Similar documents
Munsell Soil Color. Munsell Gradient No. Code

Tabla de conversión Pantone a NCS (Natural Color System)

Color Code Drug Doses L.A. County Kids

Green = 0,255,0 (Target Color for E.L. Gray Construction) CIELAB RGB Simulation Result for E.L. Gray Match (43,215,35) Equal Luminance Gray for Green

Perception of Light and Color

Working Paper Series Department of Economics Alfred Lerner College of Business & Economics University of Delaware

User Guide for Patients

PROVIDER APPLICATION FOR MEDICAL LIEN PORTFOLIO PURCHASE

Guide for Performing a Wireless Site Survey. 2.4 GHz IEEE g/802.11b/

PANTONE Solid to Process

Pantone Matching System Color Chart PMS Colors Used For Printing

CGS X X X X X

How To Color Print

Download Google Drive to windows 7

1. Let A, B and C are three events such that P(A) = 0.45, P(B) = 0.30, P(C) = 0.35,

Notes on Probability. Peter J. Cameron

Betting Duty (Amendment) Ordinance Contents

A Comprehensive Set of Image Quality Metrics

VARIABLE SPEED CONTROL CPMD CONNECTIONS

Analysis of Patient Monitor Alarms in Adult Intensive Care Units --- University of California, San Francisco April 25, 2013 Patricia Harris, RN, PhD

First Step: Have the patient describe the problem. 0 - Patient is not sure if the lights are behaving correctly:

General Product Line Quick Quote U.S. Suggested List Price

Department of Industrial Engineering IE 202: Engineering Statistics Example Questions Spring 2012

PANTONE Chart Builder File: MPC2000_2500_3000 Page: 1 of 14

Setting your session preferences

Animated example of Mr Coscia s trading

PANTONE ColorVANTAGE Process Simulations of PANTONE solid colors Page: 1 of 14

Health Science Center AirWatch Installation and Enrollment Instructions For Apple ios 8 Devices

This feature is available on select devices featuring VUDU. In order to use this feature, your VUDU device must be connected to the Internet.

Organic Pigments. Azo Pigments

Dolch Pre-Primer Sight Vocabulary Bingo. Noah s Ark Bingo. 40 Pre-Primer Dolch Sight Words. Ordered by frequency

Garment Manufacturing Through Lean Initiative-An Empirical Study On WIP Fluctuation In T-Shirt Production Unit

Effects of changing slope or y-intercept

Employee Manual Development Tool Version 7.0. User Guide

Enjoy Music over Your Wi-Fi Network

B I N G O B I N G O. Hf Cd Na Nb Lr. I Fl Fr Mo Si. Ho Bi Ce Eu Ac. Md Co P Pa Tc. Uut Rh K N. Sb At Md H. Bh Cm H Bi Es. Mo Uus Lu P F.

Section 11.4: Equations of Lines and Planes

Software Development Process

References Control and signalling units Ø 22 0

Section 6-5 Sample Spaces and Probability

Accessing vlabs using the VMware Horizon View Client for OSX

E. OZCEYLAN /International Journal of Lean Thinking Volume 1, Issue 1 (June 2010)

To add fractions we rewrite the fractions with a common denominator then add the numerators. = +

COLOR THEORY WORKSHEET

System Requirements for Dynamics GP 2013

13s Windy City Power League Schedule

WebEx Sharing Resources

AMAS Mine/ERW Clearance Marking System. Third Edition April 2011

Mekong River Cruises. Indochina Sails BRAND GUIDELINES. (*) This brand guidelines is temporary and can be changed by request

TWS SERIES TWS PLASTIC ENCLOSURE WITH SILICONE PROTECTOR FEATURE. Operating temperature -10 C~+60 C. Components. Accessories (Optional parts)

For example some Bookkeepers are using Dropbox to share the accounting files between them and their client.

HR DEPARTMENTAL SUFFIX & ORGANIZATION CODES

On Frequency Assignment in Cellular Networks

Google Sites: Site Creation and Home Page Design

User s Manual of BTGP-38KM Bluetooth GPS Data Logger V1.0

Joomla Article Advanced Topics: Table Layouts

Simplify your palette

ONLINE LEVEL ONE UMPIRES COURSE PARTICIPANT FACT SHEET

SOMERS POINT BRAND IDENTITY STYLE GUIDE

Lithichrome Stone Paint- LT Blue Gallon Lithichrome Stone Paint- Blue 2 oz Lithichrome Stone Paint- Blue 6 oz

DNS-327L Compatible List

Panda GateDefender Virtual eseries QUICK GUIDE

System Requirements for Microsoft Dynamics GP 2013

GE Lightech LED Driver Programming Tool

Week 2: Conditional Probability and Bayes formula

GOLD COAST CITY COUNCIL

More Quadratic Equations

48,00 26,50 18,00 59,50 59,50 54,50. PENNE white 20 x 50. m 2. WHITE glossy 20 x 50. PENNE inserto flower 20 x 50. szt. PENNE border flower 8 x 50

WifiEagle Channel Analyzer

Industrial Automation course

APRSstat An APRS Network Analysis Tool

Probability. Sample space: all the possible outcomes of a probability experiment, i.e., the population of outcomes

COMOS FEED Knowledge Base

Using Your Polyvision Digital Whiteboard and Walk-and-Talk

Kingston Technology DataTraveler Customization Program

THE BASICS OF COLOUR THEORY

Mobile Eyewear Recorder

' LapLink DB25 Male/Male ' Serial LapLink DB9F/DB25F (4 Connectors)

Telekom Communication Center User manual. version 4.2

Trailer Bodywork. Plywood Steel & Aluminium Sheet Safety & Security Products

Why use ColorGauge Micro Analyzer with the Micro and Nano Targets?

Graphing Equations. with Color Activity

The Foundation User Manual LED Battery Slim Par Fixture (RGBA, DMX 7Ch/4CH)

NTI INCORPORATED. 13W3M-15HDF-ID. VGA Monitor-to-SUN Video Port Adapter Installation Manual. Man078 Rev Date October 28, 2004

Industrial Automation course

Data Storage 3.1. Foundations of Computer Science Cengage Learning

Review for Test 2. Chapters 4, 5 and 6

Activities/ Resources for Unit V: Proportions, Ratios, Probability, Mean and Median

Transcription:

StaticFrequencyAssignmentinCellular LataNarayanan Networks August,998 SunilM.Shendey tice.inthestaticfrequencyassignmentproblem,eachvertexofthegraphisa Acellularnetworkisgenerallymodeledasasubgraphofthetriangularlat- Abstract distinctfrequenciespercall.theedgesofthegraphmodelinterferenceconstraintsforfrequenciesassignedtoneighboringstations.thestaticfrequency representsthenumberofcallsthatmustbeservedatthevertexbyassigning basestationinthenetwork,andhasassociatedwithitanintegerweightthat anyouterplanargraph.fortheproblemofmulticoloringanarbitraryconnected describeanecientalgorithmtooptimallymulticoloranyweightedevenorodd lengthcyclerepresentingacellularnetwork.thisresultisfurtherextendedto assignmentproblemcanbeabstractedasagraphmulticoloringproblem.we whichguaranteesthatnomorethan4=3timestheminimumnumberofrequired subgraphofthetriangularlattice,wedemonstrateanapproximationalgorithm weightsatasmallneighborhood. adistributedmanner,whereeachstationneedstohaveknowledgeonlyofthe colorsareused.further,weshowthatthisalgorithmcanbeimplementedin DepartmentofComputerScience,ConcordiaUniversity,Montreal,Quebec,Canada,H3GM8. rithms,graphmulticoloring,distributedalgorithms. Keywords:Frequencyassignment,cellularnetworks,approximationalgo- email:lata@cs.concordia.ca,fax(54)848-2830.researchsupportedbynserc,canada. shende@crab.rutgers.edu,fax(609)225-6624.thisresearchwasconductedatconcordiauniversityduringthesummerof996,withpartialsupportfromnserc,canada. ydepartmentofcomputerscience,rutgersuniversity,camden,nj0802,usa.email:

Cellulardataandcommunicationnetworkscanbemodeledasgraphswitheachver- texrepresentingabasestation(sometimescalledacell)inthenetwork.cellscan Introduction anygiventime,acertainnumberofactiveconnections(orcallsincellularnetwork terminology)areservicedbytheirnearestbasestation.thisserviceconsistsmainly communicatewiththeirneighborsinthegraphviadirectionalradiotransceivers.at interferencebetweentwodistinctcallsinthenetwork.however,cellularnetworksuse ofassigningafrequencytoeachclientcallinamannerthatminimizesoravoidsradio staticfrequencyassignmentproblem,therefore,consistsofdesigninganinterferencefreefrequencyallocationprotocolforanetworkwherethenumberofcallspercelliitedavailablebandwidthiscriticaltotheviabilityandeciencyofthenetwork.the axedspectrumofradiofrequenciesandtheecientsharedutilizationofthelim- nitetriangulargridembeddedintheplane.verticesrepresentingcellsareplaced knownapriori.thisformsthemotivationfortheproblemsstudiedinthispaper. attheapexesofsimilartriangles,andeachvertexhasatmostsixotherneighbors Inparticular,cellularnetworksareusuallymodeledasniteportionsoftheinsurroundingitinthegrid.Thereasonforadoptingthisparticulargeometrystems fromthefactthatcellsareuniformlydistributedinthegeographicareaofthenetwork,andanindividualcellgenerallyhassixdirectionaltransceivers.hence,the asaregularhexagon.thetriangulartilingrepresentingthenetworkissimplythe planardualoftheresultingvoronoidiagram.weshallrefertotheresultinggraphs Voronoiregionaroundacell(orequivalently,thatcell'scallingarea)canbeidealized ashexagongraphs. hasanassociatedintegerweight,w(v)0.aw-coloring(ormulticoloring)ofg abstractedasfollows.letg=(v;e)denoteanhexagongraph.eachvertexv2v Thefrequencyassignmentproblemincorporatinginterferenceconstraintscanbe w(v)distinctcolorswherebyforeveryedge(u;v)2e,thesetofcolorsassignedto theendpointsuandvaredisjoint.inparticular,weareinterestedinaminimum isanassignmentofsetsofcolorstotheverticessuchthateachvertexvisassigned multicoloringoraw-coloringofgthatusestheleastnumberofcolors. thecellularnetworkliterature.however,inpractice,cellularsystemstendtobemorecomplicated functions;seeforexampleborndorferet.al.[].nevertheless,thehexagongraphmodelcontinues andtherehavebeenrecentstudiesthatattempttomodelmoregeneralinterferencepatternsandcost Wenoteherethatthisisthemostcommonlyusedmodelforfrequencyassignmentproblemsin tobeofsignicancefromthehistoricalstandpointandasanabstractionthatissucientlycloseto realitytoprovideusefulinsights. 2

distinctfrequencyanditisassumedthattwocallsmayusethesamefrequencyifand ausefulabstractionoftheessentialinterferenceconstraints:eachcolorrepresentsa Inthecontextoffrequencyassignment,amulticoloringasdenedabove,provides oftheradiospectrum,andfrequencyreuseiscontrolledbyasequenceofnon-negative inpractice,theavailablecellularfrequencyspectrumisacontiguouslinearsubinterval onlyiftheyoriginateindistinctcellsthatarenotneighbors.itshouldbenotedthat incellsthatareadistanceiapartintheunderlyinggraphmustbeassignedfrequencies integers,c0c:::,withc0,calleddistancereuseconstraints.twodistinctcalls thesimplestconstraints,viz.whenc0=c=andci=0,i2.underthis andversionsofthefrequencyassignmentproblem.weformulateourproblemunder thatdierbyciinthefrequencyspectrum.hale[3]discussesmanygeneralizations formulation,theproblemreducestobeingabletocomputeaminimummulticoloring a(vertex)weightedgraphthatisanite,inducedsubgraphoftheinnitetriangular toagivenhexagongraph. grid.thus,thegraphisplanar,andeveryvertexv2vhasdegreeatmostsix Inthesequel,weassumethatG=(V;E;w)denotesahexagongraph,i.e.itis andanassociatedintegerweight,w(v)0.theweightedchromaticnumberofg, ofdetermining(g)isnon-trivial.infact,ithasbeenestablishedonlyrecentlythat forgraphswitharegularstructuresuchasthoseconsideredinthepaper,theproblem denoted(g),istheminimumnumberofcolorsrequiredinaw-coloringofg.even interesttostudyapproximationalgorithmsfortheproblem. apolynomialtimealgorithmforcomputing(g)canbedevised.naturally,itisof thecorrespondingdecisionproblemisnp-complete[7],andhenceitisunlikelythat atverticesinanymaximalcliqueinthegraphisatriviallowerboundon(g).note atanysetofmutuallyadjacentvertices.thusthemaximumoverthesumofweights Itiseasytoseethat(G)mustbegreaterthanthetotalnumberofcolorsrequired upperbounds,whilethereisavastliteratureonalgorithmsforfrequencyassignment thatforhexagongraphs,edgesandtrianglesaremaximalcliques.inthedirectionof arenoprovenboundsontheperformanceoftheproposedalgorithms,intermsofthe numberofcolorsusedinrelationtotheweightedchromaticnumber[2,5,6,8,9].we ongraphs(especiallyhexagongraphs)thatclaimtousefewcolors,generallythere noteheretwoexceptions.awell-knownalgorithm,sometimesreferredtoasfixed usescolorsfromtherstset,andavertexthatbasecolor2or3usescolorsfromthe usesthreexedsetsofcolors,oneforeachbasecolor.avertexthathasbasecolor Allocation,usesthefactthattheunderlyinggraphcanbe3-colored.Thealgorithm manyas3timesthenumberofrequiredcolors.janssenet.al.[4]proposeadierent secondorthirdsetsrespectively.itiseasytoshowthatthisalgorithmcoulduseas 3

3=2timestheminimumnumberofcolorsrequired. algorithmcalledfixedpreferenceallocationthatisguaranteedtousenomorethan graphs.insection4,weaddressthequestionofmulticoloringanarbitraryhexagon Section3,wepresentoptimalalgorithmsformulticoloringcyclesandouterplanar Inthenextsection,weformallydenesomebasicterminologyandproblems.In graph.ourmainresultisanecientapproximationalgorithmwithaperformance offuturedirectionsinsection6. guaranteeofwithin4=3oftheoptimal.finally,insection5,weshowhowtoimplementtheabovealgorithminadistributedmanner2.weconcludewithadiscussion denedontheverticesofthegraph,wherew(v)representsthenumberofcallsto 2LetG=(V;E;w)beahexagongraphwithanon-negativeintegerweightvectorw Preliminaries withverticesandedgescontainedintheinnitetriangularlattice(tessellation)ofthe beservedatvertexv.weassumehereafterthatghasaxedplanarembedding plane.thusanyvertexvcanbeconnectedtoatmost6neighbors,andforaxed edgeincidentonv,anyotheredgeincidentonvisatanangleof=3;2=3;;4=3or 5=3fromthatedge.Sincethetriangularlatticeis3-colorableintheordinarysense (i.e.wheneachvertexhasunitweight),theunderlyinggraphcorrespondingtounit weightsatverticesofgisalso3-colorable. thepalettecsuchthat C(thecolorpalette)andafunctionfthatassignstoeachv2Vasubsetf(v)of Aw-coloringormulticoloringofthegraphG=(V;E;w)consistsofasetofcolors 8(u;v)2E;f(u)\f(v)=:twoneighboringverticesgetdisjointsetsofcolors. 8v;jf(v)j=w(v):eachvertexgetsw(v)distinctcolors,and maticnumberofg,denoted(g),isthesmallestnumbermsuchthatthereexists amulticoloringofgofspanm.thusgivenahexagongraphg,ourobjectiveisto ThespanofamulticoloringisthecardinalityofthesetC.Theweightedchro- ndamulticoloringforgwhosespanisascloseto(g)aspossible. inadistributedsetting. dentlyderived,andunlikethemcdiarmid-reedalgorithm,hastheadvantageofbeingimplementable havealsodescribedadierent4=3-approximatealgorithmfortheproblem.ourresultwasindepen- 2WenotethatinadditiontoshowingtheNP-hardnessofthisproblem,McDiarmidandReed[7] 4

thattheweightofanymaximalcliqueofgisalowerboundon(g).letd[2] D[3] anedge(triangle)ingtobethesumoftheweightsofitsendpoints(apexes).note TheonlymaximalcliquesinGbeingedgesandtriangles,wedenetheweightof denedg=maxfd[2] GdenotetherespectivemaximaovertheweightsofedgesandtrianglesinG,and G;D[3] Gg.Then,ifthereexistsamulticoloringofGwithspan, Gand itfollowsthat: besuitablyorderedorpartitioned;inparticular,wewilloftenassumethatvertices Wewillassumewithoutlossofgeneralitythatanypaletteofavailablecolorscan (G)DG: areassignedcolorsfromthecircularlyorderedinterval[;m]=f;2;:::;mg,where itmeansthatthevertexiscoloredwiththesetfi;i+;:::;jginacyclicmanner Forinstance,whenavertexisassignedthesubintervalofcolors[i;j]fromthepalette, Misapositiveintegerthatdependsontheparticulargraphunderconsideration. wherecolorisassumedtofollowthecolorm. ConsiderahexagongraphG=(V;E;w)withnverticesintheformofasimple 3cycle,labeledu;u2;:::;uninclockwiseorder.Forsimplicity,letwi;in, OptimalMulticoloringofCycles onwhethern,thenumberofverticesonthecycle,isevenorodd. optimallycoloredwithexactly(g)colors.therearetwocasestoconsiderdepending denotetheweight,w(ui),ofvertexui.weshowthatanysuchhexagongraphcanbe maximalcliquesofgbeingedges,dg=d[2] Supposethatn=2m,i.e.thegraphconsistsofanevenlengthcycle.Thenall colorpalette[;dg].theideaistoassignforim,thecolors[;w2i ]tothe cycle.weobservethataverysimplegreedystrategysucestomulticolorgwiththe Gisthemaximumweightofanedgeinthe vertexu2i.notingthatforim, odd-numberedvertexu2i andthecolors[dg w2i+;dg]totheeven-numbered withsubscriptsinterpretedcyclically,itfollowsthatthegivenmulticoloringisproper. DGw2i +w2i; DGw2i+w2i+ and optimalmulticoloringofg. Byconstruction,(G)=DGandthesimpleparity-basedalgorithmthusprovidesan [8,4],butitwasonlyappliedtonetworksconsistingofsimplepaths(itiseasyto Wenotethataverysimilarideawasalreadyusedinthecellularnetworkliterature 5

unweightedodd-lengthcycleneedsatleastthreecolorsinanyordinarycoloring.for parityargumentfailstomulticolorodd-lengthcycles,preciselybecausetheunderlying seethatthisstrategyworksingeneralforanybipartitegraph).unfortunately,the Denition3.LetG=(V;E;w)beanodd-lengthsimplecycle,withverticeslabeled bemulticoloredwithdg=4colors,butneeds5colorsinstead. instance,ifgisa9-cyclewithweight2oneachvertex,itiseasytoseethatgcannot u,u2,:::,u2m+,m,inclockwiseorder.wedene D0G=maxfD[2] G;dP2m+ i=wi m eg Theorem3.2LetG=(V;E;w)beacycleofoddlengthn=2m+3.Then Proof.ItisclearthatD[2] (G)=D0GandGcanbeoptimallymulticoloredwithexactlyD0Gcolors.Further,the multicoloringcanbeobtainedintimeo(n). isatmostm,anysinglecolorcanbeusedonlyatmverticesorfewerinthecycle. andhenced0g,isalowerboundon(g).sincethesizeofanindependentseting Gisalowerboundon(G);weestablishthatd2m+ i=wi Thetotalnumberofcolorsneededatallverticesbeing2m+ m e, (G)d2m+ NextweshowthatGcanbecoloredwithD0Gcolorsusingalineartimealgorithm; i=wi m e.thus,d0gisindeedalowerboundon(g). i=wi,weconcludethat thiscompletestheproofofthetheorem.werstobservethattheremustbeasmallest indexk,km,whichsatisestheinequality 2k+ iswell-denedandcanbefoundeasilyinlineartime. NotethatthispropertyholdstruefortheindexmfromDenition3.,andhencek i=wikd0g:.verticesuthroughu2kareassignedcontiguouscolorsinacyclicmannerfrom Theverticesofthecyclearenowcoloredasfollows: thepalette[;d0g].specically,forj2k,vertexujisassignedthecolors cyclically.byconstruction,thisensuresthatthepathu;u2;:::;u2kisproperly [(+j Xi=wi);jXi=wi]; multicoloredsinced0gd[2] G. 6

2.Verticesu2k+throughu2m+arecoloredbasedontheirparity(asintheevencyclealgorithm).Inparticular,for2k+i2m+,thevertexuiisassigned thecolors[;wi]ifiiseven,orthecolors[d0g wi+;d0g]ifiisodd.again, thisensuresthatthepathu2k+;u2k+2;:::;u2m+isproperlymulticolored. theedge(u2k;u2k+)isproperlymulticolored:thisisaconsequenceoftheminimality ofk,forweknowthat2k theedge(u;u2m+)isalsoproperlymulticolored.allthatremainsistoverifythat Sincevertexuhasthecolors[;w]andvertexu2m+thecolors[D0G w2m++;d0g], amongthecolors[d0g w2k++;d0g]assignedtovertexu2k+. Weillustratethelabelingschemeusinga9-cyclewithweight2oneachvertexas i=wi>(k )D0G.Hence,nocolorassignedtou2kcanbe anexample.asd0g=maxf4;5g=5,weusethepalette[:::5].since5i=wi= assigncolorsasinabipartitegraph,fromthetwoendsoftheinterval[;5].finally, alwaystakingthenextfouravailablecolorsinthepalette.forthelastvevertices,we 02D0Gbut3i=wi=6>D0G,wecolortherstfourverticesinacyclicmanner, wenotethatouralgorithmcanactuallymulticoloranycycle,andnotjustcyclesthat arehexagongraphs(i.e.embeddedinthetriangularlattice). everyvertexofgliesontheboundaryoftheexteriorface.itisstraightforwardto graph.agraphissaidtobeouterplanarifitcanbeembeddedintheplanesothat Theorem3.2canbeusedtoderiveanoptimalmulticoloringofanyouterplanar seethattheweightedchromaticnumberofanygraphisthemaximumtakenoverthe chords.abiconnectedouterplanargraphgwithoutchordsisasimplecycleandcan siderabiconnectedouterplanargraph:anysuchgraphisacyclewithnon-intersecting weightedchromaticnumbersofitsbiconnectedcomponents.thusitsucestocon- bemulticoloredoptimallyusingtheconstructionintheproofoftheorem3.2.otherwise,let(u;v)beachordandletgandg2bethetwopartsofgonthesidesof thecolorassignmentofg2sothatthecolorsassignedtouandving2agreewith thischord,eachoneincludingtheedge(u;v).recursivelycolorgandg2.relabel coloredoptimallyusing(g)colors. Corollary3.3LetG=(V;E;w)beanarbitraryouterplanargraph.ThenGcanbe thoseassigneding.thefollowingcorollaryisimmediate: multicoloring.weremarkthatcorollary3.3appliestoanyouterplanargraph,and notjustouterplanarhexagongraphs. Acarefulimplementationofthealgorithmsketchedabove,resultsinalineartime 7

4Inthissection,weconsidertheproblemofcomputinganapproximatemulticoloring ofanarbitraryhexagongraph.sinceahexagongraphmaycontainanoddcycleasan Approximatemulticoloringofhexagongraphs inducedsubgraph,itfollowsasaconsequenceoftheorem3.2thatd[3] D[3] weighttakenoveralltriangles,isnotalwaysatightbound.forexample,consider a9-cyclewhereeveryvertexisgiventheweightk,forsomeintegerk2.while G=D[2] G,themaximum choosea9-cyclebecauseitisthesmallestoddcyclethatcanbeaninducedsubgraph ofthetriangularlattice.thisshowsthatanyalgorithmtocolorhexagongraphsmust useatleastd9d[3] G=2kforthegraph,weknowfromTheorem3.2that(G)=d9k=4e.We graphusingatmost4dd[3] demonstrateanecientapproximationalgorithmthatcanmulticoloranyhexagon G=8ecolorsonsomegraphsGwithtriangleboundD[3] G=3ecolors(andhence,atmost4d(G)=3ecolors). G.Infact,we weletm=dd[3] ponentsofgcanbeindependentlycoloredwithoutanycolorconicts.forsimplicity, Withoutlossofgenerality,weassumethatGisconnected,sincedisconnectedcom- withabasecoloringofgsothateveryvertexgetsbasecolorred,blueorgreen.with eachbasecolor,weassociateaclassofmhuesidentiedwiththeinterval[;m]. G=3eandwechoosethefollowingcolorpaletteinouralgorithm.Start withtheinterval[;m].theentirecollectionof4mdistincthuesformsourcolor palette. Inaddition,wehaveatourdisposalaclassofauxiliarypurplehues,againidentied theauxiliarypurpleclass.wedescribethealgorithmasproceedinginvephases; possiblebeforetryingtousehueseitherfromtheremainingtwobaseclassesorfrom Theideaistoleteachvertexvuseasmanyhuesfromitsbasecolorclassas wemaintaintheinvariantthatattheendofeachphase,thegraphispartiallybut completed.wealsoassignanarticialprioritytovertices:redverticesdominateover letgi=(vi;ei;wi)denotetheremaininggraphafterphasei(i4)hasbeen correctlycolored.tofacilitatereasoningaboutthecorrectnessofthealgorithm,we blueoneswhichinturndominateovergreenones.thispriorityschemeisusedin phases2and3toselect,ineachcase,asuitablesubsetofverticesforpartialcoloring. of24colorsequallydividedamongthered,blue,greenandpurplehues. forwhich3m=d[3] WeillustrateouralgorithmwitharunningexampleshowninFigure,agraphG Avertexv2Gisdenedtobelightifw(v)Mandtobeheavyotherwise.This Gcaneasilybeveriedtobe8.Hence,thecolorpaletteconsists distinctioniscriticaltoeachofthethevephasesbelow: Phase:Everyvertexvisassignedtherstw(v)huesfromitsbasecolorclass,in particular,thehues[;minfw(v);mg].allthelightverticesthusgetcompletely 8

7 7 7 7 7 0 3 8 7 Figure:Ahexagongraphwithinitialweights 2 0 Red Blue Green 3 8 8 3 4 vertexvisdecreasedbym,resultinginthegraphg. coloredandaredeletedfromthegraph.theweightofeveryremainingheavy 7 2 7 everytriangleingmustcontainatleastonelightvertexthatiseliminatedinthe rstphase.lethdenotethesubgraphofginducedbythedegree3verticesing. ItiseasytoseethatGhasnomaximalcliquesofsizegreaterthan2,because 2=3radiansinorder;furthermorethegeometryimpliesthatallthreeneighborshave embedding)ing,thentheincidentedgestotheneighborsformsuccessiveanglesof Notethatifavertexv2Ghasthreeneighbors(say,inclockwiseorderinthexed thesamebasecolor.itfollowsthateachconnectedcomponentinhcontainsvertices orblueandgreenvertices. consistsofeitheranisolatedvertex,orcontainsonlyredandblue,orredandgreen, thatbelongtoatmosttwobasecolorclasses.thus,everyconnectedcomponentofh amongitsneighbors(ifany)inh(recallthatreddominatesbluewhichdominates green).clearly,thepriorityverticesformanindependentset(infact,adominating Callavertexv2Hapriorityvertexifandonlyifithasthehighestpriority set)inh. Phase2:Withoutlossofgenerality,letvbearedpriorityvertexinHwiththree blueneighborsinh.letg(v)bethemaximumamongtheweightsofthe colortheremainingweightonvsinceallthreeblueneighborsofvareheavy ThenvcanborrowfromamongthelastM g(v)greenhues;thesesuceto threegreenneighborsofv;theseverticesmusthavebeenlightverticesing. verticesandhence,w(v)m g(v).accordingly,visassignedthegreen 9

r-6 r-6 r-6 r-6 r-6 r-6 b-6 b-6 b-6 b-6 g-6 g-3 g g g g-3 b-6 g-2 r-6 2 8 4 2 4 (a) (b) g5-6 8 2 4 x v u y t b-2 g-6 2 b-3 4 2 a c b d e Figure2:Colorassignmentduring(a)Phaseand(b)Phase2 hues,[m g(v)+;m],andeliminatedfromfurtherconsideration.notethat thepartialcolorassignmentattheendofphase2hasnocolorconictsamong neighbors,andtheremaininggraphisdesignatedg2. Figure2detailsthepartialcolorassignmentattheendofphaseandphase2 respectively.notethatthesixredhuesaredenotedasr-6andsoforthinthe gure.sincethesubsetofpriorityverticeseliminatedinphase2isadominatingset ofh(thedegree3verticesofg),everyremainingvertexing2nowhasdegreeat most2.equivalently,theconnectedcomponentsofg2consistofisolatedvertices, cyclesandpathsinthetriangulargrid.notealsothatanyedgeofg2hasaresidual weightofatmostm,aconsequenceofthedenitionofheavyvertices.ifthegraph G2containsonlyevencyclesorpaths,thenwecancoloralltheverticesusingtheM purplecolors.howeverg2maycontainisolatedverticesandoddcycles.inthenext phase,weessentiallyeliminateallpotentialcyclesing2. Callavertexv2G2acornervertexifandonlyifithastwoneighborsx;yofthe samebasecolorclassing2suchthattheanglesubtendedatvbytheincidentedges (v;x)and(v;y)isexactly2=3radians.further,acornervertexvisapriorityvertex ing2ifandonlyifvhasthehighestpriorityamongallitsneighbors,ifany,thatare alsocornervertices.itisnotdiculttoseethatthesubsetofpriorityverticesing2 formsanindependentseting2.also,everycornervertexiseitheritselfapriority vertexorisadjacenttoapriorityvertex;hence,thesubsetofpriorityverticesis adominatingsetofthesubgraphinducedbythecornerverticesing2.finally,by denitioneverycycleing2containsatleastonecornervertex.thus,coloringpriority verticesandeliminatingthemalsobreaksallcycles.forexample,infigure2-(b), 0

a x p Red t u Figure3:LocalgeometryaroundabluepriorityvertexvinPhase3 r Blue v Green priorityvertices. theverticeslabeledv,v0andxarecornerverticesofg2;amongthem,vandv0are b y q Phase3:Withoutlossofgenerality,letvbeabluepriorityvertexinG2withred neighborsxandying2asshowninfigure3.notethatudenotesthethird blueverticespandqmustbelightvertices.whileuisabsenting2,thereare priorityconsiderationsthatxandymustbenon-cornerverticesandhence,the neighborofvofthesamebasecolorclassasxandy.itisalsoeasytoseefrom (i)uwaseliminatedinphase(i.e.w(u)=0):letg(v)bethemaximum twopossibilities. overtheweightsofv'sgreenneighbors(i.e.verticesa,bandtinfigure3) ing.sinceudidnotparticipateinphase2,vcanborrowfromamongthe lastm g(v)greenhues;thesesucesincetwoofv'sredneighborsareheavy [M w2(v)+;m],andeliminated. verticesandhencew2(v)m g(v).accordingly,visassignedthegreenhues, andrecallthatuwasassignedthelastw(u)greenhues,[m w(u)+;m], greenneighborsofuandving.letwabbethemaximumamongw(a)andw(b) (ii)uwasapriorityvertexinphase2:consideraandb,thecommon(light) thegreenhues[wab+;wab+w2(v)]fromthemiddleofthegreenspectrum. However,wemustensurethatthisassignmentwillnotconictwiththegreen duringphase2.sincew2(v)+w(u)+wabm,itappearsthatvcouldborrow huesassignedtot(seefigure3). Ifw(t)wab,thenclearlythenewassignmentdoesnotconictwithprior nowusestherstwabandthelastw(t) wabgreenhues.thenewassignment colorassignments.however,ifw(t)>wab,thenwecanrecolortasfollows:its originalassignmentoftherstw(t)greenhues(inphase)ischangedsothatt

g2 p p-4 g2-5 b4-5 4 p p g3-4 g6-2 Figure4:Colorassignmentduring(a)Phase3and(b)Phase4 8 p p p-6,b5-6 sincew(t)+w2(v)m. totcannotconictwiththegreenhues,[wab+;wab+w2(v)],borrowedbyv (b) otherneighboroft.clearlysuchaconictcouldoccuronlyifsomeotherneighborof talsoborrowsgreenhuesinphases2or3.fromtheobservationsabove,weconclude Itremainstoverifythatrecoloringtdoesnotcauseacascadingconictwithany phase)andp;q(theyarelighting).theonlyremainingpossibility,viz.vertexr,is thatthiswouldbeimpossibleforverticesx;y(theyarenotpriorityverticesineither 3,itwouldhavegreenneighborsinG2andwouldborrowbluehues.Figure4-(a) notaproblemeither:rcouldnothavebeenapriorityvertexinphase2(sinceithas threeconsecutivelightneighborsp;tandq)andifitwereacornervertexinphase lightvertexlabeledtinfigure2-(b)isrecoloredasdescribedabove. depictsthecolorassignmentduringphase3fortherunningexample;notethatthe paths.itiseasytoseethatanyremainingisolatedvertexmayhavearesidualweight furthermore,theremaininggraphg3consistsonlyofisolatedverticesorstraight-line Thus,attheendofphase3,wehaveacorrectpartialassignmentofcolors,and Phase4:Toanyisolatedvertexv2G3,werstassignminfw3(v);Mgpurplehues. betweenand2m,whereastheweightoneveryremainingedgeisatmostm. Ifw3(v)>M,thenwestillneedtond=w3(v) MMadditionalcolorsto nishcoloringv.withoutlossofgenerality,weassumethatvisaredvertex, andobservethatalltheneighborsofvmusthavebeenlightverticesing. Hence,theblue(green)neighborsofvmusthavehadcolorseitherassigned totheminphaseorinphase3(asaresultofrecoloring).weclaimthat 2

vcanstillborrowcolorsfromeithertheblueorthegreenpaletteswithout beusingcolorsfromeitherendoftheirbasecolorpalettes.letb3b2bm Fromthedescriptionofphasesand3,observethatthelightneighborsofvmay conictingwithanyneighboringassignment. andg3g2gmbetheweightsofv'sblueandgreenneighborsing thebluepalette.likewise,thereareatleastm (g+g2)greenhuesavailable assignedbluehues,vhasatleastm (b+b2)huesavailableforitsusefrom respectively.clearly,regardlessofthemannerinwhichtheblueneighborsare tog.sincethegreenvertexwithweightgformsatriangleingwithvand that M (b2+g).asimilarargumentshowsthatm (g2+b).itfollows eitheroneoftheblueverticeswithweightborwithweightb2.inanyevent, orinotherwords,thatvcanobtaintheremainingcolorsbyborrowingeither onlybluehuesoronlygreenhueswithoutanycolorconictswithassignments maxfm (b+b2);m (g+g2)g; verticesinourrunningexample.attheendofphase4,theremaininggraphconsists Figure4-(b)demonstratesthecolorsassignedinPhase4totheremainingisolated priortothephase. onlyofstraight-linepaths,i.e.pathsinwhichanytwoconsecutiveedgessubtendan Phase5:Sinceeveryremainingconnectedcomponentisastraight-linepathwitha edgehasaresidualweightofatmostm. angleofdegreesatthecommonvertex.further,asnotedabove,everyremaining weightedchromaticnumberofatmostm,itsucestousethegreedyparitybasedstrategydescribedinsection3tonishcoloringthegraphusingthem sincethepurplehueswereusedonlyinphase4tocolorisolatedverticesinp; purplehues.thiscannotcauseanyconictwithpreviouslycoloredvertices runningexample.thefollowingresultisimmediate: Figure5showstheentirecolorassignmentconstructedbyouralgorithmforthe thelatteraredisconnectedfromanyremainingvertexinthecurrentphase. foranyhexagongraphg=(v;e;w)canbeecientlycomputedinlineartime. Theorem4.Anapproximatemulticoloringthatusesnomorethan4dD[3] G=3ecolors 3

r-6,g2 b-6,p b-6,p g r-6,p6 b-2 g-6,p-4 g r-6,g2-5 b-3 g-6,b4-5 b-6,p g r-6,p g-3 r-6,g5-6 g6-2 b-6,g3-4 r-6,p-6,b5-6 5 Adistributedimplementation Figure5:Completecolorassignmentintheexamplegraph b-6,p g-2 r-6,p modelinganetworkofprocessors(basestations),witheachprocessorresponsiblefor implementedinacompletelydistributedmanner.weconsiderthehexagongraphas Thealgorithmgivenintheprevioussectionhastheadditionalpropertythatitcanbe asinglevertexinthehexagongraph.thenetworkhasthesamespatialembeddingas thegraph,andprocessorsatneighboringbasestationsinthenetworkcanexchange localinformationeciently.foreaseofdescription,wewillsometimesidentifythe verticesofthehexagongraphwiththeirprocessors. arbitraryvertextobetheorigin,andthreedirectionalaxesthatintersecttheorigin: onedesignatedthehorizontalaxis,andtheremainingtwoatangles=3and2=3 Inthexedplanarembeddingoftheinnitetriangulargrid,wecanselectan fromthehorizontalaxis.itiseasytoseethatanypathinthegraph,wheretheangles subtendedbyallintermediateedgesinthepathareexactly,isorientedalongone ofthethreedirectionalaxes.foreveryvertex,wewishtoassignaparitywithrespect toeachdirectionalaxis.thiscaneasilybedoneasfollows.theparityofavertexv alongthehorizontalaxisisdenedtobetheparityofthelengthofthepathoriented alongthehorizontalaxisfromvtoavertexonthe=3axisthatintersectstheorigin. 4

axisintersectingtheorigin.thusgivenanarbitrarynitehexagongraph,anypath thepathfromvorientedalongthe=3axis(2=3axis)toavertexonthehorizontal Similarly,theparityofvalongthe=3axis(2=3axis)istheparityofthelengthof canbepre-computedaccordingtotheparitiesoftheverticesalongthepath. inthegraphthatisorientedalongoneofthedirectionalaxeshasa2-coloringthat information: Abasecoloringforthegraphisknown:eachprocessorknowswhetheritsvertex Ouralgorithmassumesthateachprocessorinitiallyhasaccesstothefollowing A2-coloringalongeachpathorientedalongadirectionalaxisisknown.This meanseachprocessorknowsthreebitscorrespondingtowhetherithasevenor isred,greenorblue,andalsothecolorofeachofitsneighbors. ThevalueofD[3] classesamongprocessorsisknown.additionally,thisimpliesthateveryvertex oddparityalongeachofthe3directionalaxes. hasaccesstoaknownsetofmpurplehuesifneeded. Gisknown;thisalsoimpliesthatthedivisionofbasecolor insection4.thealgorithmstartswiththreeroundsofinformationgathering,after participateinoneormoreofthevephasesoftheapproximationalgorithmdescribed Thedistributedalgorithmconsistsofeachprocessordeterminingwhetheritshould whichnomorecommunicationotherthaninformingneighborsofthecurrentcolor assignmentisrequired;essentially,processorscancontinueindependentlytocompute Round:psendsitsweighttoeachofitssixneighbors. thehuestoassigntothemselves.wedescribethecommunicationroundsfromthe perspectiveofaxedprocessorp. Round2:Havingreceivedtheweightsofallitsneighbors,pdecidesifitwouldbea degree3vertexafterphase,andsendsthisinformation(asinglebit)toeach Round3:Theinformationreceivedintheprevioustworoundssucesforptodecide ofitsneighbors.thiswouldbethecaseifpisitselfaheavyvertexandhas threeneighborsthatareheavyvertices. isabluevertexwithdegree3afterphaseeitherwithnoneighborsthatwill ifitwillbeapriorityvertexinphase2aswellasifanyofitsneighborswill alsobedegree3verticesafterphase,orwithgreenneighborsofdegree3after bepriorityverticesinphase2.forinstance,pwillbeapriorityvertexifit phase(seesection4). 5

Next,pdeterminesifitwillbeacornervertexinphase4andsendsthisinformation(asinglebit)toeachofitsneighbors.pisacornervertexifallthe pisaheavyvertex. phasexactlytwoneighborsofthesamecolorclassthatareheavyvertices pwillnotbeapriorityvertexinphase2. followingconditionsaremet: Round4:TheinformationderivedfromRound3enablesptodetermineifitwillbe apriorityvertexinphase3,asdescribedinsection4.ifpwouldbeapriority butnotpriorityverticesinphase2. neighbor,sayqthatwouldneedtoberecoloredinthatphase.inthiscase,p vertexinphase3,andwouldfallintocase(ii)ofphase3,thenithasalight Round5:AlightvertexthatgotamessagetorecoloritselfinRound4informs sendsamessagetoqwiththemaximumweightofitsremainingtwoneighbors allitsneighborsofhowmanycolorsitwillusefromtheendofitsbasecolor ofthesamecolorasq,sothatpcancoloritselfappropriately. Fromtheweightsofitsneighboringprocessorsanditslimitedglobalknowledge, appropriatelyinphase4. spectrum.thisenablesanisolatedvertexinitsneighborhoodtocoloritself cessorcaneasilycomputethecolorsitwilluseineachofthevephases.without lossofgenerality,consideraprocessorthatcorrespondstoabluevertexv2g.the andtheinformationcollectedinthecommunicationroundsdescribedabove,apro- Phase:Ifw(v)<=M,theprocessorassigns(toitself)theappropriatebluehues. processoremulatesthevephasesofthesequentialalgorithmasfollows: Phase2:IftheprocessorisapriorityvertexinPhase2,itsimulatesphase2and continues. Otherwise,itassignstoitselfallthebluehues,reducesitsweightbyMand wouldborrowfromoneofitsneighboringcolorclassarethelastcolorsfrom stops,orelsecontinuestophase3.recallthatthecolorsthatapriorityvertex Phase3:IftheprocessorisapriorityvertexinPhase3,itcandeterminethecolors atapriorityvertex. thatclass;thusnoconsultationisrequiredwithneighborstocomputethecolors itneedstoborrowfromtheappropriateneighboringcolorclass. 6

distributedalgorithmaswell.todoso,itusestheinformationitreceivedin inphase3ofthesequentialalgorithm,thenitcanemulatethisbehaviorinthe If,however,theprocessorisalightvertexthatwouldhaveundergonerecoloring Phase4:Ifaprocessorisanisolatedvertexinthisphase,itassignsitselfanypurple itsneighborinround4,andrecolorsitselfasdescribedinsection4sothatno huesthatitneeds.additionalcolorsthatitmayneedareborrowedfromone conictappearsamongthebluecolors. recoloreditselfinthelastphasesentinformationinround5aboutthenumber ofcolorsitwouldusefromtheendofitsbasecolorspectrum,theprocessorcan oftheneighboringcolorclasses.sinceanyofitslightneighborsthatmayhave Phase5:Anyvertexwithunassignedcolorsatthisstageliesalongsomestraight-line determineallcolorsusedbyitslightneighborsandcanborrowcolorswithout anypossibilityofconict. caneasilycomputetheidentityoftheparticularaxisfromtheinformation neighborsthatliealongexactlyoneofthethreedirectionalaxes.theprocessor pathinthegrid.inparticular,itcandetectitsoneortwoincompletelyassigned thesetofpurplehues,asinthecoloringforbipartitegraphs. alongthisaxis,itassignsitselfthenecessaryhuesfromthebeginningorendof gatheredafterround3.dependingonwhetheritisanevenoroddvertex completelydistributedmanner,afteraninitialconstanttimecommunicationprotocol foranyhexagongraphg=(v;e;w),canbeecientlycomputedinconstanttimeina Theorem5.Anapproximatemulticoloringthatusesnomorethan4dD[3] G=3ecolors whereeachprocessorexchangesvemessageswitheachofitsneighbors. 6Inthispaper,wehavecasttheproblemoffrequencyassignmentincellularnet- worksasamulticoloringproblemforhexagongraphs.forsomeparticularinduced Discussion algorithmsformulticoloringthemusinganoptimalnumberofcolors.insection4, subgraphsofhexagongraphs,i.e.cyclesandouterplanargraphs,weshowecient wedescribeamulticoloringalgorithmthatusesatmost4dd[3] themaximumweightonany3-cliqueing,isatriviallowerboundontheminimum numberofcolorsrequired.weshowedalsoahexagongraphthatrequiresd9d[3] G=3ecolorswhereD[3] G=8e colors.determininganexactboundon(g),theweightedchromaticnumberofan arbitraryhexagongraphisnp-hard;ourresultsdoestablishthatforallhexagon 7

graphsg,(g)4dd[3] forhexagongraphswhichalwaysusesatmostd9d[3] G=3e.Whetherornotthereisanapproximationalgorithm distributedmanner.incontrast,thealgorithmofmcdiarmidandreed[7],which openproblem.ausefulfeatureofouralgorithmisthatitcanbeimplementedina G=8ecolorsremainsanintriguing madedistributedinanyobviousway. hasthesameperformanceratioasours,seemsinherentlycentralizedandcannotbe quirednotmerelytobedierent,butalsotobefarenoughapart[3].anotherrecently wherethefrequenciesassignedataparticularvertexoratadjacentverticesarere- Aninterestingavenueforfutureresearchisthegeneralizedversionoftheproblem, proposedmodel[]considersarbitraryinterferencegraphswithpre-denedcostson totalcost,orequivalently,thenetinterferenceofanassignment.finally,thedynamic frequenciesareassignedtoneighboringnodes.theobjectivehereistominimizethe theedges;thesecostsreecttheinterferencepenaltieswhenthesameoradjacent inthissettingandwhatboundscanbeprovedonitsperformance. versionoftheprobleminvolveschangingweightsatvertices.itwouldbeinteresting toseeifthedistributedalgorithmwedescribeinsection5canbeadaptedtowork Acknowledgments WethankJeannetteJanssenforintroducingustotheproblemofchannelassignment, formofcorollary3.3. totheanonymousrefereesfortheirusefulcomments,andforsuggestingthecurrent andforcommentsthatgreatlyimprovedthepresentationofsection3.wearegrateful References []R.Borndorfer,A.Eisenblatter,M.Grotschel,andA.Martin.Frequencyassignmentincellularphonenetworks.Technicalreport,Knorad-ZuseZentrumfur InformationstechnikBerlin,997. [2]D.DimitrijevicandJ.Vucetic.Designandperformanceanalysisofalgorithmsfor [3]W.K.Hale.Frequencyassignment:Theoryandapplications.Proceedingsofthe ogy,42(4):526{534,993. channelallocationincellularnetworks.ieeetransactionsonvehiculartechnol- IEEE,68(2):497{54,980. 8

[5]T.KahwaandN.Georganas.Ahybridchannelassignmentschemeinlargescalecellular-structuredmobilecommunicationsystems.IEEETransactionson forcellulartelephonesystems.unpublishedmanuscript,april995. [4]J.Janssen,K.Kilakos,andO.Marcotte.Fixedpreferencefrequencyallocation [6]S.KimandS.L.Kim.Atwo-phasealgorithmforfrequencyassignmentincellular mobilesystems.ieeetransactionsonvehiculartechnology,994. Communications,4:432{438,978. [8]P.Raymond.Performanceanalysisofcellularnetworks.IEEETransactionson [7]C.McDiarmidandB.Reed.Channelassignmentandweightedcoloring.submitted Communications,39(2):787{793,99. forpublication,997. [9]W.WangandC.Rushforth.Anadaptivelocal-searchalgorithmforthechannelassignmentproblem.TechnicalReport,August995. 9