Delay of Intrusion Detection in Wireless Sensor Networks

In this paper we consider sensor networks for intrusion detection, such that node deployment, node failures and node behavior result in coverage gaps and a fraction of disconnected nodes in an otherwise dense and well-connected network. We focus on the time delay for a mobile intruder to be detected by a sensor with a connected path to the sink, in contrast to existing results for the detection time by a sensor with arbitrary connectivity. We model our network using a supercritical percolation model on the plane, implying the existence of a unique unbounded connected component, and we assume that the sink belongs to this component. We analyze the distribution of the distance traveled by a moving target until it comes within sensing range of a node in the giant component, providing analytical bounds for linear intruder mobility and thorough simulation results for other mobility models. We show that the probability that the intruder proceeds undetected exhibits non-memoryless behavior over shorter distances and an exponentially decreasing tail. We also show that the time of contact with the giant component incurs considerably more delay than the time of first contact with any node, in networks with less than 10 % of nodes without a path to the sink, which means that even a small percentage of node failures may have a drastic impact on the performance of intrusion detection by a wireless sensor network