Radiometric identification using variational mode decomposition

https://doi.org/10.1016/j.compeleceng.2019.04.014Get rights and content

Abstract

Radiometric Identification (RAI) is the identification of wireless devices through their Radio Frequency (RF) emissions. In recent years, the research community has investigated it applying different methods and sets of statistical features extracted from the digitized RF emissions. In this paper, the authors investigate the application of Variational Mode Decomposition (VMD), recently introduced as an improvement to Empirical Mode Decomposition (EMD). VMD is applied to two sets of RF emissions from: wireless devices supporting Dedicated Short Range Communications (DSRC) at 5.9 GHz and Internet of Things wireless devices transmitting in the Industrial, Scientific and Medical (ISM) band at 2.4 GHz. Various machine learning algorithms have been used for classification and results are compared. Performances of VMD are evaluated against other approaches used in literature in Line of Sight (LOS) conditions, with Additive White Gaussian Noise (AWGN) and fading effects. Results show that VMD significantly outperforms other approaches.

Introduction

Radiometric Identification (RAI) is a technique to identify and authenticate wireless devices through their Radio Frequency (RF) emissions. The effectiveness of such techniques have been demonstrated in literature in various settings and propagation conditions. RAI exploits the presence of small differences in the material and the composition of the front-ends used for wireless transmission. These differences are usually not significant to hamper the correct functioning of wireless services but they are relevant enough for the unique identification of the model or the wireless device itself [1]. A synonym of RAI is also Radio Frequency - DNA (RF-DNA) because such small differences resemble the DNA of the human beings or RF fingerprinting like the fingerprints of the human skin of the fingers [2].

RAI is suitable for various applications [1], which are briefly described here. For security applications, multi-factor authentication using RF fingerprints has been proposed by many research papers [1], [2], [3]. The idea is that radiometric identification can be used as a form of authentication and complement conventional authentication means using cryptography. Because, radiometric identification is based on intrinsic physical features of the electronic device, these features cannot be stolen and they are quite difficult to reproduce. In this context, an authentication system can first collect the RF fingerprints (e.g., before market deployment or before they are put in service) and then compares them with the RF signal from the wireless device to authenticate. In the fight against the distribution of counterfeit electronic products, the authentication of electronic devices is an important function. Most of the approaches used in this domain are destructive: the electronic device must be opened and examined in a forensics lab to determine if it is a counterfeit or not. Then, a non-destructive approach would be quite useful. Because counterfeit devices are often built with lower quality components than legitimate devices [4], they may generate different RF fingerprints and radiometric identification can detect counterfeit electronic devices from legitimated devices without using destructive means. In this context, it is important to select features and classification algorithms, which are both accurate and time effective. These objectives are often conflicting and they represent a design trade-off, since the use of sophisticated features and algorithms may require a longer processing time compared to the application of simple features and algorithms, although the former provide a better identification accuracy. For both the applications described above, the identification and verification accuracy is the most relevant goal. If the number of false alarms is very high, the practical application of RAI can be hampered. In literature, RAI was applied to emissions from devices implemented with wireless standards including WiFi [5], ZigBee [6], WiMAX [2] and to Global System for Mobile Communications (GSM) in [3]. Because the implementation of RAI can be dependent on a specific wireless communication standard, in this paper we have used two different data sets to evaluate the performance of Variational Mode Decomposition (VMD). As expressed above, the main performance metrics of RAI are: 1) the identification accuracy of one device among others, 2) the robustness against environmental disturbances (e.g., attenuation at low signal-to-noise ratio (SNR) or in presence of fading) and 3) the computing time required by the classification algorithm. In this paper, we focus on the first two metrics and we evaluate the robustness against difficult propagation conditions in low SNR scenario. The computing time of the algorithm is already known in literature by the design of VMD and Empirical Mode Decomposition (EMD).

Since a digitized RF signal is indeed a time series, various strategies have been used to implement RAI and many techniques are based on of time series classification literature. The time series obtained for different wireless devices, can be compared either directly (e.g., euclidean distance and K Nearest Neighbour (KNN) classifiers) or using other approaches. A common strategy is to extract statistical features from the RF signal and then use a machine learning algorithm to classify the obtained set of features and correlate them with the identity of the wireless device. An extensive literature on the selection of different statistical features for RAI includes variance, entropy, skewness, kurtosis and others [2], [5]. The extraction of statistical features from the RF signal has the benefit of the dimensionality reduction, which improves the classification time but it can also decrease the accuracy if the statistical features are not selected properly. Additional details on the various approaches proposed in literature are provided in Section 2.

In this paper, VMD is applied to the digitized complex signal in the form of In-phase and Quadrature components (IQ) samples obtained from the collection of RF emissions of devices based on two different wireless standards: Dedicated Short Range Communications (DSRC) at 5.9 GHz and Internet of Things (IoT) devices transmitting in the Industrial, Scientific and Medical (ISM) band at 2.4 GHz. Details on the test bed, a description of the wireless standards and the capture of the signals are in Section 4. Signals captured from the wireless devices are usually represented as bursts, which are repeated in time and they are usually composed from a transient portion and a steady-state portion. As described in [7], a transient signal can be described as a short signal (typically lasting a few microseconds) that occurs during transmitter power-on sequence. It is noted in [7] that the capture and digitization of the transient signal requires very high oversampling rates and sophisticated and expensive receiver architectures. In contrast to the transient signal, the steady-state signal portion can be much longer than the transient part of the burst, thus providing more information for classification purposes. On the other side, the steady part can depend on the content (e.g., voice or data) being transmitted. The problem occurs because the content in the signal may not be exactly the same in all the sets of collected bursts from each device. Consequently, a classification bias is introduced because the classifiers may indicate the different content rather than the devices themselves. Then, the usual approach reported in literature [7], [8] is to use only the content independent part of the steady signal, which is usually represented by the preamble. Today, almost all digital communication systems introduce a preamble at the very start of packet transmission, in order to simplify the receiver design. Another possibility is to set the transmission of the wireless devices to a test mode and the same testing sequences are transmitted by all the wireless devices under tests. Both techniques are used in this paper, because only the preamble is used in the DSRC transmitters while the entire steady portion of the signal is used in the IoT devices. The VMD is particularly suited for the analysis of the steady part of the burst. As described in [9], VMD is particularly suited for stationary signals and the steady part of the burst is usually a stationary or quasi-stationary signal. For these reasons, this paper applies VMD only to the steady part of the signal.

This paper addresses both the problem of identification and verification in a similar way to what is presented in specific references [2]. The terms identification and verification are used in a manner consistent with other references in literature: Verification is the process of confirming the claimed identity of a wireless device using the RAI. Then, a binary classification is performed. Identification is the process where the recognition system determines a wireless device’s identity by comparing the device fingerprints with reference fingerprints templates for all known devices. Identification requires a one-to-many type of comparison and multi-classification algorithms.

Our Contribution:The novelty of this paper is the application of the VMD to the problem of radiometric identification using experimental data captured from two different sets of wireless devices using two wireless standards. VMD has been applied in other contexts, where it has shown superior performance compared to classical methods and to EMD and the objective of this paper was to evaluate its performance for this type of problem. A state of the art on recent findings on radiometric identification and the application of VMD is provided in Section 2. The paper makes a comparison among different methods for radiometric identification used in literature, which are based on the Time Domain (TD) representation, Frequency Domain (FD) representation (application of the Fast Fourier Transform (FFT) to the original signal), and the application of EMD. Different machine learning algorithms are used for classification both for identification and verification problems. In addition, the performance of VMD has been evaluated in presence of low SNR or in presence of fading. As described before and contrary to many recent studies, where simulated data is used to evaluate the performance of radiometric identification, in this paper the authors have used experimental data collected in their lab for two separate sets of wireless standards.

The structure of the paper is the following: Section 2 provides an overview of the state of the art on radiometric identification and VMD in various domains. Section 3 describes the VMD and the EMD techniques. Section 4 provides a description of the methodology for collecting the data used in the evaluation. This Section includes also a description of the experimental laboratory work used to collect the RF emissions and a short summary of the main characteristics of the wireless standards (DSRC and IoT), used in the experiment. Section 5 provides the experimental results and the related analysis, where a comparison of different statistical features and machine learning algorithms is performed. Specific subsections describe the evaluation in presence of Additive White Gaussian Noise (AWGN) and fading conditions. Finally, Section 6 concludes this paper.

Section snippets

State of art

The concept to identify wireless devices through the specific characteristics of their RF emissions has been widely investigated by the research community in recent years. As described in the Introduction, the concept is based on physical differences in the hardware components of the wireless communication front end, which generates small but significant and reproducible features in the RF emissions. Even if such differences are usually within the boundaries defined by the wireless standard

Empirical mode decomposition

The EMD, originally proposed in [18], is a method to decompose a signal into a set of IMFs. For each IMF, in the whole data set, the number of extrema and the number of zero crossings is the same or differs at most by one, and, at any point, the mean value of the envelope, defined by the local maxima and the envelope defined by the local minima is zero (i.e., it is symmetric with respect to the local zero mean) [18]. The way to extract the IMFs from a given signal s(t) is the sifting process,

Materials and method

This Section is divided in two separate subsections: a) a description of the workflow from the collection of the RF emissions to the application of the machine learning algorithms and b) a description of the test bed where RF emissions were collected.

Parameters optimization

Each of the machine learning algorithms used in this paper requires the tuning and optimization of specific parameters: the K index in KNN, the number of branches in Decision Trees, the factor C and scaling factor γ of the Radial Basis Function (RBF) kernel in SVM. In the rest of this paper, we call Box Constraint the penalty factor C of SVM. The parameter C controls the trade-off between the margin and the size of the slack variables. The impact of the optimization parameters is evaluated in

Conclusions

The application of Variational Mode Decomposition to the problem of Radiometric Identification is a novelty in literature. This paper has adopted it on two data sets of wireless devices based on two different wireless standards: Dedicated Short Range Communications and Internet of Things devices (transmitting respectively at 5.9 and 2.4 GHz). The results show that the performance of the Variational Mode Decomposition is significantly better than the other representations (i.e., time domain,

Acknowledgment

This work has been partially supported by the European Commission through Project SerIoT funded by the European Union H2020 Programme under Grant Agreement No. 780139. The opinions expressed in this paper are those of the authors and do not necessarily reflect the views of the European Commission.

Conflict of interest

The authors declare that there is no conflict of interest regarding the publication of this article.

Gianmarco Baldini completed his degree in 1993 in Electronic Engineering from the University of Rome La Sapienza with specialization in Wireless Communications. He has worked in Italy, UK, Ireland and USA as Senior Technical Architect and System Engineering Manager in Ericsson, Lucent Technologies, Hughes Network Systems and Finmeccanica (now Leonardo) before joining the Joint Research Centre of the European Commission in 2007 as a Scientific Officer. His current research activities focus on

References (23)

  • Y. Wang et al.

    Research on variational mode decomposition and its application in detecting rub-impact fault of the rotor system

    Mechan Syst Signal Process

    (2015)
  • A. Upadhyay et al.

    Determination of instantaneous fundamental frequency of speech signals using variational mode decomposition

    Comput Electr Eng

    (2017)
  • Q. Xu et al.

    Device fingerprinting in wireless networks: challenges and opportunities

    IEEE Commun Surv Tutor

    (2016)
  • D.R. Reising et al.

    Gabor-based RF-DNA fingerprinting for classifying 802.16e WiMAX mobile subscribers

    Computing, Networking and Communications (ICNC), 2012 International Conference on

    (2012)
  • D.R. Reising et al.

    Improved wireless security for gmsk-based devices using rf fingerprinting

    Int J Electron Secur Digi Forens

    (2010)
  • U. Guin et al.

    A comprehensive framework for counterfeit defect coverage analysis and detection assessment

    J Electron Test

    (2014)
  • W.C. Suski II et al.

    Using spectral fingerprints to improve wireless network security

    IEEE Global Telecommunications Conference

    (2008)
  • T.J. Bihl et al.

    Dimensional reduction analysis for physical layer device fingerprints with application to zigbee and z-wave devices

    Military Communications Conference, MILCOM 2015 - 2015 IEEE

    (2015)
  • P. Scanlon et al.

    Feature extraction approaches to RF fingerprinting for device identification in femtocells

    Bell Labs Tech J

    (2010)
  • I.O. Kennedy et al.

    Radio transmitter fingerprinting: A steady state frequency domain approach

    Vehicular Technology Conference, 2008. VTC 2008-Fall. IEEE 68th

    (2008)
  • K. Dragomiretskiy et al.

    Variational mode decomposition

    IEEE Trans Signal Process

    (2014)
  • Cited by (11)

    • Channel identification with Improved Variational Mode Decomposition

      2022, Physical Communication
      Citation Excerpt :

      VMD is particularly suited for narrowband signals (like the WiFi preamble) and it has demonstrated its discriminating power in comparison to the time domain in various signal processing problems including mechanical related problems [12,13] or health related problems (e.g., ECG analysis) [10,14]. To the best knowledge of the authors, VMD has never been applied to channel identification problems even if it was applied to wireless device identification (i.e., RF fingerprinting) in [15,16]. The rationale for the application of VMD to this problem of channel identification is that the impact of environmental differences in the radio frequency communication channel has the potential to change the characteristics and the predominance of the specific modes (which can be extracted using EMD or VMD) composing the signal.

    • On the application of sensor authentication with intrinsic physical features to vehicle security

      2021, Computers and Electrical Engineering
      Citation Excerpt :

      All the techniques presented in the previous paragraphs and references focus on the physical layer identification and authentication of the ECU connected to the in-vehicle network but the tampering and replacement of the sensors is also identified as a key threat in [2] and it is present in the tachograph regulated application [6]. To the knowledge of the authors, the analysis of this technique is missing in the automotive sector literature, even if electronic device identification was analyzed in literature in other contexts like IoT [11] or for the built-in sensors of smartphones [13]. In the avionics sector, a similar approach based on fingerprinting of electronic devices for identification and authentication was proposed in the in-vehicle networks of the airplanes.

    • Primary signal phase suppression technique based on mutual information improved variational mode decomposition

      2023, Proceedings of SPIE - The International Society for Optical Engineering
    View all citing articles on Scopus

    Gianmarco Baldini completed his degree in 1993 in Electronic Engineering from the University of Rome La Sapienza with specialization in Wireless Communications. He has worked in Italy, UK, Ireland and USA as Senior Technical Architect and System Engineering Manager in Ericsson, Lucent Technologies, Hughes Network Systems and Finmeccanica (now Leonardo) before joining the Joint Research Centre of the European Commission in 2007 as a Scientific Officer. His current research activities focus on Internet of Things, navigation, wireless communications, machine learning, security and privacy.

    Gary Steri is a researcher at the Joint Research Center of the European Commission. He received his masters degree in Information Technologies in 2006 and his PhD in Computer Science in 2011. His research activity focuses on security and authentication aspects of wireless networks.

    Raimondo Giuliani was born in Bologna and received the Laurea in Ingegneria delle Telecomunicazioni in 1996 from the University of Bologna. He worked in the Fondazione Ugo Bordoni, Fondazione G. Marconi, CSELT and ITU on radio communications. In 2007, he joined the Joint Research Centre of the European Commission. Current research topics include radiometric signature identification, devices fingerprinting, distributed ledgers in electric power systems and inertial and satellite positioning systems for automotive applications.

    Franc Dimc obtained his BSc (1993), M.Sc. (1996) and Ph.D. (2010) degrees from the Faculty of Electrical Engineering, University of Ljubljana, Slovenia. In addition to work with students he is involved in the fields of geophysical exploration and maritime awareness and he attends to the needs of contemporary maritime and coastal intelligent transportation systems.

    This paper is for regular issues of CAEE. Reviews processed and recommended for publication to the Editor-in-Chief by Area Editor Dr. E. Cabal-Yepez.

    View full text