摘要
盲信号处理(BSP)是目前信号处理中最热门的新兴技术之一，它具有稳定的理论基础和许多方面的应用潜力。事实上，BSP已成为重要的研究课题，并在许多领域得到发展，尤其是在雷达、声呐、遥感、通信系统、噪声控制、医学成像、图像处理等领域。作为盲信号处理中的一部分，盲源分离或盲信号分离(BSS)也正成为信号和图像处理等领域的一个强有力的工具。BSS的目标是在没有任何或很少关于源信号和混合先验知识的前提下，从一组混合(观测)信号中恢复原始的信号。在本文中，首先研究和讨论了在线性瞬态混合BBS和卷积混合BSS中应用的各种方法，并通过采取不同技术的仿真和比较，阐述了在解决BSS问题中所采用的主要理论和方法，分析了各种算法的特点，通过采用固定性算法(FastICA)和联合近似对角化法(JADE)进行了盲源分离的仿真，验证了算法的有效性。同时，也研究了各种混合模型下的语音分离的主要算法和问题。
基于非线性函数和步长因子对算法的性能有着很大的影响，如收敛速度、均方误差、系统的稳定性等，通过对自然梯度算法的研究，提出了一种新的变步长方法(VS-NGA)，极大地提高了系统的收敛速度。为了降低算法的计算复杂性，将符号函数引进EASI算法中，产生了新的S-EASI算法。同时又派生出其余两种改进的算法：符号自然梯度算法(S-NGA)和变步长符号自然梯度算法(VS-S-NGA)，并经证明在简化复杂性和提高收敛速度上，改进的算法是成功的。在有关BSS的文献中，混合矩阵一直被假设为是固定的，而在实际应用过程中，随时间变化的BSS更具现实意义，本文首次在仿真中使用了一个随时间变化的混合矩阵，并获得了一个满意的仿真结果。
对水下混合信号的盲源分离问题进行了建模，将瞬态BSS和卷积BSS模型应用到实际采集到的水下混合信号的分离中。混合信号如何去噪，一直是很难解决的问题，本文在处理过程中提出将噪声当作是源信号中的组成部分，并在实验中验证了方法的可行性。
Blind Signal Processing (BSP) is now one of the hottest and emerging areas in signal with solid theoretical foundations and many potential applications. In fact, BSP has become a very important topic of research and development in many areas, especially radar, sonar, remote sensing, communication systems, biomedical engineering, medical imaging, and image processing, etc. As a branch of BSP, Blind source separation, or source signal separation (BSS), has also become a powerful tool in the areas of signal and image processing. The goal of BSS is to recover the original signals form a set of mixed(observed) signals with no or little a priori knowledge about the sources and mixtures. In this work, we firstly review and discuss the various approaches of linear BSS for both the instantaneous mixtures and convolutive mixtures. Simulations and comparison studies of different techniques have been undertaken to illustrate the main theories and methodologies adopted in solving the BSS problem. And then We analyze the characteristic of typical algorithms. The cases of BSS were done by FastICA and JADE, and the results of simulation proved efficiency of algorithms on separation mixed signals. At the same time, the fundamental theories of BSS and main methods for audio separation are introduced and investigated in this paper.
Based on have great effect of nonlinearity function and step-size factor on performance of algorithm, such as convergence rate, squared error, and stability of system, Natural Gradient Algorithm was studied and a new varying step-size algorithm (VS-NGA) based on NGA was proposed, which greatly improved convergence rate of system. In order to reduce computation complexity, sign function was applied to EASI algorithm to form a new algorithm, namely S-EASI. And two novel algorithms based on them were achieved. All of four algorithm were proven successful for simplifying computation and improving convergence speed.
In documents on BSS, mixture is always assumed as static, but time-varying
引文
[1] Choi S and Cichocki A. Adaptive blind separation of speech signals: Cocktail party problem. In Proc. International Conference on Speech Processing. Seoul, Korea, August 1997: 617-622P
[2] Gao X R and Liu R W. A general approach to blind source separation. IEEE Trans. Signal processing. 1996,44:562-571P
[3] Hyvarinen A, Karhunen J and Oja E. Independent Component Analysis. John Wiley & Sons, Inc. 2001
[4] Hyvarinen A and Oja E. Independent Component Analysis: a Tutorial. http://www.cis.hut.fi/projects/ica/.
[5] Haykin S, ed. Unsupervising Adatpive Filtering, Volume Ⅱ: Blind Deconvolution. John Wiley &son Inc.2000.
[6] Haykin S, ed. Unsupervising Adatpive Filtering, Volume Ⅰ: Blind Source Separation. John Wiley &son Inc.2000.
[7] Hinton G E and Sejnowski T j. Unsupervised Leaming: Foundations of Neural Computation. MIT Press, 1999.
[8] Haykin S. Adaptive Filter Theory. 3rd Edition Prentice Hall,1996.
[9] Haykin S. Neural Networkw - A Comprehensive Foundation. 2nd Edition, Prentice Hall, 1999.
[10] Lathauwer De L. Signal Processing Based on Multilinear Algebra. PhD thesis, KU leuven, Leuven, Belgium, 1997.
[11] Lathauwer De L, Comon P, Moor De B and Vandewalle J. Higher-order power method - Application in independent component analysis. In Proc. Int. Symposium Nonlin. Theory and Appl. Las Vegas, Dec 1995:91-96P
[12] Vigario R N. Extraction of ocular artifacts from EEG using independent component analysis. Electroencephalography and Clinical Neurophysiology, 1997,103:395-404P
[13] Vigario R N, Jousmaki M and Oja E. Independent component analysis for identification of artifacts in magnetoencephalographic recording. In Advances in Neural Information Processing System 10,1997: 229-235P
[14] Cichocki A, Karhunen J and Vigario R N. Neural networks for blind separation with unknown number of sources. Neurocomputing, 24(1-3),Feb 1999:55-93P
[15] Cichocki A, Cao J, Amari S, Murata N and Endo H. Enhancement and blind identification of magnetoencephalographic signals using independent component analysis. In Proc of 11~(th) Int. Conference on BiomagentismBiomag-98,Sendal, Japan, 1999: 169-172P.
[16] Hashimoto S and Saito H. Restoration of shift variant blurred image estimating the parameter distribution of point-spread function. Systems and computers in Japan, 26(1), Jan 1995:62-72P
[17] Hoyer P O and Hyvarinen A. Independent component analysis applied to feaure extraction from colour and stereo images. Network Computation in Neural Systems. 2000:191-210P
[18] Haritopoulos M, Yin H J and Allinson N M. Image Denoising using self-organizing map-based nonlinear independent component analysis. Neural Network. 2002,15:1085-1098P
[19] Funaro M, Oja E and Valpola H. Independent component analysis for artefact separationin astrophysical images. Neural Networks, 16,2003: 469-478P
[20] Kurouoglu EE, Bedini L,Paratore M t,et al. Source separation in astrophysical maps using independent component analysis. Neural Networks, 16,2003:479-491P
[21] Sato Y. Two extensional applications of the zero-forcing equalization method, IEEE Trans. Commun, Vol.23,June 1975:688-672P
[22] Lucky R W. Techniques for adaptive equalization of digital communication systems, Bell Sys. Tech. J., vol.45, Feb, 1966:255-286P
[23] Herault J and Jutten C. Space or time adaptive signal processing by neural network models, Neural Networks for Computing:AIP Conference Proceedings 151 ,New York, 1986.
[24] Cardoso J.-F. Blind identification of independent components with higher-order statistics. In Proc. Workshop on Higher-order Spect. Anal., Vail, Colorado, 1989:174-179P
[25] Compon P. Separation of Stochastic Processes. In Proc. Workshop on Higher-order Spect. Anal, Vail, Colorado,1989:157-160P
[26] Lee T W, Girolami M, Bell A J and Sejnowski T J. A unifying information-theoretic framework for independent component analysis. Computers & Mathematics with Applications,31(11), 2000: 1 -21P
[27] Pajunen P and Karhunen J. Least-squares methods for blind source separation based on nonlinear PCA. Int. Journal of Neural Systems, 6(5,6), 1998:601-612P
[28] Karhunen J and Joutsensalo J. Representation and separation of signal using non-linear PCA type learing. Neural Networks,1994:113-127P
[29] Cichocki A, Robert E and Pope K. Modified Herault-Jutten Algorithms for blind separation of sources. Digital Signal Processing,1997:80-93P
[30] Bell A.J and Sejnowski T.J. An information maximization approach to blind separation and blind deconvolution. Neural Computation,vol.7 1995:1129-1159P
[31] Amari S., Cichocki A. and Yang H.H.. A New Learning Algorithm for blind signal separation. Advances in Neural Information Processing Systems8. MIT Press, Cambridge MA, 1996:757-763P
[32] Yang H.H. and Amari S.. Adaptive on-line learning algorithms for blind separation maximum entropy and mimimum mutual information. Neural Computation, vol.9, 1997:1457-1482P
[33] Cardoso J.F. and Laheld B.. Equivariant adaptive source separation. IEEE Trans. On Signal Processing,vol.44, 1996:3017-3030P.
[34] Lee TW, Girolami M. and Sejnowski T.. Independent component analysis using extended informax algorithm for mixed sub-Gaussian and super- Gaussian sources. NeuralComputation. Vol. 11,1999:417-441P
[35] Hyvarinen A. and Oja E. A fast fixed-point algorithm for independent component analysis. Neural Computation. 1997:1487-1492P
[36] Hyvarinen A. Fast and robust fixed-point algorithm for independent component analysis. IEEE Trans. on Neural Networks. 1999:626-634P
[37] Cardoso,J. Informax and maximum likelihood for blind source separation. IEEE Signal Processing Letters. Vol.4,1997:112-114
[38] Oja E. The nonlinear PCA learning rule in independent component analysis. Neurocomputing, 1997:25-45P
[39] Xu Lei. Least mean reconstruction: A principle for self-organizing nets. Neural Networks. 1993:627-648P
[40] Torkkola T. Blind separation of convolved sources based on information maximization. In IEEE Workshop on Neural Networks for Signal Processing, Kyoto, Japan. 1996:423-432P
[41] 杨绿溪，李可，周长春，何振亚．一种用于超高斯和亚高斯混合信号盲分离的新算法．东南大学学报．1999(1)：1-7页
[42] 杨家兴，陈文．神经网络自适应均衡器研究．信息工程学院学报．1994(2)：9-17页
[43] 刘琚，何振亚．利用高阶累积量何独立分量分析网络进行盲均衡与系统辨识．数据采集与处理．1998(3)：201-205页
[44] 冯大政，史维祥．一种自适应信号盲分离和盲辨识的有效算法．西安交通人学学报．1998(5)：76-79页
[45] 何振亚，陈宇新．用 Hopfield 网络实现盲波束形成．通信学报．1999(12)：86-91页
[46] 何振亚，刘琚，杨绿溪，蔚承建．盲均衡和信道参数估计的一种ICA和进行计算方法．中国科学(E辑)．2000(2)：142-149页
[47] 刘琚，张新刚，孙建德．一种基于ICA的图像水印方法．电路与系统学报．2003(3)：55-59页
[48] 潘红，吴飞，赵雪雁，刘骏伟．基于独立关键子块的图像检索新方法．计算机研究与发展．2003(7)：956-962页
[49] 苏野平，何量．一种改进的基于高阶累积量的语音盲分离算法．电子学报．2002(7)：956-958页
[50] 倪晋平，马远良．基于高阶累积量的复数混合矩阵盲估计算法．电子与信息学报．2002(11)：1506-1511页
[51] 倪晋平，马远良，杨坤德．水声信号盲分离的实数算法与复数算法．西北工业大学学报．2002(2)：43-48页
[52] Donoho D. L. and Johnstone I. M. Projection-based approximation and a duality with kernel methods. Annals of Statistics. 1989:58-106P
[53] Comon P. Independent component analysis, a new concept?. Signal Processing. 1994(3):287-314P
[54] Hyvarinen A. and Oja E. Independent component analysis: algorithms and applications. Neural Networks. 2000:411-430P
[55] Amari A and Cichocki A. Adaptive blind signal processing: neural network approach. Proc. IEEE. 1998(10):2026-2048P
[56] Haykin S. Neural Networks. Pretice-Hall, New Jersey. 1999
[57] Amari S. Natural gradient works efficiently on learning. Neural Comput. 1998:287-314P
[58] Cardoso J. F. Source separation using higher order moments. In Proc. ICASSP, Glasgow, UK. 1989:(2109-2212P)
[59] Puntonet C.G and Prieto A. Neural net approach for blind separation of sources based on geometric properties. Neurocomputing. 1998(3):141-164P
[60] Cichocki A and Unbehauen R. A new on-line adaptive learning algorithm for blind source separation of source signals. Int. Symp. on Artificial Neural Networks. 1994:406-411P
[61] Cardoso J F. High-order contrasts for independent component analysis. Neural Computation. 1999(1): 157-192P
[62] Nikias C L and Mendel J M. Signal processing with higher-order spectra. IEEE Signal Processing Magazine. 1993(7): 10-37P
[63] Mendel J M. Tutorial on higher-order statistics in singal processing and system theory: Theoretical results and some applications. Proceedings of the IEEE. 1991(3):278-305P
[64] Molgedey L and Schuster H G. Separation of a mixture of independent signals using time delayed correlations. Phys. Rev. Lett. 1994:3634-3637P
[65] Choi S, Cichocki A and Belouchrani A. Second order non-stationary source separation. VLSL Signal Processing. 2002:93-104P
[66] Choi S, Cichocki A and Belouchrani A. Blindseparation of second-order nonstationary and temporally colored sources. In Proceedings of the 11~(th) IEEE signal Processing Workshop on Statistical Signal Processing. Singapore. 2001:444-447P
[67] Karhunen J, Wang L and Vigario R. Nonlinear PCA type approaches for source separation and independent component analysis. In Proc. Of the IEEE Int. Conf. on Neural Networks. 1995:995-1000P
[68] Oja E. Neural networks, principal components, and subspaces. Int. J. on Neural Systems. 1989:61-68P
[69] Oja E. Nonlinear PCA citerion and maximum likelihood in independent component analysis. In Proc. Int. Workshop on independent Component Analysis and Signal Separation. Aussois, France. 1999:143-148P
[70] Cichocki A and Amari S. Adaptive Blind and Image, Processing: Learning Algorithms and Application. New Work: Wiley,2002.
[71] N.Murata, K.R.Muller, A.Ziehe, S.Amari, Adaptive On-line Learning in Changing Environments, Advances in NIPS, Cambridge, MA:MIT Press, 1997:599-605P
[72] Douglas S C and Cichocki A. Adaptive Step Size Techniques for Decorrelation and Blind Source Separation, in Proc. Asilomar Conf. on Signals, Systems and Computers, Pacific Grove, CA, 1998,pp.1191-1195
[73] Shan T J, Kailath T. Adaptive algorithms with an automatic gain control feature. IEEE Trans, on Circuits and Systems. 1998:122-127P
[74] Aboulnasr T and Mayyas K T. A Robust Variable Step-size LMS-type Algorithm:Analysis and Simulations. IEEE Trans on Signal Processing. 1997:631-639P
[75] Karni S and Zeng G. A new convergence factor for adaptive filters. IEEE Trans. on Circuits and Systems. 1989:1011-1012P
[76] Yuan Lianxi, Wang W and Chambers J. Variable Sep-Size Sign Natural Gradient Algorithm for Sequential Blind Source Separation. IEEE Signal Processing Letter. 2005(8):589-592P
[77] Mathews V J and Xie Z. A stochastic gradient adaptive filterwith gradient adaptive steip size. IEEE Trans. Signal Process. 1993(6):2075-2087P
[78] Chambers J A, Jafari M G and Mclaughlim S. Variable step-size EASI algorithm for sequential blind source separation. Electon. Lett. 2004: 393-394P
[79] Cardoso J and Laheld B. Equivariant Adaptive Source Separation. IEEE Trans. On Signal Processing. 1996:3017-3030P
[80] Lehmann E L. Testing statistical hypothesis. Wiley Pub. In statistics. John Wiley. 1959
[81] Yuan Lianxi, Sang Enfeng, Wang W and Chambers J A. An Effective Method to Improve Convergence for Sequential Blind Source Separation. Lecture Notes in Computer Science. 2005:199-208P
[82] Georgiev P, Cichocki A and Amari S. On some extensions of the natural gradient algorithm. In Proc. ICA.2001:581-585P
[83] Cichocki A and Amari S. Adaptive Blind and Image, Processing: Learning Algorithms and Application. New Work: Wiley.2002
[84] Comon P. Independent Component Analysis. in Proc. Int. Workshop on Higher-Order Stat. Chamrousse, France. 1991:111-120P
[85] Cruces S, Cichocki A and Castedo L. Blind source extraction in Gaussian noise. In proceedings of the 2nd International Worshop on Independent Component Analysis and Blind Signal Separation(ICA'2000), Helsinki, Findland. 2000:63-68P
[86] Hild E, Deniz Erdogmus and Jose C P. Blind source separation using Renyi mutual information. IEEE Signal Proc. Lett. 2001:174-176P
[87] Amari S, Douglas S C and Cichocki A. Multichannel blind deconvolution and equalization using the natural gradient. In Proc IEEE Workshop on Signal Processing Advances in Wireless Communications. La Bastille,Paris, France. 1997:101-104P
[88] Inouye Y. Blind deconvolution of multichannel linear time-invariant systems of nonminimum phase. In Statistical Methods in Control and Signal Processing. 1997:375-397P
[89] Comon P. Contrasts for multichannel blind deconvolution. IEEE Signal Processing Lett. 1996:209-211P
[90] Thi H L and Jutten C. Blind sources separation for convolutive mixtures. Signal Processing. 1995:202-209P
[91] Jutten C and Herault J. Blind separation of sources, parti: An adaptive algorithm based on neuromimetic architecture. Signal Processing. 1991: 1-10P
[92] Cichocki A, Amari S and Cao J. Blind separation of delayed and convolved signals with self-=adaptive learning rate. NOLTA. 1996:229-232P
[93] Choi S, Hong H, Clotin H and Berthommier F. Multichannel signal separation for cocktail party speech recognition: A dynamic recurrent network. Neurocomputing. 2002:299-314P
[94] Choi S, Amari S and Cichocki A. Natural gradient learning for spatio- temporal decorrelation: Recurrent network. IEICE Trans. Fundamentals. 2000:2715-2722P
[95] Bucher H, Aichner R and Kellermann W.A Generalization of a class of blind source separation algorithms for convolutive Mixtures", in Proc. Int. Symp. On Independent Component Analysis and Blind Signal Separation. Nara, Japan, 2003:945-950P
[96] W. Wang, and J.A.Chambers. Blind source separation of convolutive m ixtures and its application in speech enhancement: an Overview. CDSPR at King's College London. Technical Report.2002
[97] L. Parra, and C.Spence. Convolutive Blind Source Separation of Nonstationary Sources.IEEE Trans, on Speech and Audio Processing. 2000:320-327P
[98] Capdevielle V, Serviere C and Lacoume J L.Blind Separation of Wide-band Sources in the Frequency Domain, in Proc. ICASSP95.1995:2080-2083P
[99] Anemuller J and Kollmeier B. Amplitude Modulation Decorrelation for Convolutive Blind Source, in Proceedings of the Second International Workshop on Independent Component Analysis and Blind Signal Separation. Helsinki, Finland 2000:215-220P
[100] Mitianoudis N and Davies M. Audio Source Separation of Convolutive mixtures. IEEE Trans, on Speech and Audio Processing. 2003:489-497P
[101] Saruwatari H, Kawamura T and K. SHikano. Fast-Convergence Algorithm for ICA-based Blind Source Separation Using Array Signal Processing. 11~(th) IEEE Workshop on Statistical Signal Processing. 2001:464-467P
[102] Parra L C and Alvino C V. Geometric Source Separation: Merging Convolutive Source Separation with Geometric Beamforming. IEEE Trans, on Speech and Audio Processing. 2002:352-363P
[103] Murata N, keda S and Ziehe A. An Approach to Blind Source Separation Based On Temporal Structure of Speech Signals. RIKEN Brain Science Institute, Japan. Technical Report BSIS Technical Reports. 1998
[104] Dapena A and Castedo L. Blind Separation of Convolutive mixtures of temporally white signals: A Novel Frequency Domain Approach, in ICA2001. 2001:179-184P
[105] Dapena A and Castedo L. Blind Separation of Convolutive mixtures of temporally white signals: A Novel Solution To The Amplitude and the Prmutation Indeterminacies. in 6~(th) International Work Conference on Artificial Neural Network. Granada, Spain. 2001:603-610P
[106] Zhang L and Cichocki A.Blind Deconvolution /Equalization Using State-Space Models, in Proceedings of the 1998 IEEE Workshop on Neural Networks for Signal Processing. Cambridge, UK, 1998:123-131P
[107] Mansour A, Barros A K, and Ohnishi N. Subspace Adaptive Algorithm for blind separation of convolutive mixtures by conjugate gradient method. In First International conference and Exhibition Digital Signal Processing. Moscow, Russia. 1998:252-260P
[108] Morgan D R and Thi J C. A delayless subband adaptive filter architecture. IEEE Trans. Signal Processing. 1999:1819-1830P
[109] Sahlin H and Broman H. Signal Separation Applied To real World Signals, In proceedings of 1997 Int. Workshop on Acoustic Echo and Noise Control.London, UK. 1997
[110] Schobben D and Sommen P.Transparent Communication. In Proc. IEEE Benelux Signal Processing Chapter Symposium. 1998:171-174P
[111] Lee T W, Ziehe A, Orglmeister R and Sejnowski T. Combining time delayed decorelation and ica: towards solving the cocktail party problem. In proc. ICASSP. Seattle, WA. 1998: 1249-1252P
[112] 王振刚，马远良．水声信号源的盲分离原理及自适应算法．98全国声学学术会议论文集．1998．
[113] 倪晋平，马远良，李颖红．信号盲分离的非线性主分量算法．西安工业学院学报．2001：196-203页
[114] 倪晋平，马远良．基于峰度最大化的盲信号分拣和盲波束形成复数算法．声学技术增刊．全国青年水声学术会议论文集．上海．2001
[115] 倪晋平，马远良，鄢社锋，水声信号盲分离算法性能评价．探测与控制学报．2002：40-44页
[116] 张安清，章新华．基于信息理论的舰船噪声盲分离算法．系统工程与电子技术．2002：38-40页
[117] 姜卫东，陆佶人，张宏滔．基于单个频点的水声信号盲源分离．电子与信息学报．2005：532-535页
[118] 鄢社锋，马远良．单通道带通信号的复基带盲解卷积算法．电子与信息学报．2004：702-797页
[119] Lappalainen H and Miskin J. Ensemble learning. Advances in Independent Component Analysis, M. Girolami, Ed. Berlin: Springer Verlag, 2000: 75-92P
[120] Saruwatari H, Kawamura T and Shikano K. Fast-convergence algorithm for ICA-based blind source separation using array signal processing. 1th IEEE Workshop on Statistical Signal Processing(SSP2001).2001:464-467P
[121] 倪晋平．水声信号盲源分离技术的研究．西北工业大学博士论文．2002
[122] Bernard W, Robert C. and Goodlin et al. Adaptive Noise Cancelling: Principles and Applications. Proceedings of the IEEE, 1975:1692-1716P