Classification Analysis of Simple Pulsed Radar Signal using Adaptive Short-Time Fourier Transform and Support Vector Machine

The classification of Simple Pulsed Radar Signals (SPRS) presents a significant challenge due to the presence of noise and the need for precise time-frequency representation. This study proposes an innovative approach combining Adaptive Short-Time Fourier Transform (ASTFT) based on a Log Cost function and Support Vector Machine (SVM) for effective SPRS classification. The ASTFT technique, evaluated through a detailed analysis in the presence of noise up to -10 dB, successfully separates signal characteristics in both time and frequency domains, as demonstrated by 2D contour and 3D mesh plots. The performance of various SVM kernels— polynomial, sigmoid, and radial basis function (RBF)—was rigorously tested. The results indicate that the sigmoid kernel achieves a minimum accuracy of 86% at -10 dB, reaching 100% from -8 dB onwards. The polynomial kernel maintains a 97% accuracy at -9 dB and 100% from -7 dB onwards, while the RBF kernel shows a 97% accuracy at -8 dB, achieving 100% at -6 dB. Notably, all kernels achieved SPRS classification accuracy before reaching 0 dB. This study underscores the robustness and effectiveness of combining ASTFT with SVM for radar signal classification, offering a reliable solution for signal analysis in noisy environments.