基于非参数化步长最速下降的盲源抽取算法

Blind Source Extraction Using Method of Steepest Descentwith an Parameter-free Adaptive Step-size

为了从混叠的观测信号中抽取单个源信号,提出一种基于二阶统计量和最速下降优化技术相结合的盲抽取方法。该方法首先根据源信号的二阶统计特性构造用于估计抽取权向量的代价函数,然后采用非参数化步长的最速下降更新规则进行优化,从而获得最优的抽取权向量。仿真实验结果表明,该算法能有效地逐个分离出平稳和非平稳的信号,同时也能实现在线盲信号抽取。

The blind source extraction of instantaneous mixtures is accomplished by using second-order statistics and steepest descent techniques. The cost functions for estimating weight vectors are constructed, based on the second order statistical properties of source signals, and the method of steepest descent with parameter-free adaptive step- size is used to optimize the extracted weight vector. Simulations indicate that this algorithm is able to separate sta tionary and non-stationary signals effectively and good for on-line blind signal extraction.