NVIDIA Aerial AI Radio Frameworks

NVIDIA Aerial AI Radio Frameworks provide a package of AI enhancements to enable training and inference in the RAN. The framework tools—pyAerial, NVIDIA Aerial™ Data Lake and Sionna—span the research space from AI and machine learning (AI/ML) algorithm exploration, training, and inference to simulation and real-time implementation in a GPU-accelerated, over-the-air network testbed such as NVIDIA Aerial RAN CoLab Over-the-Air (ARC-OTA).

Join the NVIDIA 6G Developer Program to access the tools and documentation.

Join 6G Developer Program

Aerial AI Radio Frameworks - a neural receiver built with pyAerial

Aerial AI Radio Frameworks - a neural receiver built with pyAerial

Framework Tools

With Aerial Data Lake, capturing data in a real-time, over-the-air (OTA) testbed has never been easier. Then, pyAerial transforms an Aerial Data Lake database into training data tailored to your machine learning task. pyAerial is also a tool for verifying NVIDIA® TensorRT™ GPU-optimized inference engines from end to end. Sionna is an open-source library for 6G physical-layer research with native AI support.

All of the tools and libraries are available as source code, either as standalone (e.g. Sionna) or bundled with Aerial CUDA-Accelerated RAN (e.g. pyAerial) so can be customized to your specific requirements.

Python Layer-1 API

pyAerial provides a Python interface to the NVIDIA cuBB layer-1 data plane functions. Used with the Aerial Data Lake capture platform to produce training data for layer-1 functions, pyAerial can also be used to evaluate the end-to-end performance of neural network physical layer functions.

Data Collection

Generate over-the-air training data with the Aerial Data Lake data collection platform. A data collection app runs on the distributed unit (DU), writing radio frequency (RF) samples to the database. Aerial Data Lake provides APIs to access the data. Used in conjunction with pyAerial, it generates datasets for intermediate nodes in the cuBB layer-1 signal processing pipeline to train neural networks for channel estimation, equalization, soft-demapping, and more.


Sionna is a GPU-accelerated open-source library for link-level simulations. It enables rapid prototyping of complex communication system architectures and provides native support for the integration of machine learning in 6G signal processing.


Join the 6G Developer Program

Developers and researchers in the 6G Developer Program get access to software, documentation, and resources.

Join Now