Source-linked AI summary
Intelligent O-RAN for Beyond 5G and 6G Wireless Networks
Solmaz Niknam, Abhishek Roy, Harpreet S. Dhillon, Sukhdeep Singh, Rahul Banerji, Jeffery H. Reed, Navrati Saxena, Seungil Yoon
TL;DR
The paper addresses the need for more open and intelligent RAN architectures for beyond 5G and 6G. It introduces O-RAN and demonstrates an LSTM-based congestion-management scheme using real-world cellular traffic data. The reported approach predicts congestion and uses cell splitting to improve network performance, while the paper also identifies security, fronthaul, and interoperability challenges.
Problem
Existing 5G RAN architecture lacks sufficient flexibility and intelligence for diverse service demands, motivating more open and intelligent architectures for beyond 5G and 6G.
Method
The paper introduces O-RAN concepts and deploys an LSTM traffic-prediction model across non-RT and near-RT RICs to trigger cell splitting for congested cells.
Results
Preemptive cell splitting of congested cells significantly improves user-perceived IP throughput in the reported real-world traffic evaluation.
Takeaways & Limitations
O-RAN combines openness and intelligence to support adaptive radio-resource management, while practical deployment must address security, fronthaul, and multi-vendor interoperability challenges.
Abstract
from arXiv · showhide
Building on the principles of openness and intelligence, there has been a concerted global effort from the operators towards enhancing the radio access network (RAN) architecture. The objective is to build an operator-defined RAN architecture (and associated interfaces) on open hardware that provides intelligent radio control for beyond fifth generation (5G) as well as future sixth generation (6G) wireless networks. Specifically, the open-radio access network (O-RAN) alliance has been formed by merging xRAN forum and C-RAN alliance to formally define the requirements that would help achieve this objective. Owing to the importance of O-RAN in the current wireless landscape, this article provides an introduction to the concepts, principles, and requirements of the Open RAN as specified by the O-RAN alliance. In order to illustrate the role of intelligence in O-RAN, we propose an intelligent radio resource management scheme to handle traffic congestion and demonstrate its efficacy on a real-world dataset obtained from a large operator. A high-level architecture of this deployment scenario that is compliant with the O-RAN requirements is also discussed. The article concludes with key technical challenges and open problems for future research and development.
I. INTRODUCTION
Beyond 5G and 6G require more flexible and intelligent RAN architectures to handle heterogeneous services and complex demands. O-RAN addresses this through openness, virtualization, and embedded intelligence, illustrated here with an intelligent congestion-management scheme.
- 5G supports eMBB, uRLLC, and mMTC, but its architecture lacks sufficient flexibility and intelligence for these diverse demands.
- Beyond 5G and 6G therefore require architectural transformation for service heterogeneity, multi-connectivity coordination, and on-demand service deployment.
- O-RAN emerged from the merger of xRAN forum and C-RAN alliance to enhance RAN performance through virtualized elements, open interfaces, and embedded intelligence.
- O-RAN treats openness and intelligence as core principles, using learning techniques for local radio-resource allocation and network-wide efficiency.
- The article introduces O-RAN concepts and proposes an O-RAN-compliant intelligent traffic-prediction and radio-resource-management scheme.The scheme trains an LSTM model at the non-RT RIC, sends it to the near-RT RIC for inference, and applies cell splitting to congested cells.
II. PRELIMINARIES AND OVERVIEW
O-RAN extends earlier cost-oriented RAN architectures by combining open interfaces, general-purpose hardware, disaggregation, virtualization, and intelligent controllers. These features target multi-vendor interoperability, flexible deployment, and more efficient network operation.
- C-RAN and vRAN can reduce costs, but they retain proprietary software, hardware, and interfaces that limit openness.
- O-RAN uses well-defined open interfaces on general-purpose hardware and permits RRU and BBU hardware and software from different vendors.
- Disaggregation lets operators select RAN components independently, while open interfaces support multi-vendor interoperability.
- RAN virtualization enables more efficient protocol-stack splits for network slicing and supports reduced RAN expenditure.
- The non-RT RIC and near-RT RIC are O-RAN modules intended to enhance traditional network functions through intelligence.
CU-CP CU-UP CU-UP CU-CP
O-RAN’s intelligent-control workflow coordinates model capability discovery, training, deployment, inference, corrective actions, and feedback across its orchestration and RIC components.
- The workflow begins with SMO model-capability queries covering processing power, ML engines, and available data sources.
- An ML training host selects and trains a model, then sends the trained model to the non-RT RIC for deployment.
- Online data is used for inference, and the resulting policy is sent to the near-RT RIC for corrective actions.
- O1, A1, and E2 provide interfaces for management, policy guidance and model management, and near-RT RIC interaction with CU/DU components.
- Model-performance feedback returns to the training host to support model redeployment or updates.
III. PROPOSED FRAMEWORK
The proposed framework applies LSTM-based traffic prediction to anticipate congestion and reallocate radio resources within an O-RAN architecture.
- The scheme learns spatial traffic patterns with an LSTM neural network to predict possible congestion and reallocate radio resources preemptively.It is designed as a concrete O-RAN example using a commercial off-the-shelf learning technique.
A. Intelligent Radio Resource Management
The scheme predicts congestion from traffic and radio measurements, then uses cell splitting to reallocate resources and improve throughput. Its simulation uses real-world LTE measurements and evaluates different splitting orders.
- Congestion identification: A cell is congested when average user-perceived IP throughput is below 1 Mbps and DL-PRB utilization exceeds 80%.These thresholds are based on the operator SLA and can be reconfigured.
- Prediction model: The RNN predicts traffic for the next hour using two layers of 12 LSTM units trained on measurements from 17 LTE eNBs over 25 days.The dataset contains user-perceived IP throughput and downlink PRB utilization from Mumbai, India.
- Congestion relief: Cell splitting divides a congested cell’s coverage into multiple sites, including by activating sleeping eNBs or using shared infrastructure.The approach can increase capacity while supporting energy-saving and infrastructure-sharing strategies.
- Performance: Preemptive cell splitting significantly improves user-perceived IP throughput, although aggressive splitting increases channel-assignment complexity and handovers.The authors therefore recommend moderation to avoid congestion with minimal degradation of these factors.
B. Deployment Architecture
The deployment collects RAN measurements through the SMO, trains a model through the non-RT RIC, and sends it to the near-RT RIC for inference and control. O-RAN interfaces connect data collection, policy guidance, and RAN actions.
- Deployment flow: The deployment architecture is illustrated as a high-level structure and end-to-end flow for the proposed O-RAN solution.O-RAN specifies three deployment scenarios depending on training and inference locations.
- Data collection: RAN counters from control and distributed units are collected by the SMO over the O1 interface.Depending on the SMO platform, data collection may use components such as ONAP VES collectors.
- Model training: Collected data are shared with the non-RT RIC, which queries an AI server to train the machine-learning model.The described implementation uses a data bus such as Kafka and Acumos AI for the training module.
IV. CHALLENGES AND OPEN PROBLEMS
The paper identifies deployment challenges involving location-specific modeling, computational cost, security, fronthaul transport, and multi-vendor interoperability. These constraints require attention to operational setting and risk mitigation.
- Modeling requirements: Location-specific traffic patterns and infrastructure requirements can affect models for congestion self-estimation and self-healing.The discussion highlights busy-hour traffic and atmospheric conditions across ultra-dense cells.
- Security: O-RAN virtualization and SDN introduce security challenges involving VM migration, VM instantiation, hypervisors, orchestration, controllers, and shared BBU data.The paper specifically notes risks to user privacy and sensitive data access in cloud-native deployments.
- Fronthaul: CPRI-based fronthaul is strained by eMBB data rates, while uRLLC slices may face insufficient transport capacity and added latency from virtual BBUs.Ethernet transport through the 7.2x specification moderates the eMBB limitation but does not remove the harder uRLLC transport requirements.
- Interoperability: Multi-vendor O-RAN products must interoperate reliably, making incompatibility risks and mitigation strategies operational concerns.
V. SUMMARY AND CONCLUDING REMARKS
The article introduces O-RAN’s openness and intelligence principles and demonstrates intelligent congestion prediction with LSTM-based cell splitting on real-world Mumbai traffic data. It also presents an O-RAN-compliant deployment architecture and discusses remaining challenges.
- Summary and concluding remarks: O-RAN emerged from the merger of the xRAN forum and C-RAN alliance to advance openness and intelligence in next-generation RAN.
- Summary and concluding remarks: The article learns temporal traffic patterns with an LSTM neural network to predict congestion in a dense urban cellular network in Mumbai, India.
- Summary and concluding remarks: A cell-splitting radio resource management scheme and its high-level end-to-end architecture are discussed as an O-RAN-compliant practical example.
- Summary and concluding remarks: The discussion focuses on the architectural subtleties of embedding intelligence in O-RAN and uses a congestion solution selected for ease of description in an actual network.