Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Learning Robust Representations for Communications over Noisy Channels

Sudharsan Senthil, Shubham Paul, Nambi Seshadri, R. David Koilpillai

TL;DR

This work investigates the impact of using various cost functions based on mutual information and pairwise distances between codewords to generate robust representations for transmission under strict power constraints to generate robust representations for transmission under strict power constraints.

Abstract

We explore the use of FCNNs (Fully Connected Neural Networks) for designing end-to-end communication systems without taking any inspiration from existing classical communications models or error control coding. This work relies solely on the tools of information theory and machine learning. We investigate the impact of using various cost functions based on mutual information and pairwise distances between codewords to generate robust representations for transmission under strict power constraints. Additionally, we introduce a novel encoder structure inspired by the Barlow Twins framework. Our results show that iterative training with randomly chosen noise power levels while minimizing block error rate provides the best error performance.

Learning Robust Representations for Communications over Noisy Channels

TL;DR

This work investigates the impact of using various cost functions based on mutual information and pairwise distances between codewords to generate robust representations for transmission under strict power constraints to generate robust representations for transmission under strict power constraints.

Abstract

We explore the use of FCNNs (Fully Connected Neural Networks) for designing end-to-end communication systems without taking any inspiration from existing classical communications models or error control coding. This work relies solely on the tools of information theory and machine learning. We investigate the impact of using various cost functions based on mutual information and pairwise distances between codewords to generate robust representations for transmission under strict power constraints. Additionally, we introduce a novel encoder structure inspired by the Barlow Twins framework. Our results show that iterative training with randomly chosen noise power levels while minimizing block error rate provides the best error performance.
Paper Structure (9 sections, 11 equations, 11 figures, 1 table, 3 algorithms)

This paper contains 9 sections, 11 equations, 11 figures, 1 table, 3 algorithms.

Table of Contents

  1. Introduction
  2. End-to-End System Model
  3. The Autoencoder Model
  4. Experiments and Results
  5. Cost Functions for Encoders
  6. Barlow Twin based Encoder Structure
  7. Randomized Training SNRs
  8. Latent Space or Codeword Analysis
  9. Conclusions and Future Work

Figures (11)

  • Figure 1: End-to-End System Model
  • Figure 2: End-to-End Network Architecture
  • Figure 3: AE models proposed in Tim_1 vs Maximum-Likelihood Decoders
  • Figure 4: BLER performance with varying Loss Functions and Customised Power Constraints
  • Figure 5: BLER performance with varying Loss Functions which are underperforming in comparison.
  • ...and 6 more figures