Digital Twin with Delay-Optimal Relay-Assisted Federated Learning

In this paper, we propose a secure federated learning (FL) system in the digital twin (DT) based on relay technique and cooperative interference, which realizes accurate real-time prediction of physical entities. First, an optimization problem aiming to minimize the total delay of each FL iteration is established, in which the local device performs model uploading by non-orthogonal multiple access (NOMA) technique. Then, in order to solve this non-convex problem, a layered optimization algorithm is proposed, the local processing rate of each device, the transmission power of the relay and the interference signal of the jammer are obtained. Finally, numerical results show that the algorithm is effective and the relay-based FL security scheme can be effectively applied to DT.