Wireless sensor networks (WSNs) are networks of self-organizing, autonomously operating, resource-constrained computing devices. Typical designs of such so called mote-class devices integrate a low-power micro-controller (MCU), a radio chip, external memory and a radio frequency (RF) front-end on an area of a few square centimeters in size. Multi-year operation from small batteries is achieved by utilizing energy-saving low-power states and switching off currently not used components, e.g., the radio chip, whenever possible.
The most prominent use of such networks are so called sense-and-send applications. In a sense-and-send application, nodes are regularly sampling data that is then transmitted to a sink node. To date, the vast majority of published sensor network deployments employs a multi-hop routing protocol for communication. Such protocols create and maintain a dynamic routing tree, packets are forwarded over multiple hops until they are eventually received by a sink.
Wireless communication is not perfect, but subject to well-known phenomena such as multi-path propagation, interference, and loss. Additionally, the resource-scarcity found in systems that are built from mote-class devices limits the amount of capabilities that can be added to a system. For example, in contrast to Internet networks, implementing services such as active network monitoring and network time synchronization can add a considerable, even harmful overhead to a low-power wireless system.
Overall, known properties of wireless communication, common properties of distributed systems, and resource constraints specific to low-power wireless networks altogether negatively impact the quality of data obtained from such a system. Similarly, mentioned characteristics also render debugging low-power wireless networks as a very demanding task. System state first of all being distributed among the network, deployed systems also lack the resources needed for making that state accessible.
In the context of a scientific, multi-year, multi-site permafrost and rock kinematics monitoring effort, this thesis presents algorithms and systems for establishing wireless data collection systems as dependable and precise scientific instruments. Theoretical results obtained are backed by strong empirical evidence, i.e., evaluated in simulation, in testbed experiments on real hardware, and on up to 270 million packets that originate from deployed wireless sensor networks.
Matthias Keller obtained his Master's degree in Computer Science from Technische Universität München in 2008. Additionally, he also graduated from the Center for Digital Technology and Management with an Honours degree in Technology Management in 2008. Matthias finished his PhD studies at the Computer Engineering and Networks Laboratory of ETH Zurich in September 2013. Recent information on the author can be found on his homepage at https://m-keller.com
