To See or Not to See -- Fingerprinting Devices in Adversarial Environments Amid Advanced Machine Learning

TL;DR

The paper tackles the challenge of securely authenticating IoT devices and detecting eavesdropping in adversarial environments where attackers wield advanced machine learning tools. It formalizes a generic device fingerprinting model and surveys RF, network, electromagnetic, and thermal modalities, along with adversarial attacks and defenses. By analyzing performance metrics (range, scalability, spoofing resistance) and highlighting methodological gaps, the work provides guidance for designing more robust, multi-modal fingerprinting schemes and realistic testbeds. The findings underscore the need for modeling adversaries, incorporating localization, and conducting real-world experiments to ensure IoT security in practice.

Abstract

The increasing use of the Internet of Things raises security concerns. To address this, device fingerprinting is often employed to authenticate devices, detect adversaries, and identify eavesdroppers in an environment. This requires the ability to discern between legitimate and malicious devices which is achieved by analyzing the unique physical and/or operational characteristics of IoT devices. In the era of the latest progress in machine learning, particularly generative models, it is crucial to methodically examine the current studies in device fingerprinting. This involves explaining their approaches and underscoring their limitations when faced with adversaries armed with these ML tools. To systematically analyze existing methods, we propose a generic, yet simplified, model for device fingerprinting. Additionally, we thoroughly investigate existing methods to authenticate devices and detect eavesdropping, using our proposed model. We further study trends and similarities between works in authentication and eavesdropping detection and present the existing threats and attacks in these domains. Finally, we discuss future directions in fingerprinting based on these trends to develop more secure IoT fingerprinting schemes.

Figures (4)

• Figure 1: In an adversarial fingerprinting setting, the adversary may target the system by designing a spoofing method to bypass the authentication and/or by installing a hidden device (eavesdropper) to steal information.
• Figure 2: Overall diagram of fingerprinting systems. The system receives a signal as input and preprocesses it. This data is then sent to a classifier. An adversarial detector detects an attack. Finally, we have a Decision which tells us 1) which device was authenticated, 2) was a hidden device detected, and 3) was an adversary found?
• Figure 3: Systematized view of fingerprinting systems. Our input takes on a certain modulation m depending on the system. For each sample, an error term $\epsilon$ is added. We then average our samples ($\Sigma$) and preprocess our data with p. During the preprocessing stage, we take on another error term $n$. Finally, we classify our device with C and attempt to localize our device with L. An adversary can utilize its own processing and model to try to fool the classifier C, receiving different types of input depending on the threat model.
• Figure 4: 1) The adversary places their Rx near the trusted Rx and receives samples from the trusted Tx and the adversarial Tx. These samples are sent to a generative model (Gen), which iteratively learns how to generate samples that mimic the trusted Tx. 2) The adversarial Tx sends samples over the air to the trusted Rx, in the presence of trusted Tx's, and fools the trusted Rx.