Ericsson, Optus Partner on AI Model for Accurate 5G SA Coverage Prediction

Key Takeaways

• AI model predicts cross‑frequency coverage with 95% accuracy
• Real‑time predictions cut device measurement overhead
• Local cell‑level retraining tailors performance per site
• Faster handovers boost battery life and reliability
• First Ericsson‑operator AI collaboration for 5G SA

Summary

Ericsson and Australian carrier Optus have launched a joint trial of an artificial‑intelligence model that predicts whether a device on one 5G Standalone frequency layer is also covered by another. The model, trained on months of live Optus data, can be retrained locally for each cell and delivers up to 95 % accuracy in real time. Early results show reduced device measurements, faster handovers and lower power consumption, directly improving user experience. This marks the first AI‑driven collaboration of its kind between Ericsson and an operator.

Pulse Analysis

The rollout of 5G Standalone (SA) has exposed the limits of static radio planning, especially when devices must switch between multiple frequency layers on the fly. Traditional hand‑over logic relies on pre‑computed maps that consume device battery and generate unnecessary measurements, leading to dropped calls and higher core‑network load. Embedding artificial intelligence directly into the radio access network (RAN) enables per‑cell, real‑time decisions that reflect actual traffic and propagation conditions, turning the network into a self‑optimizing system. These improvements are critical as operators scale dense small‑cell deployments in urban cores.

In the Ericsson‑Optus partnership, months of live network data were used to train a bespoke AI model that predicts whether a device on one frequency is simultaneously covered by another. The model, which can be retrained locally for each cell, delivers predictions with up to 95 % accuracy and operates in real time. Early trials show a measurable drop in unnecessary device measurements, faster handovers, and lower power consumption, directly translating into fewer dropped calls and extended battery life for 5G SA users. The approach also reduces signaling traffic to the core, further easing network load.

The success of this trial signals a shift toward AI‑driven orchestration across the 5G ecosystem. Operators can now automate frequency‑layer decisions without costly manual re‑optimisation, freeing engineering resources and improving overall network efficiency. As more carriers adopt similar models, the industry could see standardized AI interfaces for RAN, accelerating the rollout of advanced services such as ultra‑reliable low‑latency communications and massive IoT. Regulators and standards bodies are already evaluating how to embed such AI capabilities into future 3GPP releases. Ultimately, the blend of real‑time AI and 5G SA promises a more resilient, energy‑efficient mobile experience for both consumers and enterprises.