Semantic Base Enabled Image Transmission With Fine-Grained HARQ

Semantic communications (SemComs) which utilize the inherent meanings and relationships of data, have shown significant advantages for information transmission in recent years. In this paper, a novel semantic base (Seb) enabled SemCom framework is proposed, where Sebs, the basic units of fine-grained image semantics, are explicitly shared among transceivers to support image transmission. Specifically, first, to improve source coding efficiency, a Seb-based image codec is proposed, where semantics in each image patch are encoded with synchronized Sebs, that are generated from recent images. To ensure semantic consistency among the embeddings of Sebs, Gray coding is used to establish the projection, enhancing the framework’s robustness against channel noise. The details of images are further refined by a residual codec for high-quality reconstructions. Second, to enhance transmission reliability with overhead as small as possible, a semantic-aware fine-grained hybrid automatic repeat request (SAFG-HARQ) is proposed, where only erroneous Sebs are retransmitted to precisely refine corrupted semantics using contextual correlations. Extensive simulations demonstrate that the proposed framework outperforms state-of-the-art works, where the image reconstruction quality is improved by 20% in learned perceptual image patch similarity (LPIPS), with a 60% reduction in transmission costs.