QoS Prediction-based Radio Resource Management

Accurate predictions of the expected change of Quality of Service (QoS) and radio key performance indicators (KPIs) in the radio access network are being enabled by machine learning (ML). These future analytics can be used to support proactive adaptation of end-user applications and in-advance optimization of the radio access network. Predicted QoS information can enable radio resource management schemes provide more reliable future QoS guarantees for individual users even with poor expected performance. In this paper, we introduce a QoS prediction-based radio resource management scheme based on the predictive proportional fairness resource allocation algorithm using expected spectral efficiency information together with QoS prediction information to prioritize individual users well in-advance. Simulation results show that the usage of QoS predictions in radio resource management can lead to future user capacity guarantees for individual users while maximizing fairness and system performance on the prediction horizon. The proposed solution achieves significant system gains in sum-rate and fairness compared to reactive resource allocation in spite of significant levels of uncertainty in the predicted information.