Empirical mode decomposition to approach the problem of detecting sources from a reduced number of mixtures

The paper presents a new approach of Blind Source Separation based on the combined use of Empirical Mode Decomposition (EMD) and Factor Analysis (FA) for the case of more sources than observable signals, the so called overcomplete problem. The EMD-FA performance is tested both over artificial data and real EEG signals and compared with that of the more traditional Independent Component Analysis (ICA). The EMD-FA approach exhibited a neatly superior performance in the overcomplete problem with respect to traditional ICA. Furthermore this approach can be adopted even for nonlinear and nonstationary signals, which makes it very attractive for biomedical signal processing.