JMRPE‐Net: Joint modulation recognition and parameter estimation of cognitive radar signals with a deep multitask network

Abstract

The newly developed cognitive radar (CR) can implement flexible work modes defined with a set of mode definition parameters. Each definition parameter can employ its modulation type and corresponding optimised modulating values. Automatic recognition and analysis of CR work mode are significant challenges for electromagnetic reconnaissance applications. In this article, a deep multitask neural network is proposed for Joint automatic Modulation Recognition and modulation Parameter Estimation (JMRPE-Net) for the emerging task of CR signals analysis. The proposed JMRPE-Net consists of a fork-shaped architecture in which three cascaded convolutional layers act as a shared module for the extraction of common features followed by multiple branches of long short-term memory layers with the attention mechanism for task-specific features extraction. The proposed network can receive a sequence of CR pulse signals as input and parallelly perform automatic modulation recognition (AMR) and modulation parameter estimation tasks for multiple work mode definition parameters. Extensive experiments are conducted based on simulated radar intermediate frequency signals with consideration of imperfections of real-world electromagnetic environments. The experimental results validate the superiority of the proposed JMRPE-Net against the existing state-of-the-art single task AMR methods or parameter estimation methods.This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made.

MIDW‐Net: A multi‐tasking network architecture for radar intra‐pulse parameter description

“…When the signals had different SNRs, the algorithm consistently exhibited a higher recognition accuracy than the other algorithms, as shown in Figure 5. For the analysis of radar signals in the range of –10–0 dB SNR, it achieved a modulation recognition accuracy of >95% at an SNR of −4 dB, outperforming other algorithms in the same category with regard to the parameter estimation error [1, 9].…”

“…Extensive research has been conducted on automatic modulation recognition (AMR) algorithms for conventional radar reconnaissance systems [1][2][3]. However, more information about the intra-pulse modulation parameters is needed to analyze the cognitive radar system optimization process [1].…”

“…Cognitive radar has considerable flexibility in the selection of intra-pulse modulation methods and thus has tremendous potential for electronic reconnaissance. Extensive research has been conducted on automatic modulation recognition (AMR) algorithms for conventional radar reconnaissance systems [1][2][3]. However, more information about the intra-pulse modulation parameters is needed to analyze the cognitive radar system optimization process [1].…”

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

“…To overcome this limitation, in Refs. [ 23 , 24 , 25 ], Recurrent Neural Networks (RNNs), including their improved form, i.e., The Long Short-Term Memory (LSTM) network, were utilized to classify pulse sequences of different work modes. Moreover, RNNs introduced the concept of timing into network architecture design to achieve better adaptability in time series data analysis.…”

Multi‐label hybrid radar signal recognition based on a feature pyramid network and class activation mapping

“…Deep learning has been applied to the field of radar signal recognition [9][10][11]. However, designing and training a deep convolutional neural network (CNN) from scratch requires sufficient hardware resources and a large amount of training time.…”