Specific Emitter Identification via Feature Extraction in Hilbert-Huang Transform Domain

Abstract

Aimed at the deficiency of conventional parameter-level methods in radar specific emitter identification (SEI), which heavily rely on empirical experience and cannot adapt to the waveform change, a novel algorithm is proposed to extract specific features and identify in Hilbert-Huang transform domain. Firstly, 2-dimensional physical representation of emitters is formed with Hilbert-Huang transform (HHT). Based on this, 4 types of multi-view features are constructed, and the feature space is spanned by elaborating the extraction. Principal components, between-class similarity, spectrum entropy, and deep architecture are used to describe the subtle features. Finally, support vector machine (SVM) is selected as the classifier to realize identification to alleviate the small sample problem. Experimental results show that the proposed algorithm realizes specific identification using 4 intentional modulations of simulated data. The selected 4 types of unintentional representations are feasible to discriminate identical emitters. Additionally, the proposed algorithm obtains higher accuracy than typical signal-level methods in the signal-to-noise ratio (SNR) range [0, 20] dB.

Transfer Learning-Based Specific Emitter Identification for ADS-B over Satellite System

“…Extracting this feature requires no synchronization or prior information and is of low complexity [18,19]. In [20], specific features were extracted and recognized in the Hilbert-Huang transform domain, and a support vector machine was utilized as the classifier to address the limitations of traditional methods, which relied on expertise and struggled to adapt to waveform variations. A feature extraction method based on bispectrum transform (BST) was proposed in [21][22][23][24].…”

Open space radar specific emitter identification using MSAK‐CNN‐LSTM network

“…They believe that due to the differences in signal styles and transmitters, different preprocessing methods may be suitable for different scenarios and models, and the influence of various preprocessing methods on results should be considered during experiments. Currently, the mainstream preprocessing methods for SEI can be classified into three categories: signal state detection [1], representation transformation [11][12][13][14][15], and signal decomposition [8]. Signal state detection primarily processes the signal in the time domain.…”

“…Before the advent of deep learning, these two steps were fundamental in addressing SEI problems. Currently, the existing literature primarily focuses on feature extraction from the time domain [1], frequency domain [2,3], and transform domain [4,5], and employs classifiers such as k-nearest neighbour [3,6], support vector machine [7][8][9], and extreme learning machine [10]. These methods proficiently manage high-dimensional feature spaces and demonstrate strong non-linear radar emitter recognition.…”