Introduction
User equipment (UE) in 5G mobile networks disconnects from Next Generation Node B (gNB) in the absence of user traffic for certain period and transitions from active. Network does not track UE, and for delivering a service, Access and Mobility management Function (AMF) locates current gNB of UE through transmitting messages to collection of gNBs in a multistep paging process. Thus paging is a most prevalent control function in mobile networks, and NSA MME data from a Korean commercial 5G operator SK Telecom confirms this as it processes average 1.8 billion (21.7 %) paging messages per day out of total 8.4 billion control messages. Increasing gNBs via ultradense cell deployments in 5G and beyond [1] are further escalating paging, and it is a serious challenge for operators to increase their revenue by curtailing paging cost. Moreover, current paging operates sequentially where it moves to next step when previous fails. The first step sends a message to last known UE gNB in active mode and gNBs are increased to operator dependent Tracking Area (TA) and TA List (TAL) in the second and third steps, respectively. Resulting in exponential and linear increases in signaling traffic and paging delay
Paging delay is the time taken by the network to locate specific UE in idle mode to deliver it a particular service such as voice call, text message, or notification.
up to 1s, respectively. In particular, the paging delay is a bottleneck for ultra-low latency services like autonomous vehicles and virtual reality with 50ms and 10ms delay budget requirements, respectively [2]. In short current paging is out-dated for 5G and beyond, and it is vital to increase the paging success rate in first and second paging steps to support ultra-low latency services and increase operators' revenue. This is achievable by sending paging messages to Deep Learning (DL) driven small predicted list of gNBs with maximum likelihood of UE presence after given elapsed time instead of all gNBs in TA and TAL.