Deep Learning-Based Automatic Modulation Classification Using Robust CNN Architecture for Cognitive Radio Networks.

Area of Science:

• Electrical Engineering
• Computer Science
• Signal Processing

Background:

• Automatic modulation classification (AMC) is crucial for intelligent receivers in non-cooperative communication systems, including cognitive radio and military applications.
• Existing AMC methods face challenges in accurately classifying modulation schemes under various real-world wireless channel impairments.

Purpose of the Study:

• To propose a robust automatic modulation classification model using a novel convolutional neural network (CNN) architecture.
• To enhance feature extraction capabilities for improved modulation recognition in complex wireless environments.

Main Methods:

• Developed a new CNN architecture featuring parallel combinations of asymmetric convolutional kernels.
• Implemented skip connections within the CNN architecture to mitigate vanishing gradient problems.
• Evaluated the model's performance on nine modulation schemes under various channel impairments (AWGN, Rician fading, clock offset).

Main Results:

• The proposed CNN model achieved high classification accuracy across different SNRs, reaching 86.1% at -2 dB, 96.5% at 0 dB, and 99.8% at 10 dB.
• Demonstrated superior performance compared to existing state-of-the-art methods in modulation type recognition.
• Showcased strong feature extraction capabilities, achieving 81.02% average accuracy for challenging 16QAM and 64QAM signals.

Conclusions:

• The novel CNN architecture provides a robust and effective solution for automatic modulation classification.
• The model's enhanced performance at low SNRs makes it suitable for realistic non-cooperative communication scenarios.
• The architecture's ability to distinguish between similar modulation schemes highlights its advanced feature extraction power.