I reverse-engineered my Pixel Watch's raw PPG sensor to check for sleep apnea

A developer investigates repeated low overnight oxygen readings from a Fitbit, which raised the possibility of sleep apnea despite few typical symptoms. Because wrist-based oxygen measurements can be distorted by position and pressure, he builds an app for a Pixel Watch 1 to record movement, heart rate and raw optical pulse data, then compares nine nights of recordings with Fitbit’s sleep summaries. He stresses that this is a self-experiment by someone without medical training, not a diagnosis or substitute for a sleep study.

The first results show that wrist motion can estimate breathing rate reasonably well: across nine nights, estimates were usually within about one breath per minute of Fitbit’s, with larger differences during REM sleep. More importantly, prolonged oxygen readings below 90% often coincide with a motionless wrist and steady heart rate rather than repeated oxygen drops and heart-rate spikes. On one night, the low readings repeatedly appeared in a particular wrist position and rose within minutes after a roughly 20-degree turn, while breathing and heart rate stayed steady. That pattern strongly suggests sensor-position artifacts, though some episodes remain unexplained and movement could also change sleep posture. The experiment then turns to whether beat-level heart rate, pulse-wave signals and raw optical channels can reveal apnea patterns or measure oxygen more directly; the available findings do not establish a medical conclusion.