一种基于自然梯度的循环平稳信号盲分离方法及其应用  

Blind Source Separation of Cyclo-stationary Signals Using an Extended Natural Gradient Method and Its Application

作　　者：郭彬[1] 张永祥[1] 柯维[1] 

机构地区：[1]海军工程大学船舶与动力学院

出　　处：《船海工程》2011年第1期111-114,117,共5页Ship & Ocean Engineering

基　　金：国家自然科学基金(50979109)

摘　　要：建立瞬时混合模型下的代价函数,通过改进自然梯度算法的迭代极小化代价函数以恢复源信号。仿真分析证明该算法的有效性,与传统的ICA算法相比有更好的效果和收敛速度;滚动轴承故障实例分析证明,该算法能有效分离内圈故障信号。The method of blind source separation of signal about cyclo-stationary was derived by applying natural gradient iterative learning to the novel cost function defined according to instantaneous mixtures.The efficiency of the algorithm was supported by simulations,which showed that the proposed method has improved performance for the separation in terms of waveform similarity and convergence speed,as compared to the traditional ICA method.In the fault rolling element bearing experiment,fault signal is extracted effectively.

关 键 词：循环平稳 盲分离 自然梯度法 循环自相关函数 

分 类 号：U676.4[交通运输工程—船舶及航道工程]