Adaptive Semantic Generation and NOMA-Based Interference-Aware Transmission for 6G Networks

Existing deep learning-based semantic communication (DeepSC) systems are typically trained for specific single-channel condition, which restricts the overall adaptability and resilience to interference. To address this limitation, we propose an innovative semantic adaptive feature extraction (SAFE) network that dynamically generates and fuses multiple sub-semantics, each characterized by unique features that can be tailored to different channel conditions. This paper also introduces three advanced learning algorithms to refine and enhance the generated sub-semantics, optimizing the semantic successive refinement performance of the SAFE network. Furthermore, we integrate a novel interference-aware semantic transmission method based on non-orthogonal multiple access (NOMA) into this framework. This approach enables users to adaptively select appropriate subsets for efficient transmission and image reconstruction, tailored to the prevailing channel interference conditions. Through extensive simulation experiments, we demonstrate the framework's capability to generate and transmit semantics under diverse channel interference scenarios adaptively, and verify the effectiveness through both objective and subjective quality evaluations.