Joint Design of Transmit Waveform and Receive Filter for Suppression of Mainlobe Deceptive Jammer via Deep Transfer Siamese Network

With advances in electronic countermeasure technology, active deception jamming leveraging digital radio frequency memory (DRFM) technology poses an unprecedented threat to radar systems. Specifically, mainlobe deception jamming, which is difficult to handle by traditional radar signal processing methods (such as space-time matched filtering, etc.), has emerged as a challenge. Recently, deep learning has been effectively applied to radar target detection, recognition, waveform formulation, and other fields. However, utilizing deep learning algorithms to suppress mainlobe deception interference under limited or zero samples remains a challenge. Therefore, this study proposes a radar mainlobe jamming suppression method predicated on a deep transfer Siamese network (DTN) that leveraging the concept of transfer learning and the Siamese network. The basic concept of this method involves generating transmit waveforms and receive filters in sequences collaboratively by creating multiple pre-training networks adhering to the same architecture. To overcome the issue of zero samples, network training is conducted through unsupervised learning with a custom loss function. This loss function integrates distinct weighting metrics for the autocorrelation and crosscorrelation characteristics of the synthesized waveform and interference filtering. Simulation experiments verify that the proposed method achieves superior performance in mainlobe interference suppression and improved model generalization compared to several existing methods.