Towards Open 6G: Experimental O-RAN Framework for Predictive Uplink Slicing

Traditional cellular Radio Access Networks (RANs) are associated with significant costs and low agility due to proprietary hard- and software resulting in vendor lock-ins. Open RAN promises to change this by harnessing open source and Commercial of-the-Shelf (COTS) solutions. Hence, the Open Radio Access Network (O-RAN) Alliance, a consortium of partners from industry and research, aims to identify and close gaps in 3rd Generation Partnership Project (3GPP) specifications. It defines a software-centric RAN architecture with open interfaces to increase interoperability, strengthen innovation and lower market entry barriers towards future 6G infrastructures. A core concept in this context is the near-Real-Time RAN Intelligent Controller (RIC), a virtual platform for hosting so-called xApps. These software-based network functions provide functionalities such as monitoring or network slicing. Using proactive resource management, slicing has the potential of enabling concurrent service profiles such as Ultra-Reliable Low Latency Communication (URLLC) and Enhanced Mobile Broadband (eMBB), while rising spectral efficiency and lowering latency. Thus, this work introduces an O-RAN-based framework for predictive uplink slicing. An xApp is presented, harnessing deep learning to dynamically reconfigure RAN scheduling via the RIC's E2 interface. The evaluation is performed on the challenging example of URLLC traffic from the Smart Grid domain via an experimental laboratory setup. O-RAN introduces additional interfaces, yet the framework performs about on par with proprietary solutions with latencies down to 5 ms.