Low-Complexity Pruned Convolutional Neural Network Based Nonlinear Equalizer in Coherent Optical Communication Systems

Nonlinear impairments caused by devices and fiber transmission links in a coherent optical communication system can severely limit its distance achievable capacity. In this paper, we propose low-complexity pruned-convolutional-neural-network-(CNN)-based nonlinear equalizer, to compensate signal for systems. By increasing the size of effective receptive field with an 11 × large convolutional kernel, performance feature extraction CNNs is enhanced structure CNN simplified. And performing channel-level pruning algorithm, prune insignificant channels, complexity model dramatically reduced. These operations could save important component reduce width computation amount. The proposed CNN-based equalizer was experimentally evaluated 120 Gbit/s 64-quadrature-amplitude-modulation (64-QAM) over 375 km standard single-mode (SSMF). experimental results showed that, compared 6 normal after pruning, saved approximately 15.5% space 43.1% time complexity, without degrading equalization performance. pruned-CNN-based has great potential application realistic holds promising prospects