Federated Learning for Channel Estimation in Conventional and IRS-Assisted Massive MIMO

TLDR: A convolutional neural network (CNN) trained on the local datasets of the users without sending them to the BS is designed and the proposed CNN architecture exhibits lower estimation error than the state-of-the-art ML-based channel estimation schemes.

Abstract: Machine learning (ML) has attracted a great research interest for the problems in the physical layer of wireless communications, such as channel estimation, thanks to its low computational complexity and robustness against imperfect channel data. Channel estimation via ML requires model training on a dataset, which usually includes the received pilot signals as input and channel data as output. In previous works, model training is mostly done in a centralized manner, where the whole training dataset is collected from the users at the base station (BS). This approach introduces huge transmission overhead for data collection from the users. In this paper, to address this challenge, we propose a federated learning (FL) framework for channel estimation. We design a convolutional neural network (CNN) trained on the local datasets of the users without sending them to the BS. We develop FL-based channel estimation schemes for both conventional and IRS (intelligent reflecting surface) assisted massive MIMO (multiple-input multiple-output) systems, where a single CNN is trained for two different datasets for both scenarios. Even if the IRS-assisted massive MIMO includes two different channels, namely the direct and cascaded channels, their estimation is performed with a single CNN, without using multiple CNNs for each task. Via numerical simulations, we evaluate the performance for noisy and quantized model transmission and show that the proposed approach provides approximately 16 times lower transmission overhead than the centralized learning (CL) schemes, while maintaining satisfactory channel estimation performance close to CL. Furthermore, the proposed CNN architecture exhibits lower estimation error than the state-of-the-art ML-based channel estimation schemes.