PHYSICAL REVIEW LETTERS week ending

PRL 95, 143902 (2005) 30 SEPTEMBER 2005

Experimental Observation of Temporal Soliton Molecules

M. Stratmann,* T. Pagel, and F. Mitschke
Institut für Physik, Universität Rostock, Rostock, Germany†
(Received 12 April 2005; published 28 September 2005)
A bound state of temporal solitons in optical fibers is predicted numerically and demonstrated
experimentally. It is appropriately described as a pair of bright solitons, bound together by a dark soliton.
This structure exists only in dispersion-managed fiber and is different from bound solitons in fiber lasers.

DOI: 10.1103/PhysRevLett.95.143902 PACS numbers: 42.81.Dp, 42.65.Tg, 42.79.Sz

Several nonlinear wave equations can give rise to sol- more complex bound states for more letters (‘‘3,’’ ‘‘4,’’ and
itons, and sometimes solitons in very different fields are so on) exist.
governed by the same equation. For example, the Gross- Propagation of light pulses in optical fiber is described
Pitaevskii equation of Bose-Einstein condensates is mathe- by the nonlinear Schrödinger equation [7]. It is well known
matically equivalent to the nonlinear Schrödinger equation that in fibers with anomalous group-velocity dispersion
of nonlinear optics. In either case, both bright and dark bright solitons exist, while in fibers with normal dispersion,
solitons have been experimentally demonstrated as de- dark solitons are supported. Increasingly the telecommu-
scribed by that equation. nications industry now uses fiber lines with periodically
The arguably most relevant application of solitons is in alternating dispersion. Such ‘‘dispersion-managed fibers’’
fiber-optical telecommunication. While that technology have several advantages (see, e.g., [7]).
has seen dramatic progress over the last decades, it will A periodic dispersion map is basically characterized by
soon reach a fundamental limit. Based on entropy flow two dispersion values (


2 > 0 and
2 < 0) and their
considerations, Shannon [1] has shown that the rate of respective segment lengths (L and L ). Its effect on

information flow through some channel is bounded by pulses of duration  (full width at half maximum) is best
the channel capacity. This capacity is set by the available described by using the map period Lmap  L

L , the


 
bandwidth times a factor which depends on the signal-to- path-average dispersion
ave 2  
2 L

2 L =Lmap ,
noise ratio or, more appropriate for digital transmission, on and the map strength S, defined by
the coding format. Transmission through optical fiber with
binary encoding, which is the standard to date, will reach j

ave  ave
2 
2 jL

j
2 
2 jL
S :
the Shannon limit within a few years. Several attempts 2
have been made of finding a way around the limitation It has been established that bright solitons exist in such
[2,3] other than the trivial yet costly possibility of deploy- fibers, and that their existence is not strictly limited to the
ing more fibers in parallel. regime of anomalous path-average dispersion; rather, they
The goal is to find a physical representation of the exist in a small parameter range of normal path-average
information with coding schemes that would allow the dispersion [8]. A corresponding condition could be estab-
transmission of more than 1 bit of information per time lished for dark solitons in [9]. This together implies that the
slot [4]. However, any effort in this direction has to take regimes of existence of bright and dark solitons overlap.
fiber nonlinearity into account, which requires a reconsid- We discovered recently in numerical experiments that in
eration of Shannon’s capacity limit. It was shown recently this overlap range bright and dark solitons cannot only
that conventional coding of more than two values in either coexist, but actually form a stable bound state in which a
amplitude [5] or phase [6] provides only marginal benefits bright pulse sits on either side of a dark soliton [10]. It
due to channel crosstalk based on nonlinear mixing pro- appears that this structure, shown in Fig. 1, is the same as
cesses in the fiber. the Hermite-Gaussian asymmetric soliton discussed in
The only conceivable remedy would be a coding with an [11]. We prefer to describe it as a compound of one dark
‘‘alphabet’’ of more than two letters, where the physical and two bright solitons. Let us emphasize that such struc-
representation of each letter has a similar self-healing ture in homogenous (constant-dispersion) fiber is unstable
robustness as a soliton. So far it was not known whether under all circumstances; this entity only exists in
such representation exists. Solitons are employed in a dispersion-alternating fiber. The structure can be approxi-
format where letters ‘‘0’’ (no light) and ‘‘1’’ (a soliton mated by two Gaussian pulses in antiphase; Fig. 1 shows
pulse) exist, but no more. We describe here our experimen- the resulting asymptotic shape after lossless, perturbation-
tal proof of existence of a bound state of solitons which free propagation which was obtained numerically follow-
might possibly serve as the next letter of the alphabet (the ing the procedure decribed in [8]. Figure 2 shows that this
‘‘2’’). At this point it must be left to speculation whether structure is stable in long-distance propagation: even on the

0031-9007=05=95(14)=143902(4)$23.00 143902-1 © 2005 The American Physical Society

 PHYSICAL REVIEW LETTERS week ending
PRL 95, 143902 (2005) 30 SEPTEMBER 2005

FIG. 1. The soliton molecule consists of two bright pulses with

a point of zero power in between; the phase jumps by  at the
FIG. 2. Numerical simulation of long-distance propagation of
zero. Alternately, the central notch can be thought of as a dark
the soliton molecule. Here S  7:2 was assumed; the maximum
soliton.
distance shown corresponds to 77LD .

log scale, no change of shape is visible. We emphasize that the dark soliton in the middle, is a salient feature of the
this soliton compound resembles neither higher-order sol- soliton molecule.
itons, which have no binding energy, nor various soliton Further studies of the stability of the soliton molecule
compounds in passive resonators [12] or fiber lasers against various perturbations reveal that symmetric pertur-
[13,14], which hinge on cavity boundary conditions and/ bations (scaling of power level or time separation) heal out.
or gain dispersion or other effects that come with the gain. Asymmetric perturbation like uneven powers of the two
This soliton compound is characterized by its two main bright pulses or effects of third-order dispersion do not.
features: The bright pulses are in antiphase, and they sit at a This may, however, not be a major obstacle: Fortunately,
certain distance from each other. We numerically inves- the growth rate of the power asymmetry is slow. For an
tigated how important these features are. For the antiphase
case we launched the structure with intentionally modified
separation between the bright pulses (see Fig. 3). It turns
out that for too narrow spacing the bright pulses repel,
whereas for too wide spacing they attract; in either case
they return to the equilibrium distance. This is reminiscent
of the equilibrium separation of the two constituents of a
diatomic molecule; we therefore call this structure a soliton
molecule. The relaxation requires a sufficient propagation
distance; it is useful to measure the latter in units of the
characteristic dispersion length LD as defined, e.g., in [7],
and referred to an individual bright pulse of the same
width, at the path-average dispersion.
Propagation of the double-pulse structure for various
power levels is demonstrated in Fig. 4. We find that the
full width at half energy is a good measure of the pulse
separation since as an integral measure it is insensitive to
wiggles. The molecule is formed for either sign of the path-
average dispersion, provided the power is chosen correctly.
Below a certain threshold, dispersion leads to broadening
of the structure.
For relative phases significantly different from  the
structure is not stable, but collapses into a single pulse.
This sets it apart from the stable in-phase double-pulse
structure found numerically by Maruta et al. [15] which,
however, requires the nonlinearity to alternate between two FIG. 3. Numerical experiment on perturbation of initial sepa-
values, in addition to the sign-alternating dispersion. Our ration of the bright constituents. Whether they are launched with
soliton molecules exist in the absence of such nonlinearity too small (top) or too large (center, bottom) separation, they
modulation, but tolerate some degree of it. Certainly the relax towards their equilibrium distance (dotted lines).
antiphase condition, or in other words the binding action of Propagation distance was 201 km31L ^ D at S  5:4.

143902-2
 PHYSICAL REVIEW LETTERS week ending
PRL 95, 143902 (2005) 30 SEPTEMBER 2005

initial power asymmetry in the few-percent range, the required length for a meaningful experimental test. It con-
asymmetry grows to the point of loss of stability only after sists of alternating segments of 497 cm of anomalously
tens of dispersion lengths; however, in any real system dispersive standard single mode fiber (SSMF,
 2 
there are restoring amplifiers every few dispersion lengths. 26 ps2 =km at 1595 nm) and 183 cm of normally disper-
Amplifier saturation can easily restore the power symme- sive ‘‘inverse dispersion fiber’’ (IDF,

2
2 
62 ps =km
try. Third-order dispersion can be made quite small by at 1595 nm); the line begins and ends with a half-segment
suitable fiber design. of SSMF. OFS-Fitel [16] provided the fibers and spliced
We now turn to an experimental verification of soliton them with 0.28 dB average loss employing proprietary
molecules. The main challenge was to set up a dispersion- procedures. We find numerically that soliton molecules
managed fiber line of sufficient length in the presence of persist in the presence of such loss when the power at
several severe constraints. We settled for a scale model of a launch point is corrected upward by half that amount so
real telecommications fiber, so as to reduce tens or hun- that the correct power level exists in the fiber on average.
dreds of kilometers to tens of meters of fiber: With input The path-average dispersion of the transmission line can
pulses around 300 fs all fiber lengths involved are scaled be tuned from 2:2 . . .  3:2 ps2 =km by variation of the
down to manageable values. This pulse duration was con- operating wavelength. This implies map strengths of S 
sidered the best compromise in the presence of perturba- 2:6 . . . 3:3 for our 300 fs pulses. The total length of the
tions from the Raman effect or higher-order dispersion; transmission line corresponds to 1:5 . . . 2:3LD , which again
these will affect our scale model more than a full scale is close to the minimum meaningful value.
setup, so that conservative extrapolations from our worst- As mentioned above, soliton molecules are somewhat
case to real-sized systems can be made. Other than in unstable in the presence of third-order dispersion
3 . IDF
Figs. 1– 4, in the following we will compare experimental can be used to completely null out its path-average value
data to results from an extended model including these
ave
3 , but we find that this sensitivity is minimal at a certain
effects. finite value of
ave3 opt
0. Our fiber line approximates
In principle one would like to have a long fiber, in the this value.
sense that it consists of a large number of dispersion- As a light source we use an optical parametric oscillator,
alternation periods. Unfortunately, every such period by pumped by a Ti:sapphire laser. The light pulses are sent
necessity involves two splices between fibers that differ in through a variable attenuator to set the desired power level,
their mode field geometry. A perfect butt couple between then split in a Mach-Zehnder interferometer and recom-
the fibers used here would give an unacceptable 1.19 dB of bined into pulse pairs of adjustable temporal separation.
loss. In a commercial system such losses would be taken
Phase adjustment is performed with piezoceramic actuator
care of by inline optical amplifiers which, however, would
control. Finally the pulse pairs are launched into the
introduce gain dynamics of their own. To keep the dem-
dispersion-managed fiber line. Both an autocorrelator and
onstration experiment simple and clear, we decided to do it
an optical spectrum analyzer serve to analyze the pulse
without the extra complexity of optical amplifiers. Only in
pair; they can be switched to compare signals before and
this way can we demonstrate that dispersion management
after the fiber.
is solely responsible for the structure we describe here.
In series of measurements we varied the optical power,
However, this decision forces us to compromise on the
the separation, and the relative phase. In Fig. 5 we show the
fiber length.
output separation between the bright pulses as a function of
The dispersion-managed fiber transmission line consists
their input separation. At large separations there is virtually
of three compensation periods which is the minimum
no interaction, and the measured points fall onto the bi-
sector. As the pulses get closer to each other, one expects
attraction (in-phase) or repulsion (antiphase) between both
bright pulses. This holds until the separation is about the
pulse width because a Mach-Zehnder interferometer does
not allow us to create antiphase double pulses with peak-
to-peak separations smaller than the pulse width.
The data shown are taken at antiphase. At low power
(circles) there is repulsion just as expected. The data at
elevated power (squares) are taken in the realm of the
soliton molecule. Here the repulsion is seen to be strongly
reduced. Indeed, for the circumstances a molecule of
 450 fs separation is expected.
In Fig. 6 we varied the power level at which a double-
pulse structure (antiphase) was launched. No data could be
FIG. 4. Power dependence of bright pulse separation after a taken below  50 pJ for lack of sensitivity of the autocor-
propagation distance of 29LD . relator. At the lowest powers shown the structure is broad-
143902-3
 PHYSICAL REVIEW LETTERS week ending
PRL 95, 143902 (2005) 30 SEPTEMBER 2005

FIG. 6. Experimental results for the separation between the

bright constituents as a function of launch power. The initial
FIG. 5. Experimental data for separation between the bright separation of 525 fs is identified by an arrow.
pulse peaks: output vs input. For large separations the measured
points fall onto the bisector. At smaller separation and low power
there is repulsion. At increased power the repulsion is counter- benefited from assistance by M. Böhm and H. Hartwig.
acted. Lines are from numerical simulations based on the known
pulse and fiber parameters.

ened out. The line labeled ‘‘  0’’ indicates the expected *Electronic address: [email protected]
†
broadening in the absence of nonlinearity. When the power Electronic address: http://www.physik.uni-rostock.de/
level is increased, the broadening is reduced. As we reach Optik
the regime where we expect to find the molecules, we find [1] C. E. Shannon, Bell Syst. Tech. J. 27, 379 (1948).
[2] N. N. Akhmediev, G. Town, and S. Wabnitz, Opt.
that the repulsion turns to an attraction such that a certain
Commun. 104, 385 (1994).
mutual distance of the pulse pair is attained. In the data [3] N. N. Akhmediev and A. Ankiewicz, Chaos 10, 600
shown this separation of  460 fs is slightly closer than in (2000).
the launch condition with 525 fs. As the power level is [4] J. M. Kahn and K.-P. Ho, Nature (London) 411, 1007
further increased, we expect from simulations that compli- (2001).
cations from the Raman effect would begin to be notice- [5] P. P. Mitra and J. B. Stark, Nature (London) 411, 1027
able. However, in the experiment the available power was (2001).
insufficient to clearly cross an upper threshold where the [6] K.-P. Ho and J. M. Kahn, Proceedings of the Optical Fiber
preferred distance is lost again. The solid line is a numeri- Communications Conference 2002 paper ThGG85
cal simulation based on the same pulse and fiber parame- (Optical Society of America, Washington, DC, 2002).
ters as in Fig. 5, not a fit. The agreement with the measured [7] G. P. Agrawal, Fiber-Optic Communication Systems
(Wiley, New York, 2002).
data points is remarkable. The data in Fig. 6 were taken
[8] J. H. B. Nijhof, N. J. Doran, W. Forysiak, and F. M. Knox,
with the bright pulses in antiphase. We emphasize that at Electron. Lett. 33, 1726 (1997).
other phase angles the behavior is quite different: for zero [9] M. Stratmann, M. Böhm, and F. Mitschke, Electron. Lett.
phase difference the two bright pulses merge into one. 37, 1182 (2001).
All observations agree very well with numerical predic- [10] M. Stratmann and F. Mitschke, Proceedings of the
tions qualitatively, and they are close even in quantitative Conference on Quantum Electronics and Laser Science
terms. Therefore, our experiment clearly corroborates the 2002 paper QThI4 (Optical Society of America,
claim that soliton molecules as described above do indeed Washington, DC, 2002).
exist. While in the context of spatial solitons bound states [11] C. Paré and P.-A. Bélanger, Opt. Commun. 168, 103
have been studied very recently [17], there has been no (1999).
analogon for temporal solitons so far. It remains to be seen [12] A. Schwache and F. Mitschke, Phys. Rev. E 55, 7720
(1997).
whether the Bose-Einstein condensate analogon will also
[13] Ph. Grelu, F. Belhache, F. Gutty, and J.-M. Soto-Crespo,
be found. We here described the first experimental dem- Opt. Lett. 27, 966 (2002).
onstration of the existence of temporal optical soliton [14] D. Y. Tang et al., Phys. Rev. A 64, 033814 (2001).
molecules, and pointed out their possible relevance for [15] A. Maruta, T. Inoue, Y. Nonaka, and Y. Yoshika, IEEE J.
telecommunications technology. Sel. Top. Quantum Electron. 8, 640 (2002).
This work was financially supported in the framework of [16] OFS-Fitel Denmark I/S, Brøndby (www.ofs.dk).
DIP 6.6 (Deutsch-Israelische Projektpartnerschaft). We [17] G. I. Stegeman and M. Segev, Science 286, 1518 (1999).

143902-4