Resource Sharing and Network Deployment Games: In Open Wireless Access Markets

Network Deployment Games

In Open WirelessA ess Markets

DINA PAMELA GONZALEZ-SANCHEZ

Li entiate Thesis in

Communi ation Systems

InOpenWirelessA essMarkets

DINAPAMELA GONZALEZ-SANCHEZ

Li entiate Thesisin

Communi ation Systems

ISRNKTH/COS/R--11/07--SE SWEDEN

AkademiskavhandlingsommedtillståndavKunglTekniskahögskolanframlägges

till oentlig granskning för avläggande av teknologie li entiatexamen i radiosys-

temtekniktisdagenden04O tober2011klo kan10.00isalC1,Ele trum1,Kung-

ligaTekniskaHögskolan,Isafjordsgatan26,Kista.

©DinaPamelaGonzalez-San hez,O tober2011

Abstra t

The mostre entproblem that wirelessservi eproviders(SPs) fa e

nowadaysisrelatedtotheintrodu tionofthenewmarketrulesi.e.,at

ratepri ingpoli ies.WhileSPshavedeployedadditionalinfrastru ture

inordertoprovidehigherdataratestoin reasingnumbersof ustomers

theatratepri ingpoli ieshaveleadto de reasedprotsby theSPs.

These at rate revenue streams in ombination with rapidly growing

osts asso iated with onventional a ess deployment haveresulted in

whatis usuallyreferredto astherevenuegap. E ientutilizationof

radio resour eshas be ame thekeyenablerto satisfyboththeservi e

provider(the operator)andthe users. This pointsoutto theneedfor

betterradio resour emanagement (RRM). The hallenge is to design

me hanisms that allo ateresour es ona dynami basis (i.e., dynami

spe trum a ess (DSA) and management, power ontrol, ooperation

enfor ement, et .) in order to either redu e or lose the revenue gap

without negatively impa ting users' performan e. E ient utilization

of radio resour es is one solution to this problem, as it would allow

SPs to support high data rateswhile providing widearea overage at

relativelylow ost.

In this thesis, we study ompetitive games among revenueseeking

SPsandtheimpa tontheirrevenuesandontheuser'performan e. We

analyzetheproblem des ribed abovefromtwoperspe tives. First,we

fo usonaparti ularsystemwherelimitedavailablespe trumresour es

are allo atedamong the SPs dynami ally by aSpe trum Broker. We

studytheee tof hannelheterogeneity(frequen y hannelsthatdier

in propagation onditionsand interferen e levels) on the performan e

ofthesystemin termsofspe trumutilization,SP's prot,andenergy

onsumption. Inthese ondpart,weanalyzeseveraldierent ompeti-

tivenetworkdeploymentgames. Wefo usonas enariowherewireless

networks (dierent SPs) with limited available bandwidth ope with

the problem of how to maximize their network revenue. Non oopera-

tivegamesbetweenusersandSPsareinvestigatedandanopenwireless

a essmarketisintrodu edbasedonnetworkdeploymentstrategies.

Basedontheresultsofthestudiesthatwereperformed,weobserve

thatalthoughthereare onsiderableapproa hes(inliterature)support-

ingthat ompetitives hemesaregoodstrategiesforloadbalan ing,the

output of our study ree ts that ompetition for spe trum resour es

based on usersrequest (real-time allo ation) leadsto deadra e when

the over load ase is onsidered. A negative impa tin SPs' prots is

observedand mightthus notbethemostee tivesolutionifrevenue-

seeking SPs are onsidered. However, under the spe i assumptions

fortheanalysisinChapter2,ithasbeenshownthatthespe trumuti-

taken into a ount when devising spe trum a ess s hemes. Regard-

ing deployment strategies, providing partially overlapping overage in

a ompetitive fashion is an option for SPs to maximize their prots.

Results indi ate that the fra tion of overage overlapae ts boththe

quality ofservi e experien ed bythe usersand theprotability of the

a essproviders. However,asuitableper entofoverlappingservi ear-

easbythetwonetworksmaybebene ialtoallin thesystem. We an

alsoinferthataprioranalysisbasedonthestrategiesusedbytheoppo-

nentSPsshouldbeimplementedbyanew omerSPbeforeenteringthe

wireless market. When an in umbent pla es its network properlya -

ordingtothedemand,itislesslikelyforanentranttobeself-sustained

inthemarket.

The LORD Jesus Christ has blessed me with the opportunity to grownot only

professionallybutalsospirituallyduringthejourneyofgettingaLi entiatedegree.

IthankHIMforallinall! Mysin eregratitudegoestoProfessorJensZander,my

advisor,forgivingmetheopportunitytojointheCOSgroupandforhisguidan e

and en ouragement during these studies. I really appre iate all our dis ussions

whi hhelpedmegrowandinspiredmewithnewideas. MaytheLORDblessyou.

Spe ialthanksto everybodyinmy hur hKLIPPAN fortheirprayers. Theyhave

beenlikeafamilyto meinSwedenduringthelastfewyears.

I am thankful to Dr. Klas Johansson (Eri sson) for reviewing my Li entiate

proposalandforallvaluable ommentsthathelpedimprovethisthesis. ToProfes-

sorSeong-Lyun Kim forreviewing apreliminaryversionofthis thesis. I am very

thankfulto Dr. Claes Tidestav (Eri sson) for a epting therole asopponent on

mydefense. ManythankstoProfessorGeraldMaguireforhisvaluablefeedba kin

someoftheworkin ludedinthis thesis.

I wouldliketo takethis opportunitytothankmy olleagues attheCommuni-

ationSystemdepartmentand letyou allknowhowI haveenjoyedworkingwith

youthese past fewyears: Dr. Östen Mäkitalo, Dr. Pietro Lungaro,Dr. Ki Won

Sung,Dr. JanMarkendahl,ProfessorClaesBe kman,ProfessorBenSlimane,Jan-

OlofÅkerlund, Anders Västberg, GöranAndersson, SaltanatKhamit, Tafzeel ur

Rehman,Ali Özyagzi,EvannyObregon,LeiShi,DoHoKang,Sibel Tombaz,and

Serveh Shalmashi. Also to former olleagues: Dr. Lu a Stabellini, Dr. Bogdan

Timus,Dr. AurelianBria,Dr. ÖmerIleri, andDr. JohanHultell. Forallthesup-

port in administrativeissues and te hni al matters I thank Adriana Flores(from

Ni aragua), Anna Barkered, Irina Radules u, Ulla Eriksson, Robin Gehrke and

Ri hardAndersson.

I am grateful for the experien e, good talks, and hallenges I got from my

olleaguesinNi aragua: Dr. MarvinArias,Dr. MarvinSán hez,Os arSomarriba,

Dr. Leonel Plazaola, Mar o Munguia, Pablo Vásquez, Johnny Flores, Anayan i

López, Norman Vargas and Apolinar Pi ado. In general, I would like to thank

anyonewhohasworkedatthedepartmentofCommuni ationSystemsCOS/KTH

in Sweden andat theNational Universityof Engineering(UNI) in Ni aragua,for

providingafun andwarmworkenvironment.

Spe ialthanks tomy friendsforsharinggood momentsandfor aringforme:

KarolinaBjörsund,Anna Palbom, Islam Al-yafawi,LuisGuillermo Martinez,Mi-

urelTer ero,RaulRodriguez,AaronJosuéQuiróz,and CristobalViedma. Tomy

an é,Pethrus Gärdborn, for his love and are. I am indebted to my family for

theirun onditionalloveandsupport,spe iallytomywonderfulmotherMariaFelix

San hezforallherprayersanden ouragement. Shehasalwaysbeenthereforme,

always.

Thenan ialsupportfromtheSwedishInternationalDevelopment ooperation

Agen y(SIDA)andtheNationalUniversityofEngineering(UNI)isgratefullya -

knowledged. This was a ooperation betweenthe department of Communi ation

SystemCOS/KTHin SwedenandUNIin Managua,Ni aragua.

Pamela GonzalezSan hez

Kista,September2011

A knowledgements iii

Contents v

ListofTables vii

ListofFigures viii

ListofAbbreviations xi

ListofNotation xiii

I 1

1 Introdu tion 3

1.1 Ba kground . . . 3

1.2 ProblemFormulation. . . 8

1.3 OverviewoftheThesisContributions . . . 10

1.4 ThesisOutline . . . 12

2 CompetitiveSpe trum Sharing 13 2.1 RelatedLiterature . . . 13

2.2 WhySpe trumSharing? . . . 14

2.3 DistributedSpe trumA esswith Energy Constraintfor Heteroge- neousChannels(Paper1) . . . 15

2.4 ResultsandDis ussion . . . 17

2.5 Summary . . . 19

3 NetworkDeploymentStrategies inOpenWirelessA essMarkets 21 3.1 RelatedLiterature . . . 22

3.2 Resour eAllo ationMe hanism. . . 23

3.3 DeploymentStrategiesinCompetitiveWirelessA essNetworks(Pa- per2) . . . 24

3.4 CompetitiveGames forRevenueMaximizationwith Heterogeneous

Demand(Paper3,Paper4) . . . 28

3.5 Summary . . . 38

4 Con lusion 41

4.1 Con ludingRemarks . . . 41

4.2 FutureWork . . . 43

Bibliography 45

II Paper Reprints 51

Paper I: Distributed Spe trum A ess with Energy Constraint

for HeterogeneousChannels 53

PaperII:DeploymentStrategiesinCompetitiveWirelessA ess

Networks 59

Paper III: Competitive Pri ingwith HeterogeneousDemandin

Open WirelessA essMarket 67

PaperIV:CompetitiveA ess-pointDeploymentinMobileBroad-

band Systems 75

1.1 Thethreedimensionsofinfrastru turesharing[1℄. . . 7

1.1 TheRevenueGap . . . 4

2.1 S hemati representationoftwodierentmethodsforspe trumsharing. 14

2.2 WirelessNetworkStru ture-a ompetitivesharings enario. . . 16

2.3 ChannelO upan yasafun tionofthenumberofusersintheSystem. 18

2.4 Servi eproviderutilityasafun tionofthenumberofusersintheSystem. 18

3.1 Servi eproviders- hoosetheirtargetareasandQoSoerings . . . 21

3.2 Representationoftheau tionpro edure asso iatedwithaletransfer.

Inthisexampletrade-agentj initiatesaletransferinau tion1.. . . . 23

3.3 Basi s enario-Illustrationofawirelessnetworkar hite turewithdier-

entper entagesofoverlapping,whi hrepresentsasystemunderdierent

levelsof ompetition . . . 25

3.4 Averagerevenueperbase stationandtime slotasso iatedwiththe BS

gameunderdierentper entagesofoverlapping overage . . . 27

3.5 Average throughput experien ed by users for dierent per entages of

overlapping overageas afun tion ofthepotentiallyoeredload. . . 28

3.6 Basestationslo ation-lineargeographi alregion . . . 31

3.7 NetworkDeployment-GivenS enario1 . . . 32

3.8 Average prot per BS per se ond in the System based on a demand-

responsivepri ingme hanism onsidering the ase ofa high ompeti-

tive regime,withmarketshare =100.. . . 33

3.9 Expe ted BS prot per se ond as a fun tion of the reservation pri e

when

1

=

2

. . . 34

3.10 Theuserresponsivenessinthesystem,denotedbytheA eptan eProb-

ability asafun tion ofthe reservationpri eduring ea h au tion y le.

Weobservehowtheprobabilitythat theuser a eptsthe servi efrom

aSP variesunderdierentmarketshares,. . . 34

3.11 Illustrationofabasi s enariowhenanew omerSPsgetsintothewire-

lessmarket-S enario2. . . 35

3.12 Expe tedprotperAPperse ond(timeslot)asafun tionoftheNew-

omer's AP lo ation. The New omer SP gets into the market pri ing

theservi esequallytotheIn umbent. . . 36

3.13 Expe tedprotperAPperse ond(timeslot)asafun tionoftheNew-

omer's AP lo ation. The New omer SP gets into the market with a

lowerpri efortheservi es. . . 37

AC AlternateCurrent

AP A essPoint

APs A essPoints

BR BestResponse

BS BaseStation

BSs BaseStations

CAPEX CapitalExpenditures

COCOR A Conferen eonAdvan esin CognitiveRadio

CoR R ComputingResear hRepository

CR OWNCOM Conferen eonCognitiveRadioOrientedWirelessNetworksand

Communi ations

dB De ibel

DC Dire tCurrent

DSA Dynami Spe trumA ess

DySPAN Dynami Spe trumA essNetworks

EW EuropeanWireless

GHz Gigahertz

IEEE InstituteofEle tri alandEle troni sEngineers

ICC InternationalCommuni ationConferen e

ICMCS InternationalConferen eonMultimediaComputingandSystems

ICUMT InternationalCongressonUltraModernTele ommuni ations

KTH KungligaTekniskaHögskolan

QoS Quality-of-Servi e

MHz Megahertz

MODyS MultiOperatorDynami Spe trumA ess

MVNO MobileVirtualNetworkOperators

NEP NashEquilibriumPoint

NOMS NetworkOperationsandManagementSymposium

OPEX OperatingExpenditures

PIMR C Personal,IndoorandMobile RadioCommuni ations

R AN RadioA essNetwork

R AT RadioA essTe hnology

R R M RadioResour eManagement

SIDA SwedishInternationalDevelopmentCooperationAgen y

SNR Signal-to-NoiseRatio

SP Servi eProvider

SPs Servi eProviders

TDMA TimeDivisionMultipleA ess

UK UnitedKingdom

UNI UniversidadNa ionaldeIngenieria

USA UnitedStatedofAmeri a

VTC Vehi ularTe hnologyConferen e

WiCOM WirelessCommuni ations,NetworkingandMobile Computing

3G ThirdGeneration

Notation(anddefaultunits) usedinChapter2:

C

Cost(j)

Channel ostor ostoffrequen yj

Co ConstantbasedontheSNR

th

,noisepowerandspeedoflight

d

r;i

Distan e betweenthebasestationofproviderrandtheuseri

f

j

Frequen yj

K Energy Costinmonetary units/powerunits

SP

r

Servi eProviderr

P

(r;i;j)

Powerrequiredbyservi eproviderrtorea huserion hannelj

U

(r;i;j)

Utilityofservi eproviderrwhenservinguserion hannelj

x

user

Fixedpri ethatea huserpaysto theSP fortheservi e

Notationusedin Chapter3:

A m

i;j

A eptan eprobabilityofuserj in au tioniatbasestationm

 Lo ationofthepi kofthedemand

;; Positive onstantstoshapethea eptan eprobabilityfun tion

D

0

Potentiallyoeredloadin megabits/BS/se ond

 Reservationpri e



max

Maximumreservationpri e



m

Reservationpri eestablishedbybase stationm





m

Bestresponseasso iatedwithbasestationm



min

Minimumreservationpri e,xed ostin urredbytheSP

 Pa kagearrivalintensity

l Users'lo ation

L Longitudeofthesystemarea

N Numberofusersinthesystem

q Sizeoftheletobedownloaded

R Cellradius

R

i;j

Data-rateexperien ebyuserj inau tioni

R

z;j

Data-rateexperien ebyuserj inau tionz (lastround)

 Levelofmarketshare

 2

Themeasureofthewidthoftheuserdistribution

s

i;j

Bidin monetaryunits thatuserj pla esinau tioni

S

i; j

Bidsofalltheopponentusers,representedalsoass

i;k +

s m

i;j

Bidthat userjpla esin au tioniatbase stationm

T

A

Au tion y leequalto1se ond

U

i;j

Utilityofuserj duringau tioni, valueformoney

^

U m

i;j

Estimatedutilitybyuserj during au tioniatbasestationm

x

i;j

Portionoftransmissiontimeforuserj in au tioni

x

1

Coverageofbasestation1

x

2

Coverageofbasestation2

z Lastau tionround

Introdu tion

1.1 Ba kground

Wideareamobile ommuni ationsystemswereprimarilydesignedtoprovide ost

e ientwidearea overageforuserswithmoderatebandwidthdemands. Thehigh

demand for these servi es, ombined with de reasing terminalpri es and reason-

ableinfrastru tureinvestmentrequirementsforoperators reatedaverysu essful

evolutionenvironmentinthebeginningofthe20 th

entury[2℄. However,this om-

binationoffa torsisnotlongertrue. Therapidlyin reasingdemandforhigh-speed

wirelessdataservi eshasleadto burstsofhighbandwidthdemandsfromusersin

wideareawireless networks. Additionally withthe largenumbersof users, many

of whom want and use high speed servi es, the total tra load in reaseswhile

revenuehasattenedoutasshownin Figure1.1.

Thela kofadditionalspe trumandtheintrodu tionofnewmarketrules,i.e.,

atrate pri ing poli ies, haveprompted SPs to deploy additional infrastru ture,

i.e.,morebase stations(BSs),in ordertoprovidehigherdatarates,whi h inturn

negativelyae tstheirprots. Theatraterevenuestreamsin ombinationwith

therapidlygrowing ostsasso iatedwith onventionala essdeploymentisusually

referredto asthe revenuegap, asshown in Figure 1.1. This rapid development

ofmobile broadbanda ess servi es havevastly in reased the interestin wireless

solutionswhi h ombinehigh- apa itywithlow- ost[3℄.

E ientutilizationofradioresour es 1

,i.e.,infrastru ture,energyandspe trum,

hasalwaysbeen atool for servi e providers to improve the performan e of their

networks and lowertheir operational osts 2

. Thisproblem hasattra ted resear h

1

Generallyreferringtoe ien yofutilizationofaradio hannel(i.e., hoosingtheappropriate

timeslot,frequen yband,dynami hannelallo ation)andtransmitpower.

2

The operational osts in lude all of the annual osts of operatingthe network, in luding

ele tri alpower,personnel,taxes,et .

interestin thelast fewyears andnewapproa heshavebeenintrodu edwhere re-

sour esareallo ateddynami ally(bymeansofDynami Spe trumA ess(DSA)

andmanagement,power ontrol, ooperativeenfor ement,et .) strivingforbetter

RRMin ordertoredu etherevenuegap.

Su ientavailabilityofradiospe trumallowsservi eprovidersto redu etheir

investment osts [4,5℄. Therefore, e ient radio spe trum managementplays an

importantroleinordertosupporthighdatarateswhileoeringwidearea overage

at relatively low ost. Unfortunately, spe trumis not yet properlymanaged and

theimplementationofDSAme hanismsrequirelargeinvestmentsinsoftware,sig-

naling, oordination,et ,hen ewelldevisednetworkdeploymentstrategiesarean

importantand interestingaspe t ofproviding apra ti alsolutionto theproblem

oftherevenuegap.

Revenue Traffic

Costs Voice Dominant

Data Dominant

Time

The Revenue Gap

Figure1.1: TheRevenueGap

Sharing resour esis a promising way to e iently utilize radio resour es and

hen elowernetwork osts. Inthefollowingsubse tionanoverviewofthebenets

ofradioresour esharing isgiven.

1.1.1Resour e Sharing

Thebenetsthat anbeobtainedbysharingnetwork resour esare: ost-oriented

(i.e., lowerCAPEX 3

and OPEX 4

), ustomer-oriented(i.e., higher apa ity, more

overage,end-userbetterqualityof servi e(QoS) in termsofa hievablethrough-

put),regulatoryreasons(i.e.,satisfyingli ensingagreements),environmentalben-

ets(i.e.,redu ednumberofsites,power onsumption), et .

Inthis thesiswefo us onthe ostredu tionbenetand onsider,spe i ally,

twowaysofsharing;infrastru ture andspe trumsharing. Therstmethod refers

totheuseof ommonnetworksorpartsofnetworks,dierenttypesofroaming[6℄,

deploymentbasedondierentte hnologies,ooadingtolo al networks, sharinga

ommongeographi region,mutualservi eprovisioning(asabusinessmodel),and

othersstrategies[3℄. Spe trumsharings hemes,on theother hand,allowseveral

operatorsto share frequen y arriersfrom a ommonpool[7℄. Thisenablesthem

toserveon-demandtra whileminimizingtheirinvestmentrisksdueto redu ed

CAPEXsin ethenumberofbase stationsites anberedu edsubstantially[3℄.

ˆ Spe trumSharing: Due tothe urrentstati wayofallo atingthefrequen y

spe trum, the amount of identied available resour e is not large su ient

to support large bandwidth allo ations for many operators. Therefore, it

is of paramountimportan e for future mobile ellularsystems to share the

frequen y spe trum [8℄. Dynami me hanisms, aiming to put spe trum to

itsbest useonashort-term time-s ale,havebeenwidelyproposed in litera-

ture[913℄. Su h models, tailoredfor future systems, are designedto allow

sharingeitherona ompetitiveor ooperativefashion andinbothtemporal

andspatialdomains.

Competitivespe trumsharingisusuallydonebyservi eproviderswhoshare

a blo k of spe trum in a ompetitive fashion aiming to selshly maximize

theirindividualutilities. In ooperativespe trumsharing,ontheotherhand,

agroup of wireless operators agree to share the available spe trumand to-

gether maximize the system utility. An example of this is the se ondary

spe truma essmodelwheretheprimaryandse ondaryusers(systems) ol-

laborateand oordinatebyagreeingontermsand onditionsofsharing(i.e.,

byestablishingwhat frequen ies,whentoshare,andpri es).

Inthesis,wehavepartiallystudied ompetitivespe trumsharing;whereSPs

a essesthespe trumina ompetitivefashion. Ithasbeendebatedthat om-

3

TheCAPitalEXpendituresrepresentexpensestoupgradethephysi alnetworkorequipment.

Thistype of outlay ismade byservi e providers to maintain or in rease the quality of their

networksandservi es.

4

TheOperatingEXpensesrepresentongoing ostsforrunningthe network, supportingbusi-

petitivespe truma ess 5

putsfrequen yspe trumtoitsbestuse. Allo ating

spe trum li enses amongseveralwireless SPs reates ompetitives enarios 6

andarapidlygrowingwirelessindustry[20,21℄. Ithasbeen learthat ompe-

titiondrivesSPstoparti ipateinau tionme hanismstoget ellularspe trum

li enses,makesthemwillingtopayhighpri esinordertoassureex lusivity 7

,

andkeepsnewentrantsoutofthewirelessmarket.ThisalsoallowstheSPsto

providehigherdata-rateswithoutbeingfor edto deploymorebase stations.

ˆ Infrastru tureSharing: Fromane onomi pointofviewandbe ausewireless

networksareoften omposedofamixtureofradioa esste hnologies(RATs),

infrastru turesharing is apossible ooperativealternativewhi hoershigh

speeddataa essina oste ientmanner[22℄. Thestartingpointisusually

thesharingofpassiveinfrastru turei.e.,towers,shelters,air onditioningand

oolingsystems,ACand DCpowersupplies, anddieselgenerators. Sharing

leasedlinesandmi rowaveslinksalsohelpoptimizea esstransmission 8

. The

sharingofnetworkinfrastru turepresentse onomi optionsfor overageand

apa ity growth for new entrants as well as network onsolidation through

ostoptimizationandte hnologyupgradesforin umbentoperators[1,23℄.

Infrastru turesharing anbe ategorizedinthreedimensions;businessmodel,

geographi model,andte hnology model,aspresentedin[1℄andillustratedin

Table1.1.

The business dimension fo us on the parties involvedand their ontra tual

relationships. The se ond dimension onsiders in the dierent te hnology

models. Whilethethirddimension onsiders theoperators'geographi mar-

ket share,addressingtheirin overageareaandiftheyoverlaporsharethis

overageareadependingupontheirbusinessmodel andte hnology hoi es.

Thisthesis ontainssomestudiesof thegeographi modelwhere weanalyze

thebehaviorofthesystemunderdierentregimes;rangingfromthefullsplit

5

Thistypeof ompetitionis arriedout amongservi eproviderswiththe spe trumbroker

asamediator. Examplesofsimilarapproa hesbased on ompetitivespe trumallo ationviaa

spe trumbroker anbefoundin[1417℄

6

This ompetitionisatthe level ofthe onsumerofwireless servi es,theenduser andit

isbene ial sin ethis reatesopportunities forthem to getbetter servi es;spe i allyhigher

data-ratesatlower osts[18,19℄.

7

TheSP ouldpotentiallylimit ompetitionatthe levelofwireless usersbypur hasing ad-

ditionalspe trumthat would otherwise goto an entrant (New omer SP). Thismayrepresent

adangerbylimitingdynami evolutionofservi eif ompetitionwithotherSPsisne essaryto

speedupbuildoutanddevelopmentofnewte hnologies[20℄.

8

Table1.1: Thethreedimensionsofinfrastru turesharing[1℄

Dimensions S ope

BusinessModel -UnilateralServi eProvisioning

-MutualServi eProvisioning

-JointVenture

-3 rd

PartyNetworkProvider

Te hnology Model -SiteSharing

-A essTransmissionSharing

-A tiveRadioA essNetworks(RAN)Sharing

-3GMulti-OperatorCoreNetwork

-RoamingBasedSharing

Geographi Model -FullSplit

-CommonSharedRegion

-UnilateralSharedRegion

-FullSharing

asethroughsome ommonsharedregionuntilrea hingthefullsharing ase

(seeChapter3).

Thesharingofwirelessinfrastru ture,however,raisesthequestionofhowre-

sour esandrevenuesshould bedividedwhen multiplesubsystems,managed

bypotentially ompetinga tors,areinvolvedin deliveringthea essservi e.

Analternativewouldbetosharetheinfrastru tureimpli itlybyestablishing

anopenwirelessa essmarketwhereinnetworksnotonly ompeteforusers

onalong-term time-s ale,butalso onamu hshortertime-base. This ould

berealizedwith anar hite turewhere autonomoustrade-agents,that reside

interminalsandBSs,managetheresour esthroughnegotiations[2427℄. In

ourstudieswehaveusedsu has heme,whi hisexplainedin detaillaterin

Se tion3.3ofthethesis.

In this thesis, wefo us on ompetitive s enariosand investigate their impa t

onSPs'revenueandusers'performan e. Approa hestodynami spe trumsharing

andnetwork deployment strategieshavebeenstudied in severalparti ular forms,

spe i ally

ˆ Au tionMe hanismshavebeentailoredforallo atingtransmissionrights

onashorttermbasisinordertoprovidee ientallo ationofs ar eresour es.

Inourstudy,werst addressspe trumau tionasa ompetitiveme hanism

for spe trum sharing. Sin e their introdu tion in 1994, spe trum au tions

havebeen remarkablysu essful inassigning and pri ingspe trum. Assign-

ingspe trumli enses toprivatefor-prot ompaniesthroughoutmostofthe

world,in ludingdevelopedanddeveloping ountries,hasledtotherapidde-

velopmentof wirelesstele ommuni ations. Indeed, wireless ommuni ations

hasbe omeafa torin e onomi development[20,28℄.

ˆ Non ooperative Approa hes have been adopted to solvemany proto ol

designissuesinwirelessnetworks 9

. Inamulti-usernetwork,servi esarepro-

vided tomultiple usersin whi h ea h userisassumed toberationalenough

to a hieve their individual highest performan e. Therefore, game theoreti

formulations an be used, and a stable solution for the players an be ob-

tained through equilibrium analysis [29℄. In a resour e managementgame,

multiple players (i.e., users and network servi e providers) are assumed to

a t rationally to a hieve theirobje tives. Thesolution to the game anbe

obtainedbymaximizingthenetworkservi eproviders'protswhilesatisfying

theusers[30℄.

1.2 Problem Formulation

Inmostsystemsforwireless ommuni ations,e ientutilizationofs ar eresour es

(i.e.,frequen yspe trum,energy onsumption,et .) hasbe amethekeyenablerto

satisfyboththeservi eprovider 10

,(SP, theoperator), andtheusers. Aprovider

of wireless ommuni ation servi es wants e ient utilization of the system sin e

theproviderderivesmorerevenue by providing servi esto moreusers. The user

inturn wantsagoodquality-of-servi e(QoS) atareasonable ost[22℄. Although,

newpri ingrules 11

havebeenestablishedinthewirelessmarket,pri ingbenetsto

theusers,SPs havebeenae ted negativelybyade rease in theirrevenues. The

hallengeistodesignme hanismsthatallo ateradioresour ese ientlyaimingto

losetherevenuegapwithouthurtingusers'performan e.

Moree ientutilizationofexisting resour esandlow ostdeploymentarekey

solutionsinvestigatedinthisthesisintheformof ompetitivegames:

1. Open Spe trum sharing aims to investigate how this paradigm ae ts the

utilization of available frequen y spe trum resour es aswell as the impa t

9

Thesemethodsaremostlyknownandproposedasgametheoreti approa hes.Inourspe i

ase, the game theoreti analysis has been arried out viamean of simulationand the Nash

equilibriumpointhasbeen al ulatedonlyforsomesamples enarios.

10

Providinghigherrevenueswhi h ompensatesthe ostsofservi eprovisioning.

11

of this spe trum sharing on SPs' utility 12

and users' performan e. In this

se tion,weaddressthefollowingresear hquestion.

ˆ Howthepropagation onditionsinthefrequen y hannels 13

mayimpa t

the performan e of the system in terms of energy expenditure while

providingservi es?

Westudyaparti ularsystemwherelimitedavailable spe trumresour esare

allo ated among the SPs 14

dynami ally as short-term allo ations, through

au tionme hanismsviaaSpe trumBroker.

2. Network deployment strategies are sought that lower investment osts, in-

reasesSPs' prots,andprovideshigh usersatisfa tion. S enarioswithspa-

tially heterogeneous user distribution are examined. We rst onsider the

asewhenusersareuniformly distributed a rosstheservi earea. Following

this, the aseof highly populatedareaswith in reasingdemand for wireless

servi es,i.e.,hot-spots,ismodeled onsideringthisasadrivingfor eforen-

trantSPstodeploywirelessa essnetworksinaduopolymarket.Itisworthy

mentioningthatin umbentSPsarealwaysabletodeploynewbasestations

(BSs) -if this would yield higherprotsfor them. Weareinterestedin an-

alyzingthe ase whenin umbentSPsare notwillingto extendtheirservi e

oeringareasbydeployingmoreBSs. Studiesaddressingthe onsequen esof

thisde isionarepresented. Thefollowingresear hstatementsarestudied.

ˆ Howtheservi eprovider'srevenueisae tedbythelevelof ompetition

(whi h is later referred to asthe level of overlap) and the tra load

variationsinthesystem? Wealsoinvestigatewhethertheusers'quality

of servi e, QoS,available data rate, and ost per Megabyte is ae ted

bythesetwoparameters.

ˆ Whi hpri ingstrategyshouldtheSPsimplementinorderto maximize

theirprotsina ompetitiveenvironmentwithaheterogeneousdemand

andunderdierentmarketshares?

ˆ Throughsimulationanalysisweintendtoinvestigateandestablish,whether

or not, it is suitable for the New omer SP to deploy a network in a

duopolymarketa ordingtothelo ationofanIn umbentSP,andifso,

whereitshouldbepla ed.

12

One analsorefertoservi eproviders'revenue.

13

Thisisalsoreferredas hannelheterogeneitylaterinthisthesis.

14

Inthiss enariowehaveassumedthattheservi eprovidersoersubs ription-basedservi es

andthattheuserswhohavesubs ribedtoaservi e annot hangetheirsubs riptionuntiltheir

1.3 Overview of the Thesis Contributions

Thekey ontributions of thethesis in ea h of the hapters are summarized next,

in ludingpreviouslypublishedmaterialontheresear hideasandobtainedresults.

Chapter 2

Chapter2fo uses on ompetitivespe trum sharingforheterogeneous hannels 15

.

This heterogeneity with regardto thefrequen y hannelsae ts theprotability

oftheSPs due to thefa t that thedieren ein hannel hara teristi sinuen es

theenergyexpenditurerequiredtoservetheenduser. Therefore,asub-problemof

optimizingtheenergy onsumptionemerges. Thissub-problemhasbeenaddressed

in:

ˆ Paper 1: M.Ter ero,PamelaGonzalez-San hez,ÖmerIleri, andJensZan-

derDistributedDynami Spe trumA esswithEnergyConstraintforHet-

erogeneousChannels,inpro eedingsof thefth international onferen eon

CognitiveRadioOrientedWirelessNetworks and Communi ations(Crown-

Com),Cannes,Fran e,June2010.

More spe i ally, Paper 1 studies a ompetitive s heme that an be used to

optimizespe trum utilization and power onsumption via optimally utilizing the

heterogeneityof the hannels in adistributed manner. Thepaperdenesthedif-

ferent distributed spe trum a ess algorithms used in our model and provides a

omparativeanalysis withthe referen eme hanisms whi h provide lowerand up-

perboundsin termsofspe trumutilization. Resultsobtainedfromthesimulation

ofthemethods thathavebeenstudiedarepresented.

Inthispaper,alltheauthors ontributedin devisingtheproblem formulation.

The modeling, simulation, and writing pro ess were done by the author of this

thesisandMiurelTer ero Vargas. ProfessorJensZanderand ÖmerIleriprovided

valuableinsightregardingthedire tionofthepaper.

Chapter 3

Chapter3presentsananalysisbasedonnetworkdeploymentstrategiesunder om-

petitive settings. The ee t of ompetition and market shares on the SPs' prof-

itability andon theusers'performan ehas been studied forsomespe i sample

s enarios. Wefo usonthe asewhere wirelessnetworks(withdierentSPs)with

15

Channels are onsideredheterogeneous (dierent) interms of propagation hara teristi s,

limitedavailablebandwidth opewiththe problemof howto maximizetheirnet-

work revenues 16

. Non ooperativegames between usersand SPs are investigated

and an open wireless a ess market is introdu ed based on thedierent network

deploymentstrategies. Werstanalyzethesystemperforman eintermsofservi e

provider'srevenue,users'qualityofservi e,QoS,availabledatarate,and ostper

Megabyte,assumingthattheusertra isuniformlydistributeda rossthe over-

agearea.

Anexplanationanddetailed resultsofthis analysisarein ludedin:

ˆ Paper 2: Pamela Gonzalez-San hezandJensZander. DeploymentStrate-

gies in Competitive Wireless A ess Networks, in pro eedings of the rst

international Conferen eon Advan es in CognitiveRadio (COCORA), Bu-

dapest,Hungary,April2011.[BestPaperAward℄

Next,we onsideramorerealisti situationwheretheusertra isnonuniformly

distributedandthedemandisdire tlyinuen edbythepri eoftheresour es. We

investigate demand-based revenue maximization in ompetitive network deploy-

ment games. The ontribution in this ase is given by a demand-based pri ing

strategyasatoolforservi eproviderstomaximizetheirrevenuesina ompetitive

wirelessa ess market. The des riptionof results and on lusion of thisstudy is

in ludedin:

ˆ Paper 3: PamelaGonzalez-San hez,SaltanatM.Khamit,andJensZander,

Competitive Pri ing with Demand Heterogeneity in Open Wireless A ess

Markets, in pro eedings of the third International IEEE Congress on Ul-

tra Modern Tele ommuni ationsand Control Systems(ICUMT), Budapest,

Hungary,O tober2011.

The last study on network deployment strategies in ompetitive s enarios is

based on nonuniform tra distributions. We onsider an open wireless a ess

marketwithanIn umbentSPandaNew omerSP.Theobje tiveofbothSPsisto

attra tandretain usersin ordertomaximizetheirprots. Adetailed explanation

andresultsarein ludedin:

ˆ Paper 4: PamelaGonzalez-San hezandJensZander,CompetitiveA ess-

pointDeploymentinMobileBroadbandSystems,inpro eedingsofthetenth

16

HereweanalyzeSP'sprotratherthanrevenue. Thisshiftinmetri o ursinthepapers3

S andinavianWorkshop onWirelessAd-ho Networks (ADHOC'11),Sto k-

holm,Sweden,May2011.

Paper4addresses themain problemthat aNew omerSP fa eswhen entering

thewirelessmarket,i.e.,todetermineafeasiblelo ationforitsBS(ora ess-point,

AP)in order to apture mostof thepopulation ( ustomers)andthus be sustain-

ablein themarket. We analyzewhether ornot itis protable for theNew omer

SPtodeployitsownnetworkandifso,wheretheBSsshouldbepla ed. Thiswill

beformulatedasasimple ompetitivenetworkdeploymentstrategythatprovides

insightforreal s enariosand ontributesto improvingfuturenetworkdeployment

modelsofrevenue-seekingservi eproviders.

Papers2,3,and4,areresultsofresear hdis ussionsbetweentheauthorofthis

thesisandProfessorJensZander. Themodeling, simulation, andwritingofthese

paperswereperformedbytheauthorofthisthesis. ProfessorJensZanderprovided

valuableinsightand omments on erningthedire tionofallthepapers. Saltanat

M.Khamit ontributedin thewritingpro esstoPaper3.

1.4 Thesis Outline

This thesis is organized in two parts. The rst one, ontains the ontents from

Chapters 2through 4. Chapter 2and 3 briey summarize the studies that have

beenperformed,in ludingashortrelatedliteraturereviewhighlightingthespe i

resear hareaswehave onsidered.

InChapter 2,wemotivatewhy ompetitive spe trumsharing hasbeen onsid-

eredinthisresear h. Thespe i s enariounderinvestigationisalsoexplainedand

nallysomeresultsareintrodu edfollowedbyashort dis ussion.

Chapter 3presents studiesof network deploymentstrategies forrevenueseek-

ing servi e providers under ompetitive settings. It ontains details of the basi

assumptionswehave onsidered,theresour eallo ationme hanismthat hasbeen

used,theopen wirelessa essmodel,and threedierents enariosthathavebeen

investigated.

Chapter 4 ontains on luding remarks and suggests some ideas for possible

futureresear hideas. These ondpart onsistsofverbatim opiesofallthepapers

in ludedin thisthesis.

Competitive Spe trum Sharing

The urrently enfor ed spe trum management me hanisms for ellular networks

mostlyrely on stati allo ation of frequen y hannels a rossSPs. In this allo a-

tionmethod anSP getsdedi atedusagerightsforaspe i band foralongterm.

Su hstati allo ationme hanismsguaranteeinterferen e-freeoperationandex lu-

sive rights to oer mobile servi es. Debates on erning higher exibility for the

spe trum li enses have be ame more ommon of late and the ne essity to make

hangesintheregulatorys hemeisbe omingevident[31℄.

Resear h suggests that stati spe trum allo ation methods are ine ient in

terms of spe trum utilization for dynami tra , espe ially when the demand

hangesdrasti allyoverthetime[32℄. Therefore,dynami spe truma essme ha-

nisms(DSA),su hasreal-timeallo ationsofspe trum,wouldbeusefultobalan e

demandandsupply. Inthis hapterwestudy dynami me hanismsin thespe i

formof ompetitivefrequen y spe trumsharing amongwirelessnetworksthatoer

mobileservi esbaseduponasubs ription.

2.1 Related Literature

Therapidlygrowingdemand forwireless ommuni ationservi esand thete hno-

logi aldevelopmenthave reatedaspe trum shortage. This apparentparadox is

ommonlyreferredas spe trums ar ity. Severalauthors arguethat spe trumper

se is not s ar e - rather, it is ine iently managed, in ways that restri t users'

optionsto exploit thisresour ee iently[33℄. Inlightof this,there isaneed for

moree ientuseofavailablespe trumresour es. Forthisreasonthedevelopment

of new s hemes for dynami spe trum sharing, aiming to avoid the ine ien ies

inherentin traditionalli ensing,havere entlyattra tedsigni antinterest.

Mostoftheearlier ontributions on erningDSAs hemes on entratedons e-

narioswherethefrequen y hannelsare onsideredasidenti alresour es(i.e., ho-

mogeneous hannels),sin eallofthe hannelsareassumedtobeequalintermsof

propagation hara teristi s[14,3436℄. Insu hs enarios,DSAme hanismsareof-

tenimplementedintheformofau tionswherethepri esforthedierent hannels

donotdier. Intheseau tionstheau tioneerkeepsin reasingtheunit ostforthe

hannelsuntilthetotalbandwidthdemandislessthanthetotalbandwidthsupply.

Onlya few of the previousstudies on DSA s hemes [3739℄ have onsidered het-

erogeneous hannelsettings,assumingthatthe hannelshavethesamebandwidth

buthavedierentpropagation hara teristi s. Therefore,theusersexperien edif-

ferenttransmissionranges,andthusdierentfrequen iesaremoreorlesssuitable

fordierentuserlo ations. These studiesfo usonthedes riptionofhow hannel

heterogeneity anbeaddressedin the ontext ofDSA. However,these studies do

notevaluatetheee t ofenergy ostone ientspe trumutilization.

KhamitandZander[40℄proposeda ompetitivespe trumallo ationme hanism

forwirelessnetworksin whi hheterogeneous overagewasaddressed. Theauthors

investigatedwhetherornotitisprotableforasmallSPtodeployitsownnetwork

and to ompete against another SP who provides a wide servi e area. A DSA

me hanismwasimplemented but, the hannelswere assumedto behomogeneous

hannels.

2.2 Why Spe trum Sharing?

Measurementsofthespe trumusagehavedemonstratedthatwithstati spe trum

sharing 1

thisresour eisidle 2

mostofthetimeandthatnewte hniquesfora ess

andsharingofthespe trumresour esareneededinordertoin reaseitsutilization.

Spe trumsharing te hniques allowSPs to ee tively get larger hunks of spe -

trum(seeFigure2.1)anddierentfrequen ybands anbemoree ientlyutilized

(trunkinggains), thus enablinghigherpeakdataratesto theend users.

Figure2.1: S hemati representationoftwodierentmethodsforspe trumsharing.

The hallengethat emergesin this ontext isto establishhow to dynami ally

sharethespe trumsothatinterferen eisminimizedandspe trumise ientlyuti-

lized,whileSPs' revenuesaremaximized. Thisvaluableresour e anbemanaged

1

Thisreferstothe urrentwayofspe trummanagement.

2

FORHETEROGENEOUSCHANNELS (PAPER 1) 15

intwoindependentmanners: as ooperativesharingwhereagreementsbetweenSPs

mustbeestablishedand ompetitivesharingwherea oordinatorofthe ompetition

maybeneeded. In ompetitivesharinga esss hemesalltransa tionsandstrate-

giesare ontrolledthusfairnessinrevenuegenerationandusers'performan e an

beassured. InthisChapter,weinvestigatethelater ase,i.e., ompetitivesharing.

2.3 Distributed Spe trum A ess with Energy Constraint

for Heterogeneous Channels (Paper 1)

In this se tion we look at the s enario where SPs share a pool of limited spe -

trum resour esin a ompetitivemanner. Weanalyze thebehaviorof the system

foroverload situation, onsideringahigh(butmanageable)loadwhi hisgreater

thantheavailable supply. Twomain aspe ts arehere onsidered; hannel hetero-

geneity whi h impliesenergy onstraint andweinvestigatetheimpa t onthe

systemperforman e.

ˆ ChannelHeterogeneity : Channelsareassumedtobeheterogeneousbased

on the fa t that they are lo ated in widely separated frequen y bands and

would show dieren es in transmission ranges meaning that they dier in

propagation onditionsand hen ein inferen elevels. Ouraimis toidentify,

under ertain basi assumptions, howmu h spe trumutilization anbein-

reasedbya ountingfortheheterogeneityofthefrequen y hannels.

ˆ Energy- onstrainedWirelessNetworks:aswidelyknown,resour eman-

agementhasalwaysbeenatoolforSPsto improvetheperforman eoftheir

networks. Consequentlytransmission poweris an important resour eto be

managede iently. Energy- onstrainedwirelesssystemshavebeenanalyzed

giventhein reasedawarenessabouttheex essiveenergy onsumptionofthe

ommuni ationsystems 3

. It is likely that the energy onsumption will be-

omeamajor on ernevenfordown-linktransmissionsininfrastru turesys-

tems[41℄. Consequently,thepropagation hara teristi sofavailable hannels

should be onsidered in spe trum allo ation de ision-making. On light of

this,wealsoanalyzetheee tofenergy- onstrainedwirelessnetworksonthe

system'sperforman eunder ompetitivesettings.

Thebasi ompetitivesharing s enariounder investigation isaddressedin Figure

2.2. Weinvestigatetheimpa toffrequen y hannelsindierentbands,heterogene-

ity,withdierentspe trumpri esandwealsoanalyzedierentPower ostrequired

toprovidetheservi e.

3

Figure2.2: WirelessNetworkStru ture-a ompetitivesharings enario.

TheSPs ompeteagainstea hotherinadynami fashioninordertogetsome

frequen y hannelsfromthespe trummanagerentity(introdu edintheliterature

as the Spe trum Broker [14,16℄) and being able to serve their subs ribed users

whileselshlytrytomaximizetheirrevenues,networkutility 4

. Usersareassumed

to havea xed subs ription and theyhavethe optionto onne tonly to the SP

theybelong to. Weexplore  hannel heterogeneity, whi h is a uniquefeature of

ognitiveradionetworks,where hannelspresentdierent hara teristi s[37℄.

TwodistributedDSAme hanismswhi htakeintoa ountthepropagation har-

a teristi sandpowerrequirementsofdierent hannels;Sequential andCon urrent

spe truma ess havebeenstudied. Theobje tiveisto providedierent meansof

propagation-aware DSA me hanisms 5

, so that we might be able to understand

whi h ompetitive s hemes should be used to optimize spe trum utilization and

power onsumption as well as how the heterogeneity of the hannels should be

optimally utilized. Two regimes of interest are onsidered, the rst one is when

thetransmissionenergy ost 6

islow,potentiallyrelatedtoawirelessnetworkwith

large overage area, and the se ond one when transmitting represents high ost

orrespondingtoawirelessnetworkwithapoor overage.

Wemodelthetransmitpowerthat agivenservi eproviderr(SP

r

)requiresto

rea huseriusingthe hannelj (frequen yf

j

),asstatedin thefollowing:

P

(r;i;j)

= Co:(d

r;i )

:(f

j )

2

; (2.1)

4

Inwirelesssystemsnetworkutility anbealsoreferredtothethroughputofthesystem.

5

Referring to me hanisms that take into a ount the heterogeneity of the hannels when

makingde isions regarding spe trum allo ation. This allows to apply dis riminatory pri ing

strategies.

6

where Co is a onstant omputed from the SNR threshold, noise power and the

speedoflight,d

r;i

isthedistan ebetweenthebasestationoftheSP

r

andtheuser

i. Note, the dependen e of P

(r;i;j)

onthe propagation hara teristi of hannel j

thatisbeingutilizedandtheuserlo ation.

An important out ome from the au tion pro ess is the SPs revenue that has

beenintrodu edas theutilityfun tion, U

(r;i;j)

,whi h isusedasade ision-taking

elementbytheSPs. Basedonthistheservi eisprovidedtothoseusersthatshow

positiveutility 7

,(U

(r;i;j)

>0),andisdened asfollows:

U

(r;i;j)

= x

user C

ost(j) K :P

(r;i;j)

; (2.2)

herex

user

is thexed pri ethat ea h userpaystothe SPfor theservi e,C

ost(j)

is the hannel ost that keeps varying during the au tioning phase(in monetary

units)andK istheenergy ostpersessionin monetaryunit/powerunit.

2.4 Results and Dis ussion

Themain ndingsof this study are illustratedin Figures 2.3 and 2.4and anbe

summarizedasfollows:Theresultsfrom on urrents hemeshowthatbyau tioning

the hannelsbasedontheirpropagation onditionsa onsiderablegainisprovidedin

termsofspe trumutilization. Consideringalowvalueofenergy ost 8

is onsidered

thesequentialand on urrentme hanismspresentalmostthesameperforman eas

theCentralized in terms of hannel o upan y, allo ating themaximumpossible

numberof hannels. Byassumingthatthevalueofenergy ostin reases 9

agreater

dieren e between sequential and on urrent s hemes an be observed, showing

thatthe on urrenta essalmostapproa hestheupperbound 10

.

ThemaximumutilitythataSP anper eivebyservingauserisobtainedfromthe

Centralizedme hanismasitwasexpe tedsin ethe hannelpri eis onsidered

zero, see Fig.2.4. However, when the hannel pri e is taken into a ount in the

distributed me hanisms, sequential and the on urrenta ess, the last one gives

higherutility. Furthermore,it an beobservedthat forlowenergy ostvaluesan

overloadsituationleadstodeathra e betweenSPs. The hannel ost,C

ost ,and

subsequenttothisfa ttheSPutilitystartstode rementradi allyafter veusers.

7

TheusershaveQoSrequirementintheformofre eivedSNRthresholdvalue. Wealsoassume

thatea hSP'sbasestationemployspower ontrolsothatthetransmitpoweratagiven hannel

istheminimumpowersu ientfora hievingtherequiredSNRvalueatthedesignatedenduser.

8

Forthesimulationanalysisalowvalueofenergy ostismodeledwithK=10.

9

Highvaluesofenergy ostarerepresented withK=1000.

10

1 2 3 4 5 6 7 8 9 10 11 12 13 10

20 30 40 50 60 70 80 90 100

Number of users in the system

Channel Occupancy

Sequential K=10 Concurrent K=10 Centralized K=10 Homogeneous K=10 Sequential K=1000

Concurrent & Centralized K=1000 Homogeneous K=1000

Figure2.3: ChannelO upan yasafun tionofthenumberofusersintheSystem.

1 2 3 4 5 6 7 8 9 10 11 12 13

0 20 40 60 80 100

1 2 3 4 5 6 7 8 9 10 11 12 13

0 20 40 60 80 100

Number of users in the system

Operator Utility

Sequential K=10 Concurrent K=10 Centralized K=10 Homogeneous K=10

Sequential K=1000 Concurrent K=1000 Centralized K=1000 Homogeneous K=1000

Figure 2.4: Servi e provider utility as a fun tion of the number of users in the

System.

2.5 Summary

Paper 1

Basedonthesimulationresults,it anbe on ludedthatthe hannelheterogeneity

provesto bean important fa t to be taken into a ountin order to improve the

spe trumo upan y. IntermsofprotabilityfortheSPs someother ooperation

strategiesmaybemoree ientleadingtobetterprots. There isanee t,obvi-

ouslyobservedinau tionme hanisms,thatpointstoSPswillingtopayhighpri es

forspe trumli enses with theobje tive to in reasetheir market share aiming to

keepentrantsoutof thewireless ompetition. Toavoidthis situation ithasbeen

assumedthattheSPsoersubs ribedservi esandbiddingforafrequen y hannel

thatwillnotbeusedmayonly ausenegativeimpa tintheirutilities.

Inenergy- onstrainedwireless network theSPs haveto beawareof thepower

expenditurewhen hoosingtheuser hannelpairs. Thesequentiala essme h-

anismpresentshigherpower onsumptionperea hserveduserduetothefa tthat

allthe hannels areau tioned oneby one onse utivelyandthe SPsdonotknow

whetherornotthenext hannelisbetter. Underthis ondition,the on urrenta -

esss hemeprovestobethebetteroptionfordynami spe truma ess. Extended

explanationand resultsarepresentedin[42℄.

Network Deployment Strategies in

Open Wireless A ess Markets

Therapidin reaseofmobileinternettra hasputthespotlightonhowthefuture

wirelessbroadbanda ess systemsshould bedeployedand operatedatsigni ant

lower ostspertransmitted bit than today. Closing the revenuegap aused by

thelethal ombinationofatraterevenuestreamsandtherapidlygrowing osts

asso iatedwith onventionala essdeploymentisontopoftheprioritylistofmost

wirelessmobileservi eproviders.

Figure 3.1: Servi eproviders- hoosetheirtarget areasandQoSoerings

Basedonthete hnologi aldevelopmentandbusinessmodels,thetrendinwire-

lessa essmarketsispointingtoamarketwithplenty ofSPsutilizingnumerous

te hnologies and ompeting for users(see representation in Figure 3.1). Network

deploymentstrategiesthatallowSPs ooperateand ompeteatthesametimewhile

maximizingtheir prots should be devised. On lightof this, ompetitivesharing

me hanisms( oopetition)where ompetingSPs ooperateby hosentheirtarget

servi eareastryingtoavoid omplete overlapwiththeir ompetitor areinvesti-

gatedin thisSe tion.

The rest of this Chapter is organizedas follows: Se tion 3.1 ontains related

literaturein orderto spe ify theresear harea onsideredin this hapter. Se tion

3.2des ribesthedynami resour eallo ationme hanismappliedin alltheinvesti-

gateds enarios. In Se tion 3.3a ompetitivesharingme hanismthat allowsSPs

to identifya suitable overage overlapamong theirnetwork in order to maximize

revenuesisinvestigated. Se tion3.4addressesdierent ompetitivegamesbasedon

networkdeploymentstrategiesforprotmaximizationusingasimulationapproa h.

3.1 Related Literature

Thetraditionalwayofinfrastru turedeploymenthasbeenthateveryoperatorpro-

videshisowna esssysteminalllo ations,i.e.,a hievingfull overagebyhimself.

Thishasbeenpossibleinmostmobilephonesystemsduetotherelativelylow osts

andhighprot margins. Asthein reasing dataratesrequireamu h denser(and

moreexpensive) network of base stations, full overage is nolonger an optionto

mostoperators. Instead Infrastru ture sharing, where providers share infrastru -

turein low user density areasis one possiblealternativeto oer better overage

and QoS in a ost e ient manner [26,43℄. The sharing of wireless infrastru -

ture,however,raisesthequestionofhowresour esandrevenuesshould bedivided

whenmultiplesubsystems,managedbypotentially ompeting a tors,areinvolved

indeliveringthea essservi e. Analternativewouldbetosharetheinfrastru ture

impli itlybyestablishinganopenwirelessa essmarketwhereinnetworksnotonly

ompeteforusersonalong-termtime-s ale,butalsoonamu hshortertime-base.

This ould berealized withan ar hite turewhere autonomous trade-agents,that

residein terminalsanda ess points(APs),managetheresour esthroughnegoti-

ations[2427,44℄.

In [24℄, the authors developed a framework for studying demand-responsive

pri ingin ontexts where a esspoints (APs)with fully overlapping overage

ompeteforusers. Theyshowthatanopena essmarketresultsinbetterservi es

atlowerpri ewhi hinthelongtermalsoyieldsmoresatised ustomers ompared

to as enariowhere APs ooperate. A market-basedframework for de entralized

RRMinenvironmentspopulatedbymultiple,possiblyheterogeneousAPs, wasin-

trodu ed in [26℄. Theproblem addressedfor the user is to determine how mu h

resour estopur hasefromdierentAPsinordertomaximizeitsutility(valuefor

money).

rational, hoosingstrategiesinordertomaximizetheirownutility. Insu hagame

formulationof theresour emanagementproblem, thesystemperforman e anbe

expressedin termsoftheNashequilibrium,orsolutionofsu hagame,whennone

oftheusers anfurtherimprovetheirutility.[29,30℄.

3.2 Resour e Allo ationMe hanism

Inthisstudyweuseagame theoreti approa handtheproportionallyfairdivisi-

bleau tionme hanismintrodu edin[24,26,4547℄. Theresour esto beallo ated

amongusersistransmissiontimewhi hisdividedviaemployingtheaforementioned

au tionme hanism.

In aproportionalshare fair allo ations heme ea h user is hara terized by a

parameterthatexpressestherelativeamountoftheresour ethatitshouldre eive.

Hereafter,thebidthattheusersubmitstotheBSisusedtoexpresstheuser'sshare.

Inthis work adynami systemhas beenmodeled in whi h usersare assumed to

dynami allyjoin andleavethe ompetition (game). Therefore,theportionof the

resour edepends on boththe numberof users that enter the game and thelevel

of ompetitionatdierenttimes. Thisme hanismallowsexibility,sin etheusers

ande idewhentojoinorleavethe ompetition,andensuresfairnesswhi hfollows

from the fa t that the usersalwaysget a share of the resour eproportionallyto

theirbids(asexpressedinEquation3.1).

Figure3.2: Representationoftheau tionpro edureasso iatedwithaletransfer.

Inthisexampletrade-agentj initiates aletransferin au tion1.

We model a le download servi e, spe i ally, the download time in a wireless

areas. TheBSsareassumedto beidenti alin transmitpower,systembandwidth,

minimumre eivedsignalto noiseratio requirement, et . Weassume that there-

sour eisinnitesimallydivisibleandthatthe ostasso iatedwiththeletransfer

dependsonthetotaltime-durationandthemonetaryexpenditurerequiredforthe

ompletele download. AsitisshowninFigure3.2,thetransmissiontimeisallo-

atedin severalau tions.

Weimplement atrade-agent-basedmodelfor theau tion bidding pro ess.The

trade-agentsare entities lo ated in the BSs, who a t selshly on behalf of their

users. Themainobje tiveofea htrade-agentistomaximizeitsuser'sutility(here

omputedasvalue for money). Theportion ofthetransmissiontimeallo atedto

userj anbeexpressedasfollows:

x

i;j (s) =

s

i;j

s

i;j +S

i; j

2[0;1); (3.1)

wheres

i;j

denotesthebidin monetaryunits thatuserj pla esduringau tioniin

ordertogetaportionoftheavailabletransmissiontimex

i;j

foraletransfer. S

i; j

representsthe strategies(bids)of alltheopponenttrade-agentsandit isequalto

P

k 6=j s

i;k

+wherethereservationpri e2(

min

;

max

). Thereservationpri eisa

nonzeropri eoorbelowwhi htheresour ewillnotbesold. Notethatbydenition

thepri eoormustbenonzeroasifitwerezero,thentherewouldbenopri eoor.

Assumingthatthepeakdata-rateofasingleuserj onwhosebehalfthetrade-

agent j isa ting, remains un hangedduring the entire le transferand that this

appliesforalltheusers,i.e. R

i;j

=R

z;j

8i;z,thetotaldemandasso iatedwiththe

othertrade-agents,thus P

k 6=j s

i;k

= P

k 6=j s

z;k

8i;z. Notethat zisthelastround

ofthe au tion asrepresentedin 3.2. Due to these assumptions, ea h trade-agent

willpla eidenti albidsinalltheau tions.

3.3 Deployment Strategies in Competitive Wireless A ess

Networks (Paper 2)

Inthisse tionwestudyhow ompetitivesharing( oopetition),wherevariousa -

essprovidersprovidepartiallyoverlapping overageina ompetitivefashion, an

generateadditional gains to the SPs. Basi ally, we analyzehow the thebalan e

betweenareasofex lusive overage(asshownin Figure3.3)whereea hSP has

amonopoly situation  and overlapareas with provider ompetition, ae ts the

protability of the a ess providers. We use a game theoreti approa h and the

proportionallyfair au tion me hanism,des ribedabove (Figure3.2). Weseek to

NETWORKS(PAPER2) 25

thelevel ofoverlap? and further: Is the userQoSin termsof available data rate

and ostperMegabyteae tedbytheoverlap?

Thes enariosstudied anbeillustratedasinFigure3.3,wheres m

i;j

denotesthe

bid,thatuserj pla esin au tioni atBSm(Notethat wehaveassumedapurely

timedivisionmultiplexed link). Thelink user SP indi atesthelink provided by

theSPwhodominatesthemarketinthisarea(i.e.,thea essproviderswhoprovide

overage)andit isto thisBS thatusersshould send apositivebidin order tobe

served.

Figure 3.3: Basi s enario - Illustration of a wireless network ar hite ture with

dierentper entagesofoverlapping,whi hrepresentsasystemunderdierentlevels

of ompetition

Theuser's performan e (QoS) is quantied through theaverage user throughput

andmonetary expenditure pertransferred Megabyte. Theusers ompete against

ea h other forresour es- whiletrying to maximizetheirutility fun tion in order

to transfera le. Thisgame is expressed in Equation(3.2). Forouranalysis, we

assumethatthelesize isnite(andidenti al),q=1Megabyte.

'(s

j

) = argmax

sj U

i;j (s

j

;S

j

) (3.2)

8 j 2f1;:::;Ng;m2f1;2g:

IntheaboveequationU

i;j (s

j

;S

j

)isrelatedtothethroughput,x

i;j R

i;j

,asso iated

U

i;j

= 2

X

m=1 max



0; x

i;j R

m

i;j s

m

i;j



: (3.3)

Derivingtherstorder solution(i.e.,asalinearequation)ofEquation(3.3)with

respe t to s m

i;j

we anobtainthe best response (BR ) whi h des ribes how trade-

agent j should rea t to the strategies (optimal bid that the trade-agent should

submittheBSs)ofalltheothertrade-agentsinordertomaximizeitsuser'sutility.

Thiswouldbeexpressedasfollows:

s m

i;j

= s

R m

i;j (

X

k 6=j s

m

i;k +

m )

X

k 6=j s

m

i;k +

m

: (3.4)

Sin ethe peak transferrateforall ofthe usersisthe sameoverall au tions, and

theyallhavetotransferthesamesizele,thengivingea huserthewhole hannel

(i.e.,allofthetimeslots)enablesthisuserto ompleteandleavethesystem,hen e

leavingalloftheremainingresour esfortheremaining users.

Ase ondgametakespla eanditisdenedasopenwirelessa essmarket,when

BSs ompete forusers and selshly, tryto maximize their own expe ted revenue

perse ond,asdened inEquation(3.5).

m (

m

) = argmax



m

(

m

;

m

); (3.5)





m

= 

m (



m

) 8m2M; (3.6)

where

m (

m

)representsthebestresponse(BR ) fun tionasso iatedwithBSm.

Equation(3.6) des ribes theNEP whi h is the solutionto the ompetitivegame

amongBSs.

Wefo usinndingtheNashEquilibriumPoint(NEP)forthereservationpri e

oftheresour e,, onsideringthetwogames( ompetitionamongusersforresour es

andamong BSsfor users) in the ompetition area fordierentlevelsof overage

overlap. This NEP isrelated to theBestResponse from thetrade-agents. In the

monopolist area(non-overlapping overage) only ompetition among users is ob-

served. Byobtainingthe NEP weare ableto analyze theBS's expe ted revenue

withdierentlevelsof ompetition. These resultsenable us to predi tthe users'

NETWORKS(PAPER2) 27

le).

The stability and uniqueness of the NEP for the games have been al ulated

throughiterationsviameanofsimulation. Furtherdetailsontheseresultsandthe

pathloss model aswell as the peak data-rate version used in our s enario anbe

foundin[48℄.

Results and Dis ussion

Theaverage revenueasso iated with the BS game for dierent levelsof overage

overlap anbeobservedinFigure3.4. Astheoverlappingareabythetwowireless

networks in reasessodoes thelevel of ompetition and moreusersexperien e an

open a ess market. The reservation pri e for the resour ede reases as a onse-

quen eofthe ompetitionleadingtolowerBS'srevenue.

0 20 40 60 80 100

0 5 10 15 20 25 30 35 40 45 50

Level of coverage overlap in the system Average revenue per BS φ ( ε 1 , ε 2 )

Representation of the level of competition between base stations

% % % % %

10% of Overlap

35% of Overlap

65% of Overlap

Completely overlaping Fully competition Nonoverlap − Monopoly situation

Figure3.4: Averagerevenueperbase stationandtimeslot asso iatedwiththeBS

gameunder dierentper entagesof overlapping overage

Figures 3.5 shows that theexperien ed user'sQoS isae ted forlowdemand

density. However,we annoti ethatforloaddensityhigherthan3.2Megabits/se ond

(=0.4les/s)thedegradationisslightlysmallerleadingtolessimpa tontheuser's

experien edQoSandprovidingagreatgain(morethan50%)in theBS'srevenue.

IthasbeenshownthatastheBSsoverlaplessin overage,thelevelof ompetition

de reasesandthe BSs an hargethe usersin thenonoverlappingareasahigher

0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6 200

400 500 600 700 800 900 1000 1200 1400

File arrival rate, λ [files/BS/s]

Average user throughput [kbits/s]

Nonoverlapping 10% Overlapping 35% Overlapping 65% Overlapping 100% Overlapping Potentially offered load, D ₀ [Mbit/BS/s]

0.8 1.6 2.4 3.2 4 4.8

Figure 3.5: Average throughput experien ed by users for dierent per entages of

overlapping overageasafun tionofthepotentiallyoeredload.

loaddensityof 3.2 Megabits/se ond,and higher, the gainin throughput may be

smallergivinglessimpa tin theuser'sexperien edqualityof servi e(See Figure

3.5). Ontheotherhand,oursimulationsindi atethatawin-winsituationforboth

usersandBSs anbea hievedwithasuitableoverlap overagebytwonetworks.

A ording to our results a proper per ent of overlap might be approximately

35%dependingontheinterestorobje tivefun tionofalltheinvolvedparties. Fig-

ure3.4representsthebehavioroftherevenueperBSonea hau tionround,where

on average 0.4 les/sarrive to ea h BS, ea h le is of size q = 8 Megabits. We

a knowledgethe fa t that thesystem's performan ewasanalyzed by onsidering

onlytwowirelessnetworksin order toget insightonto whi h extent ompetition

anbebene ialforbothprovidersandusers. Explanationanddetailedresultsof

thisanalysisarein ludedin[48℄.

3.4 Competitive Games for Revenue Maximizationwith

Heterogeneous Demand (Paper 3, Paper 4)

In order to maximize network revenue and be ompetitive in the market prot-

seekingSPs shallutilizetheirresour ese ientlyandpri etheirservi esproperly

basedonthedemand responsiveness. Pri ing strategiesplayan importantrolein

HETEROGENEOUSDEMAND (PAPER 3,PAPER 4) 29

taintyin tra demand impliesriskinrevenuegenerationforSPs. Inthis se tion

weanalyze thepri e-demand relationshipand basedon thisweaim to proposea

pri ingstrategyasatoolforservi eproviderstomaximizetheirrevenue.

3.4.1System Model

Weanalyzeas enariowheretwowirelessnetworksaredeployedinadenselypopu-

latedareawithnonuniformlydistributedusers. Wemodeltheuserpopulationwith

aGaussiandistributionanditsprobabilitydensity anbeexpressedasbelow:

f(l)= 1

p

2


e (l )

2

2

2

; (3.7)

where is themean (lo ationof thepeakof thedemand),  2

isthevarian e(the

measureofthewidthofthedistribution)fortheuserpositionlwithinthesystem

area[0-L℄whereL=3R . HereR isthe ellradiusdened inTable1in[49℄.

We assume an interferen e-free system and that the twonetworks, belonging

todierentSPs, overlappartiallyin overage. Wefo uson ommuni ationin the

downlinkdire tion,i.e.,fromtheBSstothemobiles.

WirelessA ess Market- Model

Theusersexperien etwolo ation-basedwirelessmarkets: monopolya essmarket

andopen a ess market asillustratedin Figure 3.7. The monopolymarket is ob-

servedwhileusersarelo atedin thenon-overlappingareas,sin ethereisonlyone

BSprovidinga esstoitsnetwork. Whentheusersarewithintheoverlapping ov-

eragebythetwonetworks, theyexperien eanopena essmarket. A ompetitive

gametakespla eamongtheBSs,whoselshly,trytomaximizetheirownexpe ted

protperse ond,asdened inEquation(3.11).

Theusersareableto pi ktheBSthatoersthehighestpeakdata-rateat the

lowest pri e(highest estimated utility). Ea h BS broad asts its reservation pri e

amongtheuserslo atedwithin its overagearea. Note,thatapri edierentiation

isused by theBSs, meaningthat theusersin themonopoly overage areamight

experien ehigher ostsintheabsen eofany ompetition.

Asexplainedin[48℄,wemodelaledownloadinaTDMAsystemandapplya

trade-agentbasedme hanismforthebiddingpro ess.

Resour esareallo atedusingaproportionalfairau tionme hanism(thismethod

is explained in detail in [48℄) and based on their bids userswill get a portion of

Notethat

min

,thepri eoorinEquation(3.1),isarepresentationofthexed

ostin urredbytheSP,

min

=Cost. Forthisstudywehaveassumedaxed ost



min

=2,whi hhasbeenpreviously onsideredin [14℄.

UserModel

ˆ Utility Fun tion

In this analysis, we assume that a user pi ks the BS that provides the highest

estimated utility and it is given the hoi e to not enter the system if the pri e

establishedbytheBSistoohigh. Theusermaximizationproblemisintrodu edin

Equation(3.8).

maximize

^

U m

i;j (R

m

j

;

m

) (3.8)

8j2f1;:::;Ng;m2f1;2g;

where

^

U m

i;j

is theuser's estimated utilitywhi h is used asthede ision-taking pa-

rameterdened asbellow:

^

U m

i;j

= R

m

i;j



m

: (3.9)

HereR m

i;j

isthepeakdata-ratethattheuserexperien esbasedonitslo ationand



m

isthereservationpri ebroad astedbytheBSs.

ˆ A eptan e Probability

Letus dene x

1

asthe overageof BS

1 and x

2

asthat ofBS

2

. Then, it is lear

thatuserj lo atedatx

1

\x

2

,(seeFigure3.6)willthereforeprefer,initially, BS

1

ifandonlyif

^

U 1

i;j

>

^

U 2

i;j

duringau tion1.

On e theuser hasde ided whi h BS provides thebest servi e,its satisfa tion

andsothewilinesstopayfortheoeredservi e anbemeasuredwithana eptan e

probabilitydenedasfollows:

A m

i;j

= 1 e

(

^

U m

i;j )



(s m

i;j )



; (3.10)

where the , and  are appropriate positive onstantsthat determine thelevel

HETEROGENEOUSDEMAND (PAPER 3,PAPER 4) 31

Figure3.6: Basestationslo ation-lineargeographi alregion

s m

i;j

represents the pri e that the user pays during ea h au tion y le, and it is

equivalent to the bid that this user submits to the BS, (see Equation 3.1). This

a eptan eprobabilitymodel isa modiedversionof the oneused in [14,16,40℄.

After therst stage (au tion 1), if the userstarts transmitting, itshould remain

onne tedtothis basestationuntiltheletransferis ompleted.

BaseStations BSs- ProtMaximization

TheBS'sinterest,instead,istomaximizationitsprotbyservingasmanyusersas

possible. Thedemand-basedprotmaximizationproblemfortheBSsisformulated

bymaximizingthesumofallsubmittedbidsbytheusersthata epttoenterthe

gamein ea h au tion y leiandisdenedasfollow:

maximize N

X

j=1



s m

i;j :A

m

i;j



: (3.11)

Thesummation of all theusers that hoose BS m formes the generated demand

whi h a epts the servi e with a probability A m

i;j

. The average revenue per BS,

the monetary expenditure and average user throughput, all asa fun tion of the

resour epri e,havebeenusedto measurethesystemperforman e.

3.4.2Pri ing Game with Heterogeneous Demand in Open A ess

Markets (Paper 3)

In this paper we, basi ally, study the ase where N ompeting users are non-

uniformlydistributed a rossthe overagearea,x 2 [0;L℄,whi his sharedbytwo

basestationsBSs(weassumedthattheBSsbelongtodierentSPs). Thiss enario

isrepresentedinFigure 3.7.

These two networks are deployed with areasof ex lusive overage (monopoly