A Lightweight Dual-Branch Complex-Valued Neural Network for Automatic Modulation Classification of Communication...

Area of Science:

• Electrical Engineering
• Computer Science
• Signal Processing

Background:

• Deep learning, particularly complex-valued neural networks (CVNNs), is pivotal for automatic modulation classification (AMC) due to its feature extraction capabilities.
• CVNNs offer advantages in processing complex communication signals, capturing both amplitude and phase information.
• Existing CVNN models for AMC suffer from high parameter counts and computational complexity, hindering practical deployment.

Purpose of the Study:

• To introduce a novel lightweight dual-branch complex-valued neural network (LDCVNN) for efficient and accurate AMC.
• To address the limitations of existing models regarding parameter count and computational load.
• To enhance feature extraction and classification performance in complex communication signal processing.

Main Methods:

• Proposed a dual-branch architecture to separately process phase information and complex-scaling-equivariant representations.
• Utilized trainable weighted fusion to adaptively combine features from both branches.
• Extended spatial and channel reconstruction convolution (SCConv) to the complex domain, incorporating complex-valued depthwise separable convolution blocks (CBlock) and average pooling for feature optimization.

Main Results:

• The LDCVNN achieved the highest average accuracy on the RML2016.10a dataset with only 9.0 K parameters and no data augmentation.
• Demonstrated significant parameter reduction: 99.33% compared to CDSN and 97.25% compared to CSDNN.
• Showcased a superior balance between efficiency and performance across multiple datasets.

Conclusions:

• The LDCVNN offers a highly efficient and effective solution for automatic modulation classification.
• This model significantly reduces computational complexity and parameter count, making it suitable for deployment in resource-constrained environments.
• The proposed architecture advances the state-of-the-art in signal processing for wireless communications.