基于自然梯度的独立子空间盲信号处理方法*

南京大学学报(自然科学版) ›› 2011, Vol. 47 ›› Issue (4): 420–425.

基于自然梯度的独立子空间盲信号处理方法*

王法松^1,2**，张林让¹

出版日期:2015-04-16 发布日期:2015-04-16
作者简介: (1.西安电子科技大学需达信号处理国防科技重点实验室，西安,710071;
2.中国电子科技集团公司第二十七研究所，郑州，450047)
基金资助:
国家自然科学基金(61071188)，国家高技术研究发展计划(2010AAXXX1402B)

Independent subspace analysis for blind signal separation using natural gradient algorithm

Wang Fa-Song¹²,Zhang Lin-Rang¹

Online:2015-04-16 Published:2015-04-16
About author: (1 .National Key Laboratory for Radar Signal Processing，Xidian University, Xi,an,710071.china; 2. China Electronics Technology Uroup Corporation, the 27^th Research institute, Zhengzhou，450047，China)

摘要/Abstract

摘要： 作为自信号处理的独立成分分析方法的扩展，独立子空间分析具有更广阔的应用前景.本文首先给出了独立子空间分析的一般定义和正则化定义，同时把其与独立成分分析方法进行了对比.此
外，讨论了独立子空间分析的可分离性与解的唯一性问题.基于极大似然估计和自然梯度方法，木文给七了独立子空间分析的自然梯度算法.仿真实验通过二维的独立子空间分析说明木文提出算法的有效性.

Abstract: Standard blind signal separation(BSS) model and methods have been successfully applied to many areas of scienceThe basic model assumes that the observed signals arc linear superpositions of underlying hidden source
signals. Most of the blind signal separation algorithms arc based on the independent assumption of the source signals, and arc called independent component analysisICA). However, the independence property of sources may
not hold in some real-world situations, especially in biomedical signal processing and image processing, and therefore the standard independent component analysis cannot give the expected results. Some techniques have been
developed in recent years that relax the assumptions of basic independent component analysis model and generalize the independent component analysis problem. Among many extensions of the basic independent component analysis
model，several researchers have studied the case where the source signals arc not statistically independent. Related models arc generally recognized as dependent component analysis(DCA)model. As an extended independent
component analysis method for blind signal separation, independent subspace analysis has more applications than the independent component analysis.The general and detailed definition of the independent subspace analysis(ISA)
model is given at first and the relationship between independent component analysis and independent subspace analysis methods is also discussed. Moreover, the separateness and uniqueness of the independent subspace analysis
model is discussed. Based on the maximum likelihood theory and natural gradient method，the natural gradient separation algorithm for independent subspace analysis model is constructed. Simulation result shows that the
proposed algorithm is able to separate the independent subspace analysis mixed source signals.

王法松^1,2**，张林让¹. 基于自然梯度的独立子空间盲信号处理方法*

[J]. 南京大学学报(自然科学版), 2011, 47(4): 420–425.

Wang Fa-Song¹²,Zhang Lin-Rang¹
. Independent subspace analysis for blind signal separation using natural gradient algorithm[J]. Journal of Nanjing University(Natural Sciences), 2011, 47(4): 420–425.

参考文献

[1]Cichocki A，Amari S. Adaptive blind signal and image processing; Learning algorithms and ap plications. New York: John Wilcy and Sons，2002.
[2]Comon P，Jutten C. Handbook of blind source separation independent component analysis and applications. Oxford:Elscvicr, 2010.
[3]Li R，Li H W, Wang F S. Dependent compo nent analysis; Concepts and main algorithms. Journal of computers, 2010，5(4):589一597.
[4]Poczos B, Lorincz A, independent subspace a- nalysis using k-nearest neighborhood distances. Lecture Notes in Computer Science, 2005,3697:163一168.
[5]Hyvarincn A，Kostcr U. FastlSA:A fast fixed-point algorithm for independent subspace analysis. Proceedings of European Symposium on Artificial Neural Networks, Bruges, Bel- gium, 2006，371一376.
[6]Thcis F J.Towards a general independent sub space analysis. Proceedings of Neural lnforma tion Processing Systems, 2007，1361一1368.
[7]Yach F R，Jordan M 1. Kernel independent component analysis. Journal of Machine Learn- ing Research, 2002，3:1一18.
[8]Hyvarincn A，Hoycr P. Emergence of phase and shift invariant features by decomposition of natural images into independent feature sub- spaces. Neural Computation, 2000，12:170一1720.
[9]Cardoso J F. Multidimensional independent component analysis. Proceedings of the IEEE 1998 international Conference on Acoustics, Spccch，and Signal Processing, 1998，1941一1944.
[10]Thcis F J. Blind signal separation into groups of dependent signals using joint block diagonaliza- tion. Proceedings of international Symposium on Circuits and Systems, 2005，5875一5881.
[11]Blanchard U, Kawanabc M, Sugiyama M，et al. In search of norrgaussian components of a high-dimensional distribution. Journal of Ma- chine Learning R escarch，2006，7:247一282.
[12]Theis F J. Uniqueness of complex and multidi mensional independent component analysis. Sig nal Processing, 200，84(5):951一956.
[13]Choi H，Choi S. R dative gradient learning for independent subspace analysis. Proceedings of international Joint Conference on Neural Net- works，2006，3919一3924.
[14]Szabo Z, Poczos B, Lorincz A. Auto-regressive independent process analysis without combina- torial efforts. Pattern Analysis and Applica tions 2010，13:1一13.
[15]Szabo Z，Poczos B, Lorincz A. Undercomplete blind subspace deconvolution. Journal of Ma chine Learning R escarch. 2007，8:1063一1095.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed