A Multi-View Attention Hypergraph Neural Network for Radar Emitter Signal Sorting

Abstract

In complex electromagnetic environments, the deinterleaving of dense, interwoven radar pulse signals poses a formidable challenge. To address the propensity of conventional graph models for confusion and misclassification in such scenarios, this letter proposes a novel method for radar emitter signal deinterleaving: the Multi-view Attention Hypergraph Neural Network (MVA-HGNN). This model maps radar pulses to nodes in a hypergraph, leveraging hyperedges to capture higher-order correlations among pulses. This overcomes the limitation of traditional graphs, which can only describe pairwise relationships. To fully exploit the heterogeneous information within Pulse Descriptor Words (PDWs), we construct two distinct hypergraph views: “spatial” and “intrinsic.” In the MVA - HGNN model, parallel hypergraph network branches learn node representations from different views. An advanced attention - based fusion mechanism is introduced to dynamically integrate these feature representations. Simulation results demonstrate that our method achieves superior performance, particularly in small data scenarios, showing great potential for engineering applications.