QoEAwareness For TheMobileTerminationMarketMonopoly

Liberation

Doctoral Thesis

For the Degree ofDoctor of Informatics

At the Faculty of Economics,Business Administration AndInformation Technology

of theUniversity of Zurich

byChristos Tsiaras

fromGreece

Accepted on the recommendation ofProf. Dr. Burkhard Stiller

andProf. Dr. Peter Reichl

2015

2

Abstract

Nowadays, in mobile communication only the Mobile Network Op-erator (MNO) of the callee is able to terminate his calls. Thus, in theMNOs call-terminationmarket there is only one player profiting fromMo-bile Termination Rates (MTRs); in turn this market is considered to be a“de-facto” monopoly since the early days of the introduction of commer-cial mobile communication services. Given this monopoly fact, the onlysolution against a potential speculation byMNOswas the regulation of theMTRs.

A monopoly is a corporation that is the only seller of a good or a ser-vice, and thus it can define the price. However, monopolies can be dividedinto two categories, the naturally defined and the market-defined monop-olies. The power market in many countries is considered to be a naturalmonopoly, and the main reason is that there is usually only one wire reach-ing each house. Thus, only the company that owns the delivery networkcanprovidepower services. The termination service inmobile communica-tion is also considered to be amonopoly. However, this is amarket-definedmonopoly, since there is no physical limitation (e.g., wires) for reaching amobile user.

Today the technology allows for different charging mechanisms thatcould overcome the mobile voice termination-service monopoly obstacle.This thesis shows that the mobile termination-service should not be con-sidered amonopoly anymore. To reach this goal, theAuction-basedCharg-ingUser-centric System “AbaCUS” is proposed in this thesis, whereMNOswill participate in an auction. MNOs will bid on termination rates per lo-cation and per Quality-of-Service (QoS) parameters, such as the network-access priority and QoS parameters during calls, in a manner that Quality-of-Experience (QoE) of end-users is maximized.

This thesis shows that charging for Quality-of-Experience (QoE), is thekey to overcome the monopoly of theMTRsmarket. To estimate what theend-user’s QoE will be, for a certain service that competing MNOs are of-fering; and use it as a biddingmetric in the AbaCUS auction, a Determinis-

iii

tic QoEmodel (AQX), which considers simultaneously technical and non-technical parameters is created in this thesis. AQX is a generic QoE modelthat can be applied in multiple domains. Thus, the roadmap of selecting allthe AQX parameters needed, and further optimization of the AQX resultsis presented for the VoIP scenario. This scenario is selected as a referenceto present and evaluate AQX, since there are several well-studied existingQoEmodels. AQX has been proven to outperform the state of the art QoEestimation models for VoIP, IQX Hypothesis and the ITU-T E-Model, inseveral cases.

Finally in this thesis, incentives of the mobile communications stake-holders (MNOs, end-users,MNOsproviders, and regulators), to adopt theidea of a competitive termination-service environment in the mobile call-termination services market are presented.

iv

Kurzfassung

TODO: Patrick Poullie agreed to do the translation of the Abstract inGerman. Thank you Patrick! Still the English version here.

Nowadays, in mobile communication only the Mobile Network Op-erator (MNO) of the callee is able to terminate his calls. Thus, in theMNOs call-terminationmarket there is only one player profiting fromMo-bile Termination Rates (MTRs); in turn this market is considered to be a“de-facto” monopoly since the early days of the introduction of commer-cial mobile communication services. Given this monopoly fact, the onlysolution against a potential speculation byMNOswas the regulation of theMTRs.

A monopoly is a corporation that is the only seller of a good or a ser-vice, and thus it can define the price. However, monopolies can be dividedinto two categories, the naturally defined and the market-defined monop-olies. The power market in many countries is considered to be a naturalmonopoly, and the main reason is that there is usually only one wire reach-ing each house. Thus, only the company that owns the delivery networkcanprovidepower services. The termination service inmobile communica-tion is also considered to be amonopoly. However, this is amarket-definedmonopoly, since there is no physical limitation (e.g., wires) for reaching amobile user.

Today the technology allows for different charging mechanisms thatcould overcome the mobile voice termination-service monopoly obstacle.This thesis shows that the mobile termination-service should not be con-sidered amonopoly anymore. To reach this goal, theAuction-basedCharg-ingUser-centric System “AbaCUS” is proposed in this thesis, whereMNOswill participate in an auction. MNOs will bid on termination rates per lo-cation and per Quality-of-Service (QoS) parameters, such as the network-access priority and QoS parameters during calls, in a manner that Quality-of-Experience (QoE) of end-users is maximized.

v

This thesis shows that charging for Quality-of-Experience (QoE), is thekey to overcome the monopoly of theMTRsmarket. To estimate what theend-user’s QoE will be, for a certain service that competing MNOs are of-fering; and use it as a biddingmetric in the AbaCUS auction, a Determinis-tic QoEmodel (AQX), which considers simultaneously technical and non-technical parameters is created in this thesis. AQX is a generic QoE modelthat can be applied in multiple domains. Thus, the roadmap of selecting allthe AQX parameters needed, and further optimization of the AQX resultsis presented for the VoIP scenario. This scenario is selected as a referenceto present and evaluate AQX, since there are several well-studied existingQoEmodels. AQX has been proven to outperform the state of the art QoEestimation models for VoIP, IQX Hypothesis and the ITU-T E-Model, inseveral cases.

Finally in this thesis, incentives of the mobile communications stake-holders (MNOs, end-users,MNOsproviders, and regulators), to adopt theidea of a competitive termination-service environment in the mobile call-termination services market are presented.

vi

Contents

Abstract iii

Kurzfassung v

1 Introduction 11.1 QoE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31.2 Charging for QoE . . . . . . . . . . . . . . . . . . . . . . 41.3 Breaking MTRsMonopoly . . . . . . . . . . . . . . . . . 51.4 On-demandMNO Selection . . . . . . . . . . . . . . . . 61.5 Thesis Contributions . . . . . . . . . . . . . . . . . . . . 61.6 Thesis Outline . . . . . . . . . . . . . . . . . . . . . . . 7

2 RelatedWork 92.1 QoEModels . . . . . . . . . . . . . . . . . . . . . . . . 9

2.1.1 IQXHypothesis . . . . . . . . . . . . . . . . . . 112.1.2 E-model . . . . . . . . . . . . . . . . . . . . . . 13

2.2 Mobile Termination Rates Monopoly . . . . . . . . . . . 152.2.1 The CPP Principle . . . . . . . . . . . . . . . . . 162.2.2 The RPP Principle . . . . . . . . . . . . . . . . . 162.2.3 The NatRoam Principle . . . . . . . . . . . . . . 17

2.3 Flexible MNO Selection . . . . . . . . . . . . . . . . . . 182.3.1 Business Approach . . . . . . . . . . . . . . . . . 182.3.2 Hardware Approach . . . . . . . . . . . . . . . . 182.3.3 Flexible MNO Selection Use-case . . . . . . . . . 19

2.4 Contribution Opportunities . . . . . . . . . . . . . . . . 20

3 AQX: Axiomatic QoEModel 233.1 QoE Formalization in AQX . . . . . . . . . . . . . . . . . 253.2 Specific QoE Functions . . . . . . . . . . . . . . . . . . . 27

3.2.1 IVs QoE Function . . . . . . . . . . . . . . . . . 273.2.2 AVs QoE Function . . . . . . . . . . . . . . . . . 29

3.3 Influence Factors Calculation . . . . . . . . . . . . . . . . 31

vii

3.3.1 IVs Influence Factors Calculation . . . . . . . . . 323.3.2 AVs Influence Factors Calculation . . . . . . . . . 35

3.4 Generic QoE Functions . . . . . . . . . . . . . . . . . . 373.5 ITU-TMOS-Compilant QoE Functions . . . . . . . . . . 393.6 Simultaneous Economic and Technical Parameters Con-

sideration Example . . . . . . . . . . . . . . . . . . . . . 413.7 Chapter Summary . . . . . . . . . . . . . . . . . . . . . 42

4 AQX in VoIP 454.1 Experimental Setup . . . . . . . . . . . . . . . . . . . . . 46

4.1.1 Architecture H/W and S/W . . . . . . . . . . . . 474.1.2 Experimental Procedure . . . . . . . . . . . . . . 48

4.2 AQX Parameters in VoIP . . . . . . . . . . . . . . . . . . 494.2.1 Single Variables . . . . . . . . . . . . . . . . . . 504.2.2 Multiple Variables . . . . . . . . . . . . . . . . . 54

4.3 Further Calibration . . . . . . . . . . . . . . . . . . . . . 564.4 Chapter Summary . . . . . . . . . . . . . . . . . . . . . 60

5 The Design to overcomeMTRsMonopoly 635.1 Stakeholders Analysis . . . . . . . . . . . . . . . . . . . . 66

5.1.1 End-users . . . . . . . . . . . . . . . . . . . . . 665.1.2 Regulators and Governmental Entities . . . . . . 675.1.3 M(V)NOs . . . . . . . . . . . . . . . . . . . . . 68

5.2 AbaCUS: Auction-based Charging User-centric System . . 695.2.1 QoS Class (QoS-C) . . . . . . . . . . . . . . . . 715.2.2 TeR Class (TeR-C) . . . . . . . . . . . . . . . . 715.2.3 AbaCUS Bidding Metric . . . . . . . . . . . . . . 725.2.4 AbaCUS Auction and Au² . . . . . . . . . . . . . 73

5.3 AbaCUS Technical Requirements . . . . . . . . . . . . . 735.4 Chapter Summary . . . . . . . . . . . . . . . . . . . . . 74

6 Automatic and On-demandMNO SelectionMechanism 756.1 MNO Selection with AT Commands . . . . . . . . . . . . 776.2 MNO Selection Mechanism . . . . . . . . . . . . . . . . 78

6.2.1 Accessing the Android Internal API . . . . . . . . 786.2.2 Obtaining the Original Android Framework . . . 796.2.3 Creating a Customized Android Framework . . . 796.2.4 Modifying the Eclipse Access Rule . . . . . . . . 80

viii

6.2.5 Invoking the MNO Selection Mechanism . . . . . 816.3 Evaluation of the MNO Selection Mechanism . . . . . . . 82

6.3.1 Time Consumption betweenMNO Switching . . 836.3.2 MNO Switching Energy Consumption . . . . . . 88

6.4 AbaCUS E2E Calling-time . . . . . . . . . . . . . . . . . 896.5 Chapter Summary . . . . . . . . . . . . . . . . . . . . . 91

7 Conclusions 937.1 QoE Formalization . . . . . . . . . . . . . . . . . . . . . 947.2 AbaCUS . . . . . . . . . . . . . . . . . . . . . . . . . . 947.3 Future Work . . . . . . . . . . . . . . . . . . . . . . . . 95

References 97

Other Author Publications 107

Appendix 111A.1 Formal Proofs of AQX Equations . . . . . . . . . . . . . . 111

List of Figures 116

List of Tables 117

Acknowledgments 119

Curriculum Vitae 121

Glossary 123

ix

1Introduction

T he total cost of each call placed by a subscriber of a MobileNetwork Operator (MNO) is split into two parts. The first part de-

termines the amount the caller’s provider is charging to provide the serviceto the calling party. The second part includes the amount that the providerof the callee will charge the caller’s MNO to terminate his call into his net-work. End-users ofmobile voice services have topaybothparts of these ser-vice costs. Thus, either the caller, or the callee will pay for the terminationfees. Figure 1.1 illustrates the present situation in the Mobile TerminationRates (MTRs) ecosystem.

In countries where the Calling Party Pays (CPP) principle is applied,such as in European countries, MNOs subscribers rarely consider the ter-mination cost that their operator is charging other networks, when deliv-ering an incoming call to them, before they establish their contract. Fur-thermore, a significant raise of termination rates fromMNOs will increasethe communication cost for all the MNOs subscribers. However, it willrarely have a negative impact and dissatisfy the customer base of an opera-tor. The customer base of a single MNO is significantly smaller comparedto the set of total customers in every other MNO, Fixed Network Oper-

1

Callee’s MNO Caller’s network

Caller Callee

I  want  to  call    

Termination rate Service rate

Total  cost  of  the  call    =  Service  rate  +  Termina6on  rate  Calling  Party  Pays  (CPP)  principle  →  Total  cost    

Figure 1.1: The MTRs Ecosystem

ator (FNO), or Voice-over-IP (VoIP) provider. Thereby, the majority ofcalls that a MNO has to terminate in his network originate from foreignnetworks.

For many years MNOs revenue is coupled with high termination ratesapplied. Given that fact, the MNOs termination-service as well as MTRs,since the early days of the mobile communications are considered to be a“de-facto”monopoly. Thus, the national telecommunication regulation au-thorities areusually regulatingheavilyMTRsacross theworld. The Interna-tional TelecommunicationUnion (ITU) has put effort toward the solutionof this problem. There is a rich literature on how regulation of thismarket isaffecting it, as well as selected research on the topic of the proper selectionof termination rates . Furthermore, effort has been put into the analysis ofbusiness strategies, which MNOs follow, concerning their MTRs.

However, since the initiation of mobile communications, many issueson mobile terminal devices and network infrastructure have changed. To-

2

day, themobile networks are evolved in a way that their infrastructure doesnot support only voice services, but data as well. Furthermore, the mobiledevices and networks operate in several bands. Last but not least, “today,your cell phone has more computer power than all of NASA back in 1969,when it placed two astronauts on themoon” [62]. In such an environment,this work shows that multipleMNOs could terminate a call and lead to theMTRmonopoly break.

Therefore, this thesis considers those changes and proposes an Auction-based Charging and User-centric System called “AbaCUS”, which over-comes the monopoly obstacle of the MTRs market. The key character-istic of AbaCUS is that the termination service is provided by the MNOwhich maximizes the end-user’s Quality-of-Experience (QoE). MNOs arebidding in an auction and the winning MNO (the one that will maximizethe end-user’s QoE) is selected to provide the mobile call-termination ser-vice. Since QoE is used as a bidding metric in the AbaCUS auction, thisthesis proposes a novel and generic Axiomatic QoE (AQX) model to pre-dict/estimate end-user’s QoE, considering both technical and economicparameters, such as the network access priority and service price respec-tively.

1.1 QoE

In the Internet Technology (IT) ecosystem, end-users QoE is importantinformation needed by Service Providers (SPs) to improve their services.However, end-user-satisfaction, which can be reflected by QoE metrics,cannot be easily measured like technical variables, such as bandwidth andlatency. QoE can either be estimated for a specific service through mathe-matical models, or it can be measured through an experimental setup.

For the VoIP scenario existing models, such as the exponential relation-ship connecting QoS parameters, called Interdependency of the QoE andQoS (IQX) Hypothesis [31] and the E-model [53] of the Telecommu-nication Standardization Sector of the International TelecommunicationUnion (ITU-T) [59],map technical variable values toMeanOpinionScore(MOS). However, those models either can not consider simultaneously

3

multiple parameters, or represent successfully only the average user. Un-fortunately, the average user does not exist!

Measuring the impact of technical variables, or resources priority-access,onQoEof various services, demands an extensive feedback fromend-users,when those variables change. Estimating QoE in a given scenario becomesharder, when non-technical variables, such as price, need to be consideredin addition to technical ones. In any case, detailed feedback that correlatesall variables affecting QoE is needed by end-users for each service sepa-rately. In this thesis the axiomatic mathematical QoE model AQX encap-sulating user demands, user/service characteristics, and variable specifica-tions is proposed to formalize theQoE prediction/estimation, consideringone, or multiple and diverse, variables. Furthermore, the output of QoEfunctions presented here can be normalized such that results will be com-patible with the five-point scaleMOS, proposed by the ITU-T [51], where5 is “Excellent”, 4 is “Good”, 3 is “Fair”, 2 is “Poor”, and 1 is “Bad”.

To tackle the challenge of QoE estimation whenmultiple technical vari-ables are considered, and to reveal the fact that the state of the art models[31] [53] fail tomap accurately in every case the end-user QoE, in this the-sis a VoIPQoEmeasurement setup has been designed and implemented tocapture end-user’s QoE in several VoIP scenarios. The setup used in severalexperimentsweremultiple subjects participated. Thedatameasuredduringthese experiments are used to define all necessary parameters of the AQXmodel in these VoIP scenarios. Such a calibration of the model is essen-tial to adapt it to the particular service and its technical and non-technicalconditions in which it is used. Furthermore, those AQX results achievedare compared with those results of the IQX Hypothesis and the E-Model.Thus, it is finally shown that AQX can capture more accurately end-user’sQoE in VoIP scenarios.

1.2 Charging forQoE

Charging for QoE is a novel approach (a) for IP-based services [87], and(b) forCloudProvider (CP) services selectionmethods. Comparing cloudservices, such as performance cloud servers, load balancers, and file servers

4

offered by different CPs is not trivial. Large CPs, such as Rackspace [84],GoGrid [37], and Amazon Elastic Compute Cloud (EC2) [4], offer com-parable services with slightly different characteristics concerning CentralProcessingUnit (CPU), RandomAccessMemory (RAM), available band-width, Operating System (OS), and charging schemes. There is not alwaysan exact mapping of services across those CPs. Thus, it is hard for the end-user to decide which is the right product andCP to choose, when consider-ingmain characteristics and constraints of a service. Amathematicalmodelthat can generate a score for each product and CP, receiving as an inputminimal service demands and end-user priorities concerning each variable,will be a powerful tool when CPs’ products comparison is requested. AQXis shown to be suitable for CPs service’s-value indexing.

1.3 BreakingMTRsMonopoly

This thesis proposes for the MTRs domain a MNO AQX-bidding processthat will be used to define whichMNOmaximizes the end-user’s QoE, andthus it should be selected to provide the mobile call-termination service.AbaCUS [3] is an approach that enables the competition in the mobiletermination-service market [67] through an auction. In countries wheretheCalling Party Pays (CPP) principle is applied, the person that is dialling(caller) the phone number of a mobile user (callee) has to pay the termina-tion rate that the MNO of the callee demands from the MNO of the callerto deliver the call. AbaCUS proposes a Global System forMobile Commu-nications (GSM) overlay, where the caller can select another MNO oper-ating in the area of the callee to deliver the call. AbaCUS describes in thisthesis a process where the caller will define across a set of variables pref-erences concerning the network access priority, the sound quality and theprice of the service. Thus, in the AbaCUS ecosystem the caller defines hispreference selecting among a predefined set of variables (priority/soundquality and price).

5

1.4 On-demandMNO Selection

Assuming the knowledgeof aMNOthatmaximizes the end-user’sQoEandAbaCUS in place, a manual selection of the MNO to be used is essential.Such an action is possible on a mobile device through the respective userinterface. Furthermore, mobile devices can be adjusted to select automat-ically the MNO based on the strongest signal strength, among the list ofthose MNOs the Subscriber Identity Module (SIM) card is allowed to beregistered with. However, so far inmodernmobile operating systems, suchas Android and iOS, there is no available method in the public developer’sApplication Programming Interface (API), which allows for an automaticand on-demand selection of the MNO by third-party applications.

This thesis presents an automatic andon-demandMNOselectionmech-anism, that has been designed and implemented on the Android platform.For evaluation purposes the energy and end-to-end (E2E) time consump-tion while switching amongMNOs using this mechanism is evaluated andproven to be reasonable for an on-demand MNO selection for the mobilecall-termination service purposes. Thus, this thesis provides valuable inputto (a) AbaCUS, and (b) to the Android developers community.

1.5 Thesis Contributions

Motivated by the above observations, this thesis makes the following con-tributions to the field of QoE, mobile call-termination service, and auto-matic and on-demandMNO selection:

1. introducing AQX, a generic axiomatic QoE model that considerssimultaneously multiple technical and non-technical variables, aswell as user preferences and user/service characteristics to predictand/or estimate QoE in terms of MOS. The model presented inthis thesis is suitable for multiple IT domains, such as IP-based ser-vices, and similar but non identical services ranking. The generatedMOS values can also be ITU-TMOS compliant if needed. Further-more, this thesis is providing a calibration of the proposed model

6

and shows that outperforms state of the art existing models in cer-tain VoIP scenarios. Presenting methodology that show axiomaticways of selecting the respective parameters of the QoE model, andshowing further calibration of the selected parameters for more ac-curate results;

2. using end-user’s QoE maximization and charging for QoE to ter-minate the ‘’de-facto” MTRs monopoly, showing that the involvedstakeholders can benefit from it in respect to the economic footprintof the mobile call-termination service liberation, the infrastructureutilization and with additional instruments to regulator authoritiesthat aim to overcome the mobile call-termination monopoly obsta-cle in the future; and

3. presenting an energy and time efficient automatic and on-demandMNO selection mechanism, and proving that MNO hopping is fea-sible. This finding can be used an an input not only for the MTRsmonopoly cancelation, but in other research approaches assumingthis mechanism, such as electrosmog [93].

This thesis contributes in the following areas: QoE modeling[85][108]; QoE model calibration [104][105]; mobile call-terminationmonopoly termination [107]; and automatic/on-demandMNO selection[102][103]. These contributions do represent important and innovativeadvancements in both research and mobile communications marketmanagement.

1.6 Thesis Outline

The remainder of this thesis is organized as follows. Chapter 2 presents re-lated work on concepts that lay the technical foundation upon which thisthesis stands; that includesQoEmodeling andQoE estimation and predic-tion, Mobile call-termination service policies and charging for QoE in thisdomain and automatic and on-demand MNO selection time and energy“cost”.

7

Chapter 3 defines an axiomaticmathematicQoEmodel calledAQX thatconsiders simultaneously more than one technical and economic variablesaffecting end-users’ QoE. AQX considers end-user’s preferences and ser-vice characteristics, to define influence factors andweights for each variableinvolved in QoE estimation process.

Since AQX is a generic model that can be applied in many scenariosacross the IT domain it needs to be calibrated and evaluated for its per-formance. Thus, Chapter 4 presents how AQX parameters can be selected,how the model performs, and how it can be further calibrated in the well-studied, in respect to QoE, scenario of VoIP.

Chapter 5 introduces the auction-based architecture to overcome themonopoly of the MTRs market (AbaCUS). The stakeholders of this mar-ket and their incentives are briefly presented. The variables affecting thecall-termination service are illustrated, and the mapping of those variablesto QoE is defined.

AbaCUS, among other research approaches, assumes an automatic andon-demand energy and time -efficientMNOselectionmechanismwhich ispresented in Chapter 6. The prototypically implemented MNO selectionmechanism for the Android platform in this thesis, is evaluated in termsof the MNO selection time needed and its energy-efficiency, to prove thatsuch a mechanism is technically feasible. However, this mechanism is awork-around, since the modern mobile OSs do not support such actionsyet.

Finally, Chapter 7 concludes this thesis, summarizing contributions andkey findings, and suggests future work.

8

2RelatedWork

T he QoE concept is relatively young, but there is significantwork already done in this domain [31][86][87][89]. There were

even some attempts to charge for QoE in real life [70]. However, thereis not available so far a generic enough QoE model that can encapsulateseconomic and technical parameters, such as it can be used to charge forQoE in the telecommunication domain. Furthermore, a concrete scenariowhere charging for QoE can be applied in telecommunication is missingfrom the literature. This thesis proposes AbaCUS, a mechanism that usesQoE-awareness for the mobile call-termination service monopoly libera-tion. This chapter presents technologies and models which are impor-tant building blocks of AbaCUS, including (1) QoEmodeling, (2) mobiletermination service approaches, and (3) on-demand MNO selection de-mands.

2.1 QoEModels

QoE is a subjective concept. However, when end-users are asked to rate theperformance of a service in a given scenario, there is a certain alignment onresults. Thus, the quality of a conversationof aVoIP call thatwas performed

9

with a specific codec ismapped to a specificMOS[42]. Furthermore, whilecomparing similar services with different variables, such as conversationson a VoIP network with different codecs, the MOS can be used as a com-parison metric. Thus, an axiomatic MOS calculation, addressing servicesaffected by diverse variables, can be used for service-comparison purposes.

When a comparison between diverse but similar products, such as prod-ucts offered byCPs, is needed the only constant parameter is the end-user’spreference and services’ constraints and demands. Thus, considering thelatest to calculate the expected QoE before concluding the selection of aspecific product is essential. In the CPs scenario, the SLA Support Systemfor Cloud Computing (SLACC) [71] approach suggests the existence ofan SLA negotiation process (cf. Figure 2.1). CPs offer various pre-formedSLAs to end-users. However, end-users might have demands concerningthe price and technical parameters, which are not satisfied by any prede-fined SLA. Thus, during the SLA negotiation process the end-user couldadvertise service demands. The CP could consider end-user’s preferencesinput and try to maximize the estimated QoE while minimizing the SLAviolation probability. Before the CP could make an SLA offer to the end-user, SLACC, which is a decision support system for CPs, would examineif such an SLA can be fulfilled. Thus, a mathematical model like AQX canbe used for the estimated QoEmaximization process.

MNOs bid on the same set of variables. It is not flexible enoughall MNOs to agree, or be forced by the regulator, on offering thesame priority/sound-quality service plans, neither have the same pricingschemes. Thus, a comparison between different MNOs’ bids needs to bepossible. Similar to SLACC, using AQX enables comparing diverse ser-viceswithout identical variables. According to the end-user preferences theMOS for eachMNO is calculated and used as a bidding metric in the Aba-CUS auction mechanism. Furthermore, this thesis shows that MNOs willbenefit by the existence of a mobile call-termination-free market throughthe establishment of Quality-of-Service (QoS)-guaranteed services.

10

Service

CPCU

1. Offers

Pre-Formed SLA

Complete

SLA

Complete

SLA

Fulfilled by

CP

Consult

Process

2. Does not accept

the Service’s SLA

Estimate

SLA parameters

SLA Negotiation

Process

3. Propose specific

SLA parameters4. Consult SLACC and

check the feasibility

Cloud Provider

IT InfrastructureSLA tailored to CU

and CP interests

Complete

SLA

Complete

SLA

Fulfilled CP

& CU

5. Resultant SLA

6. Contract

the service upon

the negotiated SLA

Figure 2.1: SLACC Solution Overview [71]

2.1.1 IQXHypothesis

IQX assumes an exponential relationship betweenQoE andQoS [31][43].QoE = Φ(I1, I2, . . . , In) in [43] is a function of n influence factors Ij, 1 ≤j ≤ n (the equivalent term for influence factor is variable in AQX, sincethe term influence factor used for different purposes in this thesis). Tomotivate the fundamental relationship between QoE and an impairmentfactor (antitonic variable in this thesis) corresponding to QoS, the packetloss probability ploss is examined in the VoIP scenario. Similar to [90]QoE = α · e−β·ploss + γ assumed to be an exponential function. Thus, fittingthe curve of ITU-T-compilant MOS (c.f. Table 2.1) measurements con-cerning the Internet Low Bitrate Codec (iLBC) voice codec (400 bits each30 ms) at [43] provided the numerical values for the parameters α = 3.01,β = 4.473, and γ = 1.065 (cf. Figure 2.2).

11

Table 2.1: The MOS Scheme Recommended by theITU-T [51]

MOSValue Quality5 Excellent4 Good3 Fair2 Poor1 Bad

0 0.2 0.4 0.6 0.8 11

1.5

2

2.5

3

3.5

4

4.5

packet)loss)ratio

mea

n)op

inio

n)sc

ore

measurementexponential:)R)=)0.998f

exp(x))=)3.010⋅exp(−4.473⋅x)+1.065

Figure 2.2: QoE Mapping Function of Packet Loss Ratio in the IQXHypothesis [31]

12

IQX is not intended to be used as a service comparison tool, thus is notpossible to be used for charging for QoE purposes. In this thesis the con-cept of IQX is extended to more than one variable that can also affect QoEnot only negatively but positively as well, so that a high level comparisonamongdifferent services expectedQoEcanbe facilitated. Furthermore, themathematical model that is proposed, provides more flexibility in respectto the influence that a variable, e.g., ploss, might have in different services e.g.,different VoIP codecs, or for different users, e.g., business, and home users.The later is achieved by introducing influence factors for each variable af-fecting theoverallQoE.The influence factor is a poweroperator that applieson variables and shows how fast QoE will be affected at a given fluctuationof each variable. Additionally, the concept of the Expected Variable Value(eV²) for each variable is introduced here. The role of the eV² is fundamen-tal in the selection and calculation of α, β, and γ that result from the curvefitting in IQX. Finally, IQX is extended in AQX by introducing the conceptof the Expected MOS (eMOS), which is a value that is less than the maxi-mumpossibleMOS.The latest facilitates the assumption that a certain levelof QoE can be maintained even if one variable, e.g., ploss, changes.

2.1.2 E-model

The E-Model is a transmission planning tool that can be used to predictQoE for a typical telephone user in an E2E conversational scenario. Themodel takes a wide range of transmission variables into account and it canbe used to assess the voice quality of wired and wireless services, based oncircuit-switched and packet-switched technology [57]. The output of thismodel is— in contrast to other models—not in form ofMOS values. TheE-model uses the Transmission Rating Factor R as output, which can betransformed into MOS and, therefore, it becomes possible to compare theE-model to other models [53], too. The E-Model uses mathematical al-gorithms based on the analysis of a large number of subjective tests witha wide range of transmission variables. These algorithms can transformtransmission variables into so called “impairment factors”. According to

13

the E-model tutorial [57], five impairment factors are used to calculate theR value.

• Ro: Expresses thebasic signal-to-noise ratio, including various noisesources, such as circuit noise and room noise.

• Is: This term takes impairments into account that exist more orless simultaneously with the voice signal, such as, (a) too loudspeech level, non-optimum Overall Loudness Rating (OLR), (b)non-optimum Side Tone Masking Rating (STMR), and (c) impair-ment caused by quantizing distortion.

• Id: This factor represents all impairments that are caused by toolong absolute delay and potential echo effects on both talker’s andlistener’s side.

• Ie: Equipment impairment factor represents impairments that arecaused by the respective codec used and packet-loss.

• A:The advantage, or expectation factor, considers the advantage ofservice access. E.g., a user in a region which is hard to provide con-nectivity, such as regions where a satellite link is needed, expects alower quality, and therefore, tolerates more impairment.

Equation 2.1 considers all impairment factors to calculate the R value[53]:

R = R0 − Is − Id − Ie + A (2.1)

All impairment factors are calculated through algorithms that take sev-eral transmission variables as input. An overview over all variables beingused for the calculation is illustrated in Figure 2.3, where a telephone con-nection and all impairment factors affecting the quality of the conversationaccording to the E-Model is illustrated. As can be seen in Figure 2.3 theE-Model has a high complexity considering many different parameters. Adetailed calculation of each impairment factor can be found in the ITU-Trecommendation G.107 [53]. A question answered in this thesis is if the

14

Figure 2.3: Reference Connection of the E-Model [53]

QoE model proposed here (AQX), which has a comparatively low com-plexity can keep up with the E-Model.

2.2 Mobile Termination RatesMonopoly

Since the early years of mobile communications, the scientific communityas well as regulation authorities has invested a large effort [41][16][24], inorder to reduce negative effects of the termination rates monopoly. How-ever, the attempt to overcome negative effects of this monopoly is focused(a) on charging solutions mainly targeted at the paying party of the termi-nation rate or (b) on regulation rules that need to be enforced by respectiveregulation authorities at operational MNOs. Thus, (1) the CPP principlewith a strong regulator presence, (2) the Receiving Party Pays (RPP) prin-ciple, and (3) a National Roaming (NatRoam) approach, aim to eliminatenegative effects of the monopolistic termination rates market. However, inall cases the monopoly in this market still remains since only the MNO ofthe callee can terminate his calls and profit from it.

15

2.2.1 The CPP Principle

The CPP principle is the most commonly used termination charging ap-proach among MNOs around the world, especially in the European mar-kets [46]. Within CPP principle the caller has to pay call-terminationcharges and there is no contribution from the callee. This principle is theroot of the monopoly problem in the mobile termination rates market.Thereby, strict regulations are applied in order to avoid the MNOs spec-ulation due to their dominant position.

Despite the significantly lower MTRs nowadays compared to the pastdecade, the percentage of MTRs when considering the total call cost ofa call still remains high (e.g., in March 2009 the regulations in India wereamended so that termination rates for all types of domestic call, fixed ormobile, were reduced from the equivalent of some0.006US$perminute to0.004 US $ per minute [47]). Thus, the regulation demand for this markethas not been decreased. However, regulating this market simply reducesthese monopoly’s negative effects, while the heart of the problem is stillbeating, since only the callee’s MNO can terminate his calls and collect theprice for it. Defining and applying regulations is a time-costly procedurefor regulators, which MNOs often use in order to avoid/postpone a newregulation. Thus, a more efficient way to overcome negative effects of amonopolistic/regulated market is essential.

2.2.2 The RPP Principle

InNorth America and some parts of Asia the RPP instead of the CPP prin-ciple is applied. In contrast to the CPP principle, in RPP the callee is askedto pay for the termination cost or in some cases to share a part of this costwith the caller. Initially this approach sounds fair, especially in the scope ofthe callee payment for the call-receiving service, while he is mobile and notlocated in his home network. Furthermore, a subscriber is free to comparetermination rates of each MNO and to make his choice before the estab-lishment of a contract with a MNO.Thereby, the mobile termination ratesmarket seems to allow for competition. However, the question of how acallee could avoid payments for unwanted calls (e.g., advertisements, tele-

16

sales, or polls) is raised. The answer is that it is the callee’s responsibilityto distinguish, which call is important and should be accepted and whichshould be rejected. This is only one of the RPP side effects [68] that fearedto slow-down the mobile sector in the past. The RPP principle may add anextra degree of freedom in the mobile call charges, since the terminationrate is not a part of the total cost that the caller has to pay. However, it isalso adding a considerably bigoverhead for consumers such as theprovider-selection decision, while considering the callee role.

2.2.3 The NatRoam Principle

The NatRoam approach is partially used in countries of Latin America(e.g., TIM Brazil [17]). However, MNOs that offer NatRoam services in-side larger countries (e.g., within Brazil), offer a limited selection freedomamong other MNOs, which are usually branches of the same company.Furthermore, switching to a different MNO is allowed only in deferent re-gions of the country, where the subscriber’sMNOdoes not operate its ownnetwork infrastructure. Additionally, the user that is on NatRoam has alsoto pay for every incoming call, exactly like in the international roaming case.Thus, NatRoam today has no influence on the termination rate market.

Even in case that NatRoamwas broadly allowed, or enforced by the reg-ulator [27], the price that the caller would have to pay in the CPP scenariowould not be influenced by the caller, unless the caller could notify thecallee to switch on a preferableMNO. In theRPP case the calleewould alsohave to establish a contract with each MNO in his location, in order to beable to register his device to any of them. In any case the Number LookupService (NLS) queries, prior to a call (for every call), through the Signal-ing System No. 7 (SS7) network to the Home Location Register (HLR)of the MNO, in order to find the MNO that is currently reaching a Mo-bile Subscriber Integrated Services Digital Network-Number (MSISDN),would be mandatory. This procedure is costly since these NLS costs varyfrom 0.038 to 0.0038 eper look-up [69]. Therefore, an on-demand solu-tion, which addresses the termination rates monopoly, is proposed withinAbaCUS.

17

2.3 FlexibleMNO Selection

Automatic and on-demand MNO selection is nowadays technically possi-ble since there are no physical barriers anymore preventing amobile deviceto be registered to any MNO. This thesis shows that a liberal on-demandMNO selection is not only possible, but also beneficial for the end-users.Thus, the market-defined limitation of bounding the end-user with oneMNO needs to be challenged.

2.3.1 Business Approach

The first attempt towards an on-demand, but not liberal, MNO selectionapproach introduced by Apple in 2006 at [29]. In Apple’s approach theMNO that will provide a service to the en-user such as an outgoing call isdecided by Apple. MNOs are bidding in an auction initiated by Apple whoacts as a Mobile Virtual Network Operator (MVNO).This MVNO selectswhich MNO will provide the service. On one hand, in this scenario theMVNO benefits from selecting the lower price that the actual MNO offersto deliver the service. On the other hand, theMVNOwill charge a standardfee to the end user. In other words, at [29] Apple attempts to patent theright for competition in the MTRs market!

In contrast to [29], that the auction takes place between the MNO andthe MVNO, AbaCUS has two fundamental differences: (a) additionallyto the price, a set of technical preferences, such the network resources ac-cess priority, and the QoS define the MNO that will provide the serviceare considered, and (b) the end-user is the entity that defines those pref-erences. Thus, in the AbaCUS competitive environment is not only theMVNO that might increases profits, but also the end-users that maximizetheir QoE.

2.3.2 Hardware Approach

To select the MNO on-demand a SIM card that can be registered in ev-ery MNO is needed. Apple SIM [114] overcomes this obstacle. However,there is not mechanism that allows to the User Equipment (UE) to regis-

18

ter to another MNO in an automated way, on-demand, and without anyinteraction.Apple SIM supports only 3 MNOs in the US and one in UK.Additionally to the limited number of supportedMNOs by the Apple SIMcard, in the early daysof the commercial useof suchSIMs, oneof themarketleader MNO in the US, had took the decision to permanently lock AppleSIM when activating for first time, requiring the purchase of a new AppleSIM to use a different carrier. AbaCUS presents clear benefits for MNOsas an incentive to participate in the competitive termination servicemarketproposed in this thesis.

2.3.3 FlexibleMNO Selection Use-case

A novel work to make use of an automatic and on-demand network selec-tion is presented at [93]. A network switching selection model and its al-gorithms minimizes the non-ionizing radiation of devices during use. Thekey goal is to minimize the exposure of the mobile user to electromagneticradiations, while still providing a certain QoS level. Within a proof of con-cept [93] validated the model and its algorithms. Due to the fact that theAndroid Application Programming Interface (API) does not provide fora mechanism to force switching from one MNO to another, the user hasto manually select a network. This takes a lot of time, because the pro-vided mechanism by the Android platform searches first for all availablenetworks, which is a time consuming operation. This time overheadmakesit impractical to apply a MNO selection algorithm.

The available MNOs in a country are well known and do not changeoften. Thus, to avoid the MNO searching delay in the prototyped mecha-nism that was implemented in this thesis, this operation is skipped whilea MNO is selected. Available MNOs in a location are stored in a list andwhen needed the respective MNO is selected from that list. Neverthelessperiodic updates of that list, e.g., daily, when an application starts, or whena user moves in a predefined area, are essential to ensure that all currentlyavailable MNOs are stored in the list. Thus, the proposed MNO selectionmechanism in this thesis here can be used for the non-ionizing radiationminimization purpose as well. The evaluation of this mechanism, in terms

19

of energy and time consumption per MNO switching can define a thresh-old of a maximum number of hops allowed in the non-ionizing radiationminimizationMNO selection approach, so that the electrosmog reductionapproach remains both realistic and energy efficient.

2.4 ContributionOpportunities

While thefieldofQoEandmobile communications has received significantattention from the research community, the related work research presentin this chapter has revealed the following:

1. encapsulate simultaneously, multiple technical and non TechnicalParameters in QoEmodels is missing;

2. an architecture that breaks the MTR monopoly without creatingnew monopolistic markets is not in place; and

3. a Mechanism that enables automatic and on-demand MNO selec-tion in a time and energy -efficient manner is needed to achieve theMTRmonopoly breaking.

Having identified the above shortcomings, and in direct relationshipwith the four observationsmade in Section 1.5, the following opportunitiesfor scientific contribution in the area of QoE and mobile communicationshave been revealed:

1. create a generic axiomatic QoEmodel that encapsulate multiple anddiverse parameters. The outcome of the model can be used to facili-tate charging for QoE;

2. define a methodology of selecting the parameters of the QoEmodeland identify further potential improvements; and

3. present an architecture that can overcome the MTR monopoly, aswell as a time and energy -efficient automatic and on-demandMNOselection mechanism.

20

Hence, chapters 3, 4, 5, and 6 present a detailed description of the abovecontributions.

21

22

3AQX: Axiomatic QoEModel

Q uality-of-Experience is a user-centric concept reflectingthe end-user satisfaction of a service while considering various tech-

nical variables, such as latency, bandwidth, or jitter, in VoIP services of thetelecommunication field [44] or in video streaming of the entertainmentfield [19][72][111]. Furthermore, theQoE concept can also be usedwhenconsidering pricing for IP-based services [65][87][89], since a price of aservice affects the overall end-user experience. Thus, QoE can be affectedby (a) diverse technical variables and (b) by economical/non-technicalvariables.

In the IT ecosystem such variables are usually defined in the SLA be-tween the SP and its customer. When one ormore of these variables do notmeet the agreed level, an SLA violation is occurred. However, an SLA vi-olation does not mean that the end-user dissatisfaction cannot be avoided.There are certain actions that a SP can take, such as offering the serviceat a lower price, or offering a service upgrade, such as a higher bandwidthfor the same price, to maintain the QoE of an end-user at a certain level ofsatisfaction. To prevent a potential decrement of QoE in case of an SLAviolation, it is important to know which variables and how exactly they af-fect the end-user’s QoE. A proper adjustment of involved variable(s) on

23

the QoE, might counterbalance the incident that caused the SLA violationin terms of the end-user satisfaction. However, such a process assumes aformal complete and generic overview of the QoE concept that is missingnowadays. The need to illustrate QoE contributed to the creation of stan-dards, such as the MOS [55][51][56]. TheMOS reflects the end-user sat-isfaction at a numerical scale where the higher the score is the higher theend-user’s satisfaction is and vice versa. However, since theMOS defines asubjective value, a complete and formal calculation of theMOS while con-sidering all variables that might affect the QoE is the missing piece towardthe precise user satisfaction demands estimation.

This thesis formalizes QoE in AQX which considers the agreed valuesof a set of variables described in the SLA and the measured values of thosevariables when/while a service is provided. Such measurable informationthat defines the QoS is used to calculate aMOS-normalized value that rep-resents the end-user’s QoE.Thus, the definition of an SLA violation in thisthesis is the situation where there is no possible adjustment for any vari-able(s) which can result into an agreed uponMOS score between the end-user and the SP. QoE calculation equations in this thesis here can be usedto export MOS results once the set of the QoEmodel parameters and vari-ables is defined. KeyQoEparameters are: (1) theminimumandmaximumvalues of a variable (e.g., price, bandwidth, or latency) affecting QoE; (2)the expected, or agreed in the SLA values of each variable; (3) the impor-tance and the influence factor of each variable for each service; and (4) thedesired codomain of theQoE functions. Below are summarized the axiomsof the AQXmodel.

1. To predict/estimate QoE, the first action needed is to identify allvariables that affect QoE and can be measured, or can be estimated.

2. There are two types of variables: (a) Isotonic Variables (IVs). Themore you have the better it is (e.g., bandwidth). (b) Antitonic Vari-ables (AVs). The more you have the worst it is (e.g., price).

3. For each service there are two values per variable that define theworst and the best possible values.

24

4. For each service and end-user there is anideal/desired/expected/agreed value of each variable (eV²)that shows that the end-user is satisfied. This value is between thebest and the worst variable values. However, in some cases theend-user can be satisfied even more (e.g., further discount).

5. The fluctuation of the value of each variable might affect differentlyeach end-user at a given service. TheQoE affect of the fluctuation isexpressed via the influence factor of a variable. The influence factor(m) can be different for values above and below the eV².

6. Each variable affecting QoE can have different importance (w).

3.1 QoE Formalization in AQX

A precise QoE formalization demands amathematical model that is able toconsider multiple and diverse variables, such as priority, price, and band-width that can affect the end-user QoE positively or negatively on a givensituation. Furthermore, each variablemight affectQoE in a different way ineach scenario. Additionally, QoE strongly depends on the end-user sinceeach person might have different demands and priorities concerning thesame services. The high-level formalization of the QoE is illustrated below.

QoE := f(user, service, variables, mood, context) (3.1)

WhenQoE formalization is neededQoE should be treated as a boundedconcept since a user cannot be infinitely satisfied or dissatisfied. E.g., dou-bling the price of a service that is not affordable, will not satisfy less theend-user since the service was already not accessible due to high cost. Ad-ditionally, doubling the bandwidth of an average broadband plan withoutincreasing the price, will not affect QoE proportionally, since services suchas browsing, video streaming, or VoIP, perform equally well without sucha bandwidth increment. Thus, it is assumed that there is a minimum μ anda maximum M QoE that can be represented, without any loss of general-ity, by the positive parameters, μ,M ∈ (0,∞), where μ,M ̸= ∞. Since

25

μ < M, let h be a positive parameter that represents the size of the QoEinterval scale (cf. Equation 3.2).

h = M− μ > 0 (3.2)

Figure 3.1 illustratesQoEof the end-user for twodifferent variables. TheY-axis shows theMOS of a variable in the interval h and theX-axis the nor-malized value x of each variable. The value e0 on the Y-axis is the MOSthat corresponds to the expected, agreed, or defined in the SLA value x0 ofeach variable. Thus, let e0 be the eMOS and x0 the eV² (cf. Section 2.1.1).On one hand, the ei curve reflects the MOS of a variable, such as band-width. Such variables while increasing to a maximum value xmax, imply aQoE increment to the maximum MOS value M. Those variables in AQXare termed Isotonic Variables (IVs). Furthermore, when the value of anIV is minimum xmin, the MOS value is also minimum (μ). On the otherhand, the ea curve reflects the MOS of a variable, such as the price of a ser-vice. Such variables, contrary to IVs, while increasing to a maximum valuexmax, imply a QoE decrement to theminimumMOS μ. Those variables aretermed Antitonic Variables (AVs). Last but not least, when the value of aAV is xmin theMOS isM. E.g., when a service is provided without chargingthe QoE is maximum (cf. Equations 3.3 and Equations 3.4).

ei(xmin) := μ

ei(xmax) := M(3.3)

ea(xmin) := M

ea(xmax) := μ(3.4)

The IV or AV characterization of a variable describes the nature of theargument variables in Equation 3.1. The xmin and xmax values are related tothe argument service in the same equation. E.g., the maximum throughputattainable is governed by the mobile link technology. The average sectorthroughput in Long Term Evolution (LTE) multiple-input and multiple-output (MIMO) 4x4 with 20 MHz bandwidth, the most deployed formof LTE, provides a maximum of 12.7 Mbps uplink and 50.1 Mbps down-

26

link throughput [73]. Thus in this case the minimum (min) andmaximum(max) values concerning the uplink bandwidth (ULB) of the LTEMIMO4x4 2MHz technology are: ULBmin = 0Mbps andULBmax = 12.7 Mbps.Similarly, for the downlink bandwidth (DLB) themin andmax values are:DLBmin = 0 Mbps and DLBmax = 50.1 Mbps. ULBmin = DLBmin = 0Mbps because this means no connectivity.

Finally, x0 (xmin < x0 < xmax), is a value concerning a variable thataffects QoE, and reflects the effect of the argument service in the high-levelQoE Equation 3.1. The value x0 of a variable might not necessarily be re-lated to the technology. E.g., Hulu, a website and over-the-top (OTT) sub-scription service offering ad-supported on-demand video-streaming, rec-ommends a downstream throughput of at least 1.5 Mbps for smooth play-back experience of Standard Definition (SD) 480p videos [45]. Thus, thevalue x0 for bandwidth in this case is defined by the service minimum de-mands and not by the broadband technology (e.g., LTE, or xDSL).

3.2 Specific QoE Functions

Without any loss of generality unity-based normalization of IVs values x isassumed (cf. Equation 3.5), to enable plotting the MOS of multiple valu-ables in one graph.

x :=x− xmin

xmax − xmin∀ x ∈ R (3.5)

3.2.1 IVs QoE Function

The function [0, 1] ei→ [μ,M], (cf. Equation 3.6) is defined to correspondto the QoE of an IV, such as bandwidth. The ei function that is illustratedin Figure 3.2 behaves like a step function for large values ofm, like an expo-nential function for m = 1, and like a constant function for m ≪ 1. Fur-thermore, for m ∈ (1, 3], ei changes in a smooth way. The influence factorm ∈ [0,∞) of a variable denotes the end-user’s tolerance in fluctuations ofvariable’s value x. E.g., services that demand not less than a specific amount

27

Value [x]

MOS[e

i∨a(x)]

h=

M−

µ

xmin

x0

xmax

µ

e0

M

ei

ea

Figure 3.1: MOS Evolution for IVs (ei) and AVs (ea)

of bandwidth will see a high influence factor, while if the bandwidth doesnot affect QoE a lot, the influence factor will be low. Finally, parameters hand μ of Equation 3.6 result due to Equation 3.2 and Equations 3.3. Thus,Equation 3.6 is defined (a) to satisfy AQX axioms, and (b) to be able tomodel multiple QoE curves that behave almost as constant, step, or expo-nential functions.

ei (x) := h ·(1− e−λ·xm)+ μ (3.6)

The Expected Variable Value (eV²) x0 (0 < x0 < 1) corresponds toan en Expected MOS (eMOS) e0, where μ < e0 < M. Thus, given this

28

assumption the parameter λ is presented in Equation 3.7.

ei (x0) := e0(3.6)=⇒

⇒ e0 = h ·(1− e−λ·xm0

)+ μ ⇐⇒

⇔ e0 − μh

= 1− e−λ·xm0 ⇐⇒

⇔ h− e0 + μh

= e−λ·xm0 ⇐⇒

⇔ ln(h− e0 + μ

h

)= −λ · xm0 ⇐⇒

⇔ λ = x−m0 ln

(h

h− e0 + μ

)(3.7)

Replacing λ above inEquation3.6 results Equation3.8, which is theQoEfunction that is used for MOS calculations of an IV (cf. Figure 3.2).

(3.6, 3.7) ⇒ ei (x) = h ·(1− e−

(xx0

)m·ln

(h

h−e0+μ

))+ μ (3.8)

3.2.2 AVs QoE Function

Following the principles of Equation 3.6, the function [0, 1] ea→ [μ,M],m ∈[0,∞) (cf. Equation3.9) is defined to correspond to theQoEof anAV(e.g.,price). The ea function is illustrated in Figure 3.3. Similarly to ei, valuesx are unity-based normalized to enable multiple variables plotting in onegraph. Furthermore, ej behaves also like a step function for large values ofthe influence factor m, like an exponential function for m = 1, and like aconstant function for m ≪ 1. Finally, for m ∈ (1, 3] ea also changes in asmooth way.

ea (x) := h · e−λ·xm + μ (3.9)

In the AV case the eV² x0 (0 < x0 < 1) results an eMOS e0 (μ < e0 <

M). Thus, λ is presented in Equation 3.10, with the formal proof be in the

29

Value [x ]

MOS[e

i(x)]

0 x0 1

µ

e0

M

m = 1m = 10−3

m = 103

Figure 3.2: Plot for Different m Values of Equation 3.8

Appendix.

ea (x0) := e0(3.9)=⇒

... (formally proven in the Appendix A.1)

⇒ λ = x−m0 ln

(h

e0 − μ

)(3.10)

Replacing λ above in Equation 3.9 results Equation 3.11, which is theQoE function that is used for MOS calculations of a AV (cf. Figure 3.3).

(3.9, 3.10) ⇒ ea (x) = h · e−(

xx0

)m·ln

(h

e0−μ

)+ μ (3.11)

30

Value [x]

MOS[e

a(x)]

0 x0

1µ

e0

M

m = 3m = 10−3

m = 103

Figure 3.3: Plot for Different m Values of Equation 3.11

3.3 Influence Factors Calculation

Selecting different influence factors m− and m+ for a value of a variableabove and below x0 represents the flexibility of diverse end-user’s prefer-ences on different scenarios. E.g., increasing the price α > 0 of a productto α = α+ξ for ξ > 0 can effect differently end-user’sQoE than decreasing

31

Value [x ]

MOS[e

i(x)]

0 x0 1

µ

e0

M

e i

x 0 − δ ′ x 0 + δ

e0 − ǫ ′

e0 + ǫ

Figure 3.4: MOS Change for IVs

the price of the same product to α = α− ξ. Thus, Equation 3.12 shows thatthe influence factorm is not necessarily constant.

m :=

m− > 0 for x < x00 for x = x0m+ > 0 for x > x0

(3.12)

3.3.1 IVs Influence Factors Calculation

Equation 3.13 calculates m+ if x = x0 + δ ≤ 1 for δ > 0 when the MOSof an IV is calculated (IVs case). Figure 3.4 shows that e(x) = e0 + ε < Mfor ε > 0. Thus, the appropriate influence factor can be selected, if the

32

impact of a specific value change on QoE is known. E.g, the quality of avideo streaming session will drop fromHigh Definition (HD) to SD if thebandwidth will drop from from 3 Mbps to 1.5 Mbps [45]. QoE decrementwill follow accordingly.

ei (x0 + δ) = e0 + ε(3.6,3.12)⇐⇒

⇔ h ·(1− e−λ·(x0+δ)m

+)+ μ = e0 + ε ⇐⇒

⇔ 1− e−λ·(x0+δ)m+

=e0 − μ + ε

h⇐⇒

⇔ e−λ·(x0+δ)m+

=h− e0 + μ − ε

h(3.7)⇐⇒

⇔ e−(

x0+δx0

)m+·ln

(h

h−e0+μ

)=

h− e0 + μ − εh

(3.2)⇐⇒

⇔ eln(

h−e0+μh

)( x0+δx0 )

m+

=M− e0 − ε

h⇐⇒

⇔ h− e0 + μh

(x0+δx0

)m+

=M− e0 − ε

h(3.2)⇐⇒

⇔ M− e0h

(x0+δx0

)m+

=M− e0 − ε

h⇐⇒

⇔(x0 + δx0

)m+

= log(M−e0h )

(M− e0 − ε

h

)⇐⇒

⇔ m+ = log( x0+δx0

) [log(M−e0h )

(M− e0 − ε

h

)](3.13)

Thebase and the argument of each logarithm inEquation 3.13 is positiveand different than 1, because (a) 0 < x0+δ

x0 < 1, (b) 0 < M−e0h < 1, and

(c) 0 < M−e0−εh < 1. Thus, Equation 3.13 after changing the logarithm to

the natural logarithm is rewritten below (any logarithmwith a positive basedifferent than 1 can be selected for the logarithm base change). A carefulselection of logarithms’ base can simplify calculations e.g., logαα = 1 where

33

0 < α ̸= 1.

(3.13) ⇒m+ =

ln(

ln M−e0−εh

ln M−e0h

)ln x0+δ

x0

(3.14)

Similarly to the previous case, when x = x0 − δ′ ≥ 0 for δ′ > 0 ande0 − ε′ ≥ μ for ε′ > 0 (cf. Figure 3.4),m− is seen in Equation 3.15 below,with the formal proof be in the appendix (A.1).

m− =

ln(

ln M−e0+ε′h

ln M−e0h

)ln x0−δ′

x0

(3.15)

Equation 3.13 and Equation 3.15 present the m+ and m− calculationwhen x = x0 + δ and x = x0 − δ′ respectively. However, if the influ-ence factor of an IVmi,x between any pair of (x1, x2) and (ei(x1), ei(x2)) =(ei,1, ei,2) for 0 < x1 < x2 < 1 is needed, Equation 3.18 can be used.

To calculate mi,x, a relation between x1, x2, and ei,1, ei,2 is needed. Thusthe following calculations are done:

ei(x1) = ei,1(3.6)⇐⇒ −λ · x1m = ln

h− ei,1 + μh

(3.16)

Similarly for x = x2.

ei(x2) = ei,2(3.6)⇐⇒ −λ · x2m = ln

h− ei,2 + μh

(3.17)

Dividing Equation 3.16 by Equation 3.17, and since 1 ̸= x1/x2 ̸= 0:

(3.16)(3.17)

⇔(x1x2

)mi,x

=ln h−ei,1+μ

h

ln h−ei,2+μh

⇐⇒

⇔ mi,x = log( x1x2

)[ln h−ei,1+μ

h

ln h−ei,2+μh

]⇐⇒

34

⇔ mi,x =

ln(

lnM−ei,1

h

lnM−ei,2

h

)ln x1

x2(3.18)

Value [x]

MOS[e

a(x)]

0 x0

1µ

e0

M

x0 − δ′ x0 + δ

e0 − ǫ′

e0 + ǫ

ea

Figure 3.5: MOS Change for AVs

3.3.2 AVs Influence Factors Calculation

Symmetrical steps as in the IVs case are taken when considering AVs (cf.Figure 3.5). Furthermore, symmetrical assumptions are done concerningδ, δ′, x0 + δ, x0 − δ′ and ε, ε′, e0 + ε, e0 − ε′, so that the flexibility onlogarithms’ base selection applies also in the AVs case. Therefore Equation3.19, Equation 3.20, and Equation 3.21 are obtained.

For x = x0 + δ ≤ 1, where δ > 0 and e0 − ε′ > μ, where ε′ > 0 (cf.Figure 3.5),m+ is shown in Equation 3.19 below, which is formally proved

35

be in the Appendix.

ea (x0 + δ) = e0 − ε′(3.9,3.12)⇐⇒

... (formally proven in the Appendix A.1)

⇔ m+ = log( x0+δx0

) [log( e0−μh )

(e0 − μ − ε′

h

)](3.19)

Changing all the logarithms in Equation 3.19 to the natural logarithms isallowed since all the bases and arguments are positive and not equal to 1.Thus, m+ is seen in Equation 3.20. Similar to the IVs case any logarithmwith a positive base different than 1 can be selected for the logarithm basechange.

(3.19) ⇒m+ =

ln(

ln e0−μ−ε′h

ln e0−μh

)ln x0+δ

x0

(3.20)

For x = x0 − δ′ ≥ 0 where δ′ > 0 and e0 + ε ≤ M for ε > 0 (cf. Figure3.5),m− is seen in Equation 3.21 below, which is formally proved be in theAppendix.

ea(x0 − δ′

)= e0 + ε

(3.9,3.12)⇐⇒... (formally proven in the Appendix A.1)

⇔ m− =

ln(

ln e0−μ+εh

ln e0−μh

)ln x0−δ′

x0

(3.21)

Similar to Equation 3.18, any calculation of m = ma,x for AV variablescan be done for any pair of (x1, x2) and (ea(x1), ea(x2)) = (ea,1, ea,2) for0 ≥ x1 < x2 ≤ 1 (cf. Equation 3.24).

36

To calculatema,x, a relation between x1, x2, and ea,1, ea,2 is needed. Thusthe following calculations are done:

ea(x1) = ea,1(3.9)⇐⇒ −λ · x1m = ln

ea,1 − μh

(3.22)

Similarly for x = x2.

ea(x2) = ea,2(3.9)⇐⇒ −λ · x2m = ln

ea,2 − μh

(3.23)

Dividing Equation 3.22 by Equation 3.23, and since 1 ̸= x1/x2 ̸= 0:

(3.22)(3.23)

Appendix A.1=========⇒ ma,x =

ln(

ln ea,1−μh

ln ea,2−μh

)ln x1

x2(3.24)

3.4 Generic QoE Functions

Each service’s unique characteristics, define a unique affect of QoEwhen some variable(s) are changing. Thus, combining a set X =

{x1, . . . , xk, . . . , xN} of N ∈ N+ diverse variables values xk, to calculatea generic MOS, demands weights wk for each variable, since their impor-tance might be different for different services. Equation 3.25 defines thegeneric MOS function E (X). Weights wk ∈ R+ reflect the contributionof all variables. As a starting point the selection of wk = 1 is made, sinceall variables consider to equally contribute in QoE. However, wk is an ad-ditional degree of freedom that considers the diverse importance of eachvariable. Thus, wk is used to calibrate AQX. For this purpose input fromend-users can be used to extract those values of wk that reflect better QoEas reported from end-users.

Figure 3.6 illustrates the generic MOS function E(X) of a hypotheticalservice where one IV (x1) and one AV (x2) (X = {x1, x2}) with an influ-ence factor m = 1 and m = 3, ∀x1, x2 respectively, affect the QoE. In thisexample the contribution weight of both parameters is selected to be 50%to show what is the QoE effect of an equal percent fluctuation of each pa-rameter. The white area on the graph marks all the possible pairs of both

37

Figure 3.6: Generic MOS Evolution for Equally Participating IV (m = 1)and AV (m = 3)

variables values that results an eMOS E(X0 = {x1,0, x2,0}) = E0.

E (X) := μ+ h ·N∏k=1

[e(i∨a) (xk)− μ

h

]wk

(3.25)

The generic MOS in Equation 3.25 is chosen to be a weighted productof all variable-specificMOS’ instead of a weighted sum. The reason for thatis to ensure that if one variable’s MOS is very low, and cannot be compen-sated by an improvement of other variables, the genericMOSwill reflect it.In case of a weighted sum of each variable’s MOS the generic MOS underspecific circumstances can still be equal to the eMOS E0 even if the MOSof a specific variable is the minimum one μ.

38

Assume the following scenario where a MNO is offering a flat rate mo-bile data plan, and twovariables (bandwidth (IV) andprice (AV)) affectingequally the QoE of end-users. This is a realistic scenario since an end-userwhen selection such a service across different MNOs with similar networkcoverage, the standard case for urban areas, can compare only those twovariables. Due to technical problems the MNO is unable to provide theservice; considering only the bandwidth the MOS is minimum (μ) sincethere is no data connectivity. The MNO decides not to charge customersfor the service during the non-functional period; considering only the pricethe MOS is maximum (M) since there is no cost for the service. If thegenericMOSwould be an equal weighted sumof the respectiveMOS’, thenE = M/2 + μ/2. ForM = 5, μ = 1, and E0 = 3 the generic MOS wouldbe E = 3 = E0. However, the overall end-users’ QoE despite the fact nopayment is needed should be lower than the eMOS because no service isprovided. Thus, the low credibility of theMNO should also be reflected inthe generic MOS. Equation 3.25 states the necessity of an acceptable levelfor each variable affecting QoE.Thus, in the scenario mentioned above thegeneric MOS using Equation 3.25 would be E = 1 + 0 = 1, which meansthat there is no price that a MNO can offer to maintain end-users’ QoEconcerning the mobile data service if the service is not available. The latestresult illustrates that the availability of a service, which in this scenario is en-capsulated in the bandwidth variable, is an important parameter. Thus, theend-users’ dissatisfaction and theMNO’s credibility decrement is reflectedby the genericMOS for the service unavailability scenario described above.

3.5 ITU-TMOS-Compilant QoE Functions

The ITU-T has defined in recommendations P.800 [51], P.800.1 [55], andP.805 [56] a five-point scale that represents QoE of end-users. The ITU-TMOS scale is summarized in Table 2.1. In AQXmodel, the eMOS e0 whenthe value of a variable x is equal to the eV² x0 selected to be equal with theITU-T numerical representation of “Good” QoE. Thus, Equation (3.26)

39

illustrates the ITU-TMOS-compliant maximumM, expected e0, and min-imum μMOS values.

M := 5

μ := 1

}(3.2)=⇒ h = 4 and e0 := 4 (3.26)

Given this input, the ITU-T MOS-compliant equation for IVs and theinfluence factorm equations, as presented in Section 3.2 and Section 3.3 re-spectively, can be seen in Equation 3.27, Equation 3.28, and Equation 3.29.Similarly for AVs, the same results are formalized in Equation 3.30, Equa-tion 3.31, and Equation 3.32. Those equations can be used to express theQoE in the ITU-T standardize five point scale e.g. when considering tele-phony services.

(3.8, 3.14, 3.15, 3.26) ⇒ ei (x) = 4 ·(1− e−

(xx0

)m·ln 4

)+ 1 (3.27)

(3.27) ⇒ m− =

ln(

ln 1+ε4

− ln 4

)ln x0−δ

x0

(3.28)

(3.27) ⇒ m+ =

ln(

ln 1−ε4

− ln 4

)ln x0+δ

x0

(3.29)

(3.11, 3.20, 3.21, 3.26) ⇒ ea (x) = 4 · e−(

xx0

)m·ln 4/3

+ 1 (3.30)

(3.30) ⇒ m− =

ln(

ln 3+ε4

ln 3/4

)ln x0−δ

x0

(3.31)

40

(3.30) ⇒ m+ =

ln(

ln 3−ε4

ln 3/4

)ln x0+δ

x0

(3.32)

Finally, Equation 3.33 generates the generic ITU-T MOS-compliantMOS E(X) considering a set X of IVs’ and AVs’ values. Such an method isused at [106] to calculate theMOS of various services onmobile networkssuch as VoIP, video streaming, browsing, and random flow data streaming.

(3.25, 3.26) ⇒ E (X) = 1+ 4 ·N∏k=1

[e(i∨a) (xk)− 1

4

]wk

(3.33)

Table 3.1: Broadband Plans for Home-users of a Swiss ISP [18]

Uplink bandwidth Downlink bandwidth Price permonth[Mbit/s] [Mbit/s] [CHF]

15 250 8910 125 695 50 592 20 450.2 2 0

3.6 Simultaneous Economic and Technical ParametersConsideration Example

The scenario of an Internet Service Provider (ISP) is selected to presenthow the parameters of the AQXmodel should be selected when economicparameters, in addition to technical ones, are considered in QoE. Table 3.1summarizes broadband plans that are offered by an ISP in Switzerland[18]. In this scenario there are three variables that affect QoE. Those vari-ables are: (1) uplink bandwidth u ∈ [0.2, 15], (2) downlink bandwidthd ∈ [2, 250], and (3) price p ∈ [0, 89] of broadband plans. Thus the set ofvariables affectingQoE isX = {u, d, p}. On one hand, both u and d followthe rule “themore you have the better it is”. On the other hand, p is better if

41

it is low. So, u and d are considered to be IVs and Equation 3.27will be usedfor MOS calculations concerning those variables. Price p is considered tobeAVandEquation3.30 shouldbeused to calculate theprice-relatedMOS.

To export the eV² for each variable assume a customer that selected theplan offering 5 Mbit/s uplink and 50 Mbit/s uplink, for 59 CHF permonth.Thus fromEquation 3.5 derivesu0 = 5−0.2

15−0.2 = 0.324, d0 = 50−2250−2 = 0.194,

and p0 = 59−089−0 = 0.663. To calculate influence factors m assume that

50% lower than u0 and d0 values would drop the MOS concerning u and dfrom 4 (Good) to 2 (Poor). Such information can be extracted in this caseby observing when end-users report to their ISP that their broadband con-nection is underperforming, thus they get dissatisfied because of it. Thus,Equation 3.34 is used to calculatem− in this example. Following a similarway of thinkingm+ factor can be calculated using the Equation 3.29.

Concerning the QoE estimation because of the variable p the Equa-tion 3.30 has to be used. However, in this case, the influence factors m−

andm+ canbe calculatedwhile considering the percent discount and incre-ment of the price that would satisfy and dissatisfy respectively an end-userand using Equation 3.31 and Equation 3.32 respectively. Finally, weightparameters wu = wd = wp = 1 are assumed to be equal to reflect equalimportance of all QoE-related variables for end-users. However, weight pa-rameters can be adjusted to calibrate AQX, when end-users QoE feedback,in scenarios wheremore than one variables change simultaneously, is avail-able.

(3.28) ⇒ m− =

ln(

ln 1+24

− ln 4

)ln 50

100= 2.27 (3.34)

3.7 Chapter Summary

AQX contrary to a generic exponential QoE model proposes an axiomaticmathematical model that can be used to generate a MOS illustrating end-user QoE, considering one or multiple and diverse variables, such as band-width, network access priority, or price. The model requires a minimum

42

and maximum satisfaction score in the positive numbers space and a valuebetween thosenumbers to represents the end-user’s satisfaction, wheneachvariable has the desired/expected value, such as the minimum requiredbandwidth needed to achieve HD video streaming quality. Furthermore,for each variable a positive number is needed showing the effect of the pa-rameter’s fluctuation in QoE. This influence factor can be calculated fromthose equations that are provided here. In case of more than one partici-pating variable in theMOS calculation the weight of each variable needs tobe specified. Additionally, AQX is aligned with the comprehensive frame-work for QoE and User Behavior (UB) modeling presented at [85], sincethe reference value for each parameter can be specified as a parameter inAQX.E.g., maximum latency a specific end-user can tolerate at a givenVoIPscenario.

Concluding, such a model presented here is very well suited (a) to cal-culate the MOS provided the value of measurable variables and (b) to es-timate which actions are needed in case of a change, of one or more vari-able(s’) value(s), to maintain the same end-user’s satisfaction level if pos-sible. Having such a method in place can save the time needed to performa survey to observe end-user behavior. Also, the MOS generation mathe-matical model presented here, can be used to generate a comparison indexon similar services for the end-user, across different service providers, suchas CPs, with different variables and values.

AQX is beneficial for end-uses when comparing different services acrossmultiple SPs. SPs can also estimate the end-user’s QoE with such a mathe-matical model while considering (a) their available resources, (b) demandsof the end-user, and (c) minimum service requirements. Furthermore,since this model can consider variables such as the price of a service, SPscan maximize their revenues, while maintaining their customer’s QoE at acertain level. Thus, it is essential for SPs to create end-user and service pro-files by calculating all parameters of this model. Such a calculation can beinitiated through end-user’s demands (e.g., maximum price for a service),and/or service’s characteristics (e.g., maximum latency allowed on a VoIPsession).

43

44

4AQX in VoIP

Q uality-of-Service (QoS) is defined application-specifically by avalue threshold of technical variables such as latency, packet loss,

and bandwidth. These values are well known for different technologies andservices and they can be measured [56]. Furthermore, selected values ofthose variables are often used for marketing purposes, e.g., MNO and ISPadvertise “highbandwidth” or“highperformance”. However,QoSvariablesare not explicitly linked to the end-user’s satisfaction. It is naive to con-clude that end-users’ QoE can be increased by adjusting one QoS variable,because the relationship between QoS variables and end-users’ experiencedepends on the Type-of-Service (ToS). Large latency can serve as an ex-ample here, since latency has a higher negative effect on VoIP services thanon video streaming [52].

Section 3.6 uses an example of an ISP to illustrate amethodology of eco-nomic and technical AQX parameters selection. However, there is no wayto evaluate those results without having access to the ISP customer-care-related data. Unfortunately, ISPs have no incentive on revealing such in-formation. Thus, it is essential to examine the methodology presented fora well studied example. Therefore, this chapter here is focused on definingthe axiomatic relationship betweenQoS variables andQoE in theVoIP sce-

45

nario. The four step QoE formalization methodology in this chapter readsas follows: (1) Define an experimental setup allowing for the emulation ofvarious network connection performance settings on jitter, latency, packetloss, and bandwidth; (2) perform test VoIP calls in pre-defined experi-mental setups and collect QoE-related feedback from end-users in termsof MOS [55]; (3) use the feedback collected to determine through non-linear regression the AQX model parameters for different variables in thisVoIP scenario; and (4) compare AQX [108] to the two QoE-predictingmodels, the exponential relationship connecting QoS parameters, calledIQX hypothesis [31], and the E-model [53] of the ITU-T.

In support of the accurately and timely measurement of QoE for VoIPservices a Web Real-Time Communications (WebRTC) VoIP client wasnewly designed and developed such that it collects directly all necessaryuser feedback from experimental VoIP calls under different network condi-tions in various scenarios. Those varying network conditions are emulatedby the network emulation framework WANem [98], which utilizes a realnetwork. Therefore, three computers were attached to each other througha switch via LAN cables. Using such an experimental architecture guaran-tees a fully controlled network emulation that is not influenced by externaltraffic.

This experimental setup served for the collection of more than 500 datapoints and was used to show how accurate the previously mentionedQoE-predicting models [31][53] reflect these collected data points. In this ex-periment, it was shown that the AQXmodel is the most accurate model tocapture QoE in given scenarios. Moreover, there exist by now two addi-tional evaluations of the AQXmodel concerning (a) the influence factor ofa variable that affects QoE and (b) the proposed equation that estimatesQoE, when multiple variables are considered simultaneously.

4.1 Experimental Setup

The experimental setup is based on the WANem framework for networkemulation and a WebRTC messenger that was developed and used for the

46

Figure 4.1: Experimental Setup

experiments of this thesis. Moreover the ITU recommendation P.800 [51]was considered for the experimental procedure.

4.1.1 Architecture H/W and S/W

Akey element of theQoEevaluation architecture used in this thesis is aWe-bRTCmessenger. WebRTC is World WideWeb Consortium (W3C) draftstandard for real-time communication between browsers [112]. The goalof WebRTC is plug-in-free low-cost communication in real-time betweenany browser. And with communication not only audio and video commu-nication is meant, but also the direct exchange of data. So with the helpof WebRTC it is possible to create a Peer-to-Peer (P2P) connection frombrowser to browser and send audio, video and data over it. This WebRTCapproachhas been chosenbecause it is the new trend in the field of theVoIPcommunication, it is open source and there is no software necessary otherthan a browser on the client side [109].

Additionally, the WANem [98] framework for network emulation wasused. This software, which is based on the Linux OS Knoppnix [64], isconvincing because it makes use of the well-accepted open source networktool for Linux calledNetEm[61][34]. Furthermore,WANemprovides the

47

possibility to build aUI on top of it, which facilitates the flexibility to createa control panel that meets with precision the demands of an QoE evalua-tion experiment, such as the easy setup of different scenarios with diverselatency, bandwidth, packet loss and jitter settings to be tested, as well as theautomated collection of the user ratings.

ThearchitectureusingWANemis illustrated inFigure 4.1. There are fourcomputers connected in a local LAN through a switch. Two computers runthe WebRTCmessenger, one computer runs an Apache [35] web server, anode.js [60] signaling server as well as a MySQL [35] data base and thelast computer runs the WANem tool that can emulate the network. TheWANemworks as follows: The routing table of the two computers that runtheWebRTCmessenger are modified in such amanner, that all the packetsare routed to the other peer through the computer that runsWANem. Thiscomputer is responsible for the network emulation. E.g., if the packet lossis set to 50%, the WANem computer will drop every second packet that isrouted through.Such architecture with LAN cables and a switch is neces-sary to guarantee a controlled network environment without the interfer-ences that happen in a Wireless Local Area Network (WLAN) network.

4.1.2 Experimental Procedure

The experimental procedure besides the hardware and software -related in-formation, includes some important information concerning (a) the par-ticipants group (subjects) and (b) the procedure of the experiments.

The subjects of the experimental procedure of this experiment were 34volunteers at the University of Zürich and the High School of Willisau inSwitzerland. The volunteers were mainly computer science students be-tween 20 and 25 years old. However, the overall age distribution range ofthe subjects was from 16 to 63 years old. A pair of two randomly selectedsubjects participated in several voice calls with different technical parame-ters. Each subject rated the quality of each call separately after the end ofthe call.

The goal of the overall experimental procedure was to affect as little aspossible the QoE rating of each subject. Firstly, the number and the du-

48

ration of the test calls defined carefully, so that the experiment about thehuman experience would not demand from a subject to actively participatefor more than 30minutes in voice calls. Otherwise, it was assumed that thesubjects would become annoyed and/or bored and their answers would beinfluenced by emotions which would decrease the quality of the results.Thus, to avoid such situation, the total duration of each experimental ses-sion designed to not exceed one hour.

Having a fixed interview length influenced the decision concerning thenumber and the duration of the test calls. There is a trade-off between thenumber of measurements and the confidence of the results. If the test callsare longer, fewer experiments can be performed within a fixed time-frame.It was assumed that people are not able to have a free and balanced conver-sation of 45 second on their own since it does not seem to be enough timeto develop a proper conversation, especially between strangers. Thus, thefollowingmethod was used to support the conversation in the test calls: Atthe beginning of the experiment each participant got around 300 easy gen-eral knowledge questions [74] [78] and the subjects had to ask and answerthem alternately. This approach led to a fluent and balanced conversationwithout distracting the subjects from their evaluation task.

The decision about the procedure of the interview was as follows:

• 0-5min Introduction, explanation of the experiment and rating sys-tem

• 5-25min 16TestCalls, around 45 seconds calling time+ 15 secondsvoting time each

• 25-30minQuestion and Answers about the calling experience

4.2 AQX Parameters in VoIP

The following evaluation is based on the MOS of 34 subjects, which pro-duced in total more than 500 end-user’s opinion score ratings at an over-all calling time of approximately 6 hours. 80% of these ratings were col-lected in a single variable scenario, where only one variable was adjusted.

49

Table 4.1: x0 and References Used for the Evaluation

Variable x0 ReferenceLatency 150 ms ITU [52][54]Jitter 100 ms Cisco [21]

Packet loss 5% Opus documentation [80]Bandwidth 65 kbit/s WebRTC official blog [113]

The remainder of these ratings weremixed variable scenarios, wheremulti-ple variables were adjusted simultaneously. The main focus of this experi-ment’s QoEmeasurements was on the single variable scenarios, since AQXonly demands knowledge of expected variable values and influence factorsof individual variables to predict QoE considering multiple variables. Theprimary goal of this experimentwas to calibrate AQX for theVoIP scenario.Thus, the data were collected for this purpose. The secondary goal of thisexperiment was to validate the AQX prediction performance in the multi-ple variables scenario. It was assumed that less but equally distributed setof data points would be sufficient to reveal the potential inaccuracy of theAQXmodel.

Although the AQX model can be calibrated for diverse technical andnon-technical variables, in this experiment the comparison of AQX againstother state of the art models is attempted. Thus, the focus is on four techni-cal variables: (1) jitter, (2) latency, (3) packet loss, and (4) bandwidth.The AQX model needs for every technical variable an expected variablevalue x0 [108] [105]. The WebRTC technology that was used in this ex-periment is relatively young and rich literature, that can be used to findpossible expected variable values, does not yet exist. Thus, the x0 valuesand references as of Table 4.1 were used for this evaluation

4.2.1 Single Variables

Here those results of scenarios are presented, where only one variable wasaffected. E.g., in a scenario with 5% packet loss, the latency and jitter wasset to zeromiliseconds (0ms) and the bandwidth was unlimited. Table 4.2

50

Table 4.2: Results of the Single Variable Scenarios

Latency Packet loss Jitter Bandwidthm+ 0.40 0.09 1.06 4.53R2 -0.65 0.85 0.96 0.75m+ 0.32 0.73 0.59 0.47R2 0.75 0.95 0.96 0.94

Latency (ms)0 200 400 600 800 1000 1200 1400 1600 1800

Mea

n O

pini

on S

core

1

1.5

2

2.5

3

3.5

4

4.5

5AQX-Model(m-:0.4, m+:0.32)Collected MOSStandard Deviation (σ: 0.765)E-Model

Figure 4.2: AQX Model Fit and Comparison for Latency

summarizes the influence factors (m) values found for each variable andthe Goodness-of-Fit (GoF) for each m expressed as R2 value. Results ofthe Single Variable Scenarios

These high R2 values show that the AQX model is able to capture QoEof end users quite accurately.

51

The resulting graphs of the AQXmodel were also compared to the othertwo QoE-predicting models during this experiment. Figure 4.2 showstwo plots which show the comparison between the AQX and the E-model(dashed line) QoE results as a function of latency. IQX hypothesis is miss-ing in Figure 4.2 because there is not an equation proposed by IQXmodel-ing MOS and latency. Thus, there are no IQX data available for latency.

It is noticeable in Figure 4.2 that the E-Model proposes lowerMOS val-ues than the fitted AQX model most of the time. So for example for a la-tency of 1600ms theMOS for the E-Model is 1.79 and for the AQX-Model2.875. The AQX QoE value is probably too high for such a high latencyvalue. The reason for such high values could be the following. Latencyis something that is not directly annoying, like a bad audio quality. It issomething that gets more annoying the longer and faster a conversationbecomes. Latency is not that disturbing in a short conversation with smalltalk characteristics. The conversations of the experiments had exactly thesecharacteristics. Thus, it was assumed that the subjects did not report lowMOS for high latency.

Since the collected MOS values seemed rather high, some extra experi-mentswere performedwith longer experimental calls inwhich only latencywas tested. For these calls, three different conversational tasks proposed bythe ITU-T were tested: (a) a travel office role play, (b) a random numberverification task and (c) a contacts exchange task [56]. The results of theseextra tests were unexpected. The subjects rated still high. For a test sce-nario with 1500ms latency the MOS was still 3.17. Therefore, it is furtherassumed that not only the duration and the type of conversational task areresponsible for the unexpected outcome. It is possible that a cultural phe-nomenon leads to such results. As stated in [56] MOS can vary due tocultural differences. Except four subjects, all of them spoke Swiss Germanwhich is a rather slow language and therefore latency probably disturbs less.This hypothesis is supported by a test call between a Russian and an Ital-ian participant held in English which seemed to be faster and more inter-active than most of the native Swiss German speakers’ conversations, andin this experiment the latency was not tolerated. Although a proof of this

52

Packet Loss (%)0 5 10 15 20 25 30 35 40 45 50

Mea

n O

pini

on S

core

1

1.5

2

2.5

3

3.5

4

4.5

5AQX-Model(m-:0.73, m+:0.085)Collected MOSStandard Deviation (σ: 0.703)E-ModelIQX Hypothesis

Figure 4.3: AQX Model Fit and Comparison for Packet Loss

phenomenon could not be found in [88], since the sample was not largeenough it stays only a hypothesis in this experiment.

This experiment showed that E-model would be incapable to spot suchbehavior while AQX could predict the right MOS with high accuracy withthe right influence factor selection. Thus, it is shown that theE-model is notaccurate in every VoIP scenario, since it was proved to be too pessimistic inrespect to the negative effect of latency between slow-speaking subjects.The MOS depends on the service and the respective users. Therefore amodel that allows flexible calibration, such as AQX, is needed to predictQoE accurately in diverse scenarios.

In Figure 4.3 the MOS as a function of the packet loss is illustrated andcompared to the E-Model and the IQX hypothesis. The AQX model ap-pears to capture QoE better than the E-model and the IQX hypothesis.

53

In the IQX Hypothesis experiments [31] the Internet Low Bitrate Codec(iLBC) was used during the measurements. For the E-Model calculationtool [57] the G.711 Codec [58] is assumed, and the codec in those exper-iments was Opus [80]. The reason of having different codecs is that therewas no control in codecs used in the related work and the use of the samecodecs was not possible inWebRTC by the time of the experiments. How-ever, Opus has advanced error correction mechanisms similar to the mostadvanced version of G.711 that is used in the E-Model and the one of theIQX hypothesis experiments, therefore the QoE results of different mod-els, for the same variables, which are presented in Figure 4.2 and Figure 4.3are comparable.

Another important result of this experiments is the analysis of the devel-opment of the m values. During the analysis of the experiment’s results, itwas examined if them, which needs to be determined empirically, remainsconstant, or if it is necessary to further adjust it during the evolution of avariable’s value (x). This analysis has been conducted by fitting the AQXmodel through all two neighbored data points and the so determinedm val-ues were compared to each other. Figure 4.4 and Figure 4.5 illustrates thiscomparison for different variables and shows thatm should not be consid-ered to be constant in every case. On on hand, for some variables (jitter,latency, packet loss) m values appear to be almost constant, since m oscil-lates around an almost horizontal line. On the other hand, bandwidth is avariable that proves the necessity of different influence factorsm− andm+

above and below the reference value x0.

4.2.2 Multiple Variables

This part presents the AQX model for scenarios where multiple variableswere tested. Since the main focus was on single variable tests, there are notsomany data points for thesemixed scenarios. However, it was sufficient torun basic evaluations for different values of Latency (L), Packet Loss (PL),Jitter ( J), and Bandwidth (B).

54

Figure 4.4: Development of the m Values (jitter, latency, packet loss)

In those experiments 14 multiple variables scenarios were tested andtheir results are summarized in Table 4.3. The first four columns indicatewhich variable values were tested. Column five and six contain the MOScollected from the subjects in experiments as well as the standard devia-tions of these collected ratings. The last column shows theMOS calculatedby the Equation 3.33 of the AQX model using the parameters determinedby the single variable scenarios.

Comparing these results in Table 4.3 it has to be noted that Equa-tion 3.33 creates promising results, since the differences between calcu-lated and collected MOS are small. The mean of all MOS differences is0.53, which is small for an unadjusted calculation where all weights equal1. Each variable’s weight serves as another degree of freedom, allowingfurther calibration of AQX. However, there is not a sufficient amount ofdata points to make any significant statement in those experiments. Thus,in such cases the additional degree of freedom that AQX allows for, couldnot be used. Considering these high standard deviations of thosemeasure-

55

Figure 4.5: Development of the m Values (bandwidth)

ments, another thorough verification should be done in future to validatethe accuracy of the AQXmodel in VoIP scenarios.

The result of this multiple variables scenario (latency and packet loss)is illustrated in Figure 4.6, where two variables are mixed. The 3D-curve isthe calculated AQXmodel for the two variables and the large black bulletsshow the MOS collected. The size of these bullets has been chosen for vi-sualization purposes and they should be ideally cut in half by the 3D-curveof the AQXmodel.

4.3 Further Calibration

During this experiments two further calibrations of the AQXmodel for theVoIP scenario have been performed. Thefirst one defines an adjustedMOSscale. As with the E-Model under normal conditions a MOS of 4.41 is ex-

56

Table 4.3: Collected MOS for Mixed Variables Compared to the Calcu-lated MOS

L [ms] PL [%] J [ms] B [ms] CollectedMOS AQX(std. dev.)600 10 0 0 3.13 (0.64) 2.59500 7 0 0 3.56 (0.73) 2.82500 10 0 0 3.00 (0.67) 2.63500 10 0 60 3.25 (0.46) 2.05400 0 0 75 3.38 (0.74) 3.09400 7 0 0 3.25 (0.71) 2.87400 20 0 75 2.50 (0.93) 1.95250 10 0 0 2.80 (0.63) 2.770 7 0 64 3.88 (0.64) 3.080 7 0 98 3.88 (0.64) 3.260 10 0 60 3.25 (0.46) 2.600 12 0 98 3.25 (0.71) 2.890 0 300 63 3.13 (0.83) 2.640 12 400 0 2.63 (0.74) 2.21

57

Figure 4.6: 3D-Graph of the AQX Model for Multiple Variables

pected [60], similar findings were made during experiments here. With allsubjects an uninfluenced scenario was performed to observe themaximumpossible MOS. The result is a MOS of 4.432, being very close to the onefrom the E-Model. The next assumption made now is that there does notexist any MOS higher than 4.432 and, therefore, the new scale is from 1 to4.432.

The second calibration which can be done due to the AQX model is anadjustment of the x0 parameter. Such an adaptation of x0 is a contradictionto the ideaof theAQXmodel, because this parameter shouldbedeterminedbefore experiments according to the SLA, or service characteristics. How-ever, one could argue that often x0 values are not precisely defined and thereality might vary from proposed values in the literature. Therefore, as asecond, further calibration x0 values can be determined like the m valuesthrough a non-linear least square regression.

58

Latency (ms)0 200 400 600 800 1000 1200 1400 1600 1800

Mea

n O

pini

on S

core

1

1.5

2

2.5

3

3.5

4

4.5

5AQX-Model(m-:0.61, m+:0.8)Collected MOSStandard Deviation (σ: 0.765)E-Model

Figure 4.7: Adjusted AQX Model Fit for Latency

When these two further calibrations are applied to the AQXmodel, theGoF improves for all variables in single variable scenarios and this improve-ment of the GoF results in more accurateMOS calculations in themultiplevariables scenario. This means that the mean difference between the MOScollected and the MOS calculated drops from 0.53 (c.f. Subsection 4.2.2)to 0.21. Such difference is low regarding the fact that there is no calibrationof these weight factors.

The improvement through this calibration is also illustrated in Fig-ure 4.7, where the AQX model is further calibrated for the variable de-lay. Compared to Figure 4.2, which shows the uncalibrated version of themodel, the graph of the AQX model is now closer to the data points col-lected. Thus, this calibration of the minimum andmaximumMOS and the

59

Figure 4.8: 3D-Graph of the Adjusted AQX Model

expected variable’s value (x0) leads to a significant increment of the GoFR2 value, compared to the unadjusted AQXmodel.

Figure 4.8 presents the adjusted AQX model 3D-curve in the two vari-ables scenario (latency and packet loss). It can be seen that the 3D-curvecuts through more large black bullets than in Figure 4.6. Thus, it estimatesbetter the MOS collected.

4.4 Chapter Summary

This chapter presented the design and execution of a QoE measurementexperiment setup, which is able to save and replay a sequence of differentnetwork scenarios emulations. Moreover, this setup provided the possibil-ity to save user ratings and perform an application-specific analysis with ad-justable variables being emulated and encompassing jitter, latency, packetloss, and bandwidth. The VoIP messenger— developed based on the We-

60

bRTC technology— collected over 500 data points in experiments with atotal of 34 subjects.

The data collected was used to calculate AQX results for each scenarioand respective MOS results were used to define those parameters of theAQX model for VoIP services. The evaluation performed three steps foreach variable: (1) the AQX model was fitted through the MOS collectedand the resulting GoF and the value of the influence factors m were ana-lyzed; (2) the resulting AQX model was compared to the ITU-T E-Modeland the IQX hypothesis; (3) these variables were considered in a mixedscenario with other variables.

It was shown that theAQXmodel reaches a highGoF.Moreover, an out-come of the analysis ofm values is that they are not constant and further re-search is required in this area, to determine a model with a non-constantmvalue. Additionally, those experiments revealed that the formula for multi-ple variables of the AQXmodel produces promising results, specifically forthe set of measurements with mixed variables, which were performed. TheAQX formula adopted for mixed values is also promising, especially if it isused with further calibration techniques concerning the MOS co-domainand the appropriate x0 selection. All these findings lead to the conclusionthat AQX is a highly adaptable and precise model, which outperforms allother state of the art models. Having provided the influence factor m fordifferent services, the AQX model becomes a powerful and useful tool forservice providers to predict and improve their services in terms of QoE.

61

62

5TheDesign to overcomeMTRs

Monopoly

T he amount that the provider of the callee will charge thecaller’s MNO, to terminate his call into his network is termed MTR

(cf. Figure 1.1). Furthermore, theMTR is considerably big proportionallyto the total cost of each call. Subscribers of MNOs rarely consider the ter-mination cost that their operator is charging other networks, when deliver-ing an incoming call to them before they establish their contract. Further-more, a significant raise of termination rates fromMNOs will increase thecommunication cost for all the MNOs subscribers. However, it will rarelyhave a negative impact and dissatisfy the customer base of an operator.Given that fact, MNOs termination-service as well as MNOs terminationrates are considered to be a monopoly [14]. Thus, the national telecom-munication regulation authorities are usually regulating these rates acrossthe world (Europe [22][99], North and South America [30][5], and Asia[100]). The problem of the monopolistic call-termination market is alsolarge for users that wish to receive calls in a location that is different fromthe home location, where the service was registered (roaming). Roaming

63

users has to pay high prices to receive a call [110]. Thus, the ITU has puteffort toward the solution of this problem [48][49][50].

There is a rich literature on how regulation of this market is affecting it[24] as well as selected research on the topic of the proper selection of ter-mination rates [16][91]. Furthermore, effort has been put into the analysisof business strategies, which MNOs follow, concerning their terminationrates [41][68]. The customer base of a single MNO is significantly smallercompared to the set of total customers in every otherMNO, FNO, or VoIPprovider. Thereby, the majority of calls that a MNO has to terminate inhis network originate from foreign networks. Thus, for many years MNOsrevenue is coupled with high termination rates applied. This is the mainmotivation for MNOs to keep regulation in this monopolistic market aslow as possible.

Assuming that a caller has a complete and always up-to-date knowledgeof every pricing plan of each available to him MNO, the use of dual-SIMcard devices [20][115] could be an option against high termination ratesdue to the following reasons: (1) A user subscribes with his dual-SIM de-vice simultaneously in twoMNOs. Usually MNOs provide lower rates forcalls between their customers (in cases those calls might also be for free),since in that case there is no termination fee involved. Furthermore, in thepast, a MSISDN corresponds only to a subscriber of a specific MNO. So,a caller (2) identifies by the callee’s MSISDN, whichMNO is serving him,and thus (3) selects the cheaper option among his MNOs to complete thecall. However, since the Mobile Number Portability (MNP) was intro-duced, there is not a uniqueMNO that theMSISDN can belong to. Nowa-days, a callee can establish a new contract with any availableMNO and stillcan be reached by the same MSISDN.Thus, the caller needs to have priorknowledge of which MNO is serving the callee to be able to select in thedual-SIM device the MNO, which provides the cheaper calling rate for aspecificMSISDN. Beside the callee’s MNO lack of transparency, high con-tract operations costs are an additional obstacle for callers wanting to usedual-SIM devices to minimize their out coming calls cost. Thus, dual-SIMdevices require contracts with more than one MNO and do not affect thetermination charges monopoly.

64

Fortunately, there are two significant updates, compared to the earlydays of mobile communications, which enable the operation of an effi-cientMNO-independent call-termination solution. This solution will havean impact toward the termination rates monopoly, since MNOs will loosethe control of their subscriber’s call-termination procedure. Those updatesare (1) the infrastructure update and (2) the mobile data introduction. Inmore detail, the majority of the newest mobile terminal devices, e.g., SmartPhones, can equally register in almost every network across the world, ir-respective of the device vendor. Nowadays, Smart Phones are the rule andnot the exception between mobile subscribers mobile terminal choice ac-cording toNielsen [76]. Suchdevices have sufficient computational power,multiple network interfaces, provide positioning information, and can alsosupport cross-platform applications, which are fully integrated within thedevice’s UI.Thus, several procedures, like computational calculations or anexchange of data between the caller and the callee, can take place prior to acall without the calling parties experiencing any difference during the call-ing procedure. Furthermore, mobile data charges are nowadays decreasing[75],mobile data rates are higher and expected to be improved in the futurewithin the next generations mobile communication networks [94].

Thus, this thesis presents here in today’s mobile networks environment,AbaCUSwhich is (1) aMNO-independent call-termination systemagainstthe mobile termination charges monopoly, and (2) a fair charging systemwhere the one who pays can influence the price and the QoS perceivedlevel. Thus, this thesis overcomes themonopoly obstacle of themobile ter-mination rates market, providing a fair charging solution.

Furthermore, this thesis shows that the unregulated adoption byMNOsof such a solutionwill have a positive impact for both end-users andMNOs.The former by selecting theMNO to terminate a call with a given QoS at agiven price, and the latter is facilitated through the call-termination of otherMNOs subscribers.

65

5.1 Stakeholders Analysis

The purpose of the stakeholders’ analysis and stakeholders’ map presentedat Figure 5.1 is to visualize (i) the interest, (ii) the influence, and (iii) theattitude of each stakeholder that is involved in the MTRs monopoly envi-ronment. The approach to perform the stakeholders analysis and draw thestakeholders map is outlined below.

1. Create the list of all stakeholders

(a) End-Users (caller, callee)

(b) Regulators

(c) Governmental Entities

(d) M(V)NOs

2. Define their interest in the project (Low, Medium, High)

3. Define their influence in the project (Low, Medium, High)

4. Define their attitude in the project (Positive, Neutral, Negative)

5. Draw the stakeholder map

5.1.1 End-users

During this thesis here an Internet survey [101] performed among approxi-mately two hundredMNO’s subscribers , who belong to a diverge location,age, sex, social, and educational background. As results in Table 5.1 show,it is observed that 58% of subscribers would not mind to pay more for aguarantee QoS. 74% of those subscribers would not mind to experience abelow average sound quality communication, if the total price of the callwould be less than originally priced. Finally, 61% of the callers are tolerantin time delays concerning their call establishment waiting-time. On onehand, caller’s attitude is considered to be positive and the interest in Aba-CUS high since the caller will benefit either from better services, and/or

66

Figure 5.1: AbaCUS Stakeholders Map

Table 5.1: Internet Users’s Survey Results

Per cent of callers that willing to pay extra for better QoE 58%Per cent of callers that willing to accept poor QoS for a lower price 74%Per cent of delay tolerant callers 61%

lower prices. On the other hand, callee’s attitude is neutral since there isno monetary benefit neither any loss. For the same reason, the interest ofthe callee is considered to be low. The influence of a caller and a callee inAbaCUS ismedium since high participation is essential but notmandatoryfor a successful termination of the MTRs monopoly.

5.1.2 Regulators and Governmental Entities

To break a de facto market-defined monopoly, such as the MTRsmonopoly, even if it is technically feasible, it would take a lot of time from

67

the regulator to enforce newpolicies that demand from the playersmore in-vestments, either in terms of infrastructure or working-time. In this thesisan interviewwith the Swiss Federal Office of Communications (OFCOM)[79], which is the regulation authority concerning telecommunication inSwitzerland, confirmed that to break such a monopoly is a time costly pro-cess when regulatory actions are needed. However, OFCOM agreed thatAbaCUS shows a clear impact not only on social wealth fare increment butalso on MNOs opportunity to monetize unused infrastructure. The flex-ible mobile termination service proposed in this thesis will lead to betterservices in the future and more reliable services fromMNOs.

Political action might be needed in special cases, such as the Swiss mar-ket, where governmental entities, such as the parliament, might need totake an action to enforce a new law. In such environment the regulator(e.g., OFCOM) first need to recommend the changes that are needed, andthen the respective governmental entity to initiate the procedure of enforc-ing a new law. The latest complicates and slows down more a potential at-tempt to enforce a MTRs liberation. Thus, this thesis strengthens benefitsfor MNOs as an incentive of a voluntary adoption of such system with aminimum regulatory and legislative demand.

Both regulators and governmental entities have high influence in Aba-CUS since the system need to be adopted to meet any local regulation/lawdemands. The interest of the regulator is high because AbaCUS overcomesthe MTRs monopoly problem and reduces the regulation demand. How-ever, for governmental entities the interest in AbaCUSwill increase only incase that the regulator recommends specific law enforcement. Finally, theposition of regulators and governmental entities is neutral since there is nodrawback in case the MTRs monopoly cancelation is achieved.

5.1.3 M(V)NOs

M(V)NOs are themobile call termination SPs and thus the key stakeholderof the AbaCUS system. AbaCUS increases the utilization of MNOs infras-tructure and allows monetization of unused MNOs resources by offeringthe possibility to a MNO to provide mobile call-termination services to

68

end-users that do not belong tot heir customer base. However, the initialattitude of aMNO is negative since the de-factoMTRsmonopoly is termi-nated. Furthermore,MNOs influence is high since a voluntary adoption ofAbaCUS will eliminate any overhead for regulators and governmental en-tities action demand. Finally, the interest of such system is high because ofthe new revenue sources possibility from already existing infrastructure.

5.2 AbaCUS: Auction-based Charging User-centric Sys-tem

AbaCUS defines an approach where the CPP principle is applied. In Aba-CUS a call can be terminated by everyMNOwho provides network cover-age in a specific location and who is willing to terminate any mobile com-munication subscriber’s call, irrespective of the provider the callee belongsto. Since the modern mobile terminal devices are multiband-compatible,there does not exist any technological boundary for this functionality any-more. Furthermore, no SIM change is required from the callee so there isno SIM-lock [83] interference with the AbaCUS call-termination MNO-independent system. Similarly to roaming users, who can use the same de-vice for domestic as well as abroad usage without replacing their SIM card,in AbaCUS the callee can receive a call by anyMNO that provides networkcoverage in his location, without the need of additional equipment.

Figure 5.2 illustrates the key elements of AbaCUS. A caller is flexibleto use the voice-service provider of his choice, such as VoIP, MNOs, andFNOs, to place a call. The caller can reach the callee by dialing directly hisMSISDN. In this case the host MNOwill collect the call-termination rate.However, a competitive MNO may generate a virtual MSISDN and allowto the callee to register in his network. Thus, the caller may dial the virtualMSISDN and reach the callee. In that case the guestMNOwill profit fromthe termination rate. Multiple MNOs can participate in an auction, wherethe caller will request to place a call, reach a callee in a specific location,and demands a certain QoS for the specific call. This demand is expressedby QoS Classes (QoS-Cs), which contain parameters related to the soundquality and the network-accesswaiting-time. MNOsbidding in the auction

69

Figure 5.2: Key Elements of AbaCUS

will reply to this request by proposing their charging demand. The charg-ing demand is expressed by the Termination Rate Class (TeR-C), whichcontain a potential start-up cost and the desired charging rate. Finally, ona referee role during the AbaCUS auction is the Auction Authority (Au²),which receives call requests from callers and fromMNOs the selectedTeR-C preference per QoS-C. The Au² considering end-user’s preferences, willindicate which MNO will maximize QoE for a given caller. The winningMNOwill be selected to provide the call termination service.

70

Table 5.2: QoS-Cs List

QoS-C Guaranteed Network Access SoundQualityQoS-C 1 Yes HDQoS-C 2 Yes standardQoS-C 3 No HDQoS-C 4 No standard

5.2.1 QoS Class (QoS-C)

Table 5.2 summarizes the fourQoS-Cs defined inAbaCUS. EachQoS classconsider simultaneously two variables, (1) the network-resources accesspriority, and (2) the sound quality. The reason for the distinction amongthese classes has been taken to reach a compromise on a combination ofsound quality needed and the importance of a call, while taking in accountdiverse QoS demands that a caller might have. Thus, there is a binaryoption of guaranteed network access and HD voice request. To facilitateQoS-C 1 andQoS-C 2MNOs should have network resources reserved forcallers that are willing to be charged possibly at a higher rate for such ser-vice. The proposedQoS-Cs in AbaCUS are not binding and do not need tonegotiate necessarily only such binary-decision options. In the futureQoS-Cs can encapsulate more, or different options, such as decision on codecsto be used in case of full VoIP over LTE services.

5.2.2 TeR Class (TeR-C)

There is only one TeR-C defined in AbaCUS. However, there are twoeconomic variables that need to be considered for every QoS-C: (1) theset-up call cost demand Cset−up ∈ [0,Cset−up

max ], and (2) the MTR costCMTR ∈ [0,CMTR

max ]. Furthermore, for each variable a common referencevalue 0 < Cset−up

0 < Cset−upmax and 0 < CMTR

0 < CMTRmax will be also de-

fined. Themaximum and the reference values of each variable are the TeR-C characteristics and are (a) different for every QoS-C, but (b) the samefor every MNO.The exact values of TeR-C characteristics will be selected

71

by MNOs representatives and approved by the regulation authority. Thus,MNOs will compete on common TeR-C characteristics in every QoS-C.

Both Cset−up and CMTR are IV in terms of the AQX terminology. Fur-thermore, the caller will define on-demand influence factors m−, m+ andthe importance factor w for each parameter in the TeR-C, for a requestedQoS-C.Thus, the overall QoE of a caller for a givenCset−up andCMTR offerfrom an MNO can be estimated with AQX. To calculate the influence fac-tors m− andm+ for both Cset−up and CMTR, the end-user will provide theprice tolerance percentage (e.g.,±25% · Cset−up

0 ) and the influence factorswill be calculated with Equation 3.28. Concerning the importance factorsw, since the respective of the TeR-C are only two, the caller will send thefollowing quotient wset−up/wMTR.

5.2.3 AbaCUS BiddingMetric

The bidding process in AbaCUS ecosystem is described through an ex-ample. Assume that a caller request a service in the QoS-C X (X ∈{1, 2, 3, 4}). All the competing MNOs will propose a pair of Cset−up andCMTR. Considering callers influence factorsm and importance factorsw forboth variables, a MOS will be generated using Equation 3.33. The MNOwill maximize caller’s QoE will be selected to provide the service and ter-minate the call. In case of a draw aMNO among the “winning”MNOswillbe selected randomly.

Ideally, MNOs shouldmonitor their networks load and decide based onthe resources availability about the current Cset−up and the CMTR for eachQoS-C.AMNOshould increase the valuesof theTeR-Cvariableswhen theload of their networks is increased and vise versa. This approach will dis-tribute equally the load across eachMNO, resultingminimization of dopedcalls in network congestion situations. Furthermore, a MNO can reduceCset−up and CMTR cost in case of high unoccupied network resources, toattract callers from competitive MNOs, aiming at revenue increment.

72

Table 5.3: Approaches Against the MTRs Monopoly

Auctions QoS Competition Agreementsupport support is allowed needed

CPP No No No -RPP No No Partially -NatRoam No No Partially YesAbaCUS Yes Yes Yes No

5.2.4 AbaCUS Auction and Au²

In the domain of online trading and in the scope of a real-time decision-making feature, where the internally applied auction mechanism of Aba-CUS belongs to, a single-round auction – due to urgent export result de-mand – is needed. Thus, the English [23] auction is proposed and theMNO with the higherst MOS of a service request is winning and providethemobile termination service. EachMNOhas no knowledge, neither canguess, the TeR-C’s variables values selection per QoS-C chosen by otherMNOs since MNOs do not have an insight on their competitor’s networkstatus and their provisioning policies. Furthermore, MNOs cannot predictwhen the next service request for a specificQoS-Cwill occur. Thus, it is notpossible to predict a bid with a higher winning probability. To ensure thatMNOs are not interested only in specificQoS-Cs, a TeR-C proposal for ev-ery QoS-C is needed; otherwiseMNOs cannot participate in the AbaCUSauction. When a termination rate request will be received, the Au² will se-lect theMNO,whichwill maximize caller’s QoE. Table 5.3 summarizes thebenefits of AbaCUS auction achieved compared to the CPP principle, theRPP principle, and the NatRoam cost solution.

5.3 AbaCUS Technical Requirements

To facilitate AbaCUS the following technical requirements are needed. (1)The regulator should allow (if not allowed) the NatRoam principle so thatany SIM card can camp on any MNO. (2) The value of Cset−up and CMTR

73

should be included in the Call data Records (CDRs) so that MNOs cancharge the caller with the respective amounts. (3) An automatic and on-demand MNO selection mechanism is needed to facilitate callees MNOhopping, in a timely and energy efficient manner. The first two require-ments require potential action from regulatory authorities andMNOs. Thelast requirement is proven to be feasible in this thesis.

5.4 Chapter Summary

This chapter showed that technical limitations of the past, which enforcedthe MTRs monopoly, can be overcome. Also, the breaking apart of thismonopoly will be an opportunity for MNOs rather than a threat. In theproposedMTRsmarket,MNOshave the opportunity tomaximize the util-ity of their infrastructure by providing services to call receivers, who didnot have access until now. Furthermore, MNOs have the opportunity toprovide premium quality services for callers, using existing infrastructurethrough the pre-allocation of their network resources to customers that arewilling to pay more for a better and guaranteed QoS service, [36] showedthat paying for QoS can increase networks’ utilization.

Within AbaCUS the caller, who is the party paying theMTR, is also theone to chosewhichMNOwill terminatehis call, determining a significantlyfair approach, since the party paying is able to influence the total cost of thecall.

74

6Automatic andOn-demandMNO

SelectionMechanism

T he development and the evaluation of an automatic and on-demandMNOselectionmechanism for theAndroid platform, which

is presented in this thesis, is motivated due to the following 3 reasons: (1)The number of available MNOs in a certain location, (2) potential bene-fits of such a mechanism for the three main mobile communication stake-holders (MNOs, end-users, and regulation authorities), and (3) the exist-ing number as well as the future estimation of devices in themobile phonesmarket that can support such a mechanism.

In 2013, 191 MNOs, which are active in 61 countries across Europe[25], result in an average of three available MNOs per country. Addition-ally, in mobile communications there is no physical barrier (e.g., wires)that might force an end-user to stay connected with a specificMNO.Thus,MNO subscribers, due to multiple available MNOs in a location and com-monly usedmedium inmobile communications, can hop automatically be-tween different MNOs according to their preference.

From the MNO’s perspective such a hopping attitude is driven by thefact thatMNOscanbenefit byofferingon-demandpremiumservices to any

75

subscriber of any competitive MNO, e.g., high and/or guaranteed soundquality or guaranteed access to the network in case of network congestion,if the caller and/or the callee register temporarily in other network(s). Inthis case the hosting MNO can profit from collecting termination rates ofthe call. On one hand, MNOs can monetize some of their available net-work resources by attracting more users to use their services, while offer-ing a lower than the usual price in case of low network load. Accordingto the analysis [28], MNOs should increase their focus on new researchsuggestions, as the proportion of total retail telecoms revenue stemmingfrom their current mobile services is expected to drop over the next fiveyears. On the other hand, MNO subscribers can benefit from lower ser-vice charges and/or better QoS agreements. Despite economical benefitsmentioned, an automatic and on-demandMNO selection mechanism canminimize the non-ionizing radiation of the device, especially by each timeselecting the MNOwith the stronger signal strength, as recently proposedby [93]. In this thesis, the focus of the automatic and on-demand MNOselection flexibility is that an automatic and on-demand MNO selectionmechanism, can be introduced by regulating authorities and enforce thecompetition in the traditionally considered monopoly of the MTRs.

A widely used automatic and on-demand MNO selection mechanism[103] shouldbe supportedbymanymobile devices that canbe equally usedin almost everyMNO across the world. Thus, those devices should be ableto operate in multiple 3rd Generation Partnership Project (3GPP) tech-nologies such as 2/3/4G [40] and should have high market penetration.Smart phones fulfill those criteria; according to [77], since 2011 smartphones are the primary customer’s choice. SinceAndroid is one of themostpopular platform in the smart phones market [10], the decision to imple-ment and evaluate the automatic and on-demand MNO selection mecha-nism on the Android platform has been taken. Thus, the research questionanswered in this Chapter here reads as: Is it feasible and efficient to designan automatic and on-demandMNO selection mechanism, which supportsthe attempt to overcome the mobile termination rates monopoly obstacle?And the engineering question answered concerns the path in which way to

76

implement effectively and efficiently such a prototype in the Android plat-form.

6.1 MNO Selectionwith ATCommands

TheAttention (AT) commands interface has been a standard way to accessmodems as computer peripherals [97]. Generally an AT command con-sists of three parts. It starts with AT followed by a command and ends withthe line termination character [2]. There are three different types of ATcommands (Test, Read, and Set).

Test commands test the existence of a command and check its range ofparameters. The format of those commands is ATxxx=?. To get a list ofavailable MNOs the command AT+COPS=? has to be sent to the GSMmodem. The reply of the GSMmodem returns a list ofMNOswith the fol-lowing information: (a) MNO status (0 unknown, 1 available, 2 current,3 forbidden), (b) MNO short and long alphanumeric name e.g., OrangeCH or ORANGE, and (c) a five digit number that represents the three digitsMobile Country Code (MCC) followed by the two digitsMobile NetworkCode (MNC), which is the code for the network provider [92].

The readATcommands, as indicated by the name, read the current valueof parameters. Set Commands are used to set new parameter values. TheAT command interpreter will return OK in the case that the command hasbeen successful, otherwise an error or informative result code will be re-turned. The MNO set AT command reads as AT+COPS=1,2,``22801''.In this command, the first integer defines the mode, with five different val-ues (0 automatic, 1 manual, 2 deregister from network, 3 set only, 4 if man-ual selection fails, automatic mode is entered). The second integer showsout of three possible values that format the MNO is referenced to (0 longformat alphanumeric, 1 short format alphanumeric, 2 numeric). Thus, ifthe numeric format has been chosen, the last parameter identifying theMNO is theMCC plus theMNC, e.g., 22801 for Swisscom in Switzerland.

There have been many attempts to send AT commands to Android de-vices, either as peripheral from a computer or directly from the device itself[32]. But not all issues have been solved yet. Within this thesis, an attempt

77

to send AT commands as peripheral from a computer to a SamsungGalaxyS II (SGS2) smart phone was done. Furthermore, an attempt to send ATcommands from the device itself took place. The attempt to send ATCom-mands from a computer was successful. Prerequisites were that the correctGSM modem driver of SGS2 was installed on the computer. Afterwards,the modem could be addressed over the correct device port with a SecureShell (SSH) client. The outcome of it was a successful MNO selection.However, the approach to send the MNO set AT command from the de-vice itself failed. Until this thesis concluded, there was no documentationof a successful MNO switching solution via AT commands directly sentfrom an Android device.

6.2 MNO SelectionMechanism

Besides the public Android API that is accessible with the Software De-velopment Kit (SDK), there is also an API, which is located in the pack-age com.android.internal [33] that is not accessible via the SDK.While developing Android applications the android.jar library is ref-erenced. In this library all classes, enumerations, fields, and methodsthat are marked with the annotation @hide, from the internal packageare removed. When the application is launched on a device the libraryframework.jar, which is equivalent to the android.jar, is loaded.However, the framework.jar library provides access with Java reflection[81] to all internal API components from the internal package.

6.2.1 Accessing the Android Internal API

Accessing the internal Android API requires the android.jar library tobe replaced by the framework.jar. This is not immediately workingsince the Android Developer Tools (ADT) plug-in for the Integrated De-velopment Environment (IDE) Eclipse [26] forbids the usage of any in-strument in the internal package by adding an access rule to the javabuild path. Thus, a developer that need to access anything from the internalAPI has to do the following steps: 1) obtain the original Android frame-

78

work, 2) create a custom Android framework, and 3) modify the Eclipseaccess rule.

6.2.2 Obtaining the Original Android Framework

There are two different ways to obtain the original Android framework.One approach is to compile an own framework, due to the fact thatAndroid is an open source mobile OS [7]. However, there exists an-other path for getting the runtime equivalent and being loaded ontothe device at /system/framework. Within [102] the second ap-proach has been chosen, because it is less time consuming. After theframework.jar library is downloaded it has to be extracted by the com-mand jar xf framework.jar. If the extracted folder does not con-tain a file classes.dex the file framework.odex has to be down-loaded from the device as well. This file has to be disassembled withbaksmali.jar [39] by the following command: java -jar baksmaliframework.odex. If errors occur with the suggestion to download moreodex files, missing files have to be downloaded in the same directory withthe framework.odex. This will generate the Android platform relatedclasses as smali files [15] in a folder named out. This folder has to be as-sembled with the command java -jar smali out. The assembled fileis named out.dex and is equivalent to the file classes.dex, which has to beconverted to a jar file using a tool called dex2jar [38]. The resulting jar filehas to be extracted with the command jar xf framework.jar. The ex-tracted folder contains all class files of the internal package in the foldercorresponding to the package name.

6.2.3 Creating a Customized Android Framework

To access the internal API in an IDE, such as Eclipse, a custom frameworkhas to be created, which contains classes andmethods of the internal pack-age. To create the custom framework the Android’s SDK android.jarhas to be extracted. This file is located in the Android’s SDK installationfolder in SDK/platforms/android-X/android.jar, where X is theAPI Level that is targeted at to be customized [6], e.g., level 15 for Android

79

4.0.4. All files being extracted from the original Android framework have tobe copied into the previously extracted folder overwriting already existingfiles. All files in this folder have to be compressed again intoandroid.jarandadded to thebuildpath. Allmethodsof theinternalpackage arenowaccessible.

The original Android framework library has to be replaced with the cus-tom platform by replacing the original android.jar with the one cre-ated. Alternatively, the framework created can be added as new platform.To add a new platform, the entire folder of the original platform has tobe copied. The original android.jar has to be replaced with the cus-tom one. To distinguish this custom framework from the original one,a custom name and a custom API level has to be provided by adaptingthe file build.prop in the platform folder. The value under the entryro.build.version.sdk has to be replaced by a desired number, which repre-sents the API level. The ro.build.version.release value should beexpanded with .extended to indicate that this is a customized platform.

6.2.4 Modifying the Eclipse Access Rule

The last hassle is to modify the Eclipse access rule that prohibits theuse of the internal API. There are different possible ways to achievethis. The first approach is to modify the ADT source code andbuild it, which has not been investigated within this thesis. Anotherway is to modify the ADT’s bytecode. Therefore, the content of thefile com.android.ide.eclipse.adt_*.jar, which is located in thefolder plugins of the Eclipse installation has to be extracted. The valueof * in the file name depends on the ADT version. In the subfoldercom/android/ide/eclipse/adt/internal/project of the ex-tracted folder the file AndroidClasspathContainerInitializer.class has to be opened in an editor that supports non-printable charac-ters. The string com/android/internal/ needs to be replaced with an-other string such as com/android/internax/**. In turn, the folderhas to be compressed with the same name as before. It has to be en-sured that the internal root folder of the archive is the same as the original

80

one, otherwise Eclipse will not recognize it. Finally, the archived folderhas to be renamed to *.jar and the original ADT jar file has to be re-placed with the new one. After restarting Eclipse the internal API is ac-cessible. Another approach worked successfully only with ADT version 21and 22: create a new access rule that allows to use classes out of the pack-age com/android/internal/**. Since the access rule in the subentryandroid.jar cannot bemodified, a new access rule should be created di-rectly below the android platform.

6.2.5 Invoking theMNO SelectionMechanism

Although the Android API does not provide any method to change theMNO, the class GSMPhone exists within the Android 4.0.4 source code[12]. This class contains a public method selectNetworkManually.This is part of the internal package and can be used for the purposeof the automatic and on-demand MNO selection mechanism. The classPhoneFactory provides a method to get different kinds of phone ob-jects. To instantiate a GSMPhone object the method getGsmPhone hasto be invoked [95]. Afterwards, the method selectNetworkManuallycan be invoked by the GSMPhone object with the required parametersOperatorInfo and a Message. OperatorInfo contains the informa-tion about the MNO to select. This includes the operator information,similar to the AT commands case, as alpha numeric long, alpha numericshort, and numeric. Here a selection could be performed, when a newOperatorInfo object with a correct MNO was created. Other valuescan be null or empty. Before this mechanism is usable, two further stepshave to be performed: (1) run the application with a different shareduser ID and (2) run the application under the phone process. To pre-vent a SecurityException that is thrown, when protected intents [8]are sent by the methods invoked, the application has to run either with thesystem user ID: android:sharedUserId="an-droid.uid.system"or with the shared user ID android:sharedUserId="android.uid.phone" [95]. This IDhas to be set in theAndroidManifest.xmlwithinthe manifest-tag.

81

Additionally, the application has to run in the processandroid:process="com.android.phone" to ensure that the in-vocation of getGsmPhone is allowed. This attribute has to be added intothe application-tag. Due to the reason that the shared user ID is used bymore than one applications, all applications have to be signed with thesame certificate [9]. Thus, the application has to be signed with the systemsignature key. To get such a key is to run a custom Read Only Memory(ROM), which provides these certificates [96], e.g., CyanogenMod. Theprocess of signing an application according to [13] is the following: First,the application has to be exported as an unsigned application package.Second, the platform.x509.pem and platform.pk8 have to bedownloaded. Third, these files have to be put into the same folder as theapplication to be signed.

Before the application is sent to the device, it has to be signed withthe tool jarsigner and the command: java -jar signapk.jarplatform.x509.pe m platform.pk8 Application.apksignedA pp.apk. Finally, to sent the application on the device, thepartition has to be remounted from a superuser with read-write privileges.

6.3 Evaluation of theMNO SelectionMechanism

Energy is a critical resource inmobile communications. Furthermore, longdelays are critical for serviceswith a real-timenetwork access, such asphonecall establishment, and they may affect the end-user’s QoE. Thus, havingan on-demand and automaticMNO selection mechanism that consumes alot of energy, or takes a lot of time to switch between MNOs will be prac-tically unusable. Considering that, an evaluation in respect to the energyconsumption and the time needed betweenMNOs switching has been per-formed.

A realistic evaluation of theMNO selectionmechanism has to elaboratemultiple successful SIM card registrations between various MNOs in thesame location. However, the majority of MNOs accept in their networksonly SIM cards issued by them or their roaming partners. Here, a set-upwhere MNOs accept a SIM card on their network was mandatory for the

82

MNO switching process. Thus, two prepaid SIM cards issued by two dif-ferent MVNOs have been used. The SIM cards selected have been chosenwith the criterion that the registration is possible, while in roaming withevery MNO in Switzerland (Orange, Sunrise, and Swisscom).

6.3.1 Time Consumption betweenMNO Switching

There exists a certain SIM card registration time overhead with a SIM cardin roaming due to the fact that the local MNO needs time to authorize theforeign SIM card before accepting it within its network. To minimize theauthentication overhead, the twoSIMcards that have been usedwere regis-tered to each availableMNOprior to themeasurements. Thus, a record foreach SIM card would be present already during MNO switching measure-ments, especially in every Visitor Locator Register (VLR) of each MNO.Thus, the registration time measured had the minimum possible authenti-cation overhead. However, since these SIM cards used during themeasure-ments were prepaid, the available balance authorization overhead couldneither be avoided nor estimated. Furthermore, no guarantee was giventhat the registration process of the SIM card in roaming to a local MNOwas performed with high priority.

The MNO look-up process might take more than 30 s [66]. Thus, theMNO list has been gathered once and their constant availability during themeasurements was assumed. The MNO switching average time betweenthe three available MNOs in Switzerland took place for the following twoscenarios: (a)when the devicewas placed in an urban area inside a buildingand (b)when thedevicewasmoving on a train fromZürich toLucerne. Fora comprehensive test the switching took place between all possible MNOpairs. Thus, one test step consisted of 6 switches. This test was repeated 100times, which led to 600 hops in total. Finally, the time needed for the entiretest was measured and the average time needed per case was calculated.

To examine, if the MNO switching time is correlated to the signalstrength the cell id and the corresponding signal strength have been mea-sured in the device once, immediately after theMNOhop, and then 4moretimes after every 0.5 s. The reason why the signal strength has been cap-

83

Table 6.1: Signal Strength Values

RSSI value Signal strength [dBm]0 -113 dBm or less1 -111 dBm

2-30 -109 dBm to -53 dBm31 -51 dBm or greater99 not known or not detectable

tured 5 times is to confirm that its strength was stable. The correspondingcell id has been tracked to make sure that a possible signal strength oscilla-tion is not due to a change to a different cell. According to [1] the relationbetween theReceivedSignal Strength Indication (RSSI) values and the sig-nal strength in dBm is outlined in Table 6.1.

In those tests implemented the device always tries to register to anMNOuntil the attempt is successful. Themeasurements for case (a)was executed6 times during a weekday in the following time frames (8:00, 10:30, 12:00,14:00, 17:00, 2:00 hours). These time frames have been selected so thatthese measurements undertaken are spread during the day, when the net-work state (e.g., the network load) changes between rush hours. Thus, thedata of the MNO selection consists of total 6 times 100 hops collected indifferent hours, concluding a total number of 3600 hops. The data of theMNO selection for case (b) consists of a total of 100 hops per MNO pair,reaching a total number of 600 hops, resulting in MNO switching times asshown in Figure 6.1.

The first MNO, which appears in the caption below the first set of bars,is always the MNO, where the device was registered first, and the secondMNO is the one that has been switched to. Each bar corresponds to allswitches performed, from the indicated MNO to another. Error bars indi-cate the standard deviation of all measurements. However, there is a mini-mum time needed to complete the 6-step SIMnetwork registration process[82]. Thus, the assumption that the minimum MNO switching time can-

84

Sw −> Su Sw −> Or Su −> Sw Su −> Or Or −> Sw Or −> Su Mean

0.5

1.5

2.5

3.5

4.5

5.5

6.5

7.5

8.5

9.5

10.5

11.5

12.5

MNO switching scenarios

MN

O s

witc

hing

tim

e [s

]

MNO selection time (stable)

MNO selection time (moving)

Figure 6.1: Switching Time Between MNOs

not be in practice lower than the lowest valuemeasured in this results (4.36s) has been taken. Left bars present the average switching time betweenMNOs at the same location; right bars present the average switching timebetween the same MNOs while moving. The last set of bars presents themeanMNO switching time for all cases (a) and (b) in summary. The largestandarddeviation results from largemaximumvalues (cf. Figure 6.2). Dueto the unstable availability ofMNOswhilemoving on a train themaximumMNO switching values appear to be much higher compared to the experi-ments at the same location is some cases. Furthermore, theMNOselectiontime shows a quite unstable behavior in some of the cases, which might berelated to specific MNO’s infrastructure configurations or the current ca-pacity of the connected cell. However, the averageMNO switching time issimilar in both cases showing that theMNOselectionmechanismperformswell in every scenario.

Figure 6.3 correlates theMNO switching time with respective RSSI val-ues (12, 8, 5, 3, and 1) of the newMNO.These numbers on each bar indi-cate the total number of times that the respective signal strength occurredout of the total test hops for scenarios (a) on the left bar and (b) on theright bar. Thus, error-bars represent the standard deviation of these mea-surements and they are larger in cases, where the respective signal strength

85

Sw −> Su Sw −> Or Su −> Sw Su −> Or Or −> Sw Or −> Su Total0

10

20

30

40

50

MNO switching scenarios

MN

O s

witc

hing

tim

e [s

]

Max MNO switching time (stable) Min MONO switching time (stable)Max MNO switching time (moving) Min MNO switching time (moving)

Or: Orange Su: SunriseSw: Swisscom

Figure 6.2: Min and Max Values for the MNO Switching Time

12 8 5 3 10

2

4

6

8

10

12

14

16

18

Received signal strength indication

MNO

sw

itchi

ng ti

me

[s]

MNO switchig time (stable) 3600 measurmentsMNO switching time (moving) 600 measurments

4572319 668 1286042 08 411

Figure 6.3: MNO Switching Time and Signal Strength Correlation

has been captured only a few times. It can be seen that the signal strengthis not affecting significantly the total switching time from oneMNO to an-other. However, more measurements in a controlled environment, wheremore values per signal strength are captured, can lead tomore accurate con-clusions.

Finally, Figure 6.4 presents the correlation of the MNO switching timefor case (a) in those 6 time-slots that the experiment occurred in. It can beseen that the minimum MNO switching time is stable in every time-slot.However, the average and the maximum values are higher in the morningand early in the evening. A possible explanation of this is that the MNO’savailable capacity in a cell is lower when people are moving in the morning

86

08:00 10:30 12:00 14:00 17:00 02:000

5

10

15

20

25

30

35

Time of the day

MNO

sw

itchi

ng ti

me

[s]

Mean time

Min time Max time

Figure 6.4: MNO Switching Time During the Day

08:00 10:30 12:00 14:00 17:00 02:000

50

100

150

200

250

Time of the day

Num

ber o

f tim

es M

NO s

witc

hing

exc

eede

d 10

s

Figure 6.5: MNO Switching Time > 10 s

or after lunch to their offices. Furthermore, Figure 6.5 shows how manytimes the MNO switching time exceeds 10 s each time-slot. The consid-

87

erably higher values around 14:00 hours are most likely due to the highnetwork usage at that time.

6.3.2 MNO Switching Energy Consumption

The power consumption is critical in a mobile system. If a mechanismwould absorb a large amount of available energy resources within a fewnetwork hops, the MNO switching mechanism would not be usable inpractice. Hence, a detailed evaluation of the power consumption has beenmade. To measure the power consumption, the battery level was deter-mined before the test run and after the test had been performed according[11]. The difference of these levels lead to the final battery consumption inpercentage of the battery energy. The assumption is that the battery healthis in ideal condition. This assumption is appropriate, because the deviceof those measurements and its battery was new and experiments were per-formed in an ideal temperature for the battery [63]. This procedure wasapplied, since currently no Android application exists, which can measurethe real battery capacity, or no application is in place, which measures theconsumed energy per application accurately.

E = 6.11 Wh · 3600 s = 21996 J (6.1)

Ptot =E(J) · Batteryusedtexperiment(s)

=0.14 · 21996 J

5696 s= 0.5406 W (6.2)

The total energy of the battery of the device used is 6.11 Wh accordingto the manufacturer. Thus, the total energy that a new battery can produceis 21996 J (cf. Equation 6.1). During the test the display of the device re-mained turned off, as well as every irrelevant to the experiment process wasdisabled. In the test case (a), where the location was stable, the measuredbattery consumption was 14% and the total duration of the measurementswas 5696 s. This corresponds to the energy consumption of 3079.44 J. Toreach the total power for the MNO switching mechanism the consumedenergy has to be divided by the total experiment time. However, this cal-

88

Table 6.2: Mobile Device Characteristics

Consumption 2Gmax talk 2G power 3Gmax talk 3G powertime [h] [W] time [h] [W]

Voice service 18.33 0.3333 8.67 0.705Stand-by 710.00 0.0086 610.00 0.010

culation includes also the energy needed for capturing the signal strengthand the cell IDfive times for eachMNOswitchingmeasurement. This over-head does not affect significantly the result concerning theMNOswitchingmechanism, because the energy consumed on this process is small com-pared to the energy demanded by the MNO switching process. Thus, theresults show a total 0.5406 W consumed power for the MNO switchingmechanism (cf. Equation 6.2). The same test has been performed in testcase (b), while moving from Zürich to Lucerne by train. The test lasted7404 s and 22% of the battery was consumed. This corresponds to a totalpower consumption of 0.6536 W. By comparing these values of both tests,it is evident that in the case themobile device is notmoving the power con-sumption of the MNO switching mechanism appears to be approximately17.3% lower than the power consumption, when device is moving, mostlikely due to the handover energy consumption that can not be isolated.The MNO selection mechanism power value is comparable to the powerconsumption of the talk mode in 3G networks, which is calculated con-sidering manufacturer’s maximum stand-by and talk-time in 2G and 3Gnetworks, as shown within Table 6.2. Thus, the power consumption of anMNO switching compared to other services, such as a hone call is shownin Table 6.3.

6.4 AbaCUS E2E Calling-time

To estimate how long it takes to be established a phone call in AbaCUS,the Au² server, which return the MNO that will terminate a call, has beenmocked in a local network. A phone call according to the AbaCUS proto-

89

Table 6.3: Power Consumption Evaluation

Process Power [W]Talk 3G 0.7050

MNO selection moving 0.6536MNO selection stable 0.5406

Talk 2G 0.3333

col (cf. Figure 6.6) took place. The total time from the initiation of the calluntil the callee’s phone was ringing was measured. The Au² mock serversimulated that Swisscom always wins. The test call has been done 30 times.Ten times the callee’s device had to switch from Orange to Swisscom andten times it had to switch from Sunrise to Swisscom. Additionally, in tenmore cases the time was measured when the callee’s MNO did not haveto change, because it already was Swisscom. These results are summarizedin Figure 6.7. The different bars indicate the average time consumed for acall termination, where error bars are representing the standard deviationof all measurements. The average time in the case where no MNO changehappened is still comparable with the normal dialing case where AbaCUSis not involved. Thus, is shown that for the AbaCUS protocol the averagetime of a call establishment process mainly depends on the dialing time.The difference of the calling time that a MNO switching is involved, tocases where the MNO has not been switched, corresponds to the averageMNO selection time that has been evaluated in Subsection 6.3.1 A above.The first MNO in the caption below the bars indicates to which MNO thecallee’s device was registered before the call. The second MNO is the win-ning one, which the callee’s device had to switch to. The third bar meansthat the callee’s device was already registered to the winning MNO, andthe last bar shows the average time when dialing with the traditional dialerwithout AbaCUS.

90

Caller& Callee& Au2&

1.&Service&Request&

2.&Change&MNO&

3.&OK&

5.&Ready&to&call&

Figure 6.6: AbaCUS Service Request

Or −> Sw Su −> Sw No Change Normal dialing0

5

10

15

20

25

Calling scenarios

Cal

l est

ablis

hmen

t tim

e [s

] Or: Orange Su: Sunrise Sw: Swisscom

Figure 6.7: AbaCUS E2E MNO Calling-time

6.5 Chapter Summary

At [102] a prototype of an automatic and on-demand MNO selectionmechanism for the Android platform had been designed and implemented.Theevaluationof themechanismshowed that the time consumption aswellas the communication to the server was negligibly low. For both scenarios(the stable mobile user in an urban location or moving in a train) it wasshown that the MNO switching time is independent of those MNOs in-

91

volved and on average it is expected to be below 10 s. Secondly, the powerconsumption of anMNO switching is in the same dimension of the powerneeded,when aphone call is performed. Todemonstrate key requirements,an Android application has been implemented, which makes use of theMNO selection mechanism. An automatic and on-demand MNO selec-tionmechanism is proven to be a realistic and feasible requirement for Aba-CUS.

Concluding, the existing knowledge on how to use the internal AndroidAPI has been combined to gain access to methods in connection with theGSM modem, which the open Android API does not provide. Thus, anautomatic and on-demandMNO selection mechanism has been designed,implemented, and tested successfully. However, this workaround is notthe best method, since the internal API is not listed and also may changein the future. Nevertheless, this thesis showed that such a mechanism isdoable and realistic from an energy and time perspective, especially whenit is compared to other type of services, such as the traditional phone calls.Thus, the source code of the developed mechanism in [102] can be foundat [3]. However, since such a mechanism could be used in the mobile ter-mination rates monopoly break or it could open the window for additionalservices, such as on-demand QoS-guaranteed services, an automatic andon-demand MNO selection mechanism should be published in the openAPI by all smart phone vendors.

92

7Conclusions

T his thesis has proposed and investigated several key aspects in-volved in allowing a fully-competitive MTR market in countries

where the CPP principle is applied. The path to select a MNO to providethe mobile termination service in a liberal MTRs market that is proposedin this thesis, is to select theMNOwhichmaximizes caller’sQoE.Thus, it isessential to determineQoE in away that can be estimated/predicted beforea service is provided. On this path the generic Axiomatic QoEmodel AQXis developed here. AQX novelty is (1) that it can consider multiple and di-verse economic and technical parameters, and (2) that estimates a per userpersonalized QoE of a service, considering influence and importance fac-tors of each variable affecting QoE.The validity of this generic QoEmodelhas been challenged in the VoIP scenario where the comparison with stateof the art QoE models showed that AQX outperforms.

QoE estimation of the mobile termination service considers one set oftechnical and one set of economic parameters. On the technical side (a)the preferred network access priority, and (b) the sound quality is selectedby the caller. On the economic side the influence and importance factors of(1) set-up cost, and (2) charging rate of a call are considered. MNOs thatare competing to provide the mobile termination service proposes a set of

93

values for the economic variables and the one that maximized caller’s QoEis selected to provide the service and collect the MTR.

Each stakeholder involved in the MTRs market has been briefly exam-ined and all incentives, and actions needed, from each stakeholder has beenidentified. It was shown that liberating the MTRs market is essential since(1) end-users will benefit from better and/or cheaper services, (2) MNOsare able to increase their infrastructure utilization and possibly get access tonew revenue streams, and (3) the regulation and governmental authoritieswill be able to liberate a traditionally consider monopolistic market. Fi-nally, a prototype automatic and on-demand MNO selection mechanismassumed inAbaCUS, has been implemented and proved to be usable in reallife scenarios considering (1) energy consumption, and (2) time consump-tion per MNO hop. Thus, a compete MTRs environment is illustrated inthis thesis.

7.1 QoE Formalization

The QoE model proposed in this thesis has been compared to other stateof the art QoE models in the VoIP scenario and proved to capture end-users QoE more accurately. AQX is a generic model that may considermultiple technical variables, such as bandwidth, latency, and packet-loss,as well as non technical parameters, such as price. However, AQX is notservice-specific model. Thus, it is used in this thesis as a bidding metric inan auction that selects whichMNOwill maximize a caller’s QoE, consider-ing the service characteristics (network access priority, sound quality) andthe price of the set-up andMTR cost.

7.2 AbaCUS

AbaCUS supports an auction procedure for every call separately, establish-ing like that a dynamic, live, and on-demand competition in the mobiletermination rates market. QoS-guaranteed services are also supported inAbaCUS. Furthermore, the caller has to set only his preference withoutany knowledge on MNOs MTRs charging policies. MNOs can act inde-

94

pendently, since it is optional to participate in AbaCUS auctions, or adoptother approaches. The AbaCUS auction is designed in a way that MNOscan profit by using the system (1) in congested network scenarios by of-floading traffic while increasing the requested values of TeR-C’s variables,and (2) when there are enough unused resources and the TeR-C variablesare offered in lower values to “attract” callers.

Finally, regulation authorities in an AbaCUS moderated market willhave an observer role instead of their juristic role today. Less regulationdemands will result in a competitive market with all those benefits thatthe AbaCUS approach implies. Given the survey results achieved, it isshown that AbaCUS is an approach with a strong potential to be acceptedbyMNOs subscribers, since themajority of subscribers are positive towarda potential price andQoS correlation. Thus, an accuratemechanism to pro-tect dissatisfied users that paid, but never perceived the expected QoS, isneeded.

7.3 FutureWork

This thesis proposed a solution that breaks theMTRsmonopoly in theCPPenvironment. However, this is a market-established monopoly since thebeginning ofmobile telecommunication. Thus, the chance of suchmonop-olistic environment to change is relatively low. However, this thesis consid-ers new networks and technologies to show that when the parameters of aproblem change, past solutions should be reevaluated. Thus,MTRs shouldnot be considered a de-facto monopoly anymore, and future services, suchas VoIP services overmobile networks should consider amore competitiveenvironment in the principles that AbaCUS proposes.

It is essential to examine more variables that affect callers QoE in themobile termination service market. Also, define via a more extensive mar-ket analysis new potential QoS-Cs, might reveal QoE-related variables thatincrease further the social welfare andMNOs infrastructure utilization.

The exploration of further auction types to be used by Au² could fol-low. The efficiency of AbaCUS in othermarkets, like theVoIP and the fixedtelephony market, might also be examined. Thus, regulators in future mo-

95

bile networks shouldput the effort towards liberalizationofmarket-definedmonopolies, such as spectrum access, and maybe introduce AbaCUS prin-ciples at a lower layer. E.g., on the physical layer andMedia Access Control(MAC) layer by introducing more flexible spectrum access, such as cogni-tive radio approaches.

96

References

[1] Digital Cellular Telecommunications System (Phase 2+),Universal mobiletelecommunications system (umts): At command set for user equipment (ue)(3gpp ts 27.007 version 8.5.0 release 8), October 2008.

[2] 3rd Generation Partnership Project (3GPP), Technical specification groupterminals, at command set for user equipment (ue) (release 6), June 2003.

[3] AbaCUS, Auction-based charging and user-centric system, URL: http://www.abacusproject.eu/, March 2015.

[4] Amazon,Amazon elastic compute cloud (amazon ec2), URL:http://aws.amazon.com/ec2/, March 2015.

[5] ANATEL, Agencia nacional de telecomunicacoes, URL: http://www.anatel.gov.br/hotsites/anatelgrandeseventos/en/home/index.html, March 2015.

[6] Android, Api level, URL: http://developer.android.com/guide/topics/manifest/uses-sdk-element.html#ApiLevels, March2015.

[7] , Downloading and building android, URL: http://source.android.com/source/building.html, March 2015.

[8] , Intents, URL: http://developer.android.com/reference/android/content/Intent.html, March 2015.

[9] ,Manifest, URL: http://developer.android.com/guide/topics/manifest/manifest-element.html, March 2015.

[10] ,Market share, URL: http://tiny.uzh.ch/cT, March 2015.

[11] , Monitoring the battery level and charging state,URL: http://developer.android.com/training/monitoring-device-state/battery-monitoring.html, March2015.

[12] , Source code, URL: http://grepcode.com/snapshot/repository.grepcode.com/java/ext/com.google.android/

97

android/4.0.4_r2.1/, March 2015.

[13] , Your (my) way, URL: http://omri.org.il/2012/05/25/android-your-my-way/, March 2015.

[14] Nordic Competition Authorities, Telecompetition - towards a sin-gle nordic market for telecommunication services?, URL: http://www.konkurransetilsynet.no/iKnowBase/Content/395667/TELECOMPETITION.PDF, March 2015.

[15] Backsmali/Smali, Manager, URL: http://forum.xda-developers.com/showthread.php?t=2311766, March 2015.

[16] K. Binmore and D. Harbord, Bargaining over fixed-to-mobile terminationrates: countervailing buyer power as a constraint on monopoly power, Journalof Competition Law and Economics 1 (2005), no. 3, 449–472.

[17] TIM (Brazil), National roaming, URL: http://www.tim.com.br/sp/para-empresas/pequenas-e-medias-empresas/cobertura-e-roaming/nacional, March 2015.

[18] UPC cablecom, Home-user broadband plans, URL: http://www.upc-cablecom.ch/en/internet/products/, March2015.

[19] P. Casas, M. Seufert, and R. Schatz, Youqmon: A system for on-line monitor-ing of youtube qoe in operational 3g networks, SIGMETRICS Perform. Eval.Rev. 41 (2013), no. 2, 44–46.

[20] C. Chun-Lung, Dual sim (subscriber indentity module) card connector: Uspatent 6,623,305, Spetmeber 2003.

[21] Cisco, Quality of service for voice over ip, URL: http://www.cisco.com/c/en/us/td/docs/ios/solutions_docs/qos_solutions/QoSVoIP/QoSVoIP.html, July 2014.

[22] Federal Communications Commission (ComCom), Switzerland, URL:http://www.comcom.admin.ch/index.html?lang=en, March2015.

[23] E. David, R. Azoulay-Schwartz, and S. Kraus,An english auction protocol formulti-attribute items, Agent-Mediated Electronic Commerce IV. DesigningMechanisms and Systems ( J. Padget, O. Shehory, D. Parkes, N. Sadeh, andW. Walsh, eds.), Lecture Notes in Computer Science, vol. 2531, SpringerBerlin Heidelberg, 2002, pp. 52–68 (English).

98

[24] R. Dewenter and J. Haucap, The effects of regulation mobile terminationrates for asymmetric networks, European Journal of Law and Economics 20(2005), 185–197.

[25] MNO Directory, Mno directory, URL: http://www.mnodirectory.com/europe-subs.html, March 2015.

[26] Eclipse, Ide, URL: http://www.eclipse.org/, March 2015.

[27] Econstor, The regulation of national roaming, URL: https://www.econstor.eu/dspace/bitstream/10419/52213/1/672585162.pdf,March 2015.

[28] Eurocomms,Mnos must reform tariffs, URL: http://tiny.uzh.ch/cR,March 2015.

[29] T. Fadell,Dynamic carrier selection: Us patent 7,885,654, February 2011.

[30] Federal Communications Commission (FCC), U.s.a., URL: http://www.fcc.gov/, March 2015.

[31] M. Fiedler, T. Hossfeld, and P. Tran-Gia, A generic quantitative relation-ship between quality of experience and quality of service, Network, IEEE 24(2010), no. 2, 36–41.

[32] Forum, How to talk to the modem with at commands, URL: http://forum.xda-developers.com/showthread.php?t=1471241, March2015.

[33] , Using internal (com.android.internal) andhidden (@hide) apis [part 1, introduction], URL:https://devmaze.wordpress.com/2011/01/18/using-com-android-internal-part-1-introduction/, March2015.

[34] Linux Foundation, netem | the linux foundation, URL: http://www.linuxfoundation.org/collaborate/workgroups/networking/netem, July 2014.

[35] Apache Friends, Xampp installers and downloads for apache friends, URL:https://www.apachefriends.org/de/index.html, July 2014.

[36] E.W. Fulp,M.Ott, D. Reininger, andD.S. Reeves, Paying for qos: an optimaldistributed algorithm for pricing network resources, Quality of Service, 1998.(IWQoS 98) 1998 Sixth InternationalWorkshop on,May 1998, pp. 75–84.

99

[37] GoGrid, Cloud provider, URL: http://www.gogrid.com/, March2015.

[38] Google, Dex2jar, URL: http://code.google.com/p/dex2jar/downloads/list, March 2015.

[39] , Smali/baksmali, URL: https://code.google.com/p/smali/, March 2015.

[40] Gsmarena,Mobile communication technologies per country, URL: http://www.gsmarena.com/network-bands.php3, March 2015.

[41] D. Harbord and M. Pagnozzi, Network-based price discrimination and ‘bill-and-keep’ vs. ‘cost-based’ regulation of mobile termination rates, Review ofNetwork Economics 9 (2010), no. 1, 1–44.

[42] T. Hoßfeld, D. Hock, P. Tran-Gia, K. Tutschku, and M. Fiedler, Testingthe iqx hypothesis for exponential interdependency between qos and qoe ofvoice codecs ilbc and g.711, Technical report, Institute of Computer Science,2008.

[43] T. Hoßfeld, P. Tran-Gia, and M. Fiedler, Quantification of quality of expe-rience for edge-based applications, Managing Traffic Performance in Con-verged Networks (Lorne Mason, Tadeusz Drwiega, and James Yan, eds.),LectureNotes inComputer Science, vol. 4516, SpringerBerlinHeidelberg,2007, pp. 361–373 (English).

[44] Te-Yuan Huang, Polly Huang, Kuan-Ta Chen, and Po-Jung Wang, Couldskype be more satisfying? a qoe-centric study of the fec mechanism in aninternet-scale voip system, Network, IEEE 24 (2010), no. 2, 42–48.

[45] Hulu,Hulubandwidth recommendations, URL:http://www.hulu.com/support/article/197541, January 2014.

[46] ITU, Cpp in europe, URL: http://www.itu.int/osg/spu/ni/fmi/regulation/Survey2.pdf, March 2015.

[47] , Termination cost regulation, URL: http://www.itu.int/net/itunews/issues/2010/03/20.aspx, March 2015.

[48] ITU-D, Mobile termination rates and roaming, URL: https://www.itu.int/ITU-D/finance/work-cost-tariffs/events/tariff-seminars/China-10/pdf/Session7_MTR_ErikoHondo.pdf, March 2015.

100

[49] , Mobile voice service and termination rates, URL: http://www.itu.int/ITU-D/finance/work-cost-tariffs/events/tariff-seminars/elsalvador/pdf/Sesion4_Roaming_JThompson-en.pdf, March 2015.

[50] , Roaming and mobile termination rates, URL: http://www.itu.int/ITU-D/finance/work-cost-tariffs/events/tariff-seminars/elsalvador/pdf/Sesion3-Omar_de_Leon_itinerancia-en.pdf, March 2015.

[51] ITU-T,Methods for subjective determination of transmission quality, ITU-TRecommendation P.800, June 1998.

[52] , End-user multimedia qos categories, ITU-T RecommendationG.1010, November 2001.

[53] ,Thee-model, a computationalmodel for use in transmission planning,ITU-T Recommendation G.107, March 2003.

[54] ,One-way transmission time, ITU-T Recommendation G.114,May2003.

[55] , Mean opinion score (mos) terminology, ITU-T RecommendationP.800.1, November 2006.

[56] , Subjective evaluation of conversational quality, ITU-TRecommen-dation P.805, October 2007.

[57] , E-model tutorial, URL: http://www.itu.int/ITU-T/studygroups/com12/emodelv1/tut.htm, June 2014.

[58] ,Transmission impairments due to speech, ITU-TRecommendationG.113, July 2014.

[59] , International telecommunication union - telecommunication stan-dardization sector (itu-t), URL: http://www.itu.int/en/ITU-T,March 2015.

[60] Joyent,Node.js, URL: http://nodejs.org/about/, July 2014.

[61] A. Jurgelionis, J. Laulajainen, M. Hirvonen, and A.I. Wang, An empiricalstudy of netem network emulation functionalities, Computer Communica-tions and Networks (ICCCN), 2011 Proceedings of 20th InternationalConference on, July 2011, pp. 1–6.

101

[62] M. Kaku, Physics of the future: How science will shape human destiny and ourdaily lives by the year 2100, Anchor, 2012.

[63] K. Kazuhiko Takeno, M. Ichimura, K. Takano, and J. Yamaki, Influence ofcycle capacity deterioration and storage capacity deterioration on li-ion batter-ies used inmobile phones, Journal of Power Sources 142 (2005), no. 1-2, 298– 305.

[64] K. Knopper, Knoppnix - live linux filesystem on cd, URL: http://www.knopper.net/knoppix/index-en.html, July 2014.

[65] K.U.R. Laghari and K. Connelly, Toward total quality of experience: A qoemodel in a communication ecosystem, Communications Magazine, IEEE 50(2012), no. 4, 58–65.

[66] S. Liniger, Implementation of an automatic, on-demand mobile network oper-ator (mno) selection mechanism on android devices, 8 2013, Bachelor thesis.

[67] S. Littlechild,Mobile termination charges: Calling party pays versus receivingparty pays, Telecommunications Policy 30 (2006), no. 5-6, 242–277.

[68] S.C. Littlechild, “Mobile Termination Charges: Calling Party Pays versus Re-ceiving Party Pays” (original and revised versions), Cambridge Working Pa-pers in Economics 0426, Faculty of Economics, University of Cambridge,April 2004.

[69] Number Portability Lookup, Nls cost, URL: http://www.numberportabilitylookup.com/pricing?s=, March 2015.

[70] Pay Per Lough, Pay-per-laugh: the comedy club that charges punters havingfun, URL: http://tiny.uzh.ch/nH, July 2015.

[71] GuilhermeSperbMachado andBurkhardStiller, Investigations of an sla sup-port system for cloud computing (slacc), Praxis der Informationsverarbeitungund Kommunikation (PIK) 34 (2011), no. 2, 80–86.

[72] R.K.P.Mok, E.W.W.Chan, andR.K.C.Chang,Measuring the quality of expe-rience of http video streaming, IntegratedNetworkManagement (IM), 2011IFIP/IEEE International Symposium on, May 2011, pp. 485–492.

[73] Motorola, Realistic lte performance - from peak rate to subscriber experience,URL: http://www.motorolasolutions.com/web/Business/_Documents/static%20files/Realistic_LTE_Experience_White_Paper_FINAL.pdf, January 2014.

102

[74] Keloo Network, 10000 general knowledge questions and answers, URL:http://www.keloo.ro/doc/10000_intrebari.pdf, July 2014.

[75] Nielsen,Average u.s. smartphone data usage up 89% as cost per mb goes down46%, URL: http://tiny.uzh.ch/cS, March 2015.

[76] , Smartphone penetration rate by age group,URL: http://www.marketingcharts.com/online/smartphone-penetration-rate-by-age-group-in-q2-45774/,March 2015.

[77] , Smartphones penetration, URL: http://www.nielsen.com/us/en/newswire/2010/smartphones-to-overtake-feature-phones-in-u-s-by-2011.html, March 2015.

[78] Novascola, Quizfragen und antworten, URL: http://www.novascola.ch/quizfragen-und-antworten, July 2014.

[79] OFCOM, Federal office of communications, URL: http://www.bakom.admin.ch/, May 2015.

[80] Opus, Voice coding with opus, URL: http://www.opus-codec.org/presentations/opus_voice_aes135.pdf, July 2014.

[81] Oracle, Java reflection, URL: http://docs.oracle.com/javase/tutorial/reflect/, March 2015.

[82] K. Pahlavan and P. Krishnamurthy, Principles of wireless networks, ch. 7,Prentice Hall PTR, 2002, Paragraph 7.3.

[83] H.N. Park,Method for verifying personalization in mobile radio terminal: Uspatent 6,138,005, October 2000.

[84] Rackspace, Cloud provider, URL: http://www.rackspace.com/,March 2015.

[85] P. Reichl, S. Egger, S. Möller, K. Kilkki, M. Fiedler, T. Hossfeld, C. Tsiaras,and A. Asrese, Towards a comprehensive framework for qoe and user be-havior modelling, 7th International Workshop on Multimedia Experience,(QoMEX 2015), May 2015, pp. 1–6.

[86] P. Reichl, S. Egger, R. Schatz, and A. D’Alconzo,The logarithmic nature ofqoe and the role of the weber-fechner law in qoe assessment, Communications(ICC), 2010 IEEE International Conference on, May 2010, pp. 1–5.

103

[87] P. Reichl and F. Hammer, Charging for quality-of-experience: Anew paradigm for pricing ip-based service, In Proceedings of SecondISCA/DEGA Tutorial and Research Workshop on Perceptual Quality ofSystems, September 2006.

[88] P. Reichl, F. Hammer, andM. Balinova,Conversational interactivity in slow-motion: A comparison of phone dialogues and opera scenes, Proc. 2nd Interna-tional Conference on Sound andMusic Computing (SMC 2005), Salerno,November 2005.

[89] P. Reichl, P. Maillé, and P. Zwickl, On the Fixpoint Problem of QoE-basedCharging (Invited Paper), Proc. The 6th International Conference on Per-formance Evaluation Methodologies and Tools, Cargèse, France, October2012.

[90] A. Richards, M. Antoniades, V. Witana, and G. Rogers,Mapping user levelqos from a single parameter, 1998.

[91] T.E. Rockoff and M. Groves, Design of an internet-based system for remotedutch auctions, Internet Research 5 (1995), no. 4, 10–16.

[92] J. Schiller,Mobile communications, 2 ed., ch. 4, p. 114, Pearson Education,2003.

[93] Jean-Marc Seigneur, Xavier Titi, and Tewfiq El Maliki, Towards mo-bile/wearable device electrosmog reduction through careful network selection,Proceedings of the 1st AugmentedHuman InternationalConference (NewYork, NY, USA), AH ’10, ACM, 2010, pp. 21:1–21:5.

[94] I. Sesia, S.Toufik andM.Baker,Frontmatter, pp. i–xxxv, JohnWiley&Sons,Ltd, 2009.

[95] Stackoverflow,Can a telephony.phone object be instantiated through the sdk?,URL: http://tiny.uzh.ch/cQ, March 2015.

[96] , Can a telephony.phone object be instantiated through the sdk?,URL: https://github.com/CyanogenMod/android_build/tree/gingerbread/target/product/security, March 2015.

[97] VaattovaaraT.,Analysis ofmodem integration in open source smartphone plat-forms, Master’s thesis, Oulu University of Applied Sciences, 3 2011.

[98] TATA, Wanem: The wide area network emulator, URL: http://wanem.sourceforge.net/, July 2014.

104

[99] Hellenic Telecommunications and Posts Commission (EETT), Greece,URL: http://www.eett.gr/opencms/opencms/EETT_EN/index.html, March 2015.

[100] TRAI, Telecom regulatory authority of india, URL: http://www.trai.gov.in/, March 2015.

[101] C. Tsiaras, Survey, URL: http://web.trictrac.com/servlet/trictrac?e=GdIDDD8zdppIiM8Ldp, March 2015.

[102] C. Tsiaras, S. Liniger, and B. Stiller, An automatic and on-demand mnoselection mechanism, Network Operations and Management Symposium(NOMS), 2014 IEEE, May 2014, pp. 1–8.

[103] ,Automatic and on-demand mobile network operator (mno) selectionmechanism demonstration, Network Operations and Management Sympo-sium (NOMS), 2014 IEEE, May 2014, pp. 1–2.

[104] C. Tsiaras, M. Rösch, and Stiller, Voip-based calibration of the dqx model,IFIP Networking 2015, May 2015, pp. 1–9.

[105] C. Tsiaras, A. Sehgal, S. Seeber, D. Dönni, B. Stiller, J. Schonwalder,and G.D. Rodosek, Towards evaluating type of service related quality-of-experience on mobile networks, Wireless and Mobile Networking Confer-ence (WMNC), 2014 7th IFIP, May 2014, pp. 1–8.

[106] C. Tsiaras, A. Sehgal, S. Seeber, D. Dönni, B. Stiller, J. Schönwälder,and G.D. Rodosek, Towards evaluating type of service related quality-of-experience on mobile networks, Wireless and Mobile Networking Confer-ence (WMNC), 2014 7th IFIP, May 2014, pp. 1–8.

[107] C. Tsiaras and B. Stiller, Challenging the monopoly of mobile terminationcharges with an auction-based charging and user-centric system (abacus), Net-worked Systems (NetSys), 2013Conference on,March 2013, pp. 110–117.

[108] ,Adeterministic qoe formalization of user satisfaction demands (dqx),Local Computer Networks (LCN), 2014 IEEE 39th Conference on, Sept2014, pp. 227–235.

[109] J. Uberti and S. Dutton, Webrtc plugin-free realtime communication, URL:http://io13webrtc.appspot.com/, July 2014.

[110] E. Utherland, International roaming charges: over-charging and competitionlaw, Telecommunications Policy 25 (2001), no. 1-2, 5–20.

105

[111] M. Venkataraman,M. Chatterjee, and Siddhartha Chattopadhyay, Evaluat-ing quality of experience for streaming video in real time, Global Telecommu-nications Conference, 2009. GLOBECOM 2009. IEEE, Nov 2009, pp. 1–6.

[112] W3C,Webrtc 1.0: Real-time communication between browsers, URL: http://dev.w3.org/2011/webrtc/editor/webrtc.html, July 2014.

[113] WebRTC, Opus in low bandwidth conditions, URL: https://code.google.com/p/webrtc/issues/detail?id=2025, July 2014.

[114] Wikipedia, Apple sim card, URL: http://en.wikipedia.org/wiki/Apple_SIM, January 2014.

[115] , Dual-sim, URL: http://en.wikipedia.org/wiki/Dual_SIM, March 2015.

106

Other Author Publications

Journal Article

• A.G. Tasiopoulos, C. Tsiaras, and S. Toumpis: Optimal and achievablecost/delay tradeoffs in delay-tolerant networks, Elsevier, Computer Net-works, Vol. 70, No. 9, September 2014, ISSN 1389-1286, pp 59-74. URL:http://dx.doi.org/10.1016/j.comnet.2014.05.006.

Conference Papers

• C. Tsiaras, L. Hobi, F. Hofstetter, S. Liniger, and B. Stiller: parkITs-mart: Minimization of Cruising for Parking, 24th International Conferenceon Computer Communications and Networks (ICCCN 2015), “ICCCN2015”, Las Vegas, Nevada, USA, August 2015, pp 1-8.

• D. Dönni, G.S. Machado, C. Tsiaras, and B. Stiller: Schengen Routing: ACompliance Analysis, 9th International Conference on Autonomous Infras-tructure, Management and Security (AIMS 2015), “AIMS 2015”, Gent,Belgium, May 2015, pp 1-12.

• A.G. Tasiopoulos, C. Tsiaras, and S. Toumpis: On the Cost/Delay Trade-off of Wireless Delay Tolerant Geographic Routing, IEEE International Sym-posium (WoWMoM 2012), “World of Wireless, Mobile and MultimediaNetworks”, San Francisco, California, U.S.A., June 2012, pp 1-9. URL:http://dx.doi.org/10.1109/WoWMoM.2012.6263706.

• C. Tsiaras, S. Liniger, and B. Stiller: An Automatic and On-demandMNO Selection Mechanism, IEEE/IFIP Network Operations andManagement Symposium (NOMS 2014), “Management in a Soft-ware Defined World”, Krakow, Poland, May 2014, pp 1-8. URL:http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=6838236&isnumber=6838210.

• C. Tsiaras, S. Liniger, and B. Stiller: Automatic and On-demand MobileNetwork Operator (MNO) Selection Mechanism Demonstration, IEEE/IFIP

107

NetworkOperations andManagement Symposium (NOMS2014), “Man-agement in a Software Defined World”, Krakow, Poland, May 2014, pp 1-2. URL: http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=6838270&isnumber=6838210.

• C. Tsiaras, M. Rösch, and B. Stiller: VoIP-based Calibration of the DQXModel, 14th event of the series of International Conferences on Network-ing (IFIP Networking 2015), “IFIP Networking 2015”, Toulouse, France,May 2015, pp 1-9. URL: https://files.ifi.uzh.ch/CSG/staff/tsiaras/Extern/Publications/DQX-VoIP.pdf.

• C. Tsiaras, A. Sehgal, S. Seeber, D. Dönni, B. Stiller, J. Schön-wälder, and G.D. Rodosek: Towards Evaluating Type of Service Re-lated Quality-of-Experience on Mobile Networks, 7th IFIP Wireless andMobile Networking Conference (WMNC 2014), “Wireless and MobileCommunications and Networks”, Vilamoura, Portugal, May 2014, pp 1-8. URL: http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=6878848&isnumber=6878843.

• C. Tsiaras, and B. Stiller: A Deterministic QoE Formalization of User Satis-faction Demands (DQX), 39th IEEE Conference on Local Computer Net-works (LCN), “Local Computer Networks”, Edmonton, Canada, Septem-ber 2014, ISBN 978-1-4799-3779-0, pp 227-235. URL: http://dx.doi.org/10.1109/LCN.2014.6925776.

• C. Tsiaras, and B. Stiller: Challenging the Monopoly of Mobile Termina-tion Charges with an Auction-based Charging and User-centric System (Aba-CUS), Networked Systems (NetSys) 2013, “Innovations in the Realm ofNetworked Systems”, Stuttgart, Germany, March 2013, pp 110-117. URL:http://dx.doi.org/10.1109/NetSys.2013.13.

• C. Tsiaras, M. Waldburger, G.S. Machado, A. Vancea, and B. Stiller: TheDesign of a Single Funding Point Charging Architecture, 18th EUNICE Con-ference on Information and Communications Technologies, “Informationand Communication Technologies”, Budapest, Hungary, August 2012,pp 148-160. URL:http://link.springer.com/chapter/10.1007%2F978-3-642-32808-4_14.

Workshop Paper

• P. Reichl, S. Egger, S. Möller, K. Kilkki, M. Fiedler, T. Hoßfeld, C. Tsiaras,and A. Asrese: Towards a Comprehensive Framework for QoE and User Be-havior Modeling, 7th International Workshop on Multimedia Experience,“QoMEX 2015”, Costa Navarino, Messinia, Greece, May 2015 pp 1-6.

108

Technical Reports

• B. Stiller, D. Dönni, F. Hecht, A. Lareida, G.S. Machado, C. Tsiaras, A.Vancea, and M. Waldburger: Internet Economics VII; IFI Technical Report2013.01, Zürich, Switzerland, April 2013, pp 1-191. URL: http://www.csg.uzh.ch/teaching/hs12/inteco/extern/IFI-2013.01.pdf.B. Stiller, K. Farkas, F. Hecht, G.S. Machado, P. Poullie, F. Santos, C.Tsiaras, A. Vancea, and M. Waldburger: Internet Economics VI; IFI Tech-nical Report, No. IFI-2012.02, April 2012, pp 1-147. URL: http://www.csg.uzh.ch/teaching/hs11/inteco/extern/IFI-2012.02.pdf.

109

110

Appendix

A.1 Formal Proofs of AQX Equations

This appendix displays complete formal proofs of AQX equations fromChapter 3.

Equation 3.10.

ea (x0) := e0(3.9)=⇒

⇒ e0 = h · e−λ·xm0 + μ ⇐⇒

⇔ e0 − μh

= e−λ·xm0 ⇐⇒

⇔ ln(e0 − μ

h

)= −λ · xm0 ⇐⇒

⇔ λ = x−m0 ln

(h

e0 − μ

)Equation 3.15.

ei(x0 − δ′

)= e0 − ε′

(3.6,3.12)⇐⇒

⇔ h ·(1− e−λ·(x0−δ′)m

−)+ μ = e0 − ε′ ⇐⇒

⇔ 1− e−λ·(x0−δ′)m−

=e0 − μ − ε′

h⇐⇒

⇔ e−λ·(x0−δ′)m−

=h− e0 + μ + ε′

h(3.7)⇐⇒

⇔ e−(

x0−δ′x0

)m−·ln

(h

h−e0+μ

)=

h− e0 + μ + ε′

h(3.2)⇐⇒

⇔ eln(

h−e0+μh

)(x0−δ′x0

)m−

=M− e0 + ε′

h⇐⇒

111

⇔ h− e0 + μh

(x0−δ′x0

)m−

=M− e0 + ε′

h(3.2)⇐⇒

⇔ M− e0h

(x0−δ′x0

)m−

=M− e0 + ε′

h⇐⇒

⇔(x0 − δ′

x0

)m−

= log(M−e0h )

(M− e0 + ε′

h

)⇐⇒

⇔ m− = log( x0−δ′x0

) [log(M−e0h )

(M− e0 + ε′

h

)]⇐⇒

⇔ m− =

ln(

ln M−e0+ε′h

ln M−e0h

)ln x0−δ′

x0

Equation 3.19.

ea (x0 + δ) = e0 − ε′(3.9,3.12)⇐⇒

⇔ h · e−λ·(x0+δ)m+

+ μ = e0 − ε′ ⇐⇒

⇔ e−λ·(x0+δ)m+

=e0 − μ − ε′

h(3.10)⇐⇒

⇔ e−(

x0+δx0

)m+·ln

(h

e0−μ

)=

e0 − μ − ε′

h⇐⇒

⇔ eln (e0−μ

h )(x0+δx0 )

m+

=e0 − μ − ε′

h⇐⇒

⇔(e0 − μ

h

)(x0+δx0

)m+

=e0 − μ − ε′

h⇐⇒

⇔(x0 + δx0

)m+

= log( e0−μh )

(e0 − μ − ε′

h

)⇐⇒

⇔ m+ = log( x0+δx0

) [log( e0−μh )

(e0 − μ − ε′

h

)]Equation 3.21.

ea(x0 − δ′

)= e0 + ε

(3.9,3.12)⇐⇒

112

⇔ h · e−λ·(x0−δ′)m−

+ μ = e0 + ε ⇐⇒

⇔ e−λ·(x0−δ′)m−

=e0 − μ+ ε

h(3.10)⇐⇒

⇔ e−(

x0−δ′x0

)m−·ln

(h

e0−μ

)=

e0 − μ + εh

⇐⇒

⇔ eln (e0−μ

h )

(x0−δ′x0

)m−

=e0 − μ + ε

h⇐⇒

⇔(e0 − μ

h

)(x0−δ′x0

)m−

=e0 − μ + ε

h⇐⇒

⇔(x0 − δ′

x0

)m−

= log( e0−μh )

(e0 − μ + ε

h

)⇐⇒

⇔ m− = log( x0−δ′x0

) [log( e0−μh )

(e0 − μ + ε

h

)]⇐⇒

⇔ m− =

ln(

ln e0−μ+εh

ln e0−μh

)ln x0−δ′

x0

Equation 3.24.

(3.22)(3.23)

⇒(x1x2

)ma,x

=ln ea,1−μ

h

ln ea,2−μh

⇐⇒

⇔ ma,x = log( x1x2

)[ln ea,1−μ

h

ln ea,2−μh

]⇐⇒

⇔ ma,x =

ln(

ln ea,1−μh

ln ea,2−μh

)ln x1

x2

113

114

List of Figures

1.1 TheMTRs Ecosystem . . . . . . . . . . . . . . . . . . . 2

2.1 SLACC Solution Overview [71] . . . . . . . . . . . . . . 112.2 QoE Mapping Function of Packet Loss Ratio in the IQX

Hypothesis [31] . . . . . . . . . . . . . . . . . . . . . . 122.3 Reference Connection of the E-Model [53] . . . . . . . . 15

3.1 MOS Evolution for IVs (ei) and AVs (ea) . . . . . . . . . . 283.2 Plot for Differentm Values of Equation 3.8 . . . . . . . . . 303.3 Plot for Differentm Values of Equation 3.11 . . . . . . . . 313.4 MOS Change for IVs . . . . . . . . . . . . . . . . . . . . 323.5 MOS Change for AVs . . . . . . . . . . . . . . . . . . . . 353.6 Generic MOS Evolution for Equally Participating IV

(m = 1) and AV (m = 3) . . . . . . . . . . . . . . . . . . 38

4.1 Experimental Setup . . . . . . . . . . . . . . . . . . . . . 474.2 AQXModel Fit and Comparison for Latency . . . . . . . 514.3 AQXModel Fit and Comparison for Packet Loss . . . . . 534.4 Development of them Values (jitter, latency, packet loss) . 554.5 Development of them Values (bandwidth) . . . . . . . . 564.6 3D-Graph of the AQXModel for Multiple Variables . . . . 584.7 Adjusted AQXModel Fit for Latency . . . . . . . . . . . 594.8 3D-Graph of the Adjusted AQXModel . . . . . . . . . . 60

5.1 AbaCUS Stakeholders Map . . . . . . . . . . . . . . . . . 675.2 Key Elements of AbaCUS . . . . . . . . . . . . . . . . . 70

6.1 Switching Time BetweenMNOs . . . . . . . . . . . . . . 856.2 Min andMax Values for the MNO Switching Time . . . . 866.3 MNO Switching Time and Signal Strength Correlation . . 866.4 MNO Switching Time During the Day . . . . . . . . . . . 876.5 MNO Switching Time> 10 s . . . . . . . . . . . . . . . . 87

115

6.6 AbaCUS Service Request . . . . . . . . . . . . . . . . . . 916.7 AbaCUS E2EMNOCalling-time . . . . . . . . . . . . . 91

116

List of Tables

2.1 TheMOS Scheme Recommended by theITU-T [51] . . . 12

3.1 Broadband Plans for Home-users of a Swiss ISP [18] . . . 41

4.1 x0 and References Used for the Evaluation . . . . . . . . . 504.2 Results of the Single Variable Scenarios . . . . . . . . . . 514.3 CollectedMOS forMixedVariablesCompared to theCal-

culated MOS . . . . . . . . . . . . . . . . . . . . . . . . 57

5.1 Internet Users’s Survey Results . . . . . . . . . . . . . . . 675.2 QoS-Cs List . . . . . . . . . . . . . . . . . . . . . . . . . 715.3 Approaches Against the MTRsMonopoly . . . . . . . . . 73

6.1 Signal Strength Values . . . . . . . . . . . . . . . . . . . 846.2 Mobile Device Characteristics . . . . . . . . . . . . . . . 896.3 Power Consumption Evaluation . . . . . . . . . . . . . . 90

117

118

Acknowledgments

I amverygrateful toanumber of people that have helpedme, directlyor indirectly, through the arduous work that culminates with this thesis.

Firstly, I would like to thank Prof. Dr. Burkhard Stiller for believingin me, allowing me to join the Communication Systems Group (CSG) atthe University of Zurich, giving me freedom to research topics that I foundinteresting, and precious help, feedback, and understanding during severaldifficult stages. In addition, I would like to thank Prof. Dr. Peter Reichlfor being my co-supervisor introducing me to the very interesting topic ofQuality-of-Experience (QoE) and inspiring me, as well as providing mewith highly valuable feedback. Thank you!

I would also like to warmly thank “my” co-authors, Tobias Hoßfeld,Samuel Liniger, Manuel Rüsch, Sebastian Seeber, and Anuj Sehgal, for co-authoring key scientific papers together. You all have greatly contributed toboth my work and my overall development as a scientist. I am particularlygrateful to Patrick Poullie for providing the German translation to this the-sis’ abstract. Furthermore, Iwould like to thank all 14 assignment, diploma,bachelor, andmaster students that I have supervised while working on thisthesis for the rich discussions and contributions.

It has been an immense pleasure working at the CSG and enjoying thetrue friendship of my work colleagues, including warm discussions aboutdiverse subjects with current colleagues Dr. Thomas Bocek, Andri Lareida,and Patrick Poullie, as well as former colleagues Prof. Dr. Karoly Farkas,Dr. Fabio Hecht, Dr. Guilherme Sperb Machado, Dr. Oleksiy Mazhelis,Dr. Flavio Santos, and Dr. Andrei Vancea.

I would like to thank all my former professors for helping me to gainthe knowledge that helpedme to tackle the problems that I challenged dur-ing my Ph.D. Special thanks to Prof. Dr. Christos Chaldoupis, Prof. Dr.Nikolaos Kylafis, and Prof. Dr. Stavros Toumpis for being an inspiration,believing in me and supporting me when needed.

119

Finally, I would like to thank my parents for the unconditional supportall those years supporting me and providing every possible help. I want tomake you proud, like I am proud to be your son. Also, special thanks to mygirlfriendEleniVlachaki and close friends IoannisAntoniou, ApostolosBe-lokas, Nikos Chatzidimitriou, SpirosGiannatos, Dimitris Gkounis, StavrosKaragiannopoulos, Koutrolikos Konstantinos, Nikos Planas, Chara Vareli,Eleni Veletsanou, Giorgos Vlachakis, Panagiotis Vrakatselis, and StefanosPetrakis for their support, I love you all!

120

CurriculumVitae

Christos Tsiaras was born on June 3, 1981, in Athens, Greece. Chris-tos holds a four-year degree named “Ptycheio” – equivalent to a SwissDiplom – in Physics from the University of Crete (UOC). After his stud-ies in Physics Christos received the Probationary Reserved Officer (PRO)degree from the school of Signal Corps of the Hellenic Armed Forces inGreece. Christos served as an officer for 2 years in the Hellenic ArmedForces and was dismissed with the rank of the 2ⁿd Lieutenant. After thatChristos continued his studies at the master program of the deprtment ofComputer Science at the Athens University of Economics and Business(AUEB) and received his M.Sc. with the highest distinction. His thesiswas entitled “EfficientMinimization of routing cost in Delay Tolerant Net-works (DTNs)”, in which a low complexity routing protocol for DTNs hasbeen designed. After his studies, Christos has worked as a Physics teacherin high schools, a freelancer programmer and project manager, starting hisown company in Heraklion in 2006.

By mid-2009, Christos Tsiaras has moved to Copenhagen, Denmark tobecome a software engineer. However, Christos’ dream is to work in re-search and it was time for him to chase it. Thus, by mid-2011 Christoshas moved to Zurich, Switzerland to become a doctoral student until mid-2015, as well as a research assistant in the Communication Systems Groupat theDepartment of Informatics of theUniversity of Zurich. Thework hasinvolved a multitude of tasks, including wireless and mobile communica-tions andQuality-of-Experience (QoE)modeling, managing the web-pageinfrastructure, and designing a parkingmanagment andmonitoring systemthat aims to minimize the cruising for parking effect in large cities. Chris-tos has been involved in the following research projects: “ParkITsmart(pITs)”, “Coinblesk”, “ICT COST Action IC1304: Autonomous Controlfor a Reliable Internet of Services (ACROSS)”, “Auction-based ChargingandUser-centric System (AbaCUS)”, “Socially-awareManagement of NewOverlay Application Traffic combined with Energy Efficiency in the Inter-

121

net (SmartenIT)”, “Management of the Future Internet (FLAMINGO)”,“Accounting and Monitoring of AAI Services (AMAAIS)”, and “Socio-economic Services for European Research Projects (SESERV)”.

Christos’ main research interests are Quality-of-Experience (QoE),wireless and mobile communication, and Internet of Things (IoT). Hisdoctoral thesis was supervised by Prof. Dr. Burkhard Stiller (Universityof Zurich, Switzerland) and Prof. Dr. Peter Reichl (UniversitätWien, Aus-tria).

122

Glossary

3GPP 3rd Generation Partnership Project.

AbaCUS Auction-based Charging User-centric System.

ADT Android Developer Tools.

API Application Programming Interface.

AQX Axiomatic Quality-of-Experience Model.

AT Attention (Commands).

Au² Auction Authority.

AV Antitonic Variable.

CDR Call Data Record.

CP Cloud Provider.

CPP Calling Party Pays.

CPU Central Processing Unit.

CSG Communication Systems Group.

E2E end-to-end.

EC2 Elastic Compute Cloud.

eMOS Expected Mean Opinion Score.

eV² Expected Variable Value.

FNO Fixed Network Operator.

GoF Goodness-of-Fit.

123

GSM Global System for Mobile Communications.

HD High Definition.

HLR Home Location Register.

IDE Integrated Development Environment.

iLBC Internet Low Bitrate Codec.

IP Internet Protocol.

IQX Interdependency of the Quality-of-Experience and Quality-of-Service.

ISP Internet Service Provider.

IT Information Technology.

ITU International Telecommunication Union.

ITU-T International Telecommunication Union TelecommunicationStandardization Sector.

IV Isotonic Variable.

LAN Local Area Network.

LTE Long Term Evolution.

M(V)NO Mobile (Virtual) Network Operator.

MAC Media Access Control.

MCC Mobile Country Code.

MIMO multiple-input and multiple-output.

MNC Mobile Network Code.

MNO Mobile Network Operator.

MNP Mobile Number Portability.

124

MOS Mean Opinion Score.

MSISDN Mobile Subscriber Integrated Services Digital Network-Number.

MTR Mobile Termination Rate.

MVNO Mobile Virtual Network Operator.

NatRoam National Roaming.

NLS Number Lookup Service.

OFCOM Federal Office of Communications.

OLR Overall Loudness Rating.

OS Operating System.

OTT over-the-top.

P2P Peer-to-Peer.

QoE Quality-of-Experience.

QoS Quality-of-Service.

QoS-C Quality-of-Service Class.

RAM Random Access Memory.

ROM Read Only Memory.

RPP Receiving Party Pays.

RSSI Received Signal Strength Indication.

SD Standard Definition.

SDK Software Development Kit.

SGS2 Samsung Galaxy S II.

125

SIM Subscriber Identity Module.

SLA Service-level Agreement.

SLACC Service Level Agreement Support System for Cloud Computing.

SP Service Provider.

SS7 Signaling SystemNo. 7.

SSH Secure Shell.

STMR Side ToneMasking Rating.

TeR-C Termination Rate Class.

ToS Type of Service.

UB User Behavior.

UE User Equipment.

UI User Interface.

VLR Visitor Locator Register.

VoIP Voice over Internet Protocol.

W3C WorldWideWeb Consortium.

WebRTC Web Real-Time Communications.

WLAN Wireless Local Area Network.

126