Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

On AI Verification in Open RAN

Rahul Soundrarajan, Claudio Fiandrino, Michele Polese, Salvatore D'Oro, Leonardo Bonati, Tommaso Melodia

TL;DR

This paper investigates trustworthy AI in Open RAN by arguing that explainability alone is insufficient for reliable network operation. It proposes a lightweight verification strategy using a Decision Tree-based verifier to check DRL-driven slice control and scheduling in near-real time, integrated within the O-RAN ai/ml lifecycle. The authors analyze the roles of XAI and AI verification, map an architectural integration, and demonstrate feasibility with a DT-based slice-verifier trained on KPI data (augmented via CTGAN) and evaluated against DRL agents for embb, mmtc, and urllc slices. They also outline open challenges, including system-level verification, scalability-soundness trade-offs, and phased prototyping, to guide future work toward robust, trustworthy AI adoption in Open RAN.

Abstract

Open RAN introduces a flexible, cloud-based architecture for the Radio Access Network (RAN), enabling Artificial Intelligence (AI)/Machine Learning (ML)-driven automation across heterogeneous, multi-vendor deployments. While EXplainable Artificial Intelligence (XAI) helps mitigate the opacity of AI models, explainability alone does not guarantee reliable network operations. In this article, we propose a lightweight verification approach based on interpretable models to validate the behavior of Deep Reinforcement Learning (DRL) agents for RAN slicing and scheduling in Open RAN. Specifically, we use Decision Tree (DT)-based verifiers to perform near-real-time consistency checks at runtime, which would be otherwise unfeasible with computationally expensive state-of-the-art verifiers. We analyze the landscape of XAI and AI verification, propose a scalable architectural integration, and demonstrate feasibility with a DT-based slice-verifier. We also outline future challenges to ensure trustworthy AI adoption in Open RAN.

On AI Verification in Open RAN

TL;DR

This paper investigates trustworthy AI in Open RAN by arguing that explainability alone is insufficient for reliable network operation. It proposes a lightweight verification strategy using a Decision Tree-based verifier to check DRL-driven slice control and scheduling in near-real time, integrated within the O-RAN ai/ml lifecycle. The authors analyze the roles of XAI and AI verification, map an architectural integration, and demonstrate feasibility with a DT-based slice-verifier trained on KPI data (augmented via CTGAN) and evaluated against DRL agents for embb, mmtc, and urllc slices. They also outline open challenges, including system-level verification, scalability-soundness trade-offs, and phased prototyping, to guide future work toward robust, trustworthy AI adoption in Open RAN.

Abstract

Open RAN introduces a flexible, cloud-based architecture for the Radio Access Network (RAN), enabling Artificial Intelligence (AI)/Machine Learning (ML)-driven automation across heterogeneous, multi-vendor deployments. While EXplainable Artificial Intelligence (XAI) helps mitigate the opacity of AI models, explainability alone does not guarantee reliable network operations. In this article, we propose a lightweight verification approach based on interpretable models to validate the behavior of Deep Reinforcement Learning (DRL) agents for RAN slicing and scheduling in Open RAN. Specifically, we use Decision Tree (DT)-based verifiers to perform near-real-time consistency checks at runtime, which would be otherwise unfeasible with computationally expensive state-of-the-art verifiers. We analyze the landscape of XAI and AI verification, propose a scalable architectural integration, and demonstrate feasibility with a DT-based slice-verifier. We also outline future challenges to ensure trustworthy AI adoption in Open RAN.
Paper Structure (13 sections, 3 figures, 2 tables)

This paper contains 13 sections, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. AI Verification and XAI
  3. Applications and Benefits of XAI
  4. The Current Landscape of AI Verification
  5. AI Verification in the Open RAN
  6. The Verification Use Case
  7. The DRL Agents
  8. Verifying Agent Behavior with Decision Trees
  9. Open Challenges and Future Directions
  10. Open ran-Oriented Verification
  11. Tradeoff Scalability-Soundness
  12. Prototyping and Testing
  13. Concluding Remarks

Figures (3)

  • Figure 1: O-RAN architecture and ai/ml workflows, focusing on model execution, explainability, and verification.
  • Figure 2: Operation of the drl agent xApp and the DT-based verifier xApp
  • Figure 3: The slice-verifier: training and verification capabilities. Case A shows a "drift" in the feature space for a specific slice, case B highlights a misclassification, and case C illustrates conflicting predictions for data points in a similar feature space at different times.