Federated Learning for Physical Layer Design

Model-free techniques, such as machine learning (ML), have recently attracted much interest toward the physical layer design (e.g., symbol detection, channel estimation, and beamforming). Most of these ML techniques employ centralized (CLK) schemes assume availability datasets at a parameter server (PS), demanding transmission data from edge devices, mobile phones, to PS. Exploiting generated edge, federated (FL) has been proposed distributed scheme, in which each device computes model parameters sends them PS for aggregation, while are kept intact edge. Thus, FL is more communication-efficient privacy-preserving than CL applicable wireless communication scenarios, wherein devices. This article presents recent advances FL-based training problems. Compared CL, effectiveness presented terms overhead with slight performance loss accuracy. The challenges, model, data, hardware complexity, also discussed detail along possible solutions.