Nearfield RF Sensing for Feature-Detection and Algorithmic Classification of Tamper Attacks

Abstract

As digital assets grow in value tremendously, the electronic cybersystems must become more secure. Physical attacks on electronic systems have been widely used to hack, then steal or modify assets in a criminal manner. These attacks are possible if adversaries gain physical access to the secure cybersystem. Therefore, a solution to protect the system against physical attacks is required. In this article, we detect and classify certain types of tampering events, such as probing, removing, adding, or moving components of a system, using the frequency characteristics of a wide-band antenna repurposed as a nearfield probe. Our idea is based on the notion that changes in the radio channel have an influence on an antenna’s frequency characteristics, which can be utilized to detect and identify certain tampering events. We empirically validate our claim by emulating various types of physical tampering incidents and collecting the ensuing RF signals leading to learnable feature maps. Our experimental results demonstrate that we can identify specific types of tampering events with more than 81.5% accuracy using machine learning algorithms. We ensure the robustness of our proposed technique for a variety of hardware platforms and in a noisy environment such as with the presence of both electrical additive white Gaussian noise from thermal effects, and mechanical perturbations on the setup.