Detection of successive changes in dynamics of EEG time series: linear and nonlinear approach

Interesting results have been reported on the quantitative description of EEG patterns, based on the assumption either about linear stochastic dynamics, or chaotic dynamics. Thus the question arises of whether linear or non-linear methods are to be used in the EEG analysis. This study was undertaken to reveal the dynamic behavior of EEG activity during performance of a voluntary motor task. Using autoregressive models and Kalman filtering on one side and nonlinear prediction of the other side, successive changes from linear stochastic to chaotic dynamics of short-term segments of EEG time series were found in single-trial records. This suggests that EEG activity should be processed in parts since different steady states are separated by shorter chaotic transients