Year: 2026
Type: article
Abstract: Internet of Things (IoT) devices are becoming increasingly popular thanks to the conveniences brought by smart systems. However, the widespread use of IoT devices has also made them vulnerable to attackers who want to infiltrate and damage them. Unfortunately, customized security measures on IoT devices are not feasible due to the limited resources of these devices. To address this issue, it is envisioned to use RF fingerprints, physical layer features of wireless signals, to identify IoT devices and authenticate them in the future. This study aims to capture and record RF signals using low-cost software-defined radio (SDR), a single-board computer, and open-source software. The captured signals are then analyzed to determine features for classification. The transient segment of the signal is determined, and the instantaneous amplitude (IA), instantaneous phase (IP), and instantaneous frequency (IF) values are obtained using the Hilbert Transform (HT). Various statistical methods such as skewness, kurtosis, standard deviation, variance, and median values are applied to these values to select the most relevant features. The minimum redundancy and maximum relevance (MRMR) technique further reduces the feature size. For the first time in RF fingerprinting, this study uses ELM-based Meta-ELM, Meta Multi-Layer ELM (Meta-MLELM), and Meta Constrained Mixed ELM (Meta-CMELM) algorithms as classifiers for device identification. The performance of these classifiers is evaluated based on the RF fingerprints dataset originally obtained in this study, which consists of 3752 raw signals from 7 devices collected in the study. After conducting detailed experiments, the study found that the Meta-MLELM classifier has an 88% accuracy rate in identifying devices, while the Meta-ELM classifier has an accuracy rate of around 90%. The Meta-CMELM classifier outperforms both with an accuracy rate of 92%.
Authors: Huseyin PARMAKSIZ, Cihan Karakuzu
Institution: Bilecik Şeyh Edebali Üniversitesi
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W4390768412