How a Wearable Reads Heart Rate During Sleep

A wearable sleep tracker worn at the wrist or finger does not directly observe the sleeping brain. Instead, it reads a proxy signal — the pulse of blood moving through superficial vessels — and uses that signal to make inferences about sleep depth, autonomic state, and stage transitions. The physical basis of that reading is a technology called photoplethysmography, or PPG, which has been in clinical and consumer use for decades but carries specific constraints when applied to a moving, sweating body lying in variable positions for eight hours.

This piece covers the optical mechanism by which a wearable captures heart rate data during sleep, the biological systems whose activity that data reflects, the conditions under which the signal degrades or produces an unexpected result, and what a sleep record built from wrist-based heart rate actually shows compared to what a clinical recording would show.

How Optical Heart Rate Sensing Works at the Wrist

Photoplethysmography works by shining light — typically green LEDs, sometimes supplemented by red or infrared wavelengths — into the skin and measuring how much of that light bounces back to a photodetector on the same sensor module. Blood absorbs green light more strongly than the surrounding tissue does. Each time the heart contracts and a bolus of blood moves through the capillaries and arterioles beneath the sensor, the amount of reflected light drops slightly. Between beats, as blood recedes, reflected light increases. The photodetector records this oscillation continuously, and the device's onboard processor converts the frequency of those oscillations into a beats-per-minute figure.

The raw PPG waveform also contains a wealth of timing information beyond simple rate. The interval between successive peaks in the waveform corresponds to the interval between heartbeats — the R-R interval in electrocardiograph terminology. Consumer wearables extract these intervals and use them to compute heart rate variability, a measure that reflects moment-to-moment fluctuation in autonomic nervous system balance and is one of the primary signals used to infer sleep stage.

During sleep, the algorithm running on the device samples the PPG signal continuously — often at 25 Hz or higher — and combines heart rate, heart rate variability, movement data from an accelerometer, and sometimes skin temperature into a classification model. That model assigns each epoch of time (commonly 30-second windows, matching the convention used in clinical polysomnography) to a predicted sleep stage: wake, light sleep, deep sleep, or REM. The stage assignments are the foundation of the sleep score that most consumer apps display in the morning.

The heart's behavior changes substantially across different sleep stages. During slow-wave (deep) sleep, the autonomic nervous system shifts heavily toward parasympathetic dominance: heart rate falls, beat-to-beat intervals lengthen, and heart rate variability typically rises. During REM sleep, the autonomic balance is less stable; heart rate and variability fluctuate more dynamically, sometimes spiking during vivid dreaming. Light NREM sleep sits between these extremes. These physiological signatures are what the classification algorithm is trained to detect.

Biological Systems the Heart Rate Signal Reflects

The autonomic nervous system. The balance between sympathetic ("activating") and parasympathetic ("calming") branches of the autonomic nervous system is the primary driver of heart rate changes during sleep. The parasympathetic branch, acting through the vagus nerve and releasing acetylcholine at the sinoatrial node, slows the heart. The sympathetic branch, releasing norepinephrine, accelerates it. The PPG-derived heart rate and its variability are downstream readouts of this competition, making them indirect but real-time windows into autonomic state.

The sinoatrial node. The sinoatrial node, located in the right atrium, is the heart's intrinsic pacemaker. It generates the electrical impulse that triggers each beat. The rate at which it fires is modulated by autonomic input, circadian signals, body temperature, and metabolic demand — all of which change across a night of sleep. A wearable does not measure the node's electrical activity directly; it measures the mechanical consequence of that activity in the peripheral vasculature.

The circadian system. Core body temperature and heart rate both follow a circadian rhythm that is partly independent of sleep itself. The circadian rhythm drives a fall in core temperature and resting heart rate in the hours after midnight, a pattern that overlaps with but is not identical to sleep-stage-driven changes. A wearable reading heart rate at 3 a.m. is receiving a signal shaped by both the sleep stage the person is in and the phase of the circadian oscillator — the two are not always easy to separate from the wrist signal alone.

Peripheral vasculature and skin perfusion. The PPG signal is captured at the skin surface, where blood flow is regulated by local temperature, hydration, and vasomotor tone. Vasoconstriction — which occurs when the body is cold, stressed, or during certain stages of sleep — reduces perfusion at the wrist and degrades signal quality. Vasodilation, which accompanies deep sleep and warming, generally improves it.

Where the Wrist Signal Breaks Down or Misleads

Motion artifact. The most common source of error in wrist-based PPG is movement. Even small shifts in wrist position can change how firmly the sensor contacts the skin, altering the optical path and producing spurious peaks in the waveform. Most wearables use accelerometer data to detect and partially correct for motion artifact, but the correction is imperfect. Restless sleepers, people who change positions frequently, and individuals with periodic limb movements during sleep are all likely to accumulate more erroneous readings than still sleepers.

Skin tone and optical absorption. Research published in peer-reviewed literature has documented that PPG-based pulse oximetry and heart rate accuracy varies across skin tones, with darker skin absorbing more of the green light used by most sensors. This reduces the signal-to-noise ratio and can increase error rates in heart rate estimation. The FDA has noted this limitation in the context of pulse oximetry, and the same optical physics applies to heart rate sensing.

Loose fit and sensor positioning. A wearable worn loosely slides on the wrist during sleep. Even a few millimeters of displacement changes the tissue the sensor is reading. Ring-form factors, which wrap the sensor more completely around the finger, reduce this problem but do not eliminate it, since fingers also move during sleep and blood flow to extremities drops with cold ambient temperature.

Algorithmic stage classification errors. Heart rate and movement alone do not reliably distinguish all sleep stages. The gold standard for sleep staging — polysomnography — uses electroencephalography (EEG) to read brain electrical activity directly, along with electromyography, electrooculography, and respiratory monitoring. A wrist wearable has none of these channels. Studies comparing consumer wearable sleep staging to simultaneous polysomnography have found that wearables perform reasonably well at distinguishing wake from sleep, and at identifying REM versus non-REM in aggregate, but are substantially less accurate at distinguishing light NREM from deep NREM on an epoch-by-epoch basis. The numbers in a consumer sleep app's stage breakdown should be understood as estimates with meaningful uncertainty, not clinical measurements.

Individual physiological variation. Resting heart rate varies widely between individuals. A trained endurance athlete may have a resting heart rate in the low 40s; an older sedentary adult may have one in the 70s. Classification models trained on population averages can misread an individual whose physiology sits at the extremes of that distribution. A low heart rate that indicates deep sleep in one person may trigger a different classification in another.

What the Sleep Record Shows — and What It Does Not

A typical consumer sleep record derived from wrist heart rate data shows: total time in bed, estimated total sleep time, a breakdown of time assigned to each stage (light, deep, REM, and wake), a resting heart rate figure (usually the lowest sustained rate during the night), and a heart rate variability summary — often a single number representing the average or a specific epoch. Many devices also generate a composite "sleep score," which is a proprietary index that weights these variables according to the manufacturer's algorithm. A sleep score is not a medical diagnostic value; it is a summary statistic whose formula varies between devices and is not standardized across the industry.

What the record does not show: it does not capture EEG-defined sleep spindles, K-complexes, or slow-wave activity — the electrical events that define sleep stages in clinical practice. It does not measure respiratory effort or airflow, so it cannot directly detect apnea events, though some devices use heart rate pattern changes and accelerometer data to flag probable respiratory disturbance. It does not measure blood oxygen saturation unless the device includes a separate pulse oximetry sensor (a red/infrared LED pair rather than the green LED used for heart rate). Even when such a sensor is present, consumer-grade pulse oximetry carries its own accuracy limitations that the FDA has documented.

The heart rate data that does appear in the record is, in aggregate, a reasonable reflection of autonomic state across the night. Trends over many nights — a sustained elevation in resting heart rate, a persistent drop in heart rate variability — are more meaningful than any single night's figures, because the noise in individual measurements is large relative to the signal of a single night's biology.

The wrist-based optical heart rate sensor is a genuine physiological instrument operating under real physical constraints. It reads a peripheral vascular signal that is several steps removed from the brain activity that defines sleep, and the algorithms that translate that signal into stage estimates carry uncertainty that consumer interfaces rarely make visible. The data is most informative as a longitudinal record of cardiovascular and autonomic patterns across many nights rather than as a precise nightly audit of sleep architecture.

Note: This explains how sleep works as a system. It is not medical advice, it is not a diagnosis, and it is not a substitute for a licensed healthcare provider. Check the cited sources for current clinical guidance.