Identification and mitigation of attacks in trust-based distributed communication networks

Thesis

Network security is a vital component of modern wireless communications, particularly as a result of the value of data being shared within those networks. Trust inference has provided an intuitive and accessible way for distributed networks to protect themselves in lieu of a central authority, such as a base-station. However, the trust metrics themselves can be manipulated by malicious parties in order to disrupt the network.

This thesis examines the vulnerabilities that exist within distributed communication networks that rely on trust inference for security. Chapter 3 studies the landscape of trust network vulnerabilities and presents a design for a unified framework of trust network attacks from an observer's perspective, based on the comparison and collation of pre-attack symptoms. This is then used as the basis for a novel contextual characterisation method for the classification of nodes' behaviour into possible attack scenarios, which is implemented using a support vector machine.

Chapter 4 presents the design and analysis of a trust-based data management and aggregation protocol for facilitating secure self-management of distributed network nodes. This is accompanied by the proposal for a trust-based data aggregation protocol to support network functionality, which enables coexistence with malicious nodes in the network by using their input to reinforce network decisions.

In Chapter 5, we devise a comprehensive model for node behaviour using a hidden Markov model approach, and show that the Baum Welch algorithm can be used to estimate the transition and emission probabilities of a node. Subsequently, it is shown that the use of a long short-term memory RNN facilitates the early classification and extrapolation of the node's behaviour, such that its attack probability can be estimated to a high degree of accuracy even before sufficient observations are collected.

Authors: Deshprabhu, T

• DOI: 10.5287/ora-9oevdr4gb
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: trust inference; data aggregation; distributed networks; distributed ledger; security protocol; uncertainty propagation; behavioral model; sensor fusion; insider attacks; communication; hidden Markov; recurrent neural network; long short term memory; IoT; distributed systems; communication engineering; attack detection; distributed learning; reputation; self organising networks; federated learning; network security
• Subjects: Engineering; Telecommunications engineers; Behavioral cybersecurity; Hidden Markov models; Trust; Artificial intelligence