基于规范高阶累积量的盲分离算法

本文给出信号的r阶规范累积量定义,证明了在信号瞬时线性混叠情况下r阶规范累积量绝对值不超过最大源信号的r阶规范累积量,因此可以通过最大化r阶规范累积量的绝对值实现盲分离.最大化r阶规范累积量的绝对值可以得到两种盲分离算法,一种高效快速算法———特征值分解盲分离算法,另外一种盲提取算法.本文为高阶累积量盲分离奠定了严格的理论基础.仿真结果验证了理论的正确性和算法的有效性.

Normalized Cummulant of Signal and Higher Order Cumulant Algorithm of Blind Source Separation

The rth-order normalized cumulant of signal is defined in this paper,and it has a very important property that the maximal absolute value of r th-order normalized cumulant of source signals is larger or equals to that of mixing signal for instantaneous BSS.Based on it,two BSS algorithms are proposed.The first algorithm works via eigen-decomposition and is fast and is of the highly accuracy,and another is a blind-extracted algorithm.This paper establishes rigorous theory basis for BSS approaches of higher order cumulant.The simulation results demonstrate that the theory of this paper is correct and algorithms are valid.