Research On Equalization Algorithm Of Visible Light Communication Based On Deep Learning

With the rapid development of many emerging services with bandwidth consumption and higher requirements for the existing communication rate,visible light communication technology has become an indispensable part of 6G.Visible light communication(VLC)technology,which uses LED light as the carrier and transmits in the 400-800 THz band,is expected to solve the problems of spectrum scarcity and high-speed transmission.It can meet the requirements of emerging services for high-speed transmission.However,the nonlinear damage in visible light communication system seriously restricts the further improvement of transmission rate of VLC system.Compensating the nonlinear damage of the signal in the visible light system by software pre-equalization or post equalization can be used as an auxiliary means to reduce the nonlinear impact of the visible light signal and improve the transmission rate of the system without changing the composition of the system.In recent years,deep learning algorithm as a nonlinear compensation means has been applied to the digital signal processing of VLC system to reduce the nonlinear influence.This thesis will also carry out theoretical and experimental research on how to use the existing deep learning technology to design a better neural network equalizer.The research points of this thesis are as follows:1.A regression equalization algorithm based on U-Net neural network is proposed to directly compensate the nonlinear of signal from time domain to time domain.Firstly,the source of inspiration and the principle of U-Net neural network equalizer are introduced.Then,the influence of loss function,input size and network layers on the bit error rate(BER)of the model is studied by using the control variable method,and the better value under the experimental conditions is given.By adding the techniques of residual connection and jump connection to construct deep neural network model to verify the influence of the model BER.By comparing the BER effect of pseudo-random sequence and random sequence with bandwidth,the memory problem of this neural network is further evaluated.Finally,by comparing the BER under different working conditions with the traditional equalization algorithms CMA,LMS + Volterra and some existing deep learning algorithms LSTM and CNN,the experiments show that the proposed equalizer based on U-Net neural network have the best equalization effect.2.A classification and equalization algorithm of generative supported networks(GSN)based on the idea of "mutual assistance" is proposed to compensate the nonlinear distortion of signals.The neural network consists of two parts,auxiliary classifier and classifier.The function of auxiliary classifier is to pre compensate the distorted signal,so that the classifier can realize the compensation of visible light signal by classification under the relatively simple neural network composition.Through experiments,the effects of different training strategies on GSN neural network equalizer are verified.In order to verify the memory problem of GSN neural network equalizer,we take the way of compared experiment that can be divided into two group,one is random sequence,other is pseudo random sequence,to further analyze the memory problem of the equalizer for input sequence.The effect of GSN neural network equalizer on signal equalization is further verified by visualizing the equalization effect such as eye diagram and symbol error diagram.Experiments show that GSN neural network equalizer has better nonlinear compensation performance than traditional non deep learning equalization algorithm.In short,based on the deep learning method,this thesis proposes two different neural network equalization algorithms,namely U-Net neural network equalizer based on regression and GSN neural network equalizer based on classification.Compared with the traditional equalization algorithm,they have stronger ability to solve nonlinear problems and better nonlinear equalization performance.It has an important impact on further improving the transmission rate of VLC system.