Scholarlink

<p>This paper presents a transformer-based deep learning framework designed for uncertainty-aware indoor localization, utilizing Wi-Fi beamforming feedback information (BFI) from IEEE 802.11ac/ax systems. Unlike traditional methods that rely on Channel State Information (CSI), BFI can be conveniently extracted using standard MU-MIMO equipment and accurately represents spatial channel characteristics through simplified Givens rotation matrices. The proposed architecture treats BFI sequences as input tokens and employs a SimCLR-based contrastive learning strategy to enhance feature learning from unlabeled data. A Bayesian regression head is integrated to yield probabilistic location predictions, while an auxiliary signal quality estimation task reinforces model robustness against signal degradation. Extensive evaluations conducted on real-world datasets demonstrate an average localization error below 30&#xa0;cm and achieve a 60% reduction in uncertainty loss compared to baseline methods. The proposed system exhibits strong localization accuracy, resilience to dynamic indoor conditions, and robustness against signal interference, establishing its effectiveness for practical deployment in real-world indoor environments.</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

Transformer-based indoor positioning with signal robustness and calibration via Bayesian modeling

  • 期刊论文

  • 发表日期：

    2025年10月8日

查看全文
  • Osamah Abdullah,
  • Mohammed Aal-Nouman,
  • Ali Al-Saegh

摘要

This paper presents a transformer-based deep learning framework designed for uncertainty-aware indoor localization, utilizing Wi-Fi beamforming feedback information (BFI) from IEEE 802.11ac/ax systems. Unlike traditional methods that rely on Channel State Information (CSI), BFI can be conveniently extracted using standard MU-MIMO equipment and accurately represents spatial channel characteristics through simplified Givens rotation matrices. The proposed architecture treats BFI sequences as input tokens and employs a SimCLR-based contrastive learning strategy to enhance feature learning from unlabeled data. A Bayesian regression head is integrated to yield probabilistic location predictions, while an auxiliary signal quality estimation task reinforces model robustness against signal degradation. Extensive evaluations conducted on real-world datasets demonstrate an average localization error below 30 cm and achieve a 60% reduction in uncertainty loss compared to baseline methods. The proposed system exhibits strong localization accuracy, resilience to dynamic indoor conditions, and robustness against signal interference, establishing its effectiveness for practical deployment in real-world indoor environments.