Research And Design Of Low-complexity Signal Detection Scheme For B5G Massive MIMO Multi-antenna User System

B5G(Beyond 5G)is the latest cellular technology currently being developed for wireless communication systems,and signal detection has important application value in single-antenna user and multi-antenna user related channels.As a key technology in B5 G,the Massive Multiple-Input Multiple-Out(MIMO)system usually has tens to thousands of antennas in the base station,which makes the signal detection of uplink transmission very difficult and complex.Therefore,in massive MIMO multi-antenna user signal detection,how to balance detection accuracy and complexity has become a research hotspot of domestic and foreign scholars.While traditional signal detection methods exhibit good performance when the number of antennas is small at first,as the number of antennas increases,even suboptimal linear detectors bring huge complexity.For the complexity of matrix inversion,several existing research directions such as approximate inversion,channel hardening simplified inversion,iterative elimination,etc.can reduce the complexity,but their performance can only be close to the original algorithm.So in order to break the bottleneck of performance,this paper tries to combine deep learning with signal detection.Deep learning is divided into data-driven and model-driven.Data-driven requires a lot of data-driven,which is very time-consuming,so the research direction of this paper is mainly model-driven.The main contents are as follows:Firstly,In this thesis,two model-driven DL detector networks based on the Gauss-Seidel iterative method are proposed for massive MIMO systems.Deep learning(DL)has successfully helped researchers study more accurate and efficient MIMO system signal detectors.However,while existing DL-based MIMO detectors improve performance,they also increase complexity.This paper considers single-antenna user equipment(SAUE)and multi-antenna user equipment(MAUE)systems in Rayleigh channels.Firstly,the channels of SAUE and MAUE systems are analyzed,and a loop nesting method is proposed to convert the inverse of the original large matrix into the inverse of the small matrix to reduce the complexity.Therefore,a model-driven DL detector network BGSNet based on the Gauss-Seidel iterative method is designed.At the same time,in order to improve the performance of the algorithm under the MAUE system,BGSNet is improved,and a new improved network Improved BGSNet is proposed.The simulation results show that the performance of BGSNet is significantly better than the Gauss-Seidel algorithm;the proposed scheme is suitable for massive MIMO and has low complexity;the robustness of BGSNet is very good,and the performance is less affected by the number of antennas;Under the MAUE system,Improved BGSNet performs better than BGSNet,and both are suitable for low-correlation and medium-correlation MAUE systems.Secondly,In this thesis,a scheme and a network based on the expectation propagation algorithm are proposed to solve the problem of performance degradation of BGSNet and Improved BGSNet under the disappearance of channel hardening phenomenon and multi-user highly correlated channels.The MIMO detection scheme based on Expectation Propagation(EP)has received extensive attention in recent years.However,while existing EP-based MIMO detectors achieve high accuracy,they also increase complexity.In order to explore the reasons for the high accuracy of EP algorithm,this paper found that EP algorithm constructs matrix features similar to channel hardening.In order to solve the high complexity of the EP algorithm,this paper proposes a DGEP(Diagonalized Gauss-Seidel Expectation Propagation)detection method based on expectation propagation.In this paper,some hyperparameters of DGEP are converted into parameters and a DGEP-Net scheme incorporating deep learning is proposed to further improve the detection performance.The simulation results show that the performance of DGEP is significantly better than that of BGSNet and IBGSNet algorithms;the proposed scheme can effectively reduce the influence of channel hardening disappearance;The robustness of the proposed algorithm is very good.The proposed algorithm has good robustness and can fit EP algorithm well while reducing complexity;This thesis proposed DGEP-Net network even outperforms the EP algorithm in most cases.