Rate-Splitting for Intelligent Reflecting Surface-Aided Multiuser VR Streaming

This is the PyTorch implementation for the deep deterministic policy gradient with imitation learning (Deep-GRAIL) algorithm, with a specific application to solve the sum-rate maximization problem in rate-splitting multiuser systems. The code includes the implmentations of the learning algorithm, deep neural network models, system model for the rate-spliting multiuser systems, as well as a conventional optimization algorithm. If our codes and data are helpful to your research, please kindly cite the paper. The application in the multiuser VR streaming systems will be updated upon final paper acceptance. Please check this page for updates.

Prerequisites

The following libraries are required for this code base. We recommend to use the same versions as listed.

• Python v3.7.9
• PyTorch v1.9.0
• Numpy v1.19.1
• Matlab Engine for Python (R2020a) (Only required for generating the demonstration replay. How to install Matlab Engine for Python)

Usage

1. Generate the demonstration replay by running:

python demonstration_gen.py

Running the script will generate the demonstration replay and store it as NumPy array file (.npy) with names replay_*.npy.

You can skip this step by using the pre-generated demonstration replay (together with the experience replay), which can be downloaded here (File size ~1.5 GB).

2. Start the training algorithm by running:

python main.py

Structures

• main.py: Main training loop and the implementation of Markov decision process (MDP).
• DeepGRAIL.py: The learning algorithm.
• networks.py: The implementation of the deep neural networks.
• utils.py: The implementation of the replay, including adding, loading, saving, and sampling of transition tuples.
• demonstration_gen.py: Method for generating the demonstration replay. The MDP implemented here needs to be the same as main.py.
• opt_algo.m: Matlab script for the conventional optimization algorithm (e.g., alternating optimization (AO) algorithm.

Bibtex

@ARTICLE{rui2023jsac,
  author={Huang, Rui and Wong, Vincent W.S. and Schober, Robert},
  journal={IEEE Journal on Selected Areas in Communications}, 
  title={Rate-Splitting for Intelligent Reflecting Surface-Aided Multiuser VR Streaming}, 
  year={2023},
  volume={41},
  number={5},
  pages={1516-1535},
  doi={10.1109/JSAC.2023.3240710}}