This is a repository copy of A game theoretic approach for optimizing density of remote radio heads in user centric cloud-based radio access network.

Transcription

1 This is a repository copy of A game theoretic approach for optimizing density of remote radio heads in user centric coud-based radio access network. White Rose Research Onine URL for this paper: Version: Accepted Version Proceedings Paper: Romanous, B, Bitar, N, Zaidi, SAR et a. (3 more authors) (2016) A game theoretic approach for optimizing density of remote radio heads in user centric coud-based radio access network. In: 2015 IEEE Goba Communications Conference (GLOBECOM 2015) IEEE Goba Communications Conference, Dec 2015 IEEE. ISBN (c) 2016 IEEE. Persona use of this materia is permitted. Permission from IEEE must be obtained for a other users, incuding reprinting/ repubishing this materia for advertising or promotiona purposes, creating new coective works for resae or redistribution to servers or ists, or reuse of any copyrighted components of this work in other works. Reuse Uness indicated otherwise, futext items are protected by copyright with a rights reserved. The copyright exception in section 29 of the Copyright, Designs and Patents Act 1988 aows the making of a singe copy soey for the purpose of non-commercia research or private study within the imits of fair deaing. The pubisher or other rights-hoder may aow further reproduction and re-use of this version - refer to the White Rose Research Onine record for this item. Where records identify the pubisher as the copyright hoder, users can verify any specific terms of use on the pubisher s website. Takedown If you consider content in White Rose Research Onine to be in breach of UK aw, pease notify us by emaiing eprints@whiterose.ac.uk incuding the URL of the record and the reason for the withdrawa request. eprints@whiterose.ac.uk

2 A Game Theoretic Approach For Optimizing Density of Remote Radio Heads in User Centric Coud-Based Radio Access Network Bashar Romanous, Naim Bitar, Syed Ai Raza Zaidi, Ai Imran, Mounir Ghogho, Hazem Refai Schoo of Eectrica and Computer Engineering University of Okahoma,Tusa, USA University of Leeds, UK Emais: Abstract In this paper, we deveop a game theoretic formuation for empowering coud enabed HetNets with adaptive Sef Organizing Network (SON) capabiities. SON capabiities for inteigent and efficient radio resource management is a fundamenta design piar for the emerging 5G ceuar networks. The C-RAN system mode investigated in this paper consists of utra-dense remote radio heads (RRH) overaid by centra baseband units that can be coocated with much ess densey depoyed overaying macro base stations (BS). It has been recenty demonstrated that under a user centric scheduing mechanism, C-RAN inherenty manifests the trade-off between Energy Efficiency (EE) and Spectra Efficiency (SE) in terms of RRH density. The key objective of the game theoretic framework deveoped in this paper is to dynamicay optimize the trade-off between the EE and the SE of the C-RAN. More specificay, for an utra-dense C-RAN based HetNet, the density of active RRHs shoud be carefuy dimensioned to maximize the SE. However, the density of RRHs which maximizes the SE may not necessariy be optima in terms of the EE. In order to strike a baance between these two performance determinants, we deveop a game theoretic formuation by empoying a Nash bargaining framework. The two metrics of interest, SE and EE, are modeed as virtua payers in a bargaining probem and the Nash bargaining soution for RRH density is determined. In the ight of the optimization outcome we evauate corresponding key performance indicators through numerica resuts. These resuts offer insights for a C-RAN designer on how to optimay design a SON mechanism to achieve a desired trade-off eve between the SE and the EE in a dynamic fashion. Index Terms C-RAN, SON, Outage Capacity, Energy Efficiency, RRH density, Game theory, Nash Bargaining Soution. I. INTRODUCTION By the year 2020, the growth in mobie data is expected to increase by more than 1000 fods as compared to 2010 [1]. Consequenty, the emerging 5G wireess networks shoud be abe to support this massive proiferation of mobie devices and triggered exabyte food. Therefore, 5G technoogy is expected to: 1) be abe to support 1000 times of traffic density more than today s networks; 2) be capabe of serving 10 to 1000 times more terminas than today s networks; 3) achieve better network coverage. In order to make the capacity demands required for the upcoming 5G technoogy possibe, networks densification wi be an essentia part of 5G [2]. It is envisioned that such densification wi be reaized through the utra-dense depoyment of sma ces. Coud-based Radio Access Networks (C-RANs) are expected to faciitate the densification of ceuar networks through the depoyment of distributed Remote Radio Heads (RRHs) [3], [4]. The main characteristic of C-RAN architecture is that the baseband processing unit (BBU) is separate from the distributed RRHs. Each RRH is connected with the coud BBU poo via a front-hau which is often a fiber optic cabe. Such a centraized RAN architecture enabes the impementation of compex coexistence and scheduing mechanisms. The net overhead of impementing such mechanisms is ess than what woud occur in traditiona autonomous sma ce networks. Another benefit that C-RAN architecture provides is that it enabes significant energy savings. It is estabished that RRHs do not require energy expenditure compared to traditiona macro-bss where cooing and running of computing systems resuts in significant energy consumption. This distributed architecture carried by a centra management unit enabes the impementation of advanced interference mitigation schemes, such as interference aignment [5]. A of these features which can be expoited from impementing C-RAN have triggered intensive research in this area incuding [5] [8]. Another new ceuar networks design phiosophy in context of 5G is to transform the custering scheme from a base station centric approach to be user-centric [9], [10]. Benefits of foowing such an approach are dynamic coverage and higher ink success probabiity. Dynamic coverage is provisioned by turning on ony the RRHs which are needed to serve the desired user at a certain quaity of service (QoS). A higher ink success probabiity is achieved due to higher gain in the received signa strength (RSS) at the mobie user (MU). This diversity enabed gain is attained from having severa active RRHs in the usercentric custer. This custering scheme has been expored by [10], [11]. There are severa key performance indicators (KPI) that can be used to quantify C-RAN performance. The most important of these KPIs is the spectra efficiency (SE). A study on the SE in terms of outage capacity (OC) was conducted by [10], which shows how the OC reates to the density and to the empoyed transmission power of RRHs in each tier. Another

3 important KPI that needs to be taken into consideration during C-RAN s design and depoyment is the energy efficiency of the network. Power consumption in C-RAN has been a subject of intensive research. Such research attempted to characterize and reduce the power needed to perform tasks such as joint downink and upink user-ap association and beamforming [12], decoding data [13], and resource aocation [14]. A study on energy efficiency in dense sma ces networks for different on/off schemes has been studied in [15]. II. CONTRIBUTIONS & ORGANIZATION In this work, we study the inherent trade-off between the SE and the energy efficiency (EE) in C-RAN where usercentric custering is performed. In particuar, we examine how the two KPIs, the SE and the EE, vary with the density of active RRHs in a C-RAN network. Firsty, we summarize a user-centric custering scheme and present an anaytica framework that characterizes the SE in form of an OC formua. Thus, we wi use the terms SE and OC interchangeaby throughout this paper. The next step, is to characterize the cost of impementing a user-centric custering mechanism. This is achieved by deducing a formua that quantifies the EE on the hoistic eve of the network. Our formua quantifies the energy that is needed for seecting the best RRH which wi serve the MU in each user-centric custer. We demonstrate that there exists a trade-off between the EE and the SE in terms of RRH density. The key objective of this study is to provide a SON capabiity for the seection of RRH density per tier that woud provide the best trade-off between the SE and the EE. To achieve this goa, we utiize a game theocratic formuation to find the RRH density vaue by modeing the probem as a bargaining probem. We find the Nash bargaining soution (NBS) which achieves the best baance between those two KPIs. As discussed in the introduction section, a number of prior studies have expored the EE and the SE in context of macro and sma ce networks [5], [10] [15]. However, to the best of our knowedge, this paper presents the first study of its kind that investigates the trade-off between the EE and the SE in the context of C-RAN, where user centric custering is impemented, and utiizes a game theoretic framework to optimize the soution. This paper buids on very recent resuts presented in [5], [10] [15]. The rest of the paper is organized as foows: In section III, we describe the user-centric custering scheme and characterize ce OC. In section IV, we quantify the effect of RRH density on power consumption in the network. We examine the trade-off between the two performance metrics in section V. In section VI, we formuate the probem as a bargaining game and vaidate the required axioms for it to have a Nash bargaining soution. Numerica resuts of NBS and discussion are presented in section VII. The paper is concuded in section VIII. III. OUTAGE CAPACITY UNDER USER-CENTRIC CLUSTERING MECHANISM In this paper, we consider the downink operation of a arge scae ceuar network provisioned by a dynamic user-centric custering scheme. Under such mechanism, the first tier is constituted by macro BSs and the remaining (k 1) tiers correspond to sma ces thats consist of RRHs. It is assumed that dissimiar RRH densities and transmission powers are empoyed per tier. Various user-centric custers can be formed within each tier. The tota bandwidth is divided into sub-bands where each sub-band is assigned to one custer. Sub-bands are aocated to custers in a manner that cross-tier interference is eiminated. The user-centric custering mechanism is managed by the C-RAN contro center. For an arbitrary MU, the C-RAN centra controer ocates the best tier that can serve the MU under a specific QoS requirement. The QoS requirement is captured by having the MU as a center of a custer that does not contain any other schedued users except for the targeted MU. Each custer is represented by a circe with radius R and is centered around the MU. The average number of RRHs inside each custer is assumed to be greater than unity. The operation of forming a user-centric custer proceeds as foows: The macro BS that is cosest to the MU transmits a piot signa to it. The MU, in return, retransmits the piot signa to a RRHs contained within it s custer. The corresponding RRHs examine the received strength of the piot signa. The RRH seection mechanism chooses the RRH which wi be abe to provide the targeted MU with the highest RSS among the group of RRHs ocated within the custer. It is worth noting that none of the RRHs contained within the custer are aowed to concurrenty serve any other MUs unti the targeted MU finishes its current activity. The benefits attained form appying this mechanism can be summarized as: 1) RRH seection diversity enabes a higher gain in the received signa at the targeted MU. 2) Since each custer uses its own sub-band and no overapping custers are aowed, this permits an effective mitigation of both co-tier and cross-tier interference. Any users who beong to overapping custers are schedued to be served ater. 3) The dynamic scheduing empoyed in this scheme enabes energy savings. Ony the best RRH is activated, whie the rest of RRHs are put in seep mode. We characterize the reationship between RRH density and the ce outage capacity in a k-tier C-RAN, where the propagation channe suffers from Rayeigh fading compemented with arge scae power-aw path-oss, by the foowing proposition: Proposition 1. For a desired reiabiity constraint ρ, the outage capacity is defined as the maximum downink throughput which can be obtained in the network such that the outage probabiity for the MU remains beow a per-designated reiabiity threshodρ. The upper-bound on the outage capacity

4 under ρ is given as C ρ og i=1λ i P δ 2 1+(πδΓ(δ) k i )δ 1 σ 2 n(1/ρ) δ 1, (1) where the noise at the receiver is assumed to be additive white Gaussian noise (AWGN) represented by a random variabe with Gaussian distribution of N(0,σ 2 ); δ is a path oss dependent constant given as δ=2/α for path oss exponent α>2. k is the number of tiers in the C-RAN. We assume that the RRH density in each subsequent tier is denser than its antecedent tier. Thus, we denote λ i as the RRH density for tier i k. This can be stated in terms of the baseine density λ, where λ i =η i λ for η 1. P i is the transmitted power per tier i k which can be cacuated by P i = β i P, where β 1 and P is the baseine power empoyed at the parent tier. It is assumed that RRHs in each tier consumes ess power than its parent tier. Proof: Pease refer to [10] IV. POWER CONSUMPTION IN C-RAN It is important to quantify the cost of impementing the user-centric custering scheme in terms of power expenditure. This aows us to compare the attained diversity gain with the consumed power in the network. One penaty for having diversity gain is that a RRHs in the custer have to be active during the RRHs seection phase. The more active RRHs are avaiabe to choose from, the higher is the achieved diversity gain, but the more tota power is consumed per custer. This process creates a trade-off between the EE and the SE of the user-centric custering mechanism. The EE of the network can be quantified as the cost function of impementing the custering mechanism. In order for our evauation to be vaid, we ony focus on the energy consumed during RRH seection phase, as it represents the overhead caused by the user-centric custering scheme. A. Power consumption mode Power consumption of various types of wireess networks has been investigated in [16]. The authors in [17] focus on power consumption for muti-input muti-output discontinuous transmission in C-RAN. We extend the formua described in [17] in order to quantify power consumed in the network during the discovery process of the RRH, which can be quantified as: P CRAN =ξ CRAN + µ P µ +P 0µ, (2) where P µ is the transmit power empoyed by the MU. µ is a parameter that reates power consumption with the empoyed radio frequency. P 0µ is the fixed power consumed by the hardware of the MU. ξ CRAN is the C-RAN coefficient, which represents the tota consumed power by every active RRH in a k-tiers of the network. It shows the proportiona reation between power consumption on the wide network eve and the density of RRHs and their empoyed transmission power per tier. The C-RAN coefficient is aso proportionay reated to θ, which represents the efficiency of the impementation. Hence, the C-RAN coefficient ξ CRAN is expressed by the foowing equation: k ξ CRAN =θ λ i P i, 0 θ 1, (3) i=1 where θ = 0 represents the most energy efficient impementation. B. Energy Efficiency The energy efficiency measures the number of bits transmitted per unit of bandwidth at the expense of one Joue during one second. We quantify EE according to the foowing proposition: Proposition 2. We can express the energy efficiency at the network eve by the foowing anaytica expression: ω EE = BC ρ θ k i=1λ i P i + µ P µ +P 0µ, (4) where the transmission bandwidth B is normaized to unity. Proof: Proof of this proposition directy stems from the definition of EE at the network hoistic eve, thats is: The ratio of sustainabe throughput for each schedued user to the power consumed at the user mobie device and the RRHs during the RRH seection phase. In other words EE is the ratio of [bits/s/hz] over consumed units of [Joue]. V. TRADE-OFF BETWEEN OUTAGE CAPACITY AND ENERGY EFFICIENCY To examine the effect of RRH density in each tier, we start by examining the effect of baseine RRH density λ on the OC and EE. We anayze OC as expressed in (1) and the EE as expressed in (4) for various vaues of baseine RRH density λ. We perform the anaysis using the parameters from Tabe I with different variations of η and β. Figure 1a depicts the impact of RRH density per tier on OC. It can be concuded from the corresponding graphs that OC increases as RRHs become denser. Figure 1a consoidates the observation that after certain RRH density, the corresponding OC pot becomes saturated and no significant gain can be obtained from increasing RRH density any further. The optima RRH density for maximum OC and the corresponding peak OC vaues are shown in the second and third coumns of Tabe II. Figure 1b shows the changes in EE with an increase in RRH baseine density. It can be observed from figure 1b that EE increases with RRH density up to a certain RRH density threshod. Intuitivey, for a arge increment rate η in RRH density, EE drops significanty. Coumns 2 and 3 from Tabe III show the optima RRH density vaues which resut in peak EE vaues. By examining the second coumn from Tabes II and III for each case study, we notice a major difference in the optima RRH density vaues. This simpe comparison ceary demonstrates that there exists an inherent trade-off between the two performance determinants, EE and OC in C-RAN under user centric custering. A sef-organizing capabiity is essentia here to guarantee that the best throughput and energy efficiency are achieved and maintained in the C-RAN. This

5 sef-organizing feature shoud be abe to dynamicay seect the most appropriate number of RRHs that shoud be active in each tier to achieve the desired eve of baance between OC and EE, whie taking into account spatio-temporay changing channe and user distributions. In the next section we wi empoy a game theoretic framework to sove this diemma. VI. GAME THEORY FRAMEWORK In the previous section, we concuded that seecting the best baseine RRH density woud require a trade-off between the OC and the EE. Therefore, a SON mechanism for Coud BBU poo must be devised such that the baseine RRH density strikes a desired baance between those two performance determinants. As we wi see, a game theoretic approach can provide a soution to this diemma. We propose modeing the two performance metrics as virtua game payers. Ce OC is modeed as the first payer with (1) as its utiity function, and the EE is modeed as the second payer with (4) as its utiity function. A. Game Formuation Each payer s payoff is affected by the seection of baseine RRH density λ made by the other payer. Benefiting from the centraized management in C-RAN, we can define the probem as a cooperative game. The two payers wi have to negotiate for the vaue of λ. Both payers mutuay benefit from reaching an agreement over the optima baseine RRH density. Thus, both ce OC and EE can reach an optima tradeoff. We prove that this negotiation process can be modeed as two-payer Nash Equiibrium bargaining game. B. The Bargaining probem Let N={1,2} be the set of the payers, where payer i=1 denotes OC and payer i=2 denotes EE, and S i denotes the set of a feasibe payoffs to a user i, as: S i ={U i U i =U i (λ ),λ R λ >0}. (5) Let s define the space S as the set of a feasibe payoffs that payer i N can achieve when they work together is: S={U=(u 1,u 2 ) u 1 S 1,u 2 S 2 }, (6) where u 1 is the utiity of the first payer and u 2 is the utiity of the second payer where s 1 = u 1 =C ρ (λ ), (7) s 2 = u 2 =ω EE (λ ), (8) and λ R λ > 0. We aso define the disagreement space (D S) as the set of the two disagreement points d=(d 1,d 2 ), where d 1 = u 1 (D) and d 2 = u 2 (D) represent the payoff for each payer if the bargaining process faied and no outcome is reached. For our game we set d=(0,0). Therefore, we give both payers the same bargaining power in the game. Proposition 3. The probem described by (18) and (19) is a two-payer bargaining probem defined by the pair(s, D) where S R 2 and D S. Proof: For a bargaining probem to be defined, S shoud be a convex and a compact set [18]. Since it is cear that S is compact, we ony need to prove that it is a convex set: ǫ 0 ǫ 1 and if U a =(u a 1,u a 2) S and U b =(u b 1,u b 2) S, then ǫu a +(1 ǫ)u b S. Since u 1 =og 2 (1+ (πδγ(δ) k i=1 λipδ i )δ ), we denote the SIR σ 2 n(1/ρ) δ 1 as γ and re-write it as: u 1 =og 2 (1+γ) where γ ρ. Without oss of generaity we can appy the condition of convexity on 1+γ. Since taking the ogarithmic vaues of a convex set wi not change its convexity property: 1 ǫu a 1+(1 ǫ)u b 1=1+ γ, (9) 1 where1+ γ=1+ (πδγ(δ) k i=1 (ǫλa i +(1 ǫ)λb i )Pδ i )δ ), where0<λ a σ 2 n(1/ρ) δ 1 i and 0<λ b i, thus it can be easiy concuded that vaues of 1+ γ form a convex set and that the same appies to the vaues of og 2 (1+ γ). Hence, we prove that: ǫu a 1+(1 ǫ)u b 1 S 1. (10) As for the utiity of the second payer, we use the same aforementioned definition for γ and γ from above. Simiary we find: Bog u 2 = 2 (1+γ) θ k. (11) i=1λ i P i + µ P µ +P 0µ we aready proved that the outcome of the numerator is convex set, as for denominator: k θ λ i P i + µ P µ +P 0µ, (12) i=1 where k i=1 λ i P i =( k i=1ǫλ a i + k i=1(1 ǫ)λ b i )P i thus, we write ǫu a 2+(1 ǫ)u b 2= Bog 2 (1+ γ) θ k i=1 λ i P i + µ P µ +P 0µ, (13) Since 0<λ a i and 0<λb i, we find that the denominator is aso convex. Thus, we concude that: ǫu a 2+(1 ǫ)u b 2 S 2, (14) from (10) and (14) we concude that the ǫu a +(1 ǫ)u b S and the set S is convex. C. Nash Bargaining Soution In order for a bargaining probem to have a soution U = (u 1,u 2) for the disagreement space D=(d 1,d 2 ), Nash has specified four axioms that the bargaining outcome must satisfy [18]: 1) Pareto Efficiency: The Nash bargaining soution must be Pareto-optima. This means that there cannot be a soution where utiities of both payers can be improved in the same time. This concept can be mathematicay expressed as: (U 1,U 2 )>U (U 1,U 2 ) S. (15) 2) Symmetry: The soution of the bargaining probem shoud remain the same if the roes of the two payers

6 OutageCapacity(bits/s/Hz) OutageCapacityVsBaseineRRHdensity BaseineRRHdensity =3.5, =0.45 =5, =0.45 =3.5, =0.7 =5, =0.7 (a)outagecapacitywithvaryingrrhbaseinedesnity EnergyEficiency(bits/s/Hz/Joue) EnergyEficiencyVsBaseineRRHdensity =3.5, =0.45 =5, =0.45 =3.5, =0.7 =5, = BaseineRRHdensity (b)energyef ciencywithvaryingrrhbaseinedesnity Fig.1:TheefectofvaryingRRHbaseinedensityonOCandEEforvariouscasestudies.ThepeakandNBSvauesforeach graphareconsecutiveydenotedby and ofthesamecorespondingcoor weretobeswitched.inother words,thebargaining soutiondoesnotdiscriminatebetweenthepayersif theywereindistinguishabe. 3) Invariancetoequivaentutiityrepresentation:TheNash bargainingsoutionmustsatisfythefoowingcondition foranystrictyincreasinginearfunctionf U [F(S),F(D)]=F[U (S,D)]. (16) 4)Independenceofireevantaternatives:Let sconsider ŚasmaersetofSwhereU isstipartofśthenu winotchange: U (S,D) Ś S U (Ś,D)=U (S,D). (17) WenowsearchfortheuniqueNashBargainingSoutionthat satis estheaxiomsabove. Westartbyde ningthe Nash product[18]whichisexpresedas: max (s 1 d 1 )(s 2 d 2 ),s.t.(s 1,s 2 ) S (d 1,d 2 ). (18) (s 1,s 2) Weseectthedisagreementpointsas (d 1,d 2 )=(0,0). By substitution,weget: max(c ( )) ( EE ( )). (19) Equation(19)representsthe nanashproductwhichwewant to maximizefor R R>0 VI. RESULTS&DISCUSSION Nashproductde nedin(19)andispotedin gure2.the NBSvauethat weareookingforisrrhbaseinedensity, which maximizesthe Nashproduct. Usingthenetwork parameters vauesfromtabei,wecreatefourcasestudies tobeexamined.eachcasestudyrepresentsadiferentsetof vauesof and.theobtained NBSvaueof andits corespondingoutagecapacityvauesareshownincoumns4 and5oftabei.thenbsvaueof anditscoresponding energyef ciencyvauesareshownincoumns4and5of TABLEI:Listofparametersusedintheanaysis Parameter Vaue k {3.5,5} {0.45,0.7} 0.2 P 1Wats P µ 1Wats µ 4 P 0µ 4.3Wats 0.5 TABLE I:Peakand NBSoutagecapacityvauesandtheir corespondingrrhdensityvaues Case opt C peak NBS C NBS %Los =3.5& = % =5& = % =3.5& = % =5& = % TABLE I:PeakandNBSenergyef ciencyvauesandtheir corespondingrrhdensityvaues Case opt peak EE NBS NBS EE %Los =3.5& = % =5& = % =3.5& = % =5& = % Tabe I.Bycomparingfourthand fthcoumnsbetweenthe TabesIand I,we ndthat,bothocandeehavedropped

7 byasomeamountfromitspeakvaue.theospercentage ineach KPIhasbeencacuatedforeachstudycaseandis showninthesixthcoumnintabeiandtabe I.Asimpe comparisonbetweentheospercentagesinocandeeshow thattheimpactofthebargainingprocesonoutagecapacity is morepronouncedascomparedtoitsimpactonee.ifitis desiredtocreateabiasintheoutcometowardsoneofthekpis, athreshodonacceptabe-ospercentagecanbede ned.such threshodisre ectedbyanewsetofdisagreementpoints. Eachofd 1 andd 2 canbegivenavaue 0 d 1 C peak and 0 d 2 peak EE consecutivey.theprocesthencanbe repeatedbyusingthisnewdisagreementspaced.however,it isintuitiveyobviousthatanyimprovementintheutiityofone payer wicauseadeteriorationintheutiityofthesecond payer. Hence,suchthreshodsshoudbecarefuydesigned whenintegratingthisdynamictradingmechanismbetweenee andseasasef-organizingcapabiityofthec-ranbased depoymentof5g. V I. CONCLUSION Inthis work, wehavedeveopedananayticaframework forcharacterizingenergyef ciency(ee)andoutagecapacity (OC)usinggametheoryincoudradioaccesnetwork(C- RAN).C-RAN modeconsideredinthispaperexpoitsutradenseremoteradioread(rrh)depoymentstodynamicay performuser-centriccustering mechanismforradioresource scheduingandenergyef ciency(ee). Ouranaysisshows thatthereexistsvauesofactive RRHsdensities whichcan maximizeeeandoc.however,aninherenttrade-ofbetween maximizing OCand EEneedstobeaddresed. Wesove thisdiemmaofoptimizingtwocon ictingobjectives,i.e., EEandOC,by modeingtheprobemasabargaininggame. Thetwoperformanceindicators,EEand OC were modeed asvirtuagamepayersinabargaininggame. Ouranaysis showsthatanashbargainingsoutionexistsinsuchagame. Severascenariosforthegamehavebeenexamined. Weuse theospercentagebetweenpeakand NBSvauesineach payer sutiityasacomparison metric. Thus, weevauate theresutsandconcudethat RRHsdensityobtainedfrom the NBSprovidesareasonabetrade-of. Weshowthatthe bargaininggamemodeasoenabestheposibiityofshifting thesoutiontoa morepreferabetrade-ofevedictatedby networksperformancerequirements. REFERENCES [1] R. Wang, H. Hu,and X. Yang, Potentiasandchaengesofc-ran supportingmuti-ratstoward5gmobienetworks, IEEEAcces,vo.2, pp ,2014. [2] A.Imranand A.Zoha, Chaengesin5g:howtoempowerson with bigdataforenabing5g, IEEENetwork,vo.28,no.6,pp.27 33,Nov [3] C. Mobie, C-ran:theroadtowardsgreenran, WhitePaper,ver,vo. 2,2011. [4] A.ImranandR.Tafazoi, Evauationandcomparisonofcapacities andcostsof mutihopceuarnetworks, ininternationaconference onteecommunications, May2009,pp [5]J.Cao,D.Zhu,and M.Lei, Upink-downinkinterferenceaignment intdd-basedceuarnetworks, inieee24thinternationasymposium onpersonaindoorand MobieRadioCommunications(PIMRC),Sept 2013,pp NashProduct NashProductofS 1 ands 2 =3.5, =0.45 =5, =0.45 =3.5, =0.7 =5, = BaseineRRHdensity Fig.2:NashProductforeachcasestudy.Thepeakvauefor eachgraphisdenotedby ofthesamecorespondingcoor [6] W.-T.Lin,C.-H.Lee,and H.-J.Su, Downink-to-upinkinterference canceationincoudradioaccesnetworks, inieee79thvehicuar TechnoogyConference(VTCSpring), May2014,pp.1 5. [7] D. Zhuand M. Lei, Traf candinterference-awaredynamicbburu mappinginc-rantdd withcros-subframecoordinatedscheduing/beamforming, inieeeinternationaconferenceoncommunications Workshops(ICC),June2013,pp [8] A.Davydov,G. Morozov,I.Bootin,andA.Papathanasiou, Evauation ofjointtransmisioncompinc-ranbasedte-ahetnets witharge coordinationareas, inieeegobecom Workshops(GC Wkshps),Dec 2013,pp [9] D. Huang, T. Xing,and H. Wu, Mobiecoudcomputingservice modes:auser-centricapproach, IEEE Network,vo.27,no.5,pp. 6 11,September2013. [10] S.Zaidi, A.Imran, D. McLernon,and M. Ghogho, Characterizing coverageanddowninkthroughputofcoudempoweredhetnets, IEEE CommunicationsLeters,vo.PP,no.99,pp.1 1,2015. [11] Y.Zhangand Y.J.Zhang, User-centricvirtuacedesignforcoud radioaccesnetworks, inieee15thinternationa WorkshoponSigna ProcesingAdvancesinWireesCommunications(SPAWC),June2014, pp [12] S.Luo,R.Zhang,andT.J.Lim, Coordinateddowninkandupinkuser asociationandbeamformingforenergy minimizationincoudradio accesnetwork, inieeeinternationaconferenceonacousticsspeech andsignaprocesing(icassp), May2014,pp [13] L.Chen,H.Jin,H.Li,J.-B.Seo,Q.Guo,andV.Leung, Anenergy ef cientimpementationofc-raninhetnet, inieee80thvehicuar TechnoogyConference(VTCFa),Sept2014,pp.1 5. [14] P.-R.Li,T.-S.Chang,andK.-T.Feng, Energy-ef cientpoweraocation fordistributedarge-scaemimocoudradioaccesnetworks, inieee WireesCommunicationsandNetworkingConference(WCNC), Apri 2014,pp [15] Y.-N.R.Li,J.Li,H. Wu,and W.Zhang, Energyef cientsmace operationunderutradensecoudradioaccesnetworks, ingobecom Workshops(GC Wkshps),2014,Dec2014,pp [16] G. Auer, V. Giannini,C. Deset,I. Godor,P.Skiermark, M. Oson, M.Imran,D.Sabea, M.Gonzaez,O.Bume,andA.Fehske, How muchenergyisneededtoruna wireesnetwork? IEEE Wirees Communications,vo.18,no.5,pp.40 49,October2011. [17] R.Gupta,E.CavaneseStrinati,andD.Ktenas, Energyef cientjoint dtxandmimoincoudradioaccesnetworks, inieee1stinternationa ConferenceonCoudNetworking(CLOUDNET),Nov2012,pp [18] Z.Han,D.Niyato, W.Saad,T.Baar,andA.Hjrungnes,GameTheory in WireesandCommunicationNetworks:Theory, Modes,andAppications. CambridgeUniversityPres,2012.