Coordinates#

OWRT uses one set of numerical conventions everywhere: WGS84 geographic input, a topocentric East-North-Up (ENU) frame on the WGS84 ellipsoid for the scene, heights above ground for radio devices, and explicit grid orientation for every array.

The local frame#

LocalFrame.at(longitude, latitude) builds a pyproj pipeline from geodetic coordinates to ENU metres at the given origin. The scene origin is the requested centre for scenes compiled by the loader, the CLI, the explorer and dataset generation. An HTTP simulation without an explicit scene_center uses the first transmitter as the centre.

from openworld_radio_twin import LocalFrame

frame = LocalFrame.at(longitude=4.8993, latitude=52.3762)
east, north, up = frame.to_local(4.9003, 52.3770)
longitude, latitude, altitude = frame.to_geographic(east, north)
corners = frame.corners(-250, -250, 250, 250)   # WGS84 corners of the scene square

A compiled scene restores the same frame with owrt.scene_frame(scene). The provenance file records the origin, axis order, units and the ellipsoidal vertical reference of the frame separately from the terrain source datum.

Heights#

Transmitter and receiver heights are metres above the local ground (AGL). owrt.position adds the compiled terrain elevation at the horizontal position, so a transmitter requested at 25 m stands 25 m above the modelled terrain even on a hillside. With terrain disabled, the ground is the local z = 0 plane.

The local frame’s up component is ellipsoidal and relative to the geodetic origin; it is not an AGL height. scene_z_reference in the provenance says whether scene z = 0 is the modelled terrain at the origin or the flat ground plane.

Array orientation#

Array

Rows

Columns

Note

terrain/*_elevation_m.npy

south to north

west to east

Grid nodes

terrain/surface_classes.npy

south to north

west to east

Cells between nodes

Solver arrays (path_gain, rss, sinr, association)

south to north

west to east

Plot with origin="lower"

Exported PNGs, height map, voxel products

north-up

west to east

Radio PNGs repeat each cell over whole pixels; the others sample cell centres

In memory, RadioMapData carries cell_centers with shape (rows, columns, 3) in ENU metres. The written radio arrays hold values only. array_grid_mapping in radio_map_info.json and in the export’s tx_info.json locates them: the first sample is the south-west cell centre and row_step_m is positive. The grid block in vox_slices/scene_info.json and vox_depth/scene_info.json locates the voxel products with the same first-sample and step keys, north-up: its first sample is the north-west cell centre and row_step_m is negative. The east_m and north_m axes in terrain/scene_info.json locate the terrain arrays. Align products through these records rather than by index.

Scene bounds#

The scene is the square from -radius_m to +radius_m on both horizontal axes. Terrain, surfaces, receiver cells and buildings share this boundary; the WGS84 corners are written to the provenance for GIS tools.