Over-the-air federated learning with enhanced privacy protection

The researchers propose a new federated learning framework that combines efficient over-the-air parameter aggregation with strong privacy protection. The framework introduces pairwise cancellable random artificial noise (PCR-AN) at end devices, protecting both user data and the model. Compared with existing over-the-air computation (AirComp)-based federated learning schemes, this design provides stronger privacy protection; the paper's analysis gives the secrecy capacity and convergence rate of the proposed wireless aggregation algorithm.