Deep Learning-Based Radar Signal Deinterleaving Using Temporal Convolutional Attention Network
Haojian Wang, Jin He · 2025
In the domain of electronic warfare, radar signal deinterleaving emerges as the foundational and indispensable phase of electronic reconnaissance. The ever-increasing complexity of electromagnetic environments has led to the inadequacy of traditional deinterleaving techniques. To tackle these challenges, this paper introduces the Temporal Convolutional Attention Network (TCAN) framework. This framework harmoniously combines a temporal convolutional network (TCN) with the attention mechanism, thereby significantly enhancing the system’s signal sorting proficiency. Through rigorous experimental validation, we demonstrate that TCAN consistently outperforms existing baseline methods. This superiority is particularly pronounced in non-ideal environments, which are typified by pulse loss, spurious pulses, and measurement errors. Furthermore, we conduct a thorough analysis to elucidate the profound impact of various input formats and multi-parameter features on deinterleaving performance. By meticulously examining these factors, we establish TCAN as a robust and versatile solution capable of effectively navigating the heightened complexity of modern radar signal environments. Our findings highlight TCAN’s potential as a potent instrument for augmenting electronic reconnaissance capabilities amidst evolving electromagnetic challenges.