Automatic Modulation Recognition for Radio Mixed Proximity Sensor Signals Based on a Time-Frequency Image Enhancement...

Area of Science:

• Signal Processing
• Machine Learning
• Electronic Warfare

Background:

• Automatic modulation recognition (AMR) is crucial for electronic reconnaissance of low probability intercept (LPI) signals.
• Traditional methods struggle with accuracy under low signal-to-noise ratios (SNRs).
• Deep learning, particularly Convolutional Neural Networks (CNNs), shows promise for signal classification using time-frequency images (TFIs).

Purpose of the Study:

• To propose an improved AMR method for radio frequency proximity sensor signals, focusing on enhancing accuracy in low SNR conditions.
• To develop a TFI enhancement network for denoising and improving the quality of input features for classification.

Main Methods:

• Utilized a per-pixel kernel prediction network (KPN) for denoising time-frequency images (TFIs), enhancing signal quality.
• Employed a RetinalNet-based classifier to process the enhanced TFIs for modulation recognition.
• Compared denoising performance against traditional methods like sparse representation and low-rank approximation.

Main Results:

• The KPN-based TFI enhancement achieved comparable denoising performance to traditional methods with significantly lower computational overhead.
• The proposed method demonstrated high accuracy in identifying modulation types of radio frequency proximity sensors, even when aliased in the time-frequency domain.
• Achieved an average recognition accuracy rate exceeding 97% at SNRs above -10 dB.

Conclusions:

• The TFI enhancement network effectively improves the quality of input features for AMR, crucial for low SNR environments.
• The proposed deep learning-based AMR method offers a robust and computationally efficient solution for complex signal recognition tasks.
• The method shows significant potential for applications in electronic reconnaissance and signal intelligence.