Inquiry on integrating AI/ML model into cuPHY PUSCH pipeline

Accelerated Computing Aerial Forum
1

Hello Aerial team,

I am currently working on integrating AI channel estimation model into cuPHY PUSCH receiver pipeline. I have followed the provided examples and documentation, however I am experiencing difficulties when converting the PyTorch model into a TensorRT engine and integrating it into the actual cuPHY pipeline. When running with the TrtEngine enabled, the receiver does not work well, even though it shows the engine is successfully loaded.

I have read this document (AI/ML Components for PUSCH Channel Estimation — Aerial CUDA-Accelerated RAN) , but I still need additional information for TensorRT engine integration.

Any guidance would be appreciated.

Thank you

2

Hi @sion3

You can also consider using the Aerial Framework which is released recently.

documentation and tutorials available here:

NVIDIA Aerial™ Framework — Aerial Framework

if you have any questions, please post them here:

NVIDIA/aerial-framework · Discussions · GitHub

Thank you.

3

Thanks for answering. I have one question, since I cannot find exact documentations for integrating cuPHY pipeline in Aerial Framework.. I see a component called AI Channel Filter, does this replace AI Channel Estimator? (Are they the same thing?)