Making Open RAN Intelligent: The Role of Machine Learning
O-RAN is fundamentally data-driven. Its intelligent control loops—enabled by its Near-Real-Time and Non-Real-Time RAN Intelligent Controllers (RICs)— support continuous monitoring, closed-loop control, and policy-driven optimization. These capabilities create a natural foundation for machine learning, allowing networks to learn from data, adapt to dynamic conditions, and make informed decisions in real time.
In our recent survey paper, “ML-Enabled Open RAN: A Comprehensive Survey of Architectures, Challenges, and Opportunities,” we examine how ML is being used to transform O-RAN from an open architecture into an intelligent, adaptive system. In this work, Mira Chandra Kirana, Patatchona Keyela, Fatemeh Rostamian, Dr. Deemah H. Tashman, and Prof. Soumaya Cherkaoui present a comprehensive survey of how machine learning can be systematically integrated into O-RAN architectures. By connecting architectural evolution with learning paradigms and system-level challenges, the authors provide both a structured overview of existing research and a clear perspective on future directions.
The paper begins by tracing the evolution of RAN architectures—from traditional distributed and centralized RAN to virtualized and fully disaggregated O-RAN—and show how intelligence becomes a necessity rather than an optional enhancement in modern networks.
Figure 1. Structure of the survey, showing the relationship between O-RAN architecture, ML techniques, key challenges, and future research directions. Reproduced from ML-Enabled Open RAN: A Comprehensive Survey of Architectures, Challenges, and Opportunities.
The central focus of the paper is the role of ML in addressing three fundamental challenges faced by O-RAN deployments: resource allocation, spectrum management, and security. The disaggregation and multi-vendor nature of O-RAN makes coordinated resource management significantly more complex. Machine learning, particularly reinforcement learning, enables dynamic and context-aware allocation strategies that respond to fluctuating traffic, heterogeneous services, and diverse quality-of-service requirements.
A key contribution of this work is its holistic perspective. Most prior surveys focus on a single challenge or a single learning paradigm, often overlooking how different ML approaches interact with O-RAN architecture as a whole. This work addresses that gap by presenting a unified taxonomy that links ML techniques to O-RAN objectives—enhancing service quality, communication quality, and security—while also discussing practical limitations such as data availability, computational overhead, scalability, and model robustness.
Beyond summarizing existing research, the paper also identifies open research directions that are critical for the future of intelligent O-RAN. These include scalability in large-scale deployments, integration with millimeter-wave and terahertz technologies, ultra-massive MIMO, mobile edge computing, digital twins, and support for ultra-reliable low-latency communications.
By bringing together architecture, machine learning techniques, system-level challenges, and future opportunities, this survey aims to serve as a reference for researchers, industry practitioners, and ecosystem stakeholders working at the intersection of wireless communications, artificial intelligence, and network softwarization.
Read the paper (Early Access on IEEE Xplore): “ML-Enabled Open RAN: A Comprehensive Survey of Architectures, Challenges, and Opportunities”