A Comparative Study on Global Compensation Techniques for Coherent Optical Communications

This paper presents a comparative study on global compensation methods for coherent optical communications. Specifically, it compares an approach based on statistical signal processing to a parametric network that combines the model knowledge with Machine/Deep Learning techniques. Simulation results show the effectiveness of both techniques in the presence of multiple linear effects like IQ imbalance, carrier frequency offset, and laser phase noise. Furthermore, these results show that the statistical signal processing approach benefits from a lower computational cost, while the parametric network is more flexible and can be easily adapted to system requirements.