Resilient Aggregation: Statistical Approaches

In typical sensor network applications, the sensors are left unattended for a long period of time. In addition, due to cost reasons, sensor nodes are usually not tamper resistant. Consequently, sensors can be easily captured and compromised by an adversary. Once compromised, a sensor can send authentique messages to other nodes and to the base station, but those messages may contain arbitrary data created by the adversray (e.g., bogus measurments). A similar effect can be achieved by manipulating the physical environment of uncompromised sensors so that they measure false values. Bogus data introduced by the adversary may considerably distort the output of the aggregation function at the base station, and may lead to wrong decisions. The goal of resilient aggregation is to perform the aggregation correctly despite the possibility of the above mentioned attacks. In this paper, we give an overview of the state-of-the-art in resilient aggregation in sensor networks, and briefly summarize the relevant techniques in the field of mathematical statistics. In addition, we introduce a particular approach for resilient aggregation in more details. This approach is based on RANSAC (RAndom SAmple Consensus), which we adopted for our purposes. We also present some initial simulation results showing that our RANSAC based approach can tolerate a high percentage of compromised nodes.