Ensemble averaging subspace-based approach for ERP extraction

A novel approach based on Subspace methods is proposed for extracting the Event Related Potentials (ERPs) from the background Electroencephalograph (EEG) colored noise. First, the enhancement of SNR to the neighborhood of -2 dB is achieved through the ensemble averaging of the EEG data over a limited number of trials. Then a linear estimator is used to reduce further the amount of the EEG signal in the ERPs. With this estimator the EEG colored noise is first whitened using Cholesky factorization then the eigen-decomposition of the covariance matrices of pre-whitened data performed and the subspace is decomposed into signal subspace and noise subspace. The components in the noise subspace are nullified and the components in the signal subspace are retained to do the improvement. The proposed algorithm is verified with simulated data and the results shows reliable performance in terms of accuracy and failure rate.