Decentralized Federated Learning for Secure Space-Terrestrial Communication With Intelligent Reflecting Surface

Decentralized federated learning (DFL) is viewed as a promising distributed machine learning technique for developing models for wireless networks in order to protect user privacy and reduce communication traffic. In this letter, we propose a DFL framework aided by intelligent reflecting surface (IRS) for aerial-terrestrial integrated networks without a central server. The parameters of the trained model are then transmitted to the satellite. IRS is utilized to reconfigure the wireless propagation environment in order to maximize resource utilization. In this system, we devise a secrecy rate maximization problem with a time-consuming analysis and propose an alternative optimization for configuring the parameters cooperatively. The simulation results show that the proposed algorithm is effective in the IRS-assisted DFL system and can outperform the SDP algorithm by up to 20%.