ISCS: An Interactive Multimodal Semantic Communication Architecture

To meet the stringent requirements of high reliability and low latency in sixth-generation (6G) mobile communication networks, semantic communication (SC) has emerged as a paradigm with transformative potential. However, existing multimodal-oriented SC systems typically employ independent encoders for each modality, which suffer from representational imbalance caused by modality heterogeneity, leading to semantic divergence. Furthermore, feature fusion directly at the receiver introduces semantic redundancy, thereby increasing the decoding burden and compromising communication efficiency. In this paper, we propose an interactive joint source–channel coding (JSCC) architecture that facilitates a shift towards a multi-source collaborative communication paradigm. Specifically, we design an interactive encoder guided by global prompts and a cross-attention-based interactive decoder, enabling effective cross-modal alignment and redundancy reduction. Extensive experiments on multimodal sentiment analysis (MSA) and visual question answering (VQA) tasks demonstrate that, across various channel conditions, the proposed framework consistently outperforms state-of-the-art benchmarks, achieving superior performance.