Application of Linear Projection Algorithms For Reduction of Background Eeg Noise

Abstract

In the present study, linear orthogonal projection algorithms (least square sense linear mapping (LSLM), minimum variance estimation (MVE), spectral domain estimation (SDC) and time domain constraint (TDC)) have been applied to reduce the background EEG noise on small number of trials elicited by auditory stimuli. These methods are compared to each other with respect to eigendecomposition based spectral signal-to-noise-ratio (SSNR) in tests where the grand average of experimental observations is considered as the template evoked potential (EP) signal. The actual ongoing EEG series and single-sweep EP are summed in pseudosimulations. The LSLM having simplest formulation is found to be most useful pre-filter among those methods in removing large amount of the noise without loss of information about EP components since both EEG noise level and EP component variations are highly correlated with eigenspectra of the raw data.