Towards cross-domain few-shot modulation classification: a feature transformation graph neural network approach

Area of Science:

• Electrical Engineering
• Computer Science
• Signal Processing

Background:

• Automatic modulation classification (AMC) is vital for signal recognition in many applications.
• Existing deep learning (DL) methods for AMC require extensive labeled data, limiting practical use.
• Few-shot learning (FSL) methods show promise but struggle with domain generalization.

Purpose of the Study:

• To develop a novel AMC method addressing limited data and domain distribution differences.
• To enhance the generalization capabilities of few-shot learning models in AMC.

Main Methods:

• Advanced signal transformation to convert time-series data into images.
• Efficient convolutional neural network (CNN) with feature-wise transformation for domain alignment.
• Few-shot graph neural network (GNN) with task construction for robust anti-domain shift capabilities.

Main Results:

• The proposed method effectively handles limited data and domain shifts in AMC.
• Signal transformation and CNN-based feature extraction proved effective.
• The few-shot GNN demonstrated robust anti-domain shift capabilities.

Conclusions:

• The novel cross-domain few-shot AMC method significantly outperforms existing FSL approaches.
• The proposed techniques offer a viable solution for AMC in data-scarce and diverse environments.