Uplink Channel Estimation In Massive MIMO Systems

With the rapid development of multimedia services and intelligent terminals,people have higher and higher requirements for the transmission speed of mobile communications and traditional technology can no longer meet this demand.Therefore,massive multiple input multiple output(MIMO)technology has become the research focus,and it is one of the core technologies of 5G.Compared with traditional MIMO systems,there are hundreds of antennas in massive MIMO systems,which means that the spatial freedom of wireless channel is greater,and its stability and transmission speed are also higher.However,in massive MIMO systems,the wireless channel has large randomness influenced by the shadow fading,the frequency selective fading,etc.And accurate channel state information(CSI)is the key basis for analyzing a wireless communication system.Therefore,the study of channel estimation algorithms is a significant work.This thesis focuses on the channel estimation algorithms using pilots.Influenced by the large number of antennas in massive MIMO systems,channel estimation becomes more difficult and the phenomenon of pilot contamination is more obvious,which greatly affect the accuracy of channel estimation.In this thesis,a channel estimation algorithm based on machine learning theory is studied.Based on an observation of the received signals in the beam domain,the channel components are modeled as a Gaussian mixture model(GMM).The Bayesian compressed sensing(CS)is used to estimate the channel because the channel gains in the beam domain is approximately sparse.Approximate message passing(AMP)algorithm is used to solve the integral of Bayesian estimation.When determining the initial values of the algorithm,hierarchical clustering(HC)algorithm is used to preprocess the channel samples.In a multi-cell model with pilot contamination,the channel parameters of the interference links from adjacent cells are also estimated in order to reduce the influence of pilot contamination.Simulation results show that better performance of mean square error(MSE)and convergence can be achieved in this thesis compared with the existing algorithm.