Radio planning · NVIDIA Sionna

See your 4G/5G coverage before you install a radio.

We use NVIDIA Sionna, an open-source, GPU-accelerated wireless simulator, to ray-trace radio signals through a 3D model of your site. You get coverage maps, dead zones and expected data rates for labs and 4G/5G radio projects before you buy or mount any hardware.

2-minute explainer, no sound · tap the full-screen icon to watch large

What Sionna is

Sionna is a library from NVIDIA for wireless research. It has three parts, and we use all of them.

Sionna RT

Ray-traced propagation

Follows signals from each cell through a 3D scene, including reflections off walls and buildings, and turns them into coverage maps.

Sionna PHY & SYS

4G/5G link and system simulation

Simulates the 5G waveform bit by bit, so a coverage map can be turned into error rates and data rates.

GPU-accelerated

Fast on everyday hardware

Runs on NVIDIA RTX cards. Our lab runs it on a 12 GB RTX 3060.

Why it’s AI-driven

Sionna was built for machine learning as well as for maps. That changes what we can do for you.

Differentiable ray tracing

Every result has gradients, so an optimiser can move antennas and adjust tilt and power to improve coverage automatically.

Neural receivers

Train and test AI channel estimators and detectors for 5G, for research labs that want to go further.

Synthetic training data

Generate realistic channel datasets for training AI models without weeks of field captures.

Calibrated to your site

Learn wall and material properties from measurements, so predictions match what your radios see.

Real output from our lab

Everything below was produced with Sionna on the CelTech lab machine.

3D radio maps

3D city model with a ray-traced radio map; signal from a red transmitter dot spreads down the streets while buildings cast radio shadows — **Street canyons.** Signal from the transmitter (red dot) runs down the streets; buildings cast radio shadows.

Ray-traced radio map over the Munich scene; the open square around the transmitter is well covered and surrounding streets are not — **Munich scene.** The open square is well covered; the streets behind the buildings are not.

Compare spectrum on the same site

Received signal strength map at 3550 MHz, 1.5 m height, 10 dBm transmitter — 3.55 GHz · 5G band n78Received signal strength at 1.5 m height from a 10 dBm transmitter.

Received signal strength map at 1842.5 MHz, 1.5 m height, 10 dBm transmitter — 1842.5 MHz · LTE band 3Same site and transmitter, for choosing the band that suits the site.

From signal quality to data rate

Block error rate versus Eb/N0 for 5G NR uplink at MCS 6, 13, 21 and 27 — **5G NR uplink, LDPC coded.** Rates run from 14 Mbit/s (QPSK) to 84 Mbit/s (64-QAM). The fastest scheme needs about 10 dB more signal than the most robust one to reach the 10% error target.

Uncoded bit error rate curves for QPSK, 16-QAM, 64-QAM and 256-QAM — **Four modulations compared.** About 100 million bits simulated in 4.3 seconds.

Inside the 5G waveform

16-QAM constellations received at 25, 15 and 8 dB SNR, getting noisier as SNR drops — **16-QAM at 25, 15 and 8 dB SNR.** The points spread out as the signal weakens, which is what drives errors at the cell edge.

5G resource grid: 14 OFDM symbols by 612 subcarriers with pilots in the DMRS column — **One 5G slot.** 14 OFDM symbols × 612 subcarriers, with pilot (DMRS) symbols in one column.

What you get from a Sionna study

For labs and 4G/5G radio projects, each study answers practical questions about the site.

Output	What it tells you
Coverage and signal strength map	Expected signal for every part of the site, at the height people and devices actually are.
Dead zones	Where coverage drops out, found before any hardware is installed.
Multipath detail	Path delays, delay spread and angles of arrival, which matter in factories and other metal-heavy spaces.
Interference and SINR	How several cells, or a neighbour’s network, affect each other.
Expected data rates	Which modulation and coding each area supports, from link-level 5G simulation.
Optimised antenna placement	Recommended positions, tilt, power and number of cells, found with AI optimisation.
Channel data for AI	Datasets to train and test machine-learning models in research labs.

How we deliver it

From a floor plan to cells on air.

Model the site
Floor plans, CAD drawings or OpenStreetMap data become a 3D scene.
Simulate on GPU
Coverage, multipath and interference for your band and power levels.
Optimise with AI
Antenna positions, tilt, power and number of cells.
Validate
SDR walk tests with a USRP B210 or LimeSDR, compared against the prediction.
Deploy and report
Radios installed and tuned, handed over with a coverage report.

Who it’s for

University and R&D labs
Factories and warehouses
Campuses and venues
Ports, mines and remote sites
Vendors and integrators
Training and education

Good to know

It runs on modest hardware.Our lab machine is an Intel NUC with an RTX 3060, running alongside a live 5G core and SIP server.

It’s a planning tool.Sionna simulates the radio network. It doesn’t transmit or replace your radios.

Licensing still applies.Live over-the-air trials in Ireland need a ComReg test and trial licence, and we can advise on this.

Planning a lab or a new radio site?

Tell us about the site and we’ll show you what a Sionna study would cover.

Talk to CelTech

Animated panels in the explainer video are illustrative. 3D maps and charts are real Sionna output from the CelTech lab. NVIDIA and Sionna are trademarks of NVIDIA Corporation. CelTech is not affiliated with NVIDIA.