Research On Hybrid Precoding Algorithm For Millimeter Wave Massive MIMO Systems

Current mobile communication technology and available low frequency band resources are unable to keep up with the increasing communication demands of users.As a solution,nextgeneration mobile communication systems can exploit more frequency band resources through the use of millimeter wave frequency bands.The combination of millimeter wave technology and large-scale multiple-input multiple-output(MIMO)technology can provide a communication system with wide coverage,large bandwidth,fast transmission rates,high spectrum performance,high throughput,and strong reliability,making it a key technology for future mobile communication.However,in practical application of millimeter wave large-scale MIMO systems,the use of traditional fully digital precoding is limited due to hardware cost and power constraints.Therefore,a hybrid precoding scheme that combines digital precoding and analog precoding is preferred,as it can reduce the number of high-energy consumption radio frequency(RF)chains,thereby reducing power consumption and hardware cost.Consequently,the hybrid precoding scheme is a popular research direction to investigate the complexity and spectrum performance of millimeter wave large-scale MIMO systems.Based on the above discussion,this paper will delve into the following research topics:(1)To address the issue of spectral performance degradation caused by the use of limited stationary phase shifters,this paper proposes a hybrid precoding algorithm based on alternating optimization and orthogonal matching pursuit in dynamic networks.Firstly,a simulated precoding matrix is employed to determine the initial connection state between the phase shifters and antennas under constant mode constraints,which accelerates the iterative convergence rate.Subsequently,an optimal simulated precoding matrix is constructed based on the connection state,and the globally optimal index vector is solved.Finally,the digital precoding matrix is constructed using the index vector and fed back into the dynamic network,enabling the alternating optimization and update of the connection state between the phase shifters and antennas.Simulation results demonstrate that this algorithm still maintains high spectral efficiency,rapid iterative convergence,and low complexity even with a limited number of fixed phase shifters.Notably,the spectral efficiency is notably enhanced when the number of RF links exceeds the number of data streams.(2)To enhance the spectral efficiency of hybrid precoding in low resolution phase shifter scenarios,an alternative optimal hybrid precoding algorithm that combines Semidefinite Relaxation(SDR)and Babai-Lattice algorithm is proposed in this paper.Firstly,the initial simulation precoding matrix is designed using the EGT algorithm based on the sparse characteristics of millimeter wave channels,which reduces the interference among users and the convergence time of iterations.Next,the semi-positive definite relaxation method is employed to optimize the quadratic constraint non-convex problem of the objective function and update the digital precoding matrix.Finally,through multiple iterative optimization,the updated digital precoding matrix is substituted into the Babai-Lattice algorithm to optimize the updated phase shifter connection state and the simulated precoding matrix,leading to the mixed precoding matrix.Simulation results show that the proposed hybrid precoding algorithm in low resolution phase shifter scenarios can improve the frequency spectrum efficiency,convergence time,and computational complexity.(3)A convolutional neural network-based hybrid precoding scheme is proposed to reduce the dependence on the antenna array response in practical applications of hybrid precoding.Firstly,a deep learning framework based on convolutional neural network is constructed,which takes the channel matrix as input and predicts the output of digital and analog precoding matrices.Then,in the data generation stage,the coupling optimization problem is solved by combining manifold optimization algorithm,least mean square method,least squares method and orthogonal matching pursuit algorithm,to obtain the hybrid precoding matrix with optimal spectral efficiency.In the training stage of the network,the noisy channel matrix and the hybrid precoding obtained by decoupling optimization are used as input and output pairs of the network,and the feature set of the training samples is extracted.Finally,in the prediction stage,the antenna array response is not required,and the noisy channel matrix is directly input into the trained network to obtain the hybrid precoding matrix.Compared with greedy search and optimization algorithms,this scheme has higher spectral efficiency and shorter iteration calculation time.