Repository navigation
PathSolver with diffraction=True not reproducible between runs #1164
Closed
Wireless-cmyk
started this conversation in
General
Replies: 1 comment
|
Hello @Wireless-cmyk, Non-deterministic results are expected depending on the total number of paths generated and the output buffer size, determined by Closing as duplicate of #851. |
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Hi,
I am observing that PathSolver with diffraction=True does not produce consistent valid path counts between back-to-back calls with identical parameters and seed.
Setup:
Observed with diffraction=True:
Valid path counts change by ±1 between runs for several UEs. Channel response differences are a direct consequence. Increasing samples does not help — 10M samples produced larger errors than 1M, because more samples find more diffracted borderline paths that can flip between runs.
Observed with diffraction=False at 2M samples:
Zero path count differences across all 10 UEs across all runs. Perfect reproducibility.
This rules out buffer overflow as the cause — increasing samples while disabling diffraction achieves perfect reproducibility, while the issue persists with diffraction enabled regardless of sample count.
Is this behavior expected with the current diffraction implementation, and does the upcoming deterministic path generation algorithm address valid path count instability with
diffraction enabled?
Thank you.
All reactions