一种基于非平稳特性的前馈神经网络盲源分离方法

A FEEDFORWARD NEURAL NETWORK FOR BLIND SOURCE SEPARATION

利用信息论原理，推导出一个衡量输出分量独立性的目标函数，最小化该目标函数并利用信号的非平稳特性，得到一种可以进行非平稳信号的盲分离的训练算法；指出了在特定情况下，给出的两种网络结构形式是等价的，由此得到的训练算法避免了矩阵求逆．

An information theory based on objective function is proposed ,which is a measurement of statistics independent.A learning algorithm for blind separation of nonstationary signals is obtained by minimizing the function,where the nonstationarity is applied.The analysis shows the equivalence of two neural networks architectures in some special cases,which can avoid calculating the inverse in the algorithm.The computer simulation shows the validity of the proposed algorithm.The performance surface of the object function is given at the last of this paper.