基于随机梯度的变动量因子自适应白化算法  

Stochastic Gradient Based Variable Momentum Factor Algorithm for Adaptive Whitening

作　　者：欧世峰[1] 高颖[1] 赵晓晖[2] 

机构地区：[1]烟台大学光电信息科学技术学院,烟台264005 [2]吉林大学通信工程学院,长春130012

出　　处：《自动化学报》2012年第8期1370-1374,共5页Acta Automatica Sinica

基　　金：国家自然科学基金(61005021)资助~~

摘　　要：针对自适应白化技术中算法的收敛速度问题,通过融入具有变动量因子特性的动量项,提出了一种快速的自适应白化算法.该算法利用动量项来加速系统的收敛速度,并基于随机梯度方法对动量因子进行自适应更新,有效提升了白化系统的整体性能.仿真实验表明本文算法在平稳和非平稳环境下具有良好的性能.In view of the convergence speed in adaptive whiten- ing technique, by adding a momentum term with a variable mo- mentum factor, a fast adaptive whitening algorithm is proposed. First, the presented algorithm incorporates a momentum term into the adaptive whitening algorithm to accelerate the conver- gence speed. Then, a variable momentum factor to improve the allover performance of the whitening system is obtained based on the stochastic gradient of the cost function. Experimental results demonstrate the good performance of the proposed fast algorithm both in stationary and non-stationary conditions.

关 键 词：白化 动量项 自适应 盲源分离 

分 类 号：TN911.7[电子电信—通信与信息系统]