Application of higher-order statistics for the analysis of electroencephalogram in different brain functional states

Higher-order statistics are applied to the analysis of electroencephalograms (EEGs) in order to investigate their non-Gaussianility and nonlinearity. Parametric bispectral estimation is proposed in this paper for the purpose of extracting more information, beyond second-order statistics or power spectra. The EEGs of normal subjects in different brain functional states are analyzed in terms of bispectral estimation. The experimental results show that all kinds of EEGs exhibit obvious quadratic nonlinear interactions, but the bispectral structure of a normal EEG changes with different functional states of the brain. It is suggested that the bispectrum could be regarded as one of the main characteristics in the study of EEG signals.