A Multiclass Time-Series Signal Recognition Method Based on a Large Active Radar Jamming Database

Abstract

Active radar jamming recognition is a crucial technology in electronic countermeasures (ECMs). To tackle the challenge of intelligent recognition of complex radar jamming signals, we introduce the large-scale active radar jamming database (LARJD). This comprehensive database includes 19 distinct types of jamming signals and contains a total of 66500 time-series samples across five radar frequency bands, providing a robust dataset for radar jamming signal recognition. In parallel, we propose the multiclass time-series signal recognition network (MTSS-2DCNN), a deep learning architecture designed for classifying multiple types of time-series signals, including radar jamming signals. The MTSS-2DCNN architecture comprises three 2-D convolutional neural networks (2DCNNs), which extract features from both the time- and frequency-domain representations of the time-series data. By using a 2-D network structure to process 1-D signals, MTSS-2DCNN captures high-dimensional features from sequential signals while preserving the inherent characteristics of the temporal data. The model’s generalization capability is further enhanced through K-fold cross-validation and an adaptive learning rate adjustment strategy. Experimental results demonstrate that the proposed method achieves an impressive accuracy of over 99.67% on the LARJD, with significantly shorter training times compared to existing approaches. Moreover, by pretraining radar jamming signal recognition models, ECM applications can substantially improve the efficiency of intelligent recognition systems in both engineering and military contexts.