On the Use of Feedforward Neural Networks to Improve the Intercore Crosstalk Tolerance in Self-Coherent MCF Systems

Abstract

An artificial neural network is investigated to improve the performance of self-coherent weakly-coupled multicore fibre (WC-MCF) systems. Particularly, a feedforward neural network (FNN) is proposed to mitigate the performance degradation induced by the random variation of the intercore crosstalk along time in 64 Gbaud quadrature amplitude modulation WC-MCF systems. A product between the intercore skew and the symbol rate much lower than one and a self-coherent receiver based on Kramers-Kronig technique, are considered. Compared with the reference system without neural networks, an improvement of the tolerable ICXT level close to 12 dB is achieved with the proposed shallow FNN.