A low-power sensor polling for aggregated-task context on mobile devices

Current smartphones are very rich in the type of sensors to provide various user-behavior tracing and social network sharing, i.e., pervasive social networking applications. Sensor power consumption contributes a significant part of overall power budget in system since each mobile application always actives the related sensors without any restriction. Even worse, in-smartphone sensors are usually activated with full frequencies to match the flushing rates of a group of aggregated tasks, known as full polling-based detection, which results in significant unnecessary activities on sensors and fast battery sucking-up. In this work, we propose a low-power sensor polling strategy for mobile applications to dynamically remove unnecessary sensor activities. With this design, the unrelated sensors can keep in sleeping status for longer time. To schedule mobile sensors, we provide sample-based scheduler as a middleware to model the on-the-fly mathematical relationship between application invoking and sensor real activities. Thus, the scheduler is able to dynamically configure sensor flushing rates under various application context that is executed by users. We evaluate this framework with a wide range of mobile applications. The results show that our new low-power scheduler spends a tiny responding delay (97 ms) in the middleware, as the overhead, to reduce 70% sensor energy consumption, comparing with the conventional exhausting detecting operation. (C) 2019 Elsevier B.V. All rights reserved.