DevLab

Windows lab for poking at 5G/6G PHY stuff using NVIDIA AI Aerial.

I put this together because a lot of people wanting to get into Aerial / 6G development get stuck at the same place: the docs assume you already know the jargon, Docker + NGC + GPU setup is annoying, and there's no obvious "open this and just try something."

DevLab is the front door. Type one command, get Docker going, open a browser simulator, and walk through a long tutorial if you've never touched radios before.

It's a learning sandbox. Not a base station. Not a full 3GPP test bench.

---

What you need

• Windows 10 or 11
• Docker Desktop (WSL2 backend)
• NVIDIA GPU + driver if you want GPU acceleration
• NGC account so you can pull the Aerial image: https://ngc.nvidia.com

---

Setup

Clone the repo somewhere, then:

cd tools
Install-DevLab.cmd

That adds DevLab to your PATH. Open a new terminal after it finishes.

Pull the image (once):

docker login nvcr.io
docker pull nvcr.io/nvidia/aerial/aerial-cuda-accelerated-ran:26-1-cubb

Run it:

DevLab

UI lands on http://localhost:8765 by default.

The welcome screen uses whatever name Windows has for the logged-in user. If that's wrong or you want something else for a session:

set DEVLAB_NAME=Sam
DevLab

---

Commands

DevLab                 start lab, offer tutorial
DevLab -Quick          start lab, skip tutorial
DevLab -TutorialOnly   just the tutorial
DevLab -Status         is docker / UI / gpu up?
DevLab -Stop           kill the lab containers

Shell into the full Aerial image (if that container came up):

docker exec -it aerial-ran bash

If the page won't load:

DevLab -Status
docker logs aerial-5g6g-sim

---

What's in here

tools/         DevLab launcher + installer
sim/           link simulator (web UI + PHY code)
containers/    scripts for Aerial / Jupyter containers
notebooks/     notebook experiments
repos/         optional local clones of Aerial source
setup/         machine bootstrap helpers
.env.example   copy this, don't commit real keys

The simulator lets you mess with SNR, MCS, channel type, PRBs, etc. and see constellation, resource grid, BER/BLER, throughput, SNR sweeps. Code is under sim/.

There's also an interactive tutorial baked into DevLab that explains the radio chain from scratch (bits → coding → modulation → channel → decode). It's long on purpose.

---

Aerial images

What this usually pulls / uses:

Image	Notes
nvcr.io/nvidia/aerial/aerial-cuda-accelerated-ran:26-1-cubb	main one (cuPHY / pyAerial side of things)
nvcr.io/nvidia/aerial/aerial-framework-base:v0.1.0	optional, framework work

Docs if you want to go deeper than the UI:

• https://docs.nvidia.com/aerial/
• https://catalog.ngc.nvidia.com/orgs/nvidia/aerial
• https://developer.nvidia.com/aerial

---

Don't commit secrets

.env is gitignored. Use .env.example as a template for NGC keys and friends. If you paste a real API key into the repo, that's on you.

Paths in the launchers are relative to the clone location so you can drop this on another machine without editing someone else's username into the scripts.

---

Upstream

Anything under repos/ keeps its own NVIDIA / upstream licenses. This repo is just the lab wrapper + educational simulator around that stack.