Multi-representation domain attentive contrastive learning based unsupervised automatic modulation recognition.

Area of Science:

• Electrical Engineering
• Computer Science
• Signal Processing

Background:

• Radio spectrum congestion is increasing due to advancements in 5G/6G wireless technologies and user growth.
• Automatic modulation recognition (AMR) is vital for spectrum sensing in cognitive radio but typically requires supervised learning with extensive labeled data, which is difficult to obtain.
• Existing methods face challenges in acquiring reliable labels for supervised AMR.

Purpose of the Study:

• To propose an unsupervised framework for AMR that eliminates the need for labeled data.
• To extract high-quality signal features from unlabeled data using cross-domain contrastive learning.
• To enhance the adaptability and performance of AMR in dynamic wireless environments.

Main Methods:

• Developed a Multi-Representation Domain Attentive Contrastive Learning framework.
• Employed inter-domain and intra-domain contrastive learning to improve feature extraction across different signal representations.
• Integrated a domain attention module for dynamic selection of feature representation domains.

Main Results:

• The proposed unsupervised method achieved superior performance compared to existing AMR techniques on public datasets.
• The framework demonstrated effectiveness in extracting high-quality signal features from unlabeled data.
• The method showed adaptability and could be extended to various representation domains.

Conclusions:

• The study presents a significant advancement in unsupervised learning for radio signal processing.
• The proposed framework effectively bridges the gap between unsupervised and supervised learning for AMR.
• This work contributes to the development of the Internet of Things and cognitive radio technologies by improving spectrum sensing capabilities.