Integration of Machine Learning Algorithms into the Evolved Packet Core Network

Abstract

In the world of telecommunications today, there is an increasing amount of data being transmitted to and from an increasing number of mobile phones. The systems and networks handling this data are therefore seeing the need of improvement in order to keep up with this evolution. Quickly becoming the norm for handling Big Data across several industries, machine learning is gaining influence and becoming integrated into the working practices of several world leading companies. In this industrial thesis work we examine the Evolved Packet Core (EPC) network at Ericsson. This complex, distributed and real-time system handles the throughput of 4G as well as WiFi data to and from mobile phones. The purpose is to study the network with the idea of improving it using various machine learning techniques. The thesis is divided into two parts, literature review and ideation. The literature review aims to reveal what has already been done in the telecommunications industry, with respect to machine learning. The ideation phase incorporates a full study of the EPC network, which includes reading technical documentation, interviewing experts on various subsystems as well as hosting workshops at the different departments of the company. Additionally, ideas are innovated with the newfound knowledge of the EPC network. The goal of each idea is to improve the EPC network by either adding new features or improving existing workflows. The innovated ideas are categorized into areas such as subscriber profiling, performance monitoring and test analysis amongst others. The EPC network shows promising and untapped potential in terms of evolution incorporating machine learning. Finally, we also provide an outlook on the future of the telecommunications industry. The two main topics discussed in recent years in the industry are 5G and Internet of Things (IoT), to which there is potential for integrating machine learning as well.