A Study On Low Complexity Linear Detection Algorithm In Massive MIMO System

Massive Multiple-input Multiple-output(MIMO)technology has aroused great interest because it can effectively improve the system spectral efficiency and link reliability,and create more space freedom.As one of the key technologies of the fifth generation mobile communication system,Massive MIMO makes use of the characteristic of channel "hardening" to achieve the best performance of detection algorithm.However,due to the number of antennas increases dramatically,the computational complexity of the system increases significantly as well.At present,the signal detection in Massive MIMO system has become the the bottleneck of performance.In the process of searching for a low complexity detection algorithm,linear detection algorithms have received renewed attention because of their low complexity.Therefore,this paper will focus on studying the linear detection algorithm with low complexity and high performance in Massive MIMO system.First of all,this paper introduces the research status of signal detection technology in Massive MIMO system and discusses the opportunities and challenges of Massive MIMO technology as well as some current solutions.Detailed description of the three linear detection algorithm that is ML,ZF,MMSE detection algorithm,focusing on the filter matrix expression of the ZF,MMSE detection algorithm,and the impact of noise on the detection algorithm is analyzed.Finally,the soft decision method of MMSE detection algorithm in MIMO system is studied,the simulation results show that the soft decision algorithm can improve the performance of MMSE detection algorithm.Secondly,we analyze and simulate the channel "hardening" characteristics of Massive MIMO system,that is,with the increase of the number of receiving antennas in the base station,the channel tends to be deterministic and the channel state information is mainly distributed on the main diagonal.At present,the large matrix inversion algorithm of the linear detection algorithm in Massive MIMO system is mainly based on Neumann series expansion,which is converted into large matrix multiplication,but its complexity is still high.In order to further reduce the complexity,according to the characteristic of channel"hardening",we propose a new scheme for large matrix inversion to decompose the large matrix into the sum of the diagonal matrix and the hollow matrix,then Neumann series approximation is carried out to obtain the inverse operation with a lower complexity.And a new method of solving the optimization factor in the Neumann series is proposed in this scheme,so that the linear detection algorithm by using this inversion algorithm in Massive MIMO system is reduced the complexity from O(K3)to O(K2)with a small loss of the performance.The proposed scheme is applied to two linear detection algorithms of ZF and MMSE in Massive MIMO systems.The simulation results show that Neumann series approximation based on Diagonal matrix decomposition,in the case of antenna number and the number of users are the same,the detection algorithm select first two terms(L = 2)of Neumann series expansion with optimization factor scheme,its performance approximation the no optimization factors to select first three items(L = 3)of detection algorithm with the complexity is reduced by an order of magnitude.A Neumann approximation algorithm based on Tri-diagonal matrix decomposition is proposed in this paper,which decomposes the large matrix into the sum of the hollow matrix and a matrix of three diagonal elements centered on the main diagonal.The inverse matrix of Tri-diagonal matrix is obtained by using Gauss elimination method.However the Neumann series approximation algorithm based on Tri-diagonal matrix decomposition still with a high complexity in solving the inverse matrix by using Gauss elimination method,as well as the parallel processing is difficult to achieve.We propose a Neumann approximation algorithm which is suitable for computer parallel processing based on Frobenius matrix decomposition,that is,the large matrix is decomposed into the sum of the hollow matrix and a matrix with a diagonal element and a column of the matrix.By using the good properties of Frobenius matrix,the inverse matrix of Frobenius matrix can be obtained efficiently.These two schemes are applied to the linear detection algorithms of ZF and MMSE in Massive MIMO systems in the inverse of the Neumann series approximation,respectively.The simulation results show that the Neumann series approximation detection algorithm based on Tri-diagonal matrix decomposition has better detection performance than that of Neumann series approximation based on Diagonal matrix decomposition;the performance of the Neumann approximation algorithm based on Frobenius matrix decomposition is better than that of the Neumann approximation algorithm based on the Tri-diagonal matrix decomposition.Finally,we extend the above research results to the uplink of multi cell MIMO Massive system.In the condition of knowing the channel state information of all cells in advance,the SVD decomposition method is used to eliminate the inter-cell interference at first,then applying the new low complexity ZF detection algorithm.It has also been studied that the channel state information of the target cell is known and the other cell channel state information is unknown,by sloving the mean and variance of the sum of the interference terms and the noise,the multi-cell channel model can be transformed into a single-cell mode,then using the new low complexity MMSE detection algorithm for single-cell Massive MIMO system.The simulation results show that the proposed algorithm has lower computational complexity with slight loss of performance.