About

This repository implements the results found in Reinforcement Learning for Channel Coding: Learned Bit-Flipping Decoding

Preliminaries

Create a venv:

1. pip install virtualenv
2. python -m venv env
3. source env/bin/activate

Download requirements.txt:

1. pip install -r /path/to/requirements.txt

Files

• main.py driver code to train/evaluate the agent and compare results with benchmark
• bf_env.py is an OpenAI gym environment for bit flipping environment
• sarsa.py runs sarsa algorithm using e-greedy policy
• bf_decoding.py (old)
• channel.py models AWGN and BSC channels
• codes.py is used to load generator and parity check matrices to define the type of code the agent will be decoding

Folders

• /Hmat holds generator and parity check matricies for codes in .mat format; see here
• /MATLAB runs benchmark decoding algorithms in matlab
• /policies saves the Q table for a specific code after training the agent using sarsa (too large for repository)
• /benchmark holds .mat files for BER using MATLAB decoding algorithms
• /figs figures from training and BER comparison