AI-Native Orchestration in the 6G Continuum: Evolving Operator Platforms with Agentic AI
本文提出了一种基于自主智能体的AI原生编排层,用于6G网络中的无缝云-边缘-IoT连续体,以解决跨域服务自动化和管理问题。
本文提出了一种基于自主智能体的AI原生编排层,用于6G网络中的无缝云-边缘-IoT连续体,以解决跨域服务自动化和管理问题。
论文探讨了6G网络中LLM代理间因主观信息传递导致的AI幻觉问题,提出基于认知通道与细胞层束的理论框架及五项设计原则来解决。
This study addresses the high interaction costs, safety risks, and deployment challenges associated with online multi-agent reinforcement learning (MARL) in network slice resource allocation. To overcome these limitations, this work proposes X-CODE, a framework integrating explainable AI with offline MARL. By leveraging explainability-aware reward shaping, X-CODE optimizes decentralized execution policies, enabling safe and efficient management without online interactions or inter-agent communication. Experimental results demonstrate that the proposed framework achieves zero resource conflicts during testing and reduces effective inference latency by 88%. Furthermore, it significantly decreases signaling overhead and slicing latency. These findings establish X-CODE as a secure, low-latency intelligent solution for dynamic network slicing environments, effectively bridging the gap between theoretical MARL models and practical, safety-critical telecommunications deployments.
This work addresses the challenge of insufficient 5G link reliability that hinders safe beyond visual line-of-sight (BVLoS) unmanned aerial vehicle (UAV) operations. To overcome this limitation, the authors develop a BVLoS UAV operating system based on an open-source O-RAN platform, integrating end-to-end network slicing into the open 5G architecture for the first time to ensure control-link reliability. The proposed system effectively mitigates trajectory deviations caused by link congestion and consistently maintains end-to-end latency within the 3GPP standard limits. By enabling highly reliable and low-latency remote UAV control, this approach significantly enhances flight safety and operational performance in BVLoS scenarios.
This study addresses the challenge of providing deterministic guarantees for latency-sensitive interfaces in O-RAN architectures over low-cost, commodity Ethernet. It systematically reviews key Time-Sensitive Networking (TSN) standards—including IEEE 802.1CM, 802.1Qbu, and 802.1Qbv—and proposes a TSN-enabled design space for O-RAN. The work analyzes deployment pathways within virtualized RAN (vRAN) and cloud-native environments, identifying critical technical requirements for achieving low-latency, high-reliability transport. Furthermore, it validates the feasibility of TSN in enabling multi-vendor interoperability and high flexibility in open RAN deployments, offering a practical and cost-effective technical roadmap for realizing deterministic fronthaul and midhaul networks.
本文提出了一种基于自主智能体的AI原生编排层,用于6G网络中的无缝云-边缘-IoT连续体,以解决跨域服务自动化和管理问题。
论文探讨了6G网络中LLM代理间因主观信息传递导致的AI幻觉问题,提出基于认知通道与细胞层束的理论框架及五项设计原则来解决。
This study addresses the high interaction costs, safety risks, and deployment challenges associated with online multi-agent reinforcement learning (MARL) in network slice resource allocation. To overcome these limitations, this work proposes X-CODE, a framework integrating explainable AI with offline MARL. By leveraging explainability-aware reward shaping, X-CODE optimizes decentralized execution policies, enabling safe and efficient management without online interactions or inter-agent communication. Experimental results demonstrate that the proposed framework achieves zero resource conflicts during testing and reduces effective inference latency by 88%. Furthermore, it significantly decreases signaling overhead and slicing latency. These findings establish X-CODE as a secure, low-latency intelligent solution for dynamic network slicing environments, effectively bridging the gap between theoretical MARL models and practical, safety-critical telecommunications deployments.
This work addresses the challenge of insufficient 5G link reliability that hinders safe beyond visual line-of-sight (BVLoS) unmanned aerial vehicle (UAV) operations. To overcome this limitation, the authors develop a BVLoS UAV operating system based on an open-source O-RAN platform, integrating end-to-end network slicing into the open 5G architecture for the first time to ensure control-link reliability. The proposed system effectively mitigates trajectory deviations caused by link congestion and consistently maintains end-to-end latency within the 3GPP standard limits. By enabling highly reliable and low-latency remote UAV control, this approach significantly enhances flight safety and operational performance in BVLoS scenarios.
This study addresses the challenge of providing deterministic guarantees for latency-sensitive interfaces in O-RAN architectures over low-cost, commodity Ethernet. It systematically reviews key Time-Sensitive Networking (TSN) standards—including IEEE 802.1CM, 802.1Qbu, and 802.1Qbv—and proposes a TSN-enabled design space for O-RAN. The work analyzes deployment pathways within virtualized RAN (vRAN) and cloud-native environments, identifying critical technical requirements for achieving low-latency, high-reliability transport. Furthermore, it validates the feasibility of TSN in enabling multi-vendor interoperability and high flexibility in open RAN deployments, offering a practical and cost-effective technical roadmap for realizing deterministic fronthaul and midhaul networks.