Task-Oriented Multi-User Semantic Communications

While semantic communications have shown the potential in case of single-modal single-users, its applications to multi-user scenario remain limited. In this paper, we investigate deep learning (DL) based communication systems for transmitting data and multimodal data, respectively. We adopt three intelligent tasks, including, image retrieval, machine translation, visual question answering (VQA) as transmission goal systems. propose a Transformer framework unify structure transmitters different tasks. For system, two models, named, DeepSC-IR DeepSC-MT, perform retrieval case, is trained optimize distance embedding space between images DeepSC-MT minimize errors by recovering meaning sentences. develop enabled model, DeepSC-VQA, VQA task extracting text-image information at fusing it receiver. particular, novel layer-wise designed help fuse adding connection each encoder decoder layers. Numerical results show that proposed models are superior traditional terms robustness channels, computational complexity, delay, task-execution performance various task-specific metrics.