Radar Signal Pulse Train Recognition With Dual-Branch LSTM-Transformer Networks

Radar signal pulse train recognition plays a crucial role in modern electronic warfare, serving as the foundation for further analysis of radar signal sources and operating modes. Currently, most studies rely on PRI modulation features as the primary basis for recognition. However, with the continuous development of radar technology, the frequency parameters of radar pulses have become increasingly dynamic and complex; nevertheless, this aspect has not received adequate attention over the years. This paper proposes a signal pulse train recognition method based on a Dual-branch LSTM-Transformer architecture, introducing improvements in both input design and network structure. In terms of architecture, a dual-branch structure is constructed, with each branch employing an LSTM-Transformer network that effectively combines the strengths of both models. For the input, the two branches respectively receive the normalized PRI sequences, frequency sequences, and their corresponding difference sequences. Additionally, a gating mechanism is introduced during the feature fusion stage to adaptively allocate feature weights, enabling more accurate and efficient radar pulse train recognition. Simulation results show that this method achieves a recognition rate exceeding 90% even in high-noise environments, demonstrating strong robustness.