A Deep Learning Approach to Downlink User Throughput Prediction in Cellular Networks

Abstract: A majority of the global population subscribe to mobile networks, also knownas cellular networks. Thus, optimizing mobile traffic would bring benefits tomany people. The available downlink user throughput in cellular networks issubject to heavy fluctuations, which leads to inefficient use of network capacity.The underlying network protocols address this issue by making use of adaptivecontent delivery strategies. An example of such a strategy is to maximizethe video stream resolution with respect to the available bandwidth. However,the currently dominating solutions are reactive and hence take time to adaptto bandwidth changes. In this work, a deep learning framework for downlinkuser throughput prediction is proposed. Accurate throughput predictors couldprovide information about the future downlink bandwidth to the underlyingprotocols that would let them become proactive in their decision making andadapt faster to resource changes. The models are trained with novel loss functionsthat capture the di↵erent costs of overestimation and underestimation.They are based on feedfordward and long short term memory networks andachieve up to 79.4% accuracy.