Channel-Resilient Deep-Learning-Driven Device Fingerprinting Through Multiple Data Streams

Abstract

Enabling accurate and automated identification of wireless devices is critical for allowing network access monitoring and ensuring data authentication for large-scale IoT networks. RF fingerprinting has emerged as a solution for device identification by leveraging the transmitters' inevitable hardware impairments that occur during manufacturing. Although deep learning is proven efficient in classifying devices based on hardware impairments, the performance of deep learning models suffers greatly from variations of the wireless channel conditions, across time and space. To the best of our knowledge, we are the first to propose leveraging MIMO capabilities to mitigate the channel effect and provide a channelresilient device classification framework. We begin by showing that for AWGN channels, combining multiple received signals improves the testing accuracy by up to 30%. We then show that for more realistic Rayleigh channels, blind channel estimation enabled by MIMO increases the testing accuracy by up to 50% when the models are trained and tested over the same channel, and by up to 69% when the models are tested on a channel that is different from that used for training.INDEX TERMS Automated network access, deep learning, IoT device fingerprinting, multiple-input multiple-out (MIMO).

Considerations, Advances, and Challenges Associated with the Use of Specific Emitter Identification in the Security of Internet of Things Deployments: A Survey

“…In [129], the authors present a DL-based device fingerprinting approach that leverages Multiple-Input Multiple-Output (MIMO) system capabilities and Space-Time Block Codes (STBCs) to mitigate the adverse effects AWGN and Rayleigh fading channels have on SEI performance. Since SEI features are distorted by Rayleigh fading channels, the approach in [129] exploits the MIMO system's multiple received signal streams to reconstruct a less-distorted version of transmitted signals, which are later used for model training and classification. Without knowledge of the channel state information, the transmitted signal is estimated at the receiver using two blind-source-separation and blind-channel-estimation algorithms, neither relying on pilot symbol-based estimation.…”

“…The presented approach's SEI performance is evaluated by adjusting the emitters' phase noise, CFO, and IQ gain imbalance hardware impairments within a fixed range. This allows the authors of [129] to simulate up to ten emitters with three antennas and Wi-Fi communication. The approach performs emitter identification via a Convolutional Neural Network (CNN) and signals that undergo varying Average Path Gain (APG) and Doppler shift changes.…”

Comprehensive RF Dataset Collection and Release: A Deep Learning-Based Device Fingerprinting Use Case

“…Additionally, due to the intrinsic nature of wireless channels, it is challenging and impractical to devise a universal channel-augmenting model capable of significantly improving performance across a wide range of wireless channels. Other data-centric approaches mitigate the impact of the channel through channel equalization [12], [13], [14] or impairment compensation [6].…”

“…However, it is widely believed that the wireless channel, influenced by various confounding factors, plays a significant role in the failure of these approaches to adapt and generalize to different domains. To address the impact of the channel and enhance domain generalization, some studies have focused on removing channel dynamics from the raw signal through techniques like channel equalization [12], [13], [14], [15] or hardware impairment compensation [6]. However, these approaches have drawbacks.…”