Specific Emitter Identification using Deep learning Models

Abstract

In today’s world, Specific emitter identification (SEI) has become crucial task in elec tronic warfare and signal intelligence that involves identifying a particular communication device by examining the unique radio frequency (RF) signals it emits. The ability to dis criminate between different emitters is essential for ensuring the security and efficiency of communication networks, spectrum management, and electronic warfare operations. Existing emitter recognition methods often ignore the radio frequency (RF) fingerprint details carried by the waveforms which are primarily susceptible to a specific application scenario and radio environment and thus can be interfered with by unreliable RF features. It has been found that deep learning methods have demonstrated effectiveness in this task. This research introduces an innovative approach for Specific Emitter Identification (SEI) utilizing a Savitzky-Golay filter for denoising and a Stacked Multivariate Convolu tional Neural Network (SMvCNN) architecture for classification. The inputs which are fed to Stacked Multivariate Convolutional Neural Network (SMvCNN) are time domain, frequency domain and phase of signals. By using this our proposed method surpasses conventional machine learning classifiers, achieving an impressive classification accuracy of 96% even under challenging conditions with a signal to noise ratio (SNR) of 5 dB. The integration of the Savitzky-Golay filter for noise reduction and the SMvCNN model demonstrates superior performance, underscoring its potential as a robust SEI technique in real-world scenarios.