A Dynamic Subchannel Allocation Algorithm for IEEE e Networks

Transcription

1 A Dynam Subhannel Alloaton Algorthm for IEEE 86e etworks Donysos Xenaks, Dmtrs Tsolkas, Spyros Xergas, kos Passas, and Lazaros Merakos Department of Informats & Teleommunatons Unversty of Athens, Greee Abstrat The paper desrbes a system arhteture and formulaton of the Dynam Subhannel Alloaton problem n IEEE 86e networks Aordng to t, the DSA algorthm operates as an extenson of a IEEE 86 tme sheduler, tryng to explot the benefts of the OFDMA mode Smulaton results show a onsderable mprovement, ompated to the legay system, espeally n terms of hannel errors eywords: OFDMA, IEEE 86e, Dynam Resoure Alloaton (DSA) I Introduton The IEEE 86 standard [] and ts reent amendment IEEE 86e [] for moblty support, s onsdered as one of the most promsng tehnology for wreless broadband aess n metropoltan envronments as they an provde hgh data rates and QoS guarantees Aordng to the standard, the physal layer provdes four bas multple modes: sngle arrer (SC), SCa, OFDM and OFDMA whle n the MAC layer the smallest PHY layer resoure to alloate n the tme/frequeny doman s a slot [][] OFDMA multple mode s based on the orthogonal frequeny-dvson multplexng (OFDM) tehnque and t s ndependent to ntersymbol nterferene and frequeny seletve fadng Ths mmunty s a result of a frequeny band dvson nto a group of subarrers, mutually orthogonal, wth lower bandwdth than the total hannel s bandwdth whh alled subhannels [3] For the OFDMA PHY, the sze of a slot depends on the subarrer permutaton mode adopted, whh an be dstrbuted or adjaent, DSP and ASP respetvely DSP forms a subhannel by pkng the omposng subarrers pseudo-randomly n the frequeny doman and provdes frequeny dversty and nter-ell nterferene averagng ASP forms a subhannel by groupng ontguous arrers and enables mult-user dversty by hoosng the subhannel wth the best frequeny response In ths paper, a dynam resoure alloaton algorthm that adopts ASP s proposed Addtonally, the TDD mode, n whh the tmeframe s dvded n the downlnk (DL) and the uplnk (UL) subframe s onsdered Spefally the paper fouses n the DL dreton although the proposed algorthm an be easly deployed n the UL dreton as well The proposed algorthm operates as an extenson to any SC or OFDM sheduler and an explot the benefts of the OFDMA mode, aordng to the BS s hannel ondton awareness for eah MS The rest of the paper s organzed as follows Seton II presents the bas arhteture of the Dynam Subhannel Alloaton (DSA) problem as a tmeshedulng extenson and summarzes the present state of the art In seton III the proposed framework and a DSA formulaton are presented whle n seton IV the proposed algorthm s gven Seton V presents the smulaton results and the smulaton model used for the evaluaton of the proposed algorthm aganst the ntal tmesheduler Fnally, seton VI onludes ths paper and provdes plans for future work Dynam Subhannel Alloaton (DSA) In PMP mode (Pont to Multpont) and for PHY OFDMA, the BS s responsble n alloatng resoures to the MSs n a spef ell area Aordng to ts shedulng algorthm poly, the demands of eah MS s atve onnetons and the awareness of eah MS s hannel ondton, the BS alloates ontguous olleton of slots (data regon) n order to satsfy the nput traff by explotng the benefts of OFDMA OFDMA shedulng has also been onsdered n [4-7] [4] onsders all the serve lasses n 86 but grants resoures per onneton (GPC) smlar to [5], whh proposes a mxed paket shedulng and subhannel alloaton sheme n whh for eah subhannel one partular shedulng prorty queue s obtaned The benefts of a granted per MS sheme (GPMS) s the salablty that offers to a large number of MSs and onnetons Addtonally, t s redundant to shedule onnetons wth the same hannel ondtons n order to explot hannel dversty [6] adopts a low omplexty approah whh s based on subhannel and user power management Moreover, [7] presents a ross layer analyss between the MAC and PHY layer, for OFDMA shedulng, ompatble wth WBro systems In ths paper, a new approah/arhteture of the problem s onsdered The tme sheduler alloates resoures to onnetons (and so to MSs), to satsfy ther requests for bandwdth and QoS provsonng As soon as these requests are known to the BS t an dede (aordng to ts poly) whether to alloate resoures or not Its desons are about the long term robustness of the system However, hannel ondtons may vary n a pre frame bass (short term) and so do the respetve SIR measurements As a result, durng the long term tmeshedulng t would be hard to predt whh data regons wll be alloated to eah MS per frame However, a DSA algorthm that takes as an nput the tme shedulng preeded and the urrent hannel ondtons per MS, an dede n a per frame bass whh data regons are gong to maxmze the throughput of the system It s rual to pont that suh an algorthm wll not nterfere to the shedulng poly adopted, beause the same resoures wll be alloated to the same MSs Another matter that ours s how a DSA algorthm should alloate the resoures, n a GPMS or n a GPC alloaton sheme As soon as the alloatons deded by the tmesheduler satsfy the bandwdth and QoS requrements, a GPMS poly an explot the fat that onnetons belongng to the same MS wll fae the same hannel ondtons, whle the GPC poly wll not For ths reason, the GPMS poly s followed In total, the desons onernng the long term (bandwdth, QoS requrements, et) an be taken by the tme sheduler whle a DSA algorthm an assst by explotng the short term hannel ondtons per MS Fgure a summarzes the arhteture of the approah desrbed above Aordng to ths arhteture, the tme /8/$5 8 IEEE 65

2 sheduler, based on the nput traff, produes the amount of resoures alloated to eah MS These resoures are dstrbuted to atve onnetons aordng to eah onneton s bandwdth, QoS, prorty, modulaton and odng sheme (burst profle) parameters Addtonally, there s a mehansm that provdes for eah MS and subhannel, SIR measurements for the urrent hannel ondtons The ombnaton of the tme sheduler and the hannel ondton estmaton mehansm output s onsdered as the nput of the proposed DSA algorthm, whle t produes as an output the fnal alloatons Fgure : The proposed arhteture Consequently, the man objetve of a DSA algorthm aordng to ths arhteture s summarzed as follows: Based on ) the tme sheduler s deson n eah tmeframe, and ) the measurements of the urrent hannel ondtons, maxmze the overall system throughput by alloatng to eah MS the best possble subhannels III DSA formulaton The desgn of the DSA algorthm, wll be based on two bas prnples: A step-by-step resoure alloaton, where n eah step one MS gets ts fnal alloaton of slots The amount of resoures (number of slots) per MS, fnally alloated by the DSA algorthm, wll be as lose as possble to the ntal alloaton made by the tme sheduler Addtonally, the seleted modulaton and odng sheme s based on the ntal deson of the tme sheduler Subseton IIIA ontans a set of defntons requred for the formulaton, whle n subseton IIIB an exhaustve alloaton method s desrbed, to be used as a bass for the proposed DSA algorthm IIIA Defntons : umber of avalable subhannels T: umber of tmeslots n the tmeframe : umber of MSs MS :MS dentfaton (the -th MS), S(j):Subhannel dentfaton (the j-th subhannel), j d : umber of slots alloated to MS by the tme sheduler x : umber of slots alloated to MS by the DSA algorthm SIR(,j): Sgnal to Interferene plus ose Rato of user MS n subhannel S(j) Subhannel vetor, : Based on eah MS s hannel ondtons and a threshold() defned by ts urrent modulaton and odng sheme, the BS mantans a x vetor whh for eah subhannel S(j) haraterzes the effeny of reevng bts suessfully or not As a result, SIR(,j) threshold() results to [j]=true=, or else [j]=false= For MS, subhannel S(j) s haraterzed as an atve one f [j]= and as an natve one f [j]= Aordngly, every slot loated on an atve subhannel, s also alled atve Transmsson rate, r s (): For MS, r s () represents the number of reeved bts per slot aordng to the seleted modulaton and odng sheme ormalzaton fator, F : For MS, the normalzaton fator F s defned as the perentage of slots that are fnally sheduled by the DSA algorthm Its maxmum s set to, supposng that there s no extra nformaton to be sent Hene: F = mn( x / d,) () Channel ondtons fator, : If a data regon between subhannel S(a) and subhannel S(b) s alloated to MS, then s defned as the perentage of atve subhannels n ths regon Thus: S ( b) = [j] j = S ( a) α () S( b) S( a) + MS s suessfully reeved bts, B : For eah MS, the amount of suessfully reeved bts B s defned as: B = d rs ( ) α F (3) where d r s () expresses the number of reeved bts for MS, for the alloated data regon expresses the estmated effet of the hannel ondtons to the respetve transmsson and F expresses the devaton from the ntal alloaton of the d slots (f any) The total number of suessfully reeved bts for the system s equal to the sum of eah atve MS s B : B sys = B (4) = Overall system rate, r sys : Supposng that frame represents the duraton of the tmeframe, r sys s defned as the average system rate n bps and s gven as: B sys = r = = (5) sys τ frame τframe Channel bt loss (%), L : Is defned as the bt loss aused by the hannel ondtons and s equal to: = s B d rs ( ) Bsys = L = = d r ( ) = Bsys (%) d r ( ) (6) IIIB Exhaustve Alloaton Method Based on the above dsusson, maxmzng system throughput s equvalent to maxmzng r sys ote that by maxmzng r sys then B sys s also maxmzed whle L s mnmzed The IEEE 86e standard provsons the alloated data regons to be onssted from slots on ontguous subhannels Fgure depts all possble ombnatons of subhannels that may ompose a vald data regon As seen, the number of vald ombnatons s: s 66

3 ( + ) ( ) + = = = O( ) (7) = ote that there an be up to T data regons onssted from the same ombnaton of subhannels, due to the dfferentaton of ther frst tmeslot n the tmeframe In total, the number of data regons to alloate s up to: T ( + )/ = O(T ) (8) T ( + )! A( ) =! = O( ) () STEP ROOT (,,) (,j,t) (,j,t+) (,j,t) (,j+,) (,(+),T) (+,,) (,(+),T) STEP (,,) (-,(+),T) (+,,) (,(+),T) (,,) (,(+),T) (+,,) (,(+),T) LEGED : Step : Seleted path (,j,t) : Fnalze MS to the data regon that s loated n the ombnaton of subhannels j and begns from the tmeslot t Fgure : otaton of possble ombnatons Havng measured the number of data regons to alloate and based on the ntal prnples n the begnnng of the seton, every possble resoure alloaton s expressed n a tree, alled the exhaustve DSA tree (E-DSA tree) (Fgure 3) Eah level of ths tree represents the step of a DSA algorthm n whh t alloates to an MS ts fnal data regon whle, eah node represents one of all possble resoure alloatons for the respetve step Addtonally, eah alloaton s referred to as (,j,t), meanng that n ths step MS an be assgned the data regon loated on the j-th ombnaton of subhannels, begnnng from tmeslot t Apparently,, j ( +)/ and tt For the urrent tmeframe, any possble resoure alloaton soluton s represented by a path n the E-DSA tree (begnnng from the root and endng to a leaf) As a result, an algorthm that alulates the resultng r sys of every possble path n the E-DSA tree and selets the path that maxmzes r sys, gves the best resoure alloaton soluton A omputaton of ths exhaustve algorthm s omplexty follows In eah step of ths algorthm, the number of pendng MSs s dereased by one whle aordng to (8), the number of data regons for alloaton may reman the same (equal to T ( + )/ ) Let A() symbolze the number of all possble paths begnnng from the root and endng n level, then A() represents the amount of all possble paths n the tree and so, the omputatonal omplexty needed for an exhaustve examnaton Apparently, the number of nodes n the frst level s equal to A ( ) = ( + ) / That s beause one of the MSs an be assgned one of the T (+)/ possble data regons Aordngly, A ( ) = A() ( ) ( + )/ That s beause based on the alloaton made n the frst step, n the seond step one of the - pendng MSs an be assgned one of the T (+)/ possble data regons (f the prevous alloated data regon sn t onssted from all slots on a ombnaton of subhannels) In general, for step t s n effet that: A() = A( ) ( + ) ( + )/,( A( ) = ) (9) Iteraton results that the amount of all possble paths n E- DSA tree and so the omplexty of suh an algorthm, s upper bounded by: Fgure 3: E-DSA tree IV The Proposed DSA Algorthm IVA General Estmaton Algorthm (GEA) Takng nto aount the short tme perod n whh the DSA algorthm s takng plae (n a per tme frame bass), an exhaustve algorthm s nadequate However, a lower omplexty algorthm based on the same outlnes, may estmate eah step s alloaton effently Ths estmaton s based on the dea that the most effent path (soluton), an be derved by examnng the frst h nodes of ths path (a subpath), nstead of examnng the entre path of the tree Leavng h as a parameter, a General Estmaton Algorthm (GEA) s derved Ths algorthm s summarzed as follows: Set as frst head node, the root of the E-DSA tree Whle there are pendng MSs and resoures to alloate: Calulate r sys for the frst h nodes of all sub-paths below ths step s head node Selet the sub-path that maxmzes r sys tll that depth and proeed to the alloaton represented by t 3 Set as head node, the node seleted n 4 Go to step In the (-h+) step, the estmaton proedure ends as soon as the last level of the tree s reahed The benefts of suh an algorthm are that there an be a trade off between omputatonal omplexty and estmaton effeny In Seton V, t wll be lear that ths trade off sn t as expensve as ntally expeted and that suh an algorthm may lead to good results A omputaton of GEA s omplexty wll make lear the above mentoned trade off In eah step, GEA examnes E-DSA sub-trees of depth equal to h Hene, n the frst step ths algorthm examnes A(h) sub-paths, n the seond step examnes A(h+)/A() subpaths and n general, n step j A(h+j)/A(j) number subpaths are examned Eah term of the followng sum represents the number of sub-paths examned n eah step, whle the sum tself gves the sub-paths examned n total: = h j= A( h + j) = A( j) h j= h h ( h + j)! ( T ( + ) / ] ) j= j! ( h + j)![ T ( + ) / ] j j![ T ( + ) / ] h+ j Hene, the omplexty of the algorthm s upper bounded by: O([ T h h h / ] [( h+ ) (!/( h)!)]) () 67

4 It s now obvous, that the above mentoned trade off s possble and that for h= the GEA ondes wth the exhaustve algorthm outlned n Seton III IVB The Proposed DSA Algorthm In the followng, the effeny of the least omplexty nstane of the GEA s examned For h=, n eah step of the estmaton algorthm, the alloatons examned are those who orrespond to the respetve level of the tree (length one subpaths) Hene, pkng the sub-path that maxmzes r sys tll eah step (level) s now equal to pkng the alloaton wth maxmum B for eah step Although, ths algorthm s a greedy one, the effets of ts greedness an be mnmzed by examnng an approprate subset of alloatons n eah step, as analyzed below As mentoned n Seton III, there may be more than one data regons for the same ombnaton of subhannels j, whh are dfferentated by ther begnnng tmeslot Thus for h=, the examnaton of all these (,j,t) alloatons lead to the same B, so t s redundant For ths reason, only the (,j, t mn ) alloaton wll be examned for eah (,j), where t mn represents the earlest avalable tmeslot for the j-th ombnaton In ths ase, some of these (,j, t mn ) alloatons, mght lead to unalloated fragments of random shape There are two optons to manage these fragments, ether to try to alloate them n the future, or not to alloate them at all In order to avod ths phenomenon, only (,j, t mn ) alloatons wll be examned, where t mn represents the earlest avalable tmeslot n any subhannel As a result, the number of nodes examned, s dereased even more Addtonally, the data regon s sze n slots s defned by the losest nteger to the ntally alloated slots to eah MS Consequently, there an be a small devaton between d and x whh s taken under onsderaton below In total, the proposed algorthm n eah step examnes all vald (,j, t mn ) alloatons and proeeds to the alloaton that maxmzes B If there s more than one alloaton that maxmzes B, then a number of seleton rtera should be taken under onsderaton These seleton rtera an be: () the mnmzaton of the devaton between the ntally and fnally alloated slots whh s equal to x - d, () the mportane of the alloated data regon to the pendng MSs An effent measurement of the mportane of eah slot on eah subhannel, and thus the mportane of a data regon to the pendng MSs, should nvolve the followng: The number of atve slots n the respetve subhannels (onsstng the data regon), that were ntally assgned to eah pendng MS by the tme sheduler These slots are onsdered mportant Respetvely, all the natve slots that were ntally assgned to eah pendng MS, are not onsdered as mportant Slots that were not alloated to any MS by the tme sheduler (not a full tmeframe), are also not onsdered as mportant It s ponted out that, due to the alloaton proedure followed t s only mportant to know the sum of the natve or not alloated slots n eah subhannel For ths reason, a sngle-dmenson vetor wll be suffent to mantan ths nformaton Ths x vetor s symbolzed wth (k), and for step k mantans n (k) [j], the number of not mportant slots n subhannel j In order to ntalze and update (k), the predefned parameters d and (Seton III) wll be exploted Consderng that d / represents the number of slots ntally assgned to MS n eah subhannel and that all slots on an atve subhannel j ( [j]=) are also atve by defnton, then the ntalzaton of (k) (k=)should be: ( )[j] = T ( [n]) d / =, wth j Updatng (k) should dept the alloaton made n the prevous step whh means an abstraton of the number of slots fnally assgned to the respetve MS as well as an abstraton of the number of slots ntally assgned to the MS and were onsdered durng the ntalzaton Thus, supposng that MS s fnalzed n the data regon between S(a) and S(b) subhannels, updatng of (k) s desrbed as: f (subhannel j s atve and j [ S(a),S(b) ]) d (k + )[j] = (k) [j] + f (subhannel j s atve and j [ S(a),S(b) ]) x d (k+ )[j] = (k) [j] ( + ) Havng defned a measurement of mportane for eah subhannel, a measurement of the mportane of a data regon for all the pendng MS s gven by the metr = (k-) - (k), where (k) = n = (k) [n] represents the sum of the above mentoned slots for the entre tmeframe As a result, measures the effet of an alloaton to the next step Based on the above analyss, the proposed algorthm s summarzed as follows: Whle there are pendng MSs and resoures to alloate For every pendng MS For every ombnaton of subhannels j Calulate alloaton s (,j, t mn ), B and Proeed to the alloaton wth ) maxmum B ) mnmum x - d, ) maxmum As an nstane of the GEA, the omplexty of ths algorthm would be O(T ) Takng under onsderaton that updatng n eah step needs () and that only (, j, t mn ) alloatons are examned, the omputatonal omplexty of ths algorthm has as lower lmt ( ) and as an upper lmt O( ) In a later paper, t wll be shown that the proposed algorthm s lower omplexty bound an be dereased to ( ) by examnng the alloatons n an organzed manner and avodng redundant alulatons V Smulaton Model & Results In order to measure the performane of the proposed DSA algorthm and ompare t aganst a system wth no suh apablty, a smulaton model was onstruted n C++ Ths model ompares the system s fnal hannel bt loss L usng the proposed DSA algorthm aganst the ntal fnal nt hannel bt loss L of the system wthout the algorthm (by usng only the tme sheduler) The man omponents of the smulaton model follow the arhteture of Fgure a, whle ther operaton s as follows: 68

5 Traff Generator: In order to produe real traff load per MS, a set of atve onnetons are onsdered: one UGS wth onstant data rate of 64bs (eg,voe) and lateny equal to mse, one rtps wth mean data rate of 56bslot and lateny equal to 4mse, one nrtps wth mean data rate of 8bslot and, one BE wth mean data rate of 8bslot Tme Sheduler: Based on the nput traff from the Traff Generator and the QoS requrements of eah onneton, provdes the tmeframe map on a frame-by-frame bass The modulaton shemes used were QPS, 6-QAM and 64- QAM, whle the Frame Regstry Tree Sheduler (FRTS) defned n [8], was adopted as the tmeshedulng algorthm Channel Model: Emulates the hannel ondtons for eah MS and provdes a SIR vetor for eah subhannel per MS For eah subhannel and MS, SIR follows a dsrete tme Markov Chan model n whh adjaent states dffer by 5 db rangng from 85 to 5 db Two smulaton senaros were onsdered The frst measured the performane of the proposed algorthm aganst an nreasng number of MSs, assumng an average system hannel bt loss L =45% The seond senaro measured the performane of the proposed algorthm aganst an nreasng ntal hannel bt loss L nt, whle the number of MSs was onstant and equal to 5, number adequate to result n fully alloated tmeframes As shown n Fgure 4, the mprovement of the proposed algorthm on the fnal hannel bt loss s ndependent of the number of users partpatng n the tmeframe, as long as tmeframes are almost full (more than 4 MSs) A onstant mprovement of about 35% s aheved, assumng hannel ondtons that lead to ntal hannel bt loss L fnal 45% multple aess mode, keepng the hannel bt loss at muh lower values n most of the ases Fgure 5: Fnal hannel bt loss aganst hannel bt loss VI Conlusons A DSA algorthm that ats as an extenson to any IEEE 86 tme sheduler and explots the OFDMA mode provded by IEEE 86e was desrbed and evaluated To avod exhaustve alulatons that would nrease omplexty, the algorthm estmates the best possble alloaton for every user, based on a lmted number of alternatve reservatons Smulaton results reveal the effet of the algorthm, as they show an mprovement of hannel bt loss of up to 6% Future plans nlude extendng the proposed arhteture and algorthm n order to norporate the benefts of adaptve modulaton and power ontrol, presentng optmzaton methods that lead to lower omplexty bounds and fnally, more extensve smulatons Referenes Fgure 4: Fnal Channel bt loss aganst the number of MSs Fgure 5 shows that the proposed algorthm leads to muh lower fnal hannel bt loss for the system Hene, n medum nt to bad hannel ondtons ( L <75%) the proposed fnal algorthm, stablzes the system s L between and %, leadng to an mprovement of up to 6% Ths robustness n performane may also mnmze the effet of a bad burst profle seleton Addtonally, n very bad hannel ondtons (75%< L nt <9%) the performane s stll onsderably hgh Clearly, the proposed algorthm attans to explot the OFDMA [] IEEE Std 86, IEEE Standard for Loal and metropoltan area networks Part 6: Ar Interfae for Fxed Broadband Wreless Aess Systems, Otober 4 [] IEEE Std 86e, IEEE Standard for Loal and metropoltan area networks Part 6: Ar Interfae for Fxed and Moble Broadband Wreless Aess Systems Amendment : Physal and Medum Aess Control Layers for Combned Fxed and Moble Operaton n Lensed Bands and Corrgendum, February 6 [3] Syed Hussan Al, -Dong Lee, and Vtor CM Leung, Dynam Resoure Alloaton n OFDMA Wreless Metropoltan Area etworks Wreless ommunatons, IEEE, vol 4 Issue, p 6-3, Feb 7 [4] M Hawa, DW Petr Qualty of Serve Shedulng n Cable and Broadband Wreless Aess Systems th IEEE Internatonal workshop on Qualty of Serve, May, pp [5] Lhua Wan, Wenhao Ma, Zhua Guo, A Cross-layer Paket Shedulng and Subhannel Alloaton Sheme n 86e OFDMA System Wreless Communatons and etworkng Conferene, WCC 7 IEEE, p , Marh 7 [6] Ian C Wong, Zukang Shen, Bran L Evans, and Jeffrey G Andrews, A Low Complexty Algorthm for Proportonal Resoure Alloaton n OFDMA Systems Sgnal Proessng Systems, SIPS 4, IEEE Workshop 4, p - 6, Ot 4 [7] Taesoo won, Howon Lee, Sk Cho, Juyeop m, and Dong-Ho Cho, Desgn and Implementaton of a Smulator Based on a Cross-Layer Protool between MAC and PHY Layers n a WBro Compatble IEEE 86e OFDMA System Communatons Magazne, IEEE, vol 43 ssue, p 36-46, Deember 5 [8] S Xergas, Passas, and L Merakos, Flexble Resoure Alloaton n IEEE 86 Wreless Metropoltan Area etworks, Pro 4th IEEE Workshop on Loal and Metropoltan Area etworks (LAMA), Chana, Greee, September 5 69