KEY POINTSSoftBank and Ericsson validate AI in RAN software on a 5G commercial network in JapanTrial shows up to 25% higher spectral efficiency and up to 50% faster downlink user throughputProject advances AI-native mobile network development for 5G Advanced and future 6G
The trial on SoftBank’s commercial 5G network delivered up to 25% better spectral efficiency and up to 50% higher downlink throughput. Photo by Thiago Santos on Unsplash
Thiago Santos
SoftBank Corp. and Ericsson Japan said on August 20 they had demonstrated Ericsson’s “AI in RAN” software on SoftBank’s 5G commercial network, in what the companies described as the first such validation in Japan for the vendor’s technology.
The trial tested an AI-native scheduler for link adaptation, a function within Ericsson’s AI in RAN software, by applying artificial intelligence directly inside the radio access network in real time. The companies evaluated spectral efficiency, user throughput, robustness and stability in a live 5G commercial network environment.
The trial showed spectral efficiency improved by as much as about 25% and downlink user throughput by as much as about 50% versus conventional technology. Across all evaluation areas, spectral efficiency and downlink user throughput each improved by an average of about 10%.
Ericsson’s software can run on baseband equipment at mobile base stations using real-time AI decision-making. The companies said the AI model learned and inferred how to respond to complex and constantly changing radio conditions, enabling link parameter settings for users at cell edges and in areas with high radio interference.
The gains were confirmed under difficult radio conditions such as cell boundaries and congested areas, expanding traffic-handling capacity within existing spectrum bands. Spectral efficiency measures how much traffic can be transmitted within a given frequency bandwidth.
The project was carried out as part of SoftBank and Ericsson’s collaboration toward AI-native RAN. SoftBank defined requirements for evaluation areas based on traffic characteristics, while Ericsson trained the AI-native scheduler model using actual network data and implemented and tuned real-time channel-capacity prediction and selection of optimal downlink transmission rates.
The companies said the demonstration also confirmed that AI can contribute to high-performance, secure and flexible networks aimed at 5G Advanced and future 6G. 5G Advanced refers to the next evolutionary phase of 5G standards before 6G.
Mobile operators globally are exploring AI-based network optimization as data traffic rises and applications such as generative AI assistants, autonomous agents and immersive services place more varied demands on wireless networks. Radio access networks connect user devices to the core network and account for a large share of mobile network performance and capacity.