Research On Deep Learning-based Channel Prediction Method For B5G High-Speed Mobile Communication System

In recent years,with the rapid development of science,technology and economic conditions,the high-speed railways and expressways have also shown an explosive growth trend,bringing great convenience to people’s travel.Enjoying convenient travel,the demands of low-latency and high-quality communication are becoming increasingly urgent,which promotes the wireless communication technology in high-speed mobile scenarios to attract more and more attention in the world.In the B5 G system,when the mobile speed of the terminal reaches more than 500 km/h,the fast time-varing channel and the Doppler frequency spread effect will lead to a sharp drop in the data transmission rate of the communication system,which seriously affects the quality of communication service of users.Accurate acquisition of channel state information(CSI)is currently an effective way to improve the performance of the communication system and ensure the quality of user communication.At present,researchers have obtained many research achievements of high mobility channel estimation.However,due to the delay of signal transmission,channel processing and feedback processes,the CSI obtained by the channel estimation method is outdated,which also leads to a severe deterioration of the communication system performance in this scenario.Therefore,to ensure the communication quality of the high-speed mobile scenarios in the B5 G system,the channel prediction technology has attracted much attention,and has been widely used in the acquisition of CSI in this scenario.To further improve the prediction accuracy and the applicability of the prediction models,as well as reduce the prediction complexity,this thesis explores the time-varying channel prediction methods based on existing prediction methods,which are more suitable for the B5 G high-speed mobile communication system.The main content and innovation points are as follows:(1)For high-speed mobile multiple input multiple output-orthogonal frequency division multiple access(MIMO-OFDMA)systems,a novel time-varying channel prediction method joint polynomial basis expansion model(P-BEM)and back propagation(BP)neural network is proposed.To reduce the computational complexity,the P-BEM is employed to model the time-varying channel,and the channel information at the future time is obtained by the offline training and online prediction of the channel base coefficient.During offline training,the channel base coefficient is first acquired by the received pilot signal.Then to obtain the channel prediction network model,the training sample is constructed and sent into the BP neural network for training.During online prediction,based on the network model and historical base coefficient estimation obtained by training,the time domain channel at the future time is obtained.Theoretical analysis and simulation results show that the proposed method has lower computational complexity and better prediction accuracy than the existing methods,which is suitable for the efficient acquisition of time-varying channel information in the future high-speed mobile environment.(2)Aiming at the problems of traditional BEM is relatively fixed,causing large error in modeling time-varying channel in high-speed moving scenarios,a novel time-varying channel prediction method based on BP neural network under modified basis expansion model is proposed.According to the strong correlation characteristics of the channel experienced by different vehicles at the same location,firstly,the channel correlation matrix at the historical time is used to obtain the optimal basis function to improve the traditional BEM,and the modified basis expansion model is employed to model the channel.Then the base coefficient with finite parameters is also used for offline training and online prediction.During training and prediction,the prediction target is the base coefficient at the future time,rather than the channel information,and the channel information at the future time is obtained by the conversion relationship between the base coefficient and the channel.Since the dimension of the base coefficient is much smaller than channel information,the proposed method also further reduces the complexity in training and testing stages of the network.(3)In contrast to the BP neural network,the long short-term memory(LSTM)neural network can take full account of the temporal correlation of the sequence,and it is often used to predict the time sequence.Based on this,a novel time-varying channel prediction method based on LSTM neural network under modified BEM is proposed.The proposed method similarly models the channel by the modified BEM.Then,the channel information at the future time is obtained by the offline training and online prediction of the channel base coefficient via LSTM neural network.Since the ideal channel information is unknown in the actual communication system,the estimated channel coefficient is set to the approximation objective in the training stage of the network.Theoretical analysis and simulation results show that the proposed method has better prediction accuracy and lower complexity than the existing methods,which applies not only for existing high-speed mobile scenarios but also for future high-speed mobile communication scenarios.