What are the roles of the functional blocks in the AODT diagram?

Accelerated Computing Aerial Forum
1

Dear NVIDIA Aerial Team,

I came across a blog post titled “Developing Next-Gen Wireless Networks with NVIDIA Aerial Omniverse Digital Twin”, where I first encountered the AODT concept.

811×423 160 KB

While exploring the documentation, I had a few questions regarding the functional blocks presented in the diagram from the blog post:

  1. In the diagram, the CUDA-Accelerated RAN block is represented within the AODT architecture. Does this imply that CUDA-Accelerated RAN is fully supported as a software simulation (full simulation)? Based on the documentation, I believe that anything beyond L3 is not considered. Does it represent a specific hardware-accelerated implementation?

  2. The NVIDIA Aerial AI Radio Frameworks block and the AODT block are connected via arrows. Does this connection indicate the exchange of simulated RAN data? If so, could you clarify the purpose of the Real-Time, L1, and Waveform/L1/L2 arrows? What functionality do these arrows represent?

  3. Regarding the Digital Twin RAN block, it seems to represent a software-defined implementation. However, does it require an independent server to operate?

  4. If question 3 is correct, is there a way to deploy the AODT and CUDA-Accelerated RAN blocks on separate hardware systems and interconnect them for operation? We are considering deploying both products.

  5. What does the Simulation Controller represent? Could you provide more details on its role and functionality within the AODT architecture?

Thank you for your time and for providing such insightful resources. I look forward to your clarification!

Best regards,
sjhwang.

2
  1. Yes, the CUDA-Accelerated RAN is fully supported as a software simulation . The same code runs accros both platforms. Anything beyond L2 is not considered. L3 and above may be introduced in future release.
  2. Yes, the connections indicate exchange of simulation data. The arrows show three sources of data that are consumable by Aerial AI radio framework.
    a. Data generated by the Sionna simulator (L1)
    b. Data generated by the aerial software stack and exchanged via the data lake frame work (waveform/L1/L2)
    c. Data generated by the AODT, which is the equivalent of real time data since it is genrated by the same stack that runs the realtime software and accurately models the real deployment geometries (AODT is the twin and strives to be as close to realistic life scenes)
  3. AODT requires two GPUs to run at scale.
  4. AODT is not real time. What is the use case of interconnecting them?