How a Wearable Detects Breathing Rate in Sleep

Breathing rate during sleep — typically expressed as breaths per minute — is one of the physiological signals that consumer wearables now report alongside heart rate and movement data. The figure appears straightforwardly on a morning dashboard, but the path from chest expansion to a number on a screen involves several layers of inference rather than any direct measurement of airflow or lung volume.

This piece covers the hardware and signal-processing chain that produces a breathing-rate estimate on a wrist-worn or ring-form sleep tracker, what biological systems feed into that estimate, and where the chain introduces errors that a raw number does not reveal.

From Chest Wall to Wrist: How the Signal Is Derived

Consumer wearables do not contain airflow sensors. Instead, they infer breathing rate from signals that are already being collected for other purposes — primarily the photoplethysmography (PPG) optical sensor used to detect blood volume changes at the skin surface. That same sensor, already described in the context of how a wearable reads heart rate during sleep, illuminates the skin with green or infrared LEDs and measures the variation in reflected light as blood pulses through the capillaries beneath.

Each breath modulates the PPG signal in two distinct ways. First, inhalation lowers intrathoracic pressure and slightly increases venous return, altering the amplitude of the pulse waveform in a pattern called respiratory-induced intensity variation (RIIV). Second, breathing shifts the baseline of the waveform up and down at the respiratory frequency through a process called respiratory-induced baseline wander. Firmware on the device — or processing on a paired phone or cloud server — applies a bandpass filter to isolate the frequency component of the PPG signal that falls within the plausible range for human breathing (roughly 0.1 to 0.5 Hz, corresponding to 6 to 30 breaths per minute) and extracts the dominant frequency in that band as the estimated breathing rate.

Some wearables supplement the PPG-derived estimate with data from a three-axis accelerometer. The chest wall rises and falls with each breath, and when a device sits on the wrist or finger, that periodic motion — though attenuated — can still be detected if the wearer is lying still. Algorithms fuse the accelerometer signal with the PPG-derived signal, weighting each source based on signal quality at any given moment. During periods of high motion, the accelerometer contribution is typically down-weighted because movement noise overwhelms the small respiratory oscillation.

A third contributor, used in some devices, is heart rate variability (HRV). Breathing directly modulates the interval between heartbeats through a mechanism called respiratory sinus arrhythmia: heart rate rises slightly on inhalation and falls on exhalation. Because heart rate variability reveals information about autonomic state, the same inter-beat interval data used to compute HRV metrics can also be interrogated for the respiratory frequency embedded within it. This approach, called HRV-derived respiration, requires precise beat detection but is independent of skin-contact quality at the LED site.

The final output — breaths per minute averaged over a sleep epoch, typically 30 seconds to several minutes — is what appears in the user-facing app. Most devices report a nightly average or a per-stage breakdown rather than a continuous second-by-second trace.

Biological Systems That Shape the Breathing-Rate Signal

The autonomic nervous system. Breathing rate during sleep is not fixed; it is regulated moment-to-moment by the brainstem's respiratory centers, which respond to blood oxygen and carbon dioxide levels. The autonomic nervous system also governs heart rate and vascular tone, meaning that the same neural activity that modulates breathing simultaneously modulates the PPG waveform the sensor is reading. This coupling is what makes PPG-derived respiration estimation possible, but it also means that any autonomic arousal — from noise, temperature change, or stress hormones interfering with sleep architecture — will perturb both signals together.

Sleep stage physiology. Breathing pattern changes substantially across sleep stages. During non-REM light sleep, breathing is relatively regular. During slow-wave (deep) sleep, it becomes slower and more rhythmic. During REM sleep, breathing becomes irregular and faster, mirroring the heightened and variable neural activity of that stage. A wearable that correctly identifies sleep stage transitions should therefore show corresponding shifts in estimated breathing rate, though the stage classification and the breathing estimate are derived from overlapping signals, which introduces circularity into any cross-validation.

Body position and the upper airway. The geometry of the upper airway changes with body position. Supine sleep narrows the posterior pharyngeal space, increasing airflow resistance and altering the pattern of chest-wall movement. These positional changes affect both the amplitude and regularity of the respiratory signal reaching the wrist sensor.

Skin perfusion and temperature. PPG signal quality depends on blood flow to the skin surface. Cold ambient temperatures cause peripheral vasoconstriction, reducing perfusion at the wrist or finger and degrading the optical signal. Fever or exercise-induced vasodilation has the opposite effect. Either condition can shift the signal-to-noise ratio of the respiratory modulation the algorithm is trying to extract.

Where Breathing-Rate Estimates Break Down or Mislead

The most common source of error is motion artifact. Even small wrist movements — a partial positional shift, a scratch, or a partner disturbing the bed — introduce broadband noise into the PPG signal that can temporarily overwhelm the low-amplitude respiratory modulation. Algorithms attempt to detect and discard these epochs, but short bursts of artifact can produce a spurious frequency peak in the respiratory band that registers as an anomalous breathing rate rather than a data gap.

Device fit is a persistent structural problem. A wristband worn loosely allows the sensor to slide against the skin with each heartbeat, creating a mechanical artifact at the cardiac frequency and its harmonics. If those harmonics fall within the respiratory frequency band, the filter cannot cleanly separate them from genuine respiratory signal. Ring-form devices, which sit on a finger with better optical coupling, are generally less susceptible to this specific artifact, but are not immune to motion noise.

Averaging obscures clinically relevant events. A nightly average breathing rate of 15 breaths per minute is consistent with normal adult respiration, but that average could mask repeated brief episodes of very slow breathing (hypopnea) or complete cessation (apnea) that were averaged out by surrounding normal epochs. Consumer wearables are not validated diagnostic instruments for sleep-disordered breathing; a sleep study conducted in a clinical setting uses direct airflow measurement at the nose and mouth, respiratory effort belts around the chest and abdomen, and pulse oximetry simultaneously — a level of redundancy and signal fidelity that a single wrist-worn optical sensor cannot replicate.

Epoch length also shapes what the data can show. If a device averages breathing rate over five-minute windows, it cannot resolve the breath-by-breath irregularity that characterizes REM sleep or the brief respiratory pauses associated with upper-airway obstruction. The smoothed number that appears in an app is a statistical summary, not a physiological trace.

What Tracker Data Actually Shows — and What It Does Not

A consumer wearable's breathing-rate record typically shows a nightly average (often 12–20 breaths per minute in healthy adults), a trend line across nights, and in some devices a per-stage breakdown. These figures are useful for identifying large, sustained deviations from an individual's own baseline — a persistent elevation might reflect illness, elevated resting heart rate, or changes in sleep stage distribution — but the number itself carries an uncertainty range that device manufacturers rarely publish in consumer-facing materials.

Validation studies comparing PPG-derived breathing rate against clinical respiratory monitoring (the reference standard) report mean absolute errors typically in the range of 1–3 breaths per minute under controlled conditions, with larger errors during high-motion or low-perfusion epochs. That error range is acceptable for tracking population-level trends but is not sufficient for clinical screening of conditions defined by small deviations from normal, such as early-stage respiratory compromise.

The record also does not capture oxygen saturation unless the device includes a separate pulse oximetry sensor (a red-wavelength LED in addition to the green). Breathing rate and blood oxygen saturation are related but distinct measurements; a device reporting only breathing rate provides no direct information about gas exchange efficiency. Understanding how deep sleep differs from light sleep in terms of respiratory physiology requires knowing both the rate and the regularity of breathing, neither of which is fully captured by a nightly average figure.

A polysomnography record from a sleep laboratory, by contrast, captures nasal and oral airflow with thermistors or pressure transducers, chest and abdominal effort with inductive plethysmography bands, and oxygen saturation continuously — all time-stamped to the millisecond alongside EEG, EMG, and EOG channels. The consumer wearable record and the clinical record are not equivalent documents, even when they report the same unit of measurement.

The breathing-rate figure on a wearable dashboard is the end product of a signal chain that begins with photons reflecting off capillary blood and passes through several layers of algorithmic inference before becoming a single integer. The machinery that produces it is real and grounded in established respiratory physiology, but the number it yields is an estimate with a confidence interval that the display does not show.

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.