Adaptive Transform-Domain Framework for Noise-Limited Radar Modulation Recognition

Recognizing phase-coded radar waveforms at low visibility is often handled through learning-based classifiers, but such methods require training data and tend to lose reliability when the noise statistics or operating conditions change. This work follows a different path and develops a deterministic algorithm that operates directly on the received pulse. A single carrier-bin trace is extracted from either the Short-Time Fourier Transform (STFT) or the wavelet transform, and two complementary descriptors are formed from its phase and magnitude. These descriptors are combined through an SNR-dependent weight selected by a simple margin-based optimization that does not rely on learned models or prior assumptions. Using four representative waveform families, the STFT version already offers clear gains over phase-only and magnitude-only decisions, while the wavelet front-end provides sharper chip localization and stronger separation between classes, especially at the lowest SNR values. The results show consistent improvements in accuracy, decision margins, and confusion matrices, and the overall procedure remains lightweight enough for real-time or embedded receivers.