A QoS-Aware Scheduling Algorithm Based on Service Type for LTE Downlink

Transcription

1 Proceedngs of the nd Internatonal Conference on Computer Scence and Electroncs Engneerng (ICCSEE 13) A QoS-Aware Schedulng Algorthm Based on Servce Type for LTE Downlnk Wehong Fu, Qnglang Kong, Wexn Tan, Cheng Wang, Ll Ma State Key Lab. Of Integrated Servce Networks, Xdan nversty X an, Chna e-mal: whfu@mal.xdan.edu.cn Abstract In ths paper, a resource schedulng algorthm for Long Term Evoluton (LTE) downlnk s proposed. It provdes Qualty of Servce (QoS) guarantee to Guaranteed Bt Rate (GBR) servces and also sut to the Non Guaranteed Bt Rate (Non-GBR). Dfferent servces possess dfferent QoS needs n LTE system, based on ths, 3GPP dvde them nto several types. In vew of the dvson, we set prortes for dfferent knds of servces to guarantee the transportaton of GBR servces whch have hgh QoS needs. At the same tme, we take the Meda Access Control (MAC) layer logcal channel prorty nto account, and consder users channel qualty to acqure a hgher cell throughput. Smulaton results show that our scheme can mprove the QoS of emergency servces obvously at the cost of a certan amount of the non-real tmes. Keywords-resource schedulng; GBR; Non-GBR; QoS; LTE I. INTRODCTION In recent years, there appear some knds of communcaton systems usng Orthogonal Frequency Dvson Multplexng (OFDM), for example, WMAX and 3GPP LTE. They support dfferent servces and can acheve hgh throughput. On the other hand, these systems are able to mprove cell throughput further by means of combng wth technques such as Multple Input Multple Output (MIMO) and Hybrd Automatc Repeat request (HARQ). However, ntroducton of these new technques makes hgher requrement to the resource schedulng algorthms n MAC layer. Tradtonal schedulng algorthms, ncludng Maxmum Carrer to Interference (max C/I), Round Robn (RR) and Proportonal Farness (PF) [1][], and ts mproved algorthms [3][4] have a common drawback. Non of them take QoS needs of user s servces nto consderaton. Sometmes there are many users n the system and Non-GBR servces need too much resource, n ths case, ths knd of algorthm can t guarantee the performance of GBR servces. There exst some algorthms consderng user QoS for Orthogonal Frequency Dvson Multple Access (OFDMA) system. One s Earlest Deadlne Frst (EDF) [5], user wth maxmum packet delay has the hghest prorty, and ths algorthm does best n packet delay and loss rate. Whereas, t can nether guarantee system throughput nor sut to non-realtme servces. Another s Modfed Largest Weghted Delay Frst (M-LWDF) [6] and ts modfed algorthms [7] [8]. Ths knd of algorthm adds packet delay to PF, and combnes t wth channel qualty nformaton. Nonetheless, they nether consst of dfference between logcal channel prortzaton nor apply to non-real-tme servce. Therefore accordng to lmtaton of the above methods, we propose a novel schedulng algorthm. It combnes servce QoS and MAC logcal channel prortzaton wth user channel qualty ndcaton and farness between users. It apples to not only GBR bearers but also Non-GBR bearers. The paper s organzed as follows. In Secton II, t provdes the system model of resource schedulng n LTE system. Secton III descrbes the proposed resource allocaton algorthm. Secton IV presents the smulaton results and performance analyss. Fnally Secton V concludes ths paper. II. SYSTEM MODEL Packet schedulng s to manage and schedule servces accordng to ther characterstcs. When there are several packet streams watng to be served, we should determne reasonable servce rules to arrange the servce order, servce tme and transmsson rate to satsfy ther QoS needs, ths s the basc ssue of packet schedulng. In LTE system, scheduler s the functonal entty to acheve packet schedulng. To match dfferent wreless channel data rate quckly, and meet the needs of dfferent knds of hgh speed packet servce, t has been moved from Rado Network Controller (RNC) sde n WCDMA to evolved NodeB (enode B) sde n LTE [11]. A. Scheduler n LTE Scheduler s the functonal entty to acheve fast packet schedulng and the core functon of MAC layer n LTE enode B [1]. It has a major mpact on the performance of the system. It contans uplnk scheduler and downlnk scheduler, whch take charge of allocatng physcal layer resources for plnk Shared Channel (L-SCH) and Downlnk Shared Channel (DL-SCH) respectvely. Physcal layer resources manly consst of Physcal Resource Block (PRB), Modulaton and Codng Scheme (MCS), power allocaton schemes and antenna selecton schemes n the multple antenna case [9]. Fg. 1 shows the poston of scheduler n LTE protocol and ts relatonshp wth key technques lke HARQ and Adaptve Modulaton and Codng Publshed by Atlants Press, Pars, France. 1

2 Proceedngs of the nd Internatonal Conference on Computer Scence and Electroncs Engneerng (ICCSEE 13) (AMC). Its key functon s to determne how to allocate resource to users properly for the purpose of maxmzng system throughput, on condton that ensure packet delay and QoS. stuaton, we set a threshold α, called percentage of the resource that GBR servces can utlze. It sgnfes the permsson of GBR servces. When α s large, t means that there wll be more resource that GBR servces can use, and ther QoS needs s able to be satsfed suffcently; whle α s small, the stuaton s reversed. α s an experence value, and s changeable. It can be set based on the number of users and percentage of GBR servces n the system. In the stuaton that there are no GBR resource whle there stll exst GBR servces, we stop servng for them. If all of GBR servces have been served before GBR resource s exhausted, the rest belong to the non-gbr. The whole technologcal process s llustrated n Fg.3. Fgure 1. Poston of scheduler n LTE protocol B. Downlnk Schedulng Process n LTE As mentoned, channel qualty s necessary n schedulng. enode B obtans t by recevng feedback of channel qualty ndcator from user. As shown n Fg., downlnk scheduler allocates resource for users accordng to ther channel qualty, QoS needs and farness. enode B sends date and schedulng sgnalng accordng to the schedulng results. Scheduled users accept data n the lght of RBs allocaton results, MCS and antenna select optons n downlnk schedulng sgnalng. Fgure 3. Block dagram of the algorthm We wll ntroduce our resource schedulng algorthm n two steps. Fgure. Downlnk schedulng process III. DESCRIPTION OF RESORCE SCHEDLING ALGORITHM The resource allocatng algorthm that we wll descrpt accomplshes the schedulng functon n fgure. In the proposed algorthm, GBR servce s defntely pror to Non- GBR servce, and resource allocaton for Non-GBR servce s behnd of GBR. Occasonally, there may be many users n the system and these who have terrble channel qualty occupy too much resource to transport GBR data, thus lead to the reducton of system throughput. To prevent the appearance of ths A. Resource Allocaton of GBR Servces As before, f there are GBR servces, we allocate resource for them preferentally. If all of the GBR avalable resource s exhausted, then serve Non-GBR servces n turn. When allocatng resource for GBR servces, we should take nto account users channel qualty, packet delay and logcal channel prortzaton. For GBR servces, we set the prorty calculaton formula as follows. Where, ( R * E ) D F = w * * (1) H T w s the weght factor of logcal channel. s the guaranteed bt rate of servce. E s spectrum effcency of user. H s the hstory throughput of R Publshed by Atlants Press, Pars, France.

3 Proceedngs of the nd Internatonal Conference on Computer Scence and Electroncs Engneerng (ICCSEE 13) servce, and D s ts queue delay. T s the maxmum delay that servce can tolerant. We wll explan the key factors as follows n detal. Computng methods of E and H are the same as conventonal PF algorthm. E represents spectrum effcency of user and t s the reflecton of channel qualty; t can be obtaned from users Channel Qualty Indcator (CQI) as shown n Fg.. H s the hstory throughput of user servce whch s able to brng better farness to the system. We can get t by mtatng the method n conventonal PF, and the formula s as follows. H = ( 1 β ) * H + β * T () Where, T s data volume receved successfully by user servce n the last TTI, β s the factor to compute hstory throughput, generally t s.1. Several servces maybe avalable at the same tme for one user, that s, there are several downlnk Dedcated Traffc Channels (DTCH). Prorty between them s determned by MAC Logcal Channel Prortzaton (MLP). We ntroduce w to prortze DTCH. When settng factor w, both Sgnalng Rado Bearer (SRB) and Data Rado Bearer (DRB) should be taken nto account. MLP of SRB s set default n protocol, SRB and SRB1 has been set 1 whle SRB 3. Prorty of DRB s left to be set accordng to the specfc crcumstance. We put forward a method on account of the partton of servces n protocol. Suppose there are m knds of DTCH, and MLP value 1,, 3 are left to SRB, then the rest 4~m can be used by DRB. For the DRB, a DTCH corresponds to a servce whle a servce corresponds to a QCI value provded n protocol. Consequently, we set MLP=QCI+3. When MLP=, weght factor of logcal channel prorty can be calculated by w n m = (3) Where n s a postve nteger greater than 1. To reduce the delay and decrease the packet loss rate, we ntroduce a delay related factor D, whch represents wat tme of packet. The larger ts value, the hgher prorty t corresponds to. Ths factor s capable of enhancng the schedulng prorty of users who have longer wat tme. B. Resource Allocaton of Non-GBR Servces We begn to allocate resource for Non-GBR servces after all GBR servces have been served or GBR avalable resource s exhausted. In vew of ther lower demand for QoS, and to reduce the complexty, we refer to the conventonal PF algorthm as well as leave out delay and loss rate factors. Smlarly, we add w nto our calculaton to enable QCI hgh servce get hgher prorty. Prorty calculaton formula s as follows: Where, E F = w * (4) H w s the weght factor of logcal channel, E s spectrum effcency of user, H s the hstory throughput of servce. These factors can be obtaned n the same way as former. What to note here s that, to avod starvaton, whle stll servng the logcal channels accordng to ther prortes, n LTE a Prortzed Bt Rate (PBR) s confgured by the enode B for each logcal channel. That s to say, when allocatng resource for users, RBs needed by PBR wll be allocated n the frst place. Only when all logcal channels have been served up to ther PBR, then f there s stll room left n the MAC PD each logcal channel s served agan n decreasng order of prorty. In the process of secondary dstrbuton, only logcal channels that stll have data can be served n decreasng order of prorty. Prorty mentoned here s the one we calculated before. IV. SIMLATION RESLTS AND PERFORMANCE ANALYSIS A. Smulaton Parameters In ths secton, we evaluate performance of the proposed schedulng algorthm, called proposed PF. In the smulaton, a total of 7 BS (cells) are consdered n all smulatons and users are randomly dstrbuted n each cell. The avalable spectrum s MHz wth carrer frequency.ghz. Related specfcatons are lsted n Table I. TABLE I. System Structure Parameter Dstance between two stes SYSTEM SIMLATION PARAMETERS Radus of sector m BS heght Mnmum dstance from user to BS Settng/descrpton 7cells, 3sectors per cell 5m 3m 35m Number of users per sector Publshed by Atlants Press, Pars, France. 3

4 Proceedngs of the nd Internatonal Conference on Computer Scence and Electroncs Engneerng (ICCSEE 13) Frequency reuse number 1 System mode Subframe confgure 1 Carrer frequency System bandwdth TDD GHz Total RB number 1 Subcarrer number per RB 1 MHz Transportaton mode Open-loop multplexng Antenna number * Channel type SCME spatal Shadow Fadng Standard Devaton 8dB In order to show how the LTE MAC scheduler serves and handles the mxture of the dfferent QoS servces, users wth three dfferent QoS servces (VoIP, Vdeo and FTP) are defned, as shown n Table II. Some of them must be served wth hgher prorty because of ther real tme nature. We nvestgate what happens f we mx all these servces wthn one QoS class wthout dstngushng between the GBR and non-gbr, and what happens f we separate them. Meanwhle we set parameter α =.5. TABLE II. SERVICES SETTING Servce name Servce type QCI grade VoIP GBR 1 Vdeo GBR Ftp Non-GBR 8 B. Results Analyss Performance evaluaton of schedulng algorthm contans three aspects: sector throughput, servce delay and farness between users. We get sector throughput by means of calculatng transmtted data n sector, and servce delay by packet wat tme. Farness between users can be computed va the followng formula: N ( x ) = 1 = N N x = 1 F (5) Where, x s the amount of data that servce transmts successfully, N s the total number of servce. Suppose F=1, t means that the resource allocaton can meet all of the users, and the system farness s best. The hgher the value of F s, the better the farness between users. The value of F ranges from to 1. Sector Avg. Throughput(Mbps) Sector average throughput Avg. Cell Throughput For Ftp(conventonal PF) Avg. Cell Throughput For Ftp(proposed PF) Avg. Cell Throughput totalp(conventonal PF) Avg. Cell Throughput total(proposed PF) enum Fgure 4. Throughput of sector and ftp Fg. 4 shows that throughput of the sector and ftp n our proposed algorthm s lower than the conventonal PF wth the ncrease of user number. Whle there are few users n the system, all users emergency telecommuncaton servces can be served; there s no dfference n throughput. However, wth the user number n system ncreased, wreless resource s nsuffcent, n ths case, the prortzaton of VoIP and Vdeo result n the decrease of ftp throughput. On the other hand, some of the RBs are allocated to users who have poor channel qualty to transmt GBR data nstead of servng users wth hgh channel qualty; ths also leads to the reducton of the sector throughput. Sector Avg. Throughput(Mbps) Sector average throughput Avg. Cell Throughput For Vdeo(conventonal PF) Avg. Cell Throughput For Vdeo(proposed PF) Avg. Cell Throughput For Vop(conventonal PF) Avg. Cell Throughput For Vop(proposed PF) enum Fgure 5. Throughput of VoIP/Vdeo In Fg. 5, throughput of VoIP and Vdeo n the proposed algorthm s hgher than the conventonal evdently wth the ncrease of users. Ths s because we allocate RBs for GBR data pror. Ths ndcates that our algorthm s capable of mprovng throughput of GBR bearers. Publshed by Atlants Press, Pars, France. 4

5 Proceedngs of the nd Internatonal Conference on Computer Scence and Electroncs Engneerng (ICCSEE 13) e Far Indcator e Delay Ftp(conventonal PF) Ftp(proposed PF) Vdeo(proposed PF) Vdeo(conventonal PF) Vop(proposed PF) Vop(conventonal PF) e Far Indcator enum 1 1 Fgure 6. Comparson of servce farness Ftp(conventonal PF) Ftp(proposed PF) Vdeo(proposed PF) Vdeo(conventonal PF) Vop(proposed PF) Vop(conventonal PF) e Delay enum Fgure 7. Comparson of servce delay Fnally, servce farness and delay are shown n Fg. 6 and Fg. 7. Along wth the number ncrease of users, at the cost of a certan amount of performance drawback of ftp, Vdeo and VoIP are capable of keepng preferable farness and lower delay compared wth conventonal PF algorthm when applyng the proposed. The key s that GBR servce s pror to Non-GBR. Ths attans our goal that assures the QoS of real-tme servces at the cost of a certan amount of nonreal tme servces performance. V. CONCLSION In ths paper, a novel resource schedulng algorthm based on dfference of servce QoS for LTE downlnk has been proposed. The proposed MAC scheduler consders two man types of QoS bearers: Guaranteed Bt Rate (GBR) and Non Guaranteed Bt Rate (non-gbr). We assgn RBs to GBR bearers ahead of Non-GBR to satsfy QoS of real-tme servce. We desgn a parameter to prevent GBR servces from occupyng too much resource when there are too many users n the system. A MAC logcal channel prortzaton weght factor s ntroduced to optmze the performance of servce that has hgher prorty. The algorthm suts to not only real-tme servce but also the non-real-tme, and smulaton results show that t s able to mprove the performance of GBR servces at the cost of a certan amount of the non-real tmes. ACKNOWLEDGEMENT Ths paper s supported by NSFC ( , ),the Fundamental Research Funds for the Central nverstes (714669) and the 111 Project(B838) REFERENCES [1] Z. Shen, J. G. Andrews, and B. L. Evans. "Adaptve resource allocaton n multuser OFDM systems wth proportonal farness", IEEE Trans. On Wreless Comm., vol. 4, no. 6. pp , Nov. 5. [] Daqn Yang, Dongxu Shen, Wenjan Shao and L, V.O.K, Towards opportunstc far schedulng n wreless networks, IEEE ICC 6, vol. 11, pp , Jun. 6. [3] Proebster, Magnus, Mueller, Chrstan M., Bakker, Hajo, Adaptve farness control for a proportonal far LTE scheduler, IEEE PIMRC, pp , 1. [4] Ye, Fang; Zhang, Huan; Gu, Yunbao, An mproved proportonal far resource schedulng algorthm for LTE system, Journal of Computatonal Informaton Systems, vol. 8, no. 3, pp , March 1 [5] F. M. Chuss and V. Svaraman, Achevng hgh utlzaton n guaranteed servces networks usng Early-Deadlne-Frst schedulng, 6th Int l. Wksp. QoS, pp. 18-, May [6] Amegeras, Pablo; Wgard, Jeroen; Mogensen, Proben, Performance of the M-LWDF schedulng algorthm for streamng servces n HSDPA, IEEE Vehcular Technology Conference, vol. 6, no., pp , 4. [7] Xan, Yong-Ju; Tan, Feng-Chun; Xu, Chang-Bao; Yang, Yue; Analyss of M-LWDF farness and an enhanced M-LWDF packet schedulng mechansm, Journal of Chna nverstes of Posts and Telecommuncatons, vol. 18, no. 4, pp. 8-88, August 11. [8] Ju Yong Lee, Sorour, S.; Valaee, S.; Dynamc parameter adaptaton for M-LWDF/M-LWWF schedulng, IEEE Transactons on Wreless Communcatons, vol. 11, no. 3, pp , March 1. [9] 3GPP TS V8.., Evolved nversal Terrestral Rado Access (E-TRA); Physcal Channels and Modulaton, Oct, 1. [1] 3GPP TS36.3. v1.l.. Evolved nversal Terrestral Rado Access (E-TRA) and Evolved nversal Terrestral Rado Access Network (E-TRAN), Oct, 1. [11] 3GPP TS v8.5.. Evolved nversal Terrestral Rado Access (E-TRA) Medum Access Control (MAC) Protocol specfcaton (Release 8) [S], Publshed by Atlants Press, Pars, France. 5