Title:

Decentralised control of wireless sensor networks

Wireless sensor networks are receiving a considerable degree of research interest due to their deployment in an increasing number and variety of applications. However, the efficient management of the limited energy resources of such networks in a way that maximises the information value of the data collected is a significant research challenge. To date, most of these systems have adopted a centralised control mechanism, but from a system's perspective this raises concerns associated with scalability, robustness, and the ability to cope with dynamism. Given this, decentralised approaches are appealing. But, the design of efficient decentralised regimes is challenging as it introduces an additional control issue related to the dynamic interactions between the network's interconnected nodes in the absence of a central coordinator. Within this context, this thesis first concentrates on decentralised approaches to adaptive sampling as a means of focusing a node's energy consumption on obtaining the most important data. Specifically, we develop a principled information metric based upon Fisher information and Gaussian process regression that allows the information content of a node's observations to be expressed. We then use this metric to derive three novel decentralised control algorithms for informationbased adaptive sampling which represent a tradeoff in computational cost and optimality. These algorithms are evaluated in the context of a deployed sensor network in the domain of flood monitoring. The most computationally efficient of the three is shown to increase the value of information gathered by approximately 83%, 27%, and 8% per day compared to benchmarks that sample in a naive nonadaptive manner, in a uniform nonadaptive manner, and using a stateoftheart adaptive sampling heuristic (USAC) correspondingly. Moreover, our algorithm collects information whose total value is approximately 75% of the optimal solution (which requires an exponential, and thus impractical, amount of time to compute). The second major line of work then focuses on the adaptive sampling, transmitting, forwarding, and routing actions of each node in order to maximise the information value of the data collected in resourceconstrained networks. This adds additional complexity because these actions are interrelated, since each node's energy consumption must be optimally allocated between sampling and transmitting its own data, receiving and forwarding the data of other nodes, and routing any data. Thus, in this setting we develop two optimal decentralised algorithms to solve this distributed constraint optimization problem. The first assumes that the route by which data is forwarded to the base station is fixed (either because the underlying communication network is a tree, or because an arbitrary choice of route has been made) and then calculates the optimal integration of actions that each node should perform. The second deals with flexible routing, and makes optimal decisions regarding both the sampling, transmitting, and forwarding actions that each node should perform, and also the route by which this data should be forwarded to the base station. The two algorithms represent a tradeoff in optimality, communication cost, and processing time. In an empirical evaluation on sensor networks (whose underlying communication networks exhibit loops), we show that the algorithm with flexible routing delivers approximately twice the quantity of information to the base station compared to the algorithm with fixed routing. However, this gain comes at a considerable communication and computational cost (increasing both by a factor of 100 times). Thus, while the algorithm with flexible routing is suitable for networks with a small numbers of nodes, it scales poorly, and as the size of the network increases, the algorithm with fixed routing should be favoured.
