UnifiedKP: A Unified Network Knowledge Plane for Large Model-Enabled 6G Networks

Abstract

The emergence of large language models (LLMs) and agentic systems is revolutionizing the landscape of 6G networks by enabling unprecedented levels of autonomous intelligence, including self-configuration, self-optimization, and self-healing capabilities. However, current implementations face significant challenges. Individual intelligence tasks require isolated knowledge retrieval pipelines. This isolation results in redundant data flows, inconsistent interpretations, and increased operational complexity. Inspired by the service model unification efforts in Open-RAN that promote interoperability and vendor diversity, we propose UnifiedKP: a unified Network Knowledge Plane specifically designed for large model-enabled autonomous 6G network intelligence. By decoupling network knowledge acquisition and management from intelligence logic, UnifiedKP streamlines development workflows and significantly reduces maintenance complexity for intelligence engineers. Through an intuitive and consistent knowledge interface, UnifiedKP enhances interoperability for network intelligence agents while maintaining semantic consistency across diverse intelligence tasks. We demonstrate the effectiveness of UnifiedKP through two representative intelligence applications: live network knowledge question-answering and edge AI service orchestration. Experimental results show that UnifiedKP reduces knowledge retrieval latency by 47%, improves knowledge consistency by 82%, and decreases development complexity by 65% compared to traditional isolated approaches. Our framework achieves 94.3% accuracy in network anomaly detection and reduces service orchestration time by 38% in dynamic edge computing environments. These findings establish UnifiedKP as a foundational architecture for realizing truly autonomous and intelligent 6G networks.