Modeling Of Visible Light Communication Channel Based On Deep Learning

In the past ten years,the number of mobile devices has increased significantly,wireless terminals can be seen everywhere in life,wireless services have shown an exponential growth,and there is a huge demand for radio frequency-based technologies.The arrival of the Internet of Things era will allow every wireless device to have the ability to process information,which also poses a huge challenge to radio resources.At the same time,great changes have taken place in the lighting industry due to the popularization of LEDs,bringing the dual advantages of economy and efficiency to the society.Under this background,visible light communication has become a potential technology to make up for the shortage of spectrum resources in the future due to its advantages of high speed,green pollution-free and license-free bandwidth.For any communication system,an accurate channel model is essential for the study of the whole system.The traditional numerical modeling method in the visible light communication system has complex parameters,strict requirements on the number of reflections and spatial conditions,and requires a lot of calculations,so the model is not easy to update.The neural network in deep learning technology has the ability to fit any nonlinear function due to its unique structure,and it is also easy to update when the environment or related variables change.The established model is differentiable and has strong flexibility.This paper focuses on the channel modeling problem of visible light communication system,and carries out related research with the goal of visible light communication channel modeling based on deep learning technology.The training model provides a more convenient and feasible idea for the direction of visible light channel modeling.The main work and innovations are as follows:1.The composition and principle of the indoor visible light communication(VLC)system are described in detail,and the characteristics of the indoor visible light channel are discussed.Combined with the principles and characteristics of machine learning,the current application of machine learning in the visible light field is analyzed.In terms of visible light channel,the frequency response and modulation bandwidth of LEDs related to the channel are expounded,and the emphasis is on fully investigating and analyzing the principles and methods of traditional indoor visible light channel modeling and some current work of machine learning in visible light channel modeling.2.On the basis of the above-mentioned modeling and analysis of the visible light channel,a real visible light communication link platform was built with commercial components,and the time-domain waveform data at both ends of the channel was collected and processed;Based on the time domain visible light channel model of Bi-directional Long Short-Term Memory(BiLSTM)algorithm.Compared with the traditional numerical modeling method,the channel modeling method in this work can provide a channel modeling solution that consumes less spectrum resources,has no requirements on the environment,and is easy to update.after comparison and error analysis,the Mean Absolute Error(MAE)of the proposed model is stable between 0.006 and 0.013,and the error of 80%of the test points is below 0.016,which proves that the pure data-driven visible light channel model is built.It can better fit the real channel.