Specific Emitter Identification Method Based on Aggregation Model with Metric Learning in Complex Field

In the field of Specific Emitter Identification (SEI), due to the time-varying characteristics of the ionosphere, the radio short-wave communication channel is a complex variable parameter channel, and there are multi-path fading, polarization rotation, Doppler shift and other phenomena, resulting in the short-wave radiation source individual fingerprint features are covered by the channel features, and the recognition algorithm based on a single machine learning model or deep learning model is often difficult to achieve good recognition results. To solve this problem, based on the idea of metric learning and ensemble learning, this paper proposes an aggregation model of metric complex value CNN (AM-MCVCNN) for SEI and tests it on 2G-ALE signal. We generate multiple metric learning models, which can be used for SEI, and use a sort cumulative aggregation method (SCAM) to generate aggregation model. The experimental results show that the recognition accuracy of AM-MCVCNN can reach 89%, compared with the single model, the recognition accuracy is improved by 10–15%. Testing AM-MCVCNN on the data after one week, AM-MCVCNN still has a high accuracy (89.56%). For the data after two weeks, the accuracy of AM-MCVCNN decreased slightly (81.23%). For data collected after four months, some individuals still have a high identification accuracy.