/
IEEETRANSACTIONSONINFORMATIONTHEORY,VOL.54,NO.6,JUNE20082)Totransmitme IEEETRANSACTIONSONINFORMATIONTHEORY,VOL.54,NO.6,JUNE20082)Totransmitme

IEEETRANSACTIONSONINFORMATIONTHEORY,VOL.54,NO.6,JUNE20082)Totransmitme - PDF document

sherrill-nordquist
sherrill-nordquist . @sherrill-nordquist
Follow
388 views
Uploaded On 2016-10-01

IEEETRANSACTIONSONINFORMATIONTHEORY,VOL.54,NO.6,JUNE20082)Totransmitme - PPT Presentation

user ndsthetwocodewordscorrespondingtocomponents andalsouniformlychoosesacodeword from User thenaddsallthesecodewordsandtransmitstheresultingcodeword sothatitactuallytransmitsoneof codewordsLet No ID: 471324

user ndsthetwocodewordscorrespondingtocomponents andalsouniformlychoosesacodeword from .User thenaddsallthesecodewordsandtransmitstheresultingcodeword sothatitactuallytransmitsoneof codewords.Let No

Share:

Link:

Embed:

Download Presentation from below link

Download Pdf The PPT/PDF document "IEEETRANSACTIONSONINFORMATIONTHEORY,VOL...." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.


Presentation Transcript

IEEETRANSACTIONSONINFORMATIONTHEORY,VOL.54,NO.6,JUNE20082)Totransmitmessage ,user ndsthetwocodewordscorrespondingtocomponents andalsouniformlychoosesacodeword from .User thenaddsallthesecodewordsandtransmitstheresultingcodeword ,sothatitactuallytransmitsoneof codewords.Let Notethatsinceallcodewordsarechosenuniformly,user essentiallytransmitsoneof codewordsatrandomforeachmessage ,anditsoverallrateoftransmission cally,wechoosetheratestosatisfy (26) withequalityif (27) whichwecanalsowriteas (29) withequalityif (30) Notethatif(31)iszeroforagroupofusers,wecannotachievesecrecyforthoseusers.When ,ifthesumca-pacityofthemainchannelislessthanthatoftheeavesdropperchannel,i.e., ,secrecyisnotpossibleforthesystem.Assumethisquantityispositive.Toensurethatwecanmutuallysatisfyboth(30)and(31),wecanreclassifysomeopenmes-sagesassecret.Clearly,ifwecanguaranteesecrecyforalargersetofmessages,secrecyisachievedfortheoriginalmessages.Fromtherstsetofconditionsin(25)andtheGMACcodingtheorem[44]withhighprobabilitythereceivercandecodethecodewordswithlowprobabilityoferror.Toshowthesecrecyconditionin(12),rstnotethatthecodingschemedescribedisequivalenttoeachuser selectingoneof messages,andsendingauniformlychosencodewordfromamong codewordsforeach.De ,andwehave (32) (33) (34) whereweused ,andthuswehave toget(35).Wewillconsiderthetwotermsindividually.First,wehavethetrivialboundduetochannelcapacity Nowwrite Sinceuser independentlysendsoneof codewordsequallylikelyforeachsecretmessage (38) (39) Wecanalsowrite (41)where as since,withhighprobability,theeavesdroppercandecode given dueto(30)andcodegeneration.Using(36),(37),(40),and(41)in(35),weget (42) Now,letusconsidertheTDMAregiongivenin(22).Thisre-gionisobtainedwhenuserswhocanachievesingle-usersecrecyuseasingle-userwiretapcodeasin[12]inaTDMAschedule,wherethetimeshareofeachuser isgivenby and .Atransmitter whocanachievesecrecy,i.e.,having ,transmitsfor portionofthetimewhenallotherusersaresilent,using power,satisfyingitsaveragepowerconstraintovertheTDMAtimeframe.Thisapproachwasusedin[40]toachievesecrecysumcapacityforindividualconstraints.Whenthechannelisdegraded,i.e., forall ,thenforcollectiveconstraintstheTDMAregionisseentobeasubsetofthesuperpositionregion.However,thisisnotnecessarilytrueforthegeneralcase,andbytimesharingbetweenthetwoschemes,wecangenerallyachievealargerachievableregion,givenin(25). Weremarkthatitispossibletofurtherdividethesagestogetmoresetsofprivatemessages,whicharealsoper-fectlysecret,i.e.,ifwelet , ,thenaslongasweimposethesamerestrictionson as ,wecanachieveperfectsecrecyof ,asin[12].However,thisdoesnotmeanthatwehaveperfectsecrecyatchannelcapacity,asthesecrecysubcodescarryinformationabouteachother.Observethatevenfor users,aratepointinthisregionisfourdimensional,andhencecannotbeaccuratelydrawn.Wecaninsteadfocusonthesecrecyrateregion,theregionofallachievable .Thesubregions and shownfordifferentchannelgainsinFig.3forxedtransmitpowers,and users.Fig.4representshowtheseregionschangewithdifferenttransmitpowerswhenthechannelgains TEKINANDYENER:GGMAC-WTANDGTW-WT:ACHIEVABLERATESANDCOOPERATIVEJAMMING2747single-userdecodable,andifithasenoughpowertomaketheotheruserintheneweffectivechannel.B.GTW-WTOnceagain,weproposetomaximizethesecrecysumrateusingcooperativejammingwhenuseful.Thisproblemisfor-mallystatedasfollows: wherewerecallthat isgivenby(71)and (102) Notethat sincethereareonlytwoterminals.AsimilarargumenttotheGGMAC-WTcasecaneasilybeusedtoestablishthatwecanassumeausertobeeithertransmittingorjamming,butnotboth.Sincethejamminguserisalsothereceiverthattheotheruseriscommunicatingwithandknowsthetransmittedsignal,thisschemeentailsnolossofcapacityasfarasthetransmittinguserisconcerned.Theoptimumpowerallocationsaregivenasfollows.Theorem6:Theachievablesecrecysumrateforthecollabo-rativeschemedescribedis (104)where isthesetoftransmittingusersandtheoptimumpowerallocationsaregivenby(105)shownatthebottomofthepage.Proof:SimilartotheGGMAC-WT,westartwiththesub-problemofndingtheoptimalpowerallocationgivenajam-mingset.TheLagrangianisgivenby Takingthederivativewehave if if sinceauser satises ,itmusthave Consideruser .Weagainarguethatif ,then andif ,then .Nowexamineauser .Itiseasytoseethatsincesuchauseronlyharmsthejammer,theoptimaljammingstrategyshouldhave ,i.e.,themaximumpower.Thiscanalsobeseenbynotingthat(107)forthiscasesimpliesto andhencewemusthave forall Thejammingsetwillbeoneof ,sincethereisnopointinjammingwhenthereisnotransmission.Also,ifanyofthetwousersisjamming,bytheargumentabove, , .Wecaneasilyseethatjammingbyauser offersanadvantageif ,i.e., iff for .Thus,when ,bothusersshouldbetransmittinginsteadofjamming.However,whenanyuser ,jammingalwaysdoesbetterthanthecasewhenbothusersaretransmitting.Inthiscase, someuser ,andtheobjectivefunctionin(101)isminimizedwhenthisuserisjamming,andtheotheroneistransmitting.If,however, ,thenitwillnottransmit,andweshouldnotbejamming.Consolidatingalloftheseresults,wecomeupwiththepowerallocationinTheorem6. Remark4:Asufcient,butnotnecessaryconditionfortheweakerusertobethejamminguserisif ;thiscasecorrespondstohavinghighersignal-to-noiseratio(SNR)attheeavesdropperfortheoriginal,nonstandardizedmodel.Thiscanbeinterpretedasjamwithmaximumpowerifitispossibletochangeuser1seffectivechannelgainsuchthatitisnolongersingle-userdecodable.Forthesimplecaseofequalpowercon- ,itiseasilyseenthatuser1shouldneverbejamming.Theoptimalpowerallocationinthatcasereduces bothtransmit if transmits jams if VI.NESULTSInthissection,wepresentnumericalresultstoillustratetheachievableratesobtained,aswellasthecooperativejammingschemeanditseffectonachievablesecrecysumrates.Asmentionedearlierinthispaper,examplesofachievablese-crecyrateregionsaregiveninFigs.4and6fortheGGMAC-WT andGTW-WT,respectively.ComparingFigs.4and6,weseethattheGTW-WTachievesalargersecrecyrateregion bothtransmit if transmits jams if transmits jams if , transmits jams if , otherwise.(105) TEKINANDYENER:GGMAC-WTANDGTW-WT:ACHIEVABLERATESANDCOOPERATIVEJAMMING2743IV.MAXIMIZATIONOFATETheachievableregionsgiveninTheorems1and2dependonthetransmitpowers.Weare,thus,naturallyinterestedinthepowerallocation thatwouldmaximizethetotalsecrecysumrate.Recallthatthestandardizedchannelgainforuser is ,andthatthehigher is,thebetterthecorre-spondingeavesdropperchannel.Withoutlossofgenerality,as-sumethatusersareorderedintermsofincreasingstandardizedeavesdropperchannelgains,i.e., .Notethatweonlyneedtoconcernourselveswiththecase sincewecancombineuserswiththesamechannelgainsintoonesuperuser.Wecanthensplittheresultingoptimumpowerallocationforasuperuseramongtheactualconstitutingusersinanywaywechoose,sincetheywouldallresultinthesamesumrate.Inaddition,fromaphysicalpointofview,assumingthatthechannelparametersaredrawnaccordingtoacontinuousdistributionandthenxed,theprobabilitythattwouserswouldhavethesameexactstandardizedchannelgainiszero.A.GGMAC-WTrstexaminethesuperpositionregiongivenin(20).ThesecrecysumrateachievablewithsuperpositioncodingfortheGGMAC-WTwasgiveninTheorem1as - andwewouldliketondthepowerallocationthatmaximizesthisquantity.Statedformally,weareinterestedinthetransmitpowersthatsolvethefollowingoptimizationproblem: (57) (58)where (59)and yields(58).Inobtaining(58),wesimplyusedthemonotonicityofthe function.Thesolutiontothisproblemisgivenbelow.Theorem3:Thesecrecysum-ratemaximizingpoweralloca-tionfor satises if and is where issomelimitingusersatisfying andwede and .Notethatthisallocationshowsthatonlyasubsetofthestrongusersmustbetransmitting.Proof:WestartwithwritingtheLagrangiantobemini- EquatingthederivativeoftheLagrangiantozero,weget wherewede foranyset Itiseasytoseethatif ,then ,andwehave .If ,thenwesimilarlyndthat .Finally,if ,thenwealsohave (64)and doesnotdependon ,sowecanset withnoeffectonthesecrecysumrate.Thus,wehave if ,and if .Then,theoptimalsetoftransmittersisoftheform sinceifauser istransmitting,alluserssuchthat mustalsobetransmitting.Wealsonotethat .Let bethelastusersatisfyingthisproperty,i.e., and .Notethat (65) Inotherwords,allsets for alsosat-isfythispropertyandareviablecandidatesfortheoptimalsetoftransmittingusers.Therefore,wecanclaimthat istheop-timumsetoftransmittingusers,sincefromabovewecanitera-tivelyseethat forall . Notethat,forthespecialcaseof users,theoptimumpowerallocationis if , if , WealsoneedtoconsidertheTDMAregion.Inthiscase,themaximumachievablesecrecysumrateis Thisisasimplecomplexoptimizationproblemthatcaneasilybesolvednumerically.Forthedegradedcase,wecanobtainaclosedformsolution: asin[40].Ingeneral,wecannotobtainsuchasolution.However,itistrivialtonotethatuserswith shouldnotbetransmittinginthisscheme.ThesecrecysumrateisthenthemaximumofthesolutionsgivenbythesuperpositionandTDMAregions. IEEETRANSACTIONSONINFORMATIONTHEORY,VOL.54,NO.6,JUNE2008whenthewiretapcapacityiszero.AhslwedeandCsiszr[17],[18]examinedthesecretkeycapacityandcommonrandomnesscapacity,forseveralchannels.Theseresultsalsobenetfrom[14]toprovidesecretkeycapacities.Maureralsoex-aminedthecaseofactiveadversaries,wherethewiretapperhasread/writeaccesstothechannelin[19][21].Thesecretkeygenerationproblemwasinvestigatedfromamultipartypointofviewin[22]and[23].Notably,CsiszrandNarayanconsideredthecaseofmultipleterminalswhereanumberofterminalstrytodistillasecretkeyandasubsetoftheseterminalscanactashelperterminalstotherestin[24]and[25].Recently,severalnewmodelshaveemerged,examiningse-crecyforparallelchannels[26],[27],relaychannels[28],andfadingchannels[29],[30].Fadingandparallelchannelswereexaminedtogetherin[31]and[32].Broadcastandinterferencechannelswithcondentialmessageswereconsideredin[33].LiangandPoor[34]andLiuetal.[35]examinedthemultiple-accesschannelwithcondentialmessageswheretwotransmit-terstrytokeeptheirmessagessecretfromeachotherwhilecommunicatingwithacommonreceiver.In[34],anachiev-ableregionwasfoundingeneral,andthecapacityregionwasfoundforsomespecialcases.Multiple-input(MIMO)channelswereconsideredin[36]and[37].In[38][41],weinvestigatedmultiple-accesschannelswheretransmitterscommunicatewithanintendedreceiverinthepres-enceofanexternalwiretapperfromwhomthemessagesmustbekeptcondential.In[38][40],weconsideredthecasewherethewiretappergetsadegradedversionofaGMACsignal,andnedtwoseparatesecrecymeasuresextendingWynersmea-suretomultiuserchannelstoreecttheleveloftrustthenetworkmayhaveineachnode.Achievablerateregionswerefoundfordifferentsecrecyconstraints,anditwasshownthatthese-crecysumcapacitycanbeachievedusingGaussianinputsandstochasticencoders.Inaddition,time-divisionmultipleaccess(TDMA)wasshowntoalsoachievethesecrecysumcapacity.Gaussianandbinaryadditivetwo-waywiretapchannelswereexaminedin[42].Inthispaper,weconsiderthegeneralGaussianmultiple-ac-cesswiretapchannel(GGMAC-WT)andtheGaussiantwo-waywiretapchannel(GTW-WT),bothofwhichareofinterestinwirelesscommunicationsastheycorrespondtothecasewhereasinglephysicalchannelisutilizedbymultipletransmitters,suchasinanadhocnetwork.WeconsideranexternaldropperthatreceivesthetransmitterssignalsthroughageneralGaussianmultiple-accesschannel(GGMAC)inbothsystemmodels.Weutilizeasuitablesecrecyconstraintthatisthenor-malizedconditionalentropyofthetransmittedsecretmessagesgiventheeavesdropperssignal,correspondingtothetivesecrecyconstraintsusedin[40].Weshowthatsatisfyingthisconstraintimpliesthesecrecyofthemessagesforallusers.Inbothscenarios,transmittersareassumedtohaveonesecretandoneopenmessagetotransmit.Thisisdifferentfrom[40]inthatthesecrecyratesarenotconstrainedtobeatleastaxedportionoftheoverallrates.Wendanachievablesecrecyrateregion,whereuserscancommunicatewitharbitrarilysmallEventhoughwefaithfullyfollowWynersterminologyinnamingthechan-nels,admittedlyinwirelesssystemmodels,eavesdropperisamoreappropriatetermfortheadversary.probabilityoferrorwiththeintendedreceiverunderperfectse-crecyfromtheeavesdropper,whichcorrespondstotheresultof[40]forthedegradedcase.Wenotethat,inaccordancewiththerecentliterature,whenweusethetermperfectsecrecy,wearereferringtosecrecy,wheretherateofinformationleakedtotheadversaryislimited.Assuch,thiscanbethoughtofasalmostperfectsecrecy.Wealsondthesum-ratemax-imizingpowerallocationsforthegeneralcase,whichismoreinterestingfromapracticalpointofview.Itisseenthataslongastheusersarenotsingle-userdecodableattheeaves-dropper,asecrecy-ratetradeoffispossiblebetweentheusers.Next,weshowthatanontransmittingusercanhelpincreasethesecrecycapacityforatransmittinguserbyeffectivelytheeavesdropper,andevenenablesecretcommunica-tionsthatwouldnotbepossibleinasingle-userscenario.Wetermthisnewschemecooperativejamming.TheGTW-WTisshowntobeespeciallyusefulforsecretcommunications,asthemultiple-accessnatureofthechannelhurtstheeavesdropperwithoutaffectingthecommunicationrate.Thisisbecausethetransmittedmessagesofeachuseressentiallyhelphidetheotherssecretmessages,andreducetheextrarandomnessneededinwiretapchannelstoconfusetheeavesdropper.Therestofthispaperisorganizedasfollows.SectionIIde-scribesthesystemmodelfortheGGMAC-WTandGTW-WTandtheproblemstatement.SectionIIIdescribesthegeneralachievableratesfortheGGMAC-WTandtheGTW-WT.Sec-tionsIVandVgivethesecrecysumratemaximizingpowerallocations,andtheachievablerateswithcooperativejamming.SectionVIgivesournumericalresultsfollowedbyourconclu-sionsandfutureworkinSectionVII.II.SODELANDROBLEMTATEMENTWeconsider userscommunicatinginthepresenceofaneavesdropperwhohasthesamecapabilities.Eachtrans- hastwomessages, issecretand whichisopen,fromtwosetsofequallylikelymessages , .Let , , , ,and .Themessagesareencodedusing codesinto ,where .Theencodedmessages arethentransmitted.Weassumethechannelparametersareuniversallyknown,andthattheeavesdropperalsohasknowledgeofthecodebooksandthecodingscheme.Inotherwords,thereisnosharedsecret.Thetwochannelsweconsiderinthispaperaredescribednext.A.GGMAC-WTThisisascenariowheretheuserscommunicatewithacommonbasestationinthepresenceofaneavesdropper,wherebothchannelsaremodeledasGaussianmultiple-accesschannelsasshowninFig.1.Theintendedreceiverandthewiretapperreceive and ,respectively.Thereceiverdecodes togetanestimateofthetransmitted .WewouldliketocommunicatewiththeWewouldliketostressthatisnotthesameas,i.e.,wedonotimposeadecodabilityconstraintfortheopenmessagesattheeavesdropper. IEEETRANSACTIONSONINFORMATIONTHEORY,VOL.54,NO.6,JUNE2008Theorem2:Therateregiongivenbelowisachievablefortheconvexclosureof Proof:TheproofisverysimilartotheproofofTheorem1.WeusethesamecodingschemeasTheorem1,butthemaindifferenceisthatwechoosetheratestosatisfy (48) withequalityif (49) orequivalently (51) withequalityif (52) assuming(53)ispositive.Thedecodabilityof from and comesfrom(51)andthecapacityregionoftheGaussiantwo-waychannel[5].Thisgivestherstsetoftermsintheachievableregion.Thekeyhereisthatsinceeachtransmitterknowsitsowncodeword,itcansubtractitsself-interferencefromthereceivedsignalandgetaclearchannel.Therefore,theGaussiantwo-waychanneldecomposesintotwoparallelchan-Thesecondgroupoftermsin(45),resultingfromthesecrecyconstraint,canbeshownthesamewayastheproofofTheorem1,since hasthesameformforbothchannels.Inotherwords,asfarastheeavesdropperisconcerned,thechannelisstillaGMACwith users.Assuch,weneedtosend extracodewordsintotal,whichneedtobesharedbythetwoterminalsprovidedtheyarenotsingle-userdecodable. Fordifferentchannelgains,theregionofall (45)isshowninFig.5.Sincewerequirefourdimensionsforanaccuratedepictionofthecompleterateregion,weonlyfocusonourmaininterest,i.e.,thesecrecyrateregion.Fig.6showstheachievablesecrecyrateregionasafunctionoftransmitpowers.Wenotethathigherpowersalwaysresultinalargerregion.Weindicatetheconstraintontheoverallrates,correspondingtothecapacityregionoftheGaussiantwo-waychannel,bythedottedline.NotethatthesecrecyregionhasastructuresimilartotheGGMAC-WTwith .Asfarastheeavesdropperisconcerned,thereisnodifferencebetweenthetwochannels.However,sincethemainchannelbetweenusersdecomposesintotwoparallelchannels,higherratescanbeachievedbetween Fig.5.GTW-WTachievableregionsfordifferentchannelparameters (P =2) Fig.6.GTW-WTachievablesecrecyregionwhen P h 3,andh thelegitimateterminals(users).Thus,ineffect,eachusertransmittedcodewordsactasasecretkeyfortheotherusertransmittedcodewords,requiringfewerextraneouscodewordsoveralltoconfusetheeavesdropper,andalargersecrecyre-gion.Wenotethatausermayeitherachievesecrecyornot,dependingonwhetheritissingle-userdecodable.Asaresult,TDMAdoesnotenlargetheregion,sinceeachusercanatleastachievetheirsingle-usersecrecyrates.Toseethis,notethattheconstraintonthesecrecysumratecanbewrittenas (54) sothattransmittinginthetwo-waychannelalwaysprovidesanadvantageoverthesingle-userchannels. ofthetransmittedmessagegiventhereceivedsignalatthewiretapperasthesecrecymeasure,hefoundtheregionofallpos- pairs,andtheexistenceofasecrecycapacity therateuptowhichitispossibletolimittherateofinformationtransmittedtothewiretappertoarbitrarilysmallvalues. TEKINANDYENER:GGMAC-WTANDGTW-WT:ACHIEVABLERATESANDCOOPERATIVEJAMMING2741 Fig.3.GGMAC-WTachievableregionsfordifferentchannelparameters (P =2) Fig.4.GGMAC-WTachievablesecrecyregionwhen P h ,and xed.Forthecaseshown,weneedtheconvexhullopera-tion,astheachievableregionisacombinationofdifferentsu-perpositionandTDMAregions.Notealsothatthemainextraconditionforthesuperpositionregionisonthetotalextraran-domnessadded.Asaresult,itispossibleforuserstousersbycontributingmoretothenecessaryextranumberofcodewords,whichisthesumcapacityoftheeaves-dropper.Suchaweakuseronlyhastomakesurethatitisnotsingle-userdecodable,providedthestrongerusersarewillingtocesomeoftheirownrateandgeneratemoresupercodewords.Inotherwords,weseethatusersinaset arefur-therprotectedfromtheeavesdropperbythefactthatusersin arealsoundecodable,comparedtothesingle-usercase.TheTDMAregion,ontheotherhand,doesnotallowuserstohelpeachotherthisway.Assuch,onlyuserswhosechannelgainsallowthemtoachievesecrecyontheirownareallowedtoForthespecialdegradedcaseof theperfectsecrecyrateregionfor becomestheregiongivenby[40,Th.1]for .Wealsoobservethateventhoughthereisalimitonthesecrecysumrateachievedbyourscheme,itispossibletosendopenmessagestotheintendedreceiveratratessuchthatthesumofthesecrecyrateandopenrateforallusersisinthecapacityregionoftheMACtotheintendedreceiver.Eventhoughwecannotsendatcapacitywithsecrecy,thecodewordsusedtoconfusetheeavesdroppermaybeusedtocommunicatemeaningfulinformationtotheintendedreceiver.B.GTW-WTInthissection,wepresentanachievableregionfortheGTW-WTusingasuperpositioncodingsimilartothatusedtoachievetheregion fortheGGMAC-WT.Wenethefollowing.nition5(GTW-WTSuperpositionRegion ):Let .Then,theGTW-WTsuperpositionregion isgivenby whichcanbewrittenas Remark2:Wecanalsowritethisregionmorecompactlyasthefollowing: (46) TEKINANDYENER:GGMAC-WTANDGTW-WT:ACHIEVABLERATESANDCOOPERATIVEJAMMING2737 Fig.1.StandardizedGMAC-WTsystemmodel.receiverwitharbitrarilylowprobabilityoferror,whilekeepingthewiretapper(eavesdropper)ignorantofthesecretmessages .Thesignalsattheintendedreceiverandthewiretapperaregivenby (1a) (1b)where and aretheadditivewhiteGaussiannoise(AWGN), isthetransmittedcodewordofuser ,and and arethechannelgainsofuser totheintendedreceiver ),andtheeavesdropper(wiretap ),respectively.Eachcomponentof and .Wealsoassumethefollowingtransmitpowerconstraints: Similartothescalingtransformationtoobtainthestandardformoftheinterferencechannel[43],wecanrepresentanyGMAC-WTbyanequivalentstandardform[40] (3a) where,foreach ,wehavethefollowing:thecodewordsarescaledtoget thenewpowerconstraintsare thewiretappersnewchannelgainsare thenoisesarenormalizedtoget and Wecanshowthattheeavesdroppergetsastochasticallyde-gradedversionofthereceiverssignalif .Weconsideredthisspecialcasein[39]and[40].B.GTW-WTInthisscenario,twotransmitter/receiverpairscommunicatewitheachotheroveracommonchannel.Eachreceiver gets andtheeavesdroppergets .Receiver decodes togetanestimateofthetransmittedmessagesoftheotheruser.Theuserswouldliketocommunicatetheopenandsecretmessageswitharbitrarilylowprobabilityoferror,whilemaintainingsecrecyofthesecretmessages.Thesignalsattheintendedreceiverandthewiretapperaregivenby (4a) (4b) (4c)where and .Wealsoas-sumethesamepowerconstraintsgivenin(2)(with ),andagainuseanequivalentstandardformasillustratedinFig.2 (5a) (5b) wherewehavethefollowing:thecodewords arescaledtoget and themaximumpowersarescaledtoget and thetransmittersnewchannelgainsaregivenby and thewiretappersnewchannelgainsaregivenby and ; TEKINANDYENER:GGMAC-WTANDGTW-WT:ACHIEVABLERATESANDCOOPERATIVEJAMMING2745wherewerecallthat isgivenby(59),suchthat (78) Toseethatausershouldnotbesplittingitspoweramongjammingandtransmitting,itissufcienttonotethatregardlessofhowausersplitsitspower, willbethesame,andtheuseronlyaffects .Assumetheoptimumsolutionissuchthatuser splitsitspower,so and .Then,itiseasytoseethatif ,thesumrateisincreasedwhenthatuserusesitsjammingpowertotransmit,andwhen ,thesumrateisincreasedwhentheuserusesitstransmitpowertojam.When ,thenregardlessofhowitspowerissplit,thesumrateisthesame,andwecanassumeuser eithertransmitsorjams.Notethatwemusthave tohaveanonzerosecrecysumrate,and tohaveanadvantageovernotjamming.Thisschemecanbeshowntoachievethefol-lowingsecrecysumrate.Theorem5:Thesecrecysumrateusingcooperativejamming (80)where isthesetoftransmittersandtheoptimumpowerallo-cationisoftheform with (81)and (82) (83) wheneverthepositiverealrootexists,and Proof:rstsolvethesubproblemofndingtheoptimalpowerallocationforasetofgiventransmitters .Thesolutiontothiswillalsogiveusinsightintothestructureoftheoptimalsetoftransmitters .WestartwithwritingtheLagrangian ThederivativeoftheLagrangiandependsontheuser if if sinceauser satises ,itmusthave Considerauser .Thesameargumentasinthesum-ratemaximizationproofleadsto if and if .Nowexamineauser .Wecanwrite(86)as (87)where (88)Let Then,wehave iff ,and iff .Thus,weagainndthatwemusthave forall .Also,if ,then .Onlyif ,canwehave .Now,since ,wemusthave .Thus,wendthat .Then,weknowthatforagivensetoftrans- ,thesolutionissuchthatallusers transmitwithpower if .Inthesetofjammers ,allusershave ,andwhenthisinequalityisnotsatisedwithequality,thejammersjamwithmaximumpower.Iftheequalityissatisedforsomeusers theirjammingpowerscanbefoundfromsolving Byrearrangingtermsin(88),wenotethattheoptimumpowerallocationforthisuser,callituser ,isfoundbysolvingthe thesolutionofwhichisgivenin(81).Notethat(90)denesan(upright)parabola.Iftherootgivenin(90)existsandispositive,then .Thiscomes TEKINANDYENER:GGMAC-WTANDGTW-WT:ACHIEVABLERATESANDCOOPERATIVEJAMMING2749 Fig.10.GGMAC-WTcooperativejammingexampledarkershadescorrespondtohighervalues.examinethetransmit/jampowersforthisareawhentheeaves-dropperisknowntobeat usingaxedpath-lossmodelforthechannelgains,andplotthetransmit/jampowersandtheachievedsecrecysumratesasafunctionoftheeavesdropperlo-cation.ItisreadilyseenthatwhentheeavesdropperisclosetotheBS,thesecrecysumratefallstozero.Also,whentheeaves-dropperisinthevicinityofatransmitter,thattransmittercannottransmitinsecrecy.However,inthiscase,thetransmittercanjamtheeavesdroppereffectivelyandallowtheothertransmittertotransmitand/orincreaseitssecrecyratewithlittlejammingpower.ThesituationfortheGTW-WTissimilarandisshowninFig.11.Inthiscase,jammingismoreusefulascomparedtotheGGMAC-WT,andweseethatitispossibletoprovidese-crecyforamuchlargerareawheretheeavesdropperislocated,asthejammingsignaldoesnothurttheintendedreceiver.VII.CONCLUSIONANDInthispaper,wehaveconsideredtheGaussianmultiple-ac-cessandtwo-waychannelsinthepresenceofanexternaleaves-dropperwhoreceivesthetransmittedsignalsthroughamul-tiple-accesschannel,andprovidedachievablesecrecyrates.Wehaveshownthatthemultiple-accessnatureofthechannelscon-sideredcanbeutilizedtoimprovethesecrecyofthesystem.Inparticular,wehaveshownthatthetotalextrarandomnessiswhatmattersmainlyconcerningtheeavesdropper,ratherthantheindividualrandomnessinthecodes.Assuch,itmaybepos-sibleforuserswhosesingle-userwiretapcapacityarezero,tocommunicatewithnonzerosecrecyrate,aslongasitispos-sibletoputtheeavesdropperatanoveralldisadvantage.Thisisevenclearerfortwo-waychannels,whereeventhoughtheeavesdropperschannelgainmaybebetterthanaterminals,theextraknowledgeofitsowncodewordbythatterminalenablescommunicationinperfectsecrecyaslongastheeavesdropperreceivedsignalisnotstrongenoughtoallowsingle-userde-WefoundachievablesecrecyrateregionsfortheGGMAC-WTandtheGTW-WT.WealsoshowedthatfortheGGMAC-WTthesecrecysumrateismaximizedwhenonlyuserswithchannelstotheintendedreceiverasopposedtotheeavesdroppertransmit,andtheydosousingalltheiravailablepower.FortheGTW-WT,thesumrateismaximizedwhenbothterminalstransmitwithmaximumpoweraslongastheeavesdropperschannelisnotgoodenoughtodecodethemusingsingle-userdecoding.Finally,weproposedaschemetermedcooperativejammingwhereadisadvantagedusermayhelpimprovethesecrecyratebyjammingtheeavesdropper.Wefoundtheoptimumpowerallocationsforthetransmittingandjammingusers,andweshowedthatsignicantrategainsmaybeachieved,especiallywhentheeavesdropperhasmuchhigherSNRthanthereceivers TEKINANDYENER:GGMAC-WTANDGTW-WT:ACHIEVABLERATESANDCOOPERATIVEJAMMING2739 .Ourachievableratescannotguaranteese-crecyforsuchagroupofusers.III.ACHIEVABLEATEA.GGMAC-WTInthissection,wepresentourmainresultsfortheGGMAC-WT.Werstdenetwoseparateregionsandthengiveanachievableregion.nition3(GGMAC-WTSuperpositionRegion): forall .Then,thesuperpositionregion isgivenby whichcanbewrittenas nition4(GGMAC-WTTDMARegion): besuch forall and .Let forall .Then,theTDMAregion isgivenby whichisequivalentto Remark1:ThesuperpositionandTDMAregionscanalsobewrittenasfollows: (23) inaccordancewiththedenitionsin(14)Theorem1:TherateregiongivenbelowisachievablefortheGGMAC-WT convexclosureof Proof:rstshowthatthesuperpositionencodingrateregiongivenin(20)foraxedpowerallocationisachievable.Considerthefollowingcodingschemeforrates forsome SuperpositionEncodingScheme:Foreachuser ,considerthefollowingscheme.1)Generatethreecodebooks , ,and . sistsof codewords,eachcomponentofwhichisdrawnfrom .Codebook has codewordswitheachcomponentrandomlydrawnfrom and has codewordswitheachcomponentrandomlydrawnfrom where isanarbitrarilysmallnumbertoensurethatthepowerconstraintsonthecodewordsaresatisedwithhighprob-abilityand .De and . IEEETRANSACTIONSONINFORMATIONTHEORY,VOL.54,NO.6,JUNE2008B.GTW-WTNow,wewillexaminethepowerallocationthatmaximizesthesecrecysumrategiveninTheorem2as Thisproblemisformallystatedbelow (70)where (71)and yields(70).Theoptimumpowerallocationisstatedbelow.Theorem4:Thesecrecysum-ratemaximizingpoweralloca-tionfortheGTW-WTisgivenby if if Proof:TheLagrangianis EquatingthederivativeoftheLagrangiantozeroforuser weget (74)where AnargumentsimilartotheonefortheGGMAC-WTestab-lishesthatif ,orequiva-lently,if ,then .Whenequalityissatised,then regardlessof ,andassuch canbeseentonotdependon .Toconservepower,weagainset inthiscase.Ontheotherhand,if ,then Consideruser1.If ,and ,thisimpliesthat .Since ,wecannothave .Asaconsequenceofthiscontradiction,weseethat whenever .Assume ,andconsiderthetwoalternatives .Wewillhave if ;and if .Thesecasescorrespondto and ,respectively.Thus,wehave(72)asthesecrecysum-ratemaximizingpowerallocation. Remark3:ObservethatthesolutioninTheorem4hasastructuresimilartothatinTheorem3.Insummary,itisseenthataslongasauserisnotsingle-userdecodable,itshouldbetransmittingwithmaximumpower.Hence,whenbothuserscanbemadetobenonsingle-userdecodable,thenthemaximumpowerswillprovidethelargestsecrecysumrate.Ifthisisnotthecase,thentheuserwhoissingle-userdecodablecannottransmitwithnonzerosecrecyandwilljustmakethesecrecysum-rateconstrainttighterfortheremaininguserbytransmittingopenComparing(72)to(67),weseethatthesameformofsolu-tionsisfound,buttherangeofchannelgainswheretransmissionispossibleislarger,showingthatGTW-WTallowssecrecyevenwhentheeavesdropperschannelisnotveryweak.V.SHROUGHOOPERATIVEInSectionIV,wefoundthesecrecysum-ratemaxi-mizingpowerallocations.ForboththeGGMAC-WTandtheGTW-WT,iftheeavesdropperisnotdisadvantagedforsomeusers,thentheseuserstransmitpowersaresettozero.Wepositthatsuchausermaybeabletotransmittinguser,sinceitcancausemoreharmtotheeaves-dropperthantotheintendedreceiver.Weonlyconsiderthesuperpositionregion,sinceintheTDMAregionauserhasadedicatedtimeslot,andhencedoesnotaffecttheothers.Wewillnextshowthatthistypeofcooperativebehaviorisindeeduseful,notablyexploitingthefactthattheestablishedachievablesecrecysumrateisadifferenceofthesum-capacityexpressionsfortheintendedchannel(s)andtheeavesdropperchannel.Asaresult,reducingthelattermorethantheformeractuallyresultsinanincreaseintheachievablesecrecysumFormally,theschemeweareconsideringimpliespartitioningthesetofusers intoasetoftransmittingusers andasetofjammingusers .Ifauser isjamming,thenit insteadofcodewords.Inthiscase,wecanshowthatwecanachievehighersecrecyrateswhentheweakerusersarejamming.WealsoshowthattheGTW-WThasanadditionaladvantagecomparedtotheGGMAC-WT;thatisthefactthatthereceiveralreadyknowsthejammingsequence.Assuch,thisschemeonlyharmstheeavesdropperandnottheintendedreceivers,achievinganevenhighersecrecysumrate.Onceagain,withoutlossofgenerality,weconsider .Inaddition,wewillassumethatausercaneithertaketheactionoftransmittingitsinformationorjammingtheeaves-dropper,butnotboth.ItisreadilyshowninSectionV-Athatwedonotloseanygeneralitybydoingso,andthatsplittingthepowerofauserbetweenthetwoactionsissuboptimalfromthesecrecysum-ratemaximizationpointofview.A.GGMAC-WTTheproblemisformallypresentedbelow (76) (77) IEEETRANSACTIONSONINFORMATIONTHEORY,VOL.54,NO.6,JUNE2008 Fig.7.GGMAC-WTcooperativejammingsecrecysumrateasafunctionof withdifferent forP =P h .Thecirclesindicateoptimumjammingpower.thantheGGMAC-WT,andoffersmoreprotectiontousers.Inaddition,TDMAdoesnotenlargetheachievablere-gionforGTW-WTsincesuperpositioncodingalwaysallowsuserstoachievetheirsingle-usersecrecyratesforanytransmitpower.Letushaveacloserlookatthesecrecyadvantageofthetwo-waychannelovertheMACwithtwousers.FortheGGMAC-WTwith ,theachievablemaximumsecrecysumrate islimitedbythechannelparameters.Itwasshownin[39]thatforthedegradedcase ,thesecrecysumcapacity isanincreasingfunctionofthetotalsumpower .However,itislimited as .Forthegeneralcase,where ,Theorem3impliesthatthesumrateismaximizedwhenonlyuser1transmits(assuming ),andisboundedsimilarlyby .Ontheotherhand,fortheGTW-WT,unliketheGGMAC-WT,itispossibletoincreasethesecrecycapacitybyincreasingthetransmitpowers.Thismainlystemsfromthefactthattheusersnowhavetheextraadvantageovertheeavesdropperthattheyknowtheirowntransmittedcodewords.Ineffect,eachuserhelpsencrypttheotheruserstransmission.Toseethismoreclearly,considerthesymmetriccasewhere and ,whichmakesallusersreceiveasimilarlynoisyversionofthesamesummessage.Theonlydisad-vantagetheeavesdropperhasisthathedoesnotknowanyofthecodewordswhereasuser knows .Inthiscase, isachievable,andthisrateapproaches as .Thus,itispossibletoachieveasecrecyrateincreaseatthesamerateastheincreaseinchannelcapacity.Next,weexaminethesecrecysum-ratemaximizingpowerallocationsandoptimumpowersforthecooperativejammingscheme.Figs.7and8showtheachievablesecrecyrateimprove-mentforthecooperativejammingschemeforvariouschannelparametersfortheGGMAC-WTwith .Theplotsarethesecrecyratesforuser1whenuser2isjammingwithagiven Fig.8.GGMAC-WTcooperativejammingsecrecysumrateasafunctionof withdifferent forP =P =100 .Thecirclesindicateoptimumjammingpower. Fig.9.GTW-WTcooperativejammingsecrecysumrateasafunctionof withdifferent forP =P h power,whichcorrespondtouser1ssingle-usersecrecy[12],sinceonlyoneuseristransmitting.When thesecrecycapacityisseentobezero,unlessuser2hasenoughpowertoconvertuser1srestandardizedchannelgaintoless .FortheGTW-WT,itisalwaysoptimalforuser2tojamaslongasitenablesuser1totransmit,asseeninFig.9.There-sultsshow,asexpected,thatsecrecyisachievableforbothuserssolongaswecankeeptheeavesdropperfromsingle-userde-codingthetransmittedcodewordsbytreatingtheremaininguserasnoise.Sincethecodingschemesconsideredhereassumeknowledgeofeavesdropperschannelgains,applicationsarelimited.Onepracticalapplicationcouldbesecuringofaphysicallyprotectedareasuchasinsideabuilding,whentheeavesdropperisknowntobeoutside.Insuchacase,wecandesignfortheworstcasescenario.AnexampleisgiveninFig.10fortheGGMAC-WT,whereweassumeasimplepath-lossmodelandxedlocationsfortwotransmitters andonereceiver atthecenter.We IEEETRANSACTIONSONINFORMATIONTHEORY,VOL.54,NO.6,JUNE2008fromthefactthatif ,then forall ,andwemusthave .If,ontheotherhand,(90)givesacomplexornegativesolution,thentheparaboladoesnotintersectthe axis,andisalwayspositive.Hence, ,and doesnotbelongto ,i.e., Theformofthissolutionisintuitivelypleasing,sinceitmakesmoresenseforweakeruserstojamastheyharmtheeaves-droppermorethantheydotheintendedreceiver.Whatweseeisthatalltransmittingusers ,suchthat ,transmitwithmaximumpoweraslongastheirstandardizedchannelgain lessthansomelimit ,andalljammingusersmusthave Weclaimthatallusersin musthave allusersin have .Tomakethisargument,weneedtoshowthata suchthatthereexistssome with and suchthat cannotbetheop-timumset.Toseethis,let betheoptimumpowerallocationforaset .Consideranewpowerallocationandsetsuchthat ,i.e.,user isnowjamming,andlet , , and ,forsomesmall .Wethenhave (91) (92) (93) whichisalowervaluefortheobjectivefunction,provingthat isnotoptimum.Thisshowsthatallusers have forallusers .Sincethelastuserin has ,necessarily forall ,and forall Summarizing,theoptimumpowerallocationissuchthatthereisasetoftransmittingusers with for ,thereisasetofsilentusers andthereisasetofjammingusers with for and isfoundfrom ThisiswhatispresentedinthestatementinTheorem5.Notethatto and ,wecansimplydoanexhaustivesearchaswehavenarrowedthenumberofpossibleoptimalsets insteadof andfoundtheoptimalpowerallocationsforeach. Two-userGGMAC-WT:Forillustrationpurposes,letusconsiderthefamiliarcasewith transmitters.Inthiscase,weknowthateitheruser2jams,ornouserdoes.Thesolutioncanbefoundfromcomparingthetwocases.If,withoutjamming,user2cantransmit,thenitisoptimalforittocon-tinuetotransmit,andjammingwillnotimprovethesumrate.Otherwise,user2maybejammingtoimprovethesecrecyrateofuser1.Theoptimumpowerallocationforuser1isequivalent if and if Thepowerforuser2isfoundfrom(81).Fortwousers,wecansimplywrite(90)as (95)where (96) (97) (98)If ,weautomaticallyhave .Inaddition,wehave ,soweonlyneedtoconcernourselveswiththepossiblypositiveroot .Wendwhen .Weseethat forall if ,equivalenttohavingtwonegativeroots,or ,equivalenttohavingnorealrootsof .Nowexaminewhen Thisispossibleifandonlyif .Since ,thishappensonlywhen or .However,if ,wearebetterofftransmittingthanjamming.Thelastcasetoexamineiswhen .Thisimpliesthat anditissatisedwhen .Assume .Inthiscase,weareguaranteed .If ,thenwemusthave sincethesecrecyrateis .Wewouldliketondwhenwecanhave .Since ,wemusthave ,and .Thisimplies .Itiseasytosee if and Thus,for users,thesolutionsimpliesto(99)shownatthebottomofthepage,where Thissolutioncanbecheckedtobeinaccordancewiththesum-ratemaximizingpowerallocationofTheorem3.Wenotethatinthecaseunaccountedforin(99),i.e.,when and ,bothusersshouldbetransmitting.Ingeneral,thesolutionshowsthattheweakerusershouldjamifitisnot if , if if , if , (99) IEEETRANSACTIONSONINFORMATIONTHEORY,VOL.54,NO.6,JUNE2008 Fig.2.StandardizedGTW-WTsystemmodel.thenoisesarenormalizedby , and C.PreliminaryDeÞnitionsInthissection,wepresentsomeusefulpreliminarydeincludingthesecrecyconstraintwewilluse.Inparticular,thesecrecyconstraintweusedisthecollectivesecrecyconstraintwedenedin[38]and[40],anditissuitableforthemultiaccessnatureofthesystemsofinterest.DeÞnition1(CollectiveSecrecyConstraint):Weusethenor-malizedjointconditionalentropyofthetransmittedmessagesgiventheeavesdroppersreceivedsignalasoursecrecycon-straint,i.e., foranyset ofusers.Forperfectsecrecyofalltransmittedsecretmessages,wewouldlike (7)Assume forsomearbitrarilysmall asrequired. (8) (9) (10) (11)where as .If ,thenwede .Thus,theperfectsecrecyofthesystemimpliestheperfectsecrecyofanygroupofusers,guaranteeingthatwhenthesystemissecure,soiseachindividualuser.DeÞnition2(AchievableRates): .Theratevector issaidtobeachievableifforanygiven thereexistsacodeofsufcientlength suchthat (12a) (12b)and sent istheaverageprobabilityoferror.Inaddition,weneed (12d)where denotesoursecrecyconstraintandisdenedin(7).Wewillcallthesetofallachievablerates,thesecrecy-capacityregion,anddenoteit fortheGGMAC-WT,and theGTW-WT,respectively.Beforewestateourresults,wealsodenethefollowingno-tation,whichwillbeusedextensivelyintherestofthispaper: (13) (14) (15) (16) (17) Last,weinformallycallthe thuserstrong ,and .Thisisawayofindicatingwhetherthein-tendedreceiverorthewiretapperisatamoreofanadvan-tageconcerningthatuser,andisequivalenttostatingwhetherthesingle-usersecrecycapacityofthatuserispositiveorzero.Welaterextendthisconcepttorefertouserswhocanachievepositivesecrecyratesandthosewhocannot.Inaddition,wewillsaythatauserissingle-userdecodableifitsrateissuchthatitcanbedecodedbytreatingtheotheruserasnoise.Ausergroup issingle-userdecodablebytheeavesdropperif IEEETRANSACTIONSONINFORMATIONTHEORY,VOL.54,NO.6,JUNE2008TheGeneralGaussianMultiple-AccessandTwo-WayWiretapChannels:AchievableRatesandCooperativeJammingEnderTekin,StudentMember,IEEE,andAylinYener,Member,IEEEThegeneralGaussianmultiple-accesswiretapchannel(GGMAC-WT)andtheGaussiantwo-waywiretapchannel(GTW-WT)areconsidered.IntheGGMAC-WT,mul-tipleuserscommunicatewithanintendedreceiverinthepresenceofaneavesdropperwhoreceivestheirsignalsthroughanotherGMAC.IntheGTW-WT,twouserscommunicatewitheachotheroveracommonGaussianchannel,withaneavesdropper TEKINANDYENER:GGMAC-WTANDGTW-WT:ACHIEVABLERATESANDCOOPERATIVEJAMMING2751[13]I.CsiszrandJ.KBroadcastchannelswithcondentialmes-IEEETrans.Inf.Theory,vol.IT-24,no.3,pp.339348,May[14]U.MaurerandS.Wolf,Information-theoretickeyagreement:Fromweaktostrongsecrecyforfree,LectureNotesinComputer.Berlin,Germany:Springer-Verlag,2000,vol.1807,pp.[15]U.Maurer,SecretkeyagreementbypublicdiscussionfromcommonIEEETrans.Inf.Theory,vol.39,no.3,pp.733742,May[16]C.H.Bennett,G.Brassard,C.Crpeau,andU.Maurer,izedprivacyampliIEEETrans.Inf.Theory,vol.41,no.6,pp.19151923,Nov.1995.[17]R.AhslwedeandI.CsiszCommonrandomnessininformationtheoryandcryptographyPartI:Secretsharing,IEEETrans.Inf.,vol.39,no.4,pp.11211132,Jul.1993.[18]R.AhslwedeandI.CsiszCommonrandomnessininformationtheoryandcryptographyPartII:CRcapacity,IEEETrans.Inf.,vol.44,no.1,pp.225240,Jan.1998.[19]U.MaurerandS.Wolf,Secret-keyagreementoverunauthenticatedpublicchannelsPartI:Denitionsandacompletenessresult,Trans.Inf.Theory,vol.49,no.4,pp.822831,Apr.2003.[20]U.MaurerandS.Wolf,Secret-keyagreementoverunauthenticatedpublicchannelsPartII:Thesimulatabilitycondition,IEEETrans.Inform.Theory,vol.49,no.4,pp.832838,Apr.2003.[21]U.MaurerandS.Wolf,Secret-keyagreementoverunauthenticatedpublicchannelsPartIII:PrivacyampliIEEETrans.Inf.,vol.49,no.4,pp.839851,Apr.2003.[22]S.VenkatesanandV.Anantharam,Thecommonrandomnesscapacityofapairofindependentdiscretememorylesschannels,IEEETrans.Inf.Theory,vol.44,no.1,pp.215224,Jan.1998.[23]S.VenkatesanandV.Anantharam,Thecommonrandomnessca-pacityofanetworkofdiscretememorylesschannels,IEEETrans.Inf.Theory,vol.46,no.2,pp.367387,Mar.2000.[24]I.CsiszrandP.Narayan,Commonrandomnessandsecretkeygen-erationwithahelper,IEEETrans.Inf.Theory,vol.46,no.2,pp.366,Mar.2000.[25]I.CsiszrandP.Narayan,Secrecycapacitiesformultipleterminals,IEEETrans.Inf.Theory,vol.50,no.12,pp.30473061,Dec.2004.[26]H.Yamamoto,Onsecretsharingcommunicationsystemswithtwoorthreechannels,IEEETrans.Inf.Theory,vol.IT-32,no.3,pp.393,May1986.[27]H.Yamamoto,AcodingtheoremforsecretsharingcommunicationsystemswithtwoGaussianwiretapchannels,IEEETrans.Inf.Theoryvol.37,no.3,pp.634638,May1991.[28]Y.Oohama,Codingforrelaychannelswithcondentialmessages,Proc.IEEEInf.TheoryWorkshop,2001,pp.87[29]J.BarrosandM.R.D.Rodrigues,Secrecycapacityofwirelesschan-Proc.IEEEInt.Symp.Inf.Theory,Seattle,WA,Jul.92006,pp.356[30]M.Bloch,J.Barros,M.R.D.Rodrigues,andS.W.McLaughlin,Wirelessinformation-theoreticsecurityPartI:Theoreticalaspects,IEEETrans.Inf.Theory,acceptedforpublication.[31]Y.LiangandV.Poor,Securecommunicationoverfadingchannels,Proc.AllertonConf.Commun.ControlComput.,Monticello,IL,Sep.2729,2006.[32]L.Zang,R.Yates,andW.Trappe,Secrecycapacityofindependentparallelchannels,Proc.AllertonConf.Commun.ControlComput.Monticello,IL,Sep.2729,2006.[33]R.Liu,I.Maric,R.D.Yates,andP.Spasojevic,Discretememorylessinterferenceandbroadcastchannelswithcondentialmessages,Proc.AllertonConf.Commun.ControlComput.,Monticello,IL,Sep.29,2006.[34]Y.LiangandV.Poor,Generalizedmultipleaccesschannelswithcon-dentialmessages,IEEETrans.Inf.Theory,acceptedforpublication.[35]R.Liu,I.Maric,R.D.Yates,andP.Spasojevic,Thediscretemem-orylessmultipleaccesschannelwithcondentialmessages,Proc.IEEEInt.Symp.Inf.Theory,Seattle,WA,Jul.914,2006,pp.957[36]A.KhistiandG.W.Wornell,Securetransmissionwithmultiplean-tennas:TheMISOMEwiretapchannel,IEEETrans.Inf.Theory,sub-mittedforpublication.[37]S.Shaee,N.Liu,andS.Ulukus,TowardsthesecrecycapacityoftheGaussianMIMOwire-tapchannel:The2-2-1channel,IEEETrans.Inf.Theory,submittedforpublication.[38]E.Tekin,S.Serbetli,andA.Yener,OnsecuresignalingfortheGaussianmultipleaccesswire-tapchannel,Proc.AsilomarConf.SignalSyst.Comput.,Asilomar,CA,Nov.1,2005,pp.1747[39]E.TekinandA.Yener,TheGaussianmultiple-accesswire-tapchannelwithcollectivesecrecyconstraints,Proc.IEEEInt.Symp.Inf.Theory,Seattle,WA,Jul.914,2006,pp.1164[40]E.TekinandA.Yener,TheGaussianmultiple-accesswire-tapIEEETrans.Inf.Theory[Online].Available:http://arxiv.org/format/cs.IT/0605028,submittedforpublication[41]E.TekinandA.Yener,AchievableratesforthegeneralGaussianmul-tipleaccesswire-tapchannelwithcollectivesecrecy,Proc.AllertonConf.Commun.ControlComput.,Monticello,IL,Sep.2729,2006.[42]E.TekinandA.Yener,Achievableratesfortwo-waywire-tapchan-Proc.IEEEInt.Symp.Inf.Theory,Nice,France,Jun.24[43]A.B.Carleial,Interferencechannels,IEEETrans.Inf.Theory,vol.IT-24,no.1,pp.6070,Jan.1978.[44]T.M.CoverandJ.A.Thomas,ElementsofInformationTheory.NewYork:Wiley,1991.[45]E.TekinandA.Yener,Secrecysum-ratesforthemultiple-accesswire-tapchannelwithergodicblockfading,Proc.AllertonConf.Commun.ControlComput.,Monticello,IL,Sep.2628,2007.[46]A.Thangaraj,S.Dihidar,A.R.Calderbank,S.McLaughlin,andJ.-M.ApplicationsofLDPCcodestothewiretapchannel,Trans.Inf.Theory,vol.53,no.8,pp.29332945,Aug.2007.[47]M.Bloch,J.Barros,M.R.D.Rodrigues,andS.W.McLaughlin,Wirelessinformation-theoreticsecurityPartII:Practicalimple-IEEETrans.Inf.Theory,acceptedforpublication. IEEETRANSACTIONSONINFORMATIONTHEORY,VOL.54,NO.6,JUNE2008 Fig.11.GTW-WTcooperativejammingexampledarkershadescorrespondtohighervalues.andnormalsecretcommunicationsisnotpossible.ThegainscanbesignicantforboththeGGMAC-WTandtheGTW-WT.Thiscooperativebehaviorisusefulwhenthemaximumsecrecysumrateisofinterest.Wehavealsocontrastedthesecrecyratesofthetwochannelsweconsidered,notingthebenetofthetwo-waychannelswherethefactthateachreceiverhasperfectknowledgeofitstransmittedsignalbringsanadvantagewitheachusereffectivelyencryptingthecommunicationsoftheotheruser.Inthispaper,weonlypresentedachievablesecrecyratesfortheGGMAC-WTandtheGTW-WT.Thesecrecycapacityregionforthesechannelsarestillopenproblems.In[45],wealsofoundanupperboundforthesecrecysumrateoftheGGMAC-WTandnotedthattheachievablesecrecysumrateandtheupperboundwefoundonlycoincideforthedegradedcase,sothatwehavethesecrecysumcapacityforthedegradedGMAC-WT.Eventhoughthereisagapbetweentheachievablesecrecysumratesandupperbounds,cooperativejammingwasshownin[45]togiveasecrecysumratethatisclosetotheupperboundingeneral.Finally,wenotethattheresultsprovidedareofmainlythe-oreticalinterest,sinceasofyettherearenocurrentlyknownpracticalcodesformultiple-accesswiretapchannelsunlikethesingle-usercasewhereinsomecasespracticalcodeshavebeenshowntobeusefulforthewiretapchannel[46],[47].Further-more,accurateestimatesoftheeavesdropperchannelparame-tersarerequiredforcodedesignforwiretapchannelswherethechannelmodelisquasi-static,asinourmodelsconsideredinthispaper.[1]R.Ahlswede,Multi-waycommunicationchannels,Proc.2nd.Int.Symp.Inf.Theory,Tsahkadsor,ArmenianS.S.R.,1971,pp.23[2]H.Liao,Multipleaccesschannels,Ph.D.dissertation,Dept.Electr.Engr.,Univ.Hawaii,Honolulu,HI,1972.[3]C.E.Shannon,Two-waycommunicationchannels,Proc.4thBerkeleySymp.Math.Statist.Probab.,1961,vol.1,pp.611[4]G.Dueck,Thecapacityregionofthetwo-waychannelcanexceedtheinnerbound,Inf.Control,no.40,pp.258266,1979.[5]T.S.Han,Ageneralcodingschemeforthetwo-waychannel,Trans.Inf.Theory,vol.IT-30,no.1,pp.3544,Jan.1984.[6]R.Ahlswede,Thecapacityregionofachannelwithtwosendersandtworeceivers,Ann.Probab.,vol.2,no.5,pp.805814,1974.[7]H.Sato,Two-usercommunicationchannels,IEEETrans.Inf.Theoryvol.IT-23,no.3,pp.295304,May1977.[8]A.ElGamalandT.Cover,Multipleuserinformationtheory,Proc.,vol.68,no.12,pp.14661483,Dec.1980.[9]C.E.Shannon,Communicationtheoryofsecrecysystems,BellSyst.Tech.J.,vol.28,pp.656715,1949.[10]A.Wyner,Thewire-tapchannel,BellSyst.Tech.J.,vol.54,pp.1387,1975.[11]A.B.CarleialandM.E.Hellman,AnoteonWynerswiretapIEEETrans.Inf.Theory,vol.IT-23,no.3,pp.387May1977.[12]S.K.Leung-Yan-CheongandM.E.Hellman,Gaussianwire-tapIEEETrans.Inf.Theory,vol.IT-24,no.4,pp.451456,Jul.