Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SNR-EQ-JSCC: Joint Source-Channel Coding with SNR-Based Embedding and Query

Hongwei Zhang, Meixia Tao

TL;DR

This work tackles dynamic channels in semantic joint source-channel coding by introducing SNR-EQ-JSCC, a Transformer-based architecture that performs channel adaptation through channel-adaptive queries (CAQ) and SNR embedding. By conditioning attention on the SNR via CAQ and enriching CAMHA inputs with an SNR embedding, the method achieves improved reconstruction quality and perceptual fidelity while maintaining extremely low CA overhead; a penalty-driven loss stabilizes training. An average-SNR variant is provided for scenarios with imperfect instantaneous SNR feedback, avoiding retraining. Empirical results on DIV2K show that SNR-EQ-JSCC outperforms SwinJSCC in PSNR and perceptual metrics, with substantial reductions in storage and computation for channel adaptation, and the CAQ mechanism yields notable gains in perception metrics. Overall, the approach offers a lightweight, robust pathway to channel-adaptive semantic coding applicable to Transformer-based JSCC systems.

Abstract

Coping with the impact of dynamic channels is a critical issue in joint source-channel coding (JSCC)-based semantic communication systems. In this paper, we propose a lightweight channel-adaptive semantic coding architecture called SNR-EQ-JSCC. It is built upon the generic Transformer model and achieves channel adaptation (CA) by Embedding the signal-to-noise ratio (SNR) into the attention blocks and dynamically adjusting attention scores through channel-adaptive Queries. Meanwhile, penalty terms are introduced in the loss function to stabilize the training process. Considering that instantaneous SNR feedback may be imperfect, we propose an alternative method that uses only the average SNR, which requires no retraining of SNR-EQ-JSCC. Simulation results conducted on image transmission demonstrate that the proposed SNR-EQJSCC outperforms the state-of-the-art SwinJSCC in peak signal-to-noise ratio (PSNR) and perception metrics while only requiring 0.05% of the storage overhead and 6.38% of the computational complexity for CA. Moreover, the channel-adaptive query method demonstrates significant improvements in perception metrics. When instantaneous SNR feedback is imperfect, SNR-EQ-JSCC using only the average SNR still surpasses baseline schemes.

SNR-EQ-JSCC: Joint Source-Channel Coding with SNR-Based Embedding and Query

TL;DR

This work tackles dynamic channels in semantic joint source-channel coding by introducing SNR-EQ-JSCC, a Transformer-based architecture that performs channel adaptation through channel-adaptive queries (CAQ) and SNR embedding. By conditioning attention on the SNR via CAQ and enriching CAMHA inputs with an SNR embedding, the method achieves improved reconstruction quality and perceptual fidelity while maintaining extremely low CA overhead; a penalty-driven loss stabilizes training. An average-SNR variant is provided for scenarios with imperfect instantaneous SNR feedback, avoiding retraining. Empirical results on DIV2K show that SNR-EQ-JSCC outperforms SwinJSCC in PSNR and perceptual metrics, with substantial reductions in storage and computation for channel adaptation, and the CAQ mechanism yields notable gains in perception metrics. Overall, the approach offers a lightweight, robust pathway to channel-adaptive semantic coding applicable to Transformer-based JSCC systems.

Abstract

Coping with the impact of dynamic channels is a critical issue in joint source-channel coding (JSCC)-based semantic communication systems. In this paper, we propose a lightweight channel-adaptive semantic coding architecture called SNR-EQ-JSCC. It is built upon the generic Transformer model and achieves channel adaptation (CA) by Embedding the signal-to-noise ratio (SNR) into the attention blocks and dynamically adjusting attention scores through channel-adaptive Queries. Meanwhile, penalty terms are introduced in the loss function to stabilize the training process. Considering that instantaneous SNR feedback may be imperfect, we propose an alternative method that uses only the average SNR, which requires no retraining of SNR-EQ-JSCC. Simulation results conducted on image transmission demonstrate that the proposed SNR-EQJSCC outperforms the state-of-the-art SwinJSCC in peak signal-to-noise ratio (PSNR) and perception metrics while only requiring 0.05% of the storage overhead and 6.38% of the computational complexity for CA. Moreover, the channel-adaptive query method demonstrates significant improvements in perception metrics. When instantaneous SNR feedback is imperfect, SNR-EQ-JSCC using only the average SNR still surpasses baseline schemes.
Paper Structure (15 sections, 12 equations, 4 figures, 3 tables)

This paper contains 15 sections, 12 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. System Model
  3. SNR-EQ-JSCC
  4. Preliminary of Multi-Head Attention Mechanism
  5. Channel-Adaptive Query
  6. SNR Embedding
  7. Loss Function of SNR-EQ-JSCC
  8. Experiment Results
  9. Experiment Settings
  10. Effects of $\lambda$ in the Loss Function
  11. Performance Comparison of Different JSCC Schemes
  12. Performance with Imperfect SNR Feedback
  13. Visualization Results
  14. Storage Overhead and Computational Complexity
  15. Conclusion

Figures (4)

  • Figure 1: Illustration of the proposed SNR-EQ-JSCC.
  • Figure 2: Performance comparison of different $\lambda$.
  • Figure 3: Performance comparison at different $\overline{\mu}_{\rm test}$.
  • Figure 4: Illustration of the reconstructed images at $r=1/32$. The corresponding PSNR, MS-SSIM, and LPIPS results are presented beside the recovered image, respectively.