IV. MODELING FIBER NONLINEARITIES
While numerous models exist for fiber nonlinearities,
we seek a model that captures the salient features of the nonlinear
impairments, but is simple enough to permit complex
network studies. As in [4] we turn to the Gaussian noise
model [5] that has recently been shown to be accurate for
uncompensated links using digital coherent transceivers [6] as
is likely to occur in future nonlinear elastic optical networks.
Using the Gaussian noise model it can be shown [7], [8] that
if the total available spectrum B is modulated1 over a single
span having an attenuation coefficient of α and effective length
1/α, the optimum power spectral density (Ssig) is given by2
Ssig = 3
__
_
_
27π|β2|αSASE
16γ 2 ln _2B2π2|β2|
3α
_ (1)
where β2 is the dispersion coefficient and SASE = 2nsphν
(G − 1) is the power spectral density of the amplified spontaneous
emission (ASE) noise (where G is the amplifier gain,
hν is the photon energy and nsp is the population inversion
factor). Assuming incoherent addition of the noise over N
identical spans the resulting SNR is
SNR =
2Ssig
3NSASE
(2)
Rather than deal in the abstract let us consider a specific fiber
type such as standard single mode fiber with an attenuation
of 0.22 dB/km, nonlinear coefficient γ = 1.3 W−1km−1
and chromatic dispersion of 16.7 ps/nm/km. Assuming the
amplifiers have a noise figure of 5 dB with the span length
between amplifiers being 100 km then 27 mW/THz is the
optimal power density over the 5 THz bandwidth. As such the
fiber nonlinearities can be accounted for by limiting the power
spectral density to 27 mW/THz. Given that the SNR after the
first span is 24.5 dB after N spans the SNR in decibels is
SNRdB = 24.5 − 10 log10 (N) (3)
V. OPTIMAL MODULATION FORMAT
Shannon gives a relationship between the spectral efficiency
and the linear SNR such that for a polarization multiplexed
format the net spectral efficiency (NSE) is given by [10]
NSE = 2 log2 (1 + SNR) (4)
Since the Shannon limit does not indicate the modulation
format or the FEC coding overhead that should be employed,
we seek an alternative approximate bound as to what might
be realizable in practice. In order to consider this for polarization
division multiplexed quadrature amplitude modulation
(PDM-QAM) constellations we determine using analytical
expressions from [11] combined with direct simulation of the
performance in the presence of additive white Gaussian noise
1This is based on the assumption that in the link where blocking first occurs
the spectrum will be sufficiently close to fully utilized. As such the nonlinear
interference noise density corresponds to B ≈ 5 THz, similar to the local
optimum global optimum Nyquist (LOGON) strategy proposed in [8].
2This follows from [9] with an asymptotically large total modulated bandwidth
B ≈ 5 THz, with the peak power spectral density converging to the
spectrally averaged integrated nonlinear noise power indicating the validity of
the white noise approximation. Replacing the effective length by 1/α gives
an error of ≤ 0.05 dB in the launch power for span lengths of 80 km or more.
IEEE PHOTONICS TECHNOLOGY LETTERS, VOL. 26, NO. 10, MAY 15, 2014