Resource Hopping Multiple Access based Federated Learning Scheme for 6G Massive IoT with Differential Privacy

With the widespread use of mobile devices and the rapid growth of big data, the application of federated learning in decentralized data environments has attracted increasing attention. 5G and 6G mobile communication systems take the large-scale Internet of things as one of the core scenarios, which can be employed to realize the real-time data transmissions of massive terminals in federated learning. Multiple access technology is the key to achieve massive connectivity in 6G massive IoT. Resource hopping multiple access achieves efficient access by assigning different resource hopping patterns to different users. In this paper, a federated learning scheme combining RHMA and differential privacy is proposed, which divides the communication channel into multiple subchannels and allocates hopping patterns according to data characteristics and computational resources. Therein, differential privacy mechanism effectively protects the privacy of user data by adding noise to the model training process of federated learning. Experimental results show that the proposed scheme not only improves the training speed of the federated learning model, but also improves the robustness and security of the system while ensuring user privacy.