Federated Learning for Wireless Communications
Ahmet M. Elbir, Wei Shi · 2024
In the past few years, machine learning (ML) techniques have been introduced for the physical layer applications in wireless communications. In contrast to employing centralized learning (CL) techniques, federated learning (FL) presents lower communication overhead as it does not involve dataset transmission between the edge users and the server. As a result, FL is particularly useful for applications, wherein the dataset is huge. Such examples include physical layer design applications, which may require huge datasets to represent the environment accurately. This chapter is concerned with FL-based physical layer applications, e.g., channel estimation and hybrid beamforming. The channel estimation problem is investigated for both conventional and reconfigurable intelligent surface-aided millimeter wave (mmWave) and terahertz (THz) scenarios. We begin by introducing the channel models for both mmWave and THz. Then, we discuss the implementation of FL for various channel estimation problems. We also discuss near-field channel estimation, which may occur in the THz scenario, for which the operating wavelength is very small. Then, we present FL-based hybrid beamforming in mmWave wireless communications. The performance evaluation of FL is provided via several numerical simulation results to show its effectiveness.