Guide
Sleep stages are an EEG definition, not a heart-rate inference.
Light, deep and REM are categories drawn from brain activity, eye movement and muscle tone. Everything else that reports them is estimating them from something adjacent.
Guide
Light, deep and REM are categories drawn from brain activity, eye movement and muscle tone. Everything else that reports them is estimating them from something adjacent.
Sleep staging is not a thing wearables invented and it is not a marketing category. It is a scoring system, currently maintained by the American Academy of Sleep Medicine, that divides sleep into stages using three simultaneous recordings: EEG for brain activity, EOG for eye movement, and EMG for muscle tone at the chin.
A trained scorer reads it in thirty-second windows and assigns each one a stage. That is the reference standard, and it is worth knowing that two expert scorers reading the same night typically agree about 80–90% of the time. The ground truth is itself somewhat soft.
Wake. Alpha when the eyes close, faster mixed activity when open, high muscle tone, blinks and voluntary eye movement.
N1, the lightest sleep. Alpha gives way to lower-amplitude mixed-frequency activity, and slow rolling eye movements appear. Usually a few minutes on the way in. People woken from N1 often deny having been asleep.
N2. The bulk of a night, roughly half of total sleep in an adult. Defined by two distinctive features: sleep spindles, short bursts around 11–16 Hz lasting about half a second, and K-complexes, large isolated slow waves. Both are visible by eye in a good recording and both are genuinely characteristic — you can point at them.
N3, deep or slow-wave sleep. Defined by delta: high-amplitude activity below about 4 Hz occupying at least a fifth of the window. Concentrated in the first half of the night. This is the stage most associated with physical restoration, and the one most people mean when they ask whether they slept well.
REM. The odd one. EEG looks close to wakefulness, the eyes move rapidly in bursts, and muscle tone at the chin drops to its lowest of the night — the body is effectively paralysed. It is that combination, not any one channel, that identifies it, which is precisely why REM is the hardest stage to detect without EOG and EMG.
The discriminations are not equally difficult, and it is worth separating them.
Sleep versus wake is comparatively easy and can be done reasonably well from movement alone. Actigraphy has managed it for decades.
Deep sleep is the most tractable of the stages for a forehead device, because delta is large, slow and widespread — the most forgiving signal there is.
REM is hard without eye and chin channels. A forehead band does see eye movement well, which helps, but distinguishing REM from wake on EEG alone is exactly the case where the two look alike.
N1 versus N2 depends on spotting spindles and K-complexes, which needs a decent signal and enough bandwidth to see them.
Because they are estimating different things by different means, and because there is no requirement that they agree. Two devices can report deep sleep differing by an hour on the same night and neither be lying — they are running different models over different sensors against different notions of the same category. We wrote about this in more detail in how accurate sleep trackers are.
Total sleep time and timing consistency are the most reliable things almost any device gives you, and they also happen to be the two most worth acting on. The stage breakdown is more interesting and less trustworthy, and it is where night-to-night noise most often gets mistaken for a trend.
A single night's deep sleep figure is not something to change your behaviour over. A four-week pattern in when you fall asleep might be.
Further reading