Deinterleaving Pulse Trains via RPMA-TConv for Parameter-Agile Radar

• Published in: 2024 IEEE Radar Conference (RadarConf24) pp. 1 - 6
• Main Authors: Liang, Caiyu, Tang, Han, Huo, Weibo, Zhang, Yin, Pei, Jifang, Huang, Yulin
• Format: Conference Proceeding
• Language: English
• Published: IEEE 06.05.2024
• Subjects: Convolution; Deep learning; Deinterleaving; Feature extraction; Measurement errors; Multi-branch atrous convolution; Noise; Radar; Reconnaissance; Time-frequency analysis
• ISSN: 2375-5318
• DOI: 10.1109/RadarConf2458775.2024.10548661

Radar pulse trains deinterleaving is a challenging task in modern electronic reconnaissance. The RPMA-TConv model based on multi-branch atrous convolution and feature reconstruction is proposed to solve the problem of deinterleaving parameter-agile emitters. The time of arrival (TOA), center frequency (CF) and pulse width (PW) are used to characterize the relative position and variation pattern of pulse trains. Multi-branch atrous convolutions with different receptive fields are applied to extract multi-scale temporal patterns, ensuring a comprehensive representation of agile parameter characteristics. The feature reconstruction module reconstructs the pulse train information through a learnable process and attributes each pulse to the corresponding emitter. The proposed method can correlate multiple modes of pulse trains generated by the same parameter-agile emitter. Compared with the traditional methods, it will not cause the problem of more clusters than emitters. The method also performs well in scenarios with parameter overlap and noise. Experimental results and performance analysis based on interleaved parameter-agile pulse trains are provided to demonstrate the effectiveness and robustness of the method.