AI-Assisted RLF Avoidance for Smart EN-DC Activation

Abstract

In the first phase of 5G network deployment, User Equipment (UE) will camp traditionally on LTE network. Later on, if the UE requests a 5G service, it will be made to camp simultaneously on LTE and 5G. This dual-camping is enabled through a 3GPP-standardized approach known as E-UTRAN New-Radio Dual-Connectivity (EN-DC). Unlike single-networkcamping, where poor RF conditions of only one network affect user Quality-of-Experience (QoE), in EN-DC, poor RF condition in either LTE or 5G network can be detrimental to user QoE. Sub-optimal parameter configuration to activate EN-DC can hamper retainability KPI as UE may observe increased radio link failure (RLF). While the need to maximize the EN-DC activation is obvious for 5G network maximum utility, RLF avoidance is equally important to maintain the QoE requirements. We address this problem by first using Tomek Link to counter data imbalance problem and then building an AI model to predict RLF from real network low level measurements. We then propose and evaluate an RLF risk-aware EN-DC activation scheme that draws on insights from the developed RLF prediction model. Simulation using a 3GPP-compliant 5G simulator show that compared to no-conditioning on EN-DC activation, in the evaluated cell cluster, the proposed scheme can help reduce the potential RLF instances by 99%. This RLF reduction happens at the cost of 50% reduction in EN-DC activation. This is first study to present a framework and insights for operators to optimally conFigure the EN-DC activation parameters to achieve desired trade-off between maximizing 5G sites utility and QoE.