Artificial Neural Network Symbol Estimator With Enhanced Robustness to Nonlinear Phase Noise

Abstract

This letter reports a novel approach for nonlinear phase noise mitigation, based on artificial neural networks (ANNs) tailored to classification applications and a pre-processing stage of feature engineering. Starting with a set of proof-of-concept simulations, we verify that the proposed system can achieve optimal performance for the additive white Gaussian noise (AWGN) channel. Then, considering a dispersion-less channel with strong nonlinear phase noise (NLPN) distortion, we demonstrate a Q-factor increase of 0.4dB, comparing with standard carrier-phase estimation (CPE) followed by minimum distance detection. Finally, simulating the propagation of 64Gbaud PM-16QAM over standard single mode fiber (SSMF), we verify that the ANN-based solution is effective on wavelength-division multiplexing (WDM) transmission conditions, enabling to increase the maximum signal reach by approximately 1 fiber span over the legacy CPE-enabled NLPN compensation.