color:#002060;mso-ansi-language:EN-US">5G and future Radio Access Networks (RAN) are evolving toward greater openness and programmability, driven by the principles of Open RAN and initiatives such as Telecom Infra Project (TIP) and O-RAN Alliance. This transformation is enabled by two key factors: (1) open interfaces that ensure vendor-agnostic interoperability, and (2) Software-Defined RAN (SD-RAN) principles, which introduce dynamic and flexible network control. These open interfaces are standardized not only by 3rd Generation Partnership Project (3GPP) but also by the O-RAN Alliance, which plays a crucial role in integrating intelligence into the SD-RAN ecosystem.
In the O-RAN architecture, the SD-RAN controller is defined as the RAN Intelligent Controller (RIC), which enables third parties to develop and deploy SD-RAN applications, known as xApps (for near-real-time control) and rApps (for non-real-time control). Beyond enhancing RAN programmability across different time domains, O-RAN also allows these applications to leverage techniques such as machine learning, enabling more efficient, intelligent, and adaptive network optimization. While O-RAN unlocks broader capabilities, it also presents new challenges. The growing demands of emerging use cases (e.g., real-time gaming, haptic communication) and advanced features (e.g., network slicing) increase complexity, making x/rApps harder to develop and manage. Additionally, due to less standardized interfaces, x/rApps often become tied to specific RIC platforms or vendors, limiting their portability and reuse. Moreover, integrating O-RAN technologies into existing 5G networks further complicates the optimization process, adding burdens for operators.
In this work, we investigate how to enhance the flexibility and programmability of network customization for both the control and user planes of SD-RAN while addressing the complexity introduced by O-RAN technologies, with the goal of simplifying network infrastructure and optimization processes to pave the way for next-generation mobile networks. First, we design a Flexible Control plane (FlexCtrl) that allows control logic to be distributed across three levels — SD-RAN applications, controller, and RAN nodes — based on use case requirements and control loop latency constraints. This design includes an open interface and a virtualization layer for xApps to enable real-time control and simplify development. Second, we propose Integrated and programmable User Plane (IUP), a novel RAN system design with a concrete realization, integrating User Plane Function (UPF) functionalities into a RAN node to enable real-time coordination of traffic management and radio resource allocation. Third, leveraging the synergy of the proposed FlexCtrl and IUP, we introduce AUTO-RAN, an innovative concept that enables autonomous programmability through a robust SD-RAN application design, significantly reducing operational complexity for network operators in RAN optimization while ensuring seamless coordination across both 3GPP and O-RAN ecosystems.
We evaluate the proposed solutions using open-source platforms (OpenAirInterface and FlexRIC). Results demonstrate enhanced flexibility and extended programmability across both the control and user planes. In the control plane, FlexCtrl evolves control logic within SD-RAN applications, simplifying development through recursive abstraction and supporting flexible logic placement to meet use case needs. For example, a distributed control plane achieves control loop latency below 50 μs, suitable for real-time radio scheduling. In the user plane, IUP consolidates traffic management within the RAN, enabling unified control over IP flows and radio resources. This reduces control latency and cuts data delivery overhead by up to 50%. Building on these capabilities, AUTO-RAN integrates autonomous mechanisms into SD-RAN applications, significantly reducing development complexity of x/rApps by up to 90% in line of codes. Results also show that AUTO-RAN achieves real-time adaptability without requiring explicit operator intervention in RAN slicing use cases, and supports intent-driven optimization by allowing operators to declaratively express high-level intents in mobility management use cases.
This thesis simplifies network operations by reducing the development complexity of control logic within RAN functions and SD-RAN applications, streamlining the deployment of network functions in the user plane, and abstracting the intricacies of O-RAN technologies, thereby enabling real-time, unified, and autonomous optimization. Overall, the proposed methods and architectures provide the foundational infrastructure to enhance flexibility, programmability, and customizability across both planes, accelerating the transition toward intelligent next-generation mobile networks.
