Autoregressive spectral analysis in Brain Computer Interface context

Over the past decade, a number of studies have evaluated the possibility that scalp-recorded electroencephalogram (EEG) activity might be the basis for a brain-computer interface (BCI), a system able to determine the intent of the user from a variety of different electrophysiological signals. With our current EEG-based communication system, users learn over a series of training sessions to use EEG to move a cursor on a video screen: to make this possible users must learn to control the EEG features that determines cursor movement and we must improve signal processing methods to extract from background noise the EEG features that the system translates into cursor movement. Non-invasive data acquisition, makes automated feature extraction challenging, since the signals of interest are "hidden" in a highly noisy environment. It was demonstrated that the spatial filtering operations improve the signal-to-noise ratio. On the contrary, autoregressive modeling has been successfully used by many investigators for EEG signals analysis in BCI context, but to our knowledge no clear guidelines exist on how to choose the parameters of the spectral estimation. Here we present an analysis of the dependence of BCI performance on the parameters of the feature extraction algorithm. In order to optimize user performances, we observed that a different model order value had to be chosen correspondently to different EEG features used to control the system, according to the differences in the spectral power content of alpha and/or beta bands