Key points are not available for this paper at this time.
O-RAN Alliance suggests new interfaces and architectures that can bring openness and intelligence to currently vendor-dependent Radio Access Network (RAN) environments. Newly defined O-RAN specifications guarantee the interoperability of Network Entities from different vendors to enable multivendor environments, and O-RAN Working Group 2 (WG2) focuses on AI/ML workflow architectures and specifications that enable applications of artificial intelligence (AI) and machine learning (ML) in O-RAN environments. Additionally, O-RAN Software Community (SC) realizes what O-RAN Alliance defines by developing open-source software. In this paper, we introduce an AI/ML workflow that is based on WG2 AI/ML specifications and implemented with open-source software from O-RAN SC, Acumos and Open Network Automation Platform (ONAP). Acumos Framework was used to generate and package ML models to be deployed and executed in O-RAN RAN Intelligence Controller (RIC) and components of ONAP provided monitoring and arbitration needed to operate the workflow. Along with the example of operating AI/ML workflows on O-RAN environment, we discuss shortcomings we encountered during this process and potential solutions. Finally, we suggest capabilities for RIC platform to facilitate other useful ML technologies such as reinforcement learning and federated learning as well as GPU support.
Lee et al. (2020) studied this question.