A Novel Image-Classification-Based Decoding Strategy for Downlink Sparse Code Multiple Access Systems.

Area of Science:

• Wireless Communications
• Signal Processing
• Machine Learning

Background:

• Sparse Code Multiple Access (SCMA) is crucial for future cellular systems.
• Traditional Message Passing Algorithm (MPA) decoding in SCMA has high computational complexity, hindering low-latency requirements.
• Deep Learning (DL) offers potential for low-complexity, low-bit error rate (BER) signal detection.

Purpose of the Study:

• To develop a novel, efficient decoding scheme for SCMA systems.
• To leverage deep learning for enhanced SCMA receiver performance.
• To address the limitations of MPA in terms of computational complexity and latency.

Main Methods:

• A novel SCMA decoding approach using image classification with graph neural networks (GNNs).
• Utilizing eigenvalues of training images to capture signal amplitude, phase, and channel characteristics.
• Replacing complex codeword separation with a DL-based image classification task.

Main Results:

• The proposed DL-based SCMA decoding scheme achieves superior BER performance compared to existing methods.
• The novel approach demonstrates significantly lower computational complexity than traditional MPA.
• The method effectively decodes overlapping codewords from individual sub-users.

Conclusions:

• The proposed graph neural network-based image classification method offers a promising alternative for SCMA decoding.
• This approach meets the low-latency and high-efficiency demands of future wireless communication systems.
• Deep learning provides an effective solution for optimizing SCMA receiver performance.