A Deep Learning-Based Model for the Automatic Recognition an‎d Classification of Intra-pulse Modulation of Radar Signals

10/22/2026 12:00:00 AM

Intra-pulse modulation recognition of modern Low Probability of Intercept (LPI) radar signals in noisy environments is a fundamental challenge for traditional methods. This research presents a novel an‎d computationally efficient deep learning model. In this approach, the signal is first transformed into a high-resolution, cross-term-suppressed Time-Frequency Representation (TFR) using the Smoothed Pseudo Wigner-Ville Distribution (SPWVD). These images are then fed into a lightweight hybrid architecture known as the Compact Convolutional Transformer (CCT). The proposed model, eva‎luated on 13 modulation types over a -10dB to +10dB SNR range, achieved an overall accuracy of 95.48% with only 210,350 parameters. The model demonstrated high robustness in severe noise, attaining 82.92% accuracy at -10dB SNR. The comparison results showed that the Compact Convolutional Transformer model, despite having a significantly fewer number of parameters, has a fully competitive performance with the well-known architectures ResNet50, MobileNetV2, an‎d ViT. This optimal trade-off between accuracy an‎d efficiency makes the proposed model an ideal can‎didate for implementation in real-time systems with limited hardware resources. Keywords: Intra-pulse Modulation Recognition, Radar Signal, Deep Learning, Compact Convolutional Transformer (CCT), Time-Frequency Representation (TFR), Smoothed Pseudo Wigner-Ville Distribution (SPWVD).

مدولاسيون درون‌پالسي , رادار

intra-pulse modulation , radar

Naser Ebrahimi

Adib Abrishamifar