Research On Intelligent Channel Estimation And Equalization In Visible Light Communication System

Visible light communication,as a new spectrum resource,will be combined with other communication methods to shine in the future 6G communication.However,visible light communication has nonlinear distortion problems caused by various aspects,including LED nonlinear effect,background light noise,inter-symbol interference caused by multiple reflections,and nonlinear loss of each part of the device,etc.These problems cause nonlinear distortion of the received signal,and bring difficulties to the accurate decision of the signal.Therefore,it is necessary to use channel estimation and equalization technology to compensate the loss.Because of the complexity of visible light channel,the intelligent tool machine learning algorithm can better fit the nonlinear channel than the traditional algorithm and achieve better results.In this paper,indoor visible light channel estimation and equalization,modulation using ACO-OFDM system,mainly using supervised learning deep learning and unsupervised learning Kmeans clustering algorithm of machine learning two algorithms,proposed three intelligent channel estimation and equalization scheme.The first scheme is GKDNN-DE channel estimation and signal detection scheme based on deep learning,which is an improvement of the existing DNN estimation scheme assisted by Gaussian kernel.The essence of the scheme is to use the trained GKDNN model to directly recover the received signal,which is an implicit channel state estimation method.Adding Gauss kernel greatly reduces iteration times and improves system performance.The simulation results show that when 64 pilot frequencies are used in each frame,the bit error performance of the system increases by 10 dB compared with LS equalization and 5 dB compared with MMSE equalization.The number of iterations of DNN equalizer is reduced to 37.69%by Gaussian kernel assistance,and the highest bit rate can reach 1.52 Gbps.The increase of 0.5 Gbps was higher than that without Gaussian kernel.The second scheme is channel estimation and equalization scheme based on ReEsNet.Based on previous ChannelNet,a ReEsNet network for channel estimation is proposed,which has compact structure and low computational cost.The idea of the scheme is that the channel estimation problem is regarded as an image processing problem,and the pilot channel restoration process is regarded as a low resolution image restoration into a high resolution image,and the restoration process is realized by ReEsNet network.Simulation results show that ReEsNet’s gain is better than ChannelNet’s at low signal-to-noise ratio of 2 to 3 dB and at high signal-to-noise ratio of 4 to 5 dB,and the mean square deviation is stable at run time.Short computing time.The third scheme is the visible channel blind equalization technology based on IKmeans.For ACO-OFDM signals mapped by 16QAM,Kmeans clustering method is used to divide the received signals into 16 regions,and the centroid of each region is used to replace all the data in the region for unmapping.Since the center of mass is updated using the mean of all data in the region,it is affected by abnormal data,so the 3 σprinciple is introduced to improve.The simulation results show that under 16QAM mapping,the bit error rate of equalized signal increases by 2 to 3 orders of magnitude compared with that of unequalized signal.When the SNR of the system is 15 dB,the bit error rate of the system is 3 × 10-3,which improves the SNR of the system by 8dB compared with that of unequalized signal.