Message-passing Iterative Receiving Technique For Massive MIMO Systems

With the development of modern information society and the prevalence of smart mobile devices,the future mobile communication systems require the transmission rate to improve thousands of times of the existing one.Massive multi-input multi-output(MIMO)can effectively enhance the spectral efficiency of wireless communication systems,which becomes a key enabling technology for the fifth generation(5G)mobile communication systems.In order to reduce the complexity of receiver in uplink massive MIMO systems,this thesis investigates massage-passing iterative receiving technique for massive MIMO systems.Firstly,as the optimal detection/estimation problem is one of the most common problems in linear mod-els from the perspective of signal processing,we design factor graph based message-passing algorithms to compute the marginal posterior probability in the optimal detection/estimation problem based on minimum free energy criteria.We unravel the relation between minimizing free energy and computing the marginal posterior probability,and therefore convert the problem of computing the marginal posterior probability to minimizing free energy.The free energy approximation methods are given with the help of the factor graph,field approximation,region approximation,and Bethe approximation.Then,the fixed point of belief propa-gation(BP)is obtained based on minimizing the constrained Bethe free energy.The generalized approximate message passing(GAMP)algorithm with low complexity is obtained by simplifying the BP algorithm.Fur-thermore,ADMM-GAMP algorithm is derived based on minimizing the approximate Bethe free energy.And then,we analyze the convergence of message-passing algorithms with arbitrary measurement matrixes and find out that Damping-GAMP and ADMM-GAMP algorithms can converge to fixed points when the prior probability distribution function(PDF)of transmitted signals satisfies a convex function.Simulation results demonstrate that when the transmitted signals are Gaussian random variables,message-passing algorithms can yield the same performance as the minimum mean square error(MMSE)filter under the Gaussian ran-dom matrix.The computational complexity of the GAMP algorithm is lower than that of the MMSE filter without matrix inversion.Secondly,we design message-passing receivers in order to reduce the computational complexity of u-plink receiving in massive MIMO systems.Based on a physically practical channel model,we investigate the spatial correlation of the massive MIMO channels.The Damping-GAMP and ADMM-GAMP algorithms are used as the detection algorithms for massive MIMO receivers,considering the channel matrix with one or more characteristics among non-zero mean,low-rank,column-correlated,and ill-conditioned.Meanwhile,in order to improve the convergence speed and accuracy of massage-passing detection algorithms,we pro-pose to adopt singular value decomposition(SVD)as a pre-processing operation to eliminate the column correlation and ill-conditioned of the channel matrix.Then,we propose the iterative receivers based on the Soft-Input Soft-Output(SISO)message-passing detection.Simulation results show that symbol error rates of the Damping-AMP and ADMM-GAMP detectors are superior to the MMSE detector after SVD pretreatment at the receiving end.Finally,the energy spread transform(EST)based massive MIMO transmission and message-passing receiving methods,from the transceiver joint optimization perspective,are proposed to solve the existing detection error problems caused by message-passing detection algorithms.When the transmitted signals are QAM modulation symbols,the convergence of the Damping-GAMP and ADMM-GAMP algorithms with any channel matrix can not be ensured,and the Gaussian approximation conditions in the derivation process cannot be satisfied.To solve these problems,we propose massive MIMO transmission based on EST,including EST-based transmitter and the EST-based receiver.Two detection methods are proposed,i.e.,joint detection and separate detection(detection for each user on each subcarrier respectively),and the corresponding complexity analysis are also presented.Simulation results show that when the transmitted signals are QAM modulation symbols,EST-based message passing detectors significantly outperform the ones without EST.And the EST-GAMP detector is computationally efficient while achieving a comparable performance of the joint detection.