Towards a 3GPP Network-based Framework for Improving Service Assurance and Load Balancing.

As cellular networks evolve towards the 6th generation, new schemes are proposed in the area of Quality of Service (QoS) assurance. In recent years, predicting QoS gained some momentum as a way of satisfying specific connectivity requirements, supporting service assurance, and estimating the Quality of Experience (QoE). The QoS requirements to guarantee a certain QoE differ per use case, and hence depend on a multitude of factors, e.g., selecting an appropriate cell that can guarantee specific QoS requirements. Machine Learning (ML) is proposed as a method to improve network capabilities for QoE assurance by the use of predictive Quality of Service (pQoS). This in return can improve the offered QoS, reduce latency by selecting the most appropriate cell quickly, and improve the load-balancing at the network. The adoption of ML depends heavily on removing some of the roadblocks of applying ML in commercial networks. For example, ML-based algorithms are known to depend on a large amount of data, which might increase the usage of signaling and the battery consumption at the User Equipment (UE). We present an ML framework that can enable many of the aforementioned network capabilities, which does not require the introduction of new signaling types or proprietary data collection procedures. We showcase the benefits of the ML framework on an inter-frequency load balancing use case and discuss how ML can improve UE and network performance. Finally, we highlight the need to introduce the expected interference to the UE as an input feature for further improving QoS prediction performance. We test the performance of the prediction framework on data coming from a test network and evaluate the effects of e.g., different prediction thresholds.