Intra-Pulse Modulation Recognition of Radar Signals Based on Efficient Cross-Scale Aware Network

Area of Science:

• Electrical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Radar signal intra-pulse modulation recognition is crucial for radar systems.
• Existing convolutional neural networks (CNNs) face challenges with high computational complexity and poor performance in low signal-to-noise ratio (SNR) conditions.

Purpose of the Study:

• To propose a lightweight CNN, the Cross-Scale Aware Network (CSANet), for efficient and accurate radar signal intra-pulse modulation recognition.
• To enhance recognition performance, especially in low-SNR environments.

Main Methods:

• Development of the Cross-Scale Aware (CSA) module, featuring a depthwise dilated convolution group (DDConv Group), cross-channel interaction (CCI), and spatial information focus (SIF).
• Creation of a novel time-frequency fusion (TFF) feature by integrating three types of time-frequency images (TFIs) using adaptive binarization, morphological processing, and feature fusion.
• Implementation of CSANet utilizing the TFF feature for intra-pulse modulation recognition.

Main Results:

• CSANet with the proposed TFF achieved higher accuracy compared to other TFIs.
• CSANet demonstrated superior performance against state-of-the-art networks across twelve radar signal datasets.
• The proposed method proved effective for high-precision recognition in low-SNR scenarios.

Conclusions:

• CSANet offers an efficient and accurate solution for radar signal intra-pulse modulation recognition.
• The novel TFF feature significantly improves recognition performance, particularly under low-SNR conditions.
• The lightweight design of CSANet makes it suitable for practical radar applications requiring high precision.