EEG signals were continuously digitized throughout each recording session at a sampling rate of 200 Hz. Data were digitally filtered offline using a 299-tap, finite-impulse response, low pass algorithm with a cut-off of 30 Hz (Hamming window, 10 dB down). Quantitative spectral analysis was then performed [50]. To minimize any non-optimized common-mode rejection artifacts contributing to baseline offset (i.e., any twos-complement quantization plateau introduced from the AC amplifiers), a running mean was uniformly subtracted from the signal [51], [52]. The mean subtraction was a simple offline DC offset correction, and not a boxcar filter, with its inherent side effects [53]. Data records were partitioned into consecutive 512-point subunits (2.56 sec each). These partitions were then autoregressively filtered to remove low frequency subharmonics and were tapered using a cosine-bell window [50]. Fast Fourier transformation of these filtered partitions yielded power spectrum estimates in 0.39-Hz steps.
