Adaptive Federated Deep Learning-Based Semantic Communication in the Social Internet of Things

The introduction of semantic communication offers an effective solution for achieving efficient and reliable information transmission in the social Internet of Things (SIoT). SIoT combines social networks with the Internet of Things (IoT) to create a “social network of smart objects,” utilizing analytical and statistical models to provide efficient and scalable services. However, ensuring high-quality and reliable data transmission within the SIoT remains a significant challenge. Semantic communication methods can effectively address this issue. Semantic communication represents an advanced paradigm aimed at achieving reliable transmission through semantic-level data compression. In this article, we propose a semantic communication framework based on adaptive federated deep learning. This framework combines source-channel joint coding with channel bandwidth adaptation techniques to enhance transmission efficiency and promote natural and effective information exchange. Specifically, deep reinforcement learning is employed to manage dynamic bandwidth allocation, enabling the selection of optimal bandwidth under varying signal-to-noise ratios and data conditions, thereby improving transmission quality and bandwidth utilization. Additionally, we introduce a training method based on federated learning to enhance the model’s generalization ability under different channel conditions. Simulation results demonstrate that our proposed method outperforms traditional models, exhibiting excellent adaptability to low signal-to-noise ratios and low bandwidth environments, as well as higher stability. This positions our method as a valuable approach for ensuring reliable data communication in the SIoT.