A Stochastic Network Calculus (SNC)-Based Model for Planning B5G uRLLC RAN Slices

Radio Access Network (RAN) slicing involves several challenges. In particular, the Mobile Network Operator (MNO) must ensure-before deploying each slice-that corresponding requirements can be met throughout its lifetime. For ultra-Reliable Low Latency Communication (uRLLC) slices, the MNO must guarantee the packet transmission delay within a delay budget with a certain probability. Most existing solutions focus on allocating dynamically radio resources to maximize the number of packets, whose transmission delay is within the delay budget. However, these solutions do not ensure the violation probability is below a target value in the long term. In this paper, we focus on slicing from a planning perspective. Specifically, we propose a Stochastic Network Calculus (SNC)-based model, which given the amount of radio resources allocated for a uRLLC slice, the target violation probability and the traffic demand distribution, provides the delay bound for such conditions. Additionally, we propose heuristics for planning uRLLC slices. Interestingly, such heuristics benefit from the proposed SNC-based model to compute the amount of radio resources to be assigned to each slice while its delay bound, given a target violation probability, is within the delay budget. We validate the SNC-based model and demonstrate the effectiveness of the heuristics.

Automated summary

Radio Access Network (RAN) slicing faces multiple challenges, especially the need for Mobile Network Operators (MNOs) to ensure that the requirements of each slice can be met throughout its lifetime. This paper proposes a Stochastic Network Calculus (SNC)-based model for slicing, providing delay bounds based on the allocated radio resources, target violation probability, and traffic demand distribution. Additionally, heuristics are proposed for planning ultra-Reliable Low Latency Communication (uRLLC) slices to optimize resource assignment and delay bounds. The effectiveness of the proposed model and heuristics are demonstrated.