Based On Tensor Manifold Precoding And Channel Estimation Algorithm For Multi-user Millimeter Wave Massive MIMO Systems

Millimeter-wave massive MIMO technology achieves high antenna gain and spectral efficiency by equipping a large-scale antenna array in a limited space.However,with the increase of users served by the system and the increase of the number of antennas at the transceiver end,in order to reduce the power consumption of the user equipment and increase the stability of the transmission link,the precoding technology is indispensable.Precoding also requires accurate channel state information,so the study of precoding and channel estimation in multi-user millimeter wave large-scale MIMO systems is a current hot topic.In this paper,to address this problem,we make full use of the multidimensional characteristics of the signal,the processing capability of tensor and stream shape for high-dimensional data,in addition,we also combine the clustering algorithm to model the multi-user signal to solve the precoding problem and channel estimation problem in millimeter wave large-scale MIMO systems,and accordingly make the following studies:Firstly,for the high-dimensional operation of precoding matrix in multi-user millimeter-wave massive MIMO system,a dynamic user hybrid precoding based on manifold discriminant learning is designed.The similarity measurement function is used to cluster users with similar user characteristics,and a manifold discriminant analysis is proposed to reduce the high-dimensional channel matrix of multi-users in the user group,so that the channel after dimensionality reduction maintains global characteristics.Faced with the situation that users will move among different user groups,the convolution restricted Boltzmann machine framework is introduced in manifold learning.By continuously updating the manifold discrimination in the user group,the optimal state of manifold learning in the user group is obtained.The total rate maximization problem of hybrid precoding is studied using manifold quasiconjugate gradient method.The results show that the algorithm can improve the spectral efficiency and summation rate of the system.Secondly,a hybrid precoding method based on tensor manifold is proposed to solve the problems of inter-user interference and low spectral efficiency in multi-user millimeter-wave massive MIMO hybrid precoding system.Through tensor decomposition in the form of Tucker2,the optimal analog precoding and analog combiner are obtained,respectively,which can suppress inter-user interference and intra-user interference.In tensor decomposition,in order to avoid the problem that some rank-one components converge to infinity in norm,and the decomposition terms partially cancel each other,finite sensitivity is added to tensor decomposition.Then,using a stream-by-stream analysis method,a globally optimal analog precoding and analog combiner can be achieved.Finally,the manifold learning method of tensor local preserving projection is used to embed the high-dimensional data stream into the lowdimensional manifold data stream,which can reduce the number of iterations and the algorithm when solving the optimal analog precoding and analog combiner by data stream.complexity.Simulation proves that the scheme can effectively suppress interference and improve the spectral efficiency of the system.Finally,a channel estimation method based on tensor-dictionary manifold learning is proposed to solve the problems of low spatial resolution and high-dimensional matrix operation of channel estimation in multi-user millimeter-wave massive MIMO systems.Firstly,the received signal is modeled by tensor,and by dividing the received signal tensor,users with similar channel characteristics are clustered into groups using the kmeans clustering method,and the channel tensor dictionary information of the user group is established.Through manifold learning,the relationship between adjacent users in a user group is analyzed,and the high-dimensional channel in the user group is embedded in a low-dimensional space,and then the convex relaxation property of tensor alternating multiplication can be used to eliminate the interference of other user groups.Finally,under the MUSIC method,the channel parameters of the user group are obtained,and the high-precision channel estimation of the user group is realized.This method can perform channel estimation for multi-users with only a small number of pilots,and improves the performance of the system.The simulation results confirm the good performance of the method.