Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Swin Transformer-Based CSI Feedback for Massive MIMO

Jiaming Cheng, Wei Chen, Jialong Xu, Yiran Guo, Lun Li, Bo Ai

TL;DR

This work tackles the CSI feedback bottleneck in FDD massive MIMO by introducing SwinCFNet, a two-stage autoencoder that leverages Swin Transformer blocks to model long-range dependencies in the angular–delay domain CSI. The input CSI is treated as an image, with patch-based embedding and hierarchical processing through Stage 1 and Stage 2, followed by a fully connected projector to a codeword, and a symmetric decoder with upsampling. Experimental results on COST2100 show significant NMSE and cosine-similarity gains, especially in outdoor scenarios, across multiple compression ratios, with detailed hyper-parameter analyses and complexity considerations. The proposed approach advances practical CSI feedback by achieving higher accuracy at given feedback budgets, enabling more efficient FDD massive MIMO deployments.

Abstract

For massive multiple-input multiple-output systems in the frequency division duplex (FDD) mode, accurate downlink channel state information (CSI) is required at the base station (BS). However, the increasing number of transmit antennas aggravates the feedback overhead of CSI. Recently, deep learning (DL) has shown considerable potential to reduce CSI feedback overhead. In this paper, we propose a Swin Transformer-based autoencoder network called SwinCFNet for the CSI feedback task. In particular, the proposed method can effectively capture the long-range dependence information of CSI. Moreover, we explore the impact of the number of Swin Transformer blocks and the dimension of feature channels on the performance of SwinCFNet. Experimental results show that SwinCFNet significantly outperforms other DL-based methods with comparable model sizes, especially for the outdoor scenario.

Swin Transformer-Based CSI Feedback for Massive MIMO

TL;DR

This work tackles the CSI feedback bottleneck in FDD massive MIMO by introducing SwinCFNet, a two-stage autoencoder that leverages Swin Transformer blocks to model long-range dependencies in the angular–delay domain CSI. The input CSI is treated as an image, with patch-based embedding and hierarchical processing through Stage 1 and Stage 2, followed by a fully connected projector to a codeword, and a symmetric decoder with upsampling. Experimental results on COST2100 show significant NMSE and cosine-similarity gains, especially in outdoor scenarios, across multiple compression ratios, with detailed hyper-parameter analyses and complexity considerations. The proposed approach advances practical CSI feedback by achieving higher accuracy at given feedback budgets, enabling more efficient FDD massive MIMO deployments.

Abstract

For massive multiple-input multiple-output systems in the frequency division duplex (FDD) mode, accurate downlink channel state information (CSI) is required at the base station (BS). However, the increasing number of transmit antennas aggravates the feedback overhead of CSI. Recently, deep learning (DL) has shown considerable potential to reduce CSI feedback overhead. In this paper, we propose a Swin Transformer-based autoencoder network called SwinCFNet for the CSI feedback task. In particular, the proposed method can effectively capture the long-range dependence information of CSI. Moreover, we explore the impact of the number of Swin Transformer blocks and the dimension of feature channels on the performance of SwinCFNet. Experimental results show that SwinCFNet significantly outperforms other DL-based methods with comparable model sizes, especially for the outdoor scenario.
Paper Structure (12 sections, 11 equations, 4 figures, 3 tables)

This paper contains 12 sections, 11 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. System model
  3. Massive MIMO-OFDM System
  4. DL-Based CSI Feedback
  5. Design of SwinCFNet
  6. Overall Architecture
  7. Swin Transformer Blocks
  8. Experimental Results
  9. Experimental Settings
  10. Performance Comparison
  11. Performance of Different Hyper-Parameters Settings
  12. Conclusion

Figures (4)

  • Figure 1: The overall architecture of the proposed SwinCFNet, which includes the encoder and the decoder.
  • Figure 2: Two successive Swin Transformer Blocks.
  • Figure 3: (a) Shifted window approach for computing self-attention. (b) Implementation of the shifted window partition with the cyclic shift.
  • Figure 4: NMSE(dB) performance comparison under different compression ratios in the outdoor scenario.