Sensor networks have received a lot of attention due to many current and envisaged applications. Typically, sensor nodes are inexpensive and require no infrastructure for long-term deployments. They can collaboratively gather information through wireless communication. However, sensor nodes are resource constrained, and their unattended nature increases the probability of capture, modification and cloning. Therefore, defending sensor networks from potential attackers is a formidable challenge. Our research addresses various aspects of security in wireless and sensor networks. I briefly summarize three research directions below.
Revisiting Random Key Pre-distribution In this work, we found that efficiency of existing key management mechanisms could be significantly improved by using the giant component theory in Erdos-Renyi’s random graph theory. Our current research interest is to extend this result
to design group key management schemes, so that the information can be securely shared across the entire network to assist secure collaboration of sensor nodes.
Secure Localization The second line of research deals with secure localization problem. In an adversarial environment, various kinds of security attacks become possible if malicious nodes could claim fake locations. We designed a secure localization mechanism that detects the existence of these nodes, termed as phantom nodes, without relying on any trusted entities. This distributed and localized construction results in quite strong results: even when the number of phantom nodes is greater than that of honest nodes, the proposed mechanism can filter out most phantom nodes. Analytical results as well as simulations under realistic noisy settings demonstrate that the proposed scheme is effective in the presence of a large number of phantom nodes.
Physical Security One of most important goals of a sensor network is to infer events correctly, even in the presence of adversary. The primary research challenge is to differentiate irregularities from the adversarial events. This is not a simple task, since 1) physical sensing coverage is irregular, 2) environmental irregularities skew this coverage more significantly, and 3) attackers may generate events that confuse the event inference engine. As a preliminary step, we designed and implemented two Sensing Area Modeling (SAM) techniques. P-SAM provides accurate sensing area models for individual nodes using controlled or monitored events, while V-SAM[13] provides continuous sensing similarity models using natural events in the environments. To my knowledge, this research is the first investigation about the impact of sensing irregularity on application performance, such as coverage scheduling and tracking. Extensive evaluation in real-world settings reveals several serious issues concerning circular models, and significant improvements of several applications when SAM is used.