Algebraic Perspectives of Background EEG Elimination

Abstract

Least squares linear mapping (LSLM) algorithm is applied to reduce the background EEG noise on single-trial auditory evoked potentials (EPs) in the present study. Relationships between eigenvalues and spectral signal-to-noise ratio (SNR) are shown where a small number of noisy sweeps are considered as a raw matrix corrupted with additive noise. Results show that the LSLM can be assigned as a pre-filter in single trial EP estimations. Dominant eigenvectors of noisy EPs models the noiseless EP waveforms.