Power management for long-term sensing applications with energy harvesting

Power management of mobile embedded devices remains important with the slow growth of battery energy density relative to computing power. Estimating the state of charge (SOC) of battery is key for scheduling power intensive tasks, yet current approaches either require dedicated hardware, use battery voltage as a loose indicator of SOC, or track the net energy flow from the battery over time where inevitable small errors in instantaneous estimation can lead to large cumulative estimation errors and significantly degraded sampling strategies. In this paper, we propose a method for estimating SOC of a node's battery based on the conation of two noisy inputs: (1) the net current flow in the battery for instantaneous net energy flow estimation; and (2) the battery voltage as a measure of absolute SOC. Using empirical data from several weeks of flying fox tracking experiments, we validate our approach in terms of the accuracy of SOC prediction and show how SOC prediction can be used to adaptively schedule tasks for energy neutral operation of sensing applications.