基于核密度估计的信号盲处理

Blind Source Separation Based on Kernel Density Estimation

在信号的盲处理中，常用一个非线性函数来代替算法中不可知的评价函数。针对不同统计特性的源信号，需要选择不同的非线性函数。核密度估计法可以用来对评价函数直接做出估计，从而避免了对非线性函数的选择。它使得盲处理算法可以成功地恢复出包含不同统计特性的杂系混合信号。因此，它更加接近于实际的应用。

A non-linear function is usually used to replace the unknown evaluation function in the algorithm. For source signals with different statistics features different non-linear functions should be selected. The kernel density estimation can be used to evaluate directly the evaluation functions so as to avoid the selection of non-linear functions. Thus it makes the blind process algorithm possible to separate successfully hybrid mixed signals with different statistics features in practice.