Sleep Stages
Sleep Stages — The four stages your sleep cycles through (N1, N2, N3, REM) — the underlying biology that consumer sleep trackers attempt to estimate from peripheral signals like heart rate and HRV.
The four sleep stages, briefly
A healthy adult sleep cycle moves through four stages, each with distinct EEG patterns measurable in a polysomnography lab:
- N1 (light sleep, transitional): the first 1-7 minutes after sleep onset; easily disrupted.
- N2 (light sleep): the bulk of overnight sleep; characterized by sleep spindles and K-complexes on EEG.
- N3 (deep / slow-wave sleep): the restorative stage; the bulk of N3 happens in the first half of the night.
- REM (rapid eye movement): the dreaming stage; the bulk of REM happens in the second half of the night.
A typical sleep cycle is ~90 minutes and a healthy adult will go through 4-6 cycles per night. The composition of each cycle changes across the night, with more N3 early and more REM late.
Why consumer trackers can only approximate
The clinical gold standard for sleep stage classification is polysomnography (PSG): EEG, EOG (eye movement), and EMG (chin muscle) sensors recorded simultaneously for hours. The classifications from PSG are made by a human scorer applying the AASM scoring rules to 30-second epochs. This is not a thing that fits in a smart ring.
Consumer trackers infer sleep stages from peripheral physiological signals — primarily heart rate, HRV, motion (accelerometer), and sometimes blood oxygen — and apply machine-learning models to predict which stage the wearer is most likely in at each moment. The accuracy of this prediction is bounded by the strength of the relationship between the peripheral signal and the underlying brain state, which is meaningful but imperfect.
The state of the art for consumer trackers is roughly 80-85% epoch-by-epoch agreement with PSG, with REM and deep-sleep stages being the most difficult to discriminate accurately. Total sleep time is much more accurate than stage-by-stage classification — most consumer trackers get within 15 minutes of PSG total sleep time on a good night.
What you can and can’t trust
Trust the total sleep time and rough cycle structure. Trust the trends across weeks (was your average sleep duration shorter this week than usual?). Don’t trust the absolute stage durations on any single night; the noise on the per-stage classification is large enough that a single night’s REM minutes is not a meaningful number.
The quality of consumer sleep tracking has improved meaningfully in the last 5 years, mostly because the validation studies (Stone et al. 2024 on Oura, the Apple Watch validation series, Whoop’s published comparisons) have given the manufacturers feedback loops their algorithms can train against. The Oura Ring 4 represents the current state of the art for consumer-grade sleep staging at the wrist or finger.
Why this matters for our verdicts
Sleep stages are the dominant criterion in our shift-worker sleep-tracker verdict (Oura Ring 4). HRV-based staging holds up across irregular sleep schedules where actigraphy-anchored staging fails. The same sensor architecture is the reason for our smart ring verdict — finger-worn devices produce cleaner sleep-stage classifications than wrist-worn devices.
Related concepts
For the HRV signal that consumer devices use as the foundation for stage prediction, see HRV. For the machine-learning models that translate sensor signals into stage predictions, see machine learning.