Near-Field Positioning for XL-MIMO Uniform Circular Arrays: An Attention-Enhanced Deep Learning Approach

In the evolving landscape of sixth-generation (6G) mobile communication, multiple-input multiple-output (MIMO) systems are incorporating an unprecedented number of antenna elements, advancing towards Extremely large-scale multiple-input-multiple-output (XL-MIMO) systems. This enhancement significantly increases the spatial degrees of freedom, offering substantial benefits for wireless positioning. However, the expansion of the near-field range in XL-MIMO challenges the traditional far-field assumptions used in previous MIMO models. Among various configurations, uniform circular arrays (UCAs) demonstrate superior performance by maintaining constant angular resolution, unlike linear planar arrays. Addressing how to leverage the expanded aperture and harness the near-field effects in XL-MIMO systems remains an area requiring further investigation. In this paper, we introduce an attention-enhanced deep learning approach for precise positioning. We employ a dual-path channel attention mechanism and a spatial attention mechanism to effectively integrate channel-level and spatial-level features. Our comprehensive simulations show that this model surpasses existing benchmarks such as attention-based positioning networks (ABPN), near-field positioning networks (NFLnet), convolutional neural networks (CNN), and multilayer perceptrons (MLP). The proposed model achieves superior positioning accuracy by utilizing covariance metrics of the input signal. Also, simulation results reveal that covariance metric is advantageous for positioning over channel state information (CSI) in terms of positioning accuracy and model efficiency.

💡 Research Summary

The paper addresses the challenge of high‑precision three‑dimensional positioning in the near‑field region of extremely large‑scale MIMO (XL‑MIMO) systems that employ uniform circular arrays (UCAs). As antenna apertures grow to meter‑scale dimensions, the conventional far‑field plane‑wave assumption breaks down; signals arrive as spherical wavefronts, causing a strong coupling between angle and range (the so‑called angle‑range coupling). This coupling complicates traditional two‑step positioning methods (angle estimation followed by range estimation) and renders subspace‑based algorithms such as MUSIC and ESPRIT computationally prohibitive for large arrays.

To overcome these issues, the authors propose an attention‑enhanced deep learning framework specifically tailored for near‑field UCA positioning. The core of the architecture consists of two novel attention modules. First, a dual‑path channel‑level attention mechanism performs a “soft” selection of the most informative frequency‑phase channels. One path uses average pooling to preserve overall feature statistics, while the other employs max pooling to highlight the strongest responses; the combination balances expressive power with computational cost. Second, a spatial‑level attention module captures long‑range dependencies among the antenna elements, automatically emphasizing regions where the spherical wavefront exhibits rapid phase variations. Together, these modules enable the network to learn both inter‑channel correlations and global spatial patterns that are essential for disentangling angle and range information.

A key methodological choice is the use of the covariance matrix of the received signals as the network input instead of raw channel state information (CSI). The covariance metric aggregates inter‑antenna correlations, reduces dimensionality, and suppresses noise, leading to a more compact model (≈1.2 M parameters) and faster inference. Simulation results demonstrate that the covariance‑based model achieves over 30 % lower mean positioning error than a CSI‑based counterpart under identical signal‑to‑noise ratios, while also cutting training and inference time by roughly 40 %.

Extensive experiments compare the proposed approach against state‑of‑the‑art benchmarks: Attention‑Based Positioning Networks (ABPN), Near‑Field Positioning Networks (NFLnet), conventional convolutional neural networks (CNN), and multilayer perceptrons (MLP). Across a wide range of SNR values (down to 10 dB) and array sizes (up to 1024 antennas), the proposed model consistently outperforms all baselines, achieving mean positioning errors below 0.15 m—more than twice as accurate as the best competing method. Moreover, the model maintains high accuracy while using significantly fewer parameters than the CNN baseline.

The paper’s contributions are fourfold: (1) introduction of a dual‑path channel‑level attention mechanism that effectively mitigates angle‑range coupling in near‑field UCA data, (2) design of a spatial‑level attention module that exploits the omnidirectional geometry of UCAs, (3) demonstration that covariance‑based inputs provide superior accuracy and efficiency compared with raw CSI, and (4) comprehensive validation showing the method’s superiority over existing deep‑learning and model‑based positioning techniques.

In conclusion, this work presents a new paradigm for near‑field positioning in XL‑MIMO systems, leveraging attention‑enhanced deep learning to fuse physical propagation characteristics with data‑driven inference. Future directions include real‑world hardware validation, multi‑user simultaneous localization, and adaptive attention updates for mobile scenarios, paving the way for ultra‑accurate positioning services in forthcoming 6G networks.