Deep learning vs. traditional learning for radio frequency fingerprinting

Otto, ARananga, SMasonta, Moshe T2024-09-182024-09-182024-05Otto, A., Rananga, S. & Masonta, M.T. 2024. Deep learning vs. traditional learning for radio frequency fingerprinting. http://hdl.handle.net/10204/13761 .978-1-905824-73-1DOI: 10.23919/IST-Africa63983.2024.10569298http://hdl.handle.net/10204/13761Radio Frequency (RF) fingerprinting is the theory of identifying a wireless device based on its unique transmitting characteristics. RF fingerprinting uses the validated concept that the physical components and configuration of a transmitting device can result in a distinct wireless emission. This research focuses on the application of machine learning algorithms, specifically Support Vector Machines (SVMs) and Convolutional Neural Networks (CNNs) for the task of RF fingerprinting. The primary aim of this research paper is to comparatively assess the performance of SVMs and CNNs in RF fingerprinting for wireless device identification, focusing on hyperparameters, accuracy and real-world applicability. The study includes an in-depth implementation and evaluation of the SVMs and CNNs models, considering their performance in a high-dimensional dataset of multiple transmissions and wireless devices. While the CNN model slightly outperformed the SVM in terms of classification accuracy, other metrics such as inference time and training duration made the SVM equally competitive. The high accuracy and competitive inference times affirm the real-world applicability of these models, and their need to be further explored.FulltextenRadio frequency fingerprintingSupport vector machinesConvolutional neural networksPhysical layer securityDeep learning vs. traditional learning for radio frequency fingerprintingConference PresentationOtto, A., Rananga, S., & Masonta, M. T. (2024). Deep learning vs. traditional learning for radio frequency fingerprinting. http://hdl.handle.net/10204/13761Otto, A, S Rananga, and Moshe T Masonta. "Deep learning vs. traditional learning for radio frequency fingerprinting." <i>IST-Africa Conference, Virtual, 20 - 24 May 2024</i> (2024): http://hdl.handle.net/10204/13761Otto A, Rananga S, Masonta MT, Deep learning vs. traditional learning for radio frequency fingerprinting; 2024. http://hdl.handle.net/10204/13761 .TY - Conference Presentation AU - Otto, A AU - Rananga, S AU - Masonta, Moshe T AB - Radio Frequency (RF) fingerprinting is the theory of identifying a wireless device based on its unique transmitting characteristics. RF fingerprinting uses the validated concept that the physical components and configuration of a transmitting device can result in a distinct wireless emission. This research focuses on the application of machine learning algorithms, specifically Support Vector Machines (SVMs) and Convolutional Neural Networks (CNNs) for the task of RF fingerprinting. The primary aim of this research paper is to comparatively assess the performance of SVMs and CNNs in RF fingerprinting for wireless device identification, focusing on hyperparameters, accuracy and real-world applicability. The study includes an in-depth implementation and evaluation of the SVMs and CNNs models, considering their performance in a high-dimensional dataset of multiple transmissions and wireless devices. While the CNN model slightly outperformed the SVM in terms of classification accuracy, other metrics such as inference time and training duration made the SVM equally competitive. The high accuracy and competitive inference times affirm the real-world applicability of these models, and their need to be further explored. DA - 2024-05 DB - ResearchSpace DP - CSIR J1 - IST-Africa Conference, Virtual, 20 - 24 May 2024 KW - Radio frequency fingerprinting KW - Support vector machines KW - Convolutional neural networks KW - Physical layer security LK - https://researchspace.csir.co.za PY - 2024 SM - 978-1-905824-73-1 T1 - Deep learning vs. traditional learning for radio frequency fingerprinting TI - Deep learning vs. traditional learning for radio frequency fingerprinting UR - http://hdl.handle.net/10204/13761 ER -27893