一种嵌入射频指纹的半监督辐射源识别方法

Semi-supervised Specific Emitter Identification Method with Embedded RF Fingerprint

对于非协作通信场景下辐射源识别(SEI)问题,基于人工射频指纹特征(Radio Frequency Fingerprints,RFF)的识别方式准确率不高,基于深度学习的方法又对训练数据量有过高的要求。为了克服该问题,提出一种结合了人工射频指纹特征的基于贝叶斯卷积神经网络(CNN)的半监督SEI算法,将一个回归拟合信号双谱的直方图特征的CNN嵌入一个SEI的贝叶斯CNN中,并通过基于模糊度的半监督学习方法进一步降低算法对标签训练集的依赖性。在模拟数据集和真实数据集中的实验结果表明,在标签训练集规模为500~4 500条数据时,提出的方法比端到端的卷积神经网络识别方法的识别率提高了5%~20%。

For specific emitter identification problem in the scenario of non-cooperative communication,the accuracy of the identification method based on the artificial radio frequency fingerprints(RFF)feature is not high enough,and the requirements for the training data size of deep learning based method are too high.In order to overcome this problem,a new type of semi-supervised emitter identification algorithm based on Bayesian CNN combined with artificial RFF feature is proposed in this paper,the CNN of a histogram feature of the regression fitting signal bispectrum is embedded in a Bayesian CNN identified by the emitter,and the dependence of the algorithm on the label training set is further reduced by the fuzziness-based semi-supervised learning method.The experimental results in the simulated dataset and the real dataset show that the identification rate of the proposed method is 5%~20%higher than that of the end-to-end convolutional neural network identification method when the size of the label training set is 500~4 500.