Convolutional Transformer for Radar Active Jamming Classification
Hairui Zhu, Shanhong Guo, Weixing Sheng · 2023
Accurate classification of radar jamming signals can guide the application of effective anti-jamming policies to protect functions of the radar. In this paper, a novel method for radar active jamming classification based on time and range-Doppler domain is proposed. We design a lightweight convolutional Transformer network to process the data from two domains. Efficient convolutional blocks are employed to extract features from different domains. Then, the range gate tokenizer is proposed to operate a feature fusion and generate a sequence of tokens. Two Transformer encoders are responsible for further feature extraction and outputting final predictions. Simulation experiments show the proposed method has a high performance with 94% accuracy requiring only 42 ms latency on a Raspberry pie.