摘要
盲源分离是近二十年来发展起来的一门新型技术学科,在理论和实际应用上都取得了长足的发展,广泛的应用于各个学科领域,出现了许多热门方向,许多学者都致力于相应领域的理论和应用研究,也取得了很好的效果。为了机载阵列天线设备能在通信侦察中获取有价值的信息,本文主要研究了运动平台盲源分离的问题。建立了阵列接收通信信号的混合模型,研究发现天线阵列在运动时,通信信号混合模型相当于一个时变的线性瞬时混合模型。对经典的盲分离算法——自然梯度算法和EASI算法进行了研究,提出了几种改进算法并对其进行了仿真分析。同时针对盲源分离算法收敛速度和稳态性能这一矛盾性能指标,提出了改进的新算法——基于优选函数的峭度变步长EASI算法,该算法根据分离状态与峭度方差的关系,使步长随峭度方差的变化而变化,有效的减小稳态误差,同时优选函数在信号分离的初始阶段和稳态阶段使用不同的估计函数,具有收敛速度快的特点。新算法相对于固定步长EASI算法,经典变步长算法,稳定性和收敛速度都有一定的提高,适用于时变的线性瞬时混合模型,能成功的分离混合通信信号,具有广阔的应用前景。
Blind source separation is a new type of technology,which is developed in the past two decades. BSS has made considerable development in theory and practical application,there are wide range of applications in various subject areas and many popular research directions. Many scholars are committed to the corresponding the field of theoretical and applied research, also they have achieved good results. In order to obtain valuable information in communication reconnaissance by array antenna equipment on board, blind source separation problem for the moving platform is mainly studied in this paper.An array to receive communication signals mixed model is established, the study found that the communication signal mixed model is equivalent to a time-varying linear instantaneous model,when the antenna array is moving. The classical blind source separation algorithm - the natural gradient algorithm and the EASI algorithm have been studied ,and then put forward several improved algorithms, make some simulation analysis. For the contradiction between convergence speed and steady-state performance in BSS, a new algorithm is proposed - kurtosis variable step-size EASI algorithm based on optimially selected function for blind source separation. According to the relationgship between the covariance of kurtosis and the state of separation ,the algorithm’s step-size is controlled with kurtosis covariance, it decreases the steady-state error efficiently.Meanwhile optimially selected function algorithm carrys on two different estimation functions, which are used between two phases of the signal separation, as the result of the convergence speed has increased. The proposed algorithm has faster convergence speed and smaller steady-state error than EASI and VS_EASI. It applies to the time-varying linear instantaneous mixed model and separates mixed communication signals successfully.The new algorithm has extensive application prospects.
引文
[1]Herault J,Jutten C.Blind Separation of Source.Part 1:An adapative algorithm based on neuromimetic architecture,Signal Processing,1991,24(1):1-10.
[2]Comon P.Independent component analysis,a new concept Signal Pro- cessing,1994,36:287-314.
[3]Cichocki A,Unbehauen R,Moszczynski L, et al.A New On-line Adaptive Learning Algorithm for Blind Separation of SourceSignals[C]//ISANN94,Taiwan,1994:406-411.
[4]Bell A,Sejnoeski T J,An Information Maximization Approach to Blind Separation and Blind Deconvolution[J].Neural Computation,1995,7(6):1004-1034.
[5] Cichocki A,Bogner R E, Moszczynski L. Modified Herault-jutten Algorithms for Blind Separation of Sources[J].Digital Signal Processing,1997(7):80-93.
[6]Hyvarinen A. Fast and Roubst Fixed-point Algorithms for Independent Component Analysis[J].IEEE Trans On Neural Networks,1999,10(3):626-634.
[7]Binghham E,Hyvarinen A.A Fast Fixed-point Algorithm for Independent Component Analysis of Complex Valued Signals[J].NeuralSystem,2000,10(1):1-8.
[8]Zhang L, Cichocki A. Blind Deconvolution of Dynamical systems:A state Space Appoach[J].Iapanese Journal of Signal Processing,2000,4(2):111-130.
[9]冯大郑,保铮,张贤达.信号盲分离问题的多阶段分解算法.自然科学进展,2002,12(3):324-328.
[10]张贤达,保铮.通信信号处理.北京:国防工业出版社,2000.
[11]Comon P.Independent component analysis,a new concept. Signal Pro- cessing,1994,36:287-314.
[12]Papadias C B,Paulraj A. A constant modulus algorithm for util-user signal separation in presence of delay spread using antenna arrays.IEEE Signal Processing Lett.,1997,4:178-181.
[13]Hoyer P O, Hyarinen A.Independent Component Analysis Applied to Feaure From Colour and Stereo Images.Network Compuation in Neural Systems.2000:191-210.
[14]Amari S I,et al..A learning algorithm for blind signal separation.Advances in Neural Information Processing System,1996,8:757-764.
[15]Amari S I.Natural gradient works efficientlyin learning.Neural Computation,1998,10(2):251-276.
[16]Cichocki A,Amari S I.Adaptive Blind Signal and Image Processing:Learning Algorithms and Applications.John Wileyand sons,2002.
[17]Cardoso J F.Blind signal processing:statistical principles.Proc.IEEE,1998,86(10):2009-2025.
[18]Amari S I,Cichocki A and H H Yang.A new learning algorithm for blind signal separation.In Michackel C.Mozer David S.Touretzky and Michael E.Hasselmo,editors,Advances in Neural Information Processing Systems 1995(8):757-763.
[19]Amari S I, Cardoso J F.Blind source separation-semi paramrtric approach.IEEE Trans.on Signal Processing, 1997,45(11):2692-2700.
[20]Amari S I, Chen T P,and Cichocki A.Stability analysis of adaptive blind source separation.Neural Networks ,1997,10(8):1345-1351.
[21]Amari S I, Cichocki A,and H H Yang.Unsupervised Adaptive Filtering,chapter Blind Signal Separation and Extraction-Neural and Information Theoretic Approaches.John Wiley,1999.
[22]Yuan L X,Wang W,ChmabersJ.Variable step-size sign natural gradient algorithm for sequential blind source separation IEEE Singal Poreessing Letters. 2005(12):589-592.
[23]Geogriev P, Cichoeki A and Amari S I.On some extensions of the natural gradient algorithm.In Porc.ICA.2001:581-585.
[24]Murata N, Muller K R, Ziehe A, Amari S.,AdpativeOn-lineLearning in ChangingEnviromnents,In:Advnaees in NIPS’9 Cambridge,MA:MIT Press,1997:599-605.
[25]Douglas S C,Cichocki A.Adaptive Step Size Teehniques for Decorrelation and Blind oucre Separation,in Proc.Asilomar Conf.on Singals,Systems and Computers,Pacific Grove,CA,1998,1191-1195.
[26]Cardoso J F,Equivariant Adapative source separation,IEEE Transaction on Signal Processing,1996,44(12):3013-3030.
[27]朱孝龙,张贤达,基于选优估计函数的盲信号分离,西安电子科技大学学报(自然科学版),2003,30(3):335-339.
[28]Cardoso J F,on the stability of source separation algorithms[J],Joural of VLSI Signal Processing, 2000,26(1):7-14.
[29]马守科,何选森,许广廷,基于扩展Infomax算法的变步长在线盲分离,系统仿真学报,2007,19(19):4513-4516.
[30]张贤达,朱孝龙,保铮,基于分阶段学习的盲分离算法,中国科学,2002,5(32):693-703.
[31]孙守宇,郑君里,吴里江等,峭度自适应学习率的盲信源分离,电子学报,2005,3(33):474-476.
[32]Chambers J A, Jafari M G and McLaughlin S,Variable Step-size EASI Algorithm for Sequential Blind Source Separation,Electrnic Letters,2004,40 (6):393-394.