MUSIC算法的DSP实现与优化

Implementation and Optimization of MUSIC Algorithm on DSP

针对MUSIC算法在DSP实现中矩阵特征值计算和复数矩阵运算复杂的问题,本文把复矩阵转换为实矩阵并利用一种QR变换法计算特征值。该方法是一种迭代方法,在每次迭代计算的过程中,若次对角线元素小于设置精度就可以把它看作零,将矩阵分割为不可约赫申伯格矩阵(Hessen berg)即上H矩阵,以便逐步降低主子阵的阶数,从而减少计算量。此外本文针对定点DSP进行优化,对数据进行整型处理再进行运算,减少计算时间。最后本文对毫米波雷达实测数据在DSP和MATLAB中进行运算对比,运算结果证明了本文所提优化算法的有效性和实用性。

In view of the complexity of matrix eigenvalue calculation and complex matrix operation in DSP implementation of MUSIC, this paper proposes a QR transform method and transforms complex matrix into real matrix to compute eigenvalues. This method is an iterative method. In each iteration of the calculation process, the values of the secondary diagonal elements smaller than the specified value are set as zero. The matrix is divided into irreducible Hessenberg matrix, so as to gradually reduce the order of principal submatrix, thus reducing the amount of computation. In addition, the data are processed by integer, and then the computing time is reduced. In this paper, the measured data of millimeter wave radar are compared in DSP and MATLAB, and the effectiveness and practicability of the optimization algorithm are proved.