Research On Individual Identification Technology Of Communication Radiator

Radiation source identification technology is one of the key technologies in the field of communication reconnaissance and electronic countermeasures,and it is also a frontier direction in the field of electronic warfare at home and abroad.It identifies individual radiation source equipment by extracting features from the captured electromagnetic signals.In the civil field,radiation source identification plays an important role in spectrum management and communication signal identification.Facing the increasingly complex and changeable electromagnetic environment,traditional radiation source identification methods have more and more limitations.Electromagnetic signals often have multiple working modes,which require high prior knowledge of the signal,and it is difficult to extract individual characteristics.The algorithm takes a long time,has high time complexity,and is not universal.And in actual application scenarios,there are usually new communication individuals appearing,resulting in the need to retrain the entire model.In response to the above problems,through the advantages of deep learning algorithms,researches are carried out from the aspects of radiation source identification technology identification system design,parameter optimization and unknown radiation source identification.The specific work includes the following aspects:1.This thesis designs a neural network recognition algorithm based on deep learning time domain IQ path features for scenarios where there are multiple working modes of the radiator signal.The time domain IQ path map features are used as input samples.The algorithm uses deep residuals.The difference network structure realizes the extraction of the subtle features of the time domain IQ signal.The deep residual network structure can automatically extract the individual signal characteristics and the transmitter nonlinear characteristics,and compare it with other recognition algorithms for feature extraction.When the radiation source signal When the individual working mode becomes more and more complicated,it also has a high recognition accuracy,which proves the effectiveness and robustness of the algorithm in this thesis.Aiming at the network parameter optimization problem of deep learning,an algorithm for adaptively adjusting the learning rate based on the loss function is designed.The algorithm uses the current loss value to adjust the learning rate growth factor,reducing manual tuning,and solving the network model because of the learning rate.Parameter settings lead to slow training convergence and time-consuming training.The algorithm designed in this thesis improves the recognition performance of the entire system,accelerates the model convergence speed,and has high robustness.2.In view of the fact that in actual scenarios,new radiator individuals often appear,this thesis builds a recognition system based on online learning and migration learning,realizes online recognition and detection of unknown individual samples,and then uses the fine-tuning in migration learning The method updates the model,avoiding manual label addition and model retraining.The online migration learning recognition algorithm can dynamically and continuously identify different unknown radiation source individuals.Compared with the traditional algorithm,it can adapt to different radiation source data,which proves The effectiveness of the recognition system designed in this thesis is analyzed.