Robust Design In MA-RIS-Aided AirFL System

Federated Learning (FL) has gained significant attention in distributed machine learning for its ability to enhance data privacy and communication efficiency by enabling collaborative training without centralizing raw data. However, the performance of FL in wireless networks, particularly in Over-the-Air Federated Learning (AirFL) systems, is often limited by spectral inefficiency and poor channel conditions. To address these challenges, this paper proposes a novel framework that integrates Reconfigurable Intelligent Surfaces (RIS) and Movable Antennas (MA) into AirFL systems under imperfect channel state information (CSI). By leveraging RIS to reconfigure the propagation environment and MA to dynamically adjust antenna positions, the proposed approach enhances model aggregation efficiency and mitigates channel degradation. We formulate a mean squared error (MSE) minimization problem to jointly optimize the base station’s receive beamformer, RIS phase shifts, users MA positions, and transmit power. To solve this non-convex problem, a new alternating optimization (AO)-based algorithm is developed. Extensive simulations validate the effectiveness, robustness, and viability of the proposed framework, demonstrating significant performance improvements in AirFL systems even under substantial channel estimation errors.