Research On Compressive Sensing And Deep Learning-based CSI Acquisition In FDD Massive MIMO Systems

Massive Multiple Input Multiple Output(MIMO)is one of the key technologies of the Fifth Generation(5G)mobile communication,which has attracted extensive attention and research due to its high frequency spectrum and energy efficiency.In order to achieve the performance gain of massive MIMO system,the Base Station(BS)needs to obtain accurate downlink Channel State Information(CSI).In Frequency Division Duplex(FDD)system,CSI is acquired by the user’s estimation of CSI first and then feedback to the base station.However,due to the large number of antennas in massive MIMO systems,the complexity of existing CSI acquisition algorithms and the pilot and feedback overhead will be greatly increased.In order to solve this problem,this thesis uses compressed sensing and deep learning technology to acquire CSI,and mainly completes the following two tasks:First,aiming at the problem of the overhead of spectrum resources in the Orthogonal Frequency-Domain Multiplexing(OFDM)FDD massive MIMO systems when the user(UE)feedback the CSI,a CSI feedback method based on CS and DL is proposed based on the sparse characteristics of massive MIMO channels in angular time-delay domain.First,the CSI obtained by the UE are converted into the angular time-delay domain to obtain the new sparse channel matrices.By using CS technology,the sparse channel matrices can be compressed at any ratio,thus the feedback overhead is reduced.At the same time,a CSI reconstruction model,named Re Net,is designed,and the parameter setting,training scheme and performance indexes are described.With comparing the normalized mean-square error(NMSE),cosine similarity and bit error rate(BER)performance of the classical algorithms based on CS and algorithms based on DL under different compression ratio,the simulation result shows that the proposed method can reduce the cost of CSI feedback,fully guarantee the perfomance of the CSI reconstruction and reduce the computational complexity to a large extent.Secondly,aiming at the problem that CSI estimation becomes more difficult due to the fast obsolescence of CSI that obtained through downlink feedback in time-varying channels,a channel estimation scheme based on CS technology and DL is proposed in this thesis.First,an OFDM time-varying massive MIMO system is modeled.Based on the analysis of the structured sparsity of massive MIMO channel in the time-delay domain,a joint channel estimation algorithm of Adaptive Structured Joint Orthogonal Matching Pursuit(AS-JOMP)and Deep Neural Network(DNN)is designed.The reconstruction performance of the ASJOMP algorithm is analyzed firstly,and the NMSE of the proposed AS-JOMP algorithm with the classical CS-based method under different signal-to-noise ratio(SNR)and training overhead are compared.For the proposed DL-based models,Dn Net network and Dn LSTM network,we compared the NMSE of the two network architectures with AS-JOMP algorithm under different SNR,the simulation shows that the proposed joint algorithm can reduce the number of pilots and feedback overhead while reducing the computational complexity and improving the reconstruction performance.To sum up,this thesis focus on the massive MIMO CSI acquisition scheme based on compressed sensing and deep learning.The proposed schemes can effectively reduce the overhead in the channel estimation and feedback process,and improve the accuracy of the acquired CSI.