Qos-aware resource allocation for mixed multicast and unicast traffic in OFDMA networks

Transcription

1 RESEARCH Open Access Qos-aware resource allocaton for mxed multcast and uncast traffc n OFDMA networs Hu Deng 1,2*, Xaomng Tao 1 and Janhua Lu 1 Abstract Ths artcle focuses on the subchannel and power allocaton for mxed multcast and uncast traffc n wreless OFDMA networs, where the multcast data s dvded nto basc layer and enhancement layer data. Our goal s to maxmze the networ total throughput wth a total power constrant whle guaranteeng the mnmum rate requrements of both the uncast and mutcast traffc. A suboptmal allocaton algorthm s proposed, whch combnes a cost-based subchannel allocaton wth the tradtonal water fllng (TWF) and an advanced water fllng (AWF). The TWF s used for the subchannels allocated to satsfy the rate requrements of each uncast traffc and multcast traffc whle the AWF s used for the remanng subchannels. Besdes, we present an average SNR-based user selecton scheme whch selects a proper set of multcast users to serve when the mnmum rate requrements of all users can not be satsfed. Smulaton results show that our proposed algorthm can mprove the networ throughput and outage probablty compared wth other algorthms. 1 Introducton The next-generaton wreless networs are expected to provde varous broadband multmeda servces wth dverse qualty of servce (QoS) requrements. Orthogonal frequency dvson multple access (OFDMA) s a promsng technology of the next-generaton wreless broadband networs for ts hgh spectral effcency and flexble resource management. Currently, much attenton s pad to the uncast wreless OFDMA networs. In uncast OFDMA systems, dynamc resource allocaton explots multuser dversty gan by allocatng the subcarrers to the users wth good channel condtons to mprove system performance. In [1-11], resource allocaton method was proposed for the uncast streams n OFDMA systems. In [7], the resource allocaton problem was resolved n two stages. Frst, a suboptmal subchannel allocaton was proposed. Then, optmal power allocaton was done based on the pre-determned subchannel allocaton. Ths method of separatng subchannel allocaton and power allocaton s wdely used n the resource allocaton of OFDMA networs. Meanwhle, many multmeda applcatons such as Internet televson and vdeo conferencng are carred by * Correspondence: dengh7@gmal.com 1 Tsnghua Natonal Laboratory for Informaton Scence and Technology (TNLst), Tsnghua Unversty, Bejng 184, P.R.Chna Full lst of author nformaton s avalable at the end of the artcle the multcast transmsson. The 3GPP has defned multmeda broadcast multcast servces (MBMSs) for the unversal moble telecommuncatons system [12]. The 3GPP2 fnalzed the specfcatons of broadcast multcast servces (BCMCSs) n the 1xEV-DO system [13]. The moble-wmax system forum also supported multcast broadcast servce (MBS) [14]. In each tme slot, data are delvered to a sngle user n case of uncast transmsson, whle the nformaton s smultaneously delvered to multple users n case of multcast transmsson. Resource allocaton schemes for mxed multcast and uncast traffc need to be nvestgated. Seo et al. [15] developed a subchannel allocaton scheme that maxmzes the total uncast throughput whle guaranteeng the mnmum transmsson rate of multcast traffc n OFDMA networs. Lu et al. [16] proposed a dynamc subcarrer and power allocaton scheme for several multcast groups n OFDMA networs to maxmze the networ throughput gven the total power constrant. However, the scheme n [16] dd not consderng the mnmum rate requrement. The uncast group s regarded as a multcast group wth one user n [16]. In fact, uncast and multcast servces have dfferent QoS requrements. When both multcast and uncast servces exst, the QoS requrements for the two nds of servces should be consdered concurrently. Bae et al. [17] analyzed the affect of cell radus and the user number n a cell to the performance of the uncast 212 Deng et al; lcensee Sprnger. Ths s an Open Access artcle dstrbuted under the terms of the Creatve Commons Attrbuton Lcense ( whch permts unrestrcted use, dstrbuton, and reproducton n any medum, provded the orgnal wor s properly cted.

2 Page 2 of 1 and multcast transmsson schemes, and proposed a hybrd schedulng scheme whch selects the multcast or uncast transmsson n a slot accordng to SNR threshold values.in[18],thepowerallocatonformxeduncast and multcast servces s consdered. The optmzaton ams to maxmze the networ sum rate under the precondton that the subcarrer allocaton s predefned. Smlarly, the above wors [15-18] faled to guarantee the rate requrements of the multcast traffc and uncast traffc at the same tme. The conventonal multcast transmsson over the wreless channels suffers from the lmtaton problem that the multcast transmsson rate s decded by the transmsson rate of the worst-channel user n the multcast group. Ths problem lmts the gan whch can be acheved by utlzng the multuser dversty n OFDMA networs. To overcome ths problem, the herarchcal vdeo codng schemes such as H.264 and MPEG-4 [19,2] etc. whch decompose the vdeo contents nto layers, can be employed. Suh and Mo [21] proposed subcarrer allocaton and bt loadng for a sngle herarchc multmeda stream based on the assumpton that any combnaton of layers consstng of multcast data can be decoded at the recever. But t dd not consder the rate requrement. Kwac et al. [22] and Xu et al. [23] too consderaton of the multcast rate requrement and desgned subcarrer allocaton schemes for multcast servces employng layered vdeo codng n OFDMA networs. But they dd not consder the power allocaton. In [24], we developed subchannel and power allocaton for a sngle herarchc multcast traffc consderng the mnmum rate requrement. These wors [21-24] only consder multcast traffc. Our wor attempts to desgn a subchannel/power allocaton method for mxed multcast and uncast traffc n OFDMA networs to maxmze networ throughput whle guaranteeng the rate requrements of all traffc. We utlze the layered vdeo codng technque for the multcast traffc. More specfcally, we present a suboptmal algorthm whch combnes a cost-based subchannel allocaton () wth the tradtonal water fllng (TWF) and an advanced water fllng (AWF). Also, an average-snr user selecton scheme s proposed to exclude bad channel condton users to mprove system performance. Smulaton results wll show that the networ throughput by usng our proposed algorthm outperforms other algorthms. The rest of artcle s organzed as follows: In Secton 2, we present the system model and problem formulaton. In Secton 3, subcarrer/power allocaton schemes and user selecton method are proposed. Next, the smulaton results of our algorthm are showed n Secton 4. Fnally, concluson s drawn n Secton 5. 2 System model and problem formulaton Fgure 1 llustrates the downln mxed multcast and uncast OFDMA networ that we tae nto consderaton n ths artcle. Here, each orgnal multcast data s compressed nto a base layer data and an enhancement data at themultcastsourcenthenetwor.thebaselayerdata contans the most essental nformaton that provdes a mnmum level vdeo qualty. The enhancement layer data s delvered to a set of selected users n a multcast group wth better channel condtons. It provdes addtonal nformaton that enhances the vdeo qualty of the base layer n proporton to the amount of the receved data, under the condton that the requrement of the base layer s satsfed. The base staton (BS) wll carry out the resource allocaton and user selecton n a centralzed manner. We assume the channels change slowly and the channel gan s stable wthn a schedulng tme slot. The channel nformaton s sent bac from all users to the BS. BS can get all the nvolved perfect CSI to do adaptve subchannel/power allocaton and select sutable user set to serve. In our model, G downln traffc flows are transmtted to K users on L subchannels, where the traffc flows contans U uncast traffc flows and M multcast traffc flows. We assume that each user receves only one traffc flow at each tme. The user set s denoted by K = U un U1 mul UM mul contans all the users n the networ, where U un denotes the uncast user set and U mul ( =1,..., M) denotes the user set of multcast group. The symbol U mul represents the number of users ncluded n the th multcast group. The total system bandwdth s B and each subchannel have an equal bandwdth B/L. M-ary quadrature ampltude modulaton (MQAM) s adopted n our system. Wthout loss of generalty, we assume a fxed and the same BER for all users n all subchannels. The achevable transmsson data rate of user n subchannel l can be expressed as: Uncast sources 1 U Base Layer Enhancement Layer Multcast Sources 1 M Subchannel and Power Allocaton/ User Selecton Base Staton OFDM Channels Channel Feedbac Multcast group 1... Multcast group M Uncast User Uncast User U Multcast User Multcast User Fgure 1 Downln mxed multcast and uncast OFDMA networ.

3 Page 3 of 1 r l, = B L log 2 ( 1+ cp ) lg l,, (1) N where P l and G l, denote the transmsson power allocated to subchannel l and the channel gan for user n subchannel l, N s the nose power n each subchannel, c -1.5/ln(.2/BER) [25]. To mae the base layer data relably receved by all the multcast users n th multcast group, we tae the base layer transmsson rate of multcast group n subchannel l to be B l, = mn r l,. U mul The summed base layer transmsson rate of the subchannels whch are allocated to the base layer of a multcast group should meet the transmsson rate requrement of the base layer to provde the mnmum level vdeo qualty to users. For the enhancement layer of multcast group, we decde ts transmsson rate n sub-channel l such that the enhancement layer throughput n suchannel l be maxmnzed. The enhancement layer throughput of multcast group n subchannel l represents the amount of the receved enhancement layer data of all the users n multcast group n subchannel l. If the enhancement layer transmsson rate n subchannel l happens to be the rate of user n multcast group, then the enhancement layer throughput of multcast group n subchannel l s E l, = n U mul r l, 1(r l,n r l, ), U mul. 1(A) s an ndcator functon that becomes 1 when the condton A s met and otherwse. When the transmsson power n each subchannel s determned, the optmal transmsson rate for the enhancement layer of multcast group n subchannel l, U mul, U mul, s determned such that E l, can be maxmzed,.e., =argmax U mule l,. In ths case the maxmum enhancement layer throughput of multcast group n subcarrer l, E l,, becomes E l,. The optmzaton problem can be formulated as (P1): max p l,α l,β l,p l s. t. u U un ρ l,u r l,u + (2) (3) M (α l, U mul B l, + β l, E l, ) (4) =1 ρ l,u r l,u R mn u, u U un, (5) α l, B l, R b, =1,2,..., M, (6) P l P max, (7) ρ l,u + α l, + β l,j 1, l, u,, j, (8) ρ l,u + ρ l,v 1, l, u v, (9) α l, + α l,j 1, l, j, (1) β l, + β l,j 1, l, j, (11) ρ l,u, α l,, β l,j {, 1}, l, u,, j. (12) where r l, u s ether 1 or, dependng on whether the subchannel l s assgned to uncast user u or not, a l, explans whether the subchannel l s assgned to the base layer of the th multcast group, b l, j denotes whether the subchannel l s assgned to the enhancement layer of the jth multcast group. Constrans (5) and (6) guarantee the rate requrements of each uncast and mutlcast traffc where R mn u and R b represent the rate requrement of uncast user u and multcast group, respectvely. Constran (7) s the total power constrant where P max s the total power of the BS. Constrans (8-12) ensure that one subchannel can not be reused by a uncast traffc, the base layer of a multcast traffc, and the enhancement layer of a multcast traffc. The optmzaton problem (P1) s a NP-hard problem. Optmal allocaton n whch subchannels and power should be allocated jontly poses a prohbtve computatonal burden at the BS. There, low-complexty suboptmal algorthms are preferred for ts cost-effectve and delay-senstve mplementatons. Separatng the subchannel and power allocaton s a way to reduce the complexty, because the number of varables n the objectve functon can be reduced. Ths method of separatng subchannel allocaton and power allocaton s wdely used n the resource allocaton of OFDMA networs [7,16]. In the next secton, a low-complexty supoptmal allocaton scheme based on the above method s proposed. 3 Methods: Heurstc resource allocaton and user selecton scheme The proposed resource allocaton scheme s dvded nto two steps. In the frst step, the subchannels are assgned assumng that the BS s totalpowerp max s equally

4 Page 4 of 1 dstrbuted to each subchannel,.e. P l = P max /L. Ths assumpton s used only for the subchannel allocaton. Next, power allocaton s done based on the subchannel allocaton results. Subchannel allocaton Assume that the maxmum transmsson power of the BS s unformly dstrbuted among the subchannels. The problem (P1) s transformed as (P2): max p l,α l,β l s.t. u U un ρ l,u r l,u + M (α l, U mul B l, + β l, E l, ) (13) =1 ρ l,u r l,u R mn u, u U un, (14) α l, B l, R b, =1,2,..., M, (15) ρ l,u + α l, + β l,j 1, l, u,, j, (16) ρ l,u + ρ l,v 1, l, u v, (17) α l, + α l,j 1, l, j, (18) β l, + β l,j 1, l, j, (19) ρ l,u, α l,, β l,j {, 1}, l, u,, j. (2) Lemma 1. When R mn u, u U un and R b, =1,2,..., M are all zero, the problem P2 can be solved by { } M l =max max {r l, }, max {E l,}, (21) U un =1,...,M ρ l,u =1,u =argmax u U un {r l,u }, f M l =max u U un {r l,u }, (22) β l, =1, =argmax{e l, }, f M l = max {E l,}, =1,,M =1,,M (23) Proof. For multcast group, the base layer throughput U mul B l, = U mul mn U mul r l, = n U mul U mul ( mn r l, 1 r l,n mn U mul ) r l, = E b l,, where b = arg mn U mulr l,, then the enhancement layer throughput E l, =max U mul E l, E b l,. So the enhancement layer throughput of a multcast group s no less than the base layer throughput. Therefore, when the mnmum rate requrements of the traffc are all zero, the problem P2 can be smplfed as the followng: [ ] max ρ l,u r l,u + M β l, E l, (24) ρ l,β l u U un =1 s.t. ρ l,u + β l,j 1, l, u, j, (25) ρ l,u + ρ l,v 1, l, u v, (26) β l, + β l,j 1, l, j, (27) ρ l,u, β l,j {, 1}. (28) The subchannel can not be reuse. Suppose the lth subchannel s allocated to uncast user u. Assume that there exsts a multcast traffc that has E l, >r l, u.the total throughput can be mproved by reallocatng the subchannel l from uncast user u to mutlcast traffc. Smlar conclusons can be obtaned n other stuatons. Therefore, the subchannel l need to be assgned to the traffc wth the maxmum achevable throughput as the Equaton (21-23) showed. Accordng to the above lemma, wthout consderng the rate requrement, n order to maxmze the networ throughput, the optmal subchannel allocaton s to allocate each subchannel to the traffc whose achevable throughput n that subchannel s largest. There-fore, we can calculate the achevable throughput of the enhancement layer of each multcast traffc and the achevable rate of each uncast traffc n each subchannel, and then allocate each subchannel to the traffc wth maxmum achevable throughput ntally. After that, we can reallocate the subchannels to meet the rate requrement of each traffc. We hope that the total throughput reducton be mnmzed n each reallocaton and the number of reallocaton be ept as low as possble. Next, a cost functon s defned to determne whether a subchannel wll be reallocated to another traffc. Suppose that the achevable throughput of a subchannel s dstrbuted n such a pattern that t s comparatvely hgh for a specfc traffc whle beng comparatvely low for an other traffc. Then t would be more desrable to allocate that subchannel to the traffc wth hgher throughput. Furthermore, consderng to mnmze the total throughput reducton, a subchannel s more desrable to be assgned to a traffc whose achevable throughput has a small gap from the maxmum achevable throughput n that subchannel. Meanwhle, a subchannel needs to have a hgher probablty to be allocated to a traffc wth a larger remanng rate to acheve the rate requrement.

5 Page 5 of 1 Based on the above consderaton, the cost functon for a traffc can be proportonal to the decrease of overall throughput, and nverse proportonal to the traffc s achevable throughput and remanng rate to acheve the rate requrement. The cost functon s defned as: R unreq R mulreq [ max [ max R mn L R b L ] ρ l, r l,,, =1,..., U, (29) ] α l, B l,,, =1,..., M, (3) c l, = M l r l, r l, R unreq, for =1,..., U, (31) c l,+u = M l U mul B l, R mulreq B l,, for =1,..., M. (32) Fnally, we establsh a as follows: frst, we allocate each subchannel to the traffc wth the maxmum achevable throughput n t. Second, we determne a subchannel-traffc par that has the mnmum value of c l,, and assgn the subchannel to that traffc. Thrd, we exclude the selected subchannel from the set of subchannels, S, and repeat the second step untl the transmsson rate requrements for all traffc are met. We examne the complexty of proposed subchannel allocaton scheme. In our model, G downln traffc flows are transmtted on L subchannels, where the traffc flows contans U uncast traffc flows and M multcast traffc flows. To determne the enhancement layer transmsson rate of multcast group n each subchannel, the scheme needs to sort the user rate n the multcast group frst wth complexty of U mul log( U mul ). Then, we need to calculate U mul throughput correspondng to U mul dfferent transmsson rate of enhancement layer, and U mul comparsons are need to fnd the optmal rate to maxmze the enhancement layer throughput. The base layer transmsson rate can be found n the sortng process. Let Umax mul be the maxmum user number n multcast group. Therefore, at most L M (2Umax mul + Umul max log(umul max )) comparsons are needed to fnd the base layer transmsson rate and the maxmum enhancement layer throughput for all multcast groups n all subchannels. Besdes, we need to calculate LU transmsson rate for all uncast users n all subchannel. Then, L Gcomparsons are need to fnd the traffc wth maxmum achevable throughput n each subchannel. For the subchannel reallocaton procedure, we need to calculate at most L Gcosts and need at most L G comparsons to fnd the subchannel-traffc par that has the mnmum value costs n each loop. At most L loop are requred to chec whether the transmsson rate requrements for all traffc are met. The complexty s bounded by L 2 Gn the subchannel reallocaton stage. In a concluson, the subchannel allocaton scheme has a complexty of O(LMUmax mul log(umul max )+L2 G) whchslessthanthe complexty requred for the complete search over the problem space whch s O(G L ). Power allocaton After the power and subchannls whch are allocated to satsfy the rate requrement of a traffc are fxed, the TWF [26] could be done mmedately to maxmze the achevable throughput of the traffc. We do the TWF for each traffc separately and ths method can avod the mnmum requred rates not satsfed after new power allocaton. The soluton of TWF for uncast traffc u s {( P l =max μ un u N ) },, (33) G l,u where μ un u s solved by L ρ l,up l = Pu un,andpu un s the total power allocated to uncast traffc u to satsfy ts rate requrement. The soluton of TWF for multcast traffc s {( ) } P l =max μ mul u N G mn,, (34) l, where G mn l, μ mul s solved by L α l,p l = P B, = arg mn U mulg l, and P B s the total power allocated to the base layer of multcast traffc to satsfy ts rate requrement. Gven r l, u, a l,, b l, j, the total power assgned to satsfy the rate requrement of all traffc and the user whose rate has been chosen to be the transmsson rate of the enhancement layer n each subchannel for each multcast traffc from step 1, the optmzaton problem (P1) s transformed as (P3): max M ρ P l l,u r l,u + β l, E l, u U un =1 s.t. P l P max P l, l L (35) (36) where L denotes the set of subchannels whch are assgned to satsfy the rate requrements of all traffc, and L denotes the remanng subchannel set.

6 Page 6 of 1 The above problem (P3) can be transformed as (P4): ( K l B max P l L log 2 1+ P ) lg l (37) N s.t. P l P left, (38) where P left = P max - lîl P l, K l =1andg l = G l, u f subchannel l s allocated to uncast traffc u, U mul, U mul and g l = G l, f subchannel l s allocated to multcast traffc whle * s the user whose rate r l, * s chosen to be the transmsson rate for multcast traffc n subchannel l under the equal power allocaton assumpton. The soluton to the power allocaton problem (P4) can be found by usng the Lagrange multpler technque. We defne the Lagrangan functon as F = ( K l B L log 2 1+ P ) lg l N + λ (39) P l P left, where l s a Lagrange multpler and the soluton of problem (P4) can be obtaned by solvng F/ l =. Consequently, the transmsson power for each subchannel should satsfy F P l = ( Kl B L g l N + P l g l + λ ) =, l L. (4) The optmal amount of power P l allocated to subchannel l can be represented by {( Kl P l =max N ) },, l L. (41) λ g l The power allocaton n (41) satsfes P l = P left and we call t AWF. l n (41) s determned by substtutng all the P l nto the constrant equaton P l = P left. Fnally, we conclude the proposed resource allocaton scheme for mxed multcast and uncast traffc n Table 1. User selecton Snce dfferent users have dfferent channel condtons, some users may get a severely bad channel condton. In ths case, t may be not possble to meet the transmsson rate requrements of all users even by allocatng all the resource. Especally for the multcast group, a bad channel condton user restrcts the achevable transmsson rate of the base layer and may lead to the outage of the whole multcast group. Then we need to exclude some bad users and provde the servce to a lmted set of users wth a relatve good channel condtons. How to decde the user set to serve s a problem. Usng an exhaustve searchng can get a optmal user set. However, that would tae a hgh complexty. Here we present a average SNR-based user selecton (ASUS) method n Table 2 wth a reduced complexty. After the subchannel allocaton s done, the user selecton scheme excludes the user wth the worst average SNR n each multcast group whch dd not acheve the mnmum rate requrement. Once the ASUS algorthm determnes the set of users to serve, the subchannel allocaton scheme can be executed agan usng the set of selected users to maxmze the system throughput. The teraton s repeatng untl all the selected users acheve the rate requrement. In each teraton, the user selecton needs Table 1 The proposed resource allocaton scheme 1. Intalzaton: Set L = {1, 2,..., L}, R unreq = R mn 2. Allocate each subchannel to the traffc wth maxmum achevable throughput and set r l, u, b l, j accordng to (21-23) ntally. Calculate c l, for = 1,...,U + M, l L accordng to (29-32). Whle R unreq > ( U un ) or R mulreq 1) fnd a par [l, ] = arg mn,=1,,u+m, R mulreq = R b and Pleft = P max. 2) when * U, allocate subchannel l* to uncast user *, P un >,( =1,...,M) and L φ {c l, }. = Pun + P / max N, rl, * = 1, and update R unreq accordng to (29); when * >U, / allocate subchannel l* to multcast group (* -U), P B = P B + P max N,al, * = 1, and update R mulreq accordng to (3). P left = P max - P max /N. 3) Let L = L {l } and update c l, accordng to (31)(32) for l L. 3. 1) For each traffc, do calculate the assgned power for the sub-channels whch are allocated to satsfy the rate requrement by usng the TWF accordng to (33)(34). 2) If L φ, calculate the assgned power for subchannels n L by usng the AWF (41).

7 Page 7 of 1 Table 2 User selecton 1. Intalzaton: { Set U = m } L α l,mb l,m < R b m whch s the set of multcast groups whch dd not acheve the rate requrement. 2. Iteraton: Whle U ϕ for each m Î U 1) fnd a user u = arg mn u U mul SNR u, m SNR u =1 / L L P / lg l,u N. 2) exclude the user u* from set Um mul, set U m mul = Um mul {u }. end { do our proposed cost-based subchannel allocaton, reset U = m } L ρ l,mr l,m < R b m. end at most Umax mul comparson to select the worst average SNR user for each multcast group. 4 Smulaton results We conduct computer smulatons to evaluate the performance of our proposed algorthm. For comparson purpose, we also consder three other schemes: (1) subchannel assgnment algorthm proposed n [27] based on a unform power dstrbuton. Here we determne the prunng threshold for each multcast group n each subchannel such that the throughput of the enhancement layer can be maxmzed other than fxed the threshold as done n [27]; (2) our proposed based on a equal power allocaton; (3) the proposed resource allocaton scheme n Table 1 () wthout user selecton method ASUS. A sngle cell case wth several uncast users and two multcast groups s taen nto consderaton. All the users are unformly dstrbuted n the cell and request a rate-adaptve servce such as vdeo and audo servces. The cell radus s set as 3 m. A COST-WI propagaton model s adopted wth path loss L(d) = log 1 (d) whered s dstance n meters [1]. The frequency selectve fadng channel s a sx-path Raylegh model wth an exponental power profle. The number of subchannels s 32 and each subchannel s 18 Hz. The power spectral densty of AWGN s -144 db W/Hz. The BS avalable total power s 2 W and the desred BER s 1-4. Frstly, we set the number of uncast users to 4, and let one multcast group contans three users whle the other contans sx users. Fgures 2 and 3 show the average networ throughput and outage probablty versus dfferent mnmum rate requrements of uncast traffc. The rate requrement of multcast traffc s fxed as 16 bts/s. The outage probablty means the proporton of the users who do not acheve ts mnmum rate requrement. We observe that +ASUS outperforms other schemes. Comparng wth, +ASUS can mproved the networ throughput about 3% and decrease the outage probablty about 7%. Fgures 4 and 5 show the average networ throughput and outage probablty versus dfferent mnmum rate requrements of multcast traffc. The rate requrement of uncast traffc s fxed as 16 bts/s. We can fnd that + WF + ASUS also outperforms other schemes. On one hand, the superorty of +ASUA results from the waterfllng whch mproved the achevable rate of traffc. On the other hand, the superorty s due to the fact that t deletes some bad channel-condton multcast users to avod the outage of whole multcast group and saves more resource to the good-channel condton users. Fgures 6 and 7 show the average networ throughput and outage probablty when varyng the number of uncast users. The mnmum rate requrements of each uncast traffc and multcast traffc are set to 14 bt/s. We can fnd that +ASUS outperforms other schemes when varyng the numbers of uncast users. As the uncast user number ncreasng, the throughput of all the four Average system throughput (Mbts/s) ASUS Mnmum rate requrment of uncast traffc(mbts/s) Fgure 2 The average networ throughput aganst the rate requrements of uncast traffc.

8 Page 8 of Average system outage probablty ASUS Average system throughput (Mbts/s) ASUS Mnmum rate requrment of uncast traffc (Mbts/s) Fgure 3 The average networ outage probablty aganst the rate requrements of uncast traffc The number of uncast users Fgure 6 The average networ throughput aganst the number of uncast users. Average system throughput (Mbts/s) ASUS Mnmum rate requrment of multcast traffc (Mbts/s) Fgure 4 The average networ throughput aganst the rate requrements of multcast traffc. schemes gets lower and the outage gets hgher, manly snce the avalable resource to each user becomes less. Fgures 8 and 9 show the average networ throughput and outage probablty when varyng the settng of the user number n multcast groups. The mnmum rate requrements of each uncast traffc and multcast traffc are set to 18 bt/s. The user number s 3, 4, 5 n the frst multcast group, and s 6, 8, 1 n the second group, respectvely, correspondng to the settng ndex 1, 2, 3. For multcast group, both the achevable transmsson rate and the supported user number of the multcast group affect the multcast throughput. We can fnd that when the user number of the multcast group becomes larger, the networ throughput and outage probablty becomes greater under our user number settng of the multcast groups. The throughput ncrement manly because of the sharng nature of the multcast.4.35 Average system outage probablty ASUS Average system outage probablty ASUS Mnmum rate requrment of multcast traffc (Mbts/s) Fgure 5 The average networ outage probablty aganst the rate requrements of multcast traffc The number of uncast users Fgure 7 The average networ outage probablty aganst the number of uncast users.

9 Page 9 of 1 Average system throughput (Mbts/s) ASUS The ndexs of the multcast group user number settng Fgure 8 The average networ throughput when varyng the settng of the user number n multcast groups. transmsson. The outage probablty ncrement may result from the probablty that a user gets a bad channel condton gettng hgher when the user number ncreases. Also, +ASUS outperforms other schemes when varyng the numbers of multcast users. 5 Concluson Ths artcle consdered subchannel and power allocaton for mxed multcast and uncast traffc n OFDMA networs where the multcast traffc employng herarchcal vdeo codng scheme. Our goal was to maxmze the system throughput wth a total transmsson power constrant whle guaranteeng the rate requrements of all traffc. A was frstly presented. Then we ntroduced the tradtonal waterfllng for the subchannels allocated to satsfy the rate requrements of each uncast traffc and the base layer of each multcast traffc, and Average system outage probablty ASUS The ndexs of the multcast group user number settng Fgure 9 The average networ outage probablty when varyng the settng of the user number n multcast groups. proposed an advanced waterfllng method for the remanng subchannels. Besdes, an ASUS algorthm was developed to reduce the outage probablty when the rate requrements of all traffc can not be satsfed. Smulaton results show the system throughput and outage probablty mprovement over other algorthms by usng our proposed algorthm. Acnowledgements Ths wor was supported by NSFC of Chna (No , 61211, ). Ths artcle was presented n part at the IEEE 74th Vehcular Technology Conference, San Francsco, Unted States, 5-8 September 211. Author detals 1 Tsnghua Natonal Laboratory for Informaton Scence and Technology (TNLst), Tsnghua Unversty, Bejng 184, P.R.Chna 2 Electronc Technology Informaton Research Insttute, MIIT., Bejng 14, P.R.Chna Competng nterests The authors declare that they have no competng nterests. Receved: 16 September 211 Accepted: 12 June 212 Publshed: 12 June 212 References 1. CY Wong, R Cheng, K Letaef, R Murch, Multuser OFDM wth adaptve subcarrer, bt, and power allocaton. IEEE J Sel Areas Commun. 17, (1999). do:1.119/ W Rhee, JM Coff, Increase n capacty of multuser OFDM system usng dynamc subchannel allocaton, n Proc IEEE VTC 2 Sprng, Toyo, Japan, (2) 3. J Jang, KB Lee, Transmt power adaptaton for multuser OFDM systems. IEEE J Sel Areas Commun. 21, (23). do:1.119/jsac D Kvanc, G L, H Lu, Computatonally effcent bandwdth allocaton and power control for OFDMA. IEEE Trans Wrel Commun. 2, (23). do:1.119/twc G Song, Y L, Cross-layer optmzaton for OFDM wreless networs-part I, theoretcal framewor. IEEE Trans Wrel Commun. 4, (25) 6. G Song, Y L, Cross-layer optmzaton for OFDM wreless networs-part II: algorthm development. IEEE Trans Wrel Commun. 4, (25) 7. Z Shen, JG Andrews, BL Evans, Adaptve resource allocaton n multuser OFDM systems wth proportonal rate constrants. IEEE Trans Wrel Commun. 4, (25) 8. H Km, Y Ha, A proportonal far schedulng for multcarrer transmsson systems. IEEE Commun Lett. 9, (25). do:1.119/ LCOMM T Wang, L Vandendorpe, WSR maxmzed resource allocaton n multple DF relays aded OFDMA downln transmsson. IEEE Trans Sgnal Process. 59, (211) 1. L Vandendorpe, N Prasad, X Wang, A successve vonvex approxmaton algorthm for weghted sum-rate maxmzaton n downln OFDMA networs, n Proc Conf Inf Sc Systems, (Prnceton, USA, 28), pp T Wang, L Vandendorpe, Iteratve resource allocaton for maxmzng weghted sum mn-rate n downln cellular OFDMA systems. IEEE Trans Sgnal Process. 59, (211) 12. Introducton of the Multmeda Broadcast Multcast Servce (MBMS) n the Rado Access Networ (RAN); Stage 2 (Release 7), 28, 3GPP RAN, 3G TS V Interoperablty Specfcaton (IOS) for Broadcast Multcast Servces (BCMCS): (26), 3GPP2 TSG-A, A.S19-A v Moble WMAX, Part I, A techncal overvew and performance evaluaton, n WMAX Forum, Frederc, USA, 29 3 (26) 15. H Seo, S Kwac, BG Lee, Channel structurng and subchannel allocaton for effcent multcast and uncast servces ntegraton n wreless OFDM systems, n Proc IEEE Globecom 27, Washngton, USA, (27) 16. J Lu, W Chen, Z Cao, KB Lee, Dynamc power and sub-carrer allocaton for OFDMA-based wreless multcast systems, n Proc IEEE ICC 28, Bejng, Chna, (28)

10 Page 1 of SY Bae, Y Hong, DK Sung, Adaptve transmsson scheme for mxed multcast and uncast traffc n cellular systems. IEEE Trans Veh Technol. 58, (29) 18. Y Slva, A Klen, Power allocaton n mult-carrer networs wth uncast and multcast servces, n Proc IEEE ICC 27, Glasgow, Scotland, (27) 19. S McCane, M Vetterl, V Jacobson, Low-complexty vdeo codng for recever-drven layered multcast. IEEE J Sel Areas Commun. 15, (1997). do:1.119/ W L, Overvew of fne granularty scalablty n MPEG-4 vdeo standard. IEEE Trans Crc Syst Vdeo Technol. 11, (21). do:1.119/ C Suh, J Mo, Resource allocaton for multcast servces n multcarrer wreless communcatons, n Proc IEEE Infocom 26, Barcelona, Span, 1 12 (26) 22. S Kwac, H Seo, BG Lee, Sutablty-based subcarrer allocaton for multcast servces employng layered vdeo codng n wreless OFDM systems, n Proc IEEE VTC 27 Fall, Baltmore, USA, (27) 23. Y Xu, X Wu, J Lu, Cross-layer Qos schedulng for layered multcast streamng n OFDMA wreless networs. Wrel Pers Commun. 51, (29). do:1.17/s H Deng, X Tao, T Xng, J Lu, Resource allocaton for layered multcast streamng n wreless OFDMA networs, n Proc IEEE ICC 211, Kyoto, Japan, 1 5 (211) 25. S Chung, A Goldsmth, Degrees of freedom n adaptve modulaton: a unfed vew. IEEE Trans Commun. 49, (21). do:1.119/ TM Cover, JA Thomas, Elements of Informaton Theory, Wley, New Yor, (1991) 27. DS Baum, J Hansen, J Salo, An nterm channel model for beyond-3g systems: extendng the 3GPP spatal channel model (SCM), n Proc IEEE VTC 25 Sprng, Stocholm, Sweden, (25) do:1.1186/ Cte ths artcle as: Deng et al.: Qos-aware resource allocaton for mxed multcast and uncast traffc n OFDMA networs. EURASIP Journal on Wreless Communcatons and Networng :195. Submt your manuscrpt to a journal and beneft from: 7 Convenent onlne submsson 7 Rgorous peer revew 7 Immedate publcaton on acceptance 7 Open access: artcles freely avalable onlne 7 Hgh vsblty wthn the feld 7 Retanng the copyrght to your artcle Submt your next manuscrpt at 7 sprngeropen.com