Zheng, Guhan and Ni, Qiang and Navaie, Keivan (2024) Machine Learning based Semantic Communication Systems for 6G Three-dimensional Communication Networks. PhD thesis, Lancaster University.
![[thumbnail of 2024GuhanZhengPhD]](https://eprints.lancs.ac.uk/style/images/fileicons/text.png)
Abstract
The sixth generation (6G) wireless communication is anticipated as a threedimensional (3D) network with full support of aerial edge and space edge. Moreover, semantic communication (SemCom) based on machine learning (ML) is also considered a significant enabling technology for 6G systems. Nevertheless, integrating SemCom into future 3D networks introduces emerging semantic coder updating requirements and new functional challenges considering, e.g. latency, energy, and privacy. Motivated by the above observations, in this thesis, the challenges of SemCom in various 6G edge-enable network architectures are investigated. Firstly, a terrestrial vehicular SemCom system is investigated for vehicle task offloading in vehicular networks (VNs). A novel mobility-aware split-federated with transfer learning (MSFTL) framework for SemCom coder updating is then proposed. Moreover, to incorporate vehicle mobility and training delays I propose a highmobility training resource optimisation mechanism based on a Stackelberg game for MSFTL. Secondly, an air-terrestrial SemCom system is proposed for energy-efficient implementation of SemCom in aerial-aided edge networks (AENs). An energyefficient game theoretic incentive mechanism (EGTIM) is proposed for improving the energy efficiency of the AEN for SemCom. To update SemCom coders accurately and efficiently in AENs, I further present a game theoretic efficient distributed learning (GEDL) framework based on the renewed EGTIM. Finally, a space-air-terrestrial (SAT) SemCom system is proposed for the computation offloading of resource-limited users in SAT networks. An adaptive pruning-split federated learning (PSFed) method for updating the SemCom coder is then proposed. Furthermore, the users processing computational tasks strategy in presented systems is formulated as an incomplete information mixed integer nonlinear programming (MINLP). A new computational task processing scheduling (CTPS) mechanism is also proposed based on the Rubinstein bargaining game.