Deep-Learning-Based Classification of Digitally Modulated Signals Using Capsule Networks and Cyclic Cumulants

Area of Science:

• Signal Processing
• Machine Learning
• Deep Learning

Background:

• Digital modulation signal classification is crucial for various communication systems.
• Traditional methods often struggle with complex signal environments and generalization.
• Deep learning offers potential for improved classification accuracy and robustness.

Purpose of the Study:

• To propose a novel deep-learning based approach for classifying digitally modulated signals.
• To leverage capsule networks (CAPs) with cyclic cumulant (CC) features for enhanced signal classification.
• To evaluate the classification performance and generalization capabilities of the proposed method.

Main Methods:

• Blind estimation of cyclic cumulant (CC) features using cyclostationary signal processing (CSP).
• Inputting estimated CC features into capsule networks (CAPs) for training and classification.
• Testing the approach on two distinct datasets with varying signal generation parameters.

Main Results:

• The proposed CAPs and CCs approach significantly outperformed conventional CSP-based classifiers.
• It also showed better performance compared to alternative deep learning models like CNNs and RESNETs using I/Q data.
• The method demonstrated strong generalization abilities across datasets with different generation parameters.

Conclusions:

• Capsule networks combined with cyclic cumulants offer a powerful and effective method for digital modulation signal classification.
• This deep learning approach provides a robust alternative to existing classification techniques.
• The findings highlight the potential of feature engineering with advanced deep learning architectures for signal processing tasks.