Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Residual Cross-Attention Transformer-Based Multi-User CSI Feedback with Deep Joint Source-Channel Coding

Hengwei Zhang, Minghui Wu, Li Qiao, Ling Liu, Ziqi Han, Zhen Gao

TL;DR

This work tackles the challenge of high CSI feedback overhead in FDD massive MIMO by exploiting CSI correlation among nearby users. It introduces RCA-MUNet, a residual cross-attention transformer that enables joint multi-user CSI reconstruction at the BS, integrated with deep joint source-channel coding for end-to-end optimization. The key contributions include a scalable RCA-Block that extracts complementary inter-user information, a transformer-based multi-user backbone that supports any UE count, and a two-stage training strategy to adapt to varying uplink SNRs; extensive experiments show improved NMSE, reduced complexity, and robust performance under different CR, SNR, and UE settings. The results also demonstrate a clear advantage of DJSCC over conventional SSCC in multi-user CSI feedback, with better resilience to uplink noise and without cliff effects, indicating strong practical impact for scalable, high-throughput MIMO systems.

Abstract

This letter proposes a deep-learning (DL)-based multi-user channel state information (CSI) feedback framework for massive multiple-input multiple-output systems, where the deep joint source-channel coding (DJSCC) is utilized to improve the CSI reconstruction accuracy. Specifically, we design a multi-user joint CSI feedback framework, whereby the CSI correlation of nearby users is utilized to reduce the feedback overhead. Under the framework, we propose a new residual cross-attention transformer architecture, which is deployed at the base station to further improve the CSI feedback performance. Moreover, to tackle the "cliff-effect" of conventional bit-level CSI feedback approaches, we integrated DJSCC into the multi-user CSI feedback, together with utilizing a two-stage training scheme to adapt to varying uplink noise levels. Experimental results demonstrate the superiority of our methods in CSI feedback performance, with low network complexity and better scalability.

Residual Cross-Attention Transformer-Based Multi-User CSI Feedback with Deep Joint Source-Channel Coding

TL;DR

This work tackles the challenge of high CSI feedback overhead in FDD massive MIMO by exploiting CSI correlation among nearby users. It introduces RCA-MUNet, a residual cross-attention transformer that enables joint multi-user CSI reconstruction at the BS, integrated with deep joint source-channel coding for end-to-end optimization. The key contributions include a scalable RCA-Block that extracts complementary inter-user information, a transformer-based multi-user backbone that supports any UE count, and a two-stage training strategy to adapt to varying uplink SNRs; extensive experiments show improved NMSE, reduced complexity, and robust performance under different CR, SNR, and UE settings. The results also demonstrate a clear advantage of DJSCC over conventional SSCC in multi-user CSI feedback, with better resilience to uplink noise and without cliff effects, indicating strong practical impact for scalable, high-throughput MIMO systems.

Abstract

This letter proposes a deep-learning (DL)-based multi-user channel state information (CSI) feedback framework for massive multiple-input multiple-output systems, where the deep joint source-channel coding (DJSCC) is utilized to improve the CSI reconstruction accuracy. Specifically, we design a multi-user joint CSI feedback framework, whereby the CSI correlation of nearby users is utilized to reduce the feedback overhead. Under the framework, we propose a new residual cross-attention transformer architecture, which is deployed at the base station to further improve the CSI feedback performance. Moreover, to tackle the "cliff-effect" of conventional bit-level CSI feedback approaches, we integrated DJSCC into the multi-user CSI feedback, together with utilizing a two-stage training scheme to adapt to varying uplink noise levels. Experimental results demonstrate the superiority of our methods in CSI feedback performance, with low network complexity and better scalability.

Paper Structure

This paper contains 10 sections, 9 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Channel Model
  3. Proposed Multi-User CSI Feedback Scheme
  4. Proposed Multi-User CSI Feedback Network Architecture
  5. Residual Cross-Attention Multi-User CSI Joint Processing
  6. Simulation Results
  7. Comparison with State-Of-The-Art Methods
  8. Performance of Proposed Schemes Across Conditions
  9. Comparison with SSCC Feedback Scheme
  10. Conclusion

Figures (4)

  • Figure 1: Illustration of CSI correlation from nearby UEs with shared scatterers.
  • Figure 2: Proposed Transformer-based DJSCC multi-user CSI feedback framework and architecture of RCA-Block.
  • Figure 3: NMSE performance for single-user network BL1 and multi-user framework Ours with different numbers of UEs.
  • Figure 4: Performance comparison of the proposed multi-user network with DJSCC-based CSI feedback and conventional SSCC schemes under the 2-UE scenario: (a) feedback overhead 64; (b) feedback overhead 128.

Theorems & Definitions (1)

  • Remark