Multi-Symbol Output Long Short-Term Memory Neural Network Equalizer For 200+ Gbps IM/DD System

Abstract

We propose a single-lane 212Gbps IM/DD PAM-4 system with a novel Multi-Symbol Output LSTM equalizer that performs much better than FFE&VNE and single-symbol output LSTM, and reduces complexity by 49.85% at the same time, and achieves similar performance with Bi-directional LSTM with around 1/4 complexity.