Period-amplitude analysis and power spectral analysis: a comparison based on all-night sleep EEG recordings

Both period-amplitude analysis (PAA) and power spectral analysis (PSA) were performed on all-night human sleep EEG recordings obtained from 11 subjects. The comparison of the two methods was based on the PAA variables time in band (a wave incidence measure) and rectified amplitude, and on the PSA variables spectral power density and spectral amplitude (the square root of power). The mean time course of these variables was determined for the first 4 nonREM-REM sleep cycles. Spectral power density and spectral amplitude in the delta range were high in nonREM sleep and low in REM sleep, and showed a declining trend over consecutive nonREM sleep episodes. In the frequency range below 2 Hz, rectified amplitude was highly correlated with both time in band and spectral amplitude, and there was no evidence for a dissociation between wave amplitude and wave incidence measures. However, in frequencies above 2 Hz, the modulation of time in band was a mirror image of that below 2 Hz. This result does not reflect a property of the data, but is inherent to the methodology applied. The reversal point of modulation was merely shifted when the high-pass filter settings were changed. It is concluded that band-pass filtering is necessary prior to PAA even for the analysis of the lowest frequency range, and that the indiscriminate use of PAA may give rise to spurious results.