1 Topologicaldesign,routing andhand-over in satellitenetworks

AFONSOFERREIRA

JÉRÔMEGALTIER

PAOLO PENNA�

AbstractIn this chapter, we survey communicationrelatedissuesarisingin thecontext of

Low EarthOrbit (LEO) satelliteconstellations.In particular, we studythe impactof the predictablemovementof the satelliteson the techniquesusedin topologicaldesign,routing,andhand-overstrategies.

1.1 INTRODUCTION

A Low EarthOrbit (LEO) satelliteconstellationconsistsof asetof satellitesorbitingthe Earthwith high constantspeedat a relatively low altitude(a few thousandsofkilometers)[1]. Eachsatellite is equippedwith a fixed numberof antennasthatallow it to communicatewith groundtransmitters/receiversandwith othersatellites.Oneof themajoradvantagesof LEO satellites(asopposedto geostationary– GEO– satellites)is that they are closer to the Earth’s surface. This allows to reducethe communicationdelayandthe energy requiredto directly connecta userwith asatellite.

On the other hand, two major issuesarisedue to their low altitude. First, asinglesatellitecanonly coverasmallgeographicalarea(calledfootprint) attheEarthsurface,many satellitesbeingthusrequiredto provideglobalcoverage.Second,thefootprintof eachsatellitemovescontinuously, implying ahighmobility of thewholenetwork, in contrastwith othercellularsystems.

�Partially supportedby aEuropeanRTN fellowshipfrom theARACNE project.

i

ii DESIGN, ROUTING AND HAND-OVER IN SATELLITE NETWORKS

In the following, we will seehow the topologyof LEO constellationsis limitedby physicalconstraints.Then,wewill review how thesefactorshavebeentakenintoaccountin thedesignof routingandhand-overpolicies.

1.2 TOPOLOGIES

Duringthesystemsdesignphase,severalparameterscomeintoplay, suchassatellitesaltitude, numberof satellites,numberof orbits andof satellitesper orbit, how todeploy theorbits,andhow to inter-connectthesatellites.All suchfactorsdeterminethetopologyof thenetwork, asshown in thissection.

1.2.1 Orbits

A closerlook atthefeasibletypesof orbitsshowsthatunlesstheorbitshavethesamealtitudeandinclination,theirrelativepositionchangesooftenthatinter-satellitelinks(ISLs) canhardlyconnectthemfor a sufficient amountof time (for moredetailsonorbitmechanicswith respectto telecommunicationservices,see[1, 39]). Undersuchconstraints,differentkinds of constellationscanbe obtainedaccordingto how theorbitsaredeployed.

The so-called � -constellationsare the structureof the Iridium system[20, 22]andat thebasisof theoriginal plansfor Teledesic[21, 30]. Thebasicstructureof a� -constellationconsistsin a setof orbitsthataredeployedalonga semi-circle whenviewed from a pole,asshown in Figure1.1(a). The satellitesareplacedalongtheorbits soasto obtaina maximumcoverage of the Earth’s surface. In Figure1.1(c)thedeploymentof satellitesalongwith their footprintsis shown. Wecanseethatin a� -constellationtherearetwo extremeorbitswhich areadjacent,but whosesatellitesmove in oppositedirections. As a result,a seamappears,that dividesthe networkinto two parts: thosesatellitesmoving from southto northandthosemoving fromnorthto south(seeFigure1.1(a)-(b)).

PSfragreplacements

seam

orbitsmovement

satellitefootprint

PSfragreplacements

seam

orbitsmovement

satellitefootprint

���� ������ �� � ������ ��������

������ ��

���� ��

�� � !

"#$$%%&�&&�&''

PSfragreplacementsseam

orbits

movement

satellitefootprint

(a) (b) (c)

Fig. 1.1 Thestructureof � -constellations:(a) view from thenorthpole; (b) view from theequatorialplane;(c) thepositionof satellitesonadjacentorbitsandtheresultingcoverage.

TOPOLOGIES iii

From a communicationnetwork viewpoint, the seamis the main drawback of� -constellations,as it will be seenlater in the text. Also, � -constellationssufferfrom excessive polar coverage. Finally, their uniquecoveragein many areasand,therefore,sensibility to many obstacles,like treesandbuildings, doesnot alwaysensurea sufficient radiosignalquality.

In orderto avoid this kind of problems,()� -constellationshavebeenproposed.A(*� -constellation,consistsin spacingtheorbitsalonga completecircle asshown inFigure1.2. The ()� -constellationis usedin theGlobalstarconstellation[9], andhasalsobeenplannedfor thefutureSkybridgeprojectandthenow abandonedCelestri.

(a) (b)

Fig. 1.2 ()� -constellations:(a)view from thenorthpole;(b) view from theequatorialplane.

Another importantaspectconcernsthe useof “inclined” orbits, that is, orbitswhoseinclination is betweenthe equatorialinclination (0 degrees)and the polarone(90 degrees).Usually, � -constellationsusepolarorbits (informally, orbits that“roughly” crossthepolaraxis) for coveragereasons(seeSection1.2.5below), andthereforearecalled“polar” constellations.On theotherhand,inclinedorbitsallowabetteroptimizationof ()� -constellations,hencethename“inclined” constellations.The useof inclinedorbits allows to compensatefor the satellitesmobility with theEarth’sself rotation,soasto increasetheamountof time a satelliteis visible from afixedpointon theEarthsurface(seeSection1.2.5below).

It is worth to observe that thereis no technicalreasonto forbid the useof polarorbitson (*� -constellationsandreversely. Moreover, theuseof inclinedorbitsdoesnotaffectthenetwork topology(for instance,� -constellationsthatuseinclinedorbitsstill resultin themesh-like topologyshown in Figure1.4(b)).

1.2.2 Inter -SatelliteLinks

The next stepis to inter-connectthe satellitesthroughthe ISLs. In particular, wedistinguishbetweenintra-orbital and inter-orbital links: The former connectcon-secutive satelliteson thesameorbits,while the latterconnecttwo satellitesthatareon differentorbits. In Figure1.3 we show threepossiblepatternsthat canbe ob-tainedby usinginter-orbital links betweenadjacentorbits: the“W” patternandthe“inclined” patternin Figures1.3(a)-(b)usefour ISLs persatellite,while thepatternin Figure1.3(c)usesonly threeISLs.

iv DESIGN, ROUTING AND HAND-OVER IN SATELLITE NETWORKS

(a) (b) (c)

Fig. 1.3 Someinter-satellitelink patterns.

Considernow the “W” patternin Figure1.3(a). In order to obtainthe networktopologywehave to take into accounttheseamandtherelativepositionof satellitescrossingthepoles,asfollows.

For � -constellations,onehasto considertheproblemof connectingtwo satellitesmoving in oppositedirections,which is too expensive or even infeasiblewith theexisting technology(seeSection1.2.5). Hence,it is commonlyassumedthat twosuchsatellitescannotbe directly connectedover the seam,even thoughthey are“physically” closeoneto eachother. Therefore,long user-to-userdelaycanoccureven whenthe two partiesaregeographicallycloseto eachotherbut the coveringsatellitesareseparatedby the seam. Also noticethat two adjacentsatellitesswaptheir relative positionwhenever crossingthe poles(seeFigure1.4(a)). Hence,thenetwork topologycanberepresentedasa two-dimensionalmeshwherecolumnsarewrappedaround,but rowsarenot (seeFigure1.4(b)).

In [15] theimpactof theISLs architecture(for instance,theuseof antennasthatsupporthigherangularvelocity) hasbeenstudied,and further patternsto connectthesatellitesof a � -constellationhavebeenproposed.Suchpatternsuseinter-orbitallinks thatconnectsatellitesin nonadjacentorbits,typically theneighboringorbit oftheneighboringorbit. This reducestheuser-to-userdelaywhenthecommunicationtakesplacebetweentwo positionsthat arequite far (or whenthe communicationshaveto goacrosstheseam).Undertheassumptionof ISLs thatsupporthighangularvelocity, thedelayeffectsof ISLs thatcrosstheseamhave alsobeeninvestigatedin[15].

With respectto inter-satellitelinks for ()� -constellations,neitherGlobalstarnorSkybridgehave implementedISLs in their design,althoughit seemsthat they havebeenconsideredin theearlyphaseof theseprojects,asit wasthecasein Celestri.Atthatpoint in time,many designersthoughtthoseprojectswereinnovativeenoughtodelaytheintroductionof this additionalnew feature.Nevertheless,thereis a strongbelief thatfuturedesignsof (*� -constellationswill introducesuchlinks.

Fromthetopologypointof view, it is worth to observethattheregulartorusturnsinto a skewed torus if an inclined ISL patternsuchas the oneof Figure 1.3(b) isadopted[17]. Notice that (*� -constellationsdo not presentany seam. Thus, theircoveragehassmootherproperties. On the other hand,a uniqueposition may becoveredby two satellitesquite far onefrom anotherin the network topology(e.g.,

TOPOLOGIES v

PSfragreplacements

+ ,

+.-, -

northsouth

PSfragreplacementsnorth

south

south(a) (b)

Fig. 1.4 The relative positionof two adjacentsatellitescrossingthe pole andthe resultingtopologyof � -constellations.

two satellitesthatmove in oppositedirections),especiallywhentheuseris closetotheequator.

1.2.3 ISLs versusTerrestrial Gateways

The use of ISLs is intendedto implementcommunicationsthat do not use anyterrestrialinfrastructure.However, theuseof terrestrialgatewaysstill presentsomeadvantagessuchasa reducednumberof computingdeviceson-boardthesatellites.For instance,gatewayscanbeusedto computetheroutingtablesthatareusedby thesatellites.

A moreextensiveuseof thegatewayshasbeenadoptedin theGlobalstarsystem,wherethesatellitesoperatein a“bent-pipe”mode.Theirmainfunctionis to redirectusersignalsto groundgateways,andvice-versa.As aresult,theoperatorhasto buildmany gateways,onefor eachareain which theserviceis opened.Additionally, partof the radiospectrumis usedto supportthe communicationsbetweenthe satellitesandthe gateways. Unfortunately, radio resourcesarebecominga scarceresource.Currently, severalsystemssharethesamespectrumof frequencies(Globalstar, ICO,andprobablyEllipso)which is thesourceof severalinterferenceproblems.

We note that the useof ISLs presentssignificantadvantages,like reducingthecommunicationsbetweenthe satellitesand the gateways, reducingthe numberofgateways,balancingtheloadbetweenthegateways,andpreventinggateway faults.

1.2.4 Multiple Coverage

Another important issue for satellite constellationswith ISLs is risen by multi-coverage goals. From the radio and signal propagationpoint of views, a singlesatellitemaynotsufficeto ensurethereal-timeconnection,especiallyif someobsta-clesexist betweentheuserandthesatellite.Systemslike Globalstar[9] answerthis

vi DESIGN, ROUTING AND HAND-OVER IN SATELLITE NETWORKS

problemusingmulti-pathtechniques: Insteadof beingreceivedby onesatellite,thesignalis receivedby two to four satellitesandmergedto recoveraclearsignal.Whenanew satelliteis visibleto theuser, its signalcontributionis introducedprogressivelyinto theglobalmergingof thesignals.

We remarkthat routing with multi-pathtechniquesin a satelliteconstellationisverychallenging.A singleusermaybedirectlyconnectedto two (or more)satellitesthat are very far one from anotherin the network topology, mainly in inclinedconstellations. From the algorithmic point of view, this characteristicessentiallyturnsabasicnetwork routingprobleminto amulticastingproblem.

1.2.5 Physical and TechnologicalConstraints

In this sectionwe discusssomeof themainphysicalandtechnologicalfactorsthatimpacton many of theabovedesignchoices.

ISLs Geometry. The main technologicalconstraintsto take into accountin ISLdesignaretherelativeangularvelocityof theendpointsandtheirvisibility [17]. Thisis becauseantennascannotafford excessiveangularspeedandtheatmosphereis alsoa sourceof fadingof thesignal.

Mobility of ISLs. As a satellitemovesalong its orbit, the setof satellitesvisiblefrom it changescontinuously. Thishappensfor thosesatellitesthatarenotin adjacentorbitsand,in polarconstellations,whenever thesatelliteapproachesthepoles.Thisis dueto thesmalldistancebetweenadjacentsatellitesapproachingthepolewhichresultsin ahigherangularvelocity[1, 15]. Additionally, ISLsbetweenadjacentorbitsmustbeturnedoff whencrossingthepolesbecauseof thesatellitesrelativepositionswitching(seeFigure1.4). As observed in [15], ISLs that supporthigherangularvelocityallow to maintainintra-orbitallinks athigherlatitudes.An unexpectedside-effect of the angularvelocity is that the trackingsystemmay affect the stability ofthesatellitewithin its orbit andthereforeresultin anadditionalconsumptionof fuel,which in turn impactsthesatellite’sweightandtime in service.

Shortest(Delay)Path. It is worthto observethatthedistancebetweentwo adjacentpolar orbits decreasesas they get closerto the poles. Hence,for � -constellationsusing the “W” ISLs pattern,for instance,the minimum delaypath is the one thatusesaminimalnumberof ISLsandinter-orbital links whoselatitudeis themaximumlatitudebetweenthetwo satellitesto beconnected.

Notice that routing algorithmson mesh-like topologiesmay returnsub-optimaltime/delaypaths,sincesuchmodelsdonotconsiderthattheorbitsdistancevarieswiththelatitude. In [10] a modelthattakesinto accountthis issuehasbeeninvestigated.

NETWORK MOBILITY AND TRAFFIC MODELING vii

1.3 NETWORK MOBILITY AND TRAFFIC MODELING

Therearetwo mainfactorsthatshouldbetakeninto accountwhendesigningroutingalgorithmsfor LEO satelliteconstellations:

/ Users’ distribution: the fact that thepositionof theusersandthedurationofthecommunicationsarenot known in advance.

/ Networkmobility: thefactthatsatellitesmove,constantlychangingthenetworktopology.

Although the first aspecthasbeenextensively studiedfor classicalcellular net-works,suchnetworksusewired connectionsin orderto connecttwo basestations.Hence,themainissuein these“terrestrialnetworks” is to provideenoughresourcesfor theuser’sconnectionto last. Thereis alot of flexibility in thesizeof thecells,buttheusersmaymove from oneto another, andat differentspeeds.Conversely, LEOcells are big enoughto considerthe usersimmobile. However, routing problemsoccursinceon-boardresources– in particularthemaximumnumberof connectionsusingISLs– arescarce.

Thesecondaspect,namelythenetwork mobility, is adistinguishingfactorof LEOconstellations.Indeed,evenif weassumeastaticsetof communications(i.e.,pairsofusersthatwanttocommunicateonewith theother),theproblemof maintainingactivetheseconnectionsovertimeisnotatrivial task:Thesatellitesmovementtriggersbothhand-oversandconnectionsupdates(re-routing),whena topologychangeoccurs.

In both cases,mobility is the main causeof call blocking, call dropping,andunboundeddelay in communications.However, thereis a fundamentaldifferencebetweentheusers’mobility andthenetwork’smobility: Theusers’behaviour is notdeterministic, while changesto the network topologyarepredictable. Hence,twodifferentapproachesaregenerallyadopted:

/ The network’s behaviour is deterministicandcanbe “predicted” quite accu-rately(seeSection1.3.1).

/ Theusers’behaviour is usuallymodeledby meansof aprobabilitydistribution(seeSection1.3.2).

It is worth to observe that if we considerthe relative movementbetweena userandthesatellites,thenthemajorpartof suchmovementis dueto thesatellitesspeed.Hence,the probability distributionsusedto modelusers’mobility mainly focusonthe issueof managingrequestswhosepositionanddurationarenot known prior totheir arrival.

1.3.1 SatellitesMobility

Oneof themaindifferencesbetween“classical”cellularnetworksandLEO constel-lationsis thehighmobility of thesystem.ComplicatingfactorssuchasthesatellitesmovementandtheEarth’sself rotationmaketheproblemof connecting“immobile”

viii DESIGN, ROUTING AND HAND-OVER IN SATELLITE NETWORKS

usersnontrivial. In thefollowing we describethe interplaybetweenthesetwo fac-tors andthe previously mentionedaspects,andalsohow network mobility canbemodeled.

1.3.1.1 SatellitesMovement Thesatellitesmovementis themaincauseof hand-overs.Two typesof hand-overmayoccur:

/ A satellitehand-over is thetransferof a userfrom a satelliteto anotherduringacommunication.

/ A cell hand-over is thetransferof a userfrom a spot-beamto another, withinthesamesatellite. A satelliteantennadirectedto terminalsis composedby aseriesof beams.Suchadecompositionof thesatellitefootprintallowsto reusetheradiofrequenciesseveraltimesin its coveragearea.Thosehand-overshaveno consequenceon the inter-satelliterouting, but impactseriouslyon-boardcomputations.

If auseris justontheborderof thecoverageareaof asatellite,his/herconnectiontimeto an individual satellitecanbe extremelysmall. Hence,in general,constellationsaredesignedin suchawaythatthefootprintsoverlapandextremelysmallconnectiontimesto anindividualsatelliteneverhappen.Nevertheless,themaximumconnectiontime is still limited. A user’s trajectory, viewed from the satellite,will resembleastraightline crossingthecenterof thecoveragearea.Theapparent(or relative)speedof the useris thenthe speedof the satellite. This causesthe following undesirablephenomena:Visibility changes,varying topologies(ISLs changes),footprint hand-over, andneedfor re-routing.

1.3.1.2 Earth’s Self Rotation The Earth’s self-rotationintroducessomemorecomplicationin thesystem.In Figure1.5,we plot themaximumtime betweentwosatellitehand-oversagainstthealtitude 0 andtheelevationangle1 of aconstellation,in two cases:

/ The Earth’s self rotation is not taken into considerationand the satellite’sinclinationcanbearbitrary.

/ TheEarth’s self rotationis taken into accountandthe orbit of thesatelliteisequatorial.

1.3.1.3 ModelingtheNetwork Mobility Noticethatthemaximumhand-overtime,shown in Figure1.5,canvaryfrom someminutesup to severalhours.Also, inclinedorbitscanbeusedto exploit theEarth’sself rotationto increasethevisibility period.Hence,themobility of thenetworkcanalsovaryalot. Roughlyspeaking,they canbedistinguishedbetweenlow andhighmobility, dependingonthemaximumhand-overtime.

LowMobility (periodic). In [5], themobility of asatelliteconstellationis describedasa Finite StateAutomata(FSA) by a seriesof statesdescribedalongthe time in

NETWORK MOBILITY AND TRAFFIC MODELING ix

1

2

5

10

20

50

100

200

500

1000

2000

500 1000 2000 5000 10000 20000 35500

Max

imum

tim

e be

twee

n tw

o ha

ndov

ers

(in m

in)

2

Satellite altitude (in km)

720 minutes = 1/2 day

60 minutes = 1 hour

Minimum elevation angles10 degrees, without Earth’s self-rotation20 degrees, without Earth’s self-rotation30 degrees, without Earth’s self-rotation40 degrees, without Earth’s self-rotation

10 degrees, equatorial orbit20 degrees, equatorial orbit30 degrees, equatorial orbit40 degrees, equatorial orbit

Fig. 1.5 Maximumtime betweentwo satellitehand-overs.

round-robinfashion.Themainadvantageof this modelis thatwe have to consideronly afinite setof configurationsof thesatelliteconstellation(wherethesatellitesareassumedto be immobile), andprovide efficient routing solutionsfor eachof them,inspiredfrom classicaltelecommunicationproblems.

Low Mobility (aperiodic). It is worth to observe that the“periodicity” assumptionof the FSA model may be, in somecases,too strong. This is essentiallydue tothecombinationof “physical” factors,suchastheEarth’s self rotation,thesatellitesspeed,theuseof inclinedorbits,etc. They makethesystemaperiodicfor all practicalusages,i.e. asatellitewill findagainthesamepositiononlyaftersuchalongtimethattoo many intermediatestateswould benecessary. In this case,a possibleapproachconsistsin taking a seriesof snapshotsor fixed constellationtopologies,methodsometimesreferredto asdiscretization[11, 37, 38]. Then, the routing problemissolvedwith respectto thatfixed“constellation”.

High Mobility. The above two modelsare interestingwhen the mobility of thesatellitenetwork is negligible with respectto the mobility of the usersrequests,e.g. if mostof the requestshave very low duration,let ussaya few minutes,whilethehand-over time would beof onehouror more. In this case,beforethenetworkconfigurationchanges(significantly)several(many) requestswill havebeensatisfied.

Ontheotherhand,thesemodelsdonot takeinto accountthedependencebetweenconsecutive statesof the network. Thus,betweentwo statesthe completeroutingschemeof theconstellationshouldbechanged.Clearly, in thecaseof highlydynamic

x DESIGN, ROUTING AND HAND-OVER IN SATELLITE NETWORKS

constellationsand/orlongcalldurations,almostall requestsmaypassthroughseveralstatesandthusmaybere-routedseveraltimes.

1.3.2 UsersDistrib ution: CommonTraffic Assumptions

Dependingon the application,threemajor scenarioscanbe identified for satellitemarkets. The first one, and the most naturalone, statesthat satelliteswill servecountrieswherethe telecommunicationinfrastructureis insufficient or inexisting.Thesecondone,which appearsasmoreandmoreprobable,is thatthesatelliteswillprovideadditionalcapacitiesto countriesthatalreadyhavegoodtelecommunicationinfrastructures,but which suffer from anoverloadof the resources.A third marketconcernspeoplewho requirea seamlessconnectionin their internationalactivities.Of course,dependingon thescenario,thetraffic mayhave differentcharacteristics,assummarizedin Table1.1.

Type Developing Overload Inter national

Location poorcountries/oceans rich countries internationalTime distrib ution Poisson-like bursty nearlydeterministicUserconcentration sparse huge irregularCall duration short exponential longCall distance average short long

Table 1.1 Characteristicsof foreseeableusagesof satellite constellations.

Little is known on the two first classesof applications. The last onehasbeeninvestigatedin [35], whereananalysisof theinternationalactivities led to a mapofdifferentzonesworldwide. In this model,theplanisphereis divided into 288cells,with 24 bandsalongthe longitudesand12 alongthe latitudes. The intensitylevelsfrom 0 to 8 shown in Table1.2 correspondto traffic expectationsfor theyear2005,of 0, 1.6, 6.4, 16, 32, 95, 191,239, and318 millions of addressableminutes/year,respectively. In [15] thetraffic requirementmatrix is obtainedfrom tradingstatistics,namely the imports/exports betweenany two regions. Furthermarket studiesonsatellitescanbefoundin thequarterlyreport[23].

In the following we describehow the usersmobility canbe modeledby meansof sometraffic assumptions.In particular, we grouptraffic assumptionsinto threecategories:

/ Geographicaldistribution: Onwhichsatellitetheuserrequestsareexpectedtoarrive.

/ Timedistribution: How long they areexpectedto beactive.

/ Ratedistribution: How muchof theresourcesthey will require.

NETWORK MOBILITY AND TRAFFIC MODELING xi

3 3 1 1 3 4 3 2 3 3 3 3 23 4 5 1 6 5

1 1 7 2 1 1 1 8 8 4 7 7 5 11 3 5 6 7 7 1

3 5 3 5 4 6 55 2 4 5 4

2 3 46 2 3 3 2

4 3 3 3 311 34 455125552

3 5 6 1677 6761

3 3 2 1 4 5

55 4

2

6 3 2 4 2

7

763 4 6

8572

463

2

5 45

Table 1.2 Intensity levelson the planispherefor the distrib ution of users.

1.3.2.1 Geographicaldistribution Statisticalmodelshavebeendevelopedto rep-resenttheloadall over theEarth.A structurethatappearspromisingis thenotionofpointprocessoverthe2-dimensionalEuclideanspace[8, 3, 16, 18]. A pointprocessis a family 35476�398;:=<?>@:BA9C.D of non-negative integer-valuedvariables,where398E:F< denotesthe randomnumberof pointsthat lie in the set : . A homogeneousPoissonprocesswith parameterGIHKJ is a pointprocess3 asfollows.

/ Thenumberof points 3L8;:F< is Poissondistributedwith parameterGNMO8;:F< foreachboundedBorel set : .

/ Therandomvariables398E:QPR<?>TSUSUST>V398E:XWY< areindependentfor eachsequence:QP*>USUSTSZ>?:[W of disjoint Borel sets.

In fact, the homogeneousPoissonprocessreflectsquite well the traffic loadwithina countrywith uniform development.More generally, whentrying to mapa pointprocessto the entire world it would be interestingto either choosea measureMthatreflectstheeconomicdevelopmentof eachregion (i.e., for instance,themapofTable1.2),or try a modelwith differentproperties,suchasMMPP or multi-fractalmodels[2] (for moredetails,we referthereaderto thesurvey in [13]).

1.3.2.2 TimeDistribution Thetraditionalwaytomodelthedistributionof thecalldurationsconsistsin usinga Poissonlaw. In fact, thebehavior of thetraffic is thenverycloseto thatobtainedonphonesystems.However, new broadbandapplications,madepossibleby theInternet,generateothertypesof traffic. In [24] a comparativestudybetweenself-similarandPoissontraffics is donein the satelliteconstellationcontext.

1.3.2.3 RateDistribution It is quite naturalto relatethe ratedistribution to thelocationsof thedifferentpartiesof a communication.In [36], the loadof the inter-

xii DESIGN, ROUTING AND HAND-OVER IN SATELLITE NETWORKS

continentaltraffic is evaluated. It is estimatedthat between81% and85% of thetraffic is within continents,theremainingtraffic beingsharedwith theclosestand/ormost populatedareas. Another methodof generatingtraffic is suggestedin [6]:Oncea pair 8]\^>V_Y< of locationsis selected( \ and _ areviewedhereaspointson theunit sphere),associatedwith potentialrequirementdensities̀ba and `bc , the trafficrequirementbetweenthetwo nodesis givenby

d 8;\e>?_N<e4 8]`baf`gcf<ihj 8]\^>V_Y<]k >where l and m aretwo parameterssetby theuser. In [6], it is assumedthat ln4oJpS qand mr4oJpSts .

1.4 ROUTING AND HAND-OVER

A goodroutingstrategy shouldmainly preventfrom (1) thecongestionof ISLs dueto too many routespassingtroughthem;(2) routingrequestsalongpathscontainingmany links sincethis resultsinto apoorresourcesutilizationandin ahigherdelayinthecommunication;(3) droppinganongoingcall or blockinganew one.

1.4.1 Problemsand Optimization Criteria

Here,we describemorein detail the main goalsconcerningthe designof efficientroutingalgorithmsfor LEO constellations.

MaximumThroughputandISLsUsage. Maximizing thethroughputunderlimitedISL capacityappearsasoneof the main objectivesof the constellationdesigners.Clearly, becauseof thelimited ISL capacity, agoodroutingstrategyshouldminimizethemaximumlink usage(e.g.,theloaddueto theoverall traffic passingthroughsuchlink) amongall theISLs.

Shortest/BoundedDelay andJitter. Oneof themainmotivationsof LEO systemsis thereductionof thecommunicationdelay. Indeed,theminimumdelayto openaconnectionthrougha geostationarysatelliteis around240 ms, while a LEO couldconnecttwo usersin around20 ms. However, while the connectionis roughlyindependentof theparties’locationin aGEOsystem,thedelaysignificantlyincreasesfor LEO systemswhenthepartiesget furtherfrom oneanother. However, sincetheISLsofferstraightfree-spacepropagation,thedelaybetweenthesatellitesisgovernedby thespeedof thelight.

Themulti-pathtechniquesandin-the-airmergingof thesignal(seeSection1.2.4)shouldraisea new delayproblem. Indeed,merging two signalsthat arefar in thenetwork topologytakestimecomparableto theonerequiredto reachageostationarysatellite.Thesameproblemoccurswith � -constellationswhenthecommunicationshave to go acrossthe seam. In eachcase,the communicationdelaycould take, in

ROUTING AND HAND-OVER xiii

theworstcase,anadditional100msto becompleted(thetime to reachthefurthestsatellitein theconstellation).

The jitter (in other words, the delay variation) is relatively important in LEOconstellations,sincethedistancebetweentheuserandthesatellite(andalsobetweenthe satelliteand the gateway, and even betweentwo satelliteof different orbits)changescontinuouslyduringthelifetime of acommunication.Thisbehavior cannotbe avoided,but canleadto storageoptimizationissuesin the satellite(in casetheterminalis notableto handleit) for instance.

GuaranteedHand-Over. The next optimizationproblemconcernsthe quality ofserviceandmorepreciselytheguaranteethat is givento theusersthata communi-cationwill notbedroppedbecauseof ahand-over. Thiscanbedoneeitherby fixinganacceptablerateof call dropping,or by forcing thesystemto avoid call droppingat any price.

Call AdmissionandRouting. Theguaranteedhand-overfeaturegreatlyimpactsthecall admissionprocedureandcanleadto additionalcall blocking. Blockedcallsmayalsobe a consequenceof scarcelink resourceavailability. This mayhappeneitherbecauseausercannotbeconnectedto thevisiblesatellite(s)or no routebetweenthetwo satellitesusedin thecommunicationcanbefoundwithoutoverloadingtheISLscapacity.

1.4.2 Algorithmic Solutions

1.4.2.1 Call Admissionfor Hand-Overs The call admissionproceduredecideswhetheranincomingcommunicationrequestwill behandledor not. A userwill berefusedthe entry in thesystemwhenthereis not enoughavailablecapacityto taketherequestin consideration.A usermaybealsorejectedbecausethesystemis notableto guaranteethe durationof the servicesufficiently enoughto meetquality ofservicegoals.For instance,ausercouldbeimmediatelyserved,but afteroneminutethe communicationwill have to be droppedbecauseit interfereswith otherusers,moreprivilegedor who werealreadyin thesystembeforethelatteruser’sarrival.

Two main schemescan be usedto control the resourcesof the LEO satellitesystem:

/ Earth-fixedcells canbe drawn directly on the ground. Onecell hasa fixedcapacity, andis servedby the same(setof) satellite(s).Therefore,thehand-oversoccursimultaneouslyfor all theusersof thecell. This schemereducestheamountof usablecapacityby thesatellite,but simplifiesthemanagementof theuserson-boardthesatellites.This ideahasbeenusedfor theplansof theTeledesicconstellation.TheEarth’ssurfacewasdividedinto stripesparalleltotheequatorof 160km from theSouthto theNorth,eachstripebeingredividedalongthelongitudesinto squaresof approximatively160km per160km, for atotalof around20000cells. Those“super-cells” werethenfurthersubdividedinto elementarycellsof 53.3km per53.3km [21].

xiv DESIGN, ROUTING AND HAND-OVER IN SATELLITE NETWORKS

It is thento theresponsibilityof theconstellation’sdesignerto makesurethatateachinstantof time eachcell will beservedby at leastonesatellite.

/ Satellite-fixedcells is the most commonscheme,applied in particular forIridium. Theusersarehandledindividually bythesatelliteandseveralsatellitesmayservea userat thesametime.

As mentionedin Section1.3,thesatellitespeedis responsiblefor thegreatestpartof thesystemmobility. After thatcomestheEarth’s self-rotation,andonly thentheterminal’smobility.

Theadmissioncontrolandresourceallocationwork hasbeenmainly focusedonsatellite-fixed cells. Usually, systemdesignerstry to includean overlappingareabetweenany two satellite-fixed cells. Therefore,someresearchershave proposedthat,whena terminalcomesinto the overlappingarea,andno channelis availablein its new covering cell, it issuesa hand-over requestin a queue(the Hand-overQueue- HQ) that haspriority over incoming calls [28]. Although this idea mayenhancethequalityof service,thefinal resultdependson many factors,like thesizeof theoverlappingarea,andthedistributionof thelengthof thecalls. Althoughthisprocedurecertainlyenhancesthesystem,thereis no meanto reacha target qualityof service.

Anotherideaconsistsin systematicallyreservingsomechannelsto thehand-overrequests[19]. In this Hand-overGard (HG) system,if thenumberof busychannelsexceedsa given threshold,then no incoming call is acceptedand only hand-overrequestsarehandled.This systembecomesmoreefficient whenthethresholdgoesdown. However, if the thresholdis too low, the systemwill be under-loadedinmany cases,asincomingcallsthatcouldhavebeenacceptedarerejected.Hencetheneedof a tradeoff betweenquality of serviceandsystemcapacity. In [7], additionalconceptsof geographicalpositionareintegratedinto thisconcept,sothattheauthorscanevaluateacall blockingprobabilitydependingontheremainingtimetheuserhasin his/hercell, andtheexpecteddurationof a call. A new useris acceptedinto thesystemif his/herhand-over call blocking probability will meetQoSrequirements,andhis/herinclusiondoesnot degradetheexisting callsundertheQoStarget. Thisresultsin amoreaccurateacceptance/rejectionof users.

In [29], the authorsconsider, for eachterminal, the servicingtime of a satellitecell. In particular, they analyzeits movementanddeducetheinstantsuiv and uxw whenthe terminal entersthe cell and when it leaves it, as shown in Figure 1.6. As aresult,whena terminal is introducedin the system,not only the presentcoveringcell is reserved,but alsoa sufficient numberof future cells, associatedwith futureutilization,sothattheservicedurationmeetsthequalityof servicerequirements(thisalgorithmis calledGuaranteedHand-Over– HG). If thecommunicationcontinues,thenadditionalreservationsaredonein real time. In theworstcase,whenthis life-timereservationsfail, theterminalcanbenotifiedthattheconnectionwill endwithina certainamountof time.

1.4.2.2 ISL Dimensioning In [4, 15], theimpactof theISLs designphaseon theoverall performancesof the network have beenpointedout (seeSection1.2.2). It

ROUTING AND HAND-OVER xv

service initiation

service end

movementof the cell

TerminalPSfragreplacements

Fig. 1.6 Instantsof serviceinitiation andserviceendfor a satellite-fixedcell.

is worth to observe that, in theseworks,a uniquepatternis chosenonceandfor allduringthedesignphase.

A different approachhasbeenadoptedin [37, 38] wherethe authorsconsiderdifferent discretetime steps(eachtime step correspondsto a “snapshot”of theconstellation,that is, the relative positionsof the satellites). For eachtime step,an ISLs dimensioningproblemis solvedsoasto minimize the maximumISL loadwhich, in turns,yieldsthesufficient ISL capacityto routetherequests.

1.4.2.3 PrecomputedRoutesandSnapshot/FSATechniques:DeterministicRout-ing In this sectionwe survey how thedeterministicbehaviour of thenetwork canbe exploited to enhancethe overall performanceof the network. In particular, thefollowing threeaspectswill be considered:(a) designchoices: how satellitesareinterconnectedthroughISLs;(b) reservationstrategies:how to guaranteethatacom-municationwill notbedropped;and(c) “ad-hoc”routingalgorithms:algorithmsthatdependon thecurrentstateof thenetwork.

Considera discretemodel in which eachstateof the network correspondsto avisibility staterepresentingthesetof satellitesthatarevisibletoeachother(i.e.,thosepairsof satellitesthat, potentially, canbe directly connectedto eachother). Then,two oppositepolicies(off-line or on-line design)canbe adopted:Either provide auniquestrategy whichdoesnotdependonthevisibility state(i.e.,doesnotvaryovertime),or provideadifferentstrategy for everyvisibility state.

In thesequelwefirst describesomebasicroutingheuristicsthatdonotexploit theknowledgeof thevisibility states.Then,we will seehow thesesolutionscanbenefitfrom thedeterministicbehaviour of thenetwork.

Basic RoutingStrategies. In [31] different routing and reservation strategies formesh-like topologies(a qzy|{}{ meshmodel is adopted)have beencompared.Asfor the routing strategies,to eachlink is associateda costfunction andthe routingalgorithmchoosesa pathwith minimumcost. Accordingto differentdefinitionsofthecostfunctionthefollowing four strategieshavebeencompared1:

1In [31] suchstrategiesarenamedMinimumHopsAlgorithm,MinimumCostAlgorithm,MeshAlgorithmandRevisedMeshAlgorithm, respectively.

xvi DESIGN, ROUTING AND HAND-OVER IN SATELLITE NETWORKS

/ Min Hops: thecostof eachISL is1,hencetheconnectionisalwaysestablishedbychoosinganypaththatminimizesthenumberof ISLsin it (i.e. theminimumnumberof hops).

/ Min Load: to eachlink is associateda cost given by {*~R_��f�U�����Z� , where_f���Z�f����� is thenumberof freechannelsin thelink; thechosenpathminimizesthesumof thecostof theISLs in sucha path.

/ Min HopsMin Load: amongthe setof pathswith minimal numberof hops,considertheonewith lowestmaximumlink usage.Then,thecall is acceptedif andonly if suchroutecontainsno link thatis overloaded.

/ RevisedMin HopsMin Load: it is definedasthepreviousone,with theonlydifferencethat, if a requestcannotbe routedalonga min-hopsroute, thenasub-optimaldelaypathis chosen.

Noticethat thefirst strategy arbitrarily choosesany min-hop(delay)path,while theotherstrategiesaim to keepthe link usageassmall aspossible. Interestingly, theexperimentalcomparisonin [31] showsthat,atleastfor someprobabilitydistribution,it is fundamentalto avoid the useof highly loadedlinks and the useof pathswithsub-optimallength. Indeed,the worsestrategy turnsout to be the Min Hopsone,while theMin HopMin Loadstrategy performsbetterthantheRevisedone.

FromOn-lineto Off-line: PrecomputedRoutes. In [6], thedeterministicbehaviourof thenetwork hasbeenalsoexploitedto enhancetheperformance(in termsof callblockingprobability)of the routingstrategies. In particular, theauthorscompareamin-coststrategy (similar to thosementionedabove) to a static routing, in whichthe routesare computedoff-line in advance. The network is modeledas a FSA(seeSection1.3.1.3)and,for eachstate(i.e., network topology),a setof routesisprecomputed.Clearly, this reducesthecommunicationoverheaddueto theperiodicbroadcastoperationsrequiredto updatethe link state(i.e., the load) information.Somehow surprisingly, thisapproachhasalsoabettercall blockingprobability. Thisis mainly dueto the fact that,aftera network change,the solutionprovidedby themin-costalgorithmis rather“unstable”.So,severalcallsarererouted.

This idea of moving the complexity of routing from an on-line problemto anoff-line oneis alsothebasisof [38, 37]: There,for eachpossibledemandpair (i.e.,for eachpossiblecommunication),� differentshortestpathsarecomputed.Thosepathsarethenusedduringtheon-linephasein which theroutingalgorithmchoosesa routebetweensuch � candidates.Basedon this idea,anupperboundon theISLscapacityis givenby thenumberof suchpathscontainingalink. Informally speaking,it is assumedthat there is a requestfor each possiblepair of satellitesandthegoalis to computetheminimum � suchthat,if every ISL capacityis � , thenevery requestcanchoosebetween� differentpathsof minimal length.Thisproblemis formulatedby meansof a linear programmingsystem. In [37] the authorsalso considertheproblemof relatingsubsequentstatesof thenetwork, sothatre-routingis performedonly whennecessary. Moreprecisely, aroutingstatewill becomputedandoptimizedwhile keepingsomeremainingroutesof thepreviousstates.

ROUTING AND HAND-OVER xvii

1.4.2.4 PredictingLink Hand-Overs Anotherapproachconsistsin takingadvan-tageof thelink hand-oversto managethemobility of thenetwork.

TheProbabilistic RoutingProtocol. Upon a call arrival, a first route throughthesatelliteshasto beassignedto it. Sucharoutecanbechosenbasedonany criterium,like minimumnumberof hops,leastcongestion,minimumcost,etc. Thedetermin-istic movementof thesatellitesallow thepredictionof the time framewherea linkhand-over is to occur. Hence,the probabilistic routing protocol – PRP–,proposedin [33], triesto establishanarriving communicationthrougharoutewhichhasmini-mumprobabilityto becutby a link hand-over. For this, it supposestheexistenceofa probabilitydistribution function(PDF)of thecall time durationovera route.

The protocol appliesDijkstra’s algorithm to find the routes. The cost of eachISL is setto one,implying that the routewill be minimum hop. The PDF is usedto remove from considerationof Dijkstra’s algorithmthoseISLs that are likely tohand-overduringthecommunication.

It is easyto seethatthePRPworksfor veryshortcalls,sinceadirectconsequenceof its implementationis that the route just setwill not be cut (with a given targetprobability) during the call. For instance,a target probability of JNS ��� reduceslinkreroutingoperationsby 80%, when comparedto the “pure” Dijkstra’s algorithm.Unfortunately, theprobabilityof blockingnew calls increasesdueto theforbiddingof many ISLs, beingalmost15%for 3-minutecalls for thesametargetprobability,over1600calls.

TheFootprintHand-OverReroutingProtocol. A reroutingstrategy, calledfootprinthand-over rerouting – FHR–, was proposedin [32, 34], which usesinformationsaboutthenetwork predictabilityto replaceacurrentrouteby anew one,basedonthesuccessorsatellitesin theoriginal route,asfollows.

Let ��4�� P >����)>TSUSTSZ>���� bea routeconnectingsatellite � P to satellite ��� . In casebothroute-endsundergoalink hand-over,thenafootprintreroutingcantakeplaceand� isreroutedthrough�Q��4����P >����� >TSUSTSU>��O�� , where���v denotesthesuccessorof satellite��v in thesameorbit. In caseonly theroute-end� P (respectively, ��� ) hasto hand-overthecommunication,but � �� (respectively, � �P ) isnotyetvisibletotheground-user, thentheoriginal route � is simply augmentedto �Q�e4����P >���P*>�� � >USUSTSU>�� � (respectively,�Q�Y4���PR>�� � >TSUSUSU>�� � >����� ), until �O�� (respectively, ���P ) becomesvisible to theground-user. When the secondroute-endundergoesits link hand-over, then a footprintreroutingis usedandthenew routewill be �Q�Y4��O�P >��O�� >USTSUST>����� .

Evidently, if � is aminimum-hoproute,thenFHRimplementsa reroutingwhere�Q� is alsoaminimum-hoproute.Further, in thecasewherethelink-costis afunctionof traffic load, andthis latter is time-homogeneous,then � beinga minimum-costrouteimpliesthat � � is alsoa minimum-costroute.

The performanceof FHR has beenevaluatedthroughsimulation, particularlyagainsta pure augmentationapproach,wherea call is droppedif an augmentedlink cannotbe found betweenthe hand-over satelliteand the currentroute. Newcalls first routesarefound throughthe implementationof Dijkstra algorithm. Twocost functionsarestudied,namelyminimum hops,andcongestion.In the caseof

xviii DESIGN, ROUTING AND HAND-OVER IN SATELLITE NETWORKS

homogeneoustraffic with high call arrival rate, FHR blocks more new calls thanpure augmentation,but drops much less calls becauseof hand-overs, than pureaugmentation.This shows theinterestof takingthedeterministicnetwork topologyinto account.

1.4.2.5 ConstellationsViewed as Dynamic Networks Another idea consistsinmanagingmore specifically the mobility of the constellation. The more naturalstrategy consistsin taking shortest-pathalgorithms,which minimize the resourcestakenby anindividualrouteusingacostbasedonthelinks. Eachlink receivesacost(proportionalto its expectedload),andtherouteminimizesthesumof thecostsofthelinks it uses.

An elementarymodel is inspired from ad-hocnetworks [26]. The two mainstrategiesare:

/ Proactive: any changeof topology is immediatelynotified to all the livingnodes;

/ Reactive:arequestis likely to initiateafloodthatwill discovertheactualstateof thenetwork.

The choicebetweenthe two methodsdependson the dominatingactivity of thenetwork. Proactive protocolsperform betterwhen the traffic throughputis highand the topology changesseldomly, whereasthe reactive protocolsperform wellwhen the network topology often changesand the traffic is low. An exampleofproactive protocol is the ExtendedBellman-Ford (EXBF), proposedfor satelliteconstellationsin [25], whereroutingtablesaredynamicallyupdatedusingashortestpathspolicy. A reactive protocol,namedDarting,wasalsoproposedin [25]. Thealgorithmbroadcastsnetwork informationonly if it becomesabsolutelynecessary.In themeantime,thealgorithmtakesadvantageof datapacketsto updatethenetworktopology. Experimentshave shown that theDartingalgorithmrequiredasmuchas72%percentmoreoverheadwhencomparedto theEXBF algorithmin an Iridium-like, lightly loadednetwork [27].

An immediatesimplificationoccurswhenoneconsidersthat the weightsof thelinks areuniform, andthe topologyis regular. In the meshor the torus/meshcase,oneshortestpathin termsof numberof hopsis obtainedby a XY routing(that is arouteconsistingof aseriesof inter-orbital links first, andthenaseriesof intra-orbitallinks). Whenoneconsiderstheminimumdelay, insteadof theminimumnumberofhops,it is clearthatthecostof inter-orbital links aresmallerwhenthey areclosertothe poles(while the intra-orbitallink costremainsconstant).Then,somegeodesicconsiderationsmay give the advantageto routesthat take morepolar inter-orbitallinks, asexplainedin [10].

1.4.2.6 ReservationStrategies A secondsteptowardsreducingthecall blockingprobability is thechoiceof the link reservationstrategy. Indeedall thesealgorithmsattemptonly to minimizethe load of the links, without implementingcongestioncontrol mechanisms,which would give someguaranteethat a communicationwillnot bedroppedwhenoneof thelinks it usesexperiencesanhand-over.

CONCLUSIONS xix

Noticethatmostof thehand-over techniqueswe described,suchastheHQ, HG,GCAC,andGH,couldbeappliedtoenhancetherouting. A basicreservationstrategywasproposedin [31]. It essentiallyconsistsin acceptinga call if andonly if it ispossibleto reserveISLsin suchawaythatthecommunicationcanbemaintainedafterany onelink hand-overevent (this canbeseenasanextensionof theGH concept).Herethedeterministicbehaviour of thenetwork playsa key role: It is assumedthat,for eachlink, wecandetermineits next hand-overandtheoverallnetwork topology.So,if duringthecall onlyoneof thelinks in thepathhasahand-over, it is guaranteedthatsuchcall will not bedropped.

Anotherrelatedstrategy is theslack reservationpolicy, which consistsin alwaysacceptingthenew call and,if possible,alsoreservingthelinks for a“next hand-over”event. Intuitively, we are relaxing the basicreservation strategy, sincethosecallswhosedurationis sufficiently shortwill terminatebefore the link hand-over eventoccurs. However, the experimentalresultsin [31] show that the slack reservationstrategy performsworsethan the basicone. This is mainly due to the fact that,in the slack reservation, a certainnumberof calls are acceptedwith no reservedroute.Clearly, suchcallsarevulnerableto link hand-overs.On theotherhand,callswith reservedroutes“consume”moreresourcesthanthey wouldwithoutreservation.Hence,with lesssystemresourcesavailable,the“vulnerable”callsaremorelikely tobedropped.

Goingastepahead,analgorithmis presentedin [12] thatforwardsthereservationonceit is used.It is shown thatthereservationpermanentlyguaranteesthecommuni-cationprovidedthatonly Southhand-oversoccur. All kindsof hand-oversaretakeninto accountin [14], wherethe call admissioncontrol problemfor regular satellitenetworksis turnedinto aproblemof max-loadof afamily of rectanglesin atwo/threedimensionalspace.Communicationrequestsarerepresentedasa seriesof rectan-gles.Requiredcapacityis thenequivalentto themaximumnumberof rectanglesthatintersectonagivenpoint. Therefore,capacitycontrolfor call admissionpoliciescanbedonethroughgeometricalgorithms.

1.5 CONCLUSIONS

In thischapterwereviewedtheliteratureconcerningroutingandhand-overtechniquesfor LEO constellations. We showed that they are particularly complex in suchacontext becauseof both physical constraintsand the movementof the satellites.Notwithstanding,severalinterestingresultsexist and,at thetimeof thiswriting, twosatelliteconstellationsareoperational(Iridium andGlobalstar).Thesefactscertainlyspeakfor moreresearchto beundertakenin thenearfuture.

1.6 ACKNOWLEDGMENTS

This work waspartly supportedby theEuropeanRTN ARACNE, theINRIA/CNPqandtheRNRT/Constellationprojects.

References

1. E. Altman,A. Ferreira,andJ.Galtier. Lesréseauxsatellitairesdetélécommuni-cation. Dunod,1999.

2. S.Appleby. Multifractal characterizationof thedistributionpatternof thehumanpopulation.Geographicanalysis, 28(2):147–160,1996.

3. F. Baccelli, M. Klein, M. Lebourges,and S. Zuyev. Géométriealéatoireetarchitecturederéseaux.Ann.Télécommun., 51:158–179,1996.

4. P. Bergé, A. Ferreira,J. Galtier, and J.-N . Petit. A ProbabilisticStudy ofInter-SatelliteLinks Lo adin PolarOrbit SatelliteConstellations.InternationalJournalon TelecommunicationSystems, To appear.

5. H.S.Chang,B.W. Kim, C.G.Lee,Y. Choi,H.S.Yang,andC.S.Kim. Topologicaldesignandroutingfor low-earthorbit satellitenetworks.In Proc.of IEEEGlobalTelecommunicationsConference(GLOBECOM), pages529–535,1995.

6. H.S. Chang,B.W. Kim, C.G. Lee, S.L. Min, Y.C., H.S. Yang,D.N. Kim, andC.S. Kim. Performancecomparisonof optimal routing and dynamicroutingin low-earthorbit satellitenetworks. In Proc. of IEEE Vehicular TechnologyConference(VTC), pages1240–1243,1996.

7. S. Cho, I.F. Akyildiz, M.D. Bender, andH. Uzunalio�glu. A New ConnectionAdmissionControlfor SpotbeamHandoverin LEO SatelliteNetworks. In Proc.of IEEEGlobalTelecommunicationsConference(GLOBECOM),SanFrancisco,USA,, pages1156–1160,2000.

8. D. R. CoxandV. Isham.Point Processes. ChapmanandHall, 1980.

9. F.J.Dietrich. Theglobalstarsatellitecellularcommunicationsystemdesignandstatus. In Proc. of the 5th International Mobile SatelliteConference, pages139–144,Pasadena,CA, June1997.

10. E. Ekici, I. F. Akyildiz, andM. Bender. DatagramRoutingAlgorithm for LEOSatelliteNetworks. In Proc.of IEEE INFOCOM, pages500–508,2000.

11. A. Ferreira,J.Galtier, P. Mahey, G. Mat eus,andA. Oliveira. An OptimizationModel for Routingin Low EarthOr bit SatelliteConstellations.In Proceedingsof ISPAN, Dallas,USA, December2000.IEEE CSPress.

xxi

xxii REFERENCES

12. A. Ferreira,J. Galtier, J.-N. Petit, and H. Rivano. Re-routingalgorithmsina meshedsatelliteconstellation. AnnalesdesTélécommunications, 2001. Toappear.

13. A. Frey and V. Schmidt. Marked point processesin the plane I - a surveywith applicationsto spatialmodelingof communicationnetworks. AdvancesinPerformanceAnalysis, 1(1):65–110,1998.

14. J.Galtier. Geographicalreservationfor guaranteedhandoverandroutingin lowearthorbit constellations.TelecommunicationSystems, 2001.To appear.

15. B. GavishandJ.Kalvenes.Theimpactof intersatellitecommunicationlinks onLEOSperformance.TelecommunicationSystems, 8:159–190,1997.

16. N. Gerlich. On the spatialmultiplexing gainof SDMA for wirelesslocal loopaccess.TechnicalReport161,UniversitätWürzburg, January1997.

17. P. Gvozdjak. Modelingcommunicationsin Low-Earth-Orbitsatellitenetworks.PhDthesis,Schoolof ComputingScience,SimonFraserUniversity, 2000.

18. M. HellebrandtandR.Mathar. Cumulatedinterferencepowerandbit-error-ratesin mobilepacket radio. WirelessNetworks, (3):169–172,1997.

19. D. Hong andS. Rappaport.Traffic modelandperformanceanalysisfor cellu-lar mobile radio telephonesystemswith prioritizedandnonprioritizedhandoffprocedures.IEEE Transactionson VehicularTechnology, 36(3):77–92,August1986.

20. Y.C. Hubbel and L.M. Sanders. A comparisonof the IRIDIUM and AMPSsystems.IEEENetwork, 12(2):52–59,March/April 1997.

21. D.M. Kohn. Providing global broadbandinternetaccessusinglow-earth-orbitsatellites.ComputernetworksandISDNsystems, 29:1763–1768,1997.

22. R.J.Leopold.Low-earthorbit globalcellularcommunicationsnetwork. In Proc.of IEEEInternationalConferenceonCommunications(ICC), pages1108–1111,1991.

23. J. A. Musey, W. Kidd, and P. Fuhrmann. The satellite report.http://www.unterberg.com.

24. E. Papapetrou,I. Gragopoulos,and F.-N. Pavlidou. Performanceevaluationof LEO satelliteconstellationswith inter-satellitelinks underself-similarandPoissontraffic. InternationalJournal of SatelliteCommunications, 17:51–64,1999.

25. S.R.Pratt,C.E. FossaJr. R.A. Raines,andM.A. Temple. An operationalandperformanceoverview of theIRIDIUM Low EarthOrbit satellitesystem.IEEECommunicationsSurveys, pages2–10,secondquarter1999.

REFERENCES xxiii

26. A. Qayyum. Analysisand evaluationof channelaccessschemesand routingprotocols in wirelessLANs. PhD thesis,University of Paris-Sud,November2000.

27. R. Raineset al. Simulationof routing protocolsoperatingin a low earthorbitsatellitecommunicationnetwork environment.In 1997IEEEMilitary Commun.Conf., volume1, pages429–433,Monterey, CA, November1997.

28. E.D. Re,R.Fantacci,andG.Giambene.Efficientdynamicallocationtechniqueswith handoverqueuingfor mobilesatellitenetworks. IEEEJournalon SelectedAreasin Communications, 13(2):397–405,February1995.

29. J. Restrepoand G. Maral. Guaranteedhandover (GH) servicein a non-Geoconstellationwith “satellite-fixedcell” (SFC)systems.In Proc.5thInternationalMobile SatelliteConference(IMSC), pages19–24,Pasadena,CA, USA, June1997.

30. M.A. Sturza. Architectureof the Teledesicsatellitesystem. In Proc. of theInternationalMobileSatelliteConference(IMSC), pages214–218,1995.

31. P.T.S.Tam,J.C.S.Lui, H.W. Chan,Cliff C.N.Sze,andC.N.Sze.An OptimizedRouting Schemeand a ChannelReservation Strategy for a Low Earth OrbitSatelliteSystem.In Proc.of IEEEVehicularTechnologyConference(VTC1999- Fall). IEEE VTS50th,Volume5, pages2870–2874,1999.

32. H. Uzunalio�glu, I.F. Akyildiz, Y. Yesha,andW. Yen.Footprinthandoverrerout-ing protocol for low Earthorbit satellitenetworks. ACM Journal on WirelessNetworks, 5(5):327–337,October1999.

33. H. Uzunalio�glu, M.D. Bender, andI.F. Akyildiz. A RoutingAlgorithm for LowEarthOrbit (LEO) SatelliteNetworkswith DynamicConnectivity. ACM-BaltzerJournalof WirelessNetworks(WINET), 6(3):181–190,June2000.

34. H. Uzunalio�glu, W.Yen, and I.F. Akyildiz. A connectionhandover protocolfor LEO satelliteATM networks. In Proc. of the3rd ACM/IEEEInternationalConferenceon Mobile Computingand Networking(MOBICOM), pages204–214,Budapest,Hungary, September1997.

35. M. D. Violet. Thedevelopmentandapplicationof acostperminutemetricfor theevaluationof mobilesatellitesystemsin a limited-growth voicecommunicationmarket.Master’sthesis,MassachusettsInstituteof Technology,Cambridge,MA,USA, 1995.

36. M. WernerandG.Maral.Traffic flowsanddynamicroutingin LEO intersatellitelink networks. In Proc. of the 5th International Mobile SatelliteConference(IMSC), pages283–288,1997.

37. M. WernerandP. Révillon. Optimizationissuesin capacitydimensioningofLEO intersatellitelinks networks. In Proc.of ECSC5, November1999.

xxiv REFERENCES

38. M. Werner, F. Wauquiez,J.Frings,andG.Maral. Capacitydimensioningof ISLnetworks in broadbandLEO satellitesystems.In Proc. of the6th InternationalMobileSatelliteConference(IMSC), pages334–341,Ottawa,1999.

39. W.W. Wu, E.F. Miller, W.L. Pritchard,and R.L. Pickholtz. Mobile SatelliteCommunications.Proceedingsof theIEEE, 82(9):1431–1448,Sept1994.