Radar Emitter Sorting Based on One-Dimensional Convolutional Neural Network

Radar emitter sorting is a research hotspot in the field of modern electromagnetic countermeasures. However, with the increasingly complex electromagnetic environment in the battlespace and the development of radar anti-jamming technology, the traditional model-driven radar emitter sorting method is facing serious challenges. To solve the problem of radar emitter sorting with unknown parameters and numbers under the condition of high missing and spurious pulse ratios, a radar emitter sorting method based on one-dimensional convolutional neural network (1D-CNN) is proposed in this paper, which uses the time of arrival (TOA) as the sorting parameter and realizes the feature extraction and recognition of the pulse repetition interval (PRI) patterns through one-dimensional convolution operation layer by layer. For multi-emitter sorting, the cyclic iteration mechanism is introduced to sort the target pulses out in turn, and the best matching principle is further used to calculate the difference between the output sequence and the multi-target to realize parameter optimization and network training. The simulation results show that the proposed emitter sorting method based on 1D-CNN has strong robustness to overlapping, missing, and spurious pulses, and achieves better performance than the traditional histogram-based method.