Spoofing Detection in the Physical Layer with Graph Neural Networks
arxiv(2024)
摘要
In a spoofing attack, a malicious actor impersonates a legitimate user to
access or manipulate data without authorization. The vulnerability of
cryptographic security mechanisms to compromised user credentials motivates
spoofing attack detection in the physical layer, which traditionally relied on
channel features, such as the received signal strength (RSS) measured by
spatially distributed receivers or access points. However, existing methods
cannot effectively cope with the dynamic nature of channels, which change over
time as a result of user mobility and other factors. To address this
limitation, this work builds upon the intuition that the temporal pattern of
changes in RSS features can be used to detect the presence of concurrent
transmissions from multiple (possibly changing) locations, which in turn
indicates the existence of an attack. Since a localization-based approach would
require costly data collection and would suffer from low spatial resolution due
to multipath, the proposed algorithm employs a deep neural network to construct
a graph embedding of a sequence of RSS features that reflects changes in the
propagation conditions. A graph neural network then classifies these embeddings
to detect spoofing attacks. The effectiveness and robustness of the proposed
scheme are corroborated by experiments with real-data.
更多查看译文
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要