Optimization of the Capacity of a Fiber Communication Line with Nonlinear Memory

Channels with and without memory are compared using the simplest Gaussian noise distribution model. It is shown that both channels have similar dependences of the capacity on the average signal power. At the same time, the optimal input alphabet for the channel with memory changes. The applicability of the model with nonlinear memory to the QPSK communication line with a periodic dispersion compensation is demonstrated numerically.