An optimized Paging prediction model over a cell in wireless network for efficient resource planning

In a cellular network, paging is used to inform and notify user equipment (UE) about various events. Its purpose is to transmit paging information to a UE in Radio Resource Controller (RRC) IDLE or CONNECTED state using limited set of paging resources available at Radio Access Network (RAN). UE decodes the paging-content (paging-cause) of the paging message and UE has to initiate the appropriate procedure like Call/ Data/ SMS/ Emergency alerts, etc. With the growing number of devices and network infrastructures, optimizing paging resources is the most important thing. New Radio (NR) known as 5G introduce small cells and requirement to support enhanced Mobile Broadband (eMBB), Ultra Reliable Low Latency Communication (URLLC) and massive Machine Type Communication (mMTC) category type devices. Internet of things (IoT) will introduce huge number of devices may have very less data rate. Hence, there will be large amount of idle mode devices under one cell due to IoT and 5G use cases. To accommodate this huge number of idle mode devices there is a need of paging resource optimization at RAN level. When the network decides to communicate with UEs, the network may page the UEs' in idle mode in all the cells belonging to their registration area (RA). The paging is handled at RA level and hence each cell under a RA gets the same paging with similar user information. In 5G-NR max number of UEs that can be paged at a time are limited to 32. This is a bottleneck to page for max number of unique users at each time and this further adds page delay. If cell-level paging is implemented, the bottleneck can be addressed. The challenge of cell-level paging is in acquiring cell-level location information. However, with past data of location and along with last connection information, using different Machine Learning (ML) algorithm the cell-level location information can be predicted. In this paper, a comparative study has been done with various ML algorithms and simulation results depict the accuracy of each algorithm based on trained data. Simulated Results with the actual user data show 4–5 times better performance than that of conventional paging. This data will be helpful for operators to optimize their network for effective resource utilization.