Optimizing battery lifetime-fidelity tradeoffs in BSNs using personal activity profiles

Abstract

Body sensor networks (BSNs) often operate in dynamic environments, with the collected data profiles- and the resulting importance of data-varying throughout the system runtime. Therefore, the potential for power reduction fluctuates with changing user behavior, creating a dynamic battery lifetime-fidelity relationship that is subject to variations throughout the battery lifetime corresponding to an individual's daily activities-past, present, and future. This paper explores the potential for optimizing the tradeoff between meeting a desired battery lifetime and maximizing system fidelity through run-time adaptation of sensor acquisition (duty cycling) and profile-based predictions of an individual's future activities. A "personal activity profile" describes the expected behavior of an individual and is used to inform the desired battery discharge characteristics over time. Using walking activity traces collected from three human subjects wearing Fitbit® trackers over several months in order to develop such activity profiles, the approach is demonstrated in simulation based based on an analytical power model for an inertial BSN platform incorporating recent sensors. Results show improvements over statically setting a duty cycle for constant power consumption with respect to ideally setting the duty cycle based upon a priori knowledge of activities of interest throughout the system lifetime.