A Specific Emitter Identification Approach Based on Multi-Head Attention Mechanism

Specific emitter identification (SEI) is the process of identifying a unique emitter based on the internal features carried by the received signal, which has been widely used in spectrum management. In this paper, we consider the influence of modulation mode and power amplifier (PA) on radio frequency fingerprint (RFF) and propose an identification scheme based on squeeze-and-excitation network (SENet) and multi-head attention mechanism (MHA). The proposed scheme can use the abundant information of signal nonlinear characteristics for identifying emitters. Different from most existing SEI schemes based on feature transformation, our scheme can learn the feature from time domain signal, which can extract the overall characteristics hidden in the received signal. We evaluate the proposed algorithm through extensive simulation experiments, discuss the impact of frequency offset, the number of emitters, and channel conditions on the identification performance, and compare the proposed scheme with several advanced deep learning network architectures to prove its effectiveness.