In this article we propose a computationally efficient method (termed FCOMBI) to combine the strengths of non-Gaussianity-based Blind Source Separation (BSS) and cross-correlations-based BSS. This is done by fusing the separation abilities of two well-known BSS algorithms: EFICA and WASOBI. Simulations show that our approach is at least as accurate and often more accurate that other state-of-the-art approaches which also aim to separate simultaneously non-Gaussian and time-correlated components. However, in terms of computational efficiency and stability, FCOMBI is the clear winner which makes it specially suitable for the analysis of very high-dimensional datasets like high-density Electroencephalographic (EEG) or Magnetoen-cephalographic (MEG) recordings.
展开▼
