A Radar Signal Sorting Method Based on the Graph Neural Network Combined with GCN and GAT

The electromagnetic environment of modern battlefields is becoming increasingly complex, leading to sever spatial overlap in the feature distribution of radar signal parameters, which in turn reduces the accuracy of radar signal sorting. To address this challenge, this paper proposes a novel radar signal sorting algorithm, GCN&GAT-RSS, which integrates graph convolutional networks (GCNs) and graph attention networks (GATs). This approach leverages the powerful feature extraction capabilities of graph-based deep learning models. A radar signal adjacency matrix is constructed based on subspace self-expressiveness property using the generalized orthogonal matching pursuit (gOMP) algorithm, enabling a comprehensive exploration of the spatial interconnections among radar signals. By combining the global smoothness of GCNs with the local adaptability of GATs, the proposed method effectively extracts meaningful spatial features for signal sorting. The proposed method demonstrates its effectiveness in addressing missorting issues caused by the overlap of complex radar working state parameters, achieving a radar signal sorting accuracy of 94.47% for multifunctional radar systems.