Leveraging Machine Learning Classifiers in Transfer Learning for Few-Shot Modulation Recognition.

Area of Science:

• Electrical Engineering
• Computer Science
• Signal Processing

Background:

• Communication systems require efficient modulation recognition.
• Deep learning methods face challenges in few-shot scenarios due to limited labeled data.
• Few-shot modulation recognition (FSMR) is crucial for practical applications.

Purpose of the Study:

• To propose a hybrid transfer learning (HTL) approach for robust few-shot modulation recognition (FSMR).
• To evaluate the performance of traditional machine learning classifiers within the HTL framework for FSMR.
• To address the limitations of conventional deep learning in data-scarce environments.

Main Methods:

• A hybrid transfer learning (HTL) approach combining deep feature extraction and traditional machine learning (ML) classifiers.
• Knowledge transfer from large-scale datasets via pre-training.
• Few-shot adaptation using various classical ML classifiers, including K-nearest neighbor.

Main Results:

• The proposed HTL approach consistently outperforms existing baseline methods in data-scarce settings.
• K-nearest neighbor classifier demonstrated the most robust and generalizable performance within the HTL paradigm.
• Parameter analysis provided insights for practical deployment in real-world applications.

Conclusions:

• HTL offers a promising solution for reliable few-shot modulation recognition (FSMR) in practical, data-limited scenarios.
• The synergy between deep feature extraction and stable ML classifiers enhances performance.
• K-nearest neighbor is identified as a highly effective classifier for FSMR within the HTL framework.