Radar Waveform Recognition Based on Time and Phase Features Convolutional Network

Low probability of intercept (LPI) radars, distinguished by their superior anti-interception performance, are extensively used. LPI Radar Waveform Recognition (RWR) is vital for electronic support systems. However, the contemporary deep learning (DL) based RWR is afflicted with high computational complexity and expansive network parameters. To meet the demands for precise and swift recognition of LPI radar waveforms, this paper introduces a low-complexity, lightweight model for RWR. This proposed method deviates from traditional approaches that use time-frequency transformation, instead utilizing a convolutional neural network to directly extract phase and short-time features from the raw signal. Thirteen distinct LPI RWR experiments validate the effectiveness of our proposed method, achieving over 90% accuracy even at an SNR of -4dB, with minimal cost. Our approach requires only 12% of the parameters and 0.14% of the computational resources compared to the Wigner-Ville distribution plus DL method, while still achieving higher accuracy. Additionally, our model achieves comparable accuracy to complex-valued neural networks while utilizing only half the parameters and one-sixth of the computational load. It has also demonstrated a significant speed advantage in real-world hardware tests.