An efficient semi-blind source extraction algorithm and its applications to biomedical signal extraction

作     者：YE YaLan SHEU Phillip C-Y ZENG JiaZhi WANG Gang LUKe 

作者机构：School of Computer Science and Engineering University of Electronic Science and Technology of China Chengdu 610054 China Department of Electrical Engineering and Computer Science University of California Irvine CA 92697-2625 USA School of Life Science and Technology University of Electronic Science and Technology of China Chengdu 610054 China 

出 版 物：《Science in China(Series F)》 (中国科学（F辑英文版）)

年 卷 期：2009年第52卷第10期

页      面：1863-1874页

核心收录：

学科分类：12[管理学] 1201[管理学-管理科学与工程(可授管理学、工学学位)] 07[理学] 070205[理学-凝聚态物理] 08[工学] 080501[工学-材料物理与化学] 0805[工学-材料科学与工程（可授工学、理学学位）] 0702[理学-物理学] 

基　　金：Supported by the National Natural Science Foundation of China (Grant No. 60702072)  and China Scholarship Council 

主　　题：blind source extraction blind source separation independent component analysis electrocardiogram fetal ECG Atrial Fibrillation 

摘      要：In many applications, such as biomedical engineering, it is often required to extract a desired signal instead of all source signals. This can be achieved by blind source extraction （BSE） or semi-blind source extraction, which is a powerful technique emerging from the neural network field. In this paper, we propose an efficient semi-blind source extraction algorithm to extract a desired source signal as its first output signal by using a priori information about its kurtosis range. The algorithm is robust to outliers and spiky noise because of adopting a classical robust contrast function. And it is also robust to the estimation errors of the kurtosis range of the desired signal providing the estimation errors are not large. The algorithm has good extraction performance, even in some poor situations when the kurtosis values of some source signals are very close to each other. Its convergence stability and robustness are theoretically analyzed. Simulations and experiments on artificial generated data and real-world data have confirmed these results.