Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

EExApp: GNN-Based Reinforcement Learning for Radio Unit Energy Optimization in 5G O-RAN

Jie Lu, Peihao Yan, Huacheng Zeng

TL;DR

5G RAN energy consumption, especially RU power, motivates the need for intelligent joint optimization of sleep and slicing. EExAPP introduces a graph neural network–based dual-actor–dual-critic PPO xApp with a Transformer encoder for dynamic UE inputs and a bipartite Graph Attention Network to balance energy efficiency and QoS. OTA experiments on a real O-RAN testbed show significant RU energy reductions while maintaining QoS, outperforming SASC and other baselines. The work demonstrates practical, scalable near-real-time energy optimization for commercial O-RAN deployments and provides public code release.

Abstract

With over 3.5 million 5G base stations deployed globally, their collective energy consumption (projected to exceed 131 TWh annually) raises significant concerns over both operational costs and environmental impacts. In this paper, we present EExAPP, a deep reinforcement learning (DRL)-based xApp for 5G Open Radio Access Network (O-RAN) that jointly optimizes radio unit (RU) sleep scheduling and distributed unit (DU) resource slicing. EExAPP uses a dual-actor-dual-critic Proximal Policy Optimization (PPO) architecture, with dedicated actor-critic pairs targeting energy efficiency and quality-of-service (QoS) compliance. A transformer-based encoder enables scalable handling of variable user equipment (UE) populations by encoding all-UE observations into fixed-dimensional representations. To coordinate the two optimization objectives, a bipartite Graph Attention Network (GAT) is used to modulate actor updates based on both critic outputs, enabling adaptive tradeoffs between power savings and QoS. We have implemented EExAPP and deployed it on a real-world 5G O-RAN testbed with live traffic, commercial RU and smartphones. Extensive over-the-air experiments and ablation studies confirm that EExAPP significantly outperforms existing methods in reducing the energy consumption of RU while maintaining QoS.

EExApp: GNN-Based Reinforcement Learning for Radio Unit Energy Optimization in 5G O-RAN

TL;DR

5G RAN energy consumption, especially RU power, motivates the need for intelligent joint optimization of sleep and slicing. EExAPP introduces a graph neural network–based dual-actor–dual-critic PPO xApp with a Transformer encoder for dynamic UE inputs and a bipartite Graph Attention Network to balance energy efficiency and QoS. OTA experiments on a real O-RAN testbed show significant RU energy reductions while maintaining QoS, outperforming SASC and other baselines. The work demonstrates practical, scalable near-real-time energy optimization for commercial O-RAN deployments and provides public code release.

Abstract

With over 3.5 million 5G base stations deployed globally, their collective energy consumption (projected to exceed 131 TWh annually) raises significant concerns over both operational costs and environmental impacts. In this paper, we present EExAPP, a deep reinforcement learning (DRL)-based xApp for 5G Open Radio Access Network (O-RAN) that jointly optimizes radio unit (RU) sleep scheduling and distributed unit (DU) resource slicing. EExAPP uses a dual-actor-dual-critic Proximal Policy Optimization (PPO) architecture, with dedicated actor-critic pairs targeting energy efficiency and quality-of-service (QoS) compliance. A transformer-based encoder enables scalable handling of variable user equipment (UE) populations by encoding all-UE observations into fixed-dimensional representations. To coordinate the two optimization objectives, a bipartite Graph Attention Network (GAT) is used to modulate actor updates based on both critic outputs, enabling adaptive tradeoffs between power savings and QoS. We have implemented EExAPP and deployed it on a real-world 5G O-RAN testbed with live traffic, commercial RU and smartphones. Extensive over-the-air experiments and ablation studies confirm that EExAPP significantly outperforms existing methods in reducing the energy consumption of RU while maintaining QoS.
Paper Structure (18 sections, 12 equations, 9 figures, 3 tables, 1 algorithm)

This paper contains 18 sections, 12 equations, 9 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. System Modeling
  3. A Primer on 5G NR and O-RAN
  4. Problem Formulation
  5. MDP Modeling and Relaxation
  6. EExApp: Design
  7. Overview
  8. Transformer Encoder for Dynamic Input Representation
  9. Dual-Actor-Critic Architecture
  10. GAT for Critic Aggregation
  11. Training Logic
  12. Experimental Evaluation
  13. Implementation and Experimental Setup
  14. Case Studies under Diverse Network Conditions
  15. Ablation Studies
  16. ...and 3 more sections

Figures (9)

  • Figure 1: The radio transmission pattern instances of a 5G base station.
  • Figure 2: Illustrating the online policy for joint resource slicing and power-saving optimization at an O-RAN RU.
  • Figure 3: The architecture of EExAPP in O-RAN system. The EExAPP is deployed on the Near-RT RIC, interacting with the RAN (CU and DU) to perform closed-loop control. Internally, the framework utilizes a Transformer encoder to extract latent features from real-time network states $\boldsymbol{s}_t$. A dual-actor-dual-critic structure is employed to decouple the optimization of energy efficiency (EE) and resource slicing (RS), coordinated by a bipartite GAT that aggregates critic values to balance conflicting objectives in the joint action $\boldsymbol{a}_t$.
  • Figure 4: Our O-RAN system for experimental evaluation. The left subfigure shows the floor plan of our system deployment, and the right subfigure shows the network elements and architecture.
  • Figure 5: Convergence performance comparison of EExAPP and SASC under light, medium, and heavy traffic conditions.
  • ...and 4 more figures