Robust and secure semantic communication methods over integrated terrestrial and non-terrestrial networks

The integration of Terrestrial Networks (TNs) and Non-Terrestrial Networks (NTNs) is expected to play a central role in future Sixth Generation (6G) systems by providing seamless connectivity across ground, air, and space segments. This evolution supports applications such as remote sensing and disaster management, which require the reliable transmission of large volumes of data under stringent bandwidth, latency, and energy constraints. However, this trend exposes communication systems to adverse channel conditions, limited resources, and security risks due to the broadcast nature of wireless transmission. To overcome these challenges, it is necessary to develop communication and security mechanisms that are not only robust and resource-efficient, but also suitable for Artificial Intelligence (AI)-native 6G architectures. This thesis addresses the challenges through three complementary contributions that span transmission efficiency, privacy, and adaptive security. First, it proposes a lightweight Semantic Communication (SemCom) framework based on deep Joint Source-Channel Coding (JSCC) for robust video transmission over noisy wireless channels. By integrating semantic attention and Signal-to-Noise Ratio (SNR)-aware modulation, the proposed method improves the transmission of taskrelevant information while maintaining low computational cost. The proposed architecture makes the transmission suitable for resource-constrained scenarios relevant to TN–NTN systems. Second, the thesis investigates security vulnerabilities in SemCom, with a particular focus on eavesdropping and Model Inversion Eavesdropping Attacks (MIEAs). To mitigate these threats, it introduces a lightweight and reversible protection mechanism, Key-Assisted Protection for Deep Joint SourceChannel Coding (KAP-DJSCC), which secures the transmitted stream through keybased transformations. Third, the thesis develops a risk-based adaptive security framework for AI-native ’6G systems which provides a dynamic selection of countermeasures under resource and budget constraints. Overall, the contributions of this thesis address three important aspects of 6G systems which are applicable for TNNTN: communication efficiency, semantic security, and adaptive cyber-risk management.