Research On Massive MIMO Signal Detection Algorithm Based On Machine Learning

Massive MIMO technology has developed rapidly because of its data transmission rate and reliability advantages,it becomes one of the key technologies of 5G(5thGeneration)mobile communication,but with a large number of communications equipment access communication network,with the increase of the number of users and number of antennas,the complexity of the system is rising quickly,research of massive MIMO signal detection is facing great challenges,the traditional signal detection algorithm will lead to the problem of high computational complexity,thus improving the performance of signal detection algorithm and reducing its complexity has become a hot research direction at present.The main research of this paper as follows:Firstly,by introducing the theoretical basis required by MIMO signal detection algorithm,including MIMO system model,linear detection algorithm,nonlinear detection algorithm and iterative detection algorithm,the algorithm expression and its complexity are analyzed.Then the deep neural network model and back propagation method are described to lay a theoretical foundation for the establishment of deep detection network.lastly,we introduce Detection Network(Det Net)and Learned Conjugate Gradient Descent Network(Lcg Net)of MIMO detection network based on deep learning,including network model and training method,which provides some research directions for current deep MIMO signal detection network.Secondly,Richardson(RI)algorithm is a simple iterative method,but the convergence of the algorithm is insufficient.Studies show that changing the relaxation parameters can improve the convergence speed,and then improve the performance of the algorithm.So RI iterative algorithm combined with model-driven deep learning method,set the relaxation parameter for learning parameter,and in order to avoid the gradient disappears,set the residual structure,establish a deep network Richardson Network(RINet),the corresponding to different relaxation parameters of each layer network,appropriate relaxation parameters can effectively improve the convergence rate of RI,improve the performance of the algorithm,finally,the simulation experiments show that RI-Net’s performance is much better than the RI algorithm.Thirdly,RI is insufficient because of convergence and complexity,so a fast convergence detection algorithm NSR is proposed based on RI,but the performance is not ideal in the case of high dimensional MIMO.Therefore,Steepest Descent(SD)algorithm is used to update the signal in advance,which is used as input into NSR algorithm to participate in the operation,and then the deep fusion forms Steepest Descent Non-Stationary Richardson(SDNSR)algorithm.The algorithm is superior to RI algorithm in complexity and performance.Since the iteration process of the algorithm needs to calculate the step size and relaxation parameters,the step size and relaxation parameters are set as learning parameters to establish the deep network Steepest Descent Non-Stationary Richardson Network(SDNSR-Net),which avoids the calculation of the step size and relaxation parameters and further reduces the complexity.Finally,the simulation results show that the performance of RI-Net is better than that of RI-Net at high SNR.The article will research MIMO signal detection by combining MIMO detection techniques and deep learning techniques,which based on model-driven deep learning under the massive MIMO multiuser system model,by expanding algorithm iterative process to establish deep network,the parameters of influencing the algorithm performance regard as learning parameters,by training the network to find the right parameters,can effectively improve the detection performance of the algorithm.In the field of massive MIMO signal detection,the MIMO signal detection algorithm combined with deep learning technology can indeed improve the algorithm performance,which brings a new idea and direction to the field of signal detection.