SoftBank and Ericsson have validated an AI scheduler for link adaptation inside the radio access network (RAN) on a live 5G network in Japan. The two companies ran the trial on SoftBank’s commercial network rather than in a lab environment, testing a feature of Ericsson’s AI in RAN software that applies machine learning decisions directly to baseband equipment in real-time.

The pair describe the work as Japan’s first validation of this kind. The trial forms part of a broader collaboration between the two firms aimed at building what they call AI-native RAN, with an eye toward future 5G Advanced and 6G deployments.

SoftBank and Ericsson report gains of up to approximately 25 percent in spectral efficiency and up to approximately 50 percent in downlink user throughput compared with conventional technology at the measured sites. Across all evaluated locations, the companies say both spectral efficiency and downlink user throughput improved by approximately 10 percent on average.

The two firms tested four metrics during the trial: spectral efficiency, user throughput, robustness and stability. All measurements took place on SoftBank’s live commercial 5G network rather than in a controlled or simulated setting, which distinguishes the exercise from a lab demonstration but does not amount to a network-wide commercial rollout.

Real-time AI replaces rule-based link adaptation

Conventional link adaptation has relied on rule-based algorithms built from offline analysis and static parameter configuration. Ericsson’s AI-native scheduler instead runs on baseband equipment and makes decisions using a model trained to learn from, and perform inference on, radio conditions that change continuously.

The model targets three problems: it optimises link parameters for users at cell edges or in areas with high interference, stabilises user throughput in challenging radio conditions – including congested areas and cell edges – and it also aims to fit more traffic into existing frequency bands by improving spectral efficiency, rather than requiring operators to add new spectrum or hardware capacity.

Mårten Lerner, Head of Networks Strategy & Product Management at Ericsson, said: “By bringing telco-grade AI into the RAN, we are enabling real-time optimisation using Ericsson silicon in today’s installed base, improving network performance and efficiency in 5G commercial networks.

“The results show how advanced AI capabilities, combined with operator expertise and responsibly handled data, can translate AI innovation into measurable improvements and a better user experience.”

Divided responsibilities: training, tuning, and requirements

SoftBank defined the requirements for the evaluation areas based on traffic characteristics observed on its network. Ericsson trained the AI-native scheduler model using actual network data, then implemented and tuned it to predict channel capacity and select the appropriate downlink transmission rate in real time.

Jawad Manssour, President and Representative Director of Ericsson Japan, comments: “Ericsson believes that the role of mobile networks in delivering the full value of AI will become increasingly important. This field trial demonstrated that by directly applying AI to the Radio Access Network, significant gains are achievable in a network using commercially available hardware and software.

“Together with SoftBank, we are committed to further enhancing network performance and efficiency through AI-native technology, and to contributing to the realisation of a network infrastructure capable of improving the customer experience and meeting future AI traffic demands. We will continue to further strengthen the partnership between our two companies and bring advanced network solutions originating from Japan to the world.”

Rising AI traffic drives the case for RAN-level intelligence

The companies frame the trial against a backdrop of growing use of generative AI, AI assistants, autonomous agents, and immersive applications in everyday settings. As adoption of these services expands, networks face rising traffic volumes alongside new traffic patterns and more demanding connectivity requirements.

SoftBank and Ericsson argue that meeting this demand requires more than adding network capacity. Applying AI within the RAN lets operators adapt to changing radio and traffic conditions, optimise performance and efficiency, and make better use of resources already deployed. The announcement positions the link adaptation scheduler as one feature within the wider Ericsson AI in RAN software portfolio, with this validation serving as a step toward what the companies call AI-native mobile networks.

Teruyuki Oya, SVP and CNO of SoftBank, explains: “Under our growth strategy, ‘Activate AI for Society,’ SoftBank aims to bring the benefits of AI to more people and industries. As the implementation of AI in society advances rapidly, mobile networks are required to handle larger volumes of traffic with greater variability than ever before.

“Together with Ericsson, we have demonstrated the effectiveness of the AI-native scheduler for Link Adaptation on SoftBank’s 5G commercial network. This achievement marks an important milestone in advancing the RAN through AI and represents a major step toward realising networks that will support the future AI era. SoftBank will continue to accelerate its efforts toward realising AI-native networks.”

See also: EDOTCO builds Azure AI tool to speed up telecom tower planning

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events including the IoT Tech Expo and Cyber Security & Cloud Expo. Click here for more information.

Telecoms is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.