A Research Of Equalization Technology In Optical Communication Channel Based On Machine Learning

With the development of emerging technology such as cloud computing and big data,the demand of high speed and large capacity optical fiber communication network becomes stronger and stronger.Therefore,how to use the data center to transmit more information with limited resource becomes the key in this day.Compare with other transmitted system,the transmitted system based on Pulse Amplitude Modulation-4(PAM-4)single-side band signal and Dual-Drive Mach-Zehnder modulator(DDMZM)has some advantages in communication.For instance,easy to implemented,simple structure and low power consumption,which is so important in communication system.Therefore,some communication protocol has clearly stipulated the PAM-4 as the modulation format for short reach communication system.Therefore,the transmitted system based on PAM-4 single-side band signal is one of the key methods which can improve the quality of data center.Even through it has so much advantages,nonlinear distortion is still the main effect which limit the performance of transmitted system because of system characteristic.DDMZM-based optical single-side band transmission system suffer from the influence of nonlinear effect,which would degenerate the quality of signal.Ordinary linear filter cannot compensate nonlinear distortion and conventional nonlinear equalizer has some problem such as algorithm complexity is too high and the power of eliminate the nonlinear distortion is limited.Recently,machine learning technology has attracted researchers' attention because of its excellent ability in dealing with nonlinear and abstract problems.Therefore,some researchers have begun to apply machine learning algorithm to communication system.These researchers have achieved partial results,such as using deep learning techniques to design constellations,or using machine learning algorithms to design communication network structures.For the above advantages,this paper attempts to use machine learning algorithm to solve nonlinear problems in short distance transmission system.In our research,we focus on the power of machine learning technology in equalization in nonlinear channel.Firstly,we study the single side band communication system based on dual-drive Mach-Zehnder modulator(DDMZM)and square-law detection by single-ended photodiode(PD)which means it is low cost system.So that it is suitable to the short reach communication system.After that,we introduce the conventional method to mitigate the nonlinear effect and propose a kind of sparse Volterra filter to reduce the computation complexity.Then,we use the neural network to equalize the PAM-4 signal which suffer from the both linear and nonlinear distortion.At last,we propose a kind of nonlinear equalization based on the Kramers-Kronig and complex neural network.The result of experiment shown that this kind of novelty equalizer has better performance than conventional neural network equalization.