High-speed PAM4 transmission with a GeSi electro-absorption modulator and dual-path neural-network-based equalization.

Abstract

Equalization based on artificial neural networks (NN) has proved to be an effective way for nonlinearity mitigation in various kinds of optical communication systems. In this Letter, we propose a novel methodology of dual-path neural network (DP-NN)-based equalization. By combining a linear equalizer with an input-pruned NN equalizer, DP-NN can effectively reduce the computation cost compared to a conventional NN equalizer. We confirm its feasibility through 4-ary pulse amplitude modulation (PAM4) transmission at a gross(net) bitrate of 160 Gb/s (133.3 Gb/s), based on a GeSi electro-absorption modulator operating at C-band. After a 2 km transmission, the bit error rate is below the 20% hard-decision forward-error-correction threshold of 1.5×10-2 with the DP-NN equalization, which outperforms the Volterra equalization and is comparable to conventional NN-based equalization.