NVIDIA Aerial™ CUDA-Accelerated RAN is a part of NVIDIA AI Aerial™, a portfolio of accelerated computing platforms, software and tools to build, train, simulate, and deploy AI-native wireless networks.

Documentation for AI Aerial™ can be found here.

The following AI Aerial™ software is available as open source:

• NVIDIA Aerial™ CUDA-Accelerated RAN (this repository)
• NVIDIA Aerial™ Framework

Updates on new software releases, NVIDIA 6G events and technical training for AI Aerial™ are available via the NVIDIA 6G Developer Program.

The Aerial CUDA-Accelerated RAN SDK includes:

• GPU-Accelerated 5G PHY (cuPHY): CUDA-based physical layer processing for 5G NR including channel coding (LDPC, Polar), modulation/demodulation, MIMO processing, and channel estimation
• GPU-Accelerated MAC Scheduler (cuMAC): High-performance L2 scheduler acceleration for resource allocation and scheduling
• Python API (pyAerial): Python bindings for AI/ML research and integration with frameworks like TensorFlow and Sionna
• 5G Reference Models (5GModel): MATLAB-based 5G waveform generation and test vector creation based on 3GPP specifications
• Containerized Environment: Docker-based development and deployment with pre-built containers

aerial-cuda-accelerated-ran/
├── cuPHY/              # CUDA-accelerated Physical Layer (L1)
├── cuPHY-CP/           # Control Plane and integration components
│   ├── aerial-fh-driver/    # Fronthaul driver for O-RAN interfaces
│   ├── cuphycontroller/     # PHY controller
│   ├── cuphydriver/         # PHY driver
│   ├── cuphyl2adapter/      # L2 adapter
│   ├── ru-emulator/         # Radio Unit emulator
│   ├── testMAC/            # Test MAC implementation
│   └── container/          # Container build scripts and recipes
├── cuMAC/              # CUDA-accelerated L2 Layer
├── cuMAC-CP/           # MAC Control Plane components
├── pyaerial/           # Python API and ML/AI tools
├── 5GModel/            # TV generation for cuPHY and cuBB verification
├── testBenches/        # Test benches and performance measurement tools
├── testVectors/        # Test vectors for validation
└── cubb_scripts/       # Build and automation scripts

# Clone repository
git clone https://github.com/NVIDIA/aerial-cuda-accelerated-ran.git --recurse-submodules
cd aerial-cuda-accelerated-ran

# Enable git LFS (if needed for large files)
git lfs install
git lfs pull

# Pull the Aerial container from NGC
docker pull nvcr.io/nvidia/aerial/aerial-cuda-accelerated-ran:25-3-cubb

# Start interactive development container
./cuPHY-CP/container/run_aerial.sh

# Inside container: Build SDK
./testBenches/phase4_test_scripts/build_aerial_sdk.sh

• Container versions available at NVIDIA NGC

Visit the full documentation at NVIDIA Docs Hub

• Aerial is not accepting contributions at this time.

• Vulnerability disclosure: SECURITY.md
• Do not file public issues for security reports.

• Level: Maintained
• How to get help:
  • File issues on GitHub for bugs and feature requests
  • Join discussions for questions and community support

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.

Note: Some dependencies may have different licenses. See ATTRIBUTION.rst for third-party attributions in the source repository.

If you use NVIDIA Aerial™ CUDA-Accelerated RAN in your research, please cite:

@software{nvidia_aerial_cuda_accelerated_ran,
  title = {NVIDIA Aerial™ CUDA-Accelerated RAN},
  author = {NVIDIA Corporation},
  year = {2025},
  url = {https://github.com/NVIDIA/aerial-cuda-accelerated-ran}
}