Specific Emitter Identification Based on Differential Constellation

Abstract

The modulator distortion feature of the existing specific emitter identification (SEI) based on constellation contains the inherent carrier frequency offset of the receiver and transmitter, and the influence of the frequency offset on the feature distribution cannot be completely eliminated even by using high-precision frequency synchronization technology. Therefore, this paper presents a novel SEI method based on a differential constellation. On the basis of constructing the modulator distortion signal model, the demodulated signal is differentially processed to form a differential constellation. By comprehensively comparing the difference between the differential demodulation constellation and the ideal constellation, the maximum likelihood method is used to separate the frequency offset in the baseband signal from the modulator distortion feature vector. A new modulator distortion feature representation is designed, which completely eliminates the influence of the carrier frequency offset on the distortion feature distribution of the modulator. Subsequently, a random forest classifier based on a decision tree was constructed to learn the individual differences in the distortion features. Compared with existing identification methods, the fingerprint features extracted by this method are completely independent of the frequency offset of the signal examples, and the influence of the frequency offset on the recognition is eliminated. Our results show that the mean and variance of the feature vector distribution proposed in the method do not change with the frequency offset, the stable and high-precision identification of eight sources can be achieved under different carrier frequency offset conditions, and the accuracy can reach more than 90%.

Keywords

• Differential constellation
• modulator distortion
• radio-frequency fingerprinting (RFF)
• random forest
• specific emitter identification (SEI)