Secure Aggregation in Sensor Networks Using Neighborhood Watch

Many sensor networks (SN) use in-network aggregation to minimize the amount of data transmitted by sensors. Unfortunately, aggregation makes the network more vulnerable: a central server collecting the SN´s results receives less information that could be used to detect and/or mitigate cheating. In this paper we study stealth attacks where an adversary seeks to deceive the system without being detected. We propose several measures to make a sensor network resilient to attack: (a) mutual monitoring and constraint validation, (b) randomized delivery trees, and (c) redundancy in the data supplied by the network. We introduce several protocols to achieve resilience as well as a reputation system to allow detection and exclusion of compromised sensors. We conclude by reporting results of our simulations under several attack models.