Scholarlink

<p>Wi-Fi sniffers are commonly employed for silent packet capturing in wireless networks, playing a pivotal role in device localization using RSSI (Received Signal Strength Indicator). Nonetheless, RSSI’s sensitivity to environmental variability, particularly in complex indoor settings, often leads to diminished localization accuracy. In this paper, we explore the use of redundant sniffers to introduce spatial diversity, aiming to improve the accuracy of device localization. We assess the effectiveness of ensemble machine learning models with a focus on random forest to classify device locations from the feature-enhanced dataset. Our experimental results from both line-of-sight and non-line-of-sight environments demonstrate that our method substantially improves the localization accuracy and capture probability compared to conventional single-sniffer techniques, achieving robust performance across varying conditions, especially in smaller fingerprint samples. We observe that spatial diversity enables localization accuracy to exceed 99%, while also consistently reduces both the mean and standard deviation of localization error (in meters), for the same number of packets.</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

Benefits of redundancy in silent Wi-Fi fingerprinting

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月14日

查看全文
  • Pedro Videira Rubinstein,
  • Fernando Dias de Mello Silva,
  • Giuliano Prestes Fittipaldi,
  • Mohammad Imran Syed,
  • Rodrigo Souza Couto,
  • Anne Fladenmuller,
  • Marcelo Dias de Amorim,
  • Luís Henrique Maciel Kosmalski Costa

摘要

Wi-Fi sniffers are commonly employed for silent packet capturing in wireless networks, playing a pivotal role in device localization using RSSI (Received Signal Strength Indicator). Nonetheless, RSSI’s sensitivity to environmental variability, particularly in complex indoor settings, often leads to diminished localization accuracy. In this paper, we explore the use of redundant sniffers to introduce spatial diversity, aiming to improve the accuracy of device localization. We assess the effectiveness of ensemble machine learning models with a focus on random forest to classify device locations from the feature-enhanced dataset. Our experimental results from both line-of-sight and non-line-of-sight environments demonstrate that our method substantially improves the localization accuracy and capture probability compared to conventional single-sniffer techniques, achieving robust performance across varying conditions, especially in smaller fingerprint samples. We observe that spatial diversity enables localization accuracy to exceed 99%, while also consistently reduces both the mean and standard deviation of localization error (in meters), for the same number of packets.