Evaluation of inter-slice scheduling algorithms in 5G RAN through simulation and implementation

The thesis addresses the problem of resource scheduling in the 5G Radio Access Network (RAN) slicing. It focuses on allocating resources between slices with conflicting requirements: enhanced Mobile Broadband (emBB) slices, which need throughput guarantees, and Ultra-Reliable Low Latency Communications (URLLC) slices, which require both throughput and reliability guarantees. Traditional one-level MAC schedulers treat this as a complex multi-objective optimization problem, but a two-level MAC scheduler is explored to mitigate the complexity. The inter-slice scheduler allocates resources among competing slices, while the intra-slice scheduler distributes resources within each slice. A state-of-the-art review evaluates solutions for intra-slice and inter-slice scheduling. This thesis primarily investigates inter-slice scheduling algorithms, divided into data-based (e.g., heuristic and utility-function-based algorithms) and model-based algorithms that utilize machine learning techniques. A simulation phase was conducted using Py5cheSim, where basic algorithms such as Round Robin and Proportional Fair were implemented as baselines for comparison with advanced algorithms, including NVS2. The performance of the algorithms was assessed across different scenarios, focusing on isolation, efficient resource allocation, and adherence to Service Level Agreements (SLA). NVS2 was adapted to meet the design constraints of the final system and tested alongside other algorithms. In the final phase, the adapted algorithm was implemented on the OpenAirInterface framework. The two-level scheduler was compared to traditional MAC scheduling without slicing. The algorithm performed well, demonstrating excellent isolation and ensuring low latencies and zero packet loss for the URLLC slice.