一种基于负熵的信赖区域盲分离方法

A method of blind source separation based on maximizing the trust region of negative entropy

研究基于瞬时混合信号的盲分离问题。以负熵作为代价函数,提出了一种基于最大化负熵的信赖区域盲源分离方法,该方法保持了牛顿法的快速收敛特性,具有全局收敛性。分别用基于负熵的信赖区域法和牛顿算法对三路随机混合信号实施分离,仿真结果证明该方法且具有和牛顿算法一样的收敛速度,而稳态误差比牛顿算法减小约7dB,因此具有更好的稳态性能。

Blind separation of linearly mixed signals is studied in this paper. A method of blind source separation based on maximizing the trust region of negative entropy is proposed.This approach not only retains the fast convergence characteristics of Newton method,but also has global convergence.These two methods are both used for the separation of three-way random mixed signals.Simulation results show that trust region method has the same convergence rate as Newton method,while it has better steady state performance.The steady-state error can reduce about 7dB compared with that of the Newton method.