Generative Adversarial Network-Based Data Augmentation for Enhancing Wireless Physical Layer Authentication

Area of Science:

• Cybersecurity
• Machine Learning
• Wireless Communication

Background:

• Wireless physical layer authentication is crucial for robust wireless security.
• Deep learning techniques have shown significant promise in wireless node classification and recognition.
• A major challenge is the lack of sufficient datasets for training deep learning models in this domain.

Purpose of the Study:

• To develop and evaluate a data-driven approach for wireless node classification using deep learning.
• To address the dataset scarcity issue through automated data augmentation.
• To improve the accuracy of wireless physical layer authentication models.

Main Methods:

• Utilized generative adversarial networks (GANs) for automated data augmentation.
• Applied a convolutional neural network (CNN) for wireless node classification.
• Compared model performance using an original dataset and a synthetically generated dataset.

Main Results:

• The proposed data-driven models demonstrated effective wireless node classification.
• Data augmentation using GANs facilitated improved model training.
• Achieved an approximate 19% increase in classification accuracy rate compared to baseline.

Conclusions:

• Deep learning, particularly with GANs and CNNs, offers a powerful solution for wireless node classification and authentication.
• Automated data augmentation is a viable strategy to overcome dataset limitations in wireless security research.
• The developed models show significant potential for enhancing wireless security systems.