Downlin Sheduling and Radio Resoure Alloation in Adaptive OFDMA Wireless Communiation Systems for User-Individual QoS Lu Yanhui, Wang Chunming, Yin Changhuan, and Yue Guangxin Abstrat In this paper, we address the problem of adaptive radio resoure alloation (RRA) and paet sheduling in the downlin of a ellular OFDMA system, and propose a downlin multi-arrier proportional fair (MPF) sheduler and its joint with adaptive RRA algorithm to distribute radio resoures among multiple users aording to their individual QoS requirements. The alloation and sheduling objetive is to maximize the total throughput, while at the same time maintaining the fairness among users. The simulation results demonstrate that the methods presented provide for user more expliit fairness relative to RRA algorithm, but the joint sheme ahieves the higher sum-rate apaity with flexible parameters setting ompared with MPF sheduler. Keywords OFDMA, adaptive radio resoure alloation, sheduling, QoS. I. ITRODUCTIO UTURE wireless and mobile ommuniation systems are F expeted to offer higher data rates, to support a large number of subsribers and to ensure the fulfillment of quality of servie (QoS) requirements, given the limited availability of frequeny spetrum and time varying hannels. Orthogonal Frequeny Division Multiple Aess (OFDMA) is a promising andidate beause it not only inherits OFDM s immunity to inter-symbol interferene and frequeny seletive fading, but also inreases multi-user diversity by ating the hannel fading as a hannel randomizer []. The signifiant performane gain of OFDMA system, inluding frequeny diversity and multi-user diversity gain an be ahieved by using adaptive radio resoure alloation (RRA) and sheduling algorithms [2-6]. The objetive of RRA is to maximize the sum-rate apaity (i.e. total throughput) of OFDMA downlin by alloating eah sub-arrier to the user with the best hannel in eah fading state, whih has been investigated in [2]-[4] under the onstraints of the user s bit error rate (BER) and minimum data rate requirements. The sheduling algorithm is intended to provide equal hane to transmit data for all ative users. Unfortunately, the RRA algorithm maximizing the total throughput, whih is alled as K&H algorithm in single arrier system [5], always yields unfairness. Instead, fair sheduling algorithm gives rise to the loss of the total throughput [5]. In multi-user system sharing a time varying hannel, a ompromise between fairness and high system throughput is the proportional fair (PF) sheduling [7], whih is extended to multi-arrier transmission systems by H.Kim and Y.Han [6]. However, the performane of multi-arrier proportional fair (MPF) sheduling algorithm is not evaluated in [6]. In this paper, we modify the MPF sheduler in [6] to assure the user-individual QoS requirements and propose a joint K&H/MPF algorithm whih is superior to both K&H and MPF taing into aount throughput and fairness simultaneously. It provides for user fairness with expliit QoS guarantees, as well as approahes to the maximum sum-rate apaity ahieved by K&H. The remainder of this paper is organized as follows. We first desribe the system model and the assumptions used by us in setion II. In setion III, the K&H and multi-arrier proportional fair sheduling strategies for OFDMA system are addressed. Then, we propose the joint radio resoure alloation and sheduling algorithm, K&H/MPF sheduler, with different QoS provisioning. Setion IV shows the simulation results whih illustrate the performane improvement of K&H/MPF over K&H and MPF. In setion V, a onlusion is remared. Manusript reeived Deember 9, 2005. This wor was supported by the Innovation Fund for Outstanding Sholar of He nan Provine (o. 042000000), the ational Siene Foundation of China (o.60472070) and the Key Projet of Chinese Ministry of Eduation (o.05035) Lu Yanhui, Yin Changhuan and Yue Guangxin are with Beijing University of Post and Teleommuniations, Beijing 00876, China (phone: 86-0-6228246-82, e-mail: luyh70@sohu.om, yin@bupt.edu.n, gxyue@bupt.edu.n ). Wang Chunming is with ational Digital Swithing System Engineering & Tehnologial R&D Center, Zhengzhou, He nan Provine, 450002, China II. SYSTEM MODEL Fig. shows the arhiteture of downlin OFDMA system for transporting multi-user traffi over time varying fading hannels. A ellular wireless networ is assumed, but a single ell is onsidered, where the interferene from other ells is modeled as baground noise. In order to overome frequeny seletive fading ausing inter-symbol interferene, the bandwidth of eah sub-arrier is 97
Data of User Data of User 2 Data of User K buffer buffer 2 buffer K hosen to be suffiiently smaller than the oherene bandwidth of the hannel, so it is reasonable to assume that base station transmits data over parallel, independent hannels to K mobile user terminals, eah of whih requires ertain QoS guarantees. Further, we assume that eah OFDM sub-arrier n, as pereived by user, is subjet to flat Rayleigh fading, path loss and shadowing with hannel gains α,n. The signals suffer from AWG noise, whose single-sided noise power spetral density level 0 is onsidered equal to unity (i.e. 0 =), for all sub-arriers and is the same for all users. The Base Station (BS) has a perfet nowledge of the Channel State Information (CSI) i.e. α,n, and the hannel is slowly variable so that the proposed alloation algorithm onvergene within the hannel oherene time. In addition, we assume that a fluid model for paet transmission, and the paets destined to different mobile users are put into separate queues as shown in Fig.. Using nown CSI, the radio resoure alloation module and sheduler in BS assign different sub-arrier to different user, and determine the number of bits/ofdm symbol to be transmitted on every sub-arrier. Let,n denote the numbers of bits assigned to the th user on the nth sub-arrier in one OFDM symbol, so,n must be integer in the set D={0,,2,,M}. As no more than one user is allowed in a sub-arrier of an OFDM symbol, it follows that for eah n, if, n 0,,n =0 for all. We define ρ,n as the indiator of alloating the nth sub-arrier to the th user. That is, ρ,n = when the nth sub-arrier is assigned to the th user, otherwise ρ,n =0. The required QoS of the th user is desribed by its minimum data rate requirement equal to R bits per OFDM symbol, target bit error rate BER and time delay onstraints τ in OFDM symbol. The transmission power alloated to the nth sub-arrier of the th user is expressed as: g (, BER ) P Base Station, n F F 2 F IFFT RRA &Sheduler, n = () 2 α, n Where g (,n, BER ) is the required reeived power with unity hannel gain for reliable reeption of,n bits per symbol, and it depends on the target BER of the th user. Obviously,,n is the funtion of the target BER, the transmit power P,n and hannel gains α,n. That is = f BER, P α (2) P/S Fig. Multi-user OFDMA downlin arhiteture ( ), n, n,, n User User 2 User K III. JOIT K&H/MPF ALGORITHM This setion is omposed of three ontents. Setion III-A desribes the radio resoure alloation sheme ahieving maximum throughput, whih is also referred as K&H algorithm in OFDMA system as in single arrier system. In setion III-B, the MPF sheduler is presented to guarantee the QoS requirements of different users. In setion III-C, we develop the joint K&H/MPF algorithm whih utilizes both multi-user diversity and frequeny diversity to ahieve maximum apaity gain and fairness among users, by oordinating K&H and MPF. A. Radio Resoure Alloation - K&H The goal of the K&H alloation algorithm is to find the best assignment of,n so that the total system throughput is maximized for given transmitted power onstrains. Mathematially, we an formulate the K&H problem as: max K ρ (3), n, n Subjet to, n, ρ, n = n= K P, P n = n= Where P max is the total power onstraint, and ρ,n = when the nth sub-arrier is assigned to the th user, otherwise ρ,n =0. This is a non-linear optimization problems and omputationally omplex. To redue the omplexity, we onvert it into a linear one by equally distributing the transmit power over the sub-arriers, sine the benefit of having adaptive power ontrol is marginal if an adaptive rate sheme is already implemented [8,9]. Sine the transmit power P,n is fixed and α,n is nown, if orresponding modulation sheme is used in eah sub-arrier depending on the number of bits assigned to it, the,n an be pre-alulated with definite target BER [0]. Based on the aforementioned analysis, the K&H algorithm an be simplified as seleting the user to transmit data in n th sub-arrier if the user s hannel gain α,n is the best or whose,n is the largest. That is = arg max (5) max, n {,..., K } From formula (2)-(5), it is shown that the K&H strategy maximizes the total system throughput in the absene of delay or QoS onstraints. In reality, there are lateny requirements, in whih ase the average throughputs over the delay time sale is the performane metri of interest. Inevitably, the K&H algorithm produes unfairness when some users hannel is muh stronger than the others on the average, whih may be starved. B. MPF Sheduling with QoS Provisioning In this setion, a proportional fair sheduler for OFDMA system, MPF, is proposed, whih is an extension of single-arrier PF algorithm. Its ey features are that it an aommodate multiple QoS lasses and has a tunable fairness level. (4) 98
We assume that eah user belongs to a QoS lass with parameters R, BER and τ, whih is the target rate in bits/ofdm symbol, target bit error rate and averaging window time in OFDM symbols respetively. The averaging window time reflets the target delay of the QoS-lass. The basi definition of MPF is that in eah OFDM symbol the n th sub-arrier pis the user to transmit data., n = arg max (6) {,..., K } R R where,n is the numbers of bits of the user an support on the nth sub-arrier under urrent OFDM symbol, and R is the average ahieved rate defined as the average data rate effetively reeived by user, whose unit is bits/ofdm symbol. It is neessary that R are normalized by their respetive target data rate R due to different QoS requirements of eah user. All R are updated at eah OFDM symbol, aording to the following rule: R R r {,..., K} = + (7) τ τ where r is the total numbers of bits transmitted by user in urrent OFDM symbol. { K} r = ρ,..., (8), n, n n= where ρ,n = when the nth sub-arrier is assigned to the th user, otherwise ρ,n =0. From formula (6)-(8), we observe that MPF sheduler an implement satisfatory fairness among users and taes into aount individual QoS requirements of eah user, but the total throughput is disounted ompared with K&H algorithm when users average SR are quite different. To squeeze out more apaity in this ase, a possible solution is to design a joint K&H/MPF sheduler, whih attempts to provide more frequeny diversity gain and multi-user diversity gain. C. K&H/MPF Strategy The ey idea designing K&H/MPF algorithm is that a apaity gain approahing to maximum throughput an be ahieved and fairness among users an be ensured simultaneously, if radio resoure alloation ooperates with paet sheduling. The algorithm is developed as follows: First, the sub-arriers are alloated to maximize the throughput (3) without onsidering the target rates and time delay onstraints. This yields an optimum solution orresponding to K&H alloation. Afterwards, the sub-arrier alloation is adjusted by MPF sheduler to satisfy the target rates and delay onstraints. By doing so, fairness among users is fulfilled. In the following, the algorithm is desribed in detail. ) Wor out the,n aording to the urrent hannel information and target BER requirements in given OFDM symbol assuming the assigned power to eah sub-arrier is equal. = f ( BER, P, α ), n, n, n 2) Alloate the sub-arriers for eah user aording to formula (5). That is seleting the user to transmit data in eah sub-arrier. = arg max, n {,..., K } When the nth sub-arrier is assigned to the th user, ρ,n =, otherwise ρ,n =0. This is K&H alloation whih ahieves the maximum throughput defined by formula (3). 3) Determine all r using formula (8), whih is the total numbers of bits transmitted by eah user in urrent OFDM symbol. r = { K} ρ,...,, n, n n= 4) Update the average data rate of user one by one. R R r {,..., K} = + τ τ 5) Examine whether R, the target data rate of eah user, has been approahed by average rate R. For user, if R R R δ, we onsider the user s rate requirement has been guaranteed; if ( R R ) R > δ, it indiates the user s average data rate is so high that the sub-arriers belonging to them need to be realloated to the users whose ( R R ) R > δ, i.e. whose target rate onstraints have not been satisfied yet. δ is an adjustable parameter, whih affets the number of realloation operations. For onveniene, we denoted the users whose R R R > δ as, and the sub-arrier owned by the users whose ( R ) R R > δ as n. 6) Realloate the sub-arriers mared as n among the users mared as employing MPF sheduler. That is seleting the user to transmit data for eah sub-arrier n., n = arg max {,..., K } R R In this step, R is initial value at the beginning of the urrent OFDM symbol, not the ones figured out by step 4). 7) Modify r and R aording to step 3) and step 4) for all users denoted as, respetively. The K&H/MPF algorithm is arried out on an OFDM symbol by symbol, until all QoS onstrains are satisfied. IV. SIMULATIO RESULTS To evaluate the performane of K&H, MPF and K&H/MPF algorithms, we have onduted simulations using system and hannel model desribed in setion II. We onsider a ell whih radius is 500m, and the users are distributed within the ell independently and uniformly. The path loss is onsidered as a zero-mean Gaussian variable in the logarithmi sale responsible for shadowing with a standard deviation 8dB, and the path loss exponent is 4. Furthermore, we assume that the QoS requirements of all users are idential, and the target rate 99
and target bit error rate are equal to 64bits/OFDM symbol and.0e-4, respetively. The averaging window time τ is mared as τ for all users, whih is a variable parameter. In the simulation, an OFDMA downlin system with 28 sub-arriers is onsidered. The veloity of users is very slowly so the impat of the Doppler frequeny shift is negligible. Fig. 2 and Fig. 3 respetively show the total throughput ahieved by different algorithms versus the number of users and average reeived SR in db, with τ=2 and δ=0.. In Fig. 2, the average transmitted SR in base station (BS) is set to 2 db. From Fig. 2 and Fig. 3, it is demonstrated that the total apaity gain implemented by K&H/MPF algorithm is obvious ompared with MPF, but it is always under the K&H alloation. Fig. 2 Total throughput ahieved by different algorithms versus the number of users with τ=2 and δ=0. Fig. 3 Total throughput ahieved by different algorithms versus the average transmitted SR in a ten-user system, with τ=2 and δ=0. Fig. 4 Fairness ahieved by different algorithms with average transmitted SR equal to 6dB Fig. 5 Fairness ahieved by different algorithms with average transmitted SR equal to 30dB Fig. 6 A part of Fig.5 with X label from 0.25 to 0.5, and Y label from 0 to.5 Fig. 7 Total throughput ahieved by different algorithms versus various τ in a ten-user system 00
To evaluate the fairness ahieved by different algorithms, we plot the pattern of 200 users in a ten-user system with τ=2 and δ=0. in Fig. 4 and 5, where eah dot is orresponded to the normalized throughput of one user and its distane from BS. All users are distributed within the ell independently and uniformly, and their veloity is equal to zero. The normalized throughput of the th user is defined by ( ) where, R and T, ( ) R R T =, R are the average throughput and target rate of user, respetively. Obviously, the fairness of algorithm is best, when the normalized throughput of all users is equal to. From Fig. 4 and Fig. 5, we observe that the normalized throughputs of users ahieved by K&H alloation are very differene on the basis of their distane from BS, but those of MPF and K&H/MPF are approahed to, espeially when the transmitted power of BS is high (see Fig. 5). To mae out the result more learly, a part of Fig. 5 with X label from 0.25 to 0.5, and Y label from 0 to.5 is shown in Fig. 6. From Fig. 4 to Fig.6, it indiates that MPF and K&H/MPF provide transmitting rate approximating to their target rate for all users impartially, and their fairness is expliitly superior to K&H alloation. Fig. 7 shows the impat of average window time τ and δ on total throughput ahieved by different algorithms in a ten-user OFDMA system, where the values of τ are 2, 3, 4, 5, 6, 0, 50 and 00 OFDM symbol, and the average transmitted SR in base station (BS) is 2 db. From Fig. 7, it an be observed that the total throughput inreases with τ for both MPF and K&H/MPF sheduler when δ is fixed. In fat, the larger time delay requirement user an tolerant, the more window time is, so τ should be set very small for no time delay onstrain traffi. Moreover, we find that the visible apaity gain an be obtained by K&H/MPF ompared with MPF when the value of δ is more than 0.. This is mainly beause the fewer the sub-arriers are realloated by MPF in the K&H/MPF sheduler with the larger δ, where MPF sheduler dereases the total throughput. ote that δ is denoted as delta in Fig. 7. The fairness of K&H/MPF algorithm is no differene with MPF sheduler, so the urves of fairness verse τ and δ are not shown in the paper. [2] Ying Jun Zhang, K.B.Letaief, Multiuser Adaptive Subarrier-and-Bit Alloation With Adaptive Cell Seletion for OFDM Systems, IEEE Transations on Wireless Communiations, Vol.3, o.5, Sept.2004 [3] M.Ergen, S.Coleri and P.Varaiya, QoS Aware Adaptive Resoure Alloation Tehniques for Fair Sheduling in OFDMA Based Broadband Wireless Aess Systems, IEEE Transations on Broadasting, Vol.49, o.4, pp.362-370, De.2003 [4] Zuang Shen, Multiuser OFDM Capaity Analysis with Partial Channel Information, Multiuser Wireless Communiations Course Projet, Fall 2002 [5] Dapeng Wu and Rohit egi, Downlin Sheduling in a Celluar etwor for Quality of Servie Assurane, IEEE Transations on Vehiular Tehnology, Vol.53, o.5, pp.547-557, Sept.2004 [6] Hoon Kim and Youngnam Han, A Proportional Fair Sheduling for Multiarrier Transmission Systems, IEEE Commun. Letters, Vol.9, o.3, pp.20-22, Marh 2005. [7] P.Bender, P.Bla, M.Grob, et al. CDMA/HDR: A Bandwidth Effiient High Speed Data Servie for omadi Users, IEEE Commun. Mag., Vol.38, pp.70-77, July 2000 [8] S.T.Chung and A.J.Goldsmith, Degrees of Freedom in Adaptive Modulation: A Unified View, IEEE Trans. Commun., Vol.49, PP.56-57, Sept.200 [9] A.J.Goldsmith and P.Varaiya, Capaity of Fading Channels with Channel Side Information, IEEE Trans. Inform. Theory, Vol.43, pp.986-992, ov.997 [0] J.G. Proais, Digital Communiation, 3 rd ed. ew Yor: MGraw-Hill, 995. V. COCLUSIO In this paper, we modify the MPF sheduler in [6] to assure the user individual QoS requirements and propose a joint K&H/MPF algorithm with several adjustable parameters, suh as τ and δ, whose influene on the performane of algorithm are examined arefully. The results show that K&H/MPF is flexible algorithm, whih provides more fairness for users ompared with K&H, as well as improves the apaity ahieved by MPF sheduler without impat on its fairness. REFERECES [] R.Berezdivin, R.Breinig, and R.Topp, ext-generation wireless ommuniations onepts and tehnologies, IEEE Commun. Mag., Vol.40, pp.08-6, Mar.2002 0