Federated learning secrecy rate optimization for 6G intelligent reflecting surface assistance

Non-orthogonal multiple access (NOMA) based federated learning (FL) can achieve the massive connectivity of internet of thing (IoT) devices, high rate transmission, and pervasive intelligence in 6G networks. However, the stochastic channels in NOMA can lead to fluctuations in the transmission delay of model parameters. Frequent model updates may cause competition for communication and computational resources, and the uploaded model can be intercepted, leading to privacy leakage. To address these issues, an intelligent reflecting surface (IRS)-assisted NOMA-based FL communication system was proposed. In this system, IRS was leveraged to reconfigure the wireless communication environment, enhancing the signal strength at the legitimate receivers and suppressing the received signal at eavesdroppers (Eves), thereby providing a high-reliability and high-security communication foundation for federated learning. Firstly, the IRS-assisted wireless communication model was constructed. Subsequently, the minimum user-BS secrecy rate was maximized through the joint optimization of IoT device transmit power and IRS phase shifts. Finally, the deep deterministic policy gradient (DDPG) algorithm was employed to solve the optimization problem, with the action space defined as the transmit power and IRS phase shift matrix, and the channel state information and previous action constituting the state space. Simulation results demonstrated that deploying IRS in the system can greatly enhance the secrecy rate of the FL model update compared to artificial noise (AN), the block coordinate ascent method (BCAM), the scheme without IRS, and the scheme with random phase IRS, providing physical layer support for the subsequent deployment of FL algorithms in this communication system.