Towards data-driven and machine learning based mobile network automation and planning

Abstract

Mobile networks are one of the largest and most ubiquitous deployed systems in the world with growing complexity and global demand. Mainly the mobile networks are primarily operated manually, which is a time-consuming process, and human involvement is error-prone. For instance, most investigations of service-related problems in operator networks are triggered by customers’ complaints. This causes problems like customer service disappointment and churn. In addition, the manual annual network planning is inefficient to deal with the high uncertain growth, and the manual daily network operations based on historical data escalate the OPEX. Recent technological advancements in computational capabilities and machine learning have given rise to a plethora of opportunities for data-driven solutions in various industries. In this thesis, we explore the opportunity for machine learning and data-driven mobile network automation and planning, considering four important problems. This work is enabled by real-world mobile network data, including those from a multi-national mobile network operator.

First, we consider wireless jamming problems which severely deteriorate and effectively shut down the network for users, and detection of jammer activation is a manually intensive and costly process. We present a novel scalable data-driven autonomous jammer detection framework termed JADE.

Second, we study mobile network demand characterization and forecasting from the perspective of enabling cost-effective mobile network planning. To this end, we conduct, for the first time, a demand characterization and forecasting study that is based on multi-year and nationwide region-level mobile network traffic data. Specifically, we analyze the demand characteristics of mobile and fixed wireless access (FWA) services at national and regional scales.

Lastly, using a data-driven approach, we solve the problem of hotspot (underprovisioning) problem at the cell level in operational mobile networks. We present a novel scalable data-driven hotspot cell detection framework termed CellDA (Data Driven Autonomous Under-provisioning Cell Detection) for operational mobile networks.