Agentic AI-RAN for Telecom Networks Demonstrated

- The demonstrated agentic architecture uses a hierarchy of AI agents, orchestrated by a Large Language Model (LLM), to interpret natural language intents and autonomously control network functions across different time scales and protocol layers. - SoftBank, a founder of the 75-member AI-RAN Alliance, is developing its own Large Telecom Model (LTM) and an AI-RAN integrated solution called AITRAS, aiming to deploy AI servers at base stations across Japan. - The Large Telecom Model (LTM) at the core of the system is a specialized generative AI trained on vast telecom datasets, including network performance data, operational manuals, and 3GPP standards, enabling it to translate high-level goals into specific, automated actions. - zTouch Networks, a spinout from Northeastern University, provides the AI-driven orchestration framework that converts operator intents into a fabric of low-level AI modules dispatched as microservices across the Open RAN. - Northeastern University's Institute for the Wireless Internet of Things (WIoT) provides the underlying research and testing environment, including its Open6G OTIC (Open Testing and Integration Center) which facilitates multi-vendor integration. - Keysight provides the crucial validation and emulation solutions for the system, allowing for continuous testing of AI and RAN workloads on a shared cloud infrastructure and benchmarking performance and energy consumption. - This move toward agentic AI is a step toward "Autonomous Networks Level 5," aiming to shift network management from reactive, manual processes to predictive, fully autonomous operations with auditable decision-making. - The adoption of such AI systems in telecom requires robust AI governance frameworks to manage risks like data leakage and biased outputs, a challenge given that only half of organizations report having formal AI guardrails in place.