Automatic Modulation Classification Based on CNN-Transformer Graph Neural Network.

Area of Science:

• Electrical Engineering
• Computer Science
• Signal Processing

Background:

• Deep learning, including Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), shows promise for modulation classification.
• Traditional methods often use raw signals or time-frequency images as input.
• Graph Neural Networks (GNNs) offer a new paradigm by converting time series data into graph structures.

Purpose of the Study:

• To propose a novel CNN-transformer graph neural network (CTGNet) for enhanced modulation classification.
• To explore the effectiveness of transforming signal data into graph structures for complex representation learning.
• To achieve superior recognition accuracy in modulation classification tasks.

Main Methods:

• Sliding window processing was applied to original signals to create signal subsequences.
• Signal subsequences were reorganized into a signal subsequence matrix.
• The proposed CTGNet adaptively mapped signal matrices into graph structures, utilizing GraphSAGE and DMoNPool for classification.

Main Results:

• The CTGNet method achieved the highest recognition accuracy compared to advanced deep learning techniques.
• Experiments demonstrated CTGNet's significant advantage in capturing key signal features.
• The proposed approach provides an effective solution for modulation classification.

Conclusions:

• CTGNet offers a powerful new approach for modulation classification by leveraging graph neural networks.
• The method's ability to uncover complex representations in signal data is a key strength.
• This research advances the field of signal processing and machine learning applications.