Deep Learning for Channel Coding via Neural Mutual Information Estimation

This repository contains the implementation of the paper "Deep Learning for Channel Coding via Neural Mutual Information Estimation".

The paper presents novel approaches for channel coding using neural networks, focusing on estimating Mutual Information (MI) to optimize the performance of autoencoders under Rayleigh and Binary Input channels.

Citation

If you find this repository useful in your research, please consider citing the following paper:

   @article{fritschek2019deep,
      title={Deep Learning for Channel Coding via Neural Mutual Information Estimation},
      author={Rick Fritschek, Rafael F. Schaefery, and Gerhard Wunder},
      journal={IEEE Transactions on Information Theory},
      volume={65},
      number={4},
      pages={2042--2059},
      year={2019},
      publisher={IEEE}
   }

Table of Contents

• Overview
• File Descriptions
• Installation
• Usage
  • Python Scripts
  • Jupyter Notebooks
• Results
• Requirements
• Contributing
• License

Overview

This repository includes:

• The original research paper in PDF format.
• Python implementations of the models and experiments discussed in the paper.
• Jupyter Notebooks for interactive exploration of the models.

The primary focus is on using Mutual Information Neural Estimation (MINE) to enhance the performance of autoencoders in channel coding tasks. The code has been structured to facilitate ease of understanding and experimentation.

File Descriptions

Python Scripts

Filename	Description
AutoEncoder_Rayleigh_Channel_MINE.py	Implements the autoencoder architecture for Rayleigh channels with MI estimation using MINE.
Autoencoder_MINE_for_Binary_Input.py	Implements the autoencoder architecture optimized for binary input using MI estimation.
MINE_Encoder_Experimental_Model.py	Contains experimental models for testing the performance of MINE-based encoders under various channel conditions.

Jupyter Notebooks

Filename	Description
AutoEncoder_Rayleigh_Channel_MINE.ipynb	Interactive version of the Python script for Rayleigh channels, with step-by-step explanations and visualizations.
Autoencoder_MINE_for_Binary_Input.ipynb	Interactive version of the Python script optimized for binary input, allowing for parameter adjustments and result visualization.
MINE_Encoder_Experimental_Model.ipynb	Interactive notebook for experimenting with different models and channel conditions, demonstrating the efficacy of the MINE approach.

Installation

1. Clone the repository:

   git clone https://github.com/jElhamm/Deep-Learning-for-Channel-Coding-MI-Estimation
   cd Deep-Learning-for-Channel-Coding-via-Neural-Mutual-Information-Estimation

2. Install the required packages:

   pip install -r requirements.txt

3. Open Jupyter Notebooks:

   jupyter notebook

Usage

To run the Python scripts directly:

   python AutoEncoder_Rayleigh_Channel_MINE.py
   python Autoencoder_MINE_for_Binary_Input.py
   python MINE_Encoder_Experimental_Model.py

Alternatively, you can explore the Jupyter Notebooks to interactively execute and visualize the results.

Results

The results of the experiments conducted using these scripts and notebooks demonstrate the effectiveness of using MINE for channel coding tasks. The implementation shows significant improvements in Bit Error Rates (BER) compared to traditional methods, particularly in noisy channel conditions. Visualizations and detailed results are available within the respective Jupyter Notebooks.

Requirements

This project requires the following Python libraries:

Library	Version	Purpose
numpy	1.26.4	Used for numerical computations and array manipulations.
scipy	1.13.1	Used for scientific and technical computations.
keras	3.4.1	Used for building and training deep learning models.
tensorflow	2.17.0	Provides the backend for deep learning models and computations.
matplotlib	3.7.1	Used for creating visualizations and plots.

To install these dependencies, you can use the requirements.txt file included in the repository. Run the following command:

pip install -r requirements.txt

Contributing

We welcome contributions to this project! To contribute, please:

1. Fork the repository and clone your fork.
2. Create a new branch for your changes.
3. Make and test your changes, then commit them.
4. Push your changes to your fork and submit a pull request.

License

This repository is licensed under the BSD-3-Clause License. See the LICENSE file for more details.