Automatic Modulation Classification Using Hybrid Data Augmentation and Lightweight Neural Network.

Area of Science:

• Electrical Engineering
• Computer Science
• Signal Processing

Background:

• Automatic modulation classification (AMC) is crucial for intelligent wireless communications.
• Deep learning has advanced AMC, but high complexity hinders deployment on resource-constrained devices.
• Existing neural network models face challenges with low latency and storage requirements.

Purpose of the Study:

• To propose a lightweight neural network framework for AMC.
• To enable efficient deployment in low-latency and low-storage environments.
• To maintain high classification performance with reduced complexity.

Main Methods:

• Developed a lightweight neural network architecture for AMC.
• Integrated complex convolution and residual networks to enhance classification.
• Employed depthwise separable convolution for model efficiency.
• Implemented a hybrid data augmentation scheme to mitigate performance loss.

Main Results:

• Achieved an approximate 83.34% reduction in model parameters.
• Reduced floating-point operations (FLOPs) by approximately 83.77%.
• Demonstrated no degradation in classification performance compared to heavier models.

Conclusions:

• The proposed lightweight AMC framework offers significant efficiency gains.
• The method successfully balances reduced complexity with maintained performance.
• This framework is suitable for deployment in resource-limited wireless communication systems.