The Cost of Uncertainty: Impact of Overprovisioning on the Dimensioning of Machine Learning-based Network Slicing

Increasing automation of industry verticals and frequently changing production cycles require a high level of production line modularity and are locally accompanied by frequently changing disjunctive application requirements. Thus, current and future wireless communication networks need to face the challenge of providing opportunities to rapidly adapt the network to its changing application demands in order to guarantee a resilient and interference-free communication. A possible key technology for implementing such a solution is represented by private 5G networks that are additionally equipped with network slicing in order to be able to meet the versatile requirements of novel applications. However, resilient network design as well as network slice dimensioning can only be guaranteed through detailed network planning. This requires expert knowledge, which is not yet present at most companies or institutions. Accordingly, automation of the network planning process is a possible solution. Existing coverage planning frameworks are extended by capacity planning in this work, and network slicing is introduced. It is shown on the basis of a realistic scenario that the predictability of data (e.g., traffic characteristics in low-latency slices) significantly influences capacity planning and must be taken into account in the dimensioning of 5G and beyond future mobile networks.