Enhanced Decision Mechanism for RAN Subslicing in Management Closed Control Loop

5G radio access network (RAN) slicing enables better satisfaction of different quality of service (QoS) requirements than without network slicing. A handful of slice types are specified for various service verticals; however, the number and size of those slices is unspecified. Subslicing refers to grouping slice users into smaller groups, subpartitioning slice bandwidth, and allocating smaller bandwidth parts to smaller user groups. State of the art subslicing has been done to better satisfy the QoS requirements inside the service vertical. Slice performance improvement was not the purpose of subslicing, but the positive effect was noticeable. In this paper, the subslicing decision is done with the aim of improving slice performance. The decision mechanism for management closed control loop is proposed. The input dataset consists of 6 key performance indicators, namely slice bandwidth utilization, slice goodput (application level throughput) per one allocated resource block (RB), and slice block error ratio (BLER), both in uplink and downlink. This dataset is clustered, and the result is learned by a classifier to decide whether the slice is too large and should be split or too small and should be merged with another too small slice or subslice. The results show that by knowing the slice utilization and goodput per one allocated RB, the slice reconfiguration action regarding subslicing can be determined using machine learning tools.