Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Knowledge-Aided Semantic Communication Leveraging Probabilistic Graphical Modeling

Haowen Wan, Qianqian Yang, Jiancheng Tang, Zhiguo shi

TL;DR

The paper addresses the inefficiency of transmitting data-rich visuals over wireless links by introducing semantic communication anchored by a shared probabilistic graphical model (PGM). It builds a Bayesian network over disentangled latent features extracted via Semantic StyleGAN inversion, trained on CelebA-HQ to capture cross-feature dependencies and enable compression of predictable semantic components, resulting in a bandwidth compression ratio of $kn$. A compression/decompression pipeline preserves a subset of informative features (14 in the study) and reconstructs discarded details at the receiver through PGM inference, with 28 channel-encoder/decoder branches supporting variable latent sizes. Experiments on facial-image transmission show improved transmission efficiency and robustness under AWGN channels compared to Deep JSCC baselines, achieving competitive PSNR and superior perceptual quality (LPIPS).

Abstract

In this paper, we propose a semantic communication approach based on probabilistic graphical model (PGM). The proposed approach involves constructing a PGM from a training dataset, which is then shared as common knowledge between the transmitter and receiver. We evaluate the importance of various semantic features and present a PGM-based compression algorithm designed to eliminate predictable portions of semantic information. Furthermore, we introduce a technique to reconstruct the discarded semantic information at the receiver end, generating approximate results based on the PGM. Simulation results indicate a significant improvement in transmission efficiency over existing methods, while maintaining the quality of the transmitted images.

Knowledge-Aided Semantic Communication Leveraging Probabilistic Graphical Modeling

TL;DR

The paper addresses the inefficiency of transmitting data-rich visuals over wireless links by introducing semantic communication anchored by a shared probabilistic graphical model (PGM). It builds a Bayesian network over disentangled latent features extracted via Semantic StyleGAN inversion, trained on CelebA-HQ to capture cross-feature dependencies and enable compression of predictable semantic components, resulting in a bandwidth compression ratio of . A compression/decompression pipeline preserves a subset of informative features (14 in the study) and reconstructs discarded details at the receiver through PGM inference, with 28 channel-encoder/decoder branches supporting variable latent sizes. Experiments on facial-image transmission show improved transmission efficiency and robustness under AWGN channels compared to Deep JSCC baselines, achieving competitive PSNR and superior perceptual quality (LPIPS).

Abstract

In this paper, we propose a semantic communication approach based on probabilistic graphical model (PGM). The proposed approach involves constructing a PGM from a training dataset, which is then shared as common knowledge between the transmitter and receiver. We evaluate the importance of various semantic features and present a PGM-based compression algorithm designed to eliminate predictable portions of semantic information. Furthermore, we introduce a technique to reconstruct the discarded semantic information at the receiver end, generating approximate results based on the PGM. Simulation results indicate a significant improvement in transmission efficiency over existing methods, while maintaining the quality of the transmitted images.
Paper Structure (13 sections, 9 equations, 7 figures)

This paper contains 13 sections, 9 equations, 7 figures.

Table of Contents

  1. Introduction
  2. SYSTEM MODEL
  3. Knowledge Base
  4. Transmitter
  5. Receiver
  6. Metrics
  7. Proposed Method
  8. Semantic Encoder and Decoder
  9. Construction of Probabilistic Graphical Model
  10. Compression and Decompression Algorithm
  11. Channel Encoder and Decoder
  12. Simulation
  13. CONCLUSION

Figures (7)

  • Figure 1: The overall structure of the proposed semantic communication system for image transmission
  • Figure 2: Bayesian network trained on CelebA-HQ dataset
  • Figure 3: PSNR versus SNR for different approaches. The compression ratios are listed in the legend.
  • Figure 4: LPIPS versus SNR for different approaches. The compression ratios are listed in the legend.
  • Figure 5: Example of reconstructed images produced by the Deep JSCC and proposed method for image transmission for AWGN channel, and the compression ratio of Deep JSCC and proposed method is 1/48 and 1/109 respectively. From left to right, the columns correspond to SNR values of -5dB, -1dB, 1dB, 5dB.
  • ...and 2 more figures