Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SAFE: Semantic Adaptive Feature Extraction with Rate Control for 6G Wireless Communications

Yuna Yan, Lixin Li, Xin Zhang, Wensheng Lin, Wenchi Cheng, Zhu Han

TL;DR

SAFE tackles the rigidity of fixed-rate semantic communication by introducing a semantic adaptive feature extraction framework that decomposes images into multiple sub-semantics and enables adaptive transmission across channel bandwidths. It employs a U-Net–inspired architecture with two sub-semantics and a semantic contraction decoder to reconstruct images as more information is received, while defining bandwidth ratios $k_i/n$ and $k/n$ to quantify capacity usage. The authors propose three training strategies to efficiently train the SAFE network and demonstrate, on ImageNet100, that adaptive use of sub-semantics yields improved bandwidth efficiency and robustness across AWGN and Rayleigh channels. This work advances practical, bandwidth-aware semantic communication for 6G by enabling flexible, task-agnostic image transmission with progressive quality refinement.

Abstract

Most current Deep Learning-based Semantic Communication (DeepSC) systems are designed and trained exclusively for particular single-channel conditions, which restricts their adaptability and overall bandwidth utilization. To address this, we propose an innovative Semantic Adaptive Feature Extraction (SAFE) framework, which significantly improves bandwidth efficiency by allowing users to select different sub-semantic combinations based on their channel conditions. This paper also introduces three advanced learning algorithms to optimize the performance of SAFE framework as a whole. Through a series of simulation experiments, we demonstrate that the SAFE framework can effectively and adaptively extract and transmit semantics under different channel bandwidth conditions, of which effectiveness is verified through objective and subjective quality evaluations.

SAFE: Semantic Adaptive Feature Extraction with Rate Control for 6G Wireless Communications

TL;DR

SAFE tackles the rigidity of fixed-rate semantic communication by introducing a semantic adaptive feature extraction framework that decomposes images into multiple sub-semantics and enables adaptive transmission across channel bandwidths. It employs a U-Net–inspired architecture with two sub-semantics and a semantic contraction decoder to reconstruct images as more information is received, while defining bandwidth ratios and to quantify capacity usage. The authors propose three training strategies to efficiently train the SAFE network and demonstrate, on ImageNet100, that adaptive use of sub-semantics yields improved bandwidth efficiency and robustness across AWGN and Rayleigh channels. This work advances practical, bandwidth-aware semantic communication for 6G by enabling flexible, task-agnostic image transmission with progressive quality refinement.

Abstract

Most current Deep Learning-based Semantic Communication (DeepSC) systems are designed and trained exclusively for particular single-channel conditions, which restricts their adaptability and overall bandwidth utilization. To address this, we propose an innovative Semantic Adaptive Feature Extraction (SAFE) framework, which significantly improves bandwidth efficiency by allowing users to select different sub-semantic combinations based on their channel conditions. This paper also introduces three advanced learning algorithms to optimize the performance of SAFE framework as a whole. Through a series of simulation experiments, we demonstrate that the SAFE framework can effectively and adaptively extract and transmit semantics under different channel bandwidth conditions, of which effectiveness is verified through objective and subjective quality evaluations.
Paper Structure (9 sections, 1 equation, 6 figures)

This paper contains 9 sections, 1 equation, 6 figures.

Table of Contents

  1. Introduction
  2. The Semantic Adaptive Feature Extraction Based Semantic Communication Model
  3. System Model
  4. The Proposed SAFE Network Framework
  5. Training Strategies
  6. Simulation Results
  7. Training Parameters
  8. Experimental Results
  9. Conclusion

Figures (6)

  • Figure 1: The overall architecture of the proposed SAFE system.
  • Figure 2: The network framework of the proposed SAFE system.
  • Figure 3: The training flow diagram of three learning algorithms.
  • Figure 4: Comparison of PSNR and SNR in Strategy 2 under different channel conditions.
  • Figure 5: Comparison of PSNR and SNR in Strategy 3 under different channel conditions.
  • ...and 1 more figures