Physics-Informed Neural Network for Fibre Channel Modelling in Optical Communication Systems

Over 95% of the data traffic is carried over optical fibre communication links. The split-step Fourier method (SSFM) has been widely employed to model the evolution of optical signals along the fibre channels in optical communication systems. However, the split-step Fourier method requires very high computational resources, especially for ultra-long-haul and wideband communication systems. Meanwhile, deep learning techniques can be applied to investigate the evolution of optical signals along the fibre links, where the nonlinear Schrödinger equation (NLSE) can be solved directly using neural networks to avoid the huge complexity of the split-step Fourier simulations. In this work, we will discuss the application of neural networks in modelling the evolution of different types of optical pulses along fibre transmission channels.