cs.NIMar 27, 2026
SaveML-Enabled Open RAN: A Comprehensive Survey of Architectures, Challenges, and Opportunities
Organizations: The Department of Computer and Software Engineering, Polytechnique Montreal, Montreal, QC, Canada, H3T 1J4
Abstract
As wireless communication systems become more advanced, Open Radio Access Networks (O-RAN) stand out as a notable framework that promotes interoperability and cost-effectiveness. An examination of the progression of RAN architectures, as well as O-RAN's underlying principles, reveals the importance of machine learning (ML) in addressing various challenges, including spectrum management, resource allocation, and security. Hence, this survey provides a comprehensive overview of the integration of ML within O-RAN, highlighting its transformative potential in enhancing network performance and efficiency. This survey aims to describe the current status of ML applications in O-RAN while indicating possible directions for future research by analyzing existing literature. The findings aim to assist researchers and stakeholders in formulating optimal service strategies and advancing the understanding of intelligent wireless networks.
Figures & tables
| Acronym | Definition | Acronym | Definition |
| 3GPP | 3rd Generation Partnership Project | A2C | Advantage Actor-Critic |
| ACER | Actor-Critic with Experience Replay | AI | Artificial Intelligence |
| API | Application Programming Interfaces | ARIMA | AutoRegressive Integrated Moving Average |
| BBU | Baseband Unit | BS | Base Station |
| CAPEX | Capital Expenditure | CNN | Convolutional Neural Network |
| COTS | Commercial Off-The-Shelf | CPRI | Common Public Radio Interface |
TABLE I: List of acronyms and definitions
| Ref/Year | SL | UL | RL | FL | The Challenges Tackled by ML Approach | Contributions Summary | ||
| Security | Resource Allocation | Spectrum Management | ||||||
| [ 22 ] /2022 | ✓ | - | ✓ | - | - | ✓ | ✓ | Provides a thorough review of the application of DL to O-RAN architecture through case studies and demonstrates consistent performance by automating DL modeling. |
| [ 23 ] /2022 | ✓ | ✓ | ✓ | ✓ | - | ✓ | - | Classifies ML techniques used in resource slicing management, analyze each study based on the algorithms used, challenges overcome, and types of resources allocated, compare various methods based on performance and efficiency parameters in RAN slicing, and identify practical challenges and future research directions. |
| [ 24 ] /2022 | ✓ | ✓ | ✓ | - | - | ✓ | - | Develops research framework guidelines for the efficient management of resources in 5G and beyond through the use of AI/ML. |
| [ 25 ] /2023 | - | - | - | - | ✓ | - | - | Integrates of AI/ML is one of the components of the O-RAN that are susceptible to attack. |
| [ 26 ] /2024 | - | - | ✓ | ✓ | ✓ | - | - | Provides a survey of the security aspects of O-RAN, introduce a structured taxonomy of O-RAN security threats, provide an in-depth analysis of Intrusion Detection System (IDS) in O-RAN environments, and provide a case study describing security integration in O-RAN deployments. |
TABLE II: Surveys of AI/ML in O-RAN
Fig. 1: Structure of Survey.
Fig. 2: A typical RAN serving different UEs
Fig. 3: Architecture of Distributed RAN
Fig. 4: Architecture of Centralized RAN
Fig. 5: Architecture of virtual RAN
Fig. 6: Network architecture disaggregation in O-RAN
Fig. 7: Functional architecture of O-RAN, illustrating the disaggregation of baseband processing elements into distinct units.
Fig. 8: Percentage of ML research in O-RAN
Fig. 9: Percentage of ML research pertaining to O-RAN categorized by type
| Type | Algorithm | Task | Ref |
| SL | FFNN,RNN | Traffic prediction and VNF baseband allocation | [ 96 ] |
| ARIMA | Predicting the scaling of the number of VNFs | [ 110 ] | |
| LSTM | RAN Slicing | [ 95 ] | |
| Predicting the scaling of VNFs’ number; Handover organizing | [ 110 ] | ||
| Anomaly Detection | [ 111 ] [ 112 ] | ||
| Network Energy Saving | [ 113 ] |
TABLE III: ML Utilization in O-RAN
Fig. 10: A taxonomy of ML in O-RAN.
Fig. 11: Reward function over the simulation time-steps for ACER and PPO models for O-DU computation resource allocation
Fig. 12: The power consumption at the O-DU as a function of the number of incoming requests
Fig. 13: The power consumption at the O-DU over time
Fig. 14: Comparison of Accuracy for ML Algorithms
Fig. 15: Comparison of F1-score for ML Algorithms
| Security Challenges | Description | ML Solutions | References |
| Network Architecture & Interoperability | Open, standardized interfaces increase attack surfaces, making O-RAN more vulnerable to security threats. | SL for cell traffic prediction, DRL for energy-efficiency, and adversarial defense models. | [ 21 , 176 , 4 ] |
| Ecosystem & Vendor Security | Multi-vendor environments increase risks of supply chain attacks, including tampering and supply chain poisoning. | UL for anomaly detection, predictive analytics for threat intelligence, and dynamic security mechanisms. | [ 261 , 262 , 263 ] |
| Data Protection & Privacy | Ensuring privacy and protection of sensitive data within a hyperconnected, open network, especially in 6G networks. | AI/ML-based encryption methods, DL models for secure data handling, and anomaly detection for unauthorized access. | [ 252 , 256 , 279 , 251 ] |
| AI Trust & Reliability | Vulnerabilities of AI models to adversarial attacks that can degrade performance, necessitating robust defense strategies. | Adversarial defense mechanisms, Explainable AI (XAI) for transparency, Federated Learning for secure, distributed AI. | [ 253 , 212 , 271 , 258 ] |
| Network Automation & Security | Managing the complexity of multivendor and interoperable solutions while maintaining security in an open and dynamic network. | ML-driven automation for threat detection, RL for adaptive security policies, ML-enhanced access control. | [ 249 , 250 , 29 ] |
| Open-Set Device Identification | Identifying unauthorized devices and preventing unauthorized access in an open and flexible network architecture. | CNN+LSTM models for RF signal pattern recognition, open-set detection approaches for device identification. | [ 251 , 253 ] |
TABLE IV: Security Challenges in O-RAN and Corresponding ML Solutions