SoftBank Corp. develops a foundational Large Telecom Model (LTM)

SoftBank Corp. has developed a Large Telecom Model (LTM) aimed at improving cellular network operations. The LTM utilizes extensive datasets from SoftBank’s network data and operational experience, facilitating advanced inference for the design and management of cellular networks. SoftBank plans to integrate the LTM into its operational workflows and continue enhancing its research efforts. The company has fine-tuned the LTM for optimizing base station configurations. These specialized models achieved over 90% accuracy in predicting configurations for real base stations omitted from training, reducing implementation times from days to minutes while maintaining accuracy. This approach suggests potential operational efficiencies and cost savings, alongside minimizing human error. SoftBank positions the LTM as a fundamental component for its “AI for RAN” initiative, which seeks to improve Radio Access Network (RAN) performance. The LTM framework will support the development of various Artificial Intelligence (AI) models geared towards specific operational scenarios, aiding network design and optimization. Additionally, it has been optimized using NVIDIA technologies, enhancing performance capabilities and providing deployment flexibility across cloud and on-premise environments. The implementation aligns with the “Human AI” concept from SoftBank’s Research Institute of Advanced Technology (RIAT), which envisions the use of AI models in mobile networks. SoftBank aims to refine these AI models and integrate them into its AITRAS project, an AI-RAN solution under development. As the company advances these initiatives, it plans to enhance mobile network efficiency and explore new service offerings, reinforcing its commitment to next-generation networks.