Scholarlink

<p>Advancements in passive sensing have elevated indoor localization to a key research focus in home security and human-computer interaction. Researchers are increasingly prioritizing the concurrent detection of individuals across interconnected spaces during practical deployments. However, obstacles such as walls separating these different spaces can result in signal attenuation, subsequently leading to a deterioration in the system's recognition capabilities. To address this challenge, this paper proposes the WiNAL system, which leverages a feature fusion approach to process Channel State Information (CSI) and introduces an adversarial training strategy for Non-Line-of-Sight WiFi localization. In this study, we convert amplitude and phase data into images and concatenate them into a composite image for enhanced analysis. And then, (Line of Sight) LoS and (Non Line of Sight) NLoS position predictors underwent cross-adversarial training to reduce environmental interference and enhance localization accuracy. Experimental results show that WiNAL achieves 98.7% accuracy in NLoS environments, outperforming TransLoc, AutoFi, and CNNLoc by 3.2%, 5.2%, and 16.3%, respectively.</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

Winal: Joint feature fusion and adversarial training for non line of sight WiFi localization

  • 期刊论文

  • 发表日期：

    2025年9月15日

查看全文
  • Zhongcheng Wei,
  • Jiacheng Li,
  • Bin Lian,
  • Wei Wang,
  • Jijun Zhao

摘要

Advancements in passive sensing have elevated indoor localization to a key research focus in home security and human-computer interaction. Researchers are increasingly prioritizing the concurrent detection of individuals across interconnected spaces during practical deployments. However, obstacles such as walls separating these different spaces can result in signal attenuation, subsequently leading to a deterioration in the system's recognition capabilities. To address this challenge, this paper proposes the WiNAL system, which leverages a feature fusion approach to process Channel State Information (CSI) and introduces an adversarial training strategy for Non-Line-of-Sight WiFi localization. In this study, we convert amplitude and phase data into images and concatenate them into a composite image for enhanced analysis. And then, (Line of Sight) LoS and (Non Line of Sight) NLoS position predictors underwent cross-adversarial training to reduce environmental interference and enhance localization accuracy. Experimental results show that WiNAL achieves 98.7% accuracy in NLoS environments, outperforming TransLoc, AutoFi, and CNNLoc by 3.2%, 5.2%, and 16.3%, respectively.