一种基于时间相关度的盲分离方法

An Algorithm of Blind Source Separation Based on Time Correlation

本文从信号的相关性出发,提出信号时间相关度的定义,并证明了时间相关度定义具有如下性质:一组独立源信号的线性混合信号其时间相关度介于源信号中时间相关度的最小值和最大值之间。本文根据这一定义及性质,利用矩阵广义特征值理论,建立时间相关度的广义特征值问题,通过求解此广义特征值问题的特征矢量,从而达到分离信号的目的。本文算法适用于瞬时混合模型,仿真结果证明本算法可以应用与灰度图像的分离及复杂混合环境中声音的分离,计算简单,性能良好,效果真实可靠。

A concept of signal time correlation is given in this paper. It is proved that a linear mixture of statistically uncorrelated source signals has this property：the signal time correlation of any mixed signal is between the minimal and maximal value of its compo- nent source signals. Following this definition and property, a problem about generalized eigenvalue is obtained by using generalized matrix eigenvalue theory. The goal of blind separation of signals is achieved by getting the eigenvectors of the generalized eigenvalue problem. This algorithm is based on instantaneous-mixed module. This algorithm has been proved to adapt to the separation of black images and the separation of voices in complex environment. The calculation is simple;performance is all right and effect is good.