We investigated the effect of the size of an artificial neural network (ANN) on the performance when compensating for linear and nonlinear waveform distortion caused by chromatic dispersion (CD) and self-phase modulation (SPM). The results revealed that larger nonlinear waveform distortion requires more hidden-layer neurons, whereas the required number of input-layer neurons is decided by the amount of CD. The compensation performance of the ANN was evaluated by numerical simulation of 50-km standard single-mode fiber (SSMF) transmission of 16-ary quadrature amplitude modulation (16QAM) signals.
