LJCD-Net: Cross-Domain Jamming Generalization Diagnostic Network Based on Deep Adversarial Transfer.

Area of Science:

• * Signal processing and machine learning for navigation systems.
• * Robust positioning in challenging radio frequency environments.

Background:

• * Global Navigation Satellite Systems (GNSS) and 5G provide essential positioning, navigation, and timing (PNT).
• * Low signal power in 5G environments increases vulnerability to jamming, impacting PNT accuracy.
• * Existing jamming diagnosis methods struggle with adaptability and real-world distribution shifts.

Purpose of the Study:

• * To develop a robust jamming diagnosis network that generalizes across different data distributions.
• * To improve the accuracy and reliability of jamming classification for GNSS and 5G receivers.
• * To overcome limitations of traditional methods and deep learning models facing domain discrepancies.

Main Methods:

• * Introduction of LJCD-Net, a deep adversarial migration-based cross-domain network.
• * Utilizing a fully labeled source domain and multiple unlabeled auxiliary domains for feature learning.
• * Implementing an uncertainty-guided auxiliary domain labeling weighting strategy and probabilistic spatial constraints.

Main Results:

• * LJCD-Net demonstrated enhanced recognition accuracy and confidence in jamming diagnosis.
• * The proposed uncertainty weighting and spatial constraint methods improved generalization capabilities.
• * Significant performance gains were observed compared to five other diagnostic methods.

Conclusions:

• * LJCD-Net effectively addresses cross-domain generalization challenges in jamming diagnosis.
• * The network provides a more adaptable and reliable solution for spectrum sensing and mitigation.
• * This research contributes to more resilient positioning services in diverse operational environments.