摘要
利用二阶统计量(不同时延相关矩阵)的非平稳性和信号时序结构特征,能简单估计出线性瞬时混合的盲源信号。但随时延τ增大,仅利用某一个时延协方差均衡化,忽略了信号的时间变化特性,很难保证算法的性能。通过分析矩阵的平均特征,提出一种改进的基于二阶统计量盲源分离算法,对一组均衡化的时延相关函数进行等时延分段,并对等间隔段的两个时延矩阵分别求取均值,采用类似联合近似对角化,估计出最优化的酉矩阵,最终得到信源的稳健估计。性能指标分析和仿真实验证明,算法很好地解决原算法的不足,并成功分离出信号。
It is known that using non-stationary and structural characteristics of signal timing of the second-order statistics（different time-delay correlation matrix）can roughly estimate the instantaneous linear-mixture blind-source signal.With the increasing of time delayτ,only using single delay covariant equalization will ignore the time-varying characteristics of the signal,so that the performance of algorithm may not be ensured.Through the analysis of average characteristics of ma-trix,the improved algorithm for blind source separation based on the second-order statistics is presented in this paper.Segmenting a set of equalized delay-related functions,calculating the mean values of the two time-delay matrixes,and using a quasi joint approximate diagonalization,the optimal U-matrix can be estimated.Then,the robust source estimates can be obtained after all.Both performance index analysis and simulations show that the new algorithm can make up for the deficiencies of the original algorithm,and successfully separate the signal.
出处
《噪声与振动控制》
CSCD
北大核心
2011年第3期19-23,共5页
Noise and Vibration Control
关键词
振动与波
二阶统计量
均衡化
联合近似对角化
性能指标
vibration and wave
second-order statistics
equalization
joint approximate diagonalization
perfor-mance index
