Federated Generative Adversarial Networks Based Channel Estimation

Yiyu Guo,Zhijin Qin,Octavia A. Dobre
DOI: https://doi.org/10.1109/iccworkshops53468.2022.9814633
2022-01-01
Abstract:With the rapid growth of connectivity and demands on reliable communications, channel estimation becomes a critical problem. Recently, the advancement of deep learning enables real-time data-driven channel estimation. Particularly, this paper proposes a generative adversarial network (GAN) based method for channel estimation in a distributed approach. We leverage GAN to accurately estimate the orthogonal frequency division multiplexing channels with a shorter length of pilots. Considering the low complexity of the least square channel estimator, we use it as the input of GAN and train the neural network to learn the distribution of channels. To guarantee an efficient learning process, federated learning is adopted in this paper. Meanwhile, we develop dynamic regularizers to improve the federated learning performance. Simulation results demonstrate that the proposed federated GAN model provides better channel estimation performance with lower pilot overhead, compared to the conventional channel estimation methods.
What problem does this paper attempt to address?