[OutdoorNoiseRenderer] Simulation output artifacts when performing TemporalVariations turbulence filtering
BRANCH: https://git.rwth-aachen.de/ita/VACore/-/tree/feature/outdoornoise_with_turbulence
Objective and introduction:
Implementation of the IVATurbulenceFilter
and its child classes (CVATemporalVariations
and CVAHighShelfFilter
) in the OutdoorNoiseRenderer
(ON) paradigm.
This was done in order to introduce the possibility of simulating atmospheric turbulence effects in a given simulation using this renderer.
The IVATurbulenceFilter interface and child classes were originally designed for the AirTrafficNoiseRenderer
(ATN).
The ON renderer had already a framework prepared for the possibility of adding further filtering to the simulation. This made the implementation a fairly simple process.
Output and issues:
-
HighShelf filtering -> The output features filtering as expected, similarly to what was obtained using the ATN renderer: a somewhat plausible result, with some roughness due to the sudden filter change. This roughness is seen as a fixable issue, but it wasn't approached since it wasn't seen as a priority at the time of this implementation. It is also an issue observed in the ATN renderer.
-
TemporalVariations filtering -> The output filtering is also in line with the results from the ATN renderer. Some artifacts were observed: a regular "clicking" noise persists throughout the whole audio.
Isolating the issue:
It became important to determine whether the issue sourced from IVATurbulenceFilter
and its child classes. The filter sets generated by the turbulence filtering classes were recorded over the simulation and compared. The filter sets recorded in both renderers were identical, for each of the respective filtering methods.
Then, both renderers were used to process data from the same aircraft trajectory and receiver position. Doppler delays and spreading loss effects were taken out of the simulation, in order to better isolate the issue.
The "clicking" artifacts remained for the TemporalVariations
filtering with ON rendering.
Next steps:
At the moment, we believe the problem could be related to the way IIR filter coefficients are being generated in this renderer.
This somewhat sensitive process could be raising issues with the generated TemporalVariations
turbulence filters.
A possible way to either solve or isolate this problem would be to import the IIR filter generation method from the ATN renderer into the ON renderer.
This seems reasonable, since no artifacts are found when rendering via ATN renderer. Nevertheless, there is still no guarantee that this would solve the issue.
Furthermore, the two renderers generate IIR filtering in a very diferent fashion: the classes and data structures employed in each renderer differ greatly; hence, the previously described approach is rather complex and was not yet followed due to time constraints.