Wearable biosignals · Free-living activity
A free-living recording from one participant in an NIH field study — 104 hours of ENMO across 11 days — raw wrist acceleration reduced to ENMO on-device, then screened in real time for moderate-to-vigorous physical activity.
The wristband streams 15-second ENMO averages; they are plotted here on a uniform 6-second grid, so each point is a resampled value rather than a new measurement. Activity concentrates in waking hours and collapses overnight — the exact free-living structure health-sensing models need, and where signal quality usually breaks.
Clean free-living biosignals don't happen by accident — they are an engineering result.
The wristband derives gravity-adjusted acceleration (ENMO) on-device and streams compact 15-second summaries over BLE — so the phone stores motion-relevant features instead of gigabytes of raw high-rate waveform.
When 70% of ENMO in a 7-minute window exceeds 100.6 mg, the device flags moderate-to-vigorous activity and fires an in-the-moment survey — capturing context at the moment the behavior happens, not hours later from recall.
Offline-first buffering, schema-validated writes, and automatic BLE reconnection keep the record intact when phones die, networks vanish, or the OS kills the app — the platform sustains ~7.7M records a day across the deployment.
Records are keyed by a pseudonymous device ID, GPS is AES-256 encrypted on-device, and everything lives on university HPC — the research team never has access to participant identities.