A Radar Signal Open-Set Deinterleaving Method Based on Laplacian Pyramid Reconstruction and Adversarial Reciprocal Point Learning

Radar signal deinterleaving is a challenging task in complex electromagnetic environments. This work proposes a method for accurately deinterleaving known and unknown radar radiation sources in an open space using an open-set deinterleaving technique for radar signals based on the combination of Laplacian pyramid reconstruction and reciprocal point adversarial deinterleaving (PR-RPAD). The pulse description graph (PDG) representation of the intercepted pulse description word is first realized by applying a gray matrix symmetric mapping approach based on sliding windows. Second, to improve the visual characterization of the pulse description information of various radar radiation sources, PDG edges and texture structures are amplified and enhanced using the Otsu-threshold-based image amplification technique. Next, a Laplacian-pyramid-based multiresolution feature reconstruction and fusion model is put forth, which uses a jump connection and a multiplication gate to achieve tower reshaping from low-resolution to high-resolution features. The idea of reciprocal points is finally introduced to model the open space to decrease the risk of closed-set deinterleaving through confrontation with known radars. To improve the discriminatory nature of the model against unknown radars and to finish the accurate deinterleaving of radar radiation sources in the open space, an instantiated confrontation enhancement method is used to generate confusing samples for training. In addition to having better performance in closed-set deinterleaving than other deinterleaving techniques (sequential difference histogram (SDIF), pulse repetition interval transform algorithm (PRI-Tran), bidirectional long short-term memory (BLSTM), bidirectional gated recurrent unit (BGRU), and dilated convolutional network (DCN)), the PR-RPAD method also achieves deinterleaving in the open space, which offers great versatility and potential for real-world use in intricate electromagnetic environments.