1556-6013 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIFS.2018.2855665, IEEE Transactions on Information Forensics and Security 1 Specific Emitter Identification Based on Variational Mode Decomposition and Spectral Features in Single Hop and Relaying Scenarios Udit Satija * , Nikita Trivedi * , Gagarin Biswal, and Baratram.Ramkumar Abstract—Specific emitter identification is the process of iden- tifying or discriminating different emitters based on the radio fre- quency (RF) fingerprints extracted from the received signal. Due to inherent non-linearities of the power amplifiers of emitters, these fingerprints provide distinguish features for emitter identi- fication. In this paper, we develop an emitter identification based on variational mode decomposition and spectral features (VMD- SF). As VMD decomposes the received signal simultaneously into various temporal and spectral modes, we choose to explore different spectral features including spectral flatness, spectral brightness, and spectral roll-off for improving the identification accuracy contrary to existing temporal features based methods. For demonstrating the robustness of VMD in decomposing the received signal into emitter-specific modes, we also develop a VMD-EM 2 method based on existing temporal features (such as entropy and moments (EM 2 )) extracted from Hilbert Huang transform of the emitter-specific temporal modes. Our proposed method has three major steps: received signal decomposition us- ing VMD, feature extraction, and emitter identification. We eval- uate the performance of the proposed methods using probability of correct classification (Pcc) both in single hop and in relaying scenario by varying the number of emitters. To demonstrate the superior performance of our proposed methods, we compared our methods with the existing empirical mode decomposition- (entropy, first, second order moments) (EMD-EM 2 ) method both in terms of Pcc and computational complexity. Results depict that the proposed VMD-SF emitter identification method outperforms proposed VMD-EM 2 method and existing EMD-EM 2 method both in single hop and relaying scenarios for varying number of emitters. In addition, the proposed VMD-SF method has lowest computational cost as compared to aforementioned methods. Index Terms—Specific emitter identification, Feature extrac- tion, Spectral features. I. I NTRODUCTION Specific emitter identification (SEI) designates the individ- ual emitter by discriminating it from other emitters based on the features possessed by different emitters. Such features are extracted from the received signal and are termed as radio frequency (RF) fingerprints [1]–[4]. The applications of SEI technique lie in military communication, radar systems, interference source determination and traffic analysis in mili- tary spectrum management processes [5], cognitive radio [6], self-organized networks [7]. The key to emitter identification problem is the set of features that discriminates the trans- mitters/emitters. The process of emitter identification involves Udit Satija, Nikita Trivedi, Gagarin Biswal and Barathram.Ramkumar are with School of Electrical Sciences, Indian Institute of Technology Bhubaneswar, Odisha India. E-mail: us11@iitbbs.ac.in, nt10@iitbbs.ac.in, gb11@iitbbs.ac.in, barathram@iitbbs.ac.in. three major steps: extraction of features (RF fingerprints) from the received signal, comparing the features with a categorized set of features and then assigning the best matching class to these features. Based on the operating characteristics of the emitter, emitter identification can be performed on transients signals or on steady state signals. Transient signal is generally referred as power on/off signal, which produces emitter specific and discriminative information for feature extraction [8], [9]. As the duration of transient signal is very short, it is difficult to extract the start and end point of its occurrence from noise [10]. In addition, SEI based on transient signals basically depends on the features like amplitude, phase and frequency. However, these features are easily deteriorated by non-ideal and complex channel conditions, which can alter the perfor- mance significantly [11]. On the other hand, steady state signal is defined when the emitter is transmitting between the start and end of the transients over the entire signal. Though the detection of steady state signal is comparatively simple, extraction of steady state features is difficult, as these features tend to corrupt by the transmitted information. Steady state features constitute significant information therefore a number of feature extraction schemes have been developed based on bispectrum [12], bio- inspired algorithm [13], cumulant [14], preamble and time- frequency representation approaches [15] such as short-time Fourier transform (STFT) [16], Wavelet [17], and Wigner and Choi-Williams distribution [18]. In preamble based techniques, the preamble is first extracted then the RF fingerprints from the preamble are extracted [19]. Sometimes, it becomes difficult to extract the complete preamble. In bispectrum-based algo- rithms, large dimension of feature vector increases the classi- fier scale and hence reduces the robustness of the algorithm. Most of the existing SEI methods focus on time-frequency based features extraction. In time-frequency based technique, the received signal is mapped to a two-dimensional plane with one axis as time and other as frequency [20]. Such mapping provides simultaneous temporal and spectral information. A STFT based signal detection and identification was proposed in [16]. However, STFT based methods are linear in nature therefore these methods cannot be applied for analysis of non- linear signals. In Wavelet-based techniques, feature extraction is performed by utilizing Wavelet packet decomposition and dynamic Wavelet fingerprint [17]. This method involves the selection of the appropriate base wavelet functions. In [18], the Wigner and Choi-Williams distribution based radar waveform