Latest generation of Wi-Fi protocols are progressively introducing new functionalities to improve their support for the extreme reliability and low-latency requirements demanded by a plethora of new applications like extended reality (XR), 8K video, remote telepresence, digital twin for manufacturing, and cooperative mobile robots. Indeed, reliability is a pivotal key metric of next-generation Wi- Fi networks aiming at penetrating into industrial applications. However, malicious attacks produced by accessible and easy to build jammers may pose critical challenges, especially in critical industrial communications involving the use of unlicensed spectrum. It becomes of crucial importance to properly detect the presence of a jammer attack in next generation Wi-Fi networks. To tackle this problem, we propose and evaluate experimentally the use of machine learning (ML). Two scenarios are presented: the first where the jammer takes down the network, and the other where it acts as interference degrading the network performance. The study in the two presented scenarios are used to derive important conclusions on features collected for training purposes, accuracy of different ML models, ability to distinguish among the effect produced by legitimate co-channel interference, and malicious jamming attacks. Overall, we experimentally show that ML algorithms are highly capable of detecting in real time the presence of a jammer attack.
Paper
Full text
Experimental Jamming Detection Using Machine Learning in IEEE 802.11 Enterprise Networks
Semantic Scholar · Computer Science · 2023
Abstract
Latest generation of Wi-Fi protocols are progressively introducing new functionalities to improve their support for the extreme reliability and low-latency requirements demanded by a plethora of new applications like extended reality (XR), 8K video, remote telepresence, digital twin for manufacturing, and cooperative mobile robots. Indeed, reliability is a pivotal key metric of next-generation Wi- Fi networks aiming at penetrating into industrial applications. However, malicious attacks produced by accessible and easy to build jammers may pose critical challenges, especially in critical industrial communications involving the use of unlicensed spectrum. It becomes of crucial importance to properly detect the presence of a jammer attack in next generation Wi-Fi networks. To tackle this problem, we propose and evaluate experimentally the use of machine learning (ML). Two scenarios are presented: the first where the jammer takes down the network, and the other where it acts as interference degrading the network performance. The study in the two presented scenarios are used to derive important conclusions on features collected for training purposes, accuracy of different ML models, ability to distinguish among the effect produced by legitimate co-channel interference, and malicious jamming attacks. Overall, we experimentally show that ML algorithms are highly capable of detecting in real time the presence of a jammer attack.