With the foreseen rapidly changing environment approaching with the sixth generation (6G), the Open Radio Access Network (RAN) paradigm has transformed cellular networks into visualised, software-based components. Artificial Intelligence (AI) and Machine Learning (ML) are at the core of this transformation, particularly through Deep Reinforcement Learning (DRL), which has shown great promise in solving complex resource allocation challenges. However, with great potential comes significant complexity, especially when it comes to understanding the decisions made by these AI models.
EXPLORA, a new framework developed to bring explainability to DRL-based solutions for Open RAN, EXPLORA offers a transparent window into how AI makes decisions in mobile networks. This publication explores the potential of EXPLORA and how it addresses the critical need for explainable AI in modern telecom infrastructures.
A publication by Claudio Fiandrino, Leonardo Bonati, Salvatore D’Oro, Michele Polese, Tommaso Melodia, and Jeorg Widmer.
The challenge of AI in open ran
One of the biggest hurdles in applying DRL to Open RAN is the inherent opacity of its decision-making processes. Unlike more straightforward models like decision trees, which offer visible reasoning paths, DRL solutions operate within deep neural networks. These structures are often seen as ‘black boxes,’ making it difficult to discern how specific actions are chosen. This opacity poses risks in terms of deployment in live networks, as operators cannot easily predict or troubleshoot the AI’s behaviour, leading to trust issues and operational inefficiencies.
Bridging the gap between ai and human understanding
EXPLORA is designed to tackle this challenge head-on by providing explainable insights into DRL-driven decisions within Open RAN environments. It does this through an innovative use of attributed graphs that link DRL actions to their environmental impacts. In essence, EXPLORA allows operators to see not only what decisions the AI is making but also why those decisions are being made and the conditions under which they occur. This is crucial in dynamic and complex systems like Open RAN, where network conditions can change rapidly.
By synthesising network-oriented explanations, EXPLORA offers a lightweight, real-time solution that ensures AI-driven decisions are not only more transparent but also more controllable. Operators can use these explanations to perform intent-based action steering, allowing them to guide the AI’s actions more effectively, ultimately leading to better network performance and reliability.
Real-world testing – the colosseum experiment
EXPLORA has been rigorously tested on the Colosseum, the world’s largest wireless network emulator. This testing involved deploying DRL agents on an O-RAN-compliant near-real-time RAN Intelligent Controller (RIC) and measuring the performance of different resource allocation and control strategies. The results were promising, showcasing median transmission bitrate improvements of 4% and tail improvements of 10%.
Fig.1. EXPLORA interaction with a DRL agent
What sets EXPLORA apart is its ability to explain not only the outcomes of AI decisions but also the context in which these decisions are made. This deeper understanding allows operators to predict future actions and proactively adjust network policies to prevent performance degradation.
As we move towards the 6G era, the demand for more flexible, efficient, and explainable network management systems will only grow. AI will continue to play a central role in this transformation, but it must be paired with tools like EXPLORA to ensure transparency and trust. By making AI explainable, EXPLORA not only boosts operational efficiency but also paves the way for more widespread adoption of AI-driven network management solutions.
EXPLORA represents a significant leap forward in the integration of AI into mobile networks. Its ability to provide clear, network-oriented explanations for complex DRL decisions makes it a critical tool for the future of Open RAN, ensuring that AI can be trusted to manage the networks that power our increasingly connected world.