Welcome to the Geo2SigMap, this is the first work that:

• Designs an automated framework that integrates open-source tools, including geographic databases (OSM), computer graphics (Blender), and ray tracing (Sionna), and supports scalable ray tracing and RF signal mapping at-scale using real-world building information;
• Develops a novel cascaded U-Net architecture that achieves significantly improved signal strength (SS) map prediction accuracy compared to existing baseline methods based on channel models and ML.

1. Overview
2. Repo Structure
3. Installation
4. Example Usage
5. License

Geo2SigMap is an efficient framework for high-fidelity RF signal mapping leveraging geographic databases, ray tracing, and a novel cascaded U-Net model. Geo2SigMap features an automated pipeline that efficiently generates 3D building and path gain (PG) maps via the integration of a suite of open-sourced tools, including OpenStreetMap (OSM), Blender, and Sionna. Geo2SigMap also features a cascaded U-Net model, which is pre-trained on pure synthetic datasets leveraging the building map and sparse SS map as input to predict the full SS map for the target (unseen) area. The performance of Geo2SigMap has been evalauted using large-scale field measurements collected using three types of user equipment (UE) across six LTE cells operating in the citizens broadband radio service (CBRS) band deployed on the Duke University West Campus. Our results show that Geo2SigMap achieves significantly improved root-mean-square error (RMSE) in terms of of the SS map prediction accuracy compared to existing baseline methods based on channel models and ML.

If you find Geo2SigMap useful for your research, please consider citing:

@article{li2023geo2sigmap,
  title = {{Geo2SigMap}: High-Fidelity {RF} Signal Mapping Using Geographic Databases},
  author = {Li, Yiming and Li, Zeyu and Gao, Zhihui and Chen, Tingjun},
  journal={arXiv:2312.14303},
  year={2023}
}

This is an active project, if you are interested to have a community discussion, please start a new discussion thread in the discussion tab and we will get back to you as soon as possible.

We provide detailed guidelines for installation including a docker image (for quick try) and a clean install and.

Note: the docker way may suffer from performance loss due to xvfb virtual framebuffer.

Run the following command to use our pre-compiled docker image:

docker run --name g2s -it ffkshakf/geo2sigmap:latest bash

We only provide an amd64 arch docker image. If you run on a Apple Silicon or ARM64, you can add --platform linux/amd64 to run it under emulation.

docker run --name g2s --platform linux/amd64 -it ffkshakf/geo2sigmap:latest bash

Note: The following commands have been tested on a clean install Ubuntu:22.04, you can follow the Blender offcial document about compiling Blender here if you would like to run on other OS/ARCH.

sudo apt update
sudo apt install python3 python3-pip git
pip install --target=/usr/lib/python3/dist-packages Cython

mkdir ~/blender-git
cd ~/blender-git
git clone --depth 1 --branch v3.3.1 https://projects.blender.org/blender/blender.git
cd blender
git switch -c v3.3.1_geo2sigmap
perl -i -pe 'BEGIN{undef $/;} s/  bool is_output_operation\(bool \/\*rendering\*\/\) const override\n  \{\n    if \(G.background\) \{\n      return false;\n    \}\n    return is_active_viewer_output\(\);\n  \}/  bool is_output_operation\(bool \/\*rendering\*\/\) const override\n  \{\n    return is_active_viewer_output\(\);\n  \}/sm' source/blender/compositor/operations/COM_ViewerOperation.h
perl -i -pe 'BEGIN{undef $/;} s/  bool is_output_operation\(bool \/\*rendering\*\/\) const override\n  \{\n    return !G.background;\n  \}/  bool is_output_operation\(bool \/\*rendering\*\/\) const override\n  \{\n    return true;\n  \}/sm' source/blender/compositor/operations/COM_PreviewOperation.h
perl -i -pe 's/(_src=\$SRC\/)USD-\$USD_VERSION/$1OpenUSD-\$USD_VERSION/' ./build_files/build_environment/install_deps.sh
perl -i -pe 'BEGIN{undef $/;} s/(patch -d \$_src -p1 < \$SCRIPT_DIR\/patches\/usd\.diff\n\s+fi\n\n\s+cd \$_src)/$1\n\n    sed -i.bak  -e '\''s\/\.if !defined\.ARCH_OS_WINDOWS\.\/#if 0\/'\'' -e '\''s\/\.if defined\.ARCH_COMPILER_GCC\..*\/#if 0\/'\'' -e '\''s\/defined\.ARCH_COMPILER_CLANG\.\/\/'\'' -e '\''s\/\.if defined\.ARCH_OS_LINUX\.\/#if 0\/'\'' -e '\''s\/\.if !defined\.ARCH_OS_LINUX\.\/#if 1\/'\'' pxr\/base\/arch\/mallocHook.cpp/' ./build_files/build_environment/install_deps.sh

bash ./build_files/build_environment/install_deps.sh --with-embree --with-oidn

When the script complete you will see some thing like the follow:

Or even simpler, just run (in your blender-source dir):
  make -j20 BUILD_CMAKE_ARGS="-U XXXXXXXXXXXXXXXX

Or in all your build directories:
  cmake -U *SNDFILE* -U PYTHON* -U XXXXXXXXXXXXXXXX

Copy the first section and add an "release" between "make" and "-j" like the following ones and then execute it.

make update
make release -j20 BUILD_CMAKE_ARGS="-U XXXXXXXXXXXXXXXX

There are two add-ons, Blosm and mitsuba-blender. Download the zip file and place them in the root of this project */geo2sigmap/.

unzip mitsuba-blender.zip
perl -i -pe 's/  result = subprocess\.run\(\[sys\.executable, '-m', 'ensurepip'\], capture_output=True\)\n\s+return result\.returncode == 0/return True/' mitsuba-blender/__init__.py
zip -r -0 mitsuba-blender.zip mitsuba-blender

Please follow Sionna's official document to install Sionna.

Please follow Pytorch's official document to install PyTorch.

Simply run:

python3 gen_data/Step1_OSM/OSM_from_generate_nc_dataset.py

The above command will download the Duke Campus surrounding region's OSM XML file with the setting of 512m area dimension, 0.2 building to landing threshold.

You can check the detail option by the -h arguments.

Note: The public OSM sever have a query limitation around 2-10 query/second, so if you want to achieve a faster process speed, consider deploy a self host OSM server following the OSM offcial document here. A reasonable speed of self hosted server would be around 100-200 query/second on a SSD computer.

Distributed under the APACHE LICENSE, VERSION 2.0