In this article, we want to put the Sionna digital twin into practice and see if it would help RF engineers simulate the changes before implementing them. We implement coverage in a complex urban environment (Arc de Triomphe, Paris) using Sionna RT.
In part one of this series, we discussed why and how Sionna and Databricks help organizations move toward true data and AI democratization, specifically advanced network optimization and simulation capabilities. We gave a short helicopter view of telecommunication and summarized some of the processes, in particular, air interference that connects base stations to users. We introduced Sionna and the transformative impact of AI-native digital twins and open source simulation tools on the telecommunications industry, as announced at Nvidia’s GTC 2025 conference.
Let’s talk about the use case we want to implement.
This is a demo of how Sionna might be used by an RF optimization engineer.
Example project: In telco providers, there are multiple RF planning and optimization teams that plan, implement and test changes in the RF network. They track network demand and proactively or, based on customer complaints, adjust network capacity/config to meet the demands. Imagine you are an RF optimization engineer in France and receive a user complaint in the Arc de Triomphe. There are 7 hypothetical towers in Arc de Triomphe, Paris, each covering a 90-degree cell, we also add 50 users per cell; current cell configurations are:
Sometimes, the action you take is installing a new tower or adjusting physical configs permanently or temporality to meet the demands. Change variables in base station towers can be antenna location, antenna height, antenna array system, antenna orientation (tilt, azimuth or zenith), Tx power, Frequency, Bandwidth, and Antenna gain. Making these changes in real life is costly and time-consuming, so normally RF engineers simulate the change first before implementing it and affecting the outside world, once they have a good understanding of the change results, it will be implemented in the network.
The current challenge is that this simulation happens in third party software tools that are not open source and have a partial view of the network. This will result in suboptimal solutions, expensive siloed processes, an increases number of tickets back and forth between departments and customer dissatisfaction.
Here, we want to use Sionna to create a digital twin of the network and see if it helps RF engineers simulate the changes before implementing them.
In this simulation, we use Sionna to simulate the effect of upgrading the network's antenna array systems from config 1 to config 2 in 28GHz (UPA is a uniform planar array and refers to antenna design ).
- Config 1: transmitter: [8x2 UPA], user: [2x2 UPA]
- Config 2: transmitter: [16x16 UPA], user: [2x2 UPA]
to see if our target KPI (in this case, CDF of SINR or RSS) improves or not. (There are many other KPIs, and the root complaint identifies which KPI to improve.) The data used is sourced from OpenStreetMap and visualized with Blender. You can find API documnetioan here https://nvlabs.github.io/sionna/rt/api/rt.html
Sionna is an open source library developed by Nvidia. Sionna is Nvidia’s flagship software for creating digital twins of cellular mobile network. This is a very active area of development in Nvidia, they had a releaase in GTC 2025.
Implementation:
You can find the simulation code in this git repo
It is a simulation of a hypothetical RF network using Sionna RT and Mitsuba for ray-tracing-based radio propagation modelling.
Results for Config 1: 8x2 TX UPA and 2x2 RX UPA:
The cell-to-Tx association plot shows cell coverage areas in terms of best SINR. You can see we have 7 cell coverage in 7 different colours. Users are connected to cells with maximum SINR in that location, which is shown in the second plot.
- Bright green means high SINR (Good)
- Dark blue means low SINR (Bad)
you can see users assigned to each tower in Higherst SINR across all TXs plot.
Plot SINR & RSS CDFs
The last step is to benchmark current architecture and visualize network performance via cumulative distribution functions (CDFs) of:
- SINR: Signal-to-Interference-plus-Noise Ratio
- RSS: Received Signal Strength
Repeat the process for Config 2: 8x8 TX UPA and 2x2 RX UPA
The steps are the same as those in Config 1, but the transmitter array is changed from 8x2 to 8x8 to simulate a more powerful and directive beamforming setup.
Conclusion
By simulating these two configs, the RF optimizer can see that user experience has decreased (CDF moved to the left), so it is not a good change. This is a very simple example of the integration of Sionna with Databricks to read your data from your network and simulate the changes.
While NVIDIA boasts powerful AI-RAN capabilities for 5G and 6G, mainstream telecom vendors remain cautious, with some skepticism about the immediate practicality of AI-RAN schemes. However, the open-source nature of tools like Sionna is democratizing access to advanced simulation and planning capabilities, enabling startups and smaller players to innovate rapidly.
NVIDIA’s digital twin and data-driven solutions that are possible by integrating it with Databricks:
- Provides an end-to-end, software-first approach to modelling and managing networks, which is open-source and full-stack simulation.
- Enables predictive maintenance, rapid scenario testing, and streamlined troubleshooting.
- Lowers barriers for new entrants and accelerates innovation across the ecosystem.
That’s why we at Databricks are excited to help you implement Sionna and experiment with this new tool.
What Happens Next?
The telecom sector is on the edge of a shift similar to what happened in natural language processing (NLP). As deep learning bypassed traditional linguistic rules, value migrated from rule-based systems to data-driven, AI-powered platforms. A similar transition is underway in telecom:
- AI-native simulation stacks (like Sionna) are encoding telecom rules into software, shifting value from hardware-centric vendors to AI and cloud platforms.
- New skill sets are required, with telco teams needing AI/ML and software-defined radio expertise.
NVIDIA’s open-source, AI-native digital twin technology marks a turning point for the telecom industry. By enabling end-to-end simulation, automation, and democratized innovation, it challenges the traditional vendor landscape and opens new opportunities for operators and startups alike. The winners will be those who adapt quickly, invest in new skills, and leverage these tools to drive the next wave of telecom transformation
