [25]: (a) spectra for counterpumped Raman amplifier preceding TDFA
(triangles) and preceded by TDFA (squares); (b) spectra for counterpumped (triangles) and
copumped (circles) Raman amplifier preceding the TDFA. Source: J. Masum-Thomas, D.
Cappa, A. Moroney: “A 70 nm Wide S-Band Amplifier by Cascading TDFA and Raman Fiber
Amplifier” OFC Technical Digest, Postconference Edition, pg WDD9 (© 2001 OSA)
 10. S-Band Raman Amplifiers 345

1:1
pump laser

gain fibre gain fibre

5km 5km

...1420/1520 nm WDM ...isolator

442/1

Fig. 10.51. Configuration of all-Raman S-band amplifier reported in [27]. Source: J. Kani,
M. Jinno, K. Oguchi: “Fiber Raman Amplifier for 1520 nm Band WDM Transmission” IEEE
Electronics Letters, Vol. 34, Issue 17m 3 September 1998 (© 1998 IEEE)

25
net gain
net noise figure
net noise figure, dB

20
17.5nm
net gain, dB

15 (1504.5–1522.0nm)

10

0
1490 1500 1510 1520 1530
442/2

, nm

25
net noise figure, dB

20 net gain
net gain, dB

15

10
net noise figure
5

0
1505 1510 1515 1520
442/3

, nm

Fig. 10.52. Gain and noise figure spectra for −30 dBm input signal power (upper plot) and
for 16-channel WDM signal (lower plot) [27]. Source: J. Kani, M. Jinno, K. Oguchi: “Fiber
Raman Amplifier for 1520 nm Band WDM Transmission” IEEE Electronics Letters, Vol. 34,
Issue 17m 3 September 1998 (© 1998 IEEE)
346 M.N. Islam

40 S-band sources
PC AWG Booster

= 100 GHz

1 Preamplifier
–270 ps/nm +90 ps/nm
AOM BPF BPF

40 0.60 nm 0.25 nm

10 Gb/s NRZ
31
PRBS 2 –1 AOM GFF
3 dB
Rx
TW TW TW

Data

CLK
100 km 100 km 100 km
Inline 1 Inline 2 Inline 3
BERT
1410 1410 1410

(a)

TrueWave® pump recirculation

100 km

DCF DCF
GFF
2.2 km 3.8 km
+16 dBm 1393 +16 dBm
(0 dB/ch) (0 dB/ch)
1410 1427

(b)

Fig. 10.53. Experimental setup of all-Raman transmission experiment reported in [28]:

(a) configuration of loop experiment; (b) configuration of inline amplifiers. Source: J. Bro-
mage et al. “S-Band All-Raman Amplifiers for 40/spl times/ 10 Gb/s Transmission Over 6 /spl
times/ 100 km of Non-Zero Dispersion Fiber” OFC Technical Digest Postconference Edition,
pg PD4-1 (© 2001 OSA)

As shown in the figure, maximum gain greater than 20 dB and bandwidth at 3 dB

of 17.5 nm are obtained. Gain and noise figure spectra are also shown in the figure
for a 16-channel WDM signal at channel input powers of −30, −20, and −10 dBm
per channel. Application of the S-band Raman amplifier in an optically preamplified
receiver is also investigated, and sensitivity improvement from −15.6 dBm to −30
dBm at 10−9 BER for all channels is demonstrated.
An S-band transmission experiment was reported by Bromage et al. [28]. In that
experiment, transmission of 40 channels in the wavelength range from 1488 to 1518
nm was carried out over two laps of a 300 km nondispersion-shifted fiber loop. The
experimental setup, shown in Fig. 10.53(a), includes three Raman amplified spans of
100 km long nondispersion-shifted fiber with a zero dispersion wavelength lower than
that of conventional dispersion-shifted fibers. In addition, there are three inline ampli-
fiers comprising dispersion-compensating Raman amplifiers, as well as a dispersion-
compensating booster amplifier at the transmitter and a dispersion-compensating
preamplifier at the receiver. The line fiber is pumped close to transparency, and the
 10. S-Band Raman Amplifiers 347

20

18

16
Q [dB]
BER = 10 –9
14

12

10
1485 1490 1495 1500 1505 1510 1515 1520
wavelength [nm]
(a)

45

40
OSNR (in 0.1 nm) [dB]

35 After 300 km
Into first span

30

25

20
After 600 km
15
1485 1490 1495 1500 1505 1510 1515 1520
wavelength [nm]
(b)

Fig. 10.54. (a) Q and (b) OSNR spectra after 600 km transmission [28]. Source: J. Bromage et
al. “S-Band All-Raman Amplifiers for 40/spl times/ 10 Gb/s Transmission Over 6 /spl times/
100 km of Non-Zero Dispersion Fiber” OFC Technical Digest Postconference Edition, pg
PD4-1 (© 2001 OSA)

dispersion-compensating amplifiers, configured as two-stage amplifiers with mid-

stage isolators to minimize DRBS (Fig. 10.53(b)), compensate for residual span loss.
The Q-spectra and optical signal-to-noise ratio at the end of the 600 km transmission
are illustrated in Fig. 10.54. The inline amplifiers and preamplifiers provide 95% span
compensation with a residual 220 ps/nm dispersion across the signal band after 600
km. Optical signal-to-noise ratios greater than 20 dB are obtained for all channels,
and the worst-case Q value is 16 dB. Because line fiber with a low zero dispersion
wavelength is used, no significant four-wave mixing sidebands are observed.
348 M.N. Islam

10.10. Summary and Conclusions

The design, implementation, and issues associated with S-band amplification have
been discussed in this chapter, with a special emphasis on lumped Raman amplifiers.
LRAs can be used in a split-band augmentation strategy with new or already deployed
C- and/or L-band systems, which are usually amplified with EDFAs. To open up the
S-band, the key enabling technology is the appropriate optical amplifier. Most of the
other required components and subsystems are engineering modifications of existing
parts in the C- or L-band.
First, various alternative approaches to S-band optical amplification are compared.
SOAs are compact and can provide gain in ∼30 nm increments over any wavelength
in the telecommunications window. However, SOAs have challenges associated with
fast recovery time, crosstalk between channels, a high noise figure, and inadequate
power for multiple wavelength channels. Much work continues on TDFAs, and
TDFAs have the benefit of most closely matching the physical characteristics of
EDFAs, which are also rare earth-doped amplifiers. Nonetheless, the major challenge
of TDFAs is that the host fiber in which the amplifier is implemented cannot be fusion
spliced to standard telecommunications fiber. Another alternative for S-band ampli-
fication is to use EDFAs with filters distributed in the amplifier to block ASE in the
C-band. But high loss can be associated with this filtering mechanism and longer gain
fibers may be required, thereby not giving comparable performance for the EDFAs
in the S-band as in the standard C-band.
Raman amplifiers appear to be a practical solution to the S-band amplifier, and
they are a mature technology ready for deployment. Utilizing silica fiber as the gain
medium, Raman amplifiers can be readily fusion spliced with the fiber used in the
transport infrastructure. LRAs have also been demonstrated with performance on a
par with commercial C-band EDFAs in terms of gain, noise figure, and bandwidth.
In addition, LRAs can be implemented efficiently in DCF, which means that the
lumped amplifier can be integrated with the dispersion compensation. The major
challenge of Raman amplifiers has been their lower efficiency than EDFA’s, but this
discrepancy is narrowing through better gain fibers, higher laser diode pump powers,
and the inherent better slope efficiency for Raman amplifiers at higher channel count.
Other issues are also associated with DRBS and the fast response time of Raman
amplification. However, simple amplifier architectures can be used to overcome these
limitations. By using counterpropagating pumping, a longer effective upper-state
lifetime corresponding to the transit time can be implemented. In addition, by splitting
the amplifier into a multiple-stage amplifier and using isolation between stages, the
DRBS penalty can be controlled.
The bulk of this chapter focuses on the issues and experimental demonstration
of SLRAs in fiber-optic transmission systems. To start with, in Section 10.4 a de-
tailed system experiment is described that uses cascaded SLRAs. Then, Sections 10.5
through 10.8 study various key issues associated with practical deployment of SLRAs
in an augmentation strategy with C-band systems.
The system experiment of Section 10.4 uses SLRAs with integrated dispersion
compensation to transmit 20 NRZ channels over the entire S-band over 10 spans of
SSMF. The S-band channels are modulated at 10.67 Gb/s, and they are transmitted
 10. S-Band Raman Amplifiers 349

over 867 km of SSMF. Each channel achieves a BER <10−12 without FEC, and there
is negligible dispersion penalty and negligible four-wave mixing penalty measured.
The margins accumulated in this demonstration show the capability for such as system
to achieve 80-channel transmission over 10 × 25 dB SSMF with standard OOB-FEC.
Section 10.5 dissects in more detail the amplifier characteristics and control of
the SLRAs used in the above system experiment. The gain fiber is chosen to provide
both dispersion and dispersion slope compensation, and accurate modeling of the
amplifier performance requires detailed knowledge of the loss, gain, and saturation
as a function of wavelength. Also, to control the performance of the two-stage SLRA,
the amplifier unit is equipped with a closed loop control system. The closed loop
control acting on the pump laser powers can be set for various conditions such as
constant gain, constant output power, and provisional gain or power tilt. Moreover,
since a polarization-independent amplifier is desired, a PDG <0.05 dB can be obtained
for pump polarizations <20%. Finally, the stimulated Brillouin scattering threshold
is found to be more than 10 dB above the typically launched signal channel powers.
When used in an augmentation strategy, the existing C-band will be affected by
the addition of the S-band through the interband SRS, which is studied in Section 10.6.
The interband SRS is quantified experimentally, and it is confirmed that the addition of
the S-band channels improves transmission performance in the C-band. Also, because
of a large walk-off between the channels in both bands, the effect of SRS dynamic
crosstalk between the S- and C-bands is small compared to the increase in the C-band
OSNR caused by the SRS energy transfer. For example, in a three-span link an overall
improvement of ∼0.7 dB in transmission performance is measured for the C-band
signals, and the improvement grows proportionally with the number of spans.
Section 10.7 addresses the concern that there may be undesirable nonlinear trans-
mission penalties arising from the gain fiber in a SLRA, which is a small-core size
DCF. XPM is the largest nonlinear effect, and XPM measurements were conducted
on 10.67 Gb/s WDM signals over a 500 km SSMF link implemented with six SLRAs.
Systematic analysis shows that XPM presents a negligible additional penalty in LRA
systems with 50 GHz channel spacing and operating at realistic launch power levels.
The last concern addressed is the MPI penalties in LRAs. DRBS is the dominant
contributor to MPI for most transmission link configurations, and DRBS increases
with amplifier gain and fiber length. It is found that the MPI penalty for a 10-span sys-
tem with LRAs and an average span loss of 23 dB is about −34.6 dB per stage. Because
the MPI crosstalk for multispan systems accumulates linearly with the number of am-
plifiers, for the 10-span system with 11 LRAs the total MPI crosstalk is about −21.2
dB. This corresponds to a penalty of only about 0.5 dB20 for a Qbase of 15.5 dB20 .
Finally, in Section 10.9 other S-band Raman amplifier and system experiments
are briefly reviewed. For example, hybrid TDFA/Raman amplifiers have been demon-
strated with a bandwidth between 50 and 70 nm. Another experiment uses an all-
Raman, two-stage, lumped amplifier over the wavelength range between 1490 and
1530 nm. In addition, an S-band transmission experiment with a SLRA is demon-
strated using 40 channels in the wavelength range between 1488 and 1518 nm that
is carried over 600 km of nondispersion-shifted fiber. Optical signal-to-noise ratios
greater than 20 dB are obtained for all channels, and the worst-case Q value is 16 dB.
350 M.N. Islam

As the experiments on S-band amplifiers demonstrate, the SLRA appears ready

for commercial deployment, and most of the fundamental systems issues have been
addressed. There appear not to be any fundamental limitations that block the opening
up of the S-band. Rather, the issue is probably more of a practical and economic one.
Is it better to augment a C- and/or L-band system with the S-band, or is it better to
use a wideband, all-Raman amplified system (c.f. Chapter 14)?
Although the answer to this question is not straightforward, there may be at least
three scenarios where the S-band LRAs prove out for system deployment. First, when
the goal is to upgrade an already deployed C- and/or L-band system, the split-band
configuration with SLRAs can be attractive, particularly if the triband couplers are in
place. Also, in systems that have particular difficulty in using the C-band for dense
WDM, it may be better to shift transmission to the S-band. For example, it would be
much easier to use the S-band for dense WDM systems when the fiber is dispersion-
shifted fiber, which has the zero dispersion wavelength right in the middle of the
C-band. Finally, for systems using older vintage fiber with large variation and mean
bend-induced loss in the L-band, using a combination of the S- and C-bands should
be advantageous.
There may also be completely new uses of the S-band beyond simple fiber-optic
transport for increasing capacity. A different application of the S-band might be if
a traffic-bearing fiber were to be used as a backup path for another fiber [29]. For
example, consider a system with traffic only in the C-band. Rather than using an idle
fiber as the backup or redundant path to the fiber, another traffic-bearing fiber that
supports both the C- and S-bands can serve as the redundant path. If the original
C-band traffic-bearing fiber is disabled, the data can be translated to the S-band and
augment the C-band traffic on the backup fiber. This type of application would require
identical performance of the S-band as in the C-band, which has been demonstrated
for the SLRAs.

Acknowledgements
Special thanks to the Xtera staff for their excellent work on the S-band Raman am-
plifier, including A. B. Puc, M. W. Chbat, J. D. Henrie, N. A. Weaver, H. Kim, A.
Kaminski, A. Rahman, F. Barthelemy, and S. Burtsev.

References
[1] M. Islam and M. Nietubyc, WDM Solutions 3: 53, 2001.
[2] M.-C. Ho, M.E. Marhic, Y. Akasaka, and L.G. Kazovsky. In Proceedings of CLEO2000,
CThC6 (San Francisco, May), 2000.
[3] Y. Akasaka, K. K. Y. Wong, M.-C. Ho, M. E. Marhic, and L. G. Kazovsky. In OSA Trends
in Optics and Photonics (TOPS) Vol. 54, Optical Fiber Communication Conference,
Technical Digest, Postconference Edition, Washington DC: Optical Society of America,
WDD31, 2001.
[4] L. H. Spiekman, G.N. van den Hoven, T. van Dongen, M. J. H. Sander-Jochem, J. H. H.
M. Kemperman, and J. J. M. Binsma. In Proceedings of the 26th European Conference
on Optical Communication 2000, Berlin: VDE Verlag, vol.1, 35, 2000.
 10. S-Band Raman Amplifiers 351

[5] B. N. Samson, D. T. Walton, A. J. G. Ellison, J. D. Minelly, J. P. Trentleman, J. E.

Dickinson, and N. J. Traynor. In Optical Amplifiers and their Applications, OSATechnical
Digest, Washington DC: Optical Society of America, PDP-6, 2000.
[6] S. Aozasa, K. Hoshino, T. Kanamori, K. Kobayashi, T. Sakamoto, and M. Shimizu, IEEE
Photon. Technol. Lett., 12: 1331, 2000.
[7] E. Desurvire, Erbium-Doped Fiber Amplifiers Principles and Applications, New York:
Wiley, 1994.
[8] Y. Sun, A. K. Srivastava, J. Zhou, and J. W. Sulhoff, Bell Labs Tech. J. 4: 187, 1999.
[9] E. Ishikawa, M. Nishihara, Y. Sato, C. Ohshima, Y. Sugaya, and J. Kumasako. In Pro-
ceedings of ECOC2001 (Amsterdam), PD.A.1.2, 2001; M. Nishihara, Y. Sugaya, and E.
Ishikawa. In OSA Trends in Optics and Photonics (TOPS) Vol. 77, Optical Amplifiers
and Their Applications, OSA Technical Digest, Postconference Edition, Washington DC:
Optical Society of America, OWB4-1, 2002.
[10] M. A. Abore, Y. Zhou, G. Keaton, and T. Kane. In OSA Trends in Optics and Photon-
ics (TOPS) Vol. 77, Optical Amplifiers and Their Applications, OSA Technical Digest,
Postconference Edition, Washington DC: Optical Society of America, PD4-1, 2002.
[11] A. Mathur, M. Ziari, and V. Dominic. In Optical Fiber Communication Conference,
Technical Digest, Postconference Edition, Washington DC: Optical Society of America,
PD-15, 2000.
[12] A. B. Puc, M. W. Chbat, J. D. Henrie, N. A. Weaver, H. Kim, A. Kaminski, A. Rahman,
and H. Fevrier. In OSA Trends in Optics and Photonics (TOPS) Vol. 54, Optical Fiber
Communication Conference, Technical Digest, Postconference Edition, Washington DC:
Optical Society of America, PD39-1, 2001.
[13] P. Gavrilovic. In Proceedings of the 14th Annual Meeting of the IEEE Lasers and Electro-
Optics Society (Cat. No. 01CH37242) (IEEE, Piscataway, NJ), 471, 2001.
[14] A. Puc, F. Barthelemy, M. Chbat, and H. Kim. In Proceedings of the 14th Annual Meeting
of the IEEE Lasers and Electro-Optics Society (Cat. No. 01CH37242) (IEEE, Piscataway,
NJ), 413, 2001.
[15] V. J. Mazurczyk, G. Shaulov, and E. A. Golovchenko, Photon. Technol. Lett., 12: 1573,
2000.
[16] S. Bigo, S. Gauchard, A. Bertaina, and J.-P. Harmaide, Photon. Technol. Lett., 11: 671,
1999.
[17] A. G. Grandpierre, D. N. Christodoulides, W. E. Schiesser, C. M. McIntosh, and J.
Toulouse. In OSA Trends in Optics and Photonics (TOPS), Nonlinear Guided Waves and
Their Applications Topical Meeting, Technical Digest, Postconference Edition, Washing-
ton DC: Optical Society of America, MC 91, 2001.
[18] A. Puc, F. Barthelemy, H. Kim, M. W. Chbat, S. Burtsev, and Ned A. Weaver, unpublished.
[19] S. Bigo, G. Bellotti, and M. W. Chbat, IEEE Photon. Technol. Lett., 11: 605, 1999.
[20] S. Burtsev, R. Abramov, J. Hurley, S. Kumar, D. Lambert, and G. Luther. In OSA Trends
in Optics and Photonics (TOPS) Vol. 54, Optical Fiber Communication Conference,
Technical Digest, Postconference Edition, Washington DC: Optical Society of America,
MF2-1, 2001.
[21] R. Hui, K.R. Demarest, and C.T. Allen, J. Lightwave Technol. 17: 1018, 1990.
[22] S.A.E. Lewis, S.V. Chernikov, and J.R. Taylor, IEEE Photon. Technol. Lett., 12: 528,
2000.
[23] S. Burtsev, W. Pelouch, and P. Gavrilovic. In OSA Trends in Optics and Photonics (TOPS)
Vol. 70, Optical Fiber Communication Conference, Technical Digest, Postconference
Edition, Washington DC: Optical Society of America, 120, 2002.
[24] C. R.S. Fludger and R.J. Mears, J. Lightwave Technol. 19: 536, 2001.
[25] J. Masum-Thomas, D. Crippa, A. Maroney. In OSA Trends in Optics and Photonics
(TOPS) Vol. 54, Optical Fiber Communication Conference, Technical Digest, Postcon-
ference Edition, Washington DC: Optical Society of America, WDD9-1, 2001.
352 M.N. Islam

[26] J. Kani and M. Jinno, Electron. Lett., 35: 1004, 1999.

[27] J. Kani, M. Jinno, and K. Oguchi, Electron. Lett., 34: 1745, 1998.
[28] J. Bromage, J.-C. Bouteiller, H. J. Thiel, K. Brar, J. H. Park, C. Headley, L. E. Nelson, Y.
Qian, J. DeMarco, S. Stulz, L. Leng, B. Zhu, and B. J. Eggleton. In OSA Trends in Optics
and Photonics (TOPS) Vol. 54, Optical Fiber Communication Conference, Technical
Digest, Postconference Edition, Washington DC: Optical Society of America, PD4-1,
2001.
[29] M. N. Islam and O. Boyraz, IEEE J. Select. Topics Quantum Electron., 8: 527, 2002.
Chapter 11
Raman Fiber Lasers

C. Headley, M. Mermelstein, and J.-C. Bouteiller

11.1. Introduction
The use of stimulated Raman scattering (SRS) as a means of amplifying signals in
telecommunication systems has been demonstrated since 1976 [1]. Yet despite its
advantages over erbium-doped fiber, Raman amplification was not used in the first
generation of deployed optically amplified systems. One of the principal reasons for
this was the lack of reliable high-power pump sources needed for Raman amplifi-
cation. It was in this environment that the cascaded Raman fiber laser (RFL) was
invented.
A cascaded RFL uses stimulated Raman scattering (SRS) in optical fibers to shift
the wavelength of light from an input pump laser to another desired wavelength. De-
vices at almost any wavelength can be made by proper choice of a pump wavelength,
and by cascading the pump through several Raman shifts. Although RFLs had been
demonstrated since the 1970s [2–9], the advent of fiber Bragg gratings (FBG) [10, 11]
made the devices practical [12, 13]. Following the initial demonstration, there have
been numerous experimental [13–18] and theoretical results [19–23].
A broadband flat Raman gain profile can be obtained using multiple pump wave-
lengths [24, 25]. It is advantageous to have all the required wavelengths emitted from
one source. This motivated the development of multiple-wavelength RFLs [26–34].
Single cavities simultaneously lasing from two to six wavelengths have been demon-
strated. These cavities have the ability to distribute the total output power among the
different wavelengths to desired values.
Finally, distributed Raman amplification techniques have become more sophisti-
cated with the proposed use of higher-order pumping schemes [35–41]. These entail
launching two or more pump wavelengths separated by approximately one Stokes
shift from each other. This improves the system performance by more uniformly
distributing the gain along the fiber. The use of a RFL is especially suited to this
application because large amounts of power are required at the shortest wavelength.
In response to this application multiple-order RFLs have been proposed [9, 39, 41].
This chapter describes a cascaded RFL in detail. In the next section a single-
wavelength RFL is discussed both theoretically and experimentally. Section 11.3
354 C. Headley et al.

examines multiple-wavelength RFL, and finally Section 11.4 looks at multiple-order

pump sources.

11.2. Single-Wavelength Raman Fiber Lasers

11.2.1. Overview
A schematic of a complete RFL pump module is shown in Fig. 11.1. It has three parts.
The first is a set of multimode 9XX nm diodes that are the optical pumps [42]. The
light from the diodes is coupled into a single multimode fiber with a tapered fiber
bundle. This and alternative coupling approaches are discussed shortly [43–46]. The
next section of the RFL pump module is a rare earth-doped cladding-pumped fiber
laser (CPFL) [47–49], which converts the multimode diode light into single-mode
light at another wavelength. In the final section the single-mode light is converted
to the desired wavelength by a cascaded RFL. In this chapter, the whole device,
including the diodes and CPFL, is referred to as a RFL pump module, with cascaded
RFL (boxed section in Fig. 11.1) used exclusively for the wavelength-converting part
of the RFL. Early on in the literature the cascaded RFL was sometimes referred to
as a cascaded Raman resonator [12, 13]. In this section, a single-wavelength RFL
module is described.
Diodes that are coupled to multimode fibers typically emit light over a larger facet
area compared to those designed to couple into single-mode fibers. Significantly more
power can therefore be extracted from them. Multimode diodes capable of emitting
1 to 3 W in the 9XX nm region, from a 100 µm diameter fiber core with 0.22 NA
(numerical aperture) are typically used as the optical pumps in a RFL module [42].
The lower facet intensity increases the yield in the manufacturing of these diodes
thereby lowering their cost. The reduced thermal stress on the diodes due to the lower
intensity also means that these diodes have the potential to be run without thermal
electric cooling. This reduces electrical power consumption and provides a saving in
operational costs.

Tapered
Multimode Fiber
Diodes Bundles RFL
Yb CPFL

germanosilicate fiber
(500m-1km)
1100,1117nm pump
Yb 3+Cladding-pump reflector
fiber laser input 1455 1366 1288 1218 1156 1100 1156 1218 1288 1366 1455 OC

splice
Fiber Bragg gratings

Fig. 11.1. Exemplary schematic of a RFL pump module with the cascaded RFL highlighted.
 11. Raman Fiber Lasers 355

Silica Cladding Single mode

Single-Mode Core Laser Output
Yb 3 +
Low-Index Polymer Laser Diode
∆n Array

radius
(a) (b)

Fig. 11.2. (a) Shape and index profile of a CPF; and (b) schematic of light propagating down
the fiber.

The light from the multimode diode is next coupled into a CPFL. Several differ-
ent approaches have been presented in the literature, including V-groove side pump-
ing [43, 44], multimode couplers [45], and single-clad coiled fibers [46]. The approach
shown in Fig. 11.1 is a tapered fiber bundle. In this approach, several fibers are adia-
batically fused into one multimode core. Efficient coupling into the single multimode
fiber is governed by the Brightness Theorem, and is obtained if N Ao /Do > NAi /Di ,
where Di/o and N Ai/o are the input/output fiber diameters and numerical apertures
respectively.
A schematic of the index profile of a cladding-pumped fiber (CPF) is shown in
Fig. 11.2(a) [50, 51]. It consists of a rare earth-doped core surrounded by a silica glass
cladding. What differentiates the fiber from typical fiber is that surrounding the glass
is a polymer whose index is lower than that of silica. This allows light to be guided by
the silica cladding as well as the core. Light from a multimode diode is transmitted
along the cladding of the fiber as indicated in Fig. 11.2(b). As the light propagates
through the core it is absorbed by the rare-earth dopant. The light emitted by the rare
earth-doped ions can be trapped and guided in the single-mode core. By placing a
FBG with the appropriate reflectivity at either end of the single-mode core of the
CPF, a CPFL is formed. It should be noted that if the silica cladding were circular,
some modes could propagate without crossing the single-mode core, reducing the
efficiency of the device. The noncircular shape forces mode mixing, so all the modes
eventually cross the core. The CPFL acts as a brightness and wavelength converter,
coupling the high N A multimode light from the diodes into a small area low N A
fiber. Typically, the wavelengths that are chosen for the CPFL range from 1064 to
1117 nm in Yb-doped fibers.
The choice of diode pump wavelengths for a Yb-doped CPFL is based on a trade-
off, which can be seen from examining the Yb absorption spectrum shown in Fig. 11.3.
The use of 975 nm pumps allows for a much higher absorption coefficient and hence
a more efficient device. However, due to the narrow width of the absorption spectrum
around this wavelength, the diodes may have to be thermally stabilized. Alternatively,
pumping at 915 nm reduces the efficiency of the device, but alleviates concerns about
the thermal stability of the multimode diodes.
Rare-earth dopants for the single-mode core of the CPF that allow efficient tunable
lasing in the 14XX region are not available. Therefore a cascaded RFL is used to shift
356 C. Headley et al.

1.0

0.8 Absorption

Normalized Units
Emission

0.6

0.4

0.2

0.0
800 850 900 950 1000 1050 1100 1150 1200
Wavelength (nm)

Fig. 11.3. The emission and absorption spectra of Yb-doped fiber.

the light at 11XX nm to 14XX nm [12–19]. Light is shifted through multiple Stokes
shifts to the desired wavelength by stimulated Raman scattering. The RFL consists
of input and output grating sets separated by a fiber with an enhanced Raman gain
coefficient [53]. In reference to Fig. 11.1, which is an exemplary 1455 nm RFL, light
from the 1100 nm laser enters the cavity. As it propagates down the fiber it is converted
into light at the next Stokes shift, 1156 nm. Any 1100 nm light that is not converted
will be reflected by a high reflector (HR) (∼100%) on the output grating set. The
light at 1156 nm is confined in the cavity by two HR gratings on either end of the
cavity fiber. The 1156 nm light is converted to 1218 nm light. The process continues
in a similar manner with nested pairs of HR at all intermediate Stokes shifts forming
the intermediate cavities. When the desired output wavelength is reached, the output
grating set contains a grating whose reflectivity is less than 100% so that light is
coupled out of the cavity. This grating is called the output coupler (OC).
Experimental measurements of the performance of a Ge-doped RFL pump module
are shown in Fig. 11.4. This device is pumped with an 1100 nm Yb-doped CPFL,
which is in turn pumped by 915 nm multimode diodes. Figure 11.4(a) is a plot of the
output power at 1455 nm for the RFL pump module versus input power at 915 nm.
The slope efficiency and threshold of this device are 40% and 740 mW respectively.
Figure 11.4(b) shows the slope efficiency for only the cascaded RFL device is 52%,
with Pth = 425 mW. For P-doped RFL the best slope efficiency in the literature is
48% [17], though it should be pointed out that this type of laser is in an early stage
of development.
The output spectrum is shown for the same device in Figs. 11.4(c) and (d). The
intermediate Raman orders can be seen in Fig. 11.4(c). The ratio between the peak
power of the desired output wavelength and the intermediate Stokes order with the
highest output power is called the suppression ratio. This is approximately 20 dB
 11. Raman Fiber Lasers 357

1455 nm RFL output power (W)

1455 nm RFL output power (W)

2.4 1455 nm RFL vs. 915 nm Pump 2.4 1455 nm RFL vs. 1100 nm CPFL
2.0 Slope effieciency( ) = 40% 2.0 Slope effieciency( ) = 52%
Threshold (Pth) = 740 mW Threshold (Pth) = 425 mW
1.6 1.6
1.2 1.2
0.8 0.8
0.4 0.4
0.0 0.0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
915 nm Input pump power (W) 1100 nm CPFL pump Input (W)
(a) (b)

0 1.0 Power (W)

Normalized power (dB)

–10
-10 0.16

Normalized Power
0.8 0.37
–20
-20 0.57
0.78
0.6 1.15
–30
-30
0.4
–-40
40
–50
-50 0.2

–60
-60 0.0
1100 1200 1300 1400 1500 1453 1454 1455 1456
Wavelength (nm) Wavelength (nm)
(c) (d)

Fig. 11.4. Plot of the RFL ouput power versus (a) 915 nm multimode diode power and (b)
CPFL power. Output spectra of a RFL showing (c) the spectrum of the intermediate Stokes
shifts at an ouput power of 1 W, and (d) the spectral distribution around 1455 nm.

in Fig. 11.4(c), and changes with pump power. The spectral linewidth of the RFL
significantly broadens as the power is increased, as seen in Fig. 11.4(d).
Raman fiber lasers have been used in several of the pioneering experiments in
distributed Raman amplification. For example, the first demonstrations of (1) capac-
ity upgrades using Raman amplification by Hansen et al. [52], (2) multiwavelength
pumping for large bandwidth by Rottwitt and Kidorf [24], and (3) higher-order pump-
ing by Rottwitt et al. [35] all used single-wavelength Raman fiber lasers. Many other
systems results have also established RFL as a viable Raman pump source.

11.2.2. Cavity Design

The behavior of a RFL depends on parameters such as fiber length, OC reflectivity,
and splice losses. Numerical simulations are used in order to understand the trade-offs
in making these choices. The mathematical basis for the numerical model is given
below, and the results of the numerical simulations are used to elucidate the effect of
some design parameters on a RFL. Finally, experimental results are presented.
358 C. Headley et al.

The evolution of the pump and Stokes power inside a RFL can be described by a
set of nonlinear ordinary differential equations [19–23]. These are:
F /B
dPp F /B νp gR1 F /B
= ∓αp Pp ∓ (P F + P1B )Pp (11.1a)
dz ν1 A1eff 1
F /B
dPi F /B νi gRi F /B gRi−1 F /B
= ∓αi Pi ∓ F
(Pi+1 + Pi+1
B
)Pi ± F
(Pi−1 + Pi−1
B
)Pi
dz νi+1 Aieff Aieff
(11.1b)
F /B
dPn F /B gRn F /B F /B F /B
= ∓αn Pn ± (Pn−1 + Pn−1 )Pn , (11.1c)
dz Aneff

where the superscript F /B designates the power P in the forward and backward
traveling waves, respectively, gR is the Raman gain coefficient, Aeff is the effective
area of the fiber, and ν is the frequency of a given wave. The forward direction is
from the Yb-doped CPFL to the OC as shown in Fig. 11.1. The index p designates the
pump wave, i the intermediate Stokes orders, and n the lasing wavelength. The first
term on the right-hand side of each equation is the intrinsic fiber loss of that wave;
the next term in Eqs. (11.1a) and (11.1b) describes the depletion of that wave by the
forward and backward traveling wave of the next highest Stokes order through the
Raman effect. The third term in Eq. (11.1b) and the second in Eq. (11.1c) represent the
gain of the ith wave through Raman pumping by the (i − 1)th wave. The boundary
conditions for these equations are given by

PpF (0) = Pin PpB (L) = Rpb · PpF (L)

f
PiF (0) = Ri · PiB (0) PiB (L) = Rib · PiF (L)
f
PnF (0) = Rn · PnB (0) PnB (L) = Roc · PnF (L). (11.2)

where R is the reflectivity of the front/back (f/b) Bragg grating, L is the length
of the RFL cavity, and Roc is the reflectivity of the OC. For notational simplicity
Pout = (1 − Roc )PnF (L) is used.
Several important effects are omitted in Eqs. (11.1) for simplicity: (1) no terms
are included for spontaneous emission, (2) interaction between nonsequential Stokes
lines is neglected (e.g., the first Stokes line interacting with the third), as well as (3) the
possibility of generating the next Stokes line beyond the nth wave. Nonetheless these
models provide an excellent qualitative description of the effect of various parameters
on a RFL. The model used to obtain the results presented here includes all of these
effects.
In the simulations presented here a 1117 to 1480 nm RFL such as that shown in
Fig. 11.1 was modeled. Experimentally measured values of the Raman gain coefficient
and the fiber loss for a Ge-doped fiber were used. The splices between the fiber
containing the gratings and the Raman enhanced fiber, and the gratings themselves
were initially assumed to have no loss. The three measures of the RFL performance
used were the slope efficiency ηs , pump threshold power Pth , and the overall (total)
 11. Raman Fiber Lasers 359

75 5%
5% 1400 15%
15% 30%
30% 45%
70

Threshold power (W)

45%

Slope Efficiency (%)

60% 1200 60%
75% 75%
95% 95%
65 1000

60 800

55 600

400
50
200 300 400 500 600 700 200 300 400 500 600 700
Cavity Length (m) Length (m)
(a) (b)

Fig. 11.5. Simulation results showing the effect of fiber length on (a) slope efficiency, and (b)
threshold power for different output coupler reflectivities.

efficiency ηT . These quantities are defined from the linear fit of a graph of the launched
pump power Pin versus output power Pout as

Pout = ηs (Pin − Pth ),

Pout
ηT = . (11.3)
Pin
Because the focus is on improving the performance of the cascaded RFL, Pin is the
power from the CPFL entering the cascaded RFL.
Figure 11.5(a) is a plot of ηs as a function of L for different Roc . As L is increased,
ηs decreases almost linearly for the length range under consideration. This decrease
is due to the increased intrinsic fiber loss as its length is increased. The effect of
changing Roc can also be extracted from Fig. 11.5(a). By moving down a vertical line
at a constant length it is seen that ηs decreases fairly linearly as Roc is increased. The
slightly closer spacing of the lines at higher reflectivities suggests a slight curvature
to the fall-off. This decrease in ηs is expected because a higher Roc means less power
generated in the cavity is extracted from it.
The trade-offs in designing a RFL become apparent in comparing Figs. 11.5(a)
and 11.5(b). Figure 11.5(b) is a plot of the cascaded RFL’s threshold as a function
of length. All of the changes that had a negative effect on ηs now have a positive
effect on Pth . Increasing the length of fiber reduces Pth . This is simply because the
length-integrated Raman gain increases with increasing fiber length, lowering the
threshold. Eventually the benefit on Pth of increasing fiber length saturates as the
length-integrated gain cannot increase any more. Simulations show, for long enough
fiber lengths, Pth can eventually increase as the increased intrinsic fiber loss exceeds
any increased benefit from the length-integrated Raman gain. As with ηs , the effect of
Roc on Pth can also be obtained from Fig. 11.5(b). By moving down a vertical line at
a constant value of length it is seen that Pth decreases nonlinearly as Roc is increased,
360 C. Headley et al.

Table 11.1. Values of L and Roc Optimized to Produce the Maximum ηT for a Given Pp

Pin (W) L(m) Roc (%) Pout (W) ηT (%)

1 500 70 0.4 40
2 400 45 0.9 46
3 400 25 1.5 51
4 400 15 2.1 53
5 300 15 2.7 55

eventually saturating. This behavior is expected because as Roc is increased the cavity
losses are reduced, and threshold is obtained for a lower pump power. The longer the
fiber length is, the lower the value of Roc needed to saturate Pth . Lasing occurs when
gain equals loss, therefore the lower Roc needed for saturation at longer lengths is due
to the increased integrated Raman gain at these lengths. Saturation occurs because
beyond the point at which gain equals loss little value is obtained by retaining more
power in the cavity (by increasing Roc ).
The behavior described in the previous paragraphs points to the need to optimize
between the slope efficiency and threshold power with L and Roc . The overall effi-
ciency is a measure of this trade-off. The results of simulations to maximize ηT by
varying L and Roc , for different values of Pin are summarized in Table 11.1. It shows
that the optimized parameters depend on the desired operating point. For low Pout ,
it is more important to design the cavity to minimize Pth inasmuch as it represents
a large percentage of Pin . Increasing L and Roc does this. For higher powers, ηs is
more important because the device is operating far above threshold; hence smaller
values of L and Roc are required. It is also seen that the maximum ηT increases with
Pp . This is again a reflection of the decrease in the percentage of Pin light used to
reach Pth .
Further insight is obtained by examining the behavior of ηT as a function of L and
Roc for a fixed Pin . Plots of ηT versus L and Roc with Pin = 5 W and 2 W are shown
in Figs. 11.6(a) and 11.6(b). Unless otherwise stated, the parameters used are those
in Table 11.1. The plots indicate the sensitivity of the optimum design points listed in
Table 11.1. Qualitatively, it is seen that ηT is fairly insensitive to L and Roc around the
optimum parameters. It is noteworthy, however, that a laser designed to operate at 2
W(5 W) will not perform optimally at 5 W(2 W). Therefore a laser required to operate
over a wide power range will have to be a compromise between the two designs for
optimum performance.

11.2.3. Raman Fiber

The fiber choice for a RFL is based on five considerations, increasing the gain coef-
ficient, reducing the effective area, decreasing the fiber loss, reducing splice losses,
and the ability to write Bragg gratings in the fiber.
The largest influence on the Raman gain coefficient is the dopant used and the
quantity of the dopant [54, 55]. To date, those RFLs built have used either Ge or
 11. Raman Fiber Lasers 361

56
56

Overall efficiency T (%)

54

Overall Efficiency (%)

54
52
50 52
48 50
46 46
44 44
42 42
40 40
38 38
36 36
200 300 400 500 600 700 0 20 40 60 80 100
Fiber Length (m) Reflectivity (%)
(a) (b)

Fig. 11.6. Overall efficiency as a function of (a) fiber length and (b) output coupler reflectivity
for an input pump power of 5 W (solid symbols) and 2 W(open symbols). Plot (a) is for Roc =
5% (squares), 15% (circles), 45% (triangles) and 70% (diamonds). Plot (b) is for 5 W L = 200
m (squares), 300 m (diamonds), 400 m (circles), and for 2 W 300 m (squares), 400 m (circles),
and 500 m (triangles).

2.0
1.0
GeO2 Ge-doped
P2O5 P-doped
0.8 1.5
Normalized Gain

Loss dB/km

0.6
1.0
0.4
0.5
0.2

0.0 0.0
0 10 20 30 40 1000 1200 1400 1600 1800
Frequency Shift (THz) Wavelength (nm)
(a) (b)

Fig. 11.7. For P- and Ge-doped fibers (a) normalized gain, and (b) loss spectra. (P-doped data
courtesy of Professor E. M. Dianov, General Physics Institute Russia; Ge-doped fiber is RF
courtesy of OFS Fitel.)

P-doped fibers. Figure 11.7 (a) is a diagram of the normalized gain spectrum of P-
doped and Ge-doped fiber. Germanium has been the traditional dopant used in optical
fibers, and the techniques for incorporating Ge into fibers are well understood. On
the other hand, P-doped fiber with a larger Stokes shift (39.9 THz) could reduce the
number of Raman shifts needed (e.g., two shifts from 1060 to 1480 nm) as compared
to Ge-doped fiber (six shifts from 1060 to 1480 nm). Further flexibility is gained in
P-doped fiber because lasing can be achieved from the silica peak at 13.2 THz. The
362 C. Headley et al.

number of Stokes shifts needed to arrive at a particular wavelength can be minimized

by utilizing the P and Si Raman gain peaks [19].
The next goal of the fiber design is to reduce Aeff . This can be accomplished by
reducing the core area. Although the reduction in the core radius r does not have a
one-to-one correlation with the mode diameter, it is an effective means of controlling
mode diameter. There is a lower limit on r as eventually the mode diameter will
increase and the bend loss become excessive. This loss can be reduced by increasing
the index difference, n between the core and the cladding, which also more tightly
confines the mode, reducing Aeff . This is why in order to make low Aeff fibers, the core
cladding index difference is high. However, there is another design constraint and that
is the cutoff wavelength λc . It is intuitive that the most efficient Raman pumping will
take place if the pump and signal wavelengths spatially overlap. Therefore the light
at all wavelengths in the fiber should guide in a single-mode. The pump wavelength
for a RFL then determines the design λc . An expression for λc for step index fibers is
given as [56]
2π r
λc = [(n1 + n2 )n]1/2 , (11.4)
Vc
where n1 and n2 are the refractive indices of the core and cladding, respectively,
and Vc = 2.405 is the normalized frequency at λc . It is now clear that in order to
reduce Aeff , r should be reduced and n increased with the changes constrained by
Eq. (11.4). However, there are additional problems.
The use of high-index difference fibers is both beneficial and detrimental. The
dopants such as Ge and P added to increase the core index also increase the Raman
gain coefficient as outlined earlier. However, an undesirable effect is that the fiber
loss is also increased [57–60]. This is the third fiber design parameter. For low values
of GeO2 in the core it is straightforward to predict the fiber loss based on knowing the
concentration of GeO2 . It is well documented that there is an “excess” loss in highly
doped GeO2 fibers. This loss is called excess because it goes beyond the loss predicted
by merely extrapolating loss calculations at lower concentrations. Examples of the
fiber loss spectra for Ge and P-doped fibers are shown in Fig. 11.7(b).
The fourth design consideration for a Raman fiber is the ability to get low-loss
splices to the fiber. The effect of splice loss is examined in Figs. 11.8(a) and 11.8(b). For
these simulations, the loss indicated is divided evenly between the splices indicated
on the cascaded RFL in Fig. 11.1. As was indicated earlier a splice loss of 0 dB
was used in the previous simulations. Just adding 0.05 dB splice loss (considered
an excellent splice loss) to each end of the cavity produces a predicted decrease in
ηT of 4% (5%) for a RFL optimized for 5 W(2 W). The drop in overall efficiency is
worse at lower powers because the increased cavity loss raises Pth , which represents a
larger percentage of the pump power. Because in most applications a constant output
power is what is required, Fig. 11.8(b) shows the percentage of pump power increase
required to maintain the same output power compared to a 0 dB splice loss. For a 0.1 dB
splice loss in order for a 5 W (2 W) device to maintain the same operating power a 7%
(7.5%) increase in pump power is needed. The reduction in Aeff results in a significant
mode field mismatch between the Raman gain fiber and other fibers. For splices
 11. Raman Fiber Lasers 363

55 110
50 100

% increase in pump power

90

Overall Efficiency (%)

45
40 80
70
35 60
30 50
25 40
20 30
15 20
10
10
0
5
0.1 0.3 0.5 0.7 0.9 1.1 0.0 0.2 0.4 0.6 0.8 1.0
Splice Loss (dB) Splice Loss
(a) (b)

Fig. 11.8. (a) Change in ηT as a function of splice loss for a 5 W (closed squares) and a 2 W
(open squares) optimized RFL as described in Table 11.1; (b) the percentage increase in pump
power required to maintain the same output power as with 0 dB splice loss for a given splice
loss, under the same conditions described in (a).

outside the cavity, the different melting points and diffusion rates between high index-
difference fibers and standard fibers present splicing challenges. In telecommunication
applications, the RFL will have to be spliced to a standard fiber and designing a fiber
that minimizes this loss is important.
The need to minimize the splice losses in the cavity leads to the final design
parameter for a Raman fiber laser cavity, and that is the ability to write low-loss Bragg
gratings in the same fiber that is used to provide gain in the cavity [16]. Without this
ability the losses such as those shown in Fig. 11.8 would quickly accumulate leading
to significantly reduced cavity efficiency.
The practical implications of design trade-offs are now considered. Table 11.2
shows the optimized L and Roc for three different fiber types at 5 and 2 W. The fibers
are all manufactured by OFS Fitel and are high slope dispersion-compensating fiber
(HSDK), TrueWave® RS (TWRS), and a Raman enhanced fiber (RF). Fiber type RF
was used for the prior simulations. Note that the HSDK fiber has a higher gain and loss

Table 11.2. Values of L and Roc Optimized to Produce the Maximum ηT for a Given Pp and
for Fibers with Various Raman Gain and Loss Coefficients
Pin Fiber gR /Aeff α1550 nm FOM L Roc Pout ηT
(W) Type (km−1 W−1 ) (dB/km) (W−1 dB−1 ) (m) (%) (W) (%)
5 HSDK 3.3 0.64 5.1 200 25 2.6 51
RF 2.4 0.30 8.0 300 15 2.7 55
TWRS 0.7 0.21 3.4 600 55 2.4 48
2 HSDK 3.3 0.64 5.1 250 60 0.8 40
RF 2.4 0.30 8.0 400 45 0.9 46
TWRS 0.7 0.21 3.4 800 95 0.7 35
364 C. Headley et al.

coefficient whereas the reverse is true for the TWRS fiber. A figure of merit (FOM)
is defined as [53, 59]
gR
F OM = . (11.5)
Aeff α1550 nm
The best performance is obtained for the highest FOM, which is the RF fiber. Other
trends emerge. The higher the gain the shorter the length of fiber needed for the
maximum ηT . The difference is especially noticeable at the lower power level. The
optimum reflectivity increases for the HSDK (TWRS) because of the higher fiber
loss (lower gain) hence cavity losses must be reduced to reach threshold. Finally, the
best ηT is more sensitive to fiber design at the lower power because Pth is a larger
percentage of the total power.
A summary of the design issues is therefore to select the fiber with a high FOM,
choose the desired output power of the RFL, optimize the design for L and Roc and
then work hard to reduce splice losses.
The focus of this section has been on the cascaded RFL. However, in optimizing
the whole RFL pump module consideration should be given to the CPFL, in particular
to the choice of the CPFL wavelength. The efficiency of this device will depend on the
output wavelength selected. As an example, in considering whether to use P-doped
or Ge-doped fiber the starting wavelengths for a 1480 nm device are 1060 and 1117
nm, respectively, which may tip the choice of the most efficient RFL pump module
in the favor of Ge-doped fibers [61].

11.2.4. Temporal Behavior

Raman scattering is a very fast (<1 ps) process, thus any pump fluctuations occurring
on a time scale slower than 1 ps can cause fluctuations in the signal gain. This imposes
stringent requirements on the noise of a pump laser [62, 63]. In a counter-pumped
configuration, because the pump and signal travel in opposite directions there is a
strong averaging of pump power fluctuations. In the copumped configuration, the
signal and pump propagate through the fiber together, and the only averaging effect
is through the dispersive delay caused by walk-off between the pump and signal.
One measure of the noise in a pump is the relative intensity noise (RIN) of the
pump. The RIN values at a frequency ν are equal to the mean square optical power
fluctuations divided by the square of the mean signal power in a 1.0 Hz bandwidth.
It has been shown that for a 100 km span of standard single-mode fiber the RIN
required of a pump source is −119 dB/Hz and −81 dB/Hz for the co- and counter-
pumped cases, respectively, across a broad frequency range that extends up to 100 s
MHz [63]. Although the exact numbers will vary depending on fiber type and length
these numbers are good estimates of the required pump RIN.
Figure 11.9 is a plot of the RIN of a 1366 nm RFL similar to the one shown
in Fig. 11.1. The mode spacing of the CPFL (4.2 MHz) and the RFL (200 kHz)
are evident. The RIN levels measured allow counter-pumping, but are too high for
copumping. Based on work with semiconductor diodes it can be speculated that it is
due to the large number of modes beating with each other in the fiber [65, 66]. This
being the case, novel approaches are needed in order to significantly reduce the RIN
of RFL.
 11. Raman Fiber Lasers 365

–90

RIN (dB/Hz)
–100

–110

–120
1 2 3 4 5
Frequency (MHz)

Fig. 11.9. Noise spectrum of a 1366 nm RFL from 0.1 to 5 MHz.

11.3. Multiple Wavelength Raman Fiber Lasers

11.3.1. Overview

An important property of stimulated Raman scattering is that the wavelength at which

gain occurs, for a given fiber, depends only on the pump wavelength. Therefore, a
broad flat Raman gain profile can be obtained using multiple pump wavelengths
[24, 25]. Multiple-wavelength RFL (MWRFL) distribute the large amount of power
at one wavelength among multiple wavelengths. In this section MWRFL are described
[26–34]. Following a broad overview, the stability of the device at an operating point
and the ability to achieve a given operating point are examined. Finally the noise
properties are reviewed.
Multiple-wavelength lasing in RFL was first demonstrated in a dual wavelength
device. This device used a ring cavity configuration with carefully selected WDM cou-
plers and fiber Bragg gratings [26]. Linear cavities showing two and three wavelengths
were demonstrated shortly thereafter [27–31]. A three-wavelength RFL (3λRFL) sim-
ilar to that used in [29] is shown schematically in Fig. 11.10. It is pumped by a
Yb-doped CPFL at 1100 nm. The RFL consists of a spool of enhanced Raman gain
single-mode fiber. Light is shifted from 1100 nm to 1347 nm in the Raman fiber with
four nested pairs of high reflectors with fixed reflectivities of approximately 100%.
The output grating set also contains a HR for the 1100 nm pump. Resonant cavities
at 1428, 1445, and 1466 nm (λ1 , λ2 , λ3 ) are created by a set of adjustable-reflectivity
OCs [28] and a matching set of broadband HR gratings. Examples of the emission
spectra obtained from this device are shown in Fig. 11.11. The ability to partition the
power between the different wavelengths is evident.
366 C. Headley et al.

1347 nm

1211 nm 1428 nm 1466 nm

1466 nm 1428 nm HR HR
HR HR 1100 nm
1100 nm
Cladding
Pumped
Fiber Laser
Splice 1445 nm 1153 nm 1445 nm
Variable OC Variable OC
1275 nm

Fig. 11.10. Schematic of a three-wavelength cascaded RFL.

Relative Power (dBm)

0 0

–20 –20

– 40 – 40

–60 –60
Relative Power (dBm)

0 0

–20 –20

– 40 – 40

–60 –60
1420 1430 1440 1450 1460 1470 1420 1430 1440 1450 1460 1470
Wavelength (nm) Wavelength (nm)

Fig. 11.11. Output spectra of a three-wavelength RFL for various settings of the OC reflectivity.

The output power at each of the three wavelengths and the total power as a function
of the incident 1100 nm pump power is shown in Fig. 11.12. It is noteworthy that the
total output power is nonlinear. This is because of the differences in quantum energy
of the different wavelengths, and the fact that the power is redistributed among the
different wavelengths as the CPFL power is increased. In the published literature the
slope efficiency was estimated by drawing an asymptote to the curved line [30]. For
Ge-doped fibers a threshold power of 400 mW and the total slope efficiency of 38%
from 1100 to 14XX nm light has been reported [28, 29]. In work published using
P-doped fibers a threshold of 2.8 W and a slope efficiency of 50% were reported [31].
Both works held out the promise of further improvements in the device.
The ability to control the distribution of power in a MWRFL is critical for a
practical device. Two techniques for accomplishing this have been proposed. In the
 11. Raman Fiber Lasers 367

1400

1200 1428 nm
1445 nm

optical power [mW]

1000 1466 nm
total nm
800

600

400

200

0
0 500 1000 1500 2000 2500 3000 3500
CPFL power (mW)

Fig. 11.12. Total output power and power at each wavelength in a three-wavelength RFL, as a
function of input CPFL power.

first, the HR and OC at a specific wavelength are deliberately misaligned, reducing

the efficiency of the cavity at that wavelength and hence its power [26, 31]. In the
second approach, the reflectivities of the OC are varied with the same net effect
[28–30]. Spectra for the OC at various reflectivities are shown in Fig. 11.13(a), with the
corresponding output laser spectra at one of the wavelengths shown in Fig. 11.13(b).
Note that in this approach there are only small changes in the center wavelength as
the OC (power) at that wavelength is adjusted.

11.3.2. Operating Point Stability and Power Partioning

In the exemplary 3λRFL shown in Fig. 11.10, the Stokes radiation at 1347 nm pro-
vides Raman gain to all three lasing wavelengths. The radiation at 1428 nm provides
gain to the radiations at 1445 and 1466 nm and the radiation at 1445 nm pumps the
radiation at 1466 nm. Therefore the output powers at all three lasing wavelengths are
interdependent. This is seen in Fig. 11.14, which shows the power at the three wave-
lengths as a function of the controlling voltage applied to one OC. In Fig. 11.14(a), as
the voltage on the 1445 nm OC is increased, thereby decreasing the OC reflectivity,
the power emitted at 1445 nm decreases. Simultaneously there is a rise in power at
1428 nm because the depletion of its power by the 1445 nm light is decreasing. The
power at 1466 nm remains fairly constant. Similar results are seen when the OC re-
flectivity of the 1466 nm line is reduced. Now the powers in both at 1428 and 1445 nm
increase as they are no longer being depleted by the 1466 nm line. Note that in both
cases the total power remains constant.
The interdependence of the optical powers at the different wavelengths raises
two questions. If the powers emitted from the laser are set to some predetermined
values, will the MWRFL remain at that operating point? If the answer is yes the
next question is: Can a wide range of such operating points be obtained? The next
368 C. Headley et al.

0.8
0.7 0.0V

0.6 1.0V

Reflectivity (%)
0.5 2.0V
3.0V
0.4
4.0V
0.3
0.2
0.1
0
1465.8 1466.0 1466.2 1466.4
wavelength [nm]
(a)

1.0
0V
1V
Normalized Power

0.8 2V

0.6

0.4

0.2

0
1465 1465.5 1466 1466.5 1467
wavelength [nm]
(b)

Fig. 11.13. (a) Spectra of 1466 nm OC for various reflectivity settings; and (b) output spectra
at 1466 nm for various OC settings.

two subsections answer these questions by examining the stability of the source, and
the ability to properly partition the powers among the different wavelengths. These
results are not general in the sense that the stability of the source will depend on the
physical mechanism used to redistribute the wavelengths (e.g., varying reflectivity
or misaligning gratings), as well as how that mechanism is exploited (e.g., strain or
temperature tuning of gratings).

11.3.2.1. Operating Point Stability

The stability of a 3λRFL was examined in Ref. [30] to see if small changes in the oper-
ating point voltages applied to the OC do not cause large abrupt changes in the power
distribution. A 3λRFL was placed in the experimental setup shown in Figure 11.15.
 11. Raman Fiber Lasers 369

relative optical power [dBm]

0

–5

–10 1428 dBm

1445 dBm
1466 dBm
–15 total dBm

1445B
–20
0 0.5 1 1.5 2 2.5 3 3.5 4
voltage on 1445 nm OC[V]
(a)

5
relative optical power [dBm]

–5
1428 dBm
1445 dBm
–10 1466 dBm
total dBm
–15
1466B
–20
0 0.5 1 1.5 2 2.5 3 3.5 4
voltage on 1466 nm OC[V]
(b)

Fig. 11.14. Output at each of the wavelengths as the reflectivity of (a) 1445 nm OC and (b)
1466 nm OC are adjusted. Also shown is the total power as a function of OC voltage.

100 km TWRS
8 Channel isolator WDM isolator
C-band
laser source OSA-1
signal

computer pumps

3-V DC PS 3
RFL 2% tap OSA-2

V1, V2, V3
GPIB

Fig. 11.15. Experimental setup for evaluating the operating point stability and accessible power
space for a MWRFL.
370 C. Headley et al.

Eight C-band equally spaced signal lasers [1538.2 to 1560.6 nm] were launched down
a 100 km span of OFS TrueWave® RS optical fiber. Pump radiation from the 3λRFL
was launched in the counterpropagating direction with the aid of the wavelength-
division-multiplexer (WDM). A 2% tap provided radiation for pump power moni-
toring by the optical spectrum analyzer (OSA-2). The amplified signal lasers passed
through the WDM to OSA-1 for measurement of the gain and the gain ripple G. The
OSAs and all voltage controls to the OCs were interfaced to a controlling computer
for collection and storage of the gain-flattening and power-partitioning data.
The gain and power distribution stability were experimentally determined. The
total optical power and power partitioning were adjusted to achieve a gain ripple
minimum at transparency. A G of 1.4 dB at transparency was achieved at an ini-
tial voltage OP (V10 , V20 , V30 ) of (2.30, 1.30, −1.05) volts and a launch pump power
OP (P10 , P20 , P30 ) of (283, 194, 144) mW. The stability of this OP was interrogated
by fixing V1 and varying V2 and V3 in a range of ±0.5 V in increments of 25 mV. Fig-
ure 11.16(a) is a contour plot of the gain ripple at constant V1 . The rectangle indicates
a region of ±50 mV about the initial OP corresponding to 1.1 dB< G < 1.8 dB
yielding a voltage sensitivity of ∼0.01 dB/mV. The gain ripple minimum was
1.13 dB. Contour plots for constant V2 and constant V3 show similar results with
reduced voltage sensitivity.
The power space (P1 , P2 , P3 ) was explored in the vicinity of the power OP .
Deviations from the power OP were quantified by the fractional power partition
deviation:
*
(P1 − P10 )2 + (P2 − P20 )2 + (P3 − P30 )2
ρ = , (11.6)
P0
where P0 is the total power. Figure 11.15(b) shows that voltage excursions of ±50 mV
generate <4% variations in the 3λRFL spectral power distribution indicating a ρ
voltage sensitivity of ∼0.04 %/mV. Data at constant V2 and constant V3 show com-
parable results with lower voltage sensitivity. Hence, G and/or ρ can serve as an
error signal in a suitable power-partitioning control algorithm.

11.3.2.2. Accessible Optical Power Space

With the stability at a given operating point established, the complete accessible power
space was examined by varying the three OC voltages from 0 to 3 V in 0.2 V incre-
ments [30]. This was done for total output powers of 600, 860, 1120, and 1330 mW.
Figure 11.17 shows that the power points corresponding to each output power lie
on a plane in a Cartesian optical power space confirming that the total power is
constant.
The three-dimensional power state can be conveniently represented in two dimen-
sions by the simplex diagram shown in √ Figure 11.18. An equilateral triangle is con-
structed with vertices (x, y) equal to (0, 3/2), (−1/2, 0), and (1/2, 0), representing
the power states (P1 , 0, 0), (0, P2 , 0) and (0, 0, P3 ), respectively. This construction
 11. Raman Fiber Lasers 371

–0.9

–1

V3

–1.1

–1.2

1.1 1.2 1.3 1.4 1.5

V2

–0.9

–1

V3

–1.1

–1.2

1.1 1.2 1.3 1.4 1.5

V2

Fig. 11.16. (a) Gain ripple (in dB) contour map as a function of V2 and V3 about the operating
point with V1 constant. Minimum gain ripple is 1.1 dB. Box interior indicates ±50 mV; (b)
fractional power partition deviation ρ (%) as a function of V2 and V3 about the OP with V1
constant. Box interior indicates ±5 mV. The dot shows the operating point.
372 C. Headley et al.

1251

1001 1330 mW

751
P3 [mW]
501

251 P2 [mW]

1 1251
751 1001
11 251 501

251
501 860 mW
751
P1 [mW] 1001
1251

V1

V3
0
3
1
2
2 1
30 V2
0 1
2 3

1251

1001

751
P3 [mW]
501 1251
P2 [mW] 1001
251 751
501
1 251
1
1 251 501 751 1001
1251
P1 [mW]

Fig. 11.17. Two views of 3λRFL optical power planes in 3-D optical power space for 860 and
1330 mW total power, with the voltage space covered shown in the center.
 11. Raman Fiber Lasers 373

0.8

0.7
1

0.6

0.5

0.4

0.3

0.2

0.1

2
3
0
–0.4 –0.2 0 0.2 0.4

Fig. 11.18. Power simplex plot showing accessible power states and experimental and simulated
operating points. (1) Small dots represent all accessible power states; (2) open symbols are
experimental power states corresponding to C-band gain ripple minima of 0.7, 1.1, and 1.6 dB
in 60, 100, and 140 km span of TWRS optical fiber; (3) solid symbols are simulated power
states for 60, 100, and 140 km spans of TWRS and SMF28 optical fiber.

can be generalized to an arbitrary power state (P1 , P2 , P3 ) according to the following

equations.
     
P1 1 P2 1 P3
x = 0· − · + · . (11.7a)
P0 2 P0 2 P0
√      
3 P1 P2 P3
y= · +0· · . (11.7b)
2 P0 P0 P0
The small dots in Fig. 11.18 show the power states spanned by the 3λRFL at a to-
tal launch power of 620 mW. The simplex coverage is not a sensitive function of
total power. Data were taken with a voltage resolution of (0.20, 0.20, 0.05) volts
for (V1 , V2 , V3 ). The open symbols correspond to experimentally determined power
distributions corresponding to minimum gain ripple, for 60, 100, and 140 km span
lengths of OFS TrueWave® RS fiber. The adjacent solid symbols are simulation
results, for the same fiber, and for Corning SMF-28TM at similar lengths. The exper-
imental datapoints and simulation datapoints fall within the laser reach. This leads to
the conclusion that the laser can access an adequate pump combination space.
Finally, an interesting point is how quickly a given operating point can be obtained.
A complete answer to this depends on the electronics used, the physical mechanism
for partitioning the powers, and the starting and finishing operating points. One aspect
of the problem that can be answered is the number of iterations in the OC reflectiv-
ity needed in tuning the cavity at a given wavelength. Using the experimental setup
shown in Fig. 11.15, a given operating point was selected. Initially, the total power
was determined by adjusting the injection current to the multimode laser diodes. The
reflectivity of each OC was then adjusted until the desired power at the wavelength
374 C. Headley et al.

700

600
total power
500
optical powers [mW]
400
1428 nm
300
1445 nm
200

100
1466 nm
0
0 5 10 15 20 25 30 35
Step

Fig. 11.19. Evolution of the three wavelengths and the total pump power to a programmed
operating point.

corresponding to that OC is reached. The next OC was then adjusted until the power
at its wavelength was at the required power. Because changing the OC at each wave-
length changes the power in the other wavelengths, this process was repeated until
all the radiation at each wavelengths were at the desired power level. Figure 11.19 is
a plot of the number of steps required to reach the operating point. Each step corre-
sponds to one adjustment of an OC. In this example as little as 20 steps were required
to reach the desired operating point. The exact number of steps will depend on the
prior and final operating points, and it may be possible to implement faster search
algorithms, however, it is reasonable to assume the number of steps needed is on the
order of tens.

11.3.3. Temporal Behavior of a MWRFL

As was discussed in Section 11.2.4 the temporal fluctuations of a RFL are an important
consideration for Raman amplification. Of particular concern for a MWRFL is the
possible presence of additional noise due to the simultaneous lasing of the different
wavelengths. This question was examined in Ref. [29] by looking at the RIN spectra
of a 3λRFL, lasing at 1427, 1455, and 1480 nm. The three wavelengths were sepa-
rated, and individual RIN spectra measured. The results in a 100 kHz bandwidth are
shown in Fig. 11.20. This bandwidth was chosen to include the low and intermediate
frequency regime where averaging of the pump power fluctuations does not com-
pletely remove this noise source in a counter-pumped configuration. The solid lines
in Figure 11.20 exhibit the RIN in units of dB/Hz as a function of frequency for each
spectral component of the multiwavelength Raman fiber laser and for all three spectral
components simultaneously. It is seen that the RIN values lie below −90 dB/Hz and
that spectral components with lower mean powers exhibit higher noise levels. The
solid trace exhibiting the lowest noise level is that corresponding to the RIN of the
 11. Raman Fiber Lasers 375

–90

1454 nm 244 mW

–95 single
1427 nm 300 mW

RIN [dB/Hz]
1480 nm 303 mW
–100
1427 nm 553 mW

–105
3
1100 mW

–110
0 20 40 60 80 100
frequency [kHz]

Fig. 11.20. RIN spectra for the three wavelengths of a 3λRFL, as well as all three wave-
lengths measured simultaneously. The RIN spectra for a single-wavelength RFL is shown for
comparison.

total optical power. A careful examination of the corresponding time records shows
that the photovoltage standard deviation for all three spectral components is equal to
the incoherent summation of the standard deviations for the individual spectral com-
ponents, indicating that the fluctuations in each spectral component are statistically
independent. The dotted line corresponds to the RIN measured for a single wave-
length 1427 nm Raman fiber laser identical to the multi-wavelength laser but without
the 1466 and 1480 nm OCs. This laser was operated at an optical power of 300 mW
corresponding to the typical power in a single component of the multiwavelength
fiber laser. The single-wavelength noise level is comparable to that measured for the
three spectral components of the three-wavelength laser. Therefore, MWRFL exhibit
noise levels comparable to single-wavelength devices, and are therefore adequate for
counter-pumped Raman amplification.

11.3.4. Six Wavelength Raman Fiber Lasers

The discussions of the previous sections focused primarily on three-wavelength RFL.
If more pump wavelengths can be extracted from a single cavity, then a larger signal
bandwidth can be amplified with the required gain flatness. This was demonstrated
using six-wavelength RFL (6λRFL) [32–34]. These devices were able to simultaneous
amplify both the C- and L-band signals. In the MWRFL shown in Fig. 11.10, this
would require generating additional pump laser components removed in frequency
from the Raman gain peak, of the 1347 nm light, by nearly twice the gain curve
linewidth where the gain has fallen to approximately 10% of its peak value. Therefore,
significant gain must be provided to these additional longer wavelength pumps from
the shorter output wavelength pumps, suggesting that a controlled power distribution
may have been difficult to achieve.
376 C. Headley et al.

HRs: OCs:
RFL 600 m Raman Fiber
1428 1466 1494 1428 1466 1494

1100 RIG ROG 6

1153, 1211, 1100, 1153,

1445 1508 1275, 1347 1211, 1275, 1445 1508
1480 1347
1480

Fig. 11.21. A schematic of a six-wavelength RFL. RIG-Raman input grating set, ROG-Raman
output grating set.

An exemplary six-wavelength cascaded RFL is shown schematically in Fig. 11.21.

A Yb-doped cladding-pumped fiber laser (not shown) is pumped with 915 nm mul-
timode semiconductor laser diodes to generate radiation at 1100 nm. This radiation
is subsequently injected into the 6λRFL where successive Raman shifts generate ra-
diation at 1347 nm. The radiation at 1347 nm is coupled to resonant cavities at the
six lasing wavelengths: 1428, 1445, 1466, 1480, 1494, and 1508 nm. The resonant
cavities occupy a single fiber and are constructed with six broadband HR Bragg grat-
ings and six adjustable OCs. Stokes radiation at 1347 nm provides Raman gain for
lasing at all six wavelengths. Laser radiation at the shorter wavelengths provides ad-
ditional gain to the longer wavelength spectral components. Hence the spectral power
distribution is dependent upon the power in each individual spectral component.
The 6λRFL was used to counter-pump a 100 km span of TrueWave® RS fiber [33].
Figure 11.22 shows the six-wavelength pump power distribution required to achieve

350
1000
1.0 nm BW
optical power [mW]

300
100
338 mW 1508 nm 1466
250
optical power [mW]

10
1480
200 1494
1
1445 1427
150 0.1
0 20 40 60 80 100
215 mW position [km]
100 83 mW

50 30 mW 39 mW
19 mW
0
1420 1440 1460 1480 1500 1520
wavelength [nm]

Fig. 11.22. 6λRFL pump power distribution used to achieve a C+L-band gain ripple of 1.7 dB
in a 100 km span of TrueWave® RS optical fiber. Numbers adjacent to peaks indicate optical
power in mW. Inset shows simulated pump powers as a function of span position.
 11. Raman Fiber Lasers 377

an optimal gain flatness of 1.7 dB pk–pk in the C+L-bands. This power distribution
was achieved by judiciously varying the voltages of the six OCs. Note that the long-
wavelength pumps have significantly less power than the short-wavelength pumps.
This is because the short-wavelength pumps amplify the long-wavelength pumps
as the radiation propagates along the fiber length. This is illustrated by the inset in
Fig. 11.22, which shows a simulation for the pump power as a function of position in
the fiber span corresponding to the launched power distribution.

11.4. Dual-Order Raman Fiber Lasers

Second-order Raman pumping has been proposed as a way to improve optical signal-
to-noise ratio, and hence system performance, by more uniformly distributing the
Raman gain along the transmission span [30–35]. Such a pumping scheme requires
two pump wavelengths, the first separated by approximately one Stokes shift from
the signal, and the other two Stokes shifts from the signal. The former is referred to as
the first-order pump (FOP), and the latter as the second-order pump (SOP). Schemes
in which the FOP and the SOP are co- and counterpropagating with respect to each
other have been demonstrated. The arrangement that is most relevant to RFL is when
both pumps are counterpropagating relative to the signal. The SOP amplifies the FOP,
which in turn amplifies the signal. There is also some direct pumping of the signal
by the SOP. A drawback of second-order pumping schemes is that significantly more
power is needed as compared to using first-order pumping. For example, to pump an
80 km span of standard single-mode fiber to transparency, power levels into the fiber
from 0.8 to 1 W are required compared to just 580 mW for three FOP [39]. The need
for large amounts of power and multiple wavelengths makes the RFL an ideal device
for this application.
A single RFL cavity capable of providing FOP and SOP light was presented in
Ref. [39]. Conceptually, the pump source is based on the multiwavelength Raman fiber
laser technology described in Section 11.3 and an embodiment is shown in Fig. 11.23.
The multimode diodes and CPFL are the same as those shown in Fig. 11.1. As in
the previously described RFL, light from the CPFL at 1100 nm will be converted
to 1365 nm. The OC at 1365 nm allows a portion of the light at this wavelength
to exit the cavity. For a fixed-reflectivity OC at 1365 nm, the reflectivity must be
chosen carefully. There is enough gain at 1455 nm so that the lasing threshold is low.
However, only a small amount of power is required at 1455 nm. The choice of this
reflectivity ultimately limits the tuning of the ratio of the power between the two
pump wavelengths. The OC at 1455 nm is adjustable, enabling the device to control

HR HR HR OC OC LPG
1455 1156 1218 1100 1156 1365 1455 1455
915 nm 1100 nm 1288 1365 1218 1288 Output
MM diodes CPFL

Fig. 11.23. Schematic of a dual-order RFL.

378 C. Headley et al.

900 1000

800

SOP Power (mW)

SOP Power (mW)

950
700
900
600
850
500

400 800

6 7 8 9 10 11 12 13 14 5 10 15 20 25 30 35 40 45
FOP Power (mW) FOP Power (mW)
(a) (b)

Fig. 11.24. Power in the second-order pump versus first-order pump when (a) the power ratio
between the two pumps is constant, and (b) when the gain in an 80 km span of fiber is held
constant.

the ratio of second-order to first-order powers. In order to have stable operation of

the 1455 nm laser line at only 10 mW of output power, it was operated at 70 mW and
then attenuated by a narrow long period grating. The ability to manipulate the ratio
between the two pumps is shown in Fig. 11.24(a), which is a plot of the power in the
SOP versus FOP. As the total power is increased, the ratio between the power in the
two pumps is held constant.
It was shown that by careful selection of the pump wavelengths the same band-
width covered by three first-order pumps could be covered by one second-order pump
wavelength and one first-order seed wavelength [39]. In order to maximize the energy
transfer from the second-order to the first-order pump, a frequency spacing equal to
the peak of the Raman coefficient at 13.2 THz would be chosen. However, the gain
bandwidth can be maximized if the first-order wavelength is slightly offset from the
peak. This is because there is a small peak in the Raman gain spectrum at 24 THz,
which for the second-order pump falls in the signal bandwidth. The wavelength of
the first-order pump should be chosen so that the location of its peak gain at 13.2 THz
complements the direct pumping by the second-order pump.
Quantifying the performance of the dual order source in order to compare its per-
formance to that of a single or multiwavelength device is difficult. This is because for
a fixed 1455 nm OC reflectivity, as the power is increased more power is emitted at
1455 nm. With the long-period grating in place this would mean that the efficiency of
the device would decrease as a larger percentage of the power is being lost. However,
from a telecommunications standpoint, these power configurations would be irrele-
vant. As Figure 11.24(b) shows, there are several different ratios of power in the FOP
to SOP that yield the same gain.
Although the preceding paragraphs focused on one embodiment of a dual-order
pump, several configurations are possible. The OC reflectivity at 1365 nm could be
made variable too. This would increase the operating range of the device. In addition,
 11. Raman Fiber Lasers 379

other techniques to attenuate the light at 1455 nm are possible. Finally, this scheme can
be extended to even higher-order pumping, using, for example, three- or four-pump
wavelengths.

References
[1] C. Lin and R. H. Stolen, Backward Raman amplification and pulse steepening in silica
fibers, Appl. Phys. Lett., 29:428, 1976.
[2] E. P. Ippen, Low-power quasi-cw Raman oscillator, Appl. Phys. Lett., 16:303, 1970.
[3] R. H. Stolen, E. P. Ippen, and A. R. Tynes, Raman oscillation in glass optical waveguide,
Appl. Phys. Lett., 20:62, 1972.
[4] R. H. Stolen and E. P. Ippen, Raman gain in glass optical waveguides, Appl. Phys. Lett.,
22:276–278, 1973.
[5] K. O. Hill, B. S. Kawasaki, and D. C. Johnson, Low-threshold cw Raman laser, Appl.
Phys. Lett., 29:181, 1976.
[6] C. Lin, R. H. Stolen, W. Pleibel, and P. Kaiser, “A high efficiency tunable cw Raman
oscillator,” Appl. Phys. Lett., 30:162, 1977.
[7] C. Lin, R. H. Stolen, W. G. French, and T. G. Malone, A tunable 1.1 µm fiber Raman
oscillator, Appl. Phys. Lett., 31:97, 1977.
[8] C. Lin, L. G. Cohen, R. H. Stolen, G. W. Tasker, and W. G. French, Near-infrared sources
in the 1–1.3 µm region by efficient stimulated Raman emission in glass fibers, Opt.
Commun., 20:426–428, 1977.
[9] R. K. Jain, C. Lin, R. H. Stolen, and A. Ashkin, A tunable multiple Stokes cw fiber Raman
oscillator, Appl. Phys. Lett., 31:89, 1977.
[10] K. O. Hill, Y. Fujii, D. C. Johnson, and B. S. Kawasaki, Photosensitivity in optical
waveguides: Application to reflection filter fabrication, Appl. Phys. Lett., 32:10, 647,
1978.
[11] R. Kashyap, Fiber Bragg Gratings, San Diego: Academic, 1999.
[12] S. Grubb, T. Erdogan, V. Mizrahi, T. Strasser, W. Y. Cheung, W. A. Reed, P. J. Lemaire, A.
E. Miller, S. G. Kosinski, G. Nykolak, and P. C. Becker, 1.3 µm cascaded Raman amplifier
in germanosilicate fibers. In Proceedings of Optical Amplifiers and Their Applications,
PD3-1, 187, 1994.
[13] S.G. Grubb, T. Strasser, W.Y. Cheung, W.A. Reed, V. Mizrachi, T. Erdogan, P.J. Lemaire,
A.M. Vengsarkar, D. J. DiGiovanni, D. W. Peckham, and B. H. Rockney, High-power 1.48
µm cascaded Raman laser in germanosilicate fibers. In Proceedings of Optical Amplifiers
and Their Applications, SA4, 197–199, 1995.
[14] D. Innis, D. J. DiGiovanni, T. A. Strasser, A. Hale, C. Headley, A. J. Stentz, R. Pedrazzani,
D. Tipton, S. G. Kosinski, D. L. Brownlow, K. W. Quoi, K. S. Kranz, R. G. Huff, R.
Espindola, J. D. Le Grange, and G. Jacobovitz-Veselka, Ultrahigh-power single-mode
fiber lasers from 1.065 to 1.472 µm using Yb-doped cladding-pumped and cascaded
Raman lasers. In Proceedings of the Conference on Lasers and Electro-Optics, CPD-31,
1997.
[15] V. I. Karpov, E. M. Dianov, A. S. Kurkov, V. M. Paramonov, V. N. Protopopov, M.
P. Bachynski, and W. R. L. Clements, LD-pumped 1.48 µm laser based on Yb-doped
double-clad fiber and phosphorosilicate-fiber Raman converter. In Proceedings of the
Optical Fiber Communication Conference, WM3, 202–204, 1999.
380 C. Headley et al.

[16] E. M. Dianov, I. A. Bufetov, M. M. Bubnov, A. V. Shubin, S. A. Vasiliev, O. I. Medvedkov,

S. I. Semjonov, M. V. Grekov, V. M. Paramonov, A. N. Gur’yanov, V. F. Khopin, D.
Varelas, A. Iocco, D. Costantini, H. G. Limberger and R-P. Salathé, CW highly efficient
1.24 µm Raman laser based on low-loss phosphosilicate fiber. In Proceedings of the
Optical Fiber Communication Conference, PD25, 1999.
[17] V. I. Karpov, E. M. Dianov, V.M. Paramonov, O. I. Medvedkov, M. M. Bubnov, S. L. Se-
myonov, S. A. Vasiliev, V. N. Protopopov, O. N. Egorova, V. F. Hopin, A. N. Guryanov, M.
P. Bachynski, and W. R. L. Clements, Laser-diode-pumped phosphosilicate-fiber Raman
laser with an output power of 1 W at 1.48 µm, Opt. Lett., 24:13, 887–889, 1999.
[18] E. M. Dianov, I. A. Bufetov, M. M. Bubnov, M. V. Grekov, S. A. Vasiliev, and O. I.
Medvedkov, Three cascaded 1407-nm Raman laser based on phosphorus-doped silica
fiber, Opt. Lett., 25:6, 402–404, 2000.
[19] W. A. Reed, W. C. Coughran and S. G. Grubb, Numerical modeling of cascaded cw Raman
fiber amplifiers and lasers. In Optical Fiber Communication Conference, WD1, 107–109,
1995.
[20] M. Rini, I. Cristiani, and V. Degiorgio, Numerical modeling and optimization of cascaded
cw Raman fiber lasers, IEEE J. Quantum Electron., QE 36:10, 1117–1122, 2000.
[21] A. Bertoni and G. C. Reali, 1.24-µm cascaded Raman laser for 1.31-µm Raman fiber
amplifiers, Appl. Phys. B 67:5–10, 1998.
[22] G. Vareille, O. Audouin, and E. Desurvire, Numerical optimization of power conversion
efficiency in 1480 nm multi-Stokes Raman fibre lasers, Electron. Lett. 34:7, 675–676,
1998.
[23] S. D. Jackson and P. H. Muir, Theory and numerical simulation of nth-order cascaded
Raman fiber lasers, J. Opt. Soc. Am. B 18:9, 1297–1306, 2001.
[24] K. Rottwitt and H. D. Kidorf, A 92 nm bandwidth Raman amplifier. In Proceedings of
the Optical Fiber Communication Conference, PD6, 1998.
[25] Y. Emori and S. Namiki, 100 nm flat Raman amplifiers pumped and gain equalized by
12-wavelength-channel WDM high power laser diodes. In Proceedings of the Optical
Fiber Conference, PD19-1, 1999.
[26] D. I. Chang, D. S. Lim, M. Y. Jeon, H. K. Lee, K. H. Kim, and T. Park, Dual wavelength
cascaded Raman fiber laser, Electron. Lett., 36:1365–1368, 2001.
[27] S. B. Paperny, V. I. Karpov, and W. R. L. Clements, Efficient dual-wavelength Raman
fiber laser, In Proceedings of the Optical Fiber Communication Conference, WDD15-1,
2001.
[28] M. D. Mermelstein, C. Headley, J.-C. Bouteiller, P. Steinvurzel, C. Horn, K. Fedder and B.
J. Eggleton, A high-efficiency power-stable three-wavelength configurable Raman fiber
laser. In Proceedings of the Optical Fiber Communication Conference, PD3-1, 2001.
[29] M. D. Mermelstein, C. Headley, J.-C. Bouteiller, P. Steinvurzel, K. Feder, and B.J. Eggle-
ton, Configurable three-wavelength Raman fiber laser for Raman amplification and dy-
namic gain flattening, IEEE Photon. Technol. Lett., 13:12, 1286–1288, 2001.
[30] M. D. Mermelstein, C. Horn, Z. Huang, P. Steinvurzel, K. Feder, M. Luvalle, J.-C.
Bouteiller, C. Headley, and B. J. Eggleton, Configurability of a three-wavelength Ra-
man fiber laser for gain ripple minimization and power partitoning. In Proceedings of the
Optical Fiber Communication Conference, TuJ2-1, 2002.
[31] X. Normandin, F. Leplingard, E. Bourova, C. Leclère, T. Lopez, Jean-Jacques Guérin, D.
Bayart, Experimental assessment of phospho-silicate fibers for three wavelength (1427
nm, 1455 nm, 1480 nm) reconfigurable Raman lasers. In Proceedings of the Optical Fiber
Communication Conference, TuB2-1, 2002.
[32] M. D. Mermelstein, C. Horn, J.-C. Bouteiller, P. Steinvurzel, K. Feder, C. Headley, and
B. J. Eggleton, Six wavelength Raman fiber laser for C+L-band Raman amplification.
In Proceedings of the Conference on Lasers and Electro-Optics, CThJ1, 2002.
 11. Raman Fiber Lasers 381

[33] M. D. Mermelstein, C. Horn, S. Radic, and C. Headley, Six-wavelength Raman fibre

laser for C- and L-band Raman amplification and dynamic gain flattening, Electron.
Lett., 38:636–638, 2002.
[34] F. Leplingard, S. Borne, L. Lorcy, T. Lopez, J.-J. Guérin, C. Moreau, C. Martinelli, D.
Bayart, Six output wavelength Raman fibre laser for Raman amplification, Electron. Lett.,
38:886–887, 2002.
[35] K. Rottwitt, A. Stentz, T. Nielsen, P. Hansen, K. Feder, and K. Walker, Transparent 80
km bidirectionally pumped distributed Raman amplifier with second order pumping. In
Proceedings of the European Conference on Optical Communication, II-144-145, 1999.
[36] V. Dominic, A. Mathur, and M. Ziari, Second-order distributed Raman amplification
with a high-power 1370 nm laser diode, In Proceedings of Optical Amplifiers and their
Applications, OMC6, 2001.
[37] L. Labrunie, F. Boubal, E. Brandon, L. Buet, N. Darbois, D. Dufournet, V. Havard, P. Le
Roux, M. Mesic, L. Piriou, A. Tran, and J.-P. Blondel, 1.6 Terabits (160 × 10.66 Gbit/s)
unrepeated transmission over 321 km using second order pumping distributed Raman am-
plification. In Proceedings of the Optical Amplifiers and their Applications, PD3-1, 2001.
[38] P. Le Roux, F. Boubal, E. Brandon, L. Buet, N. Darbois, V. Havard, L. Labrunie, L.
Piriou, A. Tran and J.-P. Blondel, 25 GHz spaced DWDM 160 × 10.66 Gbit/s (1.6 Tbit/s)
unrepeatered transmission over 380 km. In Proceedings of the European Conference on
Optical Communication, PD. M. 1.5, 2001.
[39] J.-C. Bouteiller, K. Brar, S. Radic, J. Bromage, Z. Wang, and C. Headley, Dual-order
Raman pump providing improved noise figure and large gain bandwidth. In Proceedings
of the Optical Fiber Communication Conference, Postdeadline Paper FB3, 2002.
[40] S. B. Paperny, V. I. Karpov, and W. R. L. Clements, Third-order cascaded Raman ampli-
fication. In Proceedings of the Optical Fiber Communication Conference, FB4-1, 2002.
[41] M. Prabhu, N. S. Kim, L. Jianren, and K. Ueda, Simultaneous two-color CW Raman
fiber laser with maximum output power of 1.05 W/1239 nm and 0.95W/1484 nm using
phosphosilicate fiber, Opt. Commun. 182:305–309, 2000.
[42] Alfalight, Boston Laser Inc., IRE-Polus Group, JDSU, LaserTel, SLI, Spectra Physics
Products List.
[43] L. Goldberg, B. Cole, and E. Snitzer, V-groove side-pumped 1.5 µm fibre amplifer,
Electron. Lett., 33:2127–2129, 1997.
[44] L. Goldberg and J. Koplow, Compact, side-pumped 25 dBm Er/Yb co-doped double
cladding fibre amplifier, Electron. Lett., 34:2027–2028, 1998.
[45] IRE-Polus Group Products List.
[46] A. B. Grudin, J. Nilsson, P. W. Turner, C. C. Renaud, W. A. Clarkson, and D. N. Payne,
Single clad coiled optical fibre for high power lasers and amplifiers. In Proceedings of
the Conference on Lasers and Electro-Optics, CPD26-1, 1999.
[47] V. P. Gapontsev, P. I. Sadovsky, and I. E. Samartsev, 1.5 µm erbium glass lasers,
Proceedings of the Conference on Lasers and Electro-Optics, CPDP-38, 1990.
[48] J. D. Minelly, E. R. Taylor, K. P. Jedrzejewski, J. Wang and D. N. Payne, “Laser-diode
pumped neodymium-doped fibre laser with output power >1W,” Proceedings of the
Conference on Lasers and Electro-Optics, CWE-6, 1992.
[49] H. M. Pask, J. L. Archambault, D. C. Hanna, L Reekie, P.St.J. Russell, J. E. Townsend,
and A. C. Tropper, Operation of cladding-pumped Yb3+ -doped silica fibre lasers in 1µm
region, Electron. Lett., 30:11, 863–865, 1994.
[50] E. Snitzer, H. Po, F. Hakimi, R. Tumminelli, and B. C. McCollum, Double-clad, offset
core Nd fiber laser. In Proceedings of the Optical Fiber Communication Conference,
PD5, 1988.
382 C. Headley et al.

[51] H. Po, E. Snitzer, L. Tumminelli, F. Hakimi, N. M. Chu, and T. Haw, Doubly clad high
brightness Nd fiber laser pumped by GaAlAs phased array. In Proceedings of the Optical
Fiber Communication Conference, PD7, 1989.
[52] P. B. Hansen, L. Eskildsen, S. G. Grubb, A. J. Stentz, T. A. Strasser, J. Judkins, J. J.
DeMarco, R. Pedrazzani, and D. J. DiGiovanni, Capacity upgrades of transmission
systems by Raman amplification, IEEE Photon. Technol. Lett., 9:2, 262–264, 1997.
[53] Y. Qian, J. H. Povlsen, S. N. Knudsen, and L. Grüner-Nielsen, On Rayleigh backscat-
tering and nonlinear effects evaluations and Raman amplification characterizations of
single-mode fibers. In Proceedings of the Optical Amplifiers and their Applications
Conference, OMD18, 2000.
[54] F. L. Galeener, J. C. Mikkelsen, Jr., R.H. Geils, and W.J. Mosby, The relative Raman cross
sections of vitreous SiO2 , GeO2 , B2 O3 , and P2 O5 , Appl. Phys. Lett., 32:34–36, 1978.
[55] N. Shibata, M. Horigudhi, and T. Edahiro, Raman spectra of binary high-silica glasses and
fibers containing GeO2 , P2 O5 and B2 O3 , J. Non-Crystalline Solids, 45:115–126, 1981.
[56] D. Marcuse, Light Transmission Optics, New York: Van Nostrand Reinhold, Chap. 8,
1982.
[57] V. M. Mashinsky, E. M. Dianov, V. B. Neustruev, S. V. Lavrishchev, A. N. Guryanov, V.
F. Khopin, N. N. Vechkanov, and O. D. Sazhin, UV absorption and excess optical loss in
preforms and fibers with high germanium content. In Fiber Optic Materials and Compo-
nents, H. H. Yuce, D. K. Paul, R. A. Greenwell, ed. Proc. SPIE, 2290:105–112, 1994.
[58] E. M. Dianov, V. M. Mashinsky, V.B. Neustruev, O. D. Sazhin, A. N. Guryanov, V. F.
Khopin, N. N. Vechkanov, and S. V. Lavrishchev, Origin of excess loss in single-mode
optical fibers with high GeO2 -doped silica core, Optic. Fiber Technol. 3:77–86, 1997.
[59] L. Grüner-Nielsen, High index fibers, thesis for Industrial Ph.D., EF 546/Ph.D. No.
94-0146-ATV, Danish Academy of Technical Sciences, May, 1998.
[60] M.E. Lines, W.A. Reed, D.J. DiGiovanni, and J.R. Hamlins, Explanation of anomalous
loss in high delta singlemode fibres, Electron. Lett., 1009–1010, 1999.
[61] A. S. Kurkov, V. M. Paramonov, O. I. Medvedkov, S. A. Vasiliev, and E. M. Dianov,
Raman fiber laser at 1.45 µm: Comparison of different schemes. In Proceedings of
Optical Amplifiers and Their Applications, OMB5, 16–18, 2000.
[62] C. R. S. Fludger, B. Handerek, and R. J. Mears, Pump to signal RIN transfer in Raman
fibre amplifiers, J. Lightwave Tech., 19:8, 1140–1148, 2001.
[63] M. D. Mermelstein, C. Headley, and J.-C. Bouteiller, RIN transfer analysis in the pump
depletion regime for Raman fiber amplifiers, Electron. Lett., 38:403–405, 2002.
[64] N. Tsukiji, N. Hayamizu, H. Shimizu, Y. Ohki, T. Kimura, S. Irino, J. Yoshida, T.
Fukushima, and S. Namiki, Advantage of inner-grating-muti-mode laser (iGM-laser) for
SBS reduction in copropagating Raman amplifier. In Proceedings of Optical Amplifiers
and their Applications, OMB4, 2002.
[65] L. L. Wang, R. E. Tench, L. M. Yang, and Z. Jiang, Linewidth limitations of low noise,
wavelength stabilized Raman pumps. In Proceedings of Optical Amplifiers and their
Applications, OMB5, 2002.
[66] B. J. Eggleton, J. A. Rogers, P. B. Westbrook, and T. A. Strasser, Electrically tunable
power efficient dispersion compensating fiber Bragg grating, IEEE Photon. Technol.
Lett., 11:854–856, 1999.
Chapter 12
Distributed Raman Transmission: Applications
and Fiber Issues

Alan Evans, Andrey Kobyakov, and Michael Vasilyev

12.1. Historical Introduction

After discovery of a new physical effect, the speed to widespread commercialization
is dictated by complementary technologies and market need. In the case of the Ra-
man effect, the spectral shift of scattered light from “molecules in dust-free liquids or
gases” observed in 1928 [1, 2] has long been used for the spectroscopic characteriza-
tion of materials. Yet application of it to amplification in optical fiber has had a long
wait: both for the development of the key enabling technologies, namely, high-power
pump lasers and low-loss optical fiber, and for the commercial need for significantly
reduced cascaded optical noise buildup in fiber transmission systems.
Even before the availability of high-power sources become a critical issue, the
field had to wait for low-loss optical fiber. Raman gain followed soon after 20 dB/km
loss fiber was demonstrated and in fact was measured on that same fiber [3]. Another
early achievement was the spectral characterization of gain in different waveguide
materials [4]. In the 1980s researchers turned to distributed Raman amplification for
lossless soliton propagation: first for compensating the loss-induced pulse broaden-
ing of 10 ps solitons through 10 km of fiber [5], followed by periodic amplification
over 4000 km [6] and then 20 Gb/s single-channel transmission over 70 km [7]. Dur-
ing that same time, the rich history of Raman-preamplified repeaterless transmission
for improved receiver sensitivity began [8]. In 1987 the first semiconductor laser
diodes (at 60 mW) [9] were used as pumps and soon after greater than 100 mW semi-
conductor laser diodes were combined with a then-novel pump/signal wavelength-
division-multiplexer [10]. However, the commercial development of erbium-doped
fiber amplifiers (EDFAs) in the early 1990s quickly supplanted Raman effect as the
fiber-optic amplifier technology of choice primarily due to erbium’s high signal gain
and high pump efficiency.
Although the telecommunications industry rapidly began deployment of EDFAs
by the mid1990s, the work on Raman technology still continued, forgetting partic-
ularly the amplification in the 1310 nm wavelength band for cable television and
as a high-power fiber laser via cascaded Raman gain in fiber Bragg grating res-
onators. During that same period, the length of repeaterless submarine links was being
384 A. Evans et al.

extended by 1480 nm wavelength pumping of remotely located erbium coils. Pump-

ing through the transmission fiber from either the transmitter or receiver end or both
provided additional gain from Raman amplification translating to an increase in loss
budget and repeaterless distance. The earliest transmission of significant length and
bit rate was in 1991 demonstrating an increase in 620 Mb/s transmission reach from
301 to 375 km using a remotely pumped erbium amplifier [11]. Another milestone
was the first 10 Gb/s repeaterless transmission in 1995 using both remotely located
post- and preamplifiers and high-power laser diodes over 411 km of optical fiber [12].
These early transmission experiments improved understanding of the system ben-
efits of distributed amplification. Soon Raman-assisted repeatered transmission com-
bining distributed Raman with discrete inline EDFAs were used to control the balance
between fiber nonlinearities and noise. At the Optical Fiber Communications Confer-
ence of 1999, two system papers demonstrated distributed Raman amplification could
keep four-wave-mixing in check even for 50 GHz spaced channels across the zero
dispersion wavelength of dispersion-shifted fiber [13, 14]. Optical signal-to-noise
ratio (OSNR) was maintained through distributed Raman amplification and launch
power was lowered to −12.8 dBm [13]. The flood gate was opened with lower noise-
distributed Raman amplification pushing all transmission performance dimensions:
total capacity [15–17], spectral efficiency [16, 18, 19], data rate [20, 21], number of
spans (i.e., total system reach) [22–24], and span loss [25–27]. Of these, ultra-long
(>1500 km) 10 Gb/s and long-haul (>600 km) 40 Gb/s have become leading con-
tenders for commercial deployment due to the overall system cost savings they can
provide. Raman for 10 Gb/s transmission holds the promise of longer optical trans-
mission before the need for expensive electrical regeneration. Longer distances are
required as the Internet rather than voice dominates transmission traffic. Raman for
40 Gb/s transmission holds the promise of maintaining optical links at current values
(∼80 to 100 km) while pushing system reach well beyond 600 km.
This chapter begins by focusing on the two most important properties of an op-
tical amplifier—pump efficiency and noise figure—and compares Raman to erbium
amplification. The concept of effective noise figure is reviewed and leads to a generic
system scaling relationship that aids in the prediction of Raman-assisted, system per-
formance improvements. The next sections review Raman transmission experiments
at 10 and 40 Gb/s and cover design issues specific to these systems. The last section
of the chapter introduces dispersion-managed fiber consisting of optical fiber spans
that can be optimized for Raman transmission.

12.2. Raman Amplifier Issues

12.2.1. Pump Efficiency

Gain is the fundamental property of interest for an amplifier and high pump effi-
ciency is what makes a particular technology practical. For point-to-point, pream-
plified transmission, gain from inline amplifiers compensates the loss from the
preceding span coming from the combination of cabled fiber, splices, and connec-
tors. Span lengths in US terrestrial long-haul fiber links are between 50 and 120 km,
 12. Applications and Fiber Issues 385

requiring gain between 10 and 27 dB with 80 to 100 km being more typical with gain
between 20 and 25 dB.
EDFAs have always been able to provide this level of gain at low pump power
from readily available semiconductor laser diodes. The relevant figure of merit is
pump efficiency, measured in terms of the percentage of pump photons converted
to signal photons. The range for a typical coil of erbium-doped fiber is 65 to 80%.
Including all discrete loss elements of an EDFA, the efficiency drops to 20 to 30%. In
comparison, the transmission fiber efficiency for distributed Raman is less than 1%: it
is a weak, nonlinear scattering process with a small gain per unit length (known as the
Raman gain coefficient gR ). In addition, the need to depolarize the Raman pump light
to avoid polarization-dependent gain means only half the pump photons are available
for stimulated scattering at any given location in the fiber. Finally, there is a several
more dBs drop in efficiency to account for the insertion loss of the Raman module
at both the signal and pump wavelengths and connector/patch panel loss between
the module and the transmission fiber. As an example, to provide all the gain of an
average length of standard single-mode fiber (90 km), the total pump power within
the fiber would have to be over 1 W. Add to this the well-known issues of efficient
pump-to-signal noise transfer, pump-mediated signal crosstalk, and double-Rayleigh
backscatter (DRBS) crosstalk and it is not surprising that EDFAs rapidly displaced
Raman in the early 1990s.
Several things have changed to the benefit of Raman in the intervening years.
First, counterpropagating the Raman pump light with respect to the signals greatly
mitigates pump-to-signal noise transfer and pump-mediated signal crosstalk. Second,
better understanding of the gain, fiber effective area, and receiver bandwidth depen-
dence of DRBS crosstalk allows for system designs that avoid its onset. Third, low
degrees of polarization via polarization-multiplexing of equal intensity pump diodes
are easily implemented to control polarization-dependent gain. Finally, better pack-
aging techniques allow for much higher optical power density on optical components
in the high-power pump path.
Most important, great progress has been made on increasing the output power
Raman sources. As the available pump power has increased, so has interest in de-
ployment of Raman amplifiers. Cascaded Raman fiber lasers were the early front
runner as Raman pumps and now provide power well over 1 W and up to six inde-
pendently power-tunable wavelengths [28–30]. Most early repeaterless experiments
used these sources. Semiconductor laser diodes offer power in reliable, compact, and
standard packages. Recent reports of 350 mW per diode [31] mean 1 W into the
transmission fiber is achievable for a typical four-diode, two-wavelength figuration
and an illustrative 1.5 dB pump path loss to the fiber [32].

12.2.2. Effective Noise Figure

Progress in pump power, crosstalk, and PDG alone is not enough to reignite Raman as
a viable, economic amplification technology for fiber-optic communications. Revived
interest is due to the fact that Raman amplification in transmission fiber has lower noise
buildup than conventional EDFAs. One reason is that the distributed nature of Raman
386 A. Evans et al.

20
Post-Amp

Relative Signal Power

10 Pre-Amp
Gain = 3 dB
Increasing Raman Gain
(dB) 0 Gain = 6 dB
Gain = 9 dB
–10 Gain = 15 dB
Gain = 23 dB
–20 2X50 km
0 25 50 75 100
Distance (km)

Fig. 12.1. Signal power evolution through 100 km of transmission fiber for various amplification
configurations. The solid curves of increasing signal power near the output end of the fiber are
for counterpropagating Raman pumping at several power/gain values; 2 × 50 km equals two
fiber spans of 50 km each.

amplification allows it to act as a low-noise preamplifier to the subsequent inline

discrete amplifier.Another is that twice the number of amplifiers in a hybrid distributed
Raman/discrete EDFA system means much less gain and noise per amplifier.
All optical amplifiers have the inseparable processes of stimulated and sponta-
neous emission. The former provides the desired signal gain through coherent addition
of pump photons to signal photons; amplification of the later (i.e., amplified sponta-
neous emission, ASE) gives the unavoidable ASE–ASE and ASE–signal beat noise at
the receiver. It is well known that distributed amplification gives a value of ASE den-
sity and fiber nonlinear impairments between that of post- and preamplification [33].
The signal power evolution through a span of transmission fiber in Fig. 12.1 helps
to explain why. In postamplification, the signal undergoes gain and the added noise
before launch into the fiber. The high signal launch (and high path-average) power
yields high fiber nonlinearities but excellent OSNR because both the signal and a
fixed amount of ASE power propagate together through the same fiber loss. Postam-
plification OSNR at the end of the span is Ps /(ρASE ν) where ρASE is the ASE
power density of the amplifier before the fiber span, Ps is the signal launch power,
and ν is the bandwidth of the optical filter before the receiver. In preamplification,
the signal undergoes fiber loss before entering a discrete inline amplifier. The low sig-
nal launch power yields low fiber nonlinearities but poor OSNR because the signal
is attenuated before the addition of the same amount of ASE power. Preamplification
OSNR is reduced by the factor 10Tf /10 where Tf (dB) < 0 is the fiber transmittance.
Based on signal power evolution, it should not be surprising that distributed amplifica-
tion is intermediate between these extremes; however, nonideal, counterpropagating
Raman pumps allow for a continuous knob to retune the noise/fiber nonlinearity
balance. This is especially relevant inasmuch as today’s dense wavelength-division-
multiplexed systems are effectively preamplified because fiber nonlinearities strongly
limit launch power.
 12. Applications and Fiber Issues 387

Understanding the distributed nature of Raman comes from comparing it with

erbium. In EDFAs, the strong absorption and emission cross-section in the three-
level erbium ion transition means strongly attenuated pump power and efficient gain
conversion within a few tens of meters of erbium-doped gain fiber. In Raman, there is
no real energy transition and Raman scattering weakly attenuates the pump. Most of
the pump is lost to Rayleigh scattering which sets an effective penetration length of
the Raman gain into transmission fibers of ∼20 km effectively distributing the gain
over a large fraction of the transmission span. So, it’s the low pump efficiency that
leads to distributed gain over a substantial length of the fiber. Higher Raman pump
power raises the minimum signal power and pushes it farther into the fiber as shown
in Fig. 12.1. Although some amount of Raman gain exists at every distance within the
span, the location of this signal minimum can be considered as the input of an effective
amplifier. It has been shown that the generation of ASE in the distributed, backward-
pumped Raman amplifier can be approximated by injection of one noise photon per
mode at this effective input [34]. Given this, we expect the OSNR to improve by
approximately the ratio of minimum signal powers in Raman and preamplified EDFA
cases, i.e., higher Raman gain yields lower noise impact, as shown in Fig. 12.2.
Another simple but intuitive interpretation of the effect of Raman is that the
addition of any amplifier within a transmission span will reduce noise. Cumulative
ASE density is proportional to the product of the number of spans and the amplifier
gain and noise figure in linear units. Consider the example of a single 100 km span
of 20 dB loss compared to two 50 km spans of 10 dB each. If the loss-compensating
amplifiers have the same noise figure, the total ASE density drops by a factor of
1 × 100/(2 × 10) = 5. The equal amplifier spacing of this example is optimal. For
Raman, the farther the gain penetrates into the fiber, the more the hybrid system looks
like two equally spaced amplifiers.
It is convenient to replace the complicated distributed, distance-dependent gain
and ASE density with an equivalent lumped amplifier at a discrete location in the
fiber not only for the simple interpretation described previously, but also for the
bookkeeping of cascaded gain and noise buildup throughout a multispan optical link.
Lumping the Raman gain and noise at the end of the transmission span allows the use
of the usual cascaded noise figure equation provided an effective Raman noise figure
is used: FT = Feff + (FEDFA /TL − 1)/GR , where FT is the total noise figure, Feff is
the Raman effective noise figure, GR is the ratio of signal power at the span output
with and without the Raman pumps on (i.e., on/off gain),
1 ( ρASE,EDFA )
FEDFA = +1
GEDFA hv
is the EDFA noise figure, and TL < 1 is the fractional discrete loss between the Raman
gain at the end of the fiber and the EDFA amplifier module (in linear units). A more
descriptive name for Feff is differential noise figure because a differential gain (i.e.,
GR rather than GR × span loss) is used in its definition: ρASE,Raman ≡ hν(GR Feff − 1),
where ρASE,Raman is the Raman ASE power density, h is Plank’s constant, and ν is
the optical frequency. ρASE,Raman is the physical quantity of interest and Feff loses its
388 A. Evans et al.

1 1st (1527.5 nm) of

48 Channels

0.5
48th (1565.5nm) of
48 Channels
Noise Figure (dB)

–0.5 1st of 48, 0.05

dB/km extra fiber loss
–1

1 Channel
–1.5 (1527.5nm), 1
Pump (1432nm)

–2
0 5 10 15 20
Raman Gain (dB)
(a)

8 1st (1527.5 nm) of

48 Channels
7

6
Total F, 1st of 48,
5 F (EDFA) = 6.5dB
Noise Figure (dB)

3 Total F, 1st of 48,

F (EDFA) = 6.5dB,
2 L = 1 dB
1

–1

–2
0 5 10 15 20
Raman Gain (dB)
(b)

Fig. 12.2. (a). Raman effective noise figure as a function of Raman gain for different pump
and signal configurations. Lines with diamonds, squares, and triangles have pumps at 1432 nm
and 1461 nm wavelength. The solid line has a single pump at 1432 nm and a single signal at
1527.5nm. (b). Raman-only (line with diamonds) and total noise figure as a function of Raman
gain with added 0 dB (solid line) and 1 dB (dashed line) extra insertion loss between the Raman
and EDFA modules and assuming an EDFA noise figure of 6.5 dB.
 12. Applications and Fiber Issues 389

physical meaning as the ratio of output to input electrical signal-to-noise ratio. Hence
the effective noise figure can be less than three, or even zero, dB but the total (and
physically accurate) noise figure of a distributed Raman amplifier needs to add the
front end loss of the amplifying fiber and is much greater than the quantum noise limit.
The use of effective noise figure can sometimes lead to counterinitiative results.
Consider the numerical example shown in Fig. 12.2(a) where the effective noise for
90 km of nonzero dispersion-shifted fiber (NZDSF) pumped at 1432 and 1461 nm is
plotted versus Raman on-off gain. The solid curve is for a single 1527.5 nm channel
at the gain peak of a single 1432 nm pump. This represents the best-case, lowest noise
figure. Adding more pumps and signals decreases the pump penetration depth. The
highest noise figure occurs for the shortest wavelength, in this case the first of 48 C-
band channels from 1527.5 to 1565.5 nm, for a number of reasons: (1) pump-to-pump
Raman gain allows the higher wavelength pump to penetrate deeper into the fiber so
the higher wavelength signals have more distributed gain than the lower wavelength
signals; (2) the phonon energy distribution means that the ground state is more heavily
populated for lower wavelength signals; (3) the spectral attenuation of the pump and
C-band signals is greater for shorter wavelengths; (4) the signal-to-signal Raman gain
means that for spectrally flat gain, the gain of shorter wavelengths occurs closer to
the output end of the fiber. One effect, gain saturation, increases the noise figure for
all wavelengths but changes the tilt in noise figure in the opposite direction: to higher
values at longer signal wavelength (see Chapter 4 for more details on the spectral
dependence of the noise figure). Another interesting point about the effective noise
figure is that it decreases with increasing fiber attenuation. In the figure, the fiber
loss at both the pump and signal wavelengths has increased by 0.05 dB/km and the
resulting first channel noise figure, shown in squares, decreases by over 0.5 dB at 20
dB on-off gain.
The final issue addressed in this section is the total noise figure of hybrid dis-
tributed Raman/discrete EDFAs. Assuming a constant 6.5 dB noise figure for the
subsequent EDFA independent of input power, the total noise figure for the first chan-
nel of our 48 channel example is shown as the solid line in Fig. 12.2(b). Adding a
1 dB loss between the Raman and EDFA module to account for the 14XX/1550 nm
wavelength-division-multiplexer or patch panel connector loss gives the dashed line
in the same figure. This example illustrates that minimizing the “midstage” loss of
this hybrid amplifier is particular important at low Raman gain.

12.3. Raman Q Scaling

12.3.1. Rule of Thumb
This section describes a simple Q scaling relationship [35] that is useful for pre-
dicting the relative system performance improvement when the total amplifier noise
figure decreases due to distributed Raman amplification. As a relative scaling, it is
robust against most system parameters. However, it assumes that systems are limited
by the noise from signal–spontaneous ASE beat noise and interchannel crossphase
modulation (XPM) and four-wave-mixing (FWM). Predictions from this simple
scaling will not be accurate for systems limited by dispersion, gain nonuniformity,
390 A. Evans et al.

polarization mode dispersion, or single-channel nonlinearities. However, where the

scaling is known for these effects, the relationship can be generalized. Additional loss
due to network elements is particularly easy to add.
System Q, a measure of the electrical signal-to-noise ratio, is directly related to the
bit error rate (BER). Many of the contributions to Q can be ignored in a well-designed
system: detector thermal noise, signal and ASE shot noise, amplifier spontaneous–
spontaneous beat noise, and laser relative intensity noise. In typical 10 Gb/s long-
haul dense wavelength-division-multiplexing (WDM) systems, signal–spontaneous
emission beat noise is the dominant noise mechanism and XPM and FWM are often
the dominant nonlinear effects. Considering only ASE-generated signal–spontaneous
beat noise and treating XPM and FWM as Gaussian noise, system Q has the following
functional dependence on channel power P , referenced to the output power of the
inline optical amplifier.
P
Q= √ . (12.1)
aP + bP 4
The coefficient a governs the signal–spontaneous beat noise and the coefficient b
governs the combined XPM and FWM noise terms. As four-photon parametric pro-
cesses, the XPM and FWM can be shown to depend on P 4 . A typical Q dependence
on launch power is shown in Fig. 12.3, using Eq. (12.1). The a and b coefficients
have been set equal to 6.54 × 10−3 so that at −1 dBm launch power, the maximum
Q, in 20 log dB units, is 18 corresponding to a BER of 10−15 . Equation (12.1) can
be re-written as √
3Q0 (P /P0 )
Q= + . (12.2)
2(P /P0 ) + (P /P0 )4
The significance of this form is that the peak value of Q = Q0 occurs at P = P0 . The
relationship between a and b, and P0 and Q0 is
2P0 ( a )1/3
a= P 0 =
3Q20 2b
 1/6
1 4
b= Q0 = (12.3)
3P02 Q20 27a 2 b
with the coefficient a scaling as
a ∝ N FT GT ,
where FT is the total amplifier noise figure (linear units), N is the number of spans,
and G is the total amplifier gain (linear units); the coefficient b scales as
N
b∝ 2 . (12.4)
Aeff
The interesting aspect is that the maximum Q scales as
1 1 1 1/3
Q0 ∝ 1/3 , 1/3 , 1/2 , and Aeff .
F G N
One nonobvious conclusion is that a 1 dB improvement in the amplifier noise figure
does not give a 1 dB improvement in Q2 as one would expect from considering only
 12. Applications and Fiber Issues 391

25
Noise Limited Nonlinearity Limited

20 log(System Q )
20

15

10

0
–10 –8 –6 –4 –2 0 2 4 6
Launch Power per Channel (dBm)

Fig. 12.3. A graph of Q versus launch power per channel from Eq. (12.1) with the maximum
Q = 18 for −1 dBm launch power.

the signal–spontaneous beat noise. In fact, inclusion of nonlinear impairments implies

that the improvement is only 0.66 dB. In general, inclusion of the nonlinear Q power
scaling is important in scaling to a different number of spans, gain per span, noise
figure, channel spacing and effective area.

12.3.2. Scaling Verification

To verify the scaling for Raman applications, a number of numerical transmission

simulations were performed. The base case contained no Raman amplifiers with
other parameters as follows: 6 spans of nonzero dispersion-shifted fiber of length 100
km, 32 channels at 10 Gb/s, 100 GHz channel spacing from 1536 nm to 1562 nm,
fiber loss of 0.25 dB/km, 100% slope-compensating fiber of 0.4 dB/km loss, ignoring
polarization mode dispersion (PMD) and gain ripple. The base case 20 log Q was
20.4 dB. For this Q, coefficients a and b in Eq. (12.3) were assumed to be equal to
one. From this base case, Raman was added with 10 dB gain and −1.6 dB effective
noise figure, neglecting gain ripple and additional fiber nonlinearities that could be
caused by a change in the channel power evolution; that is, only the improvement in
cascaded amplifier noise figure was included. The system Q, with and without Raman,
was calculated changing a number of system parameters: the number of spans was
varied from 5 to 8 without Raman and 5 to 30 with Raman; the channel spacing was
changed to 50 GHz for 5 to 8 spans; the data rate was increased to 40 Gb/s with 100
and 200 GHz spacing; and span length was changed to 5 × 120 and 4 × 150 km. For
each system, the launch power per channel was reoptimized.
The success of scaling is illustrated in Fig. 12.4 which plots 17 system transmission
simulations of 10 log(Q) versus 10 log(4/(27a 2 b)) on the right y-axis and upper x-
axis and optimum launch power per channel versus 10 log(a/(2b)) on the left y-axis
and lower x-axis. As expected, all points fall on a line with slope 1/6 for the Q
dependence and 1/3 for the power dependence. As previously mentioned, this should
be valid scaling for any NRZ or RZ modulated 10 Gb/s system, limited by crosschannel
nonlinearities. It may also work for other modulation formats or data rates.
392 A. Evans et al.

10 log(4/(27a 2b))
–35 –30 –25 -20 –15 –10 –5 0 5 10
10 P 0: 100km Spans 16
P 0: Longer Spans
8 14
P 0=(a /2 b )^(1/3)
6
Q 0: 100km Spans 12
Q 0: Longer Spans
10 log 10 log
P0 4 Q 0=(4/27a ^2 b)^(1/6) 10 Q0

2 8

0 6

–2 4
–30 –25 –20 –15 –10 –5 0 5 10 15
10 log(a /(2b))

Fig. 12.4. Plot of 17 system simulations with respect to scaling parameters. The 10 logs of
Eq. (12.4) are shown as solid and dotted lines for power and Q, respectively.

12.4. 10 Gb/s
Hybrid Raman/EDFA systems combine the lower noise of distributed Raman am-
plifiers with the higher gain and pump efficiency of discrete EDFAs. The resulting
Q improvement can be used to improve various system parameters: span length, as
in repeaterless submarine experiments of the early to mid1990s; increased channel
count as in the repeatered dispersion-shifted fiber demonstrations; spectral density
as in the more recent results greater than 0.4 bit/s Hz; increased number of spans
described in the following section; and higher data rate as described in Section 12.5.
The scaling rule of thumb can be used to explain why these systems work.

12.4.1. 10 Gb/s System Design

The predominant system application of Raman amplification is increasing the number
of spans to meet the needs of today’s changing long-haul traffic patterns. Internet-
initiated connections are much less distance-dependent than are voice connections,
and IP packages are rapidly dominating total transmission bandwidth. The 10 Gb/s
NRZ system simulations described in the last section are shown in Fig. 12.5. In
Fig. 12.5(a), Q versus launch power per signal shows that the addition of Raman with
an effective noise figure of −1.6 dB shifts the maximum Q launch power lower by
2 dB. The optimum launch power has a broad maximum with Q decreasing at lower
power due to OSNR degradation and at higher power due to fiber nonlinearities. In
Fig. 12.5(b), Q versus number of spans shows that the addition of Raman increases
Q for a fixed number of spans by ∼4 dB or increases system reach for a fixed Q by
2.5 times.
A key design consideration is the appropriate gain partitioning between the dis-
tributed and discrete gain of Raman-assisted transmission. Factors include the opti-
cal signal-to-noise ratio, fiber nonlinearities, DRBS-induced multipath interference
 12. Applications and Fiber Issues 393

26 Raman Off Raman On

25

20 log Q (dB)
24
Approximate
23 Optimum Power
22
21
20
2 3 4 5 6
Launch Power per Channel (dBm)
(a)

Large 3R Spacings
26
Raman Off
25
Raman On
24

23
Q [dB]

22

21

20

19

18
0 5 10 15 20 25 30
Spans [#]
(b)

Fig. 12.5. (a). 10 Gb/s NRZ simulations showing the dependence of Q on launch power
per channel for five span NRZ transmission with (squares) and without (diamonds) Raman
amplification for effective noise figure of −1.6 dB. (b). 10 Gb/s NRZ simulations showing
the dependence of Q on the number of 100 km spans of NZDSF with (circles) and without
(diamonds) Raman amplification for effective noise figure of −1.6 dB.

(MPI) crosstalk, and even the cost of amplification. Although EDFA costs are rather
insensitive to design gain, distributed Raman gain (in dB) costs increase in propor-
tion to the pump power (in mW) because they operate in the linear, undepleted pump
regime. The lower pump efficiency of Raman means that the cost of gain is higher than
EDFAs. Therefore, total amplifier costs increase as the fractional gain in distributed
Raman increases. Cost-effective gain partitioning is set by increasing the distributed
Raman gain to the necessary level to achieve acceptable system Q, but no higher.
Although double Rayleigh backscattering is covered in another chapter, it is im-
portant enough to warrant some comments here. For the 48 channel, 0 dBm launch
394 A. Evans et al.

per channel example we have considered, the DRBS MPI crosstalk is proportional to
Raman gain independent of channel position [36]. The power evolution is substan-
tially the same for all channels of the example. Other examples could be considered
where this is not the case, however, channel position is a second-order effect. Another
point is that DRBS crosstalk can be factored into the system performance by adding it
to the ASE noise figure [37]. As a distributed interference term with random phase, its
noise variance is Gaussian: σMPI2 = r 2 PRx (1)2 /2Xtalk, where we have assumed that
the average receiver power is half the peak power (true for the pseudorandom NRZ
modulation format) and Xtalk is the crosstalk power ratio between the signal and the
double-Rayeigh backscattered signal [36b]. The total noise figure is the sum of statis-
tically independent, Gaussian contributions. Modulation rate plays an important role
because the noise variance from ASE is directly proportional to receiver bandwidth.
So the relative contribution of DRBS MPI-induced noise to total noise is larger at
10 Gb/s than at 40 Gb/s for the same span launch power. The DRBS MPI-induced
noise dependence on Raman gain is stronger than for ASE-induced noise, therefore,
the total noise of 10 Gb/s systems is degraded at lower Raman gain than in 40 Gb/s
systems. For our present example, the minimum noise figure occurs at 13.1 dB for
10 Gb/s and 16.7 dB for 40 Gb/s [36]. On the other hand, if the system nonlinearity
budget can tolerate a 6 dB increase in fiber launch power corresponding to the 6 dB
increase in receiver bandwidth for 40 Gb/s, then there would be negligible difference
in total noise figure between 10 Gb/s and 40 Gb/s.
It is worth noting that the impact of DRBS can be significantly reduced when the
wavelengths of the signal and the backward-propagating pump are on opposite sides
of the zero-dispersion wavelength. In that case, DRBS spectrum broadens very rapidly
owing to the gain- and cross-phase-modulation by a noisy Raman pump, and hence
the fraction of the total DRBS power that falls into the received signal bandwidth is
too small to cause any noticeable degradation [36a].

12.4.2. 10 Gb/s Raman Transmission Experiments

Recent improvements in transmission performance enabled by Raman amplification

are best seen in a compilation of the recent transmission experiments as shown in Ta-
ble 12.1 [13, 14, 22, 24, 26, 38–51]. Although there has been much work at lower data
rates and single-span transmission dating back into the 1980s, the first single-span
10 Gb/s Raman-assisted transmission was not until 1995 [12] and multispan with
7200 km reach followed two years later [38]. As previously mentioned, several early
papers applied distributed Raman to the problem of four-wave-mixing in dispersion-
shifted fiber where the zero dispersion wavelength is in the middle of the C-band.
Fujitsu researchers instead pushed on the reach performance dimension, showing
1 Tb/s (100 × 10 Gb/s) over 10,000 km [22], and doubling to 2 Tb/s over the some-
what shorter 7200 km reach the following year [23]. These have been followed by a
couple of S-band transmission experiments at modest total capacity and single wide-
band transmission [45, 46]. All-Raman amplification, where the transmission fiber is
pumped close to transparency by a combination of counter- and copropagating pumps
and, optionally, the dispersion-compensating fiber having separate Raman pumps,
 Table 12.1. Summary of Repeatered 10 Gb/s Raman Transmission Experiments
Total Capacity Reach
Gb/s (km) Raman Gain Claim to Fame Group Year Ref.
100 7200 2 dB higher OSNR in 514 km chain First high capacity multispan WDM Lucent 1997 [38]
at 10 Gb/s
400 5 × 120 12 dB 4 dB power margin in multispan Lucent 1999 [39]
WDM system
490 4 × 84 440 mW 50 GHz spacing in DSF Lucent 1999 [13]
250 8 × 84 440 mW 100 GHz but farther than 50 GHz Lucent 1999 [13]
320 80 × 80 15.3 dB backward, 4.3 dB forward 50 GHz spacing in DSF NTT 1999 [14]
1000 200 × 50 1.1 dB 1 Tb/s over 10,000km, C+L Fujitsu 1999 [22]
1000 4 × 80 18 dB backward, 4 dB forward 25 GHz spacing in DSF NTT 1999 [40]
320 6 × 125 10 dB 6 passes through a 640 × 640 WS × C Corning 2000 [41]
1000 3× ∼ 133 10 dB 25 GHz spacing in L-band Lucent 2000 [42]
1280 6 × 140 5.8 dB in C, 5.4 dB in L Best loss × span and Distance × Length Fujitsu 2000 [26]
320 3 × 250 1W Submarine medium haul Alcatel 2000 [43]
90 × 80 Distance × span length product using

12. Applications and Fiber Issues

2110 7 dB Fujitsu 2000 [23]
asym DMF
400 6 × 100 All-Raman 20 − 24 dB +650/250 mW Highest S-band distance × bit rate product Lucent 2001 [45]
200 10 × 87 21 dB Discrete S-band slope-DCF Xtera 2001 [46]
400 40 × 100 23.6 dB All-distributed Raman >80 km spans Lucent 2001 [47]
using sym DMF
2400 185 × 40 1.5 dB OSNR improvement Tb/s greater than 2500 km, single 74 nm Fujitsu 2001 [48]
amplification
1280 40 × 100 20 dB co + counter in NZDSF + 12 dB in DCF 53 nm all-Raman over 100 km spans Lucent 2001 [49]
800 52 × 80 20.5 dB Reconfigurable OADMS with DMF Corning 2002 [50]
and all-distributed Raman
2560 ∼270 × 40 not mentioned Widest transoceanic bandwidth, Tyco 2002 [24]
single Raman pump

395
800 52 × 80 20.5 dB Capacity × distance record (no FEC) for Corning 2002 [51]
all-distributed Raman using sym DMF
396 A. Evans et al.

offers complete bandwidth flexibility in terms of placement and width [45, 47, 49, 51].
An intriguing hybrid Raman/EDFA combination offered simple yet very wide band
amplifiers and a very long reach system [24].
A few conclusions are possible from Table 12.1. First, most major system houses
are participating in using Raman amplifiers to demonstrate proof-of-concept, record-
breaking transmission. Second, there is no standard Raman gain value: it ranges from 1
to 22 dB depending on the trade-off with other system parameters. Third, tremendous
progress has been made over the last few years in system performance: terabits per
second over much greater than 1000 km. It should be noted though, that Raman is
only one of the recently added tools; forward error correction (FEC), lower gain ripple
amplifiers, and better dispersion slope compensation and dispersion-managed fiber
spans are others.

12.5. 40 Gb/s
For 40 Gb/s bit rate systems, noise accumulation is a challenging design issue, as it
seriously limits amplifier spacing and system reach. Cost-effective terrestrial trans-
mission systems require amplifier spacing of at least 80 km and at least five to six
spans between electrical regeneration. Distributed Raman-assisted transmission of-
fers the necessary low-noise amplification for economical, field-deployed 40 Gb/s
systems. In addition, discrete Raman, possibly in the form of a pumped dispersion-
compensating fiber, overcomes the bandwidth limitations created by the discrete
energy levels of erbium in EDFAs. Lastly, carefully designed all-Raman systems
combining distributed and discrete amplifiers could reduce the number of discrete
components compared to EDFA-based systems. These components are a source of
PMD, polarization-dependent loss (PDL), and group delay ripple that impairments
need careful consideration for 40 Gb/s.

12.5.1. Noise

For a given receiver sensitivity, the 25% reduction in bit period in going from 10
to 40 Gb/s implies a 6 dB increase in launch power for an equivalent amplifier Q
(i.e., the system Q of Eq. (12.1) for b = 0). However, this increase is usually not
possible due to system degradation from increased fiber nonlinearities. In fact, as
previously mentioned, this nonlinearity-limiting launch power causes today’s long-
haul systems to operate in a preamplified mode rather than the postamplified mode.
Without any change in technology, in particular the amplification noise figure, the
scaling of amplifier Q implies that the number of spans decreases by a factor of
four when moving from 10 to 40 Gb/s. Or for a fixed number of spans, the amplifier
gain needs to decrease by a factor of four. For example, with a fiber attenuation of
0.25 dB/km, a 6 dB gain reduction is 24 km of span length. Yet with distributed
Raman amplification the increased receiver bandwidth can be counteracted by an
overall noise figure reduction. In fact, for very high Raman gain, noise improvement
greater than 6 dB can be used to increase the number of spans. Looking at the Raman
 12. Applications and Fiber Issues 397

noise figure, the solid curve of Figure 12.2(b), a total noise figure improvement greater
than 6 dB occurs for Raman gain greater than ∼10 dB, meaning a 40 Gb/s hybrid
Raman/EDFA system will have less noise buildup than a 10 Gb/s, EDFA-only system.
That noise improvement can be converted to a fractional reach increase as shown with
the Q scaling equation.

12.5.2. Other Amplifier Design Issues

Although optical amplification is inherently bit rate-independent, PMD, PDL, and

group delay ripple within the optical components, as well as differences in gain and
loss requirements, typically prevent the transparent bit rate upgrade of systems. Dif-
ferences in the gain/loss map arise from gain partitioning reoptimization among the
distributed and discrete amplifiers, different dispersion-compensation maps, and dif-
ferent required optical networking functionality and their associated insertion losses.
Whereas the PMD requirement scales inversely with bit rate and its interaction with
PDL sets specifications at the limit of current component manufacturing capability.
Some components, particularly those based on grating and dielectric thin films, have
high-frequency wavelength variations in their waveguide propagation constants. Such
group delay ripple becomes a problem for the wider signal bandwidth of 40 Gb/s.
The exact requirements for PMD, PDL, and group delay ripple is system specific
and dependent on contributions from the other elements of the optical link (trans-
mission and dispersion-compensating fiber, dynamic spectral equalizers, wavelength
add-drop multiplexers and optical switches, etc.). The contribution from Raman am-
plifiers is expected to be less than from EDFAs due to lower component counts
making them an attractive option for 40 Gb/s systems. Typical distributed Raman
amplifiers only have a single WDM to couple pump light into the transmission fiber.
Given that roughly half the gain comes from this amplifier, the discrete amplifier,
whether EDFA or Raman-based, should have fewer stages of amplification. In addi-
tion, discrete Raman-based pumping of dispersion compensating fiber eliminates the
erbium-doped fiber coils that can be a source of PDL.

12.5.3. 40 Gb/s Raman Transmission Experiments

Raman amplifier technology was well timed to be introduced into 40 Gb/s trans-
mission experiments as higher speed transmitters, electrical amplifiers, and receivers
were becoming available. A complete list (to the best of the author’s knowledge)
of repeatered, 40 Gb/s Raman transmission experiments is given in Table 12.2
[15–17, 27, 52–68]. Experiments by Lucent and NEC in 1999 proved that Raman
allowed for the same amplifier spacing as lower bit rates: multiple spans of 80 to 100
km [52, 53]. The following year, a record capacity 3 Tb/s transmission over 3 × 100
km of NZDSF was enabled by combining Raman and a well-matched dispersion slope
compensating fiber with C+L -band transmission [15]. Pushing a new performance
dimension, Nortel showed that 160 to 200 km amplifier spacing and a record 1000 km
reach for 1.28 Tb/s was possible [27]. In that same European Conference of Opti-
cal Communications 2000 session, Alcatel pushed total capacity to 5.12 Tb/s using
 Table 12.2. Summary of Repeatered 40 Gb/s Raman Transmission Experiments

398
Total Capacity Reach
(Gb/s) (km) # of λs Raman Gain Claim to Fame Group Year Ref.
4 × 100

A. Evans et al.
1600 40 23 dB, 520 mW no OTDM, Pmux, or FEC, 0.4 bit/Hz Lucent 1999 [69]
320 5 × 80 8 9 dB, 110 mW RDF, 80 km amp + 40 Gb/s NEC 1999 [53]
3280 3 × 100 82 25 dB in C, 24 dB in L Record capacity Lucent 2000 [15]
1280 3 × 100 32 5 to 8 dB Full ETDM, 0.4 bit/s/Hz spectral efficiency Alcatel 2000 [54]
5120 3 × 100 128 12 dB in C, 10 dB in L 0.64 bit/s/Hz spectral efficiency Alcatel 2000 [16]
1280 6 × 160/200 32 5 to 10 dB Record span length and loss Nortel 2000 [27]
640 6 × 120 16 7 to 10 dB CS-RZ, 0.4 bit/s/Hz over 120 km spans Nortel 2001 [55]
3080 12 × 100 77 23.5/22.5 dB 3 Tb/s using dual C and L Raman Lucent 2001 [56]
10920 2 × 58 273 5.5 dB First 10 Tb/s—Raman in S-band NEC 2001 [17]
3200 3 × 82 80 8 dB 3 Tb/s in field fiber WorldCom, Siemens 2001 [57]
640 25 × 80 16 8 dB Gain partitioning optimized (1/2 span loss) KDD 2001 [58]
1280 60 × 40 32 12.2 dB Tb/s greater than 2000 km, all-Raman + DMF Alcatel 2001 [59]
5000 12 × 100 125 15 dB in NZDSF + 8 dB in DCF 2 Tb/s higher capacity at same distance than line 9 Alcatel 2001 [60]
1600 20 × 100 40 not mentioned Longest reach over terrestrial span length OFS 2001 [61]
using dispersion-managed spans
3200 3 × 100 80 14 dB in NZDSF + 6 dB in DCF 3.2 Tb/s in C-band only with 0.8 bit/s-Hz, no PDM Alcatel 2001 [62]
1600 24 × 100 40 15.5 Combination of capacity, span length, and OFS 2002 [63]
distance over commercial available fiber
2560 40 × 100 64 not mentioned Capacity × distance record of 10 Petabit-km/s, Lucent 2002 [64]
Raman used but not main enhancement
10240 3 × 100 256 not mentioned 10 Tb/s over 100 km spans, counterpropagating Alcatel 2002 [65]
second-order Raman + pumped DCF
1280 1160 × 52 32 11.5 dB all-Raman First 40 Gb/s transoceanic distance, NEC 2002 [66]
quadruple-hydrid span
1600 36 × 100 40 7.5 dB forward Record number of 100 km spans at 40 Gb/s Mintera 2002 [67]
+ 18.5 dB backward all-Raman
3200 20 × 100 80 All-Raman: ∼6dB co-, Higher capacity/lower reach than previous OFS 2002 [68]
∼15 dB counter, ∼6 dB in DCF
 12. Applications and Fiber Issues 399

vestigial sidebandlike filtering to reach a spectral efficiency of 0.64 bits/s/Hz [16].

The large signal-to-signal Raman interaction across the entire C+L 72 nm band was
partially compensated using distributed Raman amplification. Then in 2001, 3 Tb/s
was pushed to 1200 km with FEC, tight residual dispersion control, and Raman [56];
10 Tb/s in one fiber was reached using S+C+L and 273 wavelengths [17]; and
multiple Tb/s (80 × 40 Gb/s) field trials brought the technology closer to commercial
reality [57]. Six transmission demonstrations in the first half of 2002 continued to
add new complementary technologies to enhance performance. Raman itself keeps
amplifier spacing at the commercially viable 80 to 100 km range. In combination with
other technologies such as FEC, dispersion slope compensation, dispersion-managed
fiber spans, gain equalization, and alternate modulation formats it enables up to 10
Tb/s capacity, greater than 0.4 bit/s/Hz spectral efficiency, and reach over 3000 km.
Raman has become an indispensable tool in the system designer’s toolbox.
Another trend for Raman is toward more complex pumping schemes: forward
pumping of the transmission fiber, pumping the dispersion-compensating fiber, or
even second-order co- or counterpropagating pumping. For one thing, these tricks
allow for all-Raman transmission without any EDFAs. For another, the signal power
evolution is flattened, leading to a more optimal balance between the constraining
effects of ASE beat noise for low launch power and fiber nonlinearities for high launch
power. In fact, flat signal power along the fiber is ideal. The dispersion-managed fiber
described in the following section is another approach to attain this ideal.

12.6. Raman Amplification in Dispersion-Managed Fibers

As we discussed above, the distributed optical amplification improves OSNR com-
pared to lumped preamplification and reduces fiber nonlinearities compared to lumped
postamplification. (We used the same convention for definition of pre- and post-
amplification as in Section 12.2.2.) The optimum trade-off between the nonlinearities
and noise is achieved by the distributed amplifier that balances gain and loss at ev-
ery point in the fiber, making the signal power constant versus distance. One of the
key trends in the evolution of Raman amplifiers is the progress toward approaching
the performance limit of this ideal distributed amplifier. Combining backward (i.e.,
counterpropagating with the signal) and forward (copropagating) Raman pumping, as
previously described, has recently been complemented by other methods that greatly
facilitate achieving optimum signal profile in the span. The first is the use of effective-
area management in the span (e.g., by employing the dispersion-managed fiber, or
DMF) to distribute the Raman gain more evenly along the fiber [47, 70–74]. The sec-
ond is the demonstration of feasibility of second- and higher-order Raman pumping in
a distributed amplifier, which is another way of reaching the same goal [30, 75, 76a].
This section deals with Raman amplification in DMFs that have recently gained
a significant amount of interest [23, 47, 50, 51, 66, 77, 78]. The potential of DMFs
as the fiber of choice in next-generation high-capacity transmission systems has been
demonstrated lately by several research groups (see Tables 12.1 and 12.2). Below we
discuss peculiarities of noise performance caused by inhomogeneous fibers with non-
axially uniform optical properties and show that an optimally pumped DMF closely
400 A. Evans et al.

(within a few dB) approaches the performance of an ideal Raman amplifier based on
a single fiber.
DMFs, which are hybrid fiber spans consisting of sections having positive (+D)
and negative (−D) dispersion, were originally designed for the dispersion and dis-
persion slope compensation. It turns out, however, that they also offer significant
benefits for Raman amplification. High pumping efficiency in the small effective area
−D fiber, combined with improved noise performance discussed below are the main
factors that contribute to the DMF advantage over a single fiber type. The reduction
of the midstage loss requirement of an EDFA with the removal of the dispersion-
compensation module allows for a further reduction in the effective noise figure of
a Raman-assisted transmission system. In addition, DMFs were shown to be more
robust with respect to intrachannel nonlinearities [79].

12.6.1. Improved Noise Performance of DMF-Based Raman Amplifiers

Apart from more complicated DMF configurations [66], two types of DMF design
are most common: +D/ − D (asymmetric) and +D/ − D/ + D (symmetric). As we
have described previously, a more uniformly distributed Raman gain entails less ASE
penalty. This also holds true for the DMF where the desirable deeper penetration of
the pump into the span can be achieved by proper management of the effective area.
The idea of effective area management is to extend the region of useful Raman
gain deeper into the fiber span (Fig. 12.6(a)) by using large effective area fiber (e.g.,
+D section of DMF) near the pump launch point and small effective area fiber
(e.g., −D section of DMF) at some distance from it. This results in improved OSNR
(Fig. 12.6(b)). The symmetric DMF design with −D fiber in the middle of the span is
particularly attractive. As can be seen from Fig. 12.6(b), the minimum signal power
in the symmetric configuration is higher than that in the asymmetric configuration,
leading to superior noise performance in the +D/ − D/ + D design.
The asymmetric design has other disadvantages. Although placing the −D fiber at
the end of the DMF span maximizes the Raman gain efficiency, the smaller effective
area in the −D fiber enhances intensity-dependent nonlinear impairments. In addition,
because most of the Raman gain is generated in the small effective area −D fiber at
the exit end of the span, this type of DMF is highly DRBS-prone (DRBS is inversely
proportional to the square of effective area). In contrast, for symmetric DMF only a
fraction of the total Raman gain is generated in the −D fiber, so the DRBS crosstalk
is reduced, which, in turn, reduces the minimum effective noise figure and shifts
the minimum to a higher Raman gain. These trends are illustrated in Fig. 12.7 [70]
where the measured effective lumped noise figure of a distributed Raman amplifier
is shown for single fiber, symmetric DMF, and asymmetric DMF. A dramatic noise
figure improvement (∼2.5 dB) of symmetric DMF over asymmetric DMF and single
fiber can be seen.
As discussed in Section 12.2, a more homogeneously distributed Raman gain
provides better noise performance. On the other hand, it increases the path-average
power (Fig. 12.6(b)) of the signal and therefore induces more nonlinear penalties.
The question arises as to whether the noise figure advantage of the DMF is large
 12. Applications and Fiber Issues 401

Effective Raman length

1/p

–5
Relative pump intensity (dB)

–10

–15

–20

–25 Single-mode fiber

Symmetric DMF
–30
Asymmetric DMF
–35
0 20 40 60 80 100
Distance (km)
(a)

+D –D +D

–5

–10 No Raman
Relative power (dB)

–15 With Raman

Symmetric DMF
–20
Spontaneous
emission level
–25 Signal-to-noise ratio

–55
h ∆, or “1 photon per mode”
–60
0 20 40 60 80 100
Distance (km)
(b)

Fig. 12.6. Pump (a) and signal (b) evolution in a single fiber (with and without backward
Raman pumping) and symmetric DMF with backward Raman pumping. A schematic drawing
of symmetric DMF is shown above the graph.
402 A. Evans et al.

–1
NF, dB

–2

Single fiber
–3
Asymmetric 1:1 DMF
Asymmetric 2:1 DMF
–4
Symmetric 1:1 DMF
Symmetric 2:1 DMF
–5
0 5 10 15 20
Raman gain, dB

Fig. 12.7. Noise figure versus Raman gain in single fiber and two different x:1 configurations
of the DMF.

enough to compensate for nonlinear impairments associated with increased path-

average power. The answer requires a more elaborate analysis that includes accounting
for bidirectional pumping, effects of DRBS, and postspan component loss.
To resolve this issue, we note that in the span where the fiber parameters vary with
distance, constant signal power versus distance is no longer the optimum evolution
profile. If we ignore the effect of Rayleigh scattering, the best trade-off between the
ASE and nonlinearity is achieved for [80]
-
Aeff (z)αs (z) n2 (0)
G(z) = × , (12.5)
n2 (z) Aeff (0)αs (0)

where G(z) is the signal power normalized to its launch value Pin (for a single fiber,
the optimum profile is G(z) = 1), n2 is the nonlinear refractive index, Aeff is the
effective area, and αs is the signal attenuation coefficient. If the only parameter that
changes with distance is the effective area, then the optimum signal profile given
in Eq. (12.5) is the geometric average of the constant-power and constant-intensity
profiles. For transparent (i.e., G(L) = 1) spans with the same span loss and the same
amount of nonlinear penalties characterized by the nonlinear phase shift
L
2π n2 (z)
φN L = P in G(z) dz,
λ 0 Aeff (z)
the OSNRs of the ideal distributed single-fiber amplifier and DMF with the ideal
 12. Applications and Fiber Issues 403

signal profile (12.5) are related by

OSNR idealDMF n2 A+ (x + αs− /αs+ )(x + 1)
= + eff * , (12.6)
OSNR ideal n2 Aeff [x + α − n− A+ /(α + n+ A− )]2
s 2 eff s 2 eff

where x is the length ratio of +D and −D span portions, and “+”, “−”, and no
superscripts denote +D, −D, and single-fiber parameters, respectively. For x = 2
and parameters listed in Table 12.3 below, the ideal DMF case offers 0.8 dB OSNR
advantage over the ideal single-fiber case.
When Rayleigh scattering is taken into account, a more elaborate analysis is re-
quired. The next section describes optimization of a realistic bidirectionally pumped
DMF, including the effects of DRBS, Rayleigh scattered ASE, and postspan compo-
nent loss.

12.6.2. Optimization of Raman Amplifiers Based on DMFs

The on-off Raman gain found from the undepleted pump approximation in an inho-
mogeneous fiber can be written as
 z 





GR (z) = exp γ (z )αp (z )pp (z ) dz , (12.7)
0
where piecewise functions of distance
gR (z)PR
γ (z) = ,
Aeff (z)αp (z)
αp , and gR are, respectively, the dimensionless Raman interaction strength in a
nonpolarization-maintaining fiber, the loss at the pump wavelength, and the Raman
gain coefficient. The evolution of the total (forward and backward) normalized pump
has the form
 L   z 
pp (z) = (1 − k) exp − αp (z
) dz
+ k exp − αp (z
) dz
, (12.8)
z 0
where k = Pf (0)/PR is the ratio of forward pump power to the total launched pump
power PR , and L is the span length. The total gain G(z) is related to the on-off gain
by G(z) = GR (z)ps (z). It is also assumed that linear signal attenuation
 z 
ps (z) = exp − αs (z
) dz

0

Table 12.3. Fiber Parameters Used in Calculations

Fiber αs αp αRBS × 10−4 Aeff gR /Aeff × 10−3 n2 × 10−20
Type [dB/km] [dB/km] [1/km] [µm2 ] [m−1 W−1 ] [m2 /W]
+D 0.19 0.22 0.4 110 0.2 1.8
−D 0.25 0.29 1.6 30 1.0 2.3
SF 0.21 0.26 1.0 55 0.64 2.2
404 A. Evans et al.

and pump attenuation pp (z) include lumped loss at the point of fiber splice z = zsplice .
Similarly to single fiber [81], the effective lumped noise figure of an inhomogeneous
fiber can be calculated as
 L
ASE = 1 γ (z)αp (z)pp (z)
FDMF + 2nsp TF dz
GR (L) 0 G(z)
L 
αRBS (z) L





+ γ (z )α p (z )p p (z )G(z ) dz dz ,
0 G2 (z) z
(12.9)
where nsp = {1 − exp'−hν/(kB Tfib )(}−1is the spontaneous emission factor that
depends on the fiber temperature Tfib and the pump–signal frequency difference ν (h
and kB are Planck and Boltzmann constants, respectively); TF < 1 is transmittance of
a passive fiber, and αRBS is the Rayleigh backscatter coefficient. For completeness, we
have included the effect of backscattered ASE (second term in brackets in Eq. (12.9)
that contributes at high Raman gains [82].
As a reference point for the DMF, we consider a single fiber with the same total
span loss as the DMF (equal to fiber loss plus splice loss) and noise only from
forward-propagating ASE. This means that our reference system is an ideal distributed
amplifier with optimum trade-off between ASE and nonlinearity in the absence of
DRBS (see above), that is, with a flat gain profile G(z) = 1. To ensure equal nonlinear
penalties in both systems, we scale signal power to the DMF in such a way that the
nonlinear phase shifts φNL in both spans are the same. Thus, if Pref is the input signal
power to the single fiber span, the signal power to the DMF span is PDMF = Pref /RN L ,
where for equal transmittances TF of the DMF and a single fiber
L
αs,ref n̄2 (z)
RN L = GDMF (z) dz (12.10)
ln(1/TF ) 0 Āeff (z)
In Eq. (12.10), αs,ref is the absorption coefficient at the signal wavelength in a single
fiber, n̄2 (z) = nDMF
2 (z)/n2,ref , and Āeff (z) = ADMF
eff (z)/Aeff ,ref . Under these assump-
tions, the ratio of electric signal-to-noise-ratios (SNR) is given by [81] as
SNR DMF 1 Fref − 1
RSNR = = total − 1
,
SNR ref RN L FDMF
where we have neglected the effect of electrical shot noise. For a single fiber, we
have Fref = 1 + 2nsp ln(1/TF ). In the DMF-based system, we also account for the
postspan loss TL of system components such as a gain-flattening filter or an add-
drop multiplexer, so that TF GR (L)TL = 1 and the total noise figure after one span
total = F
FDMF DMF /TF + 1 − TL . After substitution of these relations into the expression
for RSNR we obtain
1 2nsp ln(1/TF )
RSNR = , (12.11)
RN L (FDMF /TF ) − TL
where the noise figure of the DMF,
5 Pref TF X
FDMF = FDMF
ASE
+
9 RN L hνBeeff
 12. Applications and Fiber Issues 405

includes the contribution from the DRBS crosstalk,

L
αRBS (z) L
X= αRBS (z
)G2 (z
) dz
dz,
0 G2 (z) z

and Beeff is the effective electrical filter bandwidth that accounts for optical signal
bandwidth [81].

12.6.3. Single Fiber Versus DMF: A Performance Comparison

Parameters of a generic DMF [83] and single fiber used in calculations are listed in
Table 12.3. We assume the ratio between the lengths of +D and −D sections equals
two, the splice loss at signal and pump wavelengths equals 0.3 dB, Beeff = 8 GHz,
nsp = 1.13, and Pref = 0.5 mW.
First, we compare the performance of the DMF and single fiber without postspan
loss (TL = 1). As can be seen from Fig. 12.8, the optimum amount of forward pumping
weakly depends on span length and amounts to kopt ≈ 0.3 − 0.4 for the DMF case and
kopt ≈ 0.5 for the single fiber case. For short spans, the performance advantage due
to forward pumping is small (<0.3 dB). It increases with the span length and reaches
∼2 dB for L = 100 km (Fig. 12.8(a)). The corresponding RSNR for the single fiber with
realistic gain profile and impaired by DRBS is shown in Fig. 12.8(b). Comparison
of Fig. 12.8(a) and 12.8(b) shows 0.5 dB SNR advantage for an 80 km span DMF
and 1.1 dB for a 100 km span DMF. In Fig. 12.9 we plot RSNR for 100 km long
DMF (Fig. 12.9(a)) and single fiber (Fig. 12.9(b)) spans, pumped to transparency, as
a function of forward pumping ratio k and the postspan loss TL . One can see that for
low postspan loss, the SNR advantage of the DMF is about 1.2 dB (for TL = 0 dB) but
drops to about 0.7 dB for TL = 6 dB. Figure 12.9 shows that use of forward pumping
is especially important in the systems with high postspan loss.

1.0 1.0
– 3.3 – 3.9 – 4.7 –3. –3. – 4. –5.5
–2
.5 3 9 7
0.8 –3.3 0.8 –2
–2
.1
–2.5 .5 –3. –3.9 – 4.7
3
0.6 –1 –2.1 0.6
.9 –1.9
k

0.4 0.4 .5
–1.9
–2 3 –3.9 .7
–2.1
–2.1 –3. – 4 .5
0.2 –2.5 0.2 –5
–2.5 –3.3 .3 –3.94.7
–3.9 –3 –
0.0 0.0
60 70 80 90 100 110 120 60 70 80 90 100 110 120
L, km L, km
(a) (b)

Fig. 12.8. Relative electrical signal-to-noise ratio, RSNR , calculated from Eq. (12.11) for (a)
DMF and (b) single fiber without postspan loss as a function of span length L and amount of
forward pumping κ; 0 dB corresponds to ideal single-fiber amplifier with G(z) ≡ 1.
406 A. Evans et al.

1.0 – 4.8 – 5.6 – 6.8 1.0

– 6.8
– 4.0
–3.6 –5.6 – 6.8 –5.6 – 6.8
0.8 –3.2 – 4.8 0.8 – 4.8
–2.8 – 4.0 –5.6
–3.6 – 4.0 – 4.8
–2.4 –3.2 – 3.6
k 0.6 – 4.0 0.6 – 4.0

k
0.4 0.4 – 3.6 – 4.0 – 4.8
2 – 3.6
–2.
4
–2.8 – 3. – 4.0 – 4.8 – 5.6
– 4.8 – 5.6 – 6.8
0.2 0.2
– 3.2 – 3.6 – 4.0 – 5.6 – 6.8
– 4.8 – 6.8
– 5.6
0.0 0.0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
TL , dB TL , dB
(a) (a)

Fig. 12.9. Relative electrical signal-to-noise ratio, RSNR , for (a) DMF and (b) single fiber for
constant span length L = 100 km as a function of postspan loss, TL and amount of forward
pumping κ.

12.7. Conclusions
This chapter has tried to put Raman amplification in a historical context. In the early
days of high gain WDM transmission, it was not a clear winner for amplification. The
persistent demand for higher performance (capacity, system reach, data rate, etc.)
has turned system designers back to distributed Raman for its lower noise figure.
The weak nonlinear interaction of Raman that results in poor pump efficiency also
gives the distributed gain benefit of lower noise. Today’s data-dominated traffic pat-
terns require reach beyond 1000 km. Raman is one vital tool in pushing out system
reach together with more exotic modulation formats, improved cascaded forward
error-correction algorithms, lower PMD components and fiber, tighter amplifier gain
ripple specifications and gain equalizers, and better fixed and dynamic dispersion
compensaters.
Dispersion-managed fiber spans offer another vehicle for better distributing the
signal gain. From the above DMF analysis one can conclude that in systems with low
signal power when noise figure is the key performance factor, the performance of
DMF overcomes that of the single fiber because of a much better noise figure. DMF
represents a cost-effective solution particularly for short spans where backward-only
pumping is nearly as good as the optimum forward/backward pumping configura-
tion. It also avoids the pump noise issues inherent with forward pumping. For in-
creased signal power when the nonlinear penalties are important, the difference in
noise performance between DMF and single fiber is smaller. Nevertheless, the opti-
mally pumped DMF still has about 1 dB SNR advantage over the single fiber. This
advantage increases in practice because the dispersion map of DMF enables bet-
ter tolerance to nonlinearities than with single fibers, and because the single fiber
also requires dispersion compensation at the amplifier site that introduces additional
penalties.
 12. Applications and Fiber Issues 407

References
[1] C. V. Raman and K. S. Krishnan, A new type of secondary radiation, Nature, 121:3048,
501, 1928.
[2] G. Landsberg and L. Mandelstam, Naturwissen, 16:28, 557, 1928.
[3] R. H. Stolen and E. P. Ippen, Raman gain in glass optical waveguides, Appl. Phys. Lett.,
22:6, 276–281, 1973.
[4] F. L. Galeener, J. C. Mikkelsen, R. H. Geils, and W. J. Mosby, The relative Raman cross
sections of vitreous SiO2, GeO2, B2O3 and P2O5, Appl. Phys. Lett., 32:1, 34–36, 1978.
[5] L. F. Mollenauer, R. H. Stolen, and M. N. Islam, Experimental demonstration of soliton
propagation in long fibers: Loss compensated by Raman gain, Optics Lett., 10:5, 229–
231, 1985.
[6] L. F. Mollenauer and K. Smith, Demonstration of soliton transmission over more than
4000 km in fiber with loss periodically compensated by Raman gain, Optics Lett., 13:8,
675–677, 1988.
[7] K. Iwatsuki, K. Suzuki, S. Nishi, M. Saruwatari, and K. Nakagawa, 20 Gb/s optical
soliton data transmission over 70 km using distributed fiber amplifiers, IEEE Photon.
Technol. Lett., 2:12, 905–907, 1990.
[8] J. Hegarty, N. A. Olsson, and L. Goldner, CW pumped Raman preamplifier in a 45
km-long fibre transmission system operating at 1.5 µm and 1 Gbit/s, Electron. Lett.,
21:7, 290–292, 1985.
[9] N. Edagawa, K. Mochizuki, and Y. Iwamoto, Simultaneous amplification of wavelength-
division-multiplexed signals by a highly efficient fibre Raman amplifier pumped by
high-power semiconductor lasers, Electron. Lett., 23:5, 196–197, 1987.
[10] Y. Tamura, S. Shikii, H. Maeda, E. Nishimura, and M. Kawahara, Fiber Raman amplifier
module with semiconductor laser pump source. In Proceedings of the ECOC 87: Thir-
teenth European Conference on Optical Communication. Technical Digest, Helsinki,
Finland, 1987.
[11] K. Aida, J. Nakajima, and K. Nakagawa, Long-span repeaterless IM/DD optical trans-
mission experiment over 300 km using optical amplifiers. In Proceedings of the ICC 91
International Conference on Communications Conference Record, New York, 1991.
[12] P. B. Hansen, L. Eskildsen, S. G. Grubb, A. M. Vengsarkar, S. K. Korotky, T. A. Strasser,
J. E. J. Alphonsus, J. J. Veselka, D. J. DiGiovanni, D. W. Peckham, D. Truxal, W. Y.
Cheung, S. G. Kosinski, and P. F. Wysocki, 10 Gb/s, 411 km repeaterless transmission
experiment employing dispersion compensation and remote post- and pre-amplifiers.
In Proceedings of the 21st European Conference on Optical Communications, Gent,
Belgium, 1995.
[13] P. B. Hansen, A. Stentz, T. N. Nielsen, R. Espindola, L. E. Nelson, and A. A.
Abramov, Dense wavelength-division multiplexed transmission in “zero-dispersion”
DSF by means of hybrid Raman/erbium-doped fiber amplifiers. In Proceedings of the
OFC/IOOC’99, Optical Fiber Communication Conference and the International Con-
ference on Integrated Optics and Optical Fiber Communications, Piscataway, NJ, 1999.
[14] N. Takachio, H. Suzuki, H. Masuda, and M. Koga, 32*10 Gb/s distributed Raman am-
plification transmission with 50-GHz channel spacing in the zero-dispersion region over
640 km of 1.55- mu m dispersion-shifted fiber. In Proceedings of the OFC/IOOC’99, Op-
tical Fiber Communication Conference and the International Conference on Integrated
Optics and Optical Fiber Communications, Piscataway, NJ, 1999.
[15] T. N. Nielsen, A. J. Stentz, K. Rottwitt, D. S. Vengsarkar, Z. J. Chen, P. B. Hansen, J.
H. Park, K. S. Feder, T. A. Strasser, S. Cabot, S. Stulz, D. W. Peckham, L. Hsu, C. K. Kan,
408 A. Evans et al.

A. F. Judy, J. Sulhoff, S. Y. Park, L. E. Nelson, and L. Gruner-Nielsen, 3.28 Tb/s (82*40

Gb/s) transmission over 3*100 km nonzero-dispersion fiber using dual C- and L-band
hybrid Raman/erbium doped inline amplifiers. In Proceedings of the Optical Fiber
Communication Conference, Technical Digest Postconference Edition, Trends in Optics
and Photonics, Vol. 37, Washington, DC, 2000.
[16] S. Bigo et al., 5.12 Tbit/s (128 × 40 Gbit/s WDM) transmission over 3 × 100 km Ter-
alight fiber. In Proceedings of the European Conference on Optical Communications,
Munich, Germany, Sept. 3–7, 2000.
[17] K. Fukuchi et al., 10.92-Tb/s (273 × 40-Gb/s) triple-band/ultra-dense WDM optical-
repeatered transmission experiment. In Proceedings of the Optical Fiber Communica-
tions, Anaheim, CA, March 22, 2001.
[18] W. S. Lee, 2.56 Tb/s capacity, 0.8 b/Hz-s DWDM transmission over 120 km NDSF
using polarization-bit-interlevaed 80 Gb/s OTDM signal. In Proceedings of the Optical
Fiber Communications, 2001.
[19] B. Mikkelsen, G. Raybon, B. Zhu, R. J. Essiambre, P. G. Bernasconi, K. Dreyer, L. W.
Stulz, and S. N. Knudsen, High spectral efficiency (0.53 bit/s/Hz) WDM transmission of
160 Gb/s per wavelength over 400 km of fiber. In Proceedings of the OFC 2001, Optical
Fiber Communication Conference and Exhibition, Technical Digest, Washington, DC,
2001.
[20] B. Mikkelsen, G. Raybon, R.-J. Essiambre, A. J. Stentz, T. N. Nielsen, D. W. Peckham,
L. Hsu, L. Gruner-Nielsen, K. Dreyer, and J. E. Johnson, 320-Gb/s single-channel pseu-
dolinear transmission over 200 km of nonzero-dispersion fiber, IEEE Photon. Technol.
Lett., 12:10, 1400–1402, 2000.
[21] U. Feiste et al., 160 Gb/s transmission over 116 km field-installed fiber using 160 Gbit/s
OTDM and 40 Gbit/s ETDM. In Proceedings of the Optical Fiber Communications,
Anaheim, CA, 2001.
[22] T. Haito et al., 1 Terabit/s WDM transmission over 10,000 km. In Proceedings of the
European Conference on Optical Communications, 1999.
[23] T. Tanaka et al., 2.1-Tbit/s WDM transmission over 7,221 km with 80-km repeater
spacing, In Proceedings of the European Conference on Optical Communications, 2000.
[24] D. G. Foursa et al., 2.56 Tb/s (256 × 10 Gb/s) transmission over 11,000 km using hybrid
Raman/EDFAs with 80 nm of continuous bandwidth. In Proceedings of the Optical
Fiber Communications, Anaheim, CA, 2002.
[25] M. X. Ma, H. D. Kidorf, K. Rottwitt, F. W. Kerfoot, III, and C. R. Davidson, 240-
km repeater spacing in a 5280-km WDM system experiment using 8*2.5 Gb/s NRZ
transmission, IEEE Photon. Technol. Lett., 10:6, 893–895, 1998.
[26] T. Terahara, T. Hoshida, J. Kumasako, and H. Onaka, 128*10.66 Gbit/s transmission
over 840 km standard SMF with 140 km optical repeater spacing (30.4 dB loss) em-
ploying dual-band distributed Raman amplification. In Proceedings of the Optical Fiber
Communication Conference. Technical Digest Postconference Edition. Trends in Optics
and Photonics, Vol. 37, Washington, DC, 2000.
[27] Y. Zhu et al., 1.28 Tbit/s (32 × 40 Gbits/s) transmission over 1000 km with only 6 spans.
In Proceedings of the European Conference on Optical Communications, 2000.
[28] M. D. Mermelstein, C. Headley, J.-C. Bouteiller, P. Steinvurzel, C. Horn, K. Feder, and
B. J. Eggleton, A high-efficiency power-stable three-wavelength configurable Raman
fiber laser. In Proceedings of the OFC 2001, Optical Fiber Communication Conference
and Exhibition. Technical Digest, Washington, DC, 2001.
[29] M. D. Mermelstein, C. Horn, S. Radic, and C. Headley, Six-wavelength Raman fibre
laser for C- and L-band Raman amplification and dynamic gain flattening, Electron.
Lett., 38:13, 636–638, 2002.
 12. Applications and Fiber Issues 409

[30] S. B. Papernyi, V. I. Karpov, and W. R. L. Clements, Third-order cascaded Raman

amplification. In Proceedings of the Optical Fiber Communications, Anaheim, CA,
2002.
[31] Y. Qian et al., High-performance AlGaInAs/InP 14xx-nm semiconductor pump lasers
for optical amplifications. In Proceedings of the Asia-Pacific Optical Communication
(APOC) Conference, October, 2002.
[32] P. Kim, J. Park, and N. Park, Performance optimization of distributed Raman amplifier
using optical pump time domain reflectometry. In Proceedings of the Optical Fiber
Conference, 2002.
[33] A. Yariv, Signal-to-noise considerations in fiber links with periodic or distributed optical
ampification, Optics Lett., 15:9, 1064–1066, 1990.
[34] R. G. Smith, Optical power handling capacity of low loss optical fibers as determined
by stimulated Raman and Brillouin scattering, Appl. Optics, 11:2489–2494, 1972.
[35] J. Grochocinski, personal communication, 1999.
[36] A. Belahlou et al., Fiber Design Considerations for 40 Gb/s Systems, Fiber design
considerations for 40 Gb/s systems, J. Lightwave Technol., 20(12):2290–2305, 2002.
[36a] S. Gray, M. Vasilyev, and K. Jepsen, “Spectral broadening of double Rayleigh backscat-
tering in a distributed Raman amplifier,” presented at Optical Fiber Communication
conference, 2001.
[36b] A. Kobyakov, S. Gray, and M. Vasilyev, Quantitative analysis of Rayleigh cross-talk in
Raman amplifiers, Electron. Letters, 39:9, 732-733, 2003.
[37] S. Burtsev, W. Pelouch, and P. Gavrilovic, Multi-path interference nose in multi-span
transmission links using lumped Raman amplifiers. In Proceedings of the Optical Fiber
Communications, 2002.
[38] M. Nissov, C. R. Davidson, K. Rottwitt, R. Menges, P. C. Corbett, D. Innis, and N. S.
Bergano, 100 Gb/s (10*10 Gb/s) WDM transmission over 7200 km using distributed
Raman amplification. In Proceedings of the Eleventh International Conference on Inte-
grated Optics and Optical Fibre Communications 23 European Conference on Optical
Communications IOOC-ECOC97 (Conf. Publ. No.448), London, 1997.
[39] A. K. Srivastava, L. Zhang, Y. Sun, J. W. Sulhoff, and C. Wolf, System margin en-
hancement with Raman gain in multi-span WDM transmission. In Proceedings of the
OFC/IOOC’99, Optical Fiber Communication Conference and the International Con-
ference on Integrated Optics and Optical Fiber Communications, Piscataway, NJ, 1999.
[40] H. Suzuki, J. Kani, H. Masuda, N. Takachio, K. Iwatsuki, Y. Tada, and M. Sumida,
25 GHz-spaced, 1 Tb/s (100*10 Gb/s) super dense-WDM transmission in the C-band
over a dispersion-shifted fibre cable employing distributed Raman amplification. In
Proceedings of ECOC’99, 25th European Conference on Optical Communication, Paris,
1999.
[41] K. S. Jepsen, U. Gliese, B. R. Hemenway, S. Yuan, K. S. Chen, J. E. Hurley, L. Guiziou,
J. W. McCamy, N. Boos, D. J. Tebben, B. Dingel, M. J. Li, S. Gray, G. E. Kohnke,
L. Jiang, V. Srikant, A. F. Evans, and J. M. Jouanno, Network demonstration of 32
lambda *10 Gb/s across 6 nodes of 640*640 WSXCs with 750 km Raman-amplified
fiber. In Proceedings of the Optical Fiber Communication Conference. Technical Digest
postconference edition. Trends in Optics and Photonics Vol. 37, Washington, DC, 2000.
[42] A. K. Srivastava, S. Radic, C. Wolf, J. C. Centannil, J. W. Sulhoff, K. Kantor, and Y.
Sun, Ultra-dense terabit capacity WDM transmission in L-band. In Proceedings of the
Optical Fiber Communication Conference. Technical Digest postconference edition.
Trends in Optics and Photonics Vol. 37, Washington, DC, 2000.
[43] J.-P. Blondel, F. Boubal, E. Brandon, L. Buet, L. Labrunie, P. Le Roux, and D. Toullier,
Network application and system demonstration of WDM systems with very large spans
410 A. Evans et al.

(error-free 32*10 Gbit/s 750 km transmission over 3 amplified spans of 250 km). In
Proceedings of the Optical Fiber Communication Conference. Technical Digest post-
conference edition. Trends in Optics and Photonics Vol. 37, Washington, DC, 2000.
[44] T. Tanaka and et al., 2.1-Tbit/s WDM transmission over 7,221 km with 80-km repeater
spacing. In Proceedings of the European Conference on Optical Communications, 2000.
[45] J. Bromage, J.-C. Bouteiller, H. J. Thiele, K. Brar, J. H. Park, C. Headley, L. E. Nelson,
Y. Qian, J. DeMarco, S. Stulz, L. Leng, B. Zhu, and B. J. Eggleton, S-band all-Raman
amplifiers for 40 * 10 Gb/s transmission over 6 * 100 km of non-zero dispersion fiber.
In Proceedings of the OFC 2001, Optical Fiber Communication Conference and Exhi-
bition. Technical Digest, Washington, DC, 2001.
[46] A. B. Puc, M. W. Chbat, J. D. Henrie, N. A. Weaver, H. Kim, A. Kaminski, A. Rah-
man, and H. Fevrier, Long-haul WDM NRZ transmission at 10.7 Gb/s in S-band using
cascade of lumped Raman amplifiers. In Proceedings of the OFC 2001, Optical Fiber
Communication Conference and Exhibition. Technical Digest, Washington, DC, 2001.
[47] S. N. Knudsen, B. Zhu, L. E. Nelson, M. O. Pedersen, D. W. Peckham, and S. Stulz,
420 Gbit/s (42*10 Gbit/s) WDM transmission over 4000 km of UltraWave fibre with
100 km dispersion-managed spans and distributed Raman amplification, Electron. Lett.,
37:15, 965–967, 2001.
[48] N. Shimojoh et al., 2.4-Tbit/s WDM transmission over 7400 km using all Raman am-
plifier repeaters with 74-nm continuous single band. In Proceedings of the European
Conference on Optical Communications, 2001.
[49] D. F. Grosz et al., Demonstration of all-Raman ultra-wide-band transmission of 1.28
Tb/s (128 × 10Gb/s) over 4000 km of NZ-DSF with large BER margins. In Proceedings
of the European Conference on Optical Communications, 2001.
[50] I. Thomkos et al., 80 × 10.7 Gb/s ultra-long-haul (+4200 km) DWDM network with
reconfigurable “broadcast & select” OADMs. In Proceedings of the OFC 2002, Optical
Fiber Communication Conference and Exhibition. Technical Digest, Washington, DC,
2002.
[51] M. Mehendale, M. Vasilyev, A. Kobyakov, M. Williams, and S. Tsuda, All-Raman trans-
mission of 80 * 10 Gbit/s WDM signals with 50 GHz spacing over 4160 km of dispersion-
managed fibre, Electron. Lett., 38:13, 648–649, 2002.
[52] T. N. Nielsen, A. J. Stentz, P. B. Hansen, Z. J. Chen, D. S. Vengsarkar, T. A. Strasser, K.
Rottwitt, J. H. Park, S. Stulz, S. Cabot, K. S. Feder, P. S. Westbrook, and K. G. Kosin-
ski, 1.6 Tb/s (40*40 Gb/s) transmission over 4*100 km nonzero-dispersion fiber using
hybrid Raman/Er-doped inline amplifiers. In Proceedings of ECOC’99. 25th European
Conference on Optical Communication, Paris, 1999.
[53] R. Ohhira, Y. Yano, A. Noda, Y. Suzuki, C. Kurioka, M. Tachigori, S. Moribayashi, K.
Fukuchi, T. Ono, and T. Suzaki, 40 Gbit/s × 8-ch NRZ WDM transmission experiment
over 80 km*5-span using distributed Raman amplification in RDF. In Proceedings of
ECOC’99. 25th European Conference on Optical Communication, Paris, 1999.
[54] S. Bigo, E. Lach, Y. Frignac, D. Hamoir, P. Sillard, W. Idler, S. Gauchard, A. Bertaina,
S. Borne, L. Lorcy, N. Torabi, B. Franz, P. Nouchi, P. Guenot, L. Fleury, G. Wien, G. Le
Ber, R. Fritschi, B. Junginger, M. Kaiser, D. Bayart, G. Veith, J.-P. Hamaide, and J.-L.
Beylat, 1.28 Tbit/s WDM transmission of 32 ETDM channels at 40 Gbit/s over 3*100 km
distance. In Proceedings of the 26th European Conference on Optical Communication,
Berlin, 2000.
[55] Y. Zhu et al., 16-channel 40 Gb/s carrier-suppressed RZ ETDM/DWDM transmission
over 720 km NDSF without polarization channel interleaving. In Proceedings of the
Optical Fiber Communications, Anaheim, CA, 2001.
 12. Applications and Fiber Issues 411

[56] B. Zhu, L. Leng, L. E. Nelson, Y. Qian, S. Stulz, C. Doerr, L. Stulz, S. Chandrasekar, S.

Radic, D. Vengsarkar, Z. Chen, J. Park, K. Feder, H. Thiele, J. Bromage, L. Gruner-
Nielsen, and S. Knudsen, 3.08 Tb/s (77*42.7 Gb/s) transmission over 1200 km of
non-zero dispersion-shifted fiber with 100-km spans using C and L-band distributed
Raman amplification. In Proceedings of the OFC 2001. Optical Fiber Communication
Conference and Exhibition. Technical Digest, Washington, DC, 2001.
[57] D. Chen, S. Wheeler, D. Nguyen, A. Farbert, A. Schopflin, A. Richter, C.-J. Weiske,
K. Kotten, P. M. Krummrich, A. Schex, and A. S. C. Glingener, 3.2 Tb/s field trial
(80*40 Gb/s) over 3*82 km SSMF using FEC, Raman and tunable dispersion compen-
sation. In Proceedings of the OFC 2001. Optical Fiber Communication Conference and
Exhibition. Technical Digest, Washington, DC, 2001.
[58] I. Morita, K. Tanaka, and N. Edagawa,Benefit of Raman amplification in ultra-long-
distance 40 Gbit/s-based WDM transmission. In Proceedings of the OFC 2001. Optical
Fiber Communication Conference and Exhibition. Technical Digest, Washington, DC,
2001.
[59] L. du Mouza, G. Le Meur, H. Mardoyan, E. Seve, S. Cussat-Blanc, D. Hamoir, C.
Martinelli, D. Bayart, F. Raineri, L. Pierre, B. Dany, O. Leclerc, J. P. Hamaide, L.
A. de Montmorillon, F. Beaumont, P. Sillard, P. Nouchi, A. Hugbart, R. Uhel, and G.
Granpierre, 1.28 Tbit/s (32*40 Gbit/s) WDM transmission over 2400 km of TeraLight,
Reverse TeraLight(C) fibres using distributed all-Raman amplification, Electron. Lett.,
37:21, 1300–1302, 2001.
[60] S. Bigo et al., Transmission of 125 WDM channels at 42.7 Gbit/s (5 Tbit/s capacity)
over 12 × 100 km of Teralight Ultra fibre. In Proceedings of the European Conference
on Optical Communications, 2001.
[61] B. Zhu et al., 1.6 Tb/s (40 × 42.7 Gb/s) transmission over 2000 km of fiber with 100-
km dispersion-managed spans. In Proceedings of the European Conference on Optical
Communications, 2001.
[62] H. Bissessur et al., 3.2 Tb/s (80 × 40 Gb/s) C-band transmission over 3 × 100 km with
0.8 bit/s/Hz efficiency. In Proceedings of the European Conference on Optical Commu-
nications, 2001.
[63] B. Zhu, L. E. Nelson, L. Leng, S. Stulz, S. Knudsen, and D. Peckham, 1.6 Tbit/s (40 ×
42.7 Gbit/s) WDM transmission over 2400 km of fibre with 100 km dispersion-managed
spans, Electron. Lett., 38:13, 647–648, 2002.
[64] A. H. Gnauck et al., 2.5 Tb/s (64 × 42.7 Gb/s) transmission over 40 × 100 km NZDSF
using RZ-DPSK format and all-Raman-amplified spans. In Proceedings of the Optical
Fiber Communications, 2002.
[65] Y. Frignac et al., Transmission of 256 wavelength-division and polarization-division-
multiplexed channels at 42.7 Gb/s (10.2 Tb/s capacity) over 3 × 100 km of Teralight
fiber. In Proceedings of the Optical Fiber Communications, 2002.
[66] H. Sugahara, K. Fukuchi,A. Tanaka, Y. Inada, and T. Ono, 6,050 km transmission of 32 ×
42.7 Gb/s DWDM signals using Raman-amplified quadruple-hybrid span configuration.
In Proceedings of the Optical Fiber Communications, Anaheim, CA, 2002.
[67] F. Liu et al., 1.6 Tb/s (40 × 42.7 Gbit/s) transmission over 3600 km UltraWave fiber with
all-Raman amplified 100 km terrestrial spans using ETDM transmitter and receiver. In
Proceedings of the Optical Fiber Communications, 2002.
[68] B. Zhu et al., 3.2 Tb/s (80 × 42.7 Gb/s) transmission over 20 × 100 km of non-zero
dispersion fiber with simultaneous C+ L-band dispersion compensation. In Proceedings
of the Optical Fiber Communications, 2002.
[69] T. N. Nielsen et al., 1.6 Tb/s (40 × 40Gb/s) transmission over 4 × 100 km nonzero-
dispersion fiber using hybrid Raman/erbium-doped inline amplifiers. In Proceedings of
the European Conference on Optical Communications, 1999.
412 A. Evans et al.

[70] M. Vasilyev, B. Szalabofka, S. Tsuda, J. M. Grochocinski, and A. F. Evans, Reduction of

Raman MPI and noise figure in dispersion-managed fibre, Electron. Lett., 38:6, 271–272,
2002.
[71] R. Hainberger, J. Kumasako, K. Nakamura, T. Terahara, H. Osaka, and T. Hoshida, Com-
parison of span configurations of Raman-amplified dispersion-managed fibers, Photon.
Technol. Lett., 14:471, 2002.
[72] K. Shimizu, K. Kinjo, N. Suzuki, K. Ishida, S. Kajiya, K. Motoshima, and Y. Kobayashi,
Fiber-effective-area managed fiber lines with distributed Raman amplification in 1.28-
Tb/s (32 × 40 Gb/s), 202-km unrepeatered transmission. In Proceedings of the Optical
Fiber Communications, 2001.
[73] T. Miyakawa, I. Morita, K. Tanaka, H. Sakata, and N. Edagawa, 2.56 Tbit/s (40 Gbit/s*64
WDM) unrepeatered 230 km transmission with 0.8 bit/s/Hz spectral efficiency using
low-noise fiber Raman amplifier and 170 mu m/sup 2/-Aeff fiber. In Proceedings of the
OFC 2001. Optical Fiber Communication Conference and Exhibition. Technical Digest,
Washington, DC, 2001.
[74] Y. Zhu, I. Hardcastle, W. S. Lee, C. R. S. Fludger, A. Hadjifotiou, C. Li, D. Qiao, H. Sun,
K. T. Wu, and J. McNicol, Experimental comparison of Raman-amplified dispersion-
managed fibre types using 16 × 40 Gbit/s transmission over 500 km, Electron. Lett.,
38:895, 2002.
[75] Y. Hadjar and N. J. Traynor, Quantitative analysis of second order distributed Raman
amplification. In Proceedings of the Optical Fiber Communications, 2002.
[76] J.-C. Bouteiller, K. Brar, S. Radic, J. Bromage, Z. Wang, and C. Headley, Dual-order
Raman pump providing improved noise figure and large gain bandwidth. In Proceedings
of the Optical Fiber Communications, 2002.
[76a] J.-C Bouteiller, K. Brar, and C. Headley, “Quasi-constant signal power transmission,”
presented at European Conference on Optical Communication, 2002.
[77] I. Tomkos, M. Vasilyev, J.-K. Rhee, A. Kobyakov, M. Ajgaonkar, and M. Sharma, Dis-
persion map design for 10 Gb/s ultra-long-haul DWDM transparent optical networks.
In Proceedings of the Seventh Optoelectronics and Communications Conference, Yoko-
hama, 2002.
[78] I. Tomkos, M. Vasilyev, J.-K. Rhee, M. Mehendale, B. Hallock, B. Szalabofka, M.
Williams, S. Tsuda, and M. Sharma, Ultra-long-haul DWDM network with 320 × 320
wavelength-port “broadcast & select” OXCs. In Proceedings of the European Confer-
ence on Optical Communication, Copenhagen, 2002.
[79] S. Kumar, J. C. Mauro, S. Raghavan, and D. Q. Chowdhury, Intrachannel nonlinear
penalties in dispersion-managed transmission systems, IEEE J. of Select. Topics Quan-
tum Electron. 8:3, 626–631, 2002.
[80] M. Vasilyev, Raman-assisted transmission: Toward ideal distributed amplification. In
Proceedings of the Optical Fiber Communications, Atlanta, 2003.
[81] R.-J. Essiambre, P. Winzer, J. Bromage, and C. H. Kim, Design of bidirectionally pumped
fiber amplifiers generating double Rayleigh backscattering, IEEE Photon. Technol. Lett.,
14:7, 914–916, 2002.
[82] A. Kobyakov, M. Vasilyev, S. Tsuda, G. Giudice, and S. Ten, Raman noise figure in
dispersion-managed fibers. In Proceedings of the ECOC, 2002; Analytical model for
Raman noise figure in dispersion-managed fibers, IEEE Photon Technol. Lett., 15:1,
30–32, 2003.
[83] S. N. Knudsen, Design and manufacture of dispersion compensating fibers and their
performance in systems. In Proceedings of the Optical Fiber Communications, 2002.
Chapter 13
Hybrid EDFA/Raman Amplifiers

Hiroji Masuda

13.1. Introduction
This chapter describes the technologies needed for cascading an erbium-doped fiber
amplifier (EDFA) and a fiber Raman amplifier (FRA or RA) to create a hybrid am-
plifier (HA), the EDFA/Raman HA. Two kinds of HA are defined in this chapter:
the narrowband HA (NB-HA) and the seamless and wideband HA (SWB-HA). The
NB-HA employs distributed Raman amplification in the transmission fiber together
with an EDFA and provides low noise transmission in the C- or L-band. The noise fig-
ure of the transmission line is lower than it would be if only an EDFA were used. The
SWB-HA, on the other hand, employs distributed or discrete Raman amplification
together with an EDFA, and provides a low-noise and wideband transmission line or
a low-noise and wideband discrete amplifier for the C- and L-bands. The typical gain
bandwidth (λ) of the NB-HA is ∼30 to 40 nm, whereas that of the SWB-HA is ∼70
to 80 nm.
The basic configurations of these HAs are introduced in the next section (Sec-
tion 13.2) and are compared with those of other amplifiers. Section 13.3 shows the
performance limitations of EDFAs with regard to gain bandwidth and noise figure
(NF). Properties of the NB-HA are described in Section 13.4. An analysis of noise
properties and an experimental example are also shown. Section 13.5 describes the
properties of the SWB-HA. An analysis of gain bandwidth, a classification of ampli-
fier type, four amplifier configurations, and gain and noise characteristics are shown
based on both calculations and experiments. Section 13.6 introduces two other types
of hybrid amplifiers, the TDFA/Raman amplifier and the tellurite/silica Raman am-
plifier. Finally, conclusions are drawn in Section 13.7.

13.2. Basic Amplifier Configurations

Figure 13.1 shows some basic configurations of a transmission line with an inline
amplifier. An EDFA is used as the repeater between two installed transmission fibers,
and amplifies the input signal light (Fig. 13.1(a)) [1]. The signal light usually consists
414 H. Masuda

EDFA
Signal C or L Band Signal

Transmission Transmission
Fiber Fiber
(a)

EDFA
C Band

Signal Signal

Transmission Transmission
Fiber DIV CMB Fiber

(b)

EDFA
L Band

Distributed FRA Distributed FRA

C or L Band C or L Band
EDFA
Signal C or L Band Signal
Coupler

Transmission Transmission
Fiber Pump Fiber
Pump
for C or L Band
(c)

Distributed FRA EDFA Distributed FRA

C + L Band C Band C + L Band

Signal Signal
Coupler

Transmission Transmission
DIV CMB Fiber
Fiber
Pump
CMB
EDFA
L Band
Pump Pump
for C Band for L Band
(d)

Fig. 13.1. Basic configurations of a transmission line with an inline amplifier: (a) an EDFA;
(b) a two-gain band amplifier (EDFA) with C- and L-band EDFAs in parallel; (c) a hybrid
EDFA/distributed Raman amplifier with C- or L-band; and (d) a hybrid EDFA/distributed
Raman amplifier with C- and L-bands in parallel (CMB: combiner, DIV: divider).
 13. Hybrid EDFA/Raman Amplifiers 415

of wavelength-division-multiplexed (WDM) multichannels, and the EDFA offers C-

or L-gain band coverage. The typical gain bands of C- and L-gain band EDFAs are
the wavelength ranges of about 1530 to 1560 nm and 1570 to 1600 nm, respectively.
Figure 13.1(b) shows a two-gain band amplifier (EDFA) with C- and L-gain band
EDFAs in parallel with each other [2, 3]. The combiner and divider connected to
the EDFAs multiplex and demultiplex the WDM signal channels according to their
wavelengths. The two-gain band EDFA has a gain bandwidth that is about twice that
of the C- or L-band EDFA (Fig. 13.1(a)). However, its cost and the number of optical
components are about twice those of the C- or L-band EDFA. The NB-HA that offers
C- or L-band coverage is shown in Fig. 13.1(c) [4–7]. The NB-HA consists of a C- or
L-band distributed RA (DRA), which is a transmission fiber itself, and a C- or L-band
EDFA set after the transmission fiber as a repeater. Finally, Fig. 13.1(d) shows a C-
and L-two-gain band HA [8, 9]. The two-gain band HA consists of a two-wavelength
pumped DRA (C- and L-band) and a two-gain band EDFA. The pump lights for the
C- and L-bands are multiplexed by a combiner and launched into the transmission
fiber via a coupler. The transmission line with a discrete (or lumped) Raman amplifier
(LRA) instead of the EDFA of Fig. 13.1(a), which is not shown for simplicity, can be
also constructed as shown in Chapter 10. Moreover, using just a DRA is also possible
as described in Chapters 12 and 14.
We can evaluate the performance of a cascaded transmission line with multiple
transmission fibers and inline amplifiers by considering a single span of transmission
line [1]. The single transmission line with an EDFA (Figs. 13.1(a) and 13.1(b)), an
LRA, or a DRA has noise and output power characteristics determined by each of the
amplifiers. On the other hand, the noise characteristics of a single transmission line
with an EDFA/Raman HA are determined by both the DRA and EDFA as described
in detail in Section 13.4. The optical SNR (OSNR) of a transmission line with an HA
is higher than would be the case if the EDFA of Fig. 13.1(a) or 13.1(b) were used.
In other words, the HA has a lower effective NF (Feff ) than the EDFA. Moreover,
the output power characteristics of the HA are determined by the EDFA used. EDFAs
offer available output powers of up to ∼30 dBm; the main determining factor is
the pump power [1]. However, the available output power of the Raman amplifier
(DRA and LRA) can range up to ∼20 dBm, and the main determining factor is the
nonlinearity of the long silica fiber used as the gain medium [10].
The basic configurations of a transmission line with an SWB-HA are shown in
Fig. 13.2. Each of the EDFAs and RAs covers a different portion of the C- and
L-bands. The gain-slope of the EDFA is opposite to that of the RA as shown in
Fig. 13.14 of Section 13.5. The EDFA and RA are connected in a series. Figure 13.2(a)
shows the HA with an LRA followed by an EDFA, whereas Figure 13.2(b) shows the
HA with an EDFA followed by an LRA. An HA with pre- and post-EDFAs and an
intermediate LRA is shown in Fig. 13.2(c). Finally, Figure 13.2(d) shows an HA with a
DRA followed by an EDFA. The HAs of Figs. 13.2(a), 13.2(c), and 13.2(d) have high
output power because they use an EDFA at the output side. The HA of Fig. 13.2(b),
however, has moderate output power because it uses an LRA at the output side. The
NF of the HA is mainly determined by the NF of the first-stage amplifier if the gain of
the first-stage amplifier is large enough (i.e., >10 dB). Otherwise (i.e., gain <10 dB),
416 H. Masuda

Raman EDFA
Partial Partial
Signal C + L Band C + L Band Signal

Transmission Transmission
Fiber Fiber
(a)

EDFA Raman
Partial Partial
Signal C + L Band C + L Band Signal

Transmission Transmission
Fiber Fiber

(b)

EDFA Raman EDFA

Partial Partial Partial
Signal C + L Band C + L Band C + L Band Signal

Transmission Transmission
Fiber Fiber
(c)
Distributed FRA Distributed FRA
Partial C + L Band Partial C + L Band
EDFA
Partial
Signal Signal
Coupler C + L Band

Transmission Transmission
Fiber Fiber
Pump
Pump
for Partial C + L Band
(d)

Fig. 13.2. Basic configurations of a transmission line with a seamless wideband hybrid
EDFA/Raman amplifier. Each of the EDFA and Raman amplifiers covers a partial C- and
L-band with different gain slopes. The EDFA and Raman amplifier are serially connected.
The hybrid amplifiers are: (a) a discrete Raman amplifier followed by an EDFA; (b) an EDFA
followed by a discrete Raman amplifier; (c) pre- and post-EDFAs and an intermediate discrete
Raman amplifier; and (d) a distributed Raman amplifier followed by an EDFA.
 13. Hybrid EDFA/Raman Amplifiers 417

the NF is determined by the NFs of both the first- and second-stage amplifiers. The
configurations and performances of the HAs are described in detail in Section 13.5.

13.3. Performance Limitations of EDFAs

This section describes the limitations of the gain bandwidths and NFs of EDFAs.
Figure 13.3 shows the three basic EDFA configurations. An EDFA with a single-
stage amplification configuration and an external gain equalizer (GEQ) is shown in
Fig. 13.3(a). An EDFA with a two-stage (three-stage) amplification configuration and
one (two) internal GEQ is shown in Figs. 13.3(b) and 13.3(c) [11–14]. The relevant
optical components such as isolators and signal–pump couplers have been omitted
for simplicity. The gains of the EDFs and the losses of the GEQs are identified by
G, G1 , G2 , G3 , and L, L1 , L2 , respectively. We can expand the gain bandwidth of
an EDFA with the two- or three-stage amplification configuration as shown below.
The gain (in units of dB) of an EDFA at a signal wavelength of λs (GdB (λs )) is
expressed as
10
GdB (λs ) = ηρL{(σabs (λs ) + σemi (λs ) − σESA (λs ))N2ave − σabs (λs )},
ln(10)
(13.1a)
L
1
N2ave = N2 (z) dz, (13.1b)
L 0
where η is the overlap factor, ρ is the Er concentration, L is EDF length, σabs , σemi ,
σESA are the absorption, emission, and excited-state-absorption (ESA) cross-sections,

Signal G L
GEQ
EDF
(a)

Signal G1 G2
L
GEQ
EDF-1 EDF-2
(b)

Signal G1 L1 G2 L2 G3

GEQ-1 GEQ-2
EDF-1 EDF-2 EDF-3
(c)

Fig. 13.3. Basic EDFA configurations: (a) single stage configuration with an external gain
equalizer (GEQ); (b) two-stage configuration with an intermediate GEQ; (c) three-stage con-
figuration with two intermediate GEQs (G: gain, L: loss).
418 H. Masuda

N2ave =
50 50 1.0
N2ave = 0.9
40 1.0 40 0.8
0.9 0.7
Gain (dB) 30 0.8 30

Gain (dB)
0.7
20 20

10 10

0 0

–10 0.6 0.5 0.4 0.3 –10 0.6 0.5 0.4 0.3

–20 –20
1480 1520 1560 1600 1640 1480 1520 1560 1600 1640
Wavelength (nm) Wavelength (nm)
(a) (b)

Fig. 13.4. Gain spectra at a peak gain of 40 dB for (a) silica EDF (EDSF); (b) flouride EDF
(EDFF) (N2ave : average fractional population of the upper state).

respectively, and N2ave is the path-averaged fractional population of the upper state
(0 < N2ave < 1) [1]. Equation (13.1) indicates that the gain is proportional to the
concentration-length product ρL, and that the gain spectrum depends on N2ave .
Figure 13.4 shows the gain spectra of (a) a silica EDF (erbium-doped silica fiber:
EDSF) and (b) a fluoride EDF (erbium-doped fluoride fiber: EDFF) at N2ave values
ranging from 0.3 to 1 in steps of 0.1. The peak gain of 40 dB can be achieved by setting
ρL at an appropriate value. A variety of N2ave values can be generated experimentally
by altering the pumping condition (wavelength and power). When N2ave is around
0.7, flat gain spectra are obtained from about 1520 to 1570 nm (the C or 1550 nm gain
band). On the other hand, when N2ave is around 0.4, flat gain spectra are obtained
from about 1560 to 1610 nm (the L or 1580 nm gain band) [15]. The L-band EDFA is
often called the “gain-shifted EDFA” [16]. Moreover, when N2ave is from about 0.5
to 0.6, gain spectra with a poor flatness are obtained from about 1530 to 1610 nm.
The gain spectra can be flattened or equalized by multistage amplification with one
or two GEQs. An analysis of the gain spectra of Fig. 13.4 yields the following gain
bandwidths. As an example, a flat gain ideal bandwidth (λideal ) of 52 nm (1559 to
1611 nm) for an EDSF (Fig. 13.4(a)) is generated at N2ave = 0.4 by the two-stage
configuration using an ideal GEQ with a peak loss of 15 dB. The resultant flattened
gain of the EDF is 25 dB. Moreover, λideal of 58 nm (1554 to 1612 nm) for an EDFF
(Fig. 13.4(b)) is also generated at N2ave = 0.45.
The NF spectra of the EDFA in the intense-pumping condition in the 1.48 µm
band can be evaluated by applying the next equation [1]. The NF can, as a function
of λs , F (λs ), be approximated as

2R(λs ) σemi (λ)

F (λs ) = , R(λ) ≡ , (13.2)
R(λs ) − R(λp ) − σESA (λs )/σabs (λs ) σabs (λ)
 13. Hybrid EDFA/Raman Amplifiers 419

8
Silica EDF
7 Fluoride EDF

Noise Figure (dB)

6

3
1500 1520 1540 1560 1580 1600 1620
Wavelength (nm)

Fig. 13.5. Noise figure spectra of silica and flouride EDFs (EDSF and EDFF) pumped at 1.48
µm.

where λp is the pump wavelength, and R(λ), λ = λs or λp , is the ratio of cross-

sections. On the other hand, F (λs ) is about 2 (or 3 dB) in the intense-pumping condi-
tion in the 0.98 µm band. Figure 13.5 shows the NF spectra for an EDSF and EDFF
as calculated by Eq. (13.2). F (λs ) under 5 dB from 1530 to 1615 nm are obtained
with the EDSF, whereas the EDFF yields F (λs ) under 5 dB from 1530 to 1620 nm.
NFs for the multistage configurations of Fig. 13.3 are expressed as follows. When
we consider only the signal–spontaneous beat noise component as contributing to the
NF for simplicity, the NF of the single-stage EDFA, F1stg , is expressed using equiva-
lent spontaneous emission factor neq = nsp (G − 1)/G, where nsp is the spontaneous
emission factor, as
F1stg = 2neq . (13.3a)
The NF of the two-stage EDFA, F2stg , is also expressed as [1]
2neq2
F2stg = 2neq1 + . (13.3b)
G1 L
Therefore, the difference between the two NFs in dB units, FdB , is
   
2neq2 ∼ 1
FdB = 10 log 2neq1 + − 10 log(2neq ) = 10 log 1 + . (13.4)
G1 L G1 L
This assumes the approximation of neq ∼
= neq1 ∼= neq2 . As an example, FdB is 1 and
0.5 dB, when the gain–loss product G1 L is set at 6.1 dB and 9.2 dB, respectively. As
another example, FdB is 0.41 dB, when G1 = 25 dB, L = 15 dB, and G2 = 15 dB.
The NF of the three-stage EDFA can also be analyzed in the same way. As an example,
FdB is 0.34 dB, when G1 = 28 dB, L1 = 15 dB, G2 = 17 dB, L2 = 15 dB, and
G2 = 10 dB.
420 H. Masuda

The above analyses were verified by several experiments [11–14]. The results of
two experimental studies [12, 13] are introduced below as examples. The first study
reports the results gained from 1.48 µm pumped two-stage EDFAs [12]. A two-stage
EDSFA (silica EDFA) generated a 3 dB gain-reduction bandwidth (λ3dB ) of 52 nm
(1556 to 1608 nm) with a peak gain of 19.7 dB and a GEQ peak loss of 17.5 dB. A
two-stage EDFFA (fluoride EDFA) generated a λ3dB of 50 nm (1554 to 1604 nm)
with a peak gain of 22.4 dB and a GEQ peak loss of 17.0 dB. The GEQs were simple
Mach–Zehnder type filters. Therefore, there were some discrepancies, in terms of the
loss spectra, between the experimental GEQs and the ideal GEQs. The NFs in λ3dB
ranged from 4.4 to 5.7 dB for the EDSFA and EDFFA. The signal output power was
50 mW (17 dBm) for the EDSFA and 40 mW (16 dBm) for the EDFFA when the
pump power launched into the output side EDF was 79 mW. However, the second
study reports the results gained from 1.48 µm pumped three-stage EDFAs [13]. A
three-stage EDSFA generated a λ3dB of 57 nm (1550 to 1607 nm) with a peak gain
of 20.9 dB and a GEQ total peak loss of 33 dB. Moreover, a three-stage EDFFA
generated a λ3dB of 62 nm (1540.5 to 1602.5 nm) with a peak gain of 22.5 dB and a
GEQ total peak loss of 34 dB. Low NFs and high output powers were also confirmed
for the EDFAs that were comparable to those of two-stage EDFAs.
Several methods have been employed to enlarge the gain bandwidths of fiber
amplifiers. Figure 13.6 compares the gain bands of several types of wideband fiber
amplifiers reported to date. The types are (1) the multistage EDFA including one
or more gain equalizers [11–14], (2) the two-gain band EDFA [2, 3], (3) the mul-
tiwavelength pumped Raman amplifier [17–19], and (4) the hybrid EDFA/Raman
amplifier [20]. Three types of EDFAs were used in the wideband fiber amplifiers as
shown by EDXFA where X = S, F, and T (S: silica, F: fluoride, T: tellurite) [21].
The gain bands of the SWB-HA (seamless and wideband HA) are shown in detail in
Section 13.5. Typical gain bands are as follows.

1. A two-stage EDSFA with a λ3dB of about 47 nm around 1550 nm was reported

[11]. The amplifier used two EDSFs and an internal GEQ set between the EDSFs.
Another two-stage EDSFA generated a λ3dB of 52 nm (1556 to 1608 nm) as
mentioned above [12]. Moreover, a three-stage EDFFA, which used three EDFFs
and two internal GEQs set among the EDFFs, generated a λ3dB of 62 nm (1540.5
to 1602.5 nm) [13]. A three-stage EDTFA generated a λ3dB of 76 nm (about
1530 to 1606 nm) [14]. Note that the amplifier used a 0.98 mm pumped EDTF
in the first stage and achieved a low NF of 5 dB.
2. A two-gain band EDSFA was reported to generate an aggregate bandwidth of
84.3 nm (∼1527 to ∼1568 nm and ∼1569 to ∼1612.5 nm) [3]. The width of the
dead band between the two gain bands was only 1 nm or so. Although some qua-
siseamless two-gain band amplifiers yielded narrow dead bands between the two
gain bands, some interference noises around the dead bands must be treated [22].
Note that most of the two-gain band EDFAs used in transmission experiments
have dead bandwidths of about 5 to 10 nm; see the examples of [2, 8, 9].
3. Multiwavelength pumped RAs were reported as seamless and wideband fiber
amplifiers. A DRA with a λ3dB of about 100 nm (1520 to 1620 nm) with an
 13. Hybrid EDFA/Raman Amplifiers 421

EDSFA
~47nm [11]
2-Stage Amp.
EDSFA
52 nm [12]
2-Stage Amp.
Multi-Stage
EDFA EDFFA
62 nm [13]
3-Stage Amp.
EDTFA
76 nm [14]
3- Stage Amp.

Two Gain-
EDSFA
Band
84.3 nm [3]
EDFA

Distributed
Raman ~100 nm [17]
Amplifier Discrete
~85 nm [18]

EDFFA + Discrete Raman

Hybrid 80 nm [23]
EDFA/Raman
EDFFA + Distributed Raman
Amplifier
82.8 nm [24]

1500 1550 1600 1650

Wavelength (nm)

Fig. 13.6. Gain bands of wideband fiber amplifiers. ED(S, F, T)FA: erbium-doped (silica,
flouride, tellurite) fiber amplifier.

on-off gain of about 13 dB was reported [17]. Moreover, an LRA was reported
to generate a gain bandwidth of about 85 nm (1510 to 1595 nm) with a net gain
(amplifier in-out gain) of over 13 dB (on-off Raman gain of over 20 dB) [18].
4. An SWB-HA with a LRA was reported to generate a λ3dB of 80 nm (1530 to
1610 nm) [23]. Moreover, an SWB-HA with a DRA generated a λ3dB of 82.8
nm (1528.8 to 1611.6 nm) [24].
The gains of the wideband fiber amplifiers listed above are, except for the Raman
amplifiers, above 20 dB or so. Therefore, the SWB-HAs have the largest seamless
bandwidths, about 80 nm, with gain above 20 dB. The bandwidths are comparable to
those of two-gain band EDFAs.
422 H. Masuda

13.4. Narrowband Hybrid EDFA/Raman Amplifiers

13.4.1. Analysis of Noise Properties

We consider the NB-HAs (narrowband hybrid EDFA/Raman amplifiers) in this sec-

tion. First we briefly describe the noise characteristics of an LRA and a DRA.
The on-off gain (or overall distributed gain) of a DRA (Gd ) is expressed as [25]
1 − exp(−αp l)
Gd = exp(gPpin Leff ), Leff ≡ , (13.5)
αp
where g is the Raman gain coefficient, αp is the loss coefficient at the pump wave-
length, Ppin is the pump power, and Leff is the effective fiber length. Note that the
Raman gain coefficient g contains the polarization scrambling effect. The coefficient
g is related to the parallel-polarization coefficient gp as given by g ∼ = gp /2. From
the definition of Eq. (13.5), Gd = 1 (0 dB) when Ppin = 0. The Raman gain of an
LRA (G1 ) is related to Gd via the expression Gl = Gd L, where L is the fiber loss
L = exp(−αs l), and αs is the loss coefficient at the signal wavelength. The NF of
the fictitious amplifier set after the DRA with the on-off gain of Gd is denoted as the
effective NF of the DRA (Feff ). When the NF of the LRA is denoted as F, Feff is
related to F as shown by Feff = F L. Moreover, Feff can, considering the signal to
spontaneous–emission (or spontaneous–scattering) beat noise (s–sp beat noise) and
the signal shot noise [26], be expressed as
PASE 1
Feff = + , (13.6)
hννGd Gd
where h is Plank’s constant, ν is the noise bandwidth, and PASE is the amplified
spontaneous emission (ASE) power at the fiber output. Moreover, the propagation
equation of PASE can be expressed as [27]
dPASE
= gPp {PASE + 2hνn (Nphon + 1)ν} − α(νn )PASE
dz
1
Nphon = , (13.7)
exp{h(νp − νs )/kT } − 1
where Nphon is the phonon population number at finite temperature T in Kelvin, νs
and νp are the optical frequencies at the signal and pump wavelengths, respectively,
and k is Boltzmann’s constant.
Figure 13.7 shows the NF of a backward pumped LRA as a function of Gd .
The NFs were calculated by Eq. (13.7). Signal and pump wavelengths were set to
1580 and 1480 nm, respectively. The fiber loss coefficients of the LRA were set to
0.5 dB/km for both signal and pump lights for simplicity. The fiber length of the LRA
was set to 5, 7.5, or 10 km, and corresponding fiber losses were 2.5, 3.75, or 5 dB.
The NF at Gd = 10 (20) dB for the 5, 7.5, or 10 km fiber was 5.8 (5.2), 5.1 (4.7), or
4.4 (4.2) dB, respectively. Therefore, the NFs of the LRA range from 4 to 6 dB in
most operating conditions.
 13. Hybrid EDFA/Raman Amplifiers 423

Noise Figure (dB)

5

2
0 5 10 15 20 25 30
On-off Gain (dB)

Fig. 13.7. Noise figures as functions of the on-off gain for a discrete Raman amplifier. Fiber
loss coefficient is 0.5 dB/km; fiber length is 5 km (solid line), 7.5 km (broken line), or 10 km
(dotted line).

Feff of a DRA are shown as a function of Gd in Fig. 13.8 as calculated by

Eqs. (13.5) through (13.7). Signal and pump wavelengths were set to 1580 and 1480
nm, respectively. The fiber length of the DRA was set to 100 km. Loss coefficients
at the signal and pump wavelengths were set to 0.2, 0.3, or 0.4 dB/km. Feff of the
0.2 dB/km fiber at Gd = 10 (20) dB was −1 (−3 dB). The differences in Feff for the
three loss coefficients are less than 0.2 dB. Although we can lower Feff by increas-
ing Gd to more than 20 dB as shown in Fig. 13.8, there is a limitation due to the
double-Rayleigh scattering noise as shown in Chapter 3.
Next, noise characteristics of NB-HAs are analyzed. We consider an NB-HA with
the configuration of Fig. 13.1(c). The NB-HA consists of a DRA with a transmission
fiber and an EDFA after the DRA. The NF and effective NF of the NB-HA are denoted
as F and Feff , respectively. The on-off gain of the DRA and the loss of the transmission
fiber are denoted as Gd and L, respectively. Feff is, from its definition, related to F
as shown by Feff = F L. Feff can be expressed as
PASE d PASEl 1 PASE d 2nspl (Gl − 1)
Feff = + + = + + L,
hννGd Gd hννGl Gd Gl hννGd Gd G l
(13.8)
where Gd Gl L = 1, PASE d is the ASE power from the DRA, and PASE l is the ASE
power from the EDFA with no input ASE. Moreover, Gl and nspl are the gain and the
spontaneous emission factor of the EDFA (lumped amplifier), respectively. The first,
second, and third terms in the right-hand side of Eq. (13.8) originate from the s–sp
beat noise generated in the DRA, the s–sp beat noise generated in the EDFA, and the
signal shot noise, respectively. It is obvious that the EDFA can be replaced by a LRA
in the above analysis if needed.
424 H. Masuda

Effective Noise Figure (dB)

–1

–2

–3

–4

–5
0 5 10 15 20 25 30
On-off Gain (dB)

Fig. 13.8. Effective noise figures as functions of the on-off gain for a distributed Raman
amplifier with a transmission fiber of 100 km. Fiber loss coefficient is 0.2 dB/km (solid line),
0.3 dB/km (broken line), or 0.4 dB/km (dotted line).

In calculations, the fiber length and the loss coefficient of the transmission fiber
were set to 100 km and 0.2 dB/km, respectively, and 2nspl was set to 11, 7, or
3 dB. The signal and pump wavelengths were set to 1580 and 1480 nm, respectively.
Figure 13.9 shows Feff of the NB-HA as a function of Gd . Feff of the DRA, calculated
by Eq. (13.6), is also shown for comparison. Feff of the NB-HA decrease with Gd .
For example, Feff decreased in order 11, 7, and 3 dB for Gd values of 0 dB to 2.3, 0.6,
and −0.5 dB at Gd = 10 dB, respectively. The third term (L = 0.01) in the right-hand
side of Eq. (13.8) is negligible compared to the sum of the other terms. The second
term decreases with Gd . On the other hand, the first term increases with Gd in the
range of Gd = 0 to 8 dB, and stays around 0.6 or so with Gd in the range of Gd = 8
to 20 dB. The second term equals the first term at Gd ∼ 5, 8, or 12 dB for nspl values
of 3, 7, or 11 dB, respectively.
For a nonlinear impairment-limited transmission system, we have to consider
an increase in the nonlinear impairment with an increase in path-averaged signal
power (Pave ) [28]. In the calculation above, Pave is 0.15, 0.6, 1.5, and 1.9 dB for
Gd values of 5, 10, 15, and 20 dB, respectively. We can obtain higher transmission
performance with higher Gd value up to Gd = 20 dB, because the increase in Feff is
larger than the increase in Pave .

13.4.2. An Experimental Example

Several experiments have been reported on the NB-HA [4–7], including NB-HAs in
the L-band with DSFs. The improvements in the OSNR and the transmission distance
with a fixed inline repeater spacing [5] and a repeater spacing upgrade [4] with the
 13. Hybrid EDFA/Raman Amplifiers 425

12

10

Effective Noise Figure (dB)

8

–2

–4
0 5 10 15 20
On-off Gain (dB)

Fig. 13.9. Effective noise figures as functions of the on-off gain for a hybrid EDFA/distributed
Raman amplifier with a transmission fiber of 100 km. The effective noise figure for the dis-
tributed Raman amplifier is also shown by a thick line. Fiber loss coefficient is 0.2 dB/km.
Twice the spontaneous emission factor fo the EDFA is 11 dB (solid line), 7 dB (broken line),
or 3 dB (dotted line).

Distributed Raman Amp. EDSFA

EDSF EDSF
Signal Signal
Coupler Coupler 18 m Coupler 70 m
Gain
Equalizer
DSF
Isolator Isolator Isolator
120 km Pump Pump Pump
Coupler

PBC PBC LD Pump LD Pump

1480 nm 1480 nm
LD Pump LD Pump
1467 nm 1503 nm

Fig. 13.10. Configuration of a narrowband hybrid EDFA/Raman amplifier (PBC: polarization

beam combiner). (After Ref. [5]. © 1999 IEE)

NB-HAs were confirmed. Moreover, reports on NB-HAs in the C-band with DSFs
have confirmed the significant suppression of nonlinear impairments and the improve-
ment of the transmission distance [6, 7, 30].
Figure 13.10 shows the configuration of the L-band NB-HA of [5]. The HA con-
sists of a DRA and an EDSFA. The DRA has a 120 km DSF as the gain medium and
a pump light source with two LD pumps. Each of the LD pumps has two LDs and a
polarization combiner. The wavelengths of the LD pumps are 1467 and 1503 nm. On
426 H. Masuda

30

Gain, Loss (dB)

20

10

0
1560 1570 1580 1590 1600 1610
Wavelength (nm)

Fig. 13.11. Gain spectra of the hybrid amplifier and a loss spectrum of the DSF (EDFA Gain,
Raman Gain, X (EDFA + Raman) Gain, DSF Loss). (After Ref. [5]. © 1999 IEE)

30
Optical SNR (dB)

25 Hybrid Amp. System

20
EDFA System

15
1560 1570 1580 1590 1600 1610
Wavelength (nm)

Fig. 13.12. Optical signal-to-noise ratio spectra of the hybrid amplifier system with distributed
amplification and the EDFA system without distributed amplification. (After Ref. [5]. © 1999
IEE)

the other hand, the EDSFA has two EDSFs and an intermediate gain equalizer. Each
EDSF is pumped by a 1480 nm LD. The lengths of the first and second EDSFs are
18 and 70 m, respectively.
The gain characteristics of the NB-HA are shown in Fig. 13.11. The figure shows
the gain spectra of the NB-HA and the loss spectrum of the DSF. The gain spectra are
of the EDSFA, the DRA, and the sum of them. The NB-HA gains well compensate
the DSF losses over a wide wavelength range. A 0.3 dB gain reduction bandwidth of
30 nm (1572 to 1602 nm) was obtained. Figure 13.12 shows the OSNR spectra
of the NB-HA system with distributed amplification and the EDFA system without
distributed amplification (DSF + EDFA). The resolution bandwidth of the noise was
0.5 nm and the signal power launched into the DSF was 0 dBm. Gd of the NB-
HA was 12.3 dB in the gain bandwidth. The OSNRs of the NB-HA system and
the EDFA system were 26.4 ± 0.2 dB and 21.0 ± 0.2 dB, respectively. Therefore,
 13. Hybrid EDFA/Raman Amplifiers 427

the improvement in OSNRs for the NB-HA system compared to those of the EDFA
system is 5.4 dB; this is due to the Gd value (12.3 dB). The NF of the EDFA was
about 5 dB. The improvement in OSNRs calculated by Eqs. (13.5) to (13.8) is 5.8 dB,
which is close to the measured value of 5.4 dB.
The performance of an NB-HA in a WDM transmission system was tested in an
experiment [5]. The amplifier was used as an inline amplifier in a recirculating loop
transmission experiment using 2.5 Gb/s × 7 channel WDM signals and the 120 km
DSF as the transmission fiber. The average transmission distance of the WDM signals
was 570 km for the EDFA system (without distributed amplification) and 2088 km for
the NB-HA system (with distributed amplification). Therefore, the improvement in
transmission distance was 5.6 dB, which well coincides with the OSNR improvement
of 5.4 dB.

13.5. Seamless and Wideband Hybrid EDFA/Raman Amplifiers

13.5.1. Bandwidth Analysis

We can fabricate a wideband hybrid EDFA/Raman amplifier by simply combining,

in a serial cascade, a wide and flat gain band EDFA such as multistage EDFAs (Sec-
tion 13.3) with a multiwavelength pumped flat gain band RA (Chapter 14). However,
the bandwidth of the HA is limited by that of the EDFA or the RA. Moreover, each of
the EDFA and the RA needs many optical components so cost is high. This section
describes SWB-HAs (seamless and wideband hybrid EDFA/Raman amplifiers) that
have a simple structure with few optical components and so are cost effective. The
EDFA and the RA have opposite gain spectral slopes over a wide wavelength region;
the gain bandwidth of the SWB-HA is as large as about 80 nm (1530 to 1610 nm). The
80 nm gain band seamlessly covers the two EDFA gain bands (the C- and L-bands).
Figure 13.13 shows typical gain spectra of the SWB-HA. Figure 13.13(a) shows
the gain of an EDFFA as a function of the average fractional population N2ave . Fig-
ure 13.13(b) shows the on-off Raman gain of an RA with a dispersion-compensation
fiber (DCF) as a function of pump wavelength λp . Similar spectra are observed with
an EDSFA and an RA with a DSF or a standard SMF. The gain spectrum of the SWB-
HA can be synthesized as an appropriate combination of an EDFA gain spectrum
and an RA gain spectrum. We can achieve a maximized gain bandwidth with the
combination of the EDFA gain spectrum at N2ave ∼ 0.7 and the RA gain spectrum at
λp ∼ 1510 nm.
Figure 13.14 shows examples of SWB-HA gain spectra. Figure 13.14(a) shows
the gain spectra of an SWB-HA that consists of an RA pumped at a single wavelength
of λp = 1510 nm and an EDFFA at N2ave = 0.7. Moreover, Fig. 13.14(b) shows the
gain spectra of an SWB-HA that consists of an RA pumped at two wavelengths
of λp = 1510 and 1480 nm, and an EDFFA at N2ave = 0.7. For single-wavelength
Raman pumping, the EDFFA and Raman peak gains were both set at 20 dB. For
two-wavelength Raman pumping, however, the EDFFA peak gain was set to 20 dB,
and the Raman peak gains for λp = 1510 and 1480 nm were set to 22 and 11 dB,
428 H. Masuda

N2AVE =
1
0.9
0.8
EDFA Gain (dB)
0.7
0.6
0
0.5
0.4
0.3
0.2
0.1
0

1450 1450 1450 1450 1450 1450

Wavelength (nm)
(a)
On-Off Raman Gain (dB)

p (nm) = 1450 1480 1510 1540

0
1450 1500 1550 1600 1650 1700
Wavelength (nm)
(b)

Fig. 13.13. Gain spectra of a flouride EDFA and a Raman amplifier with a dispersion-
compensation fiber: (a) EDFFA gain as a function of average fractional population of the
upper state; (b) on-off Raman gain as a function of pump wavelength (λp).

respectively. The significantly wider gain bandwidth of the SWB-HA, compared to

the individual gain bandwidths of the EDFFA and the RA, was obtained without a
gain equalizer by the single-wavelength pumping approach, because the gain spectra
of the EDFFA and RA have opposite gain slopes. Moreover, significantly improved
gain flatness is obtained by the two-wavelength pumping if the optimum λp values
are selected. The gain bandwidths obtained with an ideal GEQ for the two cases are
both about 90 nm (1530 to 1620nm). The SWB-HA also offers the benefits of low
noise and high output power as shown in Fig. 13.2.
The wideband NF spectra of the single-wavelength pumped Raman amplifiers to-
gether with their gain spectra are shown in Fig. 13.15. The NF and gain spectra were
calculated by Eqs. (13.5) to (13.8). The spectra of the on-off Raman gain Gd and the
effective noise figure Feff of the DRA with 80 km DSF are shown in Fig. 13.15(a).
 13. Hybrid EDFA/Raman Amplifiers 429

EDFA

Raman (
p = 1510 nm)

Total
35

30

25

20
Gain (dB)

15

10

–5
1500 1550 1600 1650
Wavelength (nm)
(a)

EDFA

Raman (
p = 1510 nm)

Raman (
p = 1480 nm)

Total
35

30

25

20
Gain (dB)

15

10

–5
1500 1550 1600 1650
Wavelength (nm)
(b)

Fig. 13.14. Examples of gain spectra of a hybrid amplifier. λp: Raman pump wavelength;
N2ave : average fractional population of the upper state. (a) N2ave = 0.7, λp = 1510 nm;
(b) N2ave = 0.7, λp = 1510 and 1480 nm.
430 H. Masuda

On-Off Gain Net Gain

Effective NF @ 298 K NF @ 298 K
Effective NF @ 0 K NF @ 0 K

Net Gain, Noise Figure (dB)

Effective Noise Figure (dB)
15 12
10
8
On-Off Gain,
10
6
4
5
2
0
0
–2
–4
1500 1550 1600 1650 1500 1550 1600 1650
Wavelength (nm) Wavelength (nm)
(a) (b)

Fig. 13.15. Gain and noise figure spectra of single-wavelength pumped Raman amplifiers:
(a) on-off gain and effective noise figure of a distributed Raman amplifier with 80 km dispersion-
shifted fiber; (b) net gain and noise figure of a discrete Raman amplifier with 8 km dispersion-
compensation fiber (pump wavelength = 1480 nm; gain: solid line; noise figure at 298◦ K
(room temperature): broken line; noise figure at 0◦ K: dotted line).

On the other hand, the net gain Gl and noise figure F of the LRA (lumped or discrete
Raman amplifier) with 8 km DCF are shown in Fig. 13.15(b). Feff and F at the tem-
peratures of 298◦ K (room temperature) and 0◦ K are shown for the two amplifiers in
the figures. The pump wavelength is set to 1480 nm. Measured absorption coefficients
at the signal and pump wavelengths were used in the calculations. Peak Gd of the
DRA is set to 15 dB, and peak Gl of the LRA is set to 10 dB. Feff of the DRA at the
peak gain wavelength of about 1580 nm is −2.3 dB at room temperature, whereas
Feff at a wavelength near the pump wavelength of 1500 nm is 3 dB. On the other
hand, F of the LRA at about 1580 nm is 5 dB at room temperature, whereas F at
1500 nm is 9 dB. The difference in NF (Feff or F ) between the two temperatures
(room temperature, 298◦ K, and 0◦ K) for signal wavelengths over 1500 nm is less
than 3 dB or so; this reflects the temperature dependence of the phonon population
factor Nphon in Eq. (13.7). NFs of the four types of SWB-HAs can be determined by
considering the NF spectra of the RAs of Fig. 13.15 and the NF spectra of the EDFAs
of Fig. 13.5 as shown in the next section.

13.5.2. Four Types of Amplifiers

13.5.2.1. Classification of Amplifiers
Partitioning of the EDFA gain (GEDFA ) and the Raman gain (GRaman ) in an HA is
shown in Fig. 13.16. The total amplifier gain (GHybrid ) is the sum of the two gains:
GHybrid = GEDFA + GRaman , where gain has units of dB. For the SWB-HA, GRaman is
defined as the peak Raman gain and GEDFA is, in this section, defined at the wavelength
that gives the peak Raman gain GRaman . We can classify the SWB-HA into four types
according to its GRaman and gain types (distributed or discrete). Table 13.1 shows the
 13. Hybrid EDFA/Raman Amplifiers 431

Hybrid Amplifier Hybrid Amplifier

with Small Raman Gain with Large Raman Gain

Amplifier Gain
Raman Gain

EDFA Gain

Hybrid Amplifier
EDFA Raman Amplifier

Fig. 13.16. Gain partitioning in a hybrid amplifier.

Table 13.1.
Raman Gain Distributed Discrete
Small Type-1 Type-3
Large Type-2 Type-4

classification with the four types [20]. The SWB-HA with small (large) distributed
Raman gain is denoted as Type-1 (2). On the other hand, the SWB-HA with a small
(large) discrete Raman gain is denoted as Type-3 (4). Typical gain examples are
GHybrid = 25 dB, GRaman = 10 dB, and GEDFA = 15 dB, for Types -1 and -3 with
small Raman gain, and GHybrid = 25 dB, GRaman = 20 dB, and GEDFA = 5 dB, for
Types -2 and -4 with large Raman gain.
The four types of SWB-HAs have different basic configurations as shown in
Fig. 13.17 [20], and thus have different gain, noise, and output characteristics. Am-
plifier configurations of Types -1 to -4 are shown in Figs. 13.17(a) to 13.17(d), re-
spectively. Optical components such as isolators in the amplifiers are not shown in the
figures for simplicity. Note that the EDFs are forward pumped and the DCFs are back-
ward pumped in Fig. 13.17, because this approach is common. However, the opposite
pump directions can be employed if needed. The basic amplifier configurations and
the amplification characteristics of the four types are described below. Experimental
amplifier configurations and their amplification characteristics are described in the
following sections.
First, the Type-1 amplifier has a two-stage EDFA with an intermediate GEQ and
a DCF. The two-stage EDFA configuration is employed because large EDFA gain is
required. The amplifier also has a DRA with a transmission fiber as its gain medium in
front of the EDFA. The peak loss of the GEQ is almost equal to that of the wideband
two-stage EDFA described in Section 13.3. The effective NF spectrum of the amplifier
is determined by both the effective NF spectrum of the DRA and the NF spectrum of
the two-stage EDFA. The output power characteristics, however, are determined by
the two-stage EDFA.
432 H. Masuda

Signal Coupler Coupler EDF Coupler EDF Signal

Transmission DCF
Gain
Fiber Pump Pump Equalizer Pump
P P P
(a)

Signal Coupler Coupler EDF Signal

Transmission DCF
Gain
Fiber Pump
Equalizer
P P
(b)

Signal Coupler EDF Coupler Coupler EDF Signal

Transmission DCF
Gain
Fiber Pump Equalizer Pump Pump
P P P
(c)

Signal EDF Signal

Coupler Coupler Coupler

Transmission DCF DCF

Gain
Fiber Pump Equalizer Pump Pump
P P P
(d)

Fig. 13.17. Configurations of the four types of seamless and wideband hybrid amplifiers:
(a) Type-1 with small distributed Raman gain; (b) Type-2 with large distributed Raman gain;
(c) Type-3 with small discrete Raman gain; (d) Type-4 with large discrete Raman gain; (EDF:
erbium-doped fiber amplifier; DCF: dispersion compensation fiber).

Next, the Type-2 amplifier has a single-stage EDFA with a GEQ and a DCF set
in front of the EDF in the EDFA. The amplifier also has a DRA with a transmission
fiber as its gain medium. The peak loss of the GEQ is small as is expected from
the gain spectra of Fig. 13.14. The effective NF spectrum of the amplifier is mainly
determined by that of the DRA. However, both the single-stage EDFA and the DRA
determine the output power.
The Type-3 amplifier has a two-stage EDFA with intermediate GEQ and DCF.
The DCF is pumped and operates as an LRA. The peak loss of the GEQ is large. The
NF spectrum of the amplifier is mainly determined by that of the first-stage EDF of
the two-stage EDFA, but the output power is determined by the second-stage EDF.
Finally, the Type-4 amplifier has a single-stage EDFA, a two-stage LRA, and an
intermediate GEQ. The LRA has two DCFs as its gain media and generates a large
Raman gain. The peak loss of the GEQ is small. The NF spectrum of the amplifier is
determined by the NF spectra of the EDFA and the LRA.
 13. Hybrid EDFA/Raman Amplifiers 433

Distributed
Raman Amplifier EDFFA
DSF EDFF EDFF
Signal 85 km 5.3 m 8.6 m Signal
Circulator Coupler Coupler
Gain
Equalizer
Pump Isolator
Pump Pump
LD
1505 nm LD Pump LD Pump
PBC 1.48  m 1.48  m
LD Pump

Fig. 13.18. Configuration of a Type-1 seamless and wideband hybrid amplifier (DSF:
dispersion-shifted fiber; PBC: polarization beam combiner; EDFF(A): erbium-doped flouride
fiber (amplifier)). (After Ref. [31]. © 1998 IEE)

13.5.2.2. Type-1 Amplifier with a Small Distributed Raman Gain

Figure 13.18 shows the experimental configuration of a Type-1 SWB-HA with small
distributed Raman gain [31]. The amplifier has a two-stage erbium-doped fluoride
fiber amplifier and a DRA. The EDFFA consists of two EDFFs and a GEQ. Each of
the EDFFs is forward pumped by a 1.48 µm LD. The first and second EDFFs are
5.3 and 8.6 m long, respectively. The GEQ is a cascade of two Mach–Zehnder filters
(GEQ-1 and -2). The free spectral ranges of GEQ-1 and -2 are both 100 nm. The DRA
consists of an 85 km DSF as the gain medium and a 1505 nm polarization-multiplexed
pump LD source. The pump light from the LD source is launched into the DSF via
an optical circulator. The pump power launched into the DSF is 209 mW.
Gain characteristics of the HA are shown in Fig. 13.19. Spectra of the EDFFA
gain, the DRA on-off gain Gd , and the HA gain (the sum of the EDFA and Raman
gains) are shown together with the loss spectrum of the DSF in Fig. 13.19(a). Losses
of GEQ-1 and -2, and the sum of them (GEQ) are shown in Fig. 13.19(b). The peak
loss of GEQ is 26 dB at 1560 nm. The net gain is defined as the difference between
the HA gain and the DSF loss. The λ3dB value (3 dB gain reduction bandwidth) of
the net gain is 75 nm (1531 to 1606 nm) with a peak gain of 22.8 dB. The peak Gd
of the DRA is 11.1 dB at 1610 nm.
Figure 13.20 shows the spectral characteristics of the OSNR of the HA. The figure
shows measured OSNRs of the HA, measured OSNRs of the transmission line using
a 1.48 µm pumped EDFFA, and calculated OSNRs of the transmission line using a
discrete amplifier with a noise figure of 3, 6, or 9 dB. The signal power launched into
the DSF was set to −14.5 dBm per channel, and the noise bandwidth was set to 0.1 nm.
The effective NFs of the HA were 5.5 dB at 1540 nm and 0.7 dB at 1600 nm, whereas
the NFs of the EDFFA were 7.5 dB at 1540 nm and 6.7 dB at 1600 nm. Therefore, the
improvement in NF between the HA and the EDFFA transmission lines ranged from
2 to 7 dB in the 3 dB gain band. The measured effective NF spectrum of the HA can
be well explained by the analyses described in Sections 13.4.1 and 13.5.1 (Figs. 13.9
and 13.15).
434 H. Masuda

30

25 EDFA + Raman DSF Loss

Gain
20

Gain Loss (dB)

15 EDFA
Gain
10

5 Raman
Gain
0

–5
1520 1540 1560 1580 1600 1620
Wavelength (nm)
(a)

30

25 GEQ – 1 + GEQ – 2

20
Loss (dB)

GEQ – 1
15

10 GEQ – 2

0
1520 1540 1560 1580 1600 1620
Wavelength (nm)
(b)

Fig. 13.19. Characteristics of the gain spectra of a Type-1 seamless and wideband hybrid ampli-
fier; (a) Spectra of gains (symbols) and DSF loss (line), EDFA: open triangle, Raman amplifier:
filled square, EDFA + Raman amplifier: open circle; (b) Loss spectra of gain equalizers (GEQ-1
and -2); (c) Net gain spectrum. (After Ref. [31]. © 1998 IEE)

A Type-1 SWB-HA using an EDSFA was also fabricated and evaluated [32].
The SWB-HA reported in 1997 was to the author’s knowledge, the first hybrid
EDFA/Raman amplifier to be described. λ3dB of the net gain was 67 nm (1549
to 1616 nm) with a peak gain of 20 dB. The λ3dB of the SWB-HA using an EDFFA
(75 nm) is larger than that of the SWB-HA using an EDSFA (67 nm). This is because
there is a smaller difference between the gain near 1530 nm and the gain near 1600
nm in the EDFFA without GEQ than in the EDSFA without GEQ (Fig. 13.4).
 13. Hybrid EDFA/Raman Amplifiers 435

26

24

Optical SNR (dB)

22
NF =
20 3 dB

18
6 dB
16
9 dB
14
1520 1540 1560 1580 1600 1620
Wavelength (nm)

Fig. 13.20. Optical signal-to-noise ratio spectra of a Type-1 seamless and wideband hybrid
amplifier (hybrid amplifier, EDFA (measured), EDFA with noise figure of 3, 6, or 9 dB (calcu-
lated)). (After Ref. [31]. © 1998 IEE)

13.5.2.3. Type-2 Amplifier with a Large Distributed Raman Gain

A Type-2 SWB-HA with a large peak Gd of 25 dB was fabricated using two 50 km

DSFs as the Raman gain media [24]. The detailed configuration of the experimental
HA is not shown here for simplicity. Note that the basic configuration is shown in
Fig. 13.17(b). The HA has two units in the 100 km transmission line consisting of
the two DSFs. The first HA unit has a single stage 1.48 µm pumped EDFFA and a
50 km DSF. The DSF was backward pumped by a 1.51 µm polarization-multiplexed
LD module and was forward pumped by a 1.48 µm LD. However, the second HA
unit has a single-stage 1.48 µm pumped EDSFA and a 50 km DSF. The DSF was
backward pumped by a 1.51 µm polarization-multiplexed LD module. The pump
powers launched into the DSFs were about 200 mW for the two 1.51 µm LD modules,
and about 100 mW for the 1.48 µm LD. A simple Mach–Zehnder type GEQ with a
peak-to-peak loss of about 2 dB was set after the second HA unit if needed.
Figure 13.21 shows gain spectra of the Type-2 HA and loss spectrum of the 100 km
transmission line without the GEQ. The peak gain of the EDFA (the sum of the EDFFA
and EDSFA) was 14.5 dB at 1530 nm, and the peak Gd of the DRA (the sum of the
two DRAs) was 25 dB at 1610 nm. The λ3dB for the net gain was 79.9 nm (1530.6
to 1610.5 nm) without the GEQ and 82.8 nm (1528.8 to 1611.6 nm) with the GEQ.
The peak gain of the HA with the GEQ was 22 dB.
Optical SNR spectra for the HA are shown in Fig. 13.22. The figure shows a
measured OSNR spectrum of the HA without the GEQ and three calculated OSNR
spectra of the transmission line using two discrete amplifiers with the same NF of
3, 5, or 7 dB. The noise bandwidth is 0.1 nm. Feff of the HA was 3 dB at 1530 nm
and −2 dB at 1600 nm. The Type-2 HA has lower Feff values than the Type-1 HA
(Section 13.5.2.2). This is because it has larger Gd values.
Another Type-2 HA, consisting of a 1497 nm wavelength pumped DRA and
a 0.98 µm pumped EDSFA, was reported as an inline amplifier for a submarine
436 H. Masuda

30

(EDFA + Raman) Gain

Gain, Loss (dB) 20

Raman
Gain DSF Loss
10

EDFA
Gain
0

–10
1520 1540 1560 1580 1600 1620
Wavelength (nm)

Fig. 13.21. Gain spectra of a Type-2 seamless and wideband hybrid amplifier and loss spectrum
of the transmission line with two 50 km DSFs (total length is 100 km). (After Ref. [24].
© 1998 IEE)

28

26

24
Optical SNR (dB)

22 NF =
3 dB
20

18 5 dB

16 7 dB

14
1520 1540 1560 1580 1600 1620
Wavelength (nm)

Fig. 13.22. Optical signal-to-noise ratio spectra of a Type-2 seamless and wideband hybrid
amplifier (hybrid amplifier (measured), transmission line using two discrete amplifiers with a
noise figure of 3, 5, or 7 dB (calculated)). (After Ref. [24]. © 1998 IEE)
 13. Hybrid EDFA/Raman Amplifiers 437

EDFFA – 1 Raman Amp. EDFFA – 2

EDFF DCF EDFF

Signal 5.3 m Coupler 8 km Signal
Coupler 8.6 m
Gain
Equalizer
Isolator Isolator Isolator
Pump Pump Pump

PBC

LD Pump LD Pump LD Pump

1.48 m 1.51 m 1.48 m

Fig. 13.23. Configuration of a Type-3 seamless and wideband hybrid amplifier (PBC: polariza-
tion beam combiner, DCF: dispersion-compensation fiber, EDFF(A): erbium-doped flouride
fiber (amplifier)). (After Ref. [34]. © 1998 IEE)

transmission system [33]. The peak gain of the EDSFA was comparable to that of the
DRA. A seamless flat gain bandwidth of 80 nm (1527 to 1607 nm) was obtained.

13.5.2.4. Type-3 Amplifier with Small Discrete Raman Gain

The experimental configuration of a Type-3 discrete SWB-HA is shown in Fig. 13.23

[34]. The amplifier has two-stage EDFFAs (EDFFA-1 and -2), an intermediate LRA,
and an intermediate GEQ. EDFFA-1 has a 5.3 m EDFF, which is backward pumped
by a 1.48 µm LD, and EDFFA-2 has an 8.6 m EDFF, which is forward pumped by a
1.48 µm LD. The LRA uses an 8 km DCF as its gain medium, which is pumped by a
1.51 µm polarization-multiplexed LD module. The 1.51 µm pump power launched
into the DCF is 197 mW. The GEQ has the same structure of the GEQ used in the
Type-1 HA (Section 13.5.2.2). The peak loss of the GEQ is about 25 dB at 1560 nm.
Figure 13.24 shows the gain spectra of the Type-3 HA. The figure plots the gains of
the two-stage EDFFAs including the GEQ loss (EDFFA-1 + EDFFA-2 + GEQ), the
LRA, and the sum of them (HA). The peak gain of the two-stage EDFFAs including
the GEQ is 19 dB at 1535 nm, and the peak gain of the LRA is 8.9 dB (on-off gain of
12.1 dB) at 1610 nm. Moreover, the peak gain of the HA is 19.8 dB. λ3dB is 75 nm
(1531 to 1606 nm).
The NF spectrum of the HA is shown in Fig. 13.25. NFs in the 3 dB gain band
range from 4.9 to 7.4 dB at the input point of the HA. The NFs are from 3.8 to 6.3 dB
at the input point of the first EDFF when we consider the 1.1 dB insertion loss of the
input isolator. The NF spectrum is mainly determined by that of the 1.48 µm pumped
EDFF (Fig. 13.5).

13.5.2.5. Type-4 Amplifier with Large Discrete Raman Gain

A Type-4 discrete SWB-HA was fabricated using the basic amplifier configuration
of Fig. 13.17(d) [23]. The experimental detailed configuration is not shown here for
simplicity. The amplifier had a single-stage EDFFA in the input side of the amplifier,
438 H. Masuda

∆
= 75 nm

25

Hybrid Amp.
20

15
Gain (dB) EDFFA + GEQ
10

Raman Amp.
5

–5
1520 1520 1520 1520 1520 1520
Wavelength (nm)

Fig. 13.24. Gain spectra of a Type-3 seamless and wideband hybrid amplifier (Open square:
EDFFA + GEQ, EDFFA = EDFFA-1 + EDFFA-2; Open triangle: Raman amplifier; Open
circle: hybrid amplifier (EDFFA + GEQ + Raman amp.); λ: 3 dB gain reduction bandwidth).
(After Ref. [34]. © 1998 IEE)

10

9
Noise Figure (dB)

4
1520 1540 1560 1580 1600 1620
Wavelength (nm)

Fig. 13.25. Noise figure spectrum of a Type-3 seamless and wideband hybrid amplifier. (After
Ref. [34]. © 1998 IEE)

a two-stage LRA in the output side of the amplifier, and a GEQ downstream of the
Raman amplifier. The GEQ can be set between the EDFFA and LRA with negligible
variations in the noise and gain characteristics because the peak loss of the GEQ was
only 3.5 dB. The EDFFA had a short length (3.3 m) of a EDFF, which was forward
pumped by a 1465 nm LD. The LRA had two DCFs as its gain media. The lengths
of the first and second DCFs (DSF-1 and -2) were 8.0 and 8.3 km, respectively. DCF-1
 13. Hybrid EDFA/Raman Amplifiers 439

∆
= 80.0 nm

30 Total

20
Gain (dB)

Raman
EDFA + GEQ

10

1520 1540 1560 1580 1600 1620

Wavelength (nm)

Fig. 13.26. Gain spectra of a Type-4 seamless and wideband hybrid amplifier (GEQ: gain
equalizer; filled triangle: EDFA + GEQ; open circle: Raman amplifier; open square: hybrid
amplifier (EDFA + GEQ + Raman amplifier); λ: 3 dB gain reduction bandwith). (After
Ref. [23]. © 1999 IEE)

was pumped by a 1472 nm forward pump (P1) and a 1495 nm backward pump
(P2), whereas DCF-2 was pumped by a 1503 nm forward pump (P3) and a 1503 nm
backward pump (P4). Each of the pump light sources was a polarization-multiplexed
LD module. The pump powers launched into the DCFs for P1 to P4 were 188, 108,
170, and 149 mW, respectively. Therefore, the LRA was bidirectionally pumped with
three wavelengths (1472, 1495, and 1503 nm) and with a total power of 615 mW.
Figure 13.26 shows the gain spectra of the HA. The figure plots the gains of the
sum of the EDFFA and GEQ, the LRA, and the sum of them (HA). The peak gain of
the sum of the EDFFA and GEQ is 25 dB at 1531 nm, and the peak gain of the LRA
is 21 dB at 1600 nm. Moreover, the peak gain of the HA is 28.1 dB. λ3dB is 80 nm
(1530 to 1610 nm). The NF spectrum of the amplifier is shown in Fig. 13.27. The
NFs in the 3 dB gain band are less than 6.0 dB. The NFs at the wavelengths shorter
than about 1570 nm are mainly determined by the NFs of the EDFFA, because the
EDFFA gains are larger than 10 dB at the wavelengths. On the other hand, the NFs
at the wavelengths longer than about 1570 nm are determined by both the NFs of
the EDFFA and LRA. A total output power with 15 WDM signals of 13.8 dBm was
obtained under the error-free operation.
Each performance of the four types of SWB-HAs described above was tested in a
WDM transmission experiment [20, 23, 24, 31, 32, 35, 36]. Each SWB-HA was used
as an inline amplifier. There was no power penalty within the experimental accuracy
in the bit error rate curves between before and after transmission for all WDM signal
channels in the 3 dB gain band.
440 H. Masuda

Noise Figure (dB)

7

4
1520 1540 1560 1580 1600 1620
Wavelength (nm)

Fig. 13.27. Noise figure spectrum a Type-4 seamless and wideband hybrid amplifier. (After
Ref. [23]. © 1999 IEE)

The complexities and costs of the four types of SWB-HAs can be compared with
some narrowband amplifiers. The configurations of the Types -1 and -2 SWB-HAs
using distributed amplification (Figs. 13.2(d), 13.17(a), 13.17(b), 13.18) are compared
with those of the NB-HA (Figs. 13.1(c), 13.10). However, the configurations of the
Types -3 and -4 SWB-HAs without using distributed amplification (Figs. 13.2(a),
13.2(b), 13.2(c), 13.17(c), 13.17(d), 13.23) are compared with those of the C- or
L-band EDFA (Fig. 13.1(a)). Differences in the number of optical components and
the cost between each of the SWB-HAs and the NB-HA or the EDFA are small.
However, the gain bandwidths of the SWB-HAs (∼80 nm) are about twice those of
the NB-HA or the EDFA (∼30 to 40 nm). Therefore, the SWB-HAs are cost-effective
wideband amplifiers.

13.6. Hybrid TDFA/Raman Amplifiers and Hybrid

Tellurite/Silica Raman Amplifiers
Two kinds of HAs other than EDFA/Raman HAs (hybrid EDFA/Raman amplifiers)
have been reported. The first kind is an HA using a thulium-doped fiber amplifier
(TDFA) [37–39] and an RA (TDFA/Raman HA) in cascade [40–44]. The other kind
is an HA using a tellurite Raman amplifier and a silica fiber Raman amplifier (tellu-
rite/silica Raman HA) in cascade [45–48].
The TDFA/Raman HA has some types as in the case of the EDFA/Raman HA.
The first type has basic configurations such as Figs. 13.2(a) and 13.2(b). The HA
using the TDFA followed by a discrete RA or using the discrete RA followed by a
TDFA were reported [40, 41]. As an example, the HA has a TDFA peak gain of 31 dB
at 1462 nm and an RA peak gain of 15 dB at 1510 nm [40]. λ3dB of the amplifier
is about 50 nm (1460 to 1510 nm) with a peak gain of 25 dB. The bandwidths with
over 20 dB gains are about 70 nm (1450 to 1520 nm) [40, 41]. The second type
is a narrowband hybrid TDFA/distributed Raman amplifier [42]. The amplifier uses
a gain-shifted TDFA with a gain bandwidth of about 30 nm (1480 to 1510 nm).
 13. Hybrid EDFA/Raman Amplifiers 441

Finally, the third type is a wideband TDFA/Raman HA using two-stage RAs and an
intermediate TDFA [44]. The amplifier has two DCFs as Raman gain media at its
input and output sides. λ3dB is 76 nm (1462 to 1538 nm) with a peak gain of 31.5
dB, and the bandwidth with over 20 dB gains is 90 nm (1455 to 1545 nm).
The tellurite/silica Raman HAhas two types [45–48]: using a discrete silica RAand
using both discrete and distributed silica RAs. A single-wavelength pumped tellurite
RA has a two-peak gain spectrum [45], whereas a single-wavelength pumped silica
RA has a single-peak gain spectrum as is well known [25]. Moreover, a multiwave-
length pumped tellurite RA generated an ultrawide gain bandwidth of 160 nm (1490 to
1650 nm) with gains over 10 dB and NFs under 10 dB [45]. However, the relative gain
flatness is a little bit poor (100%). The flatness is defined as (Gmax − Gmin )/Gmin ,
where Gmax and Gmin are the maximum and minimum gains in dB units, respectively.
The first type, tellurite/silica Raman HA, has two-stage tellurite RAs and an interme-
diate silica RA [46]. The pump wavelengths of the silica RA were chosen so that the
gain peaks of the silica RA compensated for the gain dips of the two-stage tellurite
RAs. A flattened gain spectrum with the flatness of 52% without gain equalizer was
obtained. A seamless gain bandwidth of 135 nm (1497 to 1632 nm) with a minimum
gain of 22.8 dB was obtained.
The second type, tellurite/silica Raman HA, has two-stage tellurite RAs, an inter-
mediate discrete silica RA, an intermediate GEQ, and a distributed silica RA with a
transmission fiber (80 km standard SMF) in front of the first-stage tellurite RA [47].
A 6 dB gain-reduction bandwidth of the amplifier was 120 nm (1485 to 1605 nm)
with a peak gain of 18 dB. The effective NFs in the short wavelength region were
about 5 dB. The low effective NFs are important for the wideband and large capacity
transmission.

13.7. Conclusions

Technologies on hybrid EDFA/Raman amplifiers, each of which use an EDFA and a

DRA or LRA (distributed or discrete fiber Raman amplifier) in cascade, have been
described. The HAs (hybrid amplifiers) have two kinds, a narrowband HA (NB-HA)
and a seamless and wideband HA (SWB-HA). The typical gain bandwidths λ of the
NB-HA were ∼30 to 40 nm in the C- or L-band, whereas those of the SWB-HA were
∼70 to 80 nm in the C- and L-bands. An NB-HA showed an improvement in the opti-
cal SNR of about 5 dB both experimentally and theoretically thanks to the distributed
amplification in the amplifier. Inherent spectral characteristics of the EDFA gain
and Raman gain were utilized so that wideband and efficient gain equalization were
achieved. The large Raman gains compensate for the small EDFA gains in the long-
wavelength region, and simultaneously the large EDFA gains also compensate for
the small Raman gains in the short-wavelength region. The SWB-HA has four types
according to its manner (distributed or discrete) and magnitude (small or large) of Ra-
man amplification. The two types of SWB-HAs with a small peak Raman gain (Types
-1 and -3) generated λ of about 75 nm, whereas the two types of SWB-HAs with a
large peak Raman gain (Types -2 and -4) generated λ of about 80 nm. Each type of
442 H. Masuda

SWB-HAs had a simple configuration with the number of optical components, which
is comparable with that of the NB-HA. Therefore, the four types of SWB-HAs are
cost-effective wideband amplifiers. The SWB-HAs also showed low noise and high
output power characteristics due to their multistage amplification schemes. The hybrid
amplifiers (both NB- and SWB-HAs) were successfully used in WDM transmission
experiments with error-free operation for all signal channels in their gain bands.

References
[1] E. Desurvire, Erbium-Doped Fiber Amplifier, New York: Wiley, 1994.
[2] M. Yamada, H. Ono, T. Kanamori, S. Sudo, and Y. Ohishi, Electron. Lett., 33: 710, 1997.
[3] Y. Sun, J.W. Sulhoff, A.K. Srivastava, A. Abramov, T.A. Strasser, P.F. Wysocki, J.R.
Pedrazzani, J.B. Judkins, R.P. Espindola, C. Wolf, J.L. Zyskind, A.M. Vengsarker, and J.
Zhou, ECOC, 53, 1998.
[4] A.K. Srivastava, L. Zhang, Y. Sun, J.W. Sulhoff, and C. Wolf: OFC, FC2: 53, 1999.
[5] H. Masuda, S. Kawai, and K-I. Suzuki, Electron. Lett., 35: 411, 1999.
[6] P.B. Hansen, A. Stentz, T.N. Nielsen, R. Espindola, L.E. Nelson, and A.A. Abramov,
OFC, PD8, 1999.
[7] N. Takachio, H. Suzuki, H. Masuda, and M. Koga, OFC, PD9, 1999.
[8] Y. Frignac, G. Charlet, W. Idler, R. Discher, P. Tran, S. Lanne, S. Borne, C. Martinelli, G.
Veith, A. Jourdan, J-P. Hamaide, and S. Bigo, OFC, FC5, 2002.
[9] T.N. Nielsen, A.J. Stentz, K. Rottwitt, D.S. Vengsarkar, Z.J. Chen, P.B. Hansen, J.H. Park,
K.S. Feder, S. Cabot, S. Stult, D.W. Peckham, L. Hsu, C.K. Kan, A.F. Judy, S.Y. Park,
L.E. Nielson, and L. Gruner-Nielsen, IEEE Photon. Technol. Lett., 12: 1079, 2000.
[10] A. B. Puc, M.W. Chbat, J.D. Henrie, N.A. Weaver, H. Kim, A. Kaminski, A. Rahman,
and H. Fevrier, OFC, PD39, 2001.
[11] P. Wysocki, J.B. Judkins, R.P. Espindola, M. Andrejco, and A.M. Vengsarkar: IEEE
Photon. Technol. Lett., 9: 1343, 1997.
[12] H. Masuda, S. Kawai, K.-I. Suzuki, K. Aida, Electron. Lett., 33: 1070, 1997.
[13] H. Masuda, S. Kawai, and K. Aida, Electron. Lett., 34: 567, 1998.
[14] H. Ono, A. Mori, K. Shikano, and M. Shimizu, IEEE Photon. Technol. Lett., 14: 1073,
2002.
[15] H. Ono, M. Yamada, T. Kanamori, S. Sudo, and Y. Ohishi: J. Lightwave Technol., 17:
490, 1999.
[16] M. Jinno, T. Sakamoto, M. Fukui, S. Aisawa, J. Kani, and K. Oguchi, OECC, 16A1: 404,
1998.
[17] Y. Emori, K. Tanaka, and S. Namiki, Electron. Lett., 35: 1355, 1999.
[18] S.A.E. Lewis, S.V. Chernikov, and J.R. Taylor, OAA ThA2: 72, 1999; S.A.E. Lewis, S.V.
Chernikov, and J.R. Taylor, Electron. Lett., 35: 1761, 1999.
[19] H. Masuda and S. Kawai, ECOC, 2: 146, 1999.
[20] H. Masuda, OFC, TuA1: 2, 2000.
[21] A. Mori, Y. Ohishi, and S. Sudo, Electron. Lett., 33: 863, 1997.
[22] N. Jolley, R.D. Muro, S. Parry, K. Cordina, and J. Mun, OAA, TuD2: 124, 1998.
[23] H. Masuda and S. Kawai, IEEE Photon. Technol. Lett., 11: 647, 1999; H. Masuda, S.
Kawai, and K. Aida, OAA, PD7, 1998.
[24] H. Masuda, S. Kawai, K.-I. Suzuki, and K. Aida, ECOC, 51, 1998.
[25] G.P. Agrawal, Nonlinear Fiber Optics, San Diego: Academic Press, 1989.
 13. Hybrid EDFA/Raman Amplifiers 443

[26] P.B. Hansen, L. Eskildsen, A.J. Stentz, T.A. Strasser, J. Judkins, J.J. DeMarco, R. Pedraz-
zani, and D.J. DiGiovanni, IEEE Photon. Technol. Lett., 10: 159, 1998.
[27] K. Rottwitt, M. Nissov, and F. Kerfoot, OFC, TuG1: 30, 1998; H. Kidolf, K. Rottwitt, M.
Nissov, M. Ma, and E. Rabarijaona, IEEE Photon. Technol. Lett., 11: 530, 1999; C.R.S.
Fludger, V. Handerek, and R.J. Mears, OFC, MA5, 2001.
[28] V. Curri, NFOEC, B1.1, 2000.
[29] H. Kawakami, Y. Miyamoto, K. Yonenaga, and H. Toba: OAA, ThB5, 110, 1999.
[30] H. Suzuki, J. Kani, H. Masuda, N. Takachio, K. Iwatsuki, Y. Tada, M. Sumida, IEEE
Photon. Technol. Lett., 12: 903, 2000.
[31] H. Masuda, and S. Kawai, IEEE Photon. Technol. Lett., 10: 516, 1998.
[32] H. Masuda, K.-I. Suzuki, S. Kawai, and K. Aida, Type-1, 65 nm, Electron. Lett., 33: 753,
1997. H. Masuda, S. Kawai, K.-I. Suzuki, and K. Aida, OAA, MC3, 40, 1997.
[33] D.G. Foursa, C.R. Davidson, M. Nissov, M.A. Mills, L. Xu, J.X. Cai, A.N. Pilipetskii,
Y. Cai, C. Breverman, R.R. Cordell, T.J. Carvelli, P.C. Corbett, H.D. Kidorf, and N.S.
Bergano, OFC, FC3, 2002.
[34] S. Kawai, H. Masuda, K.-I. Suzuki, and K. Aida, Electron. Lett., 34: 897, 1998.
[35] S. Kawai, H. Masuda, K.-I. Suzuki, and K. Aida, OFC, FC3, 56, 1999.
[36] H. Suzuki, N. Takachio, Y. Hamazumi, H. Masuda, S. Kawai, and K. Araya, OFC, ThO4:
221, 1999.
[37] S. Sudo, ed., Optical Fiber Amplifiers, Boston: Artech House, 1997.
[38] T. Sakamoto, OFC, TuQ1, 2001.
[39] S. Aozasa, H. Masuda, H. Ono, T. Sakamoto, T. Kanamori, Y. Ohishi, and M. Shimizu,
OFC, PD1, 2001.
[40] J. Kani and M. Jinno, Electron. Lett., 35: 1004, 1999.
[41] J. Masum-Thomas, D. Crippa, and A. Maroney, OFC, WDD9, 2001.
[42] K. Fukuchi, T. Kasamatsu, M. Morie, R. Ohhira, T. Ito, K. Sekiya, D. Ogasahara, and T.
Ono, OFC, PD24, 2001.
[43] Y. Miyamoto, A. Hirano, S. Kuwahara, Y. Tada, H. Masuda, S. Aozasa, K. Murata, and
H. Miyazawa, OAA, PD6, 2001.
[44] H. Masuda, S. Aozasa, and M. Shimizu, Electron. Lett., 38: 500, 2002.
[45] A. Mori, H. Masuda, K. Shikano, K. Oikawa, K. Kato, and M. Shimizu, Electron. Lett.,
37: 1442, 2001.
[46] H. Masuda, A. Mori, K. Shikano, K. Oikawa, K. Kato, and M. Shimizu, Electron. Lett.,
38: 867, 2002.
[47] H. Takara, H. Masuda, K. Mori, K. Sato, Y. Inoue, T. Ohara, A. Mori, M. Koutoku, Y.
Miyamoto, T. Morioka, and S. Kawanishi, OFC, FB1, 2002.
[48] H. Masuda, A. Mori, S. Aozasa, and M. Shimizu, OAA, OTuC1, 2002.
This page intentionally left blank
Chapter 14
Wideband Raman Amplifiers

Mohammed N. Islam, Carl DeWilde, and Amos Kuditcher

14.1. Introduction
This chapter describes designs and experiments that apply the Raman effect to wide-
band amplifiers (WBAs). In the context of this chapter, wideband corresponds to a
bandwidth of approximately 50 to 100 nm or more. We start by explaining the need
for WBAs, and briefly review some of the key enabling technologies for wideband
systems. Section 14.2 describes several approaches for WBA, including the erbium-
doped fiber amplifier (EDFA) and Raman amplifier combinations as well as all-Raman
amplifiers. Section 14.3 summarizes the advantages and challenges of the all-Raman
approach, the focus of this chapter. Section 14.4 identifies the key physical principles
that need to be considered in the design of all-Raman WBAs. Then, perhaps the most
important section of this chapter, Section 14.5 describes engineering design rules for
construction of all-Raman WBAs that satisfy gain and noise figure performance re-
quirements of typical long-haul and ultra-long-haul fiber-optic transmission systems.
Several WBA experiments that use either EDFA/Raman amplifier combinations or
all-Raman amplifiers are illustrated in Section 14.6, and exemplary wideband system
experiments are described in Section 14.7. Finally, we summarize and conclude the
chapter in Section 14.8.

14.1.1. Paths to High-Capacity Systems

As demand for throughput of fiber-optic networks continues to grow, the need to ex-
pand network capacity continues to increase, fueled primarily by the growth of the
Internet. In response to the increasing demand for throughput, significant research
and development effort has been directed at technologies for accessing more of the
intrinsic capacity of optical fiber. The capacity of optical fiber is the product of chan-
nel count and bit rate per channel or, alternately, the product of spectral width and
spectral efficiency. For capacity expansion to benefit network operators, upgraded
equipment must be provided at low ownership and operating cost. A common cost
metric is the cost of transmitting unit capacity over a unit distance (measured in
units of $/(Gb/s km)) [39]. Cost reduction then entails minimizing equipment cost
446 M.N. Islam et al.

and maximizing performance. Reducing the cost metric, either by minimizing equip-
ment cost or by maximizing capacity and reach, ensures cost savings are provided
without sacrificing performance. Raman amplification in wideband systems enables
cost reduction through optimization of all three factors of the cost metric, namely,
equipment cost, capacity, and reach. First, a single Raman amplifier requires fewer
components than splitband amplifiers, thereby reducing equipment cost [2]. Second,
the wide bandwidth achievable with all-Raman amplification can provide substantial
capacity allowing, for example, a bandwidth of 100 nm and a channel count of 240
with a channel spacing of 50 GHz. Finally, Raman amplification including distributed
Raman amplification provides a lower noise figure than purely discrete amplification,
thereby maximizing system reach (typically 1500 km or more, depending on the
number of channels required).
There are several ways of increasing fiber capacity: increasing bit rate, chan-
nel density, available spectrum, or spectral efficiency. Figure 14.1 schematically il-
lustrates the evolution of these approaches. In the early years of communication
networks, the most common means of increasing capacity was to increase time-
division-multiplexing bit rates. For example, bit rates increased from 64 kb/s (the
base rate of a voice signal) through 1.5 Mb/s (so-called T1), 45 Mb/s (so-called T3),
155 Mb/s (so-called OC-3), and on up to about 2.5 Gb/s (so-called OC-48). In recent
years, a channel speed of 10 Gb/s has become common, and products supporting
40 Gb/s channel speeds have become available, although deployment has been ham-
pered to date by prohibitive costs associated with the need for polarization-mode

capacity = fibers x spectrum

x spectral efficiency

.8b/Hz
spectral
efficiency
fiber capacity

WDM
demand
capacity = fibers x
channels x bit rate

TDM bit rates

time

Fig. 14.1. Evolution of optical transmission capacity.

 14. Wideband Raman Amplifiers 447

dispersion compensation, fine-tuned dispersion and dispersion slope-compensation,

and overcoming other transmission impairments.
Since the first deployment of wavelength-division-multiplexed (WDM) systems,
the number of WDM channels has continued to increase. The number of channels in
commercially available WDM systems has increased to about 240 and the bit rate to
10 Gb/s, corresponding to a single fiber capacity of 2.4 Tb/s. Back in 1995, it was
common to use a bit rate of 2.5 Gb/s on channels spaced by 200 GHz, which, at the
time, yielded four to eight channels. Channel spacing was limited by the accuracy
and stability of the transmitting lasers and of the multiplexing and demultiplexing
components. By 1999, systems featuring 10 Gb/s channels on 100 GHz channel
spacing had become commonplace [3], and over the next three years 50 GHz spac-
ing was introduced. Up to this point, the spectral extent of the signal comprising
any channel fits well within the bandwidth available to that channel, as long as a
fairly simple modulation scheme is used. Therefore, total capacity for a single fiber
can be calculated by summing the bit rate on each of the channels over the num-
ber of channels available within the total bandwidth delivered by the amplifier or
amplifiers.
Any further reduction in channel spacing or increase in bit rate at the same channel
spacing requires that careful consideration be given to spectral efficiency. As a result,
increasing bandwidth capacity is no longer a matter of just decreasing channel spacing,
but requires more sophisticated and costlier modulation techniques. At that point an
increase in bit rate without a change in modulation technique requires a proportional
increase in channel spacing and, consequently, a proportional decrease in the total
number of channels. Overall capacity remains the same and bandwidth capacity for
a single fiber, which is determined by multiplying available channels within a given
spectrum with the bit rate on each channel, no longer increases as bit rate increases
or channel spacing decreases. Only an increase in available spectrum or an increase
in spectral efficiency will lead to a higher bandwidth capacity. It is expected that the
cost of increasing spectral efficiency above about 0.8 bps/Hz will grow substantially,
by reason of a combination of factors: complicated transmitters and receivers, active
laser stabilization, as well as passive component stabilization.
Since the commercial deployment of optical amplifiers, there has also been a
significant increase in available bandwidth. The first systems used only the flat gain
section of an EDFA, corresponding to 10 to 15 nm. With the introduction of static
gain-flattening filters, gain bandwidths in commercial systems increased to 32 nm,
with laboratory experiments reporting as much as 40 nm. With the introduction of
the long-wavelength L-band by the end of the 1990s, bandwidth increased to as
much as 80 nm in EDFA-based systems utilizing a splitband configuration. In 2002,
100 nm bandwidth commercial systems using all-Raman amplification became avail-
able, and laboratory experiments demonstrated amplifiers with bandwidth greater than
130 nm. In comparing the three techniques for increasing capacity, the path followed
in commercial systems has proved to be the most economical mature technology.
Although combinations of all of the approaches can be used, the lowest cost approach
for the next several years is expected to be bandwidth expansion.
448 M.N. Islam et al.

14.1.2. Enabling Technologies for Wideband Systems

Several technologies are required to realize practical and economical wideband sys-
tems. Consider, for example, the key components for 100 nm bandwidth systems.
(Smaller bandwidth systems still require the same components or subsystems, but
with less stringent performance requirements.) First, wideband amplifiers are re-
quired. Various approaches for WBAs are described in the next section, and the
remainder of the chapter is focused on the all-Raman amplifier approach. Second,
broadband components such as multiplexers and demultiplexers are required. The
passband for couplers and other passive components needs to be larger than the ac-
tual bandwidth being used by the channels because in a cascaded system a spectrally
dependent transfer function decreases in bandwidth as a cascade of components is
traversed.
A third component that is critical for wideband systems is gain equalizers and
gain-flattening filters. Again, in a system consisting of a cascade of spans (e.g., a
1500 km system may have 19 spans each 80 km long), any nonuniformity in the gain
spectra of the amplifiers is accentuated as more and more amplifiers are traversed. A
gain variation of 0.5 dB can grow to well over 5 dB at the end of a system, resulting
in substantial variation in signal strength between channels, which can require a
large dynamic range at the receivers and lead to unwanted gain saturation, crosstalk
between channels, and substantial variation in bit error rate performance for different
channels. From a design perspective, the worst-performing channel limits the overall
system performance. Therefore, the best design achieves roughly equally performing
channels. Gain equalizers and gain-flattening filters are typically utilized to equalize
channels. In EDFAs, a bandwidth greater than 10 to 15 nm can only be obtained with
gain equalizing devices. In Raman amplifiers, gain shape can be more flat, reducing
constraints on gain equalization. However, gain equalization in Raman amplifiers is
still beneficial and is especially useful in wideband amplifiers. For uniform signal
traffic and time-invariant components, a static gain equalizer can be utilized; if, on
the other hand, traffic or component characteristics vary with time as may be the
case in a system that utilizes dynamic optical add-drop multiplexers or optical cross-
connects, then a dynamic gain equalizer may be required. Various devices have been
used for gain equalization. Static gain equalizers have been realized with waveguide
notch filters, Mach–Zender interferometers, dielectric filters, and fiber Bragg gratings.
Dynamic gain equalizers have been realized with microelectromechanical systems
(MEMS), liquid crystal, or thermooptic switch arrays integrated with a dispersive
element such as a diffraction grating. Acoustooptic filters have also been used for
dynamic gain equalization.
Another essential-component for wideband systems operating at 10 Gb/s or higher
bit rates comprises dispersion-compensating and dispersion slope-compensating
modules. Dispersion-compensating fiber (DCF), chirped fiber Bragg gratings, higher-
order mode fibers, and virtually imaged phased array microoptic devices have been
used for dispersion compensation, but DCF is the most widely deployed. As system
bandwidth increases, it is not adequate to match the dispersion magnitude at the center
of the band, as is commonly done in systems with less than 30 nm bandwidth. The
 14. Wideband Raman Amplifiers 449

dispersion slope of the compensator must also be matched to the dispersion slope of
the transmission fiber in order to achieve compensation over a wide spectral band. The
quantity to be matched is then the relative dispersion slope RDS = S/D, the ratio of
dispersion slope S to dispersion D. Manufacturers of dispersion-compensating fiber
are aware of the need for dispersion slope-compensation, and slope-compensating
fibers are now becoming available for most deployed transmission fiber types.

14.2. Alternate Approaches for Implementing Wideband

Amplifiers
Wideband amplification has been demonstrated with EDFAs, EDFA/Raman amplifier
hybrids, and all-Raman amplifiers. Figure 14.2 shows applicable wavelength bands
for various amplifier types [4]. Silica-based EDFAs have been most successful to date
and can provide amplification over the conventional C-band (∼1530 to 1565 nm) and
the long-wavelength L-band (∼1570 to 1605 nm). Bandwidths greater than 80 nm
have been demonstrated in splitband configurations employing parallel C- and L-band
EDFAs, and a continuous EDFA gain band greater than 70 nm has been demonstrated
by using new glass compositions such as tellurite. Other rare earth dopants such as
thulium can be used to provide gain in the short-wavelength S-band or in the L+
band (Chapter 10, Section 10.2). The spectral location of the gain band in Raman
amplification only depends on the spectral location of the pump band, so Raman
amplification can provide gain anywhere in the low-loss telecommunications window
between 1450 and 1650 nm. Likewise, semiconductor optical amplifiers can be used,
but a separate amplifier is required for every 30 nm band.

14.2.1. EDFAs in Various Glass Compositions

The most common glass composition for EDFAs comprises silica doped with alu-
minum, germanium, and erbium. For typical inversion levels of 40 to 60%, the flat

EDFA “C” EDFA “L”

Er Tellurite

Thulium gain media Thulium

Raman

SOA, segmented band, each up to ~30 nm

1450 1500 1550 1600 1650

Fig. 14.2. Gain bands of optical fiber amplifiers [4].

450 M.N. Islam et al.

gain band of silica-based EDFAs (EDSFAs) extends from about 1540 to 1560 nm,
corresponding to the central portion of the C-band. Using gain-equalizing filters, the
bandwidth of C-band amplifiers can be extended to about 40 nm. Further bandwidth
extension by means of gain-equalizing filters is hampered by very low gain near the
ends of the C-band. Therefore, gain shifting has been used to extend operation to
the 1560 to 1620 nm (L-band) wavelength range and, recently, the 1450 to 1520
nm (S-band) range. Long lengths of erbium-doped fiber together with low popula-
tion inversion rates, in the neighborhood of 20 to 40%, shift the gain band to long
wavelengths, yielding bandwidths of 43 nm for L-band EDSFAs [5]. Suppressing the
principal spontaneous emission peak at 1530 nm with distributed filtering shifts the
EDFA gain band to the S-band [6]. By combining gain-flattening and gain-shifting
techniques, a composite bandwidth of 84.3 nm was demonstrated for a splitband
amplifier employing parallel C- and L-band EDSFAs [5].
Ground state absorption on the blue end and excited state absorption on the red
end of the erbium emission spectrum limit the aggregate bandwidth that can be ob-
tained with EDSFAs. However, host glasses with a broader erbium emission spectrum
can be used to achieve a wider bandwidth. For example, erbium-doped fluoride fiber
amplifiers can achieve a wider flat gain band, and bandwidth can be further extended
with the aid of gain-flattening filters [7]. A gain-flattened tellurite-based EDFA has
been demonstrated to yield a continuous bandwidth greater than 70 nm [8]. In ad-
dition, antimony–silicate multicomponent silica glasses have been shown to extend
the bandwidth of L-band EDFAs to 60 nm, with which the composite bandwidth of
a splitband EDFA could be increased to 90 nm [9].

14.2.2. Other Dopings

Other dopants can be used to modify the characteristics of EDFAs or to achieve am-
plification in wavelength regions not accessible to conventional EDFAs. For example,
the bandwidth of an L-band EDFA is extended to 50 nm when a phosphorous-doped
fiber is utilized [10], and thulium-doped fiber amplifiers have been developed for
S-band amplification [11]. A bandwidth of approximately 70 nm can be achieved for
a splitband amplifier employing a Raman amplifier and thulium-doped fiber amplifier
sections [12].

14.2.3. Parallel C-, L-, and S-Band Amplifiers

Several amplifiers can be combined in a parallel configuration to achieve a wide band-

width. In this approach, couplers are used to split the WDM signal into wavelength
bands that are amplified separately and then recombined. Different amplifier tech-
nologies can be used in each band. For example, EDFAs and other rare earth-doped
fiber amplifiers employing various host fiber types, discrete Raman amplifiers, and
semiconductor optical amplifiers can used in the wavelength bands for which they
are best suited. A wide bandwidth, potentially covering the whole of the low-loss
window of transmission fiber, can be achieved [13].
 14. Wideband Raman Amplifiers 451

14.2.4. Hybrid Raman/EDFA

Hybrid amplification employing an EDFA and a Raman amplifier in series provides

another approach for extending the bandwidth of EDFAs. The EDFA provides gain
in the C-band whereas the Raman amplifier operates in the L-band. By incorporating
gain-flattening filters, continuous bandwidths ranging from 50 to 80 nm have been
achieved [14].

14.2.5. All-Raman Amplifiers

Raman amplification in silica fiber can yield a continuous bandwidth in excess of

100 nm, limited only by the Raman frequency shift in silica [15]. Wider bandwidths
can be achieved in fiber types with a larger frequency shift. A Raman amplifier using
tellurite fiber as gain medium was shown to yield a bandwidth of 160 nm [16]. The
Raman frequency spectrum in phosphorous-doped fiber includes peaks up to 40 THz
frequency shift, and phosphorous-doped fiber can potentially provide a bandwidth of
300 nm in all-Raman amplifiers [17].
Raman amplification provides gain over a single continuous band. This charac-
teristic is advantageous because a single-band amplifier requires fewer components
than one employing multiple amplifiers in a splitband configuration, and guard bands
are not required for the couplers (c.f. Fig. 1.15 in Chapter 1) [2]. Monitoring and
management of the gain profile is less complex over a single band because only a pair
of monitor taps, one at the input and one at the output of the amplifier, is required. In
addition, the dispersion-compensation module as well as much of the control circuitry
can be consolidated, thereby reducing the cost and size of the amplifier. In compari-
son, a splitband architecture employing separate amplifiers for each subband requires
separate dispersion-compensation modules, control circuitry, and monitoring of each
subband amplifier, thereby increasing the cost and footprint of the amplifier.

14.3. Advantages and Challenges of All-Raman Wideband

Amplifiers
Current trends suggest extensive deployment of Raman amplification in the future
for a number of reasons. First, distributed Raman amplification alone or in combi-
nation with lumped Raman amplification can improve the noise figure and reduce
system penalties arising from fiber nonlinearities. In optically amplified transmis-
sion systems, the principal limitations usually arise from amplifier noise and fiber
nonlinearities. In particular, the range of allowed signal levels is determined by two
factors: the minimum signal level is determined by the signal-to-noise ratio (SNR)
required to achieve a prescribed bit error rate; and the maximum signal level is de-
termined by impairments arising from nonlinear propagation effects in optical fiber.
Raman amplification eases the constraints arising from both of these limitations and
improves performance. As shown in Chapter 1, Fig. 1.18, distributed Raman amplifi-
cation reduces the excursion of the signal power level in comparison to purely lumped
452 M.N. Islam et al.

0
Power
– Raman Noise Figure
=
OSNR

–5 Raman Gain
(On/Off)
Power Profile (dBm)

Transmission Signal
–10 w/DRA

Power

–15

Transmission Signal
Attenuation Path
w/o DRA
–20

0 20 40 60 80
Transmission Length (km)

Fig. 14.3. Variation of signal power with distance in an optical fiber with and without DRA.
The minimum signal power is higher by a factor of Power with Raman amplification than
without amplification and optical signal-to-noise ratio is improved in the Raman amplified case
by Power divided by the optical noise figure.

amplification. The low signal power swing reduces signal impairments arising from
fiber nonlinearity, and the higher minimum signal level reduces SNR degradation.
Figure 14.3 illustrates the variation of signal power with distance along a distributed
Raman amplifier (DRA) and along an unamplified span of equal length. Improve-
ment in optical SNR (OSNR) in the distributed Raman amplifier depends on the ratio
Power of the minimum signal power with Raman amplification to the minimum sig-
nal power without Raman amplification. The SNR improvement is given by Power
divided by the noise figure of the Raman amplifier. This improvement can be as much
as 7 dB for an 80 km fiber span.
A second advantage of Raman amplification is that gain can be obtained in any
spectral region and over a wide spectral band. The location and width of the gain
band are determined by the pump wavelength location and distribution. The available
gain band for Raman amplification can be compared to the gain band for EDFAs in
Fig. 14.2.Although conventional EDFAs operate over the C-band (∼1525 to 1565 nm)
and the long-wavelength L-band (∼1565 to 1605 nm), Raman amplification can be
obtained anywhere within the telecommunication band, typically defined as 1300 to
1650 nm. Raman amplifiers can easily achieve 100 nm of continuous bandwidth, and
laboratory experiments have shown much larger bandwidths in a single amplifier. In
 14. Wideband Raman Amplifiers 453

present generation Raman amplifiers, several pump laser diodes operating at different
wavelengths are used. As a rule of thumb, the Raman gain bandwidth for each pump
laser is about 20 nm. Therefore, by using approximately five properly spaced pump
wavelengths, a gain bandwidth of 100 nm can be achieved. Note that the location and
number of pump wavelengths are determined not only by the gain bandwidth, but
also by gain-flatness requirements: the flatter the gain spectrum required, the greater
the number of wavelengths required.
Athird advantage of Raman amplification is that gain and dispersion-compensation
can be integrated in a lumped amplifier. Such an amplifier can have higher power
conversion efficiencies at high channel counts than EDFAs. Most systems operating
at line rates of 10 Gb/s (OC-192) or higher require dispersion compensation, com-
monly implemented using dispersion-compensating fiber. Dispersion-compensating
fiber turns out to be an efficient fiber type for Raman amplification: the Raman gain
coefficient can be as much as an order of magnitude larger in DCF than in standard
single-mode fiber (Chapter 1, Fig. 1.10). Hence, gain and dispersion-compensation
can be integrated in the same unit. Moreover, the efficiency of Raman amplification
in DCF can exceed that of EDFAs pumped at 1480 nm for channel counts exceeding
about 200 (c.f., Chapter 1, Fig. 1.16).
Beyond the advantages of Raman amplification enumerated above, additional
features of Raman amplification enable simple practical implementation in deployed
fiber-optic systems. Some of these features include:
• All-silica components: all Raman amplifier components can be based on fused-
silica, and all amplifier parts can be spliced into existing systems;
• Simple channel equalization: because gain shape can be smooth in a Raman
amplifier, gain equalization can be considerably simplified;
• Simple adjustment of gain flatness: adjustment of gain flatness can be made by
adjusting pump power levels [18], an important feature in, for example, a trans-
mission system using optical add-drop multiplexers or optical crossconnects;
• Large gain and power levels: because Raman amplifiers are relatively weakly
saturated even at high powers, gain and output power levels can be increased by
simply increasing pump power, and output powers in the watt range have been
reported; and
• Large power dynamic range for flat gain operation: unlike an EDFA where the gain
shape changes as the pump power is changed (i.e., as inversion level is changed),
the gain shape of a Raman amplifier can be maintained flat over a large dynamic
range of pump power and amplifier gain.
Despite a significant list of desirable features, Raman amplification is beset by
numerous challenges that dictate amplifier design and architecture. Gain flatness of a
wideband Raman amplifier can be significantly affected by pump–pump interactions
as well as interband and intraband stimulated Raman scattering (i.e., transfer of en-
ergy from short wavelength signals to longer wavelength signals through the Raman
effect). In addition, as the signal band in a wideband Raman amplifier approaches
the longest wavelength pump band, the noise figure of the amplifier increases be-
cause of thermal noise as further described below (see also Chapter 8). Furthermore,
454 M.N. Islam et al.

the electrical noise figure can be poorer than the optical noise figure in a Raman
amplifier. Because the upper state lifetime associated with Raman amplification is
short and long fiber lengths are typically used in Raman amplifiers, additional noise
sources contribute to the electrical noise figure. Multipath interference (MPI) arising
from double-Rayleigh scattering, coupling of pump fluctuations to signals, and pump-
mediated signal crosstalk increase the electrical noise figure with minimal effect on
the optical noise figure as usually determined from amplified spontaneous emission
(ASE) measurements. The next section examines the physical origins of these lim-
itations. Subsequent sections illustrate amplifier design approaches to mitigate the
limitations and present exemplary implementations.

14.4. Physical Origin of All-Raman Wideband Amplifier

Limitations
Several physical principles underlie the design of wideband Raman amplifiers. The
first to be considered is the interaction between different pump waves and different
signal waves through the Raman effect itself. Consider the Raman gain curve in fused
silica, as shown in Chapter 1, Fig. 1.1. The gain curve is continuous from the pump
frequency to a frequency shift as great as 40 THz away from the pump frequency with a
significant Stokes peak near 13 THz. In general, for any reference frequency, energy is
lost to lower frequency waves and gained from higher frequency waves. This is true for
Raman interaction between any set of waves and gives rise to signal–signal as well as
pump–pump power exchange. Therefore, in addition to energy transfer from pump to
signal waves, energy is transferred from short-wavelength to long-wavelength pumps
with the result that lower launched powers are usually required for long-wavelength
pumps and higher powers are needed at short pump wavelengths to achieve a flat gain
shape at a prescribed gain level. Figure 14.4 schematically exhibits a pump launch
power distribution taking into account pump–pump interaction to achieve comparable
gain at long and short signal wavelengths. Note that the shortest wavelength pumps
have the highest power at the pump input end of the fiber, because they not only
provide gain to signal waves, but also supply energy to the longer wavelength pumps.
To model the Raman interaction between waves propagating on an optical fiber, it
is customary to use a set of coupled mode equations. For copropagating steady-state
pump and signal waves, the coupled mode equations describing stimulated Raman
scattering take the form
dPs
= gPp Ps − αs Ps
dz
dPp νp
= − gPp Ps − αp Pp ,
dz νs
where Pp and Ps are the respective pump and signal powers, αp and αs are the fiber loss
coefficients at pump and signal frequencies νp and νs (νp > νs ), and g is the Raman
gain coefficient. Accompanying the scattering process are vibrational excitations that
 14. Wideband Raman Amplifiers 455

pumps
6

Raman Gain Coefficient

5
( 10–14 m/W)

3
signal
2 bandwidth

0
0 6 12 18 24 30 36 42 48
Frequency Shift (THz)

Fig. 14.4. Multipump Raman amplification. Gain bands from different pump wavelengths
combine to yield a broad net gain band. The long-wavelength pumps are launched with lower
power than the short-wavelength pumps to achieve comparable gain across the signal band.

make up for the energy difference between the signal and the pump and represent a
loss mechanism for pump waves. The loss of pump energy to vibrations of the host
material during the scattering process is accounted for by the factor −νp /νs in the
foregoing equations. For multiple pump and signal waves, Pp and Ps are summed
over all pump and signal waves. For waves with frequency lower than the signal
frequency under consideration, the terms of the sum include the energy loss factor.
For an accurate model useful for predicting the properties of Raman amplifiers, it
is necessary to include noise terms describing processes originating from spontaneous
Raman scattering (the source of ASE in Raman amplification) and Rayleigh scatter-
ing. In particular, the temperature dependence of spontaneous Raman scattering needs
to be included to accurately predict the noise figure of wideband Raman amplifiers.
Furthermore, dispersion must be accounted for when time-dependent waves (e.g.,
modulated signals) are considered. For quasicontinuous wave signals, phenomeno-
logical propagation terms in the coupled mode equations suffice [19].Amore complete
set of coupled mode equations then takes the form [20, 21]

∂Pj± (z) ±
1 ∂Pj (z)
± = −αj Pj + SαjR Pj∓
∂z Vg ∂t

+ N(k − j )(gPk Pj + 2hνj g[n(νk − νj ) + 1]Pk Bo )
k
νj
− N(j − k) (gPk Pj + 2hνk g[n(νk − νj ) + 1]Pk Bo ),
νk
456 M.N. Islam et al.

where Pj is the sum of the power in the forward direction Pj+ and the power in
the backward direction Pj− at frequency νj , Vg is the group velocity at νj , S is the
Rayleigh capture fraction, and αjR is the Rayleigh contribution to loss at νj . The Heavi-
side unit step function N(j ) is used here to distinguish between Stokes and anti-Stokes
waves, and n(ν) = (exp[hν/kb T ] − 1)−1 accounts for the temperature dependence of
ASE in bandwidth Bo . T is the temperature of the fiber, kB is Boltzmann’s constant,
and h is Planck’s constant.
A second consideration in the design of wideband Raman amplifiers concerns
thermal noise associated with the temperature dependence of spontaneous Raman
scattering. Thermal noise is most pronounced for signal waves that reside spectrally
close to pump waves and is closely related to the density of vibrational modes in the
final state of the Raman transition. Indeed, the probability of spontaneous emission
is proportional to the density of vibrational states, which at any temperature and fre-
quency shift is given by the Bose–Einstein distribution function, hence the functional
form of the thermal factor n(ν) + 1 in the coupled mode equations. Figure 14.5 shows
the Raman gain spectrum for silica and the thermal factor calculated at room tem-
perature. The rapid rise of the thermal factor as the frequency shift approaches zero
results in an increase in spontaneous emission near the pump wavelength [22], which
leads to a higher noise figure for signals with wavelength close to the pump wave-
length. The increase in noise for short-wavelength signals close to the pump band
increases with maximum gain corresponding to the longest wavelength pumps [23].
In wideband amplifiers, pump–pump stimulated Raman scattering can lead to signif-

6 1

Normalized Raman Gain Coefficient

0.8
Thermal Factor (dB)

4
0.6

0.4
2

0.2
1

0 0
0 5 10 15 20 25 30
Pump-Signal Frequency Separation (THz)

Fig. 14.5. Room-temperature thermal factor and Raman gain as a function of pump–signal
frequency separation.
 14. Wideband Raman Amplifiers 457

icant power at the longest pump wavelength, thereby significantly increasing thermal
noise. Note that the temperature dependence of Raman scattering only affects the
noise performance of Raman amplifiers; gain is largely temperature independent.
Multipath interference due to double-Rayleigh scattering (DRS) can also be im-
portant. Rayleigh scattering corresponds to an elastic scattering event that occurs
when propagating waves encounter subwavelength density variations in the fiber
core. The fiber core captures a fraction of the scattered radiation, and scattered waves
may undergo a second scattering event. The forward component of the twice-scattered
radiation, having been recaptured by the fiber core and constituting delayed copies of
the original signal, beats with the signal at the receiver and gives rise to inband noise.
Although the recaptured fraction is small, the Rayleigh scattered radiation is amplified
as it propagates along the fiber. Therefore, MPI noise due to DRS strongly depends
on amplifier properties: it increases with gain and fiber length [24] and dominates
noise performance for pump-on/pump-off gains in excess of about 15 dB. Because
the delayed signal copies reside in the same spectral band as the primary signal, DRS
crosstalk cannot be filtered out in the same manner as ASE and can lead to severe
degradation of the electrical noise figure and system quality factor Q. To ensure
Q penalties less than 1 dB20 , DRS crosstalk less than −15 dB is required, thereby
limiting the gain achievable in single-stage Raman amplifiers.
A further consideration concerns the short upper state lifetime associated with
Raman amplification, which leads to a number of requirements on pump lasers and
amplifier architecture. Because the response time associated with the Raman effect
is very short, on the order of femtoseconds, transfer of pump intensity fluctuations
to the signals can occur. The signal degradation that results depends on the spec-
trum of pump noise, fiber properties, and amplifier architecture [25]. In addition, the
signals also impress patterns on the pumps through gain saturation. The signal pat-
terns impressed on the pump waves constitute fluctuations of pump intensity that can
in turn be transferred to other signals, leading to pump-mediated crosstalk between
signal waves. As with pump–signal noise transfer, the penalty arising from pump-
mediated signal crosstalk depends on the noise frequency spectrum, fiber properties,
and pumping configuration [26].
To counter the deleterious effects of the short upper state lifetime, purely counter-
propagating (backward pumping, where the pump and signal waves are launched into
opposite ends of the fiber) geometry can be used. In this case, the transit time through
the fiber introduces an effective upper state lifetime over which pump fluctuations
are averaged. In cases where copropagation of pumps and signals (forward pumping)
is desired, pump lasers with very low relative intensity noise (RIN) are required to
minimize signal degradation from pump fluctuations. In both forward and backward
pumping, the gain fiber behaves as a low-pass filter, but with properties that depend
on the pumping configuration. For forward pumping, the 3 dB frequency of the filter
is inversely proportional to fiber dispersion and frequency shift and proportional to
fiber loss, whereas the 3 dB frequency is proportional to loss and group velocity for
backward pumping [25]. The cut-off frequency is generally higher for forward pump-
ing than for backward pumping, indicating that lower RIN pump lasers are required
for forward than for backward pumping to achieve low noise transfer penalties. In
458 M.N. Islam et al.

practical terms, laser diodes that are wavelength stabilized with an external cavity and
a fiber Bragg grating have a RIN of approximately −140 dB/Hz, and Fabry–Perot
lasers have a RIN of approximately −160 dB/Hz. Therefore, if forward pumping
is desired, Fabry–Perot laser diodes typically need to be used. Although less noisy,
Fabry–Perot lasers are not as stable in wavelength as grating-stabilized lasers.
A final consideration in the design of wideband Raman amplifiers is connected
with the location of the zero dispersion wavelength of the gain fiber relative to the
pump and signal bands. This is typically not a problem in DCF, but can pose a problem
in the DRA section that uses the transmission fiber as the gain medium because
four-wave-mixing can be phase-matched near the zero dispersion wavelength. Four-
wave-mixing leads to generation of sidebands, which have three effects: the sidebands
deplete the pump waves, effectively increasing pump attenuation; new sidebands on
the pump waves can lead to gain spectrum distortions, which may require additional
gain equalization; and the sidebands can fall within signal bands, leading to unwanted
crosstalk. The crosstalk arising from the third of these effects can lead to severe SNR
degradation. The magnitude of the crosstalk depends on the relative propagation
directions of pump and signal waves. In forward pumping, the sidebands copropagate
and directly interfere with signals. In backward pumping, the sidebands do not directly
interfere with signals. However, Rayleigh scattering reflects the interfering waves,
and the fiber core captures a fraction of the scattered sidebands propagating in the
signal direction. In both cases, Raman amplification enhances the crosstalk level, with
the result that severe SNR degradation can occur [27].

14.5. Design of Wideband Raman Amplifiers

This section describes design techniques to overcome the physical limitations de-
scribed in the previous section. As can be seen by the end of the section, each of
the limitations can be mitigated and relatively simple implementation of wideband
Raman amplifiers can be realized.

14.5.1. Electrical Noise Figure

As mentioned in the previous section, it is important to account for the electrical noise
figure as well as the optical noise figure in Raman amplifiers. The optical noise figure
primarily depends on signal–spontaneous emission beating, whereas additional ef-
fects such as MPI from DRS and pump–signal RIN transfer contribute to the electrical
noise figure. The electrical noise figure is, therefore, generally higher than the optical
noise figure if the additional effects contribute significantly to amplifier noise perfor-
mance. However, amplifier architectures exist that enable reduction of the additional
noise contributed by DRS and pump–signal RIN transfer.
Because of MPI penalties from DRS, wideband Raman amplifiers are typically
sectioned into multiple stages, where each stage is separated by an isolator. For an
amplifier comprising discrete and distributed stages, the MPI penalty is usually larger
in the discrete stages. This is because discrete stages are usually constructed of DCF,
 14. Wideband Raman Amplifiers 459

which has a higher gain coefficient and a higher DRS coefficient than transmission
fiber. Although the MPI penalty needs to be computed for each particular design, one
rule of thumb is that it is difficult to simultaneously obtain pump-on/pump-off gain
of more than 15 dB and DRS crosstalk of less than −15 dB (corresponding to a Q
penalty of less than 1 dB20 ) in a single stage. By using isolators between stages, the
buildup of DRS is minimized, and high-gain, low MPI amplifiers can be constructed.
Reduction in pump–signal RIN transfer and other noise sources associated with
the short upper state lifetime of Raman amplifiers is achieved by employing backward
pumping, which introduces an effective lifetime equal to the transit time through the
fiber. Because amplification mostly occurs near the signal egress end of the fiber
in backward pumping, the resulting noise figure is not optimal; however, backward
pumping also reduces dependence of amplifier gain on the state of polarization of the
signal. Where forward pumping is desired (as in bidirectional systems, e.g.), low-RIN
pump lasers need to be used.

14.5.2. Pump–Pump Interactions

Stimulated Raman scattering in the gain fiber leads to pump–pump power exchange,
which must be properly accounted for to obtain a desired gain profile. If equal fre-
quency spacing and equal amplitude pumps are used in a wideband Raman amplifier,
the resulting gain increases with wavelength (c.f. Fig. 14.6) [20]. Depending on the
power and bandwidth of the signal bands, there can also be transfer of energy from
short- to long-wavelength signals, further accentuating the gain tilt.
Several techniques can be used to achieve flat-gain spectra in the presence of
pump–pump interactions. The basic idea is to increase the spectral density at short

2
Fiber Gain, dB

–2

–4

–6
1520 1530 1540 1550 1560 1570 1580 1590 1600 1610 1620
Wavelength, nm

Fig. 14.6. Gain spectrum of a Raman amplifier employing a uniform pump power spectral
density [20]. Source: H. Kidorf et al: “Pump Interactions in a 100-nm Bandwidth Raman
Amplifier,” IEEE Photonics Technology Letters, Vol. 11, pg 530 (© 1999 IEEE)
460 M.N. Islam et al.

25 Pumps Signals

20

Launched Power, dBm

15

10

–5
1410
1425
1435
1445
1455
1465
1475
1485
1495
1505
1515
1525
1535
1545
1555
1565
1575
1585
1595
1605
1615
Wavelength, nm
(a)
5

4
Fiber Gain, dB

0
1520 1540 1560 1580 1600 1620
Wavelength, nm
(b)

Fig. 14.7. (a) Pump and signal spectra; and (b) gain spectrum of a 100 nm bandwidth Raman
amplifier with 1.1 dB gain ripple. The flat-gain spectrum is achieved by concentrating pump
power at short-wavelengths of the pump band [20]. Source: H. Kidorf et al: “Pump Interactions
in a 100-nm Bandwidth Raman Amplifier,” IEEE Photonics Technology Letters, Vol. 11, pg
530 (© 1999 IEEE)

pump wavelengths. If equally spaced pump frequencies are used, the power of long-
wavelength pumps can be reduced relative to the shorter wavelength pumps (pump
power preemphasis). Alternately, the pump frequency separation can be increased
to the long-wavelength side of the pump band. In a practical system, a combination
of these two techniques is actually used. Figure 14.7 illustrates the spectral density
adjustment for the case in Fig. 14.6 to obtain a flat-gain spectrum in a 100 nm amplifier
[20]. Although there is some decrease in pump power at long-wavelengths, in this
case the spectral density at the long-wavelength end of the pump band is primarily
reduced by spacing the long-wavelength pumps farther apart.
 14. Wideband Raman Amplifiers 461

Another completely different approach to reducing the effects of pump–pump

interaction is to arrange for the pumps not to encounter each other altogether. This
can be simply achieved by segregating the pumps in separate amplifier stages such that
strongly interacting pumps are launched in different stages [28]. Another approach
involves time modulation of the pumps. If the pumps are time modulated and then
multiplexed or timed so that they do not overlap, then they are not simultaneously in
the gain fiber and do not interact. A particular realization of this approach has been
demonstrated by Fludger et al. [29], where four pump wavelengths were separated into
two bands, each pair was modulated, and the relative phase was selected so that the
most strongly interacting bands did not coincide within the gain fiber. This scheme
reduces interaction between the pumps, thereby reducing depletion and increasing
penetration into the fiber of the short-wavelength pumps. The result is amplification
of short-wavelength signals deeper into the gain fiber and reduction of the short-
wavelength noise figure. The reverse occurs for long-wavelength signals, inasmuch
as reduction of interaction between the pump bands decreases the gain experienced
by the long-wavelength pumps as well as the distance along the fiber over which
long-wavelength signals experience gain.
Gain slope in a multistage configuration, as shown in Fig. 14.8 for a two-stage
amplifier, can also be used to reduce pump–pump interactions. In the example shown
in Fig. 14.8, gain in the first stage increases with wavelength and decreases with
wavelength in the second stage. The gain slopes in the two stages are complementary
so the net gain is substantially flat. Although noise performance may be degraded
(see next section), higher power conversion efficiencies may be realized by such a
scheme. This gain tilt allocation, in which gain increases with wavelength in the first
stage, allows use of fewer pump wavelengths and lower pump powers because of
energy transfer from short to long pump wavelengths through the Raman effect. The
long-wavelength pumps provide gain in the first stage, and then the shorter pump
wavelengths are introduced with more power in the second stage. In other words, by
selecting a pump wavelength distribution that leads to a sloped up gain followed by
a sloped down gain, the effect of pump–pump interactions is actually reduced.

Gain Profiles

Isolator Isolator

Signal (in) Signal (out)

1st Stage Pump(s) 2nd Stage Pump(s)

Fig. 14.8. Block diagram of a two-stage Raman amplifier with complementary gain spectra.
The gain of the first stage increases with wavelength, and that of the second stage decreases as
wavelength increases in such a manner that the composite gain shape is flat.
462 M.N. Islam et al.

14.5.3. Thermally Induced Phonon Noise

Thermal noise is particularly important in wideband Raman amplifiers. The peak

of the Raman gain curve at 13.2 THz corresponds to a separation of approximately
100 nm between pump and signals near 1550 nm. Thus, for a 100 nm bandwidth
amplifier with gain band near 1550 nm, the short-wavelength end of the signal band
is close to the long-wavelength pumps, and the contribution of thermal noise to the
short-wavelength noise figure can be substantial when the maximum gain contributed
by the long-wavelength pumps is large [23].
To mitigate the effect of the thermally induced noise, the short-wavelength signals
can be amplified before they encounter significant noise. In other words, if the signals
susceptible to thermal noise are amplified before encountering noise from the long-
wavelength pumps, the signal-to-noise ratio remains high throughout the amplifier
because the signal strength remains high. Because the most vulnerable signals are
those closest to the pumps in wavelength (i.e., the shortest wavelength signals), gain
for the shortest wavelength signals can be made high at the input of the amplifier.
However, as discussed earlier, there is also a need to equalize gain shape across the
signal band to obtain roughly equivalent behavior for all channels. A method for
accommodating both of these requirements is to use a multistage amplifier, each with
a gain slope.
The concept of using gain tilt in multiple stages to mitigate the noise limitation is
illustrated in Fig. 14.9, showing a two-stage Raman amplifier and gain profiles associ-
ated with each stage. Gain in the first stage decreases from short to long-wavelengths,
and the second stage has an approximately complementary gain spectrum. Thus short-
wavelength signals are substantially amplified in the first stage before encountering
the long-wavelength pumps responsible for thermal noise, and the net gain shape after
the two stages is substantially flat.
Signal preemphasis, discussed in Section 14.5.7, can also be used to mitigate
the effects of thermal noise. As with the gain tilt scheme previously described, the
strength of short-wavelength signals is made high to achieve a high signal-to-noise
ratio. Thus short-wavelength signals are launched with a higher power than long-

Gain Profiles

Isolator Isolator

Signal (in) Signal (out)

1st Stage Pump(s) 2nd Stage Pump(s)

Fig. 14.9. Block diagram of a two-stage Raman amplifier with complementary gain spectra.
The gain of the first stage decreases as wavelength increases, and that of the second stage
increases with wavelength in such a manner that the composite gain shape is flat.
 14. Wideband Raman Amplifiers 463

wavelength signals. In this manner, the signal-to-noise ratio at short-wavelengths can

be made comparable to that at long-wavelengths.

14.5.4. Zero Dispersion Wavelength

Four-wave-mixing impairments in Raman amplifiers are most pronounced in

dispersion-shifted fibers for which the zero dispersion wavelength falls within the
pump band or between the pump and signal bands. In fibers for which the disper-
sion wavelength falls within the pump band, four-wave-mixing between pump waves
can generate sidebands within the signal band and severely degrade OSNR for some
channels. In fibers for which the zero dispersion wavelength falls between the pump
and signal bands, nondegenerate four-wave-mixing between pump and signal waves
generates sidebands that fall in the signal band. In both cases, the power in the side-
bands depends on how close the longest wavelength pumps are to the zero dispersion
wavelength. This is because phase matching over long lengths of fiber, necessary
for significant four-wave-mixing to occur, is easily achieved when one of the pump
wavelengths is near to or coincides with the zero dispersion wavelength of the fiber.
Four-wave-mixing presents a crosstalk problem when the frequencies of the side-
bands coincide with channel frequencies. For example, in single-band transmission
using Raman amplification and only C- or L-band signals, pump–pump four-wave-
mixing sidebands usually have wavelengths too short to fall within the signal band.
On the other hand, in wideband systems where both C- and L-band transmission are
used, the longest wavelength pumps are typically close enough to the signal band
that four-wave-mixing sidebands fall within the signal band and can interfere with
some channels [27]. In that case, several methods can be used to mitigate the effects
of four-wave-mixing. One technique is simply to delete those channels that coincide
with sidebands, although that approach reduces overall capacity. Four-wave-mixing
efficiency is strongly dependent on phase matching, therefore employing fiber with
high local dispersion tends to reduce the crosstalk level. However, in many cases,
the DRA section within which four-wave-mixing is troublesome employs preexist-
ing transmission fiber, and the associated dispersion properties are predetermined.
Increasing pump frequency spacing, particularly for pump wavelengths close to the
zero dispersion wavelength, can also be used. Increasing pump frequency separation
increases dispersive walk-off between the pump waves, reducing phase matching
and four-wave-mixing efficiency. In addition, reducing the power of long-wavelength
pumps is also effective. Four-wave-mixing efficiency is at least a quadratic function of
pump intensity. Raman amplification further increases efficiency so the functional de-
pendence of sideband intensity on pump intensity can be stronger than quadratic [30].
Therefore, decreasing pump power can significantly reduce four-wave-mixing. In re-
ducing the power at long pump wavelengths, gain and gain flatness can be kept
unchanged by maintaining the spectral power density of the long-wavelength pumps.
Another approach to mitigation of four-wave-mixing employs pump amplitude
modulation to prevent interaction between the pumps involved in pump–pump four-
wave-mixing. The offending pumps are time modulated and multiplexed in such a
manner that different pump wavelengths occupy separate time slots. For example,
464 M.N. Islam et al.

–5

–10

–15
Power (dBm) –20

–25 X

–30

–35

–40

–45
1500 1510 1520 1530 1540 1550 1560 1570 1580 1590 1600 1610 1620
Wavelength (nm)

Fig. 14.10. Signal spectra with (black curve) and without (gray curve) pump modulation.
The marked band on the spectrum obtained without pump modulation is a four-wave-mixing
sideband arising from pump–pump interactions. The sideband is suppressed in the modulated
case [29]. Source: C.R.S. Fludger et al: “Ultra-broadband high performance distributed Raman
amplifier employing pump modulation,” OFC Technical Digest Postconference Edition Vol.
70, pg 183 (© 2002 OSA)

Fig. 14.10 depicts the signal spectrum for modulated and nonmodulated pump waves
[29]. Pump–pump four-wave-mixing sidebands appear near the short-wavelength end
of the spectrum in the nonmodulated pump case, shown in gray on the figure. The
sidebands disappear when the pumps are time modulated and a phase delay of π is
introduced between interacting pumps.

14.5.5. Pump Power Adjustment to Control Gain Shape

The behavior of Raman amplifiers differs significantly from rare earth-doped ampli-
fiers such as EDFAs with regard to the change in pump power and the corresponding
noise figure shape. In an EDFA, changing the pump power has the effect of changing
the level of inversion. The noise figure in an EDFA is minimum for a completely
inverted amplifier, and the noise figure increases as the level of inversion is decreased
(i.e., the spontaneous emission factor increases). On the other hand, in a Raman ampli-
fier the gain shape and noise figure shape as a function of frequency can be controlled
by the pump powers in the various pump wavelengths.
Management of Raman gain tilt is required in almost all wideband amplified sys-
tems. Although present in any DWDM system, Raman gain tilt is most problematic
in systems spanning near or above 100 nm of gain bandwidth. This tilt management
is less complicated with an all-Raman system as monitoring of the signals can occur
across the entire band at just two positions: immediately before and directly after
amplification (Fig. 14.11). Dynamic adjustment of any measured tilt (or other gain
affecting event, such as channel drops or additions) can be made through the straight-
 14. Wideband Raman Amplifiers 465

Distributed Raman
Gain Medium
(Transmission Discrete Raman
Span Fiber) Gain Medium

Adjustable Pumps

Monitor
&
Control

Fig. 14.11. Raman amplifier configuration including monitoring and control functionality.

forward adjustment of the Raman pump power levels. The very nature of Raman
amplification permits this ability to “shape” the gain spectrum. Monitoring and ad-
justment of the banded system is more intricate, as a number of monitoring points
need to be used together with the different gain adjustments for the discrete amplifiers.
This is further complicated by the fact that the greatest amount of Raman gain tilt
will be most prevalent between the bands (intraband) and careful coordination of all
adjustments, regardless of band, needs to take place.
Figure 14.12 illustrates gain tilt control obtained simply by varying pump power
levels in a Raman amplifier pumped with 12 wavelengths [31]. Decreasing the power
of the low-frequency pumps and adjusting the other pump powers results in the tilted
gain shapes shown in the figure. Another example of gain tilt control is depicted in
Fig. 14.13 [18]. The figure shows results of simulations for flat (curve (a)), linear
(curve (b)), and nonlinear (curve (c)) gain profiles in a Raman amplifier with a band-
width of 35 nm. All gain profiles shown are achieved simply by varying the pump
power levels. A further example of gain tilt control through pump power adjustment
is given in Chapter 10, Section 10.5.

14.5.6. Signal Preemphasis to Equalize Channel Performance

In designing multispan amplified transmission links, it is usual to select signal launch

powers that assure that the worst performing channel achieves a target SNR dictated by
bit error rate requirements. The noise figure of EDFAs is spectrally flat to a large extent.
Therefore, in an EDFA-based system with negligible gain ripple, a uniform signal
power spectrum yields comparable SNR for all channels. Where gain ripple is not
negligible but accumulated gain variations are low (up to about 10 dB), preemphasis
of the input signal spectrum has been used to obtain uniform SNR at the output of
point-to-point transmission systems [32]. Total input power is redistributed among
the WDM channels in such a manner as to maintain the composite output power and
466 M.N. Islam et al.

12.0

10.0

Raman gain [dB] 8.0

6.0

4.0

2.0

0.0
1520 1540 1560 1580 1600 1620
Wavelength [nm]

Fig. 14.12. Gain tilt control by pump power adjustment in a Raman amplifier comprising 25 km
long dispersion-shifted fiber and a 12-wavelength pump module [31]. Source: S. Namiki and
Y. Emori: “Ultrabroad-band Raman Amplifiers Pumped and Gain-Equalized by Wavelength-
Division-Multiplexed High-Power Laser Diodes,” IEEE Journal of Select Topics in Quantum
Electronics, Vol 7, pg 3 (© 2001 IEEE)

10
c

8
a

6
Gain (dB)

b
4

0
1520 1530 1540 1550 1560 1570
Wavelength (nm)

Fig. 14.13. (a) Flat, (b) tilted linear, and (c) nonlinear gain profiles obtained by pump power
control [18].
 14. Wideband Raman Amplifiers 467

to yield equal SNR for all channels. The SNR of the worst channels is improved, and
a uniform bit error rate performance is obtained across the entire signal band.
The noise figure in Raman amplifiers is wavelength-dependent even when gain
ripple is negligible. Noise figure variations in Raman amplifiers arise from a num-
ber of sources. Stimulated Raman scattering transfers energy from short- to long-
wavelength signals, constitutes an additional loss mechanism at short-wavelengths,
and results in a larger noise figure at short-wavelengths than at long-wavelengths.
In addition, thermal noise in spectral bands close to pump wavelengths leads to a
higher noise figure for short-wavelength signals. Furthermore, the quartic frequency-
dependence of Rayleigh scattering results in a higher double-Rayleigh crosstalk for
short-wavelength signals than for longer wavelength signals and a correspondingly
higher electrical noise figure. Therefore, a uniform signal spectrum does not yield
comparable SNR performance for all channels even when gain variations are neg-
ligible. Although power variations at the output may be small, there may still be
significant variation of SNR of the channels because of the noise figure variations.
A nonuniform SNR at the output translates into substantial bit error rate variation
among the channels.
Input signal preemphasis can again be used to equalize channel performance in
Raman amplifiers. However, unlike the case of EDFA-based systems where preem-
phasis is applied in such a manner as to maintain the composite signal power at a
prescribed level, Raman amplification permits preemphasis to be applied to lower the
composite signal power while equalizing output SNR for all channels. The launch
power for channels with the worst noise figure is selected to achieve the target SNR
as is usually done in systems with uniform signal launch power. A higher output SNR
would be then obtained for channels with better noise figure if equal launch power
were used for all channels. To achieve equal output SNR, the launch power of chan-
nels with lower noise figure is reduced. The pump powers may need to be adjusted to
maintain the gain of the amplifier. An iterative adjustment of signal and pump powers
results in a signal spectrum that mirrors the noise figure spectrum and yields a flat
SNR spectrum. Depending on the saturation level of the Raman amplifier, reduction
of the composite signal power through preemphasis of launch signal powers based
on the magnitude of the corresponding noise figure can reduce the total pump power.
As an illustration, consider a simulation of a two-stage Raman amplifier compris-
ing an 80 km long distributed section and a dispersion-compensating discrete section.
Net gain through the two stages is nominally 0 dB with a flat-gain spectrum over a
bandwidth of 100 nm in each stage. Figure 14.14 depicts the signal and Fig. 14.15
the OSNR spectra for launch signal powers preemphasized according to the shape
of the noise figure spectrum. Also shown in the figures are spectra for a three-stage
Raman amplifier comprising an 80 km long distributed section and a two-stage dis-
crete section using 0 dBm and −2.26 dBm per channel (no preemphasis). Net gain is
again nominally 0 dB over 100 nm of bandwidth. Using complementary gain slopes
as described in Section 14.5.3, a nearly flat OSNR is obtained across the entire 100 nm
wide spectrum, as shown in Fig. 14.15 for launch powers of 0 dBm and −2.26 dBm
per channel. However, OSNR variations greater than 2 dB remain. The OSNR spec-
trum obtained for input signal preemphasis applied in conformity to the noise figure
468 M.N. Islam et al.

Pre-emphasis
0 dBm/channel
–2.26 dBm /channel

Signal Power per Channel (dBm) 0

–2

–4

–6

–8
1520 1540 1560 1580 1600
Wavelength (nm)

Fig. 14.14. Signal power spectra for Raman amplifiers with (solid curve) and without (dashed
curves) input signal preemphasis. The shape of the input signal power spectrum for the preem-
phasized case is that of the noise figure of the amplifier.

spectrum is substantially flat compared to the case of the uniform signal spectrum, as
attested by the spectra in Fig. 14.15. The small slope of the OSNR spectrum evident in
the figure for the preemphasized case arises from neglecting a wavelength-dependent
factor in the SNR calculation.

14.5.7. Other Enhancements

The foregoing techniques for wideband Raman amplification are by no means ex-
haustive, and more features and techniques can be expected to emerge as research
and development efforts continue to be directed at Raman amplification. A number
of developments in all-Raman amplification that have not been treated in detail here
include: forward pumping, higher-order pumping, and time-multiplexed or frequency-
swept pumping.
Most Raman amplifiers use a counterpropagating pump configuration to avoid
coupling of pump fluctuations to the signal. In addition, by using the counterprop-
agating geometry, other effects such as pump-mediated crosstalk, polarization de-
pendence of gain, and four-wave-mixing between pumps are also reduced. However,
forward pumping improves the noise figure and enables implementation of Raman
amplified bidirectional systems. Forward pumping improves the noise figure because
the signal is amplified at the signal input end of the fiber before it is significantly
 14. Wideband Raman Amplifiers 469

SNR @ pre-emphasis
SNR @ 0 dBm/channel
SNR @ –2.26 dBm/channel
39

38
Signal-to-Noise Ratio (dB)

37

36

35

34

33
1520 1540 1560 1580 1600
Wavelength (nm)

Fig. 14.15. Output optical signal-to-noise ratios for Raman amplifiers with (solid curve) and
without (dashed curves) signal preemphasis.

attenuated, thereby maintaining a high signal-to-noise ratio throughout the fiber. As

mentioned earlier, forward pumping requires the use of low-RIN pump lasers such as
Fabry–Perot laser diodes. More detail on forward pumping is provided in Chapter 9,
and some of the effects of the fast gain dynamics are covered in Chapter 8.
One way to obtain some of the benefits of forward pumping without the direct
coupling of the pump fluctuations to the signal is to use so-called higher-order pump-
ing. In higher-order pumping, pump waves at one or more Raman frequency shift
above the primary pump frequency are used. The higher-order pumps amplify the
primary pump waves and can reduce the amplifier noise figure. A copropagating
higher-order pump can be used to pump a counterpropagating pump, thereby ampli-
fying the counterpropagating pump at the signal input end of the fiber. Thus the signal
gain is increased at the beginning of the fiber. In general, the copropagating pump is
of shorter wavelength than the counterpropagating pump, and the counterpropagating
pump is primarily responsible for the gain to the signal wavelengths. This approach
has several limitations, however. First, the process tends to be somewhat inefficient,
because one pump amplifies the other pump before it supplies energy to the signal.
Also, the pumps are at the two ends of the fiber, and are therefore maximally attenu-
ated by the propagation loss. Second, the copropagating pump can still provide gain
to the signal directly through the high-frequency features on the gain spectrum (c.f.,
Fig. 1.1 in Chapter 1 for frequencies above 13 THz), and pump fluctuations can still
470 M.N. Islam et al.

couple to the signal. A more detailed treatment of higher-order pumping appears in

Chapter 9.
Finally, time-multiplexed or frequency-swept pumping can be used to solve a
number of problems in wideband Raman amplifiers. In time-multiplexed pumping
pump waves are modulated in time and made to either overlap or not overlap in the
fiber. Time-multiplexed pumping can be used to reduce pump–pump interactions or
to avoid four-wave-mixing between different pump wavelengths, as discussed earlier
in this chapter. The idea of frequency-swept pumping is to use a tunable wavelength
pump laser and to vary the frequency of the laser. This may have several benefits
including: (1) suppression of four-wave-mixing between pumps; (2) suppression of
Raman interaction between pumps; and (3) very precise control of the gain profile. De-
tails of both time-multiplexed and frequency-swept pumping are given in Chapter 3.

14.6. Examples of Wideband Raman Amplifiers

Numerous experiments have demonstrated wideband Raman amplification with band-
widths of 100 nm or more. We review some of the results on wideband Raman am-
plifiers in this section. Early work on wideband Raman amplifiers recognized the
need to overcome pump–pump interactions. Kidorf et al. [20] demonstrated a Raman
amplifier with 100 nm bandwidth (with the gain band between 1520 and 1620 nm)
by varying the pump frequency spacing and pump power to achieve a high pump
spectral density near the short-wavelength end of the pump band (Fig. 14.7). Fludger
et al. [29] reduced the pump–pump interactions by time-multiplexing the strongly
interacting pumps, by this means demonstrating a 100 nm amplifier with a gain band
between 1510 and 1610 nm.
For a fixed bandwidth, the gain spectrum of a wideband Raman amplifier becomes
increasingly flat as the number of pump wavelengths increases. An example of using
closely spaced pumps is provided by Namiki and Emori [31], who demonstrate gain
flatness of about 0.1 dB over the wavelength range between 1527 and 1607 nm in
25 km of dispersion-shifted fiber by using 12 pump wavelengths (Chapter 1, Fig. 1.6).
A few pump wavelengths together with gain-flattening filters can also be used to
achieve a flat-gain shape. The trade-off to be considered in electing to use a large
number of pump wavelengths or a smaller number together with gain-flattening filters
concerns cost, ease of implementation, and flexibility. For example, a gain flatness
similar to that reported in [31] can be achieved with five or six pump wavelengths
and a gain-flattening filter. Although a large number of pump wavelengths permits
adjustment of gain shape in response to changes in channel loading or other dynamic
effects, using fewer pump lasers and a gain-flattening filter may be cost effective
because amplifier cost is often dominated by pump cost.
Although commercially available wideband Raman amplifiers have been limited
to a bandwidth of 100 nm to date, laboratory experiments have demonstrated ampli-
fiers with much greater bandwidths. For example, Naito et al. [33] have demonstrated a
Raman amplifier with a bandwidth of 136.6 nm by using a pump and signal wavelength
interleaving technique. The technique intermixes a few of the pump wavelengths into
the signal bands with a guard band surrounding the pumps. The broadband DRA
 14. Wideband Raman Amplifiers 471

Power Raman amplification

13.2THz

Pump 4.4THz
WDM signals
S+ S C L L+
Wavelength
1409 1439 1470 1502 1536 1571 1608 1647 (nm)

Optical
+D –D circulator
fiber fiber Rejection
filter
Raman
amplification Pump LDs
(a)

20
Raman gain (dB)

15

10

0
1490 1530 1570 1610 1650
Wavelength (nm)
(b)

Fig. 14.16. (a) Schematic spectrum showing pump and signal wavelength allocation (upper
diagram) and experimental configuration (lower diagram) for pump–signal wavelength inter-
leaved Raman amplification; (b) gain profile of single-stage DRA [33]. Source: T. Naito et al:
“A broadband distributed Raman amplifier for bandwidths beyond 100 nm” OFC Technical
Digest, Postconference Edition Vol. 70, page 116 (© 2002 OSA)

scheme is shown in Fig. 14.16(a), where the signal and pump wavelength allocation
is shown as well as the circulator that is used to inject the pump waves. The gain
fiber consists of 28 km of positive dispersion fiber and 12 km of negative dispersion
fiber. The pump wavelengths used are 1409, 1439, 1470, 1502, and 1536 nm. The
total pump power is set at 480 mW. Figure 14.16(b) shows the Raman gain profile for
a single-stage DRA. The transmission bandwidth is 3.7 nm in the S+ band between
1496 and 1499.7 nm, 30.1 nm in the S-band between 1504.2 and 1534.3 nm, and
102.8 nm in the C- and L-bands between 1537.2 and 1640 nm. The sum of WDM
signal bandwidths is 136.6 nm, and the total width of guard bands is 7.4 nm.
Mori et al. [16] have studied Raman amplification in tellurite fiber. They show that
tellurite fiber, a nonsilica fiber, has a large intrinsic bandwidth (Stokes shift of gain
472 M.N. Islam et al.

tellurite-based fiber
signal in signal out
module (250nm)

60 Tellurite

gain coefficient, 1/Wkm gain coefficient, 1/Wkm

50
40 1430nm 1410nm 1420nm 1430nm
30
: isolator : LD : WD coupler
20
: polarisation beam combiner : circulator
170nm
10 941/2
25 30
0
4 Silica 20 25

noise figure, dB
3 15 20

gain, dB
10 15
2
5 160nm (1490–1650nm) 10
1
100nm 0 5
0 –5 0
0 50 100 150 200 250 300 1480 1520 1560 1600 1640
Raman Strokes shift, nm
, nm
941/1 941/3
(a) (b)

Fig. 14.17. (a) Raman gain coefficient spectra of tellurite fiber and silica fiber; (b) Raman
amplification in tellurite fiber: the upper diagram shows the amplifier configuration; the lower
diagram shows the small-signal gain and noise figure spectra of the amplifier [16]. Source: A.
Mori et al “Ultra-Wideband Tellurite-based Raman Amplifiers” IEEE Electronics Letters, Vol.
37, pg 1442 (© 2001 IEEE)

peak is approximately 170 nm), a large nonlinearity, and a Raman gain coefficient
about 16 times that of fused silica. These properties combine to yield a large bandwidth
in short lengths of tellurite gain fiber. Experiments with four pump wavelengths
yielded a bandwidth of 160 nm (Fig. 14.17) [16]. Figure 14.17 shows the Raman
gain spectrum in tellurite fiber along with that of silica fiber. It can be seen that the
gain spectrum of tellurite fiber includes two prominent peaks that straddle the main
peak in the spectrum for silica. This fortuitous conjunction of spectral maxima and
minima enables construction of flat-gain amplifiers by cascading tellurite and silica
fiber amplifier stages. In experiments discussed next, a bandwidth of 120 nm could
be obtained using silica and tellurite gain fibers.
Takara et al. [34] demonstrated a multistage all-Raman amplifier with a 3.5 dB
gain bandwidth of 120 nm (between 1485 and 1605 nm) and equivalent noise fig-
ure between 4 to 10 dB within that range. The multistage amplifier configuration,
consisting of a DRA in standard single-mode fiber followed by a three-stage discrete
amplifier in silica and tellurite fiber, is shown in Fig. 14.18 along with gain and noise
figure spectra. In particular, the large bandwidth is achieved by using lumped Raman
amplification in tellurite fiber, lumped Raman amplification in DCF, distributed Ra-
man amplification in the transmission fiber, and gain equalization filters. Using this
amplifier configuration, 313 channels at 10 Gb/s are transmitted over a distance of
160 km.
 14. Wideband Raman Amplifiers 473

Distributed
Amp. Section Discrete Amp. Section
1st Stage 2nd Stage 3rd Stage Px LD Pump
Light Source
Isolator
GEQ
SMF Coupler
P4 Tellurite P1 N-DCF P2 Tellurite P3
80 km Fiber-1 10km Fiber-2
150m 150m

20 40
80 km fiber loss

Net NF, Equivelent NF (dB)

15
30
Gain, Loss (dB)

Net NF
10
20
5

Gross gain 10 Equivelent NF

0
Discrete gain Distributed gain
–5 0
1480 1520 1560 1600 1480 1520 1560 1600
Wavelength (nm) Wavelength (nm)
(a) gain (b) Noise figure

Fig. 14.18. Configuration, gain spectrum, and noise figure spectrum of inline hybrid Raman
amplifier (N-DCF: negative-slope dispersion-compensating fiber. GEQ: gain equalizer) [34].
Source: H. Takahara et al: “Ultra-wideband tellurite/silica fiber Raman amplifier and super-
continuum lightwave source for 124-nm seamless bandwidth DWDM transmission,” OFC
Technical Digest, Postconference Edition Vol. 70, pg FB1-1 (© 2002 OSA)

14.7. Wideband Raman Amplified Systems

Raman amplification has been successfully used in high-performance transmission
systems in conjunction with EDFAs and other high-performance techniques to ex-
tend the reach and/or capacity of fiber-optic transmission systems. High-performance
transmission systems utilizing all-Raman amplification have also been demonstrated.
The wide bandwidth, low noise figure, and distributed gain of Raman amplification
are critical to achieving high performance.
As an example of bandwidth and reach extension using Raman amplification,
Foursa et al. [35] demonstrated 2.56 Tb/s (256 channels at 10 Gb/s per channel)
transmission over 11,000 km using hybrid Raman/EDFAs with 80 nm of continu-
ous bandwidth. There have been several earlier reports of bandwidth extension by
addition of L-band EDFAs in parallel with C-band EDFAs. Parallel C- and L-band
EDFAs are limited by inferior noise performance compared to C-band EDFAs be-
cause of additional loss from couplers used to split and recombine the subband signals.
These limitations were overcome in [35] by using a hybrid amplifier comprising an
erbium-doped fluoride fiber amplifier in series with a Raman amplifier (R-EDFA)
that provided a continuous bandwidth over the C and L. Ultra-long-distance trans-
mission was achieved by using state-of-the-art gain equalization techniques as well
as concatenated Reed–Solomon forward error correction (FEC). The configuration
474 M.N. Islam et al.

4
980 nm
Circulator 0
WDM Er-fiber ISO GFF Tap

Gain, dB
D+ D–
In Out –4
–8
PBS
–12
1497 nm 1520 1540 1560 1580 1600 1620
LD
Wavelength, nm
(a) (b)

Fig. 14.19. (a) Amplifier configuration and (b) gain spectra of Raman (circles) and EDFA (trian-
gles) sections of a hybrid DRA/EDFA [35]. Source: D.G. Foursa: “2.56 Tb/s (256/spl times/10
Gb/s) transmission over 11,000 km using hybrid Raman/EDFAs with 80 nm of continuous
bandwidth,” OFC Technical Digest, Postconference Edition Vol. 70, pg FC3-1 (© 2002 OSA)

2.5 32
2 30

SNR, dB
Gain, dB

1.5 28
1
0.5 26
0 24
–0.5 22
1520 1540 1560 1580 1600 1620 1520 1540 1560 1580 1600 1620
Wavelength, nm Wavelength, nm
(a) (b)

Fig. 14.20. (a) Hybrid Raman-EDFA gain shapes showing equalized (solid line) and unequal-
ized (diamonds) gain spectra; (b) measured SNR of a chain of 13 hybrid Raman/EDFAs [35].
Source: D.G. Foursa: “2.56 Tb/s (256/spl times/10 Gb/s) transmission over 11,000 km using
hybrid Raman/EDFAs with 80 nm of continuous bandwidth,” OFC Technical Digest, Postcon-
ference Edition Vol. 70, pg FC3-1 (© 2002 OSA)

for the R-EDFA is illustrated in Fig. 14.19(a), which shows unpolarized backward
pumping in a positive dispersion/negative dispersion (+D/ − D) fiber combination.
Figure 14.19(b) shows that the gain spectra of the Raman and EDFA sections comple-
ment each other and provide a wide continuous bandwidth with minimum gain ripple.
Typical unequalized and equalized gain shapes are shown in Fig. 14.20(a). To evalu-
ate the performance of the amplifier, a 525 km chain of 13 R-EDFAs was constructed.
The SNR measured at the output end of the chain is illustrated in Fig. 14.20(b).
In another example, transmission of 10.2 Tb/s (256 channels at 42.7 Gb/s per chan-
nel) over three 100 km spans of nonzero dispersion-shifted fiber was demonstrated
by Frignac et al. [36]. A number of techniques were used to achieve this ultra-high-
capacity transmission. First, hybrid Raman/EDFA amplifiers with improved noise
performance achieved by second-order pumping were used. Second, polarization-
multiplexing was used: two sets of 128 WDM channels were multiplexed in the
polarization domain (with adjacent channels orthogonally polarized to reduce inter-
action or crosstalk between channels). Third, as is becoming more common in almost
all system experiments, FEC was used to improve the BER at the output. Finally,
vestigial sideband (VSB) coding was used, which in combination with polarization-
 14. Wideband Raman Amplifiers 475

1x
C 32 C
1x
M-Z
L 32 L Q 3 dB
42.6 Gb/s 100 km
23
1x 2 –1 FEC DCF TeraLight™
C 32 C Q DCF DCF DCF
PBS Polar.
1x
M-Z
L 32 L demux.
C C C C C C C C
Rx
1x
C 32 C L L L DCF L L
DCF
L L L
M-Z PBS
1x
L 32 L Q DCF
42.6 Gb/s DCF
31
1x 2 –1 FEC
C 32 C Q 3 dB
1x
M-Z
L 32 L

Fig. 14.21. Experimental configuration for transmission of 256 × 42.7 Gb/s channels over
three 100 km laps of nonzero dispersion-shifted fiber [36]. Source: Y. Frignac et al “Transmis-
sion of 256 wavelength- division and polarization-division-multiplexed channels at 42.7Gb/s
(10.2Tb/s capacity) over 3/spl times/100km of TeraLight/spl trade/ fiber,” OFC Technical Di-
gest, Postconference Edition Vol. 70, pg FC5-1 (© 2002 OSA)

Ref. with FEC

C-band L-band
10 –4 10 –15
BER

10 –6
Parallel polarisation
10 –8
10 –4 10 –15
BER

10 –6
orthogonal polarisation
10 –8
31
(in 0.1nm)

29
SNR

27
25
1525 1535 1545 1555 1565 1575 1585 1595 1605
Wavelength (nm)

Fig. 14.22. Summary of experimental results for transmission of 256 × 42.7 Gb/s channels over
three laps of 100 km nonzero dispersion-shifted fiber showing bit error rate spectra for parallel
(top) and orthogonal (center) polarized signals and SNR at output after FEC (bottom) [36].
Source: Y. Frignac et al “Transmission of 256 wavelength- division and polarization-division-
multiplexed channels at 42.7Gb/s (10.2Tb/s capacity) over 3/spl times/100km of TeraLight/spl
trade/ fiber,” OFC Technical Digest, Postconference Edition Vol. 70, pg FC5-1 (© 2002 OSA)

multiplexing allowed a high spectral density of 0.64 bps/Hz to be realized. The exper-
imental configuration is shown in Fig. 14.21, where the WDM transmitter consists of
two sets of 128 DFB lasers covering the C- and L-bands.As shown in the figure, a split-
band configuration is used for the amplifiers, with 8 nm of guard band centered around
1565 nm between the C- and L-bands. The experimental results are summarized in
Fig. 14.22. The top curves correspond to the bit error rate (BER) for parallel and or-
thogonal polarizations, and the bottom curve shows the SNR at the output after FEC.
476 M.N. Islam et al.

Clock (21.3G) 42.7-Gb/s NRZ

1
EDFA
A Precompensation
W
G
Polarizer
63 Clock (21.3G) 42.7-Gb/s NRZ
2 Clock (21.3G) 42.7-Gb/s NRZ DCM GEF
Switch Switch
A
W Interleaver
G 1
64 Clock (21.3G) 42.7-Gb/s NRZ A 3-dB
W Coupler
Clock G
Variable
Recovery Postcomp DCM 63 x5
2 100km NZDSF
A
Balanced W DCM
RX/Dmux G
MZDI Turnable BPF 64 De-interleaver Raman Raman

Fig. 14.23. Experimental setup for transmission of 64 × 42.7 Gb/s channels over 40 laps of
100 km nonzero dispersion-shifted fiber [37]. Source: A.H. Gnauck et al: “2.5 Tb/s (64/spl
times/42.7 Gb/s) transmission over 40/spl times/100 km NZDSF using RZ-DPSK format and ’
all-Raman-amplified spans,” OFC Technical Digest, Postconference Edition Vol. 70, pg FC2-1
(© 2002 OSA)

14.7.1. All-Raman Amplified Systems

The foregoing examples illustrate amplified systems that employ hybrid Raman/
EDFAs for capacity extension. Numerous high-performance experiments that use
all-Raman amplification have also been reported. In one example, Gnauck et al. [37]
reported transmission of 2.5 Tb/s (64 channels at 42.7 Gb/s per channel) aggregate
capacity over 4000 km (40 × 100 km spans) of nonzero dispersion-shifted fiber in a
loop-back experiment. This performance was achieved in a single 53 nm extended
L-band using return-to-zero differential phase shift keyed modulation, balanced de-
tection, distributed Raman amplification, and forward error correction. In the return-
to-zero differential phase shift keyed format, an optical pulse appeared in each time
slot, with the binary data encoded as either a 0 or π phase shift between adjacent bits.
In principle, this coding scheme can reduce the SNR required to achieve a given BER
by 3 dB, thereby increasing system margin. The experimental setup for the loop-back
experiment is shown in Fig. 14.23. The amplifier is a DRA followed by a discrete Ra-
man amplifier implemented in a dispersion-compensating module. The experimental
results are shown in Fig. 14.24, where the SNR without FEC of the 64 channels is
shown. After FEC, an uncorrected BER of 2.4 × 10−4 corresponds to a corrected
BER of less than 10−12 .
In another experiment, Sugahara et al. [38] demonstrated transmission of 32 ×
42.7 Gb/s DWDM signals over a distance of 6050 km using a Raman-amplified
quadruple-hybrid span configuration. Transmission of the DWDM signals, which
were spaced by 100 GHz, was accomplished through suppression of fiber nonlin-
earity effects by distributed all-Raman amplification in the quadruple-hybrid span
configuration of Fig. 14.25. The dispersion map consisted of lengths of low-loss,
pure silica core fibers (PSCFs) interleaved with DCF. The experimental configura-
 14. Wideband Raman Amplifiers 477

1.E–02 21

Received OSNR (dB)

1.E–03 17

Uncorrected BER
1.E–04 13

1.E–05 9
Uncorrected BER at 4000 km
1.E–06 FEC Threshold (maps to < 1E–12) 5
Received OSNR at 4000 km
1.E–07 1
1554 1563 1572 1581 1590 1599 1608
Wavelength (nm)

Fig. 14.24. Experimental results for transmission of 64 × 42.7 Gb/s channels over 40 laps of
100 km nonzero dispersion-shifted fiber. Optical signal-to-noise ratio (triangles) and uncor-
rected bit error rate (circles) are shown [37]. Source: A.H. Gnauck et al: “2.5 Tb/s (64/spl
times/42.7 Gb/s) transmission over 40/spl times/100 km NZDSF using RZ-DPSK format and
all- Raman-amplified spans,” OFC Technical Digest, Postconference Edition Vol. 70, pg FC2-1
(© 2002 OSA)

Raman PSCF
Signal pump
Accumulated

PSCF DCF
Dispersion

DCF

Distance
–5.0 dBm/ch
–5.5 dBm/ch
Optical
power

(a) Single-hybrid (b) Quadruple-hybrid

Fig. 14.25. Dispersion maps and power profiles for (a) single hybrid and (b) quadruple-hybrid
span configurations (PSCF: pure silica core fiber; DCF: dispersion-compensating fiber) [38].
Source: H. Sugahara et al: “6,050km transmission of 32 /spl times/ 42.7 Gb/s DWDM signals
using Raman-amplified quadruple-hybrid span configuration,” OFC Technical Digest, Post-
conference Edition Vol. 70, pg FC6-1 (© 2002 OSA)
478 M.N. Islam et al.

odd 16 channels Pre DCF Post DCF

1

3 21.3 Gb/s
LN LN SW1 AWG EA
Receiver
Data Data CLK

31 21.3 GHz 42.7 Gb/s
CLK
23
2 –1 PRBS CLK Error

2 1ns delay Extraction
Data Detector

4
LN LN
Raman
Pump

32
even 16 channels
GEQ SW2
Span 1
Loop length = 208.8 km
DCF

Span 4 Span 3 Span 2

Raman Raman Raman
Pump Pump Pump

Fig. 14.26. Experimental configuration for Raman-amplified transmission using quadruple-

hybrid spans [38]. Source: H. Sugahara et al: “6,050km transmission of 32 /spl times/ 42.7 Gb/s
DWDM signals using Raman-amplified quadruple-hybrid span configuration,” OFC Technical
Digest, Postconference Edition Vol. 70, pg FC6-1 (© 2002 OSA)

14

13

12
Q value [dB]

11

Corrected BER = 3  10 –17

10

8
1535 1540 1545 1550 1555 1560 1565 1570
Wavelength [nm]

Fig. 14.27. Measured Q-values after 6050 km transmission on quadruple-hybrid spans [38].
Source: H. Sugahara et al: “6,050km transmission of 32 /spl times/ 42.7 Gb/s DWDM signals
using Raman-amplified quadruple-hybrid span configuration,” OFC Technical Digest, Post-
conference Edition Vol. 70, pg FC6-1 (© 2002 OSA)

tion is illustrated in Fig. 14.26, showing the distributed scheme used. Figure 14.27
shows the measured Q-values after 6050 km transmission. The Q-values are all above
11.7 dB, corresponding to a BER below 3 × 10−17 with FEC.
As another example, a 1.6 Tb/s system (40 × 42.7 Gb/s) transmission over
3600 km of dispersion-managed fiber was accomplished using all-Raman amplifi-
cation in 100 km terrestrial spans using a loop-back configuration [39]. The long
transmission distance and high capacity were achieved with a carrier-suppressed
 14. Wideband Raman Amplifiers 479

SLA-IDF-SLA SLA-IDF-SLA
RP RP RP RP

AWG
EDMUX
10G FEC 43G Post-comp 100 km 100 km
BERT decoder CDR

RP
RP : Raman Pump

SLA-IDF-SLA
Photo

100 km
diode
20
odd

DGFF SLA-IDF-SLA
43G ETDM

RP
28 km IDF
Transmitter Inter-
20
even

RP RP RP
43G ETDM leaver Pre-comp
Transmitter 100 km
Loop switch DCF module

Fig. 14.28. Configuration of all-Raman amplified loop experiment for 1.6 Tbit/s transmission
[39]. Source: F. Liu et al: “1.6 Tbit/s (40/spl times/42.7 Gbit/s) transmission over 3600 km
UltraWave/spl trade/ fiber with all-Raman amplified 100 km terrestrial spans using ETDM
transmitter and receiver,” OFC Technical Digest, Postconference Edition Vol. 70, page FC7-1
(© 2002 OSA)

–5 23
–10
22
–10 –20 21
–30 20
Spectrum [dBm]

OSNR [dB]

1548 1550 1552

–15 19
18

BER
–20 10–3
17 10–4
16 10–5
–25 OSNR in 0.1 nm 10–6
15 31
Measured BER for 2 –1 PRBS 10–7
–30 14 Measured BER with FEC 10–8
13 10–9
10–11
–35 12
1530153515401545155015551560 1530 15351540 1545 1550 1555 1560
Wavelength [nm] Wavelength [nm]

Fig. 14.29. BER with and without FEC and optical signal-to-noise ratio after 3600 km trans-
mission [39]. Source: F. Liu et al: “1.6 Tbit/s (40/spl times/42.7 Gbit/s) transmission over 3600
km UltraWave/spl trade/ fiber with all-Raman amplified 100 km terrestrial spans using ETDM
transmitter and receiver,” OFC Technical Digest, Postconference Edition Vol. 70, page FC7-1
(© 2002 OSA)

return-to-zero (CS-RZ) modulation format, distributed Raman amplification, and new

fiber types. Unlike many 40 Gb/s systems that use optical time-division-multiplexing,
these results were obtained using an electronic time-division-multiplexed (ETDM)
transmitter and ETDM receiver. Even without using polarization interleaving, a high
spectral density (40 Gb/s on 100 GHz channel spacing) was achieved because of
this combination of techniques. The loop-back experimental configuration is shown
in Fig. 14.28, depicting the 100 km spans comprising 35.5 km of large effective
area fiber in series with 29 km of perfectly slope-matched dispersion-compensating
fiber and an additional 35.5 km of large effective area fiber. Each link was also
bidirectionally pumped, with a forward gain of ∼7.5 dB and a backward gain of
∼18.5 dB. In addition, fixed gain-flattening filters were used after each amplifier, and
a dynamic gain-flattening filter was used in one location of the loop. Figure 14.29
480 M.N. Islam et al.

1 1x20
NRZ CSRZ
39 C-band
Q DCF
FEC encoder 21.3
OI Post
42.7Gb/s GHz
comp Pre-Amp.
2 Q WGR
1x20
NRZ &SW
C-band CSRZ 42.7Gb/s
40 C C C
EDTM
L receiver
41 1x20 L L
NRZ CSRZ A
L-band O WGR
79 Q FEC
DCF M Post &SW decoder
FEC encoder 21.3
comp Pre-Amp.
42.7Gb/s GHz OI
BERT
Q A
42 1x20 O
NRZ CSRZ
L-band M
80

x4 DCF 100-km fiber

C C-GE C

L L
L-GE DCF-RP Counter-RP CO-RP

Fig. 14.30. Experimental setup for transmission of 80 × 42.7 Gb/s channels over 20 laps of 100
km nonzero dispersion-shifted fiber [40]. Source: B. Zhu: “3.2Tb/s (80 /spl times/ 42.7Gb/s)
transmission over 20 /spl times/ 100km of non-zero dispersion fiber with simultaneous C+L-
band dispersion compensation,” OFC Technical Digest, Postconference Edition Vol. 70, pg
FC8-1 (© 2002 OSA)

shows the BER and SNR spectra with and without FEC. Using FEC, a bit error rate
lower than 10−11 was achieved for each channel.
A similar experiment was performed by Zhu et al. [40], where transmission of
3.2 Tb/s (80 × 42.7 Gb/s) over 20 × 100 km of nonzero dispersion-shifted fiber was
accomplished in a recirculating loop. All-Raman amplified spans, dispersion slope
matched DCF modules covering a 75 nm band, CS-RZ modulation format, and FEC
were employed. The low dispersion slope of the transmission fiber enables C- and
L-band dispersion and dispersion slope-compensation of each span in a single DCF
module. Also, the location of the zero dispersion wavelength of the transmission fiber
at <400 nm enables forward and backward pumping of the fiber spans for flat-gain and
noise figure across the entire C- and L-bands by eliminating pump–pump and pump–
signal four-wave-mixing. The experimental setup for the recirculating loop is shown
in Fig. 14.30, where the all-Raman amplifier is shown to be a bidirectionally pumped
DRA in the 100 km transmission spans and a backward pumped discrete Raman
amplifier in the DCF. Figure 14.31 shows the measured pump-on/pump-off gains used
in the experiment where each span is pumped to transparency. In general, the SNRs at
short wavelengths are lower than at long wavelengths because the shorter wavelength
pumps amplify the longer wavelength pumps through the Raman effect. Therefore,
the use of higher copumped Raman gain at short wavelengths and the complementary
slope for the counter-pumped Raman gain provides both flat noise figure and gain.
Gain in the DCF is only used to compensate for losses in the DCF module and other
components. With FEC, the worst-case BER corresponds to a corrected BER below
10−13 at the receiver. This experiment illustrates the importance of both dispersion
 14. Wideband Raman Amplifiers 481

25
Span total

Raman on-off gain (dB)

20

15 Counter-pump

10
DCF
5
Co-pump
0
1530 1540 1550 1560 1570 1580 1590 1600
Wavelength (nm)

Fig. 14.31. Measured on-off Raman gains in all-Raman amplified loop experiment [40]. Source:
B. Zhu: “3.2Tb/s (80 /spl times/ 42.7Gb/s) transmission over 20 /spl times/ 100km of non-
zero dispersion fiber with simultaneous C+L-band dispersion compensation,” OFC Technical
Digest, Postconference Edition Vol. 70, pg FC8-1 (© 2002 OSA)

and dispersion slope matching, as well as proper use of complementary gain slopes
to obtain flat-gain and noise figure spectra.

14.7.2. Ultra-Long-Haul WDM Systems Based on Wideband Raman

Amplification
Telecommunications operators are shifting from demand for “capacity at any price” to
demand for “capacity at the lowest price” [39]. Thus the focus of equipment suppliers
is now to achieve large throughputs on a single strand of optical fiber over long
distances at the lowest cost. The effective metric for long-haul and ultra-long-haul
WDM transmission systems thus changes from just the product of total throughput
and maximum unregenerated distance (measured in units of Gb/s km), to cost per unit
bandwidth transmitted over unit distance ($/(Gb/s km)). Raman amplifier technology
not only provides the basis for low-noise amplification of a very large continuous
bandwidth, but also contributes to a substantial reduction in transmission cost by
enabling high-capacity WDM system design utilizing low-cost terminal components.
A single, large transmission band is desirable because it minimizes amplifier cost
by lowering the total number of passive components and pump sources, reduces
the complexity of amplifier configurations, simplifies amplifier management, and
minimizes the overall terminal cost by simplifying the multiplexer/demultiplexer
structure (c.f., Chapter 1, Fig. 1.15) [2]. An amplifier structure capable of providing
gain over a single ∼100 nm wide band, with performance and total transmission
cost comparable to those of traditional C-band (∼35 nm) EDFAs with DRA and
dispersion-compensation, can yield almost a threefold reduction of the economic
metric. By allowing gain in any wavelength window with simple tailoring of the
gain bandwidth, Raman amplification presents a reliable and low-cost solution to
extending bandwidth.
An amplifier structure combining both lumped Raman and distributed ampli-
fication can give significant advantages of reach and capacity while dramatically
lowering the overall cost of the system. Such a design has been implemented and the
482 M.N. Islam et al.

Line fiber

Gain Gain
A B fiber GFF fiber C

In Out
Distributed Lumped Lumped
Pump Module Pump Module Pump Module

Fig. 14.32. Schematic diagram of all-Raman line amplifier comprising distributed and lumped
Raman amplifiers. Dispersion and dispersion slope-compensation are provided by discrete
amplifier gain fibers [15].

ARLA
#19 Demux

Mod.1
AWG

23 fixed • OSA LRA

• ARLA
sources •
3 dB #18
coupler DCF LRA Post-
DATA comp

7 spans
Mod.2 80/20
AWG

20 fixed •• coupler
sources • RX #2
DATA ARLA ARLA
10.67Gb/s NRZ
#1 #11
3 dB
Mod.3 coupler PRBS 231–1 RX #1
BERT
16 : 1

• BERT TX RX #2
15 fixed •
•
#1 ARLA
sources ARLA BERT
#2 #10 RX #1
3 fixed DATA BER
Mod.4 Sync
6 spans

+ meas.
8:1

4 tunable ••
• BERT TX LRA Q meas.
sources
DATA
#2
DSE
10.67Gb/s NRZ
Transmitter Receiver
PRBS 231–1
ARLA #9

Fig. 14.33. Experimental setup for ultra-long-haul transmission of 67 × 10.7 Gb/s channels
over 1565 km of standard single-mode fiber [15].

feasibility of transmitting 240 OC-192 channels over 1565 km standard single-mode

fiber has been demonstrated [15]. In this all-Raman experiment, a wideband 100 nm
Raman amplifier is used both for the booster amplifier and the inline amplifier (ILA).
The booster amplifier is placed after the transmitter and consists of an LRA (lumped
Raman amplifier). The ILAs use a hybrid distributed/lumped Raman amplifier in a
three-stage configuration (Fig. 14.32). Dispersion and dispersion slope-compensation
are provided by the dispersion-compensation fiber used as gain medium in the LRAs,
and a static gain-flattening filter is used to obtain a substantially flat-gain spectrum
over 100 nm. Each LRA is composed of two stages separated by isolators to reduce
double-Rayleigh scattering, so that the resulting amplifier has a low MPI level.
The experimental setup for the ultra-long-haul transmission experiment is shown
in Fig. 14.33. For this example, the transmission of 67 × 10.7 Gb/s channels span-
ning >97 nm (from 1519.9 to 1617.0 nm) was achieved over 19 spans of average loss
 14. Wideband Raman Amplifiers 483

3000

2500 1520nm
1550nm
2000 1580nm
1617nm
Dispersion [ps/nm]

1500

1000

500

–500

–1000

–1500
0 200 400 600 800 1000 1200 1400 1600 1800
Distance [km]

Fig. 14.34. Dispersion map for transmission experiment utilizing 100 nm bandwidth all-Raman
line amplifiers. A common precompensation of ∼−1000 ps/nm is provided across the full
bandwidth. Banded postcompensation of 0, −300, −300, and 0 ps/nm is provided in four
subbands [15].

19.7 dB. The total distance was 1565 km, and the average span length was 82.4 km.
Using tunable sources, each measured channel was set at the center of a group of
five 50 GHz-spaced channels, all NRZ-modulated with 231 to 1 pseudorandom bit
sequence. The overall system performance was measured using a bit error rate tester
(BERT).
The dispersion map for the experiment is illustrated in Fig. 14.34. For all of the
channels there is a common precompensation of −1000 ps/nm, which is provided
by the DCF after the transmitter block and before the booster amplifier shown in
Fig. 14.33. Because the dispersion slope match is not perfect, the continuous band
is divided at the receiver end into four subbands, which receive respective postcom-
pensation of 0, −300, −300, 0 ps/nm. Better slope matching of the DCF to the
transmission fiber would reduce the need for postcompensation.
Input and output spectra, measured with a resolution of 0.1 nm on an optical
spectrum analyzer, are illustrated in Fig. 14.35. Whereas the input spectrum is more
or less flat, the output shows some accumulation of ripple after the 19 amplifier spans.
The amplifier output power per channel is 0 dBm (1mW), and the minimum optical
SNR is 18 dB (measured from ASE floor to the peak of the signal). On the right side
of the figure, an expanded view is shown of several channels at the short-wavelength,
midrange, and long-wavelength side of the band.
The transmission performance measured with the BERT is shown in Fig. 14.36.
The end-of-life BER target is 10−15 , which corresponds to a quality factor Q of
12.5 dB20 (i.e., 20 log(Q) = 12.5). However, an additional 3 dB of end-of-life margin
484 M.N. Islam et al.

Received 'blue' edge

(1519.86nm)
Res=0.1nm, 5db/div

Received mid-band
(1567.95nm)

Res=0.1nm, 5db/div

Received 'red' edge

(1617.00nm)

Fig. 14.35. Input (top left) and output (bottom left) signal spectra in the 100 nm bandwidth
all-Raman amplified transmission experiment. The insets on the right show output spectra near
the short-wavelength (top), central (middle), and the long-wavelength portions of the signal
band [15].

22
21
20
20 log(Q) (dB20)

19
18
17
16
15
14
13
12
1510 1520 1530 1540 1550 1560 1570 1580 1590 1600 1610 1620
Wavelength (nm)

Fig. 14.36. Experimental Q value spectrum in the 100 nm bandwidth all-Raman amplified
transmission experiment. The solid horizontal line indicates the target end-of-life Q [15].

needs to be allocated for loss and other impairments. Hence the target end-of-life
Q is 15.5 dB20 . Figure 14.36 shows the measured Q across the ∼100 nm band by
using the tunable lasers to obtain five consecutive 50 GHz-spaced channels for each
measurement. All channels showed a performance exceeding the end-of-life target of
10−15 BER, given a 5.5 dB coding gain from FEC (RS 255/239).
 14. Wideband Raman Amplifiers 485

121 channels, 100nm bandwith

+21dBm launch power

Res=0.1nm, 5dB/div
(a)

L4–L3: 13.37dB
:EF

(b)

Fig. 14.37. (a) Signal spectrum after transmission of 121 channels over 1565 km of standard
single-mode fiber; (b) detail of signal spectrum showing system margin for 50 GHz channel
spacing at 0 dBm/channel. The composite launch power is 21 dBm [15].

Although this exemplary experiment uses only 67 channels, newer results do

show the performance of the system up to the full load of 240 channels. For example,
Fig. 14.37 shows the output spectrum after 1565 km for transmission of 121 chan-
nels over a 100 nm bandwidth with a composite launch power of +21 dBm. As the
lower curve shows, the system margin is tested by using the tunable lasers to create
five channels at 50 GHz channel spacing with the output from each amplifier set
at 0 dBm per channel. All channels are decoded error-free with FEC (RS 255/239).
Margin measurements on all channels from 6.3 to 7.5 dB were obtained by input
channel deemphasis from the nominal amplifier output channel power (0 dBm). Fig-
ure 14.38 shows the output spectrum for 240 channels after eight spans (∼650 km total
length) for a substantially flat input spectrum. Each ILA now has an output power of
+24 dBm, corresponding to 0 dBm per channel for the 240 channels. With the use of
dynamic gain equalizers placed periodically in the span (i.e., one or two in the entire
19-span link), transmission of all 240 channels over a distance above 1500 km can
be achieved.
486 M.N. Islam et al.

1511.00 nm 1534.0 1557.0 1580.0 1603.0 1626.0 nm

1 2
–18.0 dBm

1
2
–28.0

–38.0

–48.0 dBm
Trace Evaluation Cursors Marker
1 2
Actual
Actual Min Hold x A C x' X–Cursors (nm) 1534.94 1572.59 37.65
X – X' Max Hold B D Y–Cursors (dBm) –24.58 –26.20 1.62

17:04 07/13/2002 Comments Res. :0.2 nm default.grh

Fig. 14.38. Output spectrum for 240 channels after transmission over 650 km (eight spans) of
standard single-mode fiber [15].

14.8. Summary and Conclusions

This chapter described design and implementation of wideband Raman amplifiers.
All-Raman amplification enables the lowest cost and smallest footprint system for a
number of reasons. First, an all-Raman amplifier requires fewer components than the
alternative splitband approach. Second, broadband gain (100 nm or more) allows use
of less dense channel spacing (e.g., 10 Gb/s on 50 GHz channel spacing) and standard
nonreturn-to-zero modulation, thereby lowering transponder cost. Third, although
most long-haul and ultra-long-haul amplified systems are limited by SNR or nonlin-
earity, the use of distributed Raman amplification offers significant improvement in
noise figure and mitigation of nonlinearity compared with discrete amplification as
offered by EDFAs.
To realize wideband systems, a number of enabling technologies are required. First
among these is the wideband amplifier itself. Approaches for constructing wideband
amplifiers include: EDFAs in various glass compositions, other rare earth dopings,
splitband S/C/L amplifiers, hybrid Raman/EDFA amplifiers, and all-Raman ampli-
fiers. Second, broadband components are required, such as multiplexers and de-
multiplexers. Third, wideband systems require gain equalizers and gain-flattening
filters, which may be static or dynamic. Finally, for operation at 10 Gb/s dispersion-
compensation as well as dispersion slope-compensation is required periodically in
the system.
Raman amplification provides a simple single platform for long-haul and ultra-
long-haul fiber-optic transmission systems. Distributed Raman amplification alone
 14. Wideband Raman Amplifiers 487

or in combination with lumped Raman amplification improves the noise figure and
nonlinear penalty performance. Also, gain in Raman amplification is obtainable in
practically any wavelength band and can be over a wideband, determined primarily
by the pump wavelength and power distribution. In addition, lumped Raman am-
plification can integrate dispersion-compensation by using DCF as the gain fiber.
Other advantages of Raman amplification include simpler equalization, simple ad-
justment of gain flatness, and large power dynamic range for flat-gain operation.
Despite this significant list of advantages, a number of challenges exist for Raman
amplification, including: pump–pump interactions, interband and intraband Raman
gain tilt, noise arising from thermally induced phonons near the pump wavelengths,
MPI from double-Rayleigh scattering, coupling of pump fluctuations to the signal,
and pump-mediated signal crosstalk.
Fortunately, design techniques exist for overcoming all of these physical limi-
tations, thus allowing for the relatively simple implementation of 100 nm Raman
amplifiers. Section 14.5 on design of wideband Raman amplifiers provides in detail
the engineering design rules used to determine the architecture of wideband Raman
amplifiers. Because of MPI penalties from DRS, wideband Raman amplifiers are
typically sectioned into multiple stages, where each stage is separated by an isolator.
Also, the easiest way to deal with the short upper state lifetime of the Raman effect
is to introduce an effective lifetime equal to the transit time by making the pump and
signal counterpropagating.
For wideband Raman amplifiers, several physical effects become particularly
important. The Raman effect leads to power exchange between the pumps and between
the signals (interband and intraband). To overcome the pump–pump interactions,
the basic idea is to increase the pump spectral density at shorter wavelengths. In
addition, thermal noise is a significant limitation for wideband Raman amplifiers. To
mitigate the effect of thermal noise at room temperature, one strategy is to amplify
the vulnerable signals first. In a hybrid amplifier comprising a DRA followed by a
lumped Raman amplifier, this technique leads to gain slopes that are complementary
in the two stages with the DRA gain increasing with frequency.
Several unique features of Raman amplification also lead to simplified system
implementations. For example, in a Raman amplifier the gain shape as a function
of frequency can be controlled by varying the pump powers. This leads to a simple
system where the signal input and output spectra can be monitored and used to control
the pump powers applied to the Raman amplifier.
Section 14.6 illustrates various wideband Raman amplifiers that have been re-
ported in the literature. Although commercially available wideband Raman amplifiers
have been limited to a bandwidth of 100 nm to date (i.e., the separation of the pump
and signal at roughly the peak of the gain band), laboratory experiments have shown
amplifiers with much larger bandwidths. Bandwidths greater than 100 nm are usu-
ally achieved with such special techniques as new glass compositions or wavelength
guard bands around the pump wavelengths.
Finally, Section 14.7 describes successful application of wideband Raman am-
plification in high-performance transmission systems. For example, a hybrid Raman
amplifier structure, combining both LRA and DRA, can give significant advantages
488 M.N. Islam et al.

of reach and capacity while dramatically lowering the overall cost of the system. Such
a design has been implemented and the transmission feasibility of 240 OC-192 chan-
nels over 1565 km standard single-mode fiber has been demonstrated. The BER target
is 10−15 and a 3 dB of end-of-life margin for loss and other impairments leads to a
target Q of 15.5 dB20 . In the systems experiment all channels showed a performance
exceeding the end-of-life target, given a 5.5 dB coding gain from FEC (RS 255/239).
Although the history of Raman amplification in fibers dates back to the 1970s, only
in recent years has the technology matured to the point that practical systems can be
built and commercialized. Engineering design rules have been developed to address
almost all physical limitations of Raman amplification, and advances in pump lasers,
fiber quality, and passive components have made Raman amplifiers economical.
Raman amplification enables other system cost reductions, although amplifiers
account for about 20% of the total loaded system cost in fiber-optic transmission
systems. For example, due to the lower noise figure and reduced nonlinear penalty
with DRAs, reach can be 1500 km or more, thus reducing the cost associated with
channel-by-channel regeneration. Because of the broad bandwidth enabled by Raman
amplification, several terabit-per-second capacities can be achieved on a single strand
of fiber without exceeding a spectral density of 0.2 bps/Hz and using standard NRZ
modulation format. Hence, transponder cost can be low for wideband Raman systems.
Moreover, because a single wideband amplifier can be used instead of three amplifiers
in a splitband configuration for a 100 nm bandwidth, the amplifier cost is reduced by
eliminating band couplers and by consolidating dispersion-compensation modules
and monitoring and control circuits. With fewer parts, the footprint for all-Raman
amplification can be lowered, thereby also reducing the operating expense to carriers.
For all of these reasons, Raman amplification can be expected to be dominant in
long-haul and ultra-long-haul systems in the near future.

References
[1] S. Lin, Lightwave 19: 3, 75, 2002.
[2] M. Islam and M. Nietubyc, WDM Solutions, 4: 51, 2000.
[3] R. C. Alferness, H. Kogelnik, and T. H. Wood, Bell Labs Tech. J., 5: 188, 2000.
[4] A. Hadjifotiou. In Optical Amplifiers and their Applications, OSA Technical Digest,
Washington DC: Optical Society of America, 2, 2000.
[5] Y. Sun, A. K. Srivastava, J. Zhou, and J. W. Sulhoff, Bell Labs Tech. J. 4: 187, 1999.
[6] H. Ono, M. Yamada, and M. Shimizu, Electron. Lett., 38: 1084, 2002.
[7] T. Sugawa, T. Komukai, and Y. Miyajima, IEEE Photon. Technol. Lett., 2: 475, 1990.
[8] M. Yamada, A. Mori, K. Kobayashi, H. Ono, T. Kanamori, K. Oikawa, Y. Nishida, and
Y. Ohishi, IEEE Photon. Technol. Lett., 10: 1244, 1998.
[9] J. D. Minelly and A. J. E. Ellison. In Proceedings of the Conference on Lasers and Electro-
Optics, OSA Technical Digest, Washington DC: Optical Society of America, 352, 2001.
[10] S. Tanaka, K. Imai, T. Yazaki, and H. Tanaka. In Proceedings of OSA Trends in Optics and
Photonics (TOPS) Vol. 70, Optical Fiber Communication Conference, Technical Digest,
Postconference Edition, Washington DC: Optical Society of America, 458, 2002.
 14. Wideband Raman Amplifiers 489

[11] T. Sakamoto. In Proceedings of OSA Trends in Optics and Photonics (TOPS) Vol. 54,
Optical Fiber Communication Conference, Technical Digest, Postconference Edition,
Washington DC: Optical Society of America, TuQ1-1, 2001.
[12] J. Masum-Thomas, D. Crippa, and A. Maroney. In Proceedings of OSA Trends in Optics
and Photonics (TOPS) Vol. 54, Optical Fiber Communication Conference, Technical
Digest, Postconference Edition, Washington DC: Optical Society of America, WDD9-1,
2001.
[13] D. Bayart, P. Baniel, A. Bergonzo, J.-Y. Boniort, P. Bousselet, L. Gasca, D. Hamoir, F.
Leplingard, A. Le Sauze, P. Nouchi, F. Roy, and P. Sillard, Electron. Lett., 36: 1569, 2000.
[14] H. Masuda and S. Kawai, IEEE Photon. Technol. Lett., 11: 647, 1999.
[15] M. W. Chbat and H. A. Fevrier. In Proceedings of the 28th European Conference on
Optical Communication (COM, University of Denmark, Copenhagen, Denmark), 2002;
M. W. Chbat. In OSA Technical Digest Series, Conference on Lasers and Electro-Optics,
Washington DC: Optical Society of America, 430, 2002.
[16] A. Mori, H. Masuda, K. Shikano, K. Oikawa, K. Kato, and M. Shimizu, Electron. Lett.,
37: 1442, 2001.
[17] T. Yagi. In Proceedings of OSA Trends in Optics and Photonics (TOPS) Vol. 77, Opti-
cal Amplifiers and Their Applications, OSA Technical Digest, Postconference Edition,
Washington DC: Optical Society of America, OMC1-1, 2002.
[18] S. G. Grubb, R. Zanoni, and T. D. Stephens, Patent Number 6115174, United States Patent
Office, 2000.
[19] G. P. Agrawal, Nonlinear Fiber Optics, New York: Academic, 1995.
[20] H. Kidorf, K. Rottwitt, M. Nissov, M. Ma, and E. Rabarijaona, IEEE Photon. Technol.
Lett., 11: 530, 1999.
[21] C.-J. Chen and W. S. Wong, Electron. Lett., 37: 371, 2001.
[22] S. A. E. Lewis, S. V. Chernikov, and J. R. Taylor, Opt. Lett., 24: 1823, 1999.
[23] K. Rottwitt, J. Bromage, D. Mei, and A. Stentz. In Proceedings of the 26th European
Conference on Optical Communication Berlin: VDE Verlag, vol. 2, 67, 2000.
[24] M. Nissov, K. Rottwitt, H. D. Kidorf, and M. X. Ma, Electron. Lett., 35: 997, 1999.
[25] C. R. S. Fludger, V. Handerek, and R. J. Mears, IEEE J. Lightwave Technol., 19: 1140,
2001; 20: 316, 2002.
[26] F. Forghieri, R. W. Tkach, and A. R. Chraplyvy. In Optical Fiber Communication Vol. 4,
Technical Digest Series, Conference Edition, Washington DC: Optical Society of Amer-
ica, 294, 1994.
[27] R. E. Neuhauser, P. M. Krummrich, H. Bock, and C. Glingener. In Proceedings of the Op-
tical Fiber Communication Conference, OSA Technical Digest, Washington DC: Optical
Society of America, MA4-1, 2001.
[28] S. A. E. Lewis, S. V. Chernikov, and J. R. Taylor, Electron. Lett., 35: 1761, 1999.
[29] C. R. S. Fludger, V. Handerek, N. Jolley, and R. J. Mears. In Proceedings of OSA Trends
in Optics and Photonics (TOPS) Vol. 70, Optical Fiber Communication Conference,
Technical Digest, Postconference Edition, Washington DC: Optical Society of America,
183, 2002.
[30] J. Bromage, P. J. Winzer, L. E. Nelson, and C. J. McKinstrie. In Proceedings of OSA Trends
in Optics and Photonics (TOPS) Vol. 77, Optical Amplifiers and Their Applications, OSA
Technical Digest, Post Conference Edition, Washington DC: Optical Society of America,
OWA5-1, 2002.
[31] S. Namiki and Y. Emori, IEEE J. Select. Topics Quantum Electron., 7: 3, 2001.
[32] A. R. Chraplyvy, R. W. Tkach, K. C. Reichmann, P. D. Magill, and J. A. Nagel, IEEE
Photon. Technol. Lett., 5: 428, 1993.
490 M.N. Islam et al.

[33] T. Naito, T. Tanaka, K. Torii, N. Shimojoh, and H. Nakamoto. In Proceedings of OSA

Trends in Optics and Photonics (TOPS) Vol. 70, Optical Fiber Communication Con-
ference, Technical Digest, Postconference Edition, Washington DC: Optical Society of
America, 116, 2002.
[34] H. Takara, H. Masuda, K. Mori, K. Sato, Y. Inoue, T. Ohara, A. Mori, M. Kotoku, Y.
Miyamoto, T. Morioka, and S. Kawanishi. In Proceedings of OSA Trends in Optics and
Photonics (TOPS) Vol. 70, Optical Fiber Communication Conference, Technical Digest,
Postconference Edition, Washington DC: Optical Society of America, FB1-1, 2002.
[35] D. G. Foursa, C. R. Davidson, M. Nissov, M. A. Mills, L. Xu, J. X. Cai, A. N. Pilipetskii,
Y. Cai,C. Breverman, R. R. Cordell, T. J. Carvelli, P. C. Corbett, H. D. Kidorf, and N.
S. Bergano. In Proceedings of OSA Trends in Optics and Photonics (TOPS) Vol. 70,
Optical Fiber Communication Conference, Technical Digest, Postconference Edition,
Washington DC: Optical Society of America, FC3-1, 2002.
[36] Y. Frignac, G. Charlet, W. Idler, R. Dischler, T. Tran, S. Lanne, S. Borne, C. Martinelli, G.
Veith, A. Jourdan, J.-P. Hamaide, and S. Bigo. In Proceedings of OSA Trends in Optics and
Photonics (TOPS) Vol. 70, Optical Fiber Communication Conference, Technical Digest,
Postconference Edition, Washington DC: Optical Society of America, FC5-1, 2002.
[37] A. H. Gnauck, G. Raybon, S. Chandrasekhar, J. Leuthold, C. Doerr, L. Stulz, A. Agarwal,
S. Banerjee, D. Grosz, S. Hunsche, A. Kung, A. Marhelyuk, D. Maywar, M. Movassaghi,
X. Liu, C. Xu, X. Wei, and D. M. Gill. In Proceedings of OSA Trends in Optics and
Photonics (TOPS) Vol. 70, Optical Fiber Communication Conference, Technical Digest,
Postconference Edition, Washington DC: Optical Society of America, FC2-1, 2002.
[38] H. Sugahara, K. Fukuchi, A. Tanaka, Y. Inada, and T. Ono. In Proceedings of OSA Trends
in Optics and Photonics (TOPS) Vol. 70, Optical Fiber Communication Conference,
Technical Digest, Postconference Edition, Washington DC: Optical Society of America,
FC6-1, 2002.
[39] F. Liu, J. Bennike, S. Dey, C. Rasmussen, B. Mikkelsen, P. Mamyshev, D. Gapontsev,
and I. Ivshin. In Proceedings of OSA Trends in Optics and Photonics (TOPS) Vol. 70,
Optical Fiber Communication Conference, Technical Digest, Postconference Edition,
Washington DC: Optical Society of America, FC7-1, 2002.
[40] B. Zhu, L. Leng, L. E. Nelson, L. Gruner-Nielsen, Y. Qian, J. Bromage, S. Stulz, S. Kado,
Y. Emori, S. Namiki, P. Gaarde, A. Judy, B. Palsdottir, and R. L. Lingle, Jr. In Proceedings
of OSA Trends in Optics and Photonics (TOPS) Vol. 70, Optical Fiber Communication
Conference, Technical Digest, Postconference Edition, Washington DC: Optical Society
of America, FC8-1, 2002.
Chapter 15
Multiple Path Interference and Its Impact
on System Design

J. Bromage, P.J. Winzer, and R.-J. Essiambre

15.1. Introduction
Lightwave communication systems carry information that is encoded onto the inten-
sity, phase, or polarization of light from one point to another along an optical path.
When designing such systems, many mechanisms that degrade the transfer of infor-
mation must be taken into account. Until the late 1990s, the main causes of signal
degradation in transmission were fiber nonlinearity and amplified spontaneous emis-
sion (ASE) from optical amplifiers. More recently, however, a third type of system
degradation, involving the unwanted beating of the signal with a number of weak
interferers, has become increasingly important. With reference to Fig. 15.1(a), such
interferers can result from imperfect extinction of the drop-signal in optical cross-
connects and add-drop multiplexers, which are both key elements for flexible and
transparent optical network architectures [1, 2]. Also, single-Rayleigh backscattering
in bidirectional transmission systems [3, 4] can lead to unwanted interferers at the
receiver. Although these two examples involve interferers that are independent of the
main signal, the important class of multiple-path interference (MPI) involves inter-
ferers that are delayed replicas1 of the main signal. In the case of MPI, additional
(unwanted) optical paths, with losses orders of magnitude greater than the main path,
lead to interfering signals at the receiver, and can have a significant impact on sys-
tem performance. With reference to Figs. 15.1(b) and 15.1(c), MPI is encountered
for
• discrete reflections within or surrounding optical amplifiers [5],
• double-Rayleigh scattering in optical amplifiers [6, 7],
• double-Rayleigh scattering in the transmission span [8], or
• unwanted transverse mode mixing in higher-order mode dispersion compen-
sators [9].

1 Note that for path-dependent signal distortion, as found in highly nonlinear or dispersive sys-
tems, MPI may lead to interference between fields that have significantly different spectra.
492 J. Bromage et al.

(a) Leakage at (b) Fiber amplifier (c) Raleigh scattering

Add/Drop sites in optical fiber

Drop Add Drop Add Reflections at poor

connectors or components

Fig. 15.1. Schematic examples of possible sources of inband crosstalk: (a) discrete reflection
sites at connectors or components on either side of an optical amplifier; (b) double-Rayleigh
scattering within a fiber; (c) poor extinction of add-drop multiplexers.

MPI has become increasingly relevant with the advent of optical amplifiers, in
particular, of distributed Raman amplification [10–12]. Lightwave systems incor-
porating optical amplifiers can transmit information much farther before electronic
regeneration is necessary. Such increased network transparency, therefore, places
stricter requirements on system components, or the accumulated MPI may reach un-
acceptable levels [13, 7, 14]. Furthermore, optical amplification can exacerbate MPI
by providing gain for paths that would otherwise have too much attenuation to be
significant [5, 15]. The classic example is that of two weak partial reflectors, spaced
by some distance along a fiber link. If an amplifier is placed between the reflectors,
the double-pass gain may be enough to offset the weak reflections, increasing the
MPI. (In the most extreme case, the fiber may even lase.) Even if care is taken to
eliminate discrete reflections, unavoidable Rayleigh backscatter in sufficiently long
fibers may cause similar effects [8, 16, 17].

15.1.1. Inband Crosstalk and MPI

Inband crosstalk occurs in general when a signal beats with other interfering fields,
producing beat frequency components that fall within the receiver’s bandwidth and
add to other sources of receiver noise. This process is visualized in Fig. 15.2. The
beating between signal (black) and interferer (gray) at difference frequency δf falls
within the receiver’s electrical bandwidth Be , thus producing inband crosstalk. Note
that larger beat frequencies (e.g., f ) are also produced. These, however, exceed
Be and thus do not lead to inband crosstalk, even though the two interfering fields
occupy the same optical frequency band. This latter process is referred to as out-
of-band crosstalk. As discussed along with Fig. 15.1, inband crosstalk can either be
caused by interferers originating from separate transmitters, or by MPI, where the
signal beats with delayed replicas of itself that have reached the receiver by one or
more additional paths.
The negative impact of MPI on communication systems can also be understood
as a consequence of the conversion of phase fluctuations to intensity fluctuations
 15. Multiple Path Interference 493

Optical spectrum
δf

Be Signal

∆f

Interferer

Frequency

Fig. 15.2. The beating between signal (black) and interferer (gray) at difference frequency δf
falls within the receiver’s electrical bandwidth Be , thus producing inband crosstalk. Beating
at f exceeds Be and is thus called out-of-band crosstalk.

by interference. As an illustration, consider continuous-wave (CW) light passing

through a simple two-path interferometer. The intensity of light at an output port
depends critically on the phase relationship between the electric fields from each
path, set by the optical path difference. If the source of light is purely monochromatic
and the optical path difference is static, this phase relationship is fixed. Therefore,
the interferometer output has constant intensity, fixed between a maximum value
(constructive interference) and a minimum value (destructive interference). A more
realistic case, however, is when the CW source emits light with a finite linewidth, often
the result of time-dependent random phase fluctuations [18]. Now the phase difference
between the interfering fields also fluctuates in time, producing intensity fluctuations.
In other words, the interferometer converts phase noise to intensity noise [19, 20].
This intensity noise can mask information-carrying modulation to the point where
information is lost.

15.1.2. Classification of MPI

Sources of MPI can be classified using two parameters. The first parameter is N , the
number of paths between the transmitter and receiver. N can vary from two (e.g.,
Mach–Zehnder-like interferometers) to infinity (e.g., double-Rayleigh backscatter-
ing distributed along a fiber amplifier). The other parameter describes the mutual
coherence2 of the interfering fields. For MPI one can consider the ratio of the co-
herence time of the source to the path delay. If many paths exist, as in the case of
double-Rayleigh backscattering (DRB), there will not be a clearly defined delay time.
Instead one can consider an effective delay time of the dominant paths that produce
the maximum beat noise.
2 Mutual coherence [21] describes the amount by which two optical fields are deterministically
related to each other. Two fields are mutually coherent if amplitude and phase of one field
can be predicted by knowledge of amplitude and phase of the other field.
494 J. Bromage et al.

Double-Rayleigh backscattering Mode-coupling

∞ in higher-order mode
in Raman amplifiers
dispersion compensators
N,
number of
interfering
paths Discrete reflections in or
Pair of discrete reflections
surrounding EDFAs
within a Raman-amplified span
2

0 ∞
Coherence time / effective path delay

Fig. 15.3. Examples of MPI plotted versus the number of paths and the ratio of coherence time
to path delay.

These two parameters are useful, because two MPI-producing devices that have
very different forms and functions can have similar MPI properties. If they have the
same number of paths and mutual coherence properties for the interfering fields, the
degradation caused for the same level of MPI will be identical. Examples of MPI for
different numbers of paths and different coherence properties are shown in Fig. 15.3.
Although much of this chapter applies to all sources of inband crosstalk mentioned
along with Fig. 15.1, we mainly concentrate on MPI produced inside Raman amplifiers
by discrete reflections or by unavoidable Rayleigh backscattering in the fiber. As
Raman amplification can be used in many different parts of lightwave systems, and in
many different types of optical fibers, an understanding of this noise source is crucial
when designing and characterizing Raman-amplified systems.
In this chapter we first present some of the main physical concepts that describe
MPI and are used to quantify it. Then we discuss the impact the resulting intensity
noise has on system performance and receiver design. Finally we show how one
optimally designs Raman amplifiers to take into account MPI from double-Rayleigh
backscattering.

15.2. Properties of MPI

In this section we examine some important properties of intensity noise resulting
from multiple path interference. First we consider the intensity noise in the frequency
domain. Starting with expressions for the interfering optical fields, we derive an
expression for the spectrum of intensity noise for the case of two paths (see Sec-
tion 15.2.1). Next, the common case of sources with Lorentzian lineshapes is treated
in Section 15.2.1.2. In Section 15.2.1.3 these results are generalized to include mul-
tiple interfering fields.
In Section 15.2.2, we examine the intensity fluctuations in the time domain. Here
we compare the probability density functions for the intensity noise from different
numbers of paths. This analysis points out important differences between cases of MPI
involving a few paths to cases where there are a large or infinite number of paths.
 15. Multiple Path Interference 495

15.2.1. Intensity Noise Spectra

15.2.1.1. Two Interfering Fields

The objective of this section is to derive a general expression for the spectrum of
the intensity noise created by interference between optical fields from two paths. The
noise power spectral density is valuable for understanding how MPI leads to inband
crosstalk as well as how levels of MPI can be measured (see Section 15.4). As a result
of this analysis, we show that the spectral shape of the beat noise depends on the
optical spectra and mutual coherence properties of the interfering optical fields. The
mathematical treatment used for this simple case of two interfering fields also serves as
an introduction to the analysis of the more complex case of double-Rayleigh-induced
MPI (see Section 15.3.3), which basically uses the same approach.
First consider the following expression for the optical field of a laser at the signal
frequency fo ,
−→
E s (t) = Re[−
→
ε s (t)e−j 2πfo t ], (15.1)
−
→
where Re denotes the real part and ε (t) is the stochastic complex amplitude vector
s
given by +
−
→
ε s (t) = −
→
ps ε s (t) = −
→
ps Is ejφφ (t) . (15.2)
√ −
→
The quantity Is is the field amplitude, ps is the field polarization vector, and φ (t)
is the random instantaneous phase of the field’s complex envelope. Throughout this
chapter we use bold print (e.g., X ) to indicate a random quantity. For simplicity we
assume that the source has negligible amplitude noise. Therefore the linewidth of the
source is only due to phase noise, which is typically the case for semiconductor lasers.
The instantaneous optical intensity is Is = |− →
ε s (t)|2 .
Consider the case of two-path interference, shown in Fig. 15.4. Added to the signal
field −
→
ε s is the doubly reflected field − →
ε r , given by
*
−
→
ε r (t) = −
→
pr R1 R2 G 12 G 12 ε s (t − Td )ej 2πfo Td , (15.3)
−
→ ← −
where − →
p is the polarization vector, T is the round-trip delay, and R is the intensity
r d i
reflectivity at position zi (i = 1, 2). The quantity G 12 is the intensity gain for light
−
→

z1 z z2
εs (t )
G12 R2

R1 G12
εr (t ) = εs (t + Td )√R1R 2G12G12

2nd order reflection

Fig. 15.4. MPI caused by two discrete reflections at z1 and z2 . The variableεε s is the output field
amplitude from the straight-through path; ε r is from the two-reflection path with round-trip
delay Td . R1 and R2 are the intensity reflectivities of the mirrors and G 12 G 12 is the round-trip
−
→ ← −
intensity gain between them. If R1 R2 G 12 G 12 # 1, higher-order paths may be neglected.
−
→ ← −
496 J. Bromage et al.

propagating in the +z-direction, from z1 to z2 ; G 12 is the intensity gain for light

←−
propagating in the −z-direction, from z2 to z1 . This notation allows for cases in
which the gain is nonreciprocal, that is, for cases where the gain experienced by light
passing between two points depends on the direction of propagation. An important
case of a nonreciprocal gain is, for example, when an isolator is located between
z1 and z2 [22, 23]. Here we assume that R1 R2 G 12 G 12 # 1 so that higher-order
−
→ ← −
reflections can be neglected. This is a realistic assumption because, as we show in
Section 15.5.2.3, significant penalties already result when this term is as small as
0.003 or −25 dB. The instantaneous intensity at the output is
I (t) = Is + Ir + II (t, Td ), (15.4)
where II is the time-dependent intensity fluctuation given by
*
ps · −
II (t, Td ) = (−
→ →
pr T ) R1 R2 G 12 G 12 ε Ts (t)εε s (t − Td ) + c.c. (15.5)
−
→ ← −
(The common notation c.c. stands for the complex conjugate of the expression pre-
ε s and −
ceding it.) Here we treat the worst case where the polarization states of −
→ →
ε r are
−
→ −
→
identical ( ps · pr T = 1), producing maximum intensity noise. To obtain an expression
for the noise spectrum, we first calculate the autocorrelation function of the noise. In
general, this is given by
=X (t, τ ) = X X (t + τ ),
X T (t)X (15.6)
where · denotes an ensemble average. For a wide-sense stationary process (WSS)
[24], =X only depends on τ and not on t. The autocorrelation function of the WSS
intensity fluctuations is
=II (τ ) = R1 R2 G 12 G 12 [εε Ts (t)εε s (t − Td ) + c.c.][εε Ts (t + τ )εε s (t − Td + τ ) + c.c.].
−
→ ← −
(15.7)

To proceed further, we need to evaluate the ensemble averages using statistical

information about the phase fluctuations and therefore the optical spectrum of the
laser light. However, if we consider the case where the delay Td is much greater
than the coherence time of the source, we can obtain a general expression for the
noise spectral density that is valid for any lineshape. The coherence time of signal
lasers used in fiber telecommunications (roughly the inverse of the laser linewidth)
is 0.1 to 1 µs. These delays correspond to distances between reflection sites of 10 to
100 m. (As we show in Section 15.3.2, significant double-Rayleigh backscattering in
distributed amplifiers comes from paths 100 to 1000 times longer than this.) In the
limit of large delay, ε s (t) and ε s (t − Td ) are uncorrelated. This allows us to reduce
the intensity autocorrelation in Eq. (15.7), which contains fourth-order moments of
the form εε 1ε T2ε 3ε T4 , to a product of second-order moments of the form εε 1ε T2 εε 3ε T4 
and εε 1ε 2 εε T3ε T4  [25]. The latter product can be shown to vanish.3 If we use our
3 This can be seen by writing φ (t) +φ
φ (t + τ ) = 2φ
φ (t) + (φ
φ (t + τ ) −φ
φ (t)) = 2φ
φ (t) +
φφ (t, τ ). Because for a random walk process, φ (t) and φ φ (t, τ ) are statistically inde-
pendent, we are left with ej 2φφ (t) ej φφ (t,τ ) . The first term ej 2φφ (t)  = 0 as the phase at
time t is a random variable that is uniformly distributed between 0 and 2π .
 15. Multiple Path Interference 497

assumption of wide-sense stationarity we obtain

=II (τ ) = 2R1 R2 G 12 G 12 |=ε s (τ )|2 , (15.8)
−
→ ← −
that is, the intensity autocorrelation expressed in terms of the field autocorrelation.
According to the Wiener-Khintchine theorem, the power spectral density of a random
process is derived from the Fourier transform of the autocorrelation [24]. Therefore
we can express the intensity noise power spectral density as
SII (f ) = 2F{=II (τ )},
= 4R1 R2 G 12 G 12 F{|=ε s (τ )|2 },
−
→ ← −
= 4R1 R2 G 12 G 12 [Sε s (f ) ∗ Sε s (f )], (15.9)
→ ←
− −
where ∗ denotes a convolution. This shows that the intensity noise spectrum depends
on the autocorrelation of the field spectrum Sε s (f ). (F denotes the Fourier transform.)
Here SII (f ) is the single-sided spectral density, which is nonzero for f ∈ [0, ∞).4
The self-homodyne technique for measuring laser linewidths [26] makes use of this
relation between the spectrum of the intensity noise and the spectrum of the optical
field, which is also a critical part of one of the MPI measurement approaches discussed
in Section 15.4.
An intuitive explanation for this result goes as follows. Consider decomposing the
two interfering fields ε s (t) and ε r (t) into quasimonochromatic spectral components.
Although the two fields ε s and ε r are mutually incoherent, any two of their spectral
components will have a high degree of mutual coherence as a consequence of their
narrow bandwidth. Therefore interference between these components will produce
a strong RF intensity modulation at their difference frequency. Such modulation is
produced by all possible pairs of spectral components, which explains why the spectral
width of the SII is closely tied to the spectral width of the source. The fact that the
two beams are mutually incoherent means that all the RF components simply add on
an intensity basis, resulting in the autocorrelation of the source spectrum shown in
Eq. (15.9).

15.2.1.2. Two Fields with Lorentzian Lineshapes

Consider the case of a Lorentzian laser lineshape, typical for single-mode semicon-
ductor lasers. The field spectrum is [26]
 
2P 0 1
Sε s (f − fo ) =   ,
π f 
(15.10)
 f − fo 2 
1+
f/2

where P 0 is the average optical power, fo is the laser center frequency, and f is the
full-width at half-maximum laser linewidth. From Eq. (15.9), we can calculate the
4 This leads to an additional factor of two in Eq. (15.9) as the single-sided spectral density
is a factor of two larger than the double-sided spectral density, defined as nonzero for
f ∈ (−∞, ∞).
498 J. Bromage et al.

(single-sided) intensity noise spectrum. It is convenient to use a normalized quantity,

the relative intensity noise (RIN), defined as5
 
SII (f ) 4R1 R2 −G G 12
→12 ← − 1
RIN(f ) = 2
=    . (15.11)
P0 π f  f 2
1+
f

Because of the Lorentzian laser lineshape, the noise spectrum also has a Lorentzian
lineshape6 with a half-width at half-maximum linewidth of f . Furthermore, for this
case note that the total noise power, calculated by integrating SII (f ) over f ∈ [0, ∞),
is independent of f .
For a Lorentzian lineshape, closed-form expressions can be derived for arbitrary
delays between the two paths [27], not only the incoherent case considered above
where Td  1/f . In general, RIN is given by
 
4R1 R2 G 12 G 12 1
RIN(f ) = b −
→ ←−
   
π f  f 2
1+
f
× [sin2 (2πfo Td )(1 + e−4π f Td − 2e−2π f Td cos(2πf Td ))
f
+ cos2 (2πfo Td )(1 − e−4π f Td − 2e−2π f Td sin(2πf Td ))].
f
(15.12)
In contrast to Eq. (15.11), we see the mean optical frequency of the source fo appearing
in the expression for the RIN. This is because when Td is less than the coherence time,
the noise power spectral density depends on the average phase difference between
the mutually coherent fields. When Td is such that 2πfo Td = (n + 1/2)π where n is
any integer, RIN(f ) integrated over f ∈ [0, ∞) is maximum, and therefore there is
the maximum amount of intensity noise. In this case, the interferometer is said to be
“in-quadrature.”
Figure 15.5 shows RIN spectra, calculated for a Lorentzian source with f =
2 MHz and three different values of f Td . For all three cases the delay is such
that the interferometer is in-quadrature, R1 R2 G 12 G 12 = −40 dB, and the fields are
−
→ ← −
partially mutually coherent. However, there are clear differences between the spectra;
for both f Td = 0.2 and 0.02 we can see well-defined minima, located at harmonics
of 1/Td . This occurs when the delay between the two paths is an integer multiple
of the period of phase noise components. Then light from the two paths has phase
fluctuations at these frequencies that are in-phase and so the phase difference between
the optical fields varies less, resulting in less intensity noise. On the other hand, for
f Td = 2, we cannot resolve the oscillations, and so the RIN spectrum appears to
be identical to the case of complete mutual incoherence (f Td → ∞); that is, the
5 Note that RIN is used here as a spectral density and not as the dimensionless quantity
corresponding to the integral of RIN(f ) over frequency which is also used in the literature.
6 The convolution of two Lorentzians is itself a Lorentzian.
 15. Multiple Path Interference 499

–100
–110 ∆ f Td = 2
∆ f Td = 0.2

RIN [dB/Hz]
–120 ∆ f Td = 0.02
–130
–140
–150
–160
0 20 40 60 80 100 120
f, frequency [MHz]

Fig. 15.5. Calculated RIN for different values of f Td . The laser has a Lorentzian lineshape
with a full-width half-maximum, f = 2 MHz.

RIN spectrum is an autocorrelation of the Lorentzian source spectrum. At the other

extreme, the fields are completely mutually coherent if f Td = 0, (i.e., the source is
monochromatic) and therefore the RIN(f ) = 0 Hz−1 . In this chapter we are primarily
concerned with MPI in Raman amplifiers for which the degree of mutual coherence
is low, and so MPI between fields with a high degree of mutual coherence is not
considered further.

15.2.1.3. Multiple Interfering Fields

The result for the RIN spectrum shown in Eq. (15.11) can be applied to multiple
paths (N > 2) if the path difference between any pair of paths is longer than the
coherence length of the source [28]. Consider a total of M reflection sites, with any
pair specified by reflectivity coefficients Ri and Rj and with gain Gij between them.
This leads to N paths where N = M(M − 1)/2 + 1 (including the “straight-through”
path). Generalizing Eq. (15.11) gives [28]
  N i−1
4 1 
RIN(f ) =   2  Ri Rj G ij G ij . (15.13)
π f  i=2 j =1 −
→ ← −
1+
f

Note that for multiple incoherent paths, the frequency dependence of the intensity
noise does not depend on the number of interfering paths.

15.2.2. Intensity Distribution for N -Path Interference

In this section we discuss the time-domain view of intensity fluctuations produced by

MPI. In general, the intensity produced by N interfering fields is given as a function
of time by
, ,2
, 
N −1 ,
,− ,
I (t) = ,→ε s (t) + −
→
ε i (t), , (15.14)
, ,
i=1
500 J. Bromage et al.

20
N=2

Probability density
15
3
4
10

5
64

0
0.8 0.9 1.0 1.1 1.2
Normalized intensity

Fig. 15.6. Numerical simulations of the probability density functions for the normalized inten-
sity due to incoherent MPI. The number of paths (N ) is 2, 3, 4, 8, 16, 32, and 64. The crosstalk
ratio (RC ) is −30 dB for all cases.

where −→
ε s (t) is the main component of the signal. Each delayed, weakly interfering
component − →ε i (t) results from one of N − 1 paths. We now consider the worst case
of aligned polarizations,
, ,2
,+ 
N −1 + ,
, φ i (t) ,
I (t) = , Is e φ s (t)
jφ
+ Ii e jφ
, , (15.15)
, ,
i=1

where Ii is the intensity and φ i (t) is the time-dependent phase of light from the ith
path. If the fields are all mutually incoherent, the difference between any pair of
phases is uniformly distributed between 0 and 2π .
A statistical quantity relevant for predicting errors caused by fluctuations ofII (t) is
the probability density function (PDF). Figure 15.6 shows the resulting intensity PDF
for incoherent interference between N paths. Shown are PDFs resulting from different
numbers of paths, ranging from N = 2 to 64, which were calculated numerically
from an ensemble of 109 realizations for each case. We can express the amount of
interference in terms of a crosstalk ratio for the average intensities RC where


N −1
Ii
i=1
RC = . (15.16)
Is
Thus RC is the ratio of average intensities from all the delayed paths to the intensity
from the main signal path. In Fig. 15.6, RC = −30 dB for all cases. Also, for a
given value of N , it was assumed that Ii was the same for each path, and so RC =
(N − 1)Ii /Is .
As can be seen in Fig. 15.6, for the case of N = 2, the PDF has clearly defined
boundaries [29, 1, 30]. These correspond to cases of complete destructive or construc-
tive interference. For RC = −30 dB, the maximum and minimum possible values for
 15. Multiple Path Interference 501

log (Probability density)

0
Gaussian
–1
–2
–3
–4 N = 64
–5 Chi-square
–6
0.8 0.9 1.0 1.1 1.2
Normalized intensity

Fig. 15.7. Probability density function for 64 mutually incoherent paths for RC = −30 dB,
calculated using an ensemble of 109 realizations. Note the asymmetry of the PDF. Also shown
is the Gaussian and chi-square PDF.

√
the total intensity are (1 ± 0.001)2 Is ; values outside this range are not possible.
However, as more paths are added, the PDF extends beyond these boundaries, even
though RC is still −30 dB. The PDF spreads because RC depends on the total average
intensity of all the mutually incoherent delayed paths, but if at some instant all these
field components interfere constructively, the instantaneous intensity can be signifi-
cantly larger. Simply, the average intensity of N − 1 delayed paths with equal Ii is
I  = (N − 1)Ii , whereas the maximum instantaneous intensity, Imax = (N − 1)2 Ii , is
larger than the average by a factor of N√− 1 as it includes all beat terms. Therefore the
total intensity ranges over I = Is (1 ± (N − 1)Rc )2 . So, although in the frequency
domain the number of incoherent paths does not affect the MPI noise spectrum, in
the time domain PDF one can clearly see a dependence on the number of interfering
paths.
In the limit of large N , the PDF becomes Gaussian-like. Figure 15.7 shows the
PDF for 64 interfering paths on a logarithmic scale along with a Gaussian fit. Note that
unlike the Gaussian, the PDF is not symmetric about the mean intensity, but is skewed
to higher intensities. The exact form of this skewed PDF is called chi-square (χ 2 )
distribution (shown dotted in Fig. 15.7), and is also encountered in systems degraded
solely by amplified spontaneous emission (see, e.g., [31]), where the random ASE
optical field takes the role of the interferers’ fields. Intuitively, the skewness results
from the fact that the optical intensity, by definition, is a strictly positive quantity (cf.
Eq. (15.14)). In analogy to the two-path interference process (Eq. (15.4)), the total
intensity I (t) (Eq. (15.14)) can be written as
−
→ −
→
I (t) = |−
→
ε s (t)|2 + | n (t)|2 + 2Re[−
→
ε s (t) · n (t)], (15.17)
−
→
where n (t) denotes the sum of all interferers’ optical fields. The second and third
contributions to Eq. (15.17) can, respectively, be associated with the beating of MPI
with itself (MPI–MPI beat noise) and of the signal with MPI (signal–MPI beat noise).
−
→
In the case of a large number of interferers, n (t) takes on circularly symmetric
complex Gaussian (ccG) statistics due to the central limit theorem. If MPI–MPI beat
502 J. Bromage et al.

noise were neglected in Eq. (15.17), and if the signal field’s predominant source of
randomness is phase noise, I (t) would become the sum of the deterministic quantity
−
→
|−
→
ε s (t)|2 and the Gaussian-distributed7 real part of the ccG process − →
ε s (t) · n (t).
This superposition, again, leads to a Gaussian distribution (dashed line in Fig. 15.7).
With its tails extending infinitely in both directions, however, a Gaussian PDF would
allow for the occurrence of negative optical intensities. Small as it may seem when
comparing its variance to that of the signal–MPI beat noise,8 it is the MPI–MPI beat
noise that is responsible for positive intensities, which necessitates the observed skew
in the intensity’s exact PDF. Including the MPI–MPI beat noise does not affect the
overall appearance of the PDF, however, it significantly alters its tails. At a probability
level of 10−6 we find a deviation from the Gaussian of about an order of magnitude.
Approximating the exact PDF by a Gaussian may therefore yield slightly pessimistic
bit error predictions [32, 30].

15.2.3. Section Summary

In this section we have presented descriptions of the intensity fluctuation caused by

MPI-induced beat noise. First, in a frequency domain description, we calculated the
noise power spectral density of the beat noise for the case of interference between
two fields, showing that it can be related to the autocorrelation of the field spectrum
if the fields are mutually incoherent. We presented results for the case of Lorentzian
source optical spectra and showed how these results can be generalized for multiple
interfering fields where the number of paths N > 2. In the time domain, we calculated
the intensity probability density function, showing that its form depends on the number
of interfering fields. For the case of many paths (N
16), the PDF looks Gaussian-
like, but is more accurately represented by a chi-squared distribution.

15.3. Properties of Rayleigh Scattering

In this section we describe the nature of Rayleigh scattering, focusing on properties

that are relevant for our treatment of MPI resulting from double-Rayleigh backscat-
tering.

15.3.1. Dependence on Fiber Properties

Rayleigh scattering of light is elastic scattering caused by small-scale inhomogeneities

of the refractive index of the media. In an ideal medium that is perfectly homogeneous,
secondary radiation from material dipoles that are driven by the incident optical field
cancels in all but the propagation direction of the incident field. Inhomogeneities
7 Uniformly distributed phase noise on the signal field does not alter these statistics.
8 The ratio of signal–MPI beat noise to MPI–MPI beat noise amounts to 2000 in our example,
as can be evaluated with the help of Eqs. (15.68) and (15.69).
 15. Multiple Path Interference 503

cause imperfect cancellation and so light is scattered in other directions. The scale
over which inhomogeneities are correlated is much smaller than the wavelength of
the light, leading to the well-known wavelength dependence of 1/λ4 [24].
Multiple path interference in fibers results from Rayleigh backscattering, where a
fraction of the incident light is scattered in the opposite direction. The fraction of light
that is backscattered depends on two main factors: the composition of the glass and
the waveguide properties of the fiber. The amount of scattered optical power per unit
length is simply the loss coefficient due to Rayleigh scattering αR . For modern fibers
this is usually the dominant loss mechanism, but other mechanisms such as OH-ion
absorption, bend losses, and loss from draw-induced fluctuations of the core [33] may
also be significant depending on the fiber type and wavelengths of interest.
The recapture fraction S determines the fraction of scattered optical power that
is captured in a guided mode, propagating in the backscatter direction. To a first
approximation, we can assume that light scatters equally from all points in a fiber,
both radial and axial. Then for Gaussian transverse modes [I (r) ∼ exp(−r 2 /w 2 )],
the recapture fraction is [34]
3 3π
S= = , (15.18)
4n2 k 2 w 2 2n2 k 2 Aeff
where k = 2π/λ. From this expression we see that the larger the effective area Aeff ,
the smaller the amount of captured backscattered light. A more thorough treatment,
which takes into account the radial dependence of Rayleigh scattering and the exact
mode profiles, can be done by integrating the overlap of an optical mode with the
radial composition profile of a fiber [35–37]. This approach can be used, for example,
where fiber index profiles are produced using dopants with significantly different
Rayleigh scattering properties from the silica host, or for fibers with modes that are
far from Gaussian.
The product of the Rayleigh scattering loss and recapture fraction αR S repre-
sents the amount of Rayleigh backscattered power per unit length of the fiber, often
referred to as the Rayleigh backscatter coefficient. This parameter is important for
predicting the amount of DRB-induced MPI in fiber amplifiers and, in general, may
be z-dependent (e.g., if multiple fiber types are used within an amplifier).

15.3.2. Rayleigh Scattered Power Within Fiber Amplifiers

In this section we examine the average optical powers of the single-Rayleigh backscat-
tered (SRB) and double-Rayleigh backscattered light inside a fiber amplifier. We
present differential equations that can be used to predict these optical powers at any
point in the amplifier, given the gain, loss, and Rayleigh backscattering properties.
Then, using these equations, we identify the most significant DRB paths for DRB-
induced MPI in Raman amplifiers.
Consider a fiber amplifier extending from z = 0 to z = L. The net gain per unit
length is g (z) for light propagating in the forward direction (increasing z) and g (z)
−
→ ←−
504 J. Bromage et al.

for the backward direction (decreasing z). The Rayleigh backscattering coefficient
per unit length is αR S. Here we consider the optical powers:
1. P s (z), the average forward propagating signal,
2. PSRB (z), the single-Rayleigh backscattered light, and
3. PDRB (z), the double-Rayleigh backscattered light.
Both SRB and DRB are stationary stochastic processes and so the ensemble-averaged
powers PSRB and PDRB are constant in time (see Section 15.3.3.2). These optical
powers are governed by the following coupled differential equations.

dz P s = g (z)P s (z) (15.19)

−
→
−dz PSRB = g (z)PSRB (z) + αR SP s (z) (15.20)
←
−
dz PDRB = g (z)PDRB (z) + αR SPSRB (z), (15.21)
−
→
where dz is the operator d/dz. Higher-order scattering may be ignored if the combi-
nation of gain and backscattering is sufficiently weak. This is a realistic assumption
because significant DRB-induced beat noise occurs when the ratio of PDRB to P s
is only −25 dB (see Section 15.5.2.3). The boundary values for the scattered fields
are PSRB (L) = 0, and PDRB (0) = 0. By eliminating PSRB , we can solve for PDRB in
terms of P s at the output of the amplifier z = L. The ratio of these powers is
L z2
PDRB (L)
= dz2 dz1 (αR S)2 G (z1 , z2 ) G (z1 , z2 ), (15.22)
P s (L) 0 0 −
→ ←−

where G (z1 , z2 ) is the intensity gain going forward from z1 to z2 , and G (z1 , z2 )
−
→ ←−
is the gain going backwards from z2 to z1 . This notation allows Eq. (15.22) to be
used for arbitrary nonreciprocal gain distributions, such as occur when isolators are
positioned within(the amplifier. For
) the amplifier described
( by Eqs. (15.19))to (15.21),

z
z
G (z1 , z2 ) = exp z12 g (z
)dz
and G (z1 , z2 ) = exp − z21 g (z
)dz
.
−
→ −
→ ←− ←−
In Raman amplifiers, the gain distribution along the fiber depends on the pump
distribution. Consider a simple amplifier formed from a single fiber so that αR S does
not depend on z. When this fiber is counter-pumped using one pump wavelength, the
small-signal9 pump power distribution is Pp (z) = Pp (L) exp[−αp (L − z)] where αp
is the loss at the pump wavelength. g (z) and g (z) are given by CR Pp (z) − αs ,
−
→ ←
−
where CR is the Raman gain efficiency of the fiber [(W · km)−1 ] and αs is the signal
loss coefficient [km−1 ]. Hence we can write the round-trip gain term in the integrand
of Eq. (15.22) as

G (z , z )G (z1 , z2 ) = exp[2CR Pp (L) exp(−αp L)(eαp z2 − eαp z1 )/αp − αs (z2 − z1 )].

−
→ 1 2← −
(15.23)
9 “Small-signal” means the power of the signals is sufficiently weak that that pump power
can be assumed to be undepleted by the signals.
 15. Multiple Path Interference 505

100
500
DFB
400 ECL
80 10

G (∆z) [km]
300 Net amplifier gain = 0 dB
2 6 20
4 40
z 1 [km] 60 200 –2

100 –4
–6
–8
40 0 –20dB
0 20 40 60 80 100
∆z [km]
20
Contour plot of G (z 1,z 2) G (z 1,z 2)
=z
2
z1
0
0 20 40 60 80 100
z 2 [km]

Fig. 15.8. Small-signal round-trip gain between reflection sites z1 and z2 in a 100-km span
that is counter-pumped to transparency. Inset shows the round-trip gain integrated over z1 ,
and plotted as a function of z = z2 − z1 for net gains ranging from −20 dB (no Raman
pumping) to 0 dB (pumped to transparency). Dashed arrows indicate z values that equal half
the coherence lengths of two common sources.

This is the round-trip gain between Rayleigh backscatter sites at z1 and z2 , where
z1 ≤ z2 . By examining the round-trip gain for all possible pairs of z1 and z2 we
can find the paths that are most important. This is shown in Fig. 15.8, which is a
contour plot of G (z1 , z2 ) G (z1 , z2 ) versus z1 and z2 for an amplifier that is pumped
−
→ ←−
to transparency ( G (0, L) = 0 dB). Other parameters are L = 100 km, αs = 0.2 dB/km,
−
→
αp = 0.25 dB/km, and CR Pp (L) = 0.265 km−1 .
First note that the round-trip gain is only significant within the region of high
pump power, roughly a distance of 1/αp (with αp in km−1 ) from the pump input at
z = L. Secondly, the round-trip gain is not significant if z1 = z2 , marked by the dashed
line. The separations of z1 and z2 that give the largest contributions to Eq. (15.22) can
be found using a new variable z = z2 − z1 , which is the distance between Rayleigh
backscatter sites. Then the integrated round-trip gain can be rewritten as
L z2
dz2 dz1 G (z1 , z2 ) G (z1 , z2 )
0 0 −
→ ←−
L L−z
= d(z) dz1 G (z1 , z1 + z) G (z1 , z1 + z),
0 0 −
→ ←−
L
= d(z) G
(z). (15.24)
0 ←
→
506 J. Bromage et al.

The variable G
(z) represents the contribution to the integrated round-trip gain
←→
from all paths with the same z. From Eq. (15.23), G
(z) is given by
←
→
.
G
(z) = e2αs z Ei(2CR Pp (L)(1 − e−αp z )/αp )
←→
/
−Ei(2CR Pp (L)e−αp L (eαp z − 1)/αp ) /αp , (15.25)

∞
where Ei(z) is the exponential integral function defined by Ei(z) = − −z e−t /t dt
for z > 0. The inset in Fig. 15.8 shows G
(z) for a range of amplifier net gains
←→
corresponding to no Raman pumping ( G (0, L) = −20 dB) up to pumping to trans-
−
→
parency ( G (0, L) = 0 dB).
−
→
In Fig. 15.8, the dashed arrows indicate values of z corresponding to round-
trip distances equal to half the coherence length of two types of sources: distributed
feedback lasers (DFB) and external-cavity lasers (ECL). Typical DFB linewidths are
approximately 1 MHz, giving coherence lengths in the fiber on the order of 200 m,
whereas ECLs have narrower linewidths of approximately 25 kHz, giving coherence
lengths on the order of 8 km. For ECL sources and no Raman pumping, 60% of
the contribution to PDRB comes from round-trip distances that are greater than the
source coherence length. When the span is pumped to transparency, this fraction
increases to 96%. For DFB sources, the shorter coherence length means that more
than 98% of the contribution to PDRB comes from round-trip distances greater than
the coherence length, even without Raman pumping. Therefore, for most relevant
situations, the signal and DRB light are mutually incoherent when they interfere to
produce intensity noise.

15.3.3. Spectral Properties of Rayleigh Backscattered Light

The spectral properties of DRB light play an important role, both in the quantification
of DRB-induced MPI and in assessing its impact on system performance. In this
section, we present results showing that the intensity noise spectrum produced by
DRB-induced MPI can be identical to that of incoherent two-path interference (treated
in Section 15.2.1.1) [38, 39]. Therefore, when considering DRB-induced beat noise in
the frequency domain, DRB light can be treated as having the same optical spectrum
as the source.
Consider a fiber span from z = 0 to L as shown in Fig. 15.9. In addition to light
propagating straight through the fiber (with optical field − →
ε s (L, t)) there will be light at
the output from double-Rayleigh scattering within the span. To obtain an expression
for the optical field of the DRB light we first note that the field at z1 that is single-
Rayleigh backscattered from an element L at z2 is given by
*
−→ ρ (z2 ) G (z1 , z2 ) −
ε SRB (z1 , z2 , t) = J21 ρ →
ε s (z2 , t − (z2 − z1 )/vg )e−jβ(z1 −z2 ) ,
←
−
(15.26)
where J21 is the Jones matrix describing the polarization evolution as a result of
 15. Multiple Path Interference 507

∆L ∆L

z
z1 z2
εs
∆ εSRB

∆εDRB
∆
(z1) ∆
(z2)

Fig. 15.9. Schematic showing DRB within a fiber amplifier. Highlighted is the path showing
Rayleigh backscattering in a length L at z2 with (field) reflectivity ρ
ρ (z1 ), followed by
another similar scattering event at z1 .

propagation from z2 to z1 . The field Rayleigh backscatter coefficient of a short section

of fiber ρρ (zi ) has circular complex Gaussian statistics (ccG) when L is shorter
than the source coherence length but longer than the correlation distance of refractive
index fluctuations [38]. The variable vg is the group velocity, G (z1 , z2 ) is the intensity
←−
gain when propagating backwards from z2 to z1 , and β is the propagation constant
of the optical mode.
A contribution to the total DRB field from one particular path, shown in Fig. 15.9,
comes from backscattering in two short sections of fiber of length L at positions z1
and z2 . This contribution to the total field at z = L is
*
−→ε DRB (L, z1 , z2 , t) = J12 J21 ρ
ρ (z1 )ρ
ρ (z2 ) G (z1 , z2 ) G (z1 , z2 )
−
→ ←−
×−→
ε (L, t − 2(z − z )/v )ej 2β(z2 −z1 ) ,
s 2 1 g (15.27)

where G (z1 , z2 ) is the intensity gain when propagating forwards from z1 to z2 . The
−
→
round-trip delay is 2(z2 − z1 )/vg . This treatment neglects spectral distortion of both
the signal and DRB fields on propagating through the fiber. The total double-Rayleigh
field is given by integrating over all possible z1 and z2 as
L z2 *
−
→
ε DRB (L, t) = dz2 dz1 J12 J21ρ (z1 )ρ
ρ (z2 ) G (z1 , z2 ) G (z1 , z2 )
0 0 −
→ ←−
×−
→
ε s (L, t − 2(z2 − z1 )/vg )ej 2β(z2 −z1 ) . (15.28)

Here ρ ρ (zi ) has been replaced by a differential Rayleigh backscatter coefficient

ρ (zi ) = limL→0 [ρ ρ (zi )/L]. We can continue to assume ccG statistics for this
differential coefficient because scales over which there are changes in field amplitude
and polarization are much larger than the correlation distance of refractive index
508 J. Bromage et al.

fluctuations [38]. Denoting ρ (zi ) as ρ i , we can therefore write the following useful-
relations that follow from ccG statistics [40].

ρ
ρ T1ρ 2  = αR S δ(z1 − z2 ) (15.29)
ρ T1ρ T2ρ 3ρ 4 
ρ = (αR S) [δ(z1 − z3 )δ(z2 − z4 ) + δ(z1 − z4 )δ(z2 − z3 )]
2
(15.30)
ρ 1ρ 2ρ 3ρ 4  = 0,
ρ (15.31)

where δ(z) Dirac’s δ-functional.

15.3.3.1. Unmodulated Source

As an intermediate step in obtaining the spectrum of DRB-induced beat noise, we first

calculate the autocorrelation of intensity fluctuations =II (τ ), following the approach
used in Section 15.2.1. We can obtain a closed-form expression assuming that the
DRB field is completely polarized with the same state of polarization as the signal10
and the source field to be WSS with a correlation length much shorter than paths that
give significant contributions11 to −
→ε r (L, t). As shown in Appendix A15.A, we get
L z2
=II (τ ) = dz2 dz1 2(αR S)2 G (z1 , z2 ) G (z1 , z2 ) |=ε s (τ )|2 (15.32)
0 0 −
→ ←−

for the autocorrelation of the intensity fluctuations II . Note once again that =II (τ )
can be expressed in terms of the magnitude-squared field autocorrelation function
|=ε s (τ )|2 . Fourier transforming Eq. (15.32) gives the (single-sided) noise power spec-
tral density for DRB-induced MPI beat noise
L z2
SII (f ) = dz2 dz1 4(αR S)2 G (z1 , z2 ) G (z1 , z2 ) [Sε s (f ) ∗ Sε s (f )].
0 0 −
→ ←−
(15.33)
Note the similarity between this equation and Eq. (15.9). This shows that DRB-
induced beat noise has a noise power spectrum with the same frequency dependence
as the self-homodyne spectrum of two-path incoherent interference.
As a simple example, consider the case of a Lorentzian source with linewidth f ,
propagating in a hypothetical distributed amplifier where gain is distributed uniformly
along the length of the fiber and is independent of direction of propagation (i.e., recip-
rocal). Thus G (z1 , z2 ) = G (z1 , z2 ) = exp[g(z2 − z1 )], where g is a z-independent
−
→ ←−
coefficient that specifies the gain per unit length. Then, if αR S is also uniform, the
RIN from double-Rayleigh MPI is
 
4(αR S)2 e2gL − 2gL − 1 1
RIN(f ) = 2
. (15.34)
π f 4g 1 + ( f
f 2
)

10 This restriction is lifted in Section 15.3.4.

11 This assumption is discussed in Section 15.3.2 along with Fig. 15.8.
 15. Multiple Path Interference 509

In some situations, the optical spectrum of double-Rayleigh scattered light may

be significantly different from the source spectrum. For example, this can occur when
the single-backscattered light copropagates with a noisy pump laser. If the dispersion
properties of the fiber are such that the walk-off between the pump and scattered
signal light is small (as is the case when the fiber’s zero dispersion wavelength falls
directly between pump and signal), high-bandwidth pump noise amplitude-modulates
the single-scattered signal light, and can significantly broaden its spectrum. For certain
systems this can be beneficial as it reduces the spectral overlap between the signal
and double-Rayleigh scattered light [41] .

15.3.3.2. Modulated Source

To assess the impact of DRB on system performance in Section 15.5 we will have to
know the autocorrelation of the DRB optical field ε DRB (t) for data-modulated optical
signals,
=ε DRB (t, t + τ ) = εε TDRB (t)εε DRB (t + τ ). (15.35)
To derive an expression for =ε DRB (t, t + τ ), we start by using the scalar version of
Eq. (15.28), assuming that the state of polarization does not change during propa-
gation (J12 = J21 = I), and we neglect laser phase noise; that is, we assume a purely
deterministic signal field,12 ε s (t) = εs (t). Proceeding similarly to the derivation of
Eq. (15.32), we arrive at
L z2
=ε DRB (t, t + τ ) = (αR S)2 dz2 dz1 εsT (L, t − 2[z2 − z1 ]/vg )
0 0
× εs (L, t + τ − 2[z2 − z1 ]/vg )
× G (z1 , z2 ) G (z1 , z2 ). (15.36)
−
→ ←−
Substituting ξ = 2(z2 − z1 )/vg , we can rewrite this expression as
L 2z2 /vg
2 vg
=ε DRB (t, t + τ ) = (αR S) dz2 dξ εsT (L, t − ξ )εs (L, t + τ − ξ )
2 0 0
× G (z2 − vg ξ/2, z2 ) G (z2 − vg ξ/2, z2 ).
−
→ ←−
(15.37)
The ξ -integration can then be interpreted [39] as the convolution of the functions
x(t; τ ) = εsT (L, t)εs (L, t + τ ) and y(t; z2 ) = G (z2 − vg t/2, z2 ) G (z2 − vg t/2, z2 )
−
→ ←−
rect(t, [0, 2z2 /vg ]), with τ and z2 acting as parameters, and the windowing func-
tion defined as rect(t, [0, 2z2 /vg ]) = 1 in [0, 2z2 /vg ] and 0 elsewhere. For double-
Rayleigh backscatter in optical fibers as the source of MPI, the function y(t, z2 ) has
low-pass characteristics, with cut-off frequencies much lower than those of typical

12 Performing the more complicated derivations with phase noise included, it turns out that
the result is unaffected by phase noise for realistic data modulation bandwidths that exceed
the laser linewidth.
510 J. Bromage et al.

data signals.13 Therefore the convolution becomes time-independent and can be re-
placed by the product of the time average of x(t; τ ) and the integral over y(t; z2 ),

1 T /2 T
x(t; τ ) ∗ y(t; z2 ) ≈ lim εs (L, t)εs (L, t + τ )dt
T →∞ T −T /2
4 56 7
=εs (τ )
2z2 /vg
× G (z − vg t/2, z2 ) G (z2 − vg t/2, z2 )dt. (15.38)
0 −
→ 2 ← −

The first term on the right-hand side of this equation is identified as the temporal
autocorrelation =εs (τ ) of the deterministic signal field, and the DRB field’s autocor-
relation becomes
=ε DRB (t, t + τ ) = =ε DRB (τ ) =
2z2 /vg
vg L
=εs (τ ) · (αR S)2 dz2 dξ G (z2 − vg ξ/2, z2 ) G (z2 − vg ξ/2, z2 ).
2 0 0 −
→ ←−
(15.39)

Several interesting properties of DRB can be seen from Eq. (15.39). First, because
the DRB autocorrelation depends on the time difference τ only, the stochastic process
is (at least wide-sense) stationary [42], with power

PDRB = =ε DRB (0) =

L 2z2 /vg
vg
= P s (αR S)2 dz2 dξ G (z2 − vg ξ/2, z2 ) G (z2 − vg ξ/2, z2 ),
2 0 0 −
→ ←−
(15.40)

where P s = =εs (0) is the average signal power.

Second, owing to the Fourier relationship between =εs (τ ) and the signal’s power
spectrum,

1 T /2
F{=εs (τ )} = lim
T →∞ T −T /2
=εs (τ ) exp[−j 2πf τ ]dτ = | s (f )|2 , ε
(15.41)

with
T /2
1
F{εs (t)} = lim
T →∞ T
εs (t) exp[−j 2πf t]dt = εs (f ), (15.42)
−T /2

13 Assuming a typical exponential gain/loss profile G (z , z ) = G (z , z ) = exp[g|z − z |]

−
→ 1 2 ←
− 1 2 2 1
with −2.3 · 10−4 m−1
g
0.06m−1 , corresponding to the extreme cases of 1 dB/km fiber
loss (lower limit) and pumping a high Raman efficiency fiber with 10 W of pump power
(upper limit), we arrive at time constants on the order of 100 ns and above. Concerning the
windowing function, a 10 ns wide window corresponds to the backscatter originating from
only the last meter of optical fiber, which can be safely neglected in the scattering integral.
 15. Multiple Path Interference 511

0
RZ 33% CSRZ 67% RZ-DPSK
–20

Power density
[dBm/0.01nm]
Input
–40

–60
Rayleigh
–80
1549.5 1550.0 1550.5 1551.0 1549.5 1550.0 1550.5 1551.0 1556.5 1557.0 1557.5 1558.0

Wavelength [nm] Wavelength [nm] Wavelength [nm]

Fig. 15.10. Measured Rayleigh backscattered spectra (lower traces) and spectra entering the
fiber (upper traces) for typical 40-Gbit/s amplitude and phase modulated optical data signals.
Note that Rayleigh scattering preserves the full information content of the signals’spectral mag-
nitudes. For formats with isolated spectral components, spontaneous Brillouin scattering lines
are additionally observed in the Rayleigh scattered spectra to both sides of these discrete tones.

the magnitude of the DRB power spectrum can be obtained from the modulated sig-
nal’s power spectrum. Rayleigh backscattering thus preserves the full information
content of the data signal’s spectral magnitude, independent of whether the signal
is phase or amplitude modulated. To illustrate this point, Fig. 15.10 shows measure-
ments of optical data spectra taken at 40 Gbit/s for on-off keying (OOK) signals (33%
duty cycle return-to-zero (RZ) and 67% duty cycle carrier-suppressed return-to-zero
(CSRZ) formats), and for phase-coded signals (33% duty cycle return-to-zero dif-
ferential phase shift keying (RZ-DPSK) format). The upper traces give the signals’
spectra at the fiber input, and the lower traces refer to the single-Rayleigh backscat-
tered signals. Note that the spectral shapes are accurately preserved by Rayleigh
scattering, irrespective of whether amplitude or phase is modulated. For formats that
have isolated spectral tones (33% RZ and 67% CSRZ), it is also possible to see spon-
taneous Brillouin scattering lines positioned 10 GHz above and below the discrete
spectral tones.14
Knowing that the DRB process is stationary with power spectral density | s (f )|2 , ε
it is also possible to generate realizations in the frequency domain for use in Monte
Carlo simulations [43]. Thereby, the magnitudes of the Fourier components are
ε
weighted by | s (f )|2 , and their phases are chosen to be independent15 random vari-
ables, uniformly distributed between 0 and 2π .

15.3.4. State and Degree of Polarization

Thus far we have ignored the way the polarization state of light can change when
propagating in an optical fiber. For simplicity, we have assumed that the Jones matrices
affecting the DRB field (J12 and J21 ) are the identity matrix I. However, we must
14 The absence of stimulated Brillouin scattering (SBS) was verified by varying the optical
power going into the fiber.
15 The statistical independence of any two spectral components is a direct consequence of
stationarity.
512 J. Bromage et al.

take polarization effects into account because intensity noise from beating between
−
→
ε1 and − ε2 depends on the scalar product of their polarization vectors −
→ →
p1 · −
→
p2 T . In
this section we discuss the impact of fiber birefringence on the state and degree of
polarization of the DRB field, following the treatment of van Deventer [44].
We use the Stokes representation for describing the state of polarization (SOP)
and degree of polarization (DOP) of light [26]. For a SOP defined by a Stokes vector
−
→
on the Poincaré sphere as S = (S0 , S1 , S2 , S3 ), the DOP is given by
*
S12 + S22 + S32
DOP = . (15.43)
S0
The component S0 represents the average power, which we can use to normalize all
−
→
components of S so that S0 = 1. The evolution of the Stokes vector on the Poincaré
sphere is determined by Müller matrices. Two types are needed here: the matrix Mt
describes transmission through fiber and Mr describes a reflection. (Here we are
ignoring polarization-dependent gain or loss.) These matrices are given by
   
1 0 0 0 1 0 0 0
 0 m1 m2 m3   0 1 0 0 
Mt =   
 0 m 4 m 5 m 6  and Mr =  0 0 1 0  .
 (15.44)
0 m7 m8 m9 0 0 0 −1

The transmission matrix Mt is an orthogonal matrix (MTt = M−1 t ) with det(Mt ) = 1.

The elements of the reflection matrix Mr show that for a (lossless) reflection, only the
handedness of the state of polarization is changed; that is, S3 → −S3 . This point is
key to understanding the state and degree of polarization of double-Rayleigh scattered
light.
First consider the case of single-Rayleigh backscattered light. Light first prop-
agates forward through the fiber, is reflected at some point, and then propagates
backward through the same section of fiber. This path can be described by MSRB ,
given by
 
1 0 0 0
 0 m 2 m 7m 8 m 7m 9 
 7 
MSRB = MTt · Mr · Mt = I − 2  , (15.45)
 0 m 7m 8 m 28 m 8m 9 
0 m 7m 9 m 8m 9 m 29

where I is the identity matrix. The total Rayleigh backscattered field is the sum of
contributions from all points in the fiber. Significant contributions come from points
within the effective length of the fiber, which for Rayleigh backscattering in passive
fiber is (1 − e−2αL )/2α [34]. If the coherence length of the source is much smaller than
this, the Stokes parameters are additive and the sum over all contributions depends
only on the average of the matrices from each reflection point.
In [44], results are given for fiber with varying degrees of birefringence: zero, low,
and high. Here we present results for low birefringence fiber as this is the most relevant
 15. Multiple Path Interference 513

for telecommunications applications where polarization beat lengths are typically

on the order of 10 m [45]. For fiber where the beat length is much less than the
effective length, Müller matrices for transmission through sections of fiber are random
orthogonal matrices [46]. From MTt · Mt = I it follows that m 21 +m m 22 +m m 23 = 1, m 24 +
m 5 +m
2 m 6 = 1, and m 7 +m
2 2 m 8 +m
2 m 9 = 1. As different matrix elements are uncorrelated
2

and have the same variance, we have m m 2i  = 1/3, and m

m im j  = 0 for i , = j . Therefore,
for low-birefringence fiber,
 
1 0 0 0
 0 1 0 0 
MSRB  =  3
 0 0 1 0 .
 (15.46)
3
0 0 0 13
Here we see two important results. First, because the Müller matrix is diagonal, the
SOP for backscattered light is the same as the input SOP. Second, the DOP is reduced
by a factor of three; for completely polarized input light (DOP = 1), the DOP for the
backscattered light is 1/3.
Brinkmeyer [47] gives the following explanation for the depolarizing mechanism.
As signal light propagates down a low birefringence fiber, the SOP evolves, sampling
all points on the Poincaré sphere equally. If the light is linearly polarized when re-
flected (S3 = 0), the SOP is preserved, but if it is circularly polarized (S1 = S2 = 0),
the handedness is changed from left- to right-handed or vice versa, to give the orthog-
onal polarization. (See the sign of the elements of Mr , Eq. (15.44).) It is this change
in the polarization state that depolarizes the Rayleigh backscattered light. However,
as the polarization vector is more likely to cross the equator of the Poincaré sphere
(linearly polarized) than to arrive at the poles (circularly polarized), it is more likely
that the original state of polarization is preserved on reflection. Therefore the Rayleigh
backscattered light is not completely depolarized, but is partially polarized with the
same SOP as the source.
Next we apply these results to the case of DRB. We assume again that both the
source coherence length and the fiber beat length are much less than the effective
length over which significant contributions originate (see Section 15.3.2). Then the
average Müller matrix for DRB light is given by
 
1 0 0 0
 0 1 0 0 
MDRB  =   0 0 1 0 ;
9  (15.47)
9
0 0 0 19
that is, the SOP is the same as the output signal light, and the DOP is 1/9.
This depolarizing effect means that there is less intensity noise from DRB-induced
MPI than was derived by assuming no change in the polarization along the fiber. We
can weight the previous results (Eqs. (15.32) to (15.34), (15.39), (15.40)) with a
depolarizing factor to take this into account. If there were complete depolarization
(DOP = 0), we could decompose the double-Rayleigh light into two uncorrelated
halves, one with the same polarization as the signal and the other with the orthogonal
514 J. Bromage et al.

polarization. In this case, the factor would be 1/2. However, as the DOP equals 1/9
and the polarized component has the same SOP as the signal light, the correct factor is
1/9 plus half of the remaining 8/9 that is depolarized. Thus the total factor, as shown
in [44], is (1/9) + (1/2) · (8/9) = (5/9).

15.3.5. Section Summary

In this section we presented some of the properties of Rayleigh scattering in single-
mode optical fibers that are relevant for DRB-induced MPI. We first discussed the
fiber properties (such as composition and waveguide design) that dictate the amount
of Rayleigh backscattered light. Next, we presented differential equations showing
the coupling between the average signal power and the single- and double-Rayleigh
backscattered power at any point in a fiber. Using these equations, we showed that
the most significant paths for DRB-induced MPI in Raman amplifiers have a length
of several kilometers, much longer than a typical source coherence length. Next
we analyzed the spectral properties of Rayleigh backscattered light, with and without
signal modulation, showing that the spectrum of the intensity fluctuations due to DRB
is closely related to the source spectrum, analogous to the case of self-homodyning.
Finally, we presented a polarization analysis that shows how DRB light is depolarized
if the fiber is birefringent.

15.4. Measurement of MPI Levels

In this section we describe ways to quantify the level of multiple path interference.
Then we discuss two techniques that are used to measure the level of MPI in fiber
amplifiers.

15.4.1. Definition of Optical Signal-to-Noise Ratio and Crosstalk Ratio for MPI
Two types of quantities are used to describe the level of MPI. One type is the optical
signal-to-noise ratio due to MPI (OSNRMPI ) which is the ratio of average signal power
P s to the average MPI power PMPI (including all polarizations),
Ps
OSNRMPI = . (15.48)
PMPI
This is similar to the commonly used definition of OSNR due to ASE,
Ps
OSNRASE = , (15.49)
2NASE Bref
where NASE is the ASE power spectral density per (polarization) mode, and Bref is the
reference optical bandwidth, typically 12.5 GHz. The other type is a crosstalk ratio
RC , which is simply the inverse of OSNRMPI ,
PMPI 1
RC = = . (15.50)
Ps OSNRMPI
 15. Multiple Path Interference 515

The crosstalk ratio is predominantly used when describing the performance of a single
component, whereas OSNRMPI is generally favored when considering the cumulative
effect of many components or amplifiers in a system [14].
Care must be taken when using these quantities to state assumptions for the SOP
and DOP for the MPI light. Here we distinguish between definitions that use the total
MPI power in all polarizations, and definitions that only use MPI power with the
same polarization as the signal. For the latter we add a “P ” superscript. For example,
OSNRPMPI and RCP are the ratio of the signal power to the MPI power with the same
P ,
polarization as the signal PMPI

1 Ps
OSNRPMPI = P
= P . (15.51)
RC PMPI

As MPI only results in intensity noise from beating with the signal when there is some
degree of polarization overlap, OSNRPMPI and RCP are the more relevant quantities. To
highlight the symmetry between ASE and MPI-induced beat noises, we also define
P
OSNRASE as
Ps
OSNRASEP
= . (15.52)
NASE Bref
Example 15.1 (Two Discrete Reflectors). Consider the case of gain between two
discrete reflectors, R1 and R2 , shown in Fig. 15.4. For sufficiently weak reflec-
tions, R1 R2 G 12 G 12 # 1, so we only need to consider the path that double-passes
−
→ ← −
the gain medium. The crosstalk ratio RC for this device is then simply given by
RC = R1 R2 G 12 G 12 . On the other hand, the polarization-dependent crosstalk ratio
−
→ ← −
RCP depends on the polarization overlap between the main signal and the delayed
component. If they are both completely polarized with the same SOP, RCP = RC , but
if they are orthogonal, RCP = 0.

Example 15.2 (Double-Rayleigh Backscatter). Another useful example is the case of

distributed reflections from DRB throughout a fiber amplifier. Consider the amplifier
described in Section 15.3.2, which has a length L, and gain and Rayleigh backscat-
tering per unit length at position z are given by g(z) and αR S, respectively. From
Eqs. (15.22) and (15.50) we get
L z2
PDRB (L)
RC = = dz2 dz1 (αR S)2 G (z1 , z2 ) G (z1 , z2 ). (15.53)
P s (L) 0 0 −
→ ←−

A couple of useful analytic results can be obtained from Eq. (15.53). For the simple
case of uniform gain and backscattering per unit length, αR S and g are z-independent
constants. Then RC is given by
 2gL 
2 e − 2gL − 1
RC = (αR S) . (15.54)
4g 2
516 J. Bromage et al.

Note that this result appears as a factor in Eq. (15.34) for the RIN from such an
amplifier. This connection between RIN and RC (or OSNRMPI ) forms the basis of the
electrical beat-noise measurement technique, described in Section 15.4.2. Using the
results of Section 15.3.4, we can take into account depolarization effects in the fiber,
and write  2gL 
5 5 e − 2gL − 1
RCP = RC = (αR S)2 . (15.55)
9 9 4g 2
In Raman amplifiers, the gain distribution along the fiber depends on the pump
distribution. For an amplifier pumped with a single wavelength, g(z) is given by
CR Pp (z) − αs , where CR is the Raman gain efficiency of the fiber [(W · km)−1 ],
Pp (z) is the pump power distribution, and αs is the signal loss coefficient [km−1 ]. If
we assume the pump and signal loss are the same (αp = αs = α), in the small-signal
limit Eq. (15.53) can be solved, giving

5(αR S)2 −(αL−2θ e−αL )  2 αL

RCP = e 4θ (e + 2θ ) {Ei(2θ ) − Ei(2 θ e−αL )}
72 (αθ )2
−αL −αL

+ (1 + 2θ )(eαL+2θ e − eαL+2θ − 2θ e2θ + 2θ eαL+2θ e ) , (15.56)

where we used a dimensionless parameter θ = CR Pp (L)/α to

simplify notation, and
∞
Ei(z) is the exponential integral function defined by Ei(z) = − −z e−t /t dt for z > 0.

15.4.2. Electrical Beat-Noise Measurement Technique

In this section we describe a technique to measure the amount of DRB-induced MPI

produced by fiber amplifiers. The goal is to measure RC (or equivalently, its inverse,
OSNRMPI ) at the output of the amplifier. Unfortunately, one cannot simply use an
optical spectrum analyzer to measure OSNRMPI as is routinely done when measuring
the ratio of signal light to ASE. This is because the DRB light responsible for MPI-
induced beat noise has the same spectrum and state of polarization as the signal light
(see Sections 15.3.3 and 15.3.4) and so the DRB light cannot be distinguished from
the signal.16 A different approach must be used.
This section presents an overview of an electrical beat-noise measurement tech-
nique [48, 43, 49], so called because this approach measures the level of MPI by
quantifying beat noise added to the electrical output of a photodetector. From this,
one can infer the power ratio of the signal light to double-Rayleigh light. Because
no interference noise is produced by the polarization component of double-Rayleigh
light that is orthogonal to the signal polarization state, this technique measures the
ratio of signal light to DRB light with the same polarization state. Therefore, using
the notation of the previous section, the quantity that is measured by this technique
is RCP (or, equivalently, OSNRPMPI ).
16 In principle, one can use a polarization nulling technique, analogous to that used for measur-
ing OSNRASE [26]. By placing high-extinction polarization-selective components before
the optical spectrum analyzer, one can measure the component of the DRB power that is
orthogonally polarized to the signal.
 15. Multiple Path Interference 517

A B C

DFB DUT ESA

VOA1 EDFA VOA2

OSA

Fig. 15.11. Setup for measuring MPI in the device-under-test using the electrical beat-noise
measurement technique. RIN measurements are made using the ESA (which consists of a
square-law photodetector, low-noise RF amplifier, optical power meter, and an electrical spec-
trum analyzer). An optical spectrum analyzer measures the OSNRASE at point C to calculate
the RIN contribution from signal–ASE beating (RINs-ASE ).

A typical measurement setup is shown in Fig. 15.11 where the amplifier in ques-
tion is the device-under-test (DUT) between points A and B. The device is probed
using a low-noise CW source at the signal wavelength. Its linewidth needs to be
large enough that its coherence length is much shorter than significant round-trip
paths in the amplifier. This ensures that the DRB light is incoherent with respect to
the straight-through signal light, greatly simplifying the analysis of the noise power
spectrum (see Section 15.2.1). For Raman amplifiers made from several kilometers of
fiber, linewidths greater than 1 MHz suffice.17 Therefore, distributed feedback semi-
conductor lasers with RIN levels below −155 dB/Hz and 1 to 10 MHz linewidths are
ideal. MPI generated in shorter amplifiers, such as EDFAs made from a few meters of
erbium-doped fiber, can also be measured if the linewidth of the source is sufficiently
broad [50].
After passing through the device-under-test, the signal may need to be optically
amplified to a level suitable for the RIN-measurement apparatus. This apparatus,
labeled ESA, consists of a high-bandwidth photodetector, optical power meter, low-
noise electrical preamplifier, and electrical spectrum analyzer [51, 26]. An optical
spectrum analyzer (OSA) may be used to measure the signal-to-ASE ratio at point C
of Fig. 15.11 to determine the noise contribution from signal–ASE beating [43, 49].
The bandpass filter (<1 nm bandwidth) before the ESA reduces noise from ASE–ASE
beating to insignificance [26].
The relative intensity noise RINtotal measured by the ESA comes from a variety
of sources, one of which is the RINMPI we want to measure. Other noise contributions
must therefore be subtracted from RINtotal . These contributions are RINs-ASE from
signal–ASE beating, RINDFB from the excess noise of the DFB measured at point
A, RINshot from shot noise, and RINth from thermal noise in the ESA electronic
components.
Figure 15.12 shows results demonstrating this subtraction technique. For this case,
two-path MPI was produced using a Mach–Zehnder interferometer (Fig. 15.12(a)). A
500 m path difference between the arms was chosen to ensure incoherent interference
for a 2.5 MHz DFB linewidth. By adjusting a polarization controller (PC) in one arm
17 The linewidth of the source used to measure RIN from DRB is sufficiently large if the
DRB RIN spectral shape shown in Fig. 15.13(b) matches the incoherent self-homodyne
RIN spectra of Fig. 15.5.
518 J. Bromage et al.

50/50

500 m
PC VOA Ps
99/1

P
RC Ps
(a)

–125
–130 Measured RINtotal with ASE noise
Signal-ASE
RIN [dB/Hz]

–135
–140 Thermal
P
–145 RINMPI (with and without ASE) for R C = –60dB
DFB
–150
–155
Shot
5 10 15 20
f [MHz]
(b)

Fig. 15.12. (a) Mach–Zehnder interferometer used to test and calibrate electrical beat-noise
measurements of MPI; (b) MPI measurement on interferometer (RC P = −60 dB) demonstrating
the RIN subtraction technique. RIN spectra are shown, measured either with or without signal–
ASE beat-noise. RINMPI is the same for both cases.

to maximize the intensity noise, and varying a variable optical attenuation (VOA)
in the other arm, values of RCP ranging from −20 to −60 dB could be produced. To
determine RCP , we simply measured the ratio of optical powers at the output when
each arm of the interferometer was blocked in turn.
Figure 15.12(b) shows RINMPI curves between 2 and 22 MHz for RCP = −60 dB.
First the RIN was measured without the EDFA and VOA2, and RINMPI was ob-
tained. Next the EDFA and VOA2 (see Fig. 15.11) were added and configured to
increase RINs-ASE . This noise contribution, labeled “Signal–ASE,” was calculated
from the measured OSNRASE at point C using RINs-ASE (f ) = 2/(OSNRASE Bref ),
where Bref is the reference optical bandwidth (in Hz) used to define OSNRASE . (RIN
measurements were calibrated using a RIN transfer standard [51, 26] formed from
a broadband light-emitting diode, an EDFA, and a 1.6 nm bandpass filter.) The fact
that both RINMPI curves lie on top of each other shows the accuracy of the RIN sub-
traction technique even for very low levels of RCP , which was only −60 dB in this
demonstration.
The same interferometer can be used to calibrate RINMPI measurements made on
fiber amplifiers. The key point in this calibration technique is that the RINMPI traces
for incoherent interference do not depend on the number of interfering paths (see
Section 15.3.3). Therefore, if the RINMPI curves of an amplifier and interferometer
 15. Multiple Path Interference 519

–100
–36

C [dB]
RINMPI [dB/Hz]
–110 Theory
–40

crosstalk, R P
Experiment
–120 –44

–130 –48
–52
–140
–56
5 10 15 20 0 5 10 15 20 25
f [MHz] Raman gain (on-off) [dB]
(a) (b)

Fig. 15.13. (a) Family of RINMPI curves measured when the on-off Raman gain was varied
P plotted as a function of gain. The
from 0 dB (pumps off) to 24.3 dB; (b) crosstalk ratio RC
markers show the measured values, and the line shows the result of Eq. (15.56).

are identical, the values of RCP for both devices are also identical. Then RCP can be
obtained by simply measuring the power ratio between the arms of the interferometer.
Figure 15.13 shows the results of measurements made on a distributed Raman am-
plifier consisting of 100 km of nonzero dispersion fiber that had a Rayleigh backscatter
coefficient of 100 × 10−6 km−1 at 1550 nm. The amplifier was counter-pumped by
multiplexed semiconductor diodes at 1435 and 1450 nm. We used a 1550 nm DFB
with a linewidth of 2.5 MHz as the signal. Figure 15.13(a) shows RINMPI curves
measured for Raman gains varying from 24.3 dB (the top curve) to 0 dB (the bottom
curve, which is for the pumps turned off). Each RINMPI curve is converted (using
the interferometer calibration technique) to a corresponding RCP value, plotted versus
on-off Raman gain in Fig. 15.13(b). The theoretical curve is from Eq. (15.56). For
small on-off gains, the contribution to RCP is dominated by double-Rayleigh scatter-
ing within the entire 100 km of fiber, and so increases in gain do not increase RCP
significantly. For large gain values, however, double-Rayleigh scattering within the
effective length of the amplifier dominates, and so RCP increases more dramatically
as the gain is increased: a 1 dB increase in gain produces a 1.2 dB increase in RCP (or,
equivalently, a 1.2 dB decrease in OSNRPMPI .)

15.4.3. Optical Time-Domain Extinction Technique

More recently, an optical time-domain extinction technique has been demonstrated for
quantifying MPI in fiber amplifiers [52]. This technique measures the ratio of signal
light to double-Rayleigh light using a time-domain extinction approach similar to that
used in EDFA noise figure measurements [26].
A schematic of the setup is shown in Fig. 15.14. Light from a CW source at the
wavelength of interest first passes through an acoustooptic modulator (AOM 1) used
as a switch to produce pulses with a high extinction ratio. Next these pulses pass
through the DUT, through another high-extinction switch (AOM 2), and are then
520 J. Bromage et al.

Signal control Gate control

0.1 to 1 MHz signal

A B

Signal AOM 1 DUT AOM 2 OSA

Fig. 15.14. Setup used for the optical time-domain extinction measurement of MPI in the DUT.

20
To measure signal power:
DRB + ASE signal + ASE + DRB
Optical spectrum [dBm]

signal
0
gate
–20 dG RC = –40 dB

To measure DRB power:

–40 ASE
signal
DRB gate
–60
1549.0 1549.5 1550.0 1550.5 1551.0
Wavelength [nm]

Fig. 15.15. Calculated OSA spectra (0.1 nm resolution bandwidth) for an optical time-domain
extinction measurement of MPI. The signal and gate pulse trains for both parts of the measure-
ment are shown. When the gate pulses are out of phase with the signal pulses, the signal light
is blocked so only ASE and DRB light are detected.

measured using an OSA. The pulse trains used to control AOM 1 and AOM 2 have
the same repetition rate, but their duty cycles may be different. The combination of
AOM 2 and the OSA forms a detector that is both gated in time and wavelength
sensitive. Therefore the spectrum recorded on the OSA depends on the relative phase
of the pulse trains that control the signal and gate. If the phase of the gate pulses is set
to block the straight-through signal pulses, the OSA trace will only show light leaving
the amplifier out of phase with the signal pulses. This light includes double-Rayleigh
“echoes,” which originate from signal pulses that are delayed inside the amplifier.
The other type of light is ASE from the amplifier, but since it is usually broadband, it
can easily be distinguished from the narrowband double-Rayleigh light.
To make a MPI measurement, two OSA traces are needed (see the schematic in
Fig. 15.15). First, the signal power is measured from a trace with the gate set in phase
with the signal pulses. Next the double-Rayleigh power is determined from a trace
with the gate out of phase and the ASE noise floor subtracted by interpolation. A
calibration factor is needed to account for the duty cycle of the signal pulses (dS )
and the gate pulses (dG ), where dS + dG < 1 to ensure the signal can be blocked by
the gate. (The duty cycle is defined as the time for which the AOM passes light divided
 15. Multiple Path Interference 521

by the repetition period.) If the average power levels measured on the OSA are Ps



and PDRB for the signal and double-Rayleigh light, respectively, RC is given by


1 PDRB
RC = , (15.57)
d Ps

where d is the larger of the two duty cycles dS and dG . Note that, for the setup
described above, the measurement is not polarization sensitive and so all polarization
components of double-Raleigh light are measured. Thus, unlike the electrical beat-
noise measurement approach, this approach cannot measure RCP unless polarization-
selective components are added before the OSA [53].
Several factors must be considered for an accurate measurement. First of all, the
probe signal must be weak enough that it does not change the gain of the amplifier
when it is turned off, or complicated transient gain effects may distort the measure-
ment [54]. Secondly, the extinction ratio of the AOM switches must be high enough
to prevent leakage of the signal when the gate is set to block it. Typically, a single
switch is formed from a pair of cascaded AOMs to provide greater than 90 dB ex-
tinction. Thirdly, care must be taken when subtracting the ASE noise floor from the
double-Rayleigh signal, especially for devices with low levels of MPI, as this can
be the largest source of error. Finally, the choice of modulation rate depends on the
duration of the double-Rayleigh echo [55]; this rate must be high enough that the
double-Rayleigh light is not time-dependent. Then, the double-Rayleigh power mea-
sured on the OSA is proportional to the double-Rayleigh power leaving the amplifier
multiplied by the gate duty cycle dG . Rates on the order of 1 MHz are suitable for
amplifiers made from many kilometers of fiber.
Typically, this technique is sensitive enough to measure RC down to −50 dB when
ASE is also present, which is not as low as can be measured with the electrical beat-
noise technique (−60 dB). Also, only devices with sufficient time delay between the
signal pulses and the double-Rayleigh “echo” can be characterized. But, because the
time domain extinction technique does not require a low-noise signal source with an
appropriate linewidth, any standard tunable laser can be used. This flexibility makes
it easier to characterize the MPI properties of fiber amplifiers over their entire gain
spectrum [56].

15.4.4. Section Summary

In this section we presented ways to quantify and measure the level of MPI. Two types
of quantities, inversely proportional to each other, were defined: an optical signal-to-
noise ratio (OSNRMPI ) and a crosstalk ratio (RC ). We also defined measures that take
into account the state and degree of polarization of the MPI-induced light, OSNRPMPI
and RCP . This extension is crucial, because only the MPI copolarized with the signal
produces beat noise. Next we discussed two techniques used to measure the level of
MPI in either the electrical domain (by measuring the noise spectrum of beat noise)
or in the optical domain (by measuring the optical spectrum of gated echoes produced
by MPI).
522 J. Bromage et al.

15.5. Impact of MPI on Beat-Noise Limited Receivers

The most widely used class of optical receivers employed today uses optically pream-
plified direct detection in order to reduce the impact of electronics noise (most notably,
thermal noise) to insignificance. Thus higher receiver sensitivities can be obtained
than those achieved in practice for pin or avalanche photodiode (APD) receivers.
In an optically preamplified receiver, the dominating noise sources originate from
beating between the signal and ASE as well as from beating of ASE with itself [57].
Therefore this class of receivers is frequently termed beat-noise limited. In optical
communication systems using discrete inline optical amplifiers, ASE is added at each
amplifier site [58]. If distributed amplification is employed, ASE is generated contin-
uously along the transmission path [59]. Because ASE is present in both propagation
directions, originally counterpropagating ASE that is Rayleigh backscattered and am-
plified together with the signal can also be observed at the receiver [60, 61]. In the
presence of MPI, the beating of the signal and MPI (as well as the beating of ASE
with MPI and of MPI with itself) can give rise to appreciable levels of MPI-induced
beat noise, which additionally impairs detection.
To properly design a receiver in the presence of both ASE-induced beat noise and
MPI-induced beat noise, one has to take into account the combined influence of both
beat noise sources. This problem has been tackled in different ways: Wan and Con-
radi [17, 62, 63] derive important statistical properties of the DRB process, including
the variance of signal–DRB beat noise. However, because they consider optical in-
tensity fluctuations rather than fluctuations of the photocurrent, their analyses only
apply to the case of infinitely broadband electrical receiver front ends. For the ASE-
induced beat noise terms, they use the well-known approximations of Olsson [57],
which assume rectangular optical and electrical receive filters as well as constant
optical signal power levels (i.e., no isolated “1”-bits or “0”-bits, and no pulsed signal
formats). Along these lines, other groups [30, 43, 49, 64–68] have theoretically and
experimentally studied detection degraded by both MPI and ASE for different num-
bers of interfering signals as well as for DRB. The discussions on the impact of MPI
and ASE on receiver performance in this section are mainly based on [69], which
generalizes the work of Wan and Conradi by providing a theoretical analysis of beat
noise due to MPI and ASE that takes into account the effect of modulation formats as
well as of optical and electrical filtering. A specialization for Gaussian optical pulses
and Gaussian filter shapes allows for closed-form expressions of the noise variances.

15.5.1. General Expressions for Beat Noise Variances

Figure 15.16 shows the basic setup of a direct detection optical receiver in a
wavelength-division-multiplexed (WDM) transmission scenario impaired by ASE
and MPI. At the receiver input, we find a superposition of the optical signal field
εs (t), the ASE field ε ASE (t), and the MPI field ε MPI (t). The optical spectra of the
three field contributions are also represented in the figure. While ASE is a white pro-
cess (i.e., the ASE field’s power spectral density is constant over the frequency range
of interest), MPI (and DRB as a special case) basically has the same spectral shape as
 15. Multiple Path Interference 523

εs (t ) Received spectra
ε (t ) s (t )
Signal
εMPI (t ) ASE MPI

1

εASE (t ) f
2
B (f ) RD h (t )

1
(P) (P)
OSNR ASE , OSNR MPI
WDM
defined here Demultiplexer

Fig. 15.16. Setup of a direct detection optical receiver impaired by ASE and MPI.

the signal (see Sections 15.1 and 15.3.3.2). The optical signal-to-noise ratios for ASE
(OSNRASE and OSNRASE P ) and MPI (OSNR P
MPI and OSNRMPI ) are defined in front
of the optical filter at the receiver input according to Section 15.4.1. The WDM de-
multiplexer at the receiver input both serves to separate the WDM channels at center
wavelengths λi , and acts as a narrowband optical filter to suppress out-of-band ASE.
The demultiplexing filter has a field transfer function B(f ) (impulse response b(t)).
The filtered optical field after the demultiplexer thus reads
. /
8εε(t) = εs (t) +εε MPI (t) +εε ASE (t) ∗ b(t) =8 8
εs (t) +8 8
ε MPI (t) +8
ε ASE (t), (15.58)

where ∗ denotes a convolution. When referred to after the optical demultiplexing

filter, optical fields and powers are written with a tilde. Photodetection is followed by
electrical amplification and low-pass filtering. The overall optoelectronic conversion
factor is denoted RD ([A/W] or [V/W]). The impulse response of the entire opto-
electronic detection chain
is h(t) (transfer characteristics H (f )), and is considered
normalized to H (0) = h(t)dt = 1. The resulting electrical signalss (t) is then applied
to a sampling-and-decision gate to restore the desired digital information.
Observing that both ε ASE (t) and ε MPI (t) are statistically independent, stationary,
zero-mean stochastic processes, it can readily be shown [40] that the ensemble average
of s (t) reads
8ASE + P
ss (t) = RD (|(εs ∗ b)(t)|2 ∗ h(t) + P 8MPI ), (15.59)

where P8ASE and P 8MPI denote the ASE and MPI powers at the photodetector. Assum-
ing realistically that shot noise is negligible compared to the beat noise terms for
preamplified receivers [31], the variance σ 2 (t) of s (t) can be shown to be [40]
∞
σ 2 (t) = RD
2
CP8 (τ, τ
)h(t − τ )h(t − τ
)dτ dτ
, (15.60)
−∞

where
8(τ )P
CP8 (τ, τ
) = P 8(τ )P
8(τ
) − P 8(τ
) (15.61)
is the autocovariance function of the total optical power at the detector,
8(t) = |8
P εε(t)|2 . (15.62)
524 J. Bromage et al.

Note from the time-dependence of the variance, Eq. (15.60), that s (t) is a nonstation-
ary stochastic process.
Inserting Eqs. (15.58) and (15.62) in (15.61), and taking note of the fact that
• ε ASE (t) is a circularly symmetric complex Gaussian (CCG) process [58],
• ε MPI (t) is a ccG process,18 and
• ε MPI (t) and ε ASE (t) are statistically independent,
Eq. (15.60) can be shown to turn into the sum of signal–ASE beat noise, signal–MPI
beat noise, ASE–ASE beat noise, MPI–MPI beat noise, and ASE–MPI beat noise,

σ 2 (t) = σs-ASE
2
(t) + σs-MPI
2
(t) + σASE-ASE
2
+ σASE-MPI
2
+ σMPI-MPI
2
. (15.63)

Introducing the autocorrelation of the stationary ASE and MPI fields, =8ε N (τ, τ
) =
=8ε N (τ − τ
) = 8
8 ε N (τ
), with N standing for either ASE or MPI, the variances
8
ε TN (τ )8
of the beat noise terms involving the signal are
 ∞ 
2
σs−N (t) = 2RD 2
Re 8 εsT (τ
)=8ε N (τ − τ
)h(t − τ )h(t − τ
)dτ dτ
,
εs (τ )8
−∞
(15.64)
the N–N beat noise terms are
∞
2
σN-N = RD2
|=8ε N (τ − τ
)|2 h(τ )h(τ
)dτ dτ
, (15.65)
−∞

and the ASE–MPI beat noise term is

 ∞ 

T



σASE-MPI = 2RD Re
2 2
=8ε ASE (τ − τ )=8ε MPI (τ − τ )h(τ )h(τ )dτ dτ .
−∞
(15.66)
Note that these equations represent a generalization of the ones derived for ASE only
(see, e.g., [70–72]). By viewing the sum of ASE and MPI as a single source of noise,
it is also possible to compress the five beat noise terms in Eq. (15.63) to just two terms
(one that involves the signal, and one that does not). This is readily done by replacing
8
ε N in Eqs. (15.64) and (15.65) by8 8
ε ASE +8 8ε MPI . This approach can be advantageous
for numerical simulations, however, it does not allow for straightforward statements
on the relative impact of ASE and MPI.
As evident from Eq. (15.64), the signal-induced beat noise variances depend on
the combined influence of the filtered signal waveform8 εs (t), on the detector impulse
response h(t), and on the spectra of the optically filtered ASE and MPI, reflected in
the correlation functions =8ε ASE (τ, τ
) and =8ε MPI (τ, τ
).Assuming no electrical filtering

18 For a sufficiently large number of discrete interferers, this property follows from the central
limit theorem. For DRB, it also follows from the statistical properties of ρ (z), Eqs. (15.29)
to (15.31).
 15. Multiple Path Interference 525

(i.e., infinitely broadband receiver electronics, h(t) = δ(t), with δ(t) being Dirac’s
δ-functional), we arrive at
8s (t)
ss (t) = RD P (15.67)
for the signal, and
2 8 8P
2
σs−N (t) = 2RD Ps (t)P N (15.68)
and
2 82
2
σN-N = RD PN , (15.69)
8s (t)
for the beat noise variances, consistent with the simplified model of [63]. Here P
and P8N (t) denote the optical signal power and the ASE or MPI power at the detector;
8P (t) is the ASE or MPI power copolarized with the signal.
P N

15.5.2. From Beat Noise to Receiver Performance

The quantity of ultimate interest when assessing the performance of optical receivers
is their sensitivity, which for beat noise limited receivers is best specified in terms of a
required OSNR [73], that is, an OSNR value at the optical receiver input that guaran-
tees a certain bit error ratio (BER) at the output of the sampling and decision device.
To arrive at a required OSNR by means of analysis or simulation, we need to know
the signal waveform at the decision gate (including deterministic signal distortions
that can lead to intersymbol interference (ISI)), as well as the probability distribution
of the noise that is corrupting detection. It has been shown (see, e.g., [31, 74]) that
replacing the exact (non-Gaussian) probability density of the decision variable s (t)
by a Gaussian density results in highly accurate predictions of receiver sensitivity,
provided that the decision threshold is assumed optimum.19 Because deviations from
the optimum threshold cannot be modeled appropriately using Gaussian statistics,
we assume throughout this section that the receiver adaptively adjusts its decision
threshold for minimum BER, which is also the most relevant situation encountered
in practice.
The Gaussian distribution is determined solely by its mean and variance. Thus
these two parameters are sufficient to assess receiver performance, and we arrive at
a closed-form expression for the BER,
√
BER = 0.5 erfc[Q/ 2], (15.70)
√
∞
where erfc[x] = (2/ π ) x exp(−ξ 2 )dξ denotes the complementary error function,
and the Q-factor is defined as [76]
ss 1  − ss 0 
Q= . (15.71)
σ1 + σ0

19 Note that this statement is only true for formats that are received using intensity modulation
at a single photodetector. For example, it does not apply for balanced detection of differential
phase shift keying (DPSK) [75].
526 J. Bromage et al.

Here ss 1  and ss 0  denote the noise-free electrical signal values at the optimum sam-
pling instant. In the presence of bit-pattern-dependent deterministic signal distortions
(in particular, in the presence of ISI), taking the worst (lowest) 1-bit and the worst
(highest) 0-bit contained in the bit sequence usually yields the most accurate results.20
In Eq. (15.71), the signal variances corresponding to the signal values ss 1  and ss 0 
are denoted σ12 and σ02 . Although based on a number of approximations, the Q-factor
provides valuable analytical insight in the noise performance of optical receivers.

15.5.2.1. The Q-Factor in the Presence of ASE and MPI

If signal-induced beat noise dominates all other noise terms, the Q-factor can be
approximated by
ss 1  − ss 0 
Q= * *
2
σ1,s−ASE + σ1,s−MPI
2 + σ0,s−ASE
2 + σ0,s−MPI
2

3 9 
: 2
√  √ : f f 2
= fs (1 − r)  1 + r ; s−ASE + s−MPI  . (15.72)
P
OSNRASE OSNRPMPI

To arrive at this expression, we have factored out the absolute optical power according
to

ss 1  = fs P s /(1 + r) (15.73)

2
σ1,s−ASE = fs−ASE
2
P s NASE Bref /(1 + r) (15.74)
2
σ1,s−MPI = fs−MPI
2 P
P s PMPI /(1 + r), (15.75)

and included the inverse extinction ratio of the optical signal in front of the demulti-
plexer,
r = Ps,0 /Ps,1 . (15.76)
The 0-bit mean and beat noise variances thus read

ss 0  = rss 1  and (15.77)

2
σ0,s−N = rσ1,s−N
2
. (15.78)

In Eqs. (15.73) through (15.75), P s is the average optical signal power, leading to a
1-bit peak power of
Ps,1 = âP s /(1 + r), (15.79)

20 If only a few bits within a long bit pattern are highly distorted, as can be the case for
distortions brought by, for example, intrachannel four-wave mixing, taking the worst 1- and
0-bits will result in noticeable underestimates of the Q-factor [77].
 15. Multiple Path Interference 527

2.5 × 10 –3

1/OSNRPASE [linear units]

NRZ RZ
13
2.0 × 10 –3 14
14
1.5 × 10 –3 15
15 16
1.0 × 10 –3
16 17
17 18
0.5 × 10 –3 20 19
19 18
0 1 × 10 –3 2 × 10 –3 3 × 10 –3 4 × 10 –3 2 × 10 –3 4 × 10 –3 6 × 10 –3
P
1/OSNR MPI [linear units] 1/OSNRPMPI [linear units]

Fig. 15.17. The experimental verification of Eq. (15.72) reveals excellent agreement with
theory [78].

with â characterizing the shape of the optical pulses. (For OOK with rectangular
pulses of duty cycle d, we have â = 2/d.) In what follows, quantities with a hat refer
to the peak of the information-carrying data pulses. The symbols NASE Bref and PMPI P

in Eqs. (15.74) and (15.75) denote the power of the ASE and MPI components that
are copolarized with the signal, the former measured in a reference optical bandwidth
Bref (typ. 12.5 GHz). All powers are measured in front of the optical demultiplexer
filter, as indicated in Fig. 15.16. The constant fs reflects the attenuating influence of
2
optical and electrical filtering on the peak of the optical pulses, and fs−ASE 2
and fs−MPI
represent the influence of filtering on the beat noise variances at the pulse peak.21
In order to introduce the effect of finite extinction ratios, it has to be assumed that
the influence of optical and electrical filtering is identical for 1-bits and 0-bits, that
is, that ISI at the decision gate is negligible. Otherwise, the expressions for 1-bit and
0-bit means and variances would not be as straightforward as Eqs. (15.73) through
(15.75) and (15.77) and (15.78).
According to Eq. (15.72), the Q-factor for a system limited by both ASE and
MPI can be accurately approximated by a function involving the sum of the inverse
OSNRs. The excellent validity of this relationship has been verified experimentally
at 10 Gbit/s for NRZ and 50% duty cycle RZ on-off keying [78], and is shown in
Fig. 15.17. The figure shows contours of constant Q-factor (expressed in dB units)
as a function of OSNRASE P and OSNRPMPI together with linear fits. We now proceed
to calculate the
• MPI-induced Q-factor penalty Q, that is, the decrease in Q-factor when MPI
is added to a system that is originally limited by signal–ASE beat noise only, and
the
• OSNR-penalty OSNR, that is, the increase in OSNRASE that is required when
MPI is added, in order to maintain the same Q-factor obtained without MPI.

21 Note that f , f
s s−ASE , and fs−MPI can substantially differ from one another.
528 J. Bromage et al.

15.5.2.2. Q-Factor Penalty

In the absence of MPI, system performance is given by a reference Q-factor of (cf.

Eq. (15.72) with OSNRPMPI → ∞)
√ *
fs (1 − r)
Qref = √ OSNRASEP
, (15.80)
fs−ASE 1 + r
which allows us to specify a Q-factor penalty according to
   
P
Qref 1 OSNRASE
QdB = 20 log = 10 log 1 + · (15.81)
Q FOSNR OSNRPMPI
 
1 Q2ref 1+r
= 10 log 1 + · √ . (15.82)
FQ OSNRPMPI (1 − r)2

Equation (15.81) expresses the Q-factor penalty in terms of the two optical signal-to-
P
noise ratios OSNRASE and OSNRPMPI , using the OSNR-based MPI tolerance factor

FOSNR = fs−ASE
2 2
/fs−MPI , (15.83)

whereas Eq. (15.82) quantifies the Q-factor penalty for a given reference Qref , using
the Q-factor-based MPI tolerance factor

FQ = fs2 /fs−MPI
2
. (15.84)

Both FOSNR and FQ indicate the tolerance of a system to degradations by MPI; that
is, the larger these quantities, the more robust the system will be to impairments by
MPI. Note also from comparing Eq. (15.83) with (15.74) and (15.75) that FOSNR
gives the ratio of signal–ASE beat noise variance to signal–MPI beat noise variance
for equal copolarized OSNRs (OSNRASE P =OSNRP ).
MPI
If we neglect the influence of electronic filtering (i.e., we set h(t) = δ(t) and use
Eqs. (15.67) and (15.68)), and if we take note of the fact that OSNRPMPI does not
change upon transition through optical filters, which is true if signal and MPI have
the same spectral shape, we can approximate the two MPI tolerance factors by
P
Bo PMPI
FOSNR ≈ (15.85)
8P
Bref PMPI

and
<
8
a
FQ ≈ , (15.86)
2
where Bo = P 8P /NASE denotes the equivalent square bandwidth of the demulti-
ASE
plexer filter, and <
8
a characterizes the optical pulse shape after the demultiplexer (cf.
Eq. (15.79)). Note from Eq. (15.85) that the tolerance to MPI for fixed OSNRs in-
creases linearly with the data rate, which is a consequence of the bit-rate-independent
 15. Multiple Path Interference 529

reference bandwidth entering the definition of OSNRASE (see also Section 15.5.2.4).
Also broadband optical signaling, resulting in a reduction of the optically filtered
MPI power P 8P for fixed optical demultiplexer bandwidths, increases the tolerance
MPI
to MPI (see Section 15.5.3.1).
For perfect signal extinction ratio (r = 0), <
8
a denotes the ratio of peak to average
optical power after the demultiplexer, and we arrive at
<
8/P
FQ ≈ 0.5P 8, (15.87)

which, again, implies that broadband signaling is beneficial for a receiver’s robustness
to MPI. Disregarding a slightly different conversion of the Q-factor to dB units, and
assuming unpolarized MPI (OSNRMPI =OSNRPMPI /2), we then arrive at the results of
Fludger and Mears, who also showed good experimental agreement in a 40 Gbit/s
CSRZ experiment [100].

15.5.2.3. OSNR Penalty

Setting Q = Qref (using Eqs. (15.72) and (15.80)), we readily find for the OSNR
penalty,
   
P
OSNRASE OSNRASE
OSNRASE, dB = 10 log = 10 log P
OSNRASE, ref OSNRASE, ref
 
P
1 OSNRASE
= −10 log 1 − · (15.88)
FOSNR OSNRPMPI
 
1 Q2ref 1+r
= −10 log 1 − · √ , (15.89)
FQ OSNRPMPI (1 − r)2

with the MPI tolerance factors defined in Eqs. (15.83) and (15.84). Neglecting, again,
the influence of electronic filtering, we have FQ ≈ < 8a /2 (see Eq. (15.86)), which,
inserted into Eq. (15.89), directly reproduces the results of Liu et al. [65].
Figure 15.18 shows measurements (circles) and theoretical predictions (solid line)
of the OSNR penalty due to MPI using Eq. (15.89) under the approximation (15.86).
The measurement results are taken from Kim et al. [79], and apply for 10 Gbit/s
NRZ modulation with a measured extinction ratio of r = −13 dB and Qref = 6, corre-
sponding to BER= 10−9 . The dashed line shows the OSNR penalty that would have
been predicted for perfect extinction (r = 0). It can be clearly seen that neglecting the
extinction ratio can lead to a significant underestimation of the performance degra-
dation due to MPI, as pointed out in [64–66]. The dotted line shows the Q-factor
penalty according to Eq. (15.82) under the approximation (15.86) for reference. The
Q-factor penalty is seen to be lower than the OSNR penalty. For OSNRPMPI
25 dB,
the two penalties closely match each other. This can be understood by comparing
Eqs. (15.82) and (15.89) and taking note of the Taylor expansion 1/(1 − x 2 ) ≈ 1 + x 2
around x = 0.
530 J. Bromage et al.

7
OSNR penalty (r = –13 dB)
6
OSNR penalty (r = 0)
Q-factor penalty (r = –13 dB)
5

Penalties for Q ref = 6

4

0
50 45 40 35 30 25 20
OSNRPMPI

Fig. 15.18. Measurements (circles) and theoretical prediction (solid line) of OSNR penalty
due to MPI for r = −13 dB. The dashed line gives the predicted OSNR penalty for perfect
extinction (r = 0), and the dotted line shows the Q-factor penalty. The experimental data are
taken from [79].

A third penalty parameter occasionally used in the literature is the power penalty.
This quantity specifies the required increase in received optical signal power to main-
tain a given reference Q-factor. When using power penalties in the context of optically
preamplified receivers, it is implicitly assumed that ASE generated by the receiver’s
optical preamplifier dominates any ASE originating from other sources (e.g., from
a saturated booster amplifier at the transmitter). If the level of MPI at the receiver
input is referred to the increased signal power that, in the presence of MPI, guarantees
detection with Qref , the power penalty due to MPI can be shown to be identical to the
OSNR penalty OSNRASE derived in this section.

15.5.2.4. Bit Rate Scaling

Equations (15.82), (15.84), (15.86), and (15.89) reveal that neither the Q-factor
penalty nor the OSNR penalty due to MPI are data-rate-dependent, if the penalties
are measured with reference to a fixed (data-rate-independent) reference Q-factor to
guarantee a certain (data-rate-independent) BER.
On the other hand, Eq. (15.83) and, more explicitly, Eq. (15.85) predict a linear
increase in tolerance to MPI with data rate, if two systems at different data rates are
compared on a constant-OSNR basis. The higher MPI tolerance at increased data rates
is reflected by the linear dependence of FOSNR on the demultiplexer bandwidth Bo ,
which scales linearly with the data rate for fixed spectral efficiencies. The physical
origin of this behavior resides in the fact that the ASE power spectral density (and thus
OSNRASE ) is defined in a fixed bandwidth Bref , and is thus independent of the bit rate.
 15. Multiple Path Interference 531

Because ASE is more broadband than Bref , increasing the receiver filter bandwidths
when transitioning to higher data rates leads to more signal–ASE beat noise within
the receiver. In contrast, the MPI power spectral density has to decrease linearly with
bit rate to maintain a constant level of MPI (i.e., constant OSNRMPI ). This leads to a
bit-rate-independent signal–MPI beat noise. For fixed OSNRs, higher bit rate systems
therefore suffer more from signal–ASE beat noise than lower bit rate systems. The
impact of changes in per channel launch powers, leading to scalings in OSNR with
bit rate, as well as the influence of fiber nonlinearities on the bit rate scaling of MPI
degradations is covered in Section 15.6.

15.5.2.5. Impact of Signal Quality on Degradations due to MPI

The results of Sections 15.5.2.2 and 15.5.2.3 show that the system penalty caused by
MPI strongly depends on the quality of the optical signal prior to adding MPI, as has
been pointed out by several groups [43, 64–67]. Most importantly,
• a large amount of ASE (or, equivalently, a low reference Q-factor, as found in sys-
tems using forward error correction, FEC) mitigates the impact of MPI, whereas
• a poor signal extinction ratio dramatically reduces a system’s robustness to MPI,
and
• a high peak-to-average optical power ratio after the demultiplexer helps to reduce
the impact of MPI. This statement, which implies that RZ using short pulses is
more robust towards MPI than high duty cycle RZ or even NRZ, is discussed in
more depth in the next section.
We explicitly note at this point that the above predictions on the impact of MPI and
ASE on receiver performance critically rely on the validity of replacing the exact
(non-Gaussian) statistics of the signal at the decision gate by Gaussian probability
densities. Special caution has to be exerted whenever this assumption breaks down,
for example, for balanced detection of DPSK, for fixed-threshold optical receivers, or
for situations where only a few interferers produce inband crosstalk, as can be the case
in optical networks incorporating add-drop multiplexers with imperfect suppression
of the drop channels (see Sections 15.1 and 15.2.2).
Furthermore, if the influence of limited bandwidth electrical filtering in the re-
ceiver is to be included to overcome the assumption of infinitely broadband electron-
ics, one has to evaluate FOSNR or FQ based on the exact expressions (15.59) and
(15.64). Although this has to be done numerically in general, closed-form solutions
can be given under the assumption of Gaussian pulse shapes and Gaussian (optical
and electrical) filter characteristics. This is demonstrated in the following section.

15.5.3. Special Case: Gaussian Pulse Shapes and Filter Characteristics

To better understand and interpret the relative importance of the beat noise due to
ASE and MPI as a function of signal pulsewidth, demultiplexer optical bandwidth,
 532 J. Bromage et al.

Fig. 15.19. Definition of equivalent

square bandwidths for the optical signal
ε
spectrum s (f ), the optical filter char-
|εs(f )|2, |B (f )|2, |H (f )|2
acteristics B(f ), and the electrical filter
characteristics H (f ).

Bs , BO

2Be
f
0

and electrical receiver bandwidth, we analytically solve Eq. (15.64), which is possible
by assuming first-order Gaussian functions for all pulse and filter shapes involved,
*
εp (t) = P <s exp[−2π Bs2 t 2 ] (15.90)
√
b(t) = 2Bo exp[−2π Bo2 t 2 ] and (15.91)
√
h(t) = 2 2Be exp[−8π Be2 t 2 ], (15.92)
where P <s denotes the peak optical pulse power before demultiplexing, and Bs , Bo , and
Be are the optical signal, demultiplexer, and electrical equivalent square bandwidths,
respectively. With reference to Fig. 15.19, these are defined as the double-sided (Bs
and Bo ) and single-sided (Be ) bandwidths that define rectangles of the same area as
the respective squared spectra. For the optical and electrical filters, the + equivalent
square bandwidths + are related to the 3 dB bandwidths by B o,3dB = 2B o ln(2)/π and
Be,3dB = 2Be ln(2)/π.
With these assumptions, the ASE and MPI autocorrelation functions =8ε ASE (τ, τ
)
and =8ε MPI (τ, τ
) appearing in Eq. (15.64) can be readily evaluated. The ASE autocor-
relation is obtained from the realistic assumption that the ASE process is white with a
power spectral density NASE per polarization mode over the optical filter bandwidth,
that is, assuming =ε ASE (τ, τ
) = NASE δ(τ − τ
). This readily leads to
= ∞ ∞ >
=8ε ASE (τ, τ
) = T
ε ASE (ξ )bT (τ − ξ )dξ ε ASE (η)b(τ
− η)dη
−∞ −∞
∞
= εε ASE
T
(ξ )εε ASE (η)bT (τ − ξ )b(τ
− η)dξ dη
−∞
∞
= NASE bT (τ − ξ )b(τ
− ξ )dξ = NASE Bo exp[−π Bo2 (τ − τ
)2 ],
−∞
(15.93)
which in the spectral domain simply reads NASE |B(f )|2 : ASE is spectrally shaped by
the demultiplexer filter, as expected. To obtain a suitable expression for =8ε MPI (τ, τ
),
which relies on finding an analytically tractable approximation to the signal autocor-
 15. Multiple Path Interference 533

relation =8εs (τ − τ
) (see Eq. (15.39)), we use the method detailed in Appendix A15.B
to arrive at
*
=8εs (τ, τ
) = PMPI
P
1/(1 + Bs2 /Bo2 ) exp[−π Bs2 /(1 + Bs2 /Bo2 )(τ − τ
)2 ], (15.94)

where PMP P
I is the copolarized optical MPI power measured in front of the WDM
demultiplexer. The difference between the two autocorrelations (15.93) and (15.94)
is caused by the different spectral distributions of ASE and DRB; while the ASE
spectrum is only shaped by the optical demultiplexing filter, the spectrum of DRB
matches that of the signal. This difference fundamentally distinguishes signal–ASE
beat noise from signal–DRB beat noise.
Substituting Eqs. (15.90) and (15.92) through (15.94) into Eq. (15.64), we finally
arrive at
4RD2N <
ASE Ps Be
2
σ1,s−ASE = ,
[1 + Bs /Bo + Bs /(4Be )] [1 + 4Be /Bo2 (1 + Bs2 /Bo2 ) + 2Bs2 /Bo2 ]1/2
2 2 2 2 1/2 2

(15.95)
2R 2 PP P <s
2
σ1,s−MPI = D MPI
,
[1 + Bs2 /Bo2 ]1/2 [1 + Bs2 /Bo2 + Bs2 /(2Be2 )]1/2 [1 + Bs2 /Bo2 + Bs2 /(4Be2 )]1/2
(15.96)
for the beat noise variances at the peak of an isolated electrical pulse. The average
electrical signal at that point then becomes (cf. Eq.(15.59))
<s /[(1 + Bs2 /Bo2 )(1 + Bs2 /Bo2 + Bs2 /(2Be2 ))]1/2 .
ss 1  = RD P (15.97)
In the limit of broadband optical filtering and quasi CW signaling (Bs # Be # Bo ),
Eq. (15.95) simplifies to the well-known expression [57]
σ2 <s Be ,
≈ 4R 2 NASE P (15.98)
1,s−ASE D
whereas Eq. (15.96) becomes [63]
2 P <
2
σ1,s−MPI ≈ 2RD PMPI Ps , (15.99)
independent of any bandwidth and in agreement with Eq. (15.68). The peak electrical
<s .
signal in this case is simply ss 1  ≈ RD P
In Section 15.6.1.4, when calculating the noise figure of a Raman amplified trans-
2 (t). This
mission span, we also need an expression for the shot noise variance σshot
quantity can be generally written as [80]
2
σshot (t) = RD e(|εs |2 ∗ h)(t), (15.100)
with e = 1.602 · 10−19 As denoting the elementary charge. Using the Gaussian pulse
and filter model (Eqs. (15.90) through (15.92)), this equation becomes
<s Be 1
2
σshot (t) = 2RD eP , (15.101)
[1 + Bs /Bo ] [1 + Bs2 /Bo2 + Bs2 /(4Be2 )]1/2
2 2 1/2

with the well-known approximation

2
σshot <s Be
≈ 2RD eP (15.102)
for large filters and quasi CW signaling.
534 J. Bromage et al.

Bs = R Bs = 3R Bs = 10R
10 11.6

Optical filter bandwidth Bo /R

4 .8
3.0
2.8

4.6
11.4

2.6

4.4
8 11.2 II

2.4

4.
11.0

2.2

2
I
10.8

2.0
6 4.0

1.8
10.6
2.8 II
4 10.4
1.6 .5

2.8 I
II 10.2
1

2.4
2 1.4 10.1
I 2.2
0.5 1 1.5 0.5 1 1.5 0.5 1 1.5
Electrical bandwidth Be /R Electrical bandwidth Be /R Electrical bandwidth Be /R
(a) (b) (c)

Fig. 15.20. MPI tolerance factor FOSNR as a function of optical and electrical receiver band-
widths, both normalized to the data rate R. The reference bandwidth for defining OSNRASE
is assumed equal to the data rate (R = Bref ). (a), (b), and (c) correspond to signal bandwidths
of Bs = 1R, 3R, and 10R. Gray lines give boundaries between the two asymptotic regions
discussed in the text.

15.5.3.1. Impact of MPI for given OSNRs

We next use Eqs. (15.95) and (15.96) together with (15.74), (15.75), and (15.83) to
calculate the MPI tolerance factor FOSNR , which can be interpreted as the ratio of
signal–ASE beat noise variance to signal–MPI beat noise variance for OSNRPMPI =
P
OSNRASE ,
[1 + Bs2 /Bo2 ]1/2 [1 + Bs2 /Bo2 + Bs2 /(2Be2 )]1/2
FOSNR = 2Be /R . (15.103)
[1 + 4Be2 /Bo2 (1 + Bs2 /Bo2 ) + 2Bs2 /Bo2 ]1/2
The results are displayed in Fig. 15.20 for a signal bandwidth of (a) Bs = R, (b) 3R,
and (c) 10R. For the sake of convenience, we assume that the bit rate R equals the
reference bandwidth Bref appearing in the definition of OSNRASE P . If R deviates from

Bref , the values for FOSNR shown in Fig. 15.20 have to be rescaled by R/Bref .
It becomes obvious from Fig. 15.20 that a system’s tolerance to MPI generally
depends on the signal bandwidth (i.e., the pulse duration used for RZ signaling) as well
as on both optical and electrical receiver bandwidths. The tendency for broadband
optical signals to result in a higher tolerance to MPI (higher values of FOSNR ) lines up
with the more simplified notion that the tolerance to MPI depends on the ratio of peak
to average optical power, which is higher for lower RZ duty cycles (cf. Eq. (15.87)).
Note that broadening the spectrum by means of short pulses is only one way of
diminishing the effect of MPI. Another way, which is not captured by the simplified
model treated here, is to impose a controlled amount of phase modulation (chirp)
on the signal to broaden the spectrum, as proposed by different groups [81–85]. The
benefit of using broad signal spectra can be explained as follows.22 For sufficiently
22 This can also be seen from Eqs. (15.95) and (15.96), when taking note of the fact that the
peak optical power is proportional to the average optical power and the signal bandwidth.
 15. Multiple Path Interference 535

ASE
Signal

Optical power spectra

Electronics
be at

Be

Bs

Bo

f
0
Fig. 15.21. Visualization of the beating process. Only those spectral ASE components whose
beat frequencies with the optically filtered signal fall into the receiver electronics’ bandwidth
produce signal–ASE beat noise.

large signal bandwidths and constant OSNRASE , the signal field’s power spectral
density has to scale inversely with the signal bandwidth to maintain constant average
optical power (and thus constant OSNRASE ). Hence the signal–ASE beat noise term
decreases linearly with signal bandwidth. The signal–MPI beat noise term, on the
other hand, decreases quadratically with signal bandwidth, because both the signal
power spectral density and the MPI power spectral density are inversely proportional
to the signal bandwidth for constant OSNRMPI .
Two asymptotic regions can be identified in Figs. (15.20)(a) through (c), qualita-
tively distinguished by contour lines running perpendicular to the Bo -axis (regions I)
and contour lines running in parallel to the Bo -axis (regions II). The two regions are
separated by gray lines in Fig. 15.20. With reference to Fig. 15.21, these lines indi-
cate, for each electrical receiver bandwidth, the largest possible optical demultiplexer
bandwidth Bo that lets pass spectral ASE components which produce a beating with
the signal at a beat frequency B that still lies+within the detection bandwidth Be ;
that is, B = Bo /2 − B 8s = Bs Bo / Bs2 + Bo2 denotes the bandwidth of
8s /2 = Be . (B
the filtered optical signal.)
Region I
For demultiplexer bandwidths below the boundary indicated by the gray lines, de-
creasing Bo reduces the signal–ASE beat noise, thus reducing FOSNR and emphasizing
the influence of MPI on system noise. The reduction of FOSNR is more significant
for narrower signal spectra (long RZ pulses or even NRZ), where the spectral distri-
bution of MPI significantly differs from that of ASE; for signal spectra significantly
broader than the optical demultiplexer bandwidth (short RZ pulses or highly chirped
signals), the spectral distribution of MPI becomes constant over the optical filter
bandwidth, and is thus similar to ASE (see Eq. (15.94)). In the limit Bs  Bo we
536 J. Bromage et al.

find FOSNR → Bs /R, independent of any filter bandwidth: signal–MPI beat noise
becomes less important at large Bs , as discussed above.

Region II

For demultiplexer bandwidths exceeding the boundary indicated by the gray lines, the
contour lines eventually+ turn parallel to the Bo -axis; in the limit Bo  {Be , Bs } we fi-
nally find FOSNR → 2 Be2 + Bs2 /2/R. This expression is independent of Bo , because
for sufficiently large optical filter bandwidths no additional beat noise falling within
the electrical filter bandwidth can be generated by increasing Bo (see Fig. 15.21).
Note, however, that ASE–ASE beat noise, which is not considered in our model, will
eventually become important when going to large optical filter bandwidths, which will
lead to an additional ASE-induced performance degradation (and thus to an increase
in the system’s tolerance to MPI).

15.5.3.2. Tolerance to MPI for Fixed Q-Factors

We now evaluate the MPI tolerance factor FQ (see Eq. (15.84)), which is used to assess
the MPI-induced system degradation on a receiver initially limited by signal–ASE beat
noise and specified in terms of a reference Q-factor Qref (see Eqs. (15.82) and (15.89)).
Using Eqs. (15.96) and (15.97), and taking note of the relation P <s /P s = 4Bs /R, we
arrive at
[1 + Bs2 /Bo2 + Bs2 /(4Be2 )]1/2
FQ = 2Bs /R · . (15.104)
[1 + Bs2 /Bo2 ]1/2 [1 + Bs2 /Bo2 + Bs2 /(2Be2 )]1/2

As expected from our previous discussion along with Eq. (15.87), for Be → ∞, this
expression turns into
1 <
8s /P
8s .
FQ ≈ 2Bs /R + = 0.5P (15.105)
1 + Bs2 /Bo2

Figure 15.22 shows an evaluation of the MPI tolerance factor FQ , Eq. (15.104),
for the same parameters used in Fig. 15.20, that is, (a) Bs = R, (b) Bs = 3R, and (c)
Bs = 10R). The curves show the same qualitative behavior as those for the tolerance
factor FOSNR shown in Fig. 15.20. In particular, it is evident from Fig. 15.22 that
• the impact of MPI on receiver performance initially decreases with increasing sig-
nal bandwidth. In+the limit of large signal bandwidths (Bs → ∞), FQ approaches
the value 2Bo /R (1 + Bo /(4Be )2 )(1 + Bo /(2Be )2 ). This, again, illustrates that
RZ signaling is more robust to MPI than NRZ signaling, which has been experi-
mentally confirmed by several groups [43, 78]. Also,
• the receiver degradation due to MPI depends on the electrical receiver bandwidth
in general. The tolerance to MPI is reduced when going to lower receiver band-
widths.
 15. Multiple Path Interference 537

Bs = R Bs = 3R Bs = 10R
10
10.0

Optical filter bandwidth Bo /R

1.85

4.5
9.5

0
1.80

4.2
8 9.0

1.75

5
8.5
8.0

1.70
4.0 7.5
6 0
7.0

1.65
3.75 6.5
6.0
1.60
4 3.50
5.0
3.00
1.5

4.0
2 1.4
0

2.50 3.0
0 2.00 2.0
0.5 1 1.5 0.5 1 1.5 0.5 1 1.5
Electrical bandwidth Be /R Electrical bandwidth Be /R Electrical bandwidth Be /R
(a) (b) (c)

Fig. 15.22. MPI tolerance factor FQ as a function of optical and electrical receiver bandwidths
normalized to data rate R for (a) Bs = R, (b) Bs = 3R, and (c) Bs = 10R.

15
Signal bandwidth Bs = 10R
MPI tolerance factor FQ

10

Bs = 3R
5

Bs = R

0
1 2 3 4 5 6 7 8 9 10
Optical filter bandwidth Bo /R

Fig. 15.23. Comparison between simple approximation to FQ (dashed; see Eq. (15.105) for
infinite electrical receiver bandwidths) and exact result using Gaussian pulses and Gaussian
optical and electrical filters (solid; Be = 0.7R) for Bs = R, 3R, and 10R. The approximation
underestimates the impact of MPI.

Finally, Fig. 15.23 shows the quality of the approximation given in Eq. (15.105),
obtained by neglecting the influence of electrical filtering, for the three signal band-
widths Bs = R, 3R, and 10R. The dashed lines represent the approximation to FQ ,
Eq. (15.105), whereas the solid curves are section lines from Fig. 15.22 at a typical
electrical receiver bandwidth of Be = 0.7R, and thus follow Eq. (15.104). It can be
seen that the approximation overestimates the receiver’s tolerance to MPI (i.e., it
underestimates the impact of MPI), with increasing deviation for large signal band-
widths.
538 J. Bromage et al.

15.5.4. Section Summary

This section analyzed the impact of MPI on beat-noise limited digital receivers. Gen-
eral formulae as well as various convenient approximations for beat noise due to ASE
and MPI were given. The overall receiver performance was shown to depend on a
weighted sum of the inverse OSNRs due to ASE and MPI. Receiver degradations
due to MPI were specified in terms of Q-factor penalties and OSNR penalties. The
tolerance of a receiver to MPI was shown to depend on the employed modulation
format, on the signal quality at the receiver input (e.g., amount of ASE loading and
signal extinction ratio), as well as on the optical and electrical receiver bandwidths.
The amount of tolerable MPI power is typically some 20 dB below the signal power,
independent of data rate.

15.6. System Design Optimization

In this section, we use the knowledge developed in previous sections to optimize the
performance of fiber-optic communication systems incorporating Raman amplifica-
tion. In systems using passive fibers, the two main elements that need to be balanced
are ASE and Kerr nonlinearity. For systems making use of Raman pumping, such
optimization requires, in addition, knowledge of the nature of MPI and the way MPI
affects the signal at photodetection. Before entering into the detailed calculations
on system design and optimization we first discuss the phenomena detrimental to
transport of information through optical fibers and the way they impact system per-
formance.
A variety of phenomena imposes limits on the transport of information, some of
which can be qualified as fundamental in nature and others can be argued as being
fortuitous. A fundamental limitation to transmission of information originates from
the quantum nature of light, which results in the generation of amplified spontaneous
emission during the process of optical amplification. Optical amplification is neces-
sary to compensate signal loss from various origins during transport. As discussed
in Section 15.5, the presence of ASE leads to errors at detection, measured by the
BER. Because the BER depends on the optical signal-to-noise ratio, it can generally
be improved by increasing the optical signal power. However, when the ASE sources
are distributed throughout the transmission line, increasing the OSNR at the end of
the line requires increasing signal power throughout the line which induces nonlinear
effects in the optical fibers.
Because of the very high field confinement in the transverse direction, the inten-
sities reached inside optical fibers quickly reach values at which the signal induces
changes in the refractive index of the fiber core. For silica-core optical fibers, the
index change is virtually instantaneous and does not depend on the polarization of the
signal. It is referred to as Kerr nonlinearity [86]. For systems designed with realistic
fiber types and signal powers, the source of signal distortions from Kerr nonlinearity
on intensity-modulated formats originates almost solely from nonlinear interactions
between the signals themselves (as opposed to interactions between signal and noise).
 15. Multiple Path Interference 539

Interactions among signals are deterministic and, in the regime of powers accessible
to current fiber-optic communication systems, do not initiate chaotic transmission be-
havior. Consequently, distortions imprinted on the signal by Kerr nonlinearity do not
exhibit the stochasticity of ASE or MPI and can clearly be compensated for [87]. Nev-
ertheless, we consider here that such distortions are not compensated for because it is
technologically difficult and expensive to correct the deterministic signal distortions
from Kerr nonlinearity at the receiver. To achieve optimal performance, fiber-optic
communication systems are generally designed to operate at the maximum possible
signal power at which any further power increase results in signal distortions from
Kerr nonlinearity that outweighs the benefits from the improvement in OSNR. Con-
sequently, the optimization of system performance involves a constant balancing of
the effects of Kerr nonlinearity and sources of noise, such as ASE.
The amount of ASE an amplifier generates is not a critical function of the signal
power going through the amplifier, but depends primarily on how much gain it pro-
vides. The impact of the ASE generated by an amplifier on the OSNR at the amplifier
output depends directly on the signal input power to the amplifier. Consequently, in
a chain of amplifiers, the maximum adverse effect of ASE on OSNR takes place in
amplifiers following locations of lowest signal powers in the line.23 Because pumping
stations are generally the most expensive element in a transmission line, one wants to
minimize their number by increasing the length of the passive transmission fibers. As
a result, the passive transmission fibers are generally the most lossy elements in a line
and the signal power in the line is at its lowest at their output. Therefore, the low signal
power at the end of a passive transmission fiber is usually the origin of the largest
OSNR degradations in a transmission line. Thus one can efficiently avert this OSNR
degradation by preventing the signal power from dropping as low as it does when the
transmission fiber is left passive. Such improvement in OSNR can be accomplished
by converting the passive transmission fiber into an active fiber, for instance, by either
using light doping with erbium atoms [88] or by using Raman pumping [89]. These
active transmission fibers are called distributed fiber amplifiers (DFAs). The term
distributed amplifier is used to describe amplifiers of physical length comparable to
important characteristic transmission lengths, such as the loss length or the nonlinear
length [86].
Because of their long lengths and their nonuniform gain distribution, DFAs are
more likely to suffer from the generation of DRB than passive fiber spans.As discussed
in Section 15.3.3, DRB produces a large number of low-power replicas of the signal
with random delays and phases that propagate along with the signal. As a result, we
are confronted with MPI from a large number of interferers. Thus the PDF for the
composite intensity becomes Gaussian-like (see Section 15.2.2), which justifies the
use of variances only (see Section 15.5.2).
The goal when designing a transmission system is the maximization of system
performance by minimizing BER at the electrical decision gate (see Fig. 15.16). To
minimize BER, one simultaneously has to reduce the effect of beat noise resulting
23 Other sources of loss are, for instance, connectors, splices, optical filters, dispersion-
compensators, crossconnects, add-drop multiplexers, and variable optical attenuators con-
trolling the input power to a device.
540 J. Bromage et al.

from all sources of noise added to the signal (see Section 15.5.2) and prevent buildup
of signal distortions from fiber dispersion and Kerr nonlinearity. The widespread
availability of dispersion-compensating elements allows dispersion to be tailored
to virtually any desired value, leaving Kerr nonlinearity the major source of signal
distortion to be considered. In designing Raman-pumped systems it is then essential
to balance the effects of three phenomena limiting system performance: ASE, DRB,
and Kerr nonlinearity.

15.6.1. Raman Amplification and System Design

Raman amplification can be used to provide a large range of gain values for any fiber
type and in any spectral window (or band). Raman amplification can also be applied
to any particular fiber, regardless of its location or function in a transmission line.
Moreover, Raman gain can be obtained by pumping a given fiber in the same direction
as the signal propagation (copumping), in the opposite direction to the signal (counter-
pumping), or Raman pumping through both fiber ends (bidirectional pumping). Even
though all these Raman pumping configurations can in principle be used to improve
system performance, it is generally desirable to reduce the number of Raman pumping
sites and Raman pumps to minimize system cost.
The possible utilizations of Raman amplification in systems are represented
in Fig. 15.24, depicting a typical span for four different transmission lines. Fig-

Transmission Pumping System

fiber site type
Rare-Earth doped amp.

(a) Non
Passive transmission Raman

Discrete Raman amplifier

(b)
Discrete
Passive transmission FW BW Raman
Pumps

Rare-earth doped amp.

(c) Hybrid
Distributed Raman BW FW
amplification (DRA) pumps pumps

Discrete Raman amplifier

(d) All
Distributed Raman FW BW Raman
amplification (DRA) BW Pumps FW

Single span

Fig. 15.24. Different use of Raman amplification in simplified representations of a transmission

line and corresponding system types: (a) non-Raman; (b) discrete Raman; (c) hybrid; and (d)
all-Raman (FW: forward, BW: backward).
 15. Multiple Path Interference 541

ure 15.24(a) represents a transmission line that does not use Raman amplification.
Optical amplification occurs periodically at pumping sites following each passive
transmission fiber. Such discrete amplification is generally accomplished using a
type of rare earth-doped amplifier. By the end of 2002, commercial systems almost
universally use a single type of rare earth-doped amplifier, the erbium-doped fiber
amplifier (EDFA). One possible application of Raman amplification is to replace dis-
crete rare earth-doped amplifiers in bands where their effectiveness remains uncertain
(Fig. 15.24(b)). These bands are essentially those outside the EDFA traditional C- and
L-bands (1530 to 1570 nm and 1570 to 1610 nm, respectively), and have been la-
beled S+ -, S-, and L+ -bands (1450 to 1490 nm, 1490 to 1530 nm, and 1610-1650 nm,
respectively) [90]. A second way of using Raman amplification is pumping transmis-
sion fibers (Fig. 15.24(c)), providing gain in any desired band. As mentioned earlier,
providing distributed Raman amplification (DRA) in the transmission fiber is the
most beneficial use of Raman amplification to reduce the buildup of ASE in the trans-
mission line. In such a configuration, DRA is generally complemented with discrete
amplification based on rare earth-doped amplifiers (generally EDFAs) at pumping
sites. Such discrete amplification is usually necessary to make up for the lack of suf-
ficient gain of DRAs in the presence of loss from various optical components present
at pumping sites. The system configuration depicted in Fig. 15.24(c) is referred to as
a hybrid-span system or simply hybrid system. Lastly, if DRA is used and all sources
of gain in the transmission line come from Raman amplification, such a system is
referred to as an all-Raman system (Fig. 15.24(d)). We chose to focus our attention
on hybrid systems because they are simpler to analyze than all-Raman systems.
To determine the optimal use of Raman pumping in a hybrid system, one should
consider Raman pumping transmission fibers by both ends and varying the amount of
gain provided by pumping in each direction. The various pumping configurations of a
hybrid system will result in different amounts of ASE, DRB, and distortions from Kerr
nonlinearity. In the following sections, we describe a procedure developed for finding
the optimum Raman pumping configuration that maximizes system performance. The
optimum Raman pumping configuration is expressed in terms of a percentage of
forward pumping and net Raman gain.

15.6.1.1. Amplified Spontaneous Emission

Any optical amplification process is simultaneously accompanied by spontaneous

emission. Spontaneous emission corresponds to the spontaneous creation of photons
through the decay of an electron from an excited atomic state to a lower energy
state. This emission can occur at any time and location in an amplifier. Moreover,
transitions can cover a large range of photon energies corresponding to all pos-
sible atomic level transitions. Because spontaneous emission occurs anywhere in
a medium, spontaneous emission is amplified from the moment of its creation to
the output end of an amplifier. It follows that the amplified signal comes with ASE
that corrupts the information carried by the signal. In the case of a bidirectionally-
pumped DRA with a single Raman pump on each side, the spectral density of the
optical noise power NASE in each state of polarization is given by (undepleted
 542 J. Bromage et al.

pump approximation and assuming that the fiber is at 0 Kelvin) [59, 91],
L
NASE = hf CR (λs , λp )[Pb e−αp (L−z) + Pf e−αp z ] G (z, L)dz, (15.106)
0 −
→
where hf is the energy of a single photon at the wavelength at which ASE is to be
evaluated, and
G (z , z ) = T F (z1 , z2 ) G R (z1 , z2 ) (15.107)
−
→ 1 2 −
→ −
→
is the net gain at the signal wavelength for signals propagating from z1 to z2 in the
forward direction. The transmission T F (z1 , z2 ) = exp[−αs (z2 − z1 )] is the passive
−
→
fiber transmission at the signal wavelength for signals propagating from z1 to z2 in
the forward direction, and
 
CR (λs , λp )
G R (z1 , z2 ) = exp [Pb (e−αp (L−z2 ) − e−αp (L−z1 ) ) + Pf (e−αp z1 − e−αp z2 )] ,
−
→ αp
(15.108)
is the Raman gain at the signal wavelength for signals propagating from z1 to z2 in the
forward direction. The DRA net gain G is given by GR TF , where GR = G R (0, L) is
−
→
the Raman on-off gain at the signal wavelength for forward propagating signals and
TF = T F (0, L) is the passive fiber transmission (≤1) at the signal wavelength for
−
→
forward propagating signals. The various gain parameters are depicted in Fig. 15.25.
In Eqs. (15.106) and (15.108), CR (λs , λp ) is the Raman gain efficiency (gain factor

NASE

Forward L Backward
pumps pumps
(Pf ) (Pb )
5
Pin Transparency
Net gain
0
Distributed Raman
Signal power (dB)

amplifier (DRA) G
–5 (Any value)
TF
(< 0 dB)
–10
GR
Passive transmission fiber
(≥ 0 dB)
–15

On-off gain
–20
0 20 40 60 80 100
Distance (km)

Fig. 15.25. Power evolution for a passive and a bidirectionally Raman-pumped transmission
fiber and definition of the corresponding Raman gain parameters.
 15. Multiple Path Interference 543

per unit of length and power) for signal and pump wavelengths λs and λp , respectively,
αp and αs are the loss coefficients at the pump and signal wavelengths, and Pb and Pf
are the backward and forward pump powers at their respective launch points on either
ends of the fiber. The expressions “copumping” and “counter-pumping” are also used
to describe forward and backward pumpings, respectively. The impact of nonzero
fiber temperature on ASE generation has been neglected in Eq. (15.106) but can be
incorporated without too much difficulty [92]. Under the assumption of perfectly
depolarized pumps, the ASE from Raman amplification is randomly polarized so that
the total spectral density of ASE in all polarization states is given by 2NASE .
As for any amplifier, the spectral density of ASE at the output of a DRA depends
on the degree of forward to total pump power and on the net gain. Figure 15.26(a)
shows the spectral density of ASE per polarization state NASE at the output of a
DRA for various values of net gain and Fig. 15.26(b) shows the OSNRASE (see
Eq. (15.49)) considering only the ASE levels given in Fig. 15.26(a). For Fig. 15.26(b),
the input power to the DRA, P in , is 0 dBm. The fiber considered is of the nonzero
dispersion fiber (NZDF) type with the parameters: L = 100 km, αs = 0.21 dB/km,
αp = 0.26 dB/km, Aeff = 55 µm2 , and CR (λs , λp ) = 0.68 (W km)−1 . One observes
in Fig. 15.26(a) that the level of ASE is lower for forward pumping than for backward
pumping. This can be understood by the difference in the net gain experienced by the
ASE from the moment of its generation to the output end of the fiber (see Eq. (15.106)).
The ASE generated near the fiber input by forward pumping experiences the loss of the
full length of the fiber in addition of the Raman on-off gain. For backward pumping,
theASE generation occurs mainly near the fiber output end and thus experiences only a
fraction of the fiber loss. As a result, forward pumping is more efficient to reduce ASE.

–40 58
Pin = 0 dBm
–45 56
ASE spectral density (dBm/GHz)

–50 54
–55 52
OSNR ASE (dB)

–60
50
–65
48
–70 % Forw. Pump.
0% 46
–75
20% 44
–80 40%
–85 60% 42
80% 40
–90
100%
–95 38
–20 –16 –12 –8 –4 0 4 8 12 –20 –16 –12 –8 –4 0 4 8 12
Net gain (dB) Net gain (dB)
(a) (b)

Fig. 15.26. Generation of ASE at the output of a distributed Raman amplifier and the corre-
sponding OSNRASE as a function of net gain: (a) spectral density of ASE per polarization state
(NASE ); (b) OSNRASE for P in = 0 dBm. OSNRASE is calculated considering that the DRA is
the only source of ASE. Fiber parameters are listed in main text.
544 J. Bromage et al.

Accordingly, as seen in Fig. 15.26(b), the OSNRASE is higher for forward pumping
than backward pumping. Note that the OSNRASE increase observed in Fig. 15.26(b)
for low net gain comes from the fact that only ASE from the DRA is considered when
plotting OSNRASE . For such low net gain, an optical amplifier following the fiber
span is necessary to bring back the signal to an appropriate power level for launching
in the next span or photodetection. Such an amplifier then becomes the main source
of OSNRASE degradation. For instance, for the conditions of Fig. 15.26, a 3 dB noise
figure amplifier following the fiber span would reduce the OSNRASE to 3̃4 dB after
amplification to bring back P in to 0 dBm.

15.6.1.2. Fiber Nonlinearity

Most fiber-optic communication systems are limited in signal power by the fiber Kerr
nonlinearity, especially for long-reach systems or systems with high fiber span loss.
These systems are generally designed to operate at the maximum signal power level
that leads to a controlled amount of tolerable waveform distortions.
One can generally separate the effects of Kerr nonlinearity into two classes (see
Table 15.1). The first class (Class I) involves continuous waveform distortions without
energy transfer between interacting fields. The interacting fields correspond to indi-
vidual channels for nonlinear WDM interactions [86] and individual pulses for intra-
channel nonlinear interactions in high-speed pseudolinear transmission [93]. The main
interactions in nonlinearity of Class I are self-phase modulation (SPM) [86], cross-
phase modulation (XPM) [86], and intrachannel cross-phase modulation (IXPM) [94].
The second class (Class II) of Kerr nonlinear interactions involves energy transfer
between interacting fields. The most important interactions belonging to this class are
four-wave mixing (FWM) [86] and intrachannel four-wave mixing (IFWM) [94].
For systems limited by SPM, XPM, and IXPM, different signal power evolutions
on a transmission fiber of given constant dispersion (along the fiber length) produce
approximately the same distortions on signals having the same integrated nonlinear
phase. One should note, however, that for this to be, one must optimize the dispersion
map of the transmission line for each Raman pumping configuration. Optimization
of dispersion maps refers to the process of choosing the value of dispersion compen-
sation for individual dispersion-compensation modules and their precise location in
a transmission line to optimize system performance.

Table 15.1. Separation of Nonlinear Interactions in Kerr Media in Two Classes, I and II, Which
Have Different Scaling Laws with Aignal Powers.
Class Nonlinear Effect
I Self-Phase Modulation (SPM)
Cross-phase Modulation (XPM)
Intrachannel Cross-Phase Modulation (IXPM)
II Four-Wave Mixing (FWM)
Intrachannel Four-Wave Mixing (IFWM)
 15. Multiple Path Interference 545

The integrated nonlinear phase of a signal power evolution P (z) over a fiber
segment of length L is given by
L
φNL (L) = γ (z) P (z) dz , (15.109)
0
where γ (z) = 2π n2 f0 /(c0 Aeff ) is the fiber nonlinear coefficient, n2 the nonlinear
refractive index coefficient, f0 the central optical frequency of the signal, and Aeff is
the fiber effective area [86]. Assuming that γ (z) does not vary appreciably over the
entire fiber length (which is approximately the case for fiber spans made of fibers of a
single type), the ratio of the integrated nonlinear phase in a DFA, φDFA (L), to that of
a passive fiber, φPas (L), at fixed input power is simply given by the ratio of average
signal powers,

L L
φDFA (L) 0 P in −
G (0, z) dz
→ αs
RNL = =
L = G (0, z) dz.
φPas (L) P in exp(−αs z) dz 1 − exp(−αs L) 0 −
0
→
(15.110)
The nonlinear ratio RNL represents the increase in Kerr nonlinearity of a DFA relative
to a passive fiber span for systems limited by nonlinearity of Class I. This nonlinear
ratio can be used to rescale signal power to equalize the effects of Kerr nonlinearity
of Class I.
For a system in which transmission is limited by the nonlinear interactions of
Class II (i.e., FWM and IFWM), the scaling of nonlinear interactions is no longer
proportional to the signal power. The origin of this fundamental difference between
the two classes lies in the scaling laws of any four-wave mixing process with signal
power. In its most simplified form (ignoring, for instance, signal depletion [86]) the
power generated during a FWM process is proportional to the product of the three
signal powers involved in the FWM process. The amplitude of the beating on the
electrical signal produced by this FWM is proportional to the ratio of FWM power to
the undistorted signal power. Assuming all neighboring WDM channels have the same
3
average power P s , then FWM power is proportional to P s and the ratio of FWM to
2
signal that determines signal distortions is proportional to P s . The same scaling laws
also apply to IFWM, which corresponds to FWM interactions between neighboring
overlapping pulses from the same channel [93]. Under the assumption that all pulses
within a channel have the same shapes and powers, the ratio of powers of the “shadow”
pulses (spurious pulses created in the empty bit slot as a result of IFWM) [93] to the
2 2
undistorted pulse power is also proportional to P s , where P s is average power of
a given channel. Note that for IFWM the scaling law applies to each individual
channel independently of the power of neighboring channels. The establishment of a
nonlinear ratio RNL for Class II nonlinearity requires, in general, knowledge of the
specific FWM or IFWM processes involved in the nonlinear interaction. We leave
this problem of forming RNL for nonlinearity of Class II for further studies.
In bidirectionally-pumped DRAs, even with identical net gains, the signal power
evolution depends strongly on the ratio of bidirectional pumping. This is illustrated
546 J. Bromage et al.

12
100 % % Forward
8 80 % pumping
4 60 %

Signal power (dB)

0 40 %
20 %
–4
0%
–8 Pas
sive
tran
smis
–12 sion
fibe
r
–16
–20
0 20 40 60 80 100
Distance (km)

Fig. 15.27. Different signal power evolutions for different ratios of forward to total pump
powers in a bidirectionally-pumped fiber. Power evolution in a passive transmission fiber is
shown for comparison. Different ratios of bidirectional pumpings lead to different levels of
fiber nonlinearity (see Fig. 15.28).

28
0% % Transparency
24 20% Forward
Nonlinear ratio RNL (dB)

40% pumping
20 60%
80%
16 100%
12

0
–21 –18 –15 –12 –9 –6 –3 0 3 6 9 12
Net gain (dB)

Fig. 15.28. Nonlinearity ratio RNL (in dB) of Kerr nonlinearity in a bidirectionally Raman
pumped 100-km-long transmission fiber for Class I nonlinearity (see Table 15.1).

in Fig. 15.27, which shows the signal power evolution for different percentages of
forward pump power to total pump powers for a DRA pumped to transparency. Thus
RNL depends strongly on the percentage of forward pump power to total pump power.
Figure 15.28 shows how RNL scales for a 100-km-long Raman-pumped transmission
fiber as a function of net gain and percentage of forward pumping. The fiber loss
coefficients at the signal and pump wavelengths αs and αp are 0.21 dB and 0.26 dB,
respectively. The passive span loss is 21 dB. For 100% backward pumping (0% for-
ward pumping), it is only when the fiber starts to approach transparency (21 dB on-off
 15. Multiple Path Interference 547

0
–2 Pref
–4 Non
line
arity

Signal power (dB)

–6 of c
lass
–8 I
–10
–12 Pref /RNL
–14
Passive fiber
–16
–18
–20
–22
0 20 40 60 80 100
Distance (km)

Fig. 15.29. Different signal power evolutions inside a transmission fiber leading to identical
fiber nonlinearities in a DRA as in a passive transmission fiber. Solid curve: signal power in
passive fiber. Dashed curve: same nonlinearity as passive fiber for Class I nonlinearity. The
DRA is transparent and bidirectionally-pumped equally on both sides (50% forward pumping).
Other parameters are identical to Fig. 15.28.

gain or 0 dB net gain) that the nonlinear ratio RNL starts to increase significantly. As
soon as forward pumping (copumping) is used, a few dBs of on-off gain immediately
translates into an increase in nonlinearity ratio. As a result, ensuring constant nonlin-
earity over one fiber span, the signal launch power must be adjusted for different pump
configurations. To illustrate this, Fig. 15.29 shows the signal power evolution for a
passive transmission fiber with input power P ref and for a bidirectionally-pumped
DRA where the input signal power is reduced from P ref to P ref /RNL to make the
effects of Kerr nonlinearity identical in both systems.
One should point out that the nonlinear ratio RNL is a measure of the averaged
effects of Kerr nonlinearity of Class I. However, for very large changes in power
evolution relative to the reference case, the nonlinear ratio RNL may lose some ac-
curacy. Such a case is exemplified by an extreme case where a DFA would have a
constant signal power evolution along the fiber length. For identical values of RNL ,
such a DFA would partially suppress some specific nonlinear interactions. In the case
of dispersion-compensated solitons, for instance, constant signal power evolution
suppresses the effect of XPM in symmetrized collisions [95, 96]. On the other hand,
partial soliton collisions are less affected by having constant power evolution [95, 96].

15.6.1.3. Double-Rayleigh Backscattering

Double-Rayleigh backscattering of a signal can be significantly enhanced in DFAs be-

cause of the nonuniformity of the gain. The total DRB signal power in all polarization
states PDRB in a DFA of length L is given by (undepleted pump approximation) [7],
L L
PDRB = P in G(αR S) 2
G (z, ζ ) G (z, ζ ) dζ dz, (15.111)
0 z − → ←
−
548 J. Bromage et al.

–12 –30
Net gain Signal loss
–16 Passive –32 0.18 dB/km
–5 dB 0.21 dB/km
–20 0 dB 0.24 dB/km
5 dB –34 0.27 dB/km
–24 10 dB 0.30 dB/km
–28 15 dB –36
RC (dB)

RC (dB)
–32 –38
–36 –40
–40
–42
–44
–48 –44

–52 –46
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Forward pumping (%) Forward pumping (%)
(a) (b)

Fig. 15.30. Dependence of the RC crosstalk ratio as a function of percentage of forward

pumping in a 100-km-long bidirectionally-pumped DRA for (a) various levels of net gains and
(b) various span losses. The loss at pump wavelength αp is set 0.05 dB/km higher than the loss
at signal wavelength αs . Other fiber parameters are described in main text.

where αR S is the Rayleigh backscatter coefficient, P in is the input signal power to

the DFA, and G (0, z) is the Raman net gain at the signal wavelength propagating
←−
in the backward direction from z to the fiber input end (z = 0). Note that different
forms of Eq. (15.111) have been derived previously in this chapter (see Eqs. (15.22),
(15.40), and (15.53)). It is worth pointing out that DRB is not randomly polarized,
but, as described in Section 15.3.4, has a degree of polarization 5/9, copolarized with
the signal [97].
To illustrate the effects of bidirectional pumping on DRB generation, we show
in Fig. 15.30 how the crosstalk ratio RC (= PDRB /P in G, see Eq. (15.50)) is af-
fected by bidirectional pumping of a DRA. The fiber parameters are identical to
those of Section 15.6.1.1 (i.e., L = 100 km, αs = 0.21 dB/km, αp = 0.26 dB/km,
Aeff = 55µm2 , and CR (λs , λp ) = 0.68 (W km)−1 ) with the Rayleigh backscatter co-
efficient αR S = 1.03 × 10−4 km−1 .
As seen in Fig. 15.30(a), the crosstalk ratio increases rapidly with increasing
Raman net gain except for symmetric bidirectional pumping (50% forward pumping)
where the crosstalk ratio does not increase as rapidly with net gain increase. The impact
of various span losses on the crosstalk ratio is shown in Fig. 15.30(b). As span losses
increase, the crosstalk ratio increases only when the bidirectional Raman pumping
is asymmetric (dominated by forward or backward pumping). As the bidirectional
pumping becomes symmetric, the crosstalk ratio exhibits little dependence on the
span loss for fixed fiber length. Clearly, symmetric bidirectional pumping significantly
decreases the generation of DRB (keeping a low RC ), especially for pumping near
transparency and above and for large span losses.
 15. Multiple Path Interference 549

The values of the RC crosstalk ratio plotted in Fig. 15.30 ignore the effects of pump
depletion. However, the impact of pump depletion on RC has been experimentally
investigated [49] and it was found that the values of the RC crosstalk ratio do not
depend strongly on pump depletion even for pumps depleted by up to 6 dB where a
reduction of only 0.3 dB in RC was observed [49].

15.6.1.4. Noise Figure

A quantity commonly used to characterize noise performance in optical amplifiers is

their noise figure, defined as [58, 98],
SNRin
NF ≡ , (15.112)
SNRout
where SNRin is the electrical signal-to-noise ratio SNR at the input to the amplifiers
where the only source of detection noise is the fundamentally unavoidable shot noise.
It is given by
ss 1 2 R 2 P 2
SNRin ≡ 2 = D2 in , (15.113)
σ1 σshot
where ss 1  and σ1 are the electrical signal current and current variance at the 1-bit peak.
ss 1  corresponds to RD Pin (see Eq. (15.67)) for the input signal to the amplifier, and σ1
corresponds to σshot for a shot noise limited signal (see Section 15.5.3), respectively.
The SNR at the output of the amplifier SNRout is obtained by considering all sources of
noise in the amplifier in addition to shot noise. For an amplifier limited by signal–ASE
and signal–DRB beat noises, the output SNR is given by
2 P2
RD out
SNRout = , (15.114)
2
σs−ASE + σs−DRB
2 + G σshot
2

where Pout = G Pin is the output signal variance from the amplifier of gain G σs−ASE
2
2
and σs−DRB are beat noise variances for ASE and DRB signal–ASE and signal–DRB
2 is the shot noise variance at the
beat noises, respectively (Section 15.5.1), and G σshot
amplifier’s output. In the presence of these types of beat noise, the noise figure of an
amplifier is then given by
 
2 2
1 σs−ASE σs−DRB
NF = 2
+ 2
+1 . (15.115)
G G σshot G σshot

In the limit of large optical filter bandwidths, the ratio of the signal–ASE beat noise
2
to shot noise variances σs−ASE 2 is given by 2 N
/ σshot ASE / hf [57] where NASE is the
spectral density of the opticalASE power copolarized with the signal. The contribution
from the signal–DRB beat noise is more difficult to evaluate, as it involves beating
of a colored noise with the signal. A detailed calculation of this type of beat noise is
presented in Section 15.5 and in [69]. In the framework of Gaussian pulse and filter
550 J. Bromage et al.

shapes and in the limit of large optical filter bandwidths, the ratio of the variances is
given by (see Eqs. (15.95) and (15.96))
2
σs−DRB 5/9 PDRB
= , (15.116)
2
σshot hf (Be2 + Bs2 /2)1/2

where 5/9 PDRB is the DRB power copolarized with the signal (see Section 15.3.4
and [97]), Be is the equivalent square bandwidth of the electrical filter, and Bs is
the equivalent square bandwidth of the optical signal (see Fig. 15.19). Note that in
Eq. (15.116), the ratio of signal–DRB beat noise to shot noise decreases as either
the signal bandwidth or receiver’s electrical bandwidth increases [81, 99]. From the
above results, one can explicitly write the noise figure NFDFA of a DFA as
 
1 2 NASE 5/9 PDRB
NFDFA = + +1 . (15.117)
G hf hf (Be2 + Bs2 /2)1/2

Both NASE and PDRB refer to quantities evaluated at the amplifier’s output. Note
that the noise figure incorporating double-Rayleigh scattering now depends on signal
power through PDRB = RC P out .
Optical amplifiers are frequently concatenated. The chain of amplifiers itself can
be considered as a single amplifier having its own noise figure. In the case of concate-
nation of two amplifiers, the first one labeled 1 with noise figure NF1 followed by a
second labeled 2 and with noise figure NF2 , the noise figure NF12 of the amplifier’s
pair is given by [58, 98]
NF2 − 1
NF12 = NF1 + , (15.118)
G1
where G1 is the net gain of the first amplifier. The noise figure for concatenation of
more than two amplifiers can be obtained by using Eq. (15.118) recursively.

15.6.1.5. Output Signal-to-Noise Ratios

To quantify the benefits of using DRAs instead of passive fiber spans in terms of an
improvement of electrical SNR at the receiver, we first need to evaluate the ratio of
SNRs at the end of a transmission line operating under these two different conditions
(see Fig. 15.31). Under the first operating condition (Fig. 15.31(a)), each fiber span
remains passive, and the optimal input signal power P ref is adjusted to minimize
BER. Note that this optimum signal power typically leads to an eye closure at the
end of the line in the range of 10 to 30% relative to the line input. The noise figure
of the entire span (including the passive fiber and all other elements of the span)
is labeled NFPas , and is evaluated for the span input signal power P ref . Under the
second operating condition, using DRAs (Fig. 15.31(b)), the signal power at the input
of each DRA has to be reduced by a nonlinear ratio RNL relative to P ref to ensure
similar fiber nonlinearities (Section 15.6.1.2). Similarly to NFPas , the DRA span noise
figure is evaluated at the input signal power P ref /RNL and is labeled NFAct . The SNR
at the output of a transmission line is obtained by generalizing Eq. (15.114) to take
 15. Multiple Path Interference 551

(a) Single span using passive fiber (b) Single span using DRA
Pref G Pref
Pref G Pref
RNL RNL
NFPas NFAct

Passive span Active span

(c) Transmission line using passive fibers

Pref Pref Pref Pref
NFPas NFPas NFPas

Tx … Rx

(d) Transmission line using DRAs

Pref Pref Pref Pref Pref
RNL RNL RNL RNL RNL
NFAct NFAct NFAct

Tx … Rx

NSpan

Fig. 15.31. Schematics of two operating conditions of a pair of amplifiers and transmission
lines. In (a) and (b), two amplifiers have identical gains but different noise figures and operate
with different input signal powers. These amplifiers can be used to represent individual spans
in transmission lines using (c) passive fibers and (d) DRAs.

into account the accumulation of signal–ASE and signal–DRB variances in a chain

of NSpan identical spans, and is given by,
2 P2
RD out
SNRout = . (15.119)
NSpan (σs−ASE + σs−DRB
2 2 ) + G σshot
2

Applying Eq. (15.119) to the two operating conditions of Fig. 15.31 we can obtain
the ratio of output SNRs, RSNR , at the end of a transmission line,
Act
SNRout 1 NSpan (NFPas − 1) + 1
RSNR ≡ = . (15.120)
Pas
SNRout RNL NSpan (NFAct − 1) + 1

Note that for a single span, RSNR reduces to the simple expression NFPas /(RNL NFAct )
corresponding to the ratio of the active and passive span noise figures weighted by the
nonlinear ratio (see Fig. 15.31). One should point out that in systems not producing
signal distortions from fiber nonlinearities, Eq. (15.120) also applies if one sets the
nonlinearity ratio RNL to 1.
552 J. Bromage et al.

NFPas

Pref PrefTF PrefTF Ga PrefTF GaTL = Pref

Ga TL
Passive transmission fiber

(a)

NFAct
Pref Pref Pref Pref Pref
G T G T G G T G T =
RNL RNL R F RNL R F b RNL R F b L RNL

Gb TL
FW Distributed fiber BW
pumps amplifier (DFA) pumps
Single span
(b)

Fig. 15.32. Schematics of one span of transmission lines using (a) passive transmission fibers
and (b) distributed Raman amplification.

We now evaluate RSNR for a typical hybrid span transmission line. Figure 15.32(a)
represents a typical single span of a transmission line using passive transmission fibers
and Fig. 15.32(b) applies to a hybrid span transmission line using DRA. The fiber
nonlinearity is made equal in both lines by reducing the signal input power to the
DFA by the nonlinear ratio RNL . A discrete lossy element of transmission TL has been
inserted after the discrete amplifier to account for the loss of optical elements such
as gain equalizers, add-drop multiplexers, and crossconnects. By using Eq. (15.118)
for the concatenation of noise figures of a chain of amplifiers, one can derive the
noise figures for a passive span (NFPas ) and for a hybrid span (NFAct ) including all
elements of each span. These noise figures for the entire spans are
NFa
NFPas = + 1 − TL (15.121)
TF
and
NFb − 1
NFAct = NFDFA + + 1 − TL , (15.122)
G R TF
where NFa and NFb are the noise figures of the discrete amplifiers in passive and
hybrid span transmission lines, respectively. Assuming that the discrete amplifiers of
Fig. 15.32 are fully inverted (3-dB noise figures when operated at high gain), one can
easily show that
1 NSpan (2/TF − 2 TL ) + 1
RSNR = , (15.123)
RNL NSpan [ NFDFA + 1/(GR TF ) − 2 TL ] + 1
 15. Multiple Path Interference 553

where DRB has been neglected in the calculation of NFPas as RC is typically lower
than −50 dB for passive fiber spans. Note that our choice of fully inverted amplifiers
for transmission lines using passive transmission fibers in Eq. (15.123) corresponds
to the best possible scenario for these systems. More realistic optical amplifiers incor-
porating potentially many functionalities will most likely have higher noise figures
that typically range between 4 and 10 dB. One should emphasize that the benefits
of using DRA in systems with such high noise figure amplifiers are larger than in
systems with ideal amplifiers.
The parameter RSNR is our final measure of system performance. Our interpreta-
tion of this factor is as follows: The factor RSNR is the improvement of signal-to-noise
ratios at the end of a transmission line using distributed Raman amplification relative
to a line using passive transmission fibers and ideal optical amplifiers. This SNR ratio
is obtained using different signal input powers to the spans for the two lines corre-
sponding to equal average effects of fiber nonlinearities. Therefore the factor RSNR
gives the amount of additional dBs of SNR that become available by making use of
distributed Raman amplification in a passive transmission line.

15.6.1.6. Results and Discussion

We illustrate the technique described above by considering the optimization of a

hybrid Raman-pumped system limited by the nonlinearity of Class I (SPM, XPM,
or IXPM). This example encompasses such systems as long-haul WDM systems
operating at 10 Gb/s over NZDF and standard single-mode fibers (SSMFs) or at 40
Gb/s over NZDF.
The DRA is made of a NZDF operating at 1550 nm without any nonreciprocal
element present in the fiber path. Under such conditions, forward and backward gains
G and G are identical. The fiber parameters are identical to those in Section 15.6.1.3,
−
→ ←−
except for L = 80 km. The nonlinear coefficient γ = 1.67 W−1 km−1 .
Figure 15.33 shows isocontours of SNR ratios, RSNR (in dB), calculated using
Eq. (15.123) as a function of the DRA net gain and the percentage of forward pump-
ing. An improvement in SNR from DRA translates into a positive value for RSNR (in
dB). Figures 15.33(a) and 15.33(b) are isocontours when either ASE or DRB noise
sources are considered in the DRAs, respectively. We considered system lengths of
40 spans. For the beat noise calculation from DRB we used an electrical receiver
bandwidth Be = 5 GHz and an optical signal bandwidth Bs = 9.4 GHz corresponding
to the equivalent square optical bandwidth of a 50% duty cycle return-to-zero (RZ)
format. The DRB power in Eq. (15.111) is calculated using P in = P ref /RNL , where
P ref = 0.5 mW (−3 dBm). The lossy element has 10 dB of loss (TL = 0.1). It is worth
mentioning that for up to 10 dB of loss, the presence of this lossy element does not
have a significant impact on either the value of RSNR or the optimum pumping scheme
configuration. Figure 15.33(c) shows the isocontours when both sources of noise are
simultaneously included in the calculation of RSNR . One sees in Fig. 15.33(a) that
when ASE is the only source of noise in the DRA, the optimum pumping scheme cor-
responds to 30% forward Raman pumping power and the optimum net gain is a few
554 J. Bromage et al.

–15 1 1
–10 2 2

Net gain (dB)

3
–5
(a) ASE only
0 3 2
5 1
3 0 –1–2
10 2
0 20 40 60 80 0 100
–15 1
2
–10 3
Net gain (dB)
4
5
–5 6
7 DRB only
(b) 7
0 6
5 4
2
10
0 20 40 60 80 0 100
–15
1
–10
Net gain (dB)

2
–5
(c) ASE + DRB
0 2
2 1 0 –1
1
5 0
–1
10
0 20 40 60 80 100
Forward pumping (%)

Fig. 15.33. Ratio of output SNRs, RSNR (in dB): (a) considering ASE only in the DRA; (b)
considering DRB only in the DRA; and (c) with both sources of optical noise in the DRA
included.

dBs above transparency. When DRB is the only source of noise (Fig. 15.33(b)), the op-
timum pumping scheme corresponds to 35% forward pumping and a slightly negative
net gain (−2 dB). When both noise sources of DRAs are included (Fig. 15.33(c)), the
optimum Raman pumping scheme is 40% forward pumping with a net gain of −3 dB.
At high net gain, generation of DRB in the DRA degrades the SNR whereas for low net
gain the limitations come from the generation of ASE. One should note that generation
of DRB can be strongly reduced by inserting isolators in the fiber span while genera-
tion of ASE remains unaffected. Thus limitations from ASE seen in Fig. 15.33 can be
interpreted as an ultimate limit to improvement in system performance from DRA.
More generally, the optimal Raman pumping configuration depends on the length
of each fiber span and type(s) of fiber used, their loss and backscatter parameters, the
noise figure of the discrete amplifiers, the class of nonlinearity limiting transmission,
the data rate, and filter design. Moreover, to achieve broadband gain flatness, a few
Raman pumps are generally required and the procedure presented in this section needs
 15. Multiple Path Interference 555

to be generalized to accommodate multiple pumps. Optimization of Raman pumping

should be applied to each system that one wishes to optimize.

15.6.1.7. Bit-Rate Scaling Laws for System Design

Sections 15.6.1.1 to 15.6.1.3 described the three important parameters determining

system performance in transmission lines using Raman amplification. These impair-
ments can be summarized by the three parameters: OSNRASE , the signal power per
channel P s , and OSNRMPI . Scaling of system performance with bit rate per channel
R for fixed received OSNRASE and fixed received OSNRMPI have been described in
Section 15.5.2.4. In this section, we first consider the effect of bit rate scaling on MPI
requirements for a transmission line at fixed received OSNRASE and fixed received
OSNRMPI . We then discuss the effects of varying the signal power P s .
A transmission line using Raman amplification designed to minimize the effects
of optical noise sources will generate levels of ASE and MPI that produce comparable
signal degradation at the receiver for a given bit rate R (degradations can be expressed,
for instance, by Eqs. (15.81), (15.82), (15.88), and (15.89)). Let’s call OSNRASE
and OSNRMPI the corresponding OSNR levels at this optimum operating point. If
one wants to increase the bit rate n-fold at fixed signal power then, as discussed in
Section 15.5.2.4, the required OSNRASE increases n-fold and the required OSNRMPI
remains unchanged. Expressed differently, the electrical bandwidths Be and Bs are
naturally scaled up n-fold with the bit rate, reducing n-fold the relative impact of
the MPI (or DRB) term in the noise figure given in Eq. (15.117). Consequently, the
OSNRASE at the bit rate n R will be insufficient to provide the same BER as for bit rate
R. The situation can be corrected, for instance, by operating at higher net gain for the
DRA which will increase OSNRASE , at the expense of OSNRMPI (see Fig. 15.33(b)).
A new optimum balance between the effects of ASE and MPI on detection will be
established. Another alternative to recover system performance is to simply scale back
system reach to increase OSNRASE and to allow relaxed specifications on sources of
MPI. Additional margin on MPI specifications especially becomes of value when
sources of MPI with high variability are present in a system. An example of a source
of MPI that is difficult to predict accurately is the MPI produced by discrete reflections
by damaged or dirty connectors in the transmission path.
We now additionally consider the effect of varying the signal power P s . For a
given transmission line layout, the level of ASE is fixed and OSNRASE is determined
solely by the signal power P s . As pointed out earlier, the required OSNRMPI does
not change with P s or bit rate whereas the required OSNRASE increases n-fold with
bit rate and P s . As a result, including both the effects of ASE and MPI, an n-fold
increase in bit rate R followed by a corresponding n-fold increase in P s leads to
OSNRASE and OSNRMPI that leaves the BER unaffected (assuming identical signal
degradations from fiber nonlinearities). It is interesting to note that for a fixed signal
spectral density (= R/f , where f is the channel spacing), the n-fold scaling for
P s with R for constant BER is automatically fulfilled with an optical amplifier with
constant total output power.
556 J. Bromage et al.

Whether it is possible to raise n-fold the signal power P s in a transmission line

depends essentially on fiber nonlinearity. The maximum value of P s that can transmit
with limited degradation from nonlinear signal distortions depends on a large number
of system parameters such as fiber types, dispersion maps, modulation formats, and
so on. It is beyond the scope of this chapter to determine how P s scales with bit rate.
Nevertheless, as can be assessed from the published literature (see [93] and references
therein), we can identify some cases where it is more likely to observe a significant
increase in the maximum value of P s as the bit rate R increases. An example of such
a case is for a spectral efficiency of 0.4 bits/s/Hz where systems operating at 40 Gb/s
(100 GHz channel spacing) have generally higher P s than systems operating at 10
Gb/s (25 GHz spacing). For such spectral efficiency, just from the transmission and
detection standpoints, it may become advantageous to use 40 Gb/s per channel instead
of 10 Gb/s if, for the particular system design of interest, the maximum value of P s
at 40 Gb/s exceeds by more than 6 dB the maximum P s at 10 Gb/s.

15.6.2. Section Summary

In this section we derived a procedure to evaluate the ratio of SNRs at the end of a
transmission line using passive transmission fibers and DRAs (hybrid spans) keeping
constant the effects of fiber nonlinearity. Two classes of fiber nonlinearity with their
respective scaling laws were introduced to accurately describe the effect of various
nonlinear interactions. Using the procedure described in this section, we explored
bidirectional Raman pumping of a hybrid span transmission line and showed that
there exist an optimum net gain and percentage of forward pumping that maximize
the benefits of distributed Raman amplification in communications systems.

A15.A. Derivation of Autocorrelation of DRB-Induced

Intensity Noise
This appendix sketches important steps in the derivation of the autocorrelation =II (τ )
of the DRB-induced intensity fluctuations II (t), Eq. (15.32). The mathematical
steps involved are also needed in the derivation of the DRB field autocorrelation
=ε DRB (t, t + τ ), Eq. (15.39).
We start by writing the autocorrelation of the intensity fluctuations II (t) in terms
of the involved scalar optical fields,
=. /T
=II (t, t + τ ) = ε s (t)εε DRB (t) +εε Ts (t)εε TDRB (t)
. />
× ε s (t + τ )εε DRB (t + τ ) +εε Ts (t + τ )εε TDRB (t + τ ) , (A15.1)

where we set ε s (L, t) =εε s (t) and ε DRB (L, t) =εε DRB (t) for notational convenience.
Next we perform the multiplication, and substitute the scalar version of the integral
expression in Eq. (15.28) for ε DRB (t) (i.e., we ignore the matrices J12 and J21 ). Then
we interchange the order of integration and ensemble averaging. The result is a sum of
 15. Multiple Path Interference 557

four terms, each containing a fourfold integration. Because the differential Rayleigh
backscatter coefficient ρ (z) is statistically independent of any random fluctuations of
ε s (t), we can separate the average overρ ρ (z) and overεε s (t), and are left with ensemble
averages of the form (15.30) and (15.31) within the integrals. Due to Eq. (15.31), two
of the four integral expressions vanish. The remaining two fourfold integrals are
complex conjugated to each other, which turns their sum into a real part,
 L L z2 z4
=II (t, t + τ ) = 2Re dz2 dz4 dz1 dz3 ρρ T (z1 )ρ
ρ T (z2 )ρ
ρ (z3 )ρ
ρ (z4 )
0 0 0 0
* *
× G (z1 , z2 ) G (z1 , z2 ) G (z3 , z4 ) G (z3 , z4 )
−
→ ←− −
→ ←−
× εε Ts (t)εε Ts (t − 2(z2 − z1 )/vg )εε s (t + τ )εε s (t + τ − 2(z4 − z3 )/vg )

× e−j 2β(z2 −z1 ) ej 2β(z4 −z3 ) .
(A15.2)
Inserting the correlation relation (15.30) for ρ (z), the fourfold integral in Eq. (A15.2)
splits into a sum of two terms, =II (t, t + τ ) = 2Re{F1 (t, t + τ ) + F2 (t, t + τ )}, with
L L z2 z4
F1 (t, t + τ ) = dz2 dz4 dz1 dz3 f (z1 , z2 , z3 , z4 ; t, t + τ )δ(z1 − z4 )δ(z2 − z3 )
0 0 0 0
(A15.3)
L L z2 z4
F2 (t, t + τ ) = dz2 dz4 dz1 dz3 f (z1 , z2 , z3 , z4 ; t, t + τ )δ(z1 − z3 )δ(z2 − z4 ),
0 0 0 0
(A15.4)
* *
where f (z1 , z2 , z3 , z4 ; t, t + τ ) = (αR S)2 G (z1 , z2 ) G (z1 , z2 ) G (z3 , z4 ) G (z3 , z4 )
−
→ ←
− −
→ ←−
εε Ts (t)εε Ts (t − 2(z2 − z1 )/vg )εε s (t + τ )εε s (t + τ − 2(z4 − z3 )/vg )e−j 2β(z2 −z1 ) ej 2β(z4 −z3 ) .
We first show that F1 (t, t + τ ) ≡ 0. The innermost integral (over z3 ) results in
replacing f (z1 , z2 , z3 , z4 ; t, t + τ ) by f (z1 , z2 , z2 , z4 ; t, t + τ ) if the Dirac functional
can take action, that is, if z2 − z3 = 0 lies within the range of integration. This is only
the case if z2 < z4 . Otherwise the integration result is zero. Next we integrate over
z1 , which results in replacing f (z1 , z2 , z2 , z4 ; t, t + τ ) by f (z4 , z2 , z2 , z4 ; t, t + τ ),
if z1 − z4 = 0 lies within the range of integration. This is only the case if z2 > z4 .
Therefore, the integrals over z1 and z3 are mutually exclusive, and F1 (t, t + τ ) ≡ 0,
independent of the nature of f (z1 , z2 , z3 , z4 ; t, t + τ ).
Next we simplify F2 (t, t + τ ) ≡ 0: By the same argument as before, the integral
over z3 results in replacing f (z1 , z2 , z3 , z4 ; t, t + τ ) by f (z1 , z2 , z1 , z4 ; t, t + τ ) if
z4 > z1 . Otherwise the integration result is zero. A subsequent integration over z4
results in replacing f (z1 , z2 , z1 , z4 ; t, t + τ ) by f (z1 , z2 , z1 , z2 ; t, t + τ ), and in re-
placing the condition z4 > z1 by z2 > z1 . Because z4 covers the entire range of z2 , the
Dirac functional always takes action here. The remaining expression reads
L z2
F2 (t, t + τ ) = dz2 dz1 f (z1 , z2 , z1 , z2 ; t, t + τ ). (A15.5)
0 0
558 J. Bromage et al.

The necessary condition z2 > z1 for a nonvanishing z3 -integration is always satisfied

here, therefore this is the final result. Assuming further that the correlation length
of the stationary source is much shorter than paths that lead to significant contribu-
tions to the DRB field, we can write εε Ts (t)εε Ts (t − 2(z2 − z1 )/vg )εε s (t + τ )rεs (t + τ
− 2(z2 − z1 )/vg ) = εε Ts (t)εε s (t + τ )εε Ts (t − 2(z2 − z1 )/vg )εε s (t + τ − 2(z2 − z1 )/
vg ) = |=ε s (τ )|2 , and directly arrive at Eq. (15.32).

A15.B. Derivation of MPI Field Autocorrelation

This appendix describes a way to calculate the autocorrelation and spectrum of a
digitally modulated optical sequence, given the signaling constellation and the pulse
waveform. The approximations are tailored to meet the needs of signal–MPI beat
noise calculations.
We first write the digitally modulated, optically filtered data signal as
∞

8
ε s (t) = a k8
εp (t − k/R), (A15.6)
k=−∞
where a k denote the (possibly complex) signaling points, assumed random for math-
ematical convenience,8 εp (t) is the optically filtered field of an isolated pulse, and 1/R
denotes the bit duration. Note that this model leads to analytically tractable results,
but has its formal shortcomings caused by pulse field overlap whenever the individ-
ual pulses 8εp (t) are longer than the bit duration. This can be the case for high duty
cycle RZ signaling, for example, and, in particular, for NRZ modulation. Neverthe-
less, even in such limiting cases the sequence spectral envelope closely matches the
isolated pulse spectrum, and therefore results in accurate approximations when used
for the signal–MPI beat noise calculation.
As a next step, we calculate the sequence autocorrelation, which can generally be
shown to be [42]
∞


ε s (τ
) = R
=8ε s (τ, τ ) = 8
ε Ts (τ )8 =a (m)=8εp (τ − τ
− m/R). (A15.7)
m=−∞
This equation takes the form of a discrete-time convolution of the data autocorrelation
=a (m) = aa ka Tk+m  (A15.8)
and the pulse autocorrelation
∞
=8εp (τ − τ
) = 8 εpT (τ − τ
+ ξ )dξ.
εp (ξ )8 (A15.9)
−∞
To give some examples, the data autocorrelations for plain on-off keying and carrier-
suppressed on-off keying (CS-OOK, used widely in the form of 67% duty cycle
CS-RZ) can be shown to read
1
=a ,OOK (m) = [δ(m) + 1] and (A15.10)
4
1
=a ,CS-OOK (m) = [δ(m) + (−1)m ]. (A15.11)
4
 15. Multiple Path Interference 559

Assuming Gaussian pulses and filters (cf. Eqs. (15.90) and (15.91)), the pulse auto-
correlation reads
8p exp[−π Bs2 /(1 + Bs2 /Bo2 )(τ − τ
)2 ],
=8εp (τ − τ
) = E (A15.12)
8p is the filtered pulse energy. The autocorrelations for the two optical se-
where E
quences thus read
1 ?  @
=8ε s ,OOK (τ, τ
) = 2=8εp (τ − τ
) + =8εp (τ − τ
− m/R) and
4/R
m, =0
(A15.13)
1 ?  @
=8ε s ,CS-OOK (τ, τ
) = 2=8εp (τ − τ
) + (−1)m =8εp (τ − τ
− m/R) .
4/R
m, =0
(A15.14)
Translating these expressions to the frequency domain reveals the well-known facts
that the power spectra are composed of
• a continuous part, corresponding to the δ(m) portion of the data autocorrelations
(cf. Eqs. (A15.10) and (A15.11)), and
• a comb of tones spaced at integer multiples of R (OOK) and at integer multi-
ples of R offset by R/2 (CS-OOK), corresponding to the sum of shifted pulse
autocorrelations.
Equations (A15.13) and (A15.14) are visualized in Fig. A15.1, evaluated for Bs = 2R
and Bo = 2.2R. The figure shows that the shifted pulse autocorrelations (dotted for
OOK and dashed for CS-OOK) may overlap, suggesting that it could be insufficient to
consider only the center part of the optical signal autocorrelation (shaded). However,
because the signal–MPI beat noise integral (15.64) only takes into account a range
of the optical signal autocorrelation corresponding to −Th < τ − τ
< Th , where Th
stands for some effective duration of the detector impulse response (measured, e.g., by

OOK
Γεs ( τ,τ)

CS-OOK

–3 –2 –1 0 1 2 3
τ – τ

Fig. A15.1. Visualization of optical sequence autocorrelations for plain OOK and carrier-
suppressed OOK. Gaussian pulses and filters are assumed, with Bs = 2R and Bo = 2.2R.
560 J. Bromage et al.

a 1/e drop of h(t)), it can be justified within the frame of a reasonable approximation
to neglect all shifted replicas of the pulse autocorrelation in the beat noise integral,
and to keep only the main (centered) portion. The shorter the optically filtered pulses,
and the larger the bandwidth of the receiver electronics, the better will this approxi-
mation match realistic receivers. In general, for plain OOK, the approximation will
underestimate the signal–MPI beat noise, whereas for CS-OOK it will overestimate
the beat noise, as evident from comparing the shaded portion of the autocorrelation
in Fig. A15.1 with the full autocorrelation (solid lines).
Physically, the above approximation corresponds to neglecting the beat noise
stemming from discrete tones in the MPI field spectrum. These tones fall on top of
the tones in the signal spectrum, thus their main beating contributions will be at low
electrical frequencies (determined by the laser linewidth). Because the integral effect
of low frequencies is less pronounced for reasonably broadband receiver electronics,
the error in neglecting these tones becomes small in this regime.
Taking note of all approximations made above, we can finally write the MPI field
autocorrelation, for OOK and CS-OOK alike, as
*
=8ε s (τ, τ
) ≈ PMPI
P
1/(1 + Bs2 /Bo2 ) exp[−π Bs2 /(1 + Bs2 /Bo2 )(τ − τ
)2 ], (A15.15)
P
where the normalization to PMP I , the copolarized optical MPI power measured in
front of the WDM demultiplexer has been further introduced.

A15.C. List of Variables

General Notation Explanation Occurrence
8
x quantity x after optical demultiplexing filter Fig. 15.16
<
x peak of pulsed quantity x Eqs. (15.87), (15.90)
<
8
x optically filtered peak of pulsed quantity x Eq. (15.87)
x time average of quantity x Eq. (15.73)
8
x time average of optically filtered quantity x Eq. (15.87)
x bold print indicates random variables
· ensemble average
c.c. complex conjugate of expression preceding it
F {·} Fourier transform
Re[·] real part
Symbol Explanation
<
a characterizes the optical pulse shape after demultiplexer Eq. (15.79)
αp , αs loss at the pump and signal wavelengths Section 15.3.2
αR loss coefficient per unit length due to Rayleigh scattering Section 15.3.1
Aeff effective area of optical transverse mode Section 15.3.1
β propagation constant Eq. (15.26)
B(f ) demultiplexer’s complex baseband transfer function Fig. 15.16
b(t) demultiplexer’s complex baseband impulse response Fig. 15.16
Be electrical bandwidth of receiver Fig. 15.19
Bref reference bandwidth for OSNRASE Section 15.4.1
Bo optical bandwidth used in OSNRASE definition Fig. 15.19
Bs optical signal bandwidth Fig. 15.19
 15. Multiple Path Interference 561

Symbol Explanation Occurrence

CR Raman gain efficiency Section 15.3.2
CP8 (t1 , t2 ) autocovariance of optical power at detector Eq. (15.61)
dS , dG duty cycle of signal and gating pulses Section 15.4.3
δ(t) Dirac’s δ-functional
f linewidth of Lorentzian source spectrum
full-width, half-maximum) Section 15.2.1.2
II (t) time-dependent intensity fluctuations Eq. (15.4)
OSNRASE,dB OSNR penalty due to MPI Section 15.5.2.3
QdB Q-factor penalty due to MPI Section 15.5.2.2
z distance between scatter sites at z1 and z2 Section 15.3.2
e elementary charge
erfc(·) complementary error function Eq. (15.70)
ε amplitude of optical field at demultiplexing optical filter Fig. 15.16
−
→ε optical field amplitude vector Eq. (15.1)
ε s (f ) Fourier transform of εs (t) Eq. (15.42)
f frequency of optical or electrical quantities
fo central optical frequency of source Eq. (15.1)
fs peak signal attenuation factor due to filtering Eq. (15.73)
2
fs-N attenuation factor for variance of beating between signal
and ASE or MPI due to filtering Eqs. (15.74), (15.75)
FOSNR OSNR-based MPI tolerance factor Eqs. (15.83), (15.103)
FQ Q-based MPI tolerance factor Eq. (15.84), Fig. 15.23
g gain coefficient per unit length Section 15.3.2
γ nonlinear coefficient Section 15.3.2
=x (t, t + τ ) autocorrelation of the stochastic process x (t) Eq. (15.6)
G net gain of an amplifier Fig. 15.25
G [G] net intensity gain in the forward [backward] directions Section 15.2.1.1
−
→ ← −
G
(z) integrated round-trip gain from all paths with the same z Fig. 15.8
←→
G [ G R] on-off gain of a Raman amplifier in the forward
−
→R ← −
[backward] propagating direction Fig. 15.25
H (f ) transfer function of optoelectronic conversion chain Fig. 15.16
h(t) impulse response of optoelectronic conversion chain Fig. 15.16
h Planck’s constant
I optical intensity Section 15.2.1.1
I identity matrix
J12 Jones matrix for propagation from z1 to z2 Eq. (15.26)
L length of fiber amplifier Section 15.3.2
λs , λp wavelength of signal or pump light Section 15.3.2
M number of reflection sites Section 15.2.1.3
M Müller matrix describing polarization evolution Section 15.3.4
mi elements of Müller transmission matrix Mt Section 15.3.4
N number of interfering paths Fig. 15.3
n refractive index
n2 Kerr nonlinear coefficient Section 15.6.1.2
−
→
n (t) sum of all interferers’ optical fields Eq. (15.17)
NASE ASE power spectral density per (polarization) mode
in front of optical demultiplexing filter Eqs. (15.49), (15.106)
NF noise figure Section 15.6.1.4
NSpan number of spans in a transmission line Fig. 15.31
φ (t) instantaneous phase of optical field Eq. (15.1)
φNL (z) accumulated nonlinear phase after transmission Section 15.6.1.2
562 J. Bromage et al.

Symbol Explanation Occurrence

−
→
ps , −→
pr polarization vector for signal and doubly reflected fields Section 15.2.1.1
P optical power


Ps
, PDRB average signal and DRB power measured on gated OSA Section 15.4.3
Pp pump power Section 15.3.2
Pf , Pb forward and backward Raman pump power Fig. 15.25
P in , P out power in to and out of amplifier Fig. 15.25
Pin , Pout peak power of the 1-bit after filtering Eq. (15.113)
P ref reference average signal power for nonlinear effects Section 15.6.1.2
Q Personick’s Q-factor Eq. (15.71)
Qref reference Q-factor in the absence of MPI Eq. (15.80)
RC MPI inband crosstalk ratio Eq. (15.50)
RC P MPI inband crosstalk ratio, same polarization states Eq. (15.51)
Ri intensity reflectivity of discrete reflection at zi Fig. 15.4
ρ (zi ) differential Rayleigh backscatter coefficient at zi Eq. (15.28)
r inverse extinction ratio of optical signal Eq. (15.76)
RIN(f ) relative intensity noise Eq. (15.11)
R data rate
RD overall optoelectronic conversion factor Fig. 15.16
RNL nonlinear ratio for nonlinearity of class I Section 15.6.1.2
RSNR ratio of SNRs at the output of a transmission line Eq. (15.120)
S dimensionless backscatter recapture fraction Eq. (15.18)
s (t) electrical signal at decision gate ([A] or [V]) Fig. 15.16
ss 0,1  sampled electrical signal for 1-bit and 0-bit Eq. (15.77)
σ 2 (t) total variance of electrical signal at decision gate Eq. (15.60)
2 (t)
σshot shot noise variance at decision gate Eq. (15.100)
2 (t)
σs−N variance of electrical signal at decision gate due to Eqs. (15.64),
beating between signal and ASE, MPI, or DRB (15.68), (15.95)
2
σN-N variance of electrical signal at decision gate due to
beating of ASE, MPI, or DRB with themselves Eqs. (15.65), (15.69)
2
σ0,1 total variance of electrical signal sample for
0-bit and 1-bit at decision gate Eq. (15.71)
Sx (f ) power spectral density of random process x (t) Eq. (15.9)
−
→
S = (S0 , S1 , S2 , S3 ) Stokes vector describing polarization of light Section 15.3.4
SNRin , SNRout SNR at the input and output of a span Section 15.6.1.4
τ correlation time variable Eq. (15.6)
Td round-trip delay time Fig. 15.4
T (T F) transmission of a passive fiber in the forward
→F ←
− −
(backward) propagating direction Section 15.6.1.1
TL transmission of a lossy element Fig. 15.32
vg group velocity at signal wavelength Section 15.3.3
zi coordinate at position i
 15. Multiple Path Interference 563

A15.C. List of Acronyms

Acronym Explanation Occurrence

AOM acoustooptic modulator Fig. 15.14
ASE amplified spontaneous emission Sections 15.4.1 and 15.6.1.1
BER bit error ratio Section 15.5.2
BW backward Fig. 15.32
ccG circularly symmetric complex Gaussian Section 15.2.2
CSRZ carrier-suppressed return-to-zero Fig. 15.10
CW continuous wave
DFA distributed fiber amplifier
DFB distributed feedback laser
DOP degree of polarization Section 15.3.4
DPSK differential phase shift keying Fig. 15.10
DRA distributed Raman amplifier
DRB double-Rayleigh backscattering
DUT device under test Fig. 15.11
ECL external cavity laser
EDFA erbium-doped fiber amplifier
ESA electrical spectrum analyzer Fig. 15.11
FEC forward error correction
FW forward Fig. 15.32
FWM four-wave mixing Section 15.6.1.2
IFWM intrachannel four-wave mixing Section 15.6.1.2
ISI intersymbol interference Section 15.5.2
IXPM intrachannel cross-phase modulation Section 15.6.1.2
MPI multiple-path interference
N denotes either ASE or MPI Section 15.5.1
NRZ nonreturn-to-zero
NZDF nonzero dispersion fiber
OOK on-off keying
OSA optical spectrum analyzer Fig. 15.11
OSNR optical signal-to-noise ratio Section 15.4
PDF probability density function Section 15.2.2
PC polarization controller
RZ return-to-zero Fig. 15.10
RF radio frequency
Rx receiver
SNR signal-to-noise ratio Section 15.6.1.4
SOP state of polarization Section 15.3.4
SPM self-phase modulation Section 15.6.1.2
SRB single-Rayleigh backscattering
SSMF standard single-mode fiber
Tx transmitter
VOA variable optical attenuator Fig. 15.11
WDM wavelength-division-multiplexing
WSS wide-sense stationary Eq. (15.6)
XPM cross-phase modulation Section 15.6.1.2
564 J. Bromage et al.

References
[1] H. Takahashi, K. Oda, and H. Toba, Impact of crosstalk in an arrayed-waveguide multi-
plexer on N x N optical interconnection, J. Lightwave Technol., 14:1097–1105, 1996.
[2] R. Khosravani, M.I. Hayee, B. Hoanca, and A.E. Willner, Reduction of coherent crosstalk
in WDM add/drop multiplexing nodes by bit pattern misalignment, IEEE Photon. Technol.
Lett., 11:134–136, 1999.
[3] R. K. Staubli and P. Gysel, Crosstalk penalties due to coherent Rayleigh noise in bidi-
rectional optical communication systems, J. Lightwave Technol., 9:375–380, 1991.
[4] S. Radic and S. Chandrasekhar, Limitations in dense bidirectional transmission in absence
of optical amplification, IEEE Photon. Technol. Lett., 14:95–97, 2002.
[5] J.L. Gimlett, M.Z. Iqbal, L. Curtis, N.K. Cheung, A. Righetti, F. Fontana, and G. Grasso,
Impact of multiple reflection noise in Gbit/s lightwave systems with optical fibre ampli-
fiers, Electron. Lett., 25:1393–1394, 1989.
[6] J.C. van der Plaats and F.W. Willems, RIN increase caused by amplified-signal redi-
rected Rayleigh scattering in erbium-doped fibers. In Proceedings of the Optical Fiber
Communications Conference, WM6, 1994.
[7] M. Nissov, K. Rottwitt, H.D. Kidorf, and M.X. Ma, Rayleigh crosstalk in long cascades
of distributed unsaturated Raman amplifiers, Electron. Lett., 35:997–998, 1999.
[8] J.L. Gimlett, M.Z. Iqbal, N.K. Cheung,A. Righetti, F. Fontana, and G. Grasso, Observation
of equivalent Rayleigh scattering mirrors in lightwave systems with optical amplifiers,
IEEE Photon. Technol. Lett., 2:211–213, 1990.
[9] S. Ramachandran, B. Mikkelsen, L.C. Cowsar, M.F. Yan, G. Raybon, L. Bovin,
M. Fishteyn, W.A. Reed, P. Wisk, D. Brownlow, R.G. Huff, and L. Gruner-Nielsen, All-
fiber grating based higher order mode dispersion compensator for broad-band compen-
sation and 1000-km transmission at 40 Gb/s, IEEE Photon. Technol. Lett., 13:632–644,
2001.
[10] J.L. Gimlett, J. Young, R.E. Spicer, and N.K. Cheung, Degradations in Gbit/s DFB laser
transmission systems due to phase-to-intensity noise conversion by multiple reflection
points, Electron. Lett., 24:406–407, 1988.
[11] A.F. Judy, Intensity noise from fiber Rayleigh backscatter and mechanical splices. In
Proceedings of the Fifteenth European Conference on Optical Communication, 486–489,
1989.
[12] D.A. Fishman, D.G. Duff, and J.A. Nagel, Measurements and simulation of multipath
interference for 1.7-Gb/s lightwave transmission systems using single and multifrequency
lasers, IEEE J. Lightwave Technol., 8:894–905, 1990.
[13] E.L. Goldstein and L. Eskildsen, Scaling limitations in transparent optical networks due
to low-level crosstalk, IEEE Photon. Technol. Lett., 7:93–94, 1995.
[14] S. Burtsev, W. Pelouch, and P. Gavrilovic, Multi-path interference noise in multi-span
transmission links using lumped Raman amplifiers. In Proceedings of the Optical Fiber
Communications Conference, TuR4, 2002.
[15] L. Eskildsen and E. Goldstein, High-performance amplified optical links without isolators
or bandpass filters, IEEE Photon. Technol. Lett., 4:55–58, 1992.
[16] S. Wu, A. Yariv, H. Blauveit, and N. Kwong, Theoretical and experimental investigation
of conversion of phase noise to intensity noise by Rayleigh scattering in optical fibers,
Appl. Phys. Lett., 59:1156–1158, 1991.
[17] P. Wan and J. Conradi, Double Rayleigh backscattering in long-haul transmission systems
employing distributed and lumped fibre amplifiers, Electron. Lett., 31:383–384, 1995.
[18] K. Petermann and E. Weidel, Semiconductor laser noise in an interferometer system,
IEEE J. Quantum Electron., QE-17:1251–1256, 1981.
 15. Multiple Path Interference 565

[19] J.A. Armstrong, Theory of interferometric analysis of laser phase noise, J. Opt. Soc.
Amer., 56:1024–1031, 1966.
[20] M.M. Choy, J.L. Gimlett, R. Welter, L.G. Kazovsky, and N.K. Cheung, Interferometric
conversion of laser phase noise to intensity noise by single-mode fibre-optic components,
Electron. Lett., 23:1151–1152, 1987.
[21] J. W. Goodman, Statistical Optics, New York: Wiley, 2000.
[22] A.J. Stentz, T. Nielsen, S.G. Grubb, T.A. Strasser, and J.R. Pedrazzani, Raman ring
amplifier at 1.3 µm with analog-grade noise performance and an output power of 23
dBm. In Proceedings of the Optical Fiber Communications Conference, PD16, 1996.
[23] A. Bononi and M. Papararo, Optimal placement of isolators in Raman amplified optical
links. In Optical Amplifiers and their Applications, OTuA2, 2001.
[24] M. Born and E. Wolf, Principles of Optics, Oxford, UK: Pergamon, 1980.
[25] M. Nazarathy, W.V. Sorin, D.M. Baney, and S. A. Newton, Spectral analysis of optical
mixing measurements, J. Lightwave Technol., 7:1083–1096, 1989.
[26] D. Derickson, ed, Fiber Optic Test and Measurement, Upper Saddle River, NJ: Prentice-
Hall, 1998.
[27] R.W. Tkach and A.R. Chraplyvy, Phase noise and linewidth in an InGaAsP DFB laser,
J. Lightwave Technol., LT-4:1711–1716, 1986.
[28] J.L. Gimlett and N.K. Cheung, Effects of phase-to-intensity noise conversion by multiple
reflections on gigabit-per-second DFB laser transmission systems, J. Lightwave Technol.,
7:888–895, 1989.
[29] M. Tur and E.L. Goldstein, Probability distribution of phase-induced intensity noise
generated by distributed-feedback lasers, Optics Lett., 15:1–3, 1990.
[30] C.J. Rasmussen, F. Liu, R.J.S. Pedersen, and B.F. Jorgensen, Theoretical and experi-
mental studies of the influence of the number of crosstalk signals on the penalty caused
by incoherent optical crosstalk. In Proceedings of the Optical Fiber Communications
Conference, TuR5, 1999.
[31] G. Einarsson, Principles of Lightwave Communications, New York: Wiley, 1996.
[32] D. Marcuse, Derivation of analytical expressions for the bit-error probability in lightwave
systems with optical amplifiers, IEEE J. Lightwave Technol., 8:1816–1823, 1990.
[33] M.E. Lines, W.A. Reed, D.J. DiGiovanni, and J.R. Hamblin, Explanation of analomalous
loss in high delta singlemode fibres, Electron. Lett., 35:1009–1010, 1999.
[34] E. Brinkmeyer, Analysis of the backscattering method for single-mode optical fibers, J.
Opt. Soc. Amer., 70:1010–1012, 1980.
[35] A.H. Hartog and M.P. Gold, On the theory of backscattering in single-mode optical fibers,
J. Lightwave Technol., LT-2:76–82, 1984.
[36] S. L. Hansen, K. Dybdal, and C. C. Larsen, Gain limit in erbium-doped fiber amplifiers
due to internal Rayleigh backscattering, IEEE Photon. Technol. Lett., 4:559–561, 1992.
[37] Y. Qian, J.H. Povlsen, S.N. Knudsen, and L. Grüner-Nielsen, Analysis and characteriza-
tion of dispersion compensating fibers in fiber Raman amplification. In Proceedings of
Optical Amplifiers and Their Applications, OMB6, 2000.
[38] P. Gysel and R.K. Staubli, Statistical properties of Rayleigh backscattering in single-mode
fibers, J. Lightwave Technol., 8:561–567, 1990.
[39] P. Gysel and R.K. Staubli, Spectral properties of Rayleigh backscattered light from single-
mode fibers caused by a modulated probe signal, J. Lightwave Technol., 8:1792–1798,
1990.
[40] L. Mandel and E. Wolf, Optical Coherence and Quantum Optics, New York: Cambridge
University Press, 1995.
566 J. Bromage et al.

[41] S. Gray, M. Vasilyev, and K. Jepsen, Spectral broadening of double-Rayleigh backscatter-

ing in a distributed Raman amplifier, In Proceedings of the Optical Fiber Communication
Conference, MA2, 2001.
[42] A. Papoulis, Probability, Random Variables, and Stochastic Processes, 3d edition. Lon-
don: McGraw-Hill, 1991.
[43] C.R.S. Fludger and R.J. Mears, Electrical measurements of multipath interference in
distributed Raman amplifiers, J. Lightwave Technol., 19:536–545, 2001.
[44] M.O. van Deventer, Polarization properties of Rayleigh backscattering in single-mode
fibers, J. Lightwave Technol., 11:1895–1899, 1993.
[45] H. Kogelnik, R.M. Jopson, and L.E. Nelson, Polarization-Mode Dispersion. In Optical
Fiber Telecommunications, Volume IVB. New York: Elsevier Science Chap. 15, 725–861,
2002.
[46] G.J. Foschini and C.D. Poole, Statistical theory of polarization dispersion in single mode
fibers, IEEE J. Lightwave Technol., 9:1439–1456, 1991.
[47] E. Brinkmeyer, Forward-backward transmission in birefringent single-mode fibers: In-
terpretation of polarization-sensitive measurements, Optics Lett., 6:575–577, 1981.
[48] F.W. Willems, J.C. van der Plaats, and D.J. DiGiovanni, EDFA noise figure degradation
caused by amplified signal double Rayleigh scattering in erbium doped fibres, Electron.
Lett., 30:645–646, 1994.
[49] J. Bromage, C.H. Kim, R.M. Jopson, K. Rottwitt, and A.J. Stentz, Dependence of double-
Rayleigh backscatter noise in Raman amplifiers on gain and pump depletion. In Pro-
ceedings of Optical Amplifiers and their Applications, OTuA1, 2001.
[50] W.S. Wong, C.-J. Chen, H.K. Lee, G. Wilson, and M.-C. Ho, Multipath interference
(MPI) measurements using a laser with a variable linewidth, In Proceedings of Optical
Amplifiers and Their Applications, OME8, 2002.
[51] D.M. Baney and R.L. Jungerman, Optical noise standard for the electrical method of
optical amplifier noise figure measurement. In Proceedings of Optical Amplifiers and
Their Applications, MB3, 1997.
[52] S.A.E. Lewis, S.V. Chernikov, and J.R. Taylor, Characterization of double Rayleigh
scatter noise in Raman amplifiers, IEEE Photon. Technol. Lett., 12:528–530, 2000.
[53] V. Smokovdin, S.A.E. Lewis, and S.V. Chernikov, Direct comparison of electrical and
optical measurements of double Rayleigh scatter noise. In Proceedings of the European
Conference on Optical Communications, Symposium 3.0.5, 2002.
[54] C.-J. Chen and W.S. Wong, Transient effects in saturated Raman amplifiers, Electron.
Lett., 37:371–373, 2001.
[55] S.R. Chinn, Temporal observation and diagnostic use of double Rayleigh scattering in
distributed Raman amplifiers, IEEE Photon. Technol. Lett., 11:1632–1634, 1999.
[56] L. Gruner-Nielsen, Yujun Qian, B. Palsdottir, P.B. Gaarde, S. Dyrbol, T. Veng, Yifei Qian,
R. Boncek, and R. Lingle, Module for simultaneous C + L-band dispersion compen-
sation and Raman amplification. In Proceedings of the Optical Fiber Communications
Conference, TuJ6, 2002.
[57] N.A. Olsson, Lightwave systems with optical amplifiers, IEEE J. Lightwave Technol.,
7:1071–1082, 1989.
[58] E. Desurvire, Erbium-Doped Fiber Amplifiers. New York: Wiley, 1994.
[59] R. G. Smith, Optical power handling capacity of low loss optical fibers as determined by
stimulated Raman and Brillouin scattering, Appl. Opt., 11:2489–2494, 1972.
[60] V. Curri and G. Rizzo, Statistical properties and system impact of multi-path interference
in Raman amplifiers. In Proceedings of the European Conference on Optical Communi-
cations, 110–111, Tu.A.1.2, 2001.
 15. Multiple Path Interference 567

[61] P.B. Hansen, L. Eskildsen, A.J. Stentz, T.A. Strasser, J. Judkins, J.J. DeMarco, R. Pe-
drazzani, and D.J. DiGiovanni, Rayleigh scattering limitations in distributed Raman
pre-amplifiers, IEEE Photon. Technol. Lett., 10:159–161, 1998.
[62] P. Wan and J. Conradi, Power penalties due to Rayleigh backscattering in long-haul trans-
mission systems employing lumped and distributed EDFAs. In Proceedings of Optical
Amplifiers and Their Applications, ThC6, 1994.
[63] P. Wan and J. Conradi, Impact of double Rayleigh backscatter noise on digital and analog
fiber systems, J. Lightwave Technol., 14:288–297, 1996.
[64] L. Eskildsen and P.B. Hansen, Interferometric noise in lightwave systems with optical
amplifiers, IEEE Photon. Technol. Lett., 9:1538–1540, 1997.
[65] F. Liu, C.J. Rasmussen, and R.J.S. Pedersen, Experimental verification of a new model
describing the influence of incomplete signal extinction ratio on the sensitivity degrada-
tion due to multiple interferometric crosstalk, IEEE Photon. Technol. Lett., 11:137–139,
1999.
[66] T. Zami, L. Noirie, F. Bruyére, and A. Jourdan, Crosstalk-induced degradation in an
optical-noise-limited detection system. In Proceedings of the Optical Fiber Communi-
cations Conference, TuR4, 1999.
[67] H.K. Kim and S. Chandrasekhar, Dependence of in-band crosstalk penalty on the signal
quality in optical network systems, IEEE Photon. Technol. Lett., 12:1273–1274, 2000.
[68] M. H. Eiselt, L. D. Garrett, and V. Dominic, Optical SNR versus Q-factor improvement
with distributed Raman amplification in long amplifier chains. In Proceedings of the
European Conference on Optical Communications, 4.4.4, 2000.
[69] P. J. Winzer, R.-J. Essiambre, and J. Bromage, Combined impact of double-Rayleigh
backscatter and amplified spontaneous emission on receiver noise. In Proceedings of the
Optical Fiber Communications Conference, THGG87, 2002.
[70] L. Boivin and G. J. Pendock, Receiver sensitivity for optically amplified RZ signals
with arbitrary duty cycle. In Proceedings of Optical Amplifiers and Their Applications,
106–109, ThB4, 1999.
[71] P. J. Winzer and A. Kalmar, Sensitivity enhancement of optical receivers by impulsive
coding, J. Lightwave Technol., 17(2):171–177, 1999.
[72] J. L. Rebola and A. V. T. Cartaxo, Power penalty assessment in optically preampli-
fied receivers with arbitrary optical filtering and signal-dependent noise dominance, J.
Lightwave Technol., 20(3):401–408, 2002.
[73] P. J. Winzer, Optical transmitters, receivers, and noise. In John G. Proakis, ed., Wiley’s
Encyclopedia on Telecommunications, New York: Wiley, 2002.
[74] P. J. Winzer, Receiver noise modeling in the presence of optical amplification. In Pro-
ceedings of Optical Amplifiers and Their Applications, OTuE16, 2001.
[75] P.J. Winzer, S. Chandrasekhar, and H. Kim, Impact of filtering on DPSK reception, IEEE
Photon. Technol. Lett., 15(6):840–842, 2003.
[76] S. D. Personick, Receiver design for digital fiber optic communication systems, Bell
Syst. Tech. J., 52:843–874, 1973.
[77] P.J. Winzer and R.-J. Essiambre, Optical receiver design trade-offs. In Proceedings of
the Optical Fiber Communications Conference, ThG1, 2003.
[78] J. Bromage, C.-H. Kim, P. J. Winzer, L. E. Nelson, R.-J. Essiambre, and R. M. Jopson,
Relative impact of multiple-path interference and amplified spontaneous emission noise
on optical receiver performance. In Proceedings of the Optical Fiber Communication
Conference, TuR3, 2002.
[79] C. H. Kim, J. Bromage, and R. M. Jopson, Reflection-induced penalty in Raman amplified
systems, IEEE Photon. Technol. Lett., 14:573–575, 2002.
568 J. Bromage et al.

[80] B. E. A. Saleh and M. C. Teich, Fundamentals of Photonics. New York: Wiley, 1991.
[81] A. Yariv, H. Blauvelt, and S.-W. Wu, A reduction of interferometric phase-to-intensity
conversion noise in fiber links by large index phase modulation of the optical beam, J.
Lightwave Technol., 10:978–981, 1992.
[82] P. K. Pepeljugoski and K. Y. Lau, Interferometric noise reduction in fiber-optic links
by superposition of high frequency modulation, J. Lightwave Technol., 10(7):957–963,
1992.
[83] S.L. Woodward, T.E. Darcie, and F. Liu, A method for reducing multipath interference
noise, IEEE Photon. Technol. Lett., 6:450–452, 1994.
[84] A. Yariv, H. Blauvelt, D. Huff, and H. Zarem, An experimental and theoretical study
of the suppression of interferometric noise and distortion in AM optical links by phase
dither, J. Lightwave Technol., 15(3):437–443, 1997.
[85] P. V. Mamyshev, Rayleigh scattering impairments in systems with Raman amplification,
Bell Labs internal technical memorandum, 2000 (unpublished).
[86] G. P. Agrawal, Nonlinear Fiber Optics, 3d ed., San Diego: Academic, 2000.
[87] C. Paré, A. Villeneuve, P.-A. Bélanger, and N. J. Doran, Compensating for dispersion
and the nonlinear Kerr effect without phase conjugation, Optics Lett., 21:459–461, 1996.
[88] A. Altuncu, L. Noel, W. A. Pender, A. S. Siddiqui, T. Widdowson, A. D. Ellis, M. A.
Newhouse, A. J. Antos, G. Kar, and P. W. Chu, 40 Gbit/s error free transmission over a
68-km distributed erbium-doped fibre amplifier, Electron. Lett., 32:233–234, 1996.
[89] L. F. Mollenauer, R. H. Stolen, and M. N. Islam, Experimental demonstration of soliton
propagation in long fibers: Loss compensated by Raman gain, Optics Lett., 10:229–231,
1985.
[90] J.-I. Kani, M. Jinno, T. Sakamoto, S. Aisawa, M. Fukui, K. Hattori, and K. Oguchi,
Interwavelength-band nonlinear interactions and their suppression in multiwavelength-
band WDM transmission systems, J. Lightwave Technol., 17:2249–2260, 1999.
[91] Y. Aoki, Properties of fiber Raman amplifiers and their applicability to digital optical
communication systems, J. Lightwave Technol., 6:1225–1239, 1988.
[92] I. Kaminow and T. Li, Optical Fiber Telecommunications IV, San Diego: Academic,
Vol. A, Chap. 5, 2002.
[93] I. Kaminow and T. Li, Optical Fiber Telecommunications IV, San Diego: Academic,
Vol. B, Chap. 6, 232–304, 2002.
[94] R.-J. Essiambre, B. Mikkelsen, and G. Raybon, Intra-channel cross-phase modulation
and four-wave mixing in high-speed TDM systems, Electron. Lett., 35:1576–1578, 1999.
[95] I. Kaminow and T. Koch, Optical Fiber Telecommunication IIIA, chapter Solitons in High
Bit-Rate, Long Distance Transmission, San Diego: Academic Press, 373–460, 1997.
[96] René-Jean Essiambre and G. P. Agrawal, Soliton Communication Systems, Vol. XXXVII
of Progress in Optics, Amsterdam: Elsevier/North-Holland, Chap. Soliton Communica-
tion Systems, 185–256, 1997.
[97] M. O. Van Deventer, Polarization properties of Rayleigh backscattering in single-mode
fibers, J. Lightwave Technol., 11:1895–1899, 1993.
[98] P. C. Becker, N. A. Olsson, and J. R. Simpson, Erbium-Doped Fiber Amplifiers Funda-
mentals and Technology, San Diego: Academic Press, 1999.
[99] K. Shimizu, T. Horiguchi, and Y. Koyamada, Characteristics and reduction of coherent
fading noise in Rayleigh backscattering measurement for optical fibers and components,
J. Lightwave Technol., 10:982–987, 1992.
[100] C. R. S. Fludger, Y. Zhu, V. Handevek, and R. J. Mears, Impact of MPI and modulation
format on transmission systems employing distributed Raman amplification, Electron.
Lett., 37:970–971, 2001.
Chapter 16
Raman Impairments in WDM Systems

P. M. Krummrich

16.1. Introduction
In most chapters of this book, stimulated Raman scattering (SRS) is invoked intention-
ally. Pump radiation is coupled into the fiber carrying the signal radiation to generate
Raman gain. The Raman gain can be used very advantageously, for example, to im-
prove the optical signal-to-noise ratio (OSNR) budget by distributed amplification in
the transmission fiber. However, SRS also occurs unintentionally in WDM transmis-
sion systems. Due to the large number of channels inside the Raman gain bandwidth,
total power can add up to levels where considerable amounts of SRS are generated,
with the signal channels acting as pumps. In contrast to the beneficial effects of inten-
tional Raman pumping, the unintended generation of SRS usually degrades system
performance.
This chapter addresses effects resulting from the unintended invocation of SRS
and their impact on WDM signal transmission. Section 16.2 covers the generation of
spontaneous emission and its effect on maximum channel launch powers. A number
of system impairments result from the interaction between signal channels due to
SRS. Effects with time scales well below the bit period affect the mean values of
the individual channel powers. These slow interactions are discussed in Section 16.3,
whereas Section 16.4 addresses fast interactions between individual bits. Such inter-
actions change the variances of the respective channel powers and can be considered
as noise. The last section (16.5) provides some selection criteria for transmission
fibers with respect to Raman efficiency.

16.2. Raman Threshold

The Raman amplification process relies on stimulated emission of radiation. However,
stimulated emission is inevitably linked to the generation of spontaneous emission
[1, 2]. Consequently, launching signal radiation of WDM channels into an optical
fiber leads to generation of new photons due to spontaneous Raman scattering. The
peak intensity occurs at a frequency of one Stokes shift (approx. 13.2 Thz in silica
570 P. M. Krummrich

fibers, corresponding to a wavelength shift of 106 nm for a signal at 1550 nm) below
the frequency of the signal radiation.
The initial number of spontaneously generated photons is rather small and usually
negligible. It can grow considerably due to amplification by stimulated emission with
the signal radiation acting as a pump. The stimulated emission converts signal photons
into amplified spontaneous emission (ASE) photons, thus contributing an additional
loss mechanism for the signal radiation. The conversion rate is proportional to the
intensity of the ASE. Consequently, the additional loss of the signal radiation due to
SRS is negligible compared to other loss mechanisms in the fiber as long as the ASE
intensity is sufficiently small. If the signal launch power increases and ASE starts
to experience a very strong growth, the additional loss due to SRS will eventually
exceed the other loss mechanisms in the fiber. The depletion of the signal power due
to SRS grows so strongly that further increase of the launch power above the level of
equal loss contributions results in a decrease of the signal power at the fiber output.
This limitation of reasonable signal launch powers is called the Raman threshold [3].
There are several options to define the Raman threshold or the critical launch
power. One would be the minimum launch power resulting in an excess loss due to
SRS that is equal to other loss mechanisms at some point along the signal propagation.
Another option is the signal launch power that achieves the maximum signal power
at the fiber output. The most commonly used definition for the critical power seems
to be the launch power that results in a total ASE power equal to the signal power at
the fiber output [3, 4].
The critical signal launch power Pcrit in W according to the definition with equal
signal and ASE power at the fiber output can be estimated using the following equa-
tion [4].
Aeff αS
Pcrit = 8 , (16.1)
gR 1 − exp(−αS L)
where Aeff denotes the effective mode field area in m2 , gR the peak Raman coefficient
in m/W for random polarizations, αS the loss coefficient at the signal wavelength in
1/m, and L the length of the fiber in m. For a section of standard single-mode fiber
(SSMF) with a length of 100 km, an effective mode field area of 80 µm2 , a peak Ra-
man coefficient of 2.3 × 10−14 m/W, and a loss coefficient of 0.2 dB/km, this equation
predicts a critical launch power of 1.3 W for a single signal with a center wavelength
of 1550 nm.
According to Eq. (16.1), the critical launch power depends on the transmission
fiber type and its characteristics. Experiments were carried out to determine the critical
power for four different transmission fiber types [5]. Figure 16.1 shows the experi-
mental setup. Eight continuous wave (CW) laser diode sources with different output
wavelengths were employed to generate the signal radiation rather than a single one
in order to reduce the impact of stimulated Brillouin scattering (SBS). The signal
wavelengths were chosen on the ITU grid in the range from 1549.3 to 1560.6 nm
with 200 GHz channel spacing and output powers of 2 dBm per channel at the input
of the power combiner. SBS was further suppressed by applying a triangular ampli-
tude modulation with a frequency of 15 kHz and a modulation index around 10% to
 16. Raman Impairments in WDM Systems 571

Laser 1 Fiber
EDFA Coupler
10/90

Power Bandpass Power

Combiner Filter Meter 2
Laser 8
Optical
Power
Spectrum
Meter 1
Analyzer

Fig. 16.1. Experimental setup that was used to measure the Raman threshold of different
transmission fiber types.

Table 16.1. Characteristics of the Four Fibers Under Investigation

Fiber Length Measured Loss Coeff. Mode Field Area
Fiber Type km dB/km µm2
SSMF 41.9 0.21 78
PSCF 100.0 0.17 75
LS 50.4 0.22 53
TW 53.2 0.22 53

all laser injection currents. The modulation resulted in a shift of the SBS threshold of
each channel to launch powers above 27 dBm, corresponding to a total launch power
of 36 dBm into the fiber.
The different channels were coupled into a single fiber by an 8:1 power combiner
followed by a high-power EDFA with a maximum output power of 32 dBm. A tap
coupler with a splitting ratio of 10% allowed monitoring of the EDFA output power
as well as spectral components backscattered from the fiber input.
Table 16.1 summarizes the characteristics of the four transmission fiber samples
that have been investigated. Each sample represented a different fiber type: standard
single-mode fiber (SSMF), pure silica core fiber (PSCF), LS fiber, and TrueWave®
classic (TW) fiber. The lengths of the fibers under study were chosen to be at least
twice the effective Raman length avoiding artifacts arising from different interaction
lengths.
The output of the transmission fiber was followed by an optical bandpass filter
and an optical power meter. The purpose of the bandpass filter was to keep all spectral
components that did not belong to the eight signal channels from reaching the detector.
Figure 16.2 shows a plot of spectral components measured at the fiber input
that were propagating counterdirectionally to the signal channels. The narrow peaks
around 1550 nm correspond to a small fraction of the eight signal channels being
reflected by Rayleigh scattering and incompletely suppressed SBS. Most of the total
power is contained in the ASE generated by SRS. The peak of the ASE spectrum is
572 P. M. Krummrich

–5
–10
–15

Power in dBm
–20
–25
–30
–35
– 40
– 45
–50
1540 1560 1580 1600 1620 1640 1660 1680 1700
Wavelength in nm

Fig. 16.2. Spectral components measured at the fiber input propagating counterdirectionally to
the eight signal channels.

25
Fiber output power in dBm

20

SSMF
15
LS
True Wave
10
Pure silica

0
10 15 20 25 30 35
Fiber input power in dBm

Fig. 16.3. Total power of the eight signal channels at the fiber output versus total launch power
at the fiber input.

located approximately 110 nm above the center of the signal wavelengths. A similar
ASE spectrum can be observed at the output of the transmission fiber.
The main results of the investigation are depicted in Fig. 16.3. The plot shows the
total power of the signal channels at the fiber output versus launch power. For total
launch powers into the fiber below 25 dBm, the output power increases linearly with
input power. At a certain input power value depending on the fiber type, the output
power saturates and even decreases for increasing launch powers.
The deviation from linear loss could be attributed to the conversion of signal
radiation to ASE caused by SRS. Total ASE power at the fiber output was smaller
than the total power in the eight channels for total launch power levels below 25 dBm.
However, the total ASE power exceeded the remaining signal power at the fiber output
 16. Raman Impairments in WDM Systems 573

as soon as the signal output power started to drop with increasing input power. This is
a clear indication that the decrease of output power was caused by crossing the SRS
threshold.
Figure 16.3 allows us to determine the Raman thresholds of the fibers under
investigation. SSMF revealed a maximum signal output power for a total launch
power of about 30.2 dBm, however, lower threshold values were obtained for PSCF
(28.2 dBm) and the NZDSFs (about 26.6 dBm).
The relation of the Raman coefficients of the four transmission fiber types was
determined experimentally by measuring the SRS-induced tilt, a configuration similar
to the one described in [6]. As the method actually measures the slope of the Raman
coefficient spectrum in a wavelength range of approximately 30 nm, it is quite difficult
to determine the peak value from the experimental data very accurately. But very
similar shapes of the spectra of the different fiber types enable a reliable prediction of
the relation of the peak values. With the Raman coefficient of the SSMF as a reference,
the Raman coefficients of the PSCF, LS, and TrueWave® fiber are higher by a factor
of 1.13, 1.23, and 1.13, respectively.
Analyzing the relation of the Raman coefficients in combination with the fiber pa-
rameters given in Table 16.1 enables an interpretation of the results shown in Fig. 16.3.
The lower Raman thresholds of the LS and TrueWave® fibers compared to SSMF can
be explained by smaller effective mode areas and higher Raman coefficients, whereas
the lower Raman threshold of PSCF results from a higher Raman coefficient and a
smaller loss coefficient at the signal wavelengths.
Although the use of SSMF seems to be the best choice for applications requiring
maximum launch power, a more detailed discussion is necessary in terms of span
length within reach. It can be shown that bigger span lengths are achievable with
PSCF instead of SSMF. This fact is due to a trade-off between the different usable
launch power levels and the different loss coefficients.
Maximum launch power due to the Raman threshold is a critical parameter for
repeaterless transmission systems. The main goal of these systems is to achieve max-
imum transmission length with a single span, avoiding the use of inline repeaters.
Remotely pumped erbium-doped fiber amplifiers are a powerful technology to ex-
tend the system reach. Pump radiation at 1480 nm has to be carried from the receive
terminal to these amplifiers either through the transmission fiber or a separate fiber.
Launch power levels of the pump radiation up to 8 W could be achieved by using
very large effective mode field area fibers [7].

16.3. Tilt of the Channel Power Distribution

The previous section addressed energy transfer due to SRS from signal channels to
ASE. SRS can also cause considerable energy transfer between signal channels, if the
WDM signal channels are distributed over a wide wavelength range and total launch
power is high. Figure 16.4 tries to illustrate the concept. Many channels are launched
into the transmission fiber with equal power at the fiber input. During propagation
in the fiber, channels on the short-wavelength side act as Raman pumps for channels
574 P. M. Krummrich

gR

λ
P P
Transmission
Fiber

λ λ

WDM signal WDM signal

Fig. 16.4. Energy transfer between WDM signal channels caused by SRS in the transmission
fiber.

on the long-wavelength side. The resulting energy transfer leads to amplification of

the channels on the long-wavelength side whereas channels on the short-wavelength
side experience additional loss. As a consequence, the channel power distribution at
the fiber output looks tilted.

16.3.1. Channel Power Management

An expression for the amount of tilt at the output of a fiber section can be derived
analytically under these assumptions [8]:
• the spectrum of the Raman gain coefficient is approximated by a triangular profile
(wavelength difference of band edge channels <110 nm);
• the energy that is lost when a short-wavelength photon is converted into a long-
wavelength photon is neglected (minor approximation for a maximum spacing
<30 nm between the shortest and longest wavelength channels);
• all channels experience the same attenuation due to other loss mechanisms; and
• no energy transfer to ASE.
The difference of the power levels of the longest and shortest wavelength channels
P in dB can be calculated using the expression:
10 gR 1 fch
P = [1 − exp(−αS L)] P0 , (16.2)
ln(10) Aeff αS fR

where gR denotes the peak Raman coefficient in m/W for random polarizations, Aeff
the effective mode field area in m2 , αS the loss coefficient in 1/m, L the length
of the fiber in m, fch the difference of the center frequencies of the shortest and
longest wavelength channels in THz, fR the frequency difference between the signal
channel acting as a pump and the peak of the Raman gain spectrum, and P0 the total
launch power in W (sum of the time-averaged power of the channels).
For a fiber section with a length of 100 km, a maximum Raman coefficient of
2.3 × 10−14 m/W, an effective mode field area of 80 µm2 , a loss coefficient of
 16. Raman Impairments in WDM Systems 575

0.2 dB/km, and a peak Raman gain at a spacing of 13.2 Thz below the pump cen-
ter frequency, Eq. (16.2) predicts a tilt of 0.8 dB for 40 channels with a spacing of
100 GHz and a total launch power of 20 dBm.
It is interesting to note that for a given fiber, the amount of tilt depends on the total
launch power and the spacing between the shortest and longest wavelength channels
only. The tilt does not depend on the distribution of power among channels at the
fiber input, the number of channels, or the location of channels inside the wavelength
band. The detailed distribution of channel powers at the fiber output does depend on
these boundary conditions, of course.
According to Eq. (16.2), SRS introduces a linear tilt on a decibel/nanometer scale.
It can be corrected by a single filter with a log-linear loss spectrum. The necessary
slope of the filter depends on the fiber type, the wavelength separation between the
shortest and longest wavelength channels, and the total launch power only.
Experimental investigations of the tilt induced by SRS for bidirectional and uni-
directional propagation were presented at OFC 1999 [9, 10]. Current state-of-the-art
systems (2002) feature unidirectional transmission of 80 channels in the C-band and
the same number of channels in the L-band. With a channel spacing of 50 GHz, the
separation between the shortest and the longest wavelength channels reaches 75 nm.
Figure 16.5 shows the comparison of measured loss spectra (difference between
channel power levels at the fiber input and output) of a single span of SSMF with a
length of 80 km for different launch powers. The 80 channels in the C-band (1528.7
to 1563.9 nm) and the 80 channels in the L-band (1570.4 to 1607.5 nm) had a total
power of 23.4 and 22.4 dBm, respectively. One spectrum was measured with a fixed
attenuation of 20 dB in front of the fiber input, the other without any attenuation (0 dB).
The loss spectra with attenuator correspond to the linear loss spectrum of the fiber,
whereas the difference between the spectra with and without attenuator was caused by
the amplification and attenuation of channels due to SRS. This difference is plotted

22

21
20 dB att. in front of fiber input
20

19
Loss in dB

18

17

16

15
0 dB att.
14
1520 1540 1560 1580 1600 1620
Wavelength in nm

Fig. 16.5. Comparison of loss spectra for different launch power levels.
576 P. M. Krummrich

5
4
3
2
Loss in dB 1
C-band
0
L-band
–1
-1
–2
-2
–3
-3
–4
-4
–5
-5
1520 1540 1560 1580 1600 1620
Wavelength in nm

Fig. 16.6. Difference of the fiber loss spectra with and without attenuator in front of the fiber.

in Fig. 16.6. The tilt of the channel power distribution features a slight deviation
from a log-linear shape and a total amount of approximately 6.1 dB. Equation (16.2)
predicts a tilt of 7.1 dB for the same configuration, which is quite good considering
the assumptions that were necessary for deriving the analytical expression.
The analytical expression is quite helpful in estimating the gain tilt of WDM
systems operating with one or two wavelength bands, corresponding to a wavelength
range of up to 80 nm. SRS tilt values for systems with wider wavelength ranges
or more bands should be calculated numerically by solving the set of differential
equations for all channel powers.
Figure 16.7 shows a plot of a typical fiber loss spectrum together with potential
wavelength bands for WDM transmission. The definition of the wavelength bands and
the order of deployment depends mainly on the availability of optical amplifier tech-
nologies. Erbium-doped fiber amplifiers (EDFA) in combination with gain-flattening
filter technology have enabled transmission in the C-band (conventional band of ED-
FAs). A second wavelength band called the L-band could be opened by shifting the
gain spectrum of EDFAs to longer wavelengths (long-wavelength band of EDFAs).
Thulium-doped fiber amplifiers (TDFA) are a good candidate for lumped ampli-
fication in the S-Band. It would be convenient to deploy the LS-band first (long-
wavelength part of the S-band) because it is closer to the C-band and features
lower fiber attenuation. Gain-shifted TDFAs are necessary to provide amplification
in the LS-band. The CS-band may be deployed as a next step by using conventional
(nongain-shifted) TDFAs. Additional transmission capacity could be provided by
transmitting channels in the XL band (extended L-band of EDFAs). The search for a
good amplifier technology for the XL band is still going on.
SRS tilt values in a single span of SSMF were calculated numerically for different
combinations of wavelength bands. Transmission of 80 channels per band with a
launch power of +3 dBm per channel was assumed. The results are depicted in
 16. Raman Impairments in WDM Systems 577

0,6

0,5

0,5

Loss coeff. in dB/km 0,4

0,4
CS LS C L XL
0,3

0,3

0,2

0,2
1300 1350 1400 1450 1500 1550 1600 1650 1700
Wavelength in nm

Fig. 16.7. Wavelength bands for WDM transmission.

18
CS + LS + C + L + XL
16
14 LS + C + L
12 CS + LS + C + L
SRS tilt in dB

10
8
6 C+L

4
C
2
0
0 50 100 150 200 250
Bandwidth in nm

Fig. 16.8. SRS tilt values for combinations of different wavelength bands.

Fig. 16.8. According to the numerical results, the SRS tilt grows nearly exponentially
until the width of the wavelength band used for transmission reaches the width of the
Raman gain spectrum. Further growth of the wavelength band still increases the SRS
tilt, but with a less steep slope.
The numerical investigation has shown that transmission of WDM channels in a
wide wavelength range results in considerable tilt of the channel power distribution
due to SRS. Tilt values up to several dB can be expected even for a single span. As
the tilt accumulates from span to span [11], growth of channel power differences has
to be limited in multispan systems by compensation of the SRS-induced tilt.
578 P. M. Krummrich

A straightforward way to compensate accumulation of the SRS tilt from span

to span is to design EDFAs with an inversely tilted gain spectrum. Unfortunately,
the amount of SRS tilt changes depending on the transmission fiber type, the launch
power per channel, and the number of active channels. A dynamic adjustment of the
tilt compensation may be necessary. Such an adjustment may be performed by chang-
ing the dynamic gain tilt of the EDFAs using a variable attenuator (VOA) between
amplifier stages or a special tunable tilt filter. Another option is to use pump signals
to compensate the SRS tilt in the transmission fiber [12]. If the system uses wideband
distributed Raman amplifiers with multiple pump wavelengths, compensation of the
tilt may be possible by adjusting the ratio of the Raman pump powers.

16.3.2. OSNR Reduction

The energy transfer induced by SRS results in an effectively reduced loss compared
to the linear case. The output power of the channel with the longest wavelength can
be several dB higher if all the other channels are activated compared to the case when
it operates alone. However, the increased output power should be interpreted as a
result of optical amplification rather than as a result of reduced loss. The fundamental
difference is noise. Reducing the fiber attenuation increases the output power of a
given channel without any additional noise. It has been shown experimentally that
SRS-induced tilt contributes additional noise [13].
In the case of unidirectional transmission, the additional noise contribution can
usually be neglected. If all channels propagate in the same direction, optical ampli-
fication and generation of noise occur in a part of the fiber where the signal power
levels are relatively high. Generation of additional noise should be considered in the
case of bidirectional transmission in a single fiber. As some channels experience op-
tical amplifiction in a part of the fiber close to the output where signal power levels
are relatively low, considerable impact on the OSNR budget of the system can be
expected [13].
The following example for a bidirectional transmission system has been chosen
to illustrate the concept. The system has a total capacity of 160 channels. Half the
channels are transmitted in the C-band and propagate in one direction. The other 80
channels propagating in the opposite direction are transmitted in the L-band.
In such a configuration, the SRS-induced tilt can be interpreted as distributed
amplification of the L-band channels with the C-band channels acting as counterdi-
rectional Raman pumps. Figure 16.9 shows a plot of the gain spectrum in the L-band
for a section of SSMF with a length of 80 km. The gain spectrum was calculated using
a numerical simulation tool based on integration of a set of differential equations for
all spectral components. The channels in the C- and L-bands were launched with a
total power of 23.4 and 22.4 dBm, respectively. The tool predicts a Raman gain of
nearly 3 dB for the L-band channel with the longest wavelength.
A convenient tool to analyze the noise contribution associated with distributed
Raman gain is provided by the effective noise figure. It can be defined as
1 2SASE
N Feff = (1 + ), (16.3)
Gon/off hν
 16. Raman Impairments in WDM Systems 579

3.5

Gain in dB 2.5

1.5

0.5

0
1570 1575 1580 1585 1590 1595 1600 1605 1610
Wavelength in nm

Fig. 16.9. Raman gain of the L-band channels pumped by the counterdirectionally propagating
C-band channels.

where Gon/off denotes the on/off Raman gain (ratio of the channel power at the fiber
output with and without Raman pumping), SASE the spectral power density of the
ASE in one polarization in W/Hz, h Planck’s constant, and ν the center frequency of
the channel in Hz.
A reduction of the span loss by reducing the fiber attenuation does not contribute
additional ASE. In this case, the effective noise figure in dB is equal to (−1) times
the decrease of the fiber loss in dB. For example, reducing the span loss by 3 dB
corresponds to an effective noise figure of −3 dB.
Figure 16.10 shows a plot of the effective noise figure spectrum calculated for the
L-band channels. The numerical tool predicts an effective noise figure of −0.6 dB for
the channel with the longest wavelength. This number has to be compared with an
effective noise figure of −3 dB which corresponds to an on/off Raman gain of 3 dB
without additional ASE. The ASE generated during the amplification of the channel
has increased the effective noise figure by 2.4 dB.
The impact of additional ASE generated by SRS on the OSNR budget of the
system depends on the noise performance of the amplifier following the fiber section.
In most cases, the lumped amplification will be provided by EDFAs. Assuming a
noise figure of the lumped amplifier of 4 dB leads to a total noise figure of 2.1 dB for
an on/off Raman gain of 3 dB and an effective noise figure of the distributed Raman
amplifier of −0.6 dB. Amplification without additional ASE or reduction of the fiber
loss by 3 dB results in a total noise figure of 1 dB. In this example, the additional
580 P. M. Krummrich

0.8

0.6

Effective noise figure in dB 0.4

0.2

–0.2

–0.4

–0.6

–0.8
1570 1575 1580 1585 1590 1595 1600 1605 1610
Wavelength in nm

Fig. 16.10. Effective noise figure spectrum of the distributed Raman amplification.

ASE generated in combination with the SRS-induced tilt has deteriorated the OSNR
budget of the system by 1.1 dB.
In most cases, it will not be a problem to tolerate the additional ASE generated
by the SRS-induced tilt. The channels that are affected most are the channels on the
long-wavelength side. The net effect in this wavelength range is still positive; that is,
the OSNR improvement due to the increased power at the fiber output overcompen-
sates the OSNR degradation due to additional ASE. Furthermore, broadband EDFAs
generally feature a better noise performance on the long-wavelength side of their
gain spectrum. As a consequence, the channels on the long-wavelength side of the
transmission band will provide the best noise performance assuming equal launch
power of all channels.
However, the additional ASE should be taken into account if the OSNR bud-
get of bidirectional transmission systems has to be calculated very accurately. It has
been mentioned in the previous section that the SRS-induced tilt has to be com-
pensated to keep channel power differences below acceptable limits in wideband,
multispan transmission systems. Without additional ASE, equal noise performance
of all channels could be achieved by selecting launch power levels resulting in equal
power levels of all channels at the fiber output. This approach relies on the addi-
tional assumption that the lumped amplifiers following the fiber sections provide a
wavelength-independent noise figure spectrum. If the system OSNR budget has to
take the wavelength-dependence of the noise figure spectrum of lumped amplifiers
into account, it should also include the additional ASE generated by the SRS tilt.
 16. Raman Impairments in WDM Systems 581

4x

TX SW

TX

CG TX

TX
OZ PD

Fig. 16.11. Sketch of the experimental setup consisting of a bidirectional DWDM link and
means to analyze channel add-drop scenarios: a clock generator (CG) controls an optical
switch (SW) and triggers an oscilloscope (OZ) recording the signal of a photo diode (PD).

16.3.3. Channel Power Transients

The tilt of the channel power distribution due to SRS depends on the total launch
power. Adding or dropping channels changes the total launch power. The resulting
change of the SRS tilt introduces power transients for the surviving channels. The
impact of such transients on a system featuring bidirectional transmission in a single
fiber have been investigated experimentally [14].
Figure 16.11 shows a sketch of the experimental setup. It consists of a bidirectional
DWDM link featuring transmission of sixteen 10 Gbps channels per direction over
five spans of standard single-mode fiber. The channels were modulated with a 231 − 1
pseudorandom bit sequence. Each span had a length of 90 km and a loss of 29 dB with
contributions from the fiber and a midspan attenuator. The total power launched into
the fiber ends was 20 dBm. The channels were grouped into two counterdirectionally
propagating wavelength bands. Each band was amplified by separate EDFAs. The
channels of the blue band covered the wavelength range from 1530.3 to 1542.9 nm
and the channels of the red band the wavelength range from 1547.7 to 1560.6 nm
with a spacing of 100 GHz between channels.
An optomechanical switch (SW) was inserted between the multiplexer output and
the input of the booster. It enabled us to activate or deactivate all channels propagating
in one direction simultaneously. The switch was driven by a clock generator (CG).
The power of a given channel propagating in the counterdirection was detected by
a PIN photo diode (PD) at the system output. The electrical output signal of the PD
was recorded by an oscilloscope (OZ), which was triggered by the clock generator.
The optomechanical switch had a transition time of approximately 0.4 ms. Fig-
ure 16.12 shows the time-resolved output signal of the switch with a CW input signal
during a switching event. In a first series of experiments, the switch was used to de-
activate all channels of the blue band. Figure 16.13 shows the time-resolved output
signal of the channel with the longest wavelength in the red band (1560.6 nm) at the
system output.
582 P. M. Krummrich

Power in mW
3

0
0 2 4 6 8 10
Time in ms

Fig. 16.12. Time-resolved output signal of the optomechanical switch.

0.25

Span No. 1 2 3 4 5

0.2
Power in mW

0.15

0.1

0.05

0
0 2 4 6 8 10 12
Time in ms

Fig. 16.13. Time evolution of the power of the channel with the longest wavelength in the red
band (1560.6 nm) at the system output after activation of the switch.
 16. Raman Impairments in WDM Systems 583

–19

–20

Power in dBm
blue on
–21
blue off

–22

–23
1545 1550 1555 1560 1565
Wavelength in nm

Fig. 16.14. Comparison of power levels of the channels in the red band at the output of the first
span with and without channels in the blue band.

The deactivation of the channels in the blue band resulted in a decrease of the
output power of the channels in the red band in five steps. This decrease cannot be
caused by the erbium-doped optical amplifiers, because each counterdirectionally
propagating band is processed by separate amplifiers.
Further experimental investigation provided convincing evidence that the de-
crease of the channel powers was caused by a reduction of SRS-induced tilt in the
transmission fiber. Figure 16.14 shows a comparison of channel powers in the red
band at the output of the first span with and without presence of the blue band chan-
nels. The channel with the longest wavelength in the red band experiences a change of
the output power of approximately 0.7 dB caused by a change of the SRS-induced tilt.
Figure 16.15 shows a comparison of the power transients of three different chan-
nels in the red band after activation of the switch: the channel with the longest wave-
length (1560.6 nm), a channel close to the center of the band (1553.3 nm), and the
channel with the shortest wavelength (1547.7 nm). The amplitude of the power tran-
sients exhibits the same wavelength dependence as the SRS-induced tilt.
The structure in the time evolution of the power decrease for a given channel can
be explained by the accumulation of the effect in each span. The EDFAs in the system
were equipped with a fast gain control by pump power adjustment. A fast reduction
of the input power level results in a proportional reduction of the output power. This
means that the change of power levels in the red band generated in the first span
superimpose to changes in the second span and so on. At the system output, the total
power decrease accumulated to more than 4 dB in case of the red band channel with
the longest wavelength.
The power transients observed in the red band can be explained as follows. After
activation of the switch, a falling power slope of the blue band channels propagates
through the system with the speed of light (approx. 2 × 108 m/s in silica fibers).
The strongest energy transfer due to stimulated Raman scattering occurs at locations
584 P. M. Krummrich

0.8
Power, norm.

0.6

1547.7 nm
0.4

1553.3 nm
0.2
1560.6 nm

0
0 2 4 6 8 10
Time in ms

Fig. 16.15. Comparison of the power transients of three channels in the red band: the channel
with the longest wavelength (1560.6 nm), the shortest wavelength (1547.7 nm), and a channel
close to the center of the band (1553.3 nm).

along the fiber with the highest channel power levels in the blue band, that is, in the
first kilometers following the fiber input. Consequently, the strongest changes of the
red band power levels can be observed if the falling slope of the blue band enters
a new fiber span. The rate of decay reduces as the slope propagates along the fiber
span and reaches locations with lower channel powers in the blue band. The resulting
pulsations of the power decrease can be observed in Fig. 16.13.
A similar behavior can be observed for the power levels of the channels in the blue
band at the system output if the red band is switched off. However, in this case the
channel output power increases after activation of the switch. This can be understood
from the fact that the energy transfer from the blue band to the red band caused by
stimulated Raman scattering stops if the channels in the red band are switched off.
BER measurements for the channel with the longest wavelength in the red band
have revealed error bursts of several thousand bit errors in transmitted 10 Gbps signals
immediately after the activation of the switch. The channel was transmitted error free
(BER < 10−13 ) before and a short time after the switching event. The number of errors
clearly depends on the system configuration and the characteristics of the receiver.
Severe error burst can be expected for systems with many spans and considerable
SRS-induced tilt.
The stepwise time evolution of the power transients results from the counterdirec-
tional propagation of the switched channels and the surviving channels. A different
time evolution has to be expected for codirectionally propagating channels. In this
 16. Raman Impairments in WDM Systems 585

case, the power slope of the switched channel propagates in the same direction as the
induced transient of the surviving channel. As the response time of SRS is very short,
the rise or fall time of the transient will basically correspond to the rise or fall time
of the optical switch.
The accumulation of the contributions from the individual spans depends on
the characteristics of the optical amplifiers. The amplitude of the transient in dB at
the output of the system will be approximately the sum of the contributions from the
individual spans in dB, if EDFAs are gain controlled or if the response time of the
switching event is faster than the response time of the inversion (<1 ms). Compensa-
tion of the power transients by pump power adjustment is rather inefficient, because
it mainly alters the sum of all channel powers and does not provide the necessary tilt.
Better results can be expected from a tilt filter with a fast control loop.
An impression about the potential magnitude of SRS-induced transients in high-
capacity systems can be obtained by considering the C+L-band system mentioned
above. The unidirectional transmission of 80 channels in the C-band and the same
number of channels in the L-band results in a SRS-induced tilt of 6.1 dB in a single span
of SSMF. Deactivation of all channels in the L-band with a fall time of 1 ms increases
the power of the channel with the shortest wavelength in the C-band (1528.8 nm) at
the fiber output by 2.2 dB. After 10 spans, the amplitude of the transient will have
accumulated to approximately 22 dB. Such a huge increase of the channel power
within 1 ms will result in a severe error burst and can damage the receiver.
Fortunately, deactivation of all channels in a transmission band is an unlikely
event under normal operating conditions. It may occur in the case of a component
failure or if the service staff unplugs the wrong patchcord. Both events can result in
a fall time of the channel power below 1 ms, especially if the connector of the patch
cord is springloaded. The resulting transients of the surviving channels will have a
severe impact on system performance.
SRS-induced transients may also deteriorate the performance of transparent pho-
tonic networks. Dynamic adding and dropping of channels has been proposed to adapt
the network to a changing traffic pattern. In such networks, it is unlikely that many
channels will be added or dropped at the same time. However, even the activation or
deactivation of a few channels can result in considerable transients, if the channels
are distributed over a wide wavelength range and transmitted over many spans.

16.4. Pattern Dependent Crosstalk

The effects mentioned in the previous sections result from the impact of SRS on
the time-averaged channel power. SRS in silica fibers is a very fast effect with a
response time on the order of several 10 femtoseconds. It is much faster than the
bit length of currently installed transmission systems working with bit rates up to
10 Gbps, corresponding to a bit length of 100 ps. Consequently, SRS can cause
power fluctuations within the timescale of a single bit.
Figures 16.16 and 16.17 try to illustrate the concept. In Fig. 16.16, three WDM
channels with different wavelenths and bit patterns are launched into a section of
586 P. M. Krummrich

λ1

λ2
Transmission
Fiber
λ3

P λ1 < λ2 < λ3

Fig. 16.16. Pattern-dependent energy transfer between WDM channels induced by SRS.

Fig. 16.17. Comparison of probability densities of

p the power of a given channel at the fiber output with
with no and without SRS.
SRS SRS

Pout

transmission fiber. SRS results in an energy transfer from shorter to longer wavelength
channels. The amount of energy transfer in a given time slot depends on the number
of marks transmitted in the individual channels in this time slot. For example, the
first bit from the left side of the channel with the longest wavelength (λ3 ) does not
experience any energy transfer because the other channels are transmitting space bits
in the same time slot. The second bit gains some energy which is transferred from the
mark bits of the other channels due to SRS.
The pattern-dependence of the energy transfer due to SRS results in a very fast
modulation of the power of a given channel. As the bit patterns transmitted in the
channels of a WDM system are usually independent, the modulation can be interpreted
as a noise contribution increasing the variance of the power of a given channel.
Figure 16.17 depicts probability densities of the power level of a given channel at
the output of the fiber. Without SRS and with negligible impact of other nonlinear
effects and quantum noise, every mark bit arrives at the fiber output with the same
power. SRS results in a dependence of the output power on the bits transmitted in
the other channels. The shift of the mean of the probability distribution corresponds
to the SRS-induced tilt discussed in the previous section. This section focuses on the
broadening of the distribution.
The impact of pattern-dependent crosstalk induced by SRS (SRS-XT) on the
performance of WDM transmission systems has been analyzed by many authors
[15–34]. The initial work concentrated on a small number of channels: in many cases
 16. Raman Impairments in WDM Systems 587

only two. Furthermore, it did not differentiate between the shift of the time-averaged
channel power due to SRS and the additional variance. As the shift of the time-
averaged power can be compensated, the initial studies usually overestimated the
impact of SRS-XT on ultimate system capacity.
For deriving expressions describing the variance of the channel power, it is helpful
to start by looking at the channel with the shortest wavelength. As SRS transfers
energy from shorter to longer wavelengths, the shortest wavelength channel can only
experience depletion. The depletion D in dB can be expressed as the power P0 of a
given channel at the fiber output with the other channels switched off divided by the
output power of the same channel being transmitted together with the other channels:
P0
D = 10 log10 . (16.4)
P
The amount of depletion of the shortest wavelength channel depends on the bits
transmitted in the other channels. It reaches its maximum value if all other channels
transmit mark bits and is equal to 0 dB if all other channels transmit space bits. The
depletion can be interpreted as a random variable, because the bits transmitted in the
individual WDM channels are usually independent.
In the following discussion, we consider N equally spaced channels with a channel
spacing of f . The channel with the shortest wavelength has the index C = 1. The
channels are binary intensity modulated according to the nonreturn-to-zero (NRZ)
format. Space and mark bits have equal probabilty: p0 = p1 = 12 . The Raman gain
profile is approximated by a triangular shape. In the first step, the effect of pulse
walk-off or dispersion will be neglected. Using these assumptions, the mean µ of the
depletion of channel No. C can be calculated using the following expression.
(N − C)(N − C + 1) − (C − 1)C
µ= K. (16.5)
4
The factor K in dB is defined as
gR f 1
K = 10 log10 P [1 − exp(−αS L)], (16.6)
Aeff fR αS
where gR denotes the peak Raman coefficient in m/W for random polarizations, Aeff
the effective mode field area in m2 , f the channel spacing THz, fR the frequency
difference between the signal channel acting as a pump and the peak of the Raman
gain spectrum in THz, P the launch power per channel in W during a mark bit, αS
the loss coefficient in 1/m, and L the length of the fiber in m.
The mean of the depletion in dB corresponds to the difference of the channel
power level at the fiber output in dBm for single-channel operation and the time-
average output power level in dBm for transmission together with the other channels.
For equal launch power levels of all channels, the depletion of the channel with the
shortest wavelength (C = 1) equals half the SRS-induced tilt. Equation (16.5) can
actually be converted into Eq. (16.2) by replacing N P /2 by the total time-averaged
launch power P0 and (N − 1)f by the frequency separation of the shortest and the
longest wavelength channel fch .
Figure 16.18 shows a plot of the mean of the depletion divided by K of a given
channel C for a total number of N = 80 channels with equal launch powers. The
588 P. M. Krummrich

2000

1500

1000

500
Mean/K

–500

–1000

–1500

–2000
0 10 20 30 40 50 60 70 80
C

Fig. 16.18. Plot of the mean of the depletion divided by K of a given channel C for N = 80
channels.

channels at the band edges experience the strongest shift of the time- averaged channel
power, whereas the channels in the center of the band arrive at the fiber output with
the same time-averaged power as in the case of single-channel operation.
A different wavelength-dependence can be observed for the variance of the de-
pletion. The variance of the SRS-induced depletion of channel C for a total number
of N channels can be calculated using the following expression.
(N − C)(N − C + 1)(2N − 2C + 1) + C(C − 1)(2C − 1) 2
σ2 = K . (16.7)
24
Figure 16.19 shows a plot of the variance of the depletion devided by K 2 of a
given channel C for the same total number of channels (N = 80) as in the previous
example. The maximum variance occurs for the channels at the band edges. Note that
the variance does not drop to zero in the center of the band. This can be understood
from the fact that the output power of the channel in the center of the band does still
depend on the bits transmitted in the other channels. The output power reaches its
maximum if all channels with shorter wavelength transmit a mark bit and all channels
on the longer wavelength side transmit a space bit. Minimum output power can be
observed if the mark and the space bit switch sides.
The nonvanishing variance of the depletion in the center of the band has an
important impact on system performance. Pattern-dependent crosstalk introduced by
SRS degrades the performance of all channels, even if a filter is used to compensate for
the SRS-induced tilt. The strongest reduction of the signal quality will be observed for
the channels at the band edges. However, some impairments also have to be expected
in the center of the band.
 16. Raman Impairments in WDM Systems 589

× 104
5

4.5

3.5

Variance/K 2 3

2.5

1.5

0.5

0
0 10 20 30 40 50 60 70 80
Channel No.

Fig. 16.19. Plot of the variance of the depletion divided by K 2 of a given channel C for N = 80
channels.

The observation that the channels at the band edges experience stronger per-
formance degradations can be employed to differentiate between limiting nonlinear
effects. Other multichannel nonlinear effects such as cross-phase modulation (XPM)
and four-wave-mixing (FWM) grow stronger with the number of nearby interacting
channels. Consequently, XPM and FWM result in a stronger performance degradation
of channels in the center of the band. Stronger degradation of the channels at the band
edges is a clear indication for dominating pattern-dependent crosstalk introduced by
SRS.
In the previous discussion, the walk-off of bits in different channels due to dis-
persion was neglected. This assumption may be valid for a small number of channels
located around the zero dispersion wavelength of a fiber with a small dispersion
slope. In most fibers currently used for the transmission of WDM channels, disper-
sion will introduce a considerable walk-off between bits, especially if the channels
are distributed over a wide wavelength range.
The walk-off of the bits increases the number of bits interacting with a mark bit of
a given channel. The effect on the variance of the depletion is similar to an increase
of the number of channels. However, fiber attenuation decreases the impact of bit
interactions on the variance during propagation through the fiber.
The variance of the depletion of the channel with the longest wavelength for the
case of fast walk-off can be estimated using the following expression [32].

N (N − 1)αS T 2
σ2 = K , (16.8)
16|Dλ|
590 P. M. Krummrich

15

12
SPM
10
20 dB
No. of fiber spans

5
25 dB
4

XPM
2

30 dB SRS-XT FWM
span loss

1
0 2 4 6 8 10 12
Launch power per ch. in dBm

Fig. 16.20. Number of fiber spans as a function of launch power per channel for a WDM system
transmitting 160 × 10 Gbps in the C- and L-bands over SSMF.

where αS denotes the fiber attenuation in 1/km, T the bit length in ps, D the dispersion
coefficient in ps/(nm km), and l the channel spacing in nm.
Due to its statistical nature, pattern-dependent crosstalk introduced by SRS can
be treated as an additional noise contribution to the overall noise budget of a WDM
transmission system. The system reach is limited by quantum noise from the optical
amplifiers if launch powers per channel into the fiber sections are too low. The impact
of quantum noise decreases with increasing launch power. According to Eqs. (16.7)
and (16.8), the variance of the depletion increases with increasing launch power. As
a consequence, SRS-XT results in an upper limit of the launch power similar to other
nonlinear effects.
Figure 16.20 shows limitations imposed on the number of spans as a function of
launch power per channel. The plot was calculated for transmission of 160 channels in
the C- and L-bands with a channel spacing of 50 GHz over SSMF. The required OSNR
at the system output was set to 22 dB to enable error-free detection of 10 Gbps signals
without FEC. The dispersion-compensation scheme was optimized to minimize the
impact of XPM and FWM. Gain-flattening filters compensated for the SRS-induced
tilt and other wavelength-dependent effects.
The lines with the positive slope correspond to noise limitations for the span losses
noted below the lines. The positive slope reflects the fact that increasing the launch
power reduces the ASE noise at the system output. The lines with the negative slope
correspond to OSNR penalties of 1 dB contributed by various nonlinear effects. The
impact of these effects increases with increasing launch power.
 16. Raman Impairments in WDM Systems 591

It can be concluded from the plot that pattern-dependent crosstalk induced by SRS
is the dominating nonlinear effect. The high dispersion coefficient of SSMF helps to
reduce the impact of XPM and FWM, leaving SRS-XT as the limiting effect. A similar
situation can be observed for various other configurations featuring transmission of
many channels distributed over a wide wavelength range as well as for other fiber
types. Consequently, SRS-XT has to be considered as an important limitation for the
reach of state-of-the-art high-capacity WDM transmission systems.

16.5. Selecting Transmission Fibers

Many different transmission fiber types are currently commercially available. The
loss spectra of these fibers are quite similar. The main difference can be found in the
dispersion characteristics. The dispersion spectrum of a transmission fiber has a very
important impact on system performance. For example, a high dispersion coefficient
can help to suppress nonlinear effects. On the other hand, it generates a need for strong
dispersion compensation, which in turn may be difficult to implement, expensive, or
degrade the system performance.
The dispersion coefficient may be a very important parameter of transmission
fibers, but it is not the only one that has an impact on system performance. Another
parameter that has gained attention recently is the Raman efficiency. The Raman
efficiency is defined as the Raman gain coefficient divided by the effective mode field
area of the fiber. A high Raman efficiency can be achieved by designing a fiber with a
small mode field area. Such a fiber will also exhibit strong nonlinear effects due to the
Kerr effect, because the efficiency of these effects also grows inversely proportional
to the effective mode field area. Another way to increase the Raman efficiency is
to select a fiber material with a high Raman coefficient. Such a material does not
necessarily come with a strong Kerr effect.
The previous discussion has shown that it is possible to select the Raman efficiency
of transmission fibers within certain limits. But it did not answer the question of
whether it is more desirable for transmission fibers to feature high or low Raman
efficiencies. The answer seems to be clear, if the potential benefits of distributed
Raman amplification are taken into account. This technology can be deployed very
effectively to improve the OSNR budget of a system by virtually reducing the span
loss. Increasing the Raman efficiency reduces the amount of pump power required
for a given gain of the distributed amplifier. Thus higher Raman efficiency seems to
be beneficial.
However, increasing the Raman efficiency also increases the strength of sys-
tem impairments introduced by SRS. It has been shown in the previous section that
pattern-dependent crosstalk caused by SRS is the dominating limitation of system
reach for many high-capacity long-haul terrestrial system configurations. Means to
compensate the signal degradation caused by SRS-XT are currently not available. On
the other hand, it is quite easy to compensate smaller Raman gain of transmission
fibers with less Raman efficiency by applying more pump power. Pump sources with
sufficient output power to achieve the maximum useful gain even in fibers with a
592 P. M. Krummrich

small Raman efficiency are commercially available. As a consequence, it seems to

be more beneficial from a system design perspective to select transmission fibers
with a small Raman efficiency if system impairments introduced by SRS contribute
a potential limitation for the system reach.

References
[1] A. Einstein, Zur Quantentheorie der Strahlung, Physikalische Zeitschrift, 18:121–128,
(March), 1917.
[2] A. Yariv, Introduction to Quantum Electronics. 3d ed., New York: Holt, Rinehart and
Winston 129–132, 1985.
[3] R. G. Smith, Optical power handling capacity of low loss optical fibers as determined by
stimulated Raman and Brillouin scattering, Appl. Optics, 11:11 (Nov.), 2489–2494, 1972.
[4] G. Agrawal, Nonlinear Fiber Optics, 2d ed., New York: Academic, 319–322, 1995.
[5] P. M. Krummrich, R. Neuhauser, and G. Fischer, Experimental comparison of Raman
thresholds of different transmission fiber types. In Proceedings of the 26th European
Conference on Optical Communication (ECOC 2000), Sept. 3–7, Munich, Vol. 3, P1.5,
133–134, 2000.
[6] S. Bigo, S. Gauchard, A. Bertaina, and J.-P. Hamaide, Experimental investigation of
stimulated Raman scattering limitation on WDM transmission over various types of fiber
infrastructures, IEEE Photon. Technol. Lett., 11:6 (June), 671–673, 1999.
[7] T. Miyakawa, I. Morita, and N. Edagawa, 40 Gbit/s × 25 WDM unrepeatered trans-
mission over 362 km. In Proceedings of the Conference on Optical Amplifiers and their
Applications (OAA 2002), July 14–17, Vancouver, Canada, OTuA1, 1–3, 2002.
[8] M. Zirngibl, Analytical model of Raman gain effects in massive wavelength division
multiplexed transmission systems, Electron. Lett., 34:8, 789–790, 1998.
[9] P.M. Krummrich, E. Gottwald, A. Mayer, C.-J. Weiske, and G. Fischer, Influence of
stimulated Raman scattering on the channel power balance in bidirectional WDM trans-
mission. In Proceedings of the Conference on Optical Fiber Communication (OFC 1999),
San Diego, Feb. 21–26, Vol. 2, WJ6, 171–173, 1999.
[10] S. Bigo, S. Gauchard, A. Bertaina, and J. Hamaide, Investigation of stimulated Raman
scattering limitation on WDM transmission over various types of fiber infrastructures.
In Proceedings of the Conference on Optical Fiber Communication (OFC 1999), San
Diego, Feb. 21–26, Vol. 2, WJ7, 174–176, 1999.
[11] V. J. Mazurczuyk, G. Shaulov, and E. A. Golovchenko, Accumulation of gain tilt in
WDM amplified systems due to Raman crosstalk, IEEE Photon. Technol. Lett., 12:11,
1573–1575, 2000.
[12] M. Takeda, S. Kinoshita, Y. Sugaya, and T. Tanaka, ed. S. Kinoshita, J. C. Livas, and G.
van den Hoven, Active gain-tilt equalization by preferentially 1.43 mu m- or 1.48 mu
m-pumped Raman amplification. In OSA Trends in Optics and Photonics Series (TOPS
1999), Vol. 30, Conference on Optical Amplifiers and their Applications, Nara, Japan,
June 9–11, 101–105, 1999.
[13] P.M. Krummrich, A. Mayer, and G. Fischer, Noise penalty caused by stimulated Raman
scattering in bidirectional DWDM transmission. In Proceedings of the Conference on
Optical Amplifiers and their Applications (OAA 1999), Nara, Japan, ThB3, 102–105,
1999.
[14] P.M. Krummrich, E. Gottwald, A. Mayer, R. Neuhauser, and G. Fischer, Channel power
transients in photonic networks caused by stimulated Raman scattering. In Proceedings
of the Conference on Optical Amplifiers and their Applications (OAA 2000), Quebec, July
9–12, OTuC6, 143–145, 2000.
 16. Raman Impairments in WDM Systems 593

[15] M. Ikeda, Stimulated Raman amplification characteristics in long span single mode silica
fibers, Optics Commun. 39:148–152, 1981.
[16] A. Tomita, Cross talk caused by stimulated Raman scattering in single-mode wavelength-
division multiplexed systems, Optics Lett., 8:7 (July), 412–414, 1983.
[17] A. R. Chraplyvy and P. S. Henry, Performance degradation due to stimulated Raman
scattering in wavelength-division-multiplexed optical fibre systems, Electron. Lett. 19:16
(Aug.), 641–643, 1983.
[18] A. R. Chraplyvy, Optical power limits in multi-channel wavelength-division multiplexed
systems, Electron. Lett., 20:2 (Jan.), 58–59, 1984.
[19] D. Cotter and A. M. Hill, Stimulated Raman crosstalk in optical transmission: Effects of
group velocity dispersion, Electron. Lett., 20:4 (Feb.), 185–187, 1984.
[20] R. H. Stolen and A. M. Johnson, The effect of pulse walk-off on stimulated Raman
scattering in fibers, IEEE J. Quantum Electron., QE-22:11 (Nov.), 2154–2160, 1986.
[21] D. N. Christodoulides and R. I. Joseph, Theory of stimulated Raman scattering in optical
fibers in the pulse walk-off regime, IEEE J. Quantum Electron., QE-25:273–279, 1989.
[22] M. S. Kao, ON/OFF ratio degradation of high density WDM systems due to Raman
crosstalk, Electron. Lett., 26:14 (July), 1034–1035, 1990.
[23] R. Comuzzi, C. de Angelis, and G. Gianello, Improved analysis of the effects of stimulated
Raman scattering in a multi-channel WDM communication system, European Trans.
Telecommun. Related Technol., 3:3, 295–298, 1992.
[24] S. Tariq and J. C. Palais, A computer model of non-dispersion-limited stimulated Raman
scattering in optical fiber multiple-channel communications, IEEE J. Lightwave Technol.,
11:12 (Dec.), 1914–1924, 1993.
[25] X. Zhang, B. F. Jorgensen, F. Ebskamp, and R. J. Pedersen, Input power limits and maxi-
mum capacity in long-haul WDM lightwave systems due to stimulated Raman scattering,
Optics Commun. 107:5–6, 358–360, 1994.
[26] F. Forghieri, R. W. Tkach, and A. R. Chraplyvy, Effect of modulation statistics on Raman
crosstalk in WDM systems, IEEE Photon. Technol. Lett. 7:1, 101–103, 1995.
[27] G. P. Agrawal, Raman-induced crosstalk. In Nonlinear Fiber Optics, New York: Aca-
demic, 334–336, 1995.
[28] S. Tariq and J. C. Palais, Stimulated Raman scattering in fiber optic systems, Fiber Integ.
Optics 15:4 (April), 335–352, 1996.
[29] D. N. Christodoulides and R. B. Jander, Evolution of stimulated Raman crosstalk in
wavelength division multiplexed systems, IEEE Photon. Technol. Lett., 8:12 (Dec.),
1722–1724, 1996.
[30] J. Wang, X. Sun, and M. Zhang, Effect of group velocity dispersion on stimulated Raman
crosstalk in multichannel transmission systems, IEEE Photon. Technol. Lett., 10:4, 540–
542, 1998.
[31] L. Rapp, Impact of stimulated Raman scattering in WDM systems using different types
of fiber, Int. J. Electron. Commun., 52:5, 302–309, 1998.
[32] K.-P. Ho, Statistical properties of stimulated Raman crosstalk in WDM systems, IEEE J.
Lightwave Technol., 18:7, 915–921, 2000.
[33] S. Norimatsu and T. Yamamoto, Waveform distortion due to stimulated Raman scattering
in wide-band WDM transmission systems, IEEE J. Lightwave Technol., 19:2 (Feb.),
159–171, 2001.
[34] A. G. Grandpierre, D. N. Christodoulides, W. E. Schiesser, C. M. McIntosh, and J.
Toulouse, Stimulated Raman scattering crosstalk in massive WDM systems under the
action of group velocity dispersion, Optics Commun., 194:4–6, 319–323, 2001.
This page intentionally left blank
Chapter 17
Ultra-Long-Haul Submarine and Terrestrial
Applications

Howard Kidorf, Morten Nissov, and Dmitri Foursa

17.1. Introduction
Optical communication is the science of transmitting information over a distance
using light. The engineering difficulties vary greatly because the distances to be
traversed differ significantly. For some, the task may be the need to interconnect
electronic integrated circuits within a computer. For others, the distance to be covered
may be between computers in a building or between the buildings in a city. The
differences between these cases are the distances over which information must be
carried. These differences dictate the nature of the technologies required to implement
the connections.
Extreme situations are found in cases where communication is required between
distant cities within a land mass or between the various landmasses of the globe. These
communications links are often called ultra-long-haul (ULH) systems. As defined in
this chapter, ULH systems span from 1500 to over 12,000 km.
A very reasonable question is: why would anyone want ULH transmission sys-
tems? To cross oceans, the case is clear. The lack of intermediate land upon which to
build the equipment capable of optoelectronic regeneration creates a need for ULH
systems. The demand to transmit high-capacity data signals 6000 km across the At-
lantic and 9000 km across the Pacific Ocean drives this obvious need.
For terrestrial systems a good question is: why would anyone want ULH transmis-
sion systems when concatenating shorter systems can cover the same distances? For
decades, terrestrial digital transmission systems were built from city to city. Traffic
was reconfigured to redirect local and express traffic. Why might this network model
change? There are two reasons.
The demand patterns for traffic have changed. Twenty years ago, a small fraction of
total telecommunications traffic was data traffic. The vast majority was voice traffic.
It has been shown that demand for voice traffic falls off very fast as the distance
increases. A span distance of 600 km (as is common today) satisfies more than 60%
of the connections for voice traffic [1]. In other words, most people communicate and
do business close to where they are located. Their calling patterns reflect this.
The demand for the Internet has changed all of this. In the past few years, data
traffic (i.e., mostly the Internet) has surpassed voice traffic in the major economic
596 H. Kidorf et al.

regions of the world (Europe, North America, and SE Asia). Interestingly, Internet
traffic has a very different demand pattern than voice traffic. A span distance of 3000
km is required to satisfy 60% of the connections for Internet traffic [1]. Few people
care or are even aware where the information that they are seeking resides when they
follow a hyperlink. The result is that a higher ratio of traffic travels much farther from
the user than ever before.
The other reason for a changing network model is the technical possibility and
economic motivation for all-optical networking. With the commercial availability
of all-optical cross-connects [2] (based on microelectromechanical systems, MEMS,
technology among others), the opportunity for all-optical mesh networking has grown.
The goal of reducing the cost of large networks has driven the market to look for
technologies to allow transmission of a lightwave as far as possible so that as many
optical–electrical–optical (O-E-O) regenerators as possible can be eliminated. The
economic argument is that routing an optical signal with a simple mirror reflection
is much less expensive (and takes less space and power) than an O-E-O regenerator.
Further development of all-optical cross-connects, wavelength add-drop devices, and
dynamic gain equalization technologies all promise to enhance the appeal of ULH
transmission.

17.1.1. History of Ultra-Long-Haul

It is hard to imagine a time when rapid communications between distant points was
not possible. Yet, for 5000 years of mankind’s civilization, messages could not travel
much faster than wind-propelled ships or the gallop of a horse. Kings, queens, and
emperors, in addition to ordinary businessmen, frequently made history-changing
decisions in the absence of timely or correct information.
The development of the telegraph would change all of this. In 1861 the Western
Union Company completed the final leg of the first high-speed telegraph link between
the east and west coasts of North America. This led to the rapid growth of a web
of connectivity. Unfortunately, the web stopped at the shores of the continent. In
1866, after four previous unsuccessful attempts, entrepreneur Cyrus Field, numerous
eminent scientists, and venture capitalists succeeded in spanning the Atlantic Ocean
with the first telegraph cable [3]. This was the first introduction to a worldwide
communications network.
In 1884, the cutting edge in long distance telephony was a telephone link between
Boston and New York City [4]. However, without amplification, a transcontinental
connection would have to wait until the summer of 1914 when a New York to San
Francisco link could be established with the aid of Dr. DeForest’s triode vacuum
tube [4]. Voice connectivity between the continents would not be established until
overseas radiotelephony was introduced in 1927 [5]. Despite technical improvements,
high cost and unavoidable ionospheric disturbances kept usage low.
The world would have to wait until 1955 before the first of more than 20 trans-
Atlantic telephone cables were installed [6]. This impressive 4000 km transmis-
sion feat was accomplished with amplified, analogue, frequency-division-multiplexed
 17. Ultra-Long-Haul Applications 597

Local
Access

Fig. 17.1. National core network showing local access.

transmission over coaxial cables. Analogue ULH telephony cables would eventually
be capable of carrying over 4200 voice channels on each coaxial cable. A web of
ULH connectivity soon grew to connect the major cities of the world using these
technologies (along with digital microwave transmission).
From April 1977 1 through the mid1980s, three generations of optical fiber trans-
mission technology allowed terrestrial fiber-optic transmission to spread and to grad-
ually replace coaxial cables and digital microwave as the technology of choice for
ULH connectivity [7]. Beginning in the early 1990s erbium-doped fiber amplifiers
(EDFAs) allowed optically amplified transmission to greatly increase the capacity
of optical transmission systems and to cement the role of optical transmission in
inter-city transmission.
Throughout this evolution, terrestrial optical transmission systems used technolo-
gies that required signal regeneration at least every 600 to 1200 km (i.e., optical signals
that are converted to electronic signals and regenerated after every fiber span). This
regenerator spacing was selected partly because in Asia, Europe, and most of North
America, major cities are spaced by these distances. At each of these sites, traffic is re-
groomed (i.e., local traffic added and dropped, express traffic rerouted) with electronic
cross-connects. Therefore, electronic regeneration in metropolitan areas makes sense.
This is illustrated in Fig. 17.1, which shows an example of a national core network.
Only in submarine systems, where transoceanic transmission eliminates the need for
local traffic access and the size of regenerative equipment makes it unappealing, do
we see true ULH transmission systems deployed today.

1 General Telephone and Electronics sent the first commercial telephone traffic though fiber
optics in April 1977.
598 H. Kidorf et al.

There are many examples of ULH systems in service today. Almost all of these
systems are undersea. Dozens of optically amplified fiber-optic systems cross the At-
lantic and Pacific Oceans. These systems have ultimate capacities ranging from a few
Gb/s to more than 5 Tb/s on each cable. Their repeater (i.e., optical amplifier) spacing
ranges from 33 km to more than 50 km. The longest of these systems (the China–US
Cable Network) has two cables (to increase reliability by offering geographically
diverse routes) that connect mainland China directly to Oregon and California in the
United States. The longer of the two cables covers a distance of 12,460 km using only
optical amplification without any electronic regeneration. Figure 17.2 shows a map
of this network.

17.1.2. History of Raman Amplification in Ultra-Long-Haul

To date, there have been very few ULH commercial transmission systems that make
use of Raman amplification and none (at the time of this writing) that are ULH.
Therefore, the history of Raman in ULH is reported through work performed in the
laboratory.
The first major experiment that used Raman amplification was, interestingly, not
focused on Raman amplification, but on soliton transmission [8]. This work was the
foundation of the first ULH demonstration, which traversed approximately 4000 km
without regeneration [9]. This demonstration was made years before the EDFA was
invented and entirely used Raman amplification to overcome the 9.2 dB loss of the
41.7 km spans. This experiment foretold work to be performed more than a decade in
the future by demonstrating Raman amplification by pumping the transmission fiber
at about 1497 nm to achieve optical amplification at a wavelength of about 1600 nm.
This work also anticipated the need to avoid Brillouin backscattering of the intense
pump light by spreading out the spectrum of the pump.

Pusan Bandon
Chongming Chikura
Fangshan San Luis Obispo
Okinawa
Shantou
Guam

Fig. 17.2. The China–US cable network.

 17. Ultra-Long-Haul Applications 599

In 1997 and 1998, the availability of practical high-powered 1450 to 1500 nm

pumps spurred interest in experiments that demonstrated the value of Raman am-
plifiers in ULH transmission systems. The pump technology that became available
was the combination of a cladding-pumped fiber laser and a cascaded Raman res-
onator [11] (themselves an interesting application of Raman scattering) capable of
producing an output power in excess of 1.2 W at 1480 nm.
In the first demonstration ever of ULH WDM transmission using only Raman am-
plification to overcome the loss of the transmission fiber, experimenters successfully
demonstrated transmission over 7200 km (far enough to traverse the Atlantic Ocean
or North America) by measuring error-free (bit error rate, BER, < 10−9 ) performance
for ten 10 Gb/s channels [12]. Furthermore, the optical signal-to-noise ratio (OSNR)
of the transmission line was 1.5 dB better than equivalent EDFA chains with the same
path-averaged signal power. This demonstrated for the first time the previously the-
orized advantages of distributed amplification as having superior noise performance
to lumped amplification (as EDFAs are). It also demonstrated that some of the po-
tentially large transmission impairments found in distributed amplification systems,
such as multipath interference (MPI), could be managed effectively.
This experiment and many others are discussed in Section 17.4 (Applications of
Raman Gain).

17.2. Ultra-Long-Haul Transmission Issues

Ultra-long-haul systems have fundamental design differences from their shorter
length cousins. These differences are mostly (but not entirely) due to transmission
distance and the fact that optically amplified fiber transmission behaves somewhat
like analogue transmission where the impairments incurred in each span accumulate
over the entire length of the transmission. That is, digital regeneration occurs only at
the end of the transmission line that may be more than 10,000 km from the transmitter.

17.2.1. Differences Between Terrestrial and Undersea Systems

There are many differences between terrestrial and undersea ULH systems besides
the opportunity for large regenerator sites along terrestrial routes. These differences
have a great impact on the designs of these systems. They arise due to two primary
reasons: in undersea systems, access to the amplifiers is limited to that which can be
provided through the cable and all equipment in the transmission line must be able
to fit into a housing that is capable of withstanding the pressure found on the bottom
of the ocean.
All of the power needed for the inline amplifiers along an undersea transmission
line comes from a pair of large power supplies on the shore (although sometimes only
one is provided). This power is supplied through a power conductor in the cable to
the repeaters where it is distributed to the amplifiers. The ocean serves as the second
conductor between the two shores. Due to constraints on the dielectric material in the
cable, the maximum voltage used in these power supplies is typically 10 to 12 kV.
600 H. Kidorf et al.

The operating current is about one ampere supplying a cable that has a resistance of
0.7 to 1.6 "/km [13]. Over a 10,000 km system with repeater spacing of 50 km (200
repeaters), less than 100 W is available to power each repeater (which may contain
up to eight pairs of amplifiers with today’s technology). Whereas a terrestrial repeater
can consume a much larger amount of power (subject to availability and cooling
constraints), an undersea repeater has a significant limit on available power. The
impact on Raman amplified systems is that, because undersea systems are constrained
by available power, their system designers have reason to heavily favor use of the
most power-efficient optical amplification. In general, this is delivered by EDFAs.
Another limitation imposed by the oceans is the need to house all optical, electrical,
and mechanical equipment for powering, mechanical interconnection, amplification,
and supervision and diagnostics inside a pressure housing that can be deployed from
a ship, sink more than 7.5 km through the ocean, and rest under more than 730
atmospheres (74 MPa = 10, 700 lbs/in2 ) of pressure. Practical considerations call for
restrictions on the size of the pressure housing to 0.065 m3 (65 liters). See Figure
17.3 for an example. For comparison, the volume of a typical belowground terrestrial
repeater hut is 16 to 100 m3 .
Unlike undersea systems, terrestrial systems have limitations caused by long re-
peater spacing and fiber selection. The first restriction arises from the fact that most
terrestrial transmission systems are installed into already established rights-of-way.
Therefore the repeater spacing is already predetermined.Also, the fiber type is selected
once and deployed (often years in advance of the transmission system deployment) in
a cable frequently with hundreds of fibers. The optical system designer must contend
with the fibers that are already deployed.
Freedom to select the optimum fiber at the time of the system design is a great
advantage in submarine systems. All-optical transmission systems require a match

Fig. 17.3. View of repeater housing for undersea application. The housing shown contains up
to eight pairs of optical amplifiers.
 17. Ultra-Long-Haul Applications 601

between the optical source and the transmission fiber. In high-capacity ULH systems
that make use of optical amplifiers, this match is even more critical. Ever since the
first ULH optically amplified systems, proper selection of the chromatic dispersion
characteristics of the fiber has been critical. The advent of WDM systems brought
focus on the inherently nonlinear characteristics of the transmission fibers and the
desire to lower their impact through expanding the effective area [14]. Matching the
desire for spans with large effective area and low dispersion slope with the chirped
return-to-zero (CRZ) modulation format results in a hybrid span design (i.e., two types
of fiber in one span) that is implemented primarily in undersea transmission systems.
Perhaps the greatest difference between terrestrial and submarine systems is their
different approaches to reliability. Submarine systems are designed to be one of the
most reliable systems on earth. Not only are the repeaters usually beneath 5 to 8
km of water, but they are also often in places where it takes up to a week to get a
suitable ship into position for a repair. Unlike terrestrial equipment that can depend
on service personnel being able to access and service the equipment (often within
hours of a reported failure), submarine systems are created with a design target to
suffer less than three failures in their lifetime of 25 years. This level of reliability
requires extensive qualification testing of all components and subsystems that will
be placed undersea. The requirement for reliability further underscores the need for
simple, low-power undersea components.
Table 17.1 summarizes the design issues found in terrestrial and submarine ULH
systems.

Table 17.1. Summary of ULH Design Issues

Terrestrial Submarine
Space for inline amplifiers >1.0 m3 <0.1 m3
Powering Less important Limited due to remote supply
Transmission fiber Legacy Custom tailored
Reliability (assemblies) ∼1000 FITs #100 FITs

The challenge of introducing Raman amplification into systems is to do it in a way

that provides sufficient economic benefit for the owner/system supplier to warrant the
substantial change of technology. For submarine systems, the introduction must be
compatible with the limited physical space in the pressure vessels, limited power-feed
capability, and ultra-high system reliability.

17.2.2. ULH Impairment Accumulation

All ULH transmission systems have at least one thing in common: all the impairments
experienced by the signal along the transmission line accumulate from the transmitter
to the receiver. Obtaining successful ULH transmission between the regenerative
terminals is first a matter of engineering amplifiers and fiber spans so that the optical
602 H. Kidorf et al.

signal-to-noise ratio over the bandwidth of the system is acceptable at the receiver.
This is achieved by carefully controlling amplifier and system gain shape.
Raman amplifiers in ULH systems add several challenges to obtaining spectral
flatness. Broadband Raman amplifier designs need broadband components with a
bandwidth sometimes even in excess of the signal bandwidth. One example is the
optical component that couples the pump laser into the transmission fiber while at
the same time letting the signals pass through. For a multiwavelength-pumped Ra-
man amplifier with 100 nm of bandwidth it is not uncommon to require that this
coupler must have a low-loss bandwidth of 200 nm. Achieving low loss and little
spectral shape for both the signals and the pumps is important for good transmission
performance and good pump efficiency. This can be challenging to component manu-
facturers whose main product lines until recently have been geared towards primarily
C-band EDFAs. Another challenge affecting spectral flatness is how to manage the
pump wavelengths over the life of the system. Unlike EDFAs, Raman gain is directly
affected by deviations of pump wavelengths away from the designed values. This
gives rise to two issues for an ULH system with many cascaded amplifiers: getting
pump lasers at the correct wavelength when the system is being built, and dealing
with pump laser aging/wavelength changes over the life of the system.
Some of the other significant impairments that require consideration in the design
of any ULH system (including Raman amplified systems) are:
• polarization mode dispersion (PMD),
• polarization-dependent loss (PDL),
• chromatic dispersion,
• fiber nonlinearity, and
• multiple reflections from components and fiber (mostly Rayleigh scattering, MPI).
To control the impacts of these effects careful attention must be paid to the design of
the system as well as the specifications for all the optical components used.

17.2.2.1. PMD and PDL

Polarization mode dispersion is introduced into the transmission path through the
components in the optical amplifiers and the transmission fiber itself. This is a much
bigger issue in ULH systems, inasmuch as there are more fiber and more components.
In single-mode fiber there exists a small modal birefringence due to core defor-
mation and external stress. It has been shown that the PMD grows according to the
square root of the length of the transmission line [15]. Fortunately, for ULH systems,
this means that a 10,000 km system has only three times as much PMD as a 1000 km
system. Still, standard fiber has about 0.1 ps/km1/2 resulting in an average of 10 ps
of differential group delay (DGD) over 10,000 km.
Within the optical amplifier, the short length of the devices and their inherent
circular symmetry causes most components to exhibit low PMD. The largest contrib-
utor to PMD in the amplifier is the optical isolator/circulator. Typical isolators have
less than 0.05 ps per device, thus contributing a few more of picoseconds differential
group delay to the transmission line.
 17. Ultra-Long-Haul Applications 603

For the PMD to have a small impact on transmission, the total DGD in the trans-
mission path must be a small fraction of the bit period. At a bit rate of 10 Gb/s (a bit
period of 100 ps), a PMD of 10 to 15 ps due to both the fiber and amplifier components
would likely cause a tolerable amount of transmission penalty. Transmission at higher
bit rates would be difficult, however.
Another polarization effect associated with amplifier components is PDL. An op-
tical component exhibiting PDL where the polarization state of the input signal slowly
changes will cause the signal power to fluctuate. Due to the presence of amplifiers in
the transmission line and the fact that ASE is mostly unpolarized, signal polarization
fluctuations are converted by the amplifiers into changes in received OSNR [16]. Just
as with PMD, the impairment caused by many components each having small values
of PDL accumulates over the length of the transmission line.

17.2.2.2. Fiber Nonlinearity

This impairment is due to the index of refraction of silica fiber being dependent on
the intensity of the light. This effect is given by
P
n = n0 + n2 , (17.1)
Ae
where P is the power in the fiber, Ae is the effective area of the fiber, and n2 is
the nonlinear index of refraction. The accumulated phase change (where Le is the
effective length of each span, N is the number of spans in a system, and λ is the
wavelength) is given by
2π n2
Wnl = P NLe . (17.2)
λAe
Though the nonlinear index is small (about 2.6 × 10−20 m2 /W [18]), the resultant
phase change for light propagating in the fiber becomes significant (about π/2 radians)
when P · Le reaches 1000 mW-km [19]. This value is easily reached in ULH systems.
The nonlinear index of refraction has a number of manifestations: self-phase mod-
ulation (SPM), cross-phase modulation, and four-wave-mixing. A complete treatment
of these impairments [20] is beyond the scope of this book, but any ULH system with
or without Raman amplifiers must account for their impact.
There are two ways to minimize the impact of optical nonlinearity (as can be
seen in the equation above): reduce the power of the signal and increase the effective
area. The impact of Raman amplification on the average power level of the signal is
significant and is discussed later in this chapter. Increasing the effective area of the
transmission fiber is an ongoing effort. The effective area of transmission fibers has
been increased from 50 to 75 µm2 in practice (dispersion-shifted fibers) and to over
120 µm2 (non-dispersion-shifted fibers) in laboratories. This increase in effective area
has clearly led to improvements in transmission performance [14]. However, it should
be noted that Raman amplifiers often depend on pumping smaller effective area fibers
to achieve reasonable efficiencies. Even though increasing the effective area of the
transmission fibers improves performance it also makes Raman amplification less
efficient and, subsequently, less attractive.
604 H. Kidorf et al.

17.2.2.3. MPI
Reflections in transmission systems have long been known to cause several delete-
rious effects. Refractive index discontinuities due to components, connectors, and
splices were shown to cause large power penalties and bit error rate floors in even
simple, unamplified transmission systems [22]. The impact of the inherent Rayleigh
backscatter has also been shown to cause transmission penalty [23].
Rayleigh scattering (one of the dominant loss mechanisms in the 1.5 µm low-loss
transmission window in silica) results from random inhomogeneities in the fiber on
a scale small compared with the wavelength of light. Although somewhat reducible
through improved fabrication technique, the effect cannot be eliminated [24]. For
most single-mode fiber, the power reflected due to Rayleigh scattering results in a
return loss of about 32 dB.
If the signals were simply reflected by Rayleigh scattering, Rayleigh scattering
would have minimal effect on transmission. But because the reflected light itself is
subject to Rayleigh reflection and both of these reflections are subject to distributed
Raman amplification, the doubly reflected and amplified signal (often called multipath
interference) generates an inband noise source for the transmitted signal. In general,
analysis of this type of interference is complicated and has been ongoing since the
first evaluation of MPI in amplified systems in 1990 [25].

17.3. Raman Effects Applicable to ULH Transmission

In this section, the Raman effects that are most important for ULH systems are dis-
cussed. Several of the effects have already been discussed in greater detail in the
preceding chapters of this book. Discussion is included in this chapter relating their
importance to ULH systems. In this way, a better understanding of the use of Raman
amplification for such systems is provided. Most of the discussion in this section
is focused on single-channel Raman amplifiers. This is done to simplify the discus-
sion and present the concepts more clearly. However, all the concepts are readily
extended to also cover multiwavelength-pumped, broadband Raman amplifiers, or
hybrid combinations of Raman amplifiers and EDFAs.
This section primarily discusses the use of distributed Raman amplification.
Lumped Raman amplification for ULH is not nearly as attractive as either distributed
Raman amplification or the use of conventional EDFAs. Lumped Raman amplifica-
tion has the disadvantage of poorer noise performance than both distributed Raman
amplification and EDFAs. Furthermore, the total amount of optical fiber allocated
to the repeaters is significant compared to the spans. In submarine systems, the lim-
ited space in the repeater housings cannot easily accommodate the needed lengths of
fiber. So for ULH systems, lumped Raman amplifiers are most useful in the terminals
of systems operating at wavelengths where no other convenient lumped amplifier
technology is readily available. Another application for lumped Raman amplifiers is
in terrestrial systems where there is a strong desire and/or necessity to reuse previ-
ously installed fiber even though that fiber type might not be particularly suitable for
distributed Raman amplification.
 17. Ultra-Long-Haul Applications 605

Two fundamental concepts in ULH system design are the path-average power and
effective noise figure [26]. These are discussed first. The path-average power provides
a simple way of accounting for the significant changes to the channel power evolution
that results from the use of Raman gain. Effective noise figure allows different lumped
and/or distributed amplifier topologies to be compared with equivalent path-average
power (which causes similar nonlinear impairments).2 The main advantage of the
effective noise figure concept is to allow the majority of the amplifier design trade-offs
to be performed without having to resort to (time-consuming) split-step simulations.
The next two effects discussed are the impact of temperature on Raman ampli-
fication and noise and the additional noise effects generated by Rayleigh scattering.
After all the basic concepts have been clarified, the benefits of lumped versus dis-
tributed amplification are analyzed. The section ends by briefly discussing additional
penalties that are enhanced by Raman amplification that must be taken into account
when Raman is used for ULH transmission.

17.3.1. Path-Average Power

Performance of ULH transmission systems is almost always limited by optical non-

linearity. To reduce accumulation of amplified spontaneous emission (ASE) noise,
higher channel launch power is desired. However, higher launch power also increases
nonlinear impairments. Channel launch power is therefore chosen as a trade-off be-
tween nonlinear distortions and noise accumulation. A comparison of lumped rare
earth-doped fiber amplifier topologies is straightforward and can be performed based
on noise figure, power efficiency, cost, and so on. However, comparing lumped and
distributed amplifiers is not as easy. Noise figure is no longer an unambiguous metric.
Any comparison between distributed and lumped amplifiers in such a case must take
into account that the channel launch power of distributed amplification systems has
to be reduced to make up for the increased power in the span.
Figure 17.4 shows the power evolution of a single channel in a transparent3
90 km Raman amplifier normalized to launch power. The power evolution is shown
for various backward pump ratios (defined as the ratio between backward pump
power and total pump power) as well as for an unpumped fiber where the signal only
experiences loss. The figure shows that the power evolution is strongly affected by
the presence of Raman gain and the pump direction. The signal power for the case
of 100% backward pumping is the smallest everywhere in the span. As more and
more pump power is injected into the forward direction the signal power increases.
To avoid the system being degraded by additional nonlinear impairment, the launch
power has to be reduced. This has an impact on the received OSNR.

2 The performance of real system designs ultimately has to be optimized using nonlinear
propagation models to accurately model the constituent amplifiers and transmission impair-
ments, however.
3 A transparent Raman amplifier is one with unity gain where the internal Raman gain exactly
compensates the fiber loss.
606 H. Kidorf et al.

10
Forward
5

Normalized Power (dB)

0

–5

–10 Backward

–15
Fiber loss
–20
0 10 20 30 40 50 60 70 80 90
Position (km)

Fig. 17.4. Power evolution in a 90 km Raman amplifier for backward pump ratios of 0%
(forward), 25%, 50% (bidirectional), 75%, and 100% (backward). For comparison a fiber span
without Raman gain is also shown.

An important metric for characterizing the nonlinearity enhancement associated

with the use of Raman gain is the path-average power. The path-average power P of
a signal channel is the integrated power of the channel normalized by the span length.
L
1
P= Ps (z) dz, (17.3)
L 0

where Ps (z) is the signal power at position z and L is the span length. The normalized
path-average power η is subsequently defined as the path-average power normalized
by the launch power:
P
η= . (17.4)
Ps (0)
All simple comparisons of distributed and lumped amplifiers should be performed
for the same path-average power. This ensures similar nonlinear impairments. This
is the basis for the definition of effective noise figure in the next section.
Figure 17.5 shows the normalized path-average power versus span length for
forward, bidirectionally, and backward pumped transparent Raman amplifiers. The
normalized path-average power of bidirectionally pumped transparent amplifiers is
always very close to 0 dB. Forward pumped amplifiers have a normalized path-average
power larger than and backward pumped amplifiers smaller than the bidirectional
case. The figure also shows that the difference between the pump directions increases
significantly with span length.
Figures 17.4 and 17.5 deal with transparent Raman amplifiers (unity gain). In
all-Raman systems all amplification must be provided as Raman gain. Multiple op-
tical components are needed in practical amplifiers such as the means for coupling
pump power into the fiber, optical isolators, monitor couplers, and gain-equalizing
 17. Ultra-Long-Haul Applications 607

Normalized path-average power

Forward
6
Bidirectional
4 Backward

2
(dB)
0

–2

–4

–6
0 10 20 30 40 50 60 70 80 90
Amplifier length (km)

Fig. 17.5. Normalized path-average power versus span length for transparent Raman amplifiers.
Curves are shown for forward, bidirectional, and backward pumping.

12
Gain = 0, 1, 3, 6, 10 dB
10
Normalized power (dB)

8
Forward
6

2
Backward
0

–2

–4
0 10 20 30 40 50 60 70 80 90
Position (km)

Fig. 17.6. Power evolution in 45 km forward and backward pumped Raman amplifiers for
excess gain of 0 through 10 dB.

filters. The amplifier must therefore be able to provide excess gain to overcome these
losses in the amount of multiple dBs. This is not the case for Raman-based systems.
Figure 17.6 shows the power evolution for forward and backward pumped 45 km Ra-
man amplifiers with an end-to-end gain of 0 through 10 dB. In conventional lumped
EDFA-based systems, component losses have no effect on the power evolution of the
signals.
The path-average power of forward pumped amplifiers is especially sensitive to
component losses. But even for backward pumped amplifiers the extra gain needed
to overcome the component losses increases path-average power. To avoid nonlinear
608 H. Kidorf et al.

impairments, the launch power has to be reduced; which unfortunately increases noise
accumulation. It is therefore very important for all-Raman systems to use a topology
that minimizes excess losses due to components. Because Raman gain scales directly
with pump power, the addition of unnecessary losses also has a detrimental impact
on pump efficiency. As shown later in this section, excess gain is also undesirable
from the point of view of Rayleigh scattering (both single and double scattering). For
these reasons, hybrid combinations of Raman and EDFAs become attractive for ULH
terrestrial systems with long span lengths.

17.3.2. Effective Noise Figure

Noise performance of lumped fiber amplifiers can easily be compared based on noise
figure. For distributed amplifiers (such as Raman amplifiers) where the path-average
power of the signals is significantly affected, noise figure alone is not sufficient for
comparison. For such amplifiers, it is important to perform the comparison with sim-
ilar amounts of nonlinearity. This is achieved by performing the comparison for the
same path-average power. An effective noise figure is defined in this section to allow
direct comparison of different systems (consisting of both lumped and distributed am-
plifiers). This metric is not only useful for understanding the benefit over conventional
EDFAs, but also for design trade-offs such as pump direction and span length.
The noise figure F of any optical amplifier can be described by [27]
 
+
1 PASE
F= · +1 , (17.5)
G hνBo

+
where G is the gain, PASE is the forward propagating noise, h is Planck’s constant, ν
is the frequency, and Bo is the optical bandwidth. This definition has the advantage of
also working for transparent optical amplifiers (G = 1) and obeys the cascade formula
for cascades of linear amplifiers.
Figure 17.7 shows the noise figure for forward, bidirectionally, and backward
pumped transparent Raman amplifiers versus span length. The figure shows that
Raman amplifiers with span lengths of 40 to 90 km (typical in ULH systems) have
noise figures of 6 to 16 dB. These noise figures, however, include the loss of the span
whereas for conventional EDFAs the quantum limited 3 dB noise figure only includes
the lumped amplifier.
The figure shows that forward pumped amplifiers have the lowest noise figure and
backward pumped amplifiers the highest. This could suggest that forward pumping
would be the best choice. However, in the preceding discussion of path-average
intensity, forward pumping also had the highest relative path-average power.
To take this into account the effective noise figure is therefore introduced. The
effective noise figure Feff is defined by comparison to a lumped amplifier system with
the same received OSNR for the same path-average power after transmission through
a long chain of amplifiers. This allows direct comparison of the noise figure of the
lumped amplifier with the effective noise figure of the distributed amplifier. In other
 17. Ultra-Long-Haul Applications 609

16

14
Forward
12

Noise figure (dB)

Bidirectional
10 Backward
8

0
0 10 20 30 40 50 60 70 80 90
Amplifier length (km)

Fig. 17.7. Noise figure for transparent Raman amplifiers versus span length. Curves are shown
for forward, bidirectional, and backward pumping.

words, the effective noise figure is the noise figure that a lumped amplifier must have
to achieve the same noise performance as a distributed amplifier.
To develop the concept of effective noise figure for a distributed amplifier, we
start with the received OSNR for a transmission span followed by a lumped amplifier
with gain G (gain exactly compensates for the fiber losses). It can be expressed as

P
OSNR l = , (17.6)
ηl · (Fl Gl − 1)hνBo
where the subscript l is used to denote values associated with the lumped amplifier.
Gl and Fl are the gain and noise figure of a single lumped amplifier, respectively.
The definition of normalized path-average power incorporates the launch power of
the signal.
Similarly the received OSNR of a unity gain (amplifier gain exactly compensates
fiber and component losses) distributed amplifier can be expressed as

P
OSNR d = , (17.7)
ηd · (Fd − 1)hνBo
where the subscript d is used to denote values associated with the distributed am-
plifiers. The effective noise figure Feff is now defined as the lumped amplifier noise
figure that would result in the same received OSNR for the same path-average power
as for the distributed case. An expression for the effective noise figure can now be
written using Eqs. (17.6) and (17.7).
 
1 ηd
Feff = (Fd − 1) + 1 . (17.8)
Gl ηl
610 H. Kidorf et al.

2.5
2

Effective noise figure (dB)

1.5
1
0.5
0 Forward/backward
–0.5 Bidirectional
–1
–1.5
–2
0 10 20 30 40 50 60 70 80 90
Amplifier length (km)

Fig. 17.8. Effective noise figure for forward, bidirectionally, and backward pumped transparent
Raman amplifiers.

In the analysis, signals are assumed to experience only linear loss in the spans of
the lumped system; signal–signal Raman interactions are ignored in the equivalent
lumped system. The path-average power of the lumped system can therefore be simply
expressed in closed form and inserted in Eq. (17.8), resulting in:

ln L 1
Feff = ηd · (Fd − 1) + , (17.9)
L−1 L
where L(> 1) is the loss of the span. This is the definition of the effective noise used
in the remainder of this chapter.
Figure 17.8 shows the effective noise figure of forward, backward, and bidirec-
tionally pumped transparent Raman amplifiers versus span length. Only considering
power evolution and noise figure, the figure shows that pump direction has little im-
pact on performance (<0.2 dB) for amplifiers shorter than 50 km. For span lengths
in excess of 50 km bidirectional pumping performs best.
The additional component losses that are present in any practical Raman amplifier
were discussed in Section 17.3.1 (Path-Average Power). It can be shown that the
additional gain needed to make up for the losses actually reduces the noise figure,
while at the same time increasing path-average power. The effective noise figures
for 45 km Raman amplifiers versus gain are shown in Fig. 17.9. For 45 km Raman
amplifiers backward pumping is optimal when the gain exceeds 2.5 dB. For less
than 2.5 dB of excess gain bidirectional pumping works slightly better. Forward
pumping is especially sensitive to excess gain. Regardless of pump direction the figure
demonstrates that effective noise figure and therefore system performance degrades
with increased excess gain.
Apart from performing design trade-offs, the effective noise figure also allows di-
rect comparisons with other systems (including conventional EDFA-based systems),
 17. Ultra-Long-Haul Applications 611

Effective noise figure (dB)

Forward
5 Bidirectional
Backward
4

1
0 2 4 6 8 10
Amplifier Gain (dB)

Fig. 17.9. Normalized noise figure versus gain for 45 km Raman amplifiers.

by simple subtraction of effective noise figures. The effective noise figure of an EDFA
is the noise figure. This is used later in this chapter to quantify the performance im-
provement achieved by using distributed amplification.

17.3.3. Temperature-Dependence of Noise Figure

The noise figure of Raman amplifiers is temperature-dependent because at tem-

peratures above absolute zero there is a nonzero probability of finding molecules
thermally excited [28]. These thermally excited molecules can interact both with
signal and pump photons and create spontaneous and stimulated emission (both
Stokes and anti-Stokes). On average, anti-Stokes stimulated emission cancels out with
the temperature-enhanced Stokes stimulated emission. The Raman gain is therefore
temperature-independent. The spontaneous emission (both Stokes and anti-Stokes)
is, however, temperature-dependent. Thus the noise figure of a Raman amplifier is
temperature-dependent but the gain is not.
Figure 17.10 shows the Raman scattering cross-section both at absolute zero
and at room temperature (300◦ K). It demonstrates that the anti-Stokes scattering
as well as the enhanced noise generation disappears at absolute zero. The effect
of the temperature-dependence is seen to be significant especially close to the pump
wavelength; but even at the gain peak the effect is not insignificant. As an example: for
a single-wavelength pumped Raman amplifier with close to unity gain the noise figure
at the gain peak is increased by ∼0.5 dB. This increase might not seem significant, but
for shorter spans the inherent Raman benefit compared to EDFAs can be as small as 1
dB. Also for wideband multiwavelength pumped Raman amplifiers where the longest
wavelength pumps are very close to the shortest wavelength signals significant excess
noise can be added thereby increasing the noise figure well in excess of 0.5 dB.
Most of the simple models that have been published traditionally neglect this
temperature-dependence [29, 30]. However, to realistically predict the performance
612 H. Kidorf et al.

1.2

Raman scattering cross-section

0K
1.0 Pump wavelength: 1 µm 300 K

0.8

(×10–13 m/W)
0.6 Stokes

0.4 Anti-Stokes

0.2

0.0
–20 –10 0 10 20 30 40
Frequency difference (THz)

Fig. 17.10. Spontaneous Raman scattering at room temperature and at absolute zero.

advantage of Raman amplification over EDFAs or to compare alternate Raman am-

plifier topologies it is important to take this effect into account.

17.3.4. Noise Enhancement Through Rayleigh Scattering

Rayleigh scattering in EDFA systems mainly manifests itself as span loss. For Raman
amplified systems, however, Rayleigh scattering is also responsible for contributing
additional noise. The two main contributions are: double-Rayleigh scattering (DRS)
of the signal leading to incoherent inband crosstalk and (although not as important
as DRS) single reflections of backward traveling spontaneous noise into the forward
direction. Both effects are also present in EDFA systems, but due to the fiber loss,
they have very small impact. For distributed amplifiers with close to unity gain the
effects are much enhanced.
DRS can significantly limit system performance in ULH systems. Because, both
the single and double reflected signals are amplified, the DRS efficiency scales with the
gain squared [26, 31]. This is another reason why amplifier topologies and components
should be chosen to minimize excess gain. Figure 17.11 shows the crosstalk relative
to the signal power at the output of a single 45 and 90 km Raman amplifier versus
gain. Assuming that the crosstalk adds as power, the total crosstalk after 100 to 200
amplifiers is 20 to 23 dB higher than after a single amplifier. Virtually no penalty is
seen when the crosstalk is 30 dB lower than the signal at the receiver [32].
The figure shows that the crosstalk increases rapidly with gain and span length.
A system based on 90 km spans has half the amplifiers compared to a system with
45 km spans. Even taking this into account, the 90 km span system suffers ∼7 dB
more from double-Rayleigh scattering unless bidirectional pumping is used. This
is another reason why bidirectional pumping is very advantageous for long Raman
amplifiers. The figure also shows that even for unity gain and short repeater spacing
it is very hard to keep the double-Rayleigh scattering for each amplifier below −50
 17. Ultra-Long-Haul Applications 613

–10
Forward and backward
–20

Rayleigh crosstalk (dB)

Bi-directional

–30

– 40
90 km
–50

45 km
–60
–20 –10 0 10 20 30
Gain (dB)

Fig. 17.11. Rayleigh crosstalk relative to signal power at output of 45 and 90 km Raman
amplifiers versus net amplifier gain (assuming a fiber loss of 0.2 dB/km and no excess losses).

1.0
Normalized reflected noise (dB)

0.8
0 dB
0.6 10 dB
20 dB

0.4

0.2

0.0
10 20 30 40 50 60 70 80 90
Amplifier length (km)

Fig. 17.12. Increased noise figure through Rayleigh reflection of backward ASE for backward
pumped transparent Raman amplifiers.

dB so that hundreds of amplifiers can be cascaded with insignificant penalty. One

way to limit the penalty is to combine the Raman amplifier with an EDFA to form a
hybrid amplifier. In this way the Raman and EDFA gains can be chosen to optimize
performance by trading off ASE noise and DRS. This has been used to obtain a 7000
km transmission distance with 100 km spans using terrestrial fibers [38].
In EDFA systems, the reflection of the backward traveling ASE is only an issue
for high-gain amplifiers. In this case the extra noise can be avoided by the use of input
and output optical isolators. This solution cannot be used for distributed amplifiers
where both reflections and gain happen in the same fiber. Figure 17.12 shows the
614 H. Kidorf et al.

noise figure increase of backward pumped Raman amplifiers versus amplifier length
for different amplifier gains. For reasonable amplifier designs with only a few dB
of component losses this is not a major issue. However, for a Raman amplifier with
significant excess gain, noise figure degradations of ∼1 dB are possible.

17.3.5. Comparison of Distributed and Discrete Amplification

We have now reviewed the main concepts of effective noise figure and significant
Raman enhanced noise effects. The performance of distributed amplification can now
be compared to that of lumped amplification. To make the comparison general no de-
tails of distance or topology are included. The benchmark is an ideal lumped system
consisting of amplifiers with quantum limited noise figure (3 dB). The comparison
is performed as a function of span length and pump direction for the distributed am-
plifiers. For the distributed systems, component losses of 4 dB are assumed which
include typical optical components such as input pump signal coupler, optical iso-
lator(s), output tap, and gain-flattening filter. Because all distributed amplifiers are
assumed to be transparent the excess Raman gain must therefore also equal 4 dB. The
Raman amplifiers are assumed to be operated at room temperature where the noise
figure is degraded by approximately 0.5 dB as previously discussed. DRS penalties
are not included in this comparison. It is assumed that the chosen repeater spacing
for the desired distance does not cause significant degradations because of DRS. This
is of course important to verify for a particular application. The comparison is per-
formed by calculating the effective noise figure of the distributed system. An effective
noise figure of <3 dB means that the distributed amplifiers produce less noise than
the lumped reference system. The received OSNR scales directly with the effective
noise figure. Figure 17.13 shows the effective noise figure calculated for forward,
bidirectionally, and backward pumped Raman amplifiers.

4
Effective Noise figure (dB)

0 Forward
Bidirectional
–1
Backward
–2

–3
20 30 40 50 60 70 80 90 100
Amplifier length (km)

Fig. 17.13. Effective noise figures for forward, backward, and bidirectionally pumped Raman
amplifiers with a temperature-dependent noise figure degradation of 0.5 dB.
 17. Ultra-Long-Haul Applications 615

The figure shows that for very short amplifier spacing, lumped amplifiers work as
well as distributed amplifiers and there is therefore little reason for using distributed
amplification. For intermediate length amplifier spacings (∼50 km) backward pumped
Raman amplifiers provide the best performance, but the benefit does not exceed 1 dB
over lumped amplification until a span length of 55 km. For the longest span lengths,
bidirectional pumping significantly improves performance (up to 5 dB compared to
lumped amplification). It is important to emphasize that even though the smallest
effective noise figure is seen for the longest spans, performance is still worse than for
shorter spans. Having a lower effective noise figure just means that the distributed
benefit over lumped amplification for the same span length is larger.
This comparison illustrates why especially for terrestrial systems, where the re-
peater hut spacing is fixed and typically on the order of 90 km, Raman amplification
has been strongly embraced. On the other hand, for submarine applications where the
repeater spacing can be chosen to optimize system performance and is typically on
the order of 50 km, Raman amplification is not as competitive as EDFAs.

17.3.6. Other Degradation Mechanisms

There are several other issues and performance-degrading effects that are unique to or
enhanced by the presence of Raman gain. Most of these are discussed in much greater
detail in other parts of the book, but are briefly mentioned here for completeness.
Relative intensity noise (RIN) from the pump lasers can couple to the signals.
This pump-RIN transfer is especially strong for forward pumping and virtually in-
significant for backward pumping. For this reason all of the early experiments used
only backward pumping. As discussed in connection with effective noise figure, in
general there is no particular advantage of forward pumping as compared to back-
ward pumping. However, for long Raman amplifiers, bidirectional pumping is very
attractive. Because bidirectional pumping requires that part of the pump light propa-
gate in the forward direction, pump lasers with low RIN are needed. Recently several
pump laser designs that provide high power and low RIN have therefore been devel-
oped [33, 34].
Raman gain is polarization-dependent; maximum gain is obtained only when sig-
nal and pump are copolarized. To avoid unacceptable levels of polarization-dependent
gain, the Raman pump has to be mostly unpolarized. This is especially important for
forward pumping, but is also necessary for backward pumping. (That the backward
pump and signal travel in opposite directions and follow independent evolutions of
the polarization state helps to “scramble” the polarization-dependent interactions.)
Early experiments used fiber lasers to provide the Raman pump. These sources are
inherently unpolarized [12]. For practical use, semiconductor pump lasers are a more
desirable choice. However, these sources are polarized and must therefore be depo-
larized for acceptable system performance. One popular way of ensuring depolarized
Raman pumping is to combine two lasers with similar wavelengths using a polariza-
tion beam combiner. This also has the advantage of almost doubling available pump
power at each wavelength. The downside is that more diodes are needed, which
influences cost, space, etc.
616 H. Kidorf et al.

17.4. Applications of Raman Gain

Due to its unique properties, Raman amplification has been studied extensively since
the advent of low-loss optical fibers. Over the past few years, one of the last tech-
nological obstacles on the way to practical implementation of Raman amplifiers in
telecommunication has been overcome with the development of compact, reliable,
and powerful Bragg grating stabilized laser diodes (LDs).
The system designer can now use Raman amplification in a number of different
ways. In some designs, Raman amplifiers allow the unregenerated reach of the system
to be increased or the repeaters (i.e., amplifiers) to be spread farther apart. In other
designs, the bandwidth of the system can be increased; or, systems can be operated
at wavelengths where rare earth-doped fiber amplifiers are not available. To achieve
these goals, Raman amplification can be used either in combination with EDFAs (as
hybrid Raman/EDFA) or on their own in a transmission system that employs only
Raman amplification (all-Raman) [35, 36]. Progress toward each of these goals is
discussed in the following sections.

17.4.1. Extending Repeater Spacing Using Hybrid Raman Assisted EDFAs

The addition of Raman amplification to erbium-doped amplifiers is very much like
adding a low-noise preamplifier to the EDFA. Combining Raman amplification with
EDFAs has several benefits. Raman amplifiers are very good at providing low-noise
preamplification for the signals, whereas EDFAs provide saturation and high output
power with good pump efficiency. Saturation is desirable for long systems because the
signal power is then self-limiting and active control of the amplifier can potentially
be avoided. As shown in the previous section, it is very desirable to limit the excess
gain that the Raman amplifier has to provide. Less excess gain has an impact on DRS,
effective noise figure, pump efficiency, and so on. This way of using Raman amplifi-
cation, however, limits the bandwidth of the system to that of the lumped amplifier.
The performance improvement obtained from adding the Raman assist to the
EDFA can be used (among other things) to extend the unregenerated reach of the
system or to extend the repeater spacing (distance between amplifiers) while still
fulfilling the design targets (system length, capacity, etc.). For a given system length,
the desire to reduce the system cost is a strong motivation to extend the amplifier
spacing as much as possible while achieving the desired system capacity. This can be
particularly attractive in ULH transmission systems where in excess of 200 amplifiers
[44] are used in the transmission lines of the longest systems.
Many experiments have been conducted to explore the feasibility of adding margin
to systems by adding Raman assist to the EDFAs. In the first attempt (1998) to
demonstrate the use of distributed Raman amplification to increase the span length in
ULH WDM systems, a record 240 km between repeaters (defined as optoelectronic
units requiring power and supervision) was obtained [37]. This was achieved with
distributed Raman amplifiers and remotely pumped EDFAs. The configuration of the
240 km transmission spans is shown in Fig. 17.14(a). Remote pumping by a fiber
laser of EDF1 through a low-loss pure silica-core fiber forms the first amplifier. The
dedicated pump fiber is used to avoid interactions between the noisy copropagating
 17. Ultra-Long-Haul Applications 617

pump and the signal. A second fiber laser backward pumped the remaining three
amplifiers. After passing through EDF3, the pump power was then launched into
the transmission fiber creating a distributed Raman amplifier with approximately 14
dB of internal gain. The pump power remaining after passing through the Raman
amplifier was used to pump EDF2. Isolators were used along the transmission line
to reduce MPI. In total, two pump sources, each providing 1.2 W (at 1480 nm) of
optical power were required for each span. The signal power evolution between the
repeaters, depicted in Fig. 17.14(b), was reduced to less than 20 dB of total excursion.
Without the remote EDFAs and the Raman amplifier a signal excursion in excess of
48 dB would be expected. The impact of the remotely pumped EDFAs is similar to
that of distributed Raman amplification, that is, making the power distribution along
the propagation path more even.
Hybrid Raman/EDFA amplifiers have also been shown to be useful in creating
ULH transmission lines using fibers of a type that is already deployed in much of the
world. Raman amplification in combination with an EDFA has been used to achieve
transcontinental (7500 km) transmission with 90 km spans built from LEAFTM fiber

240 km Span

80 km. 80 km. 80 km.

DSF WDM/ DSF DSF WDM/
EDF EDF1 Isolator EDF2 EDF3 Isolator

Pump Laser Pump Laser

Pump Laser Pump Laser

80 km.
Si Core Repeater Contents

(a)

15.0 EDF1 EDF3

Signal Power, dBm

10.0
EDF2
5.0
0.0
–5.0
–10.0 Raman Gain
–15.0
0 50 100 150 200 250
Distance, km

(b)

Fig. 17.14. (a) Configuration of the 240 km repeater span; (b) power evolution in the span.
618 H. Kidorf et al.

[38]. Two unpolarized pump sources with wavelengths of 1430 and 1457 nm backward
pumped the transmission fiber to provide 10 dB of internal Raman gain. The remaining
10 dB of gain needed to overcome the loss of the transmission fiber was provided
by the EDFA. In this way, both amplifier technologies were used: distributed Raman
preamplification ensured good noise performance and the EDFA provided the output
power with good pump efficiency.
In the section on effective noise figure it was shown that bidirectional pumping,
which achieves a more distributed gain, improves noise performance for longer spans
(in excess of 55 km). This has also been demonstrated in several experiments [39, 40].
Another approach to improve the distribution of Raman gain is achieved by chang-
ing the order of the fibers in the span. In most system designs, most of the Raman
amplification comes from a small effective area (and hence better Raman pump ef-
ficiency) fiber at the end of the span; see Fig. 17.15(a). Improvements were demon-
strated by placing the fiber with smaller effective area in the middle portion of the
transmission span (Fig. 17.15(b)) [41]. This fiber arrangement improved the system

+D fiber +D fiber –D fiber

D = 20ps/nm/km D = 18ps/nm/km D = –35ps/nm/km
Aeff = 100 µm2 Aeff = 80 µm2 Aeff = 20 µm2
L = 40 km L = 10 km L = 25 km

EDF

Raman

(a)

+D fiber –D fiber +D fiber

D = 20ps/nm/km D = –35ps/nm/km D = 18ps/nm/km
Aeff = 100 µm2 Aeff = 20 µm2 Aeff = 80 µm2
L = 40 km L = 25 km L = 10 km

EDF

Raman

(b)

Fig. 17.15. (a) Conventional map (+D/ −D), (b) Aeff managed map (+D/ −D/ +D).
 17. Ultra-Long-Haul Applications 619

OSNR by 0.9 dB. By shifting the Raman gain into the middle section, a more dis-
tributed amplification was achieved. However, doing this also significantly reduces
pump efficiency.
Yet another approach to distribute the Raman gain more uniformly is the use of
higher-order pumping, where the pump signals are Raman amplified by the addition of
strong shorter wavelength pumps. This concept has the disadvantage of needing even
more pump wavelengths, which increases complexity and also suffers from reduced
efficiency. Higher-order pumping is described in detail in Chapter 10.
Following the first demonstration in 1998, numerous experiments have since
confirmed the benefits of ULH transmission using Raman-assisted EDFAs supporting
a very high transmission capacity [42]. Most recent ULH transmission experiments
performed using terrestrial span lengths (∼100 km) depend on Raman amplification
to achieve their goals.

17.4.2. Extending Bandwidth Using Raman/EDFAs

One of the main advantages that Raman amplification has over rare earth-doped
amplifiers is that Raman amplification is not limited to wavelength bands linked to
atomic transitions. Raman amplification can be realized across the whole transparency
spectrum of silica fibers. It can therefore be used to expand the transmission bandwidth
of ULH WDM systems beyond the EDFA’s C-band (and even L-band).
The first use of Raman amplification to extend the bandwidth of ULH systems
was the demonstration of a hybrid combination of Raman amplification and an
erbium-doped fluoride fiber (EDFFA) amplifier with overlapping gain spectra [43];
see Fig. 17.16(a). To construct the hybrid amplifier, the experimenters followed a
Raman amplifier (with a gain peak at 1610 nm) with a two-stage EDFFA. The re-
sultant gain bandwidth was 75 nm. This type of amplifier has the advantage that its
gain is continuous. That is, there is no break in the gain spectrum as there is in most
C+L-band EDFA amplifiers.
A similar hybrid approach resulted in the realization of a continuous bandwidth
of 80 nm in the demonstration of transmission over 11,000 km of 256 channels mod-
ulated at 10 Gb/s [44]. The 80 nm Raman/EDFA hybrid was constructed by using
a single-wavelength, unpolarized, backward pumped Raman amplifier in front of
a single-stage EDFA. The gain spectra of the Raman and EDFA sections comple-
mented each other and provided a wide continuous bandwidth with minimum gain
ripple (Fig. 17.17). A pump wavelength of 1497 nm was chosen to best comple-
ment the gain shape of the highly inverted EDFA. The pump radiation from two
polarization-multiplexed, grating stabilized lasers was coupled into the span using a
circulator to achieve simultaneous low-loss and isolation for both pump and signals.
The transmission spans used a dispersion-matched combination of large effective area
(110 µm2 , +20 ps/nm/km) and small effective area (30 µm2 , −40 ps/nm/km) fibers
in a 2:1 ratio. The pump was coupled into the fiber with the small effective area to
achieve good efficiency. A gain-flattening filter at the output was designed to equalize
the gain shape of the combined Raman/EDFA. The amplifier had a total output power
of 18.6 dBm. The OSNR as measured after 11,000 km of transmission corresponded
to an effective NF of a single Raman/EDFA ranging from 2.7 dB in the L-band to
620 H. Kidorf et al.

30

25 EDFA + Raman
20

Gain (dB)
15 EDFA

10

5
Raman
0

–5
1520 1540 1560 1580 1600 1620
Wavelength (nm)
(a)

Raman amplifier EDFA

DSF EDFF Gain EDFF
85 km equalizer
Signal Circulator Coupler Coupler Isolator

Raman pump Cascaded

Pump LDs Mach-Zehnder
1505 nm liters
Pump LD Pump LD
1480 nm 1480 nm
Polarization
beam
combiner

(b)

Fig. 17.16. (a) Overlapping gain spectra of EDFF and Raman Amplifiers; (b) block diagram
of hybrid EDFF and Raman amplifier.

4.8 dB at the shortest wavelength. The improvement of the OSNR towards longer
wavelengths is attributed to the increased contribution from distributed Raman gain
(Fig. 17.17(b)).
This ability to significantly increase the bandwidth of a system by adding only a
few extra components compares very favorably to the C+L-band EDFA alternative.
Compared to C+L-band amplifiers, wider bandwidth, better noise performance, and
potentially better electrical efficiency can be achieved. The repeater spacing, however,
is limited to what is possible using just EDFAs unless Raman assist is added to
preamplify the channels that are predominantly amplified by the EDFAs.

17.4.3. All-Raman Amplification

Despite the lure of easily constructed and pump-efficient EDFAs, many researchers
have successfully demonstrated ULH amplifier chains based entirely on Raman am-
plification. The first WDM demonstration of all-Raman amplification in an ULH
transmission line was made in 1997 with an extremely simple Raman amplifier ar-
chitecture [12]. A wavelength selective coupler (to inject the pump while passing the
 17. Ultra-Long-Haul Applications 621

EDFA
980 nm Tap
Circulator Er-fiber ISO GFF
D+ D– WDM
In Out

PBS

1497 nm
LD

(a)

4 Total gain 14
2 12
0 10
Gain (dB)

Gain (dB)
–2 8
–4 6
–6 4
–8 EDFA gain 2
–10 Raman gain 0
–12 –2
1520 1540 1560 1580 1600 1620
Wavelength (nm)
(b)

Fig. 17.17. (a) Hybrid Raman/EDFA; (b) Raman, EDFA, and total gain.

signal), isolator, and tap were used in each repeater (see Fig. 17.18). Despite the sim-
plicity of the amplifier, a chain of these amplifiers was successfully used to transmit
ten 10 Gb/s channels over 7200 km. The bandwidth limitation was due, in part, to
the narrow bandwidth caused by single-wavelength pumping and the lack of gain
equalization in the amplifier.
All-Raman amplifier chains have since matured significantly. Most notably, mul-
tiwavelength pumping has been employed and careful fiber span design has been
employed to permit wideband operation. Some recent results are shown in Table 17.2.
Apart from the unique ability of Raman amplifiers to provide broadband gain
through the use of multiple wavelength pumping, these amplifiers also offer new ways
of dealing with gain equalization. For example, 100 nm bandwidth Raman amplifiers
have been realized using a 12 wavelength pump unit comprising high-power laser
diode modules. Gain-flatness of better than ±0.5 dB was achieved without the use of
any gain-equalization filters [47]. This level of gain-flatness is, however, not sufficient
for most ULH systems and in general gain-flattening filters will be necessary even
for multiwavelength pumped Raman amplifiers.
622 H. Kidorf et al.

Wavelength
Selective Tap
Transmission Coupler
Fiber

Pump Laser
1452 nm

Fig. 17.18. Amplifier configuration used with first all-Raman ULH transmission experiment.

Table 17.2. Comparison of Recent All-Raman Transmission Results

Year Transmission Span Length Bandwidth Capacity
[Ref.] Length (km) (km) (nm) (Tb/s) Configuration
2001 [45] 7400 40 74 2.4 Backward pumped
2001 [40] 4000 100 53 1.3 Bidirectionally pumped
2002 [46] 4000 100 53 2.5 Backward pumped
42.7 Gb/s channels
2002 [39] 3600 100 31 1.6 Bidirectionally pumped
42.7 Gb/s channels

Using all-Raman amplification allows simultaneous wideband operation and long

repeater spacing. However, since all of the gain to overcome fiber and component
losses must come from Raman amplification, DRS and increased path-average power
can affect performance. Whether all-Raman amplification performs better than hybrid
Raman/EDFAs is strongly dependent on the details of the system design. It is clear
from Table 17.2 that multi-Tb/s fiber capacity is available over ULH fiber spans with
all-Raman amplification. Early concerns about insurmountable obstacles caused by
pump power demand, MPI, channel–channel crosstalk, and the like, have all been
overcome (but still need to be taken into account) to produce a technology that can
potentially compete with the EDFA.

17.5. Conclusion
Ultra-long-haul optically amplified transmission systems are some of the most tech-
nically challenging systems designed today. Raman amplifiers have proven their use-
fulness in certain categories of these systems, that is, terrestrial systems. In terrestrial
systems, the marriage of Raman amplification technology and erbium-doped fiber
amplifiers has demonstrated great benefit by expanding the bandwidth of ampli-
fiers, extending the distance between amplifiers, and allowing longer distances to
be spanned. Because of the long repeater spacing in typical terrestrial networks the
benefit of adding Raman is substantial. Many manufacturers of the terrestrial systems,
therefore, already offer a Raman product.
 17. Ultra-Long-Haul Applications 623

For submarine systems where the systems are designed to achieve a desired ca-
pacity over often the longest transmission distances (6000 to 11,000 km), shorter
span length (than for terrestrial systems) often has to be chosen. For such shorter
spans (∼50 km) the benefits of Raman amplification are not nearly as substantial.
Furthermore, submarine systems have much more stringent requirements for power
efficiency, volume, and reliability. These requirements put Raman amplification at a
further disadvantage. In addition, ongoing transmission fiber development has con-
tinuously increased the effective area of the transmission fibers to lower optical non-
linearity and increase capacity and span length. This has a negative impact on Raman
amplifier designs, because the pump efficiency is further reduced. Even faced with all
these challenges most submarine suppliers have ongoing research efforts to find appli-
cations for Raman in undersea systems. Presently, the most promising candidate use
of Raman amplification in submarine systems is the wideband hybrid Raman/EDFA.
For systems that require a very wide bandwidth this seems like an attractive way to
more than double the transmission bandwidth without doubling the component count.

References
[1] R. E. Wagner, L. Nederlof, and S. De Maesschalck, The potential of optical layer networks,
Technical Digest of Optical Fiber Communications 2001, TuT3-1, 2001.
[2] Lucent’s new all-optical router uses Bell Labs microscopic mirrors, Lucent Press Release,
November 10, 1999. http://www.bell-labs.com/news/1999/november/10/1.html.
[3] B. Dibner, The Atlantic Cable, Norwalk, CT: Burndy Library, Inc., 1959.
[4] J. Brooks, Telephone: The First Hundred Years, New York: Harper & Row, 90, 1975.
[5] E. H. Ehrbar, Undersea cables for telephony. In Undersea Lightwave Communications,
ed. P. K. Runge and P. Trischitta, New York: IEEE Press, 1986.
[6] A. Clarke, Voice Across The Sea, New York: Harper & Row, 1959.
[7] J. Hecht, City of Light: The Story of Fiber Optics, New York: Oxford University Press,
1999.
[8] L. F. Mollenauer, R. H. Stolen, and M. N. Islam, Experimental demonstration of soliton
propagation in long fibers: Loss compensated by Raman gain, Optics Lett., 10:5, (May),
229–231, 1985.
[9] L. F. Mollenauer, R. H. Stolen, and M. N. Islam, Demonstration of soliton transmission
over more than 4000 km in fiber with loss periodically compensated by Raman gain,
Optics Lett., 13:8, (August), 1988.
[10] R. H. Stolen and E. P. Ippen, Raman gain in glass optical waveguides, Appl. Phys. Lett.,
22:6, (March), 276–278, 1973.
[11] S. G. Grubb, T. Strasser, W. Y, Cheung, W. A. Reed, V. Mizrahi, T. Erdogan, P. J. Lemaire,
A. M. Vengsarkar and D. J. DiGiovanni, High power 1.48 µm cascaded Raman laser
in germanosilicate fibers. In Proceedings of Optical Amplifiers and Their Applications,
Davos, Switzerland, (June), 197–199, 1995.
[12] M. Nissov, C. R. Davidson, K. Rottwitt, R. Menges, P. C. Corbett, D. Innis, and
N. S. Bergano, 100 Gb/s (10×10Gb/s) WDM transmission over 7200 km using dis-
tributed Raman amplification. In Proceedings of ECOC’97, 1997.
[13] P. Lancaster, P. Mejasson, A. Cordier, C. Little, T. Shirley, P. Dupire, and T. Farrar,
Efficient powering of long haul and high capacity submarine networks. In Proceedings
of SubOptic 2001 International Convention, Kyoto, Japan, May 20–24, T4.5.2, 2001.
624 H. Kidorf et al.

[14] Suzuki, H. Kidorf, et al., 170 Gb/s transmission over 10,850 km using large core trans-
mission fiber. In Postdeadline Papers of OFC, 1998.
[15] F. Corti, B. Daino, G. de Marchis, and F. Matera, Statistical treatment of the evolution
of the principal states of polarization in single-mode fiber, J. Lightwave Technol., 8:8,
(August), 1162–1166, 1990.
[16] C. D. Poole and J. Nagel, Polarization effects in lightwave systems. In Optical Fiber
Telecommunications IIIA, ed. I. Kaminow and T. Koch, 153, San Diego: Academic Press,
1997.
[17] E. A. Golovchenko, A. N. Pilipetskii, N. S. Bergano, C.R Davidson, F. I. Khatri,
R. M. Kimball, and V. J. Mazurczyk, Modeling of transoceanic fiber-optic WDM commu-
nication systems, IEEE J. Select. Topics Quantum Electron., 6:2, (March/April), 337–347,
2000.
[18] K. S. Kim, W. A. Reed, R. H. Stolen, and K. W. Quoi, Measurement of the non-linear
index of silica core and dispersion shifted fibers, IEE Electron. Lett., 32:570.
[19] F. Forghieri, R. W. Tkach, and A. R. Chraplyvy, Fiber nonlinearities and their impact on
transmission systems. In Optical Fiber Telecommunications IIIA, ed. I. Kaminow and
T. Koch, San Diego: Academic Press, 1997.
[20] G. P. Agrawal, Non-Linear Fiber Optics, 2d ed., San Diego: Academic, 1995.
[21] F. Forghieri, R. W. Tkach, and A. R. Chraplyvy, Fiber nonlinearities and their impact on
transmission systems. In Optical Fiber Telecommunications IIIA, ed. I. Kaminow and
T. Koch, San Diego: Academic Press, 1997.
[22] J. L. Gimlett and N. K. Cheung, Effects of phase-to-intensity noise conversion by multiple
reflections on gigabit-per-second DFB laser transmission systems, J. Lightwave Technol.,
7:6, (June), 888–895, 1989.
[23] A. F. Judy, The generation of intensity noise from fiber Rayleigh backscatter and discrete
reflections. In Proceedings of the European Conference on Optical Communications,
TuP11, 1989.
[24] J. M. Senior, Optical Fiber Communications, Englewood Cliffs, NJ: Prentice-Hall, 69,
1985.
[25] J. L. Gimlett, M. Z. Iqbal, N. K. Cheung, A. Righetti, F. Fontana, and G. Grasso, Ob-
servations of equivalent Rayleigh scattering mirrors in lightwave systems with optical
amplifiers, IEEE Photon. Technol. Lett., 2:3, (March), 211–213, 1990.
[26] M. Nissov, Long-haul optical transmission using distributed Raman amplification, PhD
thesis, Department of Electromagnetic Systems, Technical University of Denmark, De-
cember 1997.
[27] B. Pedersen, A. Bjarklev, J. H. Povlsen, K. Dybdal, and C. C. Larsen, The design of
erbium doped fiber amplifiers, J. Lightwave Technol., 9:9, (Sept.), 1105–1112, 1991.
[28] R. Stolen, Stimulated Raman scattering (SRS). In Optical Fiber Telecommunications, ed.
S. E. Miller and A. G. Chynoweth, Chapter 5.2, 127–133, New York: Academic, 1979.
[29] G. P. Agrawal, Non Linear Fiber Optics, 2d ed. Chapter 8, San Diego: Academic, 1995.
[30] R. G. Smith, Optical power handling capability of low loss optical fibers as determined
by stimulated Raman and Brillouin scattering, Appl. Optics, 11:11, 2489–2494, 1972.
[31] M. Nissov, K. Rottwitt, H. D. Kidorf, and M. X. Ma, Rayleigh crosstalk in long cascades
of distributed unsaturated Raman amplifiers, IEE Electron. Lett., 35:12, (June), 997–998,
1999.
[32] H. Takahashi et al., Impact of crosstalk in an arrayed waveguide multiplexer on n×n
optical interconnection, J. Lightwave Technol., 14:6, (June), 1097–1105, 1996.
[33] N. Tsukiji et al., Advantage of inner-grating-multimode laser (iGM laser). In Proceedings
of OMB4, Optical Amplifiers and Their Applications, Vancouver Canada, July 14–19,
2002.
 17. Ultra-Long-Haul Applications 625

[34] L. L. Wang et al., Linewidth limitations of low noise wavelength stabilized Raman pumps.
In Proceedings of OMB5, Optical Amplifiers and Their Applications, Vancouver Canada,
July 14–19, 2002.
[35] A. Evans, Raman amplification in broadband WDM systems. In Technical Digest of
OFC’01, Anaheim, CA, TuF4: 2001.
[36] M. N. Islam, Raman amplifiers for telecommunications, IEEE J. Select. Topics Quantum
Electron., 8:3, 548–559, 2002.
[37] M. X. Ma, H. D. Kidorf, K. Rottwitt, F. W. Kerfoot, and C. R. Davidson, 240-km repeater
spacing in a 5280-km WDM system experiment using 8 2.5 Gb/s NRZ transmission,
IEEE Photon. Technol. Lett., 10:6, (June), 1998.
[38] C. B. Clausen, S. Ten, B. Cavrak, C. R. Davidson, A. N. Pilipetskii, M. Nissov, E. A.
Golovchenko, R. Ragbir, and K. Adams, Modeling and experiments of Raman assisted
ultra long-haul terrestrial transmission over 7500 km, Technical Digest of ECOC’01,
Amsterdam, The Netherlands, 315-1, 2001.
[39] F. Liu, J. Bennike, S. Dey, C. Rasmussen, B. Mikkelsen, and P. Mamyshev, 1.6 Tbit/s
(40×42.7 Gb/s) transmission over 3600 km UltraWaveTM fiber with all-Raman amplified
100 km terrestrial spans using ETDM transmitter and receiver. In Postdeadline Papers
of OFC’02, Anaheim, CA, FC7, 2002.
[40] D. F. Grosz,A. Küng, D. N. Maywar, L.Altman, M. Movassaghi, H. C. Lin, D.A. Fishman,
and T. H. Wood, Demonstration of all-Raman ultra-wide-band transmission of 1.28 Tb/s
(128×10 Gb/s) over 4000 km of NZ-DSF with large BER margins. Postdeadline Papers
of ECOC’01, Amsterdam, The Netherlands, 72–73, 2001.
[41] K. Shimizu, K. Ishida, K. Kinjo, T. Kobayashi, S. Kajiya, T. Tokura, T. Kogure, K. Mo-
toshima, and T. Mizuochi, 65 × 22.8 Gb/s WDM transmission over 8,398 km employing
symmetrically collided transmission with Aeff managed fiber. In Technical Digest of
OFC’02, Anaheim, CA, WX4, 2002.
[42] T. Tanaka, N. Shimojoh, T. Naito, H. Nakamoto, I. Yokota, T. Ueki, A. Suguyama, and
M. Suyama, 2.1-Tbit/s WDM transmission over 7,221 km with 80-km repeater spacing.
Postdeadline Papers of ECOC’00, Munich, PD1.8, 2000.
[43] H. Masuda, S. Kawai, K.-I. Suzuki, and K. Aida, 75-nm 3-dB gain-band optical ampli-
fication with erbium-doped fluoride fibre amplifiers and distributed Raman amplifiers in
9 × 2.5 Gb/s WDM transmission experiment. In Proceedings of the European Conference
on Optical Communication (ECOC’97), 73–75, 1997.
[44] D. G. Foursa, C. R. Davidson, M. Nissov, M. A. Mills, L. Xu, J. X. Cai, A. N. Pilipetskii,
Y. Cai, C. Breverman, R. R. Cordell, T. J. Carvelli, P. C. Corbett, H. D. Kidorf, and
N. S. Bergano, 2.56 Tb/s (256 × 10 Gb/s) transmission over 11,000 km using hybrid
Raman/EDFAs with 80 nm of continuous bandwidth. In Postdeadline Papers of OFC’02,
Anaheim, CA, FC8, 2002.
[45] N. Shimojoh, T. Naito, T. Tanaka, H. Nakamoto, T. Ueki, A. Sugiyama, K. Torii, and M.
Suyama, 2.4-Tbit/s WDM transmission over 7400 km using all Raman amplifier repeaters
with 74-nm continuous single band. In Postdeadline Papers of ECOC’01, Amsterdam,
The Netherlands, 8–9, 2001.
[46] A H. Gnauck, G. Raybon, S. Chandrasekhar, J. Leuthold, C. Doerr, L. Stulz, A. Agarwal, S.
Banerjee, D. Grosz, S. Hunsche, A. Kung, A. Marhelyuk, D. Maywar, M. Movassagbi, X.
Liu, C. Xu, X. Wei, and D. M. Gill, 2.5 Tb/s (64×42.7 Gb/s) transmission over 40 × 100
km NZDSF using RZ-DPSK format and all-Raman-amplified spans. In Postdeadline
Papers of OFC’02, Anaheim, CA, FC2, 2002.
[47] Y. Emori, K. Tanaka, and S. Namiki, 100nm bandwidth flat-gain Raman amplifiers
pumped and gain-equalised by 12-wavelength-channel WDM laser diode unit, IEE Elec-
tron. Lett., 35:16, 1355–1356, 1999.
This page intentionally left blank
Chapter 18
Ultra-Long-Haul, Dense WDM Using
Dispersion-Managed Solitons in an All-Raman System

Linn F. Mollenauer

18.1. All-Optical Transmission

In the late 1990s, the telecommunications industry began to see dense WDM as the
way to provide for the seemingly explosive growth in demand for transmission capac-
ity. The usual industry practice of using electronic regeneration at every node point
(typically, once every 400 to 600 km; see Fig. 18.1), however, promised to use far too
much capital equipment and office space, especially if the net transmission rates were
to be at terabit levels. For example, a system carrying one terabit/s in each direction, at
the practical and increasingly popular per-channel rate of 10 Gbit/s, would require no
less than 200 (expensive and bulky) regenerators, or OT units per node (one for each
direction and channel). In the meantime, it was already well known, principally from
undersea practice, that such dense WDM could be successfully carried out, without
regeneration, over transoceanic distances, at least under the special conditions of the
undersea environment. Thus the idea of developing an all-optical terrestrial system
(which had in fact been advanced many years ago [1–3]) began to take root and to
undergo engineering development by many firms.
The principal requirements for such ultra-long-haul, all-optical systems, are that
the growth of spontaneous emission noise be held to the minimum possible, that the
transmission mode be chosen to yield the fewest nonlinear penalties, and that the
pulse behavior be periodic. The third of these requirements stems from the needs of
optical networking, where the data must be instantly readable, and where standard
pulses can be introduced, at least at all node points along the path. With the additional
requirement to work with existing terrestrial fiber spans of typically 80 to 100 km in
length between amplifier huts, the only way to meet the first requirement is to take
advantage of the greatly reduced noise figure provided by distributed gain from the
Raman effect. The second requirement is best, and the third uniquely met with the
use of dispersion-managed solitons (henceforth abbreviated as DMS).
628 L.F. Mollenauer

= nodes, containing switching and electronic regeneration and electronic regeneration

= opt. amplified fiber spans, ~400-600km long

Fig. 18.1. Schematic of a typical terrestrial transmission system.

18.2. Dispersion-Managed Solitons

18.2.1. Introduction

With dispersion management, the transmission line consists of segments of fiber

whose individual dispersion parameters (Dlocal ) are of alternating algebraic sign and
with absolute values typically at least 3 or 4 ps/nm km (Fig. 18.2). Furthermore, this
arrangement, or dispersion map, is ideally periodic (although, in practice, it need not
be exactly so). For each map period, the accumulated dispersions of the two segments
nearly cancel, so that the path-average dispersion parameter of the map D̄ is small,
typically no greater than about 0.2 ps/nm km. To support solitons, D̄ is also positive
(anomalous dispersion).

Lmap
D loc
+
D
D-MAP: 0
–
D loc
Pulse Width

Pure Disper.
Broadening
(low intensity)
+
chirp

0 0
–
Pulse Width

+ Self Phase Mod.

(Solitions)

0
0 1 2 3
z /Lmap

Fig. 18.2. Dispersion-managed solitons in a nutshell.

 18. Dispersion-Managed Solitons 629

It is instructive first to consider pulse behavior at very low intensities, when only
the dispersive term of the nonlinear Schrödinger (NLS) equation is important. In
response to the relatively large, alternating Dlocal values, the pulse width tends to
undergo a significantly large fractional change, periodic with the map. This pulse
“breathing” is accompanied by a similarly periodic variation in the chirp parameter,
with the chirp passing through zero at or near the center of each fiber segment, and
alternating between large positive and negative peaks. But on a distance scale typically
many times greater than the map period, there is also a gradual net broadening of the
pulses, in response to the effect of D̄ (again, see Fig. 18.2).
Now, to obtain dispersion-managed solitons, we merely need to increase the pulse
intensity until self-phase modulation (SPM, from the nonlinear term in the NLS
equation) produces a phase shift across the pulse that just cancels out, at the end of
each map period, the net phase shift produced by the dispersive term. This cancellation
in turn eliminates the net pulse broadening from D̄, so that the pulse behavior now
becomes truly periodic (the bottom curve of Fig. 18.2). The cancellation of phase
shifts is similar to that obtaining with ordinary solitons (see Fig. 18.3).
Although the pulse field-envelope shape function of the ordinary soliton is sech(t),
that of the dispersion-managed soliton is essentially Gaussian, because in each seg-
ment, the large dispersive term (which scales with Dlocal ) dominates the much smaller
nonlinear term (which scales with D̄), and the solution to the NLS equation, in the
case of pure dispersion, is Gaussian.
The scheme of Fig. 18.2 is very general so that, in principle, one can have any
relation one likes between the map period and the spacing between amplifier huts;
also, so far, the exact fiber types have not been specified. In practice, however, for
terrestrial systems, the typically 80 to 100 km long spans between amplifier huts are
already filled with one or another type of nonzero dispersion-shifted fiber, such as
OFS-Fitel Raman Reduced Slope TrueWave® or Corning LEAF. In that case, the
negative D fiber is a coil of DCF, or dispersion-compensating fiber, with D ≈ −100

 NL

 NL +  Disper.

 Disper.

time

Fig. 18.3. The nonlinear and dispersive phase shifts of an ordinary soliton; note that they sum
to a constant [4]. A similar cancellation of the nonconstant parts occurs at the end of each map
period with dispersion-managed solitons.
630 L.F. Mollenauer

1.0
Mean Frequency Shift, a.u.

0.5

0
0 1 2 3 4 5 6 7 8 9 10
Normalized Pulsewidth, τ/T

Fig. 18.4. Adjacent pulse interaction (a frequency shift resulting from cross-phase modulation,
or XPM) as a function of the pulse width τ , expressed as a fraction of the bit period T . The
path-average dispersion D̄ then converts these frequency shifts into temporal displacements
(timing jitter) [5]. Note that as long as the maximum pulse width is less than approximately
T /2, the adjacent pulse interaction is negligible.

ps/nm km, conveniently located in each amplifier hut. Note that this scheme makes
the dispersion map period essentially the same as the amplifier hut spacing. This
shortest possible map period (for a given amplifier hut spacing) is in general desirable
because it produces the smallest possible extent of the pulse breathing. Limited pulse
breathing is in turn needed to control adjacent pulse interaction [5–7]; see Fig. 18.4
and its caption.

18.2.2. Why Dispersion Management?

Dispersion management was invented to meet certain needs of dense WDM [8–11].
In the first place, four-wave-mixing between adjacent channels (a potentially most
harmful effect, inasmuch as it tends to cause severe amplitude and timing jitter) is
efficiently repressed by the large phase mismatch provided by the large |Dlocal | values
of the individual fiber segments. The phase mismatch is computed as [4]

∂ 2k 2π λ2
k = ω 2
= − D(f )2 ,
∂ω2 c
where f is the frequency separation between the adjacent channels. For example,
consider a system with the parameters Dlocal ≈ 6 ps/nm km and f = 50 GHz (typical
for a 10 Gbit/s per channel system). The above equation then yields k = 0.76/km.
 18. Dispersion-Managed Solitons 631

D (ps/nm-km) S = ∂D /∂λ
TYPE D /S A eff (m)2
(@ 1560 nm) (ps/nm2-km)

low-slope T.W. +5.5 +0.037 148 55

ultra-slope D.C.F. –115 –0.78 147 18

Standard +17.0 +0.056 304 80

IDF –17.7 –0.057 310 35

Fig. 18.5. Parameters for two examples of ±D fiber pairs yielding nearly constant D̄ because
they have nearly identical ratios of D/S (dispersion to dispersion slope parameters) in the
middle of the WDM band. The first is a combination of an experimental version of low-slope
TrueWave® fiber with a matching DCF; the second is standard SMF with a matching inverse
dispersion fiber (IDF). (Note that the second combination would tend to be suitable only for
new installations, however, because the IDF segment would have to be nearly as long as the
standard SMF segment it is compensating.)

Because of this large phase mismatch, the E field of the four-wave-mixing product
spirals rapidly in tight circles in the complex plane, and hence cannot grow to sig-
nificant size. (Note that, in this example, the circle is completed (and hence nearly
closes on itself, especially in Raman-amplified systems) once every 2π/0.76 = 8.2
km.) Second, fibers having the rather small D values needed for ordinary solitons,
constant over the wide wavelength bands required for dense WDM, simply do not
exist, and probably never will. For dispersion management, however, it is possible to
use combinations of fiber for which the path-average dispersion is nearly constant.
Two examples of fiber pairs permitting this extra degree of freedom in map design
are shown in Fig. 18.5.
Figure 18.6 shows the variation in the path-average dispersion parameter typically
obtained from such fiber combinations. Note that the variation in D̄ for a 50 nm wide
band (sufficient for 125 channels at 50 GHz/channel), centered about the peak in the
curve, is only about ±27% of the median value for that band.

18.2.3. Why Dispersion-Managed Solitons?

Note that, thus far, the cited needs of dense WDM to be met by dispersion manage-
ment do not necessarily require dispersion-managed solitons. But when the goal is to
provide the backbone of an all-optical network, then the periodicity of the solitons’
behavior, unique to them, becomes vital. The many important issues surrounding the
periodic nature of soliton transmission can be best discussed in terms of Fig. 18.7.
Although the choice is arbitrary, it is convenient to let the map periods begin and
end at the unchirped pulse positions in each coil of DCF, as in Fig. 18.7. Note that
the accumulated linear dispersion values shown there are the discrete values obtain-
ing at the end of each period, so that they correspond to the product of D̄ and the
particular transmission distance. Note further that the precompensation (pre-comp.)
632 L.F. Mollenauer

0.3

0.25
Dispersion (ps/nm – km)

0.2

0.15

0.1

0.05

0
1530 1540 1550 1560 1570 1580 1590
Wavelength (nm)

Fig. 18.6. Measured path-average dispersion parameter (D̄) for a 100 km span low-slope
TrueWave® fiber, compensated with an ≈4.5 km long coil of matching DCF (see Fig. 18.5),
as a function of wavelength. The data make an excellent fit to a shallow parabola whose peak
is at the center of the intended WDM band.

map period map period map period

DCF DCF DCF DCF DCF "post-comp." coil
contains end of
final period plus
jitter reducer.
"pre-comp." coil is
really integral part n
ersio
of first map period. near disp jitter reducer
ted li ideally = 1/2
mula
accu accum. disper.
effective net dispersion with solitons

Fig. 18.7. Dispersion compensation in a DMS system. In this view, the map periods begin and
end at the unchirped pulse positions in each coil of DCF; the accumulated linear dispersion
values shown here are those discrete values obtaining at the end of each such period. Note that
the pre- and postcompensation coils are really just integral parts of the map periods (see text).

coil is really an integral part of the first map period, and that the postcompensation
(post-comp.) coil, save for an additional jitter-reducer, is likewise an integral part of
the final map period. The jitter-reducer ideally represents a dispersion equal to −1/2
of the accumulated linear dispersion, but the exact value is not at all critical and thus,
in practice, it can be set to the best value for the longest distance to be encountered
in the system. Finally, note that the effective net dispersion for solitons is always
 18. Dispersion-Managed Solitons 633

zero at the end of each map period. This scheme of dispersion compensation has the
following important consequences.
1. The pulse parameters (temporal width, bandwidth, energy, chirp, etc.), are iden-
tical at the end of each map period, and the pulses are always well resolved from
each other in time. This periodic behavior in turn means that:
(a) The data can be read instantly anywhere, or at least at the end of any map
period;
(b) Standard soliton pulses can be injected anywhere (i.e., at the beginning of
any map period).
2. The pre- and postcompensation dispersion values are independent of distance.
These properties are exactly as required for the creation of an all-optical network
and for efficient, inexpensive system monitoring. Their compatibility with the use
of standard parts (pre- and postcompensation coils) is also very important for the
reduction of system cost and for the ease of system assembly. Once again, these
properties are uniquely supplied by dispersion-managed solitons.
It is instructive to look at the dispersion compensation scheme most often used in
non-DMS systems. There it is common to use much greater precompensation disper-
sion, so that the accumulated linear dispersion tends to pass through zero somewhere
near the halfway point of the net transmission distance (see Fig. 18.8). This scheme
represents an attempt to reduce cross-phase modulation (XPM) from interchannel
collisions by greatly broadening the pulses and thus making their peak intensities
lower, over at least most of the path. Unfortunately, however, that action simultane-
ously greatly increases nonlinear penalties from certain intrachannel effects, such as
the adjacent-pulse interaction of Fig. 18.4, and intrachannel four-wave mixing. (The
four-wave-mixing tends to produce ghost pulses in the positions of zeros (bit slots
where there are no pulses) by transferring energy from adjacent ones [5].) Further-
more, in strong contrast to solitons, most of the accumulated linear dispersion is not
compensated by self-phase modulation. In consequence, one has the following.
1. Over much of the path, the pulses are strongly overlapped, so that the data are
not immediately readable.
2. The pre- and postcompensation dispersion values must be carefully tuned for
each distance. (The total of pre- and postdispersion compensation required is
roughly proportional to the total distance.)
3. Even for a fixed distance, dispersion tends to make it impossible to properly
compensate all wavelengths of a wide WDM band with just one set of pre- and
postcompensation coils.

n
persio
linear dis
mu lated post-comp.
accu

pre-comp.

Fig. 18.8. Dispersion compensation in a non-DMS system; see text.

634 L.F. Mollenauer

These facts mitigate strongly against the creation of an all-optical network and effi-
cient, inexpensive system monitoring.

18.2.4. A Shortcut for Computing DMS Behavior

Thus far the discussion of dispersion-managed solitons has been largely qualitative.
For real system design, however, we must compute exact pulse behavior, often for
many different possible dispersion maps, amplifier span gain profiles, and initial
pulse parameters. To do all this computation by exact numerical solution of the NLS
equation is tedious and time consuming. One can create an efficient shortcut, however,
by taking advantage of the fact that, as already stated, in a DMS system, the pulse
shape is Gaussian to a very good approximation. That is, by applying that assumed
pulse shape, or Ansatz, to the NLS equation, one can create an equivalent set of
ordinary differential equations, or ODEs, that are much easier and faster to solve.
Although several other ODE (largely variational [12–14]) approaches have been used,
the special ODE method (nonvariational) invented and used with great success in our
laboratory [15] is briefly outlined here.
We write the (Gaussian) signal pulse in the general form:
√ 1
u(t) = W (η/π )1/4 exp[− (η + iβ)t 2 ], (18.1)
2
√
where 1/ η is a measure of the pulse width, and β is the chirp parameter. Let η0 refer
to the unchirped pulse; that is, η = η0 when β = 0. Clearly, if we know the complex
number η + iβ, and the pulse energy W , we then know all the pulse properties. In
particular, for the pulse width in time, we have:
√ √
1/ η = τ/ 4 ln 2 = τ/1.6651 . . . , (18.2)
where τ is the intensity FWHM. The phase and frequency shifts across the pulse are,
respectively:
φ(t) = − 12 βt 2 (18.3a)

and

δω(t) = −βt. (18.3b)

Finally, the spectrum of the pulse (the Fourier transform ũ(ω) of Eq. (18.1)), yields
the spectral intensity
|ũ|2 ∝ exp[−ηω2 /(η2 + β 2 )], (18.4)
+
+ = (1.6651../2π ) (η + β )/η. Thus the time-bandwidth
which has a FWHM of f 2 2

product is τ f = 0.441 1 + (β/η)2 .

By applying the above Ansatz to the NLS equation, we get the following complex
ODE (really a pair of coupled ODEs) for η and β.
dq
= i[1 − Kq 2 ((1/q))3/2 ], (18.5)
dz
where
q(z, K) = η0 /(η + iβ). (18.6)
 18. Dispersion-Managed Solitons 635

The distance z is always measured in units of the characteristic dispersion length,

which, for a Gaussian pulse, is:

1 2π c τ02
zc = . (18.7)
4 ln 2 λ2 D
The nonlinear coefficient K is calculated as

(2π )2 n2 c τ02 P
K=√ 3
P= , (18.8)
2 4 ln 2 λ Aeff D Pc

where Aeff is the fiber core area, τ0 refers to the unchirped pulse, P is the peak pulse
power, and Pc is the peak power of an ordinary soliton of pulsewidth τ0 in fiber of
the (local) dispersion parameter D. Note that K = 1 corresponds to ordinary solitons
(although the Gaussian pulse shape is not quite right in that case), and that dispersion-
managed solitons tend to correspond to K << 1. Note also that both zc and K are
negative when D is negative. Although this convention and, in particular, the concept
of a negative dispersion length, may seem strange at first, it is self-consistent, and
avoids a certain awkwardness that would occur without it.
If we let z = 0 correspond to the unchirped pulse (η = η0 , β = 0), then q(0, K) = 1.
The solution to Eq. (18.5) then has the general form:

q(z, K) = 1 + iz + K f (z, K). (18.9)

Note that the linear solution q = 1 + iz is the well-known solution for a Gaussian
pulse subject to pure dispersion.
We have developed an efficient computer program to obtain solutions to Eq. (18.5),
based on the Maple mathematics package. From input data consisting of details of the
dispersion map, Raman pumping conditions, and input pulse parameters, our program
first calculates the “gain profile” (relative signal pulse energy versus z), and from
that, K(z). It then uses the Maple program “dsolve” to obtain solutions of Eq. (18.1),
and finally, it graphs the various pulse parameters as functions of z. The program is
very efficient, so that on a reasonably fast PC, one can obtain a full picture of pulse
behavior in a given map in just a matter of minutes. Thus our program has proven to be
a very useful engineering tool, enabling the exploration of system performance over
a wide range of map designs and pulse parameters in a relatively short time. It has
also engendered a deeper understanding of the fundamentals of dispersion-managed
soliton transmission itself. Much of the pulse behavior presented in this chapter was
computed with this program.

18.2.5. Pulse Behavior in Lossless Fiber

Before going on to study pulse behavior in “real” dispersion maps, where loss and
gain tend to complicate matters, it is instructive to survey behavior in lossless fiber.
Accordingly, Figs. 18.9 to 18.11 plot the most important pulse parameters (pulse
width, bandwidth, and chirp) as functions of distance (normalized to |zc |), with the
636 L.F. Mollenauer

K=
–1
6 –0.8
–0.6
D<0 –0.4

Pulse Width (norm.)

5
–0.2

0
4
0.2

3 0.4
D>0
0.6
2
0.8

1 1.0
Ordinary Solitons
0
–4 –3 –2 –1 0 1 2 3 4
Z /| Z c |

Fig. 18.9. Pulse width (normalized to that of the unchirped pulse) in lossless fiber as a function
of distance, for various values of the nonlinear parameter K.

K=
–1.0
1.6
–0.8
1.4 –0.6
D<0 –0.4
1.2
–0.2

0, +1

+0.2
D>0 0.8
+0.4
BW (norm.)

0.6 +0.8
+0.6
0.4

0.2

–4 –3 –2 –1 0 1 2 3 4
Z /| Z c |

Fig. 18.10. Pulse bandwidth (normalized to that of the unchirped pulse) in lossless fiber as a
function of distance, for various values of the nonlinear parameter K.
 18. Dispersion-Managed Solitons 637

–1

0.6
0
0.4
0.6
D<0
0.8 0.2
D>0
K=1
–4 –3 –2 –1 1 2 3 4

Z /| Zc | –0.2

–0.4

–0.6

Fig. 18.11. Chirp parameter (normalized to η0 ) in lossless fiber as a function of distance, for
various values of the nonlinear parameter K. Note that beginning with the unchirped pulse, the
chirp at first increases almost linearly with distance (as the frequency components of the pulse
just begin to separate), then peaks and declines as the separation becomes complete (whence
the range of frequencies is spread out over ever greater time).

nonlinear coefficient K as parameter. Note that for all three parameters, the purely
dispersive effect (i.e., that at K = 0) is always enhanced for D < 0, but tends to be
decreased for D > 0. And, of course, for the ordinary soliton (K = 1), the dispersive
effects disappear altogether.

18.2.6. Pulse Behavior When Loss and Gain Are Included

Even with Raman gain, pulse energies in the 80 to 100 km long spans typical of
terrestrial systems tend to vary by at least several dB along the span; see Fig. 18.12
for graphs of typical behavior.
Figures 18.13 to 18.15 show pulse width, chirp parameter, and pulse band-width,
respectively, as functions of z for the dispersion map and the several pumping condi-
tions of Fig. 18.12. (The color code of Fig. 18.12 is preserved.) The three figures are
worthy of close study. First, note that in each case, the minima of pulse widths and
the extrema (maxima for D > 0, minima for D < 0) of pulse bandwidths correspond
exactly to the positions of zero chirp in all cases. Second, note the asymmetry in the
pulse breathing for the first two cases (red and green curves), and the corresponding
displacements of the zero-chirp positions toward the high-intensity end of the main
span. These asymmetries are brought about by the higher intensity, hence higher non-
linear coefficient K in the first part of each span. (Recall from Fig. 18.9 that higher
K causes shallower breathing in D > 0 fiber.)
638 L.F. Mollenauer

(a)
18

16
Path Average Energies (fJ):
Pulse Energy (dBfJ)

27.7
14 23.9
21.1 (c)

12 (b)

10

0
0 20 40 60 80 100
Distance (km)

Fig. 18.12. Pulse energies as a function of distance in one period of a map consisting of 100
km of TrueWave® Reduced Slope fiber, followed by a matching coil of DCF; in all cases
the unchirped pulse width is ≈30 ps. (a) Main span backward pumped only. (b) Main span
backward and forward pumped in a 50:50 ratio. DCF (the range z > 100 km) backward pumped
for all three cases. (c) Same as curve b, but with mirror reflectivity R = 4%, free spectral range
F SR = 50 GHz, Fabry–Perot guiding filter at the beginning of each map period.

55 (a)

50
Pulse Width (ps)

(b)
45

(c)
40

35

30
0 20 40 60 80 100
Distance (km)

Fig. 18.13. Pulse widths as a function of distance for the dispersion map of Fig. 18.12. The
labels a, b, and c refer to the same pumping conditions as in Fig. 18.12.
 18. Dispersion-Managed Solitons 639

0.4

Chirp Parameter (norm) 0.2

(c)

0 (b)
(a)

–0.2

–0.4

0 20 40 60 80 100
Distance (km)

Fig. 18.14. Chirp as a function of distance for the dispersion map of Fig. 18.12. The labels a,
b, and c refer to the same pumping conditions as in Fig. 18.12.

15.1 (c)

15
BW (GHz)

(a)
14.9

14.8 (b)

14.7

0 20 40 60 80 100
Distance (km)

Fig. 18.15. Pulse bandwidths as a function of distance for the dispersion map of Fig. 18.12.
The labels a, b, and c refer to the same pumping conditions as in Fig. 18.12.
640 L.F. Mollenauer

Note the much smaller fractional changes in bandwidth, as compared with the
changes in pulse width and chirp. This smaller change reflects the fact that the band-
width changes stem entirely from the nonlinear term (which is always much smaller,
locally, than the dispersive term in DM systems).
The effects of the narrow-bandwidth (guiding) filter [4, 16] are also noteworthy
(see the blue curves of Figs. 18.12 to 18.15). First, from Fig. 18.12, note that the
filter significantly raises the path-average pulse energy. The increase is required to
supply the extra SPM that is, in turn, required to restore the pulse bandwidth from the
narrowing effect of the filter. Second, note that the filter tends move the zero-chirp
point back towards the center of the main span, and hence to restore symmetry to
the pulse breathing. The cause here is the discontinuous drop in bandwidth from the
filter at the end of the span (Fig. 18.15), which tilts the curve of bandwidth versus z
upward, thus forcing its maximum (and the corresponding position of zero chirp) to
occur farther along the main span. Both of these effects can be useful when guiding
filters are used. On the other hand, the asymmetries seen here for the cases when no
filter is used (the red and green curves) are not unacceptably large.
Note that Figs. 18.12 to 18.15 correspond to net Raman gains of unity for both
the main and DCF spans. As detailed later, however, to reduce noise from double-
Rayleigh backscattering of the signals themselves, the usual practice is to provide a
few dB less than unity gain in the main span, and to make up for that deficit with
extra Raman gain in the DCF. Nevertheless, the pulse behaviors just shown here are
not greatly affected by that change.
From time to time, someone worries that nonlinear effects in DCF will be exces-
sive, presumably on account of the relatively small core area of DCF. Fortunately,
the worry is unfounded. As far as the nonlinear effects of WDM are concerned, the
very high |D| of DCF renders four-wave-mixing truly negligible, because of the huge
phase mismatch it creates (see Sec. 18.2.2), and the relative velocities of colliding
pulses are so large that the frequency shifts from XPM are almost always much smaller
than those from the +D fiber.
SPM is necessary for the maintenance of dispersion-managed solitons and hence,
in that context, should never be considered as a harmful effect. Nevertheless, the DCF
of a map tends to contribute only a small fraction of the total required SPM. The dom-
inant factor here is that the DCF is relatively short, and that the SPM, everything else
being equal, grows in direct proportion to the propagation distance. If we temporarily
ignore signal power variation with distance, the relative fraction of the total SPM
contributed by the DCF is approximately:

SP MDCF LDCF AD + D + AD +
= = .
SP MD + LD + ADCF DDCF ADCF

For the 100 km span maps referred to here in Figs.18.12 to 18.15, for example,
this fraction is small, no more than about 15%; see Fig. 18.16, where the quantity
q − 1 − iz, or nonlinear residue (see Eq. (18.9)), is displayed in the complex plane.
Note that the real part of that quantity corresponds to the change in inverse squared
 18. Dispersion-Managed Solitons 641

Re

(b)

(c)
(a)
lm

Contributions of the DCF

are the parts between

∆ BW –2 from filter

Fig. 18.16. Nonlinear residue, the quantity q − 1 − iz, for the dispersion map of Fig.18.12. The
labels a, b, and c refer to the same pumping conditions as in Fig. 18.12.

bandwidth, whereas the imaginary part is that which cancels the residual dispersion
[15, 16].
The final but very important issue is that of the path-average pulse energies of the
dispersion-managed solitons. To anticipate a bit from the next section on noise growth,
error-free transmission over transcontinental distances, in a well-designed all-Raman
system, tends to require minimum pulse energies in the 20 to 30 fJ range. Figure 18.17
shows the energies for the systems of Fig. 18.12, as a function of D̄, both with and
without the use of guiding filters. Note that the energy of the dispersion-managed
soliton, for a given combination of pulse width and D̄, is several times greater than
that of the ordinary soliton. This energy-enhancement effect (ρ ≡ WDMS /WOS ) is a
well-known [17–20] and important advantage of dispersion-managed solitons. Note
that for the map in question here, the DMS with the pulse widths most desirable for 10
Gbit/s (≈30 to 32 ps) have adequate energies for D̄ ≈ 0.15 to 0.2 ps/nm km, whereas
the energies of ordinary solitons, for the same parameters, are several times too small.
The energy enhancement is related to the map strength parameter S, defined
as S = (|L+ | + |L− |)/2) ∼ = L+ , where the quantities L are the lengths of the map
segments as measured in units of zc (see Eq. (18.7)). For the case where no guiding
filters are used, ρ ≈ S, as shown by the results of many numerical simulations and as
borne out in Fig. 18.17, where S ∼ = 2.3 for the map it refers to and for a pulse width
of 30 ps. The energy enhancement results from the small but significant change in
pulse bandwidth across the map (see Fig. 18.15), and from the fact that the phase shift
across the pulse caused by the dispersive term scales as the product of Dlocal × BW 2 .
Thus the net dispersive phase shift across the map is larger than would exist without
the change in bandwidth. In turn, then, the compensating nonlinear phase shift and
642 L.F. Mollenauer

70

60

Path Average Pulse Energy (fJ) 50

40
With Filters
30 s
28 p
s
3 p
0 idth
32 s p 0 ps W
20
olitions of 3
s Ord inary S
28p
10
32ps
Without Filters
0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
Path Average Dispersion (ps/nm – km)

Fig. 18.17. Path-average pulse energy as a function of D̄ for the map of Fig. 18.12, and for a
range of unchirped pulse widths. In this graph, the main span is backward/forward pumped in
a 70:30 ratio. The energies of ordinary solitons refer to a map where D = D̄ everywhere.

the pulse energy required to create it must also be larger. When guiding filters are
used, the change in bandwidth across the map (again, see Fig. 18.15) and, hence, the
resulting phase shift from the dispersive term, are even greater.

18.3. Amplifier Spontaneous Emission: Growth and Effects

18.3.1. Calculating the Noise Growth

In the final analysis, the performance of any ultra-long-haul system is limited by the
growth of spontaneous emission noise. Without noise, after all, signal intensities could
be reduced to arbitrarily low levels, where all nonlinear penalties would disappear.
This section briefly outlines a unique and very clear approach, initiated many years
ago [21], to the analysis of noise growth and the problem of selection of signal
intensities that minimize the combined effects of noise and nonlinear penalties. As
that problem is still all too often surrounded by confusion, the tutorial summary given
here is an integral and important part of this chapter on ultra-long-haul transmission.
The basic facts of Raman gain, on the other hand, have been treated many times
over in this volume, so only a brief recap is needed here. The Raman effect in silica-
glass fibers begins with a pump-induced transition to a virtual state, followed by
emission from it, where the emission terminates on an excited state of the lattice;
emission of an optical phonon (which typically takes place within a few femtoseconds)
then completes the return to the ground state. Because of the extremely fast relaxation,
 18. Dispersion-Managed Solitons 643

Ps 1/G G 1/G G 1/G G 1/G G Ps

Trans. ... Detector
N amps

Fig. 18.18. Prototypical ultra-long-haul system, with N amplifiers of power gain G preceded
by N fiber spans of loss factor 1/G.

the population of the terminal state of the optical emission tends to be determined
by equilibrium with the surrounding phonon bath, and hence is almost independent
of the rates of optical pumping and emission. Thus, in contrast to erbium amplifiers,
both the shape of the Raman gain band and the excess spontaneous emission factor
are essentially independent of pump and signal levels. For gain in the neighborhood
of the very broad peak of the Raman gain band, and when the fiber is at or near room
temperature, the excess spontaneous emission factor nsp ∼ = 1.1, or about 0.5 dB.
Nevertheless, Raman gain is prized, more than anything else, for the great noise
reduction enabled by its distributed gain. In order to fully appreciate this matter, we
must begin with a general model of a long-haul system that includes the possibility
of lumped amplification. Figure 18.18 shows the prototypical system.
In the system of Fig. 18.18, each amplifier contributes a noise power per unit
bandwidth, or equipartition energy, of Weq = (G − 1)nsp hν, where hν is the photon
energy. (Note that the power/unit bandwidth has units of energy. Note further that
Weq is the expected value of the noise energy in each mode of the radiation field and
hence corresponds to just one polarization state at a time. Finally, because the unit
time and unit bandwidth can be taken as the bit period and its reciprocal, respectively,
it is often useful to think of Weq as the expected value of the noise energy in each bit
period and in the corresponding bandwidth.)
The gain from the output of each amplifier to the system output (z = Z) is unity,
therefore the noise at the detector is just
αZ
weq = N (G − 1)nsp = (G − 1)nsp , (18.10)
ln G
where α is the ln of the fiber loss rate and where, to simplify appearances, we have
substituted weq = Weq / hν.
For perfectly uniform Raman gain, we can let G → 1 + Z, where Z << 1, let N
become very large, and set nsp = 1.1, so Eq. (18.10) becomes

weq = αZ × 1.1. (18.11)

From Eq. (18.11), we can immediately see that uniform Raman gain provides
the lowest possible noise at the system output. Furthermore, from a comparison with
Eq. (18.10), it is clear that the noise with high-gain lumped amplifiers is much higher.
Consider, for example, amplifiers of 20 dB gain (as would be required for spans of
length approaching 100 km); the noise at system output is then nearly 22 times, or
13.4 dB greater than with uniform Raman gain!
644 L.F. Mollenauer

For injection of Raman pump power every distance L along the path (nonuniform
Raman gain), each dz of path contributes G − 1 = αg (z) dz, so one has
L
1
weq = αZ × 1.1 × αg (z) g(z) dz, (18.12)
αL 0

where g(z) is the net gain from z to L. Although the noise here is intermediate between
that of Eqs. (18.10) and (18.11), as we show, for fiber spans of 100 km or less, the
result is much closer to that of Eq. (18.11).
The range of acceptable signal levels is bounded on the low side by the onset of
significant errors from the inadequate signal-to-noise ratio (SNR), and on the high side
by the onset of significant errors from nonlinear effects. Because the most important
nonlinear effects (primarily SPM and XPM) tend to scale with the path-average signal
power, to facilitate comparison, we should calculate the corresponding path-average
value of the noise. Thus we must multiply the span input noise powers just calculated
by the appropriate ratios of path-average to span input power. For the case of lumped
amplifiers, that path-average factor is:

P̄ 1 L G−1
= exp(−αz) dz = .
P0 L 0 G ln G
Multiplying Eq. (18.10) by the above factor, we get the path-average noise for the
case of lumped amplifiers:
0   1
1 G−1 2
w̄eq = αZ × nsp . (18.13)
G ln G

Note that the path-average noise for lumped amplifiers, although still considerable
for high gains, is nevertheless substantially smaller than the noise at amplifier output
(Eq. (18.10)) for the same gain.
For the case of Raman gain, where the pump power is injected every distance L,
a similar calculation yields:
 L L 
1
w̄eq = αZ × 1.1 × αg (z) g(z) dz × sig.(z)/sig.(0) dz . (18.14)
αL 0 0

In both Eqs. (18.13) and (18.14), the quantities in large {} are the penalty factors,
which can be interpreted equally well as
1. The factor by which the path-average noise increases (over αZ) for constant
path-average signal power, or
2. The factor of increase in path-average signal power (hence, increase in nonlinear
penalties) required to maintain a given SNR.
Although the above expressions for these penalty factors may not be immediately
transparent, they have been evaluated numerically and are plotted in Fig. 18.19 for
lumped amplifiers and for various situations of Raman pumping. Note that although
 18. Dispersion-Managed Solitons 645

Span loss (dB) @ 0.21 dB/km

0 5 10 15 20 25
12

10
)
5 dB
8 .=
.( n.f
Penalty (dB)

mps
e da
6 mp
Lu
ped
L um
4 n+
ma
Ra
ard
ckw an
2 ba d. Ram
75% % fw
100% backward Raman 25 man
fw d. Ra
50%
0
0 20 40 60 80 100 120
Span (km)

Fig. 18.19. Noise penalty factors from Eqs. (18.13) and (18.14), plotted as a function of span
loss. The assumed noise figure for the lumped (erbium) amplifiers is assumed to be 5 dB, a
fairly typical value. All penalty factors have been normalized to the nsp = 1.1 of Raman gain,
so that the Raman penalty curves will begin at 0 dB.

the difference in penalty between the lumped and Raman amplifier curves begins with
the modest difference in their nsp values, and does not change much for span lengths
of just a first few tens of km, eventually it becomes substantial, ≈6 to 8 dB in the
neighborhood (≈100 km) of typical terrestrial amplifier hut spacings. To the extent
that the limits of error-free transmission are governed by the growth of spontaneous
emission noise, this 6 to 8 dB difference represents the factor (four to six times) by
which the maximum transmission distance is increased when all-Raman amplification
is substituted for lumped erbium amplifiers. It was this great increase in reach that first
attracted systems developers to the all-Raman approach. Note also that although the
penalty with purely backward Raman pumping is nearly 3 dB at 100 km, the addition
of a mere 25% of forward pumping cuts that penalty to about a half. Finally, note
that if the ≈100 km spans could be backward pumped at midspan as well as at their
far ends, the noise penalty is reduced to <1 dB, making it almost negligible. Such
midspan pumping would also have the practical advantage of reducing the powers
required of the individual pump lasers by a factor of two [22].

18.3.2. Experimental Test

Recently, we have been able to make an accurate experimental test of the predictions
of Eq. (18.14) by using the recirculating loop shown schematically in Fig. 18.20. As
shown there, the loop consists of six 100 km long spans of TrueWave® Extra Reduced
646 L.F. Mollenauer

×6
TWERS TWERS
DCF 100 km DCF 100 km

Sig.

+2 dB –6 dB 6 dB –6 dB +4 dB
SMF Raman fiber Raman fiber
16.35 km 8.0 km 8.0 km
A.O.Mod.
Gain –3 dB
Equaliz
–2.7 dB

SMF Raman fibers

isol.
isol.
5.0 km 5 km 8.0 km
A.O.Mod.
From To
Xmtr. –2.7 dB Receiver.
+15 - 19 dB

Two-stage, all-Raman Preamp

Fig. 18.20. All-Raman amplified recirculating loop used in test of spontaneous emission noise
growth. The net dispersion of the loop-closing amplifiers is essentially zero. The small rectan-
gular boxes are WDM couplers for the introduction of Raman pump light.

Slope fiber, each span properly compensated by a coil of DCF, with the DCF backward
pumped, and provision made for forward as well as backward pumping of the 100
km spans. (For the experimental results reported here, the forward pumping WDM
couplers were moved to allow the normally forward pumps to be used as midspan
backward pumps.) When the loss of the DCF coils, of the WDM couplers, and of all
the items (acoustooptic modulator, gain equalizer, etc.) used to close the loop on itself
is factored in, the effective loss per 100 km of transmission fiber is about 30 dB. (The
length of fiber in the DCF and loop-closing amplifiers is not included in the reported
transmission distance.)
For the noise measurements, we begin with determination, on a polarization in-
sensitive OSA (optical spectrum analyzer), of the ratio of the spectral intensity of a
10 Gbit/s data stream with the usual half-occupancy of bit periods, to the spectral
intensity of the noise in an adjacent empty channel, and with the OSA’s spectral res-
olution wide enough to completely take in the entire spectrum of the pulse stream.
That raw SNR is then corrected by adding 6 dB to correct to just one polarization
mode and for the unoccupied bit periods, and is further corrected to reflect the noise
in the bandwidth (10 GHz in this case) corresponding to the bit period. (It should be
noted that others often report the raw SNR, but without the listed corrections, and as
measured with somewhat arbitrarily chosen spectral resolution.) Our corrected mea-
surement then yields the fundamental quantity SNR = Wsol /Weq . An independent
determination of the signal pulse energy then lets us compute the noise itself; the
 18. Dispersion-Managed Solitons 647

0.1 200

Equipartition Energy (fJ)

0.08 250

S/N ratio
0.06 333

0.04 500

0.02 1000

0 ∞
0 1 2 3 4 5 6 7 8
Distance (Mm)

Fig. 18.21. Experimentally measured noise as a function of distance for the loop of Fig. 18.20.
The SNR indicated by the vertical scale on the right is based on an assumed signal pulse energy
of 20 fJ.

result (reported as equipartition energy) is plotted in Fig. 18.21 as a function of the

total transmission distance. The slope of the best-fit straight line makes an almost
perfect fit to the prediction of Eq. (18.14), based on the effective loss per 100 km
just cited, and upon the ≈1 dB penalty factor from the midspan pumping. The slight
offset at the origin represents the noise contributed by the transmitter, and from the
preamplifier shown in Fig. 18.20. In the absence of nonlinear penalties, the minimum
SNR required for a BER < 1 × 10−9 is ≈100. Thus the SNR shown in Fig. 18.21 for
8000 km, even if degraded a dB or so for double-Rayleigh backscattering of the signal
(see next section) or modest nonlinear penalties, is more than adequate for error-free
transmission.

18.3.3. Double-Rayleigh Backscattering

One disadvantage with Raman pumping of long spans is that double-Rayleigh

backscattering of the signal can add significantly to the spontaneous emission noise.
The problem arises because large Raman gain near the end of a backward pumped
span amplifies the backscattered signal just as much as it amplifies the signal itself;
thus the double backscattered signal experiences the Raman gain twice. The backscat-
tered signal appears only in active channels, therefore it is not included in the noise
measurement of Fig. 18.21, and so must be accounted for separately. Figure 18.22
shows the calculated per-span NSR (noise-to-signal ratio) from the doubly backscat-
tered signal only, calculated for various Raman gain configurations, and plotted as a
function of span length. The relative importance of the double-Rayleigh backscattered
signal is, of course, in direct proportion to the absolute signal level itself. Consider,
for example, the Weq and the 20 fJ signal pulses of Fig. 18.21. From the NSRs of
Fig. 18.22, we can calculate that for the case of 25/75% forward/backward pumping
648 L.F. Mollenauer

–35
Raman-pumped, % fwd./bkwd: Lossless: NoRaman:
Series1 Series2 Series3 Series4 Series5
0/100 25/75 50/50
–40
Noise/Signal Ratio (dB)

–45

–50

–55

–60
40 50 60 70 80 90 100 110 120
Span (km)

Fig. 18.22. Per-span NSR from a double-Rayleigh backscattered signal pulse train, for the
Raman gain configurations shown, as a function of span length. The assumed fiber core cross-
sectional area is 50 µm2 (appropriate for TrueWave® fiber) and, for the cases with Raman
gain, the spans have been pumped to unity net gain. A 3 dB correction has been made for the
half-occupancy of the bit periods in the original signal.

of 100 km spans, at any distance, the double-Rayleigh backscattering increases the

net noise by about 1/3 (so that it reduces the net SNR to about 3/4 (−1.25 dB) of
the values shown in Fig. 18.21, already a significant reduction. For the case of 100%
backward pumping (again of 100 km spans), on the other hand, the decrease in net
SNR is by a (probably intolerable) −3 dB. From this example, we can see why the
Raman pumping of long spans to complete transparency by back pumping alone is not
generally done. Rather, one either uses a combination of forward/backward pumping,
or excess gain in the following coil of DCF. Finally, note that midspan pumping (of
the same 100 km spans), and insertion of an isolator there as well, tends to make the
double-Rayleigh backscattered signal almost insignificantly small.

18.3.4. The Gordon–Haus effect

The Gordon–Haus effect is a jitter in pulse arrival times, caused by noise-induced

frequency shifts, which are then translated into timing shifts by D̄. For dispersion-
managed as well as for ordinary solitons, the standard deviation σ of the Gordon–Haus
jitter can be computed as
- 2
Weq Z D̄(ps/nm km) 2
σ (ps) = 1.005..(ps nm) Z (km2 ), (18.15)
Wsol L τ (ps)
where Weq is the equipartition energy per span, L is the span length, and τ is the
unchirped pulse width. Equation (18.15) is the original formula of Gordon and Haus
 18. Dispersion-Managed Solitons 649

Magenta
2.5

2
Std. Dev. (ps)

1.5
Blue

0.5

0
0 1 2 3 4 5 6 7 8
Distance (Mm)

Fig. 18.23. Standard deviation of the Gordon–Haus jitter versus distance for a DMS system like
that shown in Fig. 18.20. Magenta curve: Without postdispersion compensation (the full result
from Eq. (18.15)). Blue curve: With optimum postdispersion compensation (half the result of
Eq. (18.15)).

[23] rewritten to display the quantities Wsol and D̄/τ explicitly, as those two quantities
are not rigidly coupled for dispersion-managed as they are for ordinary solitons. (For
ordinary solitons, Wsol ∝ D̄/τ .) Now, due to the energy enhancement effect discussed
earlier, for the same Wsol , the value of D̄/τ is required to be several times larger for
ordinary than for dispersion-managed solitons. Then σ (by virtue of Eq. (18.15)) is
also larger for ordinary solitons by the same factor. Thus, although the Gordon–Haus
effect tends to impose a serious penalty in ultra-long-haul transmission with ordinary
solitons at 10 Gbit/s, it poses much less of a threat for the same with dispersion-
managed solitons, at least when it is the sole source of timing jitter. That is, note from
the data of Fig. 18.23, that the total spread in arrival times out to the 10−9 probability
level (≈13σ ) is a small fraction of the bit period at 10 Gbit/s, especially when the
optimal postdispersion compensation is used. (On the other hand, note that the same
spread is comparable to the bit period at 40 Gbit/s.)

18.4. Dense WDM

18.4.1. Soliton–Soliton Collisions

We now turn our attention to the most serious nonlinear penalty in dense WDM with
OOK (on-off shift keying), viz., the jitter in pulse arrival times created by the collisions
between pulses of different channels. (As noted earlier, with dispersion-managed soli-
tons, those collisions are essentially the only significant source of nonlinear penalty.)
650 L.F. Mollenauer

Briefly, the collisions, through XPM, cause frequency shifts in the colliding solitons,
which are then translated into time shifts by the dispersion of the transmission line.
Because the interacting channels carry essentially random data patterns, each soliton
in a given channel tends to see a different collisional history; hence the appearance
of a related random jitter in pulse arrival times.
With ordinary solitons, pulses in a shorter wavelength channel steadily over-
take and pass through the pulses of a longer wavelength channel. With dispersion-
management, however, the situation is more complicated [24]. In response to the large
and rapidly alternating D values, solitons from different channels race back and forth
with respect to each other in retarded time. Thus collisions between pairs of solitons
tend to consist of fast repeated minicollisions, which individually tend to produce
only small displacements of the pulses in frequency and time. But when the ratio of
local to path-average dispersion is very high (as it usually is), then the colliding pair
tends to undergo a very large number of such minicollisions before the solitons cease
to cross each other’s paths. Thus the net length for an overall collision tends to be
long, typically several thousands of kilometers.
Looked at in greater detail, a net collision tends to consist of several distinct
phases. First, it begins where the pulses tend to achieve maximum overlap at the
junction between the + and −D fibers; note that each of these half-collisions tends to
produce a net frequency shift that is approximately twice as great (when |D+ | ≈ |D− |)
as the peak shift of a collision completed in just one kind of fiber. Thus the net effect
of these half-collisions is to produce a steep wall of rise (or fall) in frequency shift.
The middle part consists of complete collisions that tend to produce only small net
effects, especially when they take place in a region of small intensity gradient. Finally,
at the end, once again we have half-collisions, but this time at the other junction of
the + and −D fibers; these produce a steep decline (or increase) of frequency, back
to zero net shift.
Figures 18.24 and 18.25 illustrate a typical case of the relative motion of a pair
of colliding, dispersion-managed solitons in adjacent channels, and the resultant fre-
quency shift of one of them, respectively.
If we define Lcoll as beginning and ending when the pulses completely overlap
during the half-collisions, a study of Fig. 18.24 reveals that
D+ L+ ∼ τeff
Lcoll ∼
= = , (18.16)
D̄ D̄ λ
where τeff ≡ L+ D+ λ is an effective pulse width for the colliding soliton, λ is
the channel separation, and where, of course, D+ and L+ refer to the +D segment of
the dispersion map. Note that when the specific parameters of the map of Fig. 18.24
are entered into it, Eq. (18.16) yields Lcoll = 3200 km, in excellent agreement with
Fig. 18.25. Also, note that τeff = 192 ps, in accord with the vertical span of the blue
lines in Fig. 18.24.
Although the absolute states of polarization of the colliding solitons tend to evolve
rapidly along the fiber (with major changes typically occurring every few meters),
in low PMD fiber, their relative states of polarization tend to remain fixed over long
distances (many thousands of km); this is especially true for pulses in adjacent chan-
 18. Dispersion-Managed Solitons 651

200
Middle
(Region of Complete Collisions)
Retarded time (ps) 100
End
(Region of Half Collisions)
0
Beginning
(Region of Half Collisions)
–100
Slope = –4.8 ps/80 km
(from D × ∆λ, where ∆λ ∆f = 50 GHz)
–200

0 80 160 240 1520 1600 1680 3120 3200 3280

Distance from beginning of collision (km)

Fig. 18.24. Relative motion in retarded time of a pair of soliton pulses from adjacent chan-
nels separated by 50 GHz; for convenience, the lower-frequency pulse (red line) is fixed in
retarded time; thus the higher-frequency pulse displays the entire relative motion (blue line).
The dispersion map consists of 80 km spans of D = 6 ps/nm km fiber, compensated by a coil
of DCF to yield D̄ = 0.15 ps/nm-km; the channel spacing is 50 GHz. Note the two breaks in
the distance scale, necessitated by the extremely great length of the overall collision.

–.2
Shift (GHz)

–.4

–.6

–.8

–1.0

0 2 4 6 8 10
Distance (Mm)

Fig. 18.25. Frequency shift, as a function of distance, of the lower-frequency pulse of the
colliding pair of dispersion-managed solitons of Fig. 18.24, as determined by exact numerical
solution of the NLS equation. (An equal but opposite frequency shift is induced into the higher-
frequency pulse.) The colliding pulses are orthogonally polarized. Midspan Raman pumping
minimizes signal intensity variation along the main span, in order to make the frequency shift
curve symmetric.
652 L.F. Mollenauer

nels that are initially either co- or orthogonally polarized [25]. The relative state of
polarization is important, because the XPM (and hence all the subsequent collisional
effects) are just half as great when the pulses are orthogonally polarized as when they
are copolarized.
The frequency shifts of the half-collisions are the sum of two shifts: that created
when the pulses come together in the +D fiber, and that created when they subse-
quently back away from each other in the −D fiber. From the known effects of XPM
on colliding pulses in fiber of constant D, we can then obtain:
 
n2 1 a Wsol
δfhalf–coll = Cpol + , (18.17)
Aeff λ D+ |D− | λ τ

where the polarization coefficient Cpol = 1/2 (1) when the colliding solitons are
orthogonally (co-) polarized, where Aeff is the effective area of the +D fiber, and
where a is the ratio of Aeff to the core area of the −D fiber. (Note that when the −D
fiber is DCF, the major contribution comes from the +D fiber.) The net frequency
shift of the overall collision can then be obtained by multiplying Eq. (18.17) by the
effective number of half-collisions:
 
n2 1 a Wsol 2τ
δf = Cpol + × . (18.18a)
Aeff λ D+ |D− | λ τ Lmap D̄ λ

For Aeff = 50 µm2 , λ = 1550 nm, and expressing Wsol in fJ and the other quantities
in the usual units of ps/nm km, nm, and ps, respectively, Eq. (18.18a) becomes
 
1 a 2Wsol
δf (GHz) = ±0.335 Cpol + . (18.18b)
D+ |D− | Lmap D̄ (λ)2

If we put the particular parameters (including Cpol = 0.5) for the collision of Fig. 18.25
into Eq. (18.18b), we get δf = 0.74 GHz, in good agreement with the exact solution
as shown by Fig. 18.25.
The time shift associated with the collision scales as the area under the frequency
shift curve, such as that of Fig. 18.25. Accordingly, the time shift can be roughly
estimated as
δt ≈ D̄ δλ Lcoll , (18.19)
where δλ = −(λ2 /c) δf is the wavelength shift corresponding to δf . (Strictly speak-
ing, the area of the sharp “needles” representing the complete minicollisions should
be multiplied by the appropriate Dlocal , and not by D̄. The error caused by the simplifi-
cation in Eq. (18.19) is usually not great, however.) For the parameters of the collision
of Fig. 18.25, Eq. (18.19) yields δt ≈ 2.5 ps. (Again, the result is for orthogonally
polarized colliding pulses; δt is twice as great when the pulses are copolarized.)
It is extremely important, as we discuss shortly, that the XPM- induced frequency
shifts scale as the inverse square of the channel separation (λ). This scaling is
nicely borne out in exact numerical simulation. For example, Fig. 18.26 shows the
same collision as Fig. 18.25, but for a three times greater channel separation. Note
 18. Dispersion-Managed Solitons 653

Shift (GHz) –.04

–.08

–.12

–.16

–.2
0 2 4 6 8 10
Distance (Mm)

Fig. 18.26. Frequency shift, as a function of distance, for a collision like that of Fig. 18.25,
except that now the channel spacing is 150 GHz.

that the peak frequency shift is indeed nine times smaller. Note further, however, that
except for even steeper walls, the collision has about the same shape and essentially
the same width as with the smaller channel spacing.
It may happen that, at or near the transmitter, the colliding solitons are on top of
each other in the middle of a span, so that the overall collision begins near the middle
of a complete collision (see Fig. 18.27). Such half-collisions can produce even greater
time shifts, inasmuch as they produce a large residual frequency shift that does not
disappear. In that case, the length factor in Eq. (18.19) is not Lcoll , but nearly the
entire distance of the transmission. Also note that the algebraic sign of the frequency
shift (and hence that of the corresponding time shift), is opposite to those produced
by the complete collision.
Finally, we note that the more usual distributions of signal intensity with distance
over the span (see Fig. 18.12) tend to introduce a certain asymmetry into the curves of
the collision-induced frequency shift (see Fig. 18.28). We note, however, that the area
under the curve does not seem to change significantly, as we have verified through
many numerical simulations.

18.4.2. Calculating the Jitter

Thus far we have discussed the collision interaction of just two solitons. In the course
of a long-haul transmission, however, each pulse of a given channel suffers many
collisions with the pulse trains of many other channels. Thus, at first, calculating the
net timing jitter from collisions would seem to be a daunting task. Nevertheless, there
are a number of factors that make the influence of all but the very closest channels of
very rapidly diminishing importance. The first of these is the fact that the maximum
possible frequency shift from simultaneous interaction with many channels is not
654 L.F. Mollenauer

1.0

0.8
Shift (GHz)

0.6

0.4

0.2

0 2 4 6 8 10
Distance (Mm)

Fig. 18.27. Frequency shift, as a function of distance, of the half-collision version of the
collision of Fig. 18.25. Note that the time shift is now proportional to the area underneath the
tail of the curve.

–.2
shift (GHz)

–.4

–.6

–.8

–1.0

0 2 4 6 8 10
Distance (Mm)

Fig. 18.28. Frequency shift, as a function of distance, of the collision like that of Fig. 18.25,
except that in this case, the lack of midspan pumping produces a considerably greater variation
of signal intensity across the main span.

much greater than that from interaction with just the nearest channel. This is because,
as already

∞ noted, the frequency shifts from collisions fall off as (λ)−2 , and the fact
−2
that n=1 n = π 2 /6 = 1.64 . . . . Thus, if we consider simultaneous collision with
just one bit from each channel, the maximum frequency shift will be no more than
about 64% greater than that from collision with just the adjacent channel. The second
is that collisions with channels of lower frequency produce effects (frequency and
 18. Dispersion-Managed Solitons 655

0 50 100 150 200 0 50 100 150 200

Time (ps) Time (ps)
(a) (b)

Fig. 18.29. Numerically simulated electrical eye diagrams after 8000 km transmission through
a system like that of Fig. 18.20, for the worst of eight channels. ASE is not included; the only
effect is that of the collisions. A different, 27 bit random pattern was used for each channel.
(a) The unaltered eye; (b) the eye with timing jitter artificially removed.

time shifts) of opposite sign from those of higher frequency. Therefore, the effects of
the lower and upper frequency sets of channels tend to cancel each other out.
Nevertheless, the minimum distance between subsequent collisions with pulses
from a given channel, lcoll , is:
T
lcoll ∼
= , (18.20)
D̄ λ
where T is the bit period. Because Eq. (18.20) implies that the number of collisions
increases in direct proportion to the channel spacing, one might at first think that
the

Nnet effect of collisions with the pulses of many channels would then fall off as
n −1 ≈ ln N, where N is the number of interacting channels. But as λ becomes
n=1
large, with high probability, the number of colliding pulses from the interfering chan-
nel will remain close to half the maximum possible number. To make a long story
short, statistics saves the day, so the net effect of collisions with many channels falls
off much faster than ln N [26]. Therefore numerical simulations involving a modest
number of channels (typically, about eight) tend to yield a rather accurate picture of
the collision-induced timing jitter. Figure 18.29 shows some results of such a simu-
lation, the eye diagram of the worst-behaved (one in the middle) of eight channels
after 8000 km transmission through a system like that of the loop of Fig. 18.20. Note
from Fig. 18.29 that artificial removal of the timing jitter almost completely opens the
eye (there remains only a very small residual closure from modest amplitude jitter).
Thus we can conclude that the rather severely penalizing closure of the unaltered eye
does indeed stem almost entirely from the collision-induced timing jitter. There is
one more important fact about this timing jitter, namely, that its statistical distribution
tends to be strongly bounded (see Fig. 18.30). There are two reasons for the bounded
nature of this distribution: as already discussed, the major damage is done by the
immediately adjacent channels, and the maximum possible number of collisions with
those adjacent channels is small. (For an adjacent channel spacing of 50 GHz and a
path-average dispersion parameter of 0.15 ps/nm km, the minimum spacing between
collisions (lcoll , Eq. (18.20)) is ≈1700 km.)
656 L.F. Mollenauer

12 50

Spread of Arrival
40

Times (ps)
Frequency
8
30
4 20
10
0 0
–30 –20 –10 0 10 20 30 0 2 4 6 8 10
Timing Jitter (ps) Distance (Mm)
(a) (b)

Fig. 18.30. Statistics of timing jitter in a numerically simulated version of the dense WDM
system of Fig. 18.20; no ASE; adjacent channels orthogonally polarized. (a) Statistical distri-
bution of pulse arrival times from the eye diagram of Fig. 18.29; (b) maximum spread in pulse
arrival times versus distance.

18.4.3. The Temporal Lens as Jitter Killer

We have just seen that the jitter in pulse arrival times tends to be bounded. Of course, in
a real system with ASE noise, Gordon–Haus jitter (which has a Gaussian distribution)
adds to the nonlinear jitter. Fortunately, in dispersion-managed soliton systems, as
discussed earlier, the Gordon–Haus jitter tends to be small. By combining the results
shown in Figs. 18.23 and 18.30, one can infer that the probability that the net jitter
will be greater than ±1–3 of a bit period is less than 1 × 10−9 up to distances of at
least 8000 km. This fact enables one to overcome the penalty from timing jitter by
using a simple device [27–29] just prior to the receiver to remove the jitter itself.
As can be seen from Fig. 18.31, the device itself is very straightforward, consisting
of nothing more than a phase modulator, appropriately driven in synchronism with
the locally recovered clock, and followed by an essentially linear dispersive element
(a coil of fiber or a fiber Bragg grating). The phase modulator gives each incoming
pulse a frequency shift proportional to its jitter displacement, and the dispersive
element then serves to translate, or “focus” each incoming pulse on to the mean
arrival time (modulo the bit period); hence we call the device a “temporal lens.” To
the extent that it can accomplish such translation for all incoming pulses, the temporal
lens completely removes the eye closure, and hence there is no longer a nonlinear
penalty. (It should be noted that in this basic principle of operation, at least, the
temporal lens of Fig. 18.31 is very similar to a device described earlier [30]. We note
further, however, that in [30], the device was apparently intended solely for use as a
compensator of higher-order PMD, and not as a jitter-killer.) It is also important to
note that the clock recovery must come before the phase modulator. (Thus, in a real
system, the jitter-killer must have its own clock, independent of the one located at
the detector, to avoid problems of varying delay with temperature in the dispersive
element.)
In the ideal mode of operation, the phase shift φ(t) produced by the modulator is
a series of truncated parabolas, centered about the middle of each bit period, such that
the corresponding frequency shift (the time derivative of φ(t)) is directly proportional
to time as measured from the center of each period (see Fig. 18.32). Nevertheless,
 18. Dispersion-Managed Solitons 657

Dispersive element
(fiber or Bragg grating)

Unjittered pulses
Jittered pulses In out to receiver
phase mod.

M.W. amp.
& pulse shaper
Locally recovered clk.

Fig. 18.31. Basic scheme of the temporal lens. The phase modulator, driven by the locally re-
covered clock, shifts the frequencies of the incoming pulses in proportion to their temporal dis-
placements; the dispersive element then converts these frequency shifts into jitter-compensating
time shifts (see text).

(t ) Produced by Modulator

Early Pulse Late Pulse

–20 0 20 40 60 80 100 120

Time (ps)

Frequency Shift

Fig. 18.32. Temporal details of the ideal mode of operation, showing: parabolic phase shift.
(saw-toothed) curve: resultant frequency shift induced on pulses, and pulses that have arrived
25 ps early and late, and which are shifted to the center of the period by the temporal lens.

even a simple sinusoidal drive of the modulator works well, at least over the middle
half of the bit period. In the following experiments, to further increase the lens’s
capture range, we have used a sinusoidal drive, but as modified by the addition of an
appropriate amount of second harmonic.
Any device used at the receiver needs to have an essentially polarization-
independent response. Unfortunately, however, most modulators, and in particular
those made of LiNbO3 , are very strongly polarization-dependent. The most straight-
forward way around this problem is to use a polarization diversity scheme. The one
658 L.F. Mollenauer

Polariz. splitter/combiners Dispersive element

(fiber or Bragg grating)
Jittered
pulses In Unjittered pulses
phase mod. out to receiver

phase mod.

M.W. amp.
& pulse shaper
Locally recovered clk.

Fig. 18.33. Polarization diversity scheme for the temporal lens. The first polarization splitter
separates the input of arbitrary polarization into its two, orthogonal, linearly polarized compo-
nents. Each of those components is then sent through its own phase modulator. The outputs of
the two phase modulators are then combined by a second polarization combiner before being
sent on to the dispersive fiber. Note that the input and output (polarization-maintaining) fiber
leads, as well as the coaxial cables to the modulators, must constitute three carefully matched
pairs in length.

we have used in our experiments (see Fig. 18.33) performed extremely well, as the
polarization-dependence of the device was nearly immeasurable.

18.4.4. Predicted System Performance from Numerical Studies

We have numerically simulated the performance of the system of Fig. 18.20, backward
Raman pumped every 50 km, and with D̄ = 0.15 ps/nm km. All simulations involved
eight channels at 10 Gbit/s each, and with adjacent channel separations of 50 GHz.
The predicted BER performance is shown in Fig. 18.34. Note that an ideal integrate
and dump receiver can do the best job of all. With the extensive development of
ultra-high-speed electronics for 40 Gbit/s now taking place, it may indeed be possible
to make a viable integrate and dump detector for 10 Gbit/s; if so, that would be the
simplest and cheapest solution.

18.4.5. Experimental Tests of Dense WDM Using DMS

To date, we have made a great many tests of ultra-long-haul, dense WDM at 10 Gbit/s
per channel using DMS. For all tests, the transmitter contained two sets of up to 80
DFB lasers, the lasers of each set on a grid of 100 GHz spacing, with their combined
outputs fed through a common, LiNbO3 , MZ modulator-based pulse carver, and then
through a second similar modulator used to impose data. The pulse carvers were
driven sinusoidally at 5 GHz, and biased to yield two pulses per cycle in a very good
approximation to Gaussian pulses of 33 ps FWHM. Each data modulator was driven
with its own independent, typically 215 bit long, pseudorandom pattern. Finally, the
 18. Dispersion-Managed Solitons 659

30
28 w/o Jitter Killer
w Jitter Killer, sin PM
26 w Jitter Killer, parabolic PM
Integrate-and-Dump Receiver
24

Q 2 (dB)
22
20
6400 km
18 6800 km
16
4400 km
14
5600 km
12
0 1 2 3 4 5 6 7 8
Distance (Mm)
(a)

30
28 w/o Jitter Killer
w Jitter Killer, sin PM
26 w Jitter Killer, parabolic PM
Integrate-and-Dump Receiver
24
Q2 (dB)

22
20 9200 km
9500 km
18
16
6100 km
14 8100 km
12
0 1 2 3 4 5 6 7 8
Distance (Mm)
(b)

Fig. 18.34. Predicted BER performance of dense WDM system with and without jitter-killer,
and with ideal integrate and dump detector: (a) all channels copolarized; (b) adjacent channels
orthogonally polarized. (Q2 = 16 should yield BER = 1 × 10−10 .)

two sets of channels, with their frequencies offset from each other by 50 GHz, were
brought together with a polarization combiner to yield an array of as many as 160
channels, with adjacent channels of orthogonal polarization.
The latest round of tests have involved the 600 km long recirculating loop shown
in Fig. 18.20, both with and without the aid of the temporal lens jitter killer. Pre-
and postdispersion compensation was carried out as discussed in Section 18.2.3; for
details of the precompensation, see Fig. 18.20. About half of the final DCF coil, plus
the SMF and 8 km of Raman fiber (D = −18 ps/nm km) shown in Fig. 18.20 form
the major part of the postcompensation; the remainder was in one or another Raman
pumped coil of DCF just ahead of the receiver (and ahead of the temporal lens jitter
killer, when used). Thus the “jitter-reducer” part of the postdispersion compensation
660 L.F. Mollenauer

(a) (b)

Fig. 18.35. Eye diagrams at 7200 km of one channel of a dense WDM transmission at 10 Gbit/s
per channel and 50 GHz channel separation, with orthogonally polarized adjacent channels:
(a) without jitter killer; (b) with jitter killer, driven by fundamental second harmonic combi-
nation. Although the directly measured log BER for the situation in (a) was in the −4 to −5
range, that corresponding to (b) was nearly error-free (log BER −9). A histogram of the eye
in (b) implies a Q2 ∼ = 19 dB, however, corresponding to an even better BER by several more
orders of magnitude; the discrepancy is also typical.

could be adjusted, but was usually at the optimum value for a distance of about 5000
km; once again, however, that value was not at all critical.
Tests involving the jitter killer were carried out with both a purely sinusoidal
drive, and later with the drive involving the addition of the second harmonic. Even
in the very first tests, with the purely sinusoidal drive, at long distances (Z > 5000
km), where the BER performance was significantly compromised by eye closure from
timing jitter, the BER performance would typically be improved from values in the log
BER = −4 to −6 range to “error-free,” that is, log BER # −9. Figure 18.35 shows a
typical result, eye diagrams at 7200 km, with and without the temporal lens jitter killer;
Fig. 18.36 shows BER versus distance, both with and without the jitter killer. From
evidence such as that shown in Figs. 18.35 and 18.36, it is clear that the jitter killer
consistently makes a substantial improvement in the BER performance of our dense
WDM system at large distances. Nevertheless, several modest discrepancies remain.
In the first place, note that the actual performance is still short of that predicted by the
numerical simulations (compare the results of Fig. 18.36, where the distance, with
jitter killer, for log BER = −10, is about 6800 km, with the graph of Fig. 18.34(b),
where the corresponding distance is almost 9000 km). The discrepancy most probably
has to do with an inadequately wide capture range of the temporal lens. That is, with
probability much too small to be seen in an eye diagram, yet large enough to be
significant relative to the 10−9 level, a few pulses are jittered outside the capture
range.
It should be noted that FEC (forward error correction) was never used in the
experiments. On the other hand, FEC is used by virtually all systems builders, because
it can, at very low additional cost, render error-free transmission that would otherwise
have error rates as great as ≈10−3 . Nevertheless, once the uncorrected error rate passes
a certain threshold, the FEC tends to completely lose control. Thus FEC can mask the
 18. Dispersion-Managed Solitons 661

–3

–4

–5

–6
log10 BER

–7

–8

–9

–10

–11

–12
5 5.5 6 6.5 7 7.5 8 8.5 9
Distance (Mm)

Fig. 18.36. Measured BER versus distance for: magenta points, with optimum postcompen-
sation coil only; blue points, with jitter killer as in Fig. 18.35; no FEC was used for either
case.

existence of dangerously thin margins, and in so doing, has no place in fundamental

investigations, such as those reported here, of the intrinsic system quality. However,
by the standard often employed today, where the performance of a system is defined
by what can be achieved with FEC, the system we have just described would have a
reach of ≈9 mm without the jitter killer, and ≈12 mm with it!

18.4.6. Test of a DMS System with Guiding Filters

Several years ago, we built a dispersion-managed recirculating loop containing guid-
ing filters [31]. Although it was originally built as a model for a transoceanic system,
it is now mostly of scientific interest as a test of the performance of an ultra-long-haul
DMS system using guiding filters. Guiding filters tend to reduce frequency jitter and,
consequently, they reduce timing jitter. They have the disadvantage, however, that
although they act to directly reduce amplitude jitter, they also act to translate fre-
quency jitter into more amplitude jitter. With ordinary solitons, because of the direct
proportionality between pulse energy and pulse bandwidth, the guiding filters reduce
both the frequency and amplitude jitter with great efficiency, so there is a net reduc-
tion of amplitude jitter as well. With DMS the pulse energy is typically several times
less dependent on the pulse bandwidth. Then while the guiding filters are reducing
frequency and timing jitter, they may actually cause a net increase in amplitude jitter.
Nevertheless, they can still provide a net benefit, as our experiments showed.
Figure 18.37 is a schematic of the dispersion map; as can be seen there, the am-
plification was a hybrid system, using backward Raman pumping to overcome about
662 L.F. Mollenauer

Raman
pump in
guid. filter (~1450 nm)

STD. DCF R. S. True Wave STD.

km
0 10 20 30 40 50
0
Rel. Sig. (dB)

–3

–6

Fig. 18.37. Dispersion map and intensity profile for the transoceanic experiment.

0.25 0.25

0.2 0.2

0.15 0.15

0.1 0.1

0.05 0.05

0 0
1535 1540 1545 1550 1555 1560

Fig. 18.38. Measured D̄ for the dispersion map of Fig. 18.37.

75% of the span loss, and a low gain, C-band erbium fiber amplifier to compensate the
rest. The path-average dispersion of the map was nearly constant over the wavelength
span of the C-band (see Fig. 18.38).
Sliding the guiding filter frequencies at the modest rate of −2.7 GHz/Mm pro-
duced the best compromise between reduction of noise growth and reduction of
frequency and timing jitter. Figure 18.39 shows a typical eye diagram at 9 Mm, and
Fig. 18.40 plots the measured BER at 9 Mm as a function of channel wavelength for
every other one out of the 27 channel wavelengths. The adjacent channel spacing,
determined by the free spectral range of the guiding filters, was 75 GHz (≈0.6 nm),
so the 27 channels spanned a good fraction of the entire C-band.
 18. Dispersion-Managed Solitons 663

Fig. 18.39. Typical eye diagram at 9072 km for one channel in a transmission involving the
setup of Fig. 18.37. The measured BER in this case is 3.7 × 10−10 .

–7 –7

–8 14 of 27 Channels Measured –8
(Every Other Channel)
log10BER

–9 –9

–10 –10

–11 –11
1542 1545 1548 1551 1554 1557 1560
Wavelength (nm)

Fig. 18.40. Measured BER versus wavelength at 9072 km using the setup of Fig. 18.37.

18.4.7. Novel Technique Using Periodic-Group-Delay-Complemented

Dispersion Compensation for Major Improvement of Dense WDM

We have just seen, both through theoretical argument and experimental verification,
that collision-induced timing jitter is the principal limiting nonlinear penalty in dense
WDM with dispersion-managed solitons, at least when the dispersion compensa-
tion involves only fiber. Very recently, however, in our laboratory, we have come
to realize that when a certain modest fraction of the transmission span’s dispersion
is compensated by one or another of the recently developed periodic-group-delay
664 L.F. Mollenauer

50

Group Delay (ps) 40

30

20

10
1569.5 1570.0 1570.5
Wavelength (nm)

Fig. 18.41. Measured group delay of a Gires–Tournois etalon-based DCM, with channel spacing
of 50 GHz (0.4 nm). Note that the mean group delays for each channel are the same. Also note
the lack of detectable delay ripple in the usable frequency regions. Insertion loss = 2.6 dB.
This extremely stable and robust device was made by Avanex, Inc. of Freemont, CA.

dispersion-compensating modules (PGD-DCMs) [32–35] and the remaining fraction

is still compensated by DCF or other fiber, that penalty is reduced by a very large
factor [36]. The PGD-DCMs have been developed primarily with an eye to reducing
cost and insertion losses, and to improving the consistency with which the disper-
sion of transmission fibers can be compensated over wide WDM bands, yet it is the
fact that their group delays are periodic and with period equal to the WDM channel
spacing that is of major interest here; see Fig. 18.41 for typical behavior.
Our analysis begins with a study of the relative motion in retarded time of a pair
of colliding pulses, similar to that we have already made in Fig. 18.24 and following.
Now, however, we must consider the more general case, where a fraction f of the
span dispersion is compensated by a PGD device, and the remainder is compensated
by fiber; see Fig. 18.42.
Note from Fig. 18.42 that the mean inverse group velocity with which the pulses
move with respect to each other (the slope of the sawtooth) ranges from the very
small value D̄λ at f = 0 to the many times greater value D + λ at f = 1, where
D̄ is the path-average dispersion, D + refers to the transmission span, and λ is the
channel spacing. Thus Lcoll decreases very rapidly as f increases.
Our formula for the collision length (Eq. (18.16)) must now be modified. From a
study of Fig. 18.42, one can easily deduce that
2τ + τeff
Lcoll = , (18.21)
D̄inter λ
 18. Dispersion-Managed Solitons 665

For f = 0, slope = D ∆λ

200

Slope ~ f D + ∆λ
100
Retarded time (ps)

Pat
h
for
–100 f=1

–200
coll. length

0 100 200 300 400 500 600

Distance from beginning of collision (km)

Fig. 18.42. Relative motion in retarded time of a pair of colliding pulses from neighboring
channels separated by 50 GHz, where 100 km spans of D = 6 ps/nm km fiber, save for a small
D̄, are fractionally compensated f by a PGD device, and (1 − f ) by a DCF module. (For the
particular behavior shown here, f = 0.2.)

where the effective pulse width

τeff = (1−f )(D + − D̄)L+ λ (18.22)

still equals the (now reduced) recovery time in the DCF, and where D̄inter = f (D + −
D̄) + D̄ is the inter-channel D̄ associated with the slope of the sawtooth of Fig. 18.42.
Our formula for the minimum possible spacing between successive collisions with
pulses of a given channel (Eq. (18.20)), must also be modified by substituting D̄inter
for D̄ in the denominator; thus we now have
T
lcoll = . (18.23)
D̄inter λ
The rapid decline in Lcoll as f is increased from zero Eq. (18.21) has profound
effects, as can be clearly seen from the curves of collision-induced frequency shifts
shown in Fig. 18.43. First, note from Fig. 18.43 that for f = 0 and f = 0.2, the fre-
quency shifts ultimately return to zero. The most striking feature of these “frequency-
conserving” collisions is the great shrinkage in all measures of the collision size as
f goes from 0 to 0.2: first, in accord with the prediction of Eq. (18.21), Lcoll is re-
duced by a factor of nearly 10 times, and the peak frequency shift is reduced by a
similar factor. From the same numerical simulations, we find that the corresponding
time displacements are 4.62 and 0.07 ps, respectively, for a 66-fold reduction. More
666 L.F. Mollenauer

–0.2

–0.4 f = 0.8 f = 0.2

Frequency Shift (GHz)
–0.6

–0.8
f=0
–1

–1.2

–1.4

–1.6

–1.8
0 2 4 6 8
Distance

Fig. 18.43. Frequency shifts of the lower frequency of two colliding pulses from adjacent
channels in the system of Fig. 18.2, for the indicated values of f ; in all cases the channels are
copolarized, the effective core area = 50 µm2 , D̄ = 0.15 ps/nm km, and the spans are backward
Raman pumped only.

generally, we find that the time shifts scale approximately as L1.85 coll as f increases
(and Lcoll decreases) throughout the region of frequency-conserving collisions.
We also find that as f becomes significantly greater than about 0.2 (so that Lcoll
ceases to be at least several span lengths long), the collisions are in general no longer
frequency-conserving, as illustrated in Fig. 18.43 for the particular case f = 0.8. The
situation is similar to that discovered many years ago for dense WDM with ordi-
nary solitons, and is principally associated with the tendency of intensity gradients to
destroy the symmetry of the collisions [37]. The effect is to be avoided, because other-
wise the residual frequency shifts, when compounded with the dispersion remaining
at the end of the transmission, can once again produce large time shifts.
With respect to the timing jitter itself, we are primarily concerned here with
the way it scales down as f increases from zero (or as Lcoll decreases), until the
collisions become no longer frequency-conserving. As noted earlier, the spread in
arrival times scales as the product of the time-shift per collision, times the number
of collisions in the transmission. the latter scaling as 1/ lcoll . From Eqs. (18.23) and
(18.21), respectively, we can see that for small f , both lcoll and Lcoll scale in nearly the
same way with f . Combining that fact with our above-given estimate for the scaling
of the per-collision time shifts, we conclude that the timing jitter should scale down
(for a given channel spacing) as about L0.85
coll . Thus inasmuch as we have already seen
that Lcoll can be reduced by a factor of ≈10 times within the region of frequency-
 18. Dispersion-Managed Solitons 667

f=0 f = 0.2 f=1

0 50 100 150 200 0 50 100 150 200 0 50 100 150 200

Time (ps) Time (ps) Time (ps)

0 50 100 150 200 0 50 100 150 200 0 50 100 150 200

Time (ps) Time (ps) Time (ps)

Fig. 18.44. Optical eye diagrams at 8000 km in dense WDM with 10 Gbit/s channels spaced
50 GHz apart, with all channels copolarized, and with no ASE, for the values of f shown. Top
row: lossless fiber; bottom row: with 25/75% forward/backward Raman pumping. All other
parameters are the same as those of the system of Figs. 18.2 and 18.3.

conserving collisions, we anticipate a similar reduction (≈7 times) in the timing jitter
itself. This prediction is well borne out both by numerical simulations and real-world
experiment.
The simulations are for 100 km spans of fiber with D = +6 ps/nm km, dispersion-
compensated with various combinations of DCF and PGD-DCMs. Other details were
essentially as earlier in this chapter (unchirped pulse width τ = 33 ps, channel spacing
50 GHz, etc). Figure 18.44 shows a representative set of results from those many
simulations, the eye diagrams as seen after 8000 km. As is immediately obvious from
the figure, the quality of the eyes changes dramatically as f is varied. Note in particular
that although the eyes are more or less uniformly bad at f = 0 (100% compensation by
DCF), independently of the Raman pumping conditions, at f = 0.2 they are excellent
for the case of uniform intensity and still very good for 25/75% forward/backward
Raman pumping. (Also please note that here the adjacent channels are copolarized,
yielding a factor of 2 times greater jitter than the orthogonally polarized case of
Fig. 18.29.) The improvement here over the behavior for f = 0 corresponds rather
well to the predictions of the simple scaling theory given earlier. Finally, note that in
the f = 1 column (100% compensation by the PGD-DCM), the behavior is also as
expected from the different intensity gradients represented there.
From simulations we have done with many other values of f , we observe that the
timing jitter tends to remain very small over a rather wide range of values of f , and
that the position of the absolute minimum depends somewhat on the exact profile of
intensity over the spans. The tolerance for considerable variation in f is important,
as it provides room to use the PGD-DCMs for complete correction of both slope
and curvature of the span dispersion. (The etalon-based PGD-DCMs are particularly
amenable to making such sophisticated corrections of the dispersion.)
668 L.F. Mollenauer

18.4.8. PGD-Complemented Dispersion Compensation in Dense WDM:

Experimental Test

Just shortly before this book went to press, we began testing the PGD-complemented
dispersion-compensation technique in dense WDM experimentally. Our tests in-
volved the all-Raman amplified recirculating loop shown in Fig. 18.20, but with
the six 100 km spans of that loop (D = 7 ps/nm km at 1575 nm), compensated ≈79%
by DCF and most of the remainder by the PGD modules referred to in Fig. 18.41. The
intrachannel path-average dispersion parameter D̄ was still 0.15 ps/nm km. To reduce
the noise penalty to the absolute minimum, the transmission spans were backward
Raman pumped every 50 km, to just a few dB less than net zero gain; the excess loss
was made up by net gain in the backward Raman pumped DCF modules. A mild etalon
guiding filter (with free spectral range = the channel spacing) was used once every
300 km to overcome the effects of a weak adjacent pulse interaction, and to reduce
the Gordon–Haus jitter (both intrachannel effects). The pre- and post-compensations
were just as before for the experiments of Section 18.4.5, as was the transmitter. We
used the temporal lens just ahead of the detector to remove any residual timing jitter
from the collisions and from the Gordon–Haus effect. (It should be noted, however,
that for the error-free distances achieved here, the net timing jitter would have been
far outside the capture range of the temporal lens without the PGD-complemented
dispersion compensation.) Finally, we inserted a polarization scrambler, operating at
about 1 MHz, into the recirculating loop, to more realistically simulate the random
PMD and PDL effects of a real system.
Figure 18.45 shows the measured BER versus distance for a typical WDM chan-
nel. First, note that the new PGD-complemented dispersion compensation technique

–3

–4

–5
log (BER)

–6

–7

–8

–9

–10
0 2 4 6 8 10 12 14 16 18 20 22
Distance (Mm)

Fig. 18.45. Measured BER vs. distance for: diamonds, a WDM channel near the middle of
a group of 37; triangles: the same, but with all other channels turned off (single channel
performance). The distances reported here are just multiples of 600 km, and thus do not include
the DCF and lumped-Raman amplifier fiber lengths. While the polarization scrambler was
turned on for both sets of measurements, it tended to make negligible increase in BER except
at the shorter distances.
 18. Dispersion-Managed Solitons 669

has allowed us to greatly increase the transmission distances achievable with the
more conventional compensation using DCF alone. (Compare with the results shown
in Fig. 18.36.) Second, note that the BER performance in dense WDM is now very
close to that for a single isolated channel. Taken together, these two facts offer clear
and dramatic evidence that our new PGD-complemented dispersion compensation
technique has succeeded in reducing the nonlinear penalty from interchannel col-
lisions to relative insignificance. Indeed, the performance with the new technique,
whether WDM or single channel, is now primarily limited by the combined effects of
the various single-channel penalties, viz., spontaneous emission noise, PMD, PDL,
Gordon–Haus jitter, and so on. Finally, note that by any measure, whether the 10−8
BER distance, or the “super-FEC distance” (where, with a mere 7% overhead, an
uncorrected BER of <10−3 can be corrected electronically to <10−15 ), the distances
here (>20,000 km with the use of super-FEC) are a factor of ≈2× greater than those
ever before achieved by any dense WDM transmission mode.

18.5. Summary
We have examined dispersion-managed solitons, their special periodic pulse behavior,
their advantages over other transmission modes, the conditions required to create and
maintain them, and we have examined closely the one serious nonlinear penalty they
suffer, viz., the timing jitter from collisions with solitons of neighboring channels.
Nevertheless, we have also seen how even that one nonlinear penalty can be made
nearly insignificant by the use of the new PGD-complemented dispersion compensa-
tion. Thus we have seen how such dispersion-managed solitons, in an all-Raman dense
WDM system at 10 G per channel, make a natural and comfortable fit with existing
terrestrial fiber spans, and can provide for transmission to distances of 20,000 km or
more. We have also seen how such DMS transmission is uniquely suited to provide
the backbone of an all-optical network.
On the other hand, dense WDM at 40 G per channel has not been discussed,
simply because the kind of DMS transmission described here is not possible over the
typical 80 to 100 km fiber spans that occupy the US terrestrial network. That is, for
the unchirped pulse widths required for 40 Gbit/s, the pulse breathing in such spans
extends far into the quasilinear regime, where each pulse overlaps many neighboring
pulses. Rather, DMS transmission at 40 G requires the use of so-called dispersion-
managed cable, transmission fiber whose dispersion changes sign frequently, at least
once every 10 km or so, and fiber which, unfortunately, does not yet exist commer-
cially. (Note from Eq. (18.7) that when the unchirped pulse width is scaled down by
a factor of 4 times, the scale length for dispersive broadening, zc , scales down by
a factor of 16 times.) In addition, transmission at 40 G is much more susceptible to
the ravages of PMD and timing jitter, such that transmission over just 3000 or 4000
km has thus far required application of the most powerful FEC technologies. Thus,
whatever merits transmission at 40 G may have, it does not fit into the purview of
this chapter, which is about the foundation of an all-optical, ultra-long-haul network.
Perhaps a few words about the relatively new technology of DPSK (differential
phase-shift keying) would be appropriate here. DPSK may well play an important
670 L.F. Mollenauer

role in quasilinear transmission at 40 G, however, it should be noted that, in general,

DPSK is seriously handicapped by the fact that the nonlinear term in the NLS equation
efficiently turns amplitude jitter into phase jitter (the so-called Gordon–Mollenauer
effect) [38]. Thus, for dense WDM at 10 G with the usual 50 GHz channel spacing,
the OOK with DMS presented here outperforms any mode that may include DPSK,
and the margin becomes exceptionally wide with use of the PGD-complemented
dispersion compensation.

Acknowledgments
A great many colleagues have made major contributions to the work that I have
tried to distill and present here; in that regard, the contributions of Jim Gordon are
of the most pervasive fundamental import; the input of my former colleague Pavel
Mamyshev has been vital as well on a number of issues. Also the extensive numerical
simulations of Nadja Mamysheva and, more recently, of Chongjin Xie have been
invaluable, and it has been highly stimulating to discuss ideas with Chris Xu and
Xiang Liu. It is especially important to note that the idea for the highly successful
PGD-complemented dispersion compensation originated with Xing Wei and Xiang
Liu. The experimental work has had the skillful input of Jay Cloonan, Andrew Grant,
and Inuk Kang. Finally, but by no means least, I have had the most patient help with
computer programming and administration from Jürgen Gripp, Andrew Grant, and
Laura Luo.

References
[1] A. Hasegawa, Numerical Study of optical soliton transmission amplified periodically by
the stimulated Raman process, Appl. Optics, 23: Oct., 3302–3305, 1984.
[2] L. F. Mollenauer, J. P. Gordon, and M. N. Islam, Soliton propagation in long fibers with
periodically compensated loss, IEEE J. Quantum Electron., QE-22, Jan., 157–173, 1986.
[3] L. F. Mollenauer and K. Smith, Demonstration of soliton transmission over more than
4000 km in fiber with loss periodically compensated by Raman gain, Optics Lett., 13,
Aug. 675–677, 1988.
[4] L. F. Mollenauer, J. P. Gordon, and P. V. Mamyshev, Solitons in high bit-rate, long-distance
transmission. In Optical Fiber Telecommunications IIIA, I. P. Kaminow and T. L. Koch,
eds., Vol. IIIA, Chap. 12, San Diego: Academic, 373–460, 1997.
[5] P. V. Mamyshev and N. A. Mamysheva, Pulse-overlapped dispersion-managed data trans-
mission and intrachannel four-wave mixing, Optics Lett., 24, Nov., 1454–1456, 1999.
[6] M. Matsumoto and H. A. Haus, Stretched-pulse optical fiber communications, Optics
Lett., 9, June, 785–787, 1997.
[7] T. Yu, E. A. Golovchenko, and A. N. Pilipetskii and C. R. Menyuk, Dispersion-managed
soliton interactions in optical fibers, Optics Lett., 22, June, 793–795, 1997.
[8] A. R. Chraplyvy, A. H. Gnauck, T. W. Tkach, and R. M. Derosier, 8 × 10 Gb/s transmission
through 280 km of dispersion-managed fiber, IEEE Photon. Technol. Lett., 5, Oct., 1233–
1235, 1993.
 18. Dispersion-Managed Solitons 671

[9] A. R. Chraplyvy, A. H. Gnauck, R. W. Tkach, and R. M. Derosier, One-third terabit/s

transmission through 150 km of dispersion-managed fiber, IEEE Photon. Technol. Lett.,
7, Jan., 98–100, 1995.
[10] M. Suzuki, I. Morita, N. Edagawa, S. Yamamoto, H. Taga, S.Akiba, Reduction of Gordon–
Haus timing jitter by periodic dispersion compensation in soliton transmission, Electron.
Lett., 31:2027–2028, 1995.
[11] N. S. Bergano and C. R. Davidson, Wavelength division multiplexing in long-haul trans-
mission systems, J. Lightwave Technol., 14, June, 1299–1308, 1996.
[12] D. Anderson, Variational approach to nonlinear pulse propagation in optical fibers, Phys.
Rev. A, 27:3135, 1983.
[13] S. K. Turitsyn and E. G. Shapiro, Variational approach to the design of optical communi-
cation systems with dispersion management, Opt. Fiber Technol., 4:151–188, 1998.
[14] J. N. Kutz, P. Holmes, S. G. Evangelides, and J. P. Gordon, Hamiltonian dynamics of
dispersion-managed breathers, J. Opt. Soc. Am. B, 15, Jan., 87, 1998.
[15] J. P. Gordon and L. F. Mollenauer, Scheme for the characterization of dispersion-managed
solitons, Optics Lett., 24, Feb., 323–325, 1999.
[16] L. F. Mollenauer, P. V. Mamyshev, and J. P. Gordon, Effect of guiding filters on the
behavior of dispersion-managed solitons, Optics Lett., 24, Feb., 220–222, 1999.
[17] N. J. Smith, F. M. Knox, N. J. Doran, K. J. Blow, and I. Bernnion, Enhanced power
solitons in optical fibres with periodic dispersion mangement, Electron. Lett., 32, Jan.,
54–55, 1996.
[18] N. J. Smith, N. J. Doran, F. M. Knox, and W. Forysiak, Energy-scaling characteristics of
solitons in strongly dispersion-managed fibers, Optics Lett., 21, Dec., 1981–1983, 1996.
[19] V. S. Grigoryan, T. Yu, E. A. Golovchenko, C. R. Menyuk, and A. N. Pilipetskii,
Dispersion-managed soliton dynamics, Optics Lett., 22, Nov., 1609–1611, 1997.
[20] T.-S. Yang and W. L. Kath, Analysis of enhanced-power solitons in dispersion-managed
optical fibers, Optics Lett., 22, July, 985–987, 1997.
[21] J. P. Gordon and L. F. Mollenauer, Effects of fiber nonlinearities and amplifier spacing
on ultra-long distance transmission, J. Lightwave Technol., 9, Feb., 170–173, 1991.
[22] L. F. Mollenauer, R. Bonney, J. P. Gordon, and P. V. Mamyshev, Dispersion-managed
solitons for terrestrial transmission, Optics Lett., 24, March, 285–287, 1999.
[23] J. P. Gordon and H. A. Haus, Random walk of coherently amplified solitons in optical
fiber transmission, Optics Lett., 11, Oct., 665–667, 1986.
[24] P. V. Mamyshev and L. F. Mollenauer, Soliton collisions in wavelength-division-
multiplexed dispersion-managed systems, Optics Lett., 24, April, 448–450, 1999.
[25] L. F. Mollenauer, J. P. Gordon, and F. Heismann, Polarization scattering by soliton–soliton
collisions, Optics Lett., 20, 2060–2062, 1995.
[26] C. Xu, and C. Xie, and L. F. Mollenauer, Analysis of soliton collisions in a wavelength-
division-multiplexed dispersion-managed soliton transmission system, Optics Lett., 27,
Aug., 1303–1305, 2002.
[27] L. F. Mollenauer and C. Xu, Time-lens timing-jitter compensator in ultra-long haul
DWDM dispersion-managed soliton transmission. In Proceedings of the Conference on
Lasers and Electro-Optics, post deadline papers, CPDB1-1, Optical Society of America,
May 19–24, 2002.
[28] L. F. Mollenauer, C. Xu, C. Xie. and I. Kang, The temporal lens as jitter-killer, In Technical
Digest of the Nonlinear Optics Conference, Optical Society of America, July 29–Aug 2,
66–68, 2002.
[29] L. A. Jiang, M. E. Grein, B. S. Robinson, E. P. Ippen, and H. A. Haus, Experimental
demonstration of a timing jitter eater. In Technical Digest of the Conference on Lasers
and Electro-Optics, Optical Society of America, May 19–24, CTuF7, 2002.
672 L.F. Mollenauer

[30] M. Romagnoli, P. Franco, R. Corsini, A. Schiffini, and M. Midrio, Time-domain Fourier

optics for polarization-mode dispersion compensation, Optics Lett., 24, Sept., 1197–1199,
1999.
[31] L. F. Mollenauer, P. V. Mamyshev, J. Gripp, M. J. Neubelt, and N. Mamysheva, Demon-
stration of massive wavelength-division multiplexing over transoceanic distances by use
of dispersion-managed solitons, Optics Lett., 25, May, 704–706, 2000.
[32] M. Shirasaki, Chromatic-dispersion compensator using virtually imaged phased array,
IEEE Photon. Technol. Lett., 9:1598–1600, 1997.
[33] C. R. Doerr, L. W. Stulz, S. Chandrasekhar, and L. Buhl, Multichannel integrated tunable
dispersion compensator employing a thermooptic lens. In Technical Digest of the Optical
Fiber Communication Conference OFC 2002, Optical Society of America and IEEE,
March 17–22, PD FA6-2, 2002.
[34] C. K. Madsen and G. Lenz, Optical all-pass filters for phase response design with appli-
cations for dispersion compensation, IEEE Photon. Technol. Lett., 10:994–996, 1998.
[35] D. J. Moss, S. McLaughlin, G. Randall, M. Lamont, M. Ardekani, P. Colbourne, S. Kiran,
and C.A. Hulse, Multichannel tunable dispersion compensation using all-pass multicavity
etalons, In Technical Digest of the Optical Fiber Communication Conference OFC 2002,
Optical Society of America and IEEE, March 17–22, 132–133, 2002.
[36] X. Wei, X. Liu, C. Xie, and L. F. Mollenauer, Reduction of collision-induced timing jitter
in dense WDM by the use of periodic group delay dispersion compensators, Optics Lett.,
28:xxx, 2003.
[37] L. F. Mollenauer, S. G. Evangelides, and J. P. Gordon, Wavelength division multiplexing
with solitons in ultra-long distance transmission using lumped amplifiers, J. Lightwave
Technol., 9, March, 362–367, 1991.
[38] J. P. Gordon and L. F. Mollenauer, Phase noise in photonic communications systems
using linear amplifiers, Optics Lett., 15, Dec., 1351–1353, 1990.
Chapter 19
40 Gb/s Raman-Amplified Transmission

L. Nelson and B. Zhu

19.1. Introduction
High-capacity terrestrial optical transmission systems are now being deployed, of-
fering aggregate capacities of 1 Tb/s or higher over distances of more than 1000 km.
These systems use dense wavelength-division-multiplexing (DWDM) operating with
more than 100 channels at a 10 Gb/s line rate and channel spacing of 50 GHz in both
the C- and L-bands. In order to offer scalable solutions for future traffic growth in
the backbone network, the 40 Gb/s line rate appears to be the natural successor to
10 Gb/s. However, with the downturn in the telecommunications industry in 2001
and 2002, natural questions to ask are: why are 40 Gb/s line rates needed, and why
are researchers pursuing 40 Gb/s technologies and transmission? The answers are
twofold: 40 Gb/s line rates will help to meet increased bandwidth demands and 40
Gb/s line rates reduce cost.
Although investments in new technologies and new network builds are low at the
present time, there has been no substantial slowdown in the growth rate of Internet
traffic [1]. According to Coffman and Odlyzko [2], Internet traffic has been approx-
imately doubling each year since 1990 and is expected to continue to grow at this
rate for the rest of this decade. Although Coffman and Odlyzko believe that supply
and demand will grow at comparable rates, and that there will be neither a bandwidth
glut nor a bandwidth shortage, it has always been difficult to predict which commu-
nications services people will accept and how these services will be used. Prices will
also be important in determining the evolution of traffic. As prices fall, usage may in-
crease. Coffman and Odlyzko predict that capacity will most likely continue to grow
at rates somewhat faster than traffic, perhaps reflecting the fact that data traffic needs
more “head-room” than voice traffic. Work on 40 Gb/s technologies today prepares
us for the anticipated future demand.
Perhaps more importantly, 40 Gb/s line rates will eventually offer carriers the
opportunity for lowering the cost-per-transmitted-bit [3]. A standard rule of thumb is
that when technologies are mature, a factor of four times the capacity will be possible
at about 2.5 times the cost. To date, many of the 40 Gb/s transmission demonstrations
674 L. Nelson and B. Zhu

have used time-division-multiplexed (TDM) systems based on noncarrier-class pro-

totype electronics or optical time-division-multiplexed (OTDM) systems based on
impractical expensive architectures. However, due to advances in high-speed elec-
tronics [4–8] and optical components, cost-effective 40 Gb/s terminal equipment will
soon become available. 40 Gb/s line rates lower the system cost for a number of
reasons. Implementing one 40 Gb/s channel versus four 10 Gb/s channels uses fewer
components, both optical and electrical, and takes up less physical space. Power con-
sumption is also reduced. In addition, with fewer components, the network has fewer
potential failure points, thus improving its reliability. It can also be argued that higher
line rates are the better path to enabling higher spectral efficiencies (or, more pre-
cisely, information spectral density), allowing more capacity in a narrower spectral
band. Extending the spectral bandwidth of the system or, for example, adding the
L-band can increase cost. Finally, 40 Gb/s channels versus 10 Gb/s channels ease the
network management, maintenance, and installation load due to the reduction in the
number of WDM channels. For the same capacity there are fewer wavelength chan-
nels to manage and route. Turn-up of a single 40 Gb/s channel instead of four 10 Gb/s
channels is faster and less expensive. Additionally, there are cost savings in sparing,
as one-quarter the number of 40 Gb/s line cards must be available compared to the
number of line cards required for 10 Gb/s. In summary, system houses are pursuing
40 Gb/s because they see significant cost advantages in the future.
If long-haul and ultra-long-haul terrestrial and submarine networks are to take
advantage of the eventual lower cost per bit, 40 Gb/s line rates must be capable of
bridging several thousand kilometers without electrical regeneration. The challenges
of DWDM transmission at 40 Gb/s are addresssed in Section 19.2, along with the
technologies enabling 40 Gb/s terrestrial transmission. Section 19.3 discusses ad-
vanced experiments and demonstrations at 40 Gb/s using Raman amplification, and
Section 19.4 provides some concluding remarks. Appendix A19 provides a list of
acronyms used in the chapter.

19.2. 40 Gb/s WDM Systems

Due to the promise of higher spectral efficiency and lower per Gb/s cost, 40 Gb/s
technologies are attractive for the construction of high-capacity WDM backbone
networks. However, when migrating from 10 to 40 Gb/s line rates, system design-
ers face a number of technical challenges. In addition to requiring a higher optical
signal-to-noise ratio than 10 Gb/s signals, 40 Gb/s signals have lower tolerance to
group velocity dispersion and polarization mode dispersion. These technical issues
must be addressed before cost-effective 40 Gb/s WDM systems can be installed.
Recent laboratory experiments have demonstrated that broadband distributed and
discrete Raman amplification, advanced modulation formats, optimized transmission
fibers and dispersion maps, and forward error correction are key technologies for 40
Gb/s DWDM terrestrial transmission. In this section, we discuss the challenges and
enabling technologies for developing multiterabit 40 Gb/s WDM systems.
 19. 40 Gb/s Raman-Amplified Transmission 675

19.2.1. Challenges of 40 Gb/s Transmission

19.2.1.1. Optical Signal-to-Noise Ratio

One of the fundamental limitations in an optically amplified transmission system is

the signal–spontaneous beat noise at the receiver caused by accumulated amplified
spontaneous emission (ASE). The signal–spontaneous noise impairment can be char-
acterized in terms of the optical signal-to-noise ratio (OSNR), defined as the ratio of
the optical signal power to the power of the ASE in a specified optical bandwidth,
often referred to as the resolution bandwidth (RBW), which is usually 0.1 nm. In
practical optical fiber transmission systems, the OSNR must be sufficiently high to
achieve the required system performance, which, for example, is a bit error rate (BER)
below 10−15 for modern commercial systems. The required OSNR should also have
sufficient margin to include any impairments arising from chromatic dispersion, po-
larization mode dispersion, fiber nonlinearities, and the distortions introduced by the
transmitter and receiver. When migrating systems from a 10 Gb/s to 40 Gb/s line rate,
the required OSNR must theoretically increase by 6 dB in order to compensate for
the four times wider receiver bandwidth. The actual increase in required OSNR with
the channel bit rate may be greater than the theoretical value because it is more diffi-
cult to achieve comparable transmitter and receiver performance at higher bit rates,
especially in 40 Gb/s systems, where there are great challenges for the development
of high-speed integrated electronic circuits.
In an optically amplified transmission system, each optical amplifier contributes
ASE, and these contributions are added cumulatively along the amplifier chain. As
the length of the system increases, the OSNR at the end of the system decreases.
The maximum unregenerated reach of an optically amplified system is the distance
where the OSNR equals the target OSNR for the required system performance. This
maximum unregenerated reach is also determined by the effective management of
transmission impairments that can generate signal distortion. For a system containing
N fiber spans, where each span is optically amplified, the OSNR of a 1550 nm signal
channel at the end of the system is approximately [9]:

OSNR[in dB / 0.1 nm RBW] = 58 + Pch − Lsp − N F − 10 log10 (N ), (19.1)

where Lsp is the span loss (in dB), N F is the noise figure of the optical amplifier
(in dB), and Pch is the per-channel power (in dBm) launched into the span. Note
that Eq. (19.1) assumes that the signals at the input to the first span have infinite
OSNR. In reality, due to amplifiers in the transmitter, channels typically have OSNRs
between 35 and 45 dB (0.1 nm RBW). This initial OSNR degradation also limits the
transmission distance. It can be seen from Eq. (19.1) that the OSNR can be increased
by one dB if Pch is increased by one dB, if the noise figure is decreased by one dB,
or if the span loss is reduced by one dB. If the OSNR is increased by three dB, the
length of the system can be doubled, assuming the optically amplified system has
fixed amplifier spacing and operates in the linear regime. Maintaining sufficiently
high OSNR is a challenge for 40 Gb/s long-haul WDM transmission.
676 L. Nelson and B. Zhu

There are a number of possible ways to improve the OSNR in long-haul WDM sys-
tems. The OSNR can be increased by using shorter spans to reduce span loss, and this
is done in commercial undersea systems where span lengths are usually 50 km or less.
However, for terrestrial systems, which usually have shorter overall reach than under-
sea systems, shortening the span length is not cost effective, and the amplifier sites are
more or less fixed by pre-existing equipment huts spaced about 80 km or more apart.
In both types of systems, reducing the loss of the transmission fiber and components
will increase the OSNR. Increasing the launch power per span is another possible
way to increase OSNR while maintaining terrestrial span lengths; however, increased
launch power can strongly increase nonlinear impairments associated with self-phase
modulation (SPM), cross-phase modulation (CPM), four-wave-mixing (FWM), and
stimulated Raman scattering (SRS) effects. Distributed Raman amplification com-
bined with erbium-doped fiber amplifiers (EDFAs) [10], and all-Raman amplification
have been demonstrated as powerful techniques for improving the OSNR for 40 Gb/s
with terrestrial span losses [11, 12]. The advantages of distributed Raman technologies
are discussed in Section 19.2.2.1. In addition, forward-error-correction coding can be
used to decrease the required OSNR at the receiver, as discussed in Section 19.2.2.4.
Most recent long-haul 40 Gb/s transmission experiments have used a combination of
Raman amplification and forward error correction.

19.2.1.2. Chromatic Dispersion

Chromatic dispersion is one of the important sources of distortion for 40 Gb/s signals.
It originates from the frequency dependence of the refractive index of the optical
fiber (material dispersion) and the frequency dependence of the mode distribution
(waveguide dispersion). As a result of chromatic dispersion, different frequencies of
light travel at different speeds. In on-off keyed data transmission, where 1s and 0s are
represented by the presence and absence of light, respectively, the 1 pulses contain a
range of different frequencies, and chromatic dispersion causes the pulses to spread
when they propagate along the fiber. The signal pulses can spread into the time slots
for adjacent bits, leading to distortion.
The effect of chromatic dispersion is cumulative and increases linearly with trans-
mission distance. More important, it increases quadratically with the data rate because
of the combination of a wider spectrum and shorter pulse width when doubling the
data rate. The dispersion-limited length, corresponding to the distance after which a
pulse has broadened by one bit interval, is inversely proportional to the chromatic
dispersion of the fiber, the bit rate, and the spectral width of signal pulses [13]. For
high bit rate long-haul transmission, external modulation of continuous-wave diode
lasers is usually preferred to direct modulation of the lasers because of the narrower
spectrum that results. For signals produced by external modulation, the spectral width
approximates the bit rate B. Assuming linear transmission using external nonreturn-
to-zero (NRZ) modulation, the maximum allowable accumulated dispersion for a 1
dB eye-closure penalty is given by
104,000
DL[ps/nm] < , (19.2)
B2
 19. 40 Gb/s Raman-Amplified Transmission 677

10000
NZDF 4.5 ps/nm-km
NZDF 7 ps/nm-km
1000
Dispersion Limit (km)

100

10

SSMF 17 ps/nm-km

1
1 2.5 10 20 40 100 160
Data Rate (Gb/s)

Fig. 19.1. Linear NRZ dispersion-limited transmission distance for optical fibers having chro-
matic dispersions of 4.5, 7, and 17 ps/nm-km.

where B is the bit rate in Gb/s, D is the chromatic dispersion in ps/nm/km, and L is the
distance in km. It can be found from Eq. (19.2) that the dispersion limit is 1040 ps/nm
for 10 Gb/s, and the dispersion limit is only 65 ps/nm for 40 Gb/s. Figure 19.1 shows
a plot of the dispersion-limited transmission distance versus data rate when using
common transmission fibers having chromatic dispersions of 4.5, 7, and 17 ps/nm-
km. The precise limit depends on the details of the modulation format and design
of the receiver circuitry. It can also be seen from Eq. (19.2) that a fourfold increase
in the bit rate causes a decrease in the dispersion limit by a factor of 16. Precise
dispersion compensation at 40 Gb/s is therefore critical, and for broadband WDM
systems, dispersion slope compensation is also necessary. Dispersion-compensating
fibers (DCF) with negative dispersion and negative dispersion slope are the most
widely used compensation method for transmission fibers with positive chromatic
dispersion, as discussed in Chapter 6. Today, slope-matching DCFs are commercially
available for all common transmission fibers, including standard single-mode fiber
(SSMF) and several nonzero dispersion fibers (NZDF).
In the field environment, the temperature-dependence of the chromatic dispersion
can cause impairments by increasing the accumulated dispersion beyond the disper-
sion limit. The temperature coefficient of dispersion is inversely proportional to the
dispersion slope, and coefficients can vary from −0.003 ps/nm/km/◦ C for high-slope
NZDF to −0.0012 ps/nm/km/◦ C for low-slope NZDF [14]. For a 2000 km system
and a 25◦ C temperature change, the dispersion for high-slope NZDF could vary by
150 ps/nm, clearly outside the 40 Gb/s dispersion window. Hence, depending on the
678 L. Nelson and B. Zhu

transmission distance and fiber type, 40 Gb/s ultra-long-haul transmission systems

may require dynamic dispersion compensation at the receiver.
Chromatic dispersion plays a dual role in modern fiber-optic communication sys-
tems: it hurts system performance by temporally spreading the pulses, leading to limi-
tations on the transmitted signals as mentioned above, but, in the absence of chromatic
dispersion, nonlinear effects such as four-wave-mixing and cross-phase modulation
make high-channel-count DWDM systems practically inoperable. Dispersion man-
agement is the reconciliation of these two opposite effects; the transmission fiber is
required to have nonzero, and typically positive, dispersion, and an equal amount of
negative dispersion is introduced to obtain a link average of zero or near-zero disper-
sion. In such a dispersion-managed link, the nonzero dispersion of the transmission
fiber significantly reduces the interchannel nonlinear effects, and the linear effects of
dispersion-induced pulse spreading can be corrected by the negative dispersion.

19.2.1.3. Polarization Mode Dispersion

Polarization mode dispersion (PMD), originating from optical birefringence and the
random variation of its orientation along the fiber length, is another critical issue in
long-haul 40 Gb/s transmission systems (see [15] and references therein). To first-
order, PMD causes a differential group delay (DGD) between the two orthogonal
principal states of polarization (PSP), and when the DGD is a significant fraction of
the bit period, pulse distortion and system penalties occur. Environmental changes
including temperature and stress cause the fiber PMD to vary stochastically in time,
making PMD particularly difficult to manage. In addition, although amplifiers or other
components such as add-drop multiplexers in an optical system may have constant
birefringence, variable polarization rotations between them due to the environment
cause these components to add randomly to the PMD of the total system.
The DGD is a random variable having a Maxwellian probability density function
[16]. Such a probability function can take on extremely high DGD values that exist in
the tail of the distribution and correspond to system outages, when the power penalty
exceeds a certain allowed value. On average, these outages may happen for only
a few seconds or minutes per year and will depend on the mean DGD value and
environment changes. In reality, the outages are likely to occur less often but with
longer duration than the average. The PMD of a fiber span is typically specified in
terms of a mean DGD, an average of the DGD over time or wavelength. Because of
the statistical nature of PMD, for fiber lengths of several km or more, the mean DGD
increases with the square root of the fiber length. The mean DGD of a fiber is often
specified using a “PMD coefficient” having units of ps/km1/2 .
A commonly used rule of thumb is that systems can tolerate a mean DGD, τ , of
approximately 10% of the bit period T . This rule arises from an outage probability
calculation assuming that the power penalty contributions from PMD are less than
Np dB for all but a specified cumulative duration per year. The pulse shape (i.e.,
modulation format) [17–19] and details of the receiver [20] are other important factors
in this calculation. An estimate for the power penalty due to first-order PMD can be
obtained from [21]
ε(dB) = A(τ/2T )2 sin2 N, (19.3)
 19. 40 Gb/s Raman-Amplified Transmission 679

where A is a dimensionless parameter depending on the pulse shape and receiver,

τ is the instantaneous DGD, and N is the angle between the Stokes vectors of the
launch polarization and input principal state of polarization of the fiber system. Know-
ing the probability density of the launch penalty factor, sin2 N, and the Maxwellian
probability density for τ , the limit for the mean DGD of the fiber can be shown to be
+ +
τ /T ≤ 4 Np / π A ln(1/Pout ), (19.4)
where the outage probability Pout is the fraction of time per year that the PMD penalty
is greater than Np dB [15]. From 10 Gb/s PMD penalty measurements with an opti-
cally preamplified receiver, NRZ signals had A = 70, and 50% duty-cycle return-to-
zero (RZ) signals had A = 30 [18]. Evaluation of Eq. (19.4) for these A values and
allowing Np = 2 dB for not more than 3 seconds per year (Pout = 10−7 ) shows that
τ ≤ 0.095T for NRZ, and τ ≤ 0.145T for RZ. It should be emphasized again that
the A values are strongly dependent on pulse shape and receiver characteristics and
must be determined for each specific system.
Fiber manufacturers currently use the link design value (LDV) to specify the PMD
of cabled fiber. The LDV defines the maximum value for the PMD coefficient in terms
of the probability QPMD for links with at least n concatenated sections, where typically
QPMD = 10−4 and n ≥ 20 (IEC SC 86A/WG1). The LDV thus serves as a statistical
upper bound for the PMD coefficient of the concatenated fibers comprising an optical
cable link. This specification allows for small variations of the PMD coefficient from
section to section. Although legacy fibers installed in the 1980s may exhibit high PMD
coefficients, a recent study [22] showed that more than 80% of post-1994 installed
fibers measured have sufficiently low PMD to transmit 40 Gb/s over at least 500 km
of fiber. In addition, PMD measurements of 432/864 fiber count ribbon cable in the
deployed configuration indicate sufficiently low PMD for 40 Gb/s [23]. Manufacturers
are now routinely specifying LDVs as low as 0.04 ps/(km)1/2 .
The total system PMD is determined by all the components through which the
signal travels, including the transmission fiber as well as the amplifiers, dispersion-
compensating modules (DCM), gain equalizers, and multiplexers/demultiplexers for
add-drop. Figure 19.2 shows a plot of the PMD-limited transmission distance for 40
Gb/s NRZ and RZ signals versus the transmission fiber’s LDV. The calculations use
Eq. (19.4) and assume A parameters of 30 and 70 for RZ and NRZ, respectively, and
a penalty of 2 dB for not more than three seconds per year on average. The solid
curves show the distance limits from transmission fiber PMD only, and the dashed
curves show the limits considering both transmission fiber and component PMD. If the
component DGD associated with each fiber span is 0.4 ps, then for 100 km fiber spans
with LDV of 0.04 ps/(km)1/2 , the mean fiber DGD and mean component DGD will
be the same. Thus the total distance over which a 40 Gb/s signal could be transmitted
before being limited by PMD is reduced by a factor of two. The component DGD
has the potential to vary significantly from one system to another, depending on the
specifications for the various amplifier components and the amplifier design as well
as the length of the DCF. Transmission fibers with lower dispersion require shorter
DCF lengths and thus tend to have lower system PMD.
Note that recent measurements of installed cables have shown that DGD spectra
are remarkably similar when the ambient temperature is the same [24, 25]. This
680 L. Nelson and B. Zhu

8000

Transmission fiber PMD only

Transmission fiber PMD + component PMD

PMD-Limited Transmission 6000

Distance at 40 Gb/s (km) PZ

4000

NRZ
RZ
2000

NRZ

0
0.04 0.06 0.08 0.10 0.12 0.14
Transmission fiber Link Design Value (ps/√km)

Fig. 19.2. Effect of fiber and component PMD on the PMD-limited transmission distance for
40 Gb/s systems, assuming 0.4 ps of component PMD per 100 km span (A = 30 for RZ, A = 70
for NRZ, and an allowed penalty of 2 dB for not more than 3 seconds per year on average).

behavior has been ascribed to the plant structure, where the buried sections of cable
act as time-stable DGD elements and the short exposed cable sections experience
the temperature variations and act as polarization rotators between two buried cable
sections. Further work is necessary, but these measurements suggest that calculation
of outage statistics may require a new nonstochastic model of DGD behavior to reflect
its dependence on temperature.

19.2.2. Enabling Technologies for 40 Gb/s WDM Transmission

19.2.2.1. Raman Amplification

As discussed in several of the preceding chapters, Raman amplification can be used in

telecommunications for both improving noise performance and expanding the WDM
signal bandwidth. Distributed Raman amplification can dramatically improve the ef-
fective noise figure of an amplifier and break the 3 dB barrier [9]. This noise figure
improvement increases system OSNR and satisfies the high OSNR requirement for 40
Gb/s transmission systems. Distributed Raman gain can be induced along the trans-
mission fiber by backward Raman pumping, typically with a laser at a wavelength
approximately 100 nm shorter than that of the signals. This distributed Raman am-
plification overcomes the attenuation in the latter part of the span, and the minimum
 19. 40 Gb/s Raman-Amplified Transmission 681

signal power in the span is increased. This improvement in noise performance is typi-
cally represented by an equivalent (or effective) noise figure, which can be evaluated
by representing the distributed Raman amplification by a discrete amplifier located
at the end of the span that would produce the same gain and the same contribution
to the accumulated ASE [26]. The theoretical (or quantum) limit of the noise figure
for a discrete optical amplifier located at the end of the span is 3 dB, and the noise
figures of commercial EDFAs are typically ∼4 to 8 dB, depending on the design. The
equivalent noise figure from a distributed Raman amplifier was measured to be as
low as −3.7 dB for silica-core fiber [26]. Thus an improvement of 7 dB or more over
a discrete EDFA can be achieved. Although this equivalent noise figure is referenced
to the end of the span, the distributed Raman amplification is not actually located at
the end of the span, but provides distributed amplification over an appreciable part of
the preceding fiber span.
The performance of Raman amplification, such as the Raman gain efficiency and
effective noise figure, depends on the properties of the transmission fibers used. The
Raman gain efficiency, determining how much Raman gain can be obtained from a
given amount of pump power, depends on a number of factors, including the Ra-
man effective area (i.e., the overlap of the pump and signal modefields) [27], the
composition of the fiber [28], and the pump and signal wavelengths [29]. A compar-
ison of Raman gain efficiencies for several fiber types is shown in Table 19.1. The
properties of a number of these fiber types are listed in Table 6.2 of Chapter 6. The
effective area (Aeff ) of the fiber should be optimized so that the best Raman gain
efficiency can be achieved while maintaining sufficiently low nonlinear effects and a
low Rayleigh backscattering coefficient, which is inversely proportional to the fiber’s
effective area [30]. Double-Rayleigh backscattering can limit the noise improvement
from distributed Raman amplification due to Raman-enhanced multiple path inter-
ference (MPI) [26], as discussed in detail in Chapter 15. The optimum Raman gain
efficiency of the transmission fiber may also depend upon the system architecture. The

Table 19.1. Comparison of Raman Gain Efficiencies for Various Fiber Typesa
Peak Raman Pump Required Pump
Gain Efficiency Ref. Wavelength Power (mW) for
Fiber Type (1/W km) (nm) 20 dB Raman Gain
SSMF 0.39 [122] 1455 680
ELEAFb 0.46 [122] 1455 580
TrueWave® REACH 0.63 [121] 1450 440
TrueWave® RS 0.71 [121] 1454 380
AllWave® 0.39 [121] 1451 680
UltraWaveTM SLA 0.29 [109] 1450 920
UltraWaveTM IDF 1.2 [109] 1450 n.a.
DCF 3.3 [121] 1450 n.a.
a Pump powers assume a 100 km span, 0.25 dB/km loss at pump wavelength, and no pump depletion.
b LEAF is a registered trademark of Corning Incorporated.
682 L. Nelson and B. Zhu

performance of very broadband WDM systems could face a limitation from stimulated
Raman scattering crosstalk from the short to long wavelength channels [31, 32].
Fibers used for distributed Raman should have low loss at the pump wavelengths,
so that less pump power is required to obtain a certain level of Raman gain. Lower
pump loss also means that the pump power penetrates farther into the span, reducing
the span noise figure [33]. In addition, fiber properties such as dispersion and dis-
persion slope determine whether Raman noise associated with pump–pump and/or
pump–signal nonlinear interactions occurs. Low dispersion slope fibers move the zero
dispersion wavelength away from the pump wavelength range, preferably below 1400
nm, to prevent signal OSNR degradation from pump–pump four-wave-mixing [34]
and from pump–signal four-wave-mixing for copumped Raman amplification [35].
Given these considerations for Raman pumping, new nonzero dispersion fibers have
been developed with λ0 < 1400 nm, loss at 1450 nm <0.26 dB/km, and Raman gain
efficiency of ∼0.6 /W km [36].
The performance of Raman amplifiers also depends on the system architecture.
There are two possible schemes for Raman amplifiers: copumped and counterpumped.
Due to the “instantaneous” response of the Raman effect (<1 ps), any pump fluctua-
tion slower than 1 ps can cause gain fluctuations, and hence signal fluctuation through
the Raman gain. In the counterpumped distributed Raman amplification scheme, the
long interaction between signal and pumps has an averaging effect, which reduces
the intensity noise induced by the pump laser’s fluctuation. In copumped schemes,
where the signal and pump propagate through the fiber together, the averaging ef-
fect occurs only through the dispersive delay caused by walk-off between pump and
signal. Therefore, low relative intensity noise pump lasers are required in order to
avoid intensity noise on the signal. In addition, the Raman gain highly depends on the
relative orientation of the pump and signal polarizations for copumping architectures
and has been observed to have some degree of polarization-dependence for counter-
pumping configurations as well [37, 38]. Depolarized pump beams are thus required
to minimize polarization-dependent gain.
Distributed Raman amplification is a powerful technique to ease noise perfor-
mance and to increase available bandwidth. However, there are many issues to be con-
sidered in the system design. For example, the ratio of Raman gain in the transmission
fiber to the gain in the EDFA or discrete Raman amplifier must be considered as well
as gain ripple, ASE noise, MPI, and nonlinearities. Higher Raman gain, smaller effec-
tive area, longer fiber length, and lower data rates will result in a higher MPI-induced
penalty from double-Rayleigh scattering. For the same launched signal power, fiber
nonlinearities are also increased due to the higher path-average powers that depend on
the Raman gain and pumping scheme for distributed Raman amplification [39]. There-
fore, complex trade-offs must be met for optimization of the system performance.
As discussed in Section 19.3, distributed Raman amplification combined with
erbium-doped fiber amplifiers has been demonstrated as a powerful technique for
improving the OSNR for 40 Gb/s systems with terrestrial span losses [10, 40–42].
All-Raman amplified spans have been demonstrated for 40 Gb/s transmission [11, 12],
where both co- and counterpumping are sometimes used in order to achieve both flat
Raman gain and flat OSNR across the entire operating band [43].
 19. 40 Gb/s Raman-Amplified Transmission 683

19.2.2.2. New Optical Fibers

The optical fiber is a critical medium to be optimized for high capacity 40 Gb/s (ultra-)
long-haul transmission. New fiber technologies should support broadband dispersion
and dispersion slope compensation, have low PMD, enhance Raman amplification,
reduce the total system penalties due to nonlinear impairments, and simplify optical
networking.
In terrestrial systems, chromatic dispersion compensation is achieved in the am-
plifier stations with lumped dispersion-compensating fibers. For such systems, the
transmission fiber must meet a complex trade-off: the chromatic dispersion should
be small enough to avoid SPM-induced penalty and to ease dispersion compensation,
but it should be large enough to avoid interchannel nonlinear effects [44]. Low dis-
persion slope of the transmission fiber is also highly desired, enabling an easier match
of the transmission fiber to a commercially available and cost-effective DCF. When
the dispersion and dispersion slope are compensated, all WDM channels experience
nearly the same dispersion map, resulting in optimum performance of all channels. In
the system design, this requirement can be met when the transmission fiber and DCF
have almost the same relative dispersion slope (RDS = ratio of dispersion slope to
dispersion at a specified wavelength). Tables 6.2 and 6.3 of Chapter 6 list the prop-
erties of common commercial transmission fibers from several companies and the
DCFs available from one company.
Fibers with moderate chromatic dispersion around +7 to +8 ps/nm/km and low
RDS have been developed recently and used in transmission experiments [45, 12].
These fibers have the advantage of sufficient dispersion for reduction of interchannel
nonlinearities (a problem particularly at 10 Gb/s with tight channel spacing), and yet
a lower dispersion than standard single-mode fibers (D = 17 ps/nm/km at 1550 nm)
to minimize costs associated with the DCF. Dispersion compensation for standard
single-mode fiber is more costly both in terms of the dispersion-compensating fiber
lengths required as well as the higher noise figure and increased design complexity
of the EDFA when up to 10 dB of interstitial loss must be compensated. In addition,
moderate fiber dispersion has been shown to be optimal for 40 Gb/s, nonreturn-to-
zero signals [46, 47]. A new NZDF fiber with dispersion of 7 ps/nm/km and a low
RDS of 0.0058/nm at 1550 nm has been used for a large-scale long-haul transmission
experiment [12], where the dispersion of each 100 km TrueWave® REACH fiber
span was compensated by a single DCF module covering the entire C+L-bands.
The variation in the span path-average dispersion was less than ±0.043 ps/nm/km
over the 75 nm band, as shown in Fig. 19.3. Furthermore, these DCF modules can
function as lumped Raman gain modules, enabling all-Raman amplified spans without
bandsplitting.
Broadband dispersion and dispersion slope compensation can also be achieved us-
ing so-called hybrid or dispersion-managed spans, where large effective area positive
dispersion fiber and negative dispersion fiber having a matching RDS are employed
in the transmission line. This technique has been used routinely in submarine links to
achieve transoceanic transmission lengths [48–50]. Use of dispersion-slope managed
fiber pairs in terrestrial ultra-long-haul systems is a natural extension of their use in
684 L. Nelson and B. Zhu

40

35

Span Path Average

Dispersion (ps/nm)
30

25

20
1525 1545 1565 1585 1605
Wavelength (nm)

Fig. 19.3. Residual dispersion of a 100 km span of TrueWave® REACH fiber compensated
by a single DCF module covering both C- and L-bands for the 40 Gb/s, WDM experiment
described in [12]. (Reprinted with permission from the Optical Society of America.)

undersea systems, and several recent experiments have been performed at 10 and 40
Gb/s [51–55], discussed in more detail in Section 19.3. The latter four experiments
employed super-large area fiber (SLA) and inverse dispersion fiber (IDF), which
currently have the properties listed in Table 19.2. Figure 19.4(b) shows the residual
dispersion obtained for a 100 km span consisting of 42.5 km of SLA, 34.5 km of IDF,
and 23 km of SLA. This span configuration (see Fig. 19.4(a)) is often referred to as an
“ABA” span, where “A” refers to the SLA fiber and “B” refers to the IDF. In earlier
submarine systems, the span configuration was typically AB due to the shorter span
lengths and EDFA-only amplification [48, 49]. For terrestrial Raman-pumped trans-
mission, placing the smaller effective area IDF in the center of the span (as in the ABA
span) can effectively reduce the nonlinear effect and improve the noise performance
for the system, as discussed in Section 19.3.2.2.
In addition to providing slope compensation across the C- and L-bands, dispersion-
managed spans eliminate several disadvantages of the lumped DCMs used in single-
fiber systems. The DCMs contribute loss, PMD, and cost to the system without in-

Table 19.2. Current Properties of UltraWaveTM Fibers [56] for the Dispersion-Managed Spans
Used in Several Experimentsa

UltraWaveTM SLA UltraWaveTM IDF

Dispersion @ 1550 nm 20 −44
(ps/nm km)
Dispersion Slope @ 1550 nm
0.062 −0.13
(ps/nm2 km)
RDS (1/nm) 0.0031 0.0031
Attenuation (dB/km) 0.187 0.234
Effective Area (µm2 ) 107 31
a [52–55].
 19. 40 Gb/s Raman-Amplified Transmission 685

SLA IDF SLA

+D, 42.5km –D, 34.5km +D, 23km

DWDM
Signals

SLA: Super Large Area fiber

IDF: Inverse Dispersion Fiber
Raman
Pumps

0.6
Average Dispersion
Residual Path-

0.5
(ps/nm/km)

0.4

0.3

0.2
1525 1545 1565 1585 1605
Wavelength (nm)

Fig. 19.4. Residual dispersion for the 100 km SLA + IDF + SLA spans used in [57].

creasing total system length. Systems based on dispersion-managed spans can be

expected to have a lower amplifier noise figure and lower system PMD. There are a
number of issues in the design of dispersion-managed spans, including the optimal
span configuration [58] and how many times to change the sign of the dispersion within
the span. Recent submarine experiments have used double (ABAB) and quadruple
(ABABABAB) hybrid spans [59, 50] in an effort to mitigate the waveform distor-
tion induced by the fiber nonlinearity. An inverse double-hybrid span configuration
(BABA) has also been proposed, where the A fiber is a pure silica-core fiber with
195 µm2 Aeff [60]. Another issue is the exact dispersion values for the positive and
negative dispersion fibers [61, 62]. Medium dispersion fibers have been proposed to
lower the accumulated dispersion within each span; however, there is still a debate
about how much accumulated dispersion a system can tolerate. Most likely, this value
depends both on the modulation format and exact details of the system. To date, a
problem with these medium dispersion fiber pairs has been that the effective areas of
both the positive and negative dispersion fibers are smaller than those of the higher
dispersion fiber pairs. Research work is now focusing on these issues to assess whether
terrestrial systems using dispersion-managed spans offer sufficient performance ad-
vantage to offset the increased complexity in the installation and management of the
outside plant.

19.2.2.3. Optimized Modulation Formats

The modulation format must be chosen so that both linear and nonlinear performance
are optimized at 40 Gb/s. Modulation formats that not only have better receiver sen-
686 L. Nelson and B. Zhu

sitivity but also provide high tolerance to nonlinearities, chromatic dispersion, and
PMD are beneficial. Furthermore, the modulation formats should offer high spec-
tral efficiency and high tolerance to imperfections of the WDM multiplexing and
demultiplexing filter characteristics. The best choice of modulation format depends
on multiple system parameters such as fiber properties, channel spacing, filter char-
acteristics, data rate, system reach and capacity, and cost. The most commonly used
formats in 40 Gb/s long-haul optical communication have been nonreturn-to-zero,
return-to-zero, and carrier-suppressed RZ (CSRZ) [63]. Recently, advanced modula-
tion formats such as RZ differential-phase-shift-keying (RZ-DPSK) [11], duobinary,
single/vestigial sideband modulation [41], and phase-shaped binary transmission [64]
have been considered for 40 Gb/s WDM transmission.
For NRZ modulation, a 1 is represented as a rectangular pulse occupying the full
bit period, and a 0 is the absence of a pulse. In practice, the NRZ-encoded optical
signal is formed by external modulation using a LiNbO3 Mach–Zehnder modulator
or electroabsorption modulator, driven by a 40 Gb/s electrical signal. RZ modulation
uses pulses that are substantially narrower than a bit period to represent 1s, so that
even for consecutive 1s, the power level returns to zero between successive pulses.
The pulses in a CSRZ-encoded optical signal are narrower than a bit period but are
typically longer than RZ pulses. Figure 19.5(a) shows optical eye diagrams for NRZ,
RZ, and CSRZ. Usually, RZ and CSRZ pulses are generated using two modulation
stages, the first to encode data, and the second to form pulses. The second modulation
stage has conventionally occurred in a second dual-electrode LiNbO3 Mach–Zehnder
modulator, driven with a pair of half bit rate clock signals and biasing either at the
transmission peak (for RZ) or the null point (for CSRZ). The carrier component of the
CSRZ signal spectrum is suppressed, and the spectral bandwidth of the CSRZ signal
is smaller than that of the conventional RZ signal, as shown by the optical spectra

NRZ RZ CSRZ
Eye pattern

(a)

–25 –20 –20

Power (dBm)
Spectrum

Power (dBm)

–30
Power (dBm)

–35 –30
–45 –40 –40
–55 –50 –50
–65 –60 –60
–75 –70 –70
1545 1545.5 1546 1546.5 1547 1545 1545.5 1546 1546.5 1547 1545 1545.5 1546 1546.5 1547
Wavelength (nm) Wavelength (nm) Wavelength (nm)
(b)

Fig. 19.5. Eye diagrams and spectra at 40 Gb/s line rate for NRZ, RZ, and CSRZ modulation
formats.
 19. 40 Gb/s Raman-Amplified Transmission 687

in Fig. 19.5(b). For practical receiver designs, RZ and CSRZ modulation result in
receiver performance superior to that for NRZ modulation by 1 to 2 dB depending
on the duty cycle. Various RZ and CSRZ formats have narrower pulse widths and
hence wider optical spectra and thus are inherently more resilient to nonlinearity and
first-order PMD [19, 63]. NRZ has a narrower optical bandwidth and is more tolerant
to dispersion slope, but less resilient to nonlinearity. CSRZ has a narrower optical
bandwidth than that of RZ, and thus it has the potential for higher spectral efficiency.
For DPSK modulation, the data information is encoded in the relative phases
of the optical pulses. Adjacent bits of the electrical NRZ data signal are compared,
and the phase of the optical signal is modulated depending on the outcome of the
comparison. When the data signal has two adjacent 1s or two adjacent 0s, the optical
signal receives a zero relative phase shift; when the data signal changes from a 0 to
1 or 1 to 0, the optical signal receives a π -phase shift. The intensity of the optical
signal is therefore constant or, for RZ-DPSK, is a constant pulse train. A high-speed
phase modulator or dual-drive intensity modulator is usually employed to generate
the PSK, and an integrated Mach–Zehnder delayed interferometer demodulator is
used at the receiver to convert the differential phase modulation back to amplitude
modulation. This allows a balanced receiver to be used to detect the signal [11, 65].
In principle, compared to amplitude shift-keyed (ASK) modulation, DPSK provides
superior receiver sensitivity when a balanced receiver is used [66]. In addition, DPSK
reduces nonlinear effects such as CPM due to its constant channel power; however,
it may suffer from nonlinear phase noise caused by amplitude-to-phase-noise con-
version from signal–ASE noise beating through SPM [67]. About 3 dB reduction in
the OSNR required to achieve a given BER has been demonstrated using balanced
detection, hence a larger increase of system reach can be achieved. For example, 2.5
Tb/s (64 × 42.7 Gb/s) transmission over 4000 km of TrueWave® RS fiber [11] and
3.2 Tb/s (80 × 42.7 Gb/s) over 5200 km of fiber with dispersion-managed spans [65]
were demonstrated recently.
In order to increase spectral efficiency in WDM systems, it is essential to reduce
the optical spectrum width without losing information. Vestigial sideband (VSB)
filtering of unequally spaced NRZ signals [41] and optical duobinary coding or phase-
shaped binary transmission (PSBT) [68, 69] have been considered recently for 40
Gb/s WDM transmission. For VSB filtering, one of the two sidebands of an NRZ
optical spectrum is filtered out, as these two sidebands generally contain redundant
information. However, VSB is difficult to implement at the transmitter because the
suppressed sidebands can rapidly reconstruct through fiber nonlinearities [41]. The
VSB filtering can be employed at the receiver to increase spectral efficiency, and
an experiment has demonstrated a spectral efficiency of 0.64 bit/s/Hz with 40 Gb/s
channels alternately spaced by 50 and 75 GHz using VSB filtering [41]. Duobinary,
also referred to as PSBT, is basically a three-level coding scheme (−1/0/ + 1) that
is optimized to reduce the channel spectral width. High spectral density thus can be
achieved; however, duobinary modulation has poor receiver sensitivity due to its poor
extinction ratio of the signal. Enhanced PSBT (EPSBT) was proposed to overcome the
low tolerance to noise by superimposing NRZ modulation on the PSBT modulation by
cascading two modulators in series [70]. The low extinction ratio also can be improved
688 L. Nelson and B. Zhu

by passing the PSBT signal through a bandwidth-limited optical filter located at the
transmitter or at the receiver [71]. A transmitter composed of a 10 Gb/s low-pass
filtered duobinary modulator followed by an optical filter having a 3 dB bandwidth
of 7 GHz was shown to enhance the spectral efficiency by ∼20% and improve the
receiver sensitivity by more than 1 dB. This bandwidth-limited PSBT (BL-PSBT)
also has been reported to reduce single-channel nonlinear effects [64] and has high
dispersion tolerance. Such advanced modulation formats have been considered to
increase overall capacity by increasing the spectral efficiency. However, this increase
in spectral efficiency is often associated with a sacrifice in receiver sensitivity and,
hence, a considerable reduction in the overall system reach.

19.2.2.4. Forward Error Correction

In order to relax the constraints in terms of OSNR and signal distortion and to meet the
target reach for a system, forward error correction (FEC) is virtually a prerequisite for
40 Gb/s systems. With FEC, extra bits are appended to the data by the FEC encoder at
the transmitter. The FEC decoder at the receiver then uses these extra bits to detect and
correct erroneously received bits. As a result, FEC enables the system to operate at a
required OSNR far below that required without FEC, while maintaining an acceptable
BER. Thus FEC enables a very large system margin that can be used to increase the
transmission distance and system capacity.
The performance of the FEC code in correcting errors is characterized in terms
of the coding gain, defined as the difference in the Q-factor (as 10 log Q2 ) between
the uncorrected and corrected BER. The coding gain is usually defined at the system
target Q or BER (e.g., 10−15 ). The line rate increases when extra bits are added due
to the FEC, and systems have additional penalties associated with impairments due
to increased bandwidths. Thus the coding gain is often quoted as net effective coding
gain (NECG) [72], defined as the coding gain corrected for the penalties due to the
higher line rate when the extra overhead is added to the system. Typically, for a given
type of error-correcting code, the stronger the FEC coding gain and the higher the
overhead, the better the FEC code will be at correcting a severely corrupted signal.
In practice, the best FEC code is the one offering the greatest NECG, but requiring
the least line rate increase. In addition to the coding gain and added overhead, the
complexity of the encoding algorithm and the feasibility of implementing it on a high-
speed integrated circuit (IC) are critical considerations in developing an FEC code.
Several types of FEC schemes have been employed and considered in optical
communication systems [72]. A commonly used FEC code is the (255,239) Reed–
Solomon (RS) code, which is the ITU G.975 standard (ITU 1999). The RS (255,239)
code increases the bit rate by 7%, from 9.953 to 10.664 Gb/s, providing a coding gain
of about 6 dB at 10−15 BER. This code was first adopted for commercial undersea
systems, and its use has now been extended to terrestrial systems. A straightforward
improvement of coding gain would be to use a stronger RS code, and a coding gain
increase of 1.2 dB has been reported by increasing the line rate from 7% for the
G.975 RS code to 14% for the RS(255,223) [73]. To further improve the coding
gain concatenated FEC codes can be used, where two FEC codes, an inner code and
 19. 40 Gb/s Raman-Amplified Transmission 689

outer code, are employed sequentially. At the transmitter the data are sequentially
encoded with the outer code and then the inner code, whereas at the receiver the
data are sequentially decoded with the inner code followed by the outer code. An
interleaver and iterative coding process are often used in concatenated FEC codes. A
concatenation of the RS (255,223) with the RS (255, 239) code was proposed for 10
Gb/s systems, providing an additional 2 dB of coding gain relative to the RS (255,239)
with an increased line rate of about 25% [74]. An optimum FEC overhead of 10% with
additional NECG increase of 1.99 dB using concatenated turbo codes and low-density
parity check codes was also reported for 20 Gb/s ultra-long-haul transmission [75].
For 40 Gb/s transmission FEC is even more critical because of the high OSNR
required at the receiver. Because the penalties from transmission and from the trans-
mitter/receiver increase dramatically for line rates near 40 Gb/s, higher coding gain
with less overhead redundancy would be very attractive. One way is to use interleav-
ing and iteration of several shorter codes to increase the FEC coding gain without
significantly increasing the overall redundancy. A trade-off has to be made between
the coding gain (and overhead) and the transmission impairments due to a higher
ratio of redundant bits. The optimum choice will be determined by the progress of
high-speed electronic ICs [6] and error-correction coding technologies.

19.3. Advanced Experiments and Demonstrations

The previous section has reviewed the issues and challenges of 40 Gb/s transmission
as well as several of the key enabling technologies. The reach and capacity of 40 Gb/s
transmission experiments have increased in large part due to advancements in Raman
amplification, particularly for terrestrial-length spans (i.e., >75 km). The lower span
noise figure afforded by Raman can be utilized for a number of different system
upgrades, including increased span loss, higher bit rates, increased number of spans,
and lower launch power per channel to decrease fiber nonlinearities. As also discussed
in previous chapters, Raman amplification opens new transmission bands and allows
seamless broadband systems. This section discusses a number of 40 Gb/s terrestrial
transmission experiments illustrating how Raman amplification has enabled multi-
Tb/s capacities over long-haul and ultra-long-haul distances. Table 19.3 at the end of
Section 19.3 summarizes the various transmission experiments discussed. Note that
this section does not attempt to cover the specific nonlinear impairments influencing
transmission at 40 Gb/s. These are primarily single-channel effects (i.e., intrachannel
cross-phase modulation and intrachannel four-wave-mixing), and are explained in
detail by Essiambre et al. [76] and Bayvel and Killey [77].

19.3.1. Typical 40 Gb/s Transmission Experiment Setup

As many of the transmission experiments reported in the literature have used similar
configurations, we show a schematic diagram of a typical 40 Gb/s transmission ex-
periment in Fig. 19.6. The WDM transmitters consist of distributed feedback (DFB)
lasers with channel spacing on the ITU grid (i.e., 200, 100, or 50 GHz). The odd
690 L. Nelson and B. Zhu

EDFA
1 1x20
1x20
NRZ
NRZ CSRZ
CSRZ
C-band
C-band
39
Q DCF
9.953 Gb/s FEC 4:1 Mux 21.3 Post-
31
PRBS 2 –1 Encoder 42.7 Gb/s GHz OI Comp
Pre-Amp.
Q WGR
2 WGR
1x20
NRZ
NRZ C & SW
&SW
C
C-band CSRZ
CSRZ C C
40 42.7Gb/s
ETDM
Receiver
41 1x20
1x20
NRZ CSRZ A L
79 L-band
L-band L O
WGR FEC
FEC
Q DCF M 3-dB & SW Decoder
decoder
Coupler Post-
9.953 Gb/s FEC 4:1 Mux 21.3 Comp
31
Encoder 42.7 Gb/s GHz OI Pre-Amp.
PRBS 2 –1
BERT
Q A
42
1x20
1x20 O
L-band NRZ CSRZ
CSRZ
80 L-band M

x4 DCF 100 km Fiber

C C-GE C

L L-GE L Counter-RP Counter-RP CO-RP

Fig. 19.6. Schematic diagram of a general 40 Gb/s, C+L-band recirculating loop experiment:
OI: optical interleaver, AOM: acousto-optic modulator, RP: Raman pump(s), C-GE: C-band
dynamic gain equalizer, L-GE: L-band dynamic gain equalizer, WGR: waveguide grating
router, SW: optical switch.

and even channels are multiplexed separately by waveguide grating routers (WGR)
and modulated independently. The optical signals pass first through LiNbO3 Mach–
Zehnder modulators (MZM) driven by 42.7 Gb/s electrical data streams. Electrical
time-division-multiplexing (ETDM) is employed to generate the 42.7-Gb/s signals
from four 10.664 Gb/s data streams, each consisting of 231 − 1 pseudorandom bit se-
quence (PRBS), 9.953 Gb/s data encoded with 255,239 Reed–Solomon FEC. Ideally,
these 10.664 Gb/s data streams should have relative delays of a quarter of the PRBS
length, so that the resulting 42.7 Gb/s signal is also a PRBS. The PRBS is chosen as
the test data signal for experiments because it closely resembles the genuinely random
signals found in real data [78]. Pseudorandom and genuinely random sequences have
nearly identical autocorrelation properties and spectral characteristics. Use of long
PRBS lengths (i.e., 223 − 1 or 231 − 1) is generally recommended for several reasons.
Because the lowest frequency content of a PRBS extends down to B/N (where B is
the bit rate and N is the PRBS length), a long PRBS length will test the bandwidth of
the receiver electronics. In addition, long PRBS lengths are the more stringent test for
transmission impairments because most penalties arise from isolated 0s within long
strings of 1s. Note that some experiments use a simple 10.664 Gb/s PRBS (231 − 1 )
to emulate the 7% overhead FEC, instead of encoding 9.953 Gb/s data with 255,239
Reed–Solomon FEC code.
The second modulator in each transmitter is used for pulse carving to generate a
42.7 GHz RZ or CSRZ pulse train by driving it with a 21.329 GHz clock. An amplifier
(possibly polarization-maintaining) can be inserted between the two modulators to
 19. 40 Gb/s Raman-Amplified Transmission 691

maintain a high OSNR at the transmission line input. The odd and even channels in
each band are then combined by an optical interleaver (OI). Such a scheme provides
effective decorrelation of the data carried by neighboring WDM channels, assuring
good emulation of real transmission where there are random data patterns on all
channels. For precompensation of dispersion, the WDM channels of each band can
be transmitted through DCF modules, prior to being amplified, combined, and finally
launched into the transmission line.
Figure 19.6 also shows the recirculating transmission loop, where the optical sig-
nals are switched into the loop and allowed to circulate a number of times through the
fiber spans before being switched out for error analysis. Until recently, recirculating
loops were used primarily for submarine system testing [79], as they allow testing of
long-haul and ultra-long-haul transmission distances with a modest number of com-
ponents and fiber spans. Loop experiments offer a significant cost advantage when
compared to straightline experiments, but these loop experiments are more complex
and must be done with care to avoid misleading results. The loop should be at least as
long as one period of the dispersion map, and preferably longer, as longer loops also
ensure that the signals pass through the loop components (acousto-optic modulators
(AOM) and coupler) as few times as possible for the target transmission distance.
The AOMs often have polarization-dependent loss (PDL), and an extra EDFA is
usually required to compensate for the loss of the AOM and output coupler. Passing
through these components less frequently minimizes their contributions to the system
impairments.
As an aside, it should be noted that recirculating loops do not necessarily correctly
reproduce the polarization behavior of real systems. Recent studies have shown that
the loop’s PDL can unrealistically improve the performance when a polarization con-
troller is used to optimize the polarization evolution in the loop [80, 81], particularly
when the loop is short and the transmission distance is very long. With polarization
adjustment the PDL can extinguish the ASE orthogonal to the signal, thus improving
the degree of polarization and OSNR [82]. The PMD in loop transmission experi-
ments is also a concern, because recirculating loops tend to exhibit some measure
of deterministic behavior that can produce unrealistic PMD distributions [83]. Polar-
ization adjustment of the signals can minimize (or maximize) the effects of PMD.
Using a polarization scrambler for the signals before the loop as well as a fast po-
larization scrambler within the loop (operating synchronously, on the time scale of
one round-trip) has recently been proposed to better emulate the results that would
be obtained for straightline transmission [83, 81, 82]. However, these polarization
scramblers add additional loss and thus lower the OSNR, a critical impairment to 40
Gb/s transmission. Due to this OSNR impairment, to date, most 40 Gb/s transmis-
sion experiments have not used polarization scrambling and have relied on long loop
lengths and relatively short transmission distances to minimize the unrealistic loop
polarization effects. But as 40 Gb/s transmission experiments increase to thousands
of km, at least some amount of polarization scrambling of the signals undoubtedly
will be required.
The recirculating loop of Fig. 19.6 contains several spans of fiber with amplifiers.
In the 40 Gb/s experiments discussed in this section, these amplifiers are comprised
692 L. Nelson and B. Zhu

of an EDFA with Raman counterpumps or the amplifiers are all-Raman, where co-
and counterpumps at multiple wavelengths provide all the gain. For single-fiber type
spans (i.e., SSMF or NZDF), DCF is also required in the amplifiers. This DCF can be
located between the two stages of the EDFA or at the input to the EDFA. In addition,
dynamic gain equalizers (GE) (see, e.g., [84] or [85]) are usually required in the loop
to achieve loop gain equalization. Figure 19.6 shows splitting of the C- and L-bands
and a separate GE for each band, followed by additional C- and L-band EDFAs to
compensate for the loss of the GE and loop components. In the following, descriptions
of the specific experiments stress the unique components or amplification techniques
used and the results.
After circulating in the loop for the targeted transmission distance, the optical
signals are switched out and sent to the receiver for error analysis. The signal bands
can be split, as shown in Fig. 19.6, before individual channels are selected by a WGR
and optical switch (SW) and sent to the preamplifier. Postdispersion compensation is
often required. This can be a fixed amount of dispersion compensation (i.e., a spool of
DCF) for all channels, if the dispersion slope of the transmission fiber and inline DCFs
match sufficiently well. Or, the postcompensation can be optimized on a channel-by-
channel basis using a tunable dispersion compensator [86–88] or different lengths of
DCF. After preamplification and filtering to suppress the ASE, the signal is sent to a
40 Gb/s (42.7 Gb/s) receiver comprised of a clock and data recovery, as well as a 1:4
demultiplexer to enable bit error rate measurements on the 10 Gb/s tributaries. 40 Gb/s
clock recovery is usually accomplished using a phase-locked loop or by selecting out
the 40 Gb/s component using a high-Q filter. In some experiments, the demultiplexing
is done optically, where an electroabsorption modulator (or lithium niobate modulator)
is used as an optical gate to choose a 10 Gb/s tributary [75] that is then sent to a 10
Gb/s receiver. More recently, commercial electronic demultiplexers (40 to 10 Gb/s)
have become available and are now frequently used in laboratory experiments. When
FEC encoding is done on the 9.953 Gb/s signals at the transmitter, the 10.664 Gb/s
electrical signals are sent to the FEC decoder before BER measurement. The BER
of each channel can be measured twice, once with the FEC decoder turned off (to
measure the “line BER” or “uncorrected BER”), and a second time with the FEC
decoder turned on (to measure the “corrected BER”). For proper analysis of the
system performance, the BERs of all four 10 Gb/s tributaries should be measured.
However, in loop experiments, during each gating period the clock recovery must
regain phase lock and thus it randomly selects a tributary. The average performance
of all four tributaries is then ascertained when the measurement time is set sufficiently
long to include a large number of gating periods.
In properly designed systems, the longest transmission distance is achieved when
the signal power is chosen to balance the OSNR impairment with the nonlinear im-
pairments. In single-channel systems, the optimum launch power for the longest reach
usually can be found fairly easily. An illustrative example is shown in Fig. 19.7 from
a single-channel 40 Gb/s transmission experiment using hybrid Raman/EDFAs and
100 km spans of NZDF [89]. The plot shows the range of allowed launch powers
for Q-factors greater than 15.6 dB (i.e., BER >10−9 ) versus transmission distance.
The Raman gain in the spans was kept constant as the launch power and distance
 19. 40 Gb/s Raman-Amplified Transmission 693

Pmax

Signal-Power Limits Pmin

10

Launched Signal Power (dBm) 8

–2

–4

–6
0 400 800 1200 1600 2000 2400
Distance (km)

Fig. 19.7. The range of allowed launch powers for Q-factors greater than 15.6 dB versus
transmission distance for the single-channel, 40 Gb/s experiment described in [89]. (Adapted
and reprinted with permission from the Optical Society of America.)

were varied. The lower curve indicates the launch power at which the performance is
OSNR-limited, and the upper curve indicates the launch power at which performance
is limited by nonlinearities. In this experiment, a launch power of +2 dBm resulted
in the longest transmission. In multichannel systems, however, the situation is more
complicated due to gain ripple from the Raman amplification and EDFAs. As the
gain ripple accumulates after transmission over multiple spans, the high-power chan-
nels can be limited by nonlinearities, whereas the low-power channels are limited by
OSNR. Both the nonlinearity and the span noise figure are determined by factors such
as the proportion of Raman gain in the transmission fiber versus EDFA gain or the
ratio of copumped gain to counterpumped gain in all-Raman systems. In WDM ex-
periments, therefore, a complicated optimization must be done involving adjustment
of the launch powers of the signals as well as Raman gain from the various pumps
to achieve a balance between the OSNR and nonlinearities on average, across all the
WDM channels.

19.3.2. 40 Gb/s Terrestrial Transmission Using Hybrid Raman/EDFAs

The subject of hybrid Raman/EDFAs, their design and performance, has been covered
in detail in Chapter 13. This section concentrates on high-capacity 40 Gb/s exper-
iments that have employed hybrid Raman/EDFAs for C- and L-band transmission
694 L. Nelson and B. Zhu

over terrestrial (80 to 100 km) spans. The section is divided according to whether
the experiments used conventional single-fiber-type spans with lumped dispersion-
compensating modules or dispersion-managed fiber spans.

19.3.2.1. Single-Fiber-Type Spans

The use of Raman amplification in 40 Gb/s terrestrial systems was first demonstrated
in 1999 with the straightline transmission of forty 100 GHz spaced C-band channels
over four 100 km spans of nonzero dispersion fiber (TrueWave RS fiber) [10]. Due to
the 40 Gb/s line rate, NRZ modulation format, and the early 40 Gb/s electronics with
limited bandwidth at that time, a received OSNR of 29 dB was required to obtain a
BER of less than 10−9 . With 21 dB loss for the 100 km spans and a launch power
of −1 dBm per channel, Eq. (19.1) shows that an amplifier noise figure of better
than 1 dB was necessary for a 400 km reach. Note that Eq. (19.1) only includes ASE
noise. Transmission impairments such as dispersion, PMD, or nonlinearities require
an even higher OSNR at the receiver and thus a lower amplifier noise figure. Raman
amplification was thus a requirement for transmitting over several terrestrial-length
spans at 40 Gb/s. The experiment of Nielsen et al. [10] used hybrid Raman/EDFAs
with a counterpropagating pump at 1450 nm, where the pump was a cladding-pumped,
cascaded Raman resonator fiber laser [90]. Due to the high Raman gain efficiency of
the transmission fiber, a peak Raman gain of 23 dB at 1550 nm was obtained with
520 mW of pump power. Figure 19.8 shows a schematic of the hybrid Raman/EDFA
along with the measured gain flatness and effective noise figure [26]. The hybrid
amplifiers consisted of distributed Raman amplification in the transmission fiber for
the first stage, followed by two erbium-doped fiber stages. A dispersion-compensating
fiber module was inserted between the Raman stage and the first EDFA stage, and a
gain-flattening filter (GFF) was placed before the second EDFA stage. The composite
gain ripple of less than 1 dB and the effective noise figure of −0.3 to −1.5 dB resulted
in OSNRs greater than 30 dB (0.1 nm RBW) and BERs of less than 1 × 10−10 for all
forty channels after 400 km transmission.
The first experiment to combine dual C+L-band transmission, distributed Raman
amplification, and 40 Gb/s line rates was the transmission of 3.28 Tb/s total capacity
over 3 × 100 km of nonzero dispersion fiber [40]. Dual-band hybrid Raman/EDFAs
were used to amplify the 40 C-band and 42 L-band NRZ channels, as shown in
Fig. 19.9. These amplifiers consisted of counterpumps at 1447 and 1485 nm for
distributed amplification in the transmission fiber, followed by separate two-stage
gain-flattened EDFAs for the C- and L-band channels. Pump powers of 600 mW at
1447 nm and 220 mW at 1485 nm provided peak Raman gains of 25 and 24 dB for
the C- and L-band channels, respectively. Also shown in Fig. 19.9 are the gain and
effective noise figure of the hybrid amplifiers. The better effective noise figure of
the L-band channels is a result of the C-band pump providing gain for the L-band
pump, thus enabling the L-band pump to penetrate farther into the transmission span
and provide gain for the L-band channels closer to the span input. In addition, the
L-band channels experience less loss through the fiber when the C-band channels are
present due to stimulated Raman scattering [44]. After transmission over the three
 19. 40 Gb/s Raman-Amplified Transmission 695

DCF EDFA EDFA

WDM
Stage 1 Stage 2
Coupler
GFF
Transmission
Fiber
1450nm Raman Pump

(a)

2
Gain Flatness
(dB)

Effective Noise Figure

–2
1530 1540 1550 1560
Wavelength (nm)
(b)

Fig. 19.8. (a) Schematic diagram of the C-band hybrid Raman/EDFA used in the experiment
demonstrating transmission of 1.6 Tb/s over 400 km of TrueWave® RS fiber [10]. (b) Gain
flatness of hybrid Raman/EDFA at output of EDFA stage 2 and effective noise figure referenced
to the end of the transmission fiber.

spans the OSNRs were greater than 29 dB for all 82 channels, and the maximum
BER was 8 × 10−11 (without FEC). The authors attributed the limitation to be OSNR
in the C-band and cross-phase modulation in the L-band. In this experiment, the
transmission fiber was a prototype TrueWave fiber with dispersion of 5.7 ps/nm/km
and dispersion slope of 0.037 ps/nm2 /km, both at 1550 nm. The low RDS of the
transmission fiber enabled good slope match with the high-slope DCF (HS-DCF)
resulting in dispersion variations of 60 and 140 ps/nm over the C- and L-bands,
respectively, after transmission. The 60 ps/nm variation in accumulated dispersion in
the C-band was within the 1 dB dispersion tolerance for 40 Gb/s NRZ signals, and
thus per-channel postdispersion trimming was not required. Further improvement of
the RDS of DCF has enabled 500 km, 40 Gb/s transmission over TrueWave RS fiber
with no per-channel postcompensation [91].
The first multi-Tb/s transmission of 40 Gb/s channels over more than 1000 km
of fiber used similar dual-band, hybrid Raman/EDFAs [42]. The demonstration of
3.08 Tb/s capacity (77 channels each at 42.7 Gb/s) was done with a recirculating
loop setup close to that shown in Fig. 19.6. FEC (7% overhead Reed–Solomon) was
implemented to achieve error-free operation of all 77 NRZ channels after twelve
100 km spans of prototype TrueWave fiber. Figure 19.10 shows the measured on-off
696 L. Nelson and B. Zhu

C / L band hybrid Raman/EDFA

C-band
Transmission HS-DCF EDFA
Fiber P/S C/L C/L
WDM WDM WDM
HS-DCF

P/P C-band pump

WDM L-band
L-band pump
EDFA

(a)

30 5
Gain (dB)

NF (dB)
25 0

20 –5
1520 1540 1560 1580 1600
Wavelength (nm)
(b)

Fig. 19.9. (a) Diagram of the C+L-band hybrid Raman/EDFA used in the experiment demon-
strating transmission of 3.28 Tb/s over 3 × 100 km of NZDF [40]. (P/S WDM: pump–signal
combiner, P/P WDM: pump–pump combiner, C/L WDM: C–L band combiner). (b) Gain flat-
ness of the hybrid Raman/EDFA at amplifier output and effective noise figure referenced to the
end of the transmission fiber. (Reprinted with permission from the Optical Society of America.)

Raman gains with only the 1447 pump on, only the 1485 nm pump on, and both pumps
on. The large increase (>10 dB) in the L-band channel powers when both pumps are
on shows the effect of the C-band pump providing gain for the L-band pump. In
this experiment, dynamic gain equalizers placed after the three spans in the loop
were important elements in minimizing the accumulated gain ripple. Improvements
in the dispersion slope match between the transmission fiber and the high-slope DCF
modules resulted in dispersion variations of 47 ps/nm across the C-band and 61
ps/nm across the L-band after 1200 km, and no per-channel postcompensation in the
C-band was required. At the time, this experiment probed the limit for the capacity
× distance product at 40 Gb/s possible with simple NRZ modulation format, hybrid
Raman/EDFAs with counterpumping only, and standard 7% overhead FEC.
 19. 40 Gb/s Raman-Amplified Transmission 697

30
1447 & 1485 nm Pump

25

On-off Raman Gain (dB)

20

1447 nm Pump
15

1485 nm Pump
10

0
1525 1545 1565 1585 1605
Wavelength (nm)

Fig. 19.10. The measured on-off Raman gains with only the 1447 pump on, only the 1485 nm
pump on, and both pumps on for the experiment reported in [42]. (Adapted and reprinted with
permission from the Optical Society of America.)

Several other hybrid Raman/EDFA experiments have demonstrated the use of

Raman amplification to increase the amplified span lengths beyond 100 km. At 10
Gb/s, Terahara et al. [92] showed transmission of 128 channels over six 140 km
spans of standard single-mode fiber (30.4 dB loss) with dual-band distributed Raman
amplification and average Raman gains of 5.8 and 5.4 dB in the C- and L-bands,
respectively. Srivastava et al. [93] transmitted one hundred, 25 GHz spaced, 10 Gb/s
channels in the L-band over three spans of 125, 132, and 140 km of TrueWave fiber.
The hybrid amplifiers used a single pump wavelength at 1480 nm, achieving a peak
gain of 10 dB at 1585 nm, and the launch power per channel was −3 dBm. In addition,
transmission of thirty-two 10 Gb/s, C-band channels over 750 km of fiber with 250
km spans was also demonstrated [94]. This experiment was targeted at a medium-haul
submarine system with land-based optical amplifiers or coastal festoons. One-watt
Raman fiber lasers at 1455 nm were used to achieve Raman preamplification in the
pure silica-core fiber spans having 44.5 dB loss.
Transmission over multiple long spans has also been achieved at 40 Gb/s using
Raman amplification. An experiment demonstrating a 1000 km reach for 32 C-band
channels with five spans of 160 km and one span of 200 km was reported by Zhu et
al. [95]. Hybrid Raman/EDFAs with pumps at 1427 and 1455 nm were used, and the
Raman gain profile was tailored to minimize the accumulated gain tilt from the ED-
FAs. The CSRZ modulation format allowed a launch power of +5 dBm/channel into
the NDSF (nondispersion-shifted fiber) spans, and analysis showed that the system
698 L. Nelson and B. Zhu

reach at 1000 km was dominated by ASE accumulation. Raw Q-factors of 10.1 to

13.1 dB were measured for the 40 Gb/s channels; FEC was not implemented.
A combination of advanced modulation formats and hybrid Raman/EDFAs has
been employed to transmit spectral efficiencies greater than 0.4 bit/s/Hz at 40 Gb/s
over terrestrial single-fiber type spans. The first demonstration of 0.64 bit/s/Hz spectral
efficiency used VSB filtering and a special wavelength allocation to transmit 128
channels over 300 km of fiber [41]. The 40 Gb/s NRZ channels were alternately spaced
by 50 and 75 GHz, and in the receiver a given channel was selected with a narrow
filter (3 dB bandwidth = 30 GHz) that was tuned from the channel’s center frequency
to isolate the sideband experiencing the smallest overlap with adjacent channels.
Subsequent studies have discussed optimization of the VSB filtering technique [96,
97]. The C- and L-band channels were polarization-interleaved (adjacent wavelength
channels have orthogonal polarizations) to reduce interchannel nonlinearities and
linear crosstalk from adjacent channels and transmitted over three 100 km spans of
Teralight1 fiber. In a similar experiment with a recirculating loop, 125 NRZ channels
at 42.7 Gb/s were transmitted over 12 × 100 km of Teralight fiber [98] with VSB
filtering. The erbium-doped fiber length in the hybrid amplifier was adjusted to deliver
higher gain to the lower-wavelength regions of the C- and L-bands to correct for the tilt
from stimulated Raman scattering. In addition, the DCF modules were counterpumped
with two wavelengths per band to mask the DCF loss and improve the noise figure
of the EDFAs. After transmission, OSNRs of better than 22.1 dB (0.1 nm RBW) and
BERs better than 1.4 × 10−4 were measured. The performance of the worst channels
was attributed to optical fiber nonlinearities.
Spectral efficiencies of 0.8 bit/s/Hz for transmission over terrestrial single-fiber
spans have been achieved using PSBT format or bandwidth-limited modulation for-
mats. In one experiment, eighty 40 Gb/s channels were placed on the 50 GHz grid
in the C-band by using polarization-interleaving (PI) and PSBT format, where the
bandwidth-limiting occurs in the electrical domain [69]. Although PSBT has dis-
advantages including a reduced extinction ratio and increased sensitivity to OSNR,
transmission over three 100 km spans of Teralight fiber was achieved. Distributed
Raman amplification providing 14 dB gain in the transmission fiber and 6 dB gain
in the DCF was a key to maintaining OSNRs greater than 29.5 dB (0.1 nm RBW).
Polarization-demultiplexing was not used at the receiver; instead an interleaver sepa-
rated the odd and even channels, followed by a tunable filter with 0.24 nm full-width–
half-maximum (FWHM). The 10 Gb/s tributaries (211 − 1 PRBS) had BERs less than
9 × 10−9 , attesting to the difficulties of implementing a duobinary format and the tight
channel spacing. Another demonstration of 0.8 bit/s/Hz spectral efficiency employed
vestigial sideband NRZ format with optimized filtering [99]. Eighty 42.7 Gb/s, C-band
channels at 50 GHz spacing were transmitted over 900 km of fiber by using narrow
periodic filters to clip the NRZ-modulated spectra and provide VSB filtering at the
transmitter side, as well as at the receiver side. An extended transmission distance of
21 × 100 km with 0.8 bit/s/Hz spectral efficiency was achieved using BL-PSBT [64].
As shown in [71], a narrow optical filter after the low-pass filtered duobinary mod-
1 Teralight is a trademark of Alcatel S.A.
 19. 40 Gb/s Raman-Amplified Transmission 699

ulator improves the tolerance to noise and also allows denser channel spacing. By
employing narrow filters in the transmitter and receiver, 159 C- and L-band channels
at 42.7 Gb/s were polarization interleaved and spaced by 50 GHz for 2100 km trans-
mission [64]. The authors reported that the neighboring 50 GHz spaced, BL-PSBT
channels caused less than a 0.3 dB penalty due to linear and nonlinear crosstalk, as
compared to their previous experiment with bandwidth-limited NRZ [99]. In both
experiments first- and second-order Raman amplification [100–102] were used in the
transmission fiber with four first-order counterpumps and a 1346 nm Raman fiber
laser counterpump acting as a secondary pump to the lower-wavelength 1427 and
1439 nm pumps. About 1 dB OSNR improvement was attributed to the second-order
pump. After the 2100 km transmission the OSNRs of the 159 WDM channels were at
least 19.4 dB (0.1 nm RBW), and the BERs were better than 7 × 10−4 . It was assumed
that this represented a 1 dB Q-factor margin above the concatenated FEC [247, 239]
+ [255,247] threshold for BERs less than 10−13 .
Spectral efficiencies of 1.28 bit/s/Hz have been demonstrated using polarization-
division-multiplexing (PDM) in combination with VSB filtering. A record capacity
of 10.2 Tb/s (256 channels at 42.7 Gb/s) was transmitted over 100 km [103] and then
over 300 km [104] of Teralight fiber. The 256 C- and L-band channels were composed
of 128 wavelengths (alternately spaced by 50 and 75 GHz) with two orthogonally
polarized channels per wavelength. PDM with a polarization-demultiplexer at the re-
ceiver suffers from low tolerance to PMD [105–107], and thus the transmission reach
is limited. PMD causes the output polarization to vary with optical frequency, result-
ing in PMD-induced coherent crosstalk at the receiver [107]. Raman amplification
was necessary even for the 100 km transmission distance to maintain sufficiently high
OSNR to overcome the linear and PMD-induced crosstalk penalties. Counterpumped
Raman amplification in the transmission fiber, as well as in the DCF modules for
pre- and postdispersion compensation, resulted in OSNRs greater than 28.4 dB (0.1
nm RBW) for all channels and BERs less than 1 × 10−4 for all channels that were
measured. To achieve the 300 km transmission distance, several improvements were
made. Analysis of the wavelength allocation revealed that a shift of 40 GHz between
the carriers of the orthogonally polarized channels would minimize the polarization
crosstalk. With this shift, the channels were, in effect, polarization-interleaved, and
the carriers of the adjacent orthogonally polarized channels were located as far as pos-
sible from the filter center. On the transmitter side, tight filtering was implemented
after the modulators, where a narrow periodic filter was inserted to limit the channel
bandwidth and reduce linear crosstalk, and VSB filtering was done at the receiver. In
addition, second-order Raman amplification was used, along with the four first-order
Raman pumps. Adding the 1346 nm counterpropagating pump for each of the three
transmission spans resulted in a reported 1 dB improvement in the received OSNR.
After 300 km transmission, the OSNRs were greater than 27 dB (0.1 nm RBW), and
the 42.7 Gb/s uncorrected BERs were better than 1 × 10−4 . Note that PDM with
polarization-demultiplexing will require fast polarization tracking in the receiver for
installed systems, creating challenges both in terms of implementation and system
costs.
700 L. Nelson and B. Zhu

19.3.2.2. Dispersion-Managed Fiber Spans

Section 19.2.2.2 and Fig. 19.4 introduced the concept of dispersion-managed spans,
where both positive and negative dispersion fibers are incorporated into the transmis-
sion path. In addition to the advantages of excellent dispersion slope match across
a broad bandwidth and elimination of lumped DCMs, dispersion-managed spans in
the ABA configuration have important advantages for terrestrial transmission using
hybrid Raman/EDFAs or all-Raman amplification. In this subsection we focus on
hybrid amplification for 10 Gb/s and 40 Gb/s systems.
The first large-scale WDM experiment using terrestrial-length dispersion-
managed spans was the transmission of 211 C- and L-band, 10.7 Gb/s channels over
7221 km of fiber [51]. The 80 km AB spans consisted of 57 km of positive dispersion
fiber and 23 km of negative dispersion fiber (effective areas were not specified). For
the same launch power per channel, 80 km spans reduce the path-average power and
the number of repeaters when compared to 40 km repeater spacing, thus reducing
the effect of fiber nonlinearities. However, Raman gain is necessary to maintain the
OSNR. In the experiment, the hybrid Raman/EDFAs had a similar configuration to
those shown in Fig. 19.9(a), except DCF was not necessary. Four counterpropagating
pumps with wavelengths between 1430 and 1505 nm were coupled into the −D fiber,
providing Raman gain for the 64.4 nm signal band. Due to the high gain in the small
Aeff , −D fiber, a total of only 180 mW was required to obtain an average Raman gain
of 7 dB, resulting in an OSNR improvement of about 3.5 dB. The 37.5 GHz spaced,
RZ channels were polarization-interleaved, and the residual dispersion of the spans
was −1.5 ps/nm/km to reduce CPM-induced waveform distortion, a particularly dif-
ficult impairment for 10 Gb/s. After 7221 km transmission, uncorrected BERs for all
channels were between 3 × 10−4 and 8 × 10−6 , and the authors identified OSNR, and
not nonlinearities, as the transmission limitation.
Transmission of 10 Gb/s channels over 100 km ABA spans has been demonstrated
using only Raman amplification to compensate for the fiber loss [52] and also with
hybrid Raman/EDFAs [53]. The latter experiment showed transmission of eighty
10.7 Gb/s, 50 GHz spaced, C-band channels over 5200 km using the same hybrid
Raman/EDFAs as in Fig. 19.8, but without the DCF. TheABAspans consisted of 38 km
of UltraWaveTM SLA, 35 km of UltraWaveTM IDF, and 27 km of SLA. This scheme
distributes the gain more evenly in the span compared to an AB fiber map, yielding
better noise performance of the system. Due to the slope match of the SLA and IDF
fiber, the accumulated residual dispersion difference across the C-band channels was
less than 260 ps/nm after transmission over 5200 km. No pre- and no postdispersion
compensation were used, and no per-channel dispersion trimming was required at the
receiver. All eighty channels had uncorrected BERs less than 1 × 10−4 , and with the
FEC enabled, no errors were detected when counting more than 1011 bits.
Transmission simulations have predicted that due to their lower span noise fig-
ure (defined as the combined noise figure of the entire span consisting of transmis-
sion fiber, discrete and/or distributed amplification, and passive components), ABA
dispersion-managed spans offer a 2 to 3 dB increase in transmission reach over single-
 19. 40 Gb/s Raman-Amplified Transmission 701

fiber type spans. This prediction has been roughly verified with 10 and 40 Gb/s exper-
iments, where similar hybrid Raman/EDFAs were used in all cases. The experiment
reported in [108] where 80 × 10.66 Gb/s NRZ channels were transmitted over 3200
km using 100 km NZDF spans was repeated using the same setup and hybrid am-
plifiers, but with the NZDF spans replaced by ABA dispersion-managed fiber spans.
With the NRZ modulation format, the transmission distance was increased to 4800
km, and for 50% duty-cycle RZ, a distance of 5200 km was achieved [53], showing
about 2 dB reach increase over NZDF.
Transmission of forty 42.7 Gb/s C-band channels over 2000 km of fiber [54]
and, later, 2400 km of fiber [57] was demonstrated using 100 km dispersion-managed
spans, consisting of the fiber lengths shown in Fig. 19.4(a), and hybrid Raman/EDFAs.
The calculated signal power evolution for Raman gain of 6 dB below transparency
is shown in Fig. 19.11(a), where it is apparent that the signal power is lowest in the
small Aeff IDF fiber, thus minimizing nonlinearities [109]. Figure 19.11(b) shows the
calculated span noise figure as a function of the starting position of the IDF fiber in
the ABA map. A more than 2 dB improvement in the span noise figure is produced
by placing the IDF near the middle of the span. The minimum noise figure appears to
occur over more than a 10 km range for the IDF starting position, thus ensuring that the
exact placement of the IDF is not critical. In the experiment of [57], the forty C-band
channels were modulated with the CSRZ format, the average on-off Raman gain was
15.5 dB, and the launched power per channel was −1 dBm. BERs for all 42.7 Gb/s
channels were better than 1 × 10−4 , which would correspond to a BER performance
well below 10−12 with standard 7% overhead Reed–Solomon FEC. This experiment
provides some verification of the 2 to 3 dB advantage of dispersion-managed spans
at 40 Gb/s if its 2400 km reach (for the C-band) is compared to the 1200 km reach
for the C- and L-bands over NZDF [42].

19.3.3. 40 Gb/s Terrestrial Transmission Using All-Raman Amplification

Before the early 1990s when EDFAs emerged as practical optical amplifiers for fiber
communication systems, distributed Raman amplification was investigated for soli-
ton transmission in simulations [110] and in experiments [111]. As interest in Raman
amplification has grown, due to its improved noise performance, for 40 Gb/s terres-
trial and ultra-long-haul transmission, a number of experiments have been reported
demonstrating the use of Raman amplification only, or all-Raman amplification, to
compensate for the transmission fiber and DCF loss. Many of these experiments have
used both co- and counterpumping. Critical issues of copumping including pump–
signal crosstalk, signal–pump–signal crosstalk, polarization-dependence of the gain,
and pump laser requirements have been covered in Chapters 5, 8, and 9. In addition,
design of pumped DCFs for discrete Raman amplifiers has been considered in Chap-
ter 6. This section describes several 40 Gb/s WDM transmission experiments using
all-Raman amplification. The focus is primarily on terrestrial span lengths for both
single-fiber-type spans as well as dispersion-managed spans.
702 L. Nelson and B. Zhu

–2 42.5km 34.5km 23km

SLA IDF SLA

–4
Signal Power (dBm)

–6

–8

–10

–12
0 20 40 60 80 100
Distance (km)
(a)

X km 34.5km (65.5 — X )km

SLA IDF SLA Pump
18-dB Gain
19.5

19.0
Span Noise Figure (dB)

18.5

18.0

17.5

17.0
0 10 20 30 40 50 60 70
Start Position of IDF, X(km)
(b)

Fig. 19.11. (a) The calculated signal power evolution for Raman gain of 6 dB below transparency
for the experiment reported in [54]. (b) The calculated span noise figure as a function of the
starting position of the IDF fiber in the ABA map for the same experiment. (Adapted and
reprinted from [109] with permission from the Optical Society of America.)
 19. 40 Gb/s Raman-Amplified Transmission 703

19.3.3.1. Single-Fiber-Type Spans

Several 10 Gb/s WDM experiments using all-Raman amplified spans consisting of a

single fiber type were reported prior to work at 40 Gb/s. For submarine applications,
Nissov et al. [112] demonstrated the first transoceanic-length transmission based
on Raman gain with 100 Gb/s (10 × 10 Gb/s) capacity over 7200 km of fiber. The
45 km dispersion-shifted fiber spans were counterpumped to transparency. Raman
amplification provided a 2 dB noise performance improvement compared to 980 nm
pumped EDFAs, assuming equal path-averaged powers. In this experiment, as in most
of the experiments described below, EDFAs were included at the end of the loop to
compensate for the loss of the gain equalizers and loop-specific elements. Copumping
was first reported in an experiment demonstrating 1 Tb/s transmission over four 80
km spans of dispersion-shifted fiber [113]. With 4 dB of forward gain and 18 dB of
backward gain at 1550 nm, the launch power for the 100, 25 GHz spaced channels
could be kept low enough (−18 dBm/ch) that four-wave-mixing was not observed,
even in the vicinity of the average zero dispersion wavelength.
All-Raman amplification of a single, 53 nm extended L-band was shown first
for transmission of 128 × 10 Gb/s over 4000 km of TrueWave RS fiber [114]. Fig-
ure 19.12 shows a typical pumping configuration. The 100 km spans were co- and
counterpumped for a total of 20 dB on-off gain, and the slope-matched DCF mod-
ule following each span was counterpumped for 12 dB on-off gain. The 10 Gb/s
dispersion-managed solitons had BERs better than 10−6 after transmission with no
per-channel postcompensation.
At 40 Gb/s, transmission of 3.2 Tb/s capacity over twenty 100 km spans of True-
Wave REACH fiber was demonstrated using distributed Raman amplification only to
compensate for the transmission fiber and DCF loss [12]. The location of the transmis-
sion fiber’s zero dispersion wavelength at <1400 nm enabled co- and counterpump-
ing for flat Raman gain and noise figure across the entire C+L-bands by eliminating
pump–pump four-wave-mixing [34] and pump–signal four-wave-mixing [35]. The
low dispersion slope of the TrueWave REACH fiber also enabled C+L-band disper-
sion and dispersion slope compensation of each span with a single Raman-pumped
DCF module, as shown in Fig. 19.3. The on-off Raman gains from the three copump
wavelengths, four counterpump wavelengths, and five DCF pump wavelengths are

Transmission
Fiber DCF

Signals WDM WDM Isolator WDM Isolator

1% tap
Co-
Pumps Counter Counter Monitor
Pumps Pumps

Fig. 19.12. Typical pumping configuration for all-Raman amplification of a single-fiber type
transmission span with a DCF module. WDM: pump–signal combiner or circulator.
704 L. Nelson and B. Zhu

25

Span Total

20
Raman On-off Gain (dB)

15 Counter-Pump

10

DCF

Co-Pump

0
1530 1540 1550 1560 1570 1580 1590 1600
Wavelength (nm)

Fig. 19.13. The on-off gains for a Raman-pumped 100 km NZDF span and DCF module for the
experiment reporting 3.2 Tb/s transmission over twenty 100 km NZDF spans [12]. (Reprinted
with permission from the Optical Society of America.)

shown in Fig. 19.13. Although in the experiment the transmitters, receivers, and dy-
namic gain equalizers required splitting into separate C- and L-wavelength bands, the
continuous Raman gain and single DCF module showed the potential for transmitting
a wide, continuous single band, and eliminating the cost and loss of band splitters.
Using a similar extended L-band as in Grosz et al. [114], a record-breaking 2.56
Tb/s (64 channels at 42.7 Gb/s) was transmitted over 4000 km of TrueWave RS fiber
[11]. This experiment was the first to demonstrate a 40 Gb/s RZ-DPSK modulation
format in transmission and used a balanced receiver, resulting in a ∼3 dB improvement
in receiver sensitivity. With only counterpumping of the 100 km spans and DCF
modules, after transmission the OSNRs of the 64 channels were 15.2 dB or higher,
and with the balanced receiver the uncorrected BERs were 1.2 × 10−4 or better.
The performance of standard Reed–Solomon FEC for the different noise statistics of
DPSK was verified.
In addition, all-Raman amplification and an extended L-band were employed to
transmit 5.12 Tb/s over 1280 km of SSMF [115]. The 80 km spans were pumped
for 4 dB of forward gain and a total of ∼16 dB on-off gain, and DCF modules were
counterpumped for 12 dB on-off gain. The spectral density of 0.8 bit/s/Hz, composed
of 42.7 Gb/s channels on the 50 GHz grid, was achieved by strong filtering of the
CSRZ signals before and after transmission and by polarization-interleaving at the
transmitter. The ability of the CSRZ signal to maintain a good waveform even in
the presence of strong optical filtering allowed better nonlinear performance and
 19. 40 Gb/s Raman-Amplified Transmission 705

–25
Input Spectrum
Channel #64 Channel #65 Channel #66

Arb. Units (dB)

–30

–35

–40

–45
1579.0 1579.5 1580.0 1580.5 1581.0 1581.5 1582.0
(a)

–30
–35
Demultiplexed Channel
Arb. Units (dB)

–40
–45
–50
–55
–60
1579.0 1579.5 1580.0 1580.5 1581.0
Wavelength (nm)
(b)

Fig. 19.14. (a) A small section of the input spectrum and (b) the spectrum of a demultiplexed
channel (in back-to-back) along with its eye diagram for the experiment demonstrating 40 Gb/s
CSRZ channels at 50 GHz spacing [115]. (Courtesy of Diego Grosz, Lucent Technologies.)

received sensitivity than other modulation formats when placed on the 50 GHz grid.
Figure 19.14 shows part of the input spectrum and the spectrum of a demultiplexed
channel along with its eye diagram.

19.3.3.2. Dispersion-managed spans

The longest WDM transmission distances at 10 and 40 Gb/s have been achieved using
a combination of dispersion-managed spans and all-Raman amplification. Knudsen
et al. [52] first investigated the possibility of extending the dispersion-managed span
length to 100 km with the transmission of forty-two 10 Gb/s channels over 4000
km. The ABA spans were counterpumped to transparency by a single-wavelength
Raman pump, and BERs less than 10−9 were achieved without FEC and without per-
channel pre- or postcompensation. The gain flatness achievable with a single pump
wavelength, and not the dispersion slope match of the SLA and IDF fibers, limited
the bandwidth of the system to the forty-two channels. All-Raman amplification over
80 km, ABA dispersion-managed spans has also been reported as part of a 10 Gb/s
network demonstration with optical add-drop multiplexers [116].
706 L. Nelson and B. Zhu

An important experiment demonstrating that all-Raman amplification can enable

a broad continuous signal bandwidth was reported by Shimojoh et al. [117]. A single
74 nm band from 1536 to 1610 nm supported 240 × 10 Gb/s channels and was 1.8
times broader than previous terabit transmission experiments. The 40 km, AB spans
were counterpumped with four pump wavelengths, and the accumulated gain ripple
in the loop was equalized by a concatenation of two Mach–Zehnder optical filters.
Two broadband discrete Raman amplifiers compensated for the loop-specific and
gain-equalizer losses. FEC with 20% overhead enabled error-free transmission of all
channels over 7400 km.
For 40 Gb/s submarine transmission, experiments have demonstrated the use of
so-called quadruple hybrid (ABABABAB) and inverse double hybrid spans (BABA)
for transatlantic and transpacific distances, respectively [50, 60]. The goal of these
span configurations is to reduce the waveform distortion induced by the interaction
of self-phase modulation with group-velocity dispersion. By alternating the sign of
the dispersion several times in the 50 km spans, the accumulated dispersion along the
transmission span is reduced. In addition, counterpumping with several sections of
the small effective area fiber interspersed in the spans allows the gain to be distributed
more evenly and can improve the span noise figure. Less input power is then required
to obtain a certain OSNR, reducing self-phase modulation and the resulting nonlinear
waveform distortion. The inverse double-hybrid spans of Sugahara et al. [60] were
composed of a new pure-silica-core fiber (PSCF) with 195 µm2 effective area (the A
fiber) and DCF with 19 µm2 effective area and −63 ps/nm/km dispersion (the B fiber).
With counterpumping of the 50 km BABA span, the signal power variation along the
span was only 3.5 dB, compared to 5 dB in the quadruple hybrid span with the new
PSCF. Simulations showed that the eye-opening penalty after 9000 km transmission
was smallest for the inverse double hybrid span, despite the fact that the signals were
launched into the DCF and the accumulated dispersion was higher in the BABA spans.
As mentioned previously, it is not yet clear how much accumulated dispersion a 40
Gb/s system can tolerate. This value could depend on the modulation format and other
details of the system. In fact, transmission over transoceanic distances (6200 km) has
been achieved using 51 km AB spans [118], where the accumulated dispersion within
each span was two to four times higher than that in the inverse double or quadruple
hybrid spans.
All-Raman amplification and dispersion-managed spans have been combined in
several experiments to meet the stringent requirements of ultra-long-haul 40 Gb/s
transmission with terrestrial span lengths. Forty C-band CSRZ channels were trans-
mitted over 3600 km using 100 km UltraWave fiber spans with the properties listed
in Table 19.2 [55]. Each symmetrical ABA span was co- and counterpumped to 3.5
dB above transparency to accommodate the WDM coupler/circulator and fixed gain-
flattening filter inserted after the span. A semiperiodical dispersion map was used,
where each span was undercompensated and an extra spool of IDF was inserted after
four spans, resulting in −100 ps/nm accumulated dispersion per loop round-trip. A
later experiment by the same authors extended the transmission distance to 5200 km
using the same 100 km dispersion-managed spans [119]. Enhanced FEC increased the
 19. 40 Gb/s Raman-Amplified Transmission 707

Total
20 20
Raman On-off Gain (dB)

Span Noise Figure (dB)

Counter-Pumps

15 15

10 10

5 5
Co-Pumps

0 0
1525 1545 1565 1585 1605
Wavelength (nm)

Fig. 19.15. The on-off Raman gains from the copumps and counterpumps, as well as the total
on-off gain and span noise figure for the experiment demonstrating 3.2 Tb/s transmission over
5200 km of dispersion-managed fiber [65].

coding gain to 7.8 dB at 10−12 BER with the same 7% overhead, assuming Gaussian
noise statistics. Back-to-back measurements showed that EFEC reduced the required
OSNR by more than 10 dB at 10−12 BER.
Ultra-long-haul 40 Gb/s terrestrial transmission over 5200 km has also been
achieved with RZ-DPSK modulation format and dispersion-managed spans [65].
The 3.2 Tb/s total capacity consisting of 80 C- and L-band channels was transmit-
ted over symmetrical 100 km UltraWave fiber spans, using co- and counterpumping.
Figure 19.15 shows the Raman on-off gains from the copumps and counterpumps,
as well as the total on-off gain and span noise figure. The higher copumped gain in
the lower C-band helped to partially offset the stimulated Raman scattering-induced
power transfer to the L-band. This experiment showed the excellent slope-match
achievable with these dispersion-managed spans, as the residual dispersion variation
was only ±30 ps/nm from 1530 to 1603 nm after the 400 km loop. The RZ-DPSK
format and balanced receiver enabled the 42.7 Gb/s channels to have BERs better
than 1 × 10−4 after 5200 km transmission. It is clear that by applying stronger FEC
and improving the transmitter waveforms and receiver sensitivities, longer 40 Gb/s
transmission distances will be possible.
Table 19.3 summarizes the transmission experiments discussed in Section 19.3,
both for hybrid Raman and all-Raman amplification, along with several other impor-
tant 40 Gb/s long-haul experiments.
 708
Table 19.3. Selected WDM 40 Gb/s Raman-Amplified Transmission Experiments
Capacity- Ch. Span
Number Capacity Distance Density Fiber Amplifier
Date Distance Spacing Format Length BER Ref

L. Nelson and B. Zhu

of Ch. (Tb/s) (km) (bit/s/Hz) Type type
(Pb.km/s) (GHz) (km)
EDFA/
09/99 40 1.6 400 0.64 100 0.4 NRZ 100 NZDF 9e–11 [10]
Raman
EDFA/
03/00 82 3.28 300 0.98 100 0.4 NRZ 100 NZDF 8e–11 [40]
Raman
09/00 128 5.12 300 1.54 50/75 0.64 NRZ/ 100 NZDF EDFA/ 1e–9 [41]
VSB Raman
09/00 32 1.28 1000 1.28 100 0.4 CSRZ 160 SSMF EDFA/ 7e–4 [95]
Raman
NRZ/
EDFA/
03/01 256 10.24 100 1.02 50/75 1.28 VSB/ 100 NZDF Raman 1e–4(FEC) [103]
PDM
03/01 77 3.08 1200 3.7 100 0.4 NRZ 100 NZDF EDFA/ 1e–4(FEC) [42]
Raman
TDFAs/
NRZ/
03/01 273 10.92 117 1.28 50 0.8 PI 58 DMFS Raman + 1e–9 [120]
EDFAs
NRZ/
EDFA/
10/01 125 5.0 1200 6.0 50/75 0.64 VSB/ 100 NZDF 1e–4(FEC) [98]
Raman
PI
10/01 40 1.6 2000 3.2 100 0.4 CSRZ 100 DMFS EDFA/ 1e–4(FEC) [54]
Raman
10/01 80 3.2 300 0.96 50 0.8 PSBT/ 100 NZDF EDFA/ 9e–9 [69]
PI Raman
RZ-
03/02 64 2.5 4000 10 100 0.4 100 NZDF All Raman 1e–4(FEC) [11]
DPSK
 Table 19.3. Continued
Capacity- Ch. Span
Number Capacity Distance Density Fiber Amplifier
Date of Ch. (Tb/s) (km) Distance Spacing (bit/s/Hz) Format Length Type type BER Ref
(Pb.km/s) (GHz) (km)
NRZ/
EDFA/
03/02 256 10.2 300 3.06 50/75 1.28 VSB/ 100 NZDF Raman 1e–4(FEC) [104]
PDM
03/02 32 1.28 6050 7.74 100 0.4 CSRZ/ 52 DMFS All Raman 1e–4(FEC) [50]
PI
03/02 40 1.6 3600 5.76 100 0.4 CSRZ 100 DMFS All Raman 1e–4(FEC) [55]
03/02 80 3.2 2000 6.4 100 0.4 CSRZ 100 NZDF All Raman 1e–4(FEC) [12]
06/02 40 1.6 2400 3.84 100 0.4 CSRZ 100 DMFS EDFA/ 1e–4(FEC) [57]

19. 40 Gb/s Raman-Amplified Transmission

Raman
CSRZ/
07/02 32 1.28 9000 11.52 100 0.4 50 DMFS All Raman 4e–4 [60]
PI
PSBT/ EDFA/
07/02 80 3.2 900 2.88 50 0.8 100 NZDF 1e–4(FEC) [99]
PDM Raman
BL-
09/02 159 6.36 2100 13.36 50 0.8 PSBT/ 100 NZDF EDFA/ 7e–4(FEC) [64]
Raman
PI
RZ-
09/02 80 3.2 5200 16.64 100 0.4 100 DMFS All-Raman 1e–4(FEC) [65]
DPSK
09/02 40 1.6 5200 8.32 100 0.4 CSRZ 100 DMFS All Raman 1.5e–3 [119]
(FEC)
CSRZ/
09/02 128 5.12 1280 6.55 50 0.8 80 SSMF All Raman 1e–4(FEC) [115]
PI
VSB: vestigial sideband; RZ-DPSK: return-to-zero differential-phase- shift-keyed; PSBT: Phased-shaped binary transmission;
PDM: polarization division multiplexing; PI: polarization interleaving; NZDF: non-zero dispersion fiber;
DMFS: dispersion-managed fiber span; SSMF: standard single-mode fiber

709
710 L. Nelson and B. Zhu

105
Single fiber, Terrestrial
Dispersion-managed, Terr.
Single fiber, Submarine
Dispersion-managed, Subm.
Capacity x Distance (Tb*km/s)

104

103
0 10 20 30 40 50
Bit Rate (Gb/s)

Fig. 19.16. Summary of recent high-capacity ETDM transmission experiments (1999–2002)

plotted as the product of total capacity and distance versus the bit rate. Results are divided
into terrestrial span lengths (>70 km) and submarine span lengths (<70 km). Also indi-
cated is whether the transmission spans were of a single-fiber type with lumped dispersion-
compensating modules at the optical amplifiers or were dispersion-managed spans.

19.4. Conclusion
Figure 19.16 shows a plot of recent laboratory transmission experiments at 10, 20,
and 40 Gb/s using the product of total capacity and transmission distance as the
performance metric. The results are divided into terrestrial span lengths (>70 km)
and submarine span lengths (<70 km). The results are also grouped by whether the
transmission spans are of a single-fiber type with lumped dispersion-compensating
modules at the optical amplifiers or are dispersion-managed spans. It is clear from the
plot that submarine experiments have concentrated primarily on dispersion-managed
spans, whereas recent experiments have begun to assess the benefit of dispersion-
managed spans for terrestrial systems. The fact that submarine experiments at 10 Gb/s
lead those at 40 Gb/s may, in part, reflect the fact that work at 40 Gb/s for transoceanic
transmission has begun more recently than work at 40 Gb/s for terrestrial transmission.
The record of 16.6 petabit km/s for 40 Gb/s terrestrial dispersion-managed spans used
the RZ-DPSK format and a balanced receiver. It can be expected that this record will
be broken, perhaps in the very near future, by utilizing new modulation formats or
fibers and/or improvements in FEC.
In summary, high-capacity 40 Gb/s transmission systems offer scalable solutions
for future traffic growth in the core network. Raman amplification is likely to be a
 19. 40 Gb/s Raman-Amplified Transmission 711

key driver to ease the noise performance and increase the available bandwidth for 40
Gb/s DWDM systems. New fiber technologies provide high system performance and
enable a simple and cost-effective dispersion compensation scheme. More system
margin can also be expected from high-coding-gain FEC. Optimized modulation
formats and high-speed optoelectronics will make practical deployment of 40 Gb/s
DWDM systems possible, facilitating multiple terabit transmission over Mm distance
at low cost per bit per kilometer.

Appendix A19. List of Acronyms

Acronym Definition Section
Aeff effective area 19.2.2.1
AOM acousto-optic modulator 19.3.1
ASE amplified spontaneous emission 19.2.1.1
ASK amplitude shift key(-ed) (-ing) 19.2.2.3
BER bit error rate 19.2.1.1
BERT bit error rate test set 19.3.1
BL-PSBT band-limited phase-shaped binary transmission 19.2.2.3
C-band conventional band of EDFA (∼1530 to 1565 nm) 19.2.2.2
CPM cross-phase modulation 19.2.1.1
CSRZ carrier-suppressed return-to-zero 19.2.2.3
D group-velocity dispersion 19.2.2.2
DCF dispersion-compensating fiber 19.2.1.2
DCM dispersion-compensating module 19.2.1.3
DFB distributed feedback laser 19.3.1
DGD differential group delay 19.2.1.3
DPSK differential phase-shift-key(-ed) (-ing) 19.2.2.3
DWDM dense wavelength-division-multiplex(-ed) (-ing) 19.1.
EDFA erbium-doped fiber amplifier 19.2.1.1
EPSBT enhanced phase-shaped binary transmission 19.2.2.3
ETDM electrical time-division-multiplex (-ed) (-ing) 19.3.1
FEC forward error correction 19.2.2.4
FWHM full-width at half-maximum 19.3.2.1
FWM four-wave-mixing 19.2.1.1
GE dynamic gain equalizer 19.3.1
GFF gain-flattening filter 19.3.2.1
HS-DCF high-slope dispersion-compensating fiber 19.3.2.1
IC integrated circuit 19.2.2.4
IDF inverse dispersion fiber 19.2.2.2
ITU International Telecommunications Union 19.2.2.4
L-band long-wavelength band of EDFA (∼1570 to 1605 nm) 19.1.
LDV link design value 19.2.1.3
MPI multiple-path interference 19.2.2.1
MZM Mach–Zehnder modulator 19.3.1
NDSF nondispersion shifted fiber 19.3.2.1
NECG net effective coding gain 19.2.2.4
NRZ nonreturn-to-zero 19.2.1.2
NZDF nonzero dispersion fiber 19.2.1.2
OI optical interleaver 19.3.1
OSNR optical signal-to-noise ratio 19.2.1.1
712 L. Nelson and B. Zhu

Acronym Definition Section

OTDM optical time-division-multiplex(-ed) (-ing) 19.1.
PDL polarization-dependent loss 19.3.1
PDM polarization-division-multiplex(-ed) (-ing) 19.3.2.1
PI polarization-interleav(-ed) (-ing) 19.3.2.1
PMD polarization mode dispersion 19.2.1.3
PRBS pseudorandom bit sequence 19.3.1
PSBT phase-shaped binary transmission 19.2.2.3
PSCF pure silica-core fiber 19.3.3.2
PSK phase-shift-key(-ed) (-ing) 19.2.2.3
PSP principal state of polarization 19.2.1.3
Q Q-factor 19.2.2.4
RBW resolution bandwidth 19.2.1.1
RDS relative dispersion slope 19.2.2.2
RS Reed–Solomon 19.2.2.4
RZ return-to-zero 19.2.2.3
RZ-DPSK return-to-zero differential-phase-shift-key(-ed) (-ing) 19.2.2.3
SLA super large area fiber 19.2.2.2
SPM self-phase modulation 19.2.1.1
SRS stimulated Raman scattering 19.2.1.1
SSMF standard single-mode fiber 19.2.1.2
SW optical switch 19.3.1
TDM time-division-multiplex(-ed) (-ing) 19.1.
VSB vestigial sideband 19.2.2.3
WDM wavelength-division-multiplex(-ed) (-ing) 19.2.2.3.
WGR waveguide grating router 19.3.1

References
[1] RHK Telecommunications Industry Analysis, Internet traffic soars, but revenues glide,
May 2002.
[2] K. Coffman and A. Odlyzko, Growth of the Internet. In Optical Fiber Telecommunica-
tions IVB. I. P. Kaminow and T. Li, eds., San Diego: Academic, 17–56, 2002.
[3] C. Videcrantz, S. Danielsen, B. Mikkelsen, and P. Mamyshev, Advances in 40G long
haul WDM. In Proceedings of the National Fiber Optics Engineers Conference 2002,
Dallas, TX, 325–332, 2002.
[4] A. Umbach, T. Engel, H. G. Bach, S. V. Waasen, E. Droge, A. Strittmatter, W. Ebert,
W. Passenberg, R. Steingruber, W. Schlaak, G. G. Mekonnen, G. Unterborsch, and D.
Bimberg, Technology of InP-based 1.55µm ultrafast OEMMICs: 40Gbit/s broadband
and 38/60GHz narrow-band photoreceivers, IEEE J. Quantum Electron., 35:7, 1024–
1031, 1999.
[5] M. Yoneyama, Y. Miyamoto, T. Otsuji, H. Toba, Y. Yamane, T. Ishibashi, and H.
Miyazawa, Fully electrical 40-Gb/s TDM system prototype based on InP HEMT digital
IC technologies, IEEE J. Lightwave Technol., 18:1, 34–42, 2000.
[6] Y. K. Chen, Y. Baeyens, C.-T. Liu, R. Kopf, C. Chen, Y. Yang, J. Frackoviak, A. Tate,
A. Leven, P. Paschke, M. Berger, J. Weiner, K. Tu, G. Georgiou, P. Roux, V. Houstma,
and U. Koc, High speed electronics for lightwave communications. In Proceedings of
the Optical Fiber Communications Conference 2002, Anaheim, CA, WN2, 2002.
 19. 40 Gb/s Raman-Amplified Transmission 713

[7] M. Heins, J. M. Carroll, M.-Y. Kao, C. F. Steinbeiser, T. R. Landon, and C. F. Campbell,

An ultra-wideband GaAs pHEMT driver amplifier for fiber optic communications at
40 Gb/s and beyond. In Proceedings of the Optical Fiber Communications Conference
2002, Anaheim, CA, WN3, 2002.
[8] B. L. Kasper, O. Mizuhara, and Y.-K. Chen, High bit-rate receivers, transmitters, and
electronics. In Optical Fiber Telecommunications IVA. I. P. Kaminow and T. Li, eds.,
San Diego: Academic, 784–851, 2002.
[9] J. L. Zyskind, J. A. Nagel, and H. D. Kidorf, Erbium-doped fiber amplifiers for optical
communications. In Optical Fiber Telecommunications IIIB. I. P. Kaminow and T. L.
Koch, eds., San Diego: Academic, 13–63, 1997.
[10] T. Nielsen, A. J. Stentz, P. B. Hansen, Z. J. Chen, D. S. Vengsarkar, T. A. Strasser, K.
Rottwitt, J. H. Park, S. Stulz, S. Cabot, K. S. Feder, P. S. Westbrook, S. G. Kosinski,
1.6 Tb/s (40 × 40 Gb/s) transmission over 4 × 100km non-zero dispersion fiber using
hybrid Raman/erbium-doped amplifiers. In Proceedings of the European Conference on
Optical Communication 1999, Nice, France, PD2-2, 1999.
[11] A. H. Gnauck, G. Raybon, S. Chandrasekhar, J. Leuthold, C. Doerr, L. Stulz, A. Agarwal,
S. Banerjee, D. Grosz, S. Hunsche, A. Kung, A. Marhelyuk, D. Maywar, M. Movassaghi,
X. Liu, C. Xu, X. Wei, and D. M. Gill, 2.5 Tb/s (64 × 42.7 Gb/s) transmission over 40 ×
100 km NZDSF using RZ-DPSK format and all-Raman-amplified spans. In Proceedings
of the Optical Fiber Communications Conference 2002, Anaheim, CA, postdeadline
paper FC2, 2002.
[12] B. Zhu, L. Leng, L. E. Nelson, L. Gruner-Nielsen, Y. Qian, J. Bromage, S. Stulz, S.
Kado, Y. Emori, S. Namiki, P. Gaarde, A. Judy, B. Palsdottir, and R. L. Lingle, Jr., 3.2
Tb/s (80 × 42.7Gb/s) transmission over 20 × 100 km of non-zero dispersion fiber with
simultaneous C+L-band dispersion compensation. In Proceedings of the Optical Fiber
Communications Conference 2002, Anaheim, CA, postdeadline paper FC8, 2002.
[13] A. H. Gnauck and R. M. Jopson, Dispersion compensation for optical fiber systems. In
Optical Fiber Telecommunications IIIA. I. P. Kaminow and T. L. Koch, eds., San Diego:
Academic, 162–190, 1997.
[14] T. Kato, Y. Koyano, and M. Nishimura, Temperature dependence of chromatic dispersion
in various types of optical fibers. In Proceedings of the Optical Fiber Communications
Conference 2000, Baltimore, MD, TuG7, 2000.
[15] H. Kogelnik, L. E. Nelson, and R. M. Jopson, Polarization mode dispersion. In Optical
Fiber Telecommunications IVB. I. P. Kaminow and T. Li, eds., San Diego: Academic,
725–861, 2002.
[16] F. Curti, B. Daino, G. De Marchis, and F. Matera, Statistical treatment of the evolution
of the principal states of polarization in single-mode fibers, IEEE J. Lightwave Technol.,
LT-8:1162–1166, 1990.
[17] H. Taga, M. Suzuki, and Y. Namihira, Polarization mode dispersion tolerance of 10
Gbit/s NRZ and RZ optical signal, Electron. Lett., 34:2098–2100, 1998.
[18] R. M. Jopson, L. E. Nelson, G. J. Pendock, and A. H. Gnauck, Polarization-mode dis-
persion impairment in return-to-zero and non-return-to-zero systems. In Proceedings of
the Optical Fiber Communication Conference 1999, San Diego, WE3, 1999.
[19] H. Sunnerud, M. Karlsson, and P. A. Andrekson, A comparison between NRZ and RZ
data formats with respect to PMD-induced system degradation. In Proceedings of the
Optical Fiber Communications Conference 2001, Anaheim, CA, WT3, 2001.
[20] P. J. Winzer, H. Kogelnik, C. H. Kim, H. Kim, R. M. Jopson, and L. E. Nelson, Effect of
receiver design on PMD outage for RZ and NRZ. In Proceedings of the Optical Fiber
Communication Conference 2002, Anaheim, CA, TuI1, 2002.
714 L. Nelson and B. Zhu

[21] C. D. Poole, R. W. Tkach, A. R. Chraplyvy, and D. A. Fishman, Fading in lightwave

systems due to polarization-mode dispersion, IEEE Photon. Technol. Lett., 3:68–70,
1991.
[22] P. Noutsios and S. Poirier, PMD assessment of installed fiber plant for 40 Gbit/s trans-
mission. In Proceedings of the National Fiber Optics Engineers Conference 2001, Bal-
timore, MD, 1342–1347, 2001.
[23] K. W. Jackson, A. Bhat, V. Chandraiah, R. E. Fangmann, L. R. Dunn, A. F. Judy, J.
Kim, H. Ly, and S. C. Mettler, Polarization mode dispersion in high fiber count outside
plant cable. In Proceedings of the National Fiber Optics Engineers Conference 2001,
Baltimore, MD: 1333–1341, 2001.
[24] R. Caponi, B. Riposati, A. Rossaro, and M. Schiano, WDM design issues with highly
correlated PMD spectra of buried optical cables. In Proceedings of the Optical Fiber
Communications Conference 2002, Anaheim, CA, ThI5, 2002.
[25] M. Brodsky, P. D. Magill, and N. J. Frigo, Evidence of parametric dependence of PMD on
temperature in installed 0.05 ps/km1/2 fiber. In Proceedings of the European Conference
on Optical Communication 2002, Copenhagen, 9.3.2, 2002.
[26] P. B. Hansen, L. Eskildsen, A. J. Stentz, T. A. Strasser, J. Judkins, J. J. DeMarco, R.
Pedrazzani, and D. J. Digiovanni, Rayleigh scattering limitation in distributed Raman
pre-amplifiers, IEEE Photon. Technol. Lett., 10:1, 159–161, 1998.
[27] W. P. Urquhart and P. J. Laybourn, Effective core area for stimulated Raman scattering
in single-mode optical fibers, Proc. Inst. Elect. Eng., 132:201–204, 1985.
[28] J. Bromage, K. Rottwitt, and M. E. Lines, A method to predict the Raman gain spectra
of germanosilicate fibers with arbitrary index profiles, IEEE Photon. Technol. Lett.,
14:24–26, 2002.
[29] K. Rottwitt, J. Bromage, and L. Leng, Scaling the Raman gain coefficient of optical
fibers. In Proceedings of the European Conference on Optical Communication 2002,
Copenhagen, symposium paper 3.03, 2002.
[30] A. H. Hartog and M. P. Gold, On the theory of backscattering in signal-mode optical
fibers, J. Lightwave Technol., LT-2:2, 76–82, 1984.
[31] A. Chraplyvy, Optical power limits in multichannel wavelength-division-multiplexed
systems due to stimulated Raman scattering, Electron. Lett., 20:58, 1984.
[32] F. Forghieri, R. W. Tkach, and A. R. Chraplyvy, Effect of modulation statistics on Raman
crosstalk, IEEE Photon. Technol. Lett., 7:1, 101–103, 1995.
[33] J. Bromage, H. Thiele, and L. E. Nelson, Raman amplification in the S-band. In Pro-
ceedings of the Optical Fiber Communications Conference 2002, Anaheim, CA, ThB3,
2002.
[34] R. E. Neuhauser, P. M. Krummrich, H. Bock, and C. Glingener, Impact of nonlinear
pump interactions on broadband distributed Raman amplification. In Proceedings of the
Optical Fiber Communications Conference 2001, Anaheim, CA, MA4, 2001.
[35] J. Bromage, P. J. Winzer, L. E. Nelson, and C. J. McKinstrie, Raman-enhanced pump-
signal four-wave mixing in bidirectionally pumped Raman amplifiers. In Proceedings
of Optical Amplifiers and their Applications 2002, Vancouver, OWA5, 2002.
[36] L. Grüner-Nielsen, Y. Qian, B. Palsdottir, P. B. Gaarde, S. Dyrbol, T. Veng, Yifei Qian,
R. Boncek, and R. Lingle, Module for simultananeous C+L-band dispersion compen-
sation and Raman amplification. In Proceedings of the Optical Fiber Communications
Conference 2002, Anaheim, CA, TuJ6, 2002.
[37] J. Zhang, V. Dominic, M. Missey, S. Sanders, and D. Mehuys, Dependence of Raman
polarization dependent gain on pump degree of polarization at high gain levels. In
Proceedings of Optical Amplifiers and their Applications 2000, Quebec City, OMB4,
2000.
 19. 40 Gb/s Raman-Amplified Transmission 715

[38] H. H. Kee, C. R. S. Fludger, and V. Handerek, Statistical properties of polarization

dependent gain in fibre Raman amplifiers. In Proceedings of the Optical Fiber Commu-
nications Conference 2002, Anaheim, CA, WB2, 2002.
[39] R.-J. Essiambre, Effects of Raman noise and double Rayleigh backscattering on bidi-
rectionally Raman-pumped systems at constant fiber nonlinearity. In Proceedings of the
European Conference on Optical Communication 2001, Amsterdam, Tu.A.1.1, 2001.
[40] T. Nielsen, A. Stentz, K. Rottwitt, D. Vengsarkar, Z. Chen, P. Hansen, J. Park, K Feder,
T. Strasser, S. Cabot, S. Stulz, D. Peckham, L. Hsu, C. Kan, A. Judy, J. Sulhoff, S. Park,
L. Nelson, and L. Gruner-Nielsen, 3.28 Tb/s (82 × 40 Gb/s) transmission over 3 × 100
km nonzero-dispersion fiber using dual C- and L-band hybrid Raman-erbium-doped
inline amplifiers. In Proceedings of the Optical Fiber Communications Conference 2000,
Baltimore, MD, PD23, 2000.
[41] S. Bigo, A. Bertaina, Y. Frignac, S. Borne, L. Lorcy, D. Hamoir, D. Bayart, J.-P. Hamaide,
W. Idler, E. Lach, B. Franz, G. Veith, P. Sillard, L. Fleury, P. Guenot, and P. Nouchi,
5.12 Tb/s (128 × 40 Gbit/s WDM) Transmission over 3 × 100 km of Teralight fiber. In
Proceedings of the European Conference on Optical Communication 2000, Munich,
PD.1.2, 2000.
[42] B. Zhu, L. Leng, L. E. Nelson, Y. Qian, L. Cowsar, S. Stulz, C. Doerr, S. Chandrasekar,
S. Radic, D. Vengsarkar, Z. Chen, H. Thiele, J. Bromage, L. Gruner-Nielsen and S.
Knudsen, 3.08 Tb/s (77 × 42.7Gb/s) transmission over 1200 km of non-zero dispersion-
shifted fiber with 100-km spans using C- and L-band distributed Raman amplification.
In Proceedings of the Optical Fiber Communications Conference 2001, Anaheim, CA,
PD23, 2001.
[43] S. Kado, Y. Emori, S. Namiki, N. Tsukiji, J. Yoshida, and T. Kimura, Broadband flat-noise
Raman amplifier using low-noise bi-directionally pumping sources. In Proceedings of
the European Conference on Optical Communication 2001, Amsterdam, PD.F.1.8, 2001.
[44] A. R. Chraplyvy, Limitations on lightwave communication imposed by optical fiber
non-linearities, J. Lightwave Technol., 8:10, 1548, 1990.
[45] S. Bigo, S. Gauchard, S. Borne, P. Bousselet, P. Poignant, L. Lorcy, A. Bertaina, J.-
P. Thiery, S. Lanne, L. Pierre, D. Bayart, L.-A. de Montmorillon, R. Sauvageon, J.-F.
Chariot, P. Nouchi, J.-P. Hamaide, and J.-L. Beylat, 1.5 Terabit/s WDM transmission
of 150 channels at 10 Gbit/s over 4 × 100km of TeraLight fiber. In Proceedings of the
European Conference on Optical Communication 1999, Nice, France, PD2-9, 1999.
[46] A. F. Judy, Optimum fiber dispersion for multi-wavelength 40 Gbit/s NRZ and RZ
transmission. In Proceedings of the European Conference on Optical Communication
1999, Vol. II, Nice, France, 280–281, 1999.
[47] B. Konrad and K. Petermann, Optimum fiber dispersion in high-speed TDM systems,
IEEE Photon. Technol. Lett., 13:4, 299–301, 2001.
[48] T. Tsuritani, N. Takeda, K. Imai, K. Tanaka, A. Agata, I. Morita, H. Yamauchi, N.
Deagawa, and M. Suzuki, 1 Tbit/s (100 × 10.7 Gb/s) transoceanic transmission using
30 nm wide broadband optical repeaters with Aeff-enlarged positive dispersion fiber
and slope-compensation DCF. In Proceedings of the European Conference on Optical
Communication 1999, Nice, France, PD2-10, 1999.
[49] J.-X. Cai, M. I. Hayee, M. Nissov, M. A. Mills, A. N. Pilipetskii, S. G. Evangelides Jr.,
N. Ramanujam, C. R. Davidson, R. Menges, P. C. Corbett, D. Sutton, G. Lenner, C.
Rivers, and N. S. Bergano, 1.12 Tb/s transmission over trans-Atlantic distance (6200
km) using fifty-six 20 Gb/s channels. In Proceedings of the European Conference on
Optical Communication 2000, Munich, postdeadline paper 1.6, 2000.
[50] H. Sugahara, K. Fukuchi, A. Tanaka, Y. Inada, and T. Ono, 6050km transmission of 32 ×
42.7 Gb/s DWDM signals using Raman-amplified quadruple-hybrid span configuration.
716 L. Nelson and B. Zhu

In Proceedings of the Optical Fiber Communications Conference 2002, Anaheim, CA,

postdeadline paper FC6, 2002.
[51] T. Tanaka, N. Shimojoh, T. Naito, H. Nakamoto, I. Yokota, T. Ueki, A. Sugiyama, and
M. Suyama, 2.1-Tbit/s WDM transmission over 7,221 km with 80-km repeater spacing.
In Proceedings of the European Conference on Optical Communication 2000, Munich,
post-deadline paper 1.8, 2000.
[52] S. Knudsen, B. Zhu, L. E. Nelson, M. O. Pedersen, D. W. Peckham, and S. Stulz, 420
Gbit/s (42 × 10Gbit/s) WDM transmission over 4000km of UltraWave fiber with 100 km
dispersion-managed spans and distributed Raman amplification, Electron. Lett., 37:15,
965–967, 2001.
[53] B. Zhu, S. N. Knudsen, L. E. Nelson, D. W. Peckham, M. O. Pedersen, and S. Stulz,
800 Gb/s (80 × 10.664Gb/s) WDM transmission over 5200 km of fibre employing 100
km dispersion-managed spans, Electron. Lett., 37:24, 1467–1469, 2001.
[54] B. Zhu, L. Leng, L. E. Nelson, S. Knudsen, J. Bromage, D. Peckham, S. Stulz, K.
Brar, C. Horn, K. Feder, H. Thiele, and T. Veng, 1.6 Tb/s (40 × 42.7 Gb/s) transmission
over 2000 km of fiber with 100-km dispersion-managed spans. In Proceedings of the
European Conference on Optical Communication 2001, Amsterdam, PD.M.1.8, 2001.
[55] F. Liu, J. Bennike, S. Dey, C. Rasmussen, B. Mikkelsen, P. Mamyshev, D. Gapontsev,
and V. Ivshin, 1.6 Tbit/s (40 × 42.7 Gb/s) transmission over 3600 km UltraWave fiber
with all-Raman amplified 100 km terrestrial spans using ETDM transmitter and receiver.
In Proceedings of the Optical Fiber Communications Conference 2002, Anaheim, CA,
postdeadline paper FC7, 2002.
[56] M. O. Pedersen, S. N. Knudsen, T. Geisler, T. Veng, and L. Gruner-Nielsen, New low-
loss inverse dispersion fiber for dispersion matched fiber sets. In Proceedings of the
European Conference on Optical Communication 2002, Copenhagen, 5.1.3, 2002.
[57] B. Zhu, L. E. Nelson, L. Leng, S. Stulz, S. Knudsen, and D. Peckham, 1.6 Tb/s (40 × 42.7
Gb/s) WDM transmission over 2400 km of fiber with 100 km dispersion-managed spans,
Electron. Lett., 38:13, 647–648, 2002.
[58] R. Hainberger, J. Kumasako, K. Nakamura, T. Terahara, H. Onaka, and T. Hoshida,
Optimum span configuration of Raman-amplified dispersion-managed fibers. In Pro-
ceedings of the Optical Fiber Communications Conference 2001, Anaheim, CA, MI5,
2001.
[59] Y. Inada, K. Fukuchi, T. Ono, T. Ogata, and H. Okamura, 2400-km transmission of
100-GHz-spaced 40-Gb/s WDM signals using a “double-hybrid” fiber configuration. In
Proceedings of the European Conference on Optical Communication 2001, Amsterdam,
We.F.1.6, 2001.
[60] H. Sugahara, M. Morisaki, T. Ito, K. Fukuchi, and T. Ono, 9000-km transmission of 32 ×
42.7 Gb/s dense-DWDM signals using 195-µm2 -Aeff fiber and inverse double-hybrid
span configuration. In Proceedings of the Optical Amplifiers and their Applications
Topical Meeting 2002, Vancouver, postdeadline paper PD3, 2002.
[61] K. Schuh, M. Schmidt, E. Lach, B. Junginger, A. Klekamp, G. Veith, and P. Sillard, 4 ×
160 Gbit/s DWDM/OTDM transmission over 3 × 80 km Teralight – Reverse Teralight
fiber. In Proceedings of the European Conference on Optical Communication 2002,
Copenhagen, 2.1.2, 2002.
[62] K. Mukasa, H. Moridaira, T. Yagi, and K. Kokura, New type of dispersion management
transmission line with MDF for long-haul 40 Gb/s transmission. In Proceedings of the
Optical Fiber Communications Conference 2002, Anaheim, CA, ThGG2, 2002.
[63] Y. Miyamoto, A. Hirano, K. Yonenaga, A. Sano, H. Toba, K. Murata, and O. Mitomi,
320 Gb/s (8 × 40 Gb/s) WDM transmission over 367 km with 120 km repeater spacing
using carrier-suppressed return-to-zero format, Electron. Lett., 35:23, 2041, 1999.
 19. 40 Gb/s Raman-Amplified Transmission 717

[64] G. Charlet, J.-C. Antona, S. Lanne, P. Tran, W. Idler, M. Gorlier, S. Borne, A. Kelkamp, C.
Simonneau, L. Pierre, Y. Frignac, M. Molina, F. Beaumont, J.-P. Hamaide, and S. Bigo,
6.4 Tb/s (159 × 42.7Gb/s) capacity over 21 × 100 km using bandwidth-limited phase-
shaped binary transmission. In Proceedings of the European Conference on Optical
Communication 2002, Copenhagen, PD4.1, 2002.
[65] B. Zhu, L. Leng, A. H. Gnauck, M. O. Pedersen, D. Peckham, L. E. Nelson, S. Stulz,
S. Kado, L. Gruner-Nielsen, R. L. Lingle, Jr., S. Knudsen, J. Leuthold, C. Doerr, S.
Chandrasekhar, G. Baynham, P. Gaarde, Y. Emori, and S. Namiki, Transmission of 3.2
Tb/s (80 × 42.7 Gb/s) over 52 × 100 km of UltraWave fiber with 100-km dispersion-
managed spans using RZ-DPSK format. In Proceedings of the European Conference on
Optical Communication 2002, Copenhagen, PD2.4, 2002.
[66] S. R. Chinn, D.M.Boroson, and J.C. Livas, Sensitivity of optically pre-amplified DPSK
receivers with Fabry-Perot filters, IEEE J. Lightwave Technol., 14:370–376, 1996.
[67] J. P. Gordon and L. F. Mollenauer, Phase noise in photonic communications systems
using linear amplifiers, Optics Lett., 15:23, 1351–1353, 1990.
[68] K. Yonenaga, Y. Miyamoto, H. Toba, K. Murata, M. Yoneyama, Y. Yamane, and H.
Miyazawa, 320 Gb/s, 100-km WDM repeaterless transmission using fully encoded
40Gbit/s optical duobinary channels with dispersion tolerance of 380 ps/nm. In Pro-
ceedings of the European Conference on Optical Communication 2000, Munich, Vol.
1, 75–79, 2000.
[69] H. Bissessur, G. Charlet, C. Simonneau, S. Borne, L. Pierre, C. De Barros, P. Tran, W.
Idler, and R. Dischler, 3.2 Tb/s (80 × 40 Gb/s) C-band transmission over 3 × 100km
with 0.8 bit/s/Hz efficiency. In Proceedings of the European Conference on Optical
Communication 2001, Amsterdam, PD.M.1.11, 2001.
[70] H. Bissessur, L. Pierre, D. Penninckx, J.-P. Thiery, and J.-P. Hamaide, Enhanced phased-
shaped binary transmission for dense WDM systems, Electron. Lett., 37:1, 45–46, 2001.
[71] H. Kim and C. X. Yu, Optical duobinary transmission system featuring improved receiver
sensitivity and reduced optical bandwidth, IEEE Photon. Technol. Lett., 7:8, 1205–1207,
2002.
[72] F. Kerfoot, Forward error correction for optical transmission systems. In Proceedings of
the Optical Fiber Communications Conference 2002, Anaheim, CA, tutorial talk WL,
2002.
[73] H. Kidorf, N. Ramanujam, I. Hayee, M. Nissov, J.-X. Cai, B. Pedersen, A. Puc, and C.
Rivers, Performance improvement in high capacity, ultra-long distance, WDM systems
using forward error correction codes. In Proceedings of the Optical Fiber Communica-
tions Conference 2000, Baltimore, MD, ThS3, 2000.
[74] S. Ait, FEC techniques in submarine transmission systems. In Proceedings of the Optical
Fiber Communications Conference 2001, Anaheim, CA, TuF1, 2001.
[75] J. X. Cai, M. Nissov, A. N. Pilipetskii, C. R. Davidson, R. M. Mu, M. A. Mills, L. Xu,
D. Foursa, R. Menges, P. C. Corbett, D. Sutton, and N. S. Bergano, 1.28 Tb/s (32 × 40
Gb/s) transmission over 4,500 km. In Proceedings of the European Conference on
Optical Communication 2001, Amsterdam, PD.M.1.2, 2001.
[76] R.-J. Essiambre, G. Raybon, and B. Mikkelsen, Pseudo-linear transmission of high-
speed TDM signals: 40 and 160 Gb/s. In Optical Fiber Telecommunications IVB. I. P.
Kaminow and T. Li, eds., San Diego: Academic, 232–304, 2002.
[77] P. Bayvel and R. Killey, Nonlinear optical effects in WDM transmission. In Optical
Fiber Telecommunications IVB. I. P. Kaminow and T. Li, eds., San Diego: Academic,
611–641, 2002.
[78] F. J. MacWilliams and N. J. A. Sloane, Pseudo-random sequences and arrays, Proc. of
the IEEE, 64:12, 1715–1729, 1976.
718 L. Nelson and B. Zhu

[79] N. S. Bergano and C. R. Davidson, Circulating loop transmission experiments for the
study of long-haul transmission systems using erbium-doped fiber amplifiers, J. Light-
wave Technol., 13: 5, 879–888, 1995.
[80] Y. Sun, I. T. Lima, Jr., H. Jiao, J. Wen, H. Xu, H. Ereifej, G. M. Carter, and C. R. Menyuk,
Study of system performance in a 107-km dispersion-managed recirculating loop due
to polarization effects, IEEE Photon. Technol. Lett., 13:9, 966–968, 2001.
[81] Y. Sun, B. S. Marks, I. T. Lima, Jr., K. Allen, G. M. Carter, and C. R. Menyuk, Po-
larization state evolution in recirculating loops. In Proceedings of the Optical Fiber
Communications Conference 2002, Anaheim, CA, ThI4, 2002.
[82] G. M. Carter and Y. Sun, Making the Q distribution in a recirculating loop resemble a
straight-line experiment. In Proceedings of the Optical Fiber Communications Confer-
ence 2002, Anaheim, CA, ThQ5, 2002.
[83] S. Lee, Q. Yu, L.-S. Yan, Y. Xie, O. H. Adamczyk, and A. E. Willner, A short recir-
culating fiber loop testbed with accurate reproduction of Maxwellian PMD statistics.
In Proceedings of the Optical Fiber Communications Conference 2001, Anaheim, CA,
WT2, 2001.
[84] H. S. Kim, S. H. Yun, H. K. Kim, N. Park, and B. Y. Kim, Actively gain-flattened
erbium-doped fiber amplifier over 35 nm by using all-fiber acoustooptic tunable filters,
IEEE Photon. Technol. Lett., 10:790–792, 1998.
[85] C. R. Doerr, R. Pafchek, and L. W. Stulz, 16-band integrated dynamic gain equalization
filter with less than 2.8-dB insertion loss, IEEE Photon. Technol. Lett., 14:334–336,
2002.
[86] B. Eggleton, A. Ahuja, P. Westbrook, J. A. Rogers, P. Kuo, T. N. Nielsen, and B. Mikkle-
sen, Integrated tunable fiber gratings for dispersion management in high bit-rate systems,
J. Lightwave Technol., 18:1418–1432, 2000.
[87] M. Shirasaki, Chromatic dispersion compensator using virtually imaged phased array,
IEEE Photon. Technol. Lett., 9:12, 1598–1600, 1997.
[88] C. R. Doerr, L. W. Stulz, S. Chandrasekhar, L. Buhl, and R. Pafchek, Multichannel inte-
grated tunable dispersion compensation employing a thermooptic lens. In Proceedings
of the Optical Fiber Communications Conference 2002, Anaheim, CA. FA6, 2002.
[89] Y. Su, G. Raybon, L. K. Wickham, R.-J. Essiambre, S. Chandrasekhar, S. Radic, L. E.
Nelson, L. Gruner-Nielsen, and B. J. Eggleton, 40-Gb/s transmission over 2000 km of
nonzero-dispersion fiber using 100-km amplifier spacing. In Proceedings of the Optical
Fiber Communications Conference 2002, Anaheim, CA, ThFF3, 2002.
[90] S. C. Grubb, T. Strasser, W.Y. Cheung, W.A. Reed, V. Mizrachi, T. Erdogan, P.J. Lemaire,
A.M. Vengsarkar, D. J. DiGiovanni, D. W. Peckham, and B. H. Rockney, High-power
1.48 µm cascaded Raman laser in germanosilicate fibers. In Proceedings of Optical
Amplifiers and their Applications 1995, Davos, Switzerland, SaA4, 1995.
[91] L. Leng, Q. Le, L. Gruner-Nielsen, B. Zhu, S. Stulz, and L. E. Nelson, 1.6 Tb/s (40 ×
40Gb/s) transmission over 500 km of non-zero-dispersion fiber with 100-km amplified
spans compensated by extra-high-slope dispersion-compensating fiber. In Proceedings
of the Optical Fiber Communications Conference 2002, Anaheim, CA, ThX2, 2002.
[92] T. Terahara, T. Hoshida, J. Kumasako, and H. Onaka, 128 × 10.66 Gb/s transmission
over 840-km standard SMF with 140-km optical repeater spacing (30.4-dB loss) em-
ploying dual-band distributed Raman amplification. In Proceedings of the Optical Fiber
Communications Conference 2000, Baltimore, MD, PD28, 2000.
[93] A. K. Srivastava, S. Radic, C. Wolf, J. C. Centanni, J. W. Sulhoff, K. Kantor, and Y.
Sun, Ultra-dense terabit capacity WDM transmission in L-band. In Proceedings of the
Optical Fiber Communications Conference 2000, Baltimore, MD, PD27, 2000.
 19. 40 Gb/s Raman-Amplified Transmission 719

[94] J. P. Blondel, F. Boubal, E. Brandon, L. Buet, L. Labrunie, P. Le Roux, and D. Toullier,

Network application and system demonstration of WDM systems with very large spans
(error-free 32 × 10 Gbit/s 750-km transmission over three amplified spans of 250 km).
In Proceedings of the Optical Fiber Communications Conference 2000, Baltimore, MD,
PD31, 2000.
[95] Y. Zhu, W. S. Lee, C. Scahill, C. Fludger, D. Watley, M. Jones, J. Homan, B. Shaw, and
A. Hadjifotiou, 1.28 Tb/s (32 × 40 Gb/s) transmission over 1000 km with only 6 spans.
In Proceedings of the European Conference on Optical Communication 2000, Munich,
post-deadline paper 1.4, 2000.
[96] W. Idler, S. Bigo, Y. Frignac, B. Franz, and G. Veith, Vestigial side-band demultiplexing
for ultra-high capacity (0.64 bit/s/Hz) transmission of 128 × 40 Gb/s channels. In Pro-
ceedings of the Optical Fiber Communications Conference 2001, Anaheim, CA, MM3,
2001.
[97] T. Tsuritani, A. Agata, I. Morita, K. Tanaka, and N. Edagawa, Performance comparison
between DSB and VSB signals in 20 Gbit/s-based ultra-long haul WDM systems. In
Proceedings of the Optical Fiber Communications Conference 2001, Anaheim, CA,
MM5, 2001.
[98] S. Bigo, W. Idler, J.-C. Antona, G. Charlet, C. Simonneau, M. Gorlier, M. Molina, S.
Borne, C. deBarros, P. Sillard, P. Tran, R. Dischler, W. Poehlmann, P. Nouchi, and Y.
Frignac, Transmission of 125 WDM channels at 42.7 Gb/s (5 Tbit/s capacity) over
12 × 100 km of TeraLight Ultra fiber. In Proceedings of the European Conference on
Optical Communication 2001, Amsterdam, PD.M.1.1, 2001.
[99] G. Charlet, W. Idler, R. Dischler, J.-C. Antona, P. Tran, and S. Bigo, 3.2 Tb/s (80 × 42.7
Gb/s) C-band transmission over 9 × 100km of TeraLight fiber with 50-GHz channel
spacing. In Proceedings of Optical Amplifiers and their Applications Topical Meeting
2002, Vancouver, PD1, 2002.
[100] K. Rottwitt, A. Stentz, T. Nielsen, P. Hansen, K. Feder, and K. Walker, Transparent 80 km
bi-directionally pumped Raman amplifier with second-order pumping. In Proceedings
of the European Conference on Optical Communication 1999, Vol. II, Nice, France,
144–145, 1999.
[101] V. Dominic, A. Mathur, and M. Ziari, Second-order distributed Raman amplification
with a high-power 1370 nm laser diode. In Proceedings of the Optical Amplifiers and
their Applications Topical Meeting 2001, Stresa, Italy, OMC6-1, 2001.
[102] L. Labrunie, F. Boubal, E. Brandon, L. Buet, N. Darbois, D. Dufournet, V. Havard, P. Le
Roux, M. Mesic, L. Piriou, A. Tran, and J.-P. Blondel, 1.6 terabit/s (160 × 10.66 Gbit/s)
unrepeatered transmission over 321 km using second order pumping distributed Raman
amplification. In Proceedings of the Optical Amplifiers and their Applications Topical
Meeting 2001, Stresa, Italy, PD3-1,
[103] S. Bigo, Y. Frignac, G. Charlet, W. Idler, S. Brone, H. Gross, R. Dischler, W. Poehlmann,
P. Tran, C. Simonneau, D. Bayart, G. Veith, A. Jourdan, and J.-P. Hamaide, 10.2 Tb/s
(256 × 42.7 Gb/s PDM/WDM) transmission over 100km Teralight fiber with 1.28-
bit/s/Hz spectral efficiency. In Proceedings of the Optical Fiber Communications Con-
ference 2001, Anaheim, CA, PD25, 2001.
[104] Y. Frignac, G. Charlet, W. Idler, R. Dischler, P. Tran, S. Lanne, S. Borne, C. Martinelli, G.
Veith, A. Jourdan, J.-P. Hamaide, and S. Bigo, Transmission of 256 wavelength-division
and polarization multiplexed channels at 42.7 Gb/s (10.2-Tb/s capacity) over 3 × 100
km of Teralight fiber. In Proceedings of the Optical Fiber Communications Conference
2002, Anaheim, CA, postdeadline paper FC5, 2002.
720 L. Nelson and B. Zhu

[105] T. Ito, K. Fukuchi, K. Sekiya, D. Ogasahara, R. Ohhira, and T. Ono, 6.4 Tb/s (160 × 40
Gb/s) WDM transmission experiment with 0.8 bit/s/Hz spectral efficiency. In Proceed-
ings of the European Conference on Optical Communication 2000, Munich, PD1.1,
2000.
[106] L. Nelson, Challenges of 40-Gb/s WDM Transmission. In Proceedings of the Optical
Fiber Communications Conference 2001, Anaheim, CA, ThF1, 2001.
[107] L. Nelson, T. Nielsen, and H. Kogelnik, Observation of PMD-induced coherent crosstalk
in polarization multiplexed transmission, IEEE Photon. Technol. Lett., 13:7, 738–740,
2001.
[108] B. Zhu, P. B. Hansen, L. Leng, S. Stulz, T. N. Nielsen, C. Doerr, A. J. Stentz, Z. J. Chen,
D. W. Peckham, E. F. Rice, L. Hsu, C. K. Kan, A. F. Judy, and L. Gruner-Nielsen, 800
Gb/s NRZ transmission over 3200 km of TrueWave fiber with 100-km amplified spans
employing distributed Raman amplification. In Proceedings of the European Conference
on Optical Communication 2000, Munich, Vol. 1, 73–74, 2000.
[109] S. Knudsen, Design and manufacture of dispersion compensating fibers and their per-
formance in systems. In Proceedings of the Optical Fiber Communications Conference
2002, Anaheim, CA, WU3, 2002.
[110] A. Hasegawa, Numerical study of optical soliton transmission amplified periodically by
the stimulated Raman process, Appl. Optics, 23:3302, 1984.
[111] L. F. Mollenauer, J. P. Gordon, and M. N. Islam, Soliton propagation in long fibers with
periodically compensated loss, IEEE J. Quantum Electron., QE-22:1, 157–173, 1986.
[112] M. Nissov, C. R. Davidson, K. Rottwitt, R. Menges, P. C. Corbett, D. Innis, and N. S.
Bergano, 100Gb/s (10 × 10 Gb/s) WDM transmission over 7200 km using distributed
Raman amplification. In Proceedings of the European Conference on Optical Commu-
nication 1997, Vol. V, Edinburgh, UK, 9–12, 1997.
[113] H. Suzuki, J. Kani, H. Masuda, N. Takachio, K. Iwatsuki, Y. Tada, and M. Sumida,
25-GHz-spaced, 1 Tb/s (100 × 10 Gb/s) super-dense WDM transmission in the C-band
over dispersion-shifted fiber cable employing distributed Raman amplification. In Pro-
ceedings of the European Conference on Optical Communication 1999, Nice, France,
PD2-4,1999.
[114] D. F. Grosz, A. Kung, D. N. Maywar, L. Altman, M. Movassaghi, H. C. Lin, D. A.
Fishman, and T. H. Wood, Demonstration of all-Raman ultra-wide-band transmission
of 1.28 Tb/s (128 × 10 Gb/s) over 4000 km of NZ-DSF with large BER margins. In
Proceedings of the European Conference on Optical Communication 2001, Amsterdam,
PD.B.1.3, 2001.
[115] D. F. Grosz, A. Agrawal, S. Banerjee, A. P. Kung, D. N. Maywar, A. Gurevich, T.
H. Wood, C. R. Lima, B. Faer, J. Black, and C. Hwu, 5.12 Tb/s (128 × 42.7 Gb/s)
transmission with 0.8 bit/s/Hz spectral efficiency over 1280 km of standard single-
mode fiber using all-Raman amplification and strong filtering. In Proceedings of the
European Conference on Optical Communication 2002, Copenhagen, PD4.3, 2002.
[116] I. Tomkos, M. Vasilyev, J.-K. Rhee, M. Mehendale, B. Hallock, B. Szalabofka, M.
Williams, S. Tsuda, and M. Sharma, 80 × 10.7 Gb/s ultra-long-haul (+4200 km) DWDM
network with reconfigurable “broadcast and select” OADMs. In Proceedings of the
Optical Fiber Communications Conference 2002, Anaheim, CA, postdeadline paper
FC1, 2002.
[117] N. Shimojoh, T. Naito, T. Tanaka, H. Nakamoto, T. Ueki, A. Sugiyama, K. Torii, and
M. Suyama, 2.4-Tbit/s WDM transmission over 7400 km using all-Raman amplifier re-
peaters with 74-nm continuous single band. In Proceedings of the European Conference
on Optical Communication 2001, Amsterdam, PD.M.1.4, 2001.
 19. 40 Gb/s Raman-Amplified Transmission 721

[118] J.-X. Cai, M. Nissov, C. R. Davidson, Y. Cai, A. N. Pilipetskii, H. Li, M. A. Mills, R.-M.
Mu, U. Feiste, L. Xu, A. J. Lucerno, D. G. Foursa, and N. S. Bergano, Transmission of
thirty-eight 40-Gb/s channels (>1.5 Tb/s) over transoceanic distance. In Proceedings of
the Optical Fiber Communications Conference 2002, Anaheim, CA, postdeadline paper
FC4, 2002.
[119] C. Rasmussen, S. Dey, F. Liu, J. Bennike, B. Mikkelsen, P. Mamyshev, M. Kimmit, K.
Springer, D. Gapontsev, and V. Ivshin, Transmission of 40 × 42.7 Gb/s over 5200 km
UltraWave fiber with terrestrial spans using turn-key ETDM transmitter and receiver. In
Proceedings of the European Conference on Optical Communication 2002, Copenhagen,
PD4.4, 2002.
[120] K. Fukuchi, T. Kasamatsu, M. Morie, R. Ohhira, T. Ito, K. Sekiya, D. Ogasahara, T. Ono,
10.92 Tb/s (273 × 40-Gb/s) triple-band/ultra-dense WDM optical-repeatered transmis-
sion experiment. In Proceedings of the Optical Fiber Communications Conference 2001,
Anaheim, CA, PD24, 2001.
[121] L. Nelson, Optical fibers for high-capacity WDM, long-haul systems, IEICE Trans.
Electron., E86–C:5, 693–698, 2003.
[122] C. Fludger, A. Maroney, N. Jolley, and R. Mears, An analysis of the improvement in
OSNR from distributed Raman amplifiers using modern transmission fibers. In Pro-
ceedings of the Optical Fiber Communication Conference 2000, Baltimore, MD, FF2,
2000.
[123] J.-K. Rhee, D. Chowdhury, K. S. Cheng, and U. Gliese, DPSK 32 × 10 Gb/s transmission
modeling on 5 × 90 km terrestrial system, IEEE Photon. Technol. Lett., 12:12, 1627–
1629, 2000.
This page intentionally left blank
Index

1310 nm wavelength band, 383 band-limited phase-shaped bi-

14XX nm pump laser diodes, 121 nary transmission, 711
14XX nm pumps, 135 bandwidth-limited PSBT (BL-PSBT), 688
40 Gb/s Raman-amplified transmission, 673 beat noise, 92
40 Gb/s bit rate systems, 396 beat noise variances, 522
40 Gb/s terrestrial transmission, 701 beat-noise limited receivers, 522
BER measurements, 584
A figure of merit (FOM), 364 bidirectional pumping, 135
absorption coefficients, 430 bidirectional systems, 459
accelerated aging tests, 127 bidirectional transmission, 253
acoustic phonons, 325 bit error rate (BER), 310, 711
acousto-optic modulator, 711 bit error rate test set, 711
all-optical cross-connects, 596 bit rate scaling, 530
all-optical mesh networking, 596 Boltzmann factor, 38
all-optical terrestrial system, 627 booster amplifier, 346
all-optical transmission, 627 Bose–Einstein distribution, 12, 456
all-Raman amplifier, 445, 621 Bose–Einstein factor, 101
all-Raman system, 17, 627 Bragg grating stabilized laser diodes, 616
AllWave, 681 Brightness Theorem, 355
amplified spontaneous emission butterfly laser module, 125
(ASE), 10, 308, 491, 711
amplitude shift key(-ed) (-ing), 711 cable television, 27
analytical expression, 576 capture fraction, 44
anti-Stokes scattering, 38 carrier injection efficiency, 124
arrayed waveguide grating (AWG), 334 carrier-suppressed return-to-zero
arrhenius relationship, 128 (CSRZ), 479, 511, 711
ASE–ASE beat noise, 524 carrier-suppressed RZ (CSRZ), 686
attenuation, 317 cascaded Raman fiber laser
(RFL), 194, 353, 385
backward pumping, 458 cascaded Raman oscillator, 29
backward Raman pumping, 62 cascaded Raman resonator, 354
backward spontaneous Raman noise level, 62 cavity design, 357
banded bidirectional signal traffic, 280 C-band, 5, 301
banded schemes, 255 channel launch powers, 569
724 Index

channel power management, 574 differential group delay (DGD), 678, 711
channel power transients, 581 differential phase-shift-key(-ed) (-ing), 711
chi-squared distribution, 502 discrete or lumped Raman am-
chromatic dispersion, 602, 676 plifier (LRA), 1, 30, 301
cladding-pump fiber lasers, 9 dispersion, 162, 301
cladding-pumped fiber (CPF), 355 dispersion compensation, 303
classical treatments, 43 dispersion curvature, 161
closed loop control system, 322 dispersion map, 628
coherence time, 493 dispersion slope, 317
complex susceptibility, 51 dispersion slope compensation, 161
constant output power mode, 324 dispersion slope compensation ratio, 165
conventional band of EDFA dispersion slope-compensation, 447, 449
(∼1530 to 1565 nm), 711 dispersion-compensating Ra-
copumped Raman amplifier, 213 man amplifiers, 186
copumping, 135, 343 dispersion-compensating fiber
corner frequency, 222 (DCF), 3, 161, 304, 711
Corning DSF, 115 dispersion-compensating
Corning LEAF, 115 module (DCM), 711
Corning NZ-DSF, 115 dispersion-compensating mod-
Corning SMF-28, 373 ules (DCM), 679
cost metric, 445 dispersion-managed fiber, 384
cost-per-transmitted-bit, 673 dispersion-managed solitons, 627
counter-pumped Raman amplifiers, 213 dispersion-shifted fibers, 463
counter-pumping, 135 distributed feedback laser, 711
counterpropagating pump, 2 distributed feedback lasers (DFB), 506
counterpropagating pumping, 310 distributed Raman amplification, 338
counterpumping, 343 distributed Raman amplifier (DRA), 1, 452
coupled mode equations, 454 distributed Raman transmission, 383
critical launch power, 570 DMS, 627
critical power, 39 double Rayleigh backscattering (DRBS), 304
crossgain modulation, 231, 235, 267 double Rayleigh scattering (DRS), 2, 10
cross-phase modulation, 711 double-heterostructure (DH) laser, 122
cross-phase modulation double-Rayleigh backscatter (DRBS), 385
(XPM), 304, 315, 589 Double-Rayleigh backscattering, 83
crosstalk ratio, 514 DPSK (differential phase-shift keying), 669
DRBS crosstalk, 340
DCF, 3, 681 DRS crosstalk, 459
decibel/nanometer scale, 575 dual-band hybrid Raman/EDFAs, 694
degrades system performance, 569 dual-order Raman fiber lasers, 377
degree of polarization (DOP), 512 duobinary, single/vestigial side-
degrees of birefringence, 512 band modulation, 686
dense wavelength-division- dynamic gain equalizer, 448, 711
multiplex(-ed) (-ing), 711
dense wavelength-division- east–west (EW), 280
multiplexing (DWDM), 673 EDFA/Raman amplifier combinations, 445
dense WDM, 627 EDFA/Raman HA, 413
depolarizer, 146 effective area, 3, 43, 711
device-under-test, 517 effective length, 24
dielectric thin film interference filters, 143 effective mode area, 196
 Index 725

effective noise figure (ENF), 95, 385 forward pumping, 458

effective Rayleigh reflection coefficient, 25 Fourier transform, 497
effective upper-state lifetime, 348 four-wave mixing (FWM), 303, 314
elastic scattering, 502 four-wave-mixing (FWM), 79, 589, 711
ELEAF, 681 frequency domain, 494
electrical beat-noise measurement, 516 frequency modulation (dithering), 157
electrical noise figure, 458 frequency-division-multiplexed, 596
electrical spectrum analyzer, 174, 517 frequency-swept pumping, 468
electrical time-division- full-width at half-maximum, 711
multiplex (-ed) (-ing), 711 fundamental 3 dB noise limit, 45
electronic regeneration, 627 fused-fiber coupler, 143
electronically sweeping the wavelength, 62
end-of-life margin, 483 gain compression term, 263
energy transfer due to SRS, 573 gain equalizer (GEQ), 417, 448
engineering design rules, 445 gain feedback, 70
enhanced phase-shaped bi-
gain flatness, 66
nary transmission, 711
gain saturation, 317
Enhanced PSBT (EPSBT), 687
gain slope, 324
epoxy pigtailing, 307
gain spectrum, 2
equipartition energy, 643
gain tilt, 324, 576
equivalent noise figure, 285
gain-flattening filter (GFF),
erbium-doped silica fiber: EDSF, 418
311, 448, 694, 711
erbium-doped fiber ampli-
gain-shifted EDFA, 418
fier (EDFA), 1, 301, 711
Gaussian pulse shapes, 531
external-cavity lasers (ECL), 506
Gaussian-like, 502
GeO2 -based fibers, 204
Fabry–Perot laser, 139, 458
germanium doping, 52
fast interactions, 569
germano–silicate fiber, 29
fast tunable laser, 66
fast-gain dynamics, 97, 213 germanosilicate, 194
Fiber Bragg Grating laser, 137 Gordon–Haus effect, 648
fiber Bragg gratings (FBG), 353 graded index separate confinement
fiber nonlinearity, 491, 602 heterostructure strained layer
fiber selection, 600 multiple quantum well structure
field autocorrelation, 497 with buried heterostructure, 122
figure of merit for the Ra- grating-stabilized lasers, 458
man amplifier fiber, 317 GRIN-SCH strained layer MQW
fine-tuned dispersion, 447 structure with BH structure, 122
first-order backward pumped group-velocity dispersion, 711
configuration, 284 guiding filters, 661
first-order PMD, 678
fluoride, 30, 306 heavy metal oxide glasses, 53
fluoride EDF (erbium-doped flu- high germanium doping, 161
oride fiber: EDFF, 418 high P2 O5 -doped optical fibers, 199
fluoride or multicomponent silicates, 304 high pump power laser diodes, 9
fluoro-zirconate (ZBLAN), 306 high slope dispersion-compensating
forward error correction (FEC), 310, 688, 711 fiber (HSDK), 363
forward pumped unidirec- higher-order pumping, 468
tional transmission, 266 higher-order Raman pumping, 284
726 Index

high-slope dispersion- Kerr nonlinearity, 538

compensating fiber, 711
homogeneous-like, 320 Landsberg–Mandelstam effect, 35
hybrid amplifier (HA), 1, 413 laser module structure, 125
hybrid bidirectional amplification, 274 L-band, 5, 301
hybrid bidirectional scheme, 255 lightwave communication systems, 491
hybrid pump, 140 LiNbO3 modulator, 313
Hybrid Raman/EDFA amplifiers, 617 linear noise characteristics, 91
hybrid TDFA/Raman amplifier, 342, 440 link design value (LDV), 679, 711
hybrid tellurite/silica fiber log-linear loss spectrum, 575
Raman amplifier, 206 long distance telephony, 596
hybrid tellurite/silica Raman amplifiers, 440 long repeater spacing, 622
long-haul, 1, 445
imbalanced Mach–Zehnder long-wavelength band of EDFA
interferometers, 143 (∼1570 to 1605 nm), 711
inband crosstalk, 492 Lorentzian laser lineshape, 497
incoherent homodyne detection, 334 Lorentzian lineshape, 71
indium gallium arsenide, 305 low degree of polarization (DOP), 135
indium gallium arsenide phosphide, 305 low-loss splices, 362
indium phosphide, 305 low-noise preamplifier, 616
InGaAsP/InP GRIN-SCH strained LS fiber, 571
layer MQW structure, 121 Lucent Allwave, 115
inhomogeneous gain saturation, 320 Lucent TrueWave, 115
inline amplifier, 329, 413 lumped Raman amplifiers, 338
Inner Grating Multimode laser, 140
insertion loss, 274 Mach–Zehnder modulator, 711
integrated circuit, 711 Mach–Zender interferometers, 448
integrated dispersion compensation, 348 Maple mathematics, 635
intensity autocorrelation, 497 material dispersion, 162
interband stimulated Raman scattering, 328 maximum RIN transfer, 214
interchannel crosstalk, 267 Maxwellian probability density function, 678
interchannel interference, 305 MBE (molecular beam epitaxy), 122
interchannel nonlinear effects, 683 MCVD technique, 199
interference noise, 92 mechanical splicing, 307
interferers, 491 metropolitan optical network, 306
interleaved bidirectional scheme, 254 microelectromechanical sys-
International Telecommu- tems (MEMS), 448
nications Union, 711 midstage loss, 389
Internet, 595 MOCVD (metal organic chemi-
intersymbol interference, 305 cal vapor deposition), 122
intraband stimulated Raman scattering, 453 mode partitioning noise (MPN), 157
intrachannel cross-phase mod- modulation technique, 447
ulation (IXPM), 544 MPI crosstalk, 340
intrachannel four-wave mixing (IFWM), 544 MPI–MPI beat noise, 502, 524
inverse dispersion fiber, 711 Müller matrices, 512
isolators, 431 multicomponent silicate (MCS), 30
multipath interference (MPI), 119, 304
jitter-killer, 656 multiple interfering fields, 499
Jones matrix, 506 multiple path interference (MPI), 136
 Index 727

multiple reflections, 602 optical time domain reflec-

multiple Stokes shifts, 356 tometry (OTDR), 170
multiple wavelength Raman fiber lasers, 365 optical time-division-
multiple-order pump sources, 354 multiplex(-ed) (-ing), 712
multiple-order RFLs, 353 optical time-division-
multiple-wavelength RFLs, 353 multiplexed (OTDM), 674
multiple-path interference (MPI), 491, 711 optical time-domain extinc-
mutually incoherent, 506 tion technique, 519
optical–electrical–optical (O-E-O), 596
narrowband HA (NB-HA), 413 optimized modulation formats, 685
ND-FWM, 282 OSNR budget, 580
NDSF, 115 OSNR penalty, 529
net effective coding gain, 711 out-of-band crosstalk, 492
new wavelength bands, 26 out-of-band forward error correction, 310
noise figure, 12, 46, 305
noise power spectral density, 495 parametric amplifier, 37
noise-induced frequency shifts, 648 parametric interactions, 303
nondegenerate FWM, 282 path delay, 493
nondispersion shifted fiber, 711 path-average noise, 644
nondispersion-shifted fiber (NDSF), 227 path-average power, 605
nonlinear distortion, 91 path-average pump powers, 71
nonlinear impairments, 180 path-average signal power, 67, 644
nonlinear index of refraction, 603 pattern dependent crosstalk, 585
nonlinear interactions, 36 pattern-dependent Raman gain, 267
nonlinear penalty, 1 periodic-group-delay-complemented
nonlinear refractive index, 37 dispersion compensation, 663
nonlinear Schrödinger (NLS) equation, 629 phase difference, 493
nonreturn-to-zero, 686, 711 phase matching, 303
non-return-to-zero (NRZ), 310 phase mismatch, 630
nonzero dispersion fiber, 711 phase-shaped binary transmission, 686, 712
nonzero dispersion-shifted phase-shift-key(-ed) (-ing), 712
fiber (NZ-DSF), 64, 226 phonon population factor, 430
phonon-stimulated optical noise, 11
on-off keyed data transmission, 676 phosphorous-doped fiber, 451
on-off keying (OOK), 511 phosphorus doping, 53
on-off Raman gain, 421 phosphosilicate Raman fibers, 194
open-loop operation, 320 photosensitivity, 196
optical add/drop, 306 Placzek model, 40
optical add-drop multiplexers, 448 planar lightwave circuit (PLC), 143
optical confinement factor, 124 Planck’s constant, 38
optical cross-connects, 448 PMD coefficient, 678
optical interleaver, 711 Poincaré sphere, 512
optical noise figure, 458 Poisson distributed, 93
optical parametric amplifiers, 303 polarization beam combiner (PBC), 144
optical phonon, 2, 325 polarization controller (PC), 517
optical signal-to-noise ratio polarization dependence, 4, 52
(OSNR), 79, 347, 711 polarization evolution, 506
optical spectrum analyzer (OSA), 114, 517 polarization interleaving, 268
optical switch, 712 polarization mode dispersion, 712
728 Index

polarization mode dispersion Raman frequency shift, 196

(PMD), 145, 602, 678 Raman gain coefficient, 45, 51, 195
polarization multiplexing, 324 Raman gain efficiency coefficient (CR ), 114
polarization-dependent gain (PDG), 324 Raman gain spectrum, 51
polarization-dependent loss Raman gain tilt, 17
(PDL), 396, 602, 712 Raman oscillator wavelength shifters, 29
polarization-division-multiplex(-ed) (-ing), Raman pumping unit, 149
712 Raman response function, 50
polarization-interleav(-ed) (-ing), 712 Raman scattering, 35
polarization-mode dispersion, 447 Raman scattering cross-section, 41
polarized Raman scattering, 44 Raman threshold, 569
power conversion efficiency, 19 Raman-aided Brillouin scattering, 271
power partioning, 367 Raman-assisted repeatered transmission, 384
power spectral density, 240 rare earth-doped cladding-pumped
power transients, 583 fiber laser (CPFL), 354
preamplifier, 346 rare earth-doped fiber amplifiers, 303
principal state of polarization, 712 Rayleigh backscatter coefficient, 503
probability density function (PDF), 500 Rayleigh loss coefficient αr , 339
profile dispersion, 162 Rayleigh scattering, 502
pseudorandom bit sequence (PRBS), 240, 712 Rayleigh scattering loss, 503
pulse arrival times, 648 Rayleigh scattering of ASE, 102
pump depletion, 112, 263, 320 recapture fraction, 503
pump efficiency, 384 receiver noise sources, 92
pump laser diodes, 121 receiver sensitivity, 383
pump–pump interactions, 453 recirculating loop, 480
pump–signal RIN transfer, 459 recirculating transmission loop, 691
pump–signal walk-off, 267 Reed–Solomon, 712
pumping efficiency, 7 Reed–Solomon (RS) code, 688
pump-mediated crosstalk, 240 Reed–Solomon 255/239, 315
pump-mediated intersymbol relative dispersion slope, 164, 449, 712
interference (ISI), 267 relative intensity noise (RIN), 135, 498
pump-mediated signal crosstalk, 385 reliability, 126
pump-to-pump Raman interaction, 76 repeater spacing, 600
pump-to-signal noise transfer, 385 repeaterless transmission systems, 573
pure silica core fiber (PSCF), 571 required OSNR, 525
pure silica-core fiber (PSCF), 712 resolution bandwidth, 712
pure-silica core, 115 responsivity, 93
return-to-zero, 686, 712
Q scaling relationship, 389 return-to-zero (RZ), 511
Q-factor, 712 return-to-zero differential phase
Q-factor-based MPI tolerance factor, 528 shift keying (RZ-DPSK), 511
Q-spectra, 347 return-to-zero differential-phase-
quality factor (Q), 228, 339 shift-key(-ed) (-ing), 712
quantum approach, 37 reverse dispersion fiber, RDF, 227
revived interest, 9
Raman efficiency, 569 RFL pump module, 354
Raman enhanced fiber (RF), 363 RIN transfer, 221
Raman fiber lasers, 353 ring amplifier, 27
Raman figure of merit, 182 round-trip gain, 505
 Index 729

RS (255, 239), 689 splitband configuration, 447

RS (255,223), 689 spontaneous Brillouin scattering, 511
rule of thumb, 673 spontaneous emission factor, 11
RZ differential-phase-shift- spontaneous noise factor, 13
keying (RZ-DPSK), 686 spontaneous Raman noise, 81
spontaneous Raman scattering, 37
S+ bands, 309 spontaneous–spontaneous beat noise, 46
same relative dispersion slope, 683 spontaneously generated photons, 570
S-band, 26, 301 SRS dynamic crosstalk, 331
S-band lumped Raman amplifier (SLRA), 310 SRS-induced transients, 585
S-band Raman amplifiers, 301 SSMF, 681
seamless and wideband HA (SWB-HA), 413 standard single-mode fiber (SSMF), 310, 712
seamless transmission bands, 161 states of polarization (PSP), 678
second-order Raman pumping, 284 static gain equalizers, 448
selecting transmission fibers, 591 statistical nature, 590
self-phase modulation (SPM), 314, 712 stimulated Brillouin scattering, 325
semiconductor laser diodes, 385 stimulated Raman scattering, 35, 38, 712
semiconductor optical am- stimulated Raman threshold, 39
plifier (SOA), 30, 303 stochastic electrodynamics, 48
sequential pulsing, 62 Stokes scattering, 38
short upper-state lifetime, 10 Stokes vector, 512
shot noise, 91 super large area fiber, 712
Shott noise limited, 45 superposition rule, 131
signal preemphasis, 462, 465
signal–ASE beat noise, 524 tapered fiber bundle, 355
signal–MPI beat noise, 502, 524 TDFA/Raman amplifier, 413
Signal–pump–signal crosstalk, 267 telecommunications, 1
signal–spontaneous beat noise, 46, 340 telegraph, 596
signal-to-noise (SNR), 92 tellurite, 449
signal-to-noise ratio, 1, 45, 451 tellurite/silica Raman amplifier, 413
signal-to-signal crosstalk, 112 Tellurite-based glasses, 205
signal-to-signal crosstalk me- temperature dependence, 100, 430
diated by the pump, 245 temperature-dependent ASE, 104
silica fiber, 309 temperature-dependence of noise figure, 611
single-mode fiber (SSMF), 571 temperature-dependence of the
single-pass analytic models, 259 chromatic dispersion, 677
single-sided spectral density, 497 temporal fluctuations, 374
slope efficiency, 356 temporal gain ripple, 77
slope-compensating fibers, 449 temporal lens, 656
slow interactions, 569 Teralight fiber, 698
small effective area, 161 terrestrial applications, 595
small-scale inhomogeneities, 502 terrestrial links, 338
SMART pump, 65 terrestrial optical transmission systems, 673
soliton propagation, 383 terrestrial systems, 600
soliton transmission, 598 thermal distribution of phonons, 112
Soliton–Soliton Collisions, 649 thermal management, 150
splice loss spectral variation, 317 thermal noise, 340
splice losses, 363 thermal occupation number, 48
split-band augmentation strategy, 348 thermally induced phonon noise, 462
730 Index

thermoelectric cooler (TEC), 122 UltraWaveTM SLA, 681

third-order Raman pumping, 286 undepleted pump approximation, 99
third-order susceptibility, 49 unidirectional and bidirectional
threshold of catastrophic damage, 196 optical transmission, 253
thulium-doped fiber ampli- unidirectional transmission lines, 253
fier (TDFA), 30, 304 unintended generation of SRS, 569
time delay, 37
time domain, 494 VAD technique, 199
time domain picture, 49 variable optical attenuator (VOA), 277
time modulation of the pumps, 461 variance of the depletion, 589
time response of the Raman effect, 213 vestigial sideband (VSB), 474, 712
time-division-multiplex(-ed) (-ing), 712 vibrational mode, 2, 35, 309
time-division-multiplexed (TDM), 305, 674 voice traffic, 595
time-division-multiplexing, 61, 446
time-multiplexed pumping, 470 walk-off, 215
total launch power, 571 waveguide, 306
transit time, 218, 348 waveguide dispersion, 162
transition rate, 37 waveguide grating router (WGR), 690, 712
wavelength add-drop devices, 596
transmission fibers, 569
wavelength agnostic, 14
transparent photonic networks, 585
wavelength converters, 306
transverse mode mixing, 491
wavelength-division-multiplex(-
triangular profile, 574
ed) (-ing), 712
TrueWave® classic (TW) fiber, 571
wavelength-division-multiplexed
TrueWave® REACH, 681
(WDM), 301, 447
TrueWave® RS, 363, 681
wavelength-division-multiplexed
two-component model, 48
(WDM) pumping, 121
two-path interference, 495
WDM backbone networks, 674
two-stage Raman amplifier, 320
WDM couplers, 142
two-way signal traffic, 253
WDM transients, 243
Type-1 SWB-HA, 434
west–east (WE) transmission, 280
Type-2 SWB-HA, 435
Wide Sense Stationary linear
Type-3 discrete SWB-HA, 437 stochastic process, 240
Type-4 discrete SWB-HA, 437 wideband amplifiers (WBAs), 445
Wiener-Khintchine theorem, 497
ultralonghaul submarine, 595
ultra-long-haul (ULH), 1, 595 Yb-doped, 355
ultra-long-haul fiber-optic trans-
mission systems, 445 zero dispersion wavelength, 303
UltraWaveTM IDF, 681 zero-point fluctuations, 48
Springer Series in
optical sciences
New editions of volumes prior to volume 70
1 Solid-State Laser Engineering
By W. Koechner, 5th revised and updated ed. 1999, 472 figs., 55 tabs., XII, 746 pages
14 Laser Crystals
Their Physics and Properties
By A.A. Kaminskii, 2nd ed. 1990, 89 figs., 56 tabs., XVI, 456 pages
15 X-Ray Spectroscopy
An Introduction
By B.K. Agarwal, 2nd ed. 1991, 239 figs., XV, 419 pages
36 Transmission Electron Microscopy
Physics of Image Formation and Microanalysis
By L. Reimer, 4th ed. 1997, 273 figs. XVI, 584 pages
45 Scanning Electron Microscopy
Physics of Image Formation and Microanalysis
By L. Reimer, 2nd completely revised and updated ed. 1998,
260 figs., XIV, 527 pages
Published titles since volume 70
70 Electron Holography
By A. Tonomura, 2nd, enlarged ed. 1999, 127 figs., XII, 162 pages
71 Energy-Filtering Transmission Electron Microscopy
By L. Reimer (Ed.), 1995, 199 figs., XIV, 424 pages
72 Nonlinear Optical Effects and Materials
By P. Günter (Ed.), 2000, 174 figs., 43 tabs., XIV, 540 pages
73 Evanescent Waves
From Newtonian Optics to Atomic Optics
By F. de Fornel, 2001, 277 figs., XVIII, 268 pages
74 International Trends in Optics and Photonics
ICO IV
By T. Asakura (Ed.), 1999, 190 figs., 14 tabs., XX, 426 pages
75 Advanced Optical Imaging Theory
By M. Gu, 2000, 93 figs., XII, 214 pages
76 Holographic Data Storage
By H.J. Coufal, D. Psaltis, G.T. Sincerbox (Eds.), 2000
228 figs., 64 in color, 12 tabs., XXVI, 486 pages
77 Solid-State Lasers for Materials Processing
Fundamental Relations and Technical Realizations
By R. Iffländer, 2001, 230 figs., 73 tabs., XVIII, 350 pages
78 Holography
The First 50 Years
By J.-M. Fournier (Ed.), 2001, 266 figs., XII, 460 pages
79 Mathematical Methods of Quantum Optics
By R.R. Puri, 2001, 13 figs., XIV, 285 pages
80 Optical Properties of Photonic Crystals
By K. Sakoda, 2001, 95 figs., 28 tabs., XII, 223 pages
81 Photonic Analog-to-Digital Conversion
By B.L. Shoop, 2001, 259 figs., 11 tabs., XIV, 330 pages
82 Spatial Solitons
By S. Trillo, W.E. Torruellas (Eds.), 2001, 194 figs., 7 tabs., XX, 454 pages
83 Nonimaging Fresnel Lenses
Design and Performance of Solar Concentrators
By R. Leutz, A. Suzuki, 2001, 139 figs., 44 tabs., XII, 272 pages
84 Nano-Optics
By S. Kawata, M. Ohtsu, M. Irie (Eds.), 2002, 258 figs., 2 tabs., XVI, 321 pages
85 Sensing with Terahertz Radiation
By D. Mittleman (Ed.), 2003, 207 figs., 14 tabs., XVI, 337 pages
Springer Series in
optical sciences
86 Progress in Nano-Electro-Optics I
Basics and Theory of Near-Field Optics
By M. Ohtsu (Ed.), 2003, 118 figs., XIV, 161 pages
87 Optical Imaging and Microscopy
Techniques and Advanced Systems
By P. Török, F.-J. Kao (Eds.), 2003, 260 figs., XVII, 395 pages
88 Optical Interference Coatings
By N. Kaiser, H.K. Pulker (Eds.), 2003, 203 figs., 50 tabs., XVI, 504 pages
89 Progress in Nano-Electro-Optics II
Novel Devices and Atom Manipulation
By M. Ohtsu (Ed.), 2003, 115 figs., XIII, 188 pages
90 Raman Amplifiers for Telecommunications
By Mohammed N. Islam, 2004, 508 figs., XXX, 740 pages