Research On Dependent Loss In Polarization Division Multiplexing System Based On Machine Learning

With the increasing development of the Internet and the information society,people have placed great demands on the transmission rate and transmission capacity of communication systems.The emergence of coherent optical communication technology and polarization division multiplexing technology has effectively improved the transmission rate and transmission capacity,but polarization dependent loss(PDL)will seriously damage the signal transmission quality.Digital signal processing(DSP)technology can compensate for some losses,but the problems that traditional DSP algorithms can solve are relatively simple,and it is incompetent for complex calculations.It is difficult to cope with the actual deterioration of the signal after high-speed and long-distance transmission.problem.This paper uses machine learning algorithms in coherent optical communication to address the problem that PDL severely restricts the performance of polarization division multiplexing systems,and proposes methods for quick estimation and compensation of PDL,which improve the measurement accuracy and compensation effect of PDL in polarization division multiplexing systems,achieve PDL damage estimation and compensation in coherent optical communication systems.The main innovations of this paper are as follows:First,to solve the problems that traditional PDL measurement methods require high-precision instruments,exist human error and DSP algorithms have poor measurement accuracy,based on Gaussian Mixture Model(GMM)and Expectation Maximization(EM)Algorithm,a method for quickly estimating PDL in a coherent optical communication DSP is proposed.The method proposed in this paper is verified on the simulation platform of 112 Gbit/s polarization division multiplexing coherent optical fiber transmission system,which not only avoids the errors caused by the accuracy and precision limitations of optical devices in traditional PDL measurement,but also compares with traditional DSP algorithms for estimating PDL reduces the average deviation by 1.4%and improves the measurement accuracy.Second,to solve the problems that traditional DSP algorithms for compensating PDL in coherent optical communication lack intelligent learning ability and poor performance,a method of PDL compensation in coherent optical communication DSP based on Deep neural network(DNN)is proposed.On the basis of 112Gbit/s polarization division multiplexing coherent optical communication simulation platform,the verification of the scheme is completed.The results show that the method can effectively compensate the signal damage caused by PDL,greatly reduce the binary error rate(BER),increase the maximum transmission distance by about 200km,and gain the error vector magnitude(EVM)by about 3.8 dB.This method can adapt to link scenarios with different transmission distances and different PDL strengths.Compared with the traditional DSP algorithm,the EVM gain is improved by 0.8dB.