Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Leveraging AI Agents for Autonomous Networks: A Reference Architecture and Empirical Studies

Binghan Wu, Shoufeng Wang, Yunxin Liu, Ya-Qin Zhang, Joseph Sifakis, Ye Ouyang

TL;DR

The paper defines Level-4 autonomous networks and TM Forum Three-Self/Three-Zero objectives, and argues that reaching L4 requires agents with proactive, self-governing capabilities beyond traditional ML-driven autonomy. It then presents a dual-driver AN Agent reference architecture, combining reactive and proactive cognition with a hybrid long-term memory, coordinated by a Workflow Coordinator Runtime, and validated through a Radio Access Network Link Adaptation case. The LA agent delivers sub-10 ms real-time control and tangible gains—approximately 4% higher downlink throughput and 85% lower BLER—versus an OLLA baseline, with ablation confirming the importance of look-ahead perception (LSTM) and world-knowledge (RAG) modules. Overall, the work demonstrates the practical viability of L4 autonomy in 5G/6G networks and outlines a path toward a Society of Agents that share knowledge to satisfy Three-Zero objectives across multi-domain telecom environments.

Abstract

The evolution toward Level 4 (L4) Autonomous Networks (AN) represents a strategic inflection point in telecommunications, where networks must transcend reactive automation to achieve genuine cognitive capabilities--fulfilling TM Forum's vision of self-configuring, self-healing, and self-optimizing systems that deliver zero-wait, zero-touch, and zero-fault services. This work bridges the gap between architectural theory and operational reality by implementing Joseph Sifakis's AN Agent reference architecture in a functional cognitive system, deploying coordinated proactive-reactive runtimes driven by hybrid knowledge representation. Through an empirical case study of a Radio Access Network (RAN) Link Adaptation (LA) Agent, we validate this framework's transformative potential: demonstrating sub-10 ms real-time control in 5G NR sub-6 GHz while achieving 4% higher downlink throughput than Outer Loop Link Adaptation (OLLA) algorithms and 85% Block Error Rate (BLER) reduction for ultra-reliable services through dynamic Modulation and Coding Scheme (MCS) optimization. These improvements confirm the architecture's viability in overcoming traditional autonomy barriers and advancing critical L4-enabling capabilities toward next-generation objectives.

Leveraging AI Agents for Autonomous Networks: A Reference Architecture and Empirical Studies

TL;DR

The paper defines Level-4 autonomous networks and TM Forum Three-Self/Three-Zero objectives, and argues that reaching L4 requires agents with proactive, self-governing capabilities beyond traditional ML-driven autonomy. It then presents a dual-driver AN Agent reference architecture, combining reactive and proactive cognition with a hybrid long-term memory, coordinated by a Workflow Coordinator Runtime, and validated through a Radio Access Network Link Adaptation case. The LA agent delivers sub-10 ms real-time control and tangible gains—approximately 4% higher downlink throughput and 85% lower BLER—versus an OLLA baseline, with ablation confirming the importance of look-ahead perception (LSTM) and world-knowledge (RAG) modules. Overall, the work demonstrates the practical viability of L4 autonomy in 5G/6G networks and outlines a path toward a Society of Agents that share knowledge to satisfy Three-Zero objectives across multi-domain telecom environments.

Abstract

The evolution toward Level 4 (L4) Autonomous Networks (AN) represents a strategic inflection point in telecommunications, where networks must transcend reactive automation to achieve genuine cognitive capabilities--fulfilling TM Forum's vision of self-configuring, self-healing, and self-optimizing systems that deliver zero-wait, zero-touch, and zero-fault services. This work bridges the gap between architectural theory and operational reality by implementing Joseph Sifakis's AN Agent reference architecture in a functional cognitive system, deploying coordinated proactive-reactive runtimes driven by hybrid knowledge representation. Through an empirical case study of a Radio Access Network (RAN) Link Adaptation (LA) Agent, we validate this framework's transformative potential: demonstrating sub-10 ms real-time control in 5G NR sub-6 GHz while achieving 4% higher downlink throughput than Outer Loop Link Adaptation (OLLA) algorithms and 85% Block Error Rate (BLER) reduction for ultra-reliable services through dynamic Modulation and Coding Scheme (MCS) optimization. These improvements confirm the architecture's viability in overcoming traditional autonomy barriers and advancing critical L4-enabling capabilities toward next-generation objectives.

Paper Structure

This paper contains 13 sections, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. From AI Agent to AN Agent
  3. Current AI Agent Architectures
  4. AN Agent Reference Architecture
  5. Building Autonomous Network Agents: Implementation Framework and Technical Realization
  6. Executable Runtime Instantiation
  7. Technology Selection for Functional Modules
  8. Case Study: Radio Access Network (RAN) Link Adaptation (LA) Agent
  9. Problem Modeling
  10. Engineering Details
  11. Experiment Environment and Results
  12. Conclusion and Future Direction
  13. Biography

Figures (4)

  • Figure 1: Sifakis's Agent architecture with proactive (processes a--g) and reactive behaviors (processes 1--6).
  • Figure 2: A flat view of an agent.
  • Figure 3: Test site environment. Lab wireless setup includes: 1x BBU, 2x RRUs, 2x test UEs, and 1x measurement/control host.
  • Figure 4: Comparative performance evaluation of LA Agent versus OLLA algorithm.