Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Discrete-Space Generative AI Pipeline for Semantic Transmission of Signals

Silvija Kokalj-Filipovic, Yagna Kaasaragadda

TL;DR

These findings demonstrate Discernment's ability to adjust to diverse physical channel conditions while maintaining spectral efficiency and low model complexity, making it well suited for IoT deployments and strongly motivating further research on this semantic channel paradigm.

Abstract

We introduce Discernment, a semantic communication system that transmits the meaning of physical signals (baseband radio and audio) over a technical channel using GenAI models operating in discrete spaces. Discernment dynamically adapts to channel impairments - modeled as erasure channels - by switching between an autoregressive or a diffusion-based generative algorithm, depending on the erasure pattern. Our results show that Discernment maintains semantic integrity even as channel capacity severely degrades, exhibiting very small and graceful performance decline in both classification accuracy and statistical fidelity of the reconstructed meaning. These findings demonstrate Discernment's ability to adjust to diverse physical channel conditions while maintaining spectral efficiency and low model complexity, making it well suited for IoT deployments and strongly motivating further research on this semantic channel paradigm.

Discrete-Space Generative AI Pipeline for Semantic Transmission of Signals

TL;DR

These findings demonstrate Discernment's ability to adjust to diverse physical channel conditions while maintaining spectral efficiency and low model complexity, making it well suited for IoT deployments and strongly motivating further research on this semantic channel paradigm.

Abstract

We introduce Discernment, a semantic communication system that transmits the meaning of physical signals (baseband radio and audio) over a technical channel using GenAI models operating in discrete spaces. Discernment dynamically adapts to channel impairments - modeled as erasure channels - by switching between an autoregressive or a diffusion-based generative algorithm, depending on the erasure pattern. Our results show that Discernment maintains semantic integrity even as channel capacity severely degrades, exhibiting very small and graceful performance decline in both classification accuracy and statistical fidelity of the reconstructed meaning. These findings demonstrate Discernment's ability to adjust to diverse physical channel conditions while maintaining spectral efficiency and low model complexity, making it well suited for IoT deployments and strongly motivating further research on this semantic channel paradigm.
Paper Structure (18 sections, 8 equations, 5 figures, 1 table)

This paper contains 18 sections, 8 equations, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. Technical Background
  3. Vector Quantized Variational Autoencoders (VQVAE)
  4. Transformer Architectures
  5. Discrete Diffusion Models
  6. Generative Modeling in RF and Audio Domains
  7. Semantic Communication System Model
  8. Methodology
  9. Datasets
  10. RF Modulated Signal
  11. Audio Data
  12. VQVAE: Discretization of Input Space
  13. Transformer: Learning the Autoregressive Prior
  14. SEDD: Score Entropy Discrete Diffusion
  15. Semantic Channel Pipeline
  16. ...and 3 more sections

Figures (5)

  • Figure 1: VQVAE Architecture with Codebook $Q$ of $N=64$ codewords
  • Figure 2: Two Discernment models: DoT and SEDD differ not only in how they get trained, but also how they generate data, both randomly and from incomplete representations.
  • Figure 3: Fidelity metrics of the DoT Discernment for the Torchsig classification task as a function of the number of received tokens $t_e=2^x$ out of $d_s=512.$ Shown are the accuracy of classification, statistical data fidelity and top-F1 score of the reconstructed data.
  • Figure 4: Accuracy of the AudioMNIST DoT-Discernement for different received context lengths, different DoT models and different codebook sizes $N$ for the same $\ell,$$K$ and $d_s$.
  • Figure 5: Accuracy, fidelity and diversity of the SEDD-Discernement Configuration 1 (N=128) shows amazing robustness to increased percentage of lost packets $\epsilon$ ( from 50% to $\sim$ 97%) with $d_s=512$ but not with $d_s=128$. Conf. 2 with $\ell=128$ shows the same behavior at $d_s=512$ and $d_s=128$, as does the Conf. 4 for $\ell=256.$