Data-Driven Proactive Uplink Slicing Enabling Real-Time Control within an Open RAN Testbed

In terms of flexibility and scalability, network slicing enables the realization of contradicting services on public mobile radio infrastructure - providing robustness and reliability for critical services. Combined with the new paradigm of Open RAN, network operators can utilize Commercial of-the-Shelf (COTS) hardware to build a fully softwarized radio network with sophisticated resource management. Using xApps built on top of a near Real-Time RAN Intelligent Controller (RIC) operating as a virtual entry point, the allocation of resources can be optimized to provide low latency and reliability to critical services while yielding high spectral efficiency. In this paper, we showcase the capability of our developed SAMUS xApp relying on Machine Learning (ML) methods in a real-world experimental setup assessed by an inverted pendulum operated with Model Predictive Control (MPC). Compared to traditional scheduling, we reduce the mean uplink one-way delay by 65% using our SAMUS xApp, while simultaneously increasing spectral efficiency by 61 % compared to static proactive allocation, maintaining similar control performance.