How do smartwatches implement continuous heart‑rate monitoring?

31 Ağu 05:31

Optical PPG (photoplethysmography) sensors in today’s smartwatches rely on three core pieces: green LEDs, a photodiode, and a high-speed SoC running continuously-updated algorithms, not on ECG-style metal contact. When the green LEDs fire every few milliseconds they light up the capillaries just beneath your skin; the returning light flickers subtly with each heartbeat because blood volume changes the reflection index. The photodiode acts like a tiny camera, sampling the reflected photons at 25–100 Hz. From that raw trace you get a “volume pulse” waveform whose peaks map 1-to-1 to systoles, giving raw BPM in real time. The heavy lifting is in noise reduction and heart-rate extraction. Every wrist movement, ambient light, or even hair on your arm injects motion artefacts that can dwarf the true PPG signal. Modern watches run a sliding FFT or wavelet transform on the raw data inside a dedicated heart-rate DSP (often an ARM Cortex-M running CMSIS-DSP). They also fuse inertial measurements from the IMU to apply a high-pass filter in three axes, cutting motion noise by >85 %. What most people don’t realize is that the SoC isn’t awake the whole second—PPG sampling fires in short 8 ms bursts every 100–150 ms, while the core sleeps in between; this gives a 20–30 mA average load instead of the 150 mA you’d get with continuous data capture. Accuracy versus chest straps is the trillion-dollar question. In controlled lab tests (resting, moderate activity) top-tier watches hit ±2 % MAE if the sensor is well-seated. Outdoors on a run, ±5 % becomes typical. Chest straps use a single-lead ECG signal that is literally measuring the heart’s electrical impulse, so they’re reference-grade. Smartwatch algorithms compensate by cross-checking with accelerometer data to detect distortions and will warn you when motion exceeds their error envelope. Apple, Garmin and Whoop all publish white-papers showing >90 % correlation during steady-state conditions, but the gap widens during intense HIIT or with darker skin tones or tattoos. If you want to dive deeper, grab the “PPG Development Guidelines” PDF from TI (their AFE4403 EVM schematics are gold) and the 2022 MIT paper “Adaptive PPG Motion Artefact Cancellation via Deep Learning”—it’s the closest public peer-review I’ve seen to what’s actually inside your watch.