ARX FILTERING OF SINGLE-SWEEP MOVEMENT-RELATED BRAIN MACROPOTENTIALS IN MONO-CHANNEL AND MULTICHANNEL RECORDINGS

A technique of stochastic parametric identification and filtering is a pplied to the analysis of single-sweep event-related potentials. This procedure, called AutoRegressive with n eXogenous inputs (ARXn), model s the recorded signal as the sum of n+l signals: the background EEG ac tivity, modeled as an autoregressive process driven by white noise, an d n signals, one of which represents a filtered version of a reference signal carrying the average information contained in each sweep. The other (n-1) signals could represent various sources of noise (i.e., ar tifacts, EOG, etc.). An evaluation of the effects of both artifact sup pression and accurate selection of the average signal on mono- or mult i-channel scalp recordings is presented.