Recognizing Non-Collaborative Radio Station Communication Behaviors Using an Ameliorated LeNet

Area of Science:

• Signal Processing
• Machine Learning
• Radio Communications

Background:

• Non-collaborative short-wave radio communication behavior recognition is challenging.
• Existing methods may lack efficiency or accuracy.

Purpose of the Study:

• To improve a LeNet algorithm for recognizing radio communication behaviors.
• To leverage bispectral features for enhanced signal analysis.

Main Methods:

• Analyzed mapping between burst waveforms and communication behaviors.
• Extracted bispectral features from simulated signals for LeNet input.
• Optimized LeNet structure and convolutional kernel size.
• Utilized improved bispectral estimation and LeNet for recognition.

Main Results:

• Achieved recognition accuracies of 81.5% (8 dB SNR), 94.5% (10 dB SNR), and 99.3% (15 dB SNR).
• Demonstrated lower training time cost compared to other algorithms.
• Showcased higher recognition accuracy than existing methods.

Conclusions:

• The improved bispectral LeNet algorithm effectively recognizes radio communication behaviors.
• The proposed method offers practical advantages in accuracy and efficiency.
• This approach holds significant value for radio signal analysis and intelligence.