Improved spectral analysis of EEG signals

Removal of low frequency trends is often a preliminary step to estimating a spectrum and failure to do so can result in serious distortion in the spectrum. In this paper, we use a 'quasi-detrending' method for classification of EEG spectrum where the level of detrending or differencing is made to vary. Differencing in time domain acts as a high pass filter in the frequency domain. Therefore the low frequency values in the delta range can be ignored and this is a saving in computation time since delta range values do not correspond to any normal conscious human mental tasks. We also show that using discrete PSD values in the range below 30 Hz gives better classification results than using the delta, theta, alpha and beta power band values used by some authors.