A RAN Optimisation Methodology That Scales Beyond Drive Tests

Optimisation teams still lean heavily on drive testing, which is expensive, slow and samples a tiny fraction of the network. A scalable methodology starts from network-wide data and uses field measurement selectively.

Four layers

Layer one is KPI monitoring with clear baselines and thresholds per cluster. Layer two is automated diagnosis — classifying degradations into coverage, interference, capacity, mobility, hardware or parameter causes using counters and cell traces. Layer three is targeted action: parameter changes, neighbour fixes, antenna changes, capacity additions. Layer four is verification, including drive tests where the change affects user experience directly.

Where AI fits

Anomaly detection and classification models accelerate layer two significantly, especially when trained on an operator's own historical resolutions. They do not replace engineering judgement at layer three — but they let engineers spend their time there.