Energy-harvesting internet of things : from theory to practice

Abstract

Powering ubiquitous sensing and actuation devices is a fundamental challenge of fulfilling the vision of the Internet of Things (IoT). This class of computational devices used to operate on tethered power supply or battery power, but these options may not be practical in the realm of IoT: Deploying IoT devices on a massive scale, often in harsh and inaccessible environment, renders human intervention and even robotic maintenance next to impossible. Even though harvesting ambient energy has long been deemed a promising solution for keeping embedded systems powered, building an effectively self-sustainable system, with the objective of maximising lifespan while minimising the on-board energy storage size, is still a challenging and non-trivial engineering process. This research focuses on the system-level design, as well as hardware and software implementation of energy-harvesting IoT sensor nodes. This thesis reviews the state-of-the-art IoT technologies and analyses the usage patterns of sensor nodes, in order to establish an evaluation framework on energy consumption of IoT sensor nodes. The framework provides insights on hardware design, which informed the development of the Self-Powered IoT Node for Environmental Sensing, or SPINES Mote. As the primary contribution of this work, SPINES Mote is a novel, open-source IoT prototyping platform for energy-harvesting sensor nodes. This thesis explains the design philosophy of SPINES Mote, elaborates its implementation process, and compares its unique power management capabilities with commercial and academic counterparts. The compact SPINES Mote design has demonstrated an uninterrupted operation under varying weather conditions: A pair of solar-powered, Wi-Fi SPINES Mote prototypes, each equipped with a 13-mW monocrystalline photovoltaic cell for recharging a 110-mAh lithium-ion battery, were deployed on the rooftop of a building in Hong Kong. The pair have sustained a periodic sensing application at one-hour interval for over 50 days and counting. However, their telemetry data indicates that a purely hardware-based, opportunistic approach to harvest ambient energy may not be robust for long-term self-sustainability. Hence, this thesis also discusses software support for smarter energy management, in particular, the use of energy prediction algorithms and load adaptation techniques. A prediction-driven coulomb counting scheme is proposed for a sensor node to measure and predict its energy availability with lower energy overhead. This can be achieved by switching off the power-hungry coulomb counter at times with no anticipated energy intake. Simulation results show that the proposed scheme can reduce the average power consumption of a sensor node in sleep mode by half, and in turn prolong its lifespan by at least 8%. This improvement allows IoT sensor nodes to be built cheaper and more compact, through using smaller-sized energy storage elements. This work showcases the feasibility of constructing perpetually-operated IoT sensor nodes using commercial off-the-shelf components available today.