基于自组织神经网络的非平稳信号盲分离  

Blind separation of non-stationary sources based on self-organizing neural network

作　　者：徐洪涛[1] 王跃钢[1] 邓卫强[1] 

机构地区：[1]第二炮兵工程学院304教研室,西安710025

出　　处：《控制与决策》2011年第5期748-752,共5页Control and Decision

基　　金：国家安全重大基础研究项目(973-61355020301)

摘　　要：针对具有二阶非平稳特性的源信号盲分离问题,提出一种基于自组织神经网络的在线盲源分离新算法.利用自组织神经网络构建一种多层盲分离网络模型,以网络输出层信号的相关性为代价函数,采用自然梯度原理对网络参数进行学习,最小化该代价函数从而实现信号分离.将多层自组织神经网络和自然梯度原理相结合,提高了分离算法的灵活性和性能.最后将该算法与其他算法进行了仿真对比,仿真结果表明该算法具有较好的收敛精度及稳定性.Aimming to the blind separation of non-stationary sources,a new on-line algorithm base on self-organizing network is proposed.The separation structure and cost function are established by utilizing the multilayer networks.The learning rule for the network’s parameters is derived from the nature gradient descent minimization of the cost function that takes the minimum only when the network outputs are uncorrelated with each other.The multilayer networks and nature gradient principal are combined,which improves the performance in non-stationary signals separation.Finally,simulation results show that the proposed algorithm has good convergence precision and stability.

关 键 词：盲源分离 自组织网络 非平稳信号 自然梯度 

分 类 号：TP911[自动化与计算机技术]