Sensor-less, Event-Driven and Fine-Grain Instantaneous Power Estimation of Android Smartphones

Abstract

Multi-core processors and displays, which have a large-size panel over five-inches in diameter and high resolution close to 300 dot-per-inch, have become commonplace in smartphone designs, and they make long battery life a very challenging task even with a high capacity (about 3,000 mAh) Li-ion battery. Such mobile platforms are comprised of several major components including a multi-core processor, display with its controller, baseband signal processor, different types of memory, radio frequency communication module, various sensors, etc., and achieving power savings in each of those components is very important to reduce total system power consumption. Accurate power estimation for each of these components is the first step required to realize system-wide power optimization in smartphones. Moreover, accurate power estimation without the need for having physical sensors is crucial because most commercial smartphones do not have current sensors for their major components to measure its power consumption. Several researches present sensor-less power estimation methods that predict the power dissipation of a smartphone without any power measurement sensors. However, previous sensor-less power estimation techniques suffer from various problems such as the aliasing, inaccurate logging timestamp, and unobservable devices.

This paper introduces a novel sensor-less, event-driven power analysis framework for providing highly accurate and nearly instantaneous estimates of power dissipation in an Android smartphone. The key idea is to collect and correctly record various events of interest within a smartphone as applications are running on the application processor within it. This is in turn done by instrumenting the Android operating system to provide information about power and performance state changes of various smartphone components at the lowest device driver layer of the kernel to avoid time stamping delays and component state observability issues. This technique then enables one to perform fine-grained (in time and space) power estimation in the smartphone. In addition, the proposed method takes account of the multi-core processor and organic light emitting diode (OLED) display in its power model while the previous power estimation techniques do not consider. Experimental results show significant accuracy improvement compared to previous approaches. The estimation error of the proposed method is lower by a factor of two with a good fidelity to actual current measurements.