针对多通道解相关算法在混合信号协方差矩阵的最大特征值较大或者最大与最小特征值的比值很大(大于106的病态混合)时收敛速度变慢且收敛误差增大的不足,本文引入归一化方法,即在迭代过程中对解混信号进行归一化,从而限制其协方差矩阵最大特征值的取值范围,且降低最大与最小特征值的比值.数值仿真表明,改进后的算法在降低迭代误差和加强稳定性方面有了明显的改善,收敛速度也得到了可观的提升,改进算法也具有更广的适应范围.%As the maximal eigenvalue of the mixture's covariance matrix becomes very large or the ratio of the maximal eigenvalue to its minimal eigenvalue is very large (which is larger than 106 as ill-conditioned mixture), the current blind multi-channel signal decorrelation algorithms tend to be slow in convergence and large in decorrelation error. In this paper, we propose a new normalization algorithm which restricts the range of the covariance matrix's maximal eigenvalue and depresses the ratio of the maximal eigenvalue to its minimal eigenvalue. The new algorithm can greatly improve the convergence speed, decorrelation error, and computational stability of the blind multi-channel signal decorrelation. The simulations verify these advantages of the proposed algorithm.
展开▼
