Research On Channel Estimation Of Massive MIMO Systems

Compared with traditional mobile communication technologies,5G communication requires higher transmission quality,faster transmission rate,and greater channel capacity.These performance enhancements cannot be separated from the exploration of large-scale MIMO systems and millimeter-wave frequency bands.Different from conventional channel estimation,for the base station has large number of antennas,the training overhead required for large-scale MIMO channel estimation is greater.Therefore,the algorithm needs to reduce the computational complexity and increase the adaptive estimation capability.Aiming at the problem of downlink multipath channel state information(Channel State Information,CSI)estimation for massive MIMO channels,this thesis proposes a mode-shared sparse Bayesian learning(Pattern Sharing-Sparse Bayesian Learning,PSsbl)method,which divides the channel vectors into multiple sets of vectors of equal length.Each group is associated with a common hyper parameter such that the coefficients of the groups of vectors have the same sparsity.Taking the channel coefficient as a hidden variable and the hyper parameter as an unknown parameter,a Bayesian inference iterative algorithm is proposed based on the expectation maximization(Expectation-maximization,EM)framework to estimate the channel posterior probability.For the downlink channel estimation problem of the multi-user and the multipleinput multiple-output system with frequency division duplex mode,assume that one base station communicates with K mobile users.Because of the limited scattering in physical propagation,each channel matrix is sparse in the virtual angular domain.Different users usually share some common scatters.Therefore,there are partial common sparse patterns for different channel matrices.In this thesis,a Gaussian mixture prior model is designed.The channel estimation method based on variational Bayesian inference can be used to effectively reconstruct the sparsity of each user.In addition,based on the GAMP algorithm,the proposed channel approximation posterior probability algorithm can effectively avoid matrix inversion,greatly reduce the computational complexity of channel estimation.In this thesis,when estimating the uplink and downlink channels,it is assumed that the sparse channel obeys the Gaussian distribution with a certain parameter,the common sparse characteristics of the channel in the spatial domain are used,and the prior parameters are learned from the measurement matrix based on the expectation maximization method,and thus rebuild sparse channels.The simulation results show that the performance of the algorithm presented in this thesis is obviously better than that of similar algorithms and can approximate the performance of the least squares(Least squares,LS)method in the ideal state.