Autoregression models of EEG. Results compared with expectations for a multilinear near-equilibrium biophysical process.

This paper considers the properties of parameters (natural frequencies and damping coefficients) obtained from segment-by-segment autoregression analysis of ECoG of rat. The use of a reference signal as control for parameter estimate errors, and multiple regression analyses indicate that the dependencies among parameters calculated from ECoG in the alert (desynchronised) state are of a form consistent with imposition of time-invariance assumptions (implicit in autoregression) on an inherently non-stationary, multimodal, linear and near-equilibrium "thermal" process.