Downlink IP Throughput Modeling and Prediction with Deep Neural Networks

Abstract

With the development of machine learning, deep neural networks are widely used in wireless communication systems for modeling and prediction. Neural networks have powerful data fitting capability and are suitable for complex multi-factor communication scenarios. The downlink IP throughput, defined as the payload data volume on IP level per elapsed time unit on the Uu interface, is an important performance metric for the quality of service experienced by the end user. In this paper, we propose a deep neural network-based modeling approach to predict the downlink IP throughput. Real-trace data of cellular systems, i.e., user-uploaded data including physical layer measurement, user scheduling information, user throughput and so on, are used for model training and testing. The experimental results show that our proposed model performs well for downlink IP throughput prediction.