Scholarlink

<p>Automatic modulation classification (AMC) of proximity detector signals is essential for effective electronic countermeasures. However, due to signal distortion and information loss, AMC becomes very challenging under interrupted sampling (IS) conditions. To tackle this, a new AMC method using the Polynomial Chirplet Transform (PCT) and a Time–Frequency Reconstruction Network (TFRNet), referred to as PCT-TFRNet, is proposed. The PCT is used to preprocess the signal and enhance the separability of its time–frequency (TF) features. TFRNet is built with an asymmetric encoder-decoder structure and incorporates an adaptive random mask algorithm to reconstruct complete TF representations from inputs under IS modes. To further boost learning efficiency with limited samples, a self-supervised pretraining strategy is employed, followed by transfer learning on small-scale labeled IS samples. Experimental results show that the proposed method achieves high classification accuracy with limited samples. Specifically, PCT-TFRNet attains 89% accuracy when the signal-to-noise ratio (SNR) is at least −14&#xa0;dB, demonstrating strong robustness and generalization in low-SNR and small-sample scenarios. This confirms the effectiveness of the approach for AMC of IS proximity detector signals.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Automatic modulation classification for interrupted sampling proximity detector signals using time–frequency reconstruction network and polynomial chirplet transform

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年8月8日

查看全文
  • Guanghua Yi,
  • Xinhong Hao,
  • Xiaopeng Yan,
  • Dan Hu,
  • Yongzhou Wang,
  • Jian Dai

摘要

Automatic modulation classification (AMC) of proximity detector signals is essential for effective electronic countermeasures. However, due to signal distortion and information loss, AMC becomes very challenging under interrupted sampling (IS) conditions. To tackle this, a new AMC method using the Polynomial Chirplet Transform (PCT) and a Time–Frequency Reconstruction Network (TFRNet), referred to as PCT-TFRNet, is proposed. The PCT is used to preprocess the signal and enhance the separability of its time–frequency (TF) features. TFRNet is built with an asymmetric encoder-decoder structure and incorporates an adaptive random mask algorithm to reconstruct complete TF representations from inputs under IS modes. To further boost learning efficiency with limited samples, a self-supervised pretraining strategy is employed, followed by transfer learning on small-scale labeled IS samples. Experimental results show that the proposed method achieves high classification accuracy with limited samples. Specifically, PCT-TFRNet attains 89% accuracy when the signal-to-noise ratio (SNR) is at least −14 dB, demonstrating strong robustness and generalization in low-SNR and small-sample scenarios. This confirms the effectiveness of the approach for AMC of IS proximity detector signals.