DIGITAL TWINS
In order to achieve our goals, we need high accuracy and high coverage digital twins of the world. At 5G FR2 and even 6G frequencies, the wavelength is in the order of millimeters. Therefore, we need models that are as accurate as possible. Moreover, to perform realistic ray-tracing, the meshes must be subdivided in different materials, i.e., they are semantically classified. They can then be assigned different EM material properties using the ITU databases or by learning the material using the open-source Sionna differentiable ray-tracer. In the European projects we are part of, a very large database of EM measurements are available around the world
Plan A
From large-coverage photogrammetry providers with automatic semantic classification algorithms or AI
As an extension to plan B, either the same or another company can provide data on the exact classification of meshes in the environment. This can start roughly by separating ground from buildings and from foliage, and can extend to advanced machine learning models. Currently, we are using a state-of-the-art ML-based model named SUMS.
Plan B
From large-coverage photogrammetry providers
Using aerial scanning and street-level images, some platforms have succeeded in creating high-fidelity digital twins of the entire world. Some of these provide access via an API or intermediary 3D rendering software such as Cesium, Omniverse, Unreal Engine, ... Tight collaboration is however necessary to improve the quality in order to be used in academic research, and more importantly, semantically classify the meshes into different EM materials.
Plan C
From OpenStreetMap
OSM is completely free and has a high-coverage database of 3D shapefiles of buildings, streets and more. Some buildings have detailed models, built by members of the community, while more ordinary buildings are usually of low Level-of-Detail (a simple footprint with a height). The database comes with crude tags that can semantically classify the urban scene. This database is used very often in literature but lacks detail for ray-tracing simulations, as it is much lower than the wavelength at mmWaves.