@article{Gao2026, 
author = {Wei Gao and Xinji Guo and Yan Wang and Jingdie Xin and Jinbing Jiang and Feng Shu},
title = {Joint power allocation and residual U-Net channel estimation in active IRS-assisted near-field multi-user uplink systems},
year = {2026},
journal = {Intelligent and Converged Networks},
volume = {7},
number = {2},
pages = {166-182},
keywords = {channel state information (CSI) sensing, active intelligent reflecting surface (IRS), near-field (NF), power allocation, multi-user},
url = {https://www.sciopen.com/article/10.23919/ICN.2026.0005},
doi = {10.23919/ICN.2026.0005},
abstract = {In this paper, the channel estimation (CE) problem in multi-user hybrid-field communication systems enhanced by an active intelligent reflecting surface (IRS) is investigated. To effectively characterize the spatially non-stationary characteristics of large-scale active IRS under hybrid-field propagation, the uniform linear array of the active IRS is partitioned into multiple sub-blocks. Each user is assumed to be in the far-field region of each sub-block while remaining in the near-field region of the entire IRS, which enables dimensionality reduction in channel modeling and alleviates computational complexity. Moreover, under the total power constraint between the users and the active IRS, a closed-form solution for the optimal power allocation factor is rigorously derived. Furthermore, a novel multi-user pilot transmission strategy is designed, based on which efficient least squares (LS) estimators are developed for both the direct and cascaded links. To further enhance estimation performance, a deep learning based CE framework is proposed by integrating residual blocks into a U-Net architecture. The proposed network leverages the encoder–decoder pathway and skip connections for hierarchical feature extraction, while the residual modules strengthen representational capability and facilitate gradient propagation. Simulation results demonstrate that the proposed deep learning based estimator significantly outperforms conventional LS and minimum mean square error (MMSE) methods in terms of estimation accuracy, while reducing pilot overhead by seven-eighths.}
}