Automatic Modulation Classification Based on Deep Learning for Unmanned Aerial Vehicles.

Area of Science:

• Artificial Intelligence
• Signal Processing
• Telecommunications

Background:

• Deep learning excels in multimedia data processing but is underutilized in automatic modulation classification (AMC).
• Traditional AMC methods struggle with complex signal disturbances.
• AMC is a critical task in communication signal recognition.

Purpose of the Study:

• To propose a unified framework for automatic modulation classification (AMC) using deep learning.
• To address the challenges posed by complex disturbances in communication signals.
• To improve the performance of AMC systems.

Main Methods:

• Developed a heterogeneous deep model fusion (HDMF) method combining Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks.
• Created a comprehensive database of eleven single-carrier modulation signal types under various noise and fading channel conditions.
• Utilized diverse signal-to-noise ratios (SNRs) and a real geographical environment for data collection.

Main Results:

• The proposed HDMF method demonstrated strong capability in solving the AMC problem.
• Experimental results showed significantly improved performance compared to independent network models.
• The fusion approach effectively handles complex signal disturbances in communication systems.

Conclusions:

• Heterogeneous deep model fusion (HDMF) offers a robust solution for automatic modulation classification (AMC).
• Combining CNN and LSTM networks enhances AMC performance beyond individual models.
• The developed database and methodology provide a valuable resource for future research in signal recognition.