466 CHAPTER 9. SOLITON SYSTEMS
somewhat but does not solve the problem. However, if sliding-frequency filters are
employed, the temporal shift is reduced to below 15 ps for copolarized solitons (curve
3) and to below 10 ps for orthogonally polarized solitons (curve 4). In these numerical
simulations, the map period and amplifier spacing are equal to 40 km. The dispersion
map consists of 36 km of anomalous-GVD fiber and 4 km of DCF (β 2 n≈130 ps^2 /km)
such that average value of dispersion is 0.1 ps/(km-nm).
Many issues need to be addressed in designing WDM systems. These include the
SNR degradation because of the accumulation of ASE, timing jitter induced by ASE
and other sources (acoustic waves, ASE, Raman-induced frequency shift, PMD, etc.),
XPM-induced intrachannel interactions, and interchannel collisions. Most of them de-
pend on the choice of the map period, local value of the GVD in each fiber section, and
average dispersion of the entire link. The choice of loss-management scheme (lumped
versus distributed amplification) also impacts the system performance. These issues
are common to all WDM systems and can only be addressed by solving the underlying
NLS equation numerically (see Appendix E).
Figure 9.29 shows the role played by local dispersion by comparing the maximum
transmission distances for the lumped (EDFA) and distributed (Raman) amplification
schemes [248]. The dispersion map consists of a long fiber section with GVD in the
range 2–17 ps/(km-nm) and a short fiber section with the dispersion of−25 ps/(km-nm)
whose length is chosen to yield an average GVD of 0.04 ps/(km-nm). The map period
and amplifier spacing are equal to 50 km. Pulse parameters at a bit rate of 40 Gb/s cor-
respond to the periodic propagation of DM solitons. The curves marked “interactions”
provide the distance at which solitons have shifted by 30% of the 25-ps bit slot because
of intrachannel pulse-to-pulse interactions. The curves marked “XPM” denote the lim-
iting distance set by the XPM-induced interchannel interactions. The curves marked
“PIM” show the improvement realized by polarization-interleaved multiplexing (PIM)
of WDM channels. The two circles marked A and B denote the optimum values of
local GVD in the cases of lumped and distributed amplification, respectively. Several
points are noteworthy in Fig. 9.29. First, Raman amplification improves the transmis-
sion distance from the standpoint of intrachannel interactions but has a negative impact
when interchannel collisions are considered. Second, polarization multiplexing helps
for both lumped and distributed amplification. Third, the local value of GVD plays an
important role and its optimum value is different for lumped and distributed amplifi-
cation. The main conclusion is that numerical simulations are essential for optimizing
any WDM system.
On the experimental side, 16 channels at 20 Gb/s were transmitted in 1997 over
1300 km of standard fiber with a map period of 100 km using a DCF that compen-
sated partially both GVD and its slope [251]. In a 1998 experiment, 20 channels at
20 Gb/s were transmitted over 2000 km using dispersion-flattened fiber with a channel
spacing of 0.8 nm [252]. A capacity of 640 Gb/s was realized in a 2000 experiment
in which 16 channels at 40 Gb/s were transmitted over 1000 km [257]. In a later ex-
periment, the system capacity was extended to 1 Tb/s by transmitting 25 channels at
40 Gb/s over 1500 km with 100-GHz channel spacing [261]. The 250-km recirculat-
ing fiber loop employed a dispersion map with the 50-km map period and a relatively
low value of average dispersion for all channels. Dispersion slope (TOD) was nearly
compensated and had a value of less than 0.005 ps/(km-nm^2 ). A BER of 10−^9 could