摘要
MUSIC算法具有很高的分辨率、估计精度和稳定性,但对接收的矩阵进行奇异值分解以及谱峰搜索时运算量巨大,CPU以及FPGA在计算速度上存在不足。利用GPU并行计算从复数奇异值分析、使用共享内存、分配锁页内存、增加流操作等方面进行改进来提高MUSIC算法的计算效率。计算机仿真实验结果表明,MUSIC算法的GPU高效实现能够实现MUSIC算法的准确快速计算。
The MUSIC algorithm has high resolution,estimation accuracy and stability,but the operating quantity is huge when the singular value decomposition and spectral peak search of the received matrix are performed,and the CPU and field programmable gatearray(FPGA)have disdvantages in the calculation speed.This paper uses GPU parallel computing to improve the computational efficiency of MUSIC algorithm by improving complex singular value analysis,using shared memory,allocating lock page memory and increasing stream operation,etc..The computer simulation experiment results show that the efficient GPU implementation of MUSIC algorithm can realize the accurate and rapid calculation of MUSIC algorithm.
引文
[1]贾维敏,姚敏立,金伟.阵列信号参数估计及应用[M].北京:北京理工大学出版社,2013.
[2]白银山,李宏.MUSIC算法的FPGA实现[J].计算机系统应用,2014,23(7):185-186.
[3]SCHMIDT R O.Multiple emitter location and signal parameter estimation[J].IEEE Trans.on Antennas and Propagation,1986,34(3):276-280.
[4]张超,黎仁刚,顾军.基于多级维纳滤波的ESPRIT算法[J].舰船电子对抗,2017,40(1):74-75.
[5]谈文韬,林明,黎仁刚.奇偶阵元数均匀圆阵测向性能研究[J].现代雷达,2016,38(11):24-29.
[6]郭强.并行JACOBI方法求解矩阵奇异值的研究[D].苏州:苏州大学,2011.
[7]胡以怀,周轶尘.一般复数矩阵SVD算法[J].武汉理工学学报,1995,19(2):150-154.
[8]COOK S.CUDA并行程序设计GPU编程指南[M].苏统华译.北京:机械工业出版社,2016.