A Unified Method for Deinterleaving and PRI Modulation Recognition of Radar Pulses Based on Deep Neural Networks

Abstract

In the modern electronic warfare signal environment, multiple radar signals of high density are mixed and received, and separating them into signals for each emitter is an essential step for emitter identification. Each radar has its own pulse repetition interval (PRI), which is a key parameter for deinterleaving pulse trains. The PRI is modulated in various forms depending on the purpose of the radar operation, and analyzing the mean PRI and the modulation type of PRI is the core of electronic warfare signal processing. Many existing papers have tried separate independent approaches for deinterleaving and for PRI modulation recognition. However, many distortions are unintentionally generated in the process of extracting the pulse train using the PRI estimated through deinterleaving for the PRI modulation recognition. This degrades the the modulation recognition performance. In this paper, we propose a unified method for the deinterleaving and PRI modulation recognition of radar pulses using deep learning-based multitasking learning. The simulation results demonstrate the good performance of the proposed method for deinterleaving and modulation recognition, compared to the conventional method, and prove that the proposed method is robust in noisy radar signal environments.

Change Point Detection for Fine-Grained MFR Work Modes with Multi-Head Attention-Based Bi-LSTM Network

“…To demonstrate the effectiveness in CPD, this section compared the performance of the proposed method with three methods: CNN [ 2 , 18 , 19 ], CNN-LSTM [ 2 ] and Bi-LSTM [ 20 , 21 , 31 ] as baselines. Ref.…”

“…[ 18 , 19 ], an Autoencoder was used to remove noise from the original pulse stream before identification. Several studies [ 20 , 21 , 22 ] have used Convolutional Neural Networks (CNNs) to recognize the modulation type of the Pulse Repetition Interval (PRI).Compared to traditional manual feature extraction, CNNs achieved significant improvements in recognition performance. These CNN-based methods improved MFR system performance and are robust to lost and spurious pulses but have the limitation of requiring inputs of fixed length.…”

Working Mode Recognition of Non-Specific Radar Based on ResNet-SVM Learning Framework

“…In [ 19 , 20 ], a clustering algorithm was used to help SVM extract features, but it could not identify complex signal styles. Since deep neural networks are widely used in radar recognition, the authors of [ 21 , 22 , 23 , 24 ] used a denoising auto-encoder (DAE), a convolutional neural network (CNN), a residual neural network (ResNet), and a recurrent neural network (RNN), respectively, to achieve their recognition rates of more than 90%, under the given conditions, which verified the advantages of the deep learning method in the field of signal recognition, but at the same time, it also exposed the weakness of a single network facing a complex environment. For the improvement of multi-objective tasks, the authors of [ 25 , 26 , 27 ] used the method of adding windows in the time domain, to process data, designed the time processing module and the threshold function of the selection window, and thus realized multi-objective classification.…”

Parametric model‐based deinterleaving of radar signals with non‐ideal observations via maximum likelihood solution

“…In Ref. [26], a unified method for the deinterleaving and PRI modulation recognition of radar pulses using DL‐based multitasking learning was proposed. The continuous wavelet transform was used to obtain images of the PRI over time, which were then used as input of the proposed convolutional neural network (CNN).…”

“…In the past few years, deep learning (DL) methods have been demonstrated for its applicability in many fields, such as computer vision and natural language processing. Recently, many methods based on deep neural networks have also been applied to deinterleaving [24][25][26]. In Ref.…”