Federated Learning and Over-the-Air Computation: A Comparative Study

As Federated Learning (FL) emerges as a privacy-preserving alternative to centralized machine learning, ensuring communication efficiency becomes increasingly crucial. This paper examines two promising strategies: orthogonal FL utilizing sequential transmissions and Over-the-Air FL, where signals are combined in the air simultaneously. Their performance is evaluated across several key parameters to uncover trade-offs between accuracy and resource usage. A simulation framework was designed to assess convergence behavior under varying dataset complexities and data distributions. While orthogonal FL achieves higher accuracy and greater stability, Over-the-Air FL offers substantial communication savings, particularly in noise-resilient scenarios. These findings highlight important trade-offs between model performance and communication efficiency.