Considerations on Adaptive Autoregressive Modelling in EEG Analysis
Alois Schlögl, Gert Pfurtscheller · 1998
INTRODUCTION For a Brain Computer Interface it is important that the electroencephalogram (EEG) can be analysed online and on a single trial basis. The EEG characteristic changes with respect to different mental activity are exploited. In other words, the event-related desynchronization (ERD) [1], which is an attenuation of frequency components, has to be evaluated. For the evaluation of the spectra an autoregressive model can be used. This offers the advantage that the spectrum can be estimated based on a Maximum entropy (MESE) [2]. The time-variation is taken into account using an adaptive model. Figure 1: An Adaptive Autoregressive model is shown. One can imagine that the model output Y(t) is a white noise process which is filtered. The characteristics of the signal are described by the AR parameters. For the evaluation of time-varying behaviour an adaptive AR model is used; the model parameters are varying with time. Y t = a 1,t Y t-