Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

LLM-hRIC: LLM-empowered Hierarchical RAN Intelligent Control for O-RAN

Lingyan Bao, Sinwoong Yun, Jemin Lee, Tony Q. S. Quek

TL;DR

This work tackles how to enhance O-RAN resource management by bridging a global, LLM-enabled non-RT RIC with a local, RL-based near-RT RIC. The proposed LLM-hRIC framework uses $y = \text{LLM}_\theta(x_1, x_2, \cdots, x_N)$ as strategic guidance and $ \boldsymbol{a} = \text{RL}_\phi(\mathbf{y}, \mathbf{o})$ for near-RT decisions, demonstrated on an IAB power-allocation use case. The results show that LLM-guided exploration and policy blending yield higher throughput and faster convergence than baselines, with insights into the trade-off between backhaul and access. The paper also discusses future challenges, including multi-modal data integration, domain-specific fine-tuning, cross-RIC collaboration, and inference efficiency, framing a path toward scalable, adaptive O-RAN systems.

Abstract

Despite recent advances in applying large language models (LLMs) and machine learning (ML) techniques to open radio access network (O-RAN), critical challenges remain, such as insufficient cooperation between radio access network (RAN) intelligent controllers (RICs), high computational demands hindering real-time decisions, and the lack of domain-specific finetuning. Therefore, this article introduces the LLM-empowered hierarchical RIC (LLM-hRIC) framework to improve the collaboration between RICs in O-RAN. The LLM-empowered non-real-time RIC (non-RT RIC) acts as a guider, offering a strategic guidance to the near-real-time RIC (near-RT RIC) using global network information. The RL-empowered near-RT RIC acts as an implementer, combining this guidance with local real-time data to make near-RT decisions. We evaluate the feasibility and performance of the LLM-hRIC framework in an integrated access and backhaul (IAB) network setting, and finally, discuss the open challenges of the LLM-hRIC framework for O-RAN.

LLM-hRIC: LLM-empowered Hierarchical RAN Intelligent Control for O-RAN

TL;DR

This work tackles how to enhance O-RAN resource management by bridging a global, LLM-enabled non-RT RIC with a local, RL-based near-RT RIC. The proposed LLM-hRIC framework uses as strategic guidance and for near-RT decisions, demonstrated on an IAB power-allocation use case. The results show that LLM-guided exploration and policy blending yield higher throughput and faster convergence than baselines, with insights into the trade-off between backhaul and access. The paper also discusses future challenges, including multi-modal data integration, domain-specific fine-tuning, cross-RIC collaboration, and inference efficiency, framing a path toward scalable, adaptive O-RAN systems.

Abstract

Despite recent advances in applying large language models (LLMs) and machine learning (ML) techniques to open radio access network (O-RAN), critical challenges remain, such as insufficient cooperation between radio access network (RAN) intelligent controllers (RICs), high computational demands hindering real-time decisions, and the lack of domain-specific finetuning. Therefore, this article introduces the LLM-empowered hierarchical RIC (LLM-hRIC) framework to improve the collaboration between RICs in O-RAN. The LLM-empowered non-real-time RIC (non-RT RIC) acts as a guider, offering a strategic guidance to the near-real-time RIC (near-RT RIC) using global network information. The RL-empowered near-RT RIC acts as an implementer, combining this guidance with local real-time data to make near-RT decisions. We evaluate the feasibility and performance of the LLM-hRIC framework in an integrated access and backhaul (IAB) network setting, and finally, discuss the open challenges of the LLM-hRIC framework for O-RAN.

Paper Structure

This paper contains 17 sections, 1 equation, 5 figures.

Table of Contents

  1. Introduction
  2. LLM-empowerd hierarchical RIC Framework
  3. LLM-Empowered non-RT RIC
  4. RL-Empowered near-RT RIC
  5. Use Case: Power Allocation in IAB Network
  6. Power Allocation in IAB Networks
  7. LLM-hRIC Framework for Power Allocation
  8. LLM-Empowered non-RT RIC
  9. RL-Empowered near-RT RIC
  10. LLM-assisted Three Phases Training for RL
  11. Experimental Results
  12. Future Challenges
  13. Leveraging Multi-Model Data in LLM-hRIC
  14. Fine-Tuning of LLM
  15. Systematic Collaboration between RIC
  16. ...and 2 more sections

Figures (5)

  • Figure 1: Structure of the proposed LLM-hRIC framework.
  • Figure 2: The Prompt Structure.
  • Figure 3: Training curve for proposed and baseline methods over 500 training epochs when $M=3, N=6, W=100\text{Mb}, P_m^{\text{max}}=44\text{dBm}$, and $\alpha=0.5$.
  • Figure 4: Average total throughput of the proposed LLM-hRIC and baseline methods according to backhaul-access bandwidth partitioning ratio $\alpha$ over 50 test examples.
  • Figure 5: Future challenges of the LLM-hRIC for O-RAN.