Low Complexity Neural Network Equalization Based on Multi-Symbol Output Technique for 200+ Gbps IM/DD Short Reach Optical System

• DOI: 10.1109/jlt.2022.3146863
• Publication Date: 2022-01-31T22:16:04Z
• Authors: Bohan Sang; Wen Zhou; Yuxuan Tan; Miao Kong; Chen Wang; Mingxu Wang; Li Zhao; Jiao Zhang; Jianjun Yu
• Keywords: Modulation (music); Symbol (formal)

Nowadays, Neural network (NN) has been proved to be an effective solution for nonlinear equalization in short reach optical systems. However, recent research mainly focused on implementing more powerful NNs equalization, while ignoring their adaptability tasks. In this paper, we propose Multi-Symbol Output (MSO)-Neural Networks high-speed interconnects. The results show that the proposed MSO design works well Deep (DNN), Long Short-Term Memory neural networks (LSTM) and Gate Recurrent Unit (GRU), which are NN-based structures. By increasing output symbols of NNs, number slide windows can sharply reduced, so complexity is reduced. same time, information brought MSO-NNs back-propagations, therefore performance gain achieved. A 212-Gb/s 1-km Pulse Amplitude Modulation (PAM)-4 link experimentally demonstrated as target system, MSO-NN equalizers used compare with traditional algorithms including Volterra Nonlinear Equalizer (VNE) single-symbol NNs. Experimental could help reduce NN required system by around 2/3, MSO-LSTM performs much better than VNE 1 dB SSO-LSTM at time. Based MSO-LSTM, transmission BER under HD-FEC over km NZDSF achieved a ROP -2 dBm. Our work expandable extended other equalizers, learn info from training data performance. further enhance reduces provides assurance future real-time short-range systems, brings new ideas equalizer design.