MAG-JAM: Jamming Detection via Magnetic Emissions

Abstract

Wireless networks inherently rely on a shared medium, making them exposed to jamming attacks. In this paper, we present MAG-JAM, a novel solution for jamming detection in static and mobile scenarios leveraging the physical layer properties of wireless communication by analyzing the magnetic emissions near the antennas of target wireless devices. To the best of our knowledge, MAG-JAM represents the first solution based on the key observation that the magnetic emissions profile of normal wireless communication between transmitter-receiver pairs is different from the magnetic emissions profile when an active jamming signal starts affecting the communication channel. MAG-JAM has several advantages: its implementation requires mainly an inexpensive magnetic sensor, it is non-invasive and privacy-preserving as it is implemented as a standalone unit, does not need access to the wireless device, and demonstrates a remarkable performance. We design and implement a proof of concept jamming detection system using a cheap magnetic sensor and test MAG-JAM on a set of different wireless devices with a perfect score in jamming detection using no more than 1 s of the magnetic emissions collected by the magnetic sensor under a normalized jamming power of 0.1–1. In addition, we also implement a more advanced jamming detection system using a specialized magnetic probe and autoencoders that, using just 150 ms of collected data, achieves a minimum of 0.91 F1-Score in detecting jamming with a normalized power of 0.2 and an F1-Score of 1 for jamming powers greater than 0.4.

Appendix A

A.1 Different Jamming Signals

We test the performance of MAG-JAM on jammers that employ different types of jamming signals, namely Gaussian and Uniform. Recall that in all of our previous experiments, we used a jammer with Gaussian noise signals. For the Uniform noise jamming depicted in Fig. 9a we analyze the magnetic emissions from a USRP transmitter-receiver pair under six distinct scenarios. Initially, the pair is observed in a normal communication state without any active jammer. After that, we introduce jamming with a uniform noise signal at varying powers ranging from 0.2 to 1, in steps of 0.2, with each scenario separated by black lines.

Fig. 9.

Uniform noise jamming and autoencoder performance.

We observe that the trends in magnetic emissions intensity with increasing jamming powers in the uniform noise scenario are similar to those seen with Gaussian noise jamming, as shown in Fig. 6a. Furthermore, Fig. 9b presents the performance of the autoencoders when trained on normal no-jamming samples and tested with samples collected under jamming with uniform signals. Notably, the performance achieved in this uniform jamming scenario is comparable to that obtained in the Gaussian jamming scenario.

A.2 MAG-JAM Robustness

To ascertain that MAG-JAM is not biased towards any particular jammer, target device, or time frame, we conduct comprehensive testing of its jamming detection capabilities across identical devices at varied time instances, using different jammers and target devices. This involved testing MAG-JAM with two distinct USRP jammers targeting six separate USRP devices. The methodology entailed training the autoencoders with data traces obtained from the first USRP jammer targeting a USRP transmitter-receiver pair. Subsequently, we employed the trained model for testing with the second jammer, which was set to target a second transmitter communicating with three other identical USRP receiver devices. These tests were conducted with jamming powers of 0.6, 0.8, and 1, with each power level being recorded independently. The results, depicted in Fig. 10, demonstrate the proficiency of the autoencoder, initially trained on data from the first USRP jammer and target pair, in distinctly identifying jamming activities from the second USRP jammer on subsequent USRP transmitter-receiver pairs (labeled 2, 3, and 4).

Fig. 10.

Autoencoder MSE in detecting jamming in identical USRP devices.

A.3 MAG-JAM Jamming Detection in a Crowded Environment

We also investigate the impact of environmental factors on magnetic emissions by placing a USRP transmitter-receiver pair in a crowded area, with people constantly moving in the path between the transmitter and receiver. We can see a noticeable difference in the observed magnetic emissions depicted in Fig. 11a compared to Fig. 6a that is recorded in a lab environment with minimal movements. However, as demonstrated in Fig. 11b, our autoencoder-based solution can detect the presence of jamming activity even in these crowded environments. We can notice various jamming powers are less distinct than in the controlled lab settings with fewer movements, as shown in Fig. 7. This reduced clarity is attributed to the fact that excessive movements around the wireless devices introduce additional fluctuations into each magnetic emissions sample, resulting in a less stable spectrum than that observed in quieter environments.

Fig. 11.

Profile of the magnetic emissions for USRP receiver placed in a crowded area and the autoencoder performance.