Tutorials#
The tutorials are Jupyter notebooks committed with their outputs; the documentation renders them without execution. Run them yourself after installing the tutorial extra:
conda activate owrt
python -m pip install -e ".[rt,tutorial]"
jupyter lab examples/tutorial_geographic_scenes.ipynb
Load a river-side district of Amsterdam with buildings, terrain and semantic surfaces; inspect materials and provenance; preview in 3D; place a transmitter by coordinates; solve and render a terrain-following radio map; compare materials, terrain and buildings cell by cell; georeference.
The same scene controls with the upstream Sionna preview and the Plotly fallback viewer, followed by a controlled radio-map comparison.
Take the antennas on one LTE 1800 downlink channel from the French register around the Opéra in Paris, see which values are defaults, add them as native transmitters, solve, and compare the registered supports with OpenCelliD’s crowdsourced sites.
Open a packaged radio-map dataset, find radio maps with its Parquet index, plot path gain, RSS, SINR and the serving transmitter, overlay building outlines, show the scene inputs, view the scene in 3D, read the provenance, solve a radio map again, and compare it with the same radio map on flat ground.
Read a package as whole radio maps or one transmitter per sample, select by split and index
query, batch random crops, train a small network on a CPU, and derive SINR and the serving
transmitter from its predictions. Needs the torch extra.
scene.preview() is Sionna’s own ipywidgets and pythreejs viewer. JupyterLab renders it
directly. VS Code must be allowed to fetch the widget’s JavaScript: add
"jupyter.widgetScriptSources": ["jsdelivr.com", "unpkg.com"] to its settings, reload the
window and restart the kernel. scene.render() produces the static figures stored in the
notebooks. Frontends that cannot load jupyter-threejs can use
openworld_radio_twin.plotly_preview.interactive_preview instead.
Benchmark scripts#
The scripts under scripts/benchmarks in the repository time provider acquisition, scene
compilation and a cold load_scene, and compare OpenCelliD site estimates with the ANFR
register (Antennas).