A Scalable Semantic Communication System Based on Meta-Learning

Industrial Internet has been proposed to realize realtime and reliable communications of various production tasks in industrial networks. However, the limitation in communication resources and the emergence of new communication tasks increase difficulties in realizing Industrial Internet. Thus, in this paper, we first propose a specific joint source-channel coding (JSCC) communication model, aiming to harness the advantages of semantic communication in resource-limited communication scenarios. Then, to accommodate to the emergence of new tasks, this paper combines meta-learning and decoding information resolution (DIR) and presents a scalable design methodology for semantic communication systems with arbitrary network structures. When a new transmission task is added in production, this design methodology can make a quicker deployment of a semantic communication system tailored for the new scalable task, efficiently meeting the communication requirements of newly tasks. Finally, the proposed design methodology is deployed within the JSCC network structure. Simulation analysis demonstrates this communication system has good performance in communication and scalability, meeting the demands of massive device connectivity and high-reliability communication in industrial internet.