Research And Application Of Nonlinear Compensation Model For Optical Wireless Based On Gaussian Process Deep Learning

Visible Light Communication(VLC)technology is a potential scheme of the last kilometer high-speed access technology in the future.Due to the rapid development of mobile Internet,Internet of Things,artificial intelligence,cloud native,virtual reality and other technologies,the global wireless data traffic is increasing exponentially.The development of 6G technology is an effective way to improve the capacity of communication terminals.However,the available radio frequency spectrum resources have been crowded to saturation,unable to support the next generation of communication,VLC technology came into being,with the carrier capacity of 380～750THz to provide a new technical solution for high-speed wireless communication.The nonlinear loss of VLC seriously affects the system performance,and machine learning and deep learning are novel digital signal processing methods to solve nonlinear problems.This thesis studies nonlinear compensation of VLC system based on machine learning and deep learning,and proposes two equalization models based on Gaussian process technology:1.A machine learning nonlinear compensation method based on gaussian feature of time domain signal is proposed.Firstly,we study the distribution of each feature of specific sequence data Pattern generated by VLC system under the influence of nonlinear factors and additive White Gaussian noise(AWGN).For the local features of complex background noise,the computation is large,so the noise distribution of digital signal features is introduced as the input of the learning model,and the nonlinear features are mined from the signal data itself.Secondly,gaussian distribution and probability statistics are used to quantify the noise between patterns to obtain prior knowledge of nonlinear signal data.Finally,the nonlinear compensator of VLC system is designed by using naive Bayes machine learning algorithm to compensate the signal and improve the transmission rate of the system.2.An end-to-end compensation model for gaussian process based on depth is proposed.Firstly,considering the influence of nonlinear factors and AWGN in THz VLC system,it is assumed that any number of signal data at the receiving end conform to random process,and the finite set of each data signal data conform to normal distribution.Then,the multilayer gaussian process is used to remember the historical data.The deep Gaussian process can describe the hidden details of the nonlinear data of THz VLC system and fit the linear and nonlinear functions.At the same time,we consider the nonlinear system as an end-to-end equalization scheme,which does not deal with a certain part of the nonlinear system,but takes the system as a black box for input and output processing.Finally,the classification prediction value of the signal is given through the depth Gaussian process,and the concrete realization form is given,and the uncertainty of prediction is measured to improve the bit error rate performance of the system.At the same time,in order to verify the equalization effect of the two models proposed in this thesis,we set up an experimental platform to verify the experimental results.The experimental results show that the equalization model proposed in this thesis has obvious compensation effect on nonlinear VLC system.