Radar PRI Modulation Recognition Under Non-ideal Observations via Joint Encoding of Gramian Angular Field

Pulse repetition interval (PRI) is a critical parameter for identifying radar emitters. However, modern agile radar systems often employ complex PRI modulation patterns, posing significant challenges for accurate recognition. Existing methods may fail to maintain robust performance in non-ideal electronic environments with high ratios of pulse loss and spurious pulses. Since different PRI modulation types have special periodicity of time sequences, this paper proposes a domain-adapted radar PRI modulation recognition framework via Gramian angular summation field (GASF) and Gramian angular difference field (GADF) encoding. The combination of harmonic elimination and Polar coordinate mapping effectively preserves temporal correlations and structural patterns during PRI sequence transformation. Then, a dual-branch feature extraction and fusion strategy is designed based on the complementary characteristics of GASF and GADF encodings. Through mid-level feature merging, the intermediate fusion architecture could well fit into practical variable-length PRI sequences and effectively learns discriminate features. Finally, extensive experiments demonstrate that the proposed method achieves satisfying recognition accuracy and maintains stable performance even under extreme non-ideal conditions. Particularly, the recognition accuracy achieves over 96% under 50% pulse loss ratio. Tests on real-measured data further evaluate the method's effectiveness in real-world radar signal scenarios.