Research On Beamforming Algorithms For Massive MIMO System Based On Beamspace

To ease the spectrum strain,mm Wave frequencies between 24.25 GHz and52.6GHz have been incorporated into the 5G standard.Massive MIMO hybrid beamforming technology can effectively overcome the problem of link loss of mm Wave,and achieve the trade-off between system performance and hardware complexity.It has become a hot technology in 5G physical layer.Most of the existing researches adopt a hybrid system architecture based on phased array,which requires a large network of phase shifters in the RF domain.In order to reduce the hardware overhead of RF domain,some scholars proposed beamspace MIMO(B-MIMO)hybrid system architecture based on discrete lens array.It consists of two parts: precoding and beam selection.And the beam selection can be easily implemented by network of switchers in RF domain.This paper will research on the beamforming algorithms in B-MIMO system.On the one hand,the beam selection under ZF precoder is studied in this paper.Most existing beam selection algorithms require that the number of beams and users must be equal.In order to overcome this limitation,the beam selection problem is transformed into a discrete combinatorial optimization problem for binary selection vectors.After two enhancements to the traditional cross entropy(CE)algorithm,a beam selection algorithm based on enhanced cross entropy(ECE)is proposed in this paper.Performance of the proposed algorithm is verified by simulation.Moreover,the convergence of the proposed algorithm is analyzed and the sufficient and necessary conditions of convergence are obtained.In addition,this paper presents a graphical convergence analysis method,and the correctness of the method is verified by simulation.On the other hand,the problem of precoding matrix optimization under fixed beam selection matrix is studied in this paper.The problem is non-convex and difficult to be solved by low complexity algorithms.Using the mathematical relationship between spectrum efficiency and the received minimum mean square error,and introducing two sets of auxiliary variables,the problem of precoding matrix optimization which is difficult to solve is transformed into the problem of weighted mean square error and minimization.This paper used block coordinate descent(BCD)algorithm to solve the problem of weighted mean square error and minimization,and proposed a precoding algorithm based on weighted minimum mean square error(WMMSE).Simulation results verify the convergence and good performance of the proposed precoding algorithm.