Communication Radiation Source Recognition Based On Semi-supervised Learning

As one of the core technologies of electronic reconnaissance system,individual emitter identification technology plays an important role in the field of electronic countermeasures.It refers to the professional receiving device to receive and the use of digital signal processing technology transformation of target signal from the source,and then extract the characterization of the characteristics of its identity(" fingerprint" feature),and compare with existing fingerprint characteristic database,and to determine the unknown source category a technology and related information.As the electromagnetic environment in the real scene becomes more and more complicated,the requirements for the radiation source identification performance and application scene in the real environment become more and more strict.If the traditional method of radiation source identification is adopted,it is time-consuming and laborious to design a specific algorithm for a specific radiation source,which obviously cannot meet the practical requirements.With the rapid development of machine learning in recent years,there are more and more examples of applying deep learning to individual identification of radiation sources,which has achieved good results so far.In order to improve the efficiency of radiation source identification,this paper comprehensively uses the method of deep learning and semi-supervised learning to carry out individual identification of radiation sources.The main work of this paper is as follows:1.In order to improve the accuracy and universality of fingerprint identification of known and unknown classes in the actual environment,a semi-supervised deep learning method for individual identification of radiation sources was proposed based on deep learning and in view of IQ distortion data generated by Matlab simulation.The combined architecture of neural network model,anomaly detection model,and Kmeans clustering model is built to realize the recognition of known and unknown classes of emitter signals.The simulation results show that the recognition rate of the two methods in this paper is above 96% for 10 kinds of distorted data,and the known class and the unknown class can be distinguished accurately.2.To test the performance of the proposed method in practical engineering application,in Chengdu,the airspace collected 260 class aircraft ads-b data,using neural network to extract fingerprint characteristics and predict,to join the fake aircraft serial number,using the center loss to detect anomaly detection model,judge whether the abnormal loss value view center.The experimental results show that the identification accuracy of the proposed method is about 95%,and the identification rate of the counterfeit signal intrusion detection is 93%,which has certain practical application value.