摘要
自然梯度算法是处理盲源分离问题的一个重要方法。自然梯度算法的分离速度与稳态性能之间存在矛盾,步长增大收敛速度加快,但是稳态误差随之增大。自适应变步长算法是解决收敛速度与稳态误差之间的矛盾的有效手段。基于原有自适应算法,提出了一种分级迭代变步长算法,更好地解决了算法存在的收敛速度与稳态误差的矛盾。仿真结果表明,该算法具有更快的分离速度和更好的稳态性能。
Natural gradient algorithm is an important method to deal with the blind source separation problem.There is a contradiction between the separation speed and the steady state performance of the natural gradient algorithm.The larger the step size is,the faster the convergence speed becomes.But the steady state error increases.The adaptive variable step size algorithm is an effective method to solve the contradiction between the convergence speed and the steady-state error.Based on the original adaptive algorithm,a hierarchical iteration variable step algorithm is presented to better solve the existing contradiction between the convergence rate and steady-state error.The simulation results show that the proposed algorithm has faster separation speed and better steady-state performance.
引文
[1]Sardouie S H,Shamsollahi M B,Albera L,et al.Interictal EEG noise cancellation;GEVD and DSS based approaches versus ICA and DCCA based methods[J].IRBM,2015,36(1):20-32.
[2]Liu Q J,Wang W W,Jackson P J B,et al.Source separation of convolutive and noisy mixtures using audio-visual dictionary learning and probabilistic time-frequency masking[J].IEEE Transactions on Signal Processing,2013,61(22);5520-5535.
[3]Abolghasemi V,Ferdowsi S,Sanei S.Fast and incoherent dictionary learning algorithms with application to fMRI[J].Signal,Image and Video Processing,2015,9(1):147-158.
[4]Tomazeli Duarte L,Moussaoui S,Jutten C.Source sepai'ation in chemical analysis;recent achievements and perspectives[J].Signal Processing Magazine,2014,31(3):135-146.
[5]Duarte L T,Donno D,Lopes R R,et al.Seismic signal processing;Some recent advances[C]//2014 IEEE International Conference on Acoustics,Speech and Signal Processing.2014:2362-2366.
[6]Zhang J,Li BY.Analysis of the communication signal processing[J].Electronic Test,2013(18):78-79.
[7]季策,杨坤,王艳茹,等.基于符号算子的变步长不完整自然梯度算法[J].模式识别与人工智能,2014,27(11):1026-1031.
[8]Cichocki A,Unbehauen R,Moszczynski L,et al.A new online adaptive learning algorithm for blind separation of source signals[C]//Proceedings of the 1994 International Symposium on Advanced Nanodevices and Nanotechnology.1994:406-411.
[9]Yang H H.Serial updating rule for blind separation derived from the method of scoring[J].IEEE Transactions on Signal Processing,1999,47(8):2279-2285.
[10]Ji C,Tang B C,Yang K,et al.Improved blind source separation based on non-holonomic natural gradient algorithm with variable step size[C]//Proceedings of 2013 Chinese Intelligent Automation Conference.2013:761-767.
[11]Tang L T,Wang J.Separation method of noisy mixed images based on wavelet tranfform and ICA[J].Journal of Huazhong Normal University,2013,52(5).
[12]Wang X,Ou S,Gao Y,et al.A new fast nonlinear principal component analysis algorithm for blind source separation[C]//2015 12th International Conference on Fuzzy Systems and Knowledge Discovery.2015:1626-1630.
[13]Schmidhuber J.Deep learning in neural networks;an overview[J].Neural Networks,2015,61:85-117.
[14]Yuan L X,Wang W W,Chambers J A.Variable step-size sign natural gradient algorithm for sequential blind source separation[J].Signal Processing Letters,2005,12(8):589-592.
[15]Amari S,Cichocki A,Yang H H.A new learning algorithm for blind signal separation[C]//Proceedings of the 8 th International Conference on Neural Information Processing Systems.1996:757-763.
[16]Xu P C,Yuan Z G,Jian W,et al.Variable step-size method based on a reference separation system for source separation[J].Journal of Sensors,2015(5):1-7.