An Efficient Network with Novel Quantization Designed for Massive MIMO CSI Feedback

Xinran Sun,Zhengming Zhang,Luxi Yang

2024-05-30

Abstract:The efficacy of massive multiple-input multiple-output (MIMO) techniques heavily relies on the accuracy of channel state information (CSI) in frequency division duplexing (FDD) systems. Many works focus on CSI compression and quantization methods to enhance CSI reconstruction accuracy with lower feedback overhead. In this letter, we propose CsiConformer, a novel CSI feedback network that combines convolutional operations and self-attention mechanisms to improve CSI feedback accuracy. Additionally, a new quantization module is developed to improve encoding efficiency. Experiment results show that CsiConformer outperforms previous state-of-the-art networks, achieving an average accuracy improvement of 17.67\% with lower computational overhead.

Signal Processing

What problem does this paper attempt to address?

The paper aims to address the issue of Channel State Information (CSI) feedback in Massive Multiple-Input Multiple-Output (Massive MIMO) systems. Specifically:

1. **Problem Background**: In Frequency Division Duplex (FDD) systems, since the uplink and downlink channels are not reciprocal, efficient compression and quantization methods are needed to improve CSI reconstruction accuracy and reduce feedback overhead.

2. **Research Objectives**:
- Propose a new network architecture, CsiConformer, which combines convolution operations and self-attention mechanisms to enhance the accuracy of CSI feedback.
- Develop a new quantization module, SVQ-V AE, to further improve encoding efficiency.

3. **Main Contributions**:
- CsiConformer efficiently integrates local features and global representations through a four-layer Conformer structure, and experiments demonstrate its performance surpasses existing state-of-the-art networks.
- The SVQ-V AE quantization module effectively transforms CSI feature maps through up-channel convolution and matches them with embedding vectors, reducing information loss.

In summary, the goal of this paper is to address the accuracy and computational complexity of CSI feedback in Massive MIMO systems by proposing new network architectures and quantization methods.