samples_per_src 's impact on accuracy of the PathSolver computation

[clinton-nyobe]

Hello,

I'm pretty new to sionna-rt, I went through the documentation (Technical Report by Aoudia and al) and I saw that samples_per_src was a parameter whose value was having an impact on the accuracy of the PathSolver computation. I actually want to understand better how it actually works and what is the basis for the comparison (we measure accuracy from a baseline right ? which one is it ? how do we measure it ? )

My goal is actually to be able to determine the optimal value that would deliver the most accurate paths computation while saving computational time and memory used.

Please let me know if my understanding is wrong.

Thank you

[merlinND]

Hello @clinton-nyobe,

In principle, the more total samples being computed, the higher the accuracy. For PathSolver, the goal is to find all possible ways to connect the transmitters to the receivers. Depending on the size of your scene, it may take more or less samples to find all specular paths.
If you enable diffuse reflections, then there is generally an infinite number of possible paths but each has a small contribution.

The only exception that I can think of is numerical error due to too many floating point values being accumulated, although that should only really become a problem for radio maps with few cells.