A Mask-Assisted Radar Signal Sorting Method Based on Digitized PDW and U1DADM

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Traditional radar signal sorting methods struggle with complex electromagnetic environments, especially signals sharing identical modulation types and overlapping parameter ranges.
• Accurate radar signal identification and sorting are crucial for electronic warfare and situational awareness.

Purpose of the Study:

• To develop an advanced deep learning model for high-precision radar signal sorting in challenging environments.
• To overcome the limitations of existing methods in distinguishing similar radar signals.

Main Methods:

• A one-dimensional convolutional neural network (1D-CNN) named U1DADM, based on semantic segmentation, was proposed.
• The network employs adaptive dilated convolution for multi-scale feature fusion and dynamic receptive field adjustment.
• A masking strategy was introduced to handle overlapping pulse regions and mitigate data defects.

Main Results:

• The U1DADM network demonstrated superior performance compared to traditional and other deep learning methods in radar signal sorting.
• The method achieved high-precision sorting even under conditions with spurious and missed pulses.
• Experimental results confirmed the model's strong robustness and generalization capabilities.

Conclusions:

• The proposed U1DADM network offers a significant advancement in radar signal sorting technology.
• The approach effectively addresses the challenge of sorting complex radar signals with identical modulation types.
• The method provides a robust and accurate solution for modern electronic warfare applications.