Bargaining Game for Effective Coexistence between LTE-U and Wi-Fi Systems

Transcription

1 Barganng Game for Effectve Coexstence between LTE-U and W-F Systems Anupam Kumar Barag, Nguyen H. Tran, Wald Saad +,, and Choong Seon Hong Dept. of Computer Scence and Engneerng, Kyung Hee Unversty, Korea + Wreless@VT, Bradley Department of Electrcal and Computer Engneerng, Vrgna Tech, Blacksburg, VA E-mal:anupam@khu.ac.kr, nguyenth@khu.ac.kr, walds@vt.edu, cshong@khu.ac.kr Abstract LTE over unlcensed band (LTE-U) has emerged as an effectve technque to overcome the challenge of spectrum scarcty. Usng LTE-U along wth advanced technques such as carrer aggregaton (CA), one can boost the performance of exstng cellular networks. However, f not properly managed, the use of LTE-U can potentally degrade the performance of coexstng W-F access ponts whch operate over the unlcensed frequency bands. Moreover, most of the exstng works consder a macro base staton (MBS) or a small cell base staton (SBS) for ther proposals. In ths paper, an effectve coexstence mechansm between LTE-U and W-F systems s studed. The goal s to enable the cellular network to use LTE-U wth CA to meet the qualty-of-servce (QoS) of the users whle protectng W-F access ponts (WAPs), consderng multple SBSs from dfferent operators n a dense deployment scenaro. Specfcally, an LTE-U sum-rate maxmzaton problem s formulated under a user QoS and WAP-LTE-U co-exstence constrants. To solve ths problem, a cooperatve Nash barganng game s proposed. Ths game allows LTE-U and WAPs to share tme resource whle protectng W-F system. For allocatng unlcensed resource among LTE-U users, a heurstc algorthm s proposed. Smulaton results show that the proposed method s better than the comparng methods regardng per user acheved rate, percentage of unsatsfed users and farness. The result also shows that the proposed method protects W-F user far better way than basc lsten-before-talk (LBT) does. I. INTRODUCTION Recent studes by Csco [1] have shown that the moble wreless traffc wll contnuously ncrease over the foreseeable future, wth moble vdeo traffc consttutng the man chunk of ths traffc. As such, cellular network operators (CNOs) and servce provders (SPs) must revst the desgn of ther network, n order to meet the qualty-of-servce (QoS) needs of ther users. In partcular, CNOs must ncrease the capacty of ther rado access networks (RANs) by explotng new spectrum bands, n conjuncton wth ther exstng lcensed spectrum. But the lcensed spectrum s both expensve and nadequate. Moreover, ths superfluous traffc does not guarantee proportonal revenue gan to the CNOs because of monstrous competton among themselves. Thus, CNOs are contnuously tryng to fnd the cost-effectve soluton to handle ths hazard and Long Term Evoluton (LTE) was a logcal evolvement to meet growng requrements of the users. Even though there are some alternatve technologes lke massve multple nputs multple outputs (MIMO) [2], co-operatve communcaton [3], /18/$ IEEE D2D (Devce to Devce) communcaton [4], etc., these were not enough to meet the ITU requrements of 4G. On the consequence of ths ongong process, LTE-Advanced (LTE-A) came, whch brought cellular networks nto true 4G era [5]. Furthermore, CNOs are also focusng the reusng technque of lcensed spectrum by deployng low-cost and low power small cell base staton (SBS) wth carrer aggregaton (CA), the technology of LTE-A to meet QoS requrements of the users. However, despte these advances, the scarcty of the lcensed spectrum wll reman a key lmtaton for the cellular networks. Consequently, moble data offloadng from cellular to W-F has ganed recent attenton [1]. In fact, some CNOs have already deployed W-F Access Ponts (WAPs) to offload part of ther cellular traffc and, n 2015, more than 50% of cellular network traffc was offloaded to the W-F [1]. However, such traffc offloadng faces major techncal and economc challenges due to coordnaton between two technologes and potental revenue losses. Moreover, the performance of W-F technology s not good wth a large number of users. Thus, the weakness of such offloadng can be brought down by augmentng the blessngs of LTE-A n the unlcensed spectrum known as LTE-U. LTE-U wll amelorate RAN capacty at a mnmal cost, and resolve the revenue ssue of the cellular system. It wll ncrease the performance of the moble network better than W-F does [6] by utlzng the already deployed network. But the transmsson range of the unlcensed spectrum s small n comparson wth lcensed one owng to low power regulaton, and hgher frequency. Hence, SBS s the approprate opton for LTE-U deployment, and CNOs can transform ther already deployed SBS nto co-located ones (works n both lcensed and unlcensed spectrum) for ths purpose. It can be techncally ensured va the utlzaton of CA technology whch was standardzed n LTE Releases LTE-U s already naugurated (part of the LTE Release 13) to allow consumers to accommodate lcensed, and unlcensed carrer under a sngle LTE network nfrastructure [7]. Moreover, nter- and ntra-band CA, lcenses asssted access to the ISM band, TV whte space, and other under-utlzed resources are urgently necessary for makng 5G realm true [8]. One of the man lmtatons of LTE-U s that t can cause consderable performance degradaton of other exstng technologes lke W-F system f t operates n the same

2 unlcensed band. Thus, an LTE-U SBS should not generate more nterference to the WAP than another WAP does n the same unlcensed band f t wants to operate n that band. On the contrary, WAPs, and other CNOs are also obstructng LTE-U users, and leads to nadequate data rate to meet the QoS who are operatng both n the same regon, and same unlcensed band because of ther ad-hoc deployment. So, there exsts mutual nterference among SBSs of dfferent CNOs, and between SBSs, and WAPs whch can dmnsh each others benefts n the unlcensed spectrum. Thus, co-exstence wth WAPs s the man challenge of LTE-U system. There are several works ([10], [11], [12]) that deals wth the co-exstence of LTE-U, and W-F system. But very few of them have consdered nter-operators nteracton n ther model and very few of them fnd guaranteed concrete closedform for W-F system protecton. As LTE-U and WF systems affect negatvely on each other, and wthn themselves n case of performance, ths nteracton can be modeled as a game theory framework namely barganng game to promote ther mutual benefts. So, we propose a coexstence mechansm that can deal multple CNOs whle protectng W-F users n the same unlcensed band. The man contrbutons of ths paper are as follows: We derve the nteracton among SBSs and WAPs mathematcally and fnd LTE-U users acheve rate. We formulate an optmzaton problem to maxmze the sum-rate n log-term of LTE-U users consderng QoS requrements, and co-exstence ssue wth WAPs. We decompose the optmzaton problem nto two subproblems namely tme sharng, and resource allocaton problem. We solve the tme sharng problem between LTE-U and W-F system wth the help of Nash barganng game (NBG), and resource allocaton problem of LTE-U system wth a heurstc algorthm. The rest of the paper s organzed as follows. In Secton II, we dscuss the system model and problem formulaton. The soluton of the problem s dscussed n Secton III. Performance evaluaton has been performed n Secton IV. Fnally, the paper s concluded n Secton V. II. SYSTEM MODEL AND PROBLEM FORMULATION Currently, small cell networks are the probably feasble soluton to meet data demand of the users, CNOs are deployng more and more SBSs to facltate growng servces. Ths ultradense nature of SBSs from dfferent operators are bound to conflct wth each other, and also wth local WAPs f they want to operate n the same unlcensed band. As each operator can control the nterference between macro base staton (MBS) and t s assocated SBSs, we are consderng a scenaro wth SBSs, and WAPs as shown n Fgure 1. Ths dense deployment scenaro conssts of a set of dual-mode (whch can act both n the lcensed and unlcensed spectrum) LTE-A SBSs, S = {1, 2,.., S} operated by S dfferent CNOs, and a set of non-overlappng WAPs denoted by W = {1, 2,..,W }. That means WAP s performance wll be affected by the SBSs only f they use the same unlcensed band. Each SBS Scan SBS-1 WAP SBS-2 Lcensed & unlcensed sgnal Unlcensed sgnal Unlcensed nterference User wth communcatng devce Fgure 1: Illustraton of the system model serve downlnk operaton of a set of LTE-U users, denoted by U. Each SBS Sowns K l orthogonal lcensed subchannel of unform bandwdth B l, denoted by SC l = {1, 2,..., K l} to support t s users. We assume that each WAP w W has V w actve users currently, and s denoted by V w = {1, 2,.., V w }. Both SBSs and WAPs operate n the same unlcensed band. As the unlcensed channel s much wder than one lcensed subchannel, and the LTE system works centrally, so each SBS dvdes ths unlcensed spectrum nto K u subchannels wth bandwdth B u each, represented by SC u = {1, 2,..., K u } for effcent management of ths resource. For relable transmsson of control sgnals from the SBS to the user, each SBS allocates at least one lcensed subchannel to ts actve LTE-U user. We assume that one subchannel can be allocated to a maxmum of one LTE-U user. SBSs work n the supplemental downlnk (SDL) mode wth CA technology. A. Data Rate of LTE-U User As SBS employs the OFDMA technque to allocate resources among t s users, there s no ntra-operator nterference n lcensed spectrum. When SBS S allocates lcensed subchannel k SC l to user j U, the acheved rate of that user s as follows: ( R l,k,j = B l log 2 1+ xk,j P l h ),j 2 σ 2. (1) where x k,j ndcates the allocaton of lcensed subchannel k SC l by SBS Sto user j U, and x k,j =1when SBS Sallocates the subchannel to user, and x k,j =0, otherwse. P l s the transmsson power from SBS to t s user j, and t s fxed for all of t s users. h,j 2 s the channel gan from SBS to user j consderng a free space propagaton pathloss model wth Raylegh fadng, and h,j 2 = Gd α,j h 0 2 where G ndcates the constant power gan factor ntroduced by amplfer and antenna, d,j s the dstance between SBS and user j, α s the path-loss exponent, and h 0 CN(0, 1)

3 Table I: Notaton Symbol Meanng S Set of SBSs wth S elements W Set of non-overlappng WAPs wth W elements U Set of users assocated wth SBS V w Set of V w actve users assocated wth WAP w SC l Set of lcensed subchannels of SBS SC u Set of unlcensed subchannels SC u Set of unlcensed subchannels of SBS B l Bandwdth of each lcensed subchannel B u Bandwdth of each unlcensed subchannel P l Transmsson power of SBS for each user n lcensed spectrum P u Transmsson power of SBS for each user n unlcensed spectrum x Resource allocaton vector for SBS n lcensed spectrum y Resource allocaton vector for SBS n unlcensed spectrum h,j 2 Channel gan between AP and recever j d,j Dstance between AP and recever j G Constant power gan factor α Path-loss exponent h 0 Raylegh fadng I S\{} Interference from S to any user of n any unlcensed subchannel I W Interference from W to any unlcensed subchannel R l,k,j Acheved rate of user j assocated wth SBS n lcensed subchannel k R u,k,j Acheved rate of user j assocated wth SBS n unlcensed subchannel k R,j R max R mn QoS,j τ Acheved rate of user j assocated wth SBS Average rate of user v assocated wth WAP w when WAP s accessng the channel Average rate of user v assocated wth WAP w when SBSs act lke WAPs QoS requrement of user j assocated wth SBS Fracton of tme that SBSs share wth WAPs s a complex Gaussan varable representng Raylegh fadng. The thermal nose has an ndependent Gaussan dstrbuton wth zero mean, and varance σ 2. LTE-A system can employ CA technology to provde a better rate to t s users for mantanng QoS f SBS has suffcent unused lcensed subchannels. When SBS S allocates more than one subchannels to user j U then the acheved rate of that user n the lcensed subchannel s as follows: R,j(x l )= x k,jr l,k,j. (2) k SC l If R,j l (x ) s large enough to meet the QoS of user j, then t needs not usng the unlcensed spectrum. On the other hand, SBS wll allocate unlcensed subchannel to user j f the acheved rate s not suffcent to provde guaranteed QoS. In case of the unlcensed subchannel, the LTE-U user perceves nterference from other SBSs, and WAP workng n the same conflctng area over the same unlcensed band. The rate obtaned by LTE-U user j U over the unlcensed subchannel k SC u s as follows: R u,k,j = B u log 2 1+ yk,j P u h,j 2 I S\ + I W + σ 2. (3) where y,j k represents the allocaton of unlcensed spectrum k SC u to LTE-U user j U by SBS S, and y k,j =1 when SBS Sallocates the unlcensed subchannel to the specfed user, and y,j k =0, otherwse. P u s the transmsson power from SBS to t s user j n case of the unlcensed spectrum, and t s fxed for all of t s users. h,j 2 s the channel gan between SBS and user j n the unlcensed n U s x k subchannel, I S\ = s S,s s,nps u h s,j 2 s the nterference perceved by LTE-U user j U from other SBSs n the same unlcensed subchannel k SC u, and I W s the nterference produced from WAP. However, the work [9] shows that W-F presence affects neglgbly to the LTE-U system performance. So we can gnore the nterference generated by WAP to the LTE-U system from (3). Moreover, n a dense deployment, I S\ >> P u h,j 2,so R u,k,j wll be neglgble, and wll not provde any beneft of usng the unlcensed spectrum to the LTE-U users. Thus, to take the advantage from ths unlcensed band, SBSs can form a grand coalton [13], and allocate the unlcensed resources n the orthogonal fashon lke the lcensed spectrum. By dong ths, they can avod nter-operators nterference I S\, generated from other SBSs n the conflctng area. Assume the unlcensed subchannels are splted as SC u = SC u 1 SC u 2... SC u S where SC u = {1, 2,..., K u}, and SCu SC u j =,, j S and dvde among the SBSs based on ther QoS requrement gaps (dfference between total mnmum QoS requrements of the assocated users to the SBS and sum-rate n the lcensed spectrum). Now, consderng all of these n (3), the obtanable rate of LTE-U user j U n the unlcensed subchannel k SC u s shown as follows:,j = B u log 2 1+ yk,j P u h,j 2. (4) R u,k Smlar to the lcensed spectrum, f SBS Sneeds to allocate multple unlcensed subchannels to user j U, then the acheved rate of that user s as follows: R u,j(y )= k SC u σ 2 y k,jr u,k,j. (5) The total acheved rate of user j U n both lcensed and unlcensed spectrum s as follows: R,j (x, y )=R l,j(x )+R u,j(y ). (6) Thus, the sum-rate of SBS S s the total acheved rate over all the users U, whch s shown as follows: R (x, y )= j U R,j (x, y ). (7) Fnally, we have a set of users U U of SBS S who needs assstance from the unlcensed resources to meet ther QoS requrements. Every user j U posses at least on lcensed subchannel as t s necessary for CA. We are also assumng that every user equpment s capable enough for CA to meet t s QoS.

4 B. Data Rate of W-F User When unlcensed channel s fully utlzed by a WAP, then t can provde maxmum rate to ts users. In ths case, average throughput of each user v V w assocated wth WAP w W can be represented as follows: R max = R w V w. (8) where R w s the overall downlnk throughput of the WAP w W. Now, when all the SBSs use the same unlcensed band of WAP, the performance of WAP wll be affected only from SBSs. If we consder that each SBS n the conflctng regon acts just lke a WAP, then normalzed throughput for each WAP w W accordng to the study [14] s as follows: Rw mn P tr P s E[P ](S +1) 1 =. (9) (1 P tr )T σ + P tr P s T s + P tr (1 P s )T c where P tr =1 (1 τ) S+1 s the transmsson probablty of at least one SBS or WAP n a tme slot wth τ denotng statonary transmsson probablty of AP. P s s the successful transmsson on the channel wth P s = (S+1)τ(1 τ)s P tr and E[P ] represents the average packet sze. T σ s the duraton of an empty slot tme, T s presents the tme duraton of a successful transmsson, and T c llustrates average tme of a collson. So, average downlnk rate acheved by each user v V w of WAP w W s represented as follows: R mn = Rmn w. (10) V w So when SBSs want to use the same unlcensed band, they must have to provde an opportunty for accessng the channel by WAP so that t can provde an average throughput to ts user that les between R mn co-exstence. and R max for the sake of far C. Problem Formulaton We observe that R max s achevable when only W-F networks access the unlcensed band. But f WAP and SBSs are deployed n the same conflctng area, and work n the same unlcensed band, then W-F users wll get almost no access n the channel, and acheve an nsgnfcant data rate. So, for far coexstence of W-F and LTE-U systems, they need to share the tme slot n such a manner that WAP can mantan a mnmum data rate for ts users and SBSs can at least mprove some of the users QoS. As the LTE-U system manages the physcal resource n a centralzed manner rather than DCF of WAPs, SBSs need to decde approprate porton of tme to acheve far amount of throughput of each W-F user. When SBSs share τ [0, 1] tme slot to WAP then the achevable rate of W-F user, and LTE-U user are shown as follows: R (τ) =R max τ. (11) R,j (τ, x, y )=R l,j(x )+(1 τ) R u,j(y ). (12) So, the sum-rate of SBS Swhen t shares τ tme slot wth WAP s as follows: R (τ, x, y )= j U R,j (τ, x, y ). (13) Now our problem s confned wth unlcensed band to maxmze the the sum-rate of SBS after sharng τ-fracton of tme wth WAP whle mantanng QoS of most of the users. For ths, we need to develop an effcent spectrum allocaton scheme for each SBS to maxmze the utlty functon R (τ, y )= j U the unlcensed spectrum where U fxed. max y,τ s.t. C 1 : R (τ, y ), S C 2 : j U j U log 2 (1 + (1 τ) R u,j (y )) n y k,j 1, k SC u k SC u y k,j K u U, consderng x s C 3 : R,j (τ, x, y ) QoS,j, j U C 4 : y,j k {0, 1}, k SC u, j U C 5 : R mn R (τ) R max, v V w C 6 :0 τ 1. (14) Here, constrants C 1 tells that one unlcensed subchannel can be utlzed by at most one LTE-U user. The lmtaton of total resources n ths spectrum are represented by constrants C 2 for each SBS. QoS requrement of LTE-U users s mtgated by constrant C 3. Every element of allocaton vector y wll be ether 0/1 that s shown n constrant C 4. W-F users are beng protected by the constrant C 5. The optmzaton problem (14) s a Mxed Integer Non-Lnear Programmng (MINLP) problem, whch s NP-hard due to ts combnatoral property. III. DECOMPOSITION OF THE PROBLEM FOR SOLVING WITH NASH BARGAINING GAME AND HEURISTIC APPROACH Now we want to decompose the problem n (14) nto two sub-problems so that ndvdual one can be solved wth approprate technques. Frstly, wth fxed τ, unlcensed resources should be allocated to the users so that the system throughput can be maxmzed wth satsfyng some constrants, and as shown n the follows: max R (τ, y ), S y (15) s.t. C 1,C 2,C 3, and C 4. Secondly, wth fxed resource allocaton (that we get from (15)), the tme sharng problem between SBSs and WAP can be represented as follows: max τ R (τ, y ), S s.t. C 5, and C 6. (16)

5 Problem (14) Algorthm 1 τ-based Resource Allocaton for SBS 1: Input: U, SCu, QoS,R l Decomposton of (14) RA Problem (15) Tme Sharng Problem (16) Solve (15) by Alg. 1, y y No Convergence? Yes Output Solve (16) by Nash Barganng Game, Fgure 2: Soluton Process of the problem (14) Sub-problems (15) and (16) have same goal wth dfferent constrants, and they are nter-connected through the parameters τ and y. The soluton (y ) of the frst sub-problem (15) s used for solvng sub-problem (16). On the contrary, the soluton (τ) of the second sub-problem (15) s used to solve sub-problem (16) and t s contnued untl convergence. Ths soluton approach s shown n the Fg. 2. Now we solve the problems (15) and (16) wth the help of heurstc approach, and NBG respectvely. The detals of these processes are represented n the followng secton. A. Soluton of Problem (15) The problem shown n (15) s stll NP-hard and cannot be effcently solved. Therefore, we propose a heurstc algorthm as presented n Alg. 1 to solve the problem. The ntuton of ths heurstc algorthm s to allocate unlcensed resources n such a way that the sum-rate s maxmzed, and can meet the QoS requrements of as many users as possble. For that, the algorthm allocates a mnmum number of unlcensed physcal resources to mtgate QoS of users accordng to ther channel gan. Thus, lnes 3-7 are responsble for fndng the mnmum number of unlcensed subchannel requred to meet the QoS of users of each SBS. Lne 8 sorts users as the descendng order accordng to ther channel gan, and lne 9 reorder the elements of subchannel requrement vector accordng to the sorted user lst. Lnes are responsble for allocatng the subchannels to users based on the lst obtaned from lne 8. The complexty of the above heurstc algorthm s O(max( U 2, SC u )). B. Nash Barganng game-based Soluton of Problem (16) From the problem n (16), f we want to maxmze R (τ, y ) for each SBS Sthen t wll suppress the performance 2: Output: y 3: for each j U do 4: Calculate QoS gap by QG,j = QoS,j R,j l 5: Fnd achevable rate of user j for a sngle unlcensed subchannel wth the help of (4).e. R u,1,j =(1 τ)ru,1,j 6: Calculate mnmum number of subchannels requrement for user j by mscr,j u = QG,j /R u,1,j 7: end for 8: Sort users from U accordng to channel gan on descendng order 9: Reorder the elements of mscr u accordng to U 10: Set nsc = SC u,j =1and k =1 11: whle nsc > 0 do 12: f nsc >= mscr,j u then 13: whle mscr,j u > 0 do 14: Set y,k = U,j 15: Set mscr u,j = mscru,j 1 16: Set k = k +1 17: end whle 18: Set nsc = nsc mscr u,j 19: Set j = j +1 20: else 21: Set j = j +1 22: end f 23: end whle of W-F users. In ths subsecton, we wll fnd a wn-wn strategy for both SBSs and WAP. As the overall tme slots on the unlcensed spectrum are constraned, t s mpossble to maxmze the benefts of both the systems smultaneously. Therefore, we need to fnd an effectve unlcensed tme slot allocaton scheme to balance the beneft between SBSs and WAP. Now redefne the problem of (16) to balance the benefts of both SBSs and WAP as follows: max R S (τ, y)r w (τ) τ s.t. R mn τ 0 τ 1 R (τ) R max, v V w (17) where R S (τ, y) = S R (τ, y ), R w (τ) =V w R (τ), and τ 0 s the tme that s necessary for mantanng R mn rate for each W-F user v V w when WAP s only usng the channel. It s a mult-objectve problem. So we can use the barganng game to dstrbute tme resource (τ) farly among the players P = {S,w} and Nash Barganng Soluton (NBS) [15] method can be a good canddate for that. Let R be a closed and convex subset that represents the set of payoff allocatons that the players can acheve f SBSs share the tme slot wth WAP and d s the set of dsagreement payoffs. Therefore, the utltes of ths game are U w = R w (τ) Rw mn = V w (R max τ R mn ) and U S = S u,j = S log{r,j (τ, x, y ) j U j U

6 R,j l (x )} = S j U log{(1 τ)r,j u (y )} respectvely, n each tme slce. Now NBS can gve us a unque soluton concept [15] from the set of payoff R that satsfes the followng : P r (τ) =φ(r, d) argmax U p. (18) r R p=1 Hence, we need such a τ that wll maxmze the value of r(τ) wth fxed y n (18). If we denote that optmal sharng tme as τ then that value s shown n Theorem 1. Theorem 1. Wth a gven allocaton y, the optmal tme slot allocaton { for WAP by a gven set of SBSs s τ (α+β+1) (α+β+1) =max 2 2α(β+δ) where α = S U, β = S j U α, τ 0 } log R,j u (y), and δ = Rmn R max The proof of ths theorem s out of space n ths paper. C. Alternatve Sum-Rate Maxmzaton for LTE-U Coexstence For a fxed set of SBSs and WAPs (wth ther assocated users), we can fnd y and τ by usng the alternatve sum-rate maxmzaton approach that s shown n Alg. 2. Wth the gven τ, each SBS can allocate ts resource (y ) to get maxmum R by usng Alg. 1 (lne 5). Now wth the gven y and other nformaton, arbtrator can fnd τ t (lne 9). The process (lnes 5-10) contnues untl t reaches to convergence. Alg. 2 wll converge after a fnte number of steps, t tres to maxmze the objectve wth lmted resources as n each step. It wll converge to some local optmums, and mght not the global optmum.. Fgure 3: Convergence of Alg. 2 for dfferent number of SBSs n the same unlcensed band Algorthm 2 Alternatve Sum-Rate Maxmzaton for LTE-U system 1: Input: S,δ,τ 0 2: Output: y, S and τ 3: Intalze: t =0, τ t =0.5 4: repeat 5: Each S determnes y t by usng Alg. 1 wth τt 6: Each S determnes R (τ t, y t ) = j U log R,j u (y ) and sends R (τ t, y t ) and U to the arbtrator 7: Arbtrator determnes α = S U, β = S R (τ t, y t ) 8: t t +1 9: Arbtrator determnes τ t wth the help of Theorem 1 consderng α, β, δ and τ 0 10: Arbtrator nforms τ t to S 11: untl convergence IV. PERFORMANCE EVALUATION In ths secton, we evaluate the performance of the proposed mechansm n terms of average acheved rate, the percentage of unsatsfed users, and farness [16]. The man parameters used n ths smulaton are shown n Table II. All SBSs, WAPs, and Fgure 4: Comparson of per user average acheved rate users are dstrbuted unformly n the conflctng area of radus 150m. W-F network operates accordng to the IEEE n protocol n the 5GHz band wth the RTS/CTS mechansm, and the W-F parameters are smlar to those of [14]. SBSs also work n the same unlcensed band as wth WAPs besdes the lcensed spectrum. We assume that SBSs use SDL wth the help of CA when the QoS of applcatons are not satsfed wth the lcensed spectrum. For our smulaton, we use typcal QoS requrements of multmeda applcatons of [17] as shown n Table III. Unlcensed resource blocks are dvded among the SBSs based on SBSs QoS gap wth lcensed resources as ndcated n Algorthm 1. In ths evaluaton secton, we have compared the performance of the proposed method wth LTE-A, LTE-U wth no coalton ndcated as LTE-U(NC), LTE-U wth the random selecton of users from user lst renamed as LTE-U(Rand), and LTE-U wth bankruptcy game

7 Table II: Value of Smulaton Parameters Symbol Value Symbol Value S 5 U, 50 B l 180 khz SC l, 50 B u 180 khz SC u 100 P l, 21 dbm P u, 17dBm σ dbm G 33.5dB V w, w 5 W 5 α 3 Table III: QoS Requrements of Multmeda Applcatons Applcaton Mn Requrement (Kbps) HD vdeo streamng 800 Vdeo conferencng 700 VoIP 512 Audo streamng 320 Fgure 5: Comparson of unsatsfed users Fgure 7: Comparson of farness among SBSs n the proposed method Fgure 6: Comparson of farness [18] known as LTE-U(BG). Fg. 3 shows the convergence of the repettve algorthm (Algo. 2). It represents that the algorthm wll convergence after a fnte number teratons, and on average t converges after 85 teratons. For comparng the performance, we take 1000 runs of all the methods. In Fg. 4, we manfest the comparson of the per-user average acheved rate among dfferent methods. It shows that the per-user acheved rate of the proposed method s hgher than all other comparng methods. The Fg. 4 also shows that LTE-A and LTE-U(NC) produce an average acheved rate of Kbps, and t s less than 440Kbps n 40% of the cases. On the contrary, ths range s Kbps for LTE- U(Rand), Kbps for LTE-U(BG), Kbps for LTE-U(Proposed), and the average acheved rate s at least 500Kbps n 40%, 60%, and 95% of cases for LTE-U(Rand), LTE-U(BG), and LTE-U(Proposed) respectvely. Moreover, the proposed method acheves 13.69%, 13.69%, 2.47%, and 1.99% more rate on average than LTE-A, LTE-U(NC), LTE- U(Rand), and LTE-U(BG) respectvely. In Fg. 5, we reveal the comparson of unsatsfed users among dfferent methods. It shows that the percentage of unsatsfed users s less n the proposed method than other methods. It also shows that the medan of unsatsfed users are 57.36%, 57.36%, 40.60%, 57.34%, and 35.08% respectvely for LTE-A, LTE-U(NC), LTE-U(Rand), LTE-U(BG), and LTE- U(Proposed) respectvely, and the proposed method acheves 63.51%, 63.51%, 15.74%, and 63.45% better than LTE-A, LTE-U(NC), LTE-U(Rand), and LTE-U(BG) respectvely. In Fg. 6, we fnd the comparson of farness among dfferent methods. It shows that most of the farness scores for LTE- A and LTE-U(NC) resde between 70% 75.50% whereas the same scores for LTE-U(Rand), LTE-U(BG), and LTE- U(Proposed) fall wthn 75% 80%, 76% 82%, and 75.5% 81% respectvely. On average, ths farness score of the proposed method s 6.74%, 6.74% and 0.63% better than LTE-A, LTE-U(NC), and LTE-U(Rand) respectvely, and 1.20% lower than LTE-U(BG).

8 ACKNOWLEDGMENT Ths work was supported by Insttute for Informaton communcatons Technology Promoton (IITP) grant funded by the Korea government (MSIT) (No , Development of Access Technology Agnostc Next-Generaton Networkng Technology for Wred-Wreless Converged Networks) *Dr. CS Hong s the correspondng author. Fgure 8: Comparson of normalzed throughput of W-F user In Fg. 7, we show the comparson of farness among the SBSs when they use a sngle unlcensed channel n the proposed method. It reveals that most of ths farness scores resde between 77% to 81% for all the SBSs and the medan of these scores are almost same (78.50%) for all the SBSs. It ndcates that the dvson of unlcensed resources s far among the SBSs. In Fg. 8, we show the comparson of normalzed throughput of W-F user between the proposed method and basc lsten-before-talk (LBT) wth the varyng number of SBSs. It represents that the proposed method shelds W-F user better than basc LBT does n all cases. Wth the ncreasng number of SBSs, the outputs reduce for both the proposed method and LBT as t ncreases the competton among the APs. The proposed method can guarantee 69.59%, and 83.85% more throughput than basc LBT n 5 SBSs and 10 SBS cases correspondngly. Moreover, the proposed method protects W- F users better than basc LBT n more dense deployment. V. CONCLUSIONS In ths paper, we have tred to meet the QoS requrements of the users by augmentng unlcensed spectrum wth lcensed one n the LTE-A network known as LTE-U after takng care of co-exstence ssue wth WAPs. Here, we have solved the problem by utlzng the NBG, and heurstc algorthm. Smulaton results show that opportunstc use of the unlcensed spectrum n the proposed method can provde better per user average acheved rate, and user satsfacton than LTE-A, LTE-U(NC), LTE-U(Rand), and LTE-U(BG) methods. The same trend also follows for farness except for the LTE-U(BG). Moreover, we fnd that the proposed method can protect W-F users fur better way than basc LBT does. In the future, we wll try to meet the QoS requrement of all the users by desgnng the mechansm carefully. REFERENCES [1] Csco, Csco Vsual Networkng Index: Global Moble Data Traffc Forecast Update, , Whte Paper, [2] M. Chan, M. Z. Wn, and H. Shn, MIMO networks: the effects of nterference, IEEE Transactons on Informaton Theory, vol. 56, no. 1, pp , January [3] A. K. Barag, N. H. Tran, N. Km, and C. S. Hong, QoS Aware Collaboratve Communcatons wth Incentves n the Downlnk of Cellular Network : A Matchng Approach, n th Asa-Pacfc Network Operatons and Management Symposum (APNOMS), pp. 1-6, October [4] S. M. A. Kazm, N. H. Tran, T. M. Ho, and D. K. Lee, and C. S. Hong, Decentralzed Spectrum Allocaton n D2D Underlyng Cellular Networks, n th Asa-Pacfc Network Operatons and Management Symposum (APNOMS), pp. 1-6, October [5] ITU, IMT Vson Framework and overall objectves of the future development of IMT for 2020 and beyond, Rep. ITU-R M , [6] F. Lu, E. Bala, E. Erkp, M. C. Belur, and R. Yang, Small-cell traffc balancng over lcensed and unlcensed bands, IEEE Transactons on Vehcular Technology, vol. 64, no. 12, pp , December [7] Noka, Noka LTE for unlcensed spectrum, Whte Paper, June [8] 3GPP, 3GPP work program, avalable at: dynareport/ganttchart-level-2.htm [9] A. M. Cavalcante, E. Almeda, R. D. Vera, S. Choudhury, E. Tuomaala, K. Doppler, F. Chaves, R. C. D. Pava, and F. Abnader, Performance evaluaton of LTE and WI-FI coexstence n unlcensed bands, n 2013 IEEE 77th Vehcular Technology Conference (VTC Sprng), pp. 1-6, June [10] Z. Guan, and T. Meloda, CU-LTE: Spectrally-effcent and far coexstence between LTE and W-F n unlcensed bands, n IEEE INFOCOM The 35th Annual IEEE Internatonal Conference on Computer Communcatons, pp. 1-9, Aprl [11] R. Yn, G. Yu, A. Maaref, and G. Y. L, LBT based adaptve channel access for LTE-U system, IEEE Transactons on Wreless Communcatons, vol. 15, no. 10, pp , October [12] Y. Gu, Y. Zhang, L. X. Ca, M. Pan, L. Song, and Z. Han, Explotng student-project allocaton matchng for spectrum sharng n LTEunlcensed, n 2015 IEEE Global Communcatons Conference (GLOBE- COM), pp. 1-6, December [13] W. Saad, Z. Han, M. Debbah, A. Hjorungnes, and T. Basar, Coaltonal game theory for communcaton networks, IEEE Sgnal Processng Magazne, vol. 26, no. 5, pp , September [14] G. Banch, Performance analyss of the IEEE dstrbuted coordnaton functon, IEEE Journal on Selected Areas n Communcatons, vol. 18, no. 3, pp , March [15] M. Maschler, E. Solan and S. Zamr, Game Theory, Cambrdge Unversty Press, ch. 15, pp , [16] R. Jan, D.M. Chu, and W.R. Hawe, A quanttatve measure of farness and dscrmnaton for resource allocaton n shared computer system, Eastern Research Laboratory, Dgtal Equpment Corporaton, vol. 38, September [17] T. Q. S. Quek, G. de la Roche, I. Güvenç, and M. Kountours, Small Cell Networks: Deployment, PHY Technques and Resource Management, Cambrdge Unversty Press, June [18] B. O Nell, A problem of rghts arbtraton from the Talmud, Mathematcal Socal Scences, vol. 2, no. 4, pp , June 1982.