利用奇异值分解的鲁棒盲信源分离

J4 ›› 2009, Vol. 36 ›› Issue (6): 985-989.

利用奇异值分解的鲁棒盲信源分离

张伟涛¹;楼顺天¹;张延良^1,2

(1. 西安电子科技大学电子工程学院，陕西西安 710071
2. 河南理工大学计算机科学与技术学院，河南焦作 454001)

收稿日期:2008-09-02 出版日期:2009-12-20 发布日期:2010-01-20
通讯作者: 张伟涛
作者简介:张伟涛(1983-)，男，西安电子科技大学博士研究生，E-mail: zhwt-work@qq.com．
基金资助:
国家自然科学基金资助(60775013)

SVD based robust approach for blind source separation

ZHANG Wei-tao¹;LOU Shun-tian¹;ZHANG Yan-liang^1,2

(1. School of Electronic Engineering, Xidian Univ., Xi'an 710071， China
2. Dept. of Computer Science and Tech., Henan Polytechnic Univ., Jiaozuo 454001， China)

Received:2008-09-02 Online:2009-12-20 Published:2010-01-20
Contact: ZHANG Wei-tao

摘要/Abstract

摘要：

利用Amari等提出的估计函数分析了在最小化非线性主分量标准的意义上分离矩阵的最优解，将其描述为归一化的输入与非线性输出的互相关，提出了一种采用奇异值分解的鲁棒盲信源分离方法．该方法将分离矩阵的估计解释为一种非线性的能量迭代问题．在求解过程中对能量项进行了奇异值分解，从而避免了归一化矩阵的求逆以及计算平方根问题，减小了运算量．由于使用了能量项的正定逆平方根来归一化分离矩阵，因此提高了算法的鲁棒性，并且在每一步迭代中都能使分离矩阵保持正交．实验仿真验证了算法的性能．

关键词: 盲信源分离, 奇异值分解, 非线性能量迭代, 预白化

Abstract:

From the point of view of the estimating function proposed by Amari, we investigate the intrinsic characteristics of the optimal separating matrix for blind source separation (BSS) in the sense of minimizing nonlinear principal component analysis criterion, which formulates the optimal solution as the normalized cross correlation between the input and nonlinearized output. We thus present a singular value decomposition (SVD) based robust scheme for BSS, which considers the estimating procedure of the separating matrix as a nonlinear power iteration problem. By performing SVD of the power term, the computational load can be significantly reduced, which results from circumventing the difficulty of solving the inverse square root of the normalization term. Since the separating matrix is properly normalized by the positive definite inverse square root of the power term, the robustness of this algorithm is greatly improved. Moreover, this guarantees the orthonormality of the separating matrix at each iteration. Some simulation results are also provided to demonstrate the performance of the proposed algorithm.

Key words: blind source separation (BSS), singular value decomposition (SVD), nonlinear power iteration (NPI), pre-whitening

中图分类号:

TN911.7

张伟涛;楼顺天;张延良. 利用奇异值分解的鲁棒盲信源分离[J]. J4, 2009, 36(6): 985-989.

ZHANG Wei-tao;LOU Shun-tian;ZHANG Yan-liang. SVD based robust approach for blind source separation[J]. J4, 2009, 36(6): 985-989.

参考文献

［1］Hyvarinen A, Karhunen J, Oja E. Independent Component Analysis ［M］. New York: Wiley, 2001.
［2］Wentrup M G, Buss M. Subspace Identification from Blind Source Separation ［J］. IEEE Signal Process Lett, 2006, 13(2): 100-103.
［3］Cardoso J F, Laheld H. Equivariant Adaptive Source Separation ［J］. IEEE Trans on Signal Process, 1996, 44(2): 3017-3029.
［4］Yang H H, Amari S. Adaptive On-line Learning Algorithms for Blind Separation-maximum Entropy and Minimum Mutual Information ［J］. Neural Computation, 1997, 9(1): 1457-1482.
［5］Aamri S, Cichocki A. Adaptive Blind Signal Processing-neural Network Approaches［J］. Proceedings of the IEEE, 1998, 86(10): 2026-2048.
［6］Yang B. Projection Approximation Subspace Tracking ［J］. IEEE Trans on Signal Process, 1995, 43(1): 95-107.
［7］Pajunen P, Karhunen J. Least-squares Methods for Blind Source Separation Based on Nonlinear PCA ［J］. Int J Neural Syst, 1998(8): 601-612.
［8］Karhunen J, Pajunen P, Oja E. The Nonlinear PCA Criterion in Blind Source Separation: Relations with Other Approaches ［J］. Neurocomputing, 1998, 22(1): 5-20.
［9］Zhu X L, Zhang X D. Adaptive RLS Algorithm for Blind Source Separation Using a Natural Gradient ［J］. IEEE Signal Process Lett, 2002, 9(12): 432-435.
［10］Abrudan T E, Eriksson J, Koivunen V. Steepest Descent Algorithms for Optimization under Unitary Matrix Constraint ［J］. IEEE Trans on Signal Process, 2008, 56(3): 1134-1147.
［11］Amari S, Cardoso J F. Blind Source Separation-semi Parametric Statistical Approach ［J］. IEEE Trans on Signal Process, 1997, 45(11): 2692-2700.
［12］Hua Y, Xiang Y, Chen T, et al. A New Look at the Power Method for Fast Subspace Tracking ［J］. Dig Signal Process, 1999, 9(2): 297-314.
［13］Badeau R, David B, Richard G. Fast Approximated Power Iteration Subspace Tracking ［J］. IEEE Trans on Signal Process, 2005, 53(8): 2931-3941.
［14］Feng D Z, Zheng W X, Cichocki A. Matrix-group Algorithm Via Improved Whitening Process for Extracting Statistically Independent Sources from Array Signals ［J］. IEEE Trans on Signal Process, 2007, 55(3): 962-977.
［15］Golub G H, Loan C F V. Matrix Computations ［M］. 3rd Ed. Baltimore: The Johns Hopkins University Press, 1996.
［16］Willink T J. Efficient Adaptive SVD Algorithm for MIMO Applications ［J］. IEEE Trans on Signal Process, 2008, 56(2): 615-622.

[1]	梁壮,温利武,丁金闪. 改进型SVD-FRFT海杂波抑制方法[J]. 西安电子科技大学学报, 2021, 48(2): 55-63.
[2]	蒋行国,万今朝,蔡晓东,李海鸥,曹艺. 奇异值分解与中心度量的细粒度车型识别算法[J]. 西安电子科技大学学报, 2019, 46(3): 82-88.
[3]	郭欣;贾海蓉;王栋. 利用子空间改进的K-SVD语音增强算法[J]. 西安电子科技大学学报, 2016, 43(6): 109-115.
[4]	陈溅来;张锐;孙光才;景国彬;邢孟道. 地球同步轨道SAR方位空变校正方法[J]. 西安电子科技大学学报, 2016, 43(5): 18-23+62.
[5]	霍雷刚;冯象初;王旭东;霍春雷. 高阶SVD和全变差正则的乘性噪声去除模型[J]. 西安电子科技大学学报, 2016, 43(3): 78-84.
[6]	刘圣恩;肖霖;杨鼎成;谭安平. MIMO双向中继GSVD线性预编码的功率优化算法[J]. J4, 2015, 42(5): 161-167.
[7]	朱凯然;何学辉;郑纪彬;苏涛. 多模块中核四极矩共振信号参数估计新方法[J]. J4, 2013, 40(3): 36-41.
[8]	吴泽民;蒋叶金;任姝婕. 拓扑序列航迹相关的高效修正算法[J]. J4, 2011, 38(2): 180-186.
[9]	张伟涛;刘宁;楼顺天. 避免奇异解联合对角化的两种高效算法[J]. J4, 2009, 36(5): 857-861.
[10]	张延良1,楼顺天2,张伟涛3. 用于统计相关源信号的盲分离方法[J]. J4, 2009, 36(3): 401-432.
[11]	保宏;赵冬竹;乔永强. 一种正交向量基结构动力模型修正 [J]. J4, 2009, 36(1): 151-155.
[12]	张伟涛1;楼顺天1;张延良1;2. 具有等变化性的最小二乘盲信源分离方法 [J]. J4, 2008, 35(6): 999-1002.
[13]	张志安;骆斌. 一种新的颜色空间特征图像检索 [J]. J4, 2008, 35(3): 577-580.
[14]	张卓奎;陈慧婵. 广义离散随机非线性系统的递推算法 [J]. J4, 2007, 34(2): 317-321.
[15]	张卓奎;陈慧婵. 广义离散随机线性系统的最优递推问题[J]. J4, 2006, 33(2): 273-277.