SoftBank Corp. and Ericsson Japan announced on August 20 that they successfully demonstrated Ericsson's "AI in RAN" software on SoftBank's commercial 5G network. The companies characterized the event as the first validation of this specific vendor technology within Japan.
The trial focused on an AI-native scheduler designed for link adaptation, a core function within the broader AI in RAN suite. By applying artificial intelligence directly inside the radio access network in real time, the system aimed to optimize performance metrics such as spectral efficiency, user throughput, robustness, and stability in a live commercial environment.
Results from the test indicated substantial improvements over conventional technology. Spectral efficiency saw gains of up to 25%, while downlink user throughput increased by as much as 50%. Across all measured evaluation areas, both spectral efficiency and downlink user throughput improved by an average of approximately 10%.
Ericsson stated that the software operates on baseband equipment at mobile base stations, utilizing real-time AI decision-making. The company explained that the AI model learned to infer responses to complex and constantly changing radio conditions. This capability allowed for adjusted link parameter settings for users located at cell edges or in areas with high radio interference.
The companies noted that these performance gains were confirmed under difficult radio conditions, including cell boundaries and congested zones. These results suggest an expansion of traffic-handling capacity within existing spectrum bands. Spectral efficiency serves as a measure of how much data traffic can be transmitted within a given frequency bandwidth.
This project was part of an ongoing collaboration between SoftBank and Ericsson toward developing AI-native RAN systems. SoftBank established requirements for evaluation areas based on specific traffic characteristics. Ericsson trained the AI-native scheduler model using actual network data, implementing and tuning real-time channel-capacity prediction and the selection of optimal downlink transmission rates.
