粒子滤波和独立分量分析的含噪信号盲分离算法研究  

Research on particle filtering and independent component analysis based noisy blind source separation

作　　者：王震[1] 丛丰裕[1] 李锋[2] 周士弘[2] 史习智[1] 奚立峰[1] 

机构地区：[1]上海交通大学机械与动力工程学院,上海200240 [2]杭州应用声学研究所声纳技术国防科技重点实验室,杭州310012

出　　处：《声学技术》2008年第3期423-427,共5页Technical Acoustics

基　　金：上海市科委基础研究项目(No:05JC14026);国家重点实验室开放式基金项目(VSN-2006-04);国防科技重点实验室基金项目(9140C2404060707)

摘　　要：研究了两阶段含噪独立分量分析算法来解决含噪信号盲分离问题。第一阶段,通过粒子滤波实现对不含噪混合信号的估计,将含噪独立分量分析转化为不含噪的独立分量分析;第二阶段用现有的FastICA算法从估计的不含嗓混合信号中提取出源信号。不含噪混合信号的时变自回归模型和含噪与不含噪混合信号之间的关系构造了动态的状态-空间方程。该方程的特点是多变量、过程和观测噪声不限于高斯分布,粒子滤波是解决该问题的有效方法。提出了解决含噪独立分量分析的PF+FastICA算法,仿真试验表明所提出的算法性能优于相关文献的结果。A two-stage approach is studied to resolve the noisy independent component analysis （ICA） for noisy blind source separation. In the first stage, particle filtering （PF） is used to estimate noise-free mixtures, and turn the noisy ICA to noise-free ICA. Thereafter, Fast ICA is adopted to extract the independent components from the estimated clean mixtures. The time-varying autoregressive model of clean mixtures and the relationships between noisy mixtures and clean mixtures compose the dynamic state space equations. The characteristics of the equations are multivariable, moreover, the process and observed noises are not restricted to be gaussian. Due to above reasons, particle filtering is applied. Thus, PF＋ FastICA is put forward. The simulation proves that PF＋FastICA outperforms Denoising Source Separation.

关 键 词：粒子滤波 含噪混合信号 独立分量分析 时变自回归模型 

分 类 号：TN911.72[电子电信—通信与信息系统]