PHYSICAL REVIEW A 72, 043816 共2005兲

Mechanism of multisoliton formation and soliton energy quantization

in passively mode-locked fiber lasers
D. Y. Tang, L. M. Zhao, B. Zhao, and A. Q. Liu
School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore
共Received 14 April 2005; published 21 October 2005兲

We report results of numerical simulations on multiple-soliton generation and soliton energy quantization in
a soliton fiber ring laser passively mode locked by using the nonlinear polarization rotation technique. We
found numerically that the formation of multiple solitons in the laser is caused by a peak-power-limiting effect
of the laser cavity. It is also the same effect that suppresses the soliton pulse collapse, an intrinsic feature of
solitons propagating in gain media, and makes the solitons stable in the laser. Furthermore, we show that the
soliton energy quantization observed in the lasers is a natural consequence of the gain competition between the
multiple solitons. Enlightened by the numerical result we speculate that multisoliton formation and soliton
energy quantization observed in other types of soliton fiber lasers could have a similar mechanism.

DOI: 10.1103/PhysRevA.72.043816 PACS number共s兲: 42.55.Wd, 42.81.Dp, 42.60.Fc, 42.65.Re

I. INTRODUCTION It was shown that when a pulse in a laser becomes so narrow

that due to the effective gain bandwidth limit, the gain could
Passively mode-locked fiber lasers as a simple and eco- no longer amplify the pulse but impose an extra loss on it,
nomic ultrashort-pulse source have been extensively investi- the pulse would split into two pulses with broader pulse
gated in the past decade 关1–9兴. By implementing the soliton width. Based on a similar mechanism and in the framework
pulse-shaping technique in the lasers it was demonstrated of a generalized complex Ginzburg-Landau equation that ex-
that optical pulses in the subpicosecond range could be rou- plicitly takes into account the effect of a bandpass filter in
tinely generated. Various passive mode-locking techniques, the cavity, Lederer et al. theoretically explained the multiple-
such as the nonlinear loop mirror method 关3,4兴, the nonlinear pulse operation of their laser 关12兴. However, we point out
polarization rotation 共NPR兲 technique 关5–7兴 and the semi- that this process of multipulse generation can be easily iden-
conductor saturable absorber method 关8,9兴, have been used tified experimentally. In the case that no bandpass filter is in
to mode-lock the lasers. Independent of the concrete mode- the cavity, a pulse splits into two pulses only when its pulse
locking techniques it was found that the soliton operation of width has become so narrow that it is limited by the gain
all the lasers exhibited a common feature, namely, under bandwidth, while in the case of fiber lasers no significant
strong pumping strength multiple-soliton pulses are always soliton pulse narrowing was ever observed before a new soli-
generated in the laser cavity, and in the steady state all the ton pulse was generated, which obviously demonstrated that
solitons have exactly the same pulse properties: the same the multiple-pulse generation in the soliton fiber lasers must
pulse energy and pulse width when they are far apart. The have a different mechanism. Agrawal has also numerically
latter property of the solitons was also called the “soliton shown multiple-pulse formation when a pulse propagates in
energy quantization effect” 关10兴. The multiple-soliton gen- a strongly pumped gain medium 关15兴. Nevertheless, it can be
eration and the soliton energy quantization effect limit the shown that multiple-pulse formation has in fact the same
generation of optical pulses with larger pulse energy and mechanism as that described by Kärtner et al. Recently,
narrower pulse width in the lasers. Therefore, in order to Grelu et al. have numerically simulated multiple-pulse op-
further improve the performance of the lasers it is essential to eration of fiber soliton lasers 关16,17兴. By using a propagation
have a clear understanding of the physical mechanism re- model and also taking into account the laser cavity effect,
sponsible for these effects. It was conjectured that the soliton they could quite well reproduce the multipulse states of the
energy quantization could be an intrinsic property of laser experimental observations. However, no analysis of the
solitons, as solitons formed in a laser are intrinsically dissi- physical mechanism of multipulse formation was given. In
pative solitons, where the requirement of soliton internal en- addition, in their simulations the multisoliton formation is
ergy balance ultimately determines the energy of a soliton only obtained for limited sets of parameters, which is not in
关11兴. However, this argument cannot explain the formation of agreement with the experimental observations. Very recently,
multiple solitons in a laser cavity. Actually multiple-pulse Komarov et al. have theoretically studied multiple-soliton
generation has also been observed in other types of soliton operation and pump hysteresis of soliton fiber lasers mode
lasers, e.g., Lederer et al. reported the multipulse operation locked by using the NPR technique 关18兴. In their model they
of a Ti:sapphire laser mode locked by an ion-implanted semi- have explicitly taken into account the nonlinear cavity effect
conductor saturable absorber mirror 关12兴, and Spielmann et so they can successfully explain the multisoliton formation
al. reported the breakup of single pulses into multiple pulses and pump hysteresis based on the nonlinear cavity feedback.
in a Kerr lens mode-locked Ti:sapphire laser 关13兴. Theoreti- However, as they ignored the linear birefringence of the fiber
cally, Kärtner et al. have proposed a mechanism of pulse and the associated linear cavity effects, their model might
splitting for multiple-pulse generation in soliton lasers 关14兴. still not accurately describe the real laser systems, e.g., in

1050-2947/2005/72共4兲/043816共9兲/$23.00 043816-1 ©2005 The American Physical Society

TANG et al. PHYSICAL REVIEW A 72, 043816 共2005兲

FIG. 1. A schematic of the soliton fiber laser. PI, Polarization- FIG. 2. A typical experimentally measured oscilloscope trace of
dependent isolator. PC, polarization controller. DSF, dispersion- the multiple-soliton operation of the laser.
shifted fiber. EDF, erbium-doped fiber. WDM, wavelength-division
used to enforce the unidirectional operation of the laser and
multiplexer.
also determine the polarization of the light at the position. A
10% output coupler was used to let out the light. The soliton
their model in order to obtain multiple-pulse operation, they pulse width of the laser was measured with a commercial
have to add phenomenologically a frequency-selective loss autocorrelator, and the average soliton output power was
term. Physically, adding the term is like adding a bandpass measured with a power meter. The soliton pulse evolution
filter in the laser cavity. inside the laser cavity was monitored with a high-speed de-
In this paper we present results of numerical simulations tector and a sampling oscilloscope.
on soliton formation and soliton energy quatization in a fiber The soliton operation of the laser was extensively inves-
ring laser passively mode locked by using the NPR tech- tigated previously 关19–21兴; various features such as the
nique. First we show that soliton formation is actually a natu- pump power hysteresis, multiple-soliton generation, various
ral consequence of a mode-locked pulse under strong pump- modes of multiple-soliton operation, and bound states of
ing if a laser is operating in the anomalous total cavity solitons were observed. Worthy of mention here again is the
dispersion regime. In particular we show how the parameters pump hysteresis effect of the soliton operation. It was found
of a laser soliton, such as the peak power and pulse width, experimentally that the laser always started mode locking at
vary with the laser operation conditions. Based on our nu- a high pump power level, and immediately after the mode
merical simulations we further show that the multiple soliton locking multiple solitons were formed in the cavity. After
formation in the laser is caused by a peak-power-limiting soliton operation was obtained, the laser pump power could
effect of the laser cavity. It is also the effect of the cavity that then be reduced to a very low level while the laser still main-
suppresses the soliton collapse and makes the solitons stable tained the soliton operation. This phenomenon of the laser
in the laser even when the laser gain is very strong. Further- soliton operation is known as pump power hysteresis 关22兴. It
more, we demonstrate numerically that the soliton energy later turned out that the pump power hysteresis effect is re-
quantization of the laser is a natural consequence of the gain lated to the multiple-soliton operation of the laser. Once mul-
competition between the solitons in the cavity. tiple solitons are generated in the cavity, on decreasing the
pump power the number of solitons is reduced. However, as
II. EXPERIMENTAL OBSERVATIONS long as one soliton remains in the cavity, the soliton opera-
tion state 共and therefore the mode locking of the laser兲 is
For the purpose of comparison and a better understanding maintained. Not only the soliton operation of the laser, but
of our numerical simulations, we present here again some of also the generation and annihilation of each individual soli-
the typical experimental results on multiple-soliton operation ton in the laser exhibited pump power hysteresis 关19兴. Ex-
and soliton energy quantization of soliton fiber lasers. We perimentally it was observed that if there were already soli-
note that although the results presented here were obtained tons in the cavity, on carefully increasing the pump power,
from a particular soliton fiber ring laser as described below, new solitons could be generated one by one in the cavity. As
similar features were also observed in other lasers 关3–5,8兴, in this case the laser is already mode locked, the generation
and are in fact independent of the concrete laser systems. A of a new soliton only requires a small increase of the pump
schematic of the fiber soliton laser we used in our experi- power.
ments is shown in Fig. 1. It contains a 1-m-long dispersion- An important characteristic of the multiple-soliton opera-
shifted fiber with group velocity dispersion 共GVD兲 of about tion of the laser is that, as long as the solitons are far apart in
−2 ps/ nm km, a 4-m-long erbium-doped fiber 共EDF兲 with the cavity, they all have exactly the same soliton parameters:
GVD of about −10 ps/ nm km, and a 1-m-long standard the same pulse width, pulse energy, and peak power. To dem-
single-mode fiber with GVD of about −18 ps/ nm km. Two onstrate the property we have shown in Fig. 2 the oscillo-
polarization controllers, one consisting of two quarter-wave scope trace of a typical experimentally measured multiple-
plates and the other one of two quarter-wave plates and one soliton operation state of our laser. The cavity round-trip
half-wave plate, were used to control the polarization of the time of the laser is about 26 ns. There are six solitons coex-
light in the cavity. A polarization-dependent isolator was isting in the cavity. It can be clearly seen that each soliton

043816-2
MECHANISM OF MULTISOLITON FORMATION AND … PHYSICAL REVIEW A 72, 043816 共2005兲

has exactly the same pulse height in the oscilloscope trace. described above. To describe the light propagation in the
Although with the electronic detection system the detailed weakly birefringent fibers, we used coupled complex nonlin-
pulse profile of the solitons cannot be resolved, nevertheless, ear Schrödinger equations of the form

冉 冊
the measured pulse height in the oscilloscope trace is directly
related to the energy of each individual soliton. Based on the ⳵u ⳵ u ik⬙ ⳵2u ik⵮ ⳵3u 2 2
= − i␤u + ␦ − 2 + 3 + i␥ 兩u兩 + 兩v兩 u
2
measured autocorrelation traces and optical spectra it was ⳵z ⳵t 2 ⳵t 6 ⳵t 3
further identified that all the solitons indeed have exactly the
i␥ 2 * g g ⳵ 2u
same soliton parameters. + v u + u+ ,
3 2 2⍀g ⳵ t2
III. THEORETICAL MODELING

To find the physical mechanism of multiple-soliton forma-

tion in our laser, we have experimentally carefully investi-
⳵v
⳵z
⳵ v ik⬙ ⳵2v ik⵮ ⳵3v
= i␤v − ␦ −
⳵t 2 ⳵t 2 +
6 ⳵t
2
3
冉
2 2
3 + i␥ 兩v兩 + 兩u兩 v 冊
gated its soliton operation and compared with those calcu- i␥ 2 * g g ⳵ 2v
lated from the conventional models of fiber soliton lasers. + u v + v+ , 共1兲
Traditionally, the soliton operation of a laser was modeled by 3 2 2⍀g ⳵ t2
the Ginzburg-Landau equation 关23兴 or the master equation where u and v are the normalized envelopes of the optical
关24兴, which also takes into account the gain, loss, and satu- pulses along the two orthogonal polarization axes of the fi-
rable absorber effects of a laser. However, a drawback of the ber. 2␤ = 2␲⌬n / ␭ is the wave-number difference between the
models is that the laser cavity effect on the soliton was either two modes. 2␦ = 2␤␭ / 2␲c is the inverse group velocity dif-
ignored or not appropriately considered. Based on results of ference. k⬙ is the second-order dispersion coefficient, k⵮ is
our experimental studies, we found that the cavity properties the third-order dispersion coefficient, and ␥ represents the
affect significantly the features of the solitons as in the case nonlinearity of the fiber. g is the saturable gain of the fiber
of soliton lasers the solitons circulate inside a cavity. There- and ⍀g is the bandwidth of the laser gain. For undoped fibers
fore, we believe that in order to accurately model the soliton g = 0. For the erbium-doped fiber, we further considered the
operation of a laser, the detailed cavity properties must also gain saturation as
be included in the model. To this end we have extended the
冕
冢 冣
conventional Ginzburg-Landau equation model through in-
corporating the cavity features. In previous papers we have 共兩u兩2 + 兩v兩2兲dt
reported results of using the model in simulating the experi- g = G exp − , 共2兲
Psat
mentally observed soliton sideband asymmetry 关25兴, subside-
band generation 关26兴, twin-pulse solitons 关27兴, and soliton where G is the small signal gain coefficient and Psat is the
pulse train nonuniformity 关28兴. We found that with our normalized saturation energy.
model we could well reproduce all the experimentally ob- To be close to the experimental conditions of our laser, we
served phenomena of our laser. have used the following fiber parameters for our simulations:
The basic idea of our model, which is fundamentally dif- ␥ = 3 W−1 km−1, k⵮ = 0.1 ps2 / nm km, ⍀g = 20 nm, gain satu-
ferent from conventional models, is that we did not make the ration energy Psat = 1000, cavity length L = 6 m, and the beat
small-pulse-variation approximation. Instead we follow the length of the fiber birefringence Lb = L / 4. To simulate the
circulation of the optical pulses in the laser cavity and con- cavity effect, we let the light circulate in the cavity. Starting
sider every action of the cavity components on the pulses. from the intracavity polarizer, which has an orientation of
Concretely, we describe the light propagation in the optical ␪ = 0.125␲ to the fiber’s fast axis, the light then propagates in
fibers by the nonlinear Schrödinger equation, or coupled the various fibers, first through the 1 m dispersion-shifted
nonlinear Schrödinger equations if the fiber is weakly bire- fiber 共DSF兲, which has a GVD coefficient of
fringent. For the erbium-doped fiber, we also incorporate k⬙ = −2 ps/ nm km, then the 4 m EDF, whose GVD coeffi-
gain effects such as the light amplification and gain band- cient k⬙ = −10 ps/ nm km, and finally the 1 m standard single-
width limitation in the equation. Whenever the pulse encoun- mode fiber whose GVD coefficient k⬙ = −18 ps/ nm km. Sub-
ters a discrete cavity component, e.g., the output coupler or sequently the light passes through the wave plates, which
polarizer, we then take account of the effect of the cavity cause a fixed polarization rotation of the light. Note that
component by multiplying its transfer matrix with the light changing the relative orientations of the wave plates is physi-
field. As the model itself is very complicated, we have to cally equivalent to adding a variable linear cavity phase de-
numerically solve it and find the eigenstate of the laser under lay bias to the cavity. Certainly the principal polarization
certain operation conditions. In our numerical simulations axes of the wave plates are not aligned with those of the
we always start the calculation with an arbitrary light field. fibers, and in general the different fibers used in the laser
After one round-trip circulation in the cavity, we then use the cavity could also have different principal polarization axes.
calculated result as the input of the next round of calculation However, for simplicity of the numerical calculations, we
until a steady state is obtained. We found that the simulations have treated them as all having the same principal polariza-
will always approach to a stable solution, which corresponds tion axes, and considered the effect caused by the principal
to a stable laser state under certain operation conditions. polarization axis change by assuming that the polarizer has
To illustrate our technique, we present here the detailed virtually a different orientation to the fast axis of the fiber
procedure in simulating the soliton operation of the laser when it acts as an analyzer. In our simulations the orientation

043816-3
TANG et al. PHYSICAL REVIEW A 72, 043816 共2005兲

FIG. 3. Numerically calculated multiple-soliton operation state

of the laser. ␦⌽l = 1.20␲, G = 350. Other parameters used are de- FIG. 4. Relationship between the soliton number in the simula-
scribed in the text. tion window and the pump strength. ␦⌽l = 1.20␲.

weak linear cavity loss we found numerically that a single-

angle of the analyzer to the fiber fast axis is set as ␸ = ␲ / 2 soliton pulse could even be directly formed from a mode-
+ ␪. We point out that in the real laser system the analyzer is locked pulse through increasing the pump strength. This nu-
also the same polarizer 共PI兲. Therefore, the light after the merical result clearly shows that the large pump hysteresis of
analyzer is also the light after the polarizer. We then used the the soliton operation of the laser is caused by the existence of
light as the input for the next round of calculation, and the the large linear cavity loss of a practical laser. A large linear
procedure repeats until a steady state is achieved. cavity loss makes the mode-locking threshold of a laser very
high, which under the existence of the cavity saturable ab-
IV. SIMULATION RESULTS sorber effect causes the effective gain of the laser after mode
locking to be very large. As will be shown below, when the
The coupled complex nonlinear Schrödinger equations 共1兲 peak power of a pulse is clamped, this large effective laser
were numerically solved by using the split-step Fourier gain will then result in the formation of multiple solitons
method 关29兴. We found that by appropriately setting the lin- immediately after the mode locking of the laser. We have
ear cavity phase delay bias of the cavity, so that an artificial mentioned in the Introduction the theoretical work of Koma-
saturable absorber effect can be generated in the laser, self- rov et al. on the multistability and hysteresis phenomena in
started mode locking can always be generated in our simu- passively mode-locked fiber lasers. In the framework of their
lations through simply increasing the small signal gain coef- model they have explained these phenomena as caused by
ficient, which corresponds to increasing the pump power in the competition between the positive nonlinear feedback and
the experiments. Exactly as in the experimental observations, the negative phase modulation effect 关18兴. It is noteworthy
multiple-soliton pulses are formed in the simulation window that in their model, in order to obtain the multiple-soliton
immediately after the mode locking. In the steady state and operation, a cavity loss term caused by the frequency-
when the solitons are far separated, all the solitons obtained selective filter has to be added, which from another aspect
have exactly the same pulse parameters such as the peak confirms our numerical result shown above.
power and pulse width. Figure 3 shows, for example, the By making the linear cavity loss small, we have numeri-
numerically calculated multiple-soliton operation of the la- cally investigated the process of how a soliton is formed in
ser. As in the experimental observations, the soliton opera- the laser cavity. Figure 5 shows the results of numerical
tion of the laser and the generation and annihilation of each simulations. In obtaining the result the linear cavity phase
individual soliton in the cavity exhibit pump hystersis. De- delay bias is set to ␦⌽l = 1.2␲. When G is less than 251, there
creasing the pump power numerically, the soliton number in is no mode locking. In the experiment this corresponds to the
the simulation window is reduced one by one, while care- case that the laser is operating below the mode-locking
fully increasing the pump strength; with at least one soliton threshold. When G is equal to 252, a mode-locked pulse
already existing in the cavity, solitons can also be generated emerges in the cavity. The mode-locked pulse has weak
one by one as shown in Fig. 4. All these numerically calcu- pulse intensity and broad pulse width. Due to the action of
lated results are in excellent agreement with the experimental the mode locker, which in the laser is the artificial saturable
observations. absorber, the mode-locked pulse circulates stably in the cav-
In a practical laser due to the existence of laser output, ity, just like any mode-locked pulse in other lasers. Although
fiber splices, etc., the linear cavity losses are unavoidable. such a mode-locked pulse has a stable pulse profile during
However, in the numerical simulations we could artificially circulation in the cavity, we emphasize that it is not a soliton
reduce the linear cavity loss and even make it zero. We found but a linear pulse. The linear nature of the pulse is also
numerically that the weaker the linear cavity loss, the smaller reflected by the fact that its optical spectrum has no side-
is the pump hysteresis of the soliton operation. With a very bands. When G is further increased, the peak power of the

043816-4
MECHANISM OF MULTISOLITON FORMATION AND … PHYSICAL REVIEW A 72, 043816 共2005兲

the soliton pulse width narrowing, the spectrum of the soli-

ton broadens, and consequently more sidebands become vis-
ible. However, the positions of the sidebands are almost
fixed. The physical mechanism of sideband generation of
laser solitons was extensively investigated previously and is
well understood now 关31兴. It is widely believed that the side-
band generation is a fundamental limitation to the soliton
pulse narrowing in a laser 关32兴. However, our numerical
simulations clearly show that the sideband generation is just
an adaptive effect, whose existence does not limit the soliton
pulse narrowing. As long as the pump power can balance the
loss caused by the sidebands, the soliton pulse width can still
be narrowed. Based on our numerical simulation and if there
is no other limitation as will be described below in the paper,
the narrowest soliton pulse that can be formed in a laser
should be ultimately only determined by the laser cavity dis-
persion property, including the net dispersion of all the cavity
components and the dispersion of the gain medium.
With already one soliton in the simulation window, we
then further increased the pump strength. Depending on the
selection of the linear cavity phase delay bias, we found that
the mechanism of further soliton generation and the features
of the multiple-soliton operation in the laser are different.
With the laser parameters as described above, we found that
when the linear cavity phase delay bias is set small, say at
about ␦⌽l = 1.2␲, on further increasing the pump power, ini-
tially the soliton pulse peak power will be increased and its
pulse width narrowed as expected. However, at a certain
fixed value these effects will stop; instead the background of
the simulation window becomes unstable and weak back-
ground pulses become visible as shown in Fig. 6共b兲. On fur-
ther slightly increasing the pump power, an additional soliton
is quickly formed in the cavity through the soliton shaping of
FIG. 5. Soliton shaping of the mode-locked pulse in the laser.
one of the weak background pulses. As the weak background
␦⌽l = 1.20␲. Top figure: Evolution of pulse profile with the pump
pulses are always initiated from the dispersive waves of the
strength. Bottom figure: Evolution of the optical spectra with the
pump strength.
solitons, we have called this type of soliton generation “soli-
ton shaping of dispersive waves” 关33兴. In the steady state
pulse quickly increases. Associated with the pulse intensity both solitons have exactly the same pulse width and peak
increase the nonlinear optical Kerr effect of the fiber also power as shown in Fig. 6共c兲. When the pump power is fur-
becomes strong and eventually starts to play a role. An effect ther increased, additional solitons are generated one by one
of the pulse self-phase-modulation 共SPM兲 is to generate a in the simulation window in exactly the same way and even-
positive frequency chirp, which in the anomalous cavity dis- tually a multiple-soliton state as shown in Fig. 3 is obtained.
persion regime counterbalances the negative frequency chirp This numerically simulated result is well in agreement with
caused by the cavity dispersion effect and compresses the the experimental observations 关19兴. Because of the additional
pulse width. When the pulse peak power has become so soliton generation, the solitons formed in the laser cannot
strong that the nonlinear SPM effect alone can balance the have large pulse energy and high peak power through simply
pulse broadening caused by the cavity dispersion effect, even increasing the pump power. The larger the laser gain, the
without the existence of the mode locker, a pulse can propa- more solitons would be formed in the cavity.
gate stably in the dispersive laser cavity. In this case a mode- When the linear cavity phase delay bias is set at a very
locked pulse then becomes a soliton. In the case of our simu- large value, say at about ␦⌽l = 1.8␲, which is still in the
lation, this corresponds to the state of G = 253. A soliton in positive cavity feedback range but close to the other end, no
the laser is also characterized by the appearance of the side- stable propagation of solitons in cavity can be obtained. With
bands in the optical spectrum as shown in Fig. 5. the linear cavity phase delay bias selection, there is a big
Once the laser gain is fixed, a soliton with fixed peak difference between the linear cavity loss and the nonlinear
power and pulse width will be formed, which is independent cavity loss. Therefore, if the gain of laser is smaller than the
of the initial conditions. The states shown in Fig. 5 are stable dynamical loss that a soliton experienced, the soliton quickly
and unique. This result confirms the autosoliton property of dies out as shown in Fig. 7共a兲. While if the gain of the laser
the laser solitons 关30兴. However, if the pump power is con- is even slightly larger than the dynamical loss that a soliton
tinuously increased, solitons with even higher peak power experienced, the soliton peak power will increase. Higher
and narrower pulse width will be generated. Associated with soliton peak power results in smaller dynamical loss and

043816-5
TANG et al. PHYSICAL REVIEW A 72, 043816 共2005兲

FIG. 7. Soliton evolutions calculated with ␦⌽l = 1.80␲. G⫽共a兲

470; 共b兲 478; 共c兲 600.

FIG. 6. Process of additional soliton generation in the laser.

␦⌽l = 1.20␲. G⫽共a兲 255; 共b兲 270; 共c兲 275. Even in the cases of stable multiple-soliton operation, de-
pending on the selection of the linear cavity phase delay, the
even larger effective gain; therefore, the soliton peak will solitons obtained have different parameters. Figure 8 shows
continuously increase. Associated with the soliton peak in- for comparison the multiple-soliton operation obtained with
crease the soliton pulse width decreases; eventually the soli- the linear cavity phase delay bias set at ␦⌽l = 1.55␲. It is
ton breaks up into two solitons with weak peak power and seen that solitons with higher peak power and narrower pulse
broad pulse width as described by Kärtner et al. 关14兴. Once a width can be formed with the linear cavity phase delay set-
soliton is broken into two solitons with weak peak power, the ting. Extensive numerical simulations have shown that the
dynamical loss experienced by each of the solitons becomes larger the linear cavity phase delay setting, the higher the
very big. Consequently the gain of the laser cannot support soliton peak and the narrower the soliton pulse achievable.
them. The new solitons are then immediately destroyed as
shown in Fig. 7共b兲. If very large gain is available in the laser, V. MECHANISM OF MULTIPLE-SOLITON GENERATION
the additional solitons may survive in the cavity temporally AND SOLITON ENERGY QUANTIZATION
and each of them repeats the same process as shown in Fig.
7共b兲, and eventually a state as shown in Fig. 7共c兲 is formed. Apparently, depending on the laser linear cavity phase
Therefore, no stable soliton propagation is possible with a delay bias setting, there exist two different mechanisms of
too large linear cavity phase delay setting in the laser. soliton generation in the laser. One is the soliton shaping of

043816-6
MECHANISM OF MULTISOLITON FORMATION AND … PHYSICAL REVIEW A 72, 043816 共2005兲

FIG. 9. An equivalent setup to Fig. 1 for determining the cavity

transmission.

T = sin2共␪兲sin2共␸兲 + cos2共␪兲cos2共␸兲
1
FIG. 8. Multiple-soliton operation of the laser calculated with + sin共2␪兲sin共2␸兲cos共⌬⌽l + ⌬⌽nl兲. 共3兲
linear cavity phase delay bias set at ␦⌽l = 1.55␲, G = 465. 2

the unstable dispersive waves or the cw components, and the Chen et al. 关34兴 and Davey et al. 关35兴 have already shown
other one is the well-known mechanism of pulse splitting. It how to select the orientations of the polarizer and the ana-
is seen that in the laser the process of soliton splitting occurs
lyzer so that the cavity would generate efficiently saturable
only in the regime where the additional solitons formed are
absorption effect. In a previous paper 关25兴 we have also
practically unstable. We have already reported previously the
phenomenon of soliton generation through unstable back- shown that the linear cavity transmission of the laser is a
ground in the lasers 关33兴. Here we further explain its physical sinusoidal function of the linear cavity phase delay ⌬⌽l with
origin. a period of 2␲. It is to point out that within one period of the
Our soliton fiber laser is mode locked by using the NPR linear cavity phase delay change, the laser cavity can provide
technique. The operation mechanism of the technique has positive 共the saturable absorber type兲 cavity feedback only in
already been analyzed by several authors 关24,34,35兴. It has half of the period, in the other half of the period it actually
been shown that through inserting a polarizer in the cavity has negative feedback.
and appropriately setting the linear cavity phase delay, the As shown in Eq. 共3兲, the actual cavity transmission for an
NPR could generate an artificial saturable absorption effect optical pulse is also nonlinear phase delay ⌬⌽nl dependent.
in the laser. It is the artificial saturable absorber effect that To illustrate the functions of this part we use our simulations
causes the self-started mode locking of the laser. After a as an example. In our simulations the orientation of the po-
soliton is formed in the laser cavity, it further stabilizes the larizer has an angle of ␪ = 0.125␲ to the fast axis of the fiber,
soliton. Although previous studies have correctly identified so light propagation in the fiber will generate a negative
the effects of NPR and the saturable absorber in the laser, nonlinear phase delay. The linear cavity beat length is 41 of
there is no further analysis on how and to what extent these the cavity length, therefore, the maximum linear cavity trans-
effects affect the soliton parameters and soliton dynamics. mission is at the positions of 共2n + 1兲␲ linear cavity phase
Here we follow the description of Chen et al. 关34兴 to com- delays, where n = 0, 1,2,… . Furthermore, when the linear
plete it. Our approach is to first determine the linear and cavity phase delay is biased within the range between 共2n
nonlinear cavity transmission of the laser, and then based on + 1兲␲ and 2共n + 1兲␲, the cavity will generate a positive feed-
the results to further find out how they affect the solitons back, as under the effect of the nonlinear polarization rota-
formed in the laser. Physically, the laser cavity shown in Fig. tion the actual cavity transmission increases. While if the
1 can be simplified to a setup as shown in Fig. 9 for the linear cavity phase delay is located in the range from 2n␲ to
purpose of determining its transmission property. Starting 共2n + 1兲␲, the cavity will generate a negative feedback. The
from the intracavity polarizer, which sets the initial polariza- maximum linear cavity transmission point also marks the
tion of light in relation to the birefringent axes of the fiber, switching position of the two feedbacks. For the soliton op-
the polarization of light after passing through the fiber is eration the laser is always initially biased in the positive
determined by both the linear and nonlinear birefringence of cavity feedback regime. It is clear to see that depending on
the fiber. The light finally passes through the analyzer, which the selection of the linear cavity phase delay and the strength
in the experimental system is the same intracavity polarizer. of the nonlinear phase delay, the cavity feedback is possible
If we assume that the polarizer has an orientation of angle ␪ to be dynamically switched from the positive feedback to the
with respect to the fast axis of the fiber, the analyzer has an negative feedback regime. For the soliton operation of a laser
angle of ␸, the phase delay between the two orthogonal po- this cavity feedback switching has the consequence that the
larization components caused by the linear fiber birefrin- peak of a soliton formed in the cavity is limited. We found
gence is ⌬⌽l, and that caused by the nonlinear birefringence that it is this soliton peak limiting effect that results in the
is ⌬⌽nl, it can be shown that the transmission coefficient of multiple soliton generation in the soliton fiber laser and the
the setup or the laser cavity is 关34兴 soliton energy quantization.

043816-7
TANG et al. PHYSICAL REVIEW A 72, 043816 共2005兲

To explain these, we assume that the peak power of a Finally we note that the multiple-soliton operation and
soliton is so strong that it switches the cavity from the posi- soliton energy quantization effect have also been observed in
tive to the negative feedback regime. In this case although other passively mode-locked soliton fiber lasers, such as in
increasing the pump strength will still cause the peak power figure-of-eight lasers and lasers passively mode locked with
of the soliton to increase, the higher the soliton peak power semiconductor saturable absorbers 关8,9兴. Even in actively
increases, the smaller the actual cavity transmission be- mode-locked fiber lasers 关2兴 these phenomena have also been
comes. To a certain fixed value of the soliton peak power, observed. Despite the fact that those soliton lasers are not
which depends on the linear cavity phase delay setting, fur- mode locked with the NPR technique and therefore their
ther increase of the soliton peak power would result in that detailed cavity transmission could not have the same feature
the actual cavity transmission that the soliton experiences as described by Eq. 共3兲, enlightened by the result obtained in
becomes smaller than the linear cavity transmission. At this
our laser, we conjecture that there must also be a certain
point the soliton peak will be clamped. Further increasing the
pulse peak-power-limiting mechanism in those lasers, which
laser gain will not amplify the soliton but the background
noise such as the dispersive waves. If the background noise causes their multiple-pulse formation. Indeed, we found that
of a certain frequency satisfies the lasing condition, it could for the figure-of-eight lasers, if the fiber birefringence of the
also start to lase and form a cw component in the soliton nonlinear loop is further considered, it will also generate a
spectrum. We note that coexistence of solitons with cw is a similar pulse-peak-clamping effect in the laser. However, bi-
generic effect of the soliton fiber lasers, and the phenomenon refringence of fibers in the lasers is normally ignored. It was
was reported by several authors 关36,37兴. Linear waves are also reported that due to the two-photon absorption effect the
intrinsically unstable in the cavity due to the modulation in- SESAM used for the passive mode locking of fiber lasers has
stability. When they are strong enough, they become modu- a pulse peak-power-limiting effect 关39兴. It is therefore not
lated. Under the effect of saturable absorption, the strongest surprising that a soliton laser mode locked with the material
background pulse will be amplified and shaped into an addi- could also exhibit multiple solitons. For the actively mode-
tional soliton. This was exactly what we have observed in the locked laser, in most cases the multiple-soliton generation is
experiments on how an additional soliton was generated. The due to harmonic mode locking. In this case as too many
two solitons in the cavity share the same laser gain. As the solitons share the limited cavity gain, the energy of each
cavity generates a positive feedback for the weak soliton and pulse is weak. Therefore, even when the net cavity disper-
a negative one for the strong soliton, under gain competition sion is negative, solitons are normally difficult to form. We
the two solitons have to adjust their strength so that the point out that for an actively mode-locked fiber laser if the
stronger one becomes weaker, and the weaker one becomes cavity is not carefully designed, the cavity birefringence
stronger; eventually they will stabilize at a state where both combined with the modulator, which is a polarizing device,
solitons have exactly the same peak power. The soliton in- could form a birefringence filter and further limit the peak
ternal energy balance further determines their other param- power of the pulses formed in the lasers.
eters. Unless there are interactions between the solitons, they
will always have identical parameters in the stable state.
It turns out that the multiple-soliton formation in the laser VI. CONCLUSIONS
is in fact caused by the peak-power-clamping effect of the
cavity. In addition, the soliton energy quantization observed In conclusion, we have numerically studied the mecha-
is also a natural consequence of the gain competition be- nism of multiple-soliton generation and soliton energy quan-
tween the solitons in the laser. Obviously the maximum tization in a soliton fiber ring laser passively mode locked by
achievable soliton peak power in the laser is linear cavity using the nonlinear polarization rotation technique. We iden-
phase delay dependent. When the linear cavity phase delay is tified that the multiple-soliton generation in the laser is
set close to the cavity feedback switching point, solitons with caused by the peak-power-clamping effect of the cavity. De-
relatively lower peak power could already dynamically pending on the linear cavity phase delay setting, the nonlin-
switch the cavity feedback. Therefore, solitons obtained at ear phase delay generated by a soliton propagating in the
this linear cavity phase delay setting have lower peak power fiber cavity could be so large that it switches the cavity feed-
and broader pulse width as shown in Fig. 6共a兲; while if the back from the initially selected positive regime into the nega-
linear cavity phase delay is set far away from the switching tive regime. And as a result of the cavity feedback change the
point, the soliton peak power is clamped at a higher value, maximum achievable soliton peak power is then limited. In
and solitons with higher peak and narrower pulse width are this case increasing the laser pump power will not increase
obtained as shown in Fig. 8. In particular, if the linear cavity the peak power of the solitons, but generate an additional
phase delay is set too close to the switching point, as the soliton. Therefore, multiple solitons are formed in the laser.
peak power of the pulse is clamped to too small a value, As the solitons share the same laser gain, gain competition
except for mode-locked pulses, no soliton can be formed in between them combined with the cavity feedback feature fur-
the laser. When the linear cavity phase delay is set too far ther results in the fact that in the steady state they have
away from the switching point as demonstrated numerically exactly the same soliton parameters. The parameters of soli-
in Fig. 7, before the soliton peak reaches the switching point, tons formed in the laser are not fixed by the laser configura-
it has already become so high and so narrow that it splits, tion but vary with the laser operation conditions, which are
and no stable soliton propagation can be obtained in the la- determined by the soliton internal energy balance between
ser. Instead only the state of so-called noiselike pulse emis- the shared laser gain and the dynamical losses of each
sion will be observed 关38兴. soliton.

043816-8
MECHANISM OF MULTISOLITON FORMATION AND … PHYSICAL REVIEW A 72, 043816 共2005兲

关1兴 K. Smith, J. R. Armitage, R. Wyatt, and N. J. Doran, Electron. and H. Y. Tam, Appl. Phys. B: Lasers Opt. 77, 585 共2003兲.
Lett. 26, 1149 共1990兲. 关21兴 D. Y. Tang, W. S. Man, H. Y. Tam, and P. Drummond, Phys.
关2兴 R. P. Davey, N. Langford, and A. I. Ferguson, Electron. Lett. Rev. A 66, 033806 共2002兲.
27, 1257 共1991兲. 关22兴 M. Nakazawa, E. Yoshida, and Y. Kimura, Appl. Phys. Lett.
关3兴 D. J. Richardson, R. I. Laming, D. N. Payne, V. J. Matsas, and 59, 2073 共1991兲.
M. W. Phillips, Electron. Lett. 27, 1451 共1991兲. 关23兴 A. K. Komarov and K. P. Komarov, Phys. Rev. E 62, R7607
关4兴 M. Nakazawa, E. Yoshida, and Y. Kimura, Electron. Lett. 29, 共2000兲.
63 共1993兲. 关24兴 H. A. Haus, J. G. Fujimoto, and E. P. Ippen, J. Opt. Soc. Am.
关5兴 V. J. Matsas, D. J. Richardson, T. P. Newson, and D. N. Payne,
B 8, 2068 共1991兲.
Opt. Lett. 18, 358 共1993兲.
关25兴 W. S. Man, H. Y. Tam, M. S. Demonkan, P. K. A. Wai, and D.
关6兴 K. Tamura, E. P. Ippen, H. A. Haus, and L. E. Nelson, Opt.
Y. Tang, J. Opt. Soc. Am. B 17, 28 共2000兲.
Lett. 18, 1080 共1993兲.
关7兴 M. Nakazawa, E. Yoshida, T. Sugawa, and Y. Kimura, Elec- 关26兴 D. Y. Tang, S. Fleming, W. S. Man, H. Y. Tam, and M. S.
tron. Lett. 29, 1327 共1993兲. Demokan, J. Opt. Soc. Am. B 18, 1443 共2001兲.
关8兴 L. A. Gomes, L. Orsila, T. Jouhti, and O. G. Okhotnikov, IEEE 关27兴 D. Y. Tang, B. Zhao, D. Y. Shen, C. Lu, W. S. Man, and H. Y.
J. Sel. Top. Quantum Electron. 10, 129 共2004兲. Tam, Phys. Rev. A 68, 013816 共2003兲.
关9兴 B. C. Collings, K. Bergman, S. T. Cundiff, S. Tsuda, J. N. 关28兴 B. Zhao, D. Y. Tang, L. M. Zhao, P. Shum, and H. Y. Tam,
Kutz, J. E. Cunningham, W. Y. Jan, M. Koch, and W. H. Knox, Phys. Rev. A 69, 043808 共2004兲.
IEEE J. Sel. Top. Quantum Electron. 3, 1065 共1997兲. 关29兴 G. P. Agrawal, Nonlinear Fiber Optics, 3rd ed. 共Academic
关10兴 A. B. Grudinin, D. J. Richardson, and D. N. Payne, Electron. Press, New York, 2001兲, pp. 51–55.
Lett. 28, 67 共1992兲. 关30兴 V. S. Grigoryan and T. S. Muradyan, J. Opt. Soc. Am. B 8,
关11兴 N. N. Akhmediev, A. Ankiewicz, and J. M. Soto-Crespo, Phys. 1757 共1991兲.
Rev. Lett. 79, 4047 共1997兲. 关31兴 S. M. J. Kelly, Electron. Lett. 28, 806 共1992兲.
关12兴 M. J. Lederer, B. Luther-Davies, H. H. Tan, C. Jagadish, N. N. 关32兴 M. L. Dennis and I. N. Duling III, IEEE J. Quantum Electron.
Akhmediev, and J. M. Soto-Crespo, J. Opt. Soc. Am. B 16, 30, 1469 共1994兲.
895 共1999兲. 关33兴 W. S. Man, H. Y. Tam, M. S. Demonkan, and D. Y. Tang, Opt.
关13兴 C. Spielmann, P. F. Curley, T. Brabec, and F. Krausz, IEEE J. Quantum Electron. 33, 1139 共2001兲.
Quantum Electron. 30, 1100 共1994兲. 关34兴 C. J. Chen, P. K. A. Wai, and C. R. Menyuk, Opt. Lett. 17, 417
关14兴 F. X. Kärtner, J. Aus der Au, and U. Keller, IEEE J. Sel. Top. 共1992兲.
Quantum Electron. 4, 159 共1998兲. 关35兴 R. P. Davey, N. Langford, and A. I. Ferguson, Electron. Lett.
关15兴 G. P. Agrawal, Phys. Rev. A 44, 7493 共1991兲. 29, 758 共1993兲.
关16兴 P. Grelu, F. Belhache, F. Gutty, and J. M. Soto-Crespo, J. Opt. 关36兴 S. Namiki, E. P. Ippen, H. A. Haus, and K. Tamura, Appl.
Soc. Am. B 20, 863 共2003兲. Phys. Lett. 69, 3969 共1996兲.
关17兴 Ph. Grelu and J. M. Soto-Crespo, J. Opt. B: Quantum Semi- 关37兴 B. Zhao, D. Y. Tang, and H. Y. Tam, Opt. Express 11, 3304
classical Opt. 6, S271 共2004兲. 共2003兲.
关18兴 A. Komarov, H. Leblond, and F. Sanchez, Phys. Rev. A 71, 关38兴 M. Horowitz, Y. Barad, and Y. Silberberg, Opt. Lett. 22, 799
053809 共2005兲. 共1997兲.
关19兴 D. Y. Tang, W. S. Man, and H. Y. Tam, Opt. Commun. 165, 关39兴 E. R. Thoen, E. M. Koontz, M. Joschko, P. Langlois, T. R.
189 共1999兲. Schibli, F. X. Kartner, E. P. Ippen, and L. A. Kolodziejski,
关20兴 B. Zhao, D. Y. Tang, P. Shum, Y. D. Gong, C. Lu, W. S. Man, Appl. Phys. Lett. 74, 3927 共1999兲.

043816-9