Study On Detection Algortihm And Efficient Hardware Implementation For Massive MIMO

The rapid growth of data volume,especially mobile data volume,puts forward higher requirements on existing networks.The fifth generation mobile wireless system(5g)has attracted a lot of attention due to its higher data rate,larger network capacity and higher spectrum efficiency,and has been put into commercial use.As a core technology in 5G,massive multiple-input multiple-output(MIMO)technology can improve link reliability,spectrum efficiency,and energy efficiency.Signal detection is a key issue in the research of massive MIMO.It is very important to estimate the transmitted signal as accurately as possible with reason-able complexity through the signal at the receiving end.Based on the review and analysis of existing detection algorithms,this paper proposes a new parallel minimize residual descent(parallel minimize residual descent).accelerate relaxation(P-MRD-AOR)detection algorithm and hardware implementation of it,which proves that the algorithm achieves a good balance between performance and complexity.In addition,this paper also implements the EPA-w NSA(expectation propagation with approximate-weighted Neumann series approxi-mation)algorithm based on EP(expectation propagation)algorithm,and verifies that it has a good balance between performance and hardware efficiency.First of all,this paper briefly introduces the massive MIMO technology and its advantages and problems.The signal detection problem of massive MIMO is the focus of this article.Aiming at this problem,taking the contradiction between performance and complexity as the breakthrough point,the current research status of this problem is introduced.Existing detection algorithms can be divided into optimal detection algorithms and sub-optimal detection algorithms,and better performance will bring higher complexity.The minimum mean square error(minimum mean square error,MMSE)algorithm in the suboptimal detection algorithm has become the focus of this article due to its better detection performance and lower complexity.For MMSE detection algorithm,the complexity of large-scale matrix inversion operation is difficult to accept when the number of antennas increases.In order to avoid the direct inversion of high-order matrix,different detection algorithms based on MMSE are proposed.Jacobi iterative algorithm,Gauss-Seidel(GS)it-erative algorithm,successive over relaxation(SOR)iterative algorithm and accelerated over relaxation(AOR)iterative algorithm are included in the category of AOR family algorithm after analysis.After that,the sim-ilarity between the iterative structure of the AOR-family algorithm and the preprocessing technology was analyzed,and the iterative structure of the AOR-family algorithm is rewritten.For such a linear iterative structure,the convergence rate is negatively related to the spectral radius of the iterative matrix.Therefore,in order to improve the convergence rate of the linear iterative algorithm,the P-MRD-AOR algorithm with the purpose of minimizing the Frobenius norm of the iterative matrix is proposed Compared with the existing AOR-family iterative algorithm,the P-MRD-AOR algorithm has better parallelism,which is very beneficial for hardware implementation.Numerical simulation results in different scenarios prove that the P-MRD-AOR algorithm has better detection performance than other AOR-family algorithms,and achieves a better balance between performance and complexity.In order to verify the hardware advantages of the P-MRD-AOR algorithm due to its full parallelism,a corresponding hardware design was proposed and implemented.In order to verify the hardware advantages of the P-MRD-AOR algorithm due to its full parallelism,a corresponding hardware design was proposed and implemented.The overall hardware structure and top-level timing diagram are proposed,each sub-module is designed for the purpose of saving hardware resources as much as possible,and the quantization scheme is determined through fixed-point simulation results.The P-MRD-AOR detector based on SIMC(semicon-ductor manufacturing international corporation)65nm process can achieve a throughput rate of 570Mbps at a frequency of 455MHz on an area of 2.22mm~2.The comparison with some existing massive MIMO detectors shows that the P-MRD-AOR detector achieves a good balance between performance and hardware efficiency.Compared with the accurate EP detection algorithm,the EPA-w NSA algorithm reduces the complexity,and can alomost achieve accurate EP detection performance under certain circumstances.In order to better evaluate the advantages of the algorithm in terms of complexity,this paper implements the algorithm in hard-ware.The comprehensive results under the SMIC 65nm process show that the detector can achieve a good balance between performance and hardware efficiency.