Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

xApp Distillation: AI-based Conflict Mitigation in B5G O-RAN

Hakan Erdol, Xiaoyang Wang, Robert Piechocki, George Oikonomou, Arjun Parekh

TL;DR

The paper addresses inter-xApp conflicts in O-RAN by introducing xApp distillation, which learns from multiple pre-trained xApps through policy distillation to yield a single, non-conflicting controller. It builds a multi-headed DQN to handle handover, RB allocation, and power control, and uses a teacher-student KL-divergence approach with a replay buffer to fuse knowledge from conflicting xApps. The method demonstrates superior QoS, higher downlink data rates, and significantly reduced network outages compared with both individually trained xApps and team-learning approaches. This approach enables robust, scalable conflict mitigation in B5G O-RAN and offers practical QoS improvements for multi-xApp deployments.

Abstract

The advancements of machine learning-based (ML) decision-making algorithms created various research and industrial opportunities. One of these areas is ML-based near-real-time network management applications (xApps) in Open-Radio Access Network (O-RAN). Normally, xApps are designed solely for the desired objectives, and fine-tuned for deployment. However, telecommunication companies can employ multiple xApps and deploy them in overlapping areas. Consider the different design objectives of xApps, the deployment might cause conflicts. To prevent such conflicts, we proposed the xApp distillation method that distills knowledge from multiple xApps, then uses this knowledge to train a single model that has retained the capabilities of Previous xApps. Performance evaluations show that compared conflict mitigation schemes can cause up to six times more network outages than xApp distillation in some cases.

xApp Distillation: AI-based Conflict Mitigation in B5G O-RAN

TL;DR

The paper addresses inter-xApp conflicts in O-RAN by introducing xApp distillation, which learns from multiple pre-trained xApps through policy distillation to yield a single, non-conflicting controller. It builds a multi-headed DQN to handle handover, RB allocation, and power control, and uses a teacher-student KL-divergence approach with a replay buffer to fuse knowledge from conflicting xApps. The method demonstrates superior QoS, higher downlink data rates, and significantly reduced network outages compared with both individually trained xApps and team-learning approaches. This approach enables robust, scalable conflict mitigation in B5G O-RAN and offers practical QoS improvements for multi-xApp deployments.

Abstract

The advancements of machine learning-based (ML) decision-making algorithms created various research and industrial opportunities. One of these areas is ML-based near-real-time network management applications (xApps) in Open-Radio Access Network (O-RAN). Normally, xApps are designed solely for the desired objectives, and fine-tuned for deployment. However, telecommunication companies can employ multiple xApps and deploy them in overlapping areas. Consider the different design objectives of xApps, the deployment might cause conflicts. To prevent such conflicts, we proposed the xApp distillation method that distills knowledge from multiple xApps, then uses this knowledge to train a single model that has retained the capabilities of Previous xApps. Performance evaluations show that compared conflict mitigation schemes can cause up to six times more network outages than xApp distillation in some cases.
Paper Structure (19 sections, 6 equations, 4 figures, 1 table)

This paper contains 19 sections, 6 equations, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. System model
  3. Connectivity model
  4. Resource management and xApp design
  5. Handover
  6. Resource block allocation
  7. Transmit power level control
  8. xApp Distillation
  9. Conflicting xApps
  10. Deep Q-Network
  11. State
  12. Action
  13. Reward
  14. Policy distillation
  15. Conflict mitigation scheme
  16. ...and 4 more sections

Figures (4)

  • Figure 1: O-RAN Architecture in the proposed system model.
  • Figure 2: Stage 3 - Distilling knowledge from pre-trained xApps to a student policy.
  • Figure 3: PDF of throughput comparison between the xApp Distillation method and O-RAN conflict mitigation with models trained either with individual or team-learning scheme.
  • Figure 4: Network outage comparison in percentage between the xApp Distillation method and O-RAN conflict mitigation with models trained either with the individual or team-learning scheme.