Next Point-of-Attachment Selection Based on Long Short Term Memory Model in Wireless Networks

Existing mobility management systems in cellular networks are ill-equipped to support Ultra-Reliable and Low Latency Communication (URLLC) requirement of next generation prevalent and real time services in dense network deployments due to their reactive approach. Proactive approach is one way to meet the URLLC requirement of these services, where resource assignment and control signaling is completed before the actual user mobility. Successful execution of proactive mobility requires accurate prediction of user next Point of Attachment (PoA) and precise estimation of user mobility instant. This paper adopts supervised deep learning approach to predict the next PoA of user. In particular, a Long Short-Term Memory (LSTM) model is developed for this purpose, which exploits the temporal characteristics of the data. We discuss different design choices of our LSTM model and show their effects on the prediction accuracy. Evaluation results reveal that proper data preprocessing and time-step increment significantly affects the prediction accuracy. The highest accuracy achieved by our model is 91% with shuffled data and stacked LSTM.