Detection of Directionality of Information Transfer in NonlinearDynamical Systems

摘要

For the quantification of strength and identification of direction of coupling be-tween two sub-systems of a complex dynamical system, observed from bivariate time series, a number of measures have been proposed that can be grouped in measures of phase synchronization, state space and information. We review all these measures and. in particular, for the information measures we examine different estimates of the probability distributions. We propose also a modifi-cation of the transfer entropy measure to span larger time windows and thus be more appropriate for flows. Simulations on systems of different types and for varying coupling strengths showed that information measures, and the modi-fied transfer entropy measure in particular, detect best the coupling strength and direction. This is also found when applying the measures to pairs of EEG channels in order to detect the propagation of pre-epileptic brain activity.