RESEARCH ARTICLE | AUGUST 18 2015

Invited Article: A compact optically coherent fiber frequency

comb 
L. C. Sinclair; J.-D. Deschênes; L. Sonderhouse; W. C. Swann; I. H. Khader ; E. Baumann; N. R. Newbury;
I. Coddington

Rev. Sci. Instrum. 86, 081301 (2015)

https://doi.org/10.1063/1.4928163
 CHORUS


View Export
Online Citation

23 September 2024 17:53:47

 REVIEW OF SCIENTIFIC INSTRUMENTS 86, 081301 (2015)

Invited Article: A compact optically coherent fiber frequency comb

L. C. Sinclair,1 J.-D. Deschênes,1,2 L. Sonderhouse,1 W. C. Swann,1 I. H. Khader,1
E. Baumann,1 N. R. Newbury,1 and I. Coddington1
1
National Institute of Standards and Technology, 325 Broadway, Boulder, Colorado 80303, USA
2
Université Laval, Québec City, Québec G1V 0A6, Canada
(Received 20 April 2015; accepted 22 July 2015; published online 18 August 2015)
We describe the design, fabrication, and performance of a self-referenced, optically coherent fre-
quency comb. The system robustness is derived from a combination of an optics package based
on polarization-maintaining fiber, saturable absorbers for mode-locking, high signal-to-noise ratio
(SNR) detection of the control signals, and digital feedback control for frequency stabilization. The
output is phase-coherent over a 1-2 µm octave-spanning spectrum with a pulse repetition rate of
∼200 MHz and a residual pulse-to-pulse timing jitter <3 fs well within the requirements of most
frequency-comb applications. Digital control enables phase coherent operation for over 90 h, critical
for phase-sensitive applications such as timekeeping. We show that this phase-slip free operation
follows the fundamental limit set by the SNR of the control signals. Performance metrics from
three nearly identical combs are presented. This laptop-sized comb should enable a wide-range of
applications beyond the laboratory. [http://dx.doi.org/10.1063/1.4928163]

I. INTRODUCTION assembled relatively quickly using common telecommunica-

tions fiber handling tools and readily available telecommu-
Optical frequency combs are powerful instruments which nications grade micro-optic components. Historically, fiber-
provide millions of mutually coherent, perfectly spaced comb based combs have been limited by changes in birefringence

23 September 2024 17:53:47

modes that are uniquely capable of phase coherently linking in the optical fiber due to strain, temperature changes, or
the optical, microwave, and terahertz regimes. Today’s fre- humidity changes. This varying environmentally induced bire-
quency combs support a range of applications and fundamental fringence results in polarization wander within the fiber. Polar-
science,1–4 and frequency combs capable of operation outside ization wander can disrupt mode-locking in several frequency
of a well-controlled laboratory environment are increasingly comb designs and can also cause significant spectral drift
necessary as the list of applications grows. This paper de- during supercontinuum generation, impacting both offset-
scribes the design and performance of a fiber-frequency comb, frequency detection and spectral coverage. The advent of high
whose development was motivated by needs for improved trace quality polarization-maintaining (PM) fiber and PM fiber-
gas measurements,5–16 high-accuracy laser detection and rang- optic components has largely solved this problem.
ing (LADAR),17–27 and optical free-space time-and-frequency The strength of fiber-based combs is clear from several
transfer28 to support future networks of mobile atomic opti- demonstrations. A very early design demonstrated turn-key
cal clocks.29–31 These applications and others necessitate the operation through the use of a semiconductor saturable absorb-
development of a new breed of compact, relatively inexpensive, ing mirror (SESAM).41,45 Previously, we demonstrated an
and environmentally robust frequency combs that can operate optically coherent frequency comb with a 27 MHz repetition
outside the laboratory in industrial settings and in moving rate that maintained its low-noise performance even when the
vehicles, all while maintaining high performance and requiring femtosecond PM-fiber laser was subjected to g-level vibra-
minimal, or even no, user intervention. tions.46 A Korean team launched an all-fiber femtosecond laser
Solid-state combs, micro-resonator-based combs, and with a 25 MHz repetition rate on a satellite, achieving mode-
fiber-based combs are all potential platforms for development locked operation in low-earth orbit for over a year.47 A German
of an environmentally robust comb. Solid-state combs, such team has developed a repetition-rate-stabilized PM-fiber fre-
as those based on Ti:sapphire, Yb:YAG, or erbium:ytterbium- quency comb for planned operation on a sounding rocket.48
doped glass, now exist with very compact cavities32–34 and More recently, our group has demonstrated a 200 MHz all-PM
with direct diode pumping.35 However, solid-state combs fiber frequency comb able to maintain optical coherence both
require precise free-space alignment, which is challenging and under vibrations of up to 0.5 g rms (integrated from 10 Hz to
expensive to implement in a robust platform. Recent advances 1 kHz with a 1/ f profile) and in a moving vehicle.49
in chip-scale micro-resonator-based combs are intriguing,36–38 In this paper, we present an optical frequency comb design
but current “micro” combs still significantly lag “macro” based on Ref. 49 in more detail and modified for improved
combs in performance, overall system size, and robustness, performance. Compared to the earlier design, the coherence is
as well as have repetition frequencies of many tens of GHz. improved by a factor of two, the output power is 50% higher,
Optical-fiber-based frequency combs have long been identified and the overall size of the optics package is reduced by a factor
as the best candidate for lower-cost fieldable systems.39–44 of six to below 1 l in volume. In a design compatible with non-
All-fiber designs are impervious to misalignment and can be laboratory operation, these compact frequency combs achieve

0034-6748/2015/86(8)/081301/15/$30.00 86, 081301-1

081301-2 Sinclair et al. Rev. Sci. Instrum. 86, 081301 (2015)

a residual pulse-to-pulse timing jitter of <3 fs and residual sub- interest here is in a fully stabilized, or phase-locked, frequency
radian optical coherence at the comb center wavelength. Ad- comb. For a fully stabilized frequency comb, all comb modes
ditionally, we have demonstrated phase-slip-free operation for are simultaneously stabilized by the stabilization of only two
over 90 h driven by the combination of high signal-to-noise- degrees of freedom. Here, the two degrees of freedom are
ratio (SNR) phase-locking signals and digitally based phase the carrier-envelope-offset frequency, f ceo, and a single optical
detection and feedback. This level of performance exceeds that comb mode whose frequency is near to an optical standard,
required to support demanding applications, such as high reso- i.e., cavity-stabilized laser (or future optical clock transition),
lution frequency-comb spectroscopy, optical time-frequency as illustrated in Fig. 1(b). This optical stabilization provides
transfer, and precision LADAR.13,16,25,28 The layout of this phase coherence across the optical spectrum, greatly increas-
paper is as follows: Section II describes the optical design ing the comb performance and range of possible applications
and fabrication procedure for the frequency comb. Section III compared to the alternative direct stabilization of the repetition
describes the feedback actuators and phase locking electronics rate.50 A detailed discussion of comb stabilization is provided
used to achieve an optically coherent, self-referenced comb. in Section III.
Section IV discusses the performance of the frequency comb in
terms of phase noise, optical coherence, and “phase-slip” free
A. Polarization-maintaining fiber design
operation. As discussed in Section IV, all of the performance
metrics represent residual, in-loop values, i.e., these represent The fiber frequency comb design consists solely of PM
the limit of comb performance in the absence of an absolute single-mode fiber. Six different types of PM fibers are used (see
optical reference. Appendices A–C address the fiber lengths Appendix A). All six have Panda-type stress rods to provide
and types, a method to build a comb with a specific repetition the difference in index between the fast and slow axes; other
rate, and packaging of the optical system. PM-fiber types with similar properties could be substituted.
The gain medium for both the femtosecond fiber-laser and
the fiber amplifier is provided by erbium (Er) doped fibers. A
II. FREQUENCY COMB OPTICAL SYSTEM
PM highly nonlinear fiber (HNLF) enables generation of the
The design presented here is based on three similar fre- octave-spanning spectrum.51 In addition, standard telecommu-

23 September 2024 17:53:47

quency combs that have been operating continuously in our lab nications Panda PM-980 and PM-1550 fibers are used. The all-
for durations up to a year. These combs, hereafter labelled “A,” PM-fiber design requires a PM fiber fusion-splicer for comb
“B,” and “C,” operate with a pulse repetition-rate (comb mode fabrication and typical splice losses for the frequency comb
spacing) near 200 MHz, but more recent versions operate optical system are <0.5 dB.
equally well at repetition rates from 100 to 300 MHz with only
modest changes to the amplifier.
B. The femtosecond laser cavity
The basic concept of the frequency comb is reviewed
in Fig. 1.1,2 While “free-running combs”—i.e., unstabilized At the heart of the frequency comb system is an all-PM-
mode-locked lasers—are adequate for some applications, our fiber femtosecond laser cavity. As a positive feedback system,

FIG. 1. (a) Generalized schematic of a fiber frequency comb. The heart of a frequency comb is the mode-locked femtosecond laser, which produces a spectrum
of discrete comb modes spaced by the repetition frequency, f rep, covering a modest spectral bandwidth. This initial comb structure is amplified and further
spectrally broadened to an octave in highly nonlinear fiber (HNLF) to produce the spectrum idealized in part (b). Periodically poled lithium niobate (PPLN)
is used to compare the two ends of this optical spectrum for stabilization of the carrier-envelope-offset frequency, f ceo. (b) Idealized frequency comb showing
relevant optical and RF frequencies. The frequency of the n-th comb mode is given by the standard comb equation ν n = nf rep + f ceo. The comb is stabilized by
phase-locking the RF heterodyne frequency, f opt, between the n 0th comb mode of frequency, ν0, and a stable cw laser of frequency, νlaser, and by phase-locking
the RF carrier-envelope-offset frequency f ceo, which is generated via the conventional 1 f -to-2 f detection scheme.50
081301-3 Sinclair et al. Rev. Sci. Instrum. 86, 081301 (2015)

FIG. 2. (a) The femtosecond laser consists of the linear cavity bounded by the semiconductor saturable absorbing mirror (SESAM) on one end and the 80%
reflective dielectric coated output coupler on the other end. (b) Micro-optic design. (c) Output spectrum with a full-width-half-maximum (FWHM) of 12.2 nm.

the femtosecond laser is generally the most delicate part of any here has a 2 ps relaxation time constant, 6% modulation depth,
frequency comb and must generate an uninterrupted, optically and a 9% low signal absorption. The lens-based design also
coherent, low timing-jitter train of pulses that will propagate reduces the risk of damage to the SESAM.55 The micro-optic
through the rest of the system. Therefore, we have selected a contains two other critical components. First, it contains a
simple femtosecond laser design, offloading as much of the polarizer aligned with the slow-axis of the PM fiber, which
power and optical-bandwidth generation as possible onto the then defines the polarization of the femtosecond laser. Second,
amplifier. it contains a dichroic mirror to allow for the injection of the
Figure 2 shows the simple linear all-PM-fiber cavity, con- 1480-nm light that pumps the Er-doped gain fiber. In our orig-
sisting of PM Er-doped fiber spliced to a PM-1550 fiber. The inal design of Ref. 49, the laser was pumped at 980 nm. How-

23 September 2024 17:53:47

output coupler is formed by a FC/PC connector junction where ever, 1480-nm pump light is preferred for the following reason:
the face of one FC/PC connector has been coated with an the frequency comb noise is driven by the pump-laser relative-
80% reflective dielectric coating. The output coupler is dis- intensity-noise (RIN). This RIN is minimized by (i) inserting
cussed in detail in Appendix B. At the other end of the cavity, an isolator between the pump diode and the femtosecond laser
the PM Er-doped fiber is coupled into a custom micro-optic cavity and (ii) operating the pump diode at maximum power
component containing a SESAM to enable self-starting mode- followed by an attenuator to reach the desired pump power
locked operation.23,34,41,45,52 The Er-doped fiber length is not for the femtosecond laser. (This is done to take advantage
critical and ranges from 8.5 cm to 16.5 cm (see Table II in of the fact that the pump RIN increases as the square root
Appendix A); however, the shorter length of 8.5 cm is close of its power—running the pump at its maximum power and
to the minimal length for laser operation. Lengths longer than attenuating to the power desired by the laser effectively results
16.5 cm tend to shift the laser spectrum to longer wavelengths in a lower pump RIN seen by the laser.) Both isolators and
making later amplification and compression stages more diffi- attenuators are lower cost, more compact, and more efficient
cult. The PM-1550 fiber forms the remainder of the cavity and at 1480 nm than at 980 nm.
has the requisite length (of ∼33.5 to 41.5 cm) to support a Both the Er-doped fiber and the PM-1550 fiber in the
repetition rate of 200 MHz. Appendix B discusses a method to cavity are anomalous dispersion fiber, yielding a net round-
achieve an exact repetition rate, which is critical in applications trip dispersion of −0.22 ps2, leading to soliton mode-locking.
such as dual-comb spectroscopy53 or microwave generation This intra-cavity dispersion is high compared to other de-
through optical frequency division.54 The femtosecond laser signs,44,56–58 which increases the carrier-envelope-offset
achieves self-starting mode-locked operation with an output of noise.44,57,59 However, this high dispersion offers several ad-
10 mW and a 12–13 nm spectral bandwidth centered around vantages. First, small changes in the net dispersion due to
1560 nm (see Fig. 2(c)) and pulse widths of 200–220 fs for fiber length variations have little overall impact on the femto-
combs A–C. second laser performance leading to a more reproducible
Manufacturing of the custom micro-optic component is design. Second, both the PM Er-doped and PM-1550 fibers
well within the capabilities of fiber component suppliers and have a similar dispersion of ∼17 (ps/nm)/km at 1550 nm,
serves several critical functions. First, the spot-size on the so modest variations in their relative lengths have nearly
SESAM must be tuned to achieve a fluence of approximately inconsequential impact on laser performance. Finally, recent
twice the SESAM saturation value to optimize the femto- theoretical studies have shown that a soliton-wake instability60
second laser spectral bandwidth and pulse duration. The use is possible for low dispersion cavities if the SESAM time
of a gradient index (GRIN) lens inside the micro-optic hous- constant is too long. This soliton-wake instability interferes
ing (Fig. 2(b)) expands the beam to a 20 µm diameter spot with mode-locking of the laser. At the relatively high net
size, twice the diameter that would result from directly butt- dispersion here for the 2 ps SESAM time constant, the mode-
coupling the fiber to the SESAM. Given our ∼50 mW intra- locking is robust. At lower anomalous dispersions of 0 to
cavity power, the pulse intensity at the SESAM is twice the −0.1 ps2, mode-locking is not observed presumably due to this
SESAM saturation intensity of 50 µJ/cm.2 The SESAM used instability.
081301-4 Sinclair et al. Rev. Sci. Instrum. 86, 081301 (2015)

FIG. 3. Detailed schematic of all-PM-fiber frequency comb. HNLF: highly nonlinear fiber, PPLN: periodically poled lithium niobate, OC: output coupler, ISO:
isolator, WDM: wavelength division multiplexer, BPF: bandpass filter, PBS: polarizing beam splitter, PZT: piezo-electric transducer, SESAM: semiconductor
saturable absorber mirror.

23 September 2024 17:53:47

C. Amplification and spectral broadening remainder of the amplifier output (∼220 mW) is compressed
in 0.5–1.0 m of PM-1550 fiber, where the length is chosen
The femtosecond laser is followed by an amplifier, a to be a few centimeters short of maximal compression. These
HNLF for spectral broadening, and a periodically poled lithi- compressed pulses are then launched into 24–28 cm length of
um niobate (PPLN) waveguide to double the wavelengths PM_HNLF to produce an octave-spanning spectrum shown in
near 2 µm for the 1 f -to-2 f method of detection of f ceo. Fig. 4. The octave-spanning spectrum of Fig. 4(b) shows the
The full fiber design is considerably more complicated than strongly varying intensity typical of supercontinua produced
Fig. 1(a) and is given in Figure 3. While the fiber and fiber- via HNLFs.51,62–65 The PM-HNLF used here had a nominal
optic components occupy less than 0.1 l in volume, they are nonlinearity of 20/W km−1 and a dispersion of 2.3 (ps/nm)/km
enclosed in 0.7 l package (Appendix C). Appendix A provides at 1550 nm.51
detailed information on fiber types and lengths. In order to generate the f ceo signal in the standard 1 f -to-
As shown in Fig. 3, the femtosecond laser described in 2 f referencing scheme,66–68 the full octave-spanning spectrum
Sec. II B has a 5% tap to both monitor its output and to generate is coupled through a commercial, fiber-coupled, ridged, PPLN
the repetition rate frequency, f rep, signal. To overcome the rela- waveguide, which frequency doubles the light at 2128 nm to
tively low-power and spectrally narrow output from the laser, 1064 nm.69 This commercial waveguide has a built-in thermo-
we use a nonlinear Er-doped fiber amplifier that both amplifies electric cooler to tune the PPLN temperature for optimized
the 10 mW laser output to 300 mW and compresses the 200 fs doubling at 2128 nm with a conversion efficiency of 5%-10%
pulses to a ∼70 fs pulse width.61 The amplifier consists of 1- including insertion losses and a bandwidth of ∼3 nm (corre-
2 meters of normal dispersion Er-doped fiber, necessary for the sponding to a pulse duration of 0.4 ps for the doubled light
later pulse compression as described in Ref. 61, that is forward pulse). As shown in Fig. 3, a wavelength division multiplexer
and backward pumped with 800–900 mW of 980 nm pump (WDM) and narrowband 1064 nm filter select the fundamental
light. This Er-doped fiber is either a combination of low-gain and doubled ∼1064-nm light pulses. f ceo is the RF frequency
(6 dB/m) fiber and high-gain (100 dB/m) fiber or solely high- from the heterodyne signal between the fundamental and
gain fiber. Note that while the high-gain Er-doped fiber has a doubled light66–68 (see Fig. 1(b)).
nominal core absorption of 100 dB/m at 1530 nm, its exact However, detection of this heterodyne signal requires
value can vary from spool to spool. Therefore, the Er-doped temporal overlap between the fundamental and doubled light
fiber length is set by optimizing the spectral width (or power) pulses. This temporal overlap can be achieved by “fine tuning”
at the amplifier output. While the combination of high-gain and the length of the PM fiber between the PM-HNLF and PPLN,45
low-gain Er-doped fiber allows us to tune the amplifier disper- but this approach is difficult, given typical fiber lead lengths
sion independently of gain, in several more recent designs, we on the fiber-coupled PPLN waveguide. Therefore, we instead
have found that use of only ∼1.3 m of high-gain Er-doped fiber use an “in-line” fiber interferometer that takes advantage of
is close to optimal and is simpler. A 20% tap at the amplifier the differential delay between the fast and slow axes of a PM
output redirects ∼70 mW of comb light for applications. The fiber.49 The in-line interferometer is created by splicing the
081301-5 Sinclair et al. Rev. Sci. Instrum. 86, 081301 (2015)

FIG. 4. Typical spectral outputs. (a) Spectrum after the amplifier with a ∼50-nm full-width at half-maximum. At the input to the PM-HNLF, the average power
is 220 mW and the pulse duration is ∼70 fs. (b) Octave-spanning spectrum after the PM-HNLF.

PM fiber output of the 1064-nm filter to a second PM fiber To fully stabilize the comb, we require a second measure-
at a 45◦ rotation, which projects the fundamental and doubled ment. As discussed in the introduction to this section, we stabi-
light onto both fast and slow axes. After a suitable delay length, lize the frequency of a single comb mode via a RF heterodyne
chosen to optimize the f ceo SNR, the light from the slow and measurement against a cavity-stabilized cw laser. Here, we use
fast axes is recombined via a micro-optic polarizer oriented at a cavity-stabilized laser at either 1535 nm or 1560 nm. These
45◦. Typically, the ratio of the fiber length in the in-line inter- lasers could equally well be replaced by an optical clock, a
ferometer to the fiber length between the PM-HNLF and PPLN robust portable cavity-stabilized laser,29 or even a freerunning
is about 10:1. The two outputs from the micro-optic polarizer diode laser70 or fiber laser typical of dual-comb spectros-
contain the temporally and spatially overlapped fundamental copy.16 The comb light from the 20% tap is combined with
and doubled light which are directed to a balanced detector to this light in a 50/50 PM fiber combiner, filtered by a 0.5 nm-

23 September 2024 17:53:47

produce the f ceo signal, shown in Fig. 5(a). wide WDM, and detected to yield the RF signal shown in
Fig. 5(b). The RF frequency of this signal, f opt, is the difference
between the frequency of the cw laser and nearest comb mode
(see Fig. 1(b)). Stabilization to cw lasers (optical standards) at
wavelengths outside of the amplifier spectrum would require a
modification of the optical design. For example, stabilization
to a laser near 1 µm would require accessing some of the su-
percontinuum light prior to f ceo detection, and stabilization at a
visible wavelength might require a second fiber-coupled wave-
guide doubler, tuned to the appropriate wavelength, located at
one of the outputs.

III. COMB STABILIZATION: PHASE-LOCKING fceo

AND fopt
A frequency comb can be stabilized either to an optical
standard or via direct stabilization of the repetition rate to a
RF standard. In either case, f ceo is phase locked to a stable RF
frequency. Here, we use optical stabilization, which is much
more challenging but also yields higher performance than RF
stabilization of the repetition rate. The challenge and improved
performance stem from the fact that acoustic or mechanical
perturbations to the laser will perturb the repetition rate but
also cause the comb structure to breathe about ∼0-1 THz.44,71
Any changes in the repetition rate will be magnified by a
factor of 1 × 106 in the position of the optical comb mode
and therefore in the optical heterodyne signal against the cw
laser. In other words, use of optical stabilization provides
an enhanced “lever-arm” to detect these environmental noise
effects. Full suppression of the noise in the optical domain
FIG. 5. (a) Free-running f ceo heterodyne signal showing a 42 dB SNR in
has a corresponding multiplication in technical difficulty but
a 300 kHz resolution bandwidth (RBW). (b) Free-running f opt heterodyne
signal between the cavity-stabilized cw laser and nearest comb mode showing leads to an optically coherent comb with femtosecond-level
a 46 dB SNR in a 300 kHz RBW. residual timing jitter. Of course, if this level of performance
081301-6 Sinclair et al. Rev. Sci. Instrum. 86, 081301 (2015)

is not required, the system described here could be easily A. Field-programmable gate array (FPGA) based
reconfigured to directly phase lock the comb spacing, f rep, to digital control
a RF standard. Digital control and real-time digital signal processing
Both f ceo and f opt are phase locked to RF frequencies implemented on FPGA platforms have enabled significant
derived from f rep, which is known as full self-referencing.72,73 advances for both real-time processing of frequency comb sig-
The optical frequency of the nth comb mode is then given by nals12,70 and phase-locking of ultra-stable cavity-stabilized cw
(n−n )
νn = νn0 + n 0 (νn0 − f ceo), where νn0 is the frequency of the lasers,30,74 overcoming many drawbacks inherent to conven-
0
comb mode at n = n0 that is locked to the optical standard.50 tional analog systems. Specifically, digital processing allows
Note that this equation is not the usual νn = n f rep + f ceo; rather for the unwrapping and continuous tracking of phase excur-
f rep is indirectly stabilized as f rep = (νno − f ceo) /n0. There- sions of 106 radians or more, rather than the ±π/2 limit imposed
fore, even if νn0 is tightly stabilized to an optical standard, by an analog phase-detector. This continuous tracking protects
any noise on the f ceo signal will lead to noise in f rep. Here, against phase-slips from environmental perturbations. In addi-
the majority of the pulse-to-pulse timing jitter does in fact tion, the system has improved immunity to electro-magnetic
arise from the phase noise of f ceo, which is best understood interference (EMI) because the raw detector signals for the f ceo
by visualizing the accordion-like motion of the comb modes and f opt heterodyne beats are digitized immediately after a RF
about νn0 in the frequency domain.44,71 In that picture, it is also anti-aliasing filter, and the digital processing is noise-free until
clear that the comb will have the highest optical coherence generation of the analog control-signal in a digital-to-analog
near n = n0 and that coherence will degrade with increasing converter (DAC). The digital system, of course, allows for sim-
distance from this lock point. As shown in Section IV, for ple fine-tuning of loop parameters as well as easy adjustments
our system, the optical coherence is preserved (less than one of the lock frequencies and filter bandwidths. Finally, while
radian) over a bandwidth of ∼110 to 200 THz, depending on not implemented here, more sophisticated state observer-
the comb, centered about the mode at n = n0. predictor control methods would further increase robustness.
We follow now the standard approach for stabilization These advantages do require expertise in FPGA programming
of frequency combs, which implements a phase-locked loop to implement the digital signal processing algorithms. Here,
around both f ceo and f opt (see Fig. 6). The RF frequency, f ceo, we seek to provide a high level view of the processing and

23 September 2024 17:53:47

is phase-locked via feedback to the pump laser current of the control algorithms that could be implemented on a range of
femtosecond laser and the RF frequency f opt is phase-locked digital processing platforms. The exact details will depend
via feedback to the cavity length, which is controlled through on the platform and are therefore beyond the scope of this
a “slow” and a “fast” piezoelectric transducer (PZT) bonded paper.
to the intracavity fiber. Figure 6 shows the overall system. Figure 7 provides a basic outline of the FPGA-based
As with any phase-locked loop, the two critical parameters control system, which consists of input analog-to-digital con-
are a large dynamic range (to avoid phase-slips) and a large verters (ADCs), an in-phase and quadrature (IQ) detection
feedback bandwidth (to suppress noise). In the past, we have module for detection of the phase, a loop filter, output DACs,
used conventional op-amp-based analog control-electronics. and built-in real-time diagnostics through zero-deadtime fre-
Here, we focus on two critical improvements to our previous quency counters, phase-noise analyzers, and a vector network
approach: (i) the implementation of digital signal processing analyzer (VNA) for transfer function measurements. This is
for the phase lock and (ii) a standardized design for high- currently implemented with hardware originally designed for
bandwidth cavity length control through a PZT bonded to the locking cw lasers to ultra-stable optical cavities30,74 and con-
fiber. We first discuss the digitally based feedback control and sists of a custom analog input/output daughterboard and an
then the PZT design. OpalKelly XEM integration module with Xilinx Spartan-6.75

FIG. 6. Schematic of the fully phase-locked frequency comb. PZT: piezo-electric transducer, FPGA: field-programmable gate array.
081301-7 Sinclair et al. Rev. Sci. Instrum. 86, 081301 (2015)

23 September 2024 17:53:47

FIG. 7. Overview of FPGA-based digital control. ADC: analog-to-digital converter, DAC: digital to analog converter, PID: proportional integral derivative.

The two 16 bit, dc-coupled ADC inputs operate with a 50 MHz has implemented two zero-deadtime, triangular averaging fre-
input bandwidth and 100 MS/s. One 16-bit DAC output is quency counters operating at a 1-s gate time to monitor the
used for the pump diode current control and two 16-bit DACs long-term performance.
are used for the PZT control, one of which (the “slow PZT The phase-lock performance is determined by the feed-
output” in Fig. 7) is a 0-64 V output. The FPGA is clocked back bandwidth. The transfer functions for feedback to both
directly from half the repetition rate, enabling self-referenced f ceo and f opt are shown in Fig. 8. The response of f ceo to
operation. Standard digital-signal-processing routines are im- pump diode power has an intrinsic pole at ∼20 kHz associated
plemented, as described qualitatively below. with the gain of the Er-doped fiber.59,76 The derivative term
The RF signals f ceo and f opt are digitized and mixed down in the PID loop compensates for this pole and extends the
via IQ detection against the phase-lock frequency, q f rep (which feedback bandwidth to ∼100 kHz, at which point it is limited
is in general different for the two phase locks). For a fully by the bandwidth of the current controller and a second pole
self-referenced system, this lock frequency, q f rep, is derived originating from the femtosecond laser dynamics. For future
from f rep and q = M/N, where M is any integer and N = 248. systems, inclusion of fast small intracavity loss modulators,
The phase of the baseband signal is calculated using an arctan- such as graphene modulators grown onto the SESAM, could
gent operation and unwrapped (not shown in the diagram) to provide even higher feedback bandwidths for f ceo.77 Feedback
±3 × 106 radians of linear range, which allows linear behavior to control f opt is achieved via tuning of the femtosecond laser
even under significant perturbations. As long as frequency/ cavity length through two PZTs glued to the intracavity fiber,
phase excursions caused by environmental perturbations to the specifics of which are discussed in Sec. III B. The stabilized
the frequency comb do not exceed the input filter bandwidth RF signals of f ceo and f opt are shown in Fig. 9 and the residual
(25 MHz in this work), they are tracked and the phase is recov- noise is discussed in detail in Section IV B.
ered without 2π ambiguities. The loop filter which follows the
phase unwrapping operation is a tunable proportional-integral-
B. Femtosecond laser cavity length control
derivative (PID) filter.
A built-in VNA allows for measurement of transfer func- In order to stabilize f opt, feedback is applied to adjust the
tions and real-time monitoring of residual lock phase-noise. femtosecond laser cavity length via fast and slow PZT fiber-
When desired, the VNA can provide a small dither to any of stretchers. The fast PZT fiber-stretcher modulates the cavity
the output signals to measure the transfer function of any given length with up to 100 kHz of feedback bandwidth while the
modulator in the frequency comb or to measure the sign of slower (<1 kHz bandwidth) long-travel PZT provides a larger
optical or CEO beats (to determine if the cw laser above or dynamic range corresponding to ∼600 MHz on f opt. Both
below the comb tooth in the f opt signal, for example). During PZTs are glued to the 250 µm cladding-diameter intracavity
normal operation, the dither is off and the VNA passively Er-doped fiber or PM 1550 fiber with consumer-grade 5-min
monitors residual phase noise on the locks. The FPGA also epoxy.
081301-8 Sinclair et al. Rev. Sci. Instrum. 86, 081301 (2015)

FIG. 9. (a) Phase-locked f ceo signal for comb “A” in a 6 Hz resolution

bandwidth (RBW). (b) Phase-locked f opt signal for comb “A” in a 6 Hz RBW.

23 September 2024 17:53:47

The inset shows that the linewidth is limited by the 0.3 s acquisition time.

FIG. 8. (a) Transfer function of the pump-diode power control for f ceo surprisingly robust and has maintained excellent damping
showing the first order pole leading to a roll-off at ∼20 kHz associated with properties even after being vibrated on a shaker table in
the femtosecond fiber-laser gain response. (b) Transfer function of the “fast”
PZT to control f opt, showing the flat response out to ∼100 kHz for both the
Ref. 49. The downside of this approach is that the best damping
“free-standing” (black line) and “pocket-clamp” (blue line) PZT mounting performance does require significant user adjustment.
methods (discussed in Section III B). Note the latter has a stronger bare PZT In the pocket-clamping scheme (Fig. 10(b)), the PZT held
resonance at 700 kHz, which limits the phase-locking bandwidth. in a small pocket and the ∼2 cm radius fiber arc is in the
horizontal plane of the laser baseplate. To avoid resonances,
For phase locks based on PZT fiber-stretchers, the feed- the two fiber arcs sit upon a double layer of electrical tape
back bandwidth is typically limited by phase-shifts associated attached to the baseplate. The fiber is clamped from above
with mechanical resonances. While the bare PZT stack has with a 2-mm-thick stiff rubber piece and thin 1.27-mm-thick
resonances at 700 kHz and higher frequencies, the mount- aluminum clamping plate, while monitoring the PZT trans-
ing of the fiber to the PZT potentially induces resonances at fer function in real time. Unsupported fiber is very prone to
much lower frequencies, on the order of 10 kHz. In order resonances, so the size of the PZT pocket does not exceed
to damp these resonances, care must be taken in the exact 2.9 mm for the 2-mm-long PZT used here. We find that the
PZT-fiber mounting geometry. We achieve 100 kHz bandwidth PZT should not be attached to the baseplate. While the pocket-
for our fast PZT fiber-stretcher using two different mounting clamping method is more straightforward and repeatable than
approaches (see Fig. 10): a free-standing scheme and a pocket- the free-standing method, the bare PZT resonance at 700 kHz
clamping scheme. is stronger (see Fig. 8) and can limit the bandwidth. However,
For both mounting schemes, the fiber (encased by its the servo bandwidth can still reach 100 kHz even with this limi-
250 µm buffer) is epoxied along the axis of a 2 mm × 2 mm tation. Higher feedback bandwidths have been achieved using
× 2 mm stack PZT. This attachment of the fiber results in a an intracavity electro-optic modulator (EOM),46,54,78,79 with
total fiber stretch that is approximately one-tenth of the total bandwidths as high as 4 MHz.79 However, EOMs are expen-
PZT throw. It is critical that there remains an arc in the fiber sive, have high third order dispersion, inconveniently located
as it leaves the PZT surface in order to preserve the dynamic fixed-points, and add intracavity loss. A high bandwidth PZT
range of the fiber-stretcher by preventing fiber compression. is a very attractive, low cost option for many applications.
However, the “violin”-like resonances between 1 kHz and 100 The low bandwidth PZT can be mounted with far less
kHz must be damped. In the free-standing design, a tight, effort as it is only driven at frequencies below 1 kHz. The fiber
∼2 cm radius of curvature, arc arises naturally as the fiber is epoxied to the PZT in the same manner as the fast PZT but
travels from the top of the PZT to the laser baseplate. Reso- no pocket is required and the PZT rests on the laser base with
nances in the suspended fiber are damped with modeling clay the fiber running over the top and gently arcing back down to
pressed up against the fiber on both sides. The modeling clay is the baseplate. The fiber and PZT are secured to the baseplate
081301-9 Sinclair et al. Rev. Sci. Instrum. 86, 081301 (2015)

FIG. 10. Methods of mounting a small PZT to achieve high feedback bandwidth. (a) Free-standing scheme side view. The ∼2 cm radius of curvature arc is
created as the fiber travels from the top of the PZT to the laser baseplate. The electrical leads (not shown) are attached to the side of the PZT. (b) Free-standing
scheme top view. (c) Pocket-clamping scheme side view. The electrical leads (not shown) are attached to the side of the PZT and emerge vertically out of the
pocket. (d) Pocket-clamping scheme top view. The in-plane ∼2 cm radius of curvature arc is formed when the fiber is clamped onto the baseplate.

with tape and epoxy. Similar to the fast PZT, the total stretch “in-loop” values. By residual, we mean that the values are all
induced in the fiber for the slow PZT is about one-tenth of the with respect to the underlying “clock,” which in this case is the
total PZT throw. cavity-stabilized laser. This cavity-stabilized laser will drift as

23 September 2024 17:53:47

Finally, active thermal control of the femtosecond laser the optical cavity length changes and it has no absolute stability
cavity length is necessary for operation in uncontrolled envi- in the sense of a true atomic clock. By “in-loop,” we mean
ronments because the optical path length of the cavity fiber that the quantities are measured from the same signals used
shifts by ∼10 ppm/˚C. Given the dynamic range of the slow to phase lock the comb. There is no separate, second comb
PZT fiber stretcher, the temperature of the cavity fiber must against which the performance was evaluated. (In Ref. 49,
remain within 0.3 ◦C. (A 0.1 ◦C change corresponds to a we did compare that the comb mode linewidth to a second
200 MHz shift in f opt.) Therefore, thermo-electric coolers cavity stabilized laser to verify the overall comb linewidth was
(TECs) are placed underneath the aluminum box that houses sub-Hz.) In general, such residual, in-loop values represent
the femtosecond laser and amplifier (see Appendix C). We an upper limit to the frequency comb performance. In any
control the aluminum box temperature based on an internal experiment, there will be varying path lengths and other effects
thermistor. For long term operation, we implement a third that can degrade the comb light that is ultimately delivered
control loop that adjusts the temperature set point to keep the to the experiment. As a result, comb-based experiments are
slow PZT within its dynamic range. For environments with typically designed to minimize these effects in order to achieve
limited temperature and humidity regulation, this third stage the best possible performance.
of control is critical.
TABLE I. Comb performance (see also Section IV B).

Property A B C
IV. PERFORMANCE
Femtosecond laser center 1563.6 1558.8 1560.0
In this section, we present the performance of combs A, wavelength (nm)
B, and C, all of which have been in continuous operation for 3 dB femtosecond laser width 12.2 [210] 11.5 [222] 12.6 [202]
at least seven months. During that period, they have routinely (nm) [equivalent pulse width (fs)]
operated on a wheeled cart near open windows from summer 3 dB amplifier width (nm) 54 52 51
20% amplifier tap output power ∼70 ∼70 ∼70
to winter demonstrating their ability to operate outside the
(mW)
well-controlled metrology laboratory. For these three combs,
f ceo integrated phase noisea (rad) 2.0 3.6 2.6
we have found that output power, spectral shape, and center
10 MHz to 6 Hz
wavelength vary little between them. Comb B does have a f opt integrated phase noisea (rad) 0.12 0.14 0.15
slightly higher intrinsic f ceo phase noise. Therefore, its esti- 10 MHz to 6 Hz
mated pulse-to-pulse timing jitter is higher and its projected Estimated pulse-to-pulse timing 1.6 2.9 2.1
coherence bandwidth is lower as discussed in Section IV B. jittera (fs)
Table I presents a summary of various performance metrics Projected coherence bandwidtha 200 110 150
for the three combs. (THz)
All the performance parameters given here—the phase a Allphase noise parameters here are residual, “in-loop” values so the coherence is with
noise, timing jitter, and frequency noise—are “residual,” respect to the reference oscillator.
081301-10 Sinclair et al. Rev. Sci. Instrum. 86, 081301 (2015)

A. Long term phase coherence

A significant limitation on comb stability is caused by
a phase unwrapping error, i.e., a phase-slip. Phase-slips can
be driven by two processes: (i) environmental perturbations
that cause the heterodyne signal to momentarily exceed the
signal processing filter bandwidth, BW, and (ii) limited SNR
that causes a (nearly) 2π error in the phase calculation. The
larger the filter bandwidth used to process the signal, the
lower the probability of an “environmentally driven” phase-
slip. However, too large filter bandwidth will decrease the SNR
and ultimately lead to “SNR-driven” phase-slips.
The probability of this SNR-driven phase-slip can be
calculated to determine the optimal filter bandwidth. For a
FIG. 11. The rate of phase slips versus SNR of the f ceo signal at a fixed
signal with additive white Gaussian
( noise,√this
) phase-slip processing bandwidth of 25 MHz. Measurements of the rate were taken by
rate is given by 12 BW × erfc 10S N R/20/20/ 2 , where erfc mixing in white noise to the signal and monitoring the system for phase slips
is the complimentary error function, and the SNR is in dB (red points), as well as by adding noise to a pre-recorded f ceo signal (dark blue
for the given filter bandwidth.80 Note that this is very similar dots). The measurements and simulation agree with the analytical calculation
assuming Gaussian noise (light blue curve). In our comb operation, the SNR
to computing the bit-error-rate for binary-phase-shift-keying is typically ∼35 dB in a 25 MHz bandwidth resulting in a negligible rate of
(BPSK). To verify that this expression accurately describes phase-slips.
our system, we applied white electrical noise to the f ceo
signal before it entered the digital-locking electronics and
monitored the un-degraded signal with a separate frequency relative to Ref. 49, comes largely from the implementation
counter to measure the phase-slip rate. A frequency jump on of the digital control. Phase-slip free operation over extended
an independent frequency counter was recorded as a phase-slip periods should enable time-based clock comparisons and
when it exceeded a 1 Hz change from the mean frequency. In synchronization of clocks over long time scales via optical

23 September 2024 17:53:47

addition, we performed a simulation by adding Gaussian noise free-space time transfer where phase-slips are deleterious.28
to a previously recorded high SNR f ceo signal and simulating
the digital processing to find the resulting phase slip rate. Both
B. Phase noise and timing jitter
the measurement and the simulation results match the formula
given above as shown in Fig. 11. The phase-noise spectra, shown in Fig. 13, provide a
From Fig. 11, it is clear that the rate of SNR-driven phase- measure of the performance of the combs on short time scales.
slips is a very strong function of the SNR and, therefore, For the three combs, the integrated phase-noise for f opt ranges
filter bandwidth (since SNR ∝1/bandwidth). A SNR of 20 dB from 0.12 to 0.15 rad from 10 MHz to 6 Hz and for f ceo
is required for long-term phase-slip-free operation. However, ranges from 2.0 to 3.6 rad from 10 MHz to 6 Hz with values
operation close to the 20 dB SNR limit results in a sensitivity to extrapolated to Nyquist at 100 MHz of 0.16–0.23 rad and
slight changes in power, noise floors, or transient RF pick-up 2.0–3.6 rad, respectively. (Note that the sub-radian phase noise
as they will cause a reduction in SNR and thus potentially a for f opt results in the sharp coherent peak in the RF signal
phase-slip. Here, we operate at a SNR of 25–35 dB, achieved shown in Fig. 9(b), while the 2 rad phase noise for f ceo of comb
in a 25 MHz filter bandwidth, for both the f ceo and f opt signals. “A” results in a coherent peak on top of a broader background
Figure 12 shows the long term stability of a locked comb, in the RF signal shown in Fig. 9(a).) For all of these combs
which operated fully stabilized without phase-slips for 91 h to achieve the lowest phase noise, the femtosecond laser was
until the measurement was stopped as the combs were needed pumped to just below the point of cw breakthrough, i.e., the
for other projects. The improved long term performance, pump power at which the output spectrum shows a narrow

FIG. 12. Long-term stability of the combs. Counted offset of f ceo (blue) and f opt (red) signals from their lock frequencies with a 1 s gate time for comb “B.” The
standard deviation for f ceo is 0.9 mHz and for f opt is 0.1 mHz. There were no phase-slips over the 91 h of the measurement (as evidenced by the lack of >1 Hz
frequency excursions).
081301-11 Sinclair et al. Rev. Sci. Instrum. 86, 081301 (2015)

pulse-to-pulse timing jitter is the residual phase-noise on f ceo.

(See the lock point diagram in Fig. 1(b) and discussion in
Section III.)

V. CONCLUSIONS
We have developed a self-referenced all-PM-fiber fre-
quency comb design capable of coherent operation outside
of a well-controlled laboratory setting within a small liter-
volume optics package. Multiple frequency combs have been
constructed and tested. With the use of digitally based control
and ∼100-kHz closed-loop bandwidths, the combs operate
with in-loop optical frequency uncertainties of <1 mHz at 1 s,
more than sufficient for use with optical clocks.83 We have
also shown that these combs operate with no phase-slips for
91 h. Finally, the residual 1.6–2.9 fs timing jitter is comfortably
below the ∼10 fs timing jitter acceptable for most precision
comb applications such as optical time transfer, LADAR, and
comb spectroscopy.5–28 It is hoped that this optically coherent
and robust comb design will facilitate “real-world” applica-
tions of these metrological tools.

ACKNOWLEDGMENTS
We acknowledge assistance in comb fabrication and
design from Ron Hui, assistance in digital–control design from

23 September 2024 17:53:47

David Leibrandt, and helpful discussions with Tara Fortier,
Gabe Ycas, and Marla Dowell. This work was funded by
the DARPA PULSE program and the National Institute of
Standards and Technology (NIST).
FIG. 13. (a) f ceo phase noise (blue, left axis) and integrated phase noise
(green, right axis) from 10 Hz to 10 MHz for comb “A.” (b) f opt phase noise APPENDIX A: CRITICAL FIBER LENGTHS
(red, left axis) and integrated phase noise (orange, right axis) from 10 Hz to
10 MHz for comb “A.” Fabrication of an octave-spanning comb requires the use
of specific fiber lengths. Critical fiber lengths are given in

frequency spike, corresponding to a cw mode, in addition to TABLE II. Critical fiber lengths for three fully characterized combs. All
the broad frequency spectrum of the comb. lengths are given in centimeters.
The comb is optically coherent around the optical lock
Section of comb A B C
point at 1535 nm. For many applications, it is important that
this optical coherence extends across all comb modes. We Ia Gain medium: Anomalous dispersion 16.0 16.5 8.5
define the projected coherence bandwidth as the frequency Er-doped fiber (Nufern PM-ESF-7/12575)
range centered at the optical lock point for which the phase Ib Total fiber in femtosecond laser cavity 50.0 50.0 50.0
noise on the comb modes remains below one radian. Here, the II. Between output coupler and amplifier 49 44 49
(PM-1550)
2.0 to 3.6 rad phase noise on f ceo corresponds to a projected
III. Gain medium: Normal dispersion 228.5a 243.0b 228.5
coherence bandwidth across the comb spectrum of 200 to
Er-doped fiber (OFS EDF08-PM and nLight
110 THz. This means that the entire 150 THz-wide (1 µm- Er80-4/125-HD-PM75)
2 µm) comb output, centered at the ∼195-THz optical lock IV. Compression between amplifier and 42 42 42
point, is coherent in most cases. Based on these numbers, even PM-HNLF (PM-980)
if the optical lock point were shifted to 1070 nm for an Al+ ion V. Compression between amplifier and 68.5 69.0 105.0
clock81 or 1156 nm for a Yb clock,82 the telecommunication PM-HNLF (PM-1550)
bands at 1550 nm used for free-space time and frequency VI. PM-HNLF (2.3 (ps/nm)/km at 1550 nm) 24.5 27.5 27.0
transfer28 would still be optically coherent. VII. PM-1550 between PM-HNLF and PPLN 52.0 48.5 43.5
The pulse-to-pulse timing jitter is calculated from the VIII. Delay length in in-line interferometer 544.5 500.0 388.0
phase noise at the f ceo and f opt lock points divided by the (PM-980)c
angular optical frequency separation, or for comb A, fs, assum- a This consists of 70 cm low gain Er-doped fiber followed by 158.5 cm of high gain
ing uncorrelated noise. For the noisier comb B, this jitter Er-doped fiber.
b This consists of 80 cm low gain Er-doped fiber followed by 163 cm of high gain
increases to 2.9 fs of pulse-to-pulse timing jitter. In all cases, Er-doped fiber.
because of the tight optical lock, the dominant source of c Generally, the fiber length ratio of VII to VIII will be about 1:10.
081301-12 Sinclair et al. Rev. Sci. Instrum. 86, 081301 (2015)

FIG. 14. Schematic of comb for determining critical fiber lengths. Critical fiber lengths are given in Table II. For the second WDM, the default fiber is PM980
on the common port, as noted in the figure. However, it is unlikely the use of PM980 or PM1550 is critical here.

23 September 2024 17:53:47

Table II with the corresponding Fig. 14 illustrating their loca- bare end of a FC/PC connector from several coating suppliers
tion. The lengths given are for the specific fiber types listed; using an array of processes. Generally speaking, the durability
however, other gain fibers or highly nonlinear fibers could be of the coating is not a critical requirement as it only ever
used with the appropriate adjustment in length as necessary. experiences compressive force. We have however found that
providing a coating supplier with FC/PC ends that have had
the boots removed and have only a 250 µm buffer on the
APPENDIX B: OUTPUT COUPLER DESIGN pigtail does reduce outgassing and leads to a higher yield on the
coating.
A range of output couplers have been tried in this laser
design. Reflections at 1560 nm have ranged from 80% to
90% with combs A-C all using an 80% which provides addi-
tional output power with no other observable change in perfor- APPENDIX C: CAVITY FABRICATION AND PRECISE
mance. (See Fig. 15.) Couplers at 75% have been tried as SELECTION OF THE FEMTOSECOND LASER
well but mode locking was unstable although such couplers REPETITION RATE
might work with a higher modulation depth SESAM. We have Many applications, such as high-precision spectroscopy,
used standard commercial dielectric coatings applied to the time-frequency transfer, and microwave generation, require
a femtosecond laser with a precisely defined repetition
rate.28,53,54 Unlike many laboratory-based frequency combs,
which can rely on a free-space variable delay-line for coarse
repetition-rate tuning,40,84 the all-fiber design must have the
optical path length set during comb fabrication. The repetition
rate can be fine-tuned by changing the temperature of the
femtosecond laser fiber (10 ppm/˚C). However, it is necessary
to have the total cavity length constructed within a few parts
in 105 of the targeted path length for operation at reason-
able temperatures. (One part in 105 translates to 5 µm for a
200 MHz linear cavity.)
We use a two-step approach to realize the desired repe-
tition rate during construction of the femtosecond laser cav-
ity. (See Fig. 16.) In the first step, the length of the two
halves of the laser cavity (the micro-optic assembly with Er-
doped fiber pigtail on one side and PM-1550 fiber pigtail with
FC/PC on the other side) is measured and co-aligned in a fiber
FIG. 15. Reflectivity curve of a typical dielectric coating. fusion-splicer, but not actually spliced. In order to measure the
081301-13 Sinclair et al. Rev. Sci. Instrum. 86, 081301 (2015)

round-trip cavity delay time, light from a coherent optical In the second fine-tuning fabrication step, the PC-
frequency-domain reflectometer (OFDR) is coupled into the connector at the end of the femtosecond laser cavity is pol-
cavity via the FC/PC connector. The OFDR signal returns the ished. Using standard diamond polishing paper, it is possible
time delay or optical path length of the PC-connector interface to remove the remaining excess cavity length in a controlled
at the near end of the femtosecond laser cavity and the SESAM fashion. As the cavity is already constructed, one can periodi-
on far end, see Fig. 16(b). Our comb-calibrated OFDR was cally measure the repetition rate by mating the newly polished
originally developed for LADAR applications25,27,85 and is end with the dielectric output coupler and pumping the cavity.
capable of measuring range-resolved optical reflections in the With some iteration, the cavity length can be tuned within
femtosecond laser cavity at a resolution of 0.4 ps (130 µm) and 10 µm of the desired length. For a 200 MHz cavity, one
a precision level of 1 fs. In general, this approach of optical can achieve the desired repetition rate with less than 2 ◦C of
length measurement could be reproduced with any high-end, temperature tuning applied to the femtosecond laser.
commercially available OFDR.86
Once the optical path length of the cavity has been deter-
mined, the appropriate length of fiber is removed via cleaving
APPENDIX D: PACKAGING
and the fibers are spliced together. Typically, this is an iterative
process. In the first pass, the cavity is left deliberately too long The optics package is housed in either two small alumi-
to allow for multiple cutbacks with OFDR measurements after num boxes of dimensions 16 cm × 11 cm × 2.5 cm (total
each. Note that the OFDR returns the round trip travel time volume of 0.7 l) connected via a 22 cm patch cable or a single
in the cavity to ∼1 fs, which directly gives the repetition rate. small aluminum box of dimensions 18 cm × 20 cm × 2.5 cm
However, the cleaving length requires knowledge of the index (total volume of 0.9 l), as shown in Fig. 17. Once the fiber is
of the cavity, which can easily be determined by the series placed inside of the boxes, large components are attached to the
of cutbacks. With care, one can achieve an optical cavity that aluminum housing via room temperature vulcanization (RTV)
is within 100 µm of the desired length, yet also deliberately silicone. Then, a two part silicone epoxy is added to the entire
longer than the desired value. volume to provide additional protection from vibrations. The

23 September 2024 17:53:47

FIG. 16. (a) Setup to measure and adjust the femtosecond laser cavity length/repetition rate during fabrication. A comb-calibrated optical frequency-domain
reflectometer (OFDR) measures the round trip delay in the cavity between reflections coming from the SESAM and the PC connector interface. (b) OFDR
signals from two different femtosecond lasers. Each peak corresponds to a reflection in the cavity with a time axis that is relative to a reference arm. Absorption
due to the Er-doped fiber in the cavity reduces the height of the SESAM peak. Curves are offset for clarity. (c) Repetition rate for a pair of femtosecond lasers
constructed using this technique after fine-tuning of the repetition rate via polishing showing the desired 4 kHz repetition rate separation.
081301-14 Sinclair et al. Rev. Sci. Instrum. 86, 081301 (2015)

TABLE III. Power consumption of comb system. The total power consump-
tion can be higher during system start-up than the values given here.

Typical current Voltage Typical power

Comb system element (mA) (V) (W)

1480 nm diode laser 700 2.7 2 (×2)

980 nm diode laser 1500 2.7 4 (×2)
Diode laser TEC 500 2.7 1 (×4)
PPLN TEC ... ... <1
Femtosecond laser housing 1000 8 8
TECs
Commercial photodetectors <250 15 <4 (×3)
for f opt, f ceo, and f rep
FPGA-based controllera 1800 5 9
Control electronicsb 200 c 5
Total for comb system ... ... 50
a These values do not include the low power consumption high-voltage output.
b This includes control electronics for the pump diode laser optical output, the pump
diode laser temperature control, the PPLN temperature control, and the femtosecond
laser housing temperature control. The actual current to the TECs is listed separately
above in the table.
c Operating voltages are 5 or 12 V.

1J. L. Hall, Rev. Mod. Phys. 78, 1279 (2006).

2T. W. Hänsch, Rev. Mod. Phys. 78, 1297 (2006).
3S. A. Diddams, J. Opt. Soc. Am. B 27, B51 (2010).
4N. R. Newbury, Nat. Photonics 5, 186 (2011).

23 September 2024 17:53:47

5A. Schliesser, M. Brehm, F. Keilmann, and D. van der Weide, Opt. Express

13, 9029 (2005).

6I. Coddington, W. C. Swann, and N. R. Newbury, Phys. Rev. Lett. 100,

013902 (2008).
7F. Adler, M. J. Thorpe, K. C. Cossel, and J. Ye, Annu. Rev. Anal. Chem. 3,

175 (2010).
8M. Godbout, J.-D. Deschênes, and J. Genest, Opt. Express 18, 15981

(2010).
9B. Bernhardt, A. Ozawa, P. Jacquet, M. Jacquey, Y. Kobayashi, T. Udem, R.

Holzwarth, G. Guelachvili, T. W. Hänsch, and N. Picqué, Nat. Photonics 4,

55 (2010).
10E. Baumann, F. R. Giorgetta, W. C. Swann, A. M. Zolot, I. Coddington, and

N. R. Newbury, Phys. Rev. A 84, 062513 (2011).

11A. M. Zolot, F. R. Giorgetta, E. Baumann, J. W. Nicholson, W. C. Swann, I.
FIG. 17. Aluminum housing designs. (a) Two small boxes each measuring
16 cm × 11 cm × 2.5 cm connected by a 22 cm patch cable. (b) Alternate Coddington, and N. R. Newbury, Opt. Lett. 37, 638 (2012).
12J. Roy, J.-D. Deschênes, S. Potvin, and J. Genest, Opt. Express 20, 21932
housing consisting of one 18 cm × 20 cm × 2.5 cm box.
(2012).
13A. M. Zolot, F. R. Giorgetta, E. Baumann, W. C. Swann, I. Coddington, and

N. R. Newbury, J. Quant. Spectrosc. Radiat. Transfer 118, 26 (2013).

14Z. Zhang, T. Gardiner, and D. T. Reid, Opt. Lett. 38, 3148 (2013).
epoxy used for potting is electrically insulating and very soft 15S. Boudreau, S. Levasseur, C. Perilla, S. Roy, and J. Genest, Opt. Express
with a Shore A hardness listed as soft gel. This simple design 21, 7411 (2013).
16G. B. Rieker, F. R. Giorgetta, W. C. Swann, J. Kofler, A. M. Zolot, L. C.
is similar, although significantly reduced in volume, from the
previous 6 l package that was shaken at 0.5 g rms while the Sinclair, E. Baumann, C. Cromer, G. Petron, C. Sweeney, P. P. Tans, I.
Coddington, and N. R. Newbury, Optica 1, 290 (2014).
comb was fully stabilized.49 17K. Minoshima and H. Matsumoto, Appl. Opt. 39, 5512 (2000).
18N. Schuhler, Y. Salvade, S. Leveque, R. Dandliker, and R. Holzwarth, Opt.

Lett. 31, 3101 (2006).

19P. Balling, P. Kren, P. Masika, and S. A. van den Berg, Opt. Express 17, 9300
APPENDIX E: POWER CONSUMPTION (2009).
20I. Coddington, W. C. Swann, L. Nenadovic, and N. R. Newbury, Nat.
Table III gives an estimate of the power consumption of Photonics 3, 351 (2009).
21M. Cui, M. G. Zeitouny, N. Bhattacharya, S. A. van den Berg, H. P. Urbach,
the comb system under typical operating conditions inside the
laboratory with the exclusion of the power consumption due to and J. J. M. Braat, Opt. Lett. 34, 1982 (2009).
22J. Lee, Y.-J. Kim, K. Lee, S. Lee, and S.-W. Kim, Nat. Photonics 4, 716
the inefficiency of the various power supplies. The total power (2010).
consumption is driven by three main categories, pump diode 23T.-A. Liu, N. R. Newbury, and I. Coddington, Opt. Express 19, 18501

laser light generation, temperature control of various sub- (2011).

24M. U. Piracha, D. Nguyen, I. Ozdur, and P. J. Delfyett, Opt. Express 19,
systems, and control electronics. Each diode laser is contained
11213 (2011).
within a butterfly package, which has a TEC to control the 25E. Baumann, F. R. Giorgetta, I. Coddington, L. C. Sinclair, K. Knabe, W. C.
diode laser temperature. Swann, and N. R. Newbury, Opt. Lett. 38, 2026 (2013).
081301-15 Sinclair et al. Rev. Sci. Instrum. 86, 081301 (2015)

26X. Wu, H. Wei, H. Zhang, L. Ren, Y. Li, and J. Zhang, Appl. Opt. 52, 2042 56B. R. Washburn, W. C. Swann, and N. R. Newbury, Opt. Express 13, 10622

(2013). (2005).
27E. Baumann, F. R. Giorgetta, J.-D. Deschênes, W. C. Swann, I. Coddington, 57R. Paschotta, Opt. Express 18, 5041 (2010).
and N. R. Newbury, Opt. Express 22, 24914 (2014). 58C. Kim, K. Jung, K. Kieu, and J. Kim, Opt. Express 20, 29524 (2012).
28F. R. Giorgetta, W. C. Swann, L. C. Sinclair, E. Baumann, I. Coddington, 59J. J. McFerran, W. C. Swann, B. R. Washburn, and N. R. Newbury, Appl.

and N. R. Newbury, Nat. Photonics 7, 434 (2013). Phys. B 86, 219 (2007).
29D. R. Leibrandt, M. J. Thorpe, J. C. Bergquist, and T. Rosenband, Opt. 60S. Wang, C. R. Menyuk, L. Sinclair, I. Coddington, and N. R. Newbury,

Express 19, 10278 (2011). in Conference on Laser and Electro-Optics (Optical Society of America,
30D. R. Leibrandt, J. C. Bergquist, and T. Rosenband, Phys. Rev. A 87, 023829 2014), p. SW3E.4.
(2013). 61K. Tamura and M. Nakazawa, Opt. Lett. 21, 68 (1996).
31Q.-F. Chen, A. Nevsky, M. Cardace, S. Schiller, T. Legero, S. Häfner, A. 62N. R. Newbury, B. R. Washburn, K. L. Corwin, and R. S. Windeler, Opt.

Uhde, and U. Sterr, Rev. Sci. Instrum. 85, 113107 (2014). Lett. 28, 944 (2002).
32L. Krainer, R. Paschotta, G. J. Spuhler, I. Klimov, C. Y. Teisset, K. J. 63K. L. Corwin, N. R. Newbury, J. M. Dudley, S. Coen, S. A. Diddams,

Weingarten, and U. Keller, Electron. Lett. 38, 225 (2002). B. R. Washburn, K. Weber, and R. S. Windeler, Appl. Phys. B 77, 269
33A. Bartels, D. Heinecke, and S. A. Diddams, Science 326, 681 (2009). (2003).
34J. Aus der Au, G. J. Spühler, T. Südmeyer, R. Paschotta, R. Hövel, M. Moser, 64K. L. Corwin, N. R. Newbury, J. M. Dudley, S. Coen, S. A. Diddams, K.

S. Erhard, M. Karszewski, A. Giesen, and U. Keller, Opt. Lett. 25, 859 Weber, and R. S. Windeler, Phys. Rev. Lett. 90, 113904 (2003).
(2000). 65J. M. Dudley, G. Genty, and S. Coen, Rev. Mod. Phys. 78, 1135 (2006).
35C. G. Durfee, T. Storz, J. Garlick, S. Hill, J. A. Squier, M. Kirchner, G. Taft, 66H. R. Telle, G. Steinmeyer, A. E. Dunlop, J. Stenger, D. H. Sutter, and U.

K. Shea, H. Kapteyn, M. Murnane, and S. Backus, Opt. Express 20, 13677 Keller, Appl. Phys. B 69, 327 (1999).
(2012). 67D. J. Jones, S. A. Diddams, J. K. Ranka, A. Stentz, R. S. Windeler, J. L. Hall,
36P. Del’Haye, A. Schliesser, O. Arcizet, T. Wilken, R. Holzwarth, and T. J. and S. T. Cundiff, Science 288, 635 (2000).
Kippenberg, Nature 450, 1214 (2007). 68R. Holzwarth, T. Udem, T. W. Hänsch, J. C. Knight, W. J. Wadsworth, and
37T. J. Kippenberg, R. Holzwarth, and S. A. Diddams, Science 332, 555 P. S. J. Russell, Phys. Rev. Lett. 85, 2264 (2000).
(2011). 69S. Kurimura, Y. Kato, M. Maruyama, Y. Usui, and H. Nakajima, Appl. Phys.
38Y. Okawachi, K. Saha, J. S. Levy, Y. H. Wen, M. Lipson, and A. L. Gaeta, Lett. 89, 191123 (2006).
Opt. Lett. 36, 3398 (2011). 70J.-D. Deschênes, P. Giaccarri, and J. Genest, Opt. Express 18, 23358
39F. Tauser, A. Leitenstorfer, and W. Zinth, Opt. Express 11, 594 (2003). (2010).
40B. Washburn, R. Fox, N. Newbury, J. Nicholson, K. Feder, P. Westbrook, 71H. R. Telle, B. Lipphardt, and J. Stenger, Appl. Phys. B 74, 1 (2002).

and C. Jørgensen, Opt. Express 12, 4999 (2004). 72J. Ye, J. L. Hall, and S. A. Diddams, Opt. Lett. 25, 1675 (2000).
41T. R. Schibli, K. Minoshima, F. L. Hong, H. Inaba, A. Onae, H. Matsumoto, 73W. C. Swann, J. J. McFerran, I. Coddington, N. R. Newbury, I. Hartl, M. E.

I. Hartl, and M. E. Fermann, Opt. Lett. 29, 2467 (2004). Fermann, P. S. Westbrook, J. W. Nicholson, K. S. Feder, C. Langrock, and

23 September 2024 17:53:47

42P. Kubina, P. Adel, F. Adler, G. Grosche, T. Hänsch, R. Holzwarth, A. M. M. Fejer, Opt. Lett. 31, 3046 (2006).
Leitenstorfer, B. Lipphardt, and H. Schnatz, Opt. Express 13, 904 (2005). 74M. J. Thorpe, D. R. Leibrandt, T. M. Fortier, and T. Rosenband, Opt. Express
43H. Inaba, Y. Daimon, F.-L. Hong, A. Onae, K. Minoshima, T. R. Schibli, 18, 18744 (2010).
H. Matsumoto, M. Hirano, T. Okuno, M. Onishi, and M. Nakazawa, Opt. 75The use of trade names is necessary to specify the experimental results

Express 14, 5223 (2006). and cannot imply endorsement by the National Institute of Standards and
44N. R. Newbury and W. C. Swann, J. Opt. Soc. Am. B 24, 1756 (2007). Technology.
45I. Hartl, G. Imeshev, M. E. Fermann, C. Langrock, and M. M. Fejer, Opt. 76J. J. McFerran, W. C. Swann, B. R. Washburn, and N. R. Newbury, Opt. Lett.

Express 13, 6490 (2005). 31, 1997 (2006).

46E. Baumann, F. R. Giorgetta, J. W. Nicholson, W. C. Swann, I. Coddington, 77C.-C. Lee, S. Suzuki, W. Xie, and T. R. Schibli, Opt. Express 20, 5264

and N. R. Newbury, Opt. Lett. 34, 638 (2009). (2012).

47J. Lee, K. Lee, Y.-S. Jang, H. Jang, S. Han, S.-H. Lee, K.-I. Kang, C.-W. 78Y. Nakajima, H. Inaba, K. Hosaka, K. Minoshima, A. Onae, M. Yasuda,

Lim, Y.-J. Kim, and S.-W. Kim, Sci. Rep. 4, 5134 (2014). T. Kohno, S. Kawato, T. Kobayashi, T. Katsuyama, and F.-L. Hong, Opt.
48T. Wilken, M. Lezius, T. W. Hansch, A. Kohfeldt, A. Wicht, V. Schkolnik, Express 18, 1667 (2010).
M. Krutzik, H. Duncker, O. Hellmig, P. Windpassinger, K. Sengstock, 79W. C. Swann, E. Baumann, F. R. Giorgetta, and N. R. Newbury, Opt. Express

A. Peters, and R. Holzwarth, in Conference on Laser and Electro-Optics 19, 24387 (2011).
(Optical Society of America, San Jose, 2013), p. AF2H.5. 80J. G. Proakis, Digital Communications, 3rd ed. (McGraw-Hill, Boston,
49L. C. Sinclair, I. Coddington, W. C. Swann, G. B. Rieker, A. Hati, K. Massachusetts, 1995).
Iwakuni, and N. R. Newbury, Opt. Express 22, 6996 (2014). 81C. W. Chou, D. B. Hume, J. C. J. Koelemeij, D. J. Wineland, and T.
50S. A. Diddams, T. Udem, J. C. Bergquist, E. A. Curtis, R. E. Drullinger, Rosenband, Phys. Rev. Lett. 104, 070802 (2010).
L. Hollberg, W. M. Itano, W. D. Lee, C. W. Oates, K. R. Vogel, and D. J. 82N. Hinkley, J. A. Sherman, N. B. Phillips, M. Schioppo, N. D. Lemke, K.

Wineland, Science 293, 825 (2001). Beloy, M. Pizzocaro, C. W. Oates, and A. D. Ludlow, Science 341, 1215
51M. Hirano, T. Nakanishi, T. Okuno, and M. Onishi, IEEE J. Sel. Top. (2013).
Quantum Electron. 15, 103 (2009). 83T. Rosenband, D. B. Hume, P. O. Schmidt, C. W. Chou, A. Brusch, L.
52H. Byun, D. Pudo, J. Chen, E. P. Ippen, and F. X. Kärtner, Opt. Lett. 33, Lorini, W. H. Oskay, R. E. Drullinger, T. M. Fortier, J. E. Stalnaker, S. A.
2221 (2008). Diddams, W. C. Swann, N. R. Newbury, W. M. Itano, D. J. Wineland, and
53I. Coddington, W. C. Swann, and N. R. Newbury, Phys. Rev. A 82, 043817 J. C. Bergquist, Science 319, 1808 (2008).
(2010). 84T. R. Schibli, K. Minoshima, F.-L. Hong, H. Inaba, Y. Bitou, A. Onae, and
54F. Quinlan, T. M. Fortier, M. S. Kirchner, J. A. Taylor, M. J. Thorpe, N. H. Matsumoto, Opt. Lett. 30, 2323 (2005).
Lemke, A. D. Ludlow, Y. Jiang, and S. A. Diddams, Opt. Lett. 36, 3260 85E. Baumann, J.-D. Deschênes, F. R. Giorgetta, W. C. Swann, I. Coddington,

(2011). and N. R. Newbury, Opt. Lett. 39, 4776 (2014).

55H. Byun, M. Y. Sander, A. Motamedi, H. Shen, G. S. Petrich, L. A. 86B. Soller, D. Gifford, M. Wolfe, and M. Froggatt, Opt. Express 13, 666

Kolodziejski, E. P. Ippen, and F. X. Kärtner, Appl. Opt. 49, 5577 (2010). (2005).