Exports#

The 16 most recent simulations run through the explorer or the HTTP API can be exported as a single-scene bundle. Unlike a dataset scene, which stores only the packed path gain, the bundle holds the full float32 radio arrays and the serving-transmitter index. It keeps raw solver values and display renderings strictly apart.

artifact/
|-- arrays/                raw radio fields and the serving-transmitter index
|-- images/                georeferenced path gain, RSS, SINR and association PNGs
|-- mesh/                  portable RF geometry (binary PLY)
|-- terrain/               source/model/local elevation, surface classes, provenance
|-- measurement_surfaces/  terrain-following receiver meshes keyed by cell size and AGL
|                          (no building rasters, which only dataset scenes hold)
|-- vox_slices/            sparse voxels and receiver-height occupancy
|-- vox_depth/             maximum-z arrays, 16-bit depth, preview, metadata
|-- 2D_Building_Height_Map.npy  north-up building height raster
|-- provenance.json        coordinate, geometry, material and bounds records
|-- scene.xml              Sionna RT / Mitsuba scene
`-- tx_info.json           simulation, geospatial, solver and export metadata

In the explorer, the download button fetches GET /api/simulations/{id}/artifacts.zip. From Python, openworld_radio_twin.artifacts.build_artifact_bundle produces the same ZIP from a request, its response, the RadioMapData, the building mask of the radio grid and the scene’s EnvironmentData.

Radio products#

Path gain

Received signal strength

Boston path-gain export

Boston RSS export

Channel path gain in dB before transmit power.

Received power in dBm after transmit power.

SINR

Transmitter association

Boston SINR export

Boston association export

Signal-to-interference-plus-noise ratio in dB.

Discrete serving-transmitter index.

Unmodified exports from one Boston Sionna RT run with three 3.5 GHz sector transmitters, 1 m cells and a 1.5 m receiver plane.

  • Arrays are linear float32, shape (n_tx, rows, columns), rows south to north. Path gain and SINR are ratios; received power is in watts. A cell without coverage holds 0; the arrays never hold NaN or infinite values. array_grid_mapping in tx_info.json locates them: its first sample is the ENU centre of the south-west cell, and row_step_m is positive.

  • PNGs are north-up, with a whole number of pixels per solver cell recorded as pixels_per_solver_cell, transparent where there is no data, and never spatially interpolated. Their metadata records the source grid, units, display range and integer display scale. The association PNG is rendered directly from integer labels.

  • arrays/association/association.npz holds those labels as int32 with shape (rows, columns) on the same grid: the index of the serving transmitter under the request’s association metric, -1 where no transmitter has a positive value.

  • SINR is computed per cell from the received powers and the scene’s thermal noise power; the bandwidth and temperature are recorded, and for Sionna runs the noise power.

Geometry products#

Receiver-height voxel slice

Voxel z-depth preview

Voxel occupancy slice

Maximum occupied voxel elevation

Binary occupancy at the receiver height on the voxel grid.

Maximum occupied voxel elevation per column; colour is a linear preview of metres above ground.

  • mesh/ holds the binary PLY meshes referenced by scene.xml: buildings with native roofs where the source provides them, and the ground, vegetation, paved and water surfaces.

  • 2D_Building_Height_Map.npy is a north-up building height raster.

  • vox_slices/ contains the sparse voxel volume and the receiver-height occupancy slice.

  • vox_depth/ contains the lossless float32 maximum-z array in metres above ground, a linearly decodable 16-bit depth PNG, a preview PNG for visualisation only, and metadata. Slice and depth products share one XY grid, pitch, bounds and orientation. They and the height map sample buildings at cell centres, as the building mask of the radio grid does; both scene_info.json files record the voxel grid in the same grid block (Scene layout), and tx_info.json repeats it as postprocess.slice_grid and postprocess.voxel_depth.grid.

Metadata#

tx_info.json records the simulation request, every transmitter’s geographic and local position, the solver details listed in Sionna RT, the building, terrain and surface provenance, the coordinate frame, the array grid, the path, key, unit and data type of each radio array, the height map’s path, data type, value, row order and pixel size, and the export parameters. Together with terrain/scene_info.json and provenance.json it is sufficient to reconstruct the scene and re-run the simulation.