Closing of 3G Sites: Model for Decision Making

INOM

EXAMENSARBETE ELEKTROTEKNIK, AVANCERAD NIVÅ, 30 HP

STOCKHOLM SVERIGE 2018 ,

Closing of 3G Sites

Model for Decision Making EMMANUEL CHAUDRON

KTH

SKOLAN FÖR ELEKTROTEKNIK OCH DATAVETENSKAP

K T H R O Y AL I N S T I T U T E O F T E C H N O L O G Y

E l e c t r i c a l E n g i n e e r i n g a n d C o m p u t e r S c i e n c e

i

Abstract

Radio access technologies evolving fast, mobile operators have to handle an increasing amount of base stations and frequency bands for their network to continue to function. This is a costly venture for mobile network operators that continuously have to keep up to date with never-ending advancements in technologies, as base stations are costly to build and to maintain. It is therefore necessary for these companies to investigate when to close down base stations that are not necessary anymore. With the upcoming release of 5G, it is expected that 3G is going to be less and less used—as of 2018, it is already less used than 4G in developed countries.

This thesis analyses the corporate data of a mobile operator, Telenor Sweden, in order to make clear which metrics are important to take into account as regards to deciding whether or not to close down a base station. It provides methods and models to help a mobile operator to take such a decision. It focuses on UMTS (3G) base stations, even though the results can be generalized for other technologies as well.

It evaluates the economic feasibility of closing a base station, with regards to how many users are still connecting to it. More importantly, it explains for what reasons users’ devices switch to 3G, and investigates what can be done to avoid switching from 4G to 3G, so as to make it easier to close down a 3G site.

It provides eventually a model to help to know when closing a site, given the traffic data of the operator.

Keywords

Mobile telephony, Base station (BS), 3G sunset, UMTS, Business data

processing, Economic feasibility analysis, Techno-economic study

ii

iii

Abstract

Radioåtkomstteknologier utvecklas snabbt, mobiloperatörer måste hantera en ökande mängd basstationer och frekvensband för att deras nätverk fortsätter att fungera. Detta är ett dyrt satsning för mobilnätoperatörer som kontinuerligt måste hålla sig uppdaterade med oändliga tekniska framsteg, eftersom basstationerna är kostsamma att bygga och underhålla. Det är därför nödvändigt för dessa företag att undersöka när man ska stänga basstationer som inte längre är nödvändiga. Med den kommande utgåvan av 5G förväntas 3G att bli mindre och mindre används. Från och med 2018 används den redan

mindre än 4G i industriländer.

Denna avhandling analyserar företagsdata från en mobiloperatör, Telenor, för att klargöra vilka mätvärden som är viktiga att ta hänsyn till när det gäller att avgöra om en basstation ska stängas eller inte. Det ger metoder och modeller för att hjälpa en mobiloperatör att fatta ett sådant beslut. Den fokuserar på UMTS (3G) basstationer, även om resultaten kan generaliseras för annan

teknik också.

Det utvärderar den ekonomiska möjligheten att stänga en basstation, med tanke på hur många användare som fortfarande ansluter till den. Viktigare är det att det förklaras av vilka anledningar användarens enheter växlar till 3G och undersöker vad som kan göras för att undvika att växla från 4G till 3G, så att det blir lättare att stänga en 3G-basstation.

Det ger så småningom en modell som hjälper till att veta när man stänger en webbplats, med tanke på operatörens trafikdata.

Nyckelord

Mobiltelefoni, Basstation, 3G sunset, UMTS, Företagsdatabehandling,

Ekonomisk genomförbarhetsanalys, Tekno-ekonomisk studie

iv

v Acknowledgments

This project, pretty innovative if we look for similar research work done in the past, could not have been realized without the following people.

First, I would like to thank Telenor Sweden for having offered me a work environment in which such a project could be achieved. Not only did I have a place to work comfortably, in an ethical environment, but also I was always able to grab a coffee (“fika”) with colleagues and ask questions to be able to progress and iterate faster in my thesis work.

In that respect, I would like to say a big thank you to my Telenor supervisor, Lars Adolfsson, for having been a great support during all the thesis work. His introduction to the rest of the Telenor team was really helpful to get to know the different stakeholders—experts in their domains—who could provide their help and insights on this work.

I would also like to thank my manager, Amin, for his sympathy and modesty in all circumstances. He was there every time I needed his help.

Without the kind help and follow-up of my KTH examiner, Slimane Ben Slimane, the thesis work could have been carried out in the perfect timeframe.

He offered me great solutions to academic-related questions I had.

Last but not least, I would like to thank my academic supervisor Cicek Cavdar, whose feedback on thesis proposal was really helpful for the rest of the thesis.

Although the academic supervisor I was first in touch with could not take on my project due to long-term sickness, she still accepted to support me for my thesis, two months after it started.

Thanks again to all these people and to those I do not mention here—

including many Telenor employees I had pleasure to work with—, without whom I could not reach this point.

Stockholm, August 2018

Emmanuel Chaudron

vi

vii List of acronyms and abbreviations

2G, 3G, 4G, 5G second, third, fourth, fifth generation ARPU Average revenue per user

BSC Base Station Controller BTS Base Transceiver Station

CS Circuit-switched

CSFB Circuit-Switched Fallback

ETSI European Telecommunications Standards Institute GSM Global System for Mobile Communications

IMSI International Mobile Subscriber Identity

IP Internet Protocol

IoT Internet of Things

LTE Long Term Evolution

M2M Machine-to-machine

MHz Megahertz

PS Packet-switched

RAB Radio Access Bearer

RAN Radio Access Network

RAT Radio Access Technology RNC Radio Network Controller

SQL Standard Query Language

UE User equipment

UMTS Universal Mobile Terrestrial System

VoLTE Voice over LTE

viii

ix

Table of contents

1 Introduction ... 1

1.1 Background ... 1

1.2 Problem ... 1

1.3 Purpose ... 2

1.4 Goal ... 2

1.4.1 Benefits, Ethics and Sustainability ... 2

1.5 Research methodology... 3

1.6 Delimitations ... 4

1.7 Outline ... 4

2 Technical Background ... 5

2.1 Challenges in the telecommunications world ... 5

2.2 Cellular networks ... 5

2.2.1 Frequency bands and ranges ... 5

2.2.2 Network architectures of today’s widespread mobile technologies ... 6

2.2.3 Spectrum license ... 9

2.3 Closure of a mobile network... 9

2.3.1 A few examples of network shutdowns—and their consequences ... 9

2.3.2 Cost savings when closing a 3G network ... 9

2.4 Big data... 10

3 Research methodology ... 12

3.1 Research process ... 12

3.2 Origin of the data ... 12

3.2.1 Protection of sensitive data ...13

3.3 Data used for this study ... 13

3.3.1 Location data ...13

3.3.2 Inter-RAT data ...13

3.3.3 Cell data ...14

3.4 Profitability threshold model ... 14

4 Data mining ... 17

4.1 Data analysis tools ... 17

4.2 Data processing ... 17

5 Results ... 19

5.1 Candidate sites for a closure ... 19

5.2 Causes distribution for switching from 4G to 3G ... 21

6 Conclusions and future work... 24

6.1 Conclusions ... 24

6.2 Future work ... 24

References ... 25

x

1

1 Introduction

The mobile cellular technologies are regularly evolving to enable improved services and data rates to mobile user, whose bandwidth consumption increases each year—with a 32% increase worldwide between 2015 and 2016 [1]. Besides, the average revenue per user has decreased significantly in the past years, falling from 1% of the global national income (GNI) per capita in 2013 to 0.7% in 2016 [1]. In such a context, mobile operators have to cut down their costs. One of the cost drivers is the ownership and maintenance of mobile networks. An opportunity to look at is to shut down legacy networks, that are still used, but only by few users. Such networks are today replaced by networks based on latest cellular technologies, and may be considered for a shutdown. However it requires a lot of previous work, including analytical work, to be achieved.

This chapter introduces the reader to the thesis project, and provides them with deeper insights of its goal, aims, and research approach.

1.1 Background

Since the inception of digital mobile networks with the second-generation cellular technology, mobile network operators keep deploying and maintaining networks that use different technologies: 2G–deployed in the early 1990’s–, 3G–deployed in the early 2000’s–, 4G–deployed in the early 2010’s–, and very soon 5G [2]–[4]. At some point in time, older technologies get underused, and it is time for mobile operators to think dismantling the equipment supporting that technology and closing down the network. This thesis focuses on dismantling a network gradually, by removing first the stations that serve a very limited amount of users.

The upcoming release of 5G and the ongoing developments of 4G networks make 3G networks more and more subject to be shut down. Whenever considering closing a network, a mobile operator has to make a painstaking investigation of the implications of such a closure. Many stakeholders have to be involved in the process. With an overview of the consequences of closing down a network, operators may take an informed decision, and will not have to start a study from scratch.

1.2 Problem

Mobile technologies are constantly evolving. Every time a new technology appears, operators need to update their cellular network infrastructure to stay up-to-date to the latest improvements in the field. By upgrading their networks, operators can offer better services to their customers, and thus either charge more for these services, or alternatively attract more users or at least avoid losing users.

Meanwhile, they still own and maintain their previous network infrastructure

as they will need to provide services to both users using the newest technology

and those not having switched to that new technology yet—e.g. because of

2 incompatible device. At some point, however, networks implementing legacy technologies will become underused and unprofitable. Moreover, older cellular networks use frequency bands that could be often reused by newer technologies to increase their capacity. Spectrum license are costly to acquire, and operators aim to exploit at their maximum potential the frequency bands for which they have rights to operate.

Given the problem stated, the main question is formulated as follows:

How may mobile operators leverage their traffic data to decide when closing down a cellular site is economically viable?

To answer this question, it is necessary to investigate the challenges of closing down a network, and spending that can be avoided. It is worth also looking at today’s telecom operators that have closed down their 2G or 3G network, to understand in which context they did it.

1.3 Purpose

The thesis illustrates the analytical work made at a telecom operator to decide when it is appropriate to shut down a 3G site in their network.

This thesis was proposed by Telenor Sweden, which owns its own 3G network in Sweden, along with a shared 3G network. However, the content of this paper was made as general as possible, considering that its conclusions and analysis should apply to the closing of a network based on any cellular technology.

1.4 Goal

The goal of this thesis is to provide hands-on models and information to mobile network operators to take faster decisions regarding their 3G network shutdown. Specifically, it aims at (a) describing the necessary knowledge prior to undertaking the closure of a site, and (b) proposing a model to help to decide when removing a site.

1.4.1 Benefits, Ethics and Sustainability

Such a work will benefit to mobile telecommunication companies considering closing down a legacy network, as no previous public research on that topic was published before this one, to the author knowledge.

Closing down sites that are underused enables telecom operators to achieve sustainable cost while focusing their effort on newer technologies, and on their future goals.

When performing processing of traffic data from a cellular network, ethical issues of such a venture must be considered. Indeed, this raises the issue of data protection. Will data be sufficiently anonymized to ensure user privacy?

The goal here is to anonymize the data so that individual users may not be

identified when analyzing the data. Such a goal is typically reached by hashing

the user International Mobile Subscriber Identity (IMSI)—which uniquely

identifies a telecom customer. However, this might not be enough, as a user

who would be physically stalked could still be found using the traffic data.

3 Therefore, additional precautions are usually taken, e.g. by changing the user hashed id every now and then, so its behavior could not be followed over a long period. The section 3.2.1 of this thesis describes in more detail the implications of business data processing from an ethical perspective.

1.5 Research methodology

A research necessarily implies a methodology to be followed, an approach to be chosen, and methods to be applied.

The research methodology is the category of the research. There are two main categories of research methodology: quantitative methodology and qualitative methodology. Quantitative methodology is applied to research that implies analyzing big amount of data to prove or refute a phenomenon. Qualitative methodology is used instead when a phenomenon has to be studied to create a theory or knowledge by probing the terrain. This is therefore the qualitative research methodology which will be applied here, as data from a mobile operator will be analyzed to deduce facts, and build a model.

The research methods are numerous. An overview of these methods is proposed in figure 1 below.

Figure 1: Portal of research methods and methodologies[5]

The research method applied in this thesis is the applied research method, as the purpose of this thesis is to solve known and practical problems. Moreover, the results are related to a particular situation, which justifies the choice of this method[5].

This thesis constitutes a case study; by looking at the case of the mobile

operator Telenor Sweden, the more global phenomenon of closing a site will

be investigated.

4 1.6 Delimitations

This thesis focuses on 3G UMTS sunset, assuming than 4G LTE is the solution used to provide service continuity. Although this thesis focuses on specific technologies, most operators around the world actually use UMTS as a 3G cellular technology, and LTE as a 4G cellular technology.

Besides, this thesis considers a case where the 2G GSM network is kept for a longer time than the 3G UMTS network, and therefore that it could be used as a fallback solution to 3G where 4G does not fit in (e.g., for IoT and M2M applications which rely on GSM).

It is worth noting that an analytical work like this one may be of interest to a single mobile network operator, or to a joint venture between several mobile network operators that own a shared network. This also means that, when a network is shared between two or more mobile operators, an agreement may not be reached between these operators, if they plan to switch technology at different times. In such a case, it might delay the decision-making process.

1.7 Outline

This thesis is divided into five chapters. Chapter 2 gives the reader the

necessary background of the field and introduces the reader to all the concepts

useful to fully understand this thesis. Chapter 3 describes the methodology to

obtain the results analysed in chapter 4. The thesis ends with conclusions,

discussion of the limitation of this work, and suggestions for future work in

chapter 5.

5

2 Technical Background

This section contains some elements of telecommunications networks that are necessary to understand when a site may be or may not be closed down, while keeping a good control of the risks involved.

2.1 Challenges in the telecommunications world

Matching increasing market expectations and a steady—if not increasing—

profitability is a must for telecom companies. Even though the Internet bandwidth per user increases significantly every year with an increase by 32%

worldwide between 2015 and 2016, telecommunication revenues decreased during the same period. As for the mobile broadband prices, they have decreased significantly as a percent of the national income per capita, falling down worldwide from 8.3% to 4.3% between 2013 and 2016 [1]. To face the financial consequences of this reality, mobile network operators have no choice but to decrease their operating costs.

2.2 Cellular networks

A cellular network is a telecommunication network where the last link is wireless. The last link is located between the mobile device and a site called a cell site or a base station. A network is made up of many base stations—

hundreds to thousands. To communicate with each other, the devices connect to the network via a base station. This station contains antennae that each serve an area called a cell. Users located in a cell may interact with the network by transmitting and receiving signals to this antenna.

In this thesis, “network” is to be understood as “set of cellular sites using the same technology and operated by a single network carrier”, meaning that a mobile operator may operate two 3G networks, for example one fully owned by itself and one operated by a joint venture with another operator.

We will see in the following sections how cellular networks are structured.

2.2.1 Frequency bands and ranges

A network operates in a given frequency band. For example, UMTS (3G) networks use the 2100 MHz frequency band, in Europe. For constant power, the higher the frequency, the lower the range. That is why, for low frequency bands, the coverage per site is wide, and the network of base stations sparser than for a high frequency network. Inversely, the higher the frequency band, the higher the frequency range and therefore the higher is the bandwidth availability for potential users to access the network. As an example, in the 1800 MHz frequency band used in GSM, the frequency range is higher—e.g.

1805–1880 MHz for downlink transfer (75 MHz)—than in the 900 MHz

band—where the range is 925–960 MHz for downlink transfer (35 MHz). As a

result, low frequency antennae may cover a wide-area cell and thus users

located far away from the antenna, but might not be able to serve a high

amount of users. Whereas high frequency antennae cover smaller areas, but

allow a higher number of users to interact with the network.

6 2.2.2 Network architectures of today’s widespread mobile

technologies

This section presents the different network architectures of currently used mobile technologies: 2G, 3G, and 4G. This will help understanding better what elements may be removed when dismantling a cell site, and what elements would instead be kept for use by another network.

2.2.2.1 General structure of a mobile network

A mobile network is constituted of radio access networks (RAN’s) and a core network. The radio access network is the part of the network that communicates directly with the mobile devices. The core network is the part of the network that provides services like Internet access and connection to the fixed telephone network; it also where the mobile users’ data is stored. Every mobile network generation (2G, 3G, etc.) is organized like this.

Let’s dive more in detail in the different mobile network generations in use today, to understand what may be removed from a network when it is shutdown.

2.2.2.2 GSM (2G)

2G networks are the first generation of digital cellular networks. A GSM network is constituted of radio access networks (RAN’s) and a core network.

Incoming traffic from mobile subscribers (calls and data) are received by the nearest base transceiver station (BTS) – or the one offering the better signal – which is controlled by a base station controller (BSC). Generally, BSC’s are remotely controlling several BTS’s – tens to hundreds of them. Traffic goes from the UE to the BTS through an air interface called Um interface, and from the BTS to the BSC through terrestrial cables, via Abis interface (cf. figure 1).

GSM networks in Europe use the 900 and 1800 MHz frequency bands.

Figure 2: Structure of a GSM network[6]

2.2.2.3 UMTS (3G)

7 UMTS is a third generation set of standards for mobile networks, the one mainly used worldwide. Unlike GSM, UMTS networks operate in the 2100 MHz band, which means that UMTS sites cover smaller areas than GSM sites, but a higher number of users per area, and/or increased bandwidth per user per area.

Although the major part of the core network is similar in 2G GSM and 3G UMTS, the migration from GSM to UMTS implies new equipment to be deployed. In particular, a UMTS network requires a NodeB for each cellular site, and radio network controllers (RNC), which handle remotely tens to hundreds of NodeB’s.

Figure 3: Structure of an UMTS network[7]

As in 2G GSM networks, 3G is both circuit-switched and packet-switched based. We will see next how 4G differs from its predecessors.

2.2.2.4 LTE (4G)

The 4G wireless service LTE differs significantly from its predecessors in that it is an all-IP mobile network; it does not rely on a circuit switching method to handle calls, as in 3G UMTS or 2G GSM technologies. However, this all-IP flat architecture requires new equipment. First, it requires only an eNodeB in the radio access network (see figure 3), which handles the functionalities of both the NodeB and the RNC in 3G UMTS. This eNodeB can be collocated with a 2G BTS or a 3G NodeB, to avoid having to rent another site. However, LTE may also work in the 2600 MHz band, which requires a denser network in the densely populated areas. Therefore new sites would be required in a LTE network.

More importantly, as an all-IP network, LTE requires many changes in the

core network. This is mainly due to the fact that the voice calls need to be

handled not anymore by the traditional circuit-switched network but by the

new LTE standard called Voice over LTE (VoLTE). This will be detailed in the

next section, as this is an important part of the 3G service shutdown.

8

Figure 4: LTE network architecture[8]

2.2.2.5 VoLTE

VoLTE is the standard that allows voice calls within the all-IP LTE network.

Because VoLTE is the main reason for an LTE network to require a new architecture, LTE networks are first deployed without voice capability. This means that, to deploy LTE faster, operators have made the choice to offer 4G service quality for data services, and keep relying on their 2G/3G networks for voice calls. Therefore, as long as the LTE network is not fully deployed, every time an LTE user wishes to make a voice call, its device will switch to the strongest available voice-capable mobile network, whether 3G or 2G.

To support natively voice and SMS services, an LTE network must implement the IP Multimedia Subsystem (IMS) architecture, which requires many changes in the mobile network. Without the IMS implemented, a “simple”

change in the Serving Gateway (SG) in the core network is necessary for LTE users to be able to make voice calls or send SMS, by switching to 2G/3G whenever it happens. This switch is called a handover or a handoff in cellular mobile communications. Handovers are treated in the next section.

Moreover, users must have a VoLTE capable phone to be able to use VoLTE

services. A list of such devices may be found at[9].

9 Also, the mobile network operator needs to make sure any VoLTE phone can actually communicates with its network. Indeed, the parameters that are sent between the device and the network vary from one operator to another.

As a consequence, to shut down their 3G network, mobile operators must consider an alternative solution to provide voice calls, which will be treated the section 2.3.

2.2.2.6 Handover

A handover is the process of transferring a session from one cell to another, without service interruption. It is used in several situations, the most common being when the device moves away from one cell to the area covered by another cell[10]. It is also used for load balancing, when a cell is overloaded.

2.2.3 Spectrum license

Every mobile operator needs to own licenses to have the rights to send and receive radio signals to cellular devices. Indeed, use of electromagnetic spectrum is regulated in most countries in the world. Every technology works on a specific frequency band—e.g. in Europe, the 2100 MHz band is used for 3G, while the 900 and 1800 MHz bands are used by 2G GSM networks—. A spectrum license is usually very expensive, depending on which band it allows to operate on, the frequency range it covers, and the geographical area where the license applies. Moreover, it varies according to government’s choice to set up a minimum price for licenses, and according to the competitive environment. Spectrum licenses are sold at auctions, where total raisings generally exceed one billion dollar in most developed countries, and may reach tens of billions of dollars[11].

2.3 Closure of a mobile network

It is common, for mobile operators, to have to close one of their mobile networks to follow technological trends. This section explores a few cases in which it happened, and provides an overview of the consequences of a closure.

2.3.1 A few examples of network shutdowns—and their consequences Even though Telenor Sweden plans to shut down their 3G network before their 2G network, this is not the case for all mobile operators. Vodafone Australia, for instance, closed down its 2G network in June 2018[12], upon announcement on September 2016, in favor of their 3G and 4G networks. This required transitioning the users from the 2G to their 3G and 4G networks, in particular to upgrade their handsets so they can make voice calls using LTE.

This also required extending and strengthening their 4G infrastructure over the past years before dismantling their 2G networks. Similarly, before shutting down a 2G network, an operator might want to make sure that they are not selling 2G-only capable phones anymore, to let the 2G traffic decline naturally[13]. In such a case, network operators have been offering free 3G/4G SIM cards to users that have a 3G or 4G capable phone but are connecting to 2G[14].

On a different point, dismantling cellular sites imply a lot of network junk, that the operator will have to recycle[15].

2.3.2 Cost savings when closing a 3G network

10 When deploying a cellular site, operators need to rent a location to install the site. It may be a greenfield, an attic in a building, a rooftop, or any other place where a base station may fit in, and where antennae may be placed. Rents generally constitute the main operating cost in a mobile network. That is why, when closing a network, operators may save money from rent. However, sites may be used for several networks at a time. For example, a space rented for a 2G site may also be used for a 4G base station. In such a case, the cost for the rent may not be saved when shutting down the 2G base station.

When removing a base station, the equipment that has been settled in that location must be removed, and technicians have to be sent on site. The cost of sending technicians must therefore be deduced from the amount saved from the rent.

Secondly, when base stations are operating, operators generally pay software licenses for the stations they operate. Once they close down a whole network, these software license fees may be saved.

In addition to software licenses, the hardware itself requires licenses to work at a certain capacity. Whenever a station is not anymore used, the hardware licenses do not need to be renewed as soon as it ends up.

Then, electricity needs to be paid for the equipment to function. This cost is low, although it does not shrink proportionally with the station’s load.

There are also fees that must be paid to cables operators that let the mobile operators send data over their cables. When closing down a network, the traffic within that network is very low and this cost is negligible—and in general transferred to other networks. We will thus neglect this cost in the following.

Finally, spectrum licenses constitute a big cost for a mobile network to operate. However, these licenses last for years (usually at least 15 years), and are paid when acquiring the license. Also, spectrum may be generally reused for newer technologies. For example, the 2100MHz band used for 3G may be reused in 4G networks, which may work in that frequency band. Reusing spectrum licenses for newer technology allows increasing the capacity of the 4G network, without having to buy a new license. This therefore may be counted as a cost saving, depending on whether these licenses will yet last some years before expiring, and if they are technology-independent, as sometimes licenses are bought for only a particular technology. This was not the case for the 2100 MHz (3G) spectrum auction in Sweden, where the spectrum sold was technology-independent.

2.4 Big data

With large amounts of data being generated and owned by telecom companies, the necessity to dive in these data and leverage them has emerged.

When traditional tools and methods do not allow querying these data in an efficient and easy way, and thus when new technologies are required to process these data, we talk about “big data”[16].

“Big data” is a term that was used first in its current meaning in the late

1990’s[17], before becoming popular in the beginning of the 2010’s[18]. It

encompasses every data-related work that concerns vast amounts of data.

11 Querying big data requires specific tools, and generally a data query language.

SQL or one of its derivatives is generally used. Section 4.1 describes in more

detail the tools used in this project.

12

3 Research methodology

The purpose of this chapter is to provide a research method used for this project. Section 3.1 describes the research process. Section 3.2 describes the origin of the data. Section 3.3 describes the data used for this study.

3.1 Research process

The research process consists in analysing the operator’s traffic data, to understand better what conditions need to be fulfilled to operate the transition from 3G to 4G networks. It is assumed that the 2G network will be continued after the 3G network, as 2G is not only used by mobile users, but also widely used by many machine-to-machine applications and the Internet of Things. Using the knowledge gained from meetings and discussions within Telenor Sweden and research done for this project, a model was built and refined during the research process. This model aims at helping in deciding whether to close or not a cellular site, hence its name, “profitability threshold model”. It is presented in last section of this chapter.

3.2 Origin of the data

The data used for this business research are collected from the mobile networks of Telenor Sweden, the second largest mobile operator in Sweden with more than 2.5 million subscribers. The data is provided by a third-party service provider, which collects them using tap devices. These tap devices capture the traffic in the backhaul part of the network (see figure 5). Mostly, this means data are collected from fibre optic links, which are the main type of link used in the backhaul part of Telenor Sweden’s network.

Figure 5: Backhaul part of a mobile network[19]

13 3.2.1 Protection of sensitive data

It was a must, for this project, to make sure the data are processed according to enforceable regulations. Therefore, only strictly necessary data were used.

Information such as the user id (or IMSI for International Mobile Subscriber Identity) was encrypted using a hash function, at the time of the data collection. Moreover, customer’s personal information such as name, address, etc., was not part of the data set either. The data used for this study are described below.

3.3 Data used for this study

The data used for this study come all from the same source—tap devices collecting traffic on fibre optic links—but they are of different nature.

On one hand, in order to locate the sites that were the least used sites in Telenor’s 3G networks, location data were used—these location data contain the main traffic data in Telenor’s network, from which can be derived at what time a device communicated with a particular cell.

On the other hand, the data describing the changes in radio access technology (RAT) were used. These “inter-RAT” data were used to figure out what are the reasons why users switch over to a 3G network when they are using 4G. By figuring this out, an operator may take the right steps that will help it towards a full 4G transition, allowing shutting down its 3G network.

The data used were formatted as follows—we will rely on this format later on to describe the data mining.

3.3.1 Location data

The location data contained the following fields:

• Timestamp: the timestamp at which the communication occurred.

• IMSI: the user id. This id is encrypted using a hash function and therefore does not allow matching to a specific person.

• Cell name: the name of the cell involved in the communication.

• Type allocation code: a code that identifies the model of the device.

3.3.2 Inter-RAT data

The inter-RAT (inter-Radio Access Technology) data were structured as follows:

• IMSI (hashed): the International Mobile Subscriber Identity

• Type allocation code: a code that allows identifying the model of the device involved in the handover.

• Start time (UNIX time, milliseconds): time from which the radio access technology has changed.

• First LAC: Location Area Code of the cellular site used for the first event; together with the Service Area Code, it identifies uniquely a cellular site. Applicable to 2G/3G only.

• First SAC: Service Area Code of the cellular site used for the first

event; together with the Location Area Code, it identifies uniquely a

cellular site. Applicable to 2G/3G only.

14

• First eNodeB: identifier of the eNodeB used during the first event (if applicable).

• First Sector ID: identifier of the sector of the eNodeB used during the first event (if applicable).

• RAT: radio access technology used since the start time; may be 2G, 3G or 4G.

• Previous RAT: radio access technology used before the radio access technology change.

• Cause type (S1AP only): the type of the cause for the radio access technology change (as per the cause table below)

• Cause: the cause for the radio access technology change (as per the cause table below)

• Voice time (milliseconds): the duration during which voice services were used, after the start time.

• Data time (milliseconds): the duration during which data services were used, after the start time

3.3.3 Cell data

The data concerning the cells were stored in table containing different types of information about the cells:



Name: the name of the cell



Site name: the name of the site on which the cell is located



Municipality: the municipality in which the site is located



Street: the name of the street where the site is located



Latitude: the latitude of the site’s geographical coordinates



Longitude: the longitude of the site’s geographical coordinates



Location area code: the code identifying the location of a group of 2G/3G sites. Together with the service area code, it uniquely identifies a site.



Service area code: the code identifying the location of a 2G/3G site within a location area. Together with the location area code, it

uniquely identifies a site.



eNobdeB id: the id of a LTE (4G) site …



Sector id: …



RAT: radio access technology (GSM/UMTS/LTE) used by the cell.



Network: the network owner: Telenor or 3GIS or Net4Mobility.



Band: band on which the cellular station operates.

3.4 Profitability threshold model

A profitability threshold model is proposed here, to suggest whether it is

relevant to close down a cellular site or not. This model is based on empirical

evidence and experience, as many parameters have to be taken into account to

take such a decision. It reminds what parameters are important when

determining when to close down a site, while keeping in mind the financial

impact of such a decision. Also, a site might have to be closed with other sites

15 in the same “cluster”. In such a case, the model must be applied to all sites in the cluster.

For this model, the following notations are used:



X

is the number of unique customers connecting i hours per station (with i ϵ {m, n, p}

and 10 ≤ m < 50, 50 ≤ n < 120 and p ≥ 120), over a 1-week period. Although the value of i is arbitrary, it was chosen to take into accounts (a) the fact that a connection does not mean necessarily activity and (b) the fact that the 2G network and the 4G network would naturally take over a good part of the 3G traffic. It may be adjusted to see how it impacts the result.



P

is the proportion of users who connect i hours to the station and who would churn if the station was shut down. P

= 2 %, P

= 7 %, P

= 20 %.



Y is the average revenue per user (ARPU) for users connecting to that specific site. If this value cannot be determined for a specific station, the nationwide ARPU may be used.



D is a (optional) parameter representing incentives to retain users that would be

impacted by such a station shutdown (users living or working within the area, typically), and that would need to upgrade their device to continue using the operator’s network (because, for example, their phone is not compatible with 4G, or is not VoLTE capable).

Such an incentive could be a discount on a 4G-capable phone.



M is the maintenance cost for the base station. It includes the site rent. This is generally known by the mobile experts working for the mobile operators and will not therefore be decomposed for the purpose of this thesis. It constitutes, together with the energy consumptions cost, the cost savings that occur when closing down a base station.



E is the energy consumption cost for the station. It constitutes, together with the station maintenance cost, the cost savings that occur when closing down a base station.

The profitability threshold is reached when the condition of the equation 1 is satisfied.

In more detail, profitability is reached when the equation 1 is satisfied.

𝑀 + 𝐸 > (∑ 𝑃

∗ 𝑋

∗ 𝑌

𝑖 ϵ { 𝑚,𝑛,𝑝 }

) + 𝐷

Equation 1: Profitability threshold model

Considering the following:



R is the total revenue generated by a site that would not be lost if the site was closed.

𝑅 = ∑ 𝑃

∗ 𝑋

∗ 𝑌

𝑖 ϵ { 𝑚,𝑛,𝑝 }



D = 0.



C = M + E, is the total cost savings.

The equation may be noted:

𝐶 > 𝑅

Equation 2: Simplified profitability threshold model

16

17

4 Data mining

This chapter describes how the data were analyzed and processed. It also describes, in its last section, the model proposed to help deciding for the closure of a 3G site.

4.1 Data analysis tools

The analysis of the data was made using a query language, a programming language to process the results, and a cloud platform to allow for querying big amount of data seamlessly.

The cloud platform used to query the data is Amazon Web Services (AWS).

AWS is widely used among big companies to store, query and process their data. It offers many data science tools among which a few have been used for this project—namely S3, Athena and EC2—, they are described below.

S3 (Simple Storage Service) is a storage service used to store big amounts of data[20]. In our case, it was used to store all the location data and the inter- RAT data, for several months during which the analysis was made. Data were uploaded there in real time for the location data, or every day for data that are less subject to change (e.g. cellular sites locations).

Athena is a query service that allows querying data using SQL. It allows querying data located on S3[21]. As it is based on Apache Presto

, a distributed SQL query engine, it is designed to process efficiently big amounts of data. It was preferred over Amazon Redshift because it was more performant with regards to the amount of data actually processed per query[22].

EC2 is a virtual server providing compute capacity in the cloud[23]. It was used to do further processing with the data, using Python frameworks pandas

and numpy

.

4.2 Data processing

To be able to iterate quickly, this thesis work presents the results for a single city in Sweden—namely Ronneby. All the 3G sites in this city were taken into account by the analysis.

To fit within the scope of this project, the data used to evaluate the sites usage in Ronneby concern the period ranging from 2018-06-05 to 2018-06-18.

Data were processed as follows:



Based on the location data, it was computed for each user how long they stayed connected to a site: each event in the location data can be mapped to a site; as a consequence, the time difference between any pair of successive events for each site gives the time spent on a site; a sum of all the differences for a given site allows to know how long a user connected to a site—including when in idle mode;

1

https://prestodb.io/

2

https://pandas.pydata.org/

3

http://www.numpy.org/

18



Then the rows corresponding to Ronneby sites were filtered (to look specifically at the Ronneby case);



The next step was to use the formula designed in house, and described

in section 5.1, equation 1.

19

5 Results

This chapter describes the results of the research project. In section 1, results concerning the sites to be closed will be described. In section 2, the reasons why users switch from 4G to 3G will be presented, and the section 3 shows the distribution of devices switching from 4G to 3G.

5.1 Candidate sites for a closure

This sections presents the sites in the municipality of Ronneby in Sweden that are potential candidates for a closure, with regards to the values of the parameters used in the formula described in section 5.1, equation 1.

For each site, recurring users were distributed into 3 classes: users having connected between 10 and 50 hours to the site, users having connected between 50 and 120 hours to the site, and users having connected more than 120 hours to the site. Each class of users was assigned a churn probability: the first class (10 ≤ 𝑡 < 50) was assigned a 2 % churn probability, the second class (50 ≤ 𝑡 < 120) was assigned a 7 % churn probability, and the third class (𝑡 ≥ 120), a 20 % churn probability.

The reason why the probabilities were chosen so low is that in any case, the 2G network will be used as a fallback solution to handle the users who cannot connect to the 4G network. And sometimes, the 4G network itself will be used, although with a poorer quality. As a consequence, if users churn, it will be most likely due to a low quality, rather than a lack of 3G network.

With such probabilities, the results are presented in Table 1.

20

Table 1: Sites to be closed (given first set of parameters)

Site name Revenue

(SEK) 3G closure

4G

colocation Users connected at least 10 hours

Users connected at least 50 hours

Users connected at least 120 hours

Skärsjön 70 985 No Yes 250 168 81

Ronneby Hamn 15 361 No No 231 86 41

Kallinge/Teracom -7 163 Yes Yes 365 201 88

Förkärla 48 601 No Yes 258 130 52

Ronneby vattentorn 179 974 No Yes 600 363 209

Soft Center -10 513 Yes Yes 227 63 29

Göholm 67 704 No No 498 271 114

Garnanäs -45 Yes No 172 54 19

Fridhemsvägen 211 142 No No 761 360 203

Bräkne Hoby C 58 796 No No 311 189 93

Hillerslätt 8 597 No No 119 66 16

Ekenäs 25 196 No No 247 115 45

Parkdala 78 250 No No 441 208 91

Galtsjön 6 001 No Yes 118 40 15

Kallinge, Häggatorp 109 139 No No 587 238 105

Bohaga -6 242 Yes No 139 84 31

Kungsgatan 103 949 No No 610 290 128

Salsjövallen -19 070 Yes No 176 54 17

Johannishus 71 968 No No 365 179 87

Långkärra 21 250 No No 199 94 21

Vieryd 79 792 No No 463 230 80

Edestad 9 460 No No 145 80 21

Leråkra -5 313 Yes No 166 75 29

Hasselstad 96 755 No No 663 181 72

Ronneby Saxemara 41 662 No Yes 247 121 51

Ronneby Aspan 7 580 No No 150 55 19

Kallinge Centrum 95 173 No No 500 266 135

Västra

Industriområdet 91 303 No No 602 205 83

Table 2: Average revenue per class of recurring users and per site

connected at least 10 hours

ARPU for users connected at least 50 hours

ARPU for users connected at least 120 hours

Skärsjön 3154 3185 3116

Ronneby Hamn 3192 3360 3607

Kallinge/Teracom 3156 3091 3036

Förkärla 3117 3186 3384

Ronneby vattentorn 3306 3381 3452

Soft Center 3311 3736 3713

Göholm 3118 3059 3035

Garnanäs 3018 2992 2890

21

Fridhemsvägen 3508 3696 3840

Bräkne Hoby C 3250 3259 3403

Hillerslätt 3044 3029 3119

Ekenäs 3143 3129 3053

Parkdala 3090 3066 3031

Galtsjön 3185 3257 3747

Kallinge, Häggatorp 3331 3314 3275

Bohaga 3128 3108 2890

Kungsgatan 3146 3187 3349

Salsjövallen 3245 3196 3322

Johannishus 3126 3146 3228

Långkärra 3120 3046 3065

Vieryd 3124 3049 2935

Edestad 3054 3050 2977

Leråkra 3089 2988 3016

Hasselstad 3436 3397 3553

Ronneby Saxemara 3061 3102 3142

Ronneby Aspan 3073 2956 2890

Kallinge Centrum 3253 3339 3461

Västra

Industriområdet 3195 3248 3399

The Table 1 shows that 6 sites out of 28 would have to be closed according to the set of input parameters. It shows only the sites located in Ronneby that are owned by Telenor.

Table 2 shows the average revenue per user per site, according to how long users connected to the network. As one can see, the variance of the revenue per site per time spent on the network is pretty low for all sites; moreover, the connection time per site is not directly proportional to the time spent making calls or using data—it also includes idle time. As a consequence, this parameter may be ignored in such an analysis.

The energy consumption cost per site was set to 2,000 SEK per year, as these data were not readily available.

5.2 Causes distribution for switching from 4G to 3G

This section shows the results concerning the reasons why 4G-capable devices switch from 4G to 3G.

The figure 6 shows the distribution of the main reasons.

22

Figure 6: Main reasons for a switch from 4G to 3G, in Telenor's network

The reasons correspond to different scenarios, as explained in Table 3.

Table 3: 4G to 3G switch causes’ distribution

Reason Explanation Importance to thesis

User inactivity The device connected to another network due to inactivity.

This situation should not cause a problem if a 3G site is closed, as the switch either will not happen or it will happen with the 2G network (if available).

Normal release

Switch required by the network rather than by the user’s device itself.

This reason will be ignored in this thesis as it does not concern a cause an operator may act upon.

Successful

handover Mainly due to coverage-related

issue This cause means that the

4G network may offer (significantly) better capabilities than the 3G network to avoid such handover.

Unknown

target RNC The communication that was erroneous between the device and the network, leading to close the connection and switch over to the closest available station.

As for “normal release”, the reason is purely technical and is hard to exploit.

“Unknown”

cause The cause is unknown, due to

corrupted data. Unknown causes are rare

and may be ignored in this

thesis.

23 Inter-RAT

redirection An inter-radio access technology redirection is another type of handover which, unlike a “successful handover”, may require a service suspension[24].

May be considered together with “successful handover”, even though this cause’s scarcity makes it easy to ignore.

Other Many other causes exist. They are ignored as they happen too rarely (less than 5 % of the time).

The table 4 shows how many switched occurred for each cause. Results were computed over a 14-day period, and filtered for Ronneby sites, for users who connect at least 10 hours a week to a site, and users who have a VoLTE- capable phone only. This allows filtering causes that are due to a circuit-switch fallback, which happen every time a user connecting to 4G with a non-VoLTE- capable phone initiates a call.

Table 4: Causes distribution for a switch from 4G to 3G, in Telenor's network

Cause Causes distribution Percentage

User inactivity 1750 32,9 %

Normal release 1354 25,4 %

Successful handover 792 14,9 %

Unknown target RNC 452 8,5 %

Unknown 372 7,0 %

Inter-RAT redirection 274 5,4 %

Authentication failure 127 2,4 %

CS fallback triggered 114 2,1 %

Results show that the most common reason for a switch from 4G to 3G is related to user inactivity, that is to say it happens when the device is in idle mode. The cause “successful handover” corresponds mainly to coverage- related issue, and represents almost as much as half of user inactivity cases, with 15% of total.

It is worth to mention here that although the causes are described in an ETSI

technical specification[24], they are hard to match to real-world situations as

their descriptions in the specification are lacking clear explanations.

24

6 Conclusions and future work

This chapter presents general conclusions about this thesis work, and the limitations faced by this project. Some future work is suggested as well.

6.1 Conclusions

Even though not any public research work about closing a cellular network was found, there have been in the past some occurrences of cellular networks being closed, and these examples could be analyzed to understand the dynamics of closing a cellular network. This thesis shows that it is possible to leverage corporate data to guess which cellular site is little used and can be considered for closing as soon as the 4G network is up and running for the vast majority of users.

With regard to the goal of creating a model, the model proposed in section 5.1 aims at inspiring for future works: it is practical and based on reflections and knowledge of the question as discussed with mobile experts and Data Scientists at Telenor Sweden. It may be improved further but should provide a better way to evaluate 3G site usage than raw data. It also provides a complementary way that the mobile team in a telecom company does not have, to guess how heavy is the traffic according to different parameters—

including, regarding the users that connect on a recurring basis to the same site, as these users require a closer look.

This project shows that depending on the criteria of what a recurring user is — in this case, it depends on how long they connect to a 3G site—, the results listing the sites to close vary a lot.

6.2 Future work

This thesis has some limitation, as it is a use case realized at a mobile network company, and therefore some future work may be realized if one wants to overcome these limitations.

First, as Telenor Sweden has not implemented 4G calls for all its users as of August, 2018, results show a low number of sites that could be closed; if more users would be able to make calls using the 4G VoLTE service, certainly more sites would have to be closed. The logical next phase to gain insights from the data would be to run the analysis again, once the VoLTE network is fully deployed within Telenor Sweden.

Another way to improve this work would be to reconsider the criteria for a recurring user: instead of taking into account just the time they were connected to a site—which was the solution that could be done in this thesis—, it would be more accurate to consider only the time they actually spent making calls, sending SMS or using data on the network—i.e. excluding the time when they are in idle mode.

Finally, it would be helpful to take a closer look at the causes why users switch from 4G to 3G, as the official explanations are lacking detailed information.

This could help not only to understand better what to fix before forcing the

transition from 3G to 4G, and perhaps even to fix some other problems that

would be pointed by such data.

25

References

[1] ITU, Ed., “ICT Facts and Figures 2017.” Jul-2017.

[2] “Vodafone UK first to test new 5G spectrum across a live network, marking a major milestone in UK telecommunications,” Vodafone UK Media Centre. . [3] Dallas, Texas, Jan 04, and 2018, “AT&T to Launch Mobile 5G in 2018.”

[Online]. Available:

http://about.att.com/story/att_to_launch_mobile_5g_in_2018.html. [Accessed: 22- Jun-2018].

[4] “KT showcases 5G innovation at the Olympics in PyeongChang.” [Online].

Available: http://news.itu.int/kt-showcase-5g-olympics/. [Accessed: 22-Jun- 2018].

[5] H. Anne, “Portal of Research Methods and Methodologies for Research Projects and Degree Projects.” Proc. Int. Conf. Front. Educ. Comput. Sci. Comput. Eng.

FECS’13.

[6] Tsaitgaist, “File:Gsm structures.svg,” Aug. 2012.

[7] Tsaitgaist, “File:UMTS structures.svg,” Aug. 2012.

[8] Crati, “File:EUTRAN arch.op.svg,” Jul. 2010.

[9] Wikipedia contributors, “Voice over LTE,” Wikipedia, The Free Encyclopedia.

13-May-2018.

[10] G. Z. J. S. K. and B. S. S. Miao, Fundamentals of Mobile Data Networks.

Cambridge: Cambridge University Press, 2016.

[11] “German Auction of 3G Mobile Phone Licenses Raises 50.5 Billion Euros.”

[Online]. Available: https://www.gartner.com/doc/314369/german-auction-g- mobile-phone. [Accessed: 25-Jun-2018].

[12] “We’ve switched off our 2G network.,” Vodafone Australia. [Online].

Available: https://www.vodafone.com.au/support/network/2G-closure. [Accessed:

06-Aug-2018].

[13] J. Taylor, “Telstra to shutter 2G network by the end of 2016,” ZDNet.

[Online]. Available: https://www.zdnet.com/article/telstra-to-shutter-2g-network- by-the-end-of-2016/. [Accessed: 06-Aug-2018].

[14] T. A. D. of PriMetrica, “Telstra switches off GSM network,”

https://www.telegeography.com. [Online]. Available:

https://www.telegeography.com/products/commsupdate/articles/2016/12/02/telstr a-switches-off-gsm-network/index.html. [Accessed: 06-Aug-2018].

[15] K. Fitchard, “Nextel shutdown will leave 45,000 tons of network junk. How Sprint plans to recycle it,” 05-Jun-2013. [Online]. Available:

https://gigaom.com/2013/06/05/nextel-shutdown-will-leave-45000-tons-of- network-junk-how-sprint-plans-to-recycle-it/. [Accessed: 06-Aug-2018].

[16] F. Provost and T. Fawcett, “Data Science and its Relationship to Big Data and Data-Driven Decision Making,” Big Data, vol. 1, no. 1, pp. 51–59, Mar. 2013.

[17] G. Press, “A Very Short History Of Big Data,” Forbes. [Online]. Available:

https://www.forbes.com/sites/gilpress/2013/05/09/a-very-short-history-of-big- data/. [Accessed: 30-Jul-2018].

[18] “Google Trends,” Google Trends. [Online]. Available: /trends/explore.

[Accessed: 30-Jul-2018].

[19] “Backhaul Networks for Broadband Mobile Communication: Tendencies and

Perspectives of Development in Russia and in the World - ШПД, GPON,

магистральные сети | RUSSIAN ANALYTICS.” [Online]. Available:

26 http://json.tv/en/ict_telecom_analytics_view/backhaul-networks-for-broadband- mobile-communication-tendencies-and-perspectives-of-development-in-russia- and-in-the-world-2014090505001818. [Accessed: 25-Jun-2018].

[20] “Cloud Object Storage | Store & Retrieve Data Anywhere | Amazon Simple Storage Service,” Amazon Web Services, Inc. [Online]. Available:

https://aws.amazon.com/s3/. [Accessed: 05-Aug-2018].

[21] “Amazon Athena — Serverless Interactive Query Service - AWS,” Amazon Web Services, Inc. [Online]. Available: https://aws.amazon.com/athena/.

[Accessed: 05-Aug-2018].

[22] A. Brody, “Amazon Athena Does Battle: Comparing Athena and Redshift.”

[Online]. Available: https://blog.panoply.io/an-amazonian-battle-comparing- athena-and-redshift. [Accessed: 05-Aug-2018].

[23] “Amazon EC2,” Amazon Web Services, Inc. [Online]. Available:

https://aws.amazon.com/ec2/. [Accessed: 05-Aug-2018].

[24] European Telecommunications Standards Institute, “ETSI TS 136 413 V11.4.0 (2013-07).” ETSI, Jul-2013.

[25] European Telecommunications Standards Institute, Ed., “ETSI TS 136 413

V12.3.0.” Sep-2014.

1

TRITA -EECS-EX-2018:565

www.kth.se