Autonomous RAN optimisation lacks propagation awareness, but AI holds the key to the solution.
Autonomous networks need to know the radio environment in near real time.
Today’s SON systems make decisions based on performance metrics — not on a continuously updated model of how radio waves are propagating.
That gap is structural. Ray-tracing models produce engineering-grade path loss predictions, but they’re too slow for operational timescales of closed-loop optimisation. Statistical models are fast, but they lack the precision required for reliable optimisation decisions.
Forsk has developed an AI-based radio propagation model that closes that gap. A convolutional neural network trained on calibrated ray-tracing outputs delivers ray-tracing accuracy at inference speed — up to 200x faster on NVIDIA RTX PRO GPUs, with less than 1 dB mean absolute error.
This isn’t a performance improvement to existing workflows. It’s a capability that unlocks a new category of RAN automation: propagation-aware closed-loop optimisation. Use cases like network self-healing and automated antenna tilt optimisation are now operationally viable at the speed autonomous systems require.
The full technical detail — CNN architecture, training methodology, benchmark results across NVIDIA GPU generations, and accuracy validation — is in the white paper.
About Forsk
Forsk is an independent software company providing operators and vendors with wireless network design and optimisation products.
Atoll, Forsk's flagship product, is a multi-technology wireless network design and optimisation software that allows operators to streamline planning and optimisation activities by combining predictions and live network data. With more than 11,000 installed licenses in over 140 countries, Atoll has become the industry standard for wireless network design and optimisation.
Naos is Forsk's automation platform dedicated to wireless network planning and optimisation. Naos is a cloud-native platform that enables operators to scale and automate planning and optimisation processes through on-premises or cloud-based deployments. Typical use cases include 5G site selection as well as massive coverage calculations and optimisation processes.
Atoll One is a dedicated version of Atoll, specifically designed for the planning and optimisation of small to medium-sized outdoor and indoor Private Mobile Networks.
Atoll and Naos provide operators with a comprehensive framework for integrated, interactive, fully automated and scalable wireless network planning and optimisation. Atoll and Naos are available from Forsk's offices and technical support centres in France, USA, and China, as well as from a worldwide network of distributors and partners.
