Optical Performance Monitoring Of High Speed Optical Signal Based On Machine Learning

Nowadays,with the rapid development of optical fiber communication network technology and the increasing transmission rate,the structure of communication network is more complex.At the same time,the problems of various optical signal impairments have followed,which affects the reliability and stability of optical network.As one of the key technical means to ensure the reliable and stable operation of optical networks,optical performance monitoring(OPM)plays a more and more important role.OPM technology can get the network status and signal transmission status in real time by detecting the performance parameters of optical network transmission link or optical network node,so as to timely handle and ensure normal network operation and signal transmission.There-fore,how to accurately monitor the parameters such as chromatic dispersion(CD),polarization mode dispersion(PMD)and optical signal-to-noise ratio(OSNR)is very im-portant for ensuring the performance of signal transmission and optimizing the allocation and management of network resources.Around parameter monitoring in nonlinear chan-nel environment,the OSNR and CD monitoring methods based on machine learning are studied in this paper and the CD and OSNR monitoring of 16QAM communication sys-tem is realized by using generalized regression neural network(GRNN)and back propagation(BP)neural network.The main research work is as follows:1.This paper describes the main research contents of OPM technology and the cur-rent research status,shortages and directions of OPM technology under the new network development trend.Then the principle of machine learning,neural network and its appli-cation in OPM are introduced.2.Combined with the general method of machine learning,a OPM model based on machine learning is proposed,which is expansibility.This model includes the simulation analysis,feature extraction and machine learning processing of communication system data.In the process of data feature extraction,the characteristic values of OSNR are de-fined by the method of high order statistical moments,and the characteristic values of the dispersion are defined by the method of density ratio.3.On the basis of data that had been extracted,the monitoring of OSNR and CD is realized by using BP neural network and GRNN respectively.Through the comparison of two kinds of methods,it is found that the training process of BP network,a long time,not easy parameter optimization,unable to efficiently process large amounts of data.Contra-rily,the monitoring results of GRNN stability and the algorithm is more simple and efficient,so this article mainly selects GRNN for OSNR and CD monitoring.The moni-toring results show that the comprehensive effect of CD on monitoring is more serious than that of OSNR.With the increase of the actual link CD,the average error of the mon-itoring results also increased.In the end,the single parameter monitoring of 40 Gb/s and 100 Gb/s systems is compared,and the increase of the monitoring error after the increase of rate is obtained.In addition,the monitoring error,range and accuracy are related to the accuracy of data extraction,mainly because the increase of CD in the process of feature extraction will make the region of data extraction unstable and affect the final result.