Opportunistic Networks

Opportunistic networks consist of low-power mobile devices that communicate over short-range wireless links. The aim of this research is to develop explicit physical models of information propagation in this type of network, and use these models to inform the design of information exchange protocols. The main focus is on the flow of information within the physical environment, rather than on the devices themselves. We have recently derived an analytic expression for the propagation rate in terms of device density, velocity distribution, transmission range and message length, extending previous research by incorporating non-instantaneous transmission. Information propagation in opportunistic networks has clear analogies with the spread of infectious diseases, and we have used spatial epidemic models to represent the macroscopic evolution of information density. By analogy with the well-known epidemic threshold, our expression for the transmission rate can be used to determine the minimum time that a device should attempt to transmit a message to ensure that the message persists in the system with high probability. The critical value is closely related to the percolation threshold for certain types of dynamic graphs. We are developing a discrete propagation model, based on an existing models of the behaviour of bacterial colonies, which incorporates non-instantaneous transmission. Cellular automata simulations have been performed to verify the model, and to investigate how various propagation characteristics scale with parameters such as device density and message length. In the continuous limit, the discrete model reduces to a reaction-advection-diffusion system; non-instantaneous transmission introduces delay terms, and non-local reaction terms arise due to wireless `action-at-a-distance'. A linear stability analysis of the this model is currently in progress.

Nonparametric Entropy Estimation

I am currently studying a class of computationally efficient methods for estimating R\'{e}yni entropy and divergence, based on nearest neighbour statistics. Applications of this research include methods for estimating the intrinsic dimension of a data set, and techniques for detecting discontinuities in the density of point clouds. The main theoretical challenge of this research is to prove a Central Limit Theorem for this class of estimators in cases where the density has unbounded support.

Adaptive Noise Estimation for Active Sonar

Using matched filter envelope data supplied by Thales Underwater Systems Ltd., we have demonstrated that the K-distribution provides an accurate model for noise in clutter-limited environments. Motivated by the problem of target detection in sonar data we have also shown, using the Anderson-Darling goodness-of-fit test, that this remains the case in the upper tail of the distribution. We are currently developing a GLRT (based on the K-distribution) for target detection under CFAR constraints. This research is being pursued as part of the Smith Institute Knowledge Transfer Network (KTN) programme.

Research Grants

2008 - 2011 Royal Society URF Information Propagation in Opportunistic Networks £253,714

2007 - 2010 EPSRC CASE Award Environmentally Adaptive Noise Estimation for Active Sonar £81,152

2003 - 2008 Royal Society URF Nonparametric Noise Estimation Using Nearest Neighbours £204,075