- The paper introduces a systematic analysis of the emerging 3GPP 6G RAN architecture by examining key design elements such as the CU-DU split and the explicit recognition of the RU.
- The paper demonstrates how architectural enhancements, including centralized processing and fine-grained scaling via CU-CP/CU-UP splits, enable flexible deployments.
- The paper details a comparative study of RAN-CN interface options, contrasting traditional P2P signaling with service-based models to optimize connectivity and support AI-driven services.
Overview of the Emerging 3GPP 6G RAN Architecture
Introduction
The paper presents a systematic analysis of the nascent 3GPP 6G RAN architecture, with specific focus on interface boundary selection, logical function splits, and architectural enhancements relative to the 5G baseline. The discussion contextualizes architectural evolution within ongoing Release-20 3GPP studies, emphasizing both the preservation of NG-RAN modularity and the adaptation towards new service requirements. Key areas include the central unit–distributed unit (CU-DU) split, recognition of the radio unit (RU) as a logical entity, and critical choices in RAN-core network (CN) interface design.
Evolution of RAN Architecture: From 5G NG-RAN to 6G
The 5G NG-RAN architecture fundamentally relies on decomposition of base station functionality into CU and DU components. This modularization, underpinned by well-defined protocols (e.g., F1, E1, Xn, NG), enables flexible deployments, computational pooling, and multi-vendor operation. The 6G architecture’s foundation is the 5G model, with the aNB (the 6G equivalent of gNB) as the main RAN node. The gNB-CU handles higher-layer functions (RRC, SDAP, PDCP), while the gNB-DU manages lower-layer radio functions (RLC, MAC, PHY), segmented by a standardized interface at PDCP/RLC.
Procedural and stateful operations that require CP-UP separation are handled with the E1 interface, providing independent CP and UP scaling. This is critical for high-density and variable-load scenarios. At the RAN-CN boundary, the NG interface supports both CP (N2 towards the AMF) and UP (N3 towards the UPF), reflecting broader 5G system separation of mobility/context handling and user data transfer.
CU-DU Split in 6G: Architectural Drivers and Enhancements
The 6G RAN architecture retains the CU-DU split for several technically grounded reasons:
- Centralized Processing and Resource Pooling: Aggregating traffic and signaling workloads across diverse DUs exploits statistical multiplexing, reducing over-provisioning and CAPEX.
- Fine-Grained Scaling: With further CU-CP/CU-UP split, operators can independently scale resources aligned with traffic, coverage, and service diversity typical of urban/rural/private deployments.
- Multi-Vendor Interoperability: A rigorously-defined CU-DU protocol boundary allows independent vendor innovation below/above the split.
- Cloud RAN Enablement: Placement flexibility (DU at edge, CU in regional clouds) aligns with diverse latency, bandwidth, and reliability requirements.
Persistent challenges include consistent UE context management, latency induced by procedural round-trips between entities, flow-control between PDCP (CU) and RLC/MAC (DU), and implementation divergence despite protocol commonality. The paper underscores that improvement in 6G will focus on clear state ownership, streamlined configuration, service-aware flow-control, and well-scoped interoperability profiles, ultimately aiming to reduce split-induced coordination costs and operational complexity.
Recognition of the Radio Unit (RU): Architectural and Implementation Implications
One of the most significant shifts is the explicit architectural recognition of the RU as a distinct logical unit within the aNB. Unlike 5G, where RU and fronthaul were outside the 3GPP logical decomposition, 6G acknowledges RU due to its central role in advanced radio processing (massive MIMO, beamforming, calibration) and deployment diversity (macro/micro/pico, public/private, O-RAN, eCPRI, proprietary).
However, 3GPP intentionally abstains from specifying RU functions or fronthaul protocols, maintaining focus on higher-level system interoperability while deferring RU protocols to ecosystem partners such as O-RAN Alliance. This layered interoperability model allows aNB implementations with O-RAN, eCPRI, or proprietary RU-baseband connectivity, decoupling system-level RAN innovation from deployment-specific radio optimizations.
RAN-Core Network (CN) Interface: P2P vs. Service-Based Models
The 6G architecture debate over RAN-CN interfaces contrasts traditional procedure-oriented P2P signaling with the newer service-based interface (SBI) paradigm. The P2P model organizes connectivity signaling (UE context, NAS, PDU session, paging, etc.) as reliable, stateful exchanges between paired RAN and CN entities, facilitating explicit ownership, deterministic behavior, and robust error handling.
Transport innovation (SCTP, QUIC, HTTP/3) is considered, leveraging improvements in stream multiplexing and security, but the architectural separation of application procedures from transport remains. The SBI model retains value for non-connectivity services (e.g., AI, sensing, service exposure), relying on service registration, discovery, and flexible producer-consumer interaction. However, for connectivity control, the operational and state ownership overhead of SBI is seen as unnecessary.
Consequently, the emerging 3GPP consensus is to retain a P2P RAN-CN interface for baseline connectivity and mobility management, with SBI potentially reserved for more dynamic or AI-centric service domains.
Open Areas and Future Directions
Several critical research and standardization questions remain as the 6G RAN matures:
- RAN-CN Interface for Non-Connectivity Services: The delineation between P2P and SBI for emerging application domains (AI, sensing) requires further refinement to avoid architectural fragmentation.
- Standardized Data Collection: A robust, common framework for data relevant to AI, sensing, telemetry, and energy optimization is necessary to enable AI-native network operation and efficient orchestration.
- AI Integration for 6G RAN: The architecture must natively support AI workflows across training, inference, and telemetry feedback while containing interface complexity and supporting both near-real-time and non-real-time operations.
Conclusion
The emerging 6G RAN architecture, as outlined, represents a disciplined and logical evolution from 5G, pairing stable standardized interfaces (CU-DU, CU-CP/CU-UP, RAN-CN P2P) with new abstractions (RU recognition) and flexible implementation boundaries. The balance between standardization for multi-vendor interoperability and freedom for deployment-specific optimization is the principal design theme. As the 3GPP process advances towards Release-21, focus must remain on avoiding architectural bloat while enabling AI-native operations, operational telemetry, and support for unprecedented service diversity. The architecture’s long-term viability will depend on disciplined boundary management and adaptable frameworks to accommodate future technology inflections (2607.10543).