Sleep Monitoring Technologies — Accelerometer, PPG, Acoustics, and mmWave Radar: A Technical Comparison

Consumer sleep monitoring is built on a small set of sensor modalities, each with distinct signal origins, failure modes, and accuracy ceilings. This article compares them at the signal level: what each one truly measures, how the major vendors combine them, where the unsolved problems lie, and how mmWave radar differs.

1. Measurement modalities

1.1 Accelerometer / IMU (actigraphy)

A 3-axis accelerometer produces an acceleration signal from which an "activity count" is derived. Periods of sustained low activity are scored as sleep (actigraphy). Sleep/wake classification is essentially a threshold plus temporal model over movement amplitude and duration. Limitation: it infers sleep from stillness, not from physiology, so it captures total sleep time far better than sleep stage.

1.2 PPG (photoplethysmography)

An LED illuminates the dermis; a photodiode measures back-scattered light modulated by pulsatile arterial blood volume. Each cardiac cycle produces one pulse → heart rate. From the inter-beat interval (IBI) series you derive HRV (e.g., RMSSD, SDNN) and, via respiratory sinus arrhythmia (RSA) or amplitude modulation, an estimated respiration rate. Crucially, respiration is inferred, not measured — the wrist carries no respiratory motion of its own.

1.3 Acoustic

A MEMS microphone captures snoring (partial upper-airway obstruction), breathing rate, and wheezing. It is a useful independent cue for breathing disruptions, but it is environmentally noisy and raises privacy considerations.

2. Vendor implementation strategies

• Accelerometer-dominant — Xiaomi / Apple: actigraphy-based total sleep time; features = time-since-last-movement and movement amplitude.
• HR-dominant — Huawei / VIVO: HRV-derived respiration amplitude used as a sleep-stage proxy and quality score.
• Sensor fusion — Samsung / Honor / Garmin / OPPO: accelerometer + PPG (+ sometimes SpO₂ / audio); fusion improves staging robustness.
• Ecosystem coupling: same-brand phone telemetry (screen-on, device motion) is fused to reject "awake-in-bed" and active phone use.

3. Unsolved problems and error sources

• NREM light/deep discrimination: the physiological signature at the wrist is weak; staging is probabilistic, not deterministic.
• REM: estimated from HR/actigraphy proxies, never from EEG.
• Ring vs wrist PPG SNR: rings (Oura-class) maintain higher contact pressure and vessel proximity → better PPG SNR. Watches suffer strap-tension variance and pressure-induced blanching / arterial-occlusion artifacts → signal loss.
• Gold standard = PSG (polysomnography): multi-channel EEG/EOG/EMG/ECG. Consumer wearables are orders of magnitude less accurate for sleep-stage classification.

4. mmWave radar

• Principle akin to actigraphy but contactless: FMCW / Doppler sensing resolves sub-millimeter displacement → motion amplitude/timing (sleep/wake) and, from thoracic micro-motion, respiration; cardiopulmonary micromotion also yields a heart-rate estimate.
• Respiration measured directly: chest-wall displacement is observed, giving a higher-fidelity respiration signal than wearables that infer it from the wrist.
• Continuous, no-wear, ambient: enables life-pattern analytics — prolonged inactivity, nocturia frequency, routine deviation (e.g., elderly / illness). No wearable to wear or charge.
• Limitation: bounded sensing volume; requires presence in the beam; no wearable fallback.

5. Comparison table

6. Takeaways

Respiration tracks sleep quality; heart rate tracks health. Each modality emphasizes a different axis, and the right choice depends on the deployment (worn vs ambient) and the question being asked.

7. Conclusion

No consumer method approaches PSG's staging authority. mmWave radar is best positioned as a contactless, continuous layer that emphasizes respiration and routine — complementary to, not a replacement for, worn sensors.