Optical Fiber Soliton Lasers

G.E. Town1 , N.N. Akhmediev2 , and J.M. Soto-Crespo3

1
Department of Electronics, Macquarie University, NSW 2109, Australia
2
Optical Sciences Centre, Research School of Physical Sciences and Engineering,
Institute of Advanced Studies, Australian National University, ACT 0200,
Australia.
3
Instituto de Optica, Consejo Superior de Investigaciones Scientific, Serrano 121,
28006 Madrid, Spain.

Abstract. Experimental and theoretical developments in optical fiber soliton lasers

are reviewed. We first review the fundamentals of optical fiber soliton lasers from
an experimental viewpoint, including soliton generation and control mechanisms, and
highlight the similarities between developments in soliton fiber lasers and soliton trans-
mission systems. We then review the mathematical theory of soliton lasers based on a
master equation description, highlighting some interesting solutions and recent results
concerning soliton stability.

1 Introduction
Optical fiber lasers are ideal systems in which to generate and observe solitons.
Whilst the optical pulses generated in what are widely called soliton lasers are
rarely solitons in the narrow mathematical sense (i.e. in which they may collide
and pass through each other with no change other than a phase shift), they nev-
ertheless can demonstrate most of the properties of Hamiltonian solitons, such
as energy quantization, phase-sensitive interactions, robustness to perturbations,
etc. Consequently in this tutorial review “optical fiber soliton laser” is defined
as any nonlinear optical oscillator containing optical fibers, and which generates
optical pulses due to a balance between various sources of linear and nonlinear
gain, loss, self phase modulation, and dispersion. The variety of such lasers is
very large, and so for the most part we shall further restrict our attention to
lasers in which the gain medium is an optical fiber amplifier.
The next section of this chapter contains a review of fundamental matters
relating to solitons and optical fiber lasers, including typical laser construc-
tion and component characteristics, the effects of nonlinearity and dispersion
on pulse propagation, etc. We also highlight parallels between the development
of optical fiber soliton lasers and soliton transmission systems. The following
section reviews experimental developments in optical fiber soliton lasers since
1984, demonstrating how the fundamental ideas are applied in practice. The
final section contains a review of theoretical methods and results of modelling
such lasers. At the time of writing it is almost 20 years since demonstration of
the first “soliton fiber laser”, however the field continues to reveal surprises in
both theory and experiment. New optical fibers and fiber components together
with ongoing discoveries of new types of soliton and soliton behavior are likely

K. Porsezian, V.C Kuriakose (Eds.): LNP 613, pp. 265–297, 2002.

c Springer-Verlag Berlin Heidelberg 2002


266 G.E. Town, N.N. Akhmediev, and J.M. Soto-Crespo

to see continued development in the field of optical fiber soliton lasers for some
time yet.

2 Fundamentals of Optical Fiber Soliton Lasers

A laser is an optical oscillator in which the presence of optical gain and optical
feedback result in the sustained generation of one or more wavelengths within
the laser cavity. In a linear cavity (i.e. in which the intensity of light does not
change the optical properties of the cavity), the wavelength of each oscillating
mode is set by the boundary conditions, and depends primarily on the optical
path length of the cavity. Optical feedback may be applied by using mirrors
(e.g. as in the Fabry-Perot cavity), or by making a recirculating loop (i.e. ring
cavity). Optical gain may be provided using one or more of the different types
of optical amplifier available. A single lasing mode produces a continuous wave
(CW) single frequency output, however in practice several wavelengths may be
present in the laser output. To obtain pulses, the phase of several lasing modes
must remain locked together (i.e. mode-locking). This is generally achieved by
placing a nonlinear and/or time-varying component in the cavity, e.g. with time
or intensity-dependent loss, or time-varying delay. Other components are also
commonly added to laser cavities for spectral control, coupling of the optical
pump and signal in and out of the laser cavity, etc.

FIBER
AMPLIFIER OPTICAL OPTICAL
ISOLATOR FILTER

WSC

PUMP LASER
INPUT MODULATOR OUTPUT

POLARIZATION
CONTROLLER
f0

Fig. 1. Typical configuration of an actively mode-locked fiber laser

2.1 Optical Fiber Laser Components and Construction

A typical optical fiber laser containing the basic elements mentioned above is
shown schematically in Fig. 1. Single mode optical fiber may be chosen to provide
the desired gain, dispersion, and Kerr nonlinearity. The wavelength selective
coupler (WSC) is required to couple the pump into the fiber amplifier without
coupling the signal out. The optical isolator is a nonreciprocal element which may
be used to force unidirectional operation in a ring laser. An optical filter may
 Optical Fiber Soliton Lasers 267

be used to help define or tune the lasing wavelength, reject noise, or modify the
cavity dispersion. An output coupler is required to out-couple a proportion of the
signal in the laser cavity for external use. A polarizatio controller is often required
to compensate for unwanted birefringence, e.g. due to bends in the fibers). The
mode-locking element, shown in Fig. 1 as an electro-optic amplitude modulator,
is required to initiate and sustain periodic pulse generation. For active mode-
locking the modulator must be driven at, or very close to, a harmonic of the
fundamental cavity frequency (i.e. the inverse of the round-trip time for light
within the cavity).

Optical Fiber Amplifiers

Amplification in optical fibers may be obtained by either stimulated scattering

in the silica comprising the fiber, or stimulated emission from rare-earth dopants
within the fiber. The main differences between amplifiers based on stimulated
scattering and rare-earth-doped amplifiers relate to the pump and gain wave-
lengths, and to the pump power and length of fiber required. In optical amplifiers
based on stimulated Raman scattering in silica fiber the gain is generated at
wavelengths approximately 100nm longer than the pump wavelength (or more
exactly, 13THz lower in frequency), and may be anywhere within the wide range
of wavelengths over which the optical fiber can be used. In rare-earth-doped fiber
amplifiers, optical gain is available only in fixed wavelength bands determined
by the rare-earth excited-state energy levels, and typically there are only one or
two fixed wavelength bands suitable for pumping. For example, erbium-doped
fiber amplifiers (EDFAs) can provide large and broadband gain (e.g. 30dB gain
over >40nm bandwidth) around 1.55µm in the anomalous dispersion regime
of standard step index fibers, and are conveniently pumped by InGaAs 980nm
semiconductor diode lasers. The bandwidth of Raman fiber amplifiers is simi-
lar to EDFA’s, but Raman amplifiers generally require higher pump power and
longer lengths of fiber for efficient amplification. An optical fiber Raman am-
plifier was used in one of the first all-fiber soliton lasers reported [1], however
rare-earth-doped fiber amplifiers are now more commonly used. For more de-
tails, the interested reader may refer to one of a number of texts on optical fiber
amplifiers and lasers [2–4].
The characteristics of fiber amplifiers have can have a significant impact on
the performance of optical fiber soliton lasers. For example, optical fiber ampli-
fiers usually have a relatively small pump absorption coefficient, and so several
meters of fiber (or even more in Raman amplifiers) are commonly required to
absorb most of the pump and produce large gain. Consequently optical fiber
lasers are relatively long (typically between 10cm and 100m), and the spacing
between modes is correspondingly small (typically between 1GHz and 1MHz).
To generate ultrashort optical pulses in such lasers at high repetition rates there-
fore requires mode-locking of many harmonically-related modes. Secondly, the
metastable level lifetime of rare-earth dopants in glass is relatively long (typi-
cally milliseconds), hence a large inversion is readily achieved, resulting in low
268 G.E. Town, N.N. Akhmediev, and J.M. Soto-Crespo

noise amplification with noise figure close to the quantum limit of 3dB. Further-
more, the energy stored in rare-earth doped fiber amplifiers is relatively large
(i.e. hundreds of microjoules), especially compared to the energy of solitons in
optical fibers. Consequently optical fiber amplifiers generally do not suffer gain
depletion from the passage of a single pulse, but rather the gain saturates in
response to the average signal power in the laser cavity, which adjusts itself so
that the round-trip gain and loss in the cavity are equal. Lastly, whilst relatively
broadband, the finite gain-bandwidth of optical fiber amplifiers usually limits the
minimum possible pulse duration that can be generated in soliton fiber lasers to
tens of femtoseconds.

Optical Feedback

Broadband optical feedback is readily achieved using either using an optical fiber
loop mirror [5] (see also Sect. 3.2), or by splicing the fibers into a loop, as in
Fig. 1. In ring cavities the isolator forces unidirectional lasing, thus avoiding
complications which can arise due to interactions between counter-propagating
waves within the gain medium and laser cavity.
Another convenient form of mirror for providing optical feedback is the Bragg
grating, i.e. an optical filter usually used in reflection, and formed by a periodic
variation in refractive index along the core of the fiber. Fiber Bragg gratings
are also commonly used for spectral and dispersion control in fiber lasers and
transmission systems. For further details on fiber Bragg gratings the interested
reader may refer to various review articles and texts on the subject, e.g. [6,7].

Additional Components

Because fiber lasers are relatively long they can be susceptible to environmental
perturbations, e.g. changes in temperature, stress, vibration, etc. Consequently
it is often necessary in practice to add more components than shown in Fig. 1
to stabilize the laser against such influences, e.g. to stabilize the pulse repetition
rate [8], or lock the cavity modes to a stable reference frequency [9].

2.2 Nonlinearity and Dispersion

The nonlinearity and dispersion in optical fibers often play a key role in deter-
mining the properties of optical fiber soliton lasers. “Nonlinearity” here refers
to the Kerr nonlinearity, i.e. the third order nonlinear susceptibility, on result of
which is an intensity-dependent refractive index, which causes spectral broaden-
ing and self-phase modulation (SPM) of light propagating in the fiber. “Disper-
sion” refers to group-velocity dispersion (GVD), associated with a wavelength-
dependent refractive index, the primary result of which is temporal broadening
of optical pulses. These effects are described in the following subsections, how-
ever for more detail the interested reader is referred to one of the excellent texts
available on nonlinear fiber optics, e.g. [10].
 Optical Fiber Soliton Lasers 269

Self-phase Modulation
The small guiding cross-section and low loss of optical fibers can result in signif-
icant nonlinear effects. One of the main nonlinear effects in soliton fiber lasers
is the intensity dependent refractive index,
n = n0 + n2 I (1)
where n2 is the nonlinear index coefficient (typically 3 × 10−20 m2 /W in sil-
ica fibers). The nonlinear index results in the accumulation of an intensity-
dependent phase during propagation, i.e. self-phase modulation, φN L = (2π/λ)n2
Iz. In the absence of dispersion the effect of the intensity-dependent refractive
index on optical pulse propagation may be calculated analytically:
u(z, t) = u(0, t) exp{iγ|u(0, t)|2 z}, (2)
in which u is the complex amplitude of the pulse envelope (in the slowly varying
envelope approximation), z is the distance of propagation, and γ = (n2 ω0 )/(cAeff )
is the nonlinearity coefficient, where ω0 is the optical carrier frequency, and Aeff
is the effective area of the core (typically γ ≈10 W−1 km−1 in single mode silica
fibers). The length scale on which nonlinearity becomes significant is LNL =
1/(γP0 ), in which P0 is the peak optical power. Equation (2) indicates that
the output does not change shape during propagation, but broadens spectrally,
developing a chirp (i.e. time-dependent optical carrier frequency) with instan-
taneous frequency offset from the carrier δω = −∂φNL (z, t)/∂t. For example,
Fig. 2 shows the spectral broadening and chirp development due to self-phase
modulation in a gaussian input pulse with RMS width T0 = 1ps.

Group Velocity Dispersion

In the absence of nonlinearity, the effect of GVD on optical pulse propagation
may be calculated analytically:
i
U (z, ω) = U (0, ω) exp{ β2 ω 2 z} (3)
2
in which U (z, ω) is the Fourier transform of the complex pulse envelope, and β2
is the group velocity dispersion coefficient (typically β2 ≈ −20ps2 /km at 1.55
µm in single mode silica fibers). The length scale on which dispersion may be
regarded as significant for an initially unchirped (i.e. transform limited) pulse
is LD = T0 /|β2 |. In this case the power spectral density does not change with
propagation, however the pulse broadens temporally, and develops a chirp. For
example, Fig. 3 shows the temporal broadening and chirp development due to
anomalous GVD on a gaussian input pulse with RMS width T0 = 1ps.

2.3 Typical Characteristics of Optical Fiber Soliton Lasers

The main performance characteristics of interest in soliton fiber lasers are; pulse
duration (typically ps or less), pulse bandwidth (typically nm), pulse energy
270 G.E. Town, N.N. Akhmediev, and J.M. Soto-Crespo

(a)

(b)

Fig. 2. The effect of self-phase modulation on (a) pulse spectral broadening, and (b)
chirp development
 Optical Fiber Soliton Lasers 271

(a)

(b)

Fig. 3. The effect of group velocity dispersion on (a) pulse temporal broadening, and
(b) chirp development
272 G.E. Town, N.N. Akhmediev, and J.M. Soto-Crespo

(typically nJ), repetition rate (typically MHz to many GHz), output power
(typically mW), laser stability and self-starting (depends on the laser config-
uration), and noise and jitter (potentially quantum-limited).
Similar performance tradeoffs apply in soliton lasers as in soliton transmission
systems. For example, to generate very short pulses with a given peak power
generally requires low average dispersion, however this reduces the energy per
pulse due to the energy quantization of solitons, which has also been observed
in lasers, [11]. Also, for a given pulse duration, the maximum repetition rate
achievable in soliton lasers can be limited by soliton-soliton interactions [12].
Furthermore, in optical fiber lasers and transmission systems the nonlinear
effects (which vary with pulse amplitude) and dispersion are often not distributed
smoothly throughout the cavity, but lumped into discrete sections, just as gain
is localized in the fiber amplifier. The rapid changes perturb the solitons propa-
gating in the cavity, and cause losses to dispersive (i.e. low intensity) waves. The
perturbations often have little effect on soliton generation, as provided the cav-
ity length is much shorter than the soliton period, z0 = πLD /2, solitons do not
change significantly as they propagate around the cavity, and mainly respond
to the average dispersion and nonlinearity in the system [13,14]. In the latter
case one may imagine that the chirps induced by the localized nonlinearity and
anomalous dispersion in Figs. 2(b) and 3(b) roughly cancel for specific values of
pulse duration and peak power. However, if the soliton period is less than sev-
eral times the cavity length (or the distance between amplifiers in a transmission
system) the dispersive waves can build up and interact with the solitons, causing
pulse jitter and instability [15–17].
Whilst the description of soliton dynamics outlined above explains the ef-
fects of lumped nonlinearity and dispersion, and how solitons might still form
in a laser where the components are lumped rather than distributed, it is not
the whole picture. For example, pulses may form in optical fiber lasers even
with normal average dispersion. The underlying principle is that there must be
a balance between the various sources of gain and loss, and between the causes
of temporal shortening and broadening, though all these elements may interact.
Consequently the dynamics of nonlinear fiber lasers, and the variety of solitons
which may be produced, are considerably richer than occurs in simple lossless
nonlinear propagation. The soliton control mechanisms mentioned in the follow-
ing section may be used to alleviate or overcome practical limitations, and to
improve the performance of soliton lasers, just as in transmission systems.

2.4 Soliton Generation Versus Soliton Transmission

In linear laser cavities, the boundary conditions restrict any stable temporal
output of the laser to periodic functions, usually regarded as a superposition of
phaselocked modes. This situation is shown schematically in Fig 4.
The latter view of pulse generation presents difficulties in the case of non-
linear soliton lasers. In the presence of nonlinearity the optical path length of
the cavity depends on the instantaneous intensity, and hence on the shape of
the pulse in time, or on the phase relationships between its Fourier components.
 Optical Fiber Soliton Lasers 273

T
t

1/T
f
Fig. 4. Periodic pulse generation

Consequently the temporal output of a pulsed fiber laser is not exactly a super-
position of the cavity’s linear modes; the modes depend on the pulse shape and
intensity, and vice-versa.
An alternative approach to understanding pulse generation in nonlinear laser
cavities is to regard each pulse as an eigenfunction of the nonlinear operator de-
scribing the cavity. In this sense the pulses are regarded as independent particles
which must self-replicate every round-trip for stable pulse generation to continue.
This view is also compatible with soliton transmission systems, particularly if
each section of the transmission link is identical. For example, if any one section
of the periodic transmission system shown schematically in Fig. 5 was looped
back on itself, it could form a laser cavity (albeit a relatively long one) like that
in Fig. 4.

Fig. 5. Periodic pulse transmission

The analogy between soliton lasers and soliton transmission systems may be
extended by noting the almost parallel development and application of soliton
control mechanisms in each. In practical soliton lasers and transmission systems
soliton control mechanisms are usually required to limit the effects of soliton
interactions [18], soliton self-frequency shift [19], noise and Gordon-Haus jitter
[20], and interactions with dispersive waves [21].
The main soliton transmission control mechanisms developed to date are
temporal regeneration (e.g. either by active modulation [22], or passively, us-
ing a fast saturable absorber [23]), spectral control (e.g. in fixed frequency [24]
and sliding-frequency [25] systems), and dispersion (or nonlinearity) manage-
ment (e.g. by alternating sections of fiber with high and low dispersion, but
with low net dispersion [26,27]). Similarities between soliton pulse formation
and control mechanisms in fiber lasers and transmission systems will be high-
lighted in subsequent sections. Table 1 summarizes the development of soliton
control mechanisms in fiber lasers and transmission systems.
274 G.E. Town, N.N. Akhmediev, and J.M. Soto-Crespo

Table 1. Development of soliton control methods in optical fiber lasers and transmis-
sion systems
Soliton Soliton Transmission Systems Soliton Fiber Lasers
Control
Temporal Active temporal regeneration [22] Actively mode-locked [28], [29]
(active)
Temporal Passive temporal regeneration [23] Passively mode-locked, with
(passive) fast or slow saturable absorber
[30,31], [32,33], [34,35]
Spectral Spectrally stabilized [24] Spectrally stabilized [36,37]
Sliding Sliding-filter/frequency [25] Sliding-frequency [38], [39]
Dispersion Dispersion-managed [26], [27] Stretched-pulse [40]

3 Experimental Developments
in Optical Fiber Soliton Lasers
Ultrashort pulse generation in optical fiber soliton lasers is typically achieved
using one of the following mode-locking techniques;
Active mode-locking: (slow) active modulation of the cavity loss or length,
together with short pulse formation due to the combined effects of nonlin-
earity and dispersion.
Passive mode-locking: fast passive modulation of the cavity loss (e.g. by a
fast saturable absorber).
Soliton mode-locking: slow passive modulation of the cavity loss (e.g. by a
slow saturable absorber), initiating short pulse formation due to the com-
bined effects of nonlinearity and dispersion.
Hybrid mode-locking: any combination of the above (i.e. aiming to combine
the advantages of each, and to avoid their respective disadvantages).
In the following sections we concentrate on short pulse lasers, i.e. in which
the effective cavity dispersion is anomalous. Nevertheless, it should be noted that
bright pulse generation is often possible even when the intracavity dispersion is
effectively normal, however the resulting pulses tend to be broad both spectrally
and temporally, i.e. chirped, with a relatively large time-bandwidth product.

3.1 Active Mode-Locking

By actively modulating the loss or delay in the laser cavity at a frequency close
to the mode spacing, the laser modes become injection locked to each other,
resulting in a periodic pulse output.
According to linear mode-locking theory, actively mode-locked lasers cannot
generate pulses as short as passively mode-locked lasers [41], however it was
discovered very early in the development of fiber lasers that in the presence of
nonlinearity and anomalous dispersion, spectral broadening and temporal short-
ening of the pulses occurred, resulting in soliton generation with much shorter
pulsewidth than otherwise possible [28,29].
 Optical Fiber Soliton Lasers 275

Actively mode-locked fiber lasers are attractive options for generating ul-
trashort optical pulses at very high repetition rates; environmentally stabi-
lized lasers generating picosecond pulses at >100GHz repetition rates have been
demonstrated, e.g. [42].

3.2 Passively Mode-Locked Soliton Fiber Lasers

Passively mode-locked optical fiber soliton lasers are very simple sources of ul-
trashort (typically sub-picosecond) optical pulses. To promote pulse generation
and shortening they use a fast saturable absorber, the function of which is to
reject low intensity light from the cavity, but to pass high intensities. The fast
saturable absorber function is readily implemented in optical fiber using one of
two methods; a nonlinear loop mirror [43], [44], or nonlinear polarization rota-
tion and a polarizer [45]. Both implementations may approximately be regarded
as linear interferometry of two pulses derived from the same source, one of which
undergoes SPM due to nonlinear propagation, whilst the other does not. It is
not difficult to show that such a process leads to pulse shortening, as shown in
Fig. 6, independent of the sign of the dispersion.

Input

+ nonlinear
no change phase (SPM)

Output

Fig. 6. Pulse shortening by linear interferometry of pulses following linear and nonlin-
ear propagation

If the shortened output pulse is fed back into the laser cavity from which it
originated, then the pulse shortening process continues until it is balanced by
temporal broadening mechanisms, such as bandwidth limiting and group velocity
dispersion. The first soliton laser relying on this pulse forming mechanism was
reported by Mollenauer and Stolen [46]. When used to promote pulse formation
in lasers, the technique is commonly called additive pulse mode-locking (APM)
276 G.E. Town, N.N. Akhmediev, and J.M. Soto-Crespo

[47]. Passively mode-locked fiber lasers employing APM are described in the next
two sections.

Figure-8 Fiber Laser

The figure-8 laser, which derives its name from its cavity geometry, shown
schematically in Fig. 7, is a passively mode-locked optical fiber laser which read-
ily produces subspicosecond optical pulses [30,31]. With reference to Fig. 7 it can
be regarded as a ring laser (to the left of the 50:50 coupler), with a nonlinear
amplifying loop mirror (NALM) [44], comprising the 50:50 coupler and amplifier
loop on the right, providing fast saturable gain. The operation of the NALM
is described in the next section, and causes additive-pulse mode-locking of the
laser.

Fig. 7. Figure-8 optical fiber soliton laser schematic

Nonlinear loop mirrors. The operation of the NALM is similar to its passive
predecessor, the nonlinear optical loop mirror (NOLM) [43]; both are antires-
onant nonlinear Sagnac interferometers, with intensity-dependent transmission.
The key to operation of devices such as the NOLM and NALM is an asymmetry
between the clockwise and anticlockwise waves propagating around the loop.
In the NOLM the asymmetry is typically provided by using a non-50:50 split
ratio in coupler. In the NALM the asymmetry is provided by placing the fiber
amplifier closer to one of the coupler arms than the other, as on the right-hand
side of in Fig. 7. In either case, light travelling clockwise around the loop has a
significantly different intensity to light travelling anticlockwise, and hence a dif-
ference in nonlinear phase is accumulated by the counterpropagating waves. On
recombining at the coupler, low intensity light (which accumulated no nonlinear
phase) is reflected back to the port from which it entered, whilst higher intensity
light is transmitted.
It can readily be shown that for CW inputs the NALM with 50:50 coupler
has a power-dependent transmission function, given by [44]
PT ≈ GPIN {1 − 0.5 [1 + cos (0.5γLPIN (1 − G))]} , (4)
 Optical Fiber Soliton Lasers 277

in which L is the nonlinear loop length, and G the amplifier power gain. The
nonlinear transmission is plotted in Fig. 8. The minimum transmission usually
occurs at low input power (i.e. zero nonlinear phase difference), however the loop-
mirror may be biased to provide any transmissivity at zero-input by adjusting
the loop birefringence [48].

Fig. 8. Plot of the transmission of nonlinear optical fiber amplifying loop mirror versus
input power, as described by Eq. 4

Apart from promoting short pulse generation in optical fiber lasers, non-
linear loop mirrors may be used for ultrafast all-optical switching [49–51], and
have been suggested for use as temporal pulse regenerators in soliton trans-
mission systems [23]. Interestingly, solitons can be switched almost completely
through nonlinear loop mirrors in the same way as CW [43,50]. It can be shown
that maximum energy transmission (i.e. >90%) for fundamental solitons occurs
when the peak difference in nonlinear loop phase is ∆φNL = 0.6π. The excellent
switching properties of solitons arise from their uniform soliton phase [52].

Laser performance and practical considerations. The typical output from

a figure-8 fiber laser in both the time and wavelength domains is shown in Figs.
9(a) and 9(b), respectively.
The results illustrate a number of practical difficulties in passively mode-
locked soliton fiber lasers. Firstly, because the pulse peak power and duration
are largely determined by the NALM peak transmission power and cavity dis-
persion, the pulse energy is quantized [11]. Consequently the number of pulses
in a passively mode-locked laser cavity varies in proportion to amplifier pump
power above threshold. Also, because the pulses circulating in the laser cavity
build up from noise, the pulse repetition rate is usually random, with pulses
often clumping together in bursts as shown in Fig. 9(a). The latter problem may
be alleviated by coupling the laser to a sub-cavity or resonator [8], thus seeding
pulse formation at a fixed delay or repetition rate.
278 G.E. Town, N.N. Akhmediev, and J.M. Soto-Crespo

(a)

(b)

Fig. 9. Typical output from a figure-8 optical fiber soliton laser (a) in the time domain
(showing pulse energy quantization, random repetition rate, and bunching), and (b)
in the frequency domain (showing characteristic of chirped solitons, and phasematched
sidebands)

For similar reasons as above, the intracavity pulse power and duration are
unlikely to correspond to those of fundamental solitons for the fiber used in the
laser cavity, hence the output pulses are likely to be slightly chirped. The latter is
apparent in the rounded spectrum shown in Fig. 9(b), which would appear more
triangular in shape for a transform-limited pulse with sech2 intensity. Of course,
the output pulse chirp, peak power, pulse width, and energy may be modified
 Optical Fiber Soliton Lasers 279

outside the laser cavity using a variety of techniques involving amplification, and
nonlinear and/or dispersive propagation [53].
Furthermore, depending on the NALM characteristics, the figure-8 laser usu-
ally has a relatively high self-starting threshold, and the cavity must be per-
turbed in some way to initiate pulse generation. The latter problem is best
solved by using hybrid mode-locking techniques, for example by incorporating a
slow saturable absorber into the cavity (see also Sects. 3.3 and 3.4).
Another potential problem is that the laser cavity incorporating the NALM
is usually relatively long (i.e. >10m), which can cause a number of difficulties
in practice. The first is that the cavity can be susceptible to slow environmental
changes. For example, changes in temperature may cause the cavity length to
vary by distances comparable to the pulse length, or may cause the net birefrin-
gence in the cavity to change and hence change the NALM switching character-
istic and the laser operating point. Consequently active control over the cavity
length and birefringence is desirable for long-term stability.
Furthermore, if a high repetition rate output is required then many pulses
must be circulating within the cavity at a time. Under these circumstances the
effects of noise, perturbations, and soliton interactions can cause the pulses in
the cavity to move relative to each other and to interact. The introduction of
a passband filter into the laser cavity has been observed to reject noise and
stabilize the lasing wavelength [36,37], i.e. by reducing noise and Gordon-Haus
jitter, preventing soliton frequency shifts, and damping soliton interactions.
Lastly, sudden changes in amplitude and dispersion around the cavity cause
the solitons to shed low intensity (i.e. dispersive) radiation [21], which may prop-
agate and interact with the solitons in the cavity [15,16]. The dispersive radiation
is phasematched to the solitons at specific wavelengths [17], and appears as side-
bands in the laser output, as in Fig. 9(b). Resonances between the soliton period
and the periodic perturbations associated with a finite cavity length ultimately
limit the minimum pulse duration achievable in any given soliton laser.

Nonlinear Polarization Rotation Fiber Laser

Nonlinear polarization rotation is the tendency of the polarization ellipse (i.e. de-
scribing the polarization state of light) to rotate when propagating in a birefrin-
gent nonlinear medium [54,55]. The amount of rotation depends on the instanta-
neous intensity of light, and results from a combination of self-phase modulation
and cross-phase modulation of the two orthogonal polarization components. A
polarizer placed after the fiber can be adjusted to reject low intensity components
of the signal, whilst passing high intensities; the resulting intensity-dependent
transmissivity is similar to the NOLM and NALM, and provides a fast saturable
absorber action which can be used to passively mode-lock fiber lasers [32,33].
Some fiber lasers with outstanding properties have been realized using this
mode-locking technique, e.g. the shortest (38fs) pulse generated in a fiber laser
[56], and high energy (0.5nJ) pulse generation in a dispersion-managed laser [57].
The latter laser is discussed further in Sect. 3.5.
280 G.E. Town, N.N. Akhmediev, and J.M. Soto-Crespo

3.3 Soliton Mode-Locking

Here we define soliton mode-locking as any pulse generation method in which

soliton formation occurs through a combination of nonlinearity and asymmetric
confinement in a temporal or spectral window that may be much wider than
that of the pulses being generated.

Mode-Locking with a Slow Saturable Absorber

It was previously shown that short pulse generation was possible in actively
mode-locked lasers, despite the weak temporal confinement (i.e. slow loss modu-
lation), due to pulse shortening and soliton formation through a balance between
nonlinearity and anomalous dispersion. A similar mechanism can also operate
in passively mode-locked lasers, in which the weak temporal confinement is pro-
vided by a slow saturable absorber [35].
The situation is shown schematically in Fig. 10. On arriving at the slow
saturable absorber, the energy in the pulse causes a rapid increase in cavity
loss. The recovery of the loss is relatively slow (typically picoseconds), however
in soliton lasers the recovery time does not directly affect the pulse duration,
which is rather determined instead by a balance between anomalous dispersion
and nonlinearity in the cavity [58].

Fig. 10. Pulse shaping due to loss dynamics in soliton fiber lasers; (a) fast saturable
absorber, (b) slow saturable absorber and soliton formation

A variety of semiconductor slow saturable absorber devices have been devel-

oped [59], and applied to mode-lock various types of laser, including fiber lasers
[34,60–62]. Careful laser design is required to prevent Q-switching instabilities
[59,63], particularly in rare-earth doped fiber lasers with long excited-state life-
times. Some novel soliton states have been observed in these types of lasers
[64,65].
 Optical Fiber Soliton Lasers 281

Sliding-Frequency Lasers

The schematic of a sliding-frequency soliton laser is shown in Fig. 11. The prin-
ciple of operation has much in common with sliding-filter soliton transmission
[25,66], the only difference being that the wavelength of the light is shifted every
round-trip (e.g. using a CW-driven acousto-optic modulator) rather than chang-
ing the wavelength of the filter. In both cases the pulse spectrum is slightly offset
from the filter’s central wavelength, as shown schematically in Fig. 12. The laser
has a relatively low self-starting threshold as CW lasing and noise are suppressed
by the frequency shifter and filter; only solitons are able to propagate contin-
uously in the cavity, with the asymmetric spectral loss each round-trip being
balanced by the generation of new frequencies through self-phase modulation
[38,39,67,68].

Fig. 11. Schematic of sliding-frequency soliton fiber laser cavity

Fig. 12. Soliton spectral confinement in the sliding-frequency laser, under the compet-
ing effects of the frequency-shifter and self-phase modulation
282 G.E. Town, N.N. Akhmediev, and J.M. Soto-Crespo

A characteristic of the sliding-frequency laser is that the spectral width of

the pulses generated is much narrower than the bandwidth of the filter confin-
ing them, and consequently it can be difficult to generate pulses much shorter
than 10ps. It can be shown by perturbation analysis that the full-width at half-
maximum duration of solitons generated in the sliding-frequency laser is given
by
1.44
∆τ =
, (5)
3
π · 10 · fs · ∆ν 2
−6

where ∆τ is the pulse width in picoseconds, fs is the frequency shift per round-
trip in megahertz, and ∆ν is the bandpass filter bandwidth in teraradians per
second. Surprisingly, both theoretical and experimental results show that in the
normal dispersion regime the pulse duration is independent of the amount of
cavity dispersion [69].
Apart from the low self-starting threshold, other useful properties of the laser
(and of sliding-frequency transmission systems) which result from the passband
filter and frequency-sliding are; damping of pulse interactions, limited noise and
Gordon-Haus jitter, and suppression of dispersive wave radiation and soliton
resonances.

3.4 Hybrid Mode-Locking

Hybrid mode-locking can be used to combine the best properties of the various
mode-locking techniques mentioned in the previous sections. For example, active
mode-locking techniques may be combined with intra-cavity fast saturable gain
or loss to promote ultrashort pulse generation at a stable repetition rate [70].
Similarly, ultrashort pulse generation in a passively mode-locked laser may be
combined with the low threshold of sliding-frequency soliton mode-locking [71].

3.5 Other Soliton Lasers

In this section we review two interesting soliton fiber lasers which use mode-
locking techniques discussed in previous sections, but which have significantly
different behaviors. Both have analogues in soliton transmission systems.

Stretched Pulse Fiber Laser

The laser shown schematically in Fig. 13 contains two equal and opposite sections
of dispersive fiber, such that the total cavity dispersion is close to zero [40]. The
laser is passively mode-locked by a fast saturable absorber based on nonlinear
polarization rotation.
A fiber laser with close to zero dispersion would normally generate very
short pulses, but with very low energy in each pulse, however in the dispersion-
balanced laser the pulse energy can be significantly larger. The reason is that
soliton “breathes” as it propagates around the cavity, undergoing large changes
in temporal width, returning to its minimum temporal width (and maximum
 Optical Fiber Soliton Lasers 283

peak power) at the midpoint of each dispersive section. Because the pulse has a
high peak power at only two points in the cavity, the average nonlinearity in the
cavity is reduced, and the pulse energy is increased accordingly. The same prin-
ciple is used in dispersion-managed soliton transmission systems to increase the
transmitted pulse energy in links with low net dispersion [72]. Other advantages
of the stretched-pulse laser in common with dispersion-managed transmission
systems are the potential for reduced pulse interactions, and reduced noise and
jitter [73–75].

Multiwavelength Sliding-Frequency Laser

If it possible to realize a WDM soliton transmission system, then by the analogy
used in Sect. 2.4 it should be possible to realize a multiwavelength soliton laser.
Such a laser has two particular requirements; a) unless the average dispersion
of the cavity is zero, only frequency-domain soliton control methods can used
for pulse generation and stabilization, and b) gain cross-saturation between the
different lasing wavelengths must be avoided.
Soliton generation at up to nine wavelengths has been observed in a sliding-
frequency soliton laser [76,77]. Multiwavelength lasing was achieved by incorpo-
rating a broadband Fabry-Perot-like Bragg grating resonator [78] in the cavity
to define the lasing wavelengths, and by inhomogeneous broadening the gain of
the erbium doped amplifier by cooling it in liquid nitrogen.
The typical output of the laser is shown in Fig. 13. Due to the non-zero cavity
dispersion, solitons at different wavelengths had different group velocities, and
hence the temporal output of the laser was essentially random, nevertheless
some beating between the different wavelength solitons was observable in the
autocorrelation.

4 Theory of Soliton Lasers

A common method for modelling mode-locked lasers which was pioneered by
Haus is the “master equation” approach [79], in which a differential equation
is derived to describe the average effect of the various components forming the
laser cavity on light propagating in the cavity. In its simplest form, this approach
assumes that any pulses generated in the laser change only slightly during each
round-trip of the cavity, which can therefore be modelled as a distributed system.
In this review we restrict our attention to passively mode-locked lasers, which
have often been described by the complex Ginzburg-Landau equation (CGLE)
[80,81], or more accurately by the quintic CGLE. Quintic terms in the equation
are essential for stable pulse solutions [82]. In the optical context the quintic
CGLE has the following form [83]:
D
iψz + ψtt + |ψ|2 ψ + ν|ψ|4 ψ = iδψ + i!|ψ|2 ψ + iβψtt + iµ|ψ|4 ψ, (6)
2
where z is the cavity round-trip number, t is the retarded time, ψ is the nor-
malized envelope of the field, D is the group velocity dispersion coefficient with
284 G.E. Town, N.N. Akhmediev, and J.M. Soto-Crespo

(a)

(b)

Fig. 13. Multiwavelength sliding-frequency soliton laser (a) output spectrum, and (b)
autocorrelation of the output showing beating between pulses at different wavelengths

D = ±1 depending on whether the group velocity dispersion (GVD) is anoma-

lous or normal, respectively, δ is the linear gain-loss coefficient, iβψtt accounts for
spectral filtering or linear parabolic gain (β > 0), !|ψ|2 ψ represents the nonlinear
gain (which arises, e.g., from saturable absorption), the term with µ represents,
if negative, the saturation of the nonlinear gain, the one with ν corresponds, also
if negative, to the saturation of the nonlinear refractive index.
The above continuous model takes into account the major physical effects
occurring in a laser cavity such as dispersion, self-phase modulation, spectral
filtering, and gain or loss (both linear and nonlinear). A delicate balance between
them gives rise to the majority of the effects observed experimentally.
 Optical Fiber Soliton Lasers 285

In the case of slow saturable absorbers, we have to take into account the de-
layed response of the absorber. Again, the laser is modelled as a distributed sys-
tem [84] if the pulse shape changes only slightly during each round-trip. The pulse
evolution is then governed by a modified CGLE with non-linear non-conservative
terms [79,35]:

D
iψz + ψtt + |ψ|2 ψ = i[g(Q) − δs (|ψ|2 )]ψ + iβψtt , (7)
2
where notations are the same as above. In addition, g(Q) is the cavity gain
∞
(Q = |ψ|2 dt), and δs (|ψ|2 ) represents the losses in the cavity and in the slow
−∞
saturable absorber.
The gain term g(Q) in Eq. (7) describes a typical solid-state laser gain
medium with a recovery time much longer than the round-trip time of the cav-
ity. For our purposes, the recovery time can be considered infinite. Therefore
g(Q), which accounts for gain depletion, depends on the total pulse energy in
the following way:
g0
g(Q) = . (8)
1 + EQL
Here EL is the saturation energy and g0 is the small signal gain. The value of
g(Q) decreases as the energy increases, so that only a limited number of pulses
can exist inside the cavity.
The loss modulation in the saturable absorber can be described by the fol-
lowing rate equation [35]:

∂δs δs − δ0 |ψ|2
=− − δs , (9)
∂t T1 EA
where T1 is the recovery time of the saturable absorber, δ0 is the loss introduced
by the absorber in the absence of pulses, and EA is the saturation energy of the
absorber.
The presence of birefringent elements in the cavity adds new effects. Phase-
locking of the two soliton components in the passive birefringent medium has
been predicted theoretically [85,86]. Polarization locking effects have been ob-
served experimentally in a laser with a slow saturable absorber [87].
In the latter case the pulse evolution in the laser is governed by a set of
two modified nonlinear Schrödinger equations (NLSE) with non-linear and non-
conservative terms [64]:

D
iφz + γφ + φtt + |φ|2 φ + A|ψ|2 φ + Bψ 2 φ∗ = i[g(Q1 ) − δs (|φ|2 )]φ + iβφtt ,
2
D
iψz − γψ + ψtt + |ψ|2 ψ + A|φ|2 ψ + Bφ2 ψ ∗ = i[g(Q2 ) − δs (|ψ|2 )]ψ + iβψtt , (10)
2
where ψ and φ are the normalized envelopes of the two components of the op-
tical field, γ is the half-difference between the phase velocities of the two field
286 G.E. Town, N.N. Akhmediev, and J.M. Soto-Crespo

components, D is the group velocity dispersion coefficient, A is the cross-phase

modulation coefficient, B is the coefficient of the energy-exchange term (four-
wave-mixing), β represents spectral filtering (β > 0), g(Qi ) is the gain in the
∞
cavity which depends on the energy, Qi = (|φ, ψ|2 ) dt, in each component of
−∞
the pulse in one round-trip, and δs (|ψ|2 ) is the total loss, including loss in the
semiconductor saturable absorber.
In situations where the discrete or lumped nature of the laser cavity cannot
be ignored, the above CGL equations can be modelled such that the parameters
D, δ, β, !, µ, and ν, vary periodically with z, the period corresponding to a cavity
round-trip. The goal of this is to capture the essential effects of periodicity and
abrupt changes of parameters inside the cavity. This model is similar to that used
for dispersion-managed optical transmission lines [88–94]. An example of the map
of the parameters is shown in Fig. 14 [95]. Dispersion, nonlinearity, and linear and
nonlinear gain act for a certain propagation length, L1 (modelling the evolution
in the gain medium). The output coupler is modelled by concentrating losses at
the end of each round-trip. For the bulk of the cavity, the pulse propagates a
distance L2 under the sole effect of the dispersion term, D2 . This model which
have distributed parameters. It gives soliton-like solutions which change shape
as they propagate within the cavity, and can be called “dispersion-managed
solitons”.

D = D2
Coupler
L2 Losses
δ2=ε 2=ν2 = 0
D = D1 D = D1
µ2= β2= 0
L1 L1
δ,ε,ν,µ,β δ,ε,ν,µ,β
0 0.5 1
roundtrip
Fig. 14. The map of the parameters of the CGLE for a single round-trip in the laser
as used in some numerical simulations

In this section the basic equations for modelling the operation of passively
mode-locked lasers have been presented. There is a wide variety of such laser
systems, and hence the exact model may also vary from one system to another.
Nevertheless, the knowledge of these models and their solutions helps to under-
stand the processes of pulse generation in laser systems in general.
 Optical Fiber Soliton Lasers 287

4.1 Exact Solutions

The CGLE Eq. (6) has no known conserved quantities. This contrasts with the
infinite number for the NLSE and the small number for Hamiltonian systems.
For an arbitrary initial condition, even the energy is not conserved, as there is
energy input exchange with from the external pump. None of the equations (6),
(7) and (10) is integrable, and only particular exact solutions can be obtained.
In general, initial value problems with arbitrary initial conditions can only be
solved numerically. The cubic CGLE, obtained by setting µ = ν = 0 in Eq. (6),
has been studied extensively [96–98]. Exact solutions to this equation can be
obtained using a special ansatz [96], the Hirota bilinear method [97], or reduction
to systems of linear PDEs [99]. However, it was realized many years ago that
the soliton-like solutions of this equation are unstable to perturbations.
We give here the stationary soliton solution of Eq. (6) with zero transverse
velocity. This occurs when β = 0:

ψ(τ, z) = A(τ ) exp(−iωz), (11)

where ω is a real constant. The complex function A(τ ) can be written as A(τ ) =
a(τ ) exp[iφ(τ )] where a and φ are real functions of τ with

φ(τ ) = φ0 + d ln[a(τ )], (12)

where d is the chirp parameter and φ0 is an arbitrary phase. We suppose φ0 = 0

for simplicity. For the cubic case, this ansatz covers all pulse-like solutions. In
the quintic case, however, Eq. (12) is a restriction imposed on φ(τ ), because the
chirp could have a more general functional dependence on τ .
For the cubic CGLE, that is Eq. (6) with ν = µ = 0:
 
3d(1 + 4β 2 ) δ
a(τ ) = BC sech(Bτ ), C = , B= , (13)
2(2β − !) d − β + βd2

and ω and d are given by

δ(1 − d2 + 4βd)
ω= − , (14)
2(d − β + βd2 )


3(1 + 2!β) ± 9(1 + 2!β)2 + 8(! − 2β)2
d = d± = (15)
2(! − 2β)
where the minus sign is chosen in front of the square root.
For the quintic case, the above formulae also give a soliton solution. However,
the solution is only valid for certain relations between the parameters of the
equation [83,100]. In general, we need to use some numerical technique to find
stationary solutions. One way to do this is to reduce Eqs. (6) to a set of ordinary
differential equations (ODEs). We achieve this by seeking solutions in the form:

ψ(t, z) = ψo (τ )exp(−iωz) = a(τ ) exp[iφ(τ ) − iωz], (16)

288 G.E. Town, N.N. Akhmediev, and J.M. Soto-Crespo

where a and φ are real functions of τ = t − vz, v is the pulse velocity and ω is the
nonlinear shift of the propagation constant. Substituting Eq. (16) into Eq. (6),
we obtain an equation for two coupled functions, a and φ. We separate real and
imaginary parts, obtain a set of two ODEs, and transform them.

2.5 2.5
5

Field A mplitu de
Field A mplitu de

2 ε=1 . 141 , 1 . 080 2 ε=1 . 09

1.5 1.5

1 2 4 1 1 3

0.5 0.5

0 0
0 2 4 6 0 2 4 6
τ τ
Fig. 15. Amplitude profiles of the stationary solutions of the quintic CGLE. The pa-
rameter values of CGLE are shown in the figure. Three types of solutions exist for the
same
= 1.09.

The resultant set of equations contains all stationary and uniformly trans-
lating solutions. The parameters v and ω are the eigenvalues of this problem. In
the (M, a) plane, where M = a2 φ (τ ), the solutions corresponding to pulses are
closed loops starting and ending at the origin. The latter happens only at certain
values of v and ω. If v and ω differ from these fixed values, the trajectory cannot
comprise a closed loop. By properly adjusting the eigenvalue ω, it is possible to
find the soliton solution with a “shooting” method.
Numerical simulations show [101] that a multitude of soliton solutions of the
CGLE exist. They have a variety of shapes and stability properties. They can
even be partly stable and partly unstable. Trajectories on the phase portrait are
deflected from their smooth motion near the singular points. As a result, the
trajectory can have additional loops and the soliton shape can become multi-
peaked [101].
Examples of the amplitude profiles are shown in Fig. 15. Only half of the
profiles are plotted as they are even functions of τ . All soliton solutions (or
at least those we know about) are interconnected, i.e. continuously changing
parameters of CGLE we can transform one type of soliton into another.
Generation of stable pulses is possible in a very narrow range of the laser
parameters and requires their careful adjustment. More generally, the pulses
may change their shape from one round-trip to another and have complicated
dynamics in time, and can have periodic behavior on a time scale larger than the
round-trip time [102]. If there are many periods involved in the dynamics, then
the pulse shape evolution in time may become chaotic. The system can enter
into a chaotic regime in various ways including the classical one through period
doubling bifurcations. The most interesting phenomenon is that stationary stable
solitons can coexist with the chaotic regime of soliton propagation.
 Optical Fiber Soliton Lasers 289

Fig. 16. Two periods of the evolution of an exploding soliton. The parameters are

= 1.0, δ = −0.1, β = 0.125, µ = −0.1, and ν = −0.6. These solutions cannot be
found in analytic form. However, they are as common as stationary solutions and exist
for a wide range of parameters. The process never repeats itself exactly in successive
“periods”. However, it always returns to the same shape.

A special class of pulsating solitons can be called “exploding” or “erupting”

[102,103] solitons. “Exploding” soliton evolution, shown in Fig. 16, starts from
a stationary localized solution which has a perfect soliton shape. After a while,
its flanks become covered with small ripples (a form of a small scale instability)
which seem to move downwards along the two sides of the soliton, and very soon
the pulse is covered with this seemingly chaotic structure. When the ripples
increase in size, the soliton cracks into pieces, like a mountain after a strong
volcanic eruption or after an earthquake. This can also look like an explosion.
This completely chaotic, but well-localized, structure then is filled with “lava”
which restores the perfect soliton shape after a “cooling” process. The process
repeats forever, although the distance between “explosions” fluctuates, and in
each of them the pulse splits into different pieces. The exploding soliton has been
observed experimentally [95].

4.2 Soliton Solutions for the Case of Slow Saturable Absorber

We consider the limiting case when the pulse amplitude is well below the satura-
tion level. The gain coefficient g is constant if we deal with stationary solutions
of Eq. (7) when Q is constant. We also assume that the relaxation time is long
in comparison to the pulse width. In this case, T1 → ∞, and the slow part of
the loss variation is given by the approximate formula

   2  t
|ψ|
δs (t) = δ0 exp − dt ≈ δ0 − α |ψ|2 dt + . . . (17)
EA
−∞
290 G.E. Town, N.N. Akhmediev, and J.M. Soto-Crespo
δ0
where α = EA . Substituting this into Eq. (6), we obtain the equation

t
D
iψz + ψtt + |ψ|2 ψ = iδψ + iβψtt + i!|ψ|2 ψ + iαψ |ψ|2 dt , (18)
2
−∞

where δ = g − δ0 . This equation is the complex Ginzburg-Landau equation

(CGLE) [83], except that it includes a non-conservative nonlinear term on the
right-hand-side of Eq. (18), with the term being nonlocal in time.
Eq. (18) has exact pulse-like solutions [104,105] which move with velocity V :

ψ = A(t − V z) eid ln[A(t−V z)] iKt−iωz

e (19)

where
A(x) = γ C sech(γx),
2
3(D + 2!β) − 9 (D + 2!β) + 8(!D − 2β)2
d= ,
2(!D − 2β)


αC 2 ± α2 C 4 − 2(2β − dD − 2C 2 !)(βK 2 − δ)
γ = γ± = , (20)
2β − dD − 2C 2 !
  
3d(D2 + 4β 2 ) αdC 2 2β
C= , K=− , V =K D− ,
2(2β − !D) 2β(1 + d2 ) d
D 2
ω= (K + γ 2 ) + βdγ 2 − (γC)2
2
We note that the form of d ensures that C is real and positive, because d and
(2β − !D) have the same sign. The parameters (α, β, δ, !, D) must be chosen
to ensure that the quantity under the square root sign in γ is positive. All the
parameters of this solution, including the velocity V , are fixed and depend on
the parameters of the equation. However, there are two branches of the solution,
as specified by the two signs in Eq. (20). An example of the solution for certain
values of parameters δ, β and α is shown in Fig. 17. Different values of loss/gain
on either side of the soliton can cause it to move relative to the reference frame.
For γ = γ− the solution amplitude is close to zero when δ = 0. For γ = γ+ it
is not close to zero when δ = 0. Note that δ is equal to the amount of loss (or
gain) experienced by the left-hand side of the pulse.
The soliton exists for a certain range of parameters. The limits of existence
are defined by the inequality

α2 C 4 − 2(2β − dD − 2C 2 !)(βK 2 − δ) > 0. (21)

An important parameter for chirped pulses is the amplitude-width product C.

It does not depend on α or δ but depends weakly on β and D. The velocity V
of the soliton does not depend directly on δ but depends linearly on α because
K depends linearly on α. The velocity varies with β and D and can be positive,
 Optical Fiber Soliton Lasers 291

t
Fig. 17. Soliton profile (solid lines) and the loss curve δ(t) = δ +
|ψ|2 + α |ψ|2 dt
−∞
(dotted line) defined by the exact solution (19) for
= 0.1, δ = −0.05, α = 0.1,
β = 0.02, and D = +1. The soliton always clings to the gradient of the absorption
curve δ(t).

negative or zero. In the case of δ positive, the solution exists for both signs in
the expression for γ. Hence, we have simultaneously two solutions for the same
set of parameters. Moreover, the solution exists for both normal and anomalous
dispersion (negative and positive D). This is not surprising [106], because in
systems with gain and loss, the pulse is the result of a balance not only of the
dispersion and nonlinearity (which is impossible at negative D) but also of gain
and loss.

4.3 Perturbation Approximation

If the coefficients δ, β, !, µ and ν on the right-hand side are all small, then soliton-
like solutions of Eq. (6) can be studied by applying perturbative theory to the
soliton solutions of the NLSE. Let us consider the right-hand side of Eq. (6),
with D = +1, as a small perturbation, and write the solution as a soliton of the
NLSE, i.e.
η
ψ(τ, z) = exp[−iΩτ + i(η 2 − Ω 2 )z/2]. (22)
cosh[η(τ + Ωz)]
In the presence of the perturbation, the parameters of the soliton, i.e. the am-
plitude η and frequency (or velocity) Ω, change adiabatically. The equations for
them can be obtained from the balance equations for the energy and momentum
[106]. Then we have the equations for the evolution of η(z) and Ω(z):
dη 1 8 dΩ 4
= 2η δ − βΩ 2 + (2! − β)η 2 + µη 4 , = − βΩ η 2 . (23)
dz 3 15 dz 3
The dynamical system of equations, (23), has two real dependent variables
and the solutions can be presented on the plane. An example is given in Fig. 18. It
292 G.E. Town, N.N. Akhmediev, and J.M. Soto-Crespo

has a line of singular points at η = 0, and, depending on the equation parameters,

may have one or two singular points on the semi-axis Ω = 0, η > 0. The values
of η 2 for singular points are defined by finding the roots of the biquadratic
polynomial in the square brackets in Eq. (23). When the roots are negative
(hence, η is imaginary), there are no singular points and hence no soliton solution.
If both roots of the quadratic polynomial (in η 2 ) are positive (so that both η are
real), then there are two fixed points and two corresponding soliton solutions.
Both roots are positive when either β < 2!, µ < 0 and δ < 0 or β > 2!, µ > 0
and δ > 0.

Fig. 18. The phase portrait of the dynamical system (23) for δ = −0.03, β = 0.1,

= 0.2 and µ = −0.11. The upper fixed point is a sink which defines the parameters
of a stable approximate soliton-like solution of the quintic CGLE. Any soliton-like
initial condition in close proximity to a fixed point will converge to a stable stationary
solution. The points on the line η = 0 are stable when δ < 0 and β > 0. This condition
is needed for the background state ψ = 0 to be stable.

The stability of at least one these fixed points requires β > 0. Moreover,
the stability of the background requires δ < 0. In the latter case, we necessarily
have β < 2!, µ < 0 and the upper fixed point is a sink (as shown in Fig. 18)
which defines the parameters of a stable approximate soliton solution of the
quintic CGLE. The background ψ = 0 is also stable, so that the whole solution
(soliton plus background) is stable. Finally, when only one of the roots is positive,
there is a singular point in the upper half-plane and there is a corresponding
soliton solution. However either the background or the soliton itself is unstable,
so that the total solution is unstable. The term with ν in the CGLE does not
 Optical Fiber Soliton Lasers 293

influence the location of the sink. It only introduces an additional phase term,
exp(8iνη 4 z/15), into the solution of Eq. (22).
In cubic CGLE, µ = 0 and ν = 0. The stationary point is then


η = 3δ/(β − 2!), Ω = 0. (24)
It is stable provided that δ > 0, β > 0 and ! < β/2. Clearly, in this perturbative
case the soliton and the background cannot be stable simultaneously. Hence,
this approach shows that to have both the soliton and the background stable,
we need to have quintic terms in the CGLE [82].
This simple approach shows that, in general, the CGLE has stationary soliton-
like solutions, and that for the same set of equation parameters two may exist
simultaneously (one stable and one unstable). Moreover, this approach results in
soliton parameters that are fixed; they do not have free parameters as solitons
of the nonlinear Schrödinger equation.
Despite its simplicity and advantages in giving stability and other properties
of solitons, the perturbative analysis has some serious limitations. Firstly, it can
only be applied if the coefficients on the right-hand-side of Eq. (6) are small,
and this is not always the case in practice. Correspondingly, it describes the
convergence correctly only for initial conditions which are close to the stationary
solution. Secondly, the standard perturbative analysis cannot be applied to the
case D < 1, when the NLSE itself does not have bright soliton solutions, though
the CGLE Eq. (6) has stable soliton solutions for this case as well.

4.4 Stability
The vast majority of soliton solutions lack stability. In fact, in some regions of
the parameter space, all of them can be unstable. Solving the whole propagation
equation only allows us to obtain stable structures [107,108]. However unstable
solitons may play an essential role in the overall dynamics when the system
starts with an arbitrary initial condition. Therefore, it is important to know all
soliton solutions when moving in the parameter space from one point to another.
Both types of stationary solitons, stable and unstable, are important and deserve
careful study.
The existence of stable and unstable solitons for the CGLE raises an impor-
tant issue: when we change the parameters of the equation, is it possible that
there are connected regions of stable and unstable solitons? If the answer is pos-
itive, when does such a transition occur? In other words, what is the stability
criterion for solitons in dissipative systems? An attempt to answer the above
questions was reported recently [101]. Previous works have dealt with the stabil-
ity criterion for ground state solitons in Hamiltonian systems [109–111]. Some
approaches for higher-order solitons have also been developed [112,113]. How-
ever, solitons in dissipative systems are qualitatively different from solitons in
Hamiltonian systems. As a result, the stability criterion for Hamiltonian systems
cannot easily be generalized to the case of dissipative systems.
Recent works by Kapitula and Sanstede consider the stability of CGLE (dis-
sipative) solitons when they are perturbations of NLSE solitons [114]. This is
294 G.E. Town, N.N. Akhmediev, and J.M. Soto-Crespo

Fig. 19. Energy of the soliton Q versus
for the quintic CGLE solitons. Stable solitons
are shown by solid line and unstable solitons by dotted line. Two consecutive magnifi-
cations of the small parts of the curve enclosed in dashed rectangles are shown in the
insets. Further magnification could reveal more detailed structure, but the accuracy is
close to the limit of the numerical method. Local maxima and minima of
(i.e. the
local edges in region of soliton existence) are the points where stability changes.

important for optical transmission lines which are governed by the perturbed
NLSE. However, in lasers, for example, dissipative terms are strong and the
approach developed in [114] is not sufficiently general.
The main result in [101] is that the points where the stability changes abruptly
coincide with the turning points of the branches representing the different fami-
lies of solitons. These branches are represented by curves which show any soliton
parameter (usually its peak amplitude or propagation constant) versus one equa-
tion parameter (!). For example, the value of soliton energy, Q, is plotted versus
! for CGLE solitons in Fig. 19. The local maxima and minima of ! in the re-
gion of soliton existence are the points where the corresponding soliton solutions
change their stability.

References
1. J.D. Kafka, T. Baer: Opt. Lett. 12, 181 (1987)
2. P.W. France, (ed.): Optical fiber Lasers and Amplifiers (CRC Press, 1991)
3. E. Desurvire: Erbium-Doped Fiber Amplifiers: Principles and Applications (Wiley,
1994)
 Optical Fiber Soliton Lasers 295

4. J.F. Digonnet, (ed.): Rare-Earth-Doped Fiber Lasers and Amplifiers, 2nd edn.
(Marcel Dekker, 2001)
5. D.B. Mortimore: J. Lightwave Technol. 6, 1217 (1988)
6. R. Kashyap: Fiber Bragg Gratings (Academic Press, 1999)
7. A. Othonos: Fiber Bragg Gratings: Fundamentals and Applications in Telecommu-
nications and Sensing (Artech House, 1999)
8. G.T. Harvey, L.F. Mollenauer: Opt. Lett. 18, 107 (1993)
9. X. Shan, D. Cleland, A. Ellis: Electron. Lett. 28, 182 (1992)
10. G. Agrawal: Nonlinear Fiber Optics 3rd edn. (Academic Press, 1989)
11. A.B. Grudinin, D.J. Richardson, D.N. Payne: Electron. Lett. 28, 67 (1992)
12. M. Romagnoli, M. Midrio, P. Franco, F. Fontana: J. Opt. Soc. Am. B 12, 1732
(1995)
13. A. Hasegawa, Y. Kodama: Opt. Lett. 15, 1443 (1990)
14. K.J. Blow, N.J. Doran NJ: IEEE Photon. Technol. Lett. 3 369 (1991)
15. S.M.J. Kelly, K. Smith, K.J. Blow, N.J. Doran: Opt. Lett. 16, 1337 (1991)
16. N. Pandit, D.U. Noske, S.M.J. Kelly, J.R. Taylor: Electron. Lett. 28, 455 (1992)
17. N.J. Smith, K.J. Blow, I. Andonovic: J. Lightwave Technol. 10, 1329 (1992)
18. J.P. Gordon: Opt. Lett. 8, 596 (1983)
19. J.P. Gordon: Opt. Lett. 11, 662 (1986)
20. J.P. Gordon, H.A. Haus: Opt. Lett. 11, 665 (1986)
21. J.P. Gordon: J. Opt. Soc. Am. B 9, 91 (1992)
22. M. Nakazawa, H. Kubota, E. Yamada, K. Suzuki: Electron. Lett. 28, 1099 (1992)
23. E. Yamada, M. Nakazawa: IEEE J. Quantum Electron. 30, 1842 (1994)
24. A. Mecozzi, J.D. Moores, H.A. Haus, Y. Lai: Opt. Lett. 16, 1841 (1991)
25. L.F. Mollenauer, J.P. Gordon, S.G. Evangelides: Opt. Lett. 17, 1575 (1992)
26. C. Kurtzke: IEEE Photon. Technol. Lett. 5, 1250 (1993)
27. W. Forysiak, F.M. Knox, N.J. Doran: Opt. Lett. 19, 174 (1994)
28. J.D. Kafka, T. Baer, D.W. Hall: Opt. Lett. 14, 1269 (1989)
29. F.X. Kärtner, D. Kopf, U. Keller: J. Opt. Soc. Am. 12, 486 (1995)
30. D.J. Richardson, R.I. Laming, D.N. Payne, V. Matsas, M.W. Phillips: Electron.
Lett. 27, 542 (1991)
31. I.N. Duling: Electron. Lett. 27, 544 (1991)
32. M. Hofer, M.E. Fermann, F. Haberl, M.H. Ober, A.J. Schmidt: Opt. Lett. 16, 502
(1991)
33. V.J. Matsas, T.P. Newson, D.J. Richardson, D.N. Payne: Electron. Lett. 28, 1391
(1992)
34. E.A. De Souza, C.E. Soccolich, W. Pleibel, R.H. Stolen, J.R. Simpson, D.J. Di-
Giovanni: Electron. Lett. 29, 447 (1993)
35. F.X. Kärtner, U. Keller: Opt. Lett. 20, 16 (1995)
36. G. Town, M. Sceats, and S. Poole: (Proceedings, 17th Australian Conference on
Optical fiber Technology, Hobart, November, 1992) pp. 154
37. K. Tamura, C.R. Doerr, H.A. Haus, E.P Ippen: IEEE Photon. Technol. Lett. 6,
697 (1994)
38. H. Sabert, E. Brinkmeyer: Electron. Lett. 29, 2122 (1993)
39. F. Fontana, L. Bossalini, P. Franco, M. Midrio, M. Romagnoli, S. Wabnitz: Elec-
tron. Lett. 30, 321 (1994)
40. K. Tamura, E.P. Ippen, H.A. Haus, L.E. Nelson: Opt. Lett. 18, 1080 (1993)
41. A.E. Siegman: Lasers, (Oxford University Press, 1986)
42. K.S. Abedin, M. Hyodo, N. Onodera: Electron. Lett. 36, 1185 (2000)
43. N.J. Doran, D. Wood: Opt. Lett. 13, 56 (1988)
296 G.E. Town, N.N. Akhmediev, and J.M. Soto-Crespo

44. M.E. Fermann, F. Haberl, M. Hofer, H. Hochreiter: Opt. Lett. 15, 752 (1990)
45. R.H. Stolen, J. Botineau, A. Ashkin: Opt. Lett. 7, 512 (1982)
46. L.F. Mollenauer, R.H. Stolen: Opt. Lett. 9, 13 (1984)
47. E.P. Ippen, H.A. Haus, L.Y. Liu: J. Opt. Soc. Am. B 6, 1736 (1989)
48. N. Finlayson, B.K. Nayar, N.J. Doran: Opt. Lett. 17, 112 (1992)
49. N.J. Doran, D.S. Forrester, B.K. Nayar: Electron. Lett. 25, 267 (1989)
50. K.J. Blow, N.J. Doran, B.K. Nayar: Opt. Lett. 14, 754 (1989)
51. B.K. Nayar, K.J. Blow, N.J. Doran: Opt. Comput. & Process. 1, 81 (1991)
52. K.J. Blow, N.J. Doran, S.J.D Phoenix: Opt. Comm. 88, 137 (1992)
53. M.E. Fermann, A. Galvanauskas, D. Harter: Appl. Phys. Lett. 64, 1315 (1994)
54. H.G. Winful: Appl. Phys. Lett. 47, 213 (1985)
55. C.R. Menyuk: IEEE J. Quantum Electron. 25, 2674 (1989)
56. M. Hofer, M.H. Ober, F. Haberl, M.E. Fermann: IEEE J. Quantum Electron. 28,
720 (1992)
57. K. Tamura, C.R. Doerr, L.E. Nelson, H.A. Haus, E.P. Ippen: Opt. Lett. 19, 46
(1994)
58. F.X. Kärtner, I.D. Jung, U. Keller: IEEE J. Select. Topics Quantum Electron. 2,
540 (1996)
59. U. Keller, K.J. Weingarten, F.X. Kärtner, D. Kopf, B. Braun, I.D. Jung, R. Fluck,
C. Honninger, N. Matuschek, J. Aus der Au: IEEE J. Select. Topics Quantum.
Electron. 2, 435 (1996)
60. W.H. Loh, D. Atkinson, P.R. Morkel, M. Hopkinson, A. Rivers, A.J. Seeds, D.N.
Payne: IEEE Photon. Technol. Lett. 5, 35 (1993)
61. W.H. Loh, D. Atkinson, P.R. Morkel, R. Grey, A.J. Seeds, D.N. Payne: Electron.
Lett. 29, 808 (1993)
62. B.C. Barnett, L. Rahman, M.N. Islam, Y.C. Chen, P. Bhattacharya, W. Riha, K.V.
Reddy, A.T. Howe, K.A. Stair, H Iwamura, S.R. Friberg, T. Mukai: Opt. Lett. 20
471 (1995)
63. R. Paschotta, U. Keller: Appl. Phys. B B73,653 (2001)
64. N.N. Akhmediev, J.M. Soto-Crespo, S.T. Cundiff, B.C. Collings, W.H. Knox: Opt.
Lett. 23, 852 (1998)
65. M.J. Lederer, B. Luther-Davies, H.H. Tan, C. Jagadish, N.N. Akhmediev, J.M.
Soto-Crespo: J. Opt. Soc. Am. B 16, 895 (1999)
66. L.F. Mollenauer, E. Lichtman, M.J. Neubelt, G.T. Harvey: Electron. Lett. 29, 910
(1993)
67. H. Sabert, E. Brinkmeyer: J. Lightwave Technol. 12, 1360 (1994)
68. M. Romagnoli, S. Wabnitz, P. Franco, M. Midrio, F. Fontana, G.E. Town: J. Opt.
Soc. Am. B 12, 72 (1995)
69. M. Romagnoli, S. Wabnitz, P. Franco, M. Midrio, L. Bossalini, F. Fontana: J. Opt.
Soc. Am. B 12, 938 (1995)
70. T.F. Carruthers, I.N. Duling, M.L. Dennis: Electron. Lett. 30, 1051 (1994)
71. G. Town, J. Chow, M. Romagnoli: Electron. Lett. 31, 1452 (1995)
72. N.J. Smith, F.M. Knox, N.J. Doran, K.J. Blow, I. Bennion: Electron. Lett. 32, 54
(1996)
73. S. Namiki, H.A. Haus: IEEE J. Quantum Electron. 33, 649 (1997)
74. B. Bakhshi, P.A. Andrekson, X. Zhang: Opt. Fiber Technol. 4,293 (1998)
75. Waiyapot S, Matsumoto M. Opt. Commun. 188,167 (2001)
76. G. Town, J. Chow, K. Sugden, I. Bennion, M. Romagnoli: J. Electric. & Electron.
Eng. Aust. 15, 267 (1995)
77. G.E. Town, J. Chow, A. Robertson, and M. Romagnoli: Electro-Optics - Europe
(CLEO-Europe ’96), Postdeadline paper CPD2.10, Hamburg, September 1996
 Optical Fiber Soliton Lasers 297

78. G.E. Town, K. Sugden, J.A.R. Williams, I. Bennion, S.B. Poole: IEEE Photon.
Technol. Lett. 7, 78 (1995)
79. H. Haus: J. Appl. Phys. 46, 3049 (1975)
80. P.A. Belanger: J. Opt. Soc. Am. B 8, 2077 (1991)
81. A.I. Chernykh, S.K. Turitsyn: Opt. Lett. 20, 398 (1995)
82. J.D. Moores: Opt. Comm. 96, 65 (1993)
83. N.N. Akhmediev, A. Ankiewicz: Solitons: Nonlinear Pulses and Beams (Chapman
& Hall, 1997)
84. C.-J. Chen, P.K.A. Wai, C.R. Menyuk: Opt. Lett. 19, 198 (1994)
85. N.N Akhmediev, A.V. Buryak, J.M. Soto - Crespo: Opt. Comm. 112, 278 (1994)
86. N.N Akhmediev, A.V. Buryak, J.M. Soto-Crespo, D.R. Andersen: J. Opt. Soc. Am.
B 12, 434 (1995)
87. S.T. Cundiff, B.C. Collings, W.H. Knox: Opt. Express 1, (1997)
88. I. Gabitov, E.G. Shapiro, S.K. Turitsyn: Phys. Rev. E 55, 3624 (1997)
89. A. Hasegawa: Physica D 123, 267 (1998)
90. Y. Kodama, S. Kumar, A. Maruta: Opt. Lett. 22, 1689 (1997)
91. J.N. Kutz, P. Holms, S. Evangelides, J. Gordon: J. Opt. Soc. Am. B 15, 87 (1998)
92. J. Nijhof, N.J. Doran, W. Forysiak, F.M. Knox: Elect. Lett. 33, 1726 (1997)
93. M.J. Ablowitz, G. Biondini: Opt. Lett. 23, 1668 (1998)
94. J.P. Gordon, L.F. Mollenauer: Opt. Lett. 24, 223 (1999)
95. S.T. Cundiff, J.M. Soto-Crespo, N. Akhmediev: Phys. Rev. Lett. 88, 073903 (2002)
96. N.R. Pereira, L. Stenflo: Phys. Fluids 20, 1733 (1977)
97. K. Nozaki, N. Bekki: Phys. Soc. Japan. 53, 1581 (1984)
98. P.-A. Bélanger, L. Gagnon and C. Paré: Opt. Lett. 14, 943 (1989)
99. R. Conte, M. Musette: Physica D 69, 1 (1993)
100. N.N. Akhmediev, V.V. Afanasjev, J.M. Soto-Crespo: Phys. Rev. E 53, 1190 (1996)
101. J.M. Soto-Crespo, N. Akhmediev, G. Town: Opt. Comm. 199, 283 (2001)
102. N. Akhmediev, J.M. Soto-Crespo, G. Town: Phys. Rev. E 63, 056602 (2001)
103. J.M. Soto-Crespo, N. Akhmediev, A. Ankiewicz: Phys. Rev. Lett. 85, 2937 (2000)
104. N.N. Akhmediev, A. Ankiewicz, M.J. Lederer, B. Luther-Davies: Opt. Lett. 23,
280 (1998)
105. V.S. Grigoryan, T.C. Muradyan: J. Opt. Soc. Am. B 8, 1757 (1991)
106. N.N. Akhmediev: General Theory of Solitons. In: Soliton-driven Photonics, ed.
by A.D. Boardman A.P. Sukhorukov (Kluver, 2001) pp. 371
107. N. Akhmediev, J.M. Soto-Crespo: In: Proceedings of SPIE, 3666, 307 (1999)
108. V.V. Afanasjev, N. Akhmediev, J.M. Soto-Crespo: Phys. Rev. E 53, 1931 (1996)
109. A.A. Kolokolov: Zh. Prikl. Mekh. Tekh. Fiz. 3, 426 (1973)
110. C.K.R.T. Jones, J.V. Moloney: Phys. Lett. A 117, 175 (1986)
111. M. Grillakis, J. Shatah, W. Strauss: J. Funct. Analysis 74, 160 (1987)
112. A.A. Kolokolov, A.I. Sukov: J. Appl. Mech. Tech. Phys. 4, 519 (1975)
113. N.N. Akhmediev, A. Ankiewicz, H.T. Tran: J. Opt. Soc. Am. B 10, 230-236 (1993)
114. T. Kapitula, B. Sandstede: Physica D 124, 58 (1998); Physica D 116, 95 (1998)