The Tradeoff between Energy Efficiency and User State Estimation Accuracy in Mobile Sensing

People-centric sensing and user state recognition can provide rich contextual information for various mobile applications and services. However, continuously capturing this contextual information on mobile devices drains device battery very quickly. In this paper, we study the tradeoff between device energy consumption and user state recognition accuracy from a novel perspective. We assume the user state evolves as a hidden discrete time Markov chain (DTMC) and an embedded sensor on mobile device discovers user state by performing a sensing observation. We investigate a stationary deterministic sensor sampling policy which assigns different sensor duty cycles based on different user states, and propose two state estimation mechanisms providing the best "guess" of user state sequence when observations are missing. We analyze the effect of varying sensor duty cycles on (a) device energy consumption and (b) user state estimation error, and visualize the tradeoff between the two numerically for a two-state setting.