Perturbative Machine Learning Technique for Nonlinear Impairments Compensation in WDM Systems

We propose a perturbation-based receiver-side machine-learning equalizer for inter- and intra-channel nonlinearity compensation in WDM systems. We show 1.6 dB and 0.6 dB $Q^{2}$ -factor improvement compared with linear equalization and DBP respectively for 1000km transmission of $3\times 128Gbit/s$ DP-16QAM signal.