一类新的独立性度量及其在盲信号分离中的应用

A Class of New Independence Measures and Their Application to Blind Signal Separation

提出了一类无穷多种称为准熵的新的独立性度量 ,它们用严格凸函数对原变量经分布函数变换再量化后得到的变量的联合概率的均匀性进行度量 ,并提出了基于准熵的盲分离算法 ,可分离任意连续分布的信号 ,包括峭度为零的信号。通过与前人算法的对比试验 。

A class of new infinitely many independence measures named quasi\|entropy (QE) is proposed, in which strictly convex functions are used to evaluate the uniformity of the joint probability of the variables. In QE, none of priori assumptions is made on the shape or statistical features of the distribution functions of continuous variables but unbiased estimates of the values of distribution functions are obtained from the samples. Therefore, blind separation algorithms based on QE can separate signals with arbitrary continuous distributions, including those with zero kurtoses. The superior performance of the QE\|based algorithms is verified by comparison experiments with previous algorithms.