Research On Signal Compensation Technology Based On Photonic Reservoir Computing

With the rapid development of the information age,various applications of the Internet have become the basic needs of people in their daily lives and work.The surge in activities such as telecommuting,online conferences,and video lectures has led to the rapid growth of network traffic data.The optical network prediction report released in 2022 indicates that the future optical network will develop towards a faster rate and greater capacity to meet people’s needs.These requirements will be the mainstream direction and technical challenge for the future development of optical fiber communication.In order to meet this demand,through theoretical analysis,the problem of signal damage in optical fiber communication is one of the main issues limiting the development of this technology.The main goal of this article is to balance the problems caused by linear and nonlinear effects in optical fiber communication and improve the system’s processing ability for these effects through equalization technology.Currently,there are many solutions proposed by scientific researchers to solve such problems.Such as digital back propagation,optical phase conjugation and machine learning.However,these technologies have some application defects and theoretical deficiencies.Therefore,this paper uses a photonic reservoir computing model based on optical domain to achieve signal equalization for optical fiber communication.The photonic reservoir computing does not need to rely on channel parameters such as nonlinear coefficient,dispersion coefficient,and loss coefficient in the channel for equalizing signals.Photonic reservoir computing can avoid the high computational load in theoretical balancing technology and optimize the redundant training amount in the model.In addition,the photonic reservoir computing also has the advantages of overcoming power limitations of electronic devices,higher computing rates,low latency,and low loss.Firstly,this paper builds an optical fiber communication system.Analyze the theoretical basis of dispersion and nonlinear effects and establish corresponding mathematical models.Through this system,the signals affected by dispersion and nonlinear effects are obtained as training and verification data sets for the photonic reservoir computing.By using the photonic reservoir computing to balance the linear and nonlinear effects,the main work includes:(1)Based on the principle of the photonic reservoir model,we explored the role and significance of time mask sequences previously used in the model input layer.We explained how mask sequences improves dimensions and increase spatial linear separability.Then photonic reservoir computing was built and the effectiveness of the model was preliminarily verified.(2)We explored the role of the virtual node number N of the hidden layer in the photonic reservoir computing.Analyze and simulate the impact on signal equalization when the number of virtual nodes N is different.Through simulation analysis,it is found that the increase of the number of virtual nodes N will make the BER of the system lower and lower.When N is more than 500,BER remains at a low level and tends to be stable.(3)In the photonic reservoir computing,the output layer is the main training component.Therefore,this article analyzes and compares three training methods:the Bayesian regulation algorithm,the scaled conversion gradient algorithm,and the Levenberg-Marquardt algorithm.Finally,through comparative analysis,we found that when the number of virtual nodes N in the reservoir is greater than 500,the L-M method has obvious advantages.When N is more than 600,the BER obtained by L-M method is reduced by more than 80%compared to the other two methods.