Investigating the effects of destructive factors on pulse repetition interval modulation type recognition using deep convolutional neural networks based on transfer learning

Abstract

Automation and self-sufficiency in the complex environment of modern electronic warfare (EW) are critical and necessary issues in electronic intelligence and support systems to detect real-time and accurate threat radars. The task of these systems is to search, discover, analyse, and identify the parameters of radar signals. However, recognition pulse repetition interval (PRI) modulation is challenging in natural environments due to destructive factors, including missing pulses (MP), spurious pulses (SP), and large outliers (LO) (caused by antenna scanning), which lead to noisy sequences of PRI variation patterns. The current article examines the effects of destructive factors on recognising PRI modulation in radar signals using deep convolutional neural networks (DCNNs). The article uses simulations based on the actual environment to generate data and consider destructive factors with different percentages. The number of images obtained by applying the sum of destructive factors for each range of destructive factors (with different percentages) considered is 30,000. It is common for six types of modulation. Then, the DCNN models, including VGG16, ResNet50V2, InceptionV3, Xception, and MobileNetV2, are trained using the transfer learning method. The simulation results show that the accuracy of training and testing the models decreases significantly with the increase in the percentage of destructive factors. Also, the effects of the model type on the performance of the models have been investigated, and the results have shown that some models are more resistant to destruction and retain more accuracy. Finally, this analysis shows that to improve the performance of deep neural network (DNN) techniques in the face of changes caused by destructive factors, it is necessary to pay attention to these factors and apply appropriate strategies.