Dynamic Resource Sharing in 5G with LSA: Criteria-Based Management Framework

Resea ch A icle Dynamic Resou ce Sha ing in 5G wi h LSA: C i e ia-Based Managemen F amewo k Zhaleh Sad eddini ,1Pa el Masek ,2,3 Tug ul Ca da ,1 Aleksand Ome o ,4Ji i Hosek ,2,3 I ina Gudko a,3,5 and Se gey And ee 4 1Depa men o Compu e Enginee ing, Ka adeniz Technical Uni e si y, T abzon, Tu key 2Depa men o Telecommunica ions, B no Uni e si y o Technology, B no, Czech Republic 3Applied P obabili y and In o ma ics Depa men , Peoples’ F iendship Uni e si y o Russia (RUDN Uni e si y), Moscow, Russia 4Labo a o y o Elec onics and Communica ions Enginee ing, Tampe e Uni e si y o Technology, Tampe e, Finland 5Fede al Resea ch Cen e “Compu e Science and Con ol” o he Russian Academy o Sciences, Moscow, Russia Co espondence should be add essed o Pa el Masek; masekpa[email p o ec ed] .cz Recei ed 27 July 2017; Re ised 18 Janua y 2018; Accep ed 5 Ma ch 2018; Published 24 May 2018 Academic Edi o : Osca Espa za Copy igh © 2018 Zhaleh Sad eddini e al. This is an open access a icle dis ibu ed unde he C ea i e Commons A ibu ion License, which pe mi s un es ic ed use, dis ibu ion, and ep oduc ion in any medium, p o ided he o iginal wo k is p ope ly ci ed. Owing o a s eadily inc easing demand o e icien spec um u iliza ion as pa o he i h-gene a ion (5G) cellula concep , i becomes c ucial o e ise he exis ing adio spec um managemen echniques and p o ide mo e lexible solu ions o he co esponding challenges. A new wa e o spec um policy e o ms can hus be en isaged by p oducing a pa adigm shi om s a ic o dynamic o ches a ion o sha ed esou ces. The eme ging Licensed Sha ed Access (LSA) egula o y amewo k enables lexible spec um sha ing be ween a limi ed numbe o use s ha access he same equency bands, while gua an eeing be e in e e ence mi iga ion. In his wo k, an ad anced use sa is ac ion-awa e spec um managemen s a egy o dynamic LSA managemen in 5G ne wo ks is p oposed o balance bo h he connec ed use sa is ac ion and he Mobile Ne wo k Ope a o (MNO) esou ce u iliza ion. The app oach is based on he MNO decision policy ha combines bo h p icing and ejec ion ules in he implemen ed p ocesses. Ou s udy o e s a classi ica ion buil o e se e al ypes o use s, di e en co esponding a ibu es, and a numbe o MNO’s decision scena ios. Ou in es iga ions a e buil on C i e ia-Based Resou ce Managemen (CBRM) amewo k, which has been speci ically designed o acili a e dynamic LSA managemen in 5G mobile ne wo ks. To e i y he p oposed model, he esul s (spec um u iliza ion, es ima ed Seconda y Use p ice o he u u e connec ion, and use selec ion me hodology in case o use ejec ion p ocess) a e alida ed nume ically as we yield impo an conclusions on he applicabili y o ou app oach, which may o e aluable guidelines o e icien adio spec um managemen in highly dynamic and he e ogeneous 5G en i onmen s. 1. In oduc ion 1.1. Backg ound on Spec um Sca ci y and Sha ing. The emendous g ow h in mobile da a olume has become a key d i e o he de elopmen o eme ging i h-gene a ion (5G) cellula sys ems. As i is o ecas ed by Cisco [1], he equi ed capaci y will be challenged by he g owing u iliza ion o mobile de ices (e.g., sma phones) ha access a di e se se o se ices and applica ions in o de o manage inc easing amoun s o da a a ic. Following his end, ce ain esea ch and de elopmen conside a ions on he nex -gene a ion wi eless sys ems migh be ega ded in he ligh o s eady inc ease o he o e all demand o adio spec um as well as he deploymen o a a ie y o echnologies ha compe e o u ilize he sha ed equency bands [2]. The adi ional dis inc ion be ween di e en ca ego ies o applica ions, such as da a se ices, oice communica ions, and b oadcas ing, becomes blu ed due o con e gence in unc ionali y o hese se ices deli e ed by new echnologies [3, 4]. A he same ime, adio spec um has ans o med in o a c i ical esou ce om economic, cul u al, and socie al pe spec i es [5, 6]. I s sca ci y has become a majo limi ing ac o in pa icula equency anges, spanning 100 MHz o 6 GHz, wi h desi ed p opaga ion cha ac e is ics o a Hindawi Wi eless Communica ions and Mobile Compu ing Volume 2018, A icle ID 7302025, 12 pages h ps://doi.o g/10.1155/2018/7302025 2 Wi eless Communica ions and Mobile Compu ing wide ange o s a ic spec um use s (e.g., mili a y, ada , TV b oadcas ing, and medical and e en p oduc ion) [7, 8]. Up o now, mobile ope a o s p e e ed o ollow he adi ional exclusi e spec um access egime when o e ing hei se ices. Howe e , his led o u ilizing la ge po ions o adiospec umonlyspo adically,wi hhighspace, ime, and equency a ia ions in usage ha anged om 15% o 85% [9]. As he needs o ex a spec um ha e been g owing o e he pas decades, al e na i e solu ions we e p oposed, especially o exploi ing small agmen s o spec um ha a e seldom used by hei license holde s. Wi h espec o spec um managemen , se e al concep s ha e been de eloped and deployed o e he yea s [10, 11]. Acco dingly, he bene i s o spec um sha ing a e wo old. Fi s , i allows imp o emen s in spec um u iliza ion and second i can p o ide addi ional capaci y o he use s who equi e mo e spec um o hei se ices. A wide ange o spec um sha ing schemes can acili a e he u iliza ion o di e en equency bands (bo h licensed and unlicensed), as adop ed by a ious ca ie -g ade se ice p o ide s. The deploymen o spec um sha ing schemes is subjec o mee - ing a se o p ede ined egula ions and equi emen s. I can also in ol e dissimila coo dina ion echniques. Ac i ely conside ed by elecom ope a o s, dynamic spec um sha ing can enable in e echnology coexis ence in licensed bands [12– 14]. Cu en ly, he e iciency o spec um sha ing mechanisms emains limi ed by p ese p io i y cons ain s o he p ima y license holde . He e, he b oade concep o spec um sha ing e e s o equal spec um access igh s bu may o e look economical aspec s and use sa is ac ion conside a ions [15]. 1.2. Spec um Sha ing wi h Licensed Sha ed Access. To o e - come he p essing e ec s o spec um agmen a ion, a demand o no el amewo ks allowing o e icien sha - ing o a ailable bu unde u ilized spec um eme ges. A no el Licensed Sha ed Access (LSA) [16, 17] concep enables ad anced spec um sha ing be ween a limi ed numbe o end-use s wi h a leas wo en i ies: he incumben (i.e., he cu en holde o he spec um usage igh s) and he LSA licensee (i.e., he empo a y use o spec um, which could be a Mobile Ne wo k Ope a o , MNO) [18]. Mo e speci ically, he LSA amewo k [19] pe mi s o con olled spec um sha ing be ween a P ima y Use (PU) and a Seconda y Use (SU) bo h ha ing access o a po ion o spec um a a gi en loca ion and ime [20, 21], as i is demons a ed in Figu e 1. The co esponding LSA sha ing ag eemen s need o gua an ee high uni o mi y in e ms o spec um access o all in ol ed pa icipan s. To his end, sa is ying end-use se ice equi emen s is a majo goal in he mobile ope a o ’s sys em design. The MNOs should maximize he ac i e SUs (ASUs) sa is ac ion by shi ing he emphasis om Quali y o Se ice (QoS) o Quali y o Expe ience (QoE). The e o e, decisions made by he MNOs play an impo an ole in he discussed ecosys em and conside a ion o app op ia e pa ame e s a ec s he o e all ne wo k pe o mance. To his end, he ne wo k pa ame e s and use a ibu es should be e alua ed join ly o he op imized MNO decision-making [22, 23], which becomes he ocus o his wo k. Incumben use Mac o base s a ion LSA licensee LSA con olle Small cell Incumben use Mac o base s a ion Figu e 1: Scena io o LSA amewo k ope a ion. The es o he pape is o ganized as ollows. Sec ion 2 p o ides an o e iew o exis ing solu ions. Nex , Sec ion 3 desc ibes he p oposed sys em model o he use expe ience- awa e spec um managemen oge he wi h de eloped Deci- sionBasedRejec ionPolicy algo i hm. Fu he , in Sec ion 4, he selec ed use-cases a e discussed. In Sec ion 5, he ob ained simula ion esul s a e desc ibed. Finally, he lessons lea ned andconclusionsa esumma izedinSec ion6.See“Lis o F equen ly Used Ac onyms” in Nomencla u e. 2. Rela ed Wo k and Ou Focus The eme ging LSA amewo k equi es MNOs o de elop e icien spec um managemen s a egies in o de o inc ease he ne wo k capaci y and p o ide high use expe ience le els o he PUs and SUs. As a esul , highe MNO e enues a e expec ed o be eached. As i has been al eady demons a ed in [24], decision-making is one o he main unc ions o he adio spec um managemen p ocess allowing he MNOs o de e mine he adio pa ame e s ha con ol he spec um u iliza ion. In [25], dynamic billing and p icing s a egy is in oduced o he LTE ne wo ks. The calcula ed esou ce cos a ies acco ding o he use ca ego y and conges ion h eshold speci ied by he MNO. The conside ed pa ame e o he end-use s is p io i y class (e.g., gold, sil e , o b onze). In [24], he au ho s p oposed an Adap i e Decision-making Op imiza ion Scheme (ADMS) o Cogni i e Radio Ne wo k (CRN) wi h mul iple subca ie s. All o he a iables in he i ness unc ion a e ela ed o he ne wo k pa ame e s. Besides ha , he au ho s conside ed he weigh o pa ame e s in he i ness unc ion being adap i e. Howe e , he weigh adap i e unc ion is no de ined in addi ion o he use o ixed weigh s in he simula ion esul s. The wo k in [26] p o ides a connec i i y managemen pla o m o e icien Wi-Fi o loading in he e ogeneous ne wo ks. Technique o O de o P e e ence by Simila i y o Ideal Solu ion (TOPSIS) wi h he conside a ion o wo a ibu es, Recei ed Signal S eng h Indica o (RSSI) and numbe o connec ed use s, was applied. The spec um u ili y model was de eloped in [27] u ilizing he mul ia ibu e u ili y heo y o o m he use ’s decision model. The co e- sponding model allows a use o balance he a ibu es, such Wi eless Communica ions and Mobile Compu ing 3 as channel capaci y, mone a y cos , and in e e ence po en- ial. Fu he , in [28], he impac o end-use decision-making on p icing and managemen o adio esou ces in CRN was s udied. In his wo k, C i e ia-Based Resou ce Managemen (CBRM) amewo k o LSA is p esen ed as a ade-o be ween he MNO’s decision on ne wo k managemen and he QoE o use s. In ou p e ious esea ch [29], Ins an O e booking F amewo k o Cogni i e Radio (IOFCR) was o mula ed o enable e icien managemen o spec um in CRN. Th ee di e en p icing policies o he booking pe iod ha e been conside ed. In his s udy, we implemen CBRM wi h espec o hec ucialuse andne wo ka ibu esinLSA. We u ilize c i e ia-based p icing and ejec ion policies, whe e he weigh s o all he pa ame e s a e upda ed pe iodically and au oma ically based on he MNO’s decision and ollowing he gi en QoS equi emen s as well as he spec um usage a io. The use sa is ac ion can be maximized by manipula ing esou ce alloca ion ia adap i e p icing and ejec ion policies basedon heQoE. In mo e de ail, i he MNO is no able o p o ide ASUs wi h adequa e QoE, he ejec ed his o y a io o he ejec ed ASUs (RASUs) will be high. The ejec ed his o y a io as a Compensa ion Cos (CC) has s ong impac on he p icing policy and i could con ince RASUs o eques spec um esou ces om he ne wo k as RSUs in he nea es u u e. Following ou p e ious wo k [30], we u ilize wo se e pool sys ems: hey ca ego ize RSUs based on he QoE ia he ejec ed his o y pa ame e s. Hence, we s udy use - cen ic QoE-based esou ce alloca ion p oblem wi h se e al MNO’s decisions, se e al use c i e ia, and di e en classes o se ices. Ou simula ion esul s indica e he spec um si ua- ion in o -peak and peak hou s. 3. Sys em Model In his sec ion, he To al Ne wo k Re enue aspec is de ailed h ough (i) esou ce block leasing, (ii) compensa ion paid o he ejec ed ASUs, (iii) spec um u iliza ion, (i ) numbe o accep ed and denied RSUs, ( ) numbe o ejec ed and inished ASUs (RASUs/FASUs), and ( i) a e age se ice ime o bo h RASUs and FASUs. To p o ide a comp ehensi e unde s anding o he p o- posed CBRM amewo k o LSA, we con inue his sec ion by de ailing he de eloped model. A en ion is speci ically paid o se e al ypes o cus ome s, di e en use a ibu es, and se e al MNO decision use-cases, while highligh ing he impo an simula ion esul s and discussing he s a egies o each highe QoE le els o ASUs. A de ailed desc ip ion o he p oposed sys em model in his sec ion is di ided in o 3 subsec ions. (i) Ne wo k A chi ec u e and Scena io: in his subsec ion, he en i ies o he p oposed LSA sys em and he in e ac ions be ween hem a e in oduced. (ii) U ilized A ibu es:in hispa ,wi h espec o he impac o decision-making on p icing and managemen o URB in he CBRM amewo k o LSA, impo an used a ibu es o he p oposed sys em model a e de ined. A e u ilizing he a ibu es o SUs, he o al numbe o RSUs, ASUs, FASUs, and RASUs can be shown Unde u ilized LSA band Incumben Alloca ed LSA license Compensa ion LSA eques LSA license Compensa ion Regula o LSA licensee S a us Condi ions Rules PU1 PU2PUn x３５1 x３５2x３５m Figu e 2: P oposed sys em model o LSA ope a ion. in one lis as 𝐿𝑥SU(𝑡𝑖).A imeins an 𝑡𝑖, depending on he URB numbe , MNO is aced wi h he ou main use-cases, which a e desc ibed in his subsec ion. (iii) C i e ia-Based Rejec ion Model:incaseo 0≤URB(𝑡𝑖) < |LPU(𝑡𝑖)|, MNO should ejec he app op ia e ASUs. In his subsec ion, DecisionBasedRejec ionPolicy is implemen ed by AHP a he side o he LSA licensee o ejec he app op ia e ASUs. Finally, heCCiscalcula ed o he ejec edASUs.Ino de o ob ain highe ne wo k e enue, while mo i a ing use s o eques he spec um esou ces mo e equen ly, he expec ed p ice o he ejec edASUsiscalcula edbasedon hede ined RR. 3.1. Ne wo k A chi ec u e and Scena io. In o de o ep esen he eal-wo ld use-cases, we u ilize LTE-Ad anced (Rel. 10) ne wo k in as uc u e wi h he implemen ed LSA ame- wo k, as de ailed in ou p e ious wo k [30]. The in e ac ion be ween he co esponding en i ies o he LSA sys em is depic ed in Figu e 2. The conside ed ne wo k a chi ec u e consis s o he ollowing building blocks: (i) Incumben : LSA is he con olled sha ed access amewo k based on an exclusi e egime o adio spec um sha ing among he incumben s, ha is, PUs, ha ha e he igh o comme cially exploi a gi en po ion o wi eless spec um (cellula band). (ii) LSA licensee: licensed use s ha lease he incum- ben ’s spec um band, which can po en ially be used when a pe mission is g an ed. The leased esou ces (i.e., esou ce blocks) a e u he o e ed o he LSA Licensee’s SUs unde he con ol o he en i y espon- sible o g an ing pe missions o e he cellula band (i.e., LSA con olle ). In his wo k, we assume ou di e en s a es o SUs (i.e., 𝑥SU1,𝑚): (1) Reques ing Seconda y Use s (RSUs): a ime ins an 𝑡𝑖, he ea eRSUs ha sendspec um a ailabili y eques s o he LSA licensees (mod- eling is de ailed in ou p e ious pape [29]). The LSA con olle moni o s he sha ing ac i i ies in hecellula ne wo kand heco esponding 4 Wi eless Communica ions and Mobile Compu ing decisions (accep o ejec he RSUs) a e made basedonspec uma ailabili yp o idedby incumben s o he LSA con olle . (2) Ac i e Seconda y Use s (ASUs): once he RSUs ob ain spec um esou ces (alloca ed esou ce blocks) om he LSA con olle , hey should deli e a compensa ion ee based on he eques ed esou ces. Then, he LSA licensee will lease he unde u ilized LSA spec um esou ces o he RSUs. A e he leasing p ocedu e is inished success ully, he RSUs o mally change hei s a us o ASUs. (3) Finished ASUs (FASUs): a each ime ins an 𝑡𝑖, some o he ASUs inish hei ope a ions and elease he alloca ed spec um esou ces back o he incumben . (4) Rejec ed ASUs (RASUs): in case he incumben eques s he leased spec um esou ces back, he ASUs a e o ced o aca e he cellula band immedia ely, e en i hei ac i e ansmission is no inished. (iii) Regula o : i is he en i y esponsible o g an ing pe missionso e heincumben ’scellula band.I ha monizes spec um usage and opens a pa h o spec um op imiza ion ia con olled sha ing, which is he key esponsibili y o he Na ional Regula o y Au ho i y (NRA). 3.2. U ilized A ibu es. The main u ilized a ibu es in he conside ed sys em model a e linked wi h RSUs and ASUs, since hose wos a eso heSUsin heleasingp ocessa e c ucial o ou CBRM amewo k.Thea ibu eso RSUs include (i) eques ed bi a e (BR) and (ii) ejec ed his o y (RH).Themain eal- imea ibu es o heASUsa elis ed as ollows: (i) Recei ed Signal S eng h Indica o (RSSI): an indica- ion o he powe le el being ecei ed by he ASUs (ii) Usage pa e n (A ST): a e age se ice ime calcu- la edbyu ilizing heASU’sse ices a ed ime(SST) and he cu en ime ins an 𝑡𝑖 (iii) Ac i i y ac o (DUR): he a e age download/upload a ioo eachASU,calcula ed om heASU’sse ice s a ed ime up o he cu en ime ins an (i ) Mobili y pa e n (MP): mo emen di ec ion o each ASU wi hin he cell co e age whe e “1” ep esen s di ec ion owa ds he eNodeB; “2” s ands o he opposi e di ec ion [30]. The SUs send a eques as RSUs; 𝐿RSU(𝑡𝑖)is u he upda edacco ding o he ollowing: 𝐿RSU (𝑡𝑖) ={RSU𝑗=(BR𝑗,RH𝑗)|1≤𝑗≤𝛼(𝑡𝑖),1≤𝑖≤𝑇}, (1) whe e 𝑇is he o al ime in e al, 𝑡𝑖 ep esen s he 𝑖 h ime ins an , and 𝛼(𝑡𝑖)shows he numbe o RSUs modeled as a Poisson dis ibu ion a ime ins an 𝑡𝑖, ha is,RSU ShowRa e. The lis o ASUs, 𝐿ASU(𝑡𝑖),iscalcula edas ollows: 𝐿ASU (𝑡𝑖)={ASU𝑗|0≤𝑗≤󵄨󵄨󵄨󵄨𝐿ASU (𝑡𝑖−1)󵄨󵄨󵄨󵄨−󵄨󵄨󵄨󵄨𝐿FASU (𝑡𝑖)󵄨󵄨󵄨󵄨 −󵄨󵄨󵄨󵄨𝐿RASU (𝑡𝑖)󵄨󵄨󵄨󵄨+󵄨󵄨󵄨󵄨𝐿ASU (𝑡𝑖)󵄨󵄨󵄨󵄨,1≤𝑖≤𝑇}. (2) The numbe o FASUs in 𝐿FASU (𝑡𝑖)isassumed obe uppe -bounded by he numbe o ASUs om p eceding ime in e al. Nex , he lis o FASUs is de i ed as ollows: 𝐿FASU (𝑡𝑖) ={FASU𝑗|0≤𝑗≤󵄨󵄨󵄨󵄨𝐿ASU (𝑡𝑖−1)󵄨󵄨󵄨󵄨,1≤𝑖≤𝑇}. (3) Fu he , he numbe o RASUs in 𝐿RASU(𝑡𝑖)is also uppe - bounded by he numbe o sensed PUs in he cu en ime in e al 𝑡𝑖 ha exceeds he numbe o URB(𝑡𝑖): 𝐿RASU (𝑡𝑖)={RASU𝑗|0≤𝑗≤󵄨󵄨󵄨󵄨𝐿PU (𝑡𝑖)󵄨󵄨󵄨󵄨 −URB (𝑡𝑖),1≤𝑖≤𝑇}, (4) whe e URB(𝑡𝑖)s ands o he numbe o unde u ilized blocks (URBs) and 𝐿PU(𝑡𝑖)is he lis o PUs de i ed om [29]. I is assumed ha a PU is cha ac e ized by an on- a e, which is deno ed by 𝛽.I is hep obabili yo PU’sac i i ype ime in e al (PUUsageRa e). In [31], i is s a ed ha he pa ame e 𝛽 a ies be ween 15% and 85%. In ou scena io, we conside ha 𝛽 ollows a binomial dis ibu ion and 𝛽(𝑡𝑖)deno es he on- a e o a PU a ime ins an 𝑡𝑖wi h 0≤𝛽(𝑡𝑖)≤1. A each ime in e al, he size o 𝐿PU(𝑡𝑖+1)is upda ed wi h espec o he o al numbe o esou ce blocks 𝜑and 𝛽(𝑡𝑖)as ollows: 𝐿PU (𝑡𝑖)=𝜑×𝛽(𝑡𝑖). (5) Following he abo e, he o al numbe o RSUs, ASUs, FASUs, and RASUs can be calcula ed as ollows: 𝐿𝑥SU (𝑡𝑖)={𝑥SU𝑚|0≤𝑚≤𝜙,1≤𝑖≤𝑇}, (6) whe e 𝑇is he o al ime in e al and 𝑡𝑖 ep esen s he 𝑖 h ime ins an and 𝜙co esponds o he o al numbe o 𝑥SU𝑠,by adding henumbe o RSUs,ASUs,FASUs,andRASUs om he ela ed lis a 𝑡𝑖as ollows: 𝜙=󵄨󵄨󵄨󵄨𝐿RSU (𝑡𝑖)󵄨󵄨󵄨󵄨+󵄨󵄨󵄨󵄨𝐿ASU (𝑡𝑖)󵄨󵄨󵄨󵄨+󵄨󵄨󵄨󵄨𝐿FASU (𝑡𝑖)󵄨󵄨󵄨󵄨 +󵄨󵄨󵄨󵄨𝐿RASU (𝑡𝑖)󵄨󵄨󵄨󵄨.(7) Fo he sake o comple eness, we assume alloca ion o exac lyone esou ceblock ooneuse a agi en imeins an 𝑡𝑖. Finally, a e he URB calcula ion a a gi en ime ins an 𝑡𝑖, ou mainuse-casescanbein oduced[31]: (i) Case 1: |𝐿RSU(𝑡𝑖)|≤URB(𝑡𝑖). (ii) Case 2: 0≤URB(𝑡𝑖)<|𝐿RSU(𝑡𝑖)|. (iii) Case 3: 0≤URB(𝑡𝑖)<|𝐿PU(𝑡𝑖)|. (i ) Case 4: |𝐿PU(𝑡𝑖)|=𝜑. Wi eless Communica ions and Mobile Compu ing 5 Table 1: Pai wise compa ison ma ix o ASU a ibu es. Use a ibu es RSSI A ST DUR MP RSSI 1 1/Θ21 1/Θ31 1/Θ41 A ST Θ21 11/Θ32 1/Θ42 DUR Θ31 Θ32 11/Θ43 MP Θ41 Θ42 Θ43 1 3.3. C i e ia-Based Rejec ion Model. The p oposed CBRM amewo k a ge s he hi d use-case men ioned in he p e ious ex , ha is, 0≤URB(𝑡𝑖)<|𝐿PU(𝑡𝑖)|.Following his condi ion, he LSA licensee mus aca e he leased spec um esou ces, since he LSA con olle has o main ain he a ge QoS and QoE pa ame e s o he incumben s (PUs). Gene ally, he p obabili y o ASU ejec ion inc eases du ing peak hou s. In his wo k, he ejec ion unc- ionali y is modeled o wo scena ios: (i) inc easing ASU’s a ge sa is ac ion, while selec ing he app op ia e ASUs o be ejec ed, and (ii) inc easing he ne wo k e enue p o ided o RASUs, paying he so-called CC, which can be used du ing u he a emp s o alloca e he unde u ilized spec um esou ces, ha is, p io i iza- ion o ea lie ejec ed ASUs. To his end, he unc ion 𝑅𝑒𝑗𝑒𝑐𝑡(𝐿ASUs(𝑡𝑖));𝐷𝑒𝑐𝑖𝑠𝑖𝑜𝑛𝐵𝑎𝑠𝑒𝑑𝑅𝑒𝑗𝑒𝑐𝑡𝑖𝑜𝑛𝑃𝑜𝑙𝑖𝑐𝑦 ep esen s he gene alized o m o ejec ion p ocess men ioned as he hi duse-caseabo e.Asanou pu ,i e u ns𝐿RASU(𝑡𝑖). 𝐷𝑒𝑐𝑖𝑠𝑖𝑜𝑛𝐵𝑎𝑠𝑒𝑑𝑅𝑒𝑗𝑒𝑐𝑡𝑖𝑜𝑛𝑃𝑜𝑙𝑖𝑐𝑦is composed o he ollowing s eps: (1) Iden i ying he decision c i e ion o SUs and hei p io i y (2) No maliza ion p ocess o decision c i e ion (3) Impo ance deg ee alloca ion (p io i y ec o ) (4) E alua ing he decision made (5) Selec ion o he app op ia e ASUs ha will be ejec ed (as s a ed in (10)). The p inciples o 𝐷𝑒𝑐𝑖𝑠𝑖𝑜𝑛𝐵𝑎𝑠𝑒𝑑𝑅𝑒𝑗𝑒𝑐𝑡𝑖𝑜𝑛𝑃𝑜𝑙𝑖𝑐𝑦 algo- i hm, p oposed in his wo k, a e de ailed in Algo i hm 1. A hebeginningo hedecisionp ocess, heMNOcollec sallo he inpu decision c i e ia (i.e., RSSI, A ST, DUR, and MP) om 𝐿ASU.Theob ained alue o eachASUisno malized based on he cos /bene i ype o a pa icula c i e ion (see line (1)in Algo i hm 1) [32]. The e o e, each c i e ion o ASUs has o be compa ed wi h he c i e ia o o he ASUs, conside ing he lis o c i e ia o ASUs. This compa ison leads o s abili y among ASUs, which is he c ucial s ep in he p ocess o inding a ejec ion a io (RR) o each and e e y ASU. Fu he , a nume ical weigh (p io i y) is compu ed o each c i e ion o he ASUs by he LSA licensee; see line (7) in Algo i hm 1. The P io i yVec o shows he impo ance deg ee o each c i e ion as compa ed o o he c i e ia, whe e ∑|C i e ion| 𝑗=1 𝑃𝑟𝑖𝑜𝑟𝑖𝑡𝑦𝑉𝑒𝑐𝑡𝑜𝑟(𝑗)=1.Thedecisionon hep io i y assignmen , om he LSA licensee side, can be adap i e a each 𝑡𝑖acco ding o he peak and he o -peak hou s. In his wo k,weha eimplemen ed heAnaly icHie a chyP ocess (AHP) a he side o he LSA licensee. AHP is one o he mul ic i e ia decision-making me hods ha was conside ed in [33–35]. De eloping a single pai wise compa ison ma ix o he c i e ia is he i s s ep when he AHP is u ilized. Table 1 shows he pai wise compa ison ma ix composed o ou ASU’s a ibu es men ioned in de ail wi hin Sec ion 3.2. The alue Θ𝑖𝑗 ep esen s a se wi h espec o he scale o absolu e judgmen in AHP me hod [22, 32, 34]. As commonly used in AHP, he alues Θ𝑖𝑗 a y be ween 1and 9.The ecip ocal ela ionship be ween he 𝑖 h and 𝑗 h a ibu es is gi en by Θ𝑖𝑗 =1/Θ𝑖𝑗. Nex , he 𝑛 h oo -o -p oduc alues in each ow is calcula ed. The Consis ency Ra io (CR) is u he de i ed as shown in (8), whe e consis ency o he pai wise compa ison om he LSA licensee side is p o ided. CR =Consis ency Index (CI) Random Index (RI).(8) The Consis ency Index (CI) is p oduced as CI =(𝜆max −𝑛) (𝑛−1),(9) whe e 𝑛is henumbe o compa edc i e ion.In hedesc ibed use-case, ou di e en c i e ia a e compa ed. Fu he , 𝜆max shows he sum o he p io i y ec o alues, as men ioned a S ep (8)in Algo i hm 1. Con inuing ou desc ip ion o he p oposed CBRM amewo k, he RI s ands o he di ec unc ion o he numbe o c i e ia aken in o conside a ion [33]. Finally, he CR (see (8)) p o ides he decision-make wi h he in o ma- ion on how consis en was he p ocess o pai wise compa - ison. A e he CR is calcula ed, wo main (and di e en ) condi ionsa e obechecked: (i) CR <0.10:pai wisecompa isonsmadeby heMNOs we e ound obe ela i elyconsis en (ii) CR >0.10: ee alua iono hecomple edpai wise compa ison p ocesses should be aken in o accoun by he MNOs. When hecondi ionis eached, heMNOcanmakei s decision and choose he app op ia e ASUs; hence, we ha e he ollowing: RRASU (𝑖)=|𝐿ASU(𝑡𝑖)| ∑ 𝑖=1 |c i e ion| ∑ 𝑗=1 𝑟(𝑖,𝑗)×𝑃𝑟𝑖𝑜𝑟𝑖𝑡𝑦𝑉𝑒𝑐𝑡𝑜𝑟𝑗,(10) 6 Wi eless Communica ions and Mobile Compu ing (1) o 𝑖=1 o |𝐿ASU(𝑡𝑖)|do (2) o 𝑖=1 o |c i e ion|do (3) 𝑟(𝑖,𝑗)←󳨀No maliza ion(ASU𝑖,C i e ion𝑗) (4) end o (5) end o (6) 𝑆𝑢𝑚𝑅𝑜𝑜𝑡𝑂𝑓𝑃𝑟𝑜𝑑𝑢𝑐𝑡←󳨀(𝑛 ∑ 𝑖=1 𝑛 ∏ 𝑗=1𝑃𝑎𝑖𝑟𝑊𝑖𝑠𝑒𝑀𝑎𝑡𝑟𝑖𝑥(𝑖,𝑗))1/𝑛 (7) 𝑃𝑟𝑖𝑜𝑟𝑖𝑡𝑦𝑉𝑒𝑐𝑡𝑜𝑟𝑗←󳨀 (∏𝑛 𝑗=1𝑃𝑎𝑖𝑟𝑊𝑖𝑠𝑒𝑀𝑎𝑡𝑟𝑖𝑥(𝑖,𝑗))1/𝑛 𝑆𝑢𝑚𝑅𝑜𝑜𝑡𝑂𝑓𝑃𝑟𝑜𝑑𝑢𝑐𝑡 (8) 𝜆max ←∑((𝑛 ∑ 𝑖=1𝑃𝑎𝑖𝑟𝑊𝑖𝑠𝑒𝑀𝑎𝑡𝑟𝑖𝑥(𝑖,𝑗))×𝑃𝑟𝑖𝑜𝑟𝑖𝑡𝑦𝑉𝑒𝑐𝑡𝑜𝑟𝑗) (9) CI ←󳨀 (𝜆max −𝑛) (𝑛−1) (10) CR ←󳨀 CI RI (11) i C >0.1 hen (12) go o (6) (13) else (14) RRASU(𝑖)←󳨀 |𝐿ASU(𝑡𝑖)| ∑ 𝑖=1 |c i e ion| ∑ 𝑗=1 𝑟(𝑖,𝑗)×𝑃𝑟𝑖𝑜𝑟𝑖𝑡𝑦𝑉𝑒𝑐𝑡𝑜𝑟𝑗 (15) CCRASU(𝑡𝑖)=1−RRASU(𝑖) (16) 𝐿󸀠 ASU(𝑡𝑖)←󳨀𝐿ASU(𝑡𝑖)−𝐿RASU(𝑡𝑖) (17) 𝐿󸀠 RASU(𝑡𝑖)←󳨀𝐿RASU(𝑡𝑖−1)+𝐿RASU(𝑡𝑖) (18) end i Algo i hm 1: DecisionBasedRejec ionPolicy algo i hm. whe e |c i e ion|is conside ed as a lis o 4pa ame e s men ionedinSec ion3.2.TheASUswillbe ejec edacco ding o he inc easing o de o RRASU(𝑡𝑖),whe e henumbe o RASUs can be ound as ollows: 󵄨󵄨󵄨󵄨𝐿RASU (𝑖)󵄨󵄨󵄨󵄨=󵄨󵄨󵄨󵄨𝐿ASU (𝑡𝑖)󵄨󵄨󵄨󵄨−(𝜑−󵄨󵄨󵄨󵄨𝐿PU (𝑡𝑖)󵄨󵄨󵄨󵄨).(11) I e e y RASU is willing o eques new spec um esou ces in he u u e as he RSU, he CC a io will be applied o he p ice unc ion as CC𝑖in (13) and hese RSUs will pay a lowe p ice han o he RSUs based on he CC a io. This beha io is connec ed wi h use sa is ac ion in bo h (i) ejec ed and (ii) spec um eques ing s eps. The CC a io is calcula ed as ollows: CCRASU (𝑡𝑖)=1−RRASU (𝑡𝑖). (12) A e assigning he calcula ed CC a io o RASUs a he ime ins an 𝑡𝑖, i should be (i) emo ed om he ASUs lis and (ii) added o he exis ing RASUs lis . Fu he , he lis s o ASUs and RASUs should be upda ed. Recalling Sec ion 3, RSUs will ha e an oppo uni y o lease unde u ilized spec um esou ces (URBs) i and only i hey pay he eques ed esou ce block’s ee. The goal o his app oach is o ob ain highe ne wo k e enue, while mo i a ing use s o eques he unde u ilized spec um mo e equen ly. The cons uc ed p icing unc ion o e s an oppo - uni y o he LSA licensee(s) o each he men ioned goal. The e o e, a each ime ins an 𝑡𝑖(a e he RSUs ecei e an app o al o access he unde -u ilized spec um), he RSUs willbecha ged o he esou cealloca ion.Thep icepaidby he RSUs can be measu ed in mone a y uni s o al e na i ely i canbemeasu edin i ualcu ency[27].Thep icing unc ion is u he calcula ed as ollows: P iceRSU (𝑡𝑖)=𝑃𝑖×(1−CCRSU),(13) whe e 𝑃𝑖is he p ice paid by he RSUs and CC s ands o he compensa ion cos in (12). Fo he in es iga ed p icing unc ion, he e a e wo impo an cases: (i) I CCRSU ==0 hen P iceRSU(𝑡𝑖)=𝑝𝑖. (ii) I CCRSU ==1 hen P iceRSU(𝑡𝑖)=0. The i s case akes place o he RSUs wi hou he ejec ed his o y in he ne wo k up o he cu en ime ins an . The e o e, he de aul p ice will be se o he RSU(s). The second case e e s o he si ua ion whe e he RSU had 100% ejec ion his o y in he cellula ne wo k. E en hough he second case dec eases he ne wo k e enue a he ime ins an 𝑡𝑖, in he long- e m pe spec i e i can be expec ed o obse e epea ed a emp s om he RSUs o use he unde u ilized spec um. Finally, Figu e 3 summa izes he decision low o he CBRM amewo k a ime ins an 𝑡𝑖.Fo hesakeo comple eness, a summa y o he a iables used in ou sys em model is gi en in “Lis o Used Va iables” in Nomencla u e. Wi eless Communica ions and Mobile Compu ing 7 No Yes i Moni o ing PU and xSU ac i i y Checking he LSA band a ailabi li y (URB) Case 3 o 4 Alloca ing he URB Decision based ejec ion unc ionali y (Sec ion 3.3) Upda ing LxSU (equa ion (6)) Figu e 3: Decision low diag am o he sys em model a 𝑡𝑖. Table 2: Simula ion pa ame e s. Pa ame e Value Simula ion ime 100 s Numbe o esou ce blocks 25 RB PUUsageRa e [0.25;0.75]% RSUShowRa e [0.25;0.75]% RSSI [−100;−30]dBm Mobili y pa e n [1;2] Downlink channel bi a e [1; 33] Mbps Uplink/downlink a io [0.01; 1] 4. Simula ion Resul s To adequa ely e alua e he p oposed sys em model o mu- la ed in Sec ion 3, we decided o u ilize as ou simula ion pa ame e s he da a ob ained om he expe imen al 3GPP LTE-A sys em loca ed a he Depa men o Telecommunica- ions, B no Uni e si y o Technology (BUT), Czech Republic. I suppo s unc ionali y o LTE Release 10 communica ions sys em. The lis o employed simula ion pa ame e s in he de eloped amewo kisshowninTable2. In his wo k, we assume PUUsageRa e = 0.25 and RSUShowRa e = 0.25 o he o -peak hou s. Fu he , PUUsageRa e and RSUShowRa e a e se o 0.75 ep esen he peak- hou pa ame e s. Acco ding o he LTE-A sys em design, he RSSI is con igu ed in he ange [−100;−30]dBm and he Table 3: Simula ion esul s o o -peak and peak hou s. Pa ame e Values o o -peak and peak hou O -peak Peak To al numbe o RSUs 622 1878 To al numbe o ASUs 357 (57.39%) 166 (8.86%) To al numbe o FASUs 321 (89.91%) 76 (45.78%) To al numbe o RASUs 17 (4.76%) 82 (49.39%) Numbe o emaining ASUs 19 (5.32%) 8 (4.82%) To al ne wo k e enue 147603 78532 BR o he downlink channel is se in he ange [1; 33] Mbi /s. The mobili y pa e n is deno ed as “1” and “2,” whe e “1” ep esen s he di ec ion owa ds he eNodeB; “2” s ands o heopposi edi ec ion.The o alnumbe o usable esou ce blocks in he sys em is conside ed o be 25 (since he a ailable sys em bandwid h is 5 MHz) o all ime in e als. The simula ion cons uc ed in Ma lab en i onmen is se o 100 ime in e als wi h a ce ain ime du a ion. The base p ice is se o 𝑝=500. In his modeled scena io, he URB, PUs, and ASUs a e in ol ed in he ne wo king. As shown in Figu e 4(a), mos o heRSUsa egoing obechanged oASUs.Howe e , he e a e s ill URBs a speci ic ime ins an s. Du ing he peak hou s (see Figu e 4(b)), almos 75% o he esou ce blocks belong o he PUs. In his case, no only is he limi ed numbe o URBs a ailable bu also he RASUs a e encoun e ed. In mo e de ail, PUs and RSUs exis in he ne wo k a he same ime, which is based on PUUsageRa e and RSUShowRa e; hisbeha io is acco ding o he men ioned use-cases; see Sec ion 3.2. As simula ion ime p og esses, he ASUs (𝐿ASU)and he s a us o each o hem will be changed o FASU, RASU, o he e will s ill be ASUs a each ime ins an ; o example, some o he ASUs will lea e he mobile ne wo k as FASUs andsomeuse smaybe ejec ed(RASU)acco ding ocase 3: 0≤URB(𝑡𝑖)<|𝐿PU(𝑡𝑖)|; see Sec ion 3.2. The impo an simula ion esul s a e de ailed in Table 3 o he o -peak and peak hou s, espec i ely. The Numbe o Remaining ASUs is u he calcula ed as ollows: NRASUs =𝐿ASU −𝐿FASU −𝐿RASU.(14) The numbe o RSUs du ing he o -peak hou s (622) is lowe han ha o e he peak hou s (1878). Fu he , 57.39% o heRSUsha ebecomeASUsdu ing heo -peakhou sand abou 8.87% did ha du ing he peak hou s. This beha io is due o he ac ha he LTE ne wo k ope a es wi h 25 RB (5 MHz bandwid h). Fu he , he numbe o RASUs du ing he peak hou s (49.39%) e lec s he To al Ne wo k Re enue (TNR), which is lowe du ing he o -peak ime. Howe e , o e he o -peak hou s, 89.91% o he ASUs inished hei da a ansmission success ully (FASUs = 21) and 4.76% o ASUs a e ejec ed (RASUs) in o al. Table 3 shows he o al numbe o RASUs (17) du ing he o -peak hou s and he peak hou s (82). Figu e 5 epo s he o alnumbe o imein e alswhen heRASUswe eac i- a ed in he ne wo k, see Figu e 5(a), and he es ima ed p ice 8 Wi eless Communica ions and Mobile Compu ing 0 20 40 60 80 100 Time cycles [—] 0 20 40 60 80 100 Spec um u iliza ion [%] URB PUs ASUs (a) 0 20406080100 Time cycles [—] 0 20 40 60 80 100 Spec um u iliza ion [%] URB PUs ASUs (b) Figu e 4: Spec um u iliza ion du ing o -peak (a) and peak (b) hou s. o hei nex connec ion a emp (Figu e 5(b)) du ing he o -peak hou s. This e lec s he ac ha whene e he RASUs a e willing o eques se ice ime, o example, a connec ion o he mobile ne wo k, he paymen will be as shown in Figu e 5(b) o each o hem. Fu he , Figu e 6 de ails he si ua ion o he peak hou s: as can be seen, i he only c i e ion o he ejec ed ASUs was he A ST, he e should be ag eemen be ween (i) Figu es 5(a) and 5(b), as well as (ii) Figu es 6(a) and 6(b). Howe e , he ob ained esul s a y e en o he ASUs wi h he same A ST. Fo example, A ST (ASU6)andA ST(ASU 7)a ese o 9 ime in e als. On he o he hand, he es ima ed p ice o he selec ed ASUs di e s (ASU6= 334.083;ASU 7= 320.629). This si ua ion clea ly indica es he impo ance o ano he c i e ion o he CC calcula ion in he p oposed CBRM model. We p oceed wi h conside ing he said c i e ion in he ollowing sec ion. 5. Use Selec ion Me hodology In his sec ion, we con inue ou desc ip ion o he con- s uc ed model ( o mula ed in he p e ious Sec ion 4). We ocus on selec ing he app op ia e RASUs among he ASUs. S ep-by-s ep selec ion unc ionali y o he se ed use s u i- lizes wo key p inciples: (i) a ailabili y o he LSA band and (ii) decision-based ejec ion unc ionali y as de ailed in Figu e 7. In o de o s udy pe o mance o he DecisionBase- dRejec ionPolicy algo i hm, he second ime in e al (𝑡2)is cap u ed. A he beginning, he LSA licensee checks whe he he spec um esou ces a e a ailable a 𝑡2. Acco ding o he URB (𝑡2)and|𝐿PU(𝑡2)|, henumbe o RASUsis|𝐿RASU(𝑡2)|= 2; see also he condi ion gi en in case 3: 0≤URB(𝑡𝑖)< |𝐿PU(𝑡𝑖)|; see Sec ion 3.2. The ollowing s ep in ou conside a ion o he ASU’s c i e ia and he impo ance deg ee calcula ion is o employ he AHP: he LSA licensee conside s all 6 c i e ia. The co e- sponding impo ance deg ees a e shown in Table 4. The LSA licensee calcula es he RR based on he condi ion o ASUs a he ime ins an 𝑡2, wi h he co esponding pa ame e s shown in Table 5. The A ST is he mos impo an a ibu e o bo h sides ( he LSA licensee and he incumben (PU)). The o de o he emaining a ibu es (c i e ia) is he ollowing: (i) RSSI, (ii) ejec ed his o y (RH), (iii) mobili y pa e n, (i ) bi a e (BR), and ( ) download/upload a e. As i was discussed in Sec ion 4, a compa ison be ween he ASUs based only on he A ST a ibu e is no easible (see Table 5, whe e he A ST a ibu e is se o 1 o all he ASUs). To his end, he second a ibu e (RSSI) was added o ou conside a ion. In his case, wo ASUs nomina ed o ejec ion will be ASUID=3 and ASUID=5 wi h RSSI = −78 dBm and RSSI = −82 dBm, espec i ely. Ano he impo an c i e- ion is RH, whe e ASUID=1 and ASUID=4 a e nomina ed o be ejec ed, since hey ha e he lowes ejec ion a io. Consid- e ing he MP, he ASUs selec ed o ejec ion will be ASUID=1 and ASUID=2. Fu he , he BR a ibu e is conside ed and, by ollowing his c i e ion, ASUID=1 and ASUID=5 a e he candida es o ejec ion. Finally, when DUR is aken in o accoun , ASUID=4 and ASUID=5 a e going o be ejec ed, since hey ha e he maximum download/upload a io a a gi en ime ins an 𝑡2. Con inuing he use selec ion p ocess, Table 6 p o ides in o ma ionabou heRR, ejec edo de ,andCCo heASUs. The LSA licensee selec s wo ASUs o ejec ion acco ding o |𝐿RASU(𝑡2)|→ASUID=1 and ASUID=5,whichha e helowes CC based on (12) (CC ASUID=1 = 0.1392;CCASU ID=5 = 0.1429). Fu he , i he LSA licensee ejec ed ASUID=4 ( hi d ASUselec ed o ejec ion,seeTable5), heLSAlicenseemus Wi eless Communica ions and Mobile Compu ing 9 Table 4: Impo ance deg ee o each c i e ion o he DecisionBasedRejec i onPolicy based on AHP me hod. C i e ia AHP Impo ance o de (MNO iew) Impo ance deg ee (AHP iew) BR 5 0.0573 RH 3 0.2005 RSSI 2 0.2526 A ST 1 0.3460 DUR 6 0.0422 MP 4 0.1010 02 4 6 8 1012141618 RASUs [ID] 1 2 3 4 5 6 7 8 9 10 Se ice ime cycle [—] (a) Es ima ed p ice o nex access in 0 2 4 6 8 1012141618 RASUs [ID] 200 220 240 260 280 300 320 340 360 380 o - peak hou s [—] (b) Figu e 5: Numbe o ime in e als ha RASUs a e ac i e (a) and es ima ed p ice o hei nex connec ion (b) du ing o -peak hou s. 0 102030405060708090 RASUs [ID] 0 1 2 3 4 5 6 7 8 9 Se ice ime cycles [—] (a) 0 102030405060708090 RASUs [ID] 50 100 150 200 250 300 350 400 450 Es ima ed p ice o nex access in peak hou s [—] (b) Figu e 6: Numbe o ime in e als ha RASUs a e ac i e (a) and es ima ed p ice o hei nex connec ion (b) du ing peak hou s. cha ge his ASU o 336.80. Since he base p ice is assumed o be 𝑝 = 500,ASU ID=1 and ASUID=5 will cha ge 430.40 and428.55 o hespec um esou ces, espec i ely.Thiscon- side a ion keeps ASUID=4 mo e sui able o he LSA licensee (i.e., inc easing he ne wo k e enue). On he o he hand, his decision will keep ASUID=4 sa is ied as i s DUR and BR a e lowe han hose o o he wo ejec ed ASUs. Fu he , a a iable spec um esou ce p ice (P iceRSU)inTable5con- i ms ha he p ice paid o using he LSA band by e e y accep ed RSU depends on hei ejec ion his o y (RH) ha is calcula ed ac oss p e ious ime in e als. As an ou pu o he decision p ocess, DecisionBasedRejec ionPolicy algo i hm has hepo en ial obalanceall hep obabili iesandmaximize he ne wo k pe o mance oge he wi h he use sa is ac ion. 6. Conclusion A C i e ia-Based Resou ce Managemen amewo k o he MNO LSA-based ope a ion was p oposed in his wo k. The p esen ed app oach is buil on he MNO’s decisions ega ding bo h p icing and ejec ion policies wi h se e al ypes o use s, he co esponding a ibu es, and a numbe o he MNO decision scena ios. The cons uc ed sys em model illus a es he MNO decisions ac oss se e al si ua ions in e ms o