Semantic Communication and Security over Cloud-Network-End Infrastructure: An Effective Architecture for Intelligent Mobile Systems

With the advancement of artificial intelligence (AI) technology, semantic communication is rapidly emerging as one of the next-generation communication technologies. In this article, we first present the architecture to integrate semantic communication with the cloud-network-end (C-N-E) infrastructure. By deploying AI models and domain knowledge on the end and the cloud appropriately, we can efficiently manage, train, and apply AI models to significantly facilitate the extraction, transmission, and inference of semantic information over error-prone physical channels. This architecture realizes the integration of communication and computation. To demonstrate the advantages of this architecture, we design a drone recognition system utilizing the You Only Look Once (YOLO)-based Joint Source-Channel Coding (JSCC). The encoder and decoder are deployed at the detective drone and the cloud, respectively. We transform the optimization of the JSCC into an end-to-end autoencoder task while incorporating the physical channels as an untrainable component. We have also introduced additional convolutional layers to perform data consistency checks and enhance semantic security. The experiment results show that the JSCC effectively mitigates the performance cliff effect commonly observed in traditional communication systems under low signal-to-noise ratio (SNR) conditions. The JSCCs trained in harsh channel conditions have strong robustness against channel fading and severe data impairment.