Perpetual Sensing: Experiences with Energy-Harvesting Sensor Systems

Abstract

Industry forecasts project the number of connected devices will outpace the global population by orders of magnitude in the next decade or two. These projections are application driven: smart cities, implantable health monitors, responsive buildings, autonomous robots, driverless cars, and instrumented infrastructure are all expected to be drivers for the growth of networked devices. Achieving this immense scale---potentially trillions of smart and connected sensors and computers, popularly called the "Internet of Things"---raises a host of challenges including operating system design, networking protocols, and orchestration methodologies. However, another critical issue may be the most fundamental: If embedded computers outnumber people by a factor of a thousand, how are we going to keep all of these devices powered?

In this dissertation, we show that energy-harvesting operation, by which devices scavenge energy from their surroundings to power themselves after they are deployed, is a viable answer to this question. In particular, we examine a range of energy-harvesting sensor node designs for a specific application: smart buildings. In this application setting, the devices must be small and sleek to be unobtrusively and widely deployed, yet shrinking the devices also reduces their energy budgets as energy storage often dominates their volume. Additionally, energy-harvesting introduces new challenges for these devices due to the intermittent access to power that stems from relying on unpredictable ambient energy sources. To address these challenges, we present several techniques for realizing effective sensors despite the size and energy constraints. First is Monjolo, an energy metering system that exploits rather than attempts to mask the variability in energy-harvesting by using the energy harvester itself as the sensor. Building on Monjolo, we show how simple time synchronization and an application specific sensor can enable accurate, building-scale submetering while remaining energy-harvesting. We also show how energy-harvesting can be the foundation for highly deployable power metering, as well as indoor monitoring and event detection. With these sensors as a guide, we present an architecture for energy-harvesting systems that provides layered abstractions and enables modular component reuse. We also couple these sensors with a generic and reusable gateway platform and an application-layer cloud service to form an easy-to-deploy building sensing toolkit, and demonstrate its effectiveness by performing and analyzing several modest-scale deployments.