GCN-WMMSE

This library contains the code for the research paper Coordinated Multicell MU-MIMO Beamforming Using Deep WMMSE Algorithm Unrolling.

Requirements

Python 3.9
PyTorch 1.10.0 (newer version might lead to errors)
tensorboard 2.7.0
numpy 1.21.4
scipy 1.8.0
matplotlib 3.5.0
pandas 1.3.4

Description

Quickstart

An example script to perform a training run for a GCN-WMMSE network and a validation on a test set is provided in main.py.

Structure

main is an examplary script to use the library.
comm\trainer contains the class GeneralTrainer which abstracts model creation, training with checkpoints, model validation. See the docstring of the class constructor for possible arguments. Call the methods run_learning_upto_step(max_num_training_steps) to run training for max_num_training_steps and evaluate_on_test_set(test_data_path) to evaluate the model.
comm\gcnwmmse supplies the architecture of this work.
comm\algorithm contains various implementations of the WMMSE algorithm and various helper functions to generate baseline data.
comm\reference_wmmse_unrolls supplies extended PyTorch implementations of architectures reported in reference works.
comm\architectures supplies get_model(model_name, *args, **kwargs) to return any architecture of this library.
comm\channel supplies functions to generate wireless scenarios (mimoifc_randcn, mimoifc_triangle, siso_adhoc_2d, deepmimo) and to compute the achievable downlink rate given some beamformers.
comm\lossfun supplies loss functions for training.
comm\mathutil supplies various helper functions used throughout the rest of the library.
comm\network supplies a general GCN implementation.
datawriter contains classes to export data to CSV-format.
data serves as run and data directory for the example script.

DeepMIMO

The code supports the DeepMIMO data set. The script deepmimo_channel_generation generates pickled channel matrix data that can be imported by comm\channels.deepmimo. The DeepMIMO package must be installed for the script. After creation of the channel matrix data, use the deepmimo data type/channel type and supply the path to the channel data to generate a scenario batch that can be used for training or testing.

Usage

Please cite the paper L. Schynol and M. Pesavento, "Coordinated Sum-Rate Maximization in Multicell MU-MIMO with Deep Unrolling," in IEEE Journal on Selected Areas in Communications, (doi: 10.1109/JSAC.2023.3242716) if you apply the provided code in you own work. If you use the reference architectures, please cite the respective works as well.

References

This paper provides extended PyTorch implementations for the algorithms and architectures of the following works:

Shi et al. - 2011 - An Iteratively Weighted MMSE Approach to Distributed Sum-Utility Maximization for a MIMO Interfering Broadcast Channel
Chowdhury et al. - 2020 - Unfolding WMMSE using Graph Neural Networks for Efficient Power Allocation
Pellaco et al. - 2020 - Iterative Algorithm Induced Deep-Unfolding Neural Networks: Precoding Design for Multiuser MIMO Systems
Hu et al. - 2021 - Iterative Algorithm Induced Deep-Unfolding Neural Networks: Precoding Design for Multiuser MIMO Systems

lsky96 / gcnwmmse