Abstract

Paging is a critical operation in 5G mobile networks for localization of User Equipment (UE) in idle mode. Current multistep paging method increases the signaling exponentially and delay linearly, and analysis of a commercial Non Stand Alone Mobility Management Entity (NSA MME) reports that paging accounts for 21.7% of NSA MME total traffic. To reduce signaling overhead and service delay by increasing the paging success rate, this paper presents a Deep Learning (DL) driven intelligent Paging as a Service (iPaaS) over a distributed NWDAF architecture based on serverless computing. iPaaS clusters UEs based on their movement patterns and feed cluster ID, movement history, and elapsed time in idle mode of a UE into DL model for UE localization prediction. Prediction results are used in two proposed schemes: iPaaS for Reduced Signaling (iPRS) and iPaaS for Reduced Delay (iPRD), which focus on cutting down signaling overhead and delay, respectively. Performance of iPaaS with two proposed schemes is evaluated on 110 million handover logs that are collected over 15 days from a Korean commercial 5G operator SK Telecom. Results show that iPRS reduces the signaling overhead up to 94% and iPRD reduces the delay up to 43% comparing to current method.