Neural Network-Assisted Self-Coherent MCF Systems Impaired by ICXT and Laser Phase Noise

The performance of 64 Gbaud self-coherent MCF systems assisted by neural networks to mitigate the combined effect of the intercore crosstalk (ICXT) and laser phase noise is assessed by numerical simulation. In particular, the impact of the fast fluctuations of the phase noise on the training phase of the neural network and on the system performance is discussed. For this, two different cases are analysed: one in which the phase noise inside the training phase of the neural network is highly correlated with the phase noise of the transmission scenario, and the other in which the phase noise of the training phase is weakly correlated with that one of the transmission scenario. This is performed for a product between the intercore skew and the symbol rate much lower than one to avoid the need for neural networks with memory. For the case with correlated phase noise, results show an outage probability improvement from 36.5% to 14.5% after applying the neural network. With weakly correlated or uncorrelated phase noise, the neural network is not able to mitigate the combined effect of the ICXT and laser phase noise and no system improvement is observed.