Comparison of Algorithms Used for Training of Artificial Neural Network Based Nonlinear Equalizer in a Coherent Optical Orthogonal Frequency Division Multiplexing System

Abstract

Artificial neural network based nonlinear equalizers (ANN-NLEs) are one of the efficient techniques of nonlinearities mitigation in coherent optical orthogonal frequency division multiplexing (CO-OFDM) systems. Until now, many techniques have been used for the performance enhancement of a CO-OFDM system. Despite this, no one has studied different training algo-rithms of ANN in ANN-NLE. This paper compares training algorithms under five categories i.e., Gradient Descent, Resilient Backpropagation, Conjugate Gradient, Quasi-Newton and Levenberg Marquardt to facilitate the choice of the optimum training algorithm for nonlinearities mitigation. The comparison is shown on the basis of different parameters i.e., Q-Factor, OSNR, and BER.