Quantitative EEG Analysis
Quantitative EEG analysis has been widely used in the assessment of certain neurophysiological states and disorders. The frequency content of the EEG is obtained from the power spectral density (PSD) estimate. Traditionally the EEG is divided into four bands: (δ 0 - 3.5 Hz), (θ 3.5 - 7 Hz), (α 7 - 13 Hz) and (β 13 - 30 Hz). PSD estimates can be calculated using either nonparametric methods (e.g. methods based on FFT) or parametric methods (e.g. methods based on autoregressive time series modelling). The sample for which PSD is obtained is traditionally assumed to be stationary, even though real EEG is usually nonstationary.
Time-Varying PSD Estimation
Time-varying PSD estimation is achieved by allowing the AR/ARMA coefficients to be time-varying. The time-varying parameter estimation problem can be solved with adaptive algorithms such as LMS, NLMS, RLS or the Kalman filter. All these algorithms suffer slightly from tracking lag. This can be avoided, however, by using the Kalman filter along with fixed interval smoothing.
ERS/ERD test: Interest in the transition dynamics of EEG has been increasing. One such application is the event-related synchronization/desynchronization (ERS/ERD) test of alpha waves. Time-varying PSD, ob-tained using the Kalman smoother approach, for an EEG signal recorded from channel O2 with eyes open (desyn-chronization) and eyes closed (synchronization) is pre-sented here.