Federated Channel Learning for Intelligent Reflecting Surfaces With Fewer Pilot Signals

TL;DR

This paper introduces a federated learning framework for IRS-assisted wireless systems that significantly reduces pilot signals and transmission overhead while maintaining high channel estimation accuracy.

Contribution

It proposes a novel federated learning approach for channel estimation in IRS systems, reducing pilot signals and overhead compared to centralized methods.

Findings

Requires 60% fewer pilot signals

12 times lower transmission overhead than centralized learning

Achieves lower estimation error than existing DL schemes

Abstract

Channel estimation is a critical task in intelligent reflecting surface (IRS)-assisted wireless systems due to the uncertainties imposed by environment dynamics and rapid changes in the IRS configuration. To deal with these uncertainties, deep learning (DL) approaches have been proposed. Previous works consider centralized learning (CL) approach for model training, which entails the collection of the whole training dataset from the users at the base station (BS), hence introducing huge transmission overhead for data collection. To address this challenge, this paper proposes a federated learning (FL) framework to jointly estimate both direct and cascaded channels in IRS-assisted wireless systems. We design a single convolutional neural network trained on the local datasets of the users without sending them to the BS. We show that the proposed FL-based channel estimation approach requires approximately 60% fewer pilot signals and it exhibits 12 times lower transmission overhead than CL, while maintaining satisfactory performance close to CL. In addition, it provides lower estimation error than the state-of-the-art DL-based schemes.