530 OURAL OF COMMUICATIOS, VOL. 5, O. 7, ULY 00 A Radio Resource Allocatio Algorithm for QoS Provisio i PMP-based Systems Pig Wag Broadbad Wireless commuicatios ad Multimedia laboratory, Key Laboratory of Embedded System ad Service Computig supported by Miistry of Educatio, Togi Uiversity, Shaghai, Chia. Email: pwag@togi.edu.c Liu Zu, Fuqiag Liu ad Yilig Wag Broadbad Wireless commuicatios ad Multimedia laboratory, Key Laboratory of Embedded System ad Service Computig supported by Miistry of Educatio, Togi Uiversity, Shaghai, Chia. Email: {ihaozuliu, liufuqiag, zagigi}@63.com Abstract Based o the studies o dowli resource allocatio i poit to multi-poit (PMP) mode i the 80.6e systems, a efficiet dowli resource allocatio algorithm with low complexity is proposed to maximize the system throughput, which taes the advatage of frequecyselective fadig property i orthogoal frequecy divisio multiple (OFDM) etwors. The proposed algorithm ot oly provides the traffic rate as high as possible, but also satisfies the miimum reserved traffic rate for o-real-time services. The simulatio shows that the proposed algorithm performs better i terms of assurig idividual QoS ad offerig fairess amog users at the cost of slight degradatio i throughput. Idex Terms radio resource maagemet, dowli resource allocatio, QoS, PMP, 80.6e I. ITRODUCTIO It is well ow that oe of the most challegig techical problems for future wireless systems is to provide various services to meet the requiremet of a icreasig umber of cosumers while esurig idividual quality of service (QoS). Owig to the scarce frequecy spectrum, resource allocatio has bee cosidered as oe of the ey techologies for icreasig utilizatio rate of the limited power ad spectrum i future wireless etwors. Accordig to the IEEE 80.6e stadard, the medium access cotrol (MAC) layer supports both poit to multi-poit (PMP) mode ad mesh mode [, ], betwee which the former is the primary topology structure. The dowli trasmissio operates o a PMP basis, ad data for MSs eed to be trasferred by a cetral BS via a wireless li. The PMP mode provides a comparatively high traffic rate, sice etwor resources ca be shared amog users. As a result, broadbad wireless access systems widely adopted the PMP mode i the last few years. The MAC layer i IEEE 80.6e etwor uses the coceptio of service flow, which provides uidirectioal Mauscript received auary 3, 00; revised April 8, 00; accepted May, 00. pacet trasmissio []. A service flow is characterized by a set of quality of service (QoS) parameters such as latecy, itter, ad throughput assuraces. Owig to the low traffic rate ad high error rate i a wireless li, as well as user's mobility, a QoS-guarateeig resource allocatio algorithm becomes oe of the ey techiques i broadbad wireless etwor. Cosequetly, to support various applicatios uder limited radio resources ad time-varyig chael, a dyamic resource allocatio, which ca achieve both higher system spectral efficiecy ad better QoS, has bee idetified as oe of the imperative tass i wireless commuicatio sice a few years ago. Letaief ad Zhag provide a overview of recet research o dyamic resource allocatio, especially for MIMO ad orthogoal frequecy divisio multiple (OFDM) systems [3]. They propose a algorithm to maximize data rate for a give power budget with a target BER. The proposed scheme i [4] adusts the power to meet the predefied delay requiremet, which is oe of the decisive factors to guaratee ed-to-ed delay. Ref. [5] proposes a efficiet subcarrier ad power allocatio algorithm, which formulates ecessary coditios of dowli schedulig for proportioal fairess i orthogoal frequecy divisio multiple access (OFDMA) systems. However, sice it cosiders power allocatio after subcarriers allocatio for assurig idividual QoS ad subsequet proportioal fairess (PF) allocatio, it adds a lot of complexity. As we ow, i a sceario with high SR, the performace of allocatig power evely to each subcarrier is almost equivalet to that of the classic optimal power allocatio algorithm, at the same time the complexity is reduced greatly. Ref. [6] utilizes a user ad coectio based scheme to improve trasmissio efficiecy ad guaratee quality of service. It schedules the user with the highest priority first ad the allocatio starts from the maximum deviatio chael, with the result that the QoS requiremet of the user with low priority may ot be met. The target QoS of resource allocatio i [7] correspods to a miimum user data rate, a target bit-error rate ad a maximum BER-outage 00 ACADEMY PUBLISHER doi:0.4304/cm.5.7.530-536
OURAL OF COMMUICATIOS, VOL. 5, O. 7, ULY 00 53 probability. Ref. [8] distributes subcarriers ad bits amog users based o their differet quality of service requiremets ad traffic type. Though they provide a efficiet resource allocatio for the users with differet traffic class, the system throughput may reduce greatly. I order to meet the requiremets of o-real-time pollig service (rtps) i IEEE80.6e etwor, a dowli resource allocatio algorithm is proposed to maximize the system throughput, which taes the advatage of frequecy-selective fadig property of OFDM etwors. What s more, the proposed Delta algorithm combies the ey QoS parameters of rtps ad assures the required miimum data rate for each user. Uder the circumstaces of usig o-real-time pollig service i PMP-based OFDM etwors, such items as subcarrier/bit allocatio statistics, throughput ad fairess idex are compared ad aalyzed by usig four differet resource allocatio algorithms. The four algorithms are Delta, MT [3], fixed modulatio ad code scheme (FMCS) algorithm based o 6 quadrature amplitude modulatio (QAM) ad maximum throughput cofied by miimum traffic rate (MTMR) [9]. The simulatio shows that though the proposed Delta algorithm sacrifices few throughput, it comes close to the best MT algorithm i terms of system throughput ad the best FMCS algorithm i terms of fairess idex. Ad Delta algorithm meets QoS demad of the miimum reserved traffic rate for all users at the same time. Besides, its complexity is reduced greatly, which maes it appropriate for applicatio i o-real-time busiess. The remaider of this paper is orgaized as follows. Sectio briefly describes the basic cosideratio of a adaptive radio resource allocatio algorithm based o OFDM system. I view of QoS characteristics of rtps defied i 80.6e protocol, a relevat adaptive allocatio algorithm i the multi-carrier system is discussed ad a dowli resource allocatio algorithm to maximize the system throughput is itroduced i sectio 3. Sectio 4 presets simulatio result for the proposed Delta ad other three algorithms. Fially, the paper is cocluded i sectio 5. II. BASIC COSIDERATIOS Power is oe of the importat resources i wireless commuicatio system. The system capacity directly depeds o BS s trasmittig power. I OFDM system, a scheduler eeds to provide the power allocatio of each subcarrier as well as subcarrier assigmet. The classic optimal power allocatio algorithm is called Water-fillig theorem [0], of which the basic idea is allocatig to the chaels of good quality as more power as possible, the oes of bad quality relatively less. However, accordig to iformatio theory, it is easily proved that allocatig the power to each subcarrier o average ca dramatically reduce the complexity of the resource allocatio ust at the cost of slight degradatio i system capacity. Assume that there are idepedet parallel Gaussia chaels, of which oise power are respectively. Accordig to Shao s formula, the capacity of chaels is C Max Pi P log ( ) where P deotes the sigal power of chael, ad satisfies the equatio of power limitatio P P. By the method of Lagrage multipliers, we ca get the power allocatio method which ca maximize the system capacity: P ( γ ) () where (x) represets positive fuctio, γ costat is P to be determied by (γ ). Thus the whole capacity of parallel Gaussia chaels equals to C ( ) log ( γ ) (3) From (), the less the oise power of each chael is, the more power each chael ca obtai ad vice versa. I order to simplify the power allocatio scheme, we assume that the power is high eough. Hece, P (γ ) retrogresses to: P (γ ) (4) The it is easy to get the equatio: P γ ( P ) (5) Substitutig (5) ito (3) yields the chael capacity: P C log ( ) (6) From (6), the chael capacities obtaied from average power allocatio algorithm ad water-fillig power allocatio algorithm are roughly equivalet at high SRs. As a result, i order to simplify the algorithm, it is geerally regarded that the chael capacity with the average allocatio algorithm approximately equals to the best allocatio result at high SRs. I OFDM system, we ca allocate the power to each subcarrier o average, ad the cosider the assigmet of carriers ad bits together, which ca reduce the complexity of adaptive allocatio algorithm to a large extet. Ⅲ. RESOURCE ALLOCATIO ALGORITHM BASED O MAXIMUM THROUGHPUT A. Algorithm Descriptio Geerally speaig, the target of a algorithm i multicarrier system is maximum throughput, or best fairess idex, or tradeoff betwee them. This paper focuses o maximizig system throughput ad guarateeig the miimum reserved traffic rate of o-real-time pollig service, ad oly dowli OFDM system is cosidered. Assume that the umber of users served by a BS i a cell () 00 ACADEMY PUBLISHER
53 OURAL OF COMMUICATIOS, VOL. 5, O. 7, ULY 00 is K, ad the umber of subcarriers available is. At trasmittig termial, data from K users will be mapped to subcarriers ad bits, i.e., differet subcarriers with correspodig bits will be allocated to differet users. A adaptive algorithm ca allocate resource for each subcarrier at trasmittig termial o coditio that chael state is always give. We defie ρ, as the occupyig idetifier of a subcarrier, whose value is whe the th subcarrier is occupied by the th user, otherwise ρ, equals 0. Besides, c, sigifies the umber of bits available for the th user o the th subcarrier. After iverse fast Fourier trasform ad addig cyclic prefix, sigal is trasmitted through the dowli. Geerally, whe the duratio of cyclic prefix is larger tha maximum multipath delay spread, iter sigal iterferece ca be elimiated ad the every subcarrier is supposed to be experiecig flat fadig i the chael. It is assumed that chael state iformatio is acquired through dedicated cotrol chael. With cyclic prefix removed ad fast Fourier trasform, users ca obtai their ow data iformatio from relevat subcarrier, accordig to subcarrier allocatio ad modulatio iformatio. Whe a system has a target bit error rate (BER), there exists a relatioship betwee BER ad c, as follows: c f BER, p, h ) (7), (,, where p, is the trasmittig power eeded whe allocatig the th subcarrier to the th user, h, is the magitude of chael gai, ad fuctio f idicates a mappig relatioship. For every OFDM subcarrier, there is: T K ρ c (8),, For user, the umber of bits trasmitted o this subcarrier is R c ρ (9),, Based o the aalysis metioed above, the problem that the proposed algorithm eeds to solve ca be cocluded as: K V Max ρ c (0) c,, ρ,,, while it should also submit to the followig restrait:, If ρ c r (),, ρ,the ρ 0 ' ', () c, ca be obtaied from (0), ad the miimum traffic rate for every user is cofied i (). From (), we ca see that a subcarrier ca be occupied by at most oe user. The target of the proposed algorithm is to obtai proper ρ, ad c,. Sice c, is a iteger ad has o-liear relatioship with ρ,, the resolutio of (0) belogs to o-liear problem. Here, ρ, is supposed to be the average value of trasmittig power, i.e., power is distributed evely, as a result that calculatio complexity is reduced ad sub-optimal solutio is achieved at the same time. Whe ρ, is fixed, c, ca be solved from (7). Firstly, we igore the restrait of (), ad tae oly the target of maximum throughput ito cosideratio. Accordig to (0), subcarriers are oly allocated to users who ca provide maximum throughput to the system. The, bit allocatio is adusted accordig to the miimum reserved rate of each user. The subcarriers occupied by QoS satisfied users eeds to be allocated to those usatisfied oes till QoS requiremets of all users are satisfied. However, modulatio order of some subcarriers must be set lower to satisfy QoS requiremets of all users durig the re-adustmet processig of subcarriers. What s more, adustmet of subcarriers i this algorithm is supposed to obey the followig rule, that is, always choosig the oe that least affects the system throughput. Delta, c * c,, c, (3) where Delta, is the criteria of udgmet whe adustig bits, ad c idicates the traffic rate available *, whe subcarrier is iitially allocated to user *. The flow chart of this algorithm is illustrated i figure. Iitializatio: Available Subcarrier Set {,...}, User Set {,...K}, First Processig User :, ρ c,, o r Solve Delta, [ belogs to available subcarrier set ] ' argmi(delta,) Whether it will affects its QoS whe the old user loses this subcarrier? R * C * r ' ', ' ' Yes Update Subcarrier Occupy Idicator: ρ ρ 0, ' *, ' ' Update Available Subcarrier Set: Delete ' Yes Figure. Flow chart of the Delta algorithm Processig the ext user, Update Available Subcarrier Set: Delete ' o 00 ACADEMY PUBLISHER
OURAL OF COMMUICATIOS, VOL. 5, O. 7, ULY 00 533 B. Descriptio of Compariso Algorithms Here the three traditioal resource allocatio algorithms used for compariso with the proposed algorithm are itroduced. Solutio : Every subcarrier utilizes fixed modulatio mode, that is, 6QAM. Ideally, each subcarrier carries 4 bits, thus a OFDM sigal carries 4 bits all together. We call this FMCS algorithm. Solutio : Adaptive resource allocatio targeted at maximum throughput, which has the same result as iitial allocatio of the proposed algorithm, is reputed as MT algorithm [3]. Solutio 3: The allocatio algorithm targeted at maximum throughput, toe ivolved ad restraied by miimum rate, with precedece udgmet criteria is simplified, is called as MTMR algorithm [9]. It ca be represeted as follows: at time slot t, allocatig subcarrier * to user. * arg max( T c ( t)),,... (4), Every time a subcarrier is allocated, users demad of miimum rate ad the umber of toes should be updated. For example, whe subcarrier is allocated, the umber of toes for user is updated as follows: T () t max{0, T ( t ) r c ()} t (5) where c ( t) c ( t represets the umber of, ) bits which are already allocated to user at time slot t, T (t) represets the service demad of the user, ad c, (t) is the real trasmissio capability of subcarrier allocated to user. simulatio, the radius of a cell is 000m, the maximal trasmittig power of a BS is 43dbm ad the system target BER is set to 0-4. For the service model, assume that each user has oly oe rtps coectio, of whose pacet the size rages from 65bit/symbol to 90bit/symbol. Table shows the parameters i detail. B. Results ad Discussio At first the fairess idex amog users of the four algorithms (Delta, MT, FMCS ad MTMR) is aalyzed. Here the Mi/Max fairess idex is used to measure the fairess of differet algorithms. Its defiitio is as follows: mi{ si} F (6) max{ s } where F represets the fairess idex, Si deotes the throughput achieved by user i. Figure illustrates the relatioship betwee Mi/Max fairess idex ad the umber of users by usig the four differet algorithms. i Ⅳ. SIMULATIO AD AALYSIS I order to evaluate the performace of the proposed allocatio algorithm, the simulatio sceario maily focuses o o-real-time pollig service (rtps). The proposed Delta algorithm will be compared with other three algorithms (i.e., FMCS, MT ad MTMR) i fairess idex amog users, allocatio of subcarriers ad bits, ad system throughput. A. Simulatio Eviromet TABLE I. PARAMETERS I SIMULATIO Parameter Value The umber of users (K) 6 ~ Users The umber of subcarriers () 56 Modulatio mode allowed BPSK / QPSK / 6QAM / 64QAM Target BER (BER) 0e-4 Service model rtps QoS property: Mi traffic rate (bits/symbol) Uiform distributio over the iterval (65, 90) This simulatio sceario cotais 56 subcarriers for data trasmissio ad supports four modulatio techiques icludig BPSK, QPSK, 6QAM ad 64QAM, where the highest modulatio order is six. I Figure. Relatioship betwee Mi/Max fairess idex ad the umber of users I simulatio, the umber of users chages from 4 to. As show i figure, with the umber of users i a cell icreasig, the Mi/Max fairess idex of the four algorithms all treds to risig. Especially, for FMCS algorithm, whe the user umber goes up to a certai value (K6 i this simulatio), the fairess idex reaches. This is because ot oly it uses a fixed modulatio, but also its resource allocatio is similar to the pollig of pacet schedulig algorithm. Thus the fairess idex is supposed to be the highest. The fairess of the other three asceds steadily. Comparatively, the fairess idex of MTMR algorithm is slightly lower tha Delta ad MT algorithms, by % uder the worst circumstaces (K4-6). The proposed algorithm, Delta algorithm, is rather equivalet to MT algorithm i system fairess, but whe the user umber is risig (K>8), Delta algorithm shows superior performace. 00 ACADEMY PUBLISHER
534 OURAL OF COMMUICATIOS, VOL. 5, O. 7, ULY 00 Through several simulatios, the statistics idicate that Delta algorithm ad MT algorithm perform approximately the same i scearios whe users are relatively few. This is because every user i this sceario ca be met QoS demad of miimum traffic rate, that is to say, there are always eough subcarriers to allocate ad thus the adustmet of bits allocatio for QoS provisio is uecessary. However, Delta algorithm reaches a better result for bit allocatio whe the umber of users icreases to a certai amout. Figure 3 illustrates the umber of bits per symbol allocated to each user i a sceario of 0 users. The results of Delta algorithm ad MT algorithm hardly differ from each other. What s more, both of them meet QoS demad of miimum traffic rate. Theoretically, if QoS demad of each user differs, FMCS algorithm, which focuses o fairess idex amog users, will igore the restrait of QoS provisio. Therefore, whe adoptig FMCS algorithm, the traffic rates of user 4 to user 9 are a little bit lower tha the required miimum traffic rate. The statistics acquired by usig MTMR algorithm fluctuate largely. For example, the traffic rates of user ad user 4 are comparatively high, while those of user 3 ad 0 are too low to satisfy the requiremet of miimum traffic rate. The reasos why the result of subcarriers allocatio differs so much are mutual ifluece durig the allocatio of adacet subcarriers, as well as the update of the umber of toes. users who have ot obtaied the miimum reserved traffic rate o the premise of maximizig the system throughput. Figure 4. umber of bits per sigal allocated to users i 5 users sceario Figure 5 shows system throughput i sceario of 0 users. As FMCS algorithm requires special modulatio mode, it gets a relatively small throughput, approximately 00 bit/symbol. Figure 5 also proves the coclusio metioed before, that the performaces of Delta algorithm ad MT algorithm are quite the same whe the umber of users i the sceario is small. However, the throughput of MTMR is slightly smaller tha that of Delta algorithm. Figure 3. umber of bits per sigal allocated to users i 0 users sceario Figure 4 shows the umber of bits per symbol allocated to each user i 5 users sceario. The result of MTMR algorithm still fluctuates largely. Whe the user umber is icreasig, there are some users experiecig hugry by usig whether MTMR, FMCS, or MT algorithm, whose target is to maximize the system throughput. While Delta algorithm, a improvemet of MT algorithm, allocates fewer bits to certai users tha MT algorithm, it satisfies QoS requiremets of miimum traffic rate of all users. I additio, Delta algorithm has geerally achieved as high traffic rate as possible, this is because Delta algorithm is desiged to adust allocated subcarrier appropriately to meet the requiremets of the Figure 5. System throughput i 0 users sceario Figure 6 idicates system throughput i sceario of 5 users. The tedecy is early the same as that of 0 users sceario. ote that the performaces of Delta algorithm ad MT algorithm are both better tha that of MTMR algorithm. 00 ACADEMY PUBLISHER
OURAL OF COMMUICATIOS, VOL. 5, O. 7, ULY 00 535 Figure 6. System throughput i 5 users sceario Ⅴ. COCLUSIO Based o MT algorithm, this paper proposes a ew resource allocatio algorithm with low complexity, which ot oly provides the maximum traffic rate as much as possible, but also guaratees the miimum reserved traffic rate for o-real-time services. Through simulatio, the proposed Delta algorithm is compared with MT, FMCS ad MTMR algorithms i fairess idex amog users, allocatio of subcarriers ad bits, ad system throughput. The simulatio idicates that Delta algorithm, as a promotio of MT algorithm, achieves QoS provisio at the cost of slight degradatio i throughput, at the same time does well i fairess idex amog users too. Particularly, i sceario with a relatively large umber of users (above 0 i this simulatio), Delta algorithm exceeds over the other three i compariso. ACKOWLEDGMET This wor was supported i part by the atioal atural Sciece Foudatio of Chia uder grat (o. 706300), the atioal Sciece ad Techology Maor Proect of Chia uder Grat (o. 009ZX03003-005 ad o. 008BAH30B0) ad program for Youg Excellet Talets i Togi Uiversity (o. 08009099). REFERECES [] IEEE Std. 80.6-004. IEEE Stadard for Local ad Metropolita Area etwors-part 6: Air Iterface for Fixed Broadbad Wireless Access Systems, Oct. 004. [] IEEE Std. 80.6e-005. IEEE Stadard for Local ad Metropolita Area etwors-part 6: Air Iterface for Fixed ad Mobile Broadbad Wireless Access Systems, Feb. 006. [3] K. B. Letaief, Y.. Zhag, Dyamic multiuser resource allocatio ad adaptatio for wireless systems, IEEE Trasactios o Wireless Commuicatios, Vol. 3(4), pp. 38-47, Aug. 006. [4] G. X. Xu, Y. i,. H. Zhag ad P. Zhag, Adaptive OFDMA subcarrier assigmet for QoS guarateed services, IEEE 6st Vehicular Techology Coferece (VTC 005-Sprig), Vol. 3, pp. 87-80, May 005. [5] T. D. guye ad Y. Ha, A Proportioal Fairess Algorithm with QoS Provisio i Dowli OFDMA Systems, IEEE Commuicatios Letters, Vol. 0(), pp. 760-76, ov. 006. [6] X. Zhu,. Huo, X. Xu, C. Xu, W. Dig, QoS-Guarateed Schedulig ad Resource Allocatio Algorithm for IEEE 80.6 OFDMA System, IEEE Iteratioal Coferece o Commuicatios (ICC08), pp. 3463-3468, May 008. [7] A. Alsawah ad I. Fialow, Practical Radio Li Resource Allocatio for Fair QoS-Provisio o OFDMA Dowli with Partial Chael-State Iformatio, EURASIP oural o Advaces i Sigal Processig, Vol. 009, pp. -6, 009. [8] Y. Lu, T. Luo, C. Yi ad G. Yue, Adaptive radio resource allocatio for multiple traffic OFDMA broadbad wireless access system, The oural of Chia Uiversities of Posts ad Telecommuicatios, Vol. 3(4), pp. -6, December 006. [9] G. X. Xu, A study o resource allocatio strategy i the ext geeratio of mobile commuicatios, Doctor thesis of Beiig Uiversity of Posts ad Telecommuicatios, pp. 57-86, 004. [0] S. T. Chug ad A.. Goldsmith, Degrees of freedom i adaptive modulatio: A uified view, IEEE Trasactios o Commuicatios, Vol. 49, pp. 56-57, Sept. 00. Pig Wag, bor i Chia, 978--8. He graduated from the departmet of computer sciece ad egieerig at Shaghai iaotog Uiversity, Chia ad received Ph. D. degree i 007. His maor field of study is wireless commuicatio. He oied the college of electroic ad iformatio egieerig at Togi Uiversity i 007 ad ow is a lecturer. He has published several papers i Computer Commuicatios, Wireless Persoal Commuicatios, IEICE Trasactios o Iformatio ad Systems, IEICE Trasactios o Commuicatios, etc. His curret ad previous iterests iclude routig algorithms ad resource maagemet i wireless etwors, vehicular ad hoc etwor ad video trascodig. Liu Zu, bor i Chia, 986-8-5. He graduated from the departmet of iformatio ad commuicatio egieerig at Togi Uiversity ad received B.S. degree i 008. His maor field of study is wireless commuicatio. ow he is is a graduate i the departmet of iformatio ad commuicatio egieerig at Togi Uiversity. His mai research iterests are i mobility modelig ad radio resource maagemet for the ext geeratio mobile commuicatios. Fuqiag Liu, bor i Chia, 963-3-7. He graduated from the departmet of automatio at Chia Uiversity of Miig ad received Ph. D. degree i 996. His maor field of study is sigal processig. ow he is a professor i the departmet of iformatio ad commuicatio egieerig at Togi Uiversity. His mai research iterests are i routig algorithms i wireless broadbad access ad image maipulatio. 00 ACADEMY PUBLISHER
536 OURAL OF COMMUICATIOS, VOL. 5, O. 7, ULY 00 Yilig Wag, bor i Chia, 984-7-4. She graduated from the departmet of iformatio ad commuicatio egieerig at Togi Uiversity ad received B.S. degree i 006. His maor field of study is sigal processig. ow she is a graduate i the departmet of iformatio ad commuicatio egieerig at Togi Uiversity. Her mai research iterests are i cross-layer resource allocatio i OFDM systems. 00 ACADEMY PUBLISHER