Analytical antenna based 3GPP propagation models for 3G, 4G and 5G

Our current version of the algorithm is based on an analytical 3GPP propagation model.

In a nutshell, it takes into account :
- the mobile technologies (3G, 4G, 4G+ and 5G), 
- the different frequencies of the broadcasting antennas, 
- the density of population (urban, suburban, rural areas).

The model is calibrated each year thanks to 25k mobile performance tests  that the regulator ARCEP performs on the field.

From an analytical model to semi-deterministic model

3D model and topography - Taking into account the Line of Sight (LoS)  in an analytical model

Next version of the algorithm includes topography in the model. Moving from 2d to 3d distances and also checking whether or not there’s LoS (Line of Sight) between a given point and the antenna.

When: “LoS maps” will go live by the end of November 2023.

Building our own deterministic Ray-tracing model for high density areas

Meanwhile our R&D team is also partnering with the French research institute INRIA to implement a deterministic model based on Ray Tracing. This model is aimed to build a precise QoS map in the high density areas where the analytical model can encounter some limits.

Latest published paper dealing with our research : https://www.sciencedirect.com/
When: 2024

Calibration of the analytical model and measuring its accuracy

Calibration of the model is critical

Our algorithm needs to be fed with real data measurements in order to be more accurate. The model is calibrated each year thanks to 25k mobile performance tests  that the regulator ARCEP performs on the field.

Each configuration  “Operator/Technology/Type of zone” has its own calibration factor that is calculating by minimizing the root mean square error or RMSE.

Measuring the accuracy of the model

There are 2 ways to determine the accuracy of each version of the algorithm:
1 - calculating the difference of rankings of the operateur between the model (Sog) and the real speedtests (Arcep), using MAE or Hamming distance
2 - calculating the “error” between  the model results in Mb/s with the real speedtests on the field