LPI radar waveform recognition model based on multiple feature image and quasi-residual attention module

Abstract

Detecting and classifying the modulation type of the intercepted noisy LPI radar waveform has become a hot topic in the field of Electronic Countermeasures (ECM). In this paper, we propose a recognition model based on multiple feature images (MFI) and convolutional neural networks (CNN) with the quasi-residual attention module (QRAM). The core technologies of this model are divided into two parts. One is the MFI that combines the ambiguity function image (AFI) of the complete signal, covariance matrix image (CMI) of the reconstructed signal, and short-time Fourier transform image (STFTI) of the truncated signal into RGB color image in signal processing as the input for recognition network. The other is CNN with a QRAM recognition network, in which the input is sliced into two mappings called identity mapping and residual mapping to build the quasi-residual module, and the attention module is embedded in each mapping to denoise and enhance the feature. The performance of the MFI-QRAM model is demonstrated by recognizing 12 modulation types of the signals defined in this paper. The simulation experiments show that the model has strong robustness with the number of train sets and the variation of SNR. The overall probability of successful recognition (PSR) is 96.17% when the SNR is −6 dB.