Optical Network OSNR Estimation And Soft-failures Link Location Based On Neural Network

The development of the Internet has driven up the demand for network services.As an important infrastructure of the Internet,optical networks are showing more and more flexible and efficient development,and the deployment of optical paths and real-time signal quality monitoring in optical networks are becoming more and more complicated.By predicting the signal quality of the optical network not established in the optical network,it can facilitate the construction and dynamic configuration of the optical network.In addition,the real-time signal quality monitoring of the optical path in the optical network,especially the monitoring and positioning of optical path faults,is also the focus of survivability guarantee for high-speed dynamic optical networks.In recent years,the application of artificial intelligence technology in the field of optical networks is endless.This paper will focus on the use of Machine Learning(Machine Learning,ML)algorithm for analysis and research of optical network transmission quality evaluation and fault monitoring in optical networks.The main research contents and main innovations of this article are:1.This paper uses a large number of historical optical path samples and uses neural network algorithm to accurately estimate the optical signal noise ratio(OSNR)of the optical path to be built in the optical network.This paper first considers the influence of the optical path in the optical network by the nonlinear interference of adjacent optical paths and the uncertainty of optical monitoring,and proposes a new model of OSNR estimation based on neural network.Through extensive simulation on NSFNET,research and analysis of multiple neural network model parameters and network parameters,when the allowable estimation error is within 0.5 dB,the neural network estimation optical path OSNR model can obtain an estimation accuracy of more than 98%.2.On the basis of considering the effects of non-linear interference and monitoring uncertainty,further analysis of the uneven gain of Erbium Doped Fiber Amplifier(EDFA),which has not been considered in the existing research,proposes a A new neural network model to estimate the OSNR of the optical path to be built.Through the simulation verification of the model,when the sample size is sufficient,the estimation accuracy of more than 98% can also be obtained.Compared with other methods of optical path signal quality estimation,the OSNR estimation model proposed in this paper takes into account more physical layer factors and no additional equipment and network costs.3.This paper studies the problem of optical path soft fault monitoring in optical networks.Considering the robustness of optical networks,that is,the small number of faulty optical path samples,this paper proposes to select the support vector domain description algorithm in the unsupervised learning algorithm(Support Vector Domain Description,SVDD)to determine whether a link soft fault has occurred in the optical path of the optical network,and the training data is historical normal optical path samples.On this basis,this paper optimizes the kernel function and parameters of the SVDD model,which can improve the feasibility and prediction accuracy of the model.Compared with the fault monitoring methods in other optical networks,the model selected in this paper does not require other hardware components or allocate dedicated monitoring optical paths,and ensures the timeliness and robustness of the optical network.Through extensive simulation on NSFNET,for the problem of monitoring whether the optical path in the dynamic optical network has a soft fault,the fault prediction accuracy of 98% can be obtained.