/charm_code

Channel Aware Reactive Mechanism (ChARM)

This repo hosts the source code we have used for the ChARM project [1] at the Institute for the Wireless Internet of Things at Northeastern University. ChARM is a framework for a O-RAN reacting 3GPP base station. Please see [1] for system design and a comprehensive list of features. ChARM leverages machine learning (specifically, a residual neural network) to sense and classify the spectrum usage, and employs customizable policies for dynamically reconfigure the 3GPP compliant network. The goal of ChARM is to intelligently optimize the wireless spectrum usage, avoiding interference and exploiting the available unused bandwidth in mobile O-RAN/3GPP networks.

This repo has the following submodules:

  • srsRAN: a modified version of srsRAN, extended with an API for dynamic cell reconfiguration (branch charm_api). It also sports a colosseum feature branch which allows the execution of srsRAN on the Colosseum wireless emulator. If you want to use it on Colosseum, simply merge the branch colosseum with the branch charm_api.
  • srsRAN_config: a collection of configuration files for srsRAN. Use the charm folder for creating a base station with two cells; this configuration is specifically tweaked for Colosseum.
  • charm_mander: this is the AI-enabled driver for the base station, implementing the intelligence. Refer to its README for further information.
  • charm_mander/charm_trainer: this is the framework for training the machine learning model anew.

Initialize repos

git submodule init
git submodule update
cd charm_mander
git submodule init
git submodule update
cd ..

ChARM work

ChARM stems from a research project at the Northeastern University [1], if you use this code, please cite our work:

@inproceedings{Baldesi2022Charm,
  author = {Baldesi, Luca and Restuccia, Francesco and Melodia, Tommaso},
  booktitle = {{IEEE INFOCOM 2022 - IEEE Conference on Computer Communications}},
  title = {{ChARM: NextG Spectrum Sharing Through Data-Driven Real-Time O-RAN Dynamic Control}},
  year = {2022},
  month = may 
}

ChARM dataset

ChARM machine learning model has been trained with a dataset for training, validation, and testing, which is publicly available [2]. The dataset was collected using Xilinx and USRP SDRs running the OpenWiFi and srsRAN software stacks.

References

  1. L. Baldesi, F. Restuccia and T. Melodia. "ChARM: NextG Spectrum Sharing Through Data-Driven Real-Time O-RAN Dynamic Control", IEEE INFOCOM 2022 - IEEE Conference on Computer Communications, May 2022.
  2. http://hdl.handle.net/2047/D20423481