Title: Radio frequency fingerprinting commercial communication devices to enhance electronic security

Authors: William C. Suski II, Michael A. Temple, Michael J. Mendenhall, Robert F. Mills

Addresses: Department of Electrical and Computer Engineering, School of Engineering and Management, US Air Force Institute of Technology (AFIT), Wright-Patterson AFB, Dayton OH 45433, USA. ' Department of Electrical and Computer Engineering, School of Engineering and Management, US Air Force Institute of Technology (AFIT), Wright-Patterson AFB, Dayton OH 45433, USA. ' Department of Electrical and Computer Engineering, School of Engineering and Management, US Air Force Institute of Technology (AFIT), Wright-Patterson AFB, Dayton OH 45433, USA. ' Department of Electrical and Computer Engineering, School of Engineering and Management, US Air Force Institute of Technology (AFIT), Wright-Patterson AFB, Dayton OH 45433, USA

Abstract: There is a current shift toward protecting against unauthorised network access at the open systems interconnection physical layer by exploiting radio frequency characteristics that are difficult to mimic. This work addresses the use of RF |fingerprints| to uniquely identify emissions from commercial devices. The goal is to exploit inherent signal features using a four step process that includes: 1. feature generation, 2. transient detection, 3. fingerprint extraction and 4. classification. Reliable transient detection is perhaps the most important step and is addressed here using a variance trajectory approach. Following transient detection, two fingerprinting and classification methods are considered, including 1. power spectral density (PSD) fingerprints with spectral correlation and 2. statistical fingerprints with multiple discriminant analysis-maximum likelihood (MDA-ML) classification. Each of these methods is evaluated using the 802.11a orthogonal frequency-division multiplexing (OFDM) signal. For minimal transient detection error, results show that amplitude-based detection is most effective for 802.11a OFDM signals. It is shown that MDA-ML classification provides approximately 8.5-9.0% better classification performance than spectral correlation over a range of analysis signal-to-noise ratios (SNRA) using three hardware devices from two manufacturers. Overall, greater than 80% classification accuracy is achieved for spectral correlation at SNRA > 6 dB and for MDA-ML classification at SNRA > −3 dB.

Keywords: anti-spoofing; burst detection; electronic security; Fisher linear discriminant; multiple discriminant analysis; PHY layer authentication; physical layer authentication; RF fingerprinting; radio frequency fingerprinting; RF forensics; SEI; specific emitter identification; transient detection; commercial communication devices; signal features; feature generation; OSI; network access; access authorisation.

DOI: 10.1504/IJESDF.2008.020946

International Journal of Electronic Security and Digital Forensics, 2008 Vol.1 No.3, pp.301 - 322

Published online: 25 Oct 2008 *

Full-text access for editors Access for subscribers Purchase this article Comment on this article