High-Speed Users’ Mobility Prediction Scheme Based on Deep Learning for Small Cell and Femtocell Networks

Users’ mobility has a huge impact on the performance of cellular networks. Acknowledge users’ multiple next locations plays an important role in various aspects which can be mentioned as helping the base stations to pre-calculate and allocate the resource to users faster and more efficiently, shortening the duration of the handover process, reducing significantly the network data congestion, and increasing the overall users’ satisfaction. In our article, we focus our attention on multiple users and multi-position ahead prediction for femtocells and small cells, typical of 5G infrastructure. We use Autoregressive Gated Recurrent Units (AR-GRU) to perform the prediction based on acknowledging users’ trajectories. We use Simulation of Urban MObility (SUMO) to create our own users’ trajectory datasets to train and test the models. In order to prove the effectiveness of the model, we compare its performance with Autoregressive Long Short-Term Memory (AR-LSTM), Deep Learning Neural Network (DNN), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU) models. Then we use the models in two more different datasets from two different simulated regions to prove the ability to work in different contexts.