EEG Preprocessing

SWT was chosen because it preserves translation invariance and temporal localization better than alternatives cited in the article. All datasets were standardized to a 19-channel international 10-20 montage. Stationary wavelet transform with symlet-4 was used for frequency-band decomposition. Signals were filtered from 0.5 to 45 Hz and resampled to 128 Hz. Each rhythm-specific signal was segmented into 1-second windows with 50% overlap. Each segment was normalized with per-channel z-scores computed within the segment.