ColO-RAN: Developing Machine Learning-based xApps for Open RAN Closed-loop Control on Programmable Experimental Platforms

Cellular networks are undergoing a radical transformation toward disaggregated, fully virtualized, and programmable architectures with increasingly heterogeneous devices applications. In this context, the open architecture standardized by O-RAN Alliance enables algorithmic hardware-independent Radio Access Network (RAN) adaptation through closed-loop control. introduces Machine Learning (ML)-based network control automation algorithms as socalled xApps running on RAN Intelligent Controllers. However, in spite of new opportunities brought about Open RAN, advances ML-based have been slow, mainly because unavailability large-scale datasets experimental testing infrastructure. This slows down development widespread adoption Deep Reinforcement (DRL) agents real networks, delaying progress intelligent autonomous paper, we address these challenges discussing insights practical solutions for design, training, testing, evaluation DRLbased RAN. To end, introduce ColO-RAN, first publicly-available framework software-defined radios-in-the-loop. Building scale computational capabilities Colosseum wireless emulator, ColO-RAN ML research at using components, base stations, ”wireless data factory”. Specifically, design develop three exemplary DRL-based slicing, scheduling online model evaluate their performance cellular 7 softwarized stations 42 users. Finally, showcase portability to different platforms deploying it Arena, an indoor testbed. The lessons learned from implementation extensive results our first-of-its-kind highlight importance frameworks end-toend pipelines, analysis DRL agents. They also provide benefits adaptive control, trade-offs associated training live collected dataset publicly available community.