Prospector: Multiscale energy measurement of networked embedded systems with wideband power signals

Today's wirelessly networked embedded systems underlie a vast array of electronic devices, performing computation, communication, and input/output. A major design goal of these systems is energy efficiency. To achieve this goal, these systems are based on processors with numerous power and clock domains, variable clock rates, voltage scaling, and multiple hibernation states. These processors are designed into systems with sophisticated wireless transceivers and a diverse array of off-chip peripherals, and are linked through regulators to increasingly complex energy supplies. As a result, modern networked embedded systems are characterized by myriad power consumption states and significant power signal transients. Moreover, their power demands are multiscale in both magnitude and time, combining short bursts of high demand with long intervals of power-sipping sleep states. Thus the power supply signals have wideband spectra. In addition, due to noise, uniform relative precision across magnitude scales requires that measurement time increases with decreasing power. Tools are needed that support modeling, hardware/ software optimization, and debugging for energy-centric embedded systems. This paper describes Prospector, an energy data acquisition system architecture for embedded systems that allows rapid, accurate, and precise assessment of system-level power usage. Prospector uses a distributed control architecture; each component contributes efficiently to control, precision and accuracy, analysis, and visualization. It is based on computerbased control of multimeters to maximize accuracy, precision, flexibility, and minimize target system overhead. Experimental results for a prototype Prospector system with a contemporary 16-bit ultra-low power microcontroller show that it can effectively measure power over the extreme time and magnitude scales found in today's embedded systems.