Low-Complexity Channel Estimation for Spatial Non-Stationary XL-MIMO Systems: A Model-Based Deep Learning Approach
IEEE Transactions on Communications · 2026 · 74 · 8900-8915
DOI: 10.1109/tcomm.2026.3693181 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Communications, 2026. Verfasst von Hao Lei und 6 weiteren Autor:innen. Die Arbeit nennt 45 Literaturangaben. Gefördert von: Fundamental Research Funds for the Central Universities, National Natural Science Foundation of China, ZTE Industry-University-Institute Cooperation Funds.
Abstract (Originalsprache)
In this paper, we investigate the channel estimation problem in near-field extremely large-scale multiple-input multiple-output (XL-MIMO) systems, explicitly accounting for both spherical-wave propagation characteristics and spatial non-stationary effects. Building on these properties, we propose a novel model-based deep learning framework that delivers high-accuracy channel estimation with low computational complexity by tightly integrating domain knowledge and data-driven learning. Specifically, the proposed framework comprises three key unfolding networks: a sparse channel recovery network, a codebook update network, and an error cancellation network. The first network, referred to as variational Bayesian inference (VBI)-Net, is derived by unfolding the inverse-free VBI (IF-VBI) algorithm. It enables high-precision sparse channel reconstruction without requiring explicit prior assumptions, by learning the underlying precision distribution directly from data. The second network, gradient (Grad)-Net, is developed by unfolding the gradient ascent procedure, where learnable step sizes are introduced to adaptively refine the parameters of the polar-domain grids. Moreover, Grad-Net captures spatial non-stationary characteristics associated with the polar-domain representation by jointly exploiting gradient information and estimated path parameters. The third network, termed projected gradient descent (PGD)-Net, is constructed by unfolding the PGD algorithm. It iteratively refines the channel estimates and effectively suppresses residual estimation errors induced by spherical-wave propagation and spatial non-stationarity. Extensive numerical simulations demonstrate that the proposed framework significantly outperforms existing methods in both estimation accuracy and computational efficiency. Furthermore, the proposed framework achieves a superior accuracy-complexity tradeoff for practical XL-MIMO systems, delivering enhanced performance while maintaining very low computational complexity.
Förderung
- Fundamental Research Funds for the Central Universities
- National Natural Science Foundation of China
- ZTE Industry-University-Institute Cooperation Funds
Zitieren
APA
Lei, H., Zhang, J., Liu, Z., Xiao, H., Ai, B., Ng, D. W. K., & Nallanathan, A. (2026). Low-Complexity Channel Estimation for Spatial Non-Stationary XL-MIMO Systems: A Model-Based Deep Learning Approach. IEEE Transactions on Communications, 74, 8900-8915. https://doi.org/10.1109/tcomm.2026.3693181
Vancouver
Lei H, Zhang J, Liu Z, Xiao H, Ai B, Ng DWK, et al. Low-Complexity Channel Estimation for Spatial Non-Stationary XL-MIMO Systems: A Model-Based Deep Learning Approach. IEEE Transactions on Communications. 2026;74:8900-8915. doi:10.1109/tcomm.2026.3693181
BibTeX
@article{Lei2026,
author = {Lei, Hao and Zhang, Jiayi and Liu, Zhilong and Xiao, Huahua and Ai, Bo and Ng, Derrick Wing Kwan and Nallanathan, Arumugam},
title = {Low-Complexity Channel Estimation for Spatial Non-Stationary XL-MIMO Systems: A Model-Based Deep Learning Approach},
journal = {IEEE Transactions on Communications},
year = {2026},
volume = {74},
pages = {8900--8915},
doi = {10.1109/tcomm.2026.3693181}
}
EUVIMED