Detecting passive attacks is always considered difficult in vehicular networks. Passive attackers can eavesdrop on the wireless medium to collect beacons. These beacons can be exploited to track the positions of vehicles not only to violate their location privacy but also for criminal purposes. In this paper, we propose a novel federated learning-based scheme for detecting passive mobile attackers in 5G vehicular edge computing. We first identify a set of strategies that can be used by attackers to efficiently track vehicles without being visually detected. We then build an efficient machine learning (ML) model to detect tracking attacks based only on the receiving beacons. Our scheme enables federated learning (FL) at the edge to ensure collaborative learning while preserving the privacy of vehicles. Moreover, FL clients use a semi-supervised learning approach to ensure accurate self-labeling. Our experiments demonstrate the effectiveness of our proposed scheme to detect passive mobile attackers quickly and with high accuracy. Indeed, only 20 received beacons are required to achieve 95% accuracy. This accuracy can be achieved within 60 FL rounds using 5 FL clients in each FL round. The obtained results are also validated through simulations.

在车载网络中检测被动攻击总是被认为是困难的。被动攻击者可以窃听无线介质以收集信标。这些信标可用于跟踪车辆的位置，不仅会侵犯其位置隐私，还会用于犯罪目的。在本文中，我们提出了一种新的基于联合学习的方案，用于检测 5G 车辆边缘计算中的被动移动攻击者。我们首先确定一组策略，攻击者可以使用这些策略来有效跟踪车辆而不会被视觉检测到。然后，我们构建了一个高效的机器学习 (ML) 模型，以仅基于接收信标来检测跟踪攻击。我们的方案在边缘启用联邦学习 (FL)，以确保协作学习，同时保护车辆的隐私。而且，FL 客户端使用半监督学习方法来确保准确的自我标记。我们的实验证明了我们提出的方案在快速、高精度地检测被动移动攻击者方面的有效性。事实上，只需要 20 个接收到的信标就可以达到 95% 的准确度。在每轮 FL 中使用 5 个 FL 客户端可以在 60 轮 FL 内实现这种准确性。所获得的结果也通过模拟得到验证。