How to interpret areas of no signal in Radio Maps

[anderspearson206]

I am working with the radio map feature of Sionna, specifically the mesh version. I am computing a radio map where my scene is a terrain mesh, and the computational mesh is just the terrain mesh shifted up (the goal is to see what the signal strength is as you move around the terrain essentially). I have noticed on almost every map, there are areas with no signal (I assume a ray never hits this part of the mesh). I'm wondering how to interpret these areas if I need a map with complete values, even if it is slightly approximated. For reference, after computing the radio map on the mesh, I use the path_gain and cell_centers to convert the radio map into a 2d image (an example is attached below). The areas where no signal is present are white in the image. My issue is mostly with areas where only a few pixels receive no signal, while the rest do. In places where large portions do not receive signal, it makes sense to me to just set these to a low value. But in other areas it seems like they should be getting at least some signal.

The specific settings for the rm computation are below, as well as an example image of a radio map.

rm = rm_solver(scene,
                max_depth = 10, 
                diffuse_reflection=True,
                refraction=True,
                measurement_surface=comp_surface,
                samples_per_tx=10**9)

[oxelarnal]

Hi @anderspearson206,

The white areas just mean no ray reached those cells with the current settings (max depth, reflections, materials, etc.). It doesn’t mean there’s “no signal in reality,” just that the solver didn’t find a path.

If you want fewer white spots, you can:
-Increase max_depth

-Tweak material BSDF (more diffuse reflection)

-Assign a very low RSS value (e.g. −100 dB) to empty cells

Basically, it’s a simulation coverage artifact, not a bug.

[anderspearson206]

Thank you for the reply.

I tweaked some of the settings to generate fewer white spaces (max_depth is now 100, and samples_per_tx is 4e9). Increasing max depth did not seem to effect the results significantly, but increasing the samples did. I tried to increase samples beyond 4e9, but I do not think it will work.

I still need data that is entirely filled. I could do what you recommended and set the value at something like -100dB, or some kind of interpolation to fill. I'm wondering if anyone has any recommendations on what method will provide a more accurate estimation (if there is one). I attached an image below showing raw data, interpolated fill, and a set fill, for both the previous radio map computation settings and the new ones.

[merlinND]

Hello @anderspearson206,

I agree with @oxelarnal's assessment that empty cells is simply due to no paths being found. Another couple of things you could try:

1. If you cannot go beyond 4e9 samples on your hardware, you can start a second simulation with a different random seed and merge the results.
2. Run a separate simulation with larger cell sizes and use those results to fill-in the missing areas. Although this is not very convenient with mesh-based radio maps (you would need to update the measurement surface's geometry) and a similar result can probably be achieved with the interpolation scheme you have already implemented.

[anderspearson206]

Hi @merlinND,

Combining maps with different seeds works great. Here are some example results for anyone else looking to use this method.