中文摘要：

从盲源信号分离后非高斯性最大化出发，提出了一种基于经验特征函数的盲源信号分离方法。该方法把经验特征函数与概率密度函数一一对应，并以混合信号与高斯信号的经验特征函数的欧氏距离最大化作为判据，以固定点算法为优化算法进行盲源分离。该方法克服了FastICA算法中选取不同的近似函数对不同概率密度分布的信号效果不佳的问题。仿真实验结果表明，与常用的几种FastICA算法相比，该方法具有更好的收敛效果。采用新的盲源信号分离方法对管道破坏产生的实际声发射信号进行分离，可将破坏点互相关定位精度提高到3％以上。

英文摘要：

Started off with the view of non-Gauss＇s maximum of the blind acoustic sources, a new approach of blind source separation （BSS） is put forward. This method is based on the experience character function （ECF）, which one-to one correspond with probability density function. And more, the maximum of Euclidean distance of blind source＇s ECF and Gauss function＇s ECF used as the criterion, the fixed-point algorithm used to iteration. Compared with the approach of FastICA, the baffle of choice of different constraint functions is unnecessary. The performance of the method is verified by the application to separating the mixing signals of pipe damage and the accuracy is increased to less than 3%.