A Lightweight Intelligent Bracelet for Real-Time Emergency Detection Using PPG Signals in Solo-Living Elderly

This is the authors' abstract. We don't add key points for this paper.

Solo-living older adults face a specific risk: a fall or a sudden cardiac event may go unnoticed for days. Existing wearables combine accelerometers, electrocardiogram electrodes and gyroscopes, which increases bulk, shortens battery life and reduces compliance. This paper presents a bracelet that detects emergency events from wrist photoplethysmography (PPG) alone. A hybrid convolutional, long short-term memory and transformer classifier, quantised to 8 bits, runs entirely on an nRF52840 microcontroller paired with a MAX30102 sensor and alerts caregivers over Bluetooth Low Energy without user interaction. On a held-out evaluation subset of public and proprietary wrist-PPG recordings, the on-device model reached 98.6% accuracy, 98.8% recall and a 0.80% false alarm rate at 96 ms latency per 5 s window; measured average power of 1.3mW implies roughly 24 days on a 200mAh cell. Because the emergency class rests on surrogate labels, the results establish technical feasibility rather than clinical accuracy.