A Radar Signal Deinterleaving Method Based on Multiscale With Attention Mechanism

Area of Science:

• Electronic Warfare and Signal Processing
• Artificial Intelligence in Radar Systems

Background:

• Deinterleaving radar signals is crucial for electronic warfare reconnaissance.
• Current single-feature algorithms face performance bottlenecks due to evolving adaptive waveforms and diverse modulation styles in modern radar systems.
• There's a need for advanced deinterleaving methods that can handle complex, multidimensional radar signal characteristics.

Purpose of the Study:

• To propose a novel intelligent deinterleaving paradigm, Multiscale Attention Deinterleaving (MSAD), for radar signals.
• To address the challenges of depicting multidomain coupling characteristics, modeling feature contribution differences across scales, and improving generalization for complex modulated signals.
• To develop a method that effectively fuses multidimensional signal properties for enhanced deinterleaving accuracy.

Main Methods:

• Expanded Pulse Description Word (PDW) data converted into a Pulse Description Graph (PDG) using Gramian Angular Fields (GAF) for joint graphical description of time, frequency, space, and energy.
• Implemented a Laplace Pyramid multiscale feature extraction framework with Deep Convolutional Networks (DCNs) to capture hierarchical signal patterns.
• Employed an Attention Mechanism (AM) to dynamically fuse feature weights from different scales for interpretable deinterleaving decisions.

Main Results:

• The MSAD method significantly outperforms existing algorithms like BLSTM, BGRU, DCN, SDIF, and PRI-Tran in radar signal deinterleaving.
• Demonstrated superior performance by leveraging multiscale image representations and dynamic attention-based feature weighting.
• Achieved competitive performance gains in challenging scenarios, including multifunctional radar and jittered Pulse Repetition Interval (PRI) deinterleaving.

Conclusions:

• The proposed MSAD method offers a robust and effective solution for radar signal deinterleaving, particularly for complex and evolving signal environments.
• The fusion of multidimensional signal properties and multiscale feature extraction with attention mechanisms provides significant advantages over traditional methods.
• MSAD shows strong potential for practical applications in contemporary electronic warfare reconnaissance.