Deep Neural Networks approach for Power Head-Room Predictions in 5G Networks and Beyond

The enhanced Interference Mitigation and Traffic Adaptation (eIMTA) mechanism is a key enabler for 5G networks and beyond. Knowing that a User Equipment (UE) cannot transmit more than the maximum power allowed by its power class. The level of available Transmission Power (TP) in each UE is an essential input for the Uplink (UL) scheduler of the next Generation Node-B (gNB). Scheduling higher data rate than what is supported by the available TP is a waste of resources. In the Downlink (DL), the power level is known by the gNB that manages the power amplifier and the DL-scheduler. Conversely, in the UL, the available power is estimated by the UE and sent to the gNB as a key input for eIMTA, known as Power Head-Room (PHR). In this context, we propose in this paper, a Deep Neural Network (DNN) based model to predict the PHR and reduce dependency on reported measures. We evaluate the effectiveness of our proposal in a 5G experimental prototype, based on Open Air Interface (OAI). Obtained results show that using DNN outperforms state-of-the-art machine-learning approaches in terms of prediction accuracy, computation complexity and throughput.