When Large Language Models Meet Optical Networks: Paving the Way for Automation

Danshi Wang, Yidi Wang, Xiaotian Jiang, Yao Zhang, Yue Pang, Min Zhang

Abstract

Since the advent of GPT, large language models (LLMs) have brought about revolutionary advancements in all walks of life. As a superior natural language processing (NLP) technology, LLMs have consistently achieved state-of-the-art performance on numerous areas. However, LLMs are considered to be general-purpose models for NLP tasks, which may encounter challenges when applied to complex tasks in specialized fields such as optical networks. In this study, we propose a framework of LLM-empowered optical networks, facilitating intelligent control of the physical layer and efficient interaction with the application layer through an LLM-driven agent (AI-Agent) deployed in the control layer. The AI-Agent can leverage external tools and extract domain knowledge from a comprehensive resource library specifically established for optical networks. This is achieved through user input and well-crafted prompts, enabling the generation of control instructions and result representations for autonomous operation and maintenance in optical networks. To improve LLM's capability in professional fields and stimulate its potential on complex tasks, the details of performing prompt engineering, establishing domain knowledge library, and implementing complex tasks are illustrated in this study. Moreover, the proposed framework is verified on two typical tasks: network alarm analysis and network performance optimization. The good response accuracies and semantic similarities of 2,400 test situations exhibit the great potential of LLM in optical networks.

Introduction

Optical networks serve as the critical backbone of global communication infrastructure, comprising a complex array of physical components such as transceivers, optical amplifiers, and wavelength selective switches. Currently, the operation and maintenance (O&M) of these networks remain largely manual and labor-intensive, requiring specialized experts to monitor telemetry, troubleshoot failures, and optimize performance. While traditional "small-scale" AI models have addressed specific tasks, they often lack the general reasoning capabilities and cross-domain flexibility required for end-to-end automation.

Large Language Models (LLMs) like GPT-4 offer significant reasoning power but face hurdles in specialized fields. They are prone to "hallucinations" and struggle with niche technical problems, such as fiber-optic link optimization or complex alarm correlation, due to a lack of specific domain knowledge. This research bridges that gap by introducing a specialized AI-Agent framework designed to navigate the complexities of the optical physical layer.

The LLM-Empowered Framework

The proposed architecture organizes the network into three distinct layers: the Application Layer, the Control Layer, and the Physical Layer. The core of this system is the AI-Agent, situated within the Control Layer, which acts as the intelligent intermediary between high-level user requirements and low-level physical configurations.

Overview of the LLM-empowered optical network framework, illustrating the interaction between the Application, Control, and Physical layers.

In this framework, the AI-Agent performs several critical functions:

  • Intentional Analysis: Interpreting natural language requests from network engineers or application-layer services.
  • Task Decomposition: Breaking down complex network management goals into manageable sub-tasks.
  • Resource Selection: Identifying the necessary real-time telemetry data or expert knowledge required for a specific task.
  • Instruction Generation: Producing precise control instructions for the physical layer and summarizing outcomes for the application layer.

AI-Agent Workflow and Domain Resource Library

To transform a general-purpose LLM into a domain expert, the framework utilizes Retrieval-Augmented Generation (RAG). This involves connecting the LLM to a specialized Domain Resource Library containing both a real-time database (telemetry, device states) and a knowledge base (manuals, expert rules, and protocols).

Detailed workflow of the AI-Agent, showing the process of text segmentation, embedding, vector storage, and retrieval to support the LLM in generating accurate responses.

As shown in the workflow, domain knowledge is segmented, embedded into vectors, and stored. When a user query is received, the system performs vector similarity searches to retrieve the most relevant text chunks. These chunks are then integrated into a structured prompt template. This ensures the LLM bases its reasoning on factual, domain-specific data rather than internal weights alone, significantly reducing errors.

Furthermore, advanced prompt engineering techniques such as Chain-of-Thought (CoT) and Tree-of-Thoughts (ToT) are employed. These methods guide the model to solve problems step-by-step or explore multiple reasoning paths, which is essential for complex tasks like root-cause analysis in alarm logs.

Experimental Validation and Results

The framework was validated across 2,400 test scenarios involving two primary tasks: network alarm analysis and autonomous performance optimization.

  1. Network Alarm Analysis: Using 2,500 real-world alarms from an Operational Transport Network (OTN), the agent demonstrated the ability to compress redundant logs and accurately identify root causes, matching the performance of human experts.
  2. Autonomous Optimization: Tested on the CONUS (Continental US) network topology, the agent performed Quality of Transmission (QoT) estimation and power profile optimization. While "raw" GPT-4 struggled with these specialized tasks, the AI-Agent equipped with the Domain Library and advanced prompting achieved an accuracy score of approximately 96%.

Semantic similarity tests using Sentence-BERT confirmed that the agent's outputs were not just plausible, but technically aligned with expert solutions.

Conclusion and Future Outlook

This study demonstrates that LLMs, when integrated into a structured AI-Agent framework with RAG and advanced prompting, can effectively automate complex optical network management. However, several challenges remain. Security is paramount; giving an LLM direct control over physical infrastructure necessitates robust verification modules to ensure generated instructions do not compromise network stability. Future work will focus on developing these automated "checkers" and refining the AI-Agent's ability to handle even more dynamic, multi-vendor network environments.

Summary updated

Abstract

Since the advent of GPT, large language models (LLMs) have brought about revolutionary advancements in all walks of life. As a superior natural language processing (NLP) technology, LLMs have consistently achieved state-of-the-art performance on numerous areas. However, LLMs are considered to be general-purpose models for NLP tasks, which may encounter challenges when applied to complex tasks in specialized fields such as optical networks. In this study, we propose a framework of LLM-empowered optical networks, facilitating intelligent control of the physical layer and efficient interaction with the application layer through an LLM-driven agent (AI-Agent) deployed in the control layer. The AI-Agent can leverage external tools and extract domain knowledge from a comprehensive resource library specifically established for optical networks. This is achieved through user input and well-crafted prompts, enabling the generation of control instructions and result representations for autonomous operation and maintenance in optical networks. To improve LLM's capability in professional fields and stimulate its potential on complex tasks, the details of performing prompt engineering, establishing domain knowledge library, and implementing complex tasks are illustrated in this study. Moreover, the proposed framework is verified on two typical tasks: network alarm analysis and network performance optimization. The good response accuracies and sematic similarities of 2,400 test situations exhibit the great potential of LLM in optical networks.
BibTeX
@misc{wang2024when,
  author = {Danshi Wang and Yidi Wang and Xiaotian Jiang and Yao Zhang and Yue Pang and Min Zhang},
  title = {{When Large Language Models Meet Optical Networks: Paving the Way for Automation}},
  year = {2024},
  eprint = {2405.17441},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2405.17441},
}
Figure