基于LSTM非线性均衡的64QAM波分模分复用通信相干传输系统  

64-QAM WDM-MDM Coherent Transmission System Based on LSTM Nonlinear Equalization

作　　者：张尧 桑博涵 王晨 龙健宇 王凯辉 周雯 刘博 余建军 Zhang Yao;Sang Bohan;Wang Chen;Long Jianyu;Wang Kaihui;Zhou Wen;Liu Bo;Yu Jianjun(State Key Laboratory of ASIC and System,School of Information Science and Technology,Fudan University,Shanghai 200433,China)

机构地区：[1]复旦大学信息科学与工程学院专用集成电路与系统国家重点实验室,上海200433

出　　处：《中国激光》2024年第17期206-212,共7页Chinese Journal of Lasers

基　　金：国家自然科学基金(61835002,61935005,62305067,62375219,62331004)。

摘　　要：为满足新兴应用对通信能力日益增长的要求,提出一种基于长短时记忆网络(LSTM)的非线性均衡器,并演示一个长距离大容量波分模分复用(MDM-WDM)通信系统。该实验在650 km的少模光纤(FMF)上,实现80通道波分复用双极化48 GBaud 64QAM信号在LP11a、LP11b、LP21a、LP21b四种模式上传输。为减小系统中非线性效应的影响,在接收端数字信号处理(DSP)中,采用基于LSTM的非线性均衡技术,显著提升系统性能。实验结果表明,经650 km的少模光纤传输,系统的比特误码率(BER)能满足25%软判决前项纠错阈值(SD-FEC),净速率达到147.4 Tbit/s,属于国内领先水平。Objective Owing to the rapid growth of Internet traffic,the demand for large transmission capacities has increased significantly.To address the issues of limited device bandwidths and high hardware-update costs,researchers have adopted expansion methods,including wavelength division multiplexing(WDM)and mode division multiplexing(MDM).Optical signal impairments caused by devices and fiber-optic channels render advanced digital signal-processing technology crucial for achieving high-speed fiber-optic communications.Parallel with the emergence and development of deep learning,machine learning-based equalization has been widely adopted in optical communications.Deep neural networks(DNNs),convolutional neural networks(CNNs),and recurrent neural networks(RNNs)have been applied for the nonlinear compensation of fiber-optic communication systems.In recent years,nonlinear equalizers(NLE)based on RNNs have been investigated extensively.Long short-term memory(LSTM)neural networks,as a type of RNN,can solve the issues of gradient disappearance and explosion during training.In this study,WDM,MDM,polarization multiplexing,and advanced digital-signal processing technology were used to construct a homodyne coherent transmission system based on an LSTM neural network equalizer(LSTM-NNE).We successfully equalized 80 channels of 48 GBaud 64-quadrature amplitude modulation(QAM)signals after transmitting 650 km of few-mode fibers(FMFs)on four modes:LP11a,LP11b,LP21a,and LP21b.The bit error rate(BER)of the MDM-WDM system can satisfy the 25%soft-decision forward-error correction(SD-FEC)threshold of 4×10^(-2).Methods At the transmitter end,50 GHz spaced 80-channel signals distributed at 1529.9-1561.4 nm are generated.The channels are separated into two groups,where each group used 40 external cavity lasers with 100 GHz spacing(corresponding to C20-C59 and H20-H59 in the ITU standards),and the optical carriers are coupled by two polarization-maintaining arrayed waveguide gratings(PM-AWGs).The coupled carriers are modulated using two in

关 键 词：光纤通信 模分复用 波分复用 少模光纤 神经网络均衡 

分 类 号：O436[机械工程—光学工程]