Statistical verification using learned usages for evaluating energy-efficient mobile device design

Abstract

Energy efficiency is an important factor that mobile device designers must consider at the beginning of the design because the hardware is difficult to change once it is implemented. Many researchers have studied energy-efficient mobile designs, but not focused on actual mobile device usages. Verification on the energy efficiency of mobile device using actual usages is needed. However, two problems may arise in the verification based on mobile device usage: the difficulty in acquiring a large amount of usage data due to invasion of privacy, and the state explosion problem due to the amount of mobile device usage data. We propose statistical verification using learned usages that can overcome both problems to evaluate energy-efficient mobile device design. We generated thousands of learned usages based on 400 actual usages. The generated usages are given in verifying the target mobile device for energy efficiency, and the state explosion problem is bypassed using statistical verification. The verification results allow the designers to optimize the module with a high probability of being used or to evaluate whether the battery is used as expected.