Research On Residual Neural Network Equalization For OAM Mode Division Multiplexing Optical Fiber Communication System

With the continuous increase of network users and exponential growth of network data,optical fiber communication systems are required to have larger transmission capacity and higher transmission rate.Orbital angular momentum mode division multiplexing is an important technology to improve the transmission capacity of optical fiber communication system because of its superior orthogonality.However,because the optical fiber communication system with orbital angular momentum mode division multiplexing uses more complex and more kinds of photoelectric devices,the system has more nonlinear interference and thus affects the transmission ability.Therefore,the nonlinear equalization research of optical fiber communication system with orbital angular momentum mode division multiplexing is a hot topic at home and abroad.In this thesis,the nonlinear equalization method of orbital angular momentum mode division multiplexing optical fiber communication system is studied based on three characteristic requirements of low bit error rate,low transmission power and high transmission mode to improve compensation effect and low complexity equalizer.The main research work of this thesis is as follows:(1)In order to reduce the bit error rate of orbital angular momentum mode division multiplexing optical fiber communication system,a nonlinear equalization method based on residual neural network is proposed.This equalization method uses residual learning to increase the depth of the network model while maintaining the continuous learning of the network,so as to better simulate and compensate the nonlinear effects of the system.Experimental results show that the compensation performance of residual neural network equalizer is better than other equalizers,and the bit error rate reaches-4.63dB.At the same time,the parameters and structure of the residual neural network equalizer are studied experimentally,and the performance of the equalizer under different network models is tested and analyzed.(2)In order to improve the compensation performance of residual neural network equalizer under low transmission power and high transmission mode,equalization method based on data mapping is proposed in this thesis.Three data mapping methods,namely,Kcoefficient normalization,mixed power data and mixed mode data,are proposed to improve the learning efficiency of residual neural network equalizer.The experimental results show that the three data mapping methods can effectively improve the compensation ability of the equalizer,and bring 0.87dB,1.5dB and 2dB bit error gains respectively,Compared with the original residual neural network equalizer.(3)In order to reduce the complexity of residual neural network equalizer,a low complexity equalizer based on computational data cache is proposed in this thesis.The design saves the same calculation result between the adjacent input data by computing cache so as to avoid a large number of repeated calculations by the equalizer.Experimental results show that the proposed scheme can effectively reduce the time complexity of the residual neural network equalizer,and the average time complexity is 0.08,Compared with the original residual neural network equalizer.