An investigation into passive information gathering for uncooperative Ad-Hoc Mesh Network assessment

Abstract

An Ad-Hoc Mesh Network is a wireless network formed spontaneously and dynamically by a group of wireless devices without the need for infrastructure or centralised administration. Its life cycle is divided into four phases: formation, operation, maintenance, and disbandment. This thesis aims to illustrate the information available to an adversary of an Ad-Hoc Mesh Network during phases 1 and 2 of the life cycle.
The first phase focuses on the formation stages of the life cycle and investigates the nodes’ initial spatial distribution. The goal is to establish a foundation for a methodology that predicts the locations of missing nodes. It is assumed that the adversary is aware of the Network Topology Generator (TG) being used, enabling them to estimate the placement of unobserved nodes based on observed node positions. The outcome for this phase of this thesis demonstrates 78% accuracy in correctly classifying the TG that produces select topology using a Gaussian Naive Bayes probabilistic clustering.
The second phase identifies the routing algorithm responsible for distributing network traffic throughout an Ad-Hoc Mesh network. From an adversarial perspective, this knowledge will enable them to follow data paths from a target node to network collection locations. The goal is to examine variations in network traffic that an outside party can identify using data available to an adversary or encrypted inter-node interactions. The research shows that of the various machine learning techniques, Support Vector Machine, Decision Tree and Random Forest gave the most accurate results of 99%. These accuracies were maintained as the sample size was reduced to 5 consecutive packets from 5 randomly chosen nodes.
In the final phase, a methodology is created to track influential nodes and investigate an extension of the operation stage of the life cycle. By predicting influential nodes, an adversary can target those nodes that will carry the most data, potentially causing data interception or network disruption. This research identifies that a Temporal Graph Networks link prediction methodology gave a 90% accuracy when determining which nodes will maintain or gain influence in the near future (roughly 9 mins of its operation). It also demonstrates that the mobility model used to generate the data does not statistically affect the outcome.

More information

Identifiers

ORCID for Michael O'Sullivan: orcid.org/0000-0002-4216-4287

ORCID for Leonardo Aniello: orcid.org/0000-0003-2886-8445

ORCID for vladi Sassone: orcid.org/0000-0002-6432-1482

Catalogue record

Export record