一种受噪声干扰的基于矩阵群白化的改进ICA算法  

An Improved ICA Algorithm Based on Matrix Group Robust Whitening under Noise Interference

作　　者：朱义勇[1] 熊焕宇[1] 

机构地区：[1]国防信息学院,武汉430010

出　　处：《现代军事通信》2013年第4期20-24,共5页

摘　　要：对于混合信号中同时存在超高斯与次高斯信号的盲源分离问题，经典的FastICA、ICAML算法的激活函数只能对应于一种概率分布，难以实现混合信号的有效分离；Pearson—ICA和JADE算法可以分离一大类对称或非对称分布的源信号，但在受噪声干扰时，如果采用常规的PCA算法进行白化，分离结果会有较大的误差。论文将基于矩阵群的白化算法与EFICA算法相结合，推导得到了一种适用于正定与超定混合的情况的灵活ICA算法，仿真表明在混合信号受白噪声干扰时，这种算法相对于其他批处理算法拥有更好的分离性能。算法的研究对于探索盲信号分离技术在军事通信抗干扰中的应用具有一定的借鉴作用。In the application of blind signal separation in which super-Gaussian and sub-Gaussian signals si-multaneously exist, it is difficult for classical algorithms such as FastICA and ICAML to separate signalseffectively. Pearson-ICA and JADE can separate many kinds of signals of which the probability densityfunctions are asymmetric or symmetric, but the separation results are inaccurate if the conventional PCAalgorithm is used for whitening processing when noise exists. A novel flexible ICA algorithm is deducedbased on the combination of matrix group whitening and EFICA algorithm which is fit for well-determinedand over-determined mixture. Simulation results indicate that the proposed algorithm is superior to otherbatch algorithms when white noise exists in mixture signals. The research result has some contributions toblind signal separation techniques used in military anti-jamming communications.

关 键 词：超高斯 次高斯 矩阵群 鲁棒白化 灵活ICA 

分 类 号：TN911.7[电子电信—通信与信息系统]