Identification of prominent features of sensed processes in wireless sensor networks: a spatial interpolation based approach

Wireless sensor networks have received much attention due to their wide range of civilian and military application areas and technical challenges raised by their use. In this paper, we focus on environmental parameter monitoring applications of densely and randomly deployed sensors that are constrained in terms of power and communication bandwidth consumption, and storage costs. Our work involves processing sensed values from a subset of sensors at a central station (or "sink") to interpolate a sensed process over a field, and to a certain extent, approximately but efficiently locate and identify prominent feature areas such as flooded or polluted sites. For such applications, a sensor management algorithm is considered where sensors with the minimum Voronoi cell area are iteratively turned off to reduce resource consumption. The management algorithm is analyzed in terms of mean-square interpolation error of the sensed process, and compared to one that randomly turns off sensing devices. Extensive simulations demonstrate that the considered sensor management algorithm achieves less interpolation errors compared to the one based on random sensor turn-offs even when up to 90% of the sensing devices per unit area are turned off. A network connectivity metric, average sensor neighbor distance, is also proposed to assess the required radio range when designing sensing devices. Simulation results exhibit that, in the considered sensor management algorithm, the average sensor neighbor distance does not exceed 20% of the width of the sensed field even when up to 60% of the sensors per unit area are turned off. Therefore, significant resource savings can potentially be achieved without compromising too much transmission power for prominent feature identification applications in sensor networks