The Research Of Key Technical Issues Of Underwater Wireless Optical Communication Based On End-to-End Deep Learning

Underwater Wireless Optical Communication(UWOC)technology has the advantages of low delay,high speed,large bandwidth and low cost,which has broad application prospects in short-range underwater communication.However,the underwater optical channel will affect the transmission optical signal of UWOC,such as absorption and multiple scattering,which causes the problem of inter-symbol inference.In addition,due to the variable and changeable characteristics of underwater optical channels,traditional channel modeling methods are difficult to model the UWOC channel quickly and accurately,which brings new challenges to the design of reliable UWOC systems.Data-driven based deep learning technology has been proven to be applicable to wireless communication systems,and can optimize the system.In this paper,the end-to-end deep neural network model of deep learning technology is applied to the UWOC system,which can learn the optimal communication parameters by training the channel impulse response matrix in different seawater types.The main work of this paper is as follows: First,design a new end-to-end neural network architecture CNN-AE,which combines UWOC-oriented Convolution Neural Network(CNN)and Autoencoder(AE)neural network.Basing on the one-dimensional convolution kernel operation,it can improve the feature extraction and generalization ability of traditional AE neural network for one-dimensional input sequence.This architecture has the characteristics of strong generalization,strong feature learning ability and fast training convergence characteristics of CNN and the characteristics of end-to-end joint optimization of AE neural network;secondly,an on-off keying modulation scheme based on adjustable soft binarization function is designed to prevent the problem of gradient disappearance in the CNN-AE model,and can also realize the transmission of optical signals power constrains and the non-negativity of signal;thirdly,for the time-varying underwater optical channel,the corresponding channel impulse responses of optical signals transmitted in different time slots are also different.Considering the optical signal is transmitted in different time slots in the on-off keying modulation and the inter-symbol inference effect in the underwater optical channel,this paper designs the corresponding channel impulse response matrix based on the Monte Carlo simulation calculation method according to the type of seawater,the length of the optical signal bit sequence,and the data transmission rate.In this paper,the basic function and underwater optical communication simulation test of the UWOC system based on the CNN-AE neural network is carried out,and the communication performance of the CNN-AE neural network is verified in the basic function simulation test.The experimental results show that it meets the communication requirements.In the underwater optical communication simulation test,the experimental results show that the CNN-AE neural network has the lowest bit error rate under the conditions of pure seawater,coding rate of 50% and data transmission rate of 1 Gpbs.In addition,compared with the communication performance of other UWOC systems,the UWOC system based on CNN-AE neural network designed in this paper can achieve a lower bit error rate in any range of signal-to-noise ratio under the same conditions.