A Radar Signal Deinterleaving Method Based on Multiscale With Attention Mechanism

Deinterleaving radar signals is a significant task in contemporary electronic warfare reconnaissance. However, as adaptive waveforms and multifunctional radar technologies evolve, the dynamic range of signal parameters broadens and modulation styles diversify, resulting in critical performance bottlenecks for single-feature deinterleaving algorithms. Therefore, this article explores a novel intelligent deinterleaving paradigm that fuses multidimensional signal properties. A radar signal deinterleaving method based on the multiscale attention mechanism, Multiscale attention deinterleaving (MSAD), is proposed to address issues like how to depict the multidomain coupling characteristics of radar signals, how to model the contribution differences of features at different scales, and how to improve the generalization ability of algorithms for complex modulated signals. The approach first expands the dimensionality of the pulse description word (PDW) data, which is then converted using a Gramian angular field (GAF) into a pulse description graph (PDG). This allows for the joint graphical description of multidimensional characteristics that include time, frequency, space, and energy. Next, build a Laplace Pyramid multiscale feature extraction framework and employ deep convolutional networks (DCNs) to hierarchically capture signal patterns at various granularities. Last, a physically interpretable deinterleaving decision is created by dynamically fusing the feature weights of each scale using the attention mechanism (AM). According to experiments, the MSAD method outperforms the current approaches [bidirectional long short-term memory (BLSTM), bidirectional gated recurrent unit (BGRU), DCN, sequential difference histogram (SDIF), and pulse repetition interval transform (PRI-Tran)] in deinterleaving by taking advantage of the enhancement effect of multiscale image representations on electromagnetic signal features and the dynamic weight assignment by the AM. In addition, in multifunctional radar and jittered PRI deinterleaving, the MSAD approach demonstrates competitive performance gains.