Classical sleep stages and the spectral content of the EEG signal

Polygraphic sleep recordings during 12 nights in 5 healthy volunteers were classified manually into waking and the various sleep stages. The smoothed power spectra of EEG signal segments defined as waking or one of the sleep stages were calculated via segmentation of the EEG signal, using the autoregressive model, and time-dependent fuzzy clustering. The spectra were derived from the prediction coefficients of the segments. The relative power in the delta frequency band was found to increase monotonically with increasing depth of sleep, together with a parallel decrease in the alpha relative power. In most cases alpha relative power had a small peak during REM sleep, and on average the relative power in the sigma frequency band during REM sleep was smaller than the beta relative power. The power spectra from subjects with no waking alpha differed from those of subjects with abundant waking alpha mostly in the relative spectral content of stages 1 and REM. The significance of these findings is discussed in relation to future standardisation of automatic analysis of sleep recordings.