Analyzing the power consumption of smartphones is difficult because of the complex interplay between soft- and hardware. Currently, researchers rely on mainly two options: external measurement tools, which are precise but constrain the mobility of the device and require the annotation of power traces; or modelling methods, which allow mobility and consider explicitly the state of events, but have less accuracy and lower sampling rates than external tools.

We address the challenges of mobile power analysis with a novel power metering toolkit, called NEAT, which comprises a coin-sized power measurement board that fits inside a typical smartphone, and analysis software that automatically fuses the event logs taken from the phone with the obtained power trace. The combination of high-fidelity power measurements and detailed information about the state of the phone's hardware and software components allows for fine-grained analysis of complex and short-lived energy patterns.

We equipped smartphones with NEAT and conducted various experiments to highlight (i) its accuracy with respect to model-based approaches, showing errors upwards of 20%; (ii) its ability to gather accurate and well annotated user-data "in the wild", which would be hard to do with current external meters; and (iii) the importance of having fine-granular and expressive traces by resolving kernel energy bugs.