Deep learning and recognition of radar jamming based on CNN

Abstract

The expert feature based radar jamming recognition methods are widely used today, however the appropriate features are hard to find and its generalization ability is not good which results in the difficulty of the classification for novel active jamming. In order to realize the smart classification of radar jamming, this paper proposes a method based on CNN (convolutional neural network) for deep learning and recognition of radar jamming. In the process of learning, the time-frequency maps of various jamming models are used to form the training dataset, and then the CNN is used to learn features from dataset. In the process of classification, firstly the jamming parameters are measured from the time frequency maps of the echo signal based on the OS-CFAR algorithm (ordered statistics constant false alarm rate), then the jamming signals are extracted accurately from the echo. Secondly, the time-frequency images of extracted jamming signal are obtained. Finally, the images are feed into the pretrained CNN model for classification. The simulation results show that the classification accuracy can reach 98.667% under 0dB~ 8dB JNR (jammer-to-noise ratio) for 9 typical jamming, and the generalization ability for the jamming with different parameters is good.