Semi-Markov Decision Process Based Power Management For Smartphones

Mobile devices,such as smartphones and tablets,have become part of our daily life.With more sophisticated hardware components being equipped on the mobile devices,the developers are exploiting them to provide state-of-the-art user experience.These come at the cost of high drain of battery.However,today's battery technology is not advanced enough for us to make heavy use of the mobile devices,and the energy density of battery has grown at a comparably insignificant rate.At the same time,various power management techniques are used to design energy-efficient mobile systems.Existing power management approaches have more or less drawbacks:(1)focusing on certain sub-system and neglecting global optimization,(2)having a great burden on computation or data collection,(3)the risk of exposing user's privacy information.Energy is such a scarce resource that power management is an essential part of mobile devices.To overcome the defects and shortcomings on existing power management solutions,a semi-Markov decision process based dynamic power management solution is presented that it balances user experience and power consumption.This approach has less state number and needs less computation time in comparison with the proposed solution Boe.We monitors the usage status of Huawei G610-T00 by using Power monitor and records the power consumption of different components.A control software is designed for the change of GPS sampling rate and LCD brightness.An Android app based on Power Tutor is designed for data collection of phones' users.Besides,two kinds of online algorithms,Q learning and policy gradient,are presented and the simulation results are discussed in detail.Various kinds of criteria are also developed to evaluate the performance of the policies and the effectiveness of the energy-saving algorithms.In this dissertation,the feasibility of the proposed model and the effectiveness of the energy-saving algorithms are veryfied by simulation.The simulation results indicate that our solution can prolong usage time for about 53% and increase total experience by about 51% in comparison with the general fixed policy.The policy gradient estimation algorithm can update policy without the use of cloud technology.It does not rely on model parameters and needs small computational time.The users can freely set the update time of the policy and the algorithm can be easily implemented.The proposed power management solution is instructive to the research of the energy-saving for smartphones.