基于极大似然偏差去除的FastICA算法  

FastICA Algorithm Based on Maximum Likelihood Estimation and Bias Removal Technique

作　　者：刘克 田学民[1] 蔡连芳 张银雪[1] 

机构地区：[1]中国石油大学(华东)信息与控制工程学院,山东青岛266580

出　　处：《计算机仿真》2013年第12期178-181,252,共5页Computer Simulation

基　　金：国家自然科学基金项目(61273160);山东省自然科学基金项目(ZR2011FM014);中央高校基本科研业务费专项资金(27R1205005A)

摘　　要：在信源去噪优化问题的研究中,噪声会直接影响ICA的分离性能。针对有噪情况下混合矩阵估计准确性下降的问题,提出了一种极大似然偏差去除的FastICA算法,考虑含高斯噪声的瞬时混合ICA模型,以观测信号的似然度为目标函数,采用偏差去除技术对目标函数进行修正,以减少由噪声引起的偏差,然后采用固定点算法对混合矩阵进行寻优。仿真结果表明,与其它两种常用算法相比,上述算法可以更为精确地估计混合矩阵,显著减少迭代次数,能够较好地解决有噪独立分量分析问题。Most ICA algorithms are based on the noise -free model. However, the noise will affect the separation performance of ICA directly. Aiming at the problem of poor estimation of mixing matrix in noisy conditions, a FastI- CA algorithm was proposed based on the maximum likelihood estimation and bias removal technique. Considering the instantaneous mixture ICA model with Gaussian noise, it regards likelihood of the observed signals as the objective function. Through the modification to the objective function by bias removal technique, it can reduce the bias caused by the noise. And the mixing matrix was optimized by the fixed - point algorithm. Simulation results illustrate that, in contrast to the other two common algorithms, this method can estimate the mixing matrix parameters more accurately, reduce the number of iterations dramatically and solve the noisy ICA problems.

关 键 词：独立分量分析 极大似然估计 偏差去除 固定点算法 混合矩阵 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]