University of North Dakota
Domain Adapted Signal Classification For Simulated And Real-World Signals
Abstract
dc:description.abstractSignal classification models based on deep neural networks are trained on a set of data either simulated or over-the-air that are restricted to specific channel environments, varying distortion conditions such as SNR offsets. The consequence are models that unexpectedly do not do well in scenarios when the feature distribution between domains are too large such as when simulated models are deployed in real world environments. In this regard, we propose signal classification framework leveraging Domain-Adversarial Neural Networks to address SNR variability for simulated data Unsupervised Domain Adaptation frameworks for simulated to real world signals. Our first approach employs domain adversarial learning techniques to align feature distributions across different SNR levels, mitigating domain shifts and enhancing modulation recognition robustness. Extensive experiments demonstrate significant improvements in classification accuracy compared to existing techniques, highlighting the potential of domain adversarial methods in overcoming domain discrepancies in signal classification. The second approach using unsupervised adaptation techniques based on adversarial learning, distance and stochasticity are used to counteract such feature distribution differences in order to bridge the generalization gap towards a targeted real-world domain. Many adaptation methods are analyzed in contrast to the baseline approach where isolated experiments cross-SNR and SNR-matched alongside true domain adaptation are evaluated.
Degree
thesis:*- Name thesis:degree_name
- Master of Science (MS)
- Level thesis:degree_level
- Thesis
- Discipline thesis:degree_discipline
- Electrical Engineering
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Ali, Mohammad
- Contributors dc:contributor
-
- Jielun Zhang
Subjects
dc:subject × 1Identifiers
dc:identifier.*- Repository record dc:identifier
- https://commons.und.edu/theses/7092