A Fast and Accurate Convolutional Neural Network for LPI Radar Waveform Recognition
Do-Hyun Park, Jong-Hyeon Bang, Jihun Park, Hyoung-Nam Kim · 2022 19th European Radar Conference (EuRAD) · 2022
Recognizing low-probability-of-intercept (LPI) radar waveforms is essential in modern electronic warfare support systems. Recently, there has been a growing body of research about a convolutional neural network (CNN) to classify modulation schemes of the intercepted LPI radar signal. This paper proposes a CNN model composed of main and sub classifier for recognizing the modulation scheme of the intercepted radar signal. Unlike the existing waveform recognition methods that uses a single CNN classifier, our proposed model exploits a two-step classification technique for effective recognition by considering the modulation scheme recognition complexity. Specifically, the proposed recognition model predicts a modulation scheme through the main classifier in the first step. When the main classifier regards modulation scheme as indistinguishable, the proposed one additionally utilize the sub classifier. Simulation results show that the proposed recognition model has a high classification accuracy while having a low overall computational cost.