Research On The BFGS Detection Algorithm And Implementation In Massive MIMO Systems

With the rapid development of communication technologies and the fast growth of communication service demands,the wireless communication will face many challenges such as the faster data rate,higher spectrum efficiency and stronger communication quality in the future.Massive multiple-input multiple-output(MIMO)can be used to greatly improve the system capacity and spectrum energy efficiency without increasing spectrum resources and transmission power by deploying a large number of transceiver antennas.It is the core physical-layer technology of the current fifth generation mobile communication system(5G)and future wireless communication.The close combination with millimeter wave communication technology also has broad application prospects.Due to the large-scale expansion of the antennas at the base station,the difficulty of signal detection in the uplink is bound to increase,which becomes an urgent problem to be solved in the practical application of the massive MIMO technology.Fortunately,the linear minimum mean square error(MMSE)detection algorithm for massive MIMO has been proved to achieve near-optimal detection performance with reduced computational complexity.However,with the increasing number of the base station antennas,the MMSE detection algorithm will involve complex high-dimensional matrix inversion,which makes it difficult to implement efficiently.In this dissertation,the massive MIMO linear signal detection technology is taken as the research object with the low complexity and high performance as the main goal.A comprehensive and in-depth research is carried out from three aspects of overcoming the defect of high computational complexity and large storage requirements of the BroydenFletcher-Goldfarb-Shanno(BFGS)quasi-Newton method applied to massive MIMO MMSE detection algorithm,further improving the detection performance of lowcomplexity BFGS detection algorithm especially in bad channel propagation environment,and designing very large scale integration(VLSI)hardware architecture corresponding to the proposed algorithm.A novel iterative linear signal detection algorithm based on BFGS quasi-Newton method and its efficient hardware implementation are proposed.The main work and innovations of this dissertation can be summarized as follows:(1)In order to avoid the high-dimensional matrix inversion involved in the MMSE detection algorithm,it is modeled as a strictly convex quadratic optimization problem,and BFGS quasi-Newton method is utilized to iteratively obtain its solution for the first time.Due to the high complexity and the large storage demand of the BFGS detection algorithm,limited-memory BFGS(L-BFGS)quasi-Newton method is introduced.By further simplifying the number of correction vector pairs required by L-BFGS method and effectively initializing the approximate matrix of Hessian inverse,an I-LBGFS detection algorithm based on the improved L-BFGS method is proposed.Simulation results show that the I-LBFGS detection algorithm can approach the accurate MMSE performance with lower computational complexity compared with the traditional BFGS detection algorithm,conventional L-BFGS detection algorithm and other low-complexity linear detection algorithms.(2)Although the improved L-BFGS quasi-Newton method has been suitable for the massive MIMO signal detection,it does not fundamentally solve the defects of the traditional BFGS method in the application of massive MIMO communication systems.Combined with the theoretical derivation of the I-LBFGS detection algorithm and the essential characteristics of BFGS quasi-Newton method,three low-complexity BFGS detection algorithms are proposed from the simplification of the search direction and step size,including I-BFGS detection with the most robust performance,S-BFGS detection with the lowest complexity and U-BFGS detection with the good performance and complexity trade-off,which can be efficiently applied to massive MIMO signal detection.In addition,the corresponding improvements are proposed to avoid the matrix multiplication involved in the Gram matrix calculation and the matrix inversion involved in the log-likelihood ratio(LLR)calculation,which further reduces the computational complexity.Simulation results show that the proposed low-complexity BFGS detection algorithms can effectively reduce the computational complexity while ensuring the detection performance in coded massive MIMO systems.(3)When the loading factor increases or the channels are correlated in massive MIMO systems,the performance of the linear signal detection algorithms will be affected.Since the selection of the initial solution is very important in the iterative algorithm,based on the low-complexity BFGS detection algorithms,this dissertation adopts the simple diagonal matrix initialization methods and the approximate inverse matrix initialization methods,respectively,which can improve the convergence speed and detection performance with the acceptable computational complexity increasement.In addition,through in-depth analysis of the I-BFGS iterative process,SDBB that combines the Steepest Descent method(SD)and Barzilai-Borwein method(BB)is taken as the first iteration to obtain a more accurate search direction.Simulation results show that the proposed high-performance BFGS detection algorithm is superior to other joint detection algorithms in terms of computational complexity and detection performance under the harsh channel propagation environment.(4)The key to the practical application of the massive MIMO detection technology is not only to determine the appropriate detection algorithm,but more importantly,to map the algorithm into a high-efficient hardware architecture.Based on the I-BFGS detection algorithm,a VLSI implementation scheme is designed for massive MIMO signal detection with 128×8 antenna scale and 64-QAM modulation mode.Firstly,the overall hardware architecture is built,and then the sub-modules are divided and designed according to different functions in order to save hardware resources as much as possible.The key operations including the matrix-vector multiplication and the reciprocal calculation are realized by the regular systolic array with the dual operating modes and Newton-Rapson method of the unified hardware circuit,respectively.The calculation accuracy and hardware overhead are evaluated by MATLAB simulation so as to determine the appropriate fixed-point bit width.FPGA implementation results show that the proposed low-complexity BFGS hardware implementation has advantages in both detection performance and hardware efficiency compared with the existing linear detection hardware implementation.Through the above work,this dissertation solves the problem that BFGS quasiNewton method,as the most popular and efficient unconstrained optimization algorithm,cannot be effectively applied to the massive MIMO signal detection,which provides a new idea and lays a foundation for the practical application of the massive MIMO linear signal detection technology with low complexity and high performance.