Building Aerial gNB using USRP

[vikkicoder]

HI everyone,

In our lab we have two USRP N310s and one B210 with a remote server with A40 GPU. I wanted to know, if it is possible to build Aerial gNB (OAI + Aerial L1) with the current configuration. Please provide necessary guidance regarding this issue. Also, let me know, if any other necessary components are required.

[roytimo]

Hi,

Thanks for the question. I'll need to check with the US team. We'll get back to you.

Are you asking specifically about the L1 from Aerial CUDA-Accelerated RAN (https://github.com/NVIDIA/aerial-cuda-accelerated-ran)?

[vikkicoder]

Thanks for the reply @roytimo. Yes, I would like to know, if L1 (using A40 server) from CUDA accelerated RAN plus OAI upper layers (other servers) can be used to build Aerial gNB with USRP N310 (as it supports only split 8 and not 7.2x split). If there is any other possible way, please let me know.

[roytimo]

Hi @vikkicoder

Thanks for the clarification.

If you would like a complete 5G L1 stack, then you can look at Aerial CUDA-Accelerated RAN platform (https://github.com/NVIDIA/aerial-cuda-accelerated-ran).

The Aerial Framework (this repo) is a new platform for building / prototyping realtime applications. It has a runtime, example PUSCH pipeline, and developer tools. Unfortunately, it is new and does not have a complete L1 stack. There is no USRP support.

So, for your question, perhaps it's good to look at https://github.com/NVIDIA/aerial-cuda-accelerated-ran?

Does this help?

[Ramin-nv-support]

@vikkicoder
Please note that Aerial sdk supports ORAN split 7.2x and therefore not compatible with USRPs that only support split 8.
Also, cuBB installation/performance has not been tested internally on the A40 GPUs yet, so if you choose to run Aerial software on it, it would be experimental.

[vikkicoder]

Thanks for the reply, Please let me know, even in emulator mode (not real time), will the build work with a server containing A40 GPU or a Dell R750 server.