Scholarlink

<p>Massive Multiple-Input Multiple-Output (m-MIMO) systems with Millimeter-Wave (mm-Wave) promise high data rates and high spectral efficiency (SE) in beyond fifth-generation (B5G) wireless networks. This paper proposes a novel deep learning approach that synergizes Graph Neural Networks (GNNs) and U-Net architectures to efficiently learn optimal hybrid beamforming strategies from channel state information (CSI). The GNN is used to capture the spatial and topological relationships between antennas and users for generalizing the system in various configurations and distributions. Simultaneously, the U-Net is used to process the structured CSI representations for extracting spatial features for enhanced precoder rebuilding. The proposed approach is trained using a loss function for optimizing SE and Bit Error Rate (BER). Simulation results show that the integrated GNN and U-Net approach achieves superior performance to conventional methods, with significantly reduced computational overhead during inference. Across SNR values from − 18 dB to 6 dB, the method yields 9% to 54% SE improvement over state-of-the-art hybrid beamforming techniques while maintaining lower BER.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Integration of a GNN and U-Net for Hybrid Beamforming in mm-Wave m-MIMO Systems

  • 期刊论文

  • 发表日期：

    2025年12月15日

查看全文
  • Gurpreet Kaur,
  • Gurmeet Kaur

摘要

Massive Multiple-Input Multiple-Output (m-MIMO) systems with Millimeter-Wave (mm-Wave) promise high data rates and high spectral efficiency (SE) in beyond fifth-generation (B5G) wireless networks. This paper proposes a novel deep learning approach that synergizes Graph Neural Networks (GNNs) and U-Net architectures to efficiently learn optimal hybrid beamforming strategies from channel state information (CSI). The GNN is used to capture the spatial and topological relationships between antennas and users for generalizing the system in various configurations and distributions. Simultaneously, the U-Net is used to process the structured CSI representations for extracting spatial features for enhanced precoder rebuilding. The proposed approach is trained using a loss function for optimizing SE and Bit Error Rate (BER). Simulation results show that the integrated GNN and U-Net approach achieves superior performance to conventional methods, with significantly reduced computational overhead during inference. Across SNR values from − 18 dB to 6 dB, the method yields 9% to 54% SE improvement over state-of-the-art hybrid beamforming techniques while maintaining lower BER.