Dynamics of human sleep EEG.

Several investigators of EEG time series reported a rejection of the null hypothesis of linear stochastic dynamics for epochs longer than 10 s. We examine whether this rejection is related to nonlinearity or to nonstationarity. Our approach is a combination of autoregressive (AR-) modeling and surrogate data testing. It is shown that the fraction of subsegments, for which the null hypothesis has to be rejected, increases with the length of the subsegments and can be related to fluctuations of the AR-coefficients on time scales in the range from 2 to 30 s. (C) 2003 Elsevier Science B.V. All rights reserved.