Dr. Albertario,

You bring up some excellent points. I used a Davis, Davis & Loomis approach to scoring as an undergrad in 1963-64, then when starting grad school in 1971 I switched to R&K (1968) and the 30-second epoch. I've used (arbitrarily) a 20-sec epoch for EEG recorded during open highway driving. Of course, many projects by others in past decades have used analog and digital filtering to automate scoring, but new attempts like this one are always interesting.

I seek your opinion on the R&K epoch length, i.e., 30 seconds. I was a student helper and then tech in Ross Adey's lab, 1958-1965, and I recall discussions and concerns among the principals there about the appropriateness of using Fourier analysis as a data reduction technique for the EEG. One output of these discussions was Don Walters' 1963 and 1965 papers on the EEG time series. I know that Don worried about the stability of the EEG signal with regard to the Fourier reduction. So, considering that during SWS we must deal with a 0.5-Hz signal that is not always very sinusoidal, might there be a minimum epoch length during which we may say that the 0.5-Hz EEG time series has stabilized? Has that question been addressed?

Also, on the more behavioral side, how long should an epoch extend in time before we declare that the sleeper is in a given stage? For example, we use half of a 30-sec epoch in “delta” (visual scoring) as a cut-off for stage 4. Should we perhaps use a 15-sec epoch?

I’m just curious. Mostly retired now. Thanks.

Disclaimer: I am not a doctor. I am a sleep tech by trade. 

.5hz is VERY VERY slow almost DC in nature. But, that aspect to the EEG is really only of possible import during slow wave sleep. If we were to filter it out of all other times, it would not alter the decision making for that "staging." But is staging what we need to "electronify?" As, per your query :"...on the more behavioral side, how long should an epoch extend in time before we declare that the sleeper is in a given stage?"

http://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1003866 (attached)

The linked paper has attempted to elucidate that. I bring up this paper because it also justifies my usage of  delta+theta (as a single entity) within my z-ratio analysis (www.z-eeg.com). They found: "Only when the power in their brain waves at the theta and delta frequencies had risen did both the behavioural and physiological measures indicate that they were asleep." Delta+Theta are my SLOW BIN and ALpha +Beta is my "FAST BIN." Put them in the simple ratio formula (Slow-Fast)/(Slow+Fast) and that is z-ratio. It is limited from -1 to +1 and zero is where the two bins are equal. As the power in the eeg (each second) transitions from fast to slow, the z-ratio changes from negative (-) to positive (+). It is very simple and elegant.  I have since now started to make it a CSA (compressed spectral array) on top of the z-ratio envelope and the output shows not only the shifting of fast and slow but also the underlying standard frequency bins contribution. 

So, if we could move away from static 20 or 30 sec "images" of EEG and instead move toward fluid, dynamic, real-time second by second analysis we could find a faster means by which to determine state. I used z-ratio is a backwards averaging scheme to look at applying older majority scoring rubric to this new second by second analysis and as long as you wait the initial period of your epoch of interest, you can use analyses in a backward averaging capacity and still allow the usage of majority rule, but in a second by second updated fashion. 

The move I speak of is akin from the world moving from single photographs every 30 seconds to a smoother hi-frame rate change analysis updated every second. It could even get faster if the FFT was to be shortened and go intra-second. I stay away from doing that because the FFT is an analysis that has its justification in measuring power in Cycles per second. We are best to stay at per second increments. This is my opinion. 

it is my opinion that we need to move away from "pages" of sleep recordings and move to seconds.

I do not think I answered your queries fully.

As to acceptable rubrics for studying