BLIND separation of independent sources has receiveda great deal of attention due to various applications inscience and technology. The problem of blind source separation(BSS) and/or ICA has been studied by many researchers inthe fields of neural networks and statistical signal processing[1]“[5], [9], [15], [16], [18], [22], [26], [31], [36] during thepast ten years, and many interesting theoretical and practicalresults have been achieved.This paper is organized as follows. A robust prewhitening technique with noise reduction and a cross-validation technique with optimal dimensionality reduction are presented in Section II. The parameterized t-distribution model and its robust properties, as well as the stability of the developed algorithm are presented in Section III. Experimental results using this new approach on artificially synthesized data and real-world unaveraged single-trial MEG data are presented in Section Invite MEG data are from an experiment studying the auditory evoked fields (AEF) task.

ABSTRACT
In this paper, we propose a robust approach for independent component analysis (ICA) of signals that observationsare contaminated with high-level additive noise and/or outliers.The source signals may contain mixtures of both sub-Gaussianand super-Gaussian components, and the number of sources is unknown. Our robust approach includes two procedures. In the first procedure, a robust prewhitening technique is used to reduce the power of additive noise, the dimensionality and the correlation among sources. A cross-validation technique is introduced to estimate the number of sources in this first procedure. In the second procedure, a nonlinear function is derived using the parameterizedt-distribution density model. This nonlinear function is robust against the undue influence of outliers fundamentally. Moreover, the stability of the proposed algorithm and the robust property of misestimating the parameters (kurtosis) have been studied. By combining the t-distribution model with a family flight-tailed distributions (sub-Gaussian) model, we can separate the mixture of sub-Gaussian and super-Gaussian source components. Through the analysis of artificially synthesized data and real-world magneto encephalographic (MEG) data, we illustrate the efficacy of this robust approach.
Index Terms”Cross-validation method, independent component analysis (ICA), parametric estimation method, principal component analysis (PCA), robust prewhitening, t-distribution density model, unvaried single-trial MEG data analysis.