| With the rapid development of wireless communication technology,5G communication technology has been widely used.Among them,technologies such as Multiple Input Multiple Output(MIMO)and Millimeter Wave Communications greatly improved spectrum utilization and provided rich spectrum resources for communication systems.At the same time,it also increases the difficulty of channel estimation and channel feedback.In a time-varying communication environment,the Channel changes in real time,and the channel has timevarying correlation in coherent time,which further increases the difficulty of obtaining Channel State Information(CSI)at both transceiving and receiving ends of the communication system.In recent years,the research on deep learning is in full swing.The powerful feature extraction capability of deep learning technology provides a new path for channel estimation and channel feedback technology.In the time-varying large-scale MIMO millimeter-wave communication system,this paper studies the channel estimation and channel feedback technology based on deep learning,so as to achieve the purpose of obtaining accurate CSI at the transceivers and receivers of the time-varying communication system.Specific research contents are as follows:Firstly,the principle,advantages and disadvantages of large-scale MIMO millimeter wave communication system are briefly described in this paper,and the Saleh-Valenzuela channel model is introduced.In a time-varying environment,Markov process is introduced into the complex gain of the channel model to reflect the time-varying correlation of the wireless channel in coherent time,so as to describe the time-varying channel more accurately.On this basis,the channel estimation technology in time-varying communication environment is studied,and the performance of traditional channel estimation algorithm in time-varying communication environment is studied.In addition,it briefly expounds the relevant theoretical knowledge of deep learning,which lays the foundation for the subsequent relevant research.Secondly,based on the Least Squares(LS)channel estimation algorithm,this paper uses Convolutional Neural Network(CNN)to design an improved LS algorithm based on CNN in a time-varying large-scale MIMO millimeter-wave communication system.After the LS channel estimation of the received pilot signal,the algorithm uses CNN to carry out noise reduction and feature extraction for the LS channel estimation,and makes full use of the timevarying correlation of the channel,thus accurately obtaining CSI.Then,pilot design and channel estimation are studied in time-varying large-scale MIMO millimeter wave communication systems.Aiming at the idea of designing pilot sequence with full connection layer and realizing channel estimation with multi-layer convolutional neural network,this paper extends the joint pilot design channel estimator network based on the timevarying dependence of wireless channels in coherent time,and proposes a time-varying joint pilot design channel estimator network by using primary and secondary pilot working mechanism.This network model is composed of multiple joint pilot design channel estimator networks.It can accurately estimate the channel of time-varying large-scale MIMO millimeterwave communication system with less pilot overhead.Compared with traditional channel estimation algorithms and joint pilot design channel estimator network,this network model has superior channel estimation performance.Finally,this paper studies the channel feedback problem in time-varying large-scale MIMO millimeter wave communication system,improves the traditional Recon Net neural network model,and design a neural network model named Reconnet-LSTM,which can selfencode and self-encode the CSI of time-varying channel.With a small amount of codeword overhead,CSI of time-varying channels can be accurately restored.The simulation results show that Recon Net-LSTM has better channel feedback performance than other traditional neural network models. |
