The pressure to automate RAN PLANNING and optimisation is increasing across mobile network operators.
White Paper: AI-Based Radio Propagation Modelling for Autonomous RAN Optimisation
This White Paper is a technical document produced by Forsk's RF propagation modelling and software engineering teams, written for RF engineers and RAN architects who need to evaluate the computational and accuracy credentials of AI-based propagation modelling before integrating it into planning or automation workflows. It describes the model architecture, training methodology, GPU acceleration approach, benchmark results on NVIDIA hardware, and accuracy validation against calibrated ray-tracing outputs and CW measurement data.
What's inside:
- The case for AI-based radio propagation modelling — why existing approaches cannot meet the speed requirements of closed-loop autonomous RAN optimisation, and the architectural reasoning behind a neural network trained on calibrated ray-tracing outputs
- CNN architecture and training methodology — how the encoder-decoder network processes geospatial inputs including terrain, clutter, and building data to produce path loss predictions
- Full benchmark results across hardware configurations — computation performance across CPU and multiple NVIDIA GPUs, including the NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs
- Accuracy analysis — mean absolute error and RMS comparisons against calibrated Aster model predictions and CW measurement data
- Proof-of-concept validation — methodology and results from a collaboration with a tier-1 US mobile network operator
- Use cases for autonomous optimisation — how near real-time propagation modelling enables network self-healing and automated antenna tilt optimisation in a closed-loop architecture
Who it's for:
This white paper is written for RF engineers, RAN architects, and technical leads at mobile network operators and systems integrators who are evaluating AI-based radio propagation modelling approaches for network planning, digital twin development, or autonomous optimisation solutions. It assumes familiarity with propagation modelling fundamentals, GPU compute concepts, and RAN automation architecture.
Naos Unique Value proposition
Cloud-native RAN digital twin enabling near real-time, propagation-aware, closed-loop optimisation
End-to-End
Automation
- Streamlined automated workflows
Accelerated RAN planning and optimisation cycles
Cloud-Native
Scalability
- Massive parallel processing across nationwide territories
- Accurate high-resolution radio wavepropagation & predictions
AI-powered closed-loop Optimisation
- Integration with RAN automation ecosystems and SMO frameworks
- AI-powered radio access network
Our products
A Complete AI-powered RAN Planning
& Optimisation Software Solution
for Mobile Operators and Private Networks
Naos — Cloud-Native Automation Platform for RAN Planning and Optimisation
RAN planning automation is becoming a must to scale nation-wide predictions and to make massive site-data based tasks such as 5G site selection faster. Naos is a cloud-native automation platform that provides MNOs with a framework for designing their own automated RAN planning and optimisation workflows and delivers off-the-shelf scalable computation capabilities for massive network-wide calculations.
Atoll — the industry standard for radio network design
From initial frequency planning to live-network optimisation — Atoll is the single platform trusted by RF engineers in 140 countries to design 2G, 3G, 4G and 5G networks with precision.
Atoll One — purpose-built for private network design
Atoll One delivers the full planning and simulation power of Atoll in a package built specifically for private network projects. Indoor and outdoor environments are modelled in a single environment, with full interaction between both. The Aster propagation model is built in to ensure accuracy across complex sites — from multi-level terminals to underground tunnels and industrial halls.
Supports 4G LTE, 5G NR, Wi-Fi, and microwave backhaul. Installs in minutes. Designed for engineers working on a project basis and operators managing ongoing network evolution.
Typical Automation Use Cases
Large-scale path loss calculations
Site selection
Automatic neighbour planning
Country-wise coverage predictions
Antenna RET optimisation
Automatic PCI planning
Incremental coverage map updates
Massive MIMO antenna selection
...
Frequently Asked Questions
This white paper is written for RF engineers, RAN architects, and technical leads at mobile network operators and systems integrators. It assumes working knowledge of propagation modelling, network planning workflows, and RAN automation concepts. It is not an introductory document — it is a technical reference for practitioners evaluating AI-based propagation modelling for RAN planning and autonomous optimisation.
The white paper covers the full technical basis for Forsk's AI propagation model: the CNN architecture and training methodology, the GPU acceleration approach developed in partnership with NVIDIA, benchmark results across CPU and multiple GPU configurations, and accuracy validation against CW measurement data and calibrated Aster ray-tracing outputs. It also includes results from a proof-of-concept deployment with a tier-1 US mobile network operator.
The AI propagation model described in this white paper is a compute component of Naos, Forsk's cloud-native platform for RAN planning and network engineering automation. To find out how it applies to your network planning or optimisation workflow, request a demonstration at forsk.com/contact-us.
Trusted by operators worldwide
A Complete AI-powered RAN Planning
& Optimisation Software Solution for Mobile Operators and Private Networks
