Research On Signal Visible Light Channel Modeling And Signal Processing Technology Based On Machine Learning

Visible Light Communication technology refers to the transmission of data through visible light generated by diodes as a signal medium.It has a series of characteristics such as no electromagnetic radiation,environmental protection and safety.In recent years,high-speed visible light communication technology has become the focus of research and development in the wireless communication industry,and its light source has gradually been replaced by LEDs.In indoor VLC,accurate channel estimation is helpful for coherent demodulation and interference cancellation of receivers,but communication systems are generally accompanied by nonlinear factors.By reproducing these problems from a deep learning(DL)perspective,communication systems can be directly optimized based on data.In communication systems,before signal detection,and after the receiver receives the transmitted information,it is usually necessary to undergo a filter process.These filters can typically include low-pass filters,matched filters,and equalizers,and then be handed over to the decision maker.After sampling,the decision maker performs signal detection and decision functions.As the communication rate increases,the probability of code errors occurring in messages increases,making it difficult for the decision circuit to accurately determine them.However,reinforcement learning can adapt to the specific conditions of various channels and accurately determine them through repeated training,After determining a sequence of information,error correction can be accomplished through a Recurrent Neural Network(RNN).The research content of this paper is experimental research based on machine learning in visible light communication channel modeling,signal decision,error correction,and other aspects.The main innovation points are as follows:(1)Recurrent neural network technologies such as Long Short Term Memory,Gated Recurrent Unit,and Sparse Auto Encoder-s can simulate various frequency band information for channel modeling.In this paper,an RGB LED multi wavelength visible light communication experimental platform has been built,and relevant experimental tests have been completed.The experimental results show that compared with real channels,the SAEs algorithm has the best channel fitting effect,with a mean square error in a stable range of about 0.000003.In order to draw more conclusions,experimental designs have been made for channel scenarios with different communication distances,different signal processing methods,and different light sources interacting,The error is within a reasonable range.(2)Design and implement a reception decision algorithm for RGB LED multi wavelength visible light communication.Using the data collected from the above experimental platform,use the Q-learning algorithm to perform signal decision on the data from the receiver,and finally obtain a digital signal.The experiment uses different data distributions for decision making,with a decision error of around 0.0108.And explore the impact of different communication distances,large and small spacing,different state division methods,and different reinforcement learning rates on the conclusion.(3)A bit error correction scheme for RGB LED multi wavelength visible light communication is proposed and implemented,verifying the possibility of bit error correction using RNN algorithm.Relevant tests on bit error correction are conducted for digital signals of 0 and 1,and evaluating the different effects of bit error correction under different environments of large and small noise.The effect of bit error correction is good for small noise.And the difference in error correction results in the case of interference with data preprocessing.