Automatic modulation classification for interrupted sampling proximity detector signals using time-frequency...

Area of Science:

• Signal Processing
• Machine Learning
• Electronic Warfare

Background:

• Automatic modulation classification (AMC) is crucial for electronic countermeasures.
• Interrupted sampling (IS) conditions severely challenge AMC due to signal distortion and data loss.

Purpose of the Study:

• To propose a novel AMC method, PCT-TFRNet, robust to IS conditions and limited data.
• To enhance the accuracy and generalization of AMC for proximity detector signals under challenging environments.

Main Methods:

• Utilizing Polynomial Chirplet Transform (PCT) for signal preprocessing and time-frequency (TF) feature enhancement.
• Employing a Time-Frequency Reconstruction Network (TFRNet) with an asymmetric encoder-decoder and adaptive random masking for TF representation reconstruction.
• Implementing a self-supervised pretraining strategy followed by transfer learning to optimize performance with limited labeled IS samples.

Main Results:

• The PCT-TFRNet method demonstrates high classification accuracy under IS conditions.
• Achieved 89% accuracy at a signal-to-noise ratio (SNR) of -14 dB, showcasing robustness in low-SNR and small-sample scenarios.
• The approach effectively reconstructs complete TF representations from incomplete IS data.

Conclusions:

• PCT-TFRNet offers an effective solution for AMC of proximity detector signals under interrupted sampling.
• The proposed method shows significant promise for improving the performance of electronic countermeasures in challenging signal environments.
• The combination of PCT, TFRNet, and advanced learning strategies leads to superior robustness and generalization capabilities.