Research On MIMO Signal Detection Algorithm Based On Expectation Propagation

Multiple-input multiple-output(MIMO)technology can multiply the spectrum utilization of wireless communication systems,and plays an important role in coping with the exponentially increasing demand for wireless access.Since the advent of MIMO technology,low-complexity and high-performance detection algorithms have been a research hotspot for researchers and engineers.Expected Propagation(EP)specifically solves the complex probability estimation,and has a wide range of applications in the signal detection of communication systems.The research focus of this paper is the MIMO signal detection algorithm based on expected propagation.The EP algorithm is a deterministic probability approximation algorithm which iteratively refines the approximation.First,a factorable approximate probability distribution is selected,and each factor item is updated in parallel to gradually converge to the best approximation iteratively.When applying EP algorithm to the MIMO detection,in one case,the joint posterior probability of the transmitted symbol can be directly calculated via Bayesian formula,while in the other case,only the marginal posterior probability of the symbol component needs to be calculated,so this paper derives two MIMO signal detection algorithms based on expected propagation.For the sake of distinction,the above two algorithms are refered as Joint posterior probability Expectation Propagation(JEP)and Marginal posterior probability Expectation Propagation(MEP),respectively.The JEP algorithm is proved to be Bayesian optimal under the theoretical large-scale system conditions,but its optimality cannot be achieved in practical applications,which limits performance;in addition,there are a large number of division operations in the iterations of the MEP algorithm,which is the main obstacle that limits its practical application.In view of the shortcomings and problems of JEP and MEP algorithms,this article proposes the following three innovations:1.In view of the high convergence error rate of the JEP algorithm and the problem that the optimality of the algorithm cannot be achieved under the actual MIMO system,the improved moment matching scheme makes full use of the self-correction capability of the EP algorithm,thereby reducing the error convergence rate of the algorithm;in addition,by analyzing two key parameters(initial variance and oscillation Factor)on the JEP algorithm convergence speed and detection performance,with the help of machine learning methods to select the best parameters for each iteration.Finally,the simulation results show that the method in this paper has achieved better detection performance and stability.2.In order to allievate the high computational complexity caused by matrix inversion,this paper derives the approximate joint posterior probability expectation propagation algorithm(Approximated JEP,AJEP).The AJEP algorithm eliminates the matrix inversion in the iterative process,but there is a problem of slow convergence or even non-convergence during zero-value initialization.To make full use of the ultra-low computational complexity of the zero-value initialization scheme,a parameter selection scheme based on machine learning is adopted.The simulation results show that the above-mentioned key parameter optimization effectively solves the convergence problem of the AJEP algorithm,and achieves the performance beyond the MMSE detector with lower computational complexity.3.For the purpose of eliminating a large number of division operations in the MEP algorithm,this paper proposes a low-complexity marginal posterior expectation propagation algorithm(Approximated MEP,AMEP)based on numerical approximation.In which an approximate division scheme using addition of the shifted dividend and an approximate update scheme of prior information ignoring the small value are proposed.Benefiting from the approximate nature of EP algorithm,the AMEP algorithm only has a maximum performance loss of less than 1dB compared to the MEP algorithm when the MIMO system has a large transceiver ratio,but the overall computational complexity is less than 20% of the MEP algorithm.