Extraction of superimposed evoked potentials by combination of independent component analysis and cumulant-based matched filtering

A novel approach is proposed for the efficient separation of mixed evoked potentials (EPs) presented simultaneously by different stimuli. We first apply a robust independent component analysis (ICA) approach to the observed sensor signals for the separation of the superimposed EP signals. Next, the desired EP components are estimated by matched-filtering of the separated signals. The impulse response of such a matched filter can be computed based on third-order cumulants of the filter input signal. Therefore, due to the tolerance of the third-order cumulants to both Gaussian and any symmetrically distributed non-Gaussian noise or interference, the filter impulse response will be matched with the desired signal alone. It is demonstrated by extensive computer simulations that applying the cumulant-based ICA and filtering improves dramatically the SNR of the final estimation of the EP components