IEEE JOURNAL O F QUANTUM ELECTRONICS, VOL. QE-20, NO.

9, SEPTEMBER 1984 1023

Stability Analysis for a Semiconductor Laser

in a n External Cavity
BJARNE TROMBORG, JENS HENRIK OSMUNDSEN, AND HENNING OLESEN

operating conditions and stability properties for external

Abstract-A detailed theoretical analysis of stability is presented for a cavity
semiconductor laser in an external cavity. The limits of stable operation
lasers. The paper maybe considered as an extension of the work
are determined as a function of the external cavity parameters and the
of R. Lang and K. Kobayashi [6] with emphasis on the deter-
linewidth enhancement factor a. Instability is related to jumps of the
laser frequency between external cavity modes (frequency bistability) mination of the limits of stability. These limits are obtained
or to feedback-induced intensity pulsations due to the carrier densityas functions of thea-parameterandtheparameterscharac-
dependence of the refractive index. The limit of bistability is derived
terizing the external cavity, Le., the level of optical feedback,
from the steady-state solutions of the rate equations and the intensity
the cavity length, and the phase of the reflected field.
pulsation limit is obtained from a small-signal analysis. This analysis
The analysis is based on the rate equations for the electric
also gives the locationof zeros in the system determinant and the
resulting FM noise spectrum. For practical applications we emphasize field amplitudeand phase,and forthe carrier density.The
phase condition for the steady-state solutions determines the
the determination of the stable tuning range for the phase in the exter-
nal cavity and the classification of the possible types of instability for
possible oscillation frequencies(external cavity modes)and
various feedback levels. among these the laser will tend to oscillate in the one with the
lowest threshold gain. The laser will only be stable for a certain
range of values of a and the external cavity parameters. The
I. INTRODUCTION limit of stable operation is reached when one of the following
PTICAL feedback from an external cavity is widely usedto three situations occurs:
0 improve the performance of semiconductorlaser sources. A ) The frequency jumps to another external cavity mode
The external cavity may consist of, e.g., a mirror, a grating, or because the actual mode ceases to be a solution to the phase
a short piece of single-mode fiber, and the cavity lengths are condition.
typically in the range 100 pm-1 m. The main advantages of B ) Thefrequencyjumps to another external cavity mode
this configuration are the enhancement of single longitudinal with lower thresholdgain.
modeoperation,thespectral line-narrowing, the improved C) The laser shows feedback-induced intensity pulsations.
frequency stability, and the wavelength tunability [ 11-[4] . The limits A and B are bistability limits that can be deter-
However, the mechanical tolerances for optimum performance mined from the steady-state solutiqns. They will be denoted
can be very critical, in particular with respect to variations in as the phase- and gain-condition limit, respectively. The inten-
external cavity length and alignment [SI. In addition, the laser sitypulsation limit C is a dynamicstabilitylimit,which is
mirrors and the external reflector form a double cavity config- obtained from a small-signal analysis of the rate equations.
uration, which is more likely to exhibit unstable behavior than In Section I1 we establish the basic theoretical model for a
the solitary laser [ 6 ] , 171. semiconductor laser in anexternal cavity. The phase- and
For a semiconductor laser the carrier density dependence of gain-condition limits are derived in Section 111 and illustrated
the refractive index plays an essential role.Itintroduces a for typical cases. In Section IV we perform a small-signal anal-
coupling between the amplitude and the phase fluctuations of ysis of the rate equations and examine the positions of zeros
the electric field. The coupling is described by the parameter for the determinant in the complex s-plane. We discuss the
R [6] or a(= - R ) [8] -[11], and gives rise to an enhanced relation between the locationof zeros and the fine-structurein
spectral linewidth and to spectral fine-structure. In the case of the FM noise spectrum [21], [22]. The limit C i s determined
injection locking it causes asymmetric locking characteristics as the limit where a zero passes the imaginary axis and moves
[12] and instabilities [ 131 , and in a semiconductor laser ampli- into the right half s-plane. This represents the onset of feed-
fier optical bistability may occur [ 141 . For a given laser a has back-induced intensity pulsations. The limits A , B, and C are
a definite value, but for different laser structures a-values from then combined to give the final stability region.
0.5 to 6.6 have been reported [15] -[20]. Recent measure- In Section V we show the stability region as a function of
ments indicate a = 3 for an AlCaAs CSP laser [I91 - [ 2 0 ] and feedback level and phase for a fixed a-value. In the experimen-
a 2: 6.6 for a 1.3 pmInGaAsP BH laser [ 191 . tal situation the phase of the reflected signal is often adjusted
In this paper we shall show how the a-parameter affects the to give optimum gain, i.e., maximum output power. Our anal-
ysis shows how much thephase can be allowed to deviate from
Manuscript received February 2, 1984. this value before one of the three stability limits, A , B , or C,is
B. Tromborg is with Telecommunication Research Laboratory, DK- reached. From the results the value of a and the absolute level
2200 Copenhagen, Denmark.
J. H. Osmundsen and H. Olesen are with the Electromagnetics Insti- of optical feedback can be determined, as reported in anearlier
tute, Technical University of Denmark, DK-2800 Lyngby, Denmark. paper [20].

00 18-9197/84/0900-1023$01 .OO 0 1984 IEEE

JOURNAL
IEEE
1024 OF QUANTUM ELECTRONICS, VOL. QE-20, NO. 9 , SEPTEMBER 1984

11. THE BASIC MODEL expression for a is

A . Equations for the Laser in External Cavity
We shall describe the semiconductor laser in an external cavity
by the simple model of Lang and Kobayashi [6]. The assump-
tion is here that the complex electric field E ( t ) satisfies the where X,. and x i are the real and imaginary parts of the com-
equation plex susceptibility.
We shall write E ( t ) as

dt
= zE(t - 7). (1)where Eo(t)and $ ( t ) are real-valued and supposed to be slowly
varying. From (l), (3), and (6) we derive the following equa-
To simplify the discussion we have not included the spontane- tions for Eo and $.
ous emission term in (1). However, it will be included in a later
section (Section IV). The model does not take into account
the spatial distribution of the field amplitude and the carrier
density N. The effects of lateral carrier diffusion and spatial
hole burning are therefore not included.
The term on the right of (1) is the field reflected from the
external cavity with time delay r. The factor z is given by z = where
K f d where fd is the frequency spacing between the longitudinal
modes of the solitary laser, and K* is the level of optical feed- A ( t ) = 007 + $ ( t ) $ ( t - 7 ) . (11)
back, i.e., the fraction of power coupled back into the lasing Inadditiontothe field equations(lOa)-(lOb) we have the
mode. We assume K << 1, so that multiple reflections can be rate equation for the carrier density
ignored.
Themodal gain per second andthephotonlifetime are d
- N =J -
denoted by G ( N ) and T ~ respectively.
, In the absence of feed- dt
back the threshold carrier density Nth is determined by
J is the pumping term and r, is the carrier lifetime. The equa-
1 tions (loa)-( lOcJ constitute our basic model.
G(Nth)= ~.
TP B. Stationary Solutions
G ( N ) is assumed to be linear, i.e., Any stationary solution to (lOa)-(lOc)can be written as
-
Eo(t)= Eo, $ ( t ) = A u t , N ( t ) = N ,
G ( N ) = G d N - No),
GN and No being constants.
(3)
- -
where E o , A o , and N are time-independent.
(12)
Aw is the sta-
The angular optical frequency o ( N ) is given by the Fabry- tionary shift in angular frequency due to the external cavity.
Perot resonance condition By inserting (1 2) in (loa)-( 1Oc) we obtain the following equa-
tions for the stationary solutions.
n(w,N)o(N)=pncl-' (4) -
G N A N = - 2 z COS OT (13)
forthepth longitudinal mode. 1 is thelength of the laser
diode, c is the speed of light: and n(o,N ) is the effective w - w o = - z ( a cos o r f sin or). (14)
refractive index.
We shall assume that the solitary laser oscillatesin a single
longitudinal mode with ahgular frequency
where
a 0 = U(Nth) (5)
W = W ~+ Ao (16)
for some fixed value of p . The carrier density N will in general
be close to thethreshold value N t h , so that we can expand
-
is- the angular frequency of the composite system. Also, A N =
o ( N ) to first order in A N = N - Nth. By (4) N - Nth, and E, is the stationary field amplitude for the soli-
tary laser. The stationary solutions are obtained by first solving
-
the phase condition(14) for o. Equations (13) and(15)
give the corresponding values for N and Eo.
-
where n* = n + oo(anlaw) is the effective group index, and The phase condition (14) may have multiple solutions corre-
sponding to multiple externalcavity modes. The laser tends to
2w0 an oscillate in the mode which has thelowest threshold gain. This
a = - -__ (7) mode is therefore the dominant mode. The threshold gain is
n*GN aN
given by
for N = Nth. a! is the parameter that gives the strength of the
coupling between phase and amplitude of the electric field as - 1
G ( N )= - - 22 cos or. (17)
discussed in the introduction. An equivalent but more basic TP
TROMBORG et al.: SEMICONDUCTOR LASER IN AN EXTERNAL CAVITY 1025

Hence the dominant mode is the mode for which the excess
gain
A G = G,AN=
- -22 COS 07 (18)
is minimum.

111. LIMITSOF FREQUENCYBISTABILITY

The frequencies of the external cavity modes depend on the
parameters a, z, T , and oo in (14). In the present section we
shall determine the region where the frequency of the domi-
nant mode is a continuous function of the parameters. At the
region boundary the laser becomes bistable and the frequency
of the dominant mode jumps to another externalcavity mode.
The jumps occur either in order to fulfill the phase condition
(14) or because there is anotherexternal'cavitymodewith Fig. 1. (a) Graphical representation of the phase condition (19) for ct =
lower threshold gain. These are the two cases denoted by A 3 and ZT = 2. (b) Corresponding excess gain (18), AGmax = 2 2 . The
and B in the Introduction and the correspondinglimits are the vertical dashed lines are drawn at W T = ( 2 p + 1) x .
phase- and gain-condition limits,respectively.
plum of 2n. By inserting in (14) and (19), the phase condition
A . The Phase Condition becomes
The phase condition (14) can be rewritten in the form B=ytz~(sinyta!cosy) (25)
woT - wr = z T d " G 7 sin (UT + $1, (19) or
where e =y t z T d G 0 1 2 sin ( y t J / ) (26)
$ = arctan a ! , (20)
where 6 andy satisfy the conditions (24) and (23), respectively.
The parameter w0r may be considered as the change of phase The graphical representation of the phase condition using
in the external cavity in the limit of no optical feedback. From the parameters 8 and y is shown in Fig. 2(a) for the same situa-
(19) ofie derives the actual change of phase in the external cav- tion as in Fig. l(a). The condition (24) is represented as a set
ity W T as a function of the phase W ~ T the, effective feedback of horizontal lines with spacing 2n. Eachline represents the
zr, and the &-parameter. In apractical setup the phase W ~ T situation in a particular interval between vertical dashed lines
can be varied by small-scale changes of the cavity length or by in Fig. l(a). The intersections between the lines and the curve
changing wo through the injection current or temperature. (26) in Fig. 2(a) determine the solutions to the phase condi-
The graphical representation of (19) is shown in Fig. l(a). tion. Since
The solutions to (19) appear as the intersections between the e(y = n) = e(Y = - n) t 2n (27)
line and the sinusoidal function. The excess gain of the solu-
at least one of the lines will intersect. The shift in frequency
tions can be read off on Fig. l(b).
of the solutionsis
In the approximation wherewe neglect the frequency depen-
dence of the modal gain, (13), (19), and (15) are invariant AU ( y - 0) T-* (28)
under the transformation
i.e., A o increases as we read off the solutions line by line. The
G J ~ T+ W O T f 'Lnq (2 1 4 solutions are cyclic in the sense that y = n represents the same
mode as y = -n. A change in w07 causes a parallel translation
07 + 07 f h q (21b) of the horizontal lines in Fig. 2(a). The graphical solution of
for integer q. This translation invariance makes it convenient the phase condition therefore gives a simplemethod to examine
to use a slightly different representation of (19) whichallows a the behavior of the solutions for varying w07. The solutions
simpler registration of the gain of the solutions. We introduce are obviously periodic in 0 0 7 with period 2n.
the principal value y of o r as the parameter instead of o r . It From Fig. 2(b) we can directly register the excess gain
is given by AGz-22 COSY (29)
WT=y + 2np (22) for the various solutions. Since AG is minimum for y = 0, the
dominant mode is the mode for which Iy I is minimum.
where
-n<y<n (23)
B. The Phase-Condition Limit
The phase-condition limit is determined by the local extrema
and p is an integer. We also define a parameter e by of (26). The function (26) bas at most onelocal maximum and
wor = e f 2np (24) one local minimum for - 7~ < y < n. Fig. 2(a) illustrates the
situation where the local maximum is also a global maximum.
where p is the same integer as in (22). This means that e may As e approaches the maximum em from below, the dominant
deviate from the principal value of wor by an integer multi- mode approaches y,, and as 0 passes Om the dominant mode
1026 JOURNAL
IEEE OF QUANTUM ELECTRONICS, VOL. QE-20, NO. 9, SEPTEMBER 1984

-lr 0 7T

Y (rad)
(b)
Fig. 2. (a) Graphical representation of the phase condition (26) using
the parameters 6 and y . (b) Corresponding excess gain. Eachhori-
zontal line in (a) represents the situation in a particularintervalof
Fig. 1. The point ( y m , e,) is a local maximum (cf. Section 111-B).

jumps to a solution with 0 = Om - 27r. If the local maximum

is not a global maximum it follows from (27) that only one of
the horizontal lines intersects the curve (26) for -n < y < 7r.
In that case, the dominant mode jumps to the other solution
with 0 = e m as 0 passes thelocalmaximum 0,. A similar
argument holds for thelocal minimum.
The local extrema are given by
-720 -360 0 360 720
a0 e (deg.)
- = 0 = 1t z r ( c o s y - a s i n y ) . (30)
aY (c)
Fig. 3. Phase-conditionlimits ( A ) and asymptotes ( y = 0, i n ) in the
Together with (25) this gives the parametric representation (a,@)-plane. (a) ZT = 0.245; (b) ZT = 1; (c) ZT = 2.45. For points in
the shaded area the phase condition (26) will have solutions satisfying
-n < y < 71.
a = (cosy + $)/sin y

Since the excess gain is minimum for y = 0 the line (33a) corre-
e=y+zT(siny+acosy) 1b) (3
sponds to solutions of optimum gain. The asymptotes (33b)
for the phase-condition limit in the (a, 0)-plane for - 7r < y < correspondtothe same external cavity modeforthe same
w. The curves are shown in Fig. 3(a)-(c) for the three cases value of a.
z7 < 1, ZT = 1, and ZT > 1. The region a < 0 is unphysical, but As y o varies from - 7r to 7r the phase-condition limit appears
is included in the figures to emphasize the symmetry. as the envelope to the lines(32), Le., it is acaustic. At the
For fixed y = y o the phase condition same timethe lines will sweep over the shadedareain Fig.
3(a)-(c). We can thereforeconcludethat (25) will have a
0 = y o + z~ (sin yo + a cos yo) (32) solution in the range - n < y < 7r if and only if (a,0) lies in
describes a line in the (a,0)-plane. The line can be shown to the shaded region. From the form (26) it is clear that it has
be a tangent tothephase-condition limit equations(31a)- at most three solutions. For each point (a,e ) in the shaded
(31b)atthepoint given by y = y o . It follows from(31a) area we shall denote the solution with minimum excess gain
as the first sheet solution. In Fig. 3(a)-(c) the curve sections
that I a I -+ ~0 for y + 0, k7r. The lines
corresponding to first sheetsolutions are drawnwiththick
fore = zra y o = 0, (334 line while the others are drawn with thin line.
0 = + nfor
- z7a yo = + n (33b) Thepropermathematicaldescriptionofthephase-condi-
tion limit is within the field of catastrophe theory. Thus the
arethereforetheasymptotestothephase-conditionlimit. limit in Fig. 3(a) shows a cusp catastrophe while the limit in
TROMBORG et al.: SEMICONDUCTOR LASER IN AN EXTERNAL CAVITY 1027

Fig. 3(c) shows afoldcatastrophe.Theconnectionbetween

elementary catastrophes and caustics is known to be a general
feature [23].

C. The Gain-Condition Limit

In this section we shall determine the subset of the shaded
area in Fig. 3(a)-(c) for which the dominant mode is a solu-
tion to (25).
In the case shown in Fig. 2(a) it is clear that the phase 0 cor-
responding to the dominant mode lies between 6, - 271 and
em. The limit 8, - 27r is a gain-condition limit. If 0 decreases
below this limit there will be a solution near y m with a more
favorablegain. A gain-condition limit wherethedominant
mode jumps to the phase-condition limit will be labeled B 1 .
If the local maximum is not an absolutemaximumthe
dominant mode will jump to a solution with the same 0. The
corresponding limitin the (a, 8)-plane is aphase-condition
limitwhenapproachedfromone side andagain-condition
limit when approached from the other side.
Another type of gain-condition limit arises when there are
two solutions y’ and y” with the same excess gain. By (29)
this requires cos y’ = cos y”, i.e., y” = -y’. We assume y‘ to
be the positive solution. The solutions y’ and y” must satisfy
the equation

e( y ‘ ) - e( y “ ) = 27rq
for some integer q. By (25) this implies that

sin ( q_
_ ’ ) -~(- 1)4+’
r - y_
(34)
q7r - Y’ zr
The limit for the dominant mode is given by the solution to
(34) with minimum excess gain, i.e., with minimum y’. It can
be shown that this is the solution for q = 1. This means that -720 -360 0 360 720
the frequency of the dominant mode will jump for y = ‘y’ e (deg.)
where
(C)

sin (n - y ‘ ) - 1 Fig. 4. Limits of bistability in the (a,8)-plane. (a) ZT = 0.245, (b) ZT =

...-
(35) 1, (c) ZT = 2.45. The shaded area is the region of the dominant mode.
71- y’ 27 It is boundedbythe phase-conditionlimit ( A ) , the gain-condition
limit (B1or B2), and the asymptotes( y = * x ) .
The equation has a solutionif and only if zr 2 1. In the (a,0)-
plane thecorrespondinglimits are the lines (32)with y o =
ky’. By using (35) these can be written as the phase-condition limit that lies in theshaded region will act as
a gain-condition limit B1 when it is approached for decreasing
e = + n + (36)
azr cosy’. e.
The lines are tangents to the phase-conditionlimit (31a)-(31b) The phase-condition limit is a strict limit in the sense that
at the points y = ky’. The points divide each line into two the frequency of the dominant mode must jump at this limit.
branches where only one of the branches corresponds to first At thegain-condition limit one may observehysteresis [ 6 ] .
sheet solutions. This is particularly pronounced at the limit of type Bz where
The phase- and gain-condition limits determine the region in the two solutionshave the same excess gain.
the ( a , 0)-plane which corresponds to a dominant mode. The In relation to the phase- and gain-condition limit it is inter-
region is shown as the shaded area in Fig. 4(a)-(c) for the same esting to consider the variation of the spectral linewidth. The
conditions as in Fig. 3(a)-(c). We have only shown the branches linewidth of the dominant mode is given by [24], [25]
ofthelimitswhichcorrespondtofirst sheet solutions.The
limit B1 is obtained by shifting the phase-condition limitA by Av = Av, (F)ae -z (37)
360’. The limit labeled Bz is the gain-condition limit given by
(36). It is tangential to the limit B1 [see Fig. 4(c)]. The tan- where Av, is the linewidth of the solitary laser. Since ae/ay =
gent point divides each limit into two branches where only one 0 at the phase-condition limit, the linewidth of the dominant
of the branches is a limit of bistability. In Fig. 4(a) the part of mode becomes infinite at this limit. The minimum linewidth
 1028 IEEE JOURNAL OF QUANTUM ELECTRONICS,
VOL. QE-20, NO. 9, SEPTEMBER 1984

appears fory = - $, i.e., for duced the parameter wR where

a=-tan0. (38) GN E2
a& = - 0 .
- (42)
Inthe (a,0)-planethe curve ofminimumlinewidth (38) TP
only corresponds to dominant mode solutions for small values In the following we approximate W R and rR by their values
of a. This fact limits the amount of linewidth reduction that for the solitary laser. In this approximation wR is the relaxa-
can be obtainedbymeansof an external cavity [25]. This tion resonance angular frequency and rR represents the carrier
will be further discussed in a forthcoming paper. density relaxation time. Assuming that
IV. SMALL-SIGNALANALYSIS Z
i>> - and w i >>z2
a (43)
The stationary solutions to the laser equations (loa) -(lOc) TR
that have been considered so far need not be stable solutions.
the determinant can be approximated by
Inthis section we shall performa small-signal analysis and
determine the limitsof stable operation. 1
~ ( s=
) s3 + s2 - + 22 cosy(1 - e-''
A. l%e Stability Criterion TR

Let &E,6q5, and 6N denote the deviation from the stationary +so& + z a g ( c o s y a- s i n y ) ( l - e-"). (44)
solution. To first order they obey the equations
We have the limiting case when a zero of D(s) passes the imagi-
S&(t) + z COS y [6E(t)- 6E(t - T)] nary axis, i.e., when s = ja is a zero of D(s) for R # 0. The
1 fixed zero of D(s) for s = 0 does not influence the stability.
+ zEo siny [s@(t>
- s@(t- r)] - - G~E,,:oN(~) By inserting s = jL! in (44) the real and imaginary parts give the
2
following two conditions forhaving a zero ofD(s) on theimag-
inary axis

Z
- - sin y [6E(t)- 6E(t - r)]
E O
1
Z COSY
-I [6$(t)- 6 @ ( t- T)] + 6d(t) - - G~a6N(t)
2
= F@J
(t), (39b)
2E0 1
- 6E(t) + Siir(t) -I - 6N(t) Equation (46) gives an explicit expression for r as a function
TP TR of R
= FN (t>
where F,, FQ,and FN are Langevin noise terms, which account
(39c)
7=
2
-
R
{ arccot [% (a2 -
1 1+pn (47)
for the spontaneous emission noise [26] . In the following we
for integer p ( p > 0). The curves are shown in Fig. 5 for rR =
neglect the carrier noise FN ( t ) which is usually less important
1 ns and OR 1271 = 2.55 GHz as derived from the parameters in
[ 1 11 , [26]. The parameter rR in (39c) is given by
Table I. For fixed r , the solutions to (46) give the positions
1 1 - +~ ~ 2 2 . (40) wherethe zeros of D(s) may pass the imaginaryaxis. It is
interesting that these positions depend only on r and not on 01
TR rs
or the external cavity parameters K and mor.
The criterion for stable operation is that the determinant for The limit of stable operation is reached when a zero of D(s)
the Laplace transform of (39a)-(39c) first passes the imaginary axis. Supposethishappens at s =
jRo . The solution Ro to (46) will generally be the one which
D(s) = s3 + s2 ;[ 1 + 2 z cosy(l - e-s7)] is closest to O R . Thus L!O = W R for W R r = ( 2 p + 1) 71. When
there are two solutions at about the same distance from O R
one may determine the limit corresponding to both solutions.
I 1
Z
+ s z2(1 - e-")' + 2 - cosy(1 - e - S T ) + w$ In Fig. 6 we have shown the positions of the zeros ofD(s) for
TR
Q = 3, K = 2 . lo-', r = 5 n / w R [i.e., z r = 2.45, which is the

+ [$ (1 - e-87)'
case shown in Figs. 3(c) and 4(c)] and for two values of mor.
The parameter d is the detuning definedas the change in phase
mor fromits value atoptimum gain (i.e., for y = 0). The
+ zwi (cosy - a sin y ) (1 - e-")

has no zeros in the right half s-plane. In (41) we have intro-

1 (41)
curves that connect the zeros are obtained from the equation
D(s) = 0 by solving itforexp ( - S T ) andtakingthenorm
squared. On the imaginary axis we have indicated the points
TROMBORG et al.: SEMICONDUCTOR LASER IN AN EXTERNAL CAVITY 1029

up as resonance peaks in the FM noise spectrum S4(f ) near

p=9
the solutions to (46). The FM noise spectrum can be derived
p=8
from (39a)-(39c) by using the methods of [9]. The result of
p=7
a calculation is shown in Fig. 7 for thesame parameters as used in
p=6
p=5
Fig. 6. (Similar curves have been presented by M.Tamburrini
p=4 et al. [2 11 , [ 2 2 ] . It is, however, not clear to us which detuning
p=3 parameters were used in their calculations.) The figure shows
p=2
bumps in the spectrum near the solutions to (46). The value
p=1
at f = 0 gives the relative linewidth Av/Av, due to optical
p=o
feedback [cf. ( 3 7 ) ] .
0 2 4 6 8 10 B. The Limit for Onset of Feedback-Induced Pulsations
WRT/T For s2 = no,(45) and (25) give the limit in the (a, O)-plane
Fig. 5 . The angular frequencies (relative to W R ) , where a zero ofD(s) where a zero of D(s) moves into the right half of the s-plane.
may pass the imaginary axis, versus normalized time delay inthe We consider this as the limit foronsetoffeedback-induced
external cavity.
intensity pulsations and introduce the label C for the limit.
TABLE I The parametric representation becomes
LASERPARAMETERS
A N D THEIR NUMERICAL VALUES

@=[(I - 2 (-)i)cosy

I G a i n coefficient GN

O=y+zr(siny+acosy) (48b)

I
~~~

for -77 < y 5 71. It is seen to have the same form as the repre-
, Photonlifetime

Longitudinal mode spacing

1 fd
I
125 GHz
1 sentationforthephase-condition limit(31a)-(31b).
consequence, it has the same asymptotes (33a)-(33b). If the
As a

I, curves (48a) and (31a) intersect for some value o f y the limits
' Normalized bias current I/I, 1.3
i (48a)-(48b) and (31a)-(31b) have a common tangent at the
point (a, 0) given by that value of y . As for the gain-condition
Threshold carrler denslty 1.55.10*~~-~
Nth
limit Bz the tangent point divides the pulsation limit into two
branches where only one of the branches corresponds t o first
sheet solutions.
If we introduce the pulsation limit in Fig. 4(a)-(c) we get
the results shown in Fig. 8(a)-(c). As in Fig. 4(a)-(c) we have
only included the limits corresponding to first sheet solutions.
For the limit B1 we only show the branches, which appear as
bistabilitylimits.The final stability region is shown as the
shaded area. It is ageneral featurethatthestability region
decreases with increasing a, due to the asymptotic approach
of the phase-condition and pulsation limitsto the line of opti-
mum gain. In Fig. 8(c) we notice that the stability region is
bounded by the gain-condition limits on bothsides in an inter-
-3 -2 -1 0 1
val of CY around 0. This interval increases with increasing feed-
Re(s) rR

5n. -
Fig. 6 . The complex zeros of D ( s ) for a = 3, K = 2 . lo-' and WRT =
: d = 0 " ; --=-- : d = -100". The frequencies obtained
from Fig. 5 are marked on the imaginary axis.
back level. For values of a in the interval the width of the
stability region is 360" in the 0-direction.

V. THESTABLETUNINGRANGE
where the zeros areallowed to pass according to (46). The solid Inthe previous sections we have determinedthestability
curve corresponds to the case of optimum gain and the dashed region in the (a, 0)-plane for fixed cavity length and for three
curve corresponds to a situation near the limiting case where a characteristic levels ofopticalfeedback. However, it is the
zero enters the right half s-plane. The zeros are seen to move region of stable operation for fixed a , but varying phase m o r ,
almost horizontally. The zeros with imaginary part less than cavity length, and optical feedback whichis directly related to
W R move toward the imaginaryaxis for decreasing detuning the experimental situation.
while the other zeros move away from the axis. In practice, one may adjust the phase o 0 r to a value of opti-
The zeros of D(s) give rise to resonance poles in the transfer mum gain. As mentioned in the previous section we denote
functions for the system (39a)-(39c). Theywill therefore show the deviation ofmor from this value as the detuningd . Within
1030 JOURNAL
IEEE OF QUANTUM ELECTRONICS, VOL. QE-20, NO. 9, SEPTEMBER 1984

the limits of bistability

d=e-B(y=o)=e-z7a (49)
and otherwise the limitsare periodic in d with period 271.
The stability limits for the detuning d as a function of level
1.OE 01
K of optical feedback are shown in Fig. 9 for a = 3 and a cavity
Fi
& 1.OE 00 length of 15 cm (corresponding to 7 = 5 r r / w ~ ) The
. labels on
2 the curves indicate the type of limit. For levels of optical feed-
1.OE-01 back below -58 dB there is no limit onthedetuning.The
I.OE-02 I { f iL f ,C f ,L ,L 1 laser shows intensity pulsations for points in the shaded area. At
0 2 4 6 8 1 0
the curve sections labeled A ?B 1 , and B2 the dominant mode
jumps to another external cavity mode, i.e., here the laser is
f (GH4 bistable. The curve section A is a phase-condition limit when
Fig. 7. Normalized FM noise spectrum S&(f)/(2n Av,) for the same approachedfor increasing detuning. When approachedfor
parameters as in Fig. 6 . __ : d = O o ; - - - - : d = -100". The decreasing detuning it appearsas a gain condition limit B 1 . (We
arrows again indicate the frequencies obtained from Fig. 5 and (46). note that thescale for K in Fig, 9 deviates by 6 dB from the scale
used in [20] . The discrepancy is due to the use offd instead
~ definition of z. We consider z = Kfd to be the
of 1 / 2 in~ the
correct form as supported by a traveling wave description.)
There are two characteristic feedback levels indicated by K
and K~ which can be easily be identified experimentally. For
feedback levels below K the dominant mode no longer jumps
in frequency. It follows from (3 la) that
K1 = [fdT=]-'. (50)
The feedback level K~ is the point where the pulsation limit
intersects the gain-condition limit B 2 . K~ is identified experi-
mentally as the feedback level above which the laser shows no
intensity pulsations. The equations (35) and (48a)with z =
~~f~ and y = y' give a parametric representation of K~ as a
function of a.
The relationship between a and K ~ / K(see ~ Fig. 10) has been
used for an experimental determination of a [20] . For a CSP
laser (Hitachi HLP 1400) we obtained the value a = 3.0 ir 0.3.
This is in good agreement with the result a = 2.8 of [ 191 for
the same type of laser.

VI. CONCLUSION
We have presented a detailed theoretical analysis of the sta-
bility properties for a semiconductorlaser in an externalcavity.
The theory is based on three position-independent rate equa-
tions for the carrier density and the amplitude and phase of
the laser field. I t takes intoaccounttheimportant carrier
density dependence of the refractive index, expressed by the
parameter a which is also known as thelinewidthenhance-
ment factor.
The stability region has been calculated as a function of the
feedback power level, the external cavity length, thephase cd07
of the reflected signal (in the limit of no optical feedback),
and the a-parameter. Three types of stability limits have been
-720 -360 0 360 720 identified: A is the phase-condition limit where the oscillation
8 (de57.J
frequency jumps to another external cavity mode because the
actual mode ceases to be a solution, B is the gam-condition
(C)
limit where the oscillation frequency jumps to another exter-
Fig. 8. The final stability region in the (a,8)-plane for the same param- nal cavity mode with a more favorable gain, and Cis the limit
eters as in Fig. 4 and W R T = 5n. (a) z r = 0.245; (b) zr = 1;(c) 2 7 =
2.45. The pulsation limit C depends explicitly on the value of z and for the onset of feedback-induced pulsations. The former two
r , whereas the limits A , B1, and B2 depend on the product27. limitsrepresenta bistable situation,wheremorethanone
TROMBORG et al.: SEMICONDUCTOR
ANLASER IN EXTERNAL CAVITY 1031

and we have demonstrated this by calculating the FM noise

spectra for two particular configurations ofzeros.
In a forthcoming paper we will treat the spectral properties
inmoredetail, in particulartheconsequencesforlinewidth
reductionbyopticalfeedback.Thetheorycould also be
extended to treat the modulation properties and the effects
of lateral carrier diffusion.
REFERENCES
-80 -60 -40 -20 0 [ I ] M. W. Fleming and A. Mooradian,“Spectralcharacteristics of
Feedbacklevel (dB) external-cavity controlled semiconductor lasers,” IEEE J. Quan-
Fig. 9. Stability limits for the detuning versus the absolute feedback tum Electron., vol. QE-17, pp. 44-59, Jan. 1981.
level (CY = 3, w ~ =7 5n). The labelsindicate the various types of [ 21K. R. Preston, K. C. Woollard, and K. H. Cameron, “External cav-
limits. In the shaded area the laser exhibits feedback-induced pulsa- ity controlled single-longitudinal mode laser transmitter module,”
tions. ~1 and ~2 arethe characteristicfeedback levels described in Electron. Lett., vol. 17, pp. 931-933, Nov. 1981.
the text. [3] S. Saito, 0. Nilsson, and Y. Yamamoto, “Oscillation center fre-
quency tuning, quantum FM noise, and direct frequency modula-
tioncharacteristics in external grating loadedsemiconductor
lasers.” IEEE J. Quantum
- Electron., vol. QE-18,pp. ..961-970,
June 1982.
141 F. Favre. D. le Guen. and J. C. Simon. “Outical feedback effects
I _

upon laser diode oscillation field spectrum,” IEEE J. Quantum

Electron., vol. QE-18, pp. 1712-1717, Oct. 1982.
[ 5 ] F. Favre and D. le Guen, “Emission frequency stability in single-
mode-fiber optical feedback controlledsemiconductor lasers,”
Electron. Lett., vol. 19, pp. 663-665, Aug. 1983.
[6] R. Lang and K. Kobayashi, “External optical feedback effects on
semiconductor injection laser properties,” IEEE J. Quantum
0 10 20 30 40 Electron., vol. QE-16, pp. 347-555, Mar. 1980.
[ 7 ] D. M. Fye, “Relationship between carrier-induced index change
2 (dB) and feedback noise in diode lasers,” IEEE J. Quantum Electron.,
V O ~ .QE-18, pp. 1675-1678, Oct. 1982.
Fig. 10. The relationship between the wparameter and the ratio K ~ / K I
[8] C.H. Henry, “Theory of the linewidth of semiconductor lasers,”
(cf. Fig. 9).
IEEEJ. Quantum Electron., vol.QE-18,pp. 259-264,Feb. 1982.
191 K. Vahala and A. Yariv, “Semiclassical theory of noise in semi-
external cavity mode may oscillate, whereas the latter repre- conductor lasers-Part 11,” IEEE J. Quantum Electron., vol. QE-
sents a small-signal instability. 19, pp. 1102-1109, June 1983.
Within the stable region the phase w07can always be adjusted 101 P. Spano, S. Piazzola, and M. Tamburrini, “Phase noise in semi-
conductor lasers: A theoreticalapproach,” IEEE J. Quantum
to minimize the threshold gain for the oscillating external cav- Electron., vol. QE-19, pp. 1195-1 199, July 1983.
ity mode. This corresponds to the optimum situation with the 111 C.H. Henry, “Theory of the phase noise and power spectrum of
maximum output power and the most favorable gain. In gen- a single mode injection laser,” IEEE J. Quantum Electron., vol.
QE-19, pp. 1391-1397, Sept. 1983.
eral we find that the oscillation frequency at the optimum is 121 R. Lang, “Injection locking properties of a semiconductor laser,”
lower than the free-running frequency and the shift increases IEEE J. QuantumElectron., vol. QE-18,pp.976-983, June
with increasing feedback level and a-value. 1982.
131 F. Mogensen, H. Olesen, and G. Jacobsen,“Theinfluence of
The optimum gain condition can be used as a reference-also asymmetriclockingcharacteristics onthecoherentmodtilation
experimentally-and our analysis shows how much the phase behaviour of an injection locked semiconductor laser,” in Proc.
u07can be allowed to deviate from the optimum, before one 9th ECOC, Geneva, Switzerland, Oct. 1983, pp. 83-86.
141 K. Otsuka and S. Kobayashi, “Optical bistability and nonlinear
of the three stability limits is reached. For very low levels of resonance in a resonant-typesemiconductor laser amplifier,”
feedback there are no limits, and the laser oscillates stably in Electron. Lett,, vol. 19, pp. 262-263, Mar. 1983.
the only possible external cavity mode. For intermediate levels 151 P. A. Kirkby, A. R. Goodwin, G . H. B. Thompson,and P. R.
Selway,“Observationof self-focusing instripegeometry semi-
feedback-inducedpulsationsmay easily occurand severely conductor lasers and the development of a comprehensive model
limit the stable range. For large levels only the gain-condition of their operation,” IEEE J. Quantum Electron., vol. QE-13, pp.
limits are present and the laser will gradually start t o oscillate 705-719, A u ~ 1977.
.
in several external cavity modes. Measurements of the stabil-
161 C. H. Henry, e. A. Logan, and K. A. Bertness,“Spectral depen-
dence of the change in refractive index due to carrier injection
ity region versus feedback level have been reported in anearlier in GaAs lasers,” J. Appl. Phys., vol. 52, pp. 4457-4461, July
paper [20] together with a determination of the @-value. 1981.
171 K. Vahala, L. C. Chiu, S. Margalit, and A. Yariv, “On the line-
The small-signal stability analysis of the rate equationsallows width enhancement factor 01 in semiconductor injection lasers,”
a determination of the locations of complex zeros in the sys- A p p l . Phys. Lett.,vol. 42,pp. 631-633, Apr. 1983.
temdeterminant.Feedback-inducedpulsationsoccurwhen [18] K. Kikuchi and T. Okoshi, “Measurement of spectra of and cor-
relation between FM and AM noises in GaAlAs lasers,” Electron.
one of the zeros move across the imaginary axis into the right Lett., vol. 19, pp. 812-813, Sept. 1983.
half of the s-plane. In general, this will happen close to s = 1191 I. D. Henning and J. V. Collins, “Measurement of the semicon-
jwR where OR is the relaxation resonance frequency of the ductor laser linewidth broadening factor,” Electron. Lett., vol.
19, pp. 927-929, Oct. 1983.
free-running laser. However, other zeros in the left half-plane [20] J. H. Osmundsen, B. Tromborg,and H. Qlesen, “Experimental
but close to the imaginary axis may have an influence as well, investigationof stabilitypropertiesfor a semiconductor laser
 1032 JOURNAL
IEEE OF QUANTUM ELECTRONICS, VOL.NO.
QE-20, 9, SEPTEMBER 1984

with optical feedback,” Electron. Lett., vol. 19, pp. 1068-1070, Member of the Telecommunication Research Laboratory, Copenhagen,
the at stationed
Dec. 1983. Electromagnetics Institute, Technical University of
M. Tamburrini, P. Spano,and S. Piazzola, “Influence of anexter- Denmark. His presentwork is on OTDR-technique and physics of
nal cavity on semiconductor laser phase noise,” Appl. Phys. Lett., optoelectronic devices.
vol. 43, pp. 410-412, Sept. 1983.
P. Spano. “High frequency phase-noise in semiconductor lasers,” in
Pro;. 9thEC6C, G&eva,-Switzerland, Oct. 1983, pp. 131-133. Jens Henrik Osmundsen was born on August 8,
A. E. R. Woodcock and T. Poston, A Geometrical Study of the 1958. He received the M.Sc. degree in electrical
Elementary Catastrophes. New York: Springer-Verlag, 1974. engineering from the ElectromagneticsInsti-
K. Kikuchi, T. Okoshi, and H. Ishigami, “Measurement of spec- tute, Technical University of Denmark, Lyngby,
tral purity of GaAlAs lasers” (in Japanese), Rec. Opt. Quantum in 1982, and is now working towards the Ph.D.
Electron. IECEJapan, vol. OQE83-23, pp. 25-32, June 1983. degree.
E. Patzak, H. Olesen, A. Sugimura, S. Saito, and T. Mukai, “Spec- His research interests are primarily within the
tral linewidth reduction in semiconductor lasers by an external field of semiconductor laser stabilization and
cavity with weak optical feedback,” Electron. Lett., vol. 19, pp. optical feedback effects on laser properties.
938-940, Oct. 1983.
R. Schimpe and W. Harth, “Theory of FM noise of single-mode
injection lasers.” Electron. Lett., vol. 19, pp. 136-137,Feb.
1983.

Bjarne Tromborg was born in 1940 in Den-

mark. He received the M.Sc. degree in physics
and mathematics from the Niels Bohr Institute,
Copenhagen, Denmark, in 1968.
From 1968 to 1977 he was a Research Asso-
ciate at NORDITA and the Niels Bohr Institute.
His research field was theoreticalelementary
particle physics; in particular, dispersion theory
and electromagnetic corrections to
hadron
scattering. From 1977 to 1979 he taught at a
high school.Since 1979he has been a Staff

On Fluctuations and Transients in Injection Lasers

STEWART E. MILLER, FELLOW: IEEE, AND DIETRICH MARCUSE, FELLOW, IEEE

Abstract-A new computer model of the semiconductor injection la- I. INTRODUCTION

ser is described. The electron-photon interactions are introduced using
INCE the recognition that longitudinal-mode jumping in
a Monte Carlo technique in combination with multimode rate equations
to compute laser output versus time and to compute output power
probability density. Laser turn-on initially in a side mode, followed by
S single-transverse mode injection lasers can cause significant
pulse-to-pulse timejitterattheendofafiber transmission
evolution to the normal equilibrium modal power distribution, is shown
to be intrinsic and a natural consequence of fluctuations in the modal
link [ l ] , increased attentionhasbeenfocusedoninjection
powers at the moment the laser passes through threshold. Calculations laser outputfluctuations.The systemeffect called “mode-
of fluctuations in the equilibriumpower levels show that occasional partition noise” is due to the fact that in a digital system, the
dropouts of the main-modepower are also a natural consequence of directlymodulatedinjection laser mayturnonindifferent
electron-photon interactions. Reduction in equilibrium fluctuations longitudinal modes on successive pulses, and because there is
and in turn-onfluctuations can be achieved by increasing the laser
power-output level, by increasing the mirror reflectivities, and by de-
wavelengthdispersionin the fiber link,thedifferentfre-
creasing the laser length. quencies result indifferent relativedelays to the receiver.
This work indicates that conventionalinjection lasers must beoperated There results a limit on repeater spacing due to intersymbol
at or abovethe 1 mW level inorder t o assure anerrorrate of in interference which is not increased by improved receiver noise
digital systems. The well-known transition from thermal-noise statistics performanceor increased transmitterpower. It is important
below threshold to Poisson statistics well above threshold is still evolv-
t o understand and alleviate this phenomenon.
ing in the 0.5-2 mW region in the injection laser.
Fluctuations in the continuously running (CW) laser were
studied theoretically by D. E. McCumber in, 1966 and many
Manuscript received October 24, 1983; revised March 9, 1984.
The authors are with AT&T Bell Laboratories, Crawford Laboratory, additions to his theory have appeared [ 2 ] . Recently, numerous
Holmdel, NJ 07733. experimentalstudies of injection lasers have beenreported,

0018-9197/84/0900-lO32$01.00 0 1984 IEEE