Research On Underwater Visible Light Signal Detection And Channel Estimation Method Based On Machine Learning

The exploitation of the oceans has become more and more important in the 21st century.Underwater visible light communication with high bandwidth,low delay,and good confidentiality has become a research hotspot in underwater scene of short-distance high speed communication.However,the complex channel environment and the nonlinear effect caused by devices in underwater visible light communication bring many difficulties to the signal processing at the receiving end.For the traditional judgment method,the signal with serious amplitude and phase distortion cannot be correctly judged when the Euclidean distance is used as the threshold.Traditional estimation algorithms are difficult to obtain accurate channel state information for subsequent compensation.At present,some scholars have solved the nonlinear distortion problem in visible light communication with the strong fitting ability of machine learning.This dissertation uses machine learning technology for signal detection and channel estimation based on underwater DCO-OFDM system.The main contents of this dissertation are as follows:(1)Firstly,the basic principle of DCO-OFDM system is introduced.Secondly,the effects of devices,clipping and underwater media on system performance are analyzed respectively,and the underwater channel model is established according to the Bill Lambert radiation model.(2)Research on underwater visible light signal detection method based on machine learning.In underwater communication scenarios with various media,the decision accuracy of support vector machine is low,a signal detection algorithm based on Back Propagation Neural Network is presented in this dissertation.The algorithm takes the amplitude and phase of the received signal as the input characteristics of the network,and by giving multidimensional labels to the training data,it can directly recover multi-bit original binary signals through the network in parallel.The bit error rate of the algorithm was simulated and analyzed in different water environment.The performance of the algorithm under different communication link distances,different modulation orders and different turbidity was tested by hardware platform.The results show that the proposed algorithm can obtain higher decision accuracy than SVM.(3)Research on underwater visible light channel estimation method based on machine learning.The estimation of LS algorithm is not high in underwater multipath channel environment,and the neural network cannot model the noise signal,a BPNN channel estimation algorithm based on time domain denoising is proposed in this dissertation.Firstly,the channel frequency domain response at the effective subcarrier is estimated by LS algorithm.Secondly,the channel frequency domain response of all subcarriers is obtained through Hermitian symmetric transformation.Thirdly,perform secondary denoising in the time domain.Finally it is sent to the network for iterative training.In this dissertation,the algorithm is simulated with different number of underwater channels paths.The performance of the algorithm was tested at different distances,different modulation orders and different turbidity of water based on the hardware platform,and the feasibility of its application in the estimation of underwater DCO-OFDM visible light communication system is verified.Experimental results show that BPNN algorithm can obtain better estimation performance than traditional LS algorithm.