Source-linked AI summary
Toward Fully Autonomous 6G Networks: AI-driven Operational Efficiency and Optimization
David Reiss, Oriol Sallent, Miguel Catalan-Cid, Daniel Camps-Mur
TL;DR
Future mobile networks face growing complexity as heterogeneous services and NaaS requests must be managed efficiently. The paper combines AI-guided RAN optimization, O-RAN-compliant emulation, feasibility assessment, and agentic orchestration to interpret intent-based policies and coordinate external requests with internal management. It identifies agent coordination and scalable network digital twins as open challenges for reliable autonomous 6G RAN management.
Problem
Growing heterogeneous services and NaaS requests increase RAN management complexity, creating a need to integrate external network demands with internal management policies.
Method
The paper combines AI-driven RAN optimization, O-RAN-compliant emulation, NaaS feasibility assessment, and agents that interpret and exchange intent-based policies.
Results
The work demonstrates AI-driven RAN efficiency mechanisms in an O-RAN-compliant framework and details an agentic environment for reliable NaaS adoption within 6G RAN management.
Takeaways & Limitations
AI agents and supporting assessment environments provide a proposed basis for integrating NaaS workflows with autonomous, intent-driven RAN management.
Takeaways & Limitations
The paper identifies unresolved challenges in coordinating AI agents and designing network digital twins that balance lightweight interaction with robust network assessment.
Abstract
from arXiv · showhide
Mobile networks evolution is characterized by a substantial increase in system complexity, driven by the need to accommodate a growing number of heterogeneous services on top of the digital infrastructure. This growth in service accommodation is expected to accelerate with the adoption of the Network as a Service (NaaS) paradigm, which has emerged as a promising approach to accelerate network innovation while enabling new revenue streams for operators. Although it is fundamental to abstract network capabilities for third-party developers, it poses significant challenges in terms of efficient network operation. To address this increased complexity, future mobile networks are envisioned to be inherently Artificial Intelligence (AI)-native. In particular, the integration of AI within the Radio Access Network (RAN) becomes a key enabler for optimizing operation, energy consumption, and autonomous network control. In this context, this research explores the convergence of AI-native RAN and NaaS ecosystems to enable autonomous 6G RAN management. We propose an Agentic-based orchestration framework capable of interpreting intent-based policies. The proposed framework becomes key to integrate external NaaS requests with internal network management policies.
I. INTRODUCTION
Future B5G/6G networks face rising operational complexity from heterogeneous services and infrastructure, motivating AI-native RAN management and NaaS integration. The paper therefore develops AI-guided, intent-driven management that can reconcile external NaaS requests with internal network policies.
- I. INTRODUCTION: Growing service diversity, infrastructure densification, and virtualization make future network orchestration increasingly challenging.The paper identifies network slicing and Quality on Demand as examples of mechanisms needed to support diverse requirements.
- I. INTRODUCTION: AI-native 6G networks are expected to provide autonomous and self-optimizable operation, particularly through AI integration in the RAN.The targeted areas include intelligent resource management, energy efficiency, and autonomous network control.
- I. INTRODUCTION: NaaS exposes network capabilities through developer-facing interfaces so third parties can request and consume network resources dynamically.The CAMARA ecosystem is presented as an example whose roadmap anticipates Agentic APIs.
- I. INTRODUCTION: AI agents can interpret high-level intents and support intent-driven orchestration of dynamic network environments.The related MAESTRO example uses LLM-based agents to automate stakeholder negotiations and ground business intents in technical requirements.
- I. INTRODUCTION: The research combines AI-guided RAN automation with NaaS workflows so external requests can coexist with internal management policies.The proposed direction includes an Agentic orchestration framework in which agents interpret and exchange policies expressed in natural language.
II. RESEARCH TOPICS AND CHALLENGES
The research addresses current RAN efficiency and future NaaS management challenges through AI techniques and an O-RAN-compliant evaluation environment. It also examines feasibility issues arising when external service requests interact with RAN operations.
- II. RESEARCH TOPICS AND CHALLENGES: The section covers AI-based RAN efficiency improvement, O-RAN-compliant evaluation, and feasibility challenges in NaaS ecosystems.It separates the current optimization work from future NaaS-related management concerns.
A. Current stage: AI for RAN optimization
The initial research stage improves 5G RAN operational efficiency using AI mechanisms that identify low-utilization conditions while accounting for QoS. An O-RAN-compliant emulation framework supports model deployment, control, telemetry exchange, and validation before commercial deployment.
- A. Current stage: AI for RAN optimization: A real European MNO dataset revealed over-provisioning in some 5G NSA areas during the transition to a new mobile generation.The finding motivated mechanisms for reducing unnecessary energy consumption.
- A. Current stage: AI for RAN optimization: AI-driven mechanisms identify low-utilization 5G scenarios and incorporate QoS awareness to balance energy savings with user service requirements.They were trained with standardized Performance Measurement counters supplied by the infrastructure provider.
- A. Current stage: AI for RAN optimization: The O-RAN emulation framework comprises a decoupled AI management platform, O-RAN control modules, and a Keysight RIC Test emulation block.A data-ingestion pipeline adapts emulation parameters to the MNO dataset.
- A. Current stage: AI for RAN optimization: The framework supports model inference through RIC-hosted xApps/rApps, standardized O-RAN interfaces, telemetry exchange, and enforcement of AI-driven management policies.It also enables high-accuracy emulation of real RAN scenarios.
B. Future directions: AI for RAN management in NaaS ecosystems
NaaS requests complicate RAN management because prioritized external services must be translated into feasible RAN actions alongside existing policies and users. The proposed research addresses this through context-aware assessment, long-term feasibility analysis, and adaptive resource management.
- B. Future directions: AI for RAN management in NaaS ecosystems: CAMARA-style APIs let external consumers request prioritized network resources, but implementing those requests may require network slices or dedicated capacity.The paper therefore calls for feasibility analysis before implementation.
- B. Future directions: AI for RAN management in NaaS ecosystems: The future objective is to transform high-level service API calls into detailed RAN-level assessments while managing conflicts among network requirements.This supports autonomous management of NaaS workflows within the RAN.
- B. Future directions: AI for RAN management in NaaS ecosystems: Network context intelligence assesses time-critical requests by combining current or historical RAN information with runtime intent descriptions such as events or weather.The AI management platform is intended to execute these rapid, context-aware assessments.
- B. Future directions: AI for RAN management in NaaS ecosystems: Long-term feasibility assessment evaluates advance QoS bookings for risks including predicted cell saturation and degradation of other users’ QoE.Potential mitigations include adjusting neighboring-cell transmit power or activating additional coverage carriers.
- B. Future directions: AI for RAN management in NaaS ecosystems: Dynamic scheduling and resource allocation must adapt to each CAMARA service’s characteristics, duration, and criticality while preserving guarantees.The challenge arises because prioritized and nonprioritized users may share the cellular network, including cases of unused reserved capacity.
III. PROPOSED AGENTIC-BASED ARCHITECTURAL
The proposed framework uses three AI agents to merge external NaaS workflows with intent-driven 6G RAN management. Their responsibilities span RAN control, NaaS request feasibility and execution, and policy conflict resolution.
- Three AI agents coordinate NaaS workflows with intent-driven RAN management: RAN Control Agent, NaaS TF Agent, and Policy/Conflict Manager.The framework is presented as an initial design whose elements may be modified or redesigned.
- The RAN Control Agent executes SON policies, coordinates accepted CAMARA requests, and can interact with lower-layer agents for radio power control and scheduling.It also uses the RAN database to collect future actions associated with CAMARA requests or SON policies.
- The NaaS TF Agent assesses CAMARA request feasibility, establishes execution requirements, and coordinates network status and accepted-request execution with the RAN Control Agent.Its operation relies on context intelligence and feasibility assessment tools.
- The Policy/Conflict Manager interprets operator policies and resolves conflicts between internal RAN optimization objectives and external CAMARA demands.Conflicts may involve overlapping time, geographic area, or resource utilization, with intent-based policies supporting coherent management.
IV. CONCLUSIONS
The paper demonstrates an initial AI-agent architecture for integrating NaaS workflows into 6G RAN management, illustrated through a policy-prioritized QoS booking interaction. It identifies unresolved challenges in multi-agent coordination and digital-twin design.
- Conclusions: The initial thesis stage addresses integration of external NaaS workflows into 6G RAN management through AI-powered algorithms validated in an O-RAN-compliant framework.The framework is described as an initial NDT for future thesis stages and as an environment for reliable NaaS adoption.
- Open challenges: Complex conflict scenarios involving dynamically evolving policies or overlapping objectives remain unexplored and require advanced coordination and negotiation mechanisms among agents.The illustrative scenario assumes a straightforward operator-policy resolution.
- Example agent interaction: The example interaction prioritizes revenue over SON policies, evaluates a CAMARA QoS booking, and accepts it when the request is feasible under operator policy.The booking concerns 15 devices with a GBR of 10 Mbps for one hour; the request is then sent to the RAN Control Agent for slice creation.
- Example agent interaction: After acceptance, the RAN Control Agent reactivates an energy-saving cell when necessary, configures a dedicated slice, and returns confirmation through the NaaS TF Agent to CAMARA.The example specifies allocation for 15 UEs with a GBR of 10 Mbps in cell Y for one hour.
- Open challenges: The NDT must balance simplicity and robustness because a full O-RAN stack may impose prohibitive scalability and computational overheads.The paper motivates adaptive NDT designs tailored to AI-driven RAN management.