The analysis of physical model used in simulation and source code

[wj_zhou]

Does NVIDIA Omniverse provide analysis of the physical models used in simulations, as well as access to the source code? I notice that Sionna’s documentation offers detailed analysis of propagation physical models and API parameter breakdowns.

[rfarhoodi]

Hi @wj_zhou
For detailed AODT features and capabilities, please refer to the AODT 1.2.0 documentations here:
Aerial Omniverse Digital Twin 1.2

[kpasad1]

@wj_zhou

1. What is your use case?
2. The Source code for the electromagnetic solver (EM Solver) is not available. EM solver is packaged as a binary.
3. The API to call the EM solver is available and defined in the documentation
   See : Additional Information — Aerial Omniverse Digital Twin.
4. You can call the emsolver from your application if you need. You can set the material properties and and diffusion properties via the stage view.
5. AODT and Sionna are diffrent products and targeted towards diffrent applications. The underlying EM solver models could be diffrent. If you can share your use case, we can assist with finding the right product.

[wj_zhou]

Thank you for your response.

The questions may be relevant to our research. We think that when the scenario setup is entirely identical (e.g., UE positions, mobility patterns, and RU placements), repeated experiments under noise-free conditions should yield identical CFR results under ideal conditions. However, we observed that the scattering model referenced in Omniverse incorporates a random phase component, which may cause variations in the CFR across simulation runs. Additionally, other stochastic factors might also contribute to this variations. I am uncertain whether our previous assumptions are correct. Though, we want to identify when and where random variables are introduced during CFR data generation or whether we can control these random variables in Omniverse, which may be useful for our research.

To investigate this, we conducted the following tests:

1. Enabled the Enable Seeded Mobility property in the Scenario, set the Mobility Seed to 0, and maintained all other parameters consistent with the Installation → Validation section of the documentation.
2. Repeated multiple simulations in EM Simulation Mode, storing the CFR results from each run in separate databases.

Results: The CFR data differed across all simulation runs. This raises the following questions:

a. With the above configuration, was the scenario layout truly identical for every simulation run?
b. If the answer to (a) is “yes”, does the observed inconsistency in CFR data stem from stochastic components embedded in the underlying physical model (e.g., random phase in scattering)?
c. Is there a corresponding random seed to control the stochastic components in these underlying physical models?

[wj_zhou]

1. For the difference between two CFR measurements:
   After calculating the magnitude of the difference between the two measured CFRs, averaging the magnitudes, and then taking the natural logarithm (ln). We obtain four measurement results of the cfrs, the difference between the first cfr and the remaining cfrs are [-15.22517488 -18.28014752 -15.22566514].

2. For a single CFR measurement:
   After calculating the magnitude of one CFR measurement, averaging the result, and applying the natural logarithm (ln), the four measurement results are [-4.8904623 -4.89048654 -4.8904623 -4.89048653] , respectively.

Based on the observed results, could this difference also be attributed to computational errors introduced during each simulation run?

Additionally, I have another question.
When calculating the CFR , given that the number of subcarriers is N and the number of raypaths is K, which of the following methods is used?

1. Each subcarrier signal independently calculates its amplitude and phase based on ray-tracing (requiring N×K ray-tracing computations in total).
2. First, calculate the amplitude and phase for the center frequency signal, then let all subcarriers share the same amplitude, while compensating the phase based on the frequency difference between the subcarriers and the center frequency, as well as the delay of the raypaths (requiring only K ray-tracing computations in total).
3. Other methods.

[kpasad1]

@wj_zhou If the seeded mobility is enabled, the CFRs should be identical. Can you compare the contents of the database and ensure that all the database tables are identical? A quick way to identify the diffrence is to use checksum to compare the databases. I will share more information for the second part soon

[kpasad1]

@wj_zhou Are you able to find the cause of diffrence?

[wj_zhou]

I enabled seed mobility, saved the results in compare_cfr1 , then re-clicked “Generate UE->copy” and copied the results to compare_cfr2 .

I compared the cheksum but found them different. I generated CFR data multiple times, but the checksum results were different each time.

Is there an issue with my operations?

[kpasad1]

@wj_zhou Minor fixed point differences in CFRS are expected. Can you please share the cfrs from the two instances? A partial list is sufficient.

[wj_zhou]

These are two CFRs from the two instances at the same time slot.

[kpasad1]

@wj_zhou This seems to be a fixed point discrepancy. I will recreate this deployment and give you an update on what is the expected diffrence.

[wj_zhou]

Thank you for your help!