Secure semantic communication system design and optimisation

Liu, Xuesong (2026) Secure semantic communication system design and optimisation. PhD thesis, University of Glasgow.

Abstract

Semantic communication (SemCom) has emerged as a promising communication paradigm that shifts the communication objective from accurate bit transmission to effective semantic delivery. By transmitting semantic information instead of raw data, SemCom significantly improves communication efficiency and enables intelligent communication services for future networks. With the rapid development of artificial intelligence and foundation models, knowledge has become a fundamental element for semantic understanding, semantic reasoning, and intelligent decision-making, driving the evolution of SemCom toward knowledge-centric SemCom. However, the increasing dependence on knowledge also introduces new challenges, such as trustworthy knowledge utilisation, and efficient knowledge management. Existing studies mainly focus on improving SemCom performance, while knowledge security and lifecycle management are not sufficiently explored. Therefore, the development of secure knowledge-enabled SemCom has become an important research topic for future intelligent communication systems.

This thesis investigates secure knowledge-enabled SemCom from four complementary perspectives, namely knowledge-assisted SemCom architecture, privacy preserving transceiver design, knowledge utilisation, and knowledge management. First, this thesis investigates knowledge-assisted privacy preservation in SemCom from a knowledge-centric perspective. Then, potential privacy threats introduced by knowledge utilisation are analyzed. Based on these observations, a knowledge-assisted privacy-preserving SemCom architecture is proposed, consisting of a semantic transmission layer and a knowledge management layer that jointly support secure semantic communications. The corresponding transceiver architecture is then presented to illustrate the integration of knowledge into semantic encoding and decoding. Second, we proposed a knowledge-based differential privacy method, which combines differential privacy with shared knowledge to improve the trade-off between privacy and utility. At the same time, we analyzed the privacy boundaries in SemCom to characterize the impact of differential privacy and knowledge mismatch on semantic reconstruction and sensitive information inference, providing a strict privacy proof for the proposed method. Furthermore, we proposed an adversarial optimisation strategy, which alternately minimizes and maximizes joint optimisation of semantic reconstruction and privacy protection, enabling the semantic encoder to learn a potential representation that is invariant to privacy. Third, this thesis focuses on jointly addressing three core issues of power allocation, knowledge caching, and device-to-device user pairing in secure SemCom. a novel performance metric is developed, namely semantic secrecy throughput, to quantify the information security level that can be achieved at each pair of SemCom users. Next, a semantic secrecy throughput maximisation problem is formulated subject to secure SemCom-related delay and reliability constraints. Afterward, a security-aware knowledge management solution is proposed using the Lagrange primal-dual method and a two-stage method.

The studies presented in this thesis establish a unified research framework for knowledge-assisted secure SemCom by integrating knowledge-assisted architecture, knowledge utilisation, and knowledge management into a coherent perspective. The proposed methods not only improve the security, reliability, and intelligence of SemCom but also demonstrate the importance of treating knowledge as a core element rather than an auxiliary component.

• Item Type: Thesis (PhD)
• Qualification Level: Doctoral
• Subjects: T Technology > TK Electrical engineering. Electronics Nuclear engineering
• Colleges/Schools: College of Science and Engineering > School of Engineering
• Supervisor's Name: Zhao, Professor Dezong and Sun, Mr. Yao
• Date of Award: 2026
• Depositing User: Theses Team
• Unique ID: glathesis:2026-86231
• Copyright: Copyright of this thesis is held by the author.
• Date Deposited: 10 Sep 2026 12:32
• Last Modified: 10 Sep 2026 14:00
• Thesis DOI: 10.5525/gla.thesis.86231
• URI: https://theses.gla.ac.uk/id/eprint/86231

Related URLs

• Article DOI
• Article DOI
• Article DOI
• Article DOI
• Article DOI
• Article DOI
• Article DOI
• Pre-print
• Conference proceeding
• Conference proceeding