A Generative Adversarial Network Based Framework for Specific Emitter Characterization and Identification

Abstract

Specific emitter identification (SEI) enables the classification of various unique emitters based on received waveforms using some external feature measurements from their transmit signals and has shown its potential for military and civil applications. However, the characterization of the received waveform is susceptible to various factors in propagation process, resulting in inaccurate representations for the individual emitters, so the discriminative performances of existing methods are usually challenging. To remedy these shortcomings, this paper presents a novel semi-supervised SEI using generative adversarial networks (GAN). We mitigated a representation deep network into Triple-GAN and construct a quadruple-structured framework. The overall feature information hidden in the original signals can be extracted by representation network to improve identification performance while the classification results of Triple-GAN can in turn help the representer learning more discriminative characteristics of individual emitters. Results from both simulations and real world data experiments are provided. Numerical performances indicate our conclusion that our proposed framework constantly outperforms other existing schemes in terms of classification accuracy.