Understand SGrand Decoding via visualization

A Python project for learning and implementing SGrand (Soft Maximum Likelihood Decoding using GRAND with abandonment) decoding algorithms for error-correcting codes.

Overview

Provides a simple and easy-to-use framework for quickly combining codes, channels, and decoders through a runner, offering both single-run execution and Monte Carlo simulation.

Most importantly, it includes an intuitive visualization page for SGrand decoding.

This project for Jaist I437E Coding Theory Final Project Report.

Project Structure

Learn_SGrand_Decoding/
├── main.py                     # Entry point for the application
├── runner.py                   # Test runner and execution orchestration
├── channel.py                  # Channel models (e.g., AWGN)
├── codes.py                    # Error-correcting code implementations
├── decoder.py                  # SGrand decoder implementation
├── sgrand_visualization.html   # Visualization output for decoding results
├── traces/                     # Trace files and logs from experiments
│   ├── awgn_monte_carlo.log
│   └── trace_*.json
└── requirements.txt            # Python dependencies

Supported Codes

• Hamming codes (e.g., Hamming(8,4))
• Golay codes (e.g., Golay(24,12))

Installation

[1. Clone the repository:

git clone <repository-url>
cd Learn_SGrand_Decoding

2. Install dependencies:

   pip install -r requirements.txt

3. visualization:

   You can view the decoding process using sgrand_visualization.html.

   Either deploy the HTML file locally, or directly open:

   https://www.jaist.ac.jp/~s2510082/sgrand_visualization.html

   Then load one of the example JSON traces in the traces folder, or generate new traces with the generate_SGRAND_trace function.

Features

• Soft-decision decoding using SGrand algorithm
• AWGN channel simulation
• Monte Carlo simulations for performance analysis
• Query tracing and logging with timestamps
• HTML visualization of decoding results
• Support for multiple error-correcting codes

Output

• Trace logs: JSON files in traces/ directory with format trace_<code>_queries<N>_<timestamp>.json
• Monte Carlo logs: awgn_monte_carlo.log in traces/ directory
• Visualization: Interactive HTML report in sgrand_visualization.html
• Performance metrics: BER, decoding success rates, and query counts

Thanks for Gemini3 Pro, GPT5.1