Poster: A Novel Adaptor Approach for O-RAN Interoperability

O-RAN, or Open-RAN, is a major revolution in mobile networks to promote an open system design for Radio Access Networks (RAN) subsystems [1]. Unlike traditional RAN with proprietary, single-vendor hardware, O-RAN separates RAN into multiple components, interconnecting them by standardized interfaces. The openness facilitates innovation and cost-effectiveness for 5G and beyond.

Since O-RAN interconnects components (e.g., radio units, baseband processing, and software apps) from various vendors, ensuring interoperability is critical for O-RAN success. Incompatibility could be caused by differences in versions, unsynchronized updates, or implementation discrepancies.

We demonstrate these deficiencies with E2, an interface that allows Near-Real-Time RAN Intelligent Controller (RIC) to interact with other O-RAN components. RIC allows the operation of AI-powered applications (called xAPPs) to optimize other RAN components. For this purpose, E2 Application Protocol (E2AP) enables general communication between RIC and RAN. E2 service models (E2SM) define RAN functions that can be invoked with E2AP. Firstly, vendors may include various versions of E2AP (e.g., v1.0, v1.01, v2.0) alongside customized service models. Using different versions of E2AP requires the handling of new message flow, modified parameters, and customized processing of service models. Besides, although backward compatibility is mentioned in the current O-RAN E2AP specification, there is no standardized means to efficiently test and patch it.

Despite extensive efforts, how to address the interoperability issues remains largely unresolved. O-RAN specifications standardize interoperability testing [2], while they are limited in scope and offer no effective solutions for identified problems in post-deployment scenarios. Another approach is to select all O-RAN components from a single vendor. However, this solution deviates from the O-RAN's initiative for a dynamic and diverse ecosystem.

To address the interoperability issue, we propose AOR in this poster, a novel, automated, and intelligent Adaptor for O-RAN interoperability. AOR will be deployed as a middle layer between O-RAN components that need to be interoperable. It automatically translates requests/responses to mask discrepancies due to versioning and/or implementation. We exemplify how AOR works for E2 interface in Fig. 1.

We propose a three-stage procedure to synthesize AOR. It first extracts a formal model from specifications using Large Language Model (LLM). The extracted model will describe the standardized behavior of an O-RAN component on initiating a new procedure, and the behavior when receiving a message in accordance to an input message under different scenarios. Second, to capture discrepancies in implementation, AOR further cross-validates the model through online testing. The novel idea is to learn the differences in terms of implementation and service models by analyzing the real-time traffic and recording any behavior that deviates from standardized model. Finally, with fine-tuned models of each component, we derive possible compatibility issues as described above. For each interoperability issue, AOR bridges them by searching for adapting paths from state machines. Such an automated approach facilitates fast-evolving O-RAN specifications, making operators able to incrementally stay abreast of the most recent specifications.

We have validated the feasibility of using an adaptor for O-RAN interoperability. Our implementation targeted E2 in srsRAN [3] and FlexRIC [4], both of which are open-source O-RAN-compliant implementations. We create two nodes on E2AP v1.0 and v1.0.1, respectively. We model the operation manually with our study on specification and test cases, and use the model to automatically capture two protocol updates: new flow addition and message field changes. The adaptor masks the new flow and changes the semantics of the updated fields. The results show the promise of a non-intrusive, automated solution for interoperability without access to source code. We are working on developing and implementing automated algorithms for synthesizing AOR.