Beyond Connectivity: An Open Architecture for AI-RAN Convergence in 6G

TL;DR

The paper argues that data-intensive AI workloads must be co-located with the RAN to enable distributed edge AI and new monetization models. It proposes a converged O-RAN and AI-RAN architecture featuring an AI-RAN Orchestrator (AI-SMO) and AI-RAN edge sites, enabling unified management of heterogeneous AI and RAN workloads across distributed infrastructure. It defines AI-RAN principles (AI-for-RAN, AI-on-RAN, AI-and-RAN), open interfaces, and two orchestration workflows (batch and real-time), and provides a prototype evaluation demonstrating feasible coexistence with measurable throughput and deployment latency. The work lays out a concrete blueprint for open, modular, cloud-native AI-RAN integration, and discusses next steps toward end-to-end prototyping and standardization alignment.

Abstract

Data-intensive Artificial Intelligence (AI) applications at the network edge demand a fundamental shift in Radio Access Network (RAN) design, from merely consuming AI for network optimization, to actively enabling distributed AI workloads. This presents a significant opportunity for network operators to monetize AI while leveraging existing infrastructure. To realize this vision, this article presents a novel converged O-RAN and AI-RAN architecture for unified orchestration and management of telecommunications and AI workloads on shared infrastructure. The proposed architecture extends the Open RAN principles of modularity, disaggregation, and cloud-nativeness to support heterogeneous AI deployments. We introduce two key architectural innovations: (i) the AI-RAN Orchestrator, which extends the O-RAN Service Management and Orchestration (SMO) to enable integrated resource and allocation across RAN and AI workloads; and (ii) AI-RAN sites that provide distributed edge AI platforms with real-time processing capabilities. The proposed architecture enables flexible orchestration, meeting requirements for managing heterogeneous workloads at different time scales while maintaining open, standardized interfaces and multi-vendor interoperability.

Figures (6)

• Figure 1: Architecture enabling AI and RAN coexistence, integrated within the proposed extension of the O-RAN framework. The figure also includes an example of use case for AI-and-RAN coexistence. An application developer submits an AI-on-RAN workload to enable XR processing at the edge. In the meantime, a vendor provides an updated AI-for-RAN slicing solution. The operator defines the policy that the infrastructure should follow to enable both services. The AI-RAN orchestrator translates the policy into resource allocation, and coordinates with the automation workflows to deploy the services in the AI-RAN sites.
• Figure 2: Extension of the O-RAN to support AI-RAN orchestration. The components in light blue, together with the , are part of the O-RAN architecture. Components in dark blue represent extensions to accommodate AI-RAN requirements.
• Figure 3: Architecture for an edge cloud site that supports AI-RAN solutions on a compute cluster managed by the AI-O-Cloud. The figure shows the AI workloads and RAN workloads (top), the orchestration components (middle), and the infrastructure and interfaces (bottom).
• Figure 4: Procedure for the authentication and deployment of real-time AI tasks on the AI-RAN infrastructure.
• Figure 5: Median values for throughput and CRC error rate with RAN only (solid bars) or AI-RAN coexistence with LLM prompts and CNN/ResNet (dotted bars).