O-RAN architecture, interfaces, and standardization: Study and application to user intelligent admission control

Open radio access network (O-RAN), driven by O-RAN Alliance is based on the disaggregation of the traditional RAN systems into radio unit (RU), distributed unit (DU) and central unit (CU) components. It provides a unique opportunity to reduce the cost of wireless network deployment by using open-source software, serving as a foundation for O-RAN compliant functions, and by utilizing low-cost, generic white-box hardware for radio components. Relying on the two core pillars of openness and intelligence, there has been a coordinated global effort from operators and equipment providers to enhance the RAN architecture and improve its performance through virtualized network elements and open interfaces that incorporate intelligence in RAN. With the increased complexity of 5G networks and the demand to fulfill requirements, intelligence is becoming a key factor for automated deployment, operation, and optimization of open wireless networks. The first thrust of this paper surveys the AI/ML architecture in O-RAN specifications, key discussion points and future standardization directions, respectively. In the second part, we introduce a proof-of-concept use case on AI-driven network optimization within the near real-time RAN intelligent controller (near-RT RIC) and non-real time RIC (non-RT RIC). In particular, we investigate the user admission control problem, led by a deep learning-based algorithm, implemented as an xApp for network performance enhancement. Extensive system-level simulations are performed with NS-3 LTE to assess the proposed admission control algorithm. Accordingly, the proposed dynamic algorithm shows a significant admission control performance improvement and flexibility, compared to existing admission control static techniques, while satisfying the stringent quality of service targets of admitted devices. Finally, the paper offers insightful conclusions and findings on the AI-based modeling, model inference performance, key performance challenges and future research directions, respectively.