Signs of a Bad Neighborhood: A Lightweight Metric for Anomaly Detection in Mobile Ad Hoc Networks

Anomaly detection in wireless multihop networks is notoriously difficult: the wireless channel causes random errors in transmission and node mobility leads to constantly changing node neighborhoods. The Neighbor Variation Rate (NVR) introduced in this paper is a metric that quantitatively describes how the topology of the neighborhood of a node in a wireless multihop network evolves over time. We analyze the expressiveness of this metric under different speeds of nodes and measuring intervals and we employ it to detect anomalies in the network caused by malicious node activity. We validate our detection model and investigate its parameterization by means of simulation. We build a proof-of-concept and deploy it in a real-world IEEE 802.11s wireless mesh network composed of several static nodes and some mobile nodes. In real-world experiments, we mount attacks against the mesh network and analyze the expressiveness of NVR to characterize these attacks. In addition, we analyze the behavior of NVR when applied to an external dataset obtained from measurements of a real-world dynamic AODV-based mobile ad hoc network. Our results show that our metric is lightweight yet effective for anomaly detection in both stationary and mobile wireless multihop networks.