Optimizing the Duty Cycle in Smartwatches for Sleep Sensing

Abstract

Smartwatch devices worn during sleep can improve people's sleep or alertness upon waking through a range of wellness-oriented, non-diagnostic, data-dependent, and time-sensitive interventions. However, higher charging burdens affect long-term compliance with smartwatch health systems and operations that wake the CPU, such as gathering sensor data and running intervention algorithms, and non-trivially consume device batteries. Here, we assess methods to reduce battery consumption while preserving the device's ability to perform these time-sensitive interventions. We propose a novel system to optimally interpolate the key sensor values used to deliver interventions while only waking the CPU when necessary. Our approach achieves an absolute 13% reduction in battery consumption with only a 5.4% median absolute error rate over 8 experiments with a population of people with sleep disorders, a more challenging population to estimate sensor values. This work illustrates that sleep sensing and wellness intervention systems can be optimized for battery efficiency without significantly compromising data quality, potentially enhancing smartwatch-based sleep monitoring solutions' long-term adoption and effectiveness.