Optmal Packetzato Iterval for VoIP Applcatos Over IEEE 802.16 Networks Sheha Perera Harsha Srsea Krzysztof Pawlkowsk Departmet of Electrcal & Computer Egeerg Uversty of Caterbury New Zealad sheha@elec.caterbury.ac.z Departmet of Electrcal & Computer Egeerg Uversty of Caterbury New Zealad srsehr@elec.caterbury.ac.z Departmet of Computer Scece Uversty of Caterbury New Zealad krys@caterbury.ac.z Abstract A aalyss of the mpact of the packetzato terval for costat bt rate traffc has bee doe the cotext of IEEE 802.16 MAC layer. Badwdth used for overheads whch clude lower layer headers as well as retrasmssos at the MAC layer are cosdered. A optmal packetzato terval selecto method for delay sestve applcatos such as VoIP s proposed. Ehacemets to the Usolcted Grat Servce retrasmsso strategy are proposed to further mprove delay ad mmze packet loss whle makg effcet use of the lmted badwdth resource. I. Itroducto T HE eed for moble as well as fxed pot wreless coectvty has creased greatly recet years ad as such ew protocols ad access techologes are emergg to satsfy ths eed. Oe of most spoke of as of late s IEEE 802.16 also kow as WMAX. Ths stadard was orgally teded for pot-to-multpot fxed broadbad wreless access. It ams to provde hgh data rates for a large umber of users a wde area. Oe of the applcatos whch WMAX specfcally caters for s VoIP. Accordg to the stadard [1] four classes of schedulg have bee defed of whch usolcted grat servce (UGS) ad real-tme pollg servce (rtps) ca potetally be used to schedule uplk VoIP traffc. Eve so there are several ssues whe t comes to ths kd of servce flow. Sychrozg the grat of uplk badwdth ether of the above schedule types ca be a problem. For UGS several solutos to ths problem are proposed [2]. These solutos propose adjustg the grat tme steps ragg from 15 frames dow to the maxmum tolerated jtter. Smaller steps of adjustmet troduce lower jtter but ca take up to hudreds of mllsecods to reach the desred level of latecy. Durg slet perods of a voce coversato badwdth s wasted UGS schedulg. Ths wastage s mmzed by usg voce actvty detecto/slece detecto at the subscrber stato (SS) ad formg the base stato (BS) of the voce state trastos [3, 4]. Ths allows more voce users to be accommodated the system tha would be usg pure UGS. Usage of rtps schedulg has also bee vestgated but does have the drawbacks of added delay due to the request-grat mechasm ad overhead due to uused request slots. Sce the protocol data uts (PDUs) ca vary cosderably sze, from tes of bytes to tes of klobytes t s mportat to vestgate the effect o effcecy ad lk utlzato. I [5] the optmal PDU sze has bee calculated for dfferet resdual bt error rates (BER) order to mmze system overhead. Ths wastage of badwdth due to overhead s more as the payload sze decreases. I the case of VoIP packets wth packet szes the tes of bytes, t s very mportat to cosder the Orthogoal Frequecy Dvso Multplexg (OFDM) symbol parameters whe decdg the optmal packetzato terval. The packet sze s drectly proportoal to the packetzato terval. The rest of ths paper s orgazed as follows. Secto II gves a bref overvew of IEEE 802.16 ad the UGS schedule type. Next we wll provde a aalyss of the effect of packetzato terval selecto ad propose a method to dyamcally select the best value durg ru tme. II. Overvew of IEEE 802.16 A. IEEE 802.16 MAC Protocol IEEE 802.16 assumes a pot-to-multpot archtecture wth a cetral base stato (BS) whch acts as gateway to coect the subscrber statos (SS) the cell to other publc etworks. The MAC operato s based o
MAP messages trasmtted perodcally (oce per frame) by the BS. The MAP defes the tmes the dowlk (DL) ad uplk (UL) whch are used for ragg, coteto based badwdth requests, allocated polled type badwdth requests, DL PDUs ad UL PDUs for SS to sed data to the BS. Ragg s a process whch s doe by the SS at tal etry to the system ad perodcally at the request of the BS or the SS tself to optmze sgal qualty. The resposblty of schedulg UL/DL data s etrely up to the BS. Depedg o the Qualty of Servce (QoS) requremets of a partcular flow t wll be classfed to oe of the four schedule classes [6]. The MAC defes Dyamc Servce Addto/Chage/Deleto messages (DSA, DSC, ad DSD) whch are used to agree upo the flow parameters usg a request/respose/ackowledge (REQ, RSP, ad ACK) 3 way hadshake process. Out of the two classes whch ca be used for real tme flows we wll cosder oly UGS ths work although the results ca be applcable to the other classes. B. Operato of UGS UGS s desged to provde fxed sze data grats at perodc tervals to real tme costat bt rate (CBR) lke traffc flows. Ths reserves a guarateed badwdth for flows wthout the overhead ad latecy of the request grat mechasm. Sce the data grats are provded o a perodc bass, the BS ca estmate the applcato s requremet wth respect to ts QoS level durg coecto talzato. The stadard defes four QoS parameters for UGS flows. 1) Maxmum sustaed traffc rate: defes the peak data rate, whch the case of UGS s also the mmum reserved rate. 2) Maxmum latecy: the delay betwee recevg a packet from the etwork layer ad forwardg the packet to the physcal layer at the trasmtter. Ths s bascally the tme take for the MAC layer to process the packet ad get t oto the ar terface. 3) Tolerated jtter: a upper boud o the amout of delay varato that ca be tolerated at the applcato level. 4) Request/Trasmsso polcy: defes the rules of uplk badwdth request ad PDU formattg. All forms of uplk requests are prohbted for the UGS coecto. III. Aalyss of Packetzato Iterval A. Relevat Equatos Cosder a VoIP applcato whch produces a voce data stream of r bts-per-secod (bps). The overhead due to headers OH headers s the sum of the RTP, UDP, IP ad MAC layers headers bts. s the packet sze as see at the MAC layer. t s the packetzato terval of the VoIP applcato. = r t + OH (1) headers pdu ad bps are the sze of the PDU ad bts per symbol respectvely. The cel fucto rouds upwards towards the closest teger. pdu = cel bps (2) bps The symbol error rate, SER s gve equato 3. Here m gves the maxmum umber of bt errors whch ca be tolerated. PER s the packet error rate ad spp s symbols per packet. m ps j SER = 1 BER j = j 1 0 ( ) spp bps j ( BER) PER = 1 1 SER (4) Overhead due to retrasmssos ( bts) s gve (5), where s the MAC layer retrasmt lmt for ths partcular traffc class. Badwdth used for overhead s foud by dvdg the total overhead by the packetzato terval, (6). E f s the effcecy of the system. payload beg the actual voce data from the applcato layer. OH ret = pdu { PER ( 1 PER) } + PER (5) = 1 ( pdu + OH ret payload) OHbw = (6) t E f pdu ret (3) payload = () + OH Whe the retrasmt lmt has bee exceeded the packet s cosdered lost. Ths probablty s P loss. +1 P = PER (8) loss
(a) (a) (b) (b) (c) (c) Fg. 1. (a) Effcecy of the system (b) Overhead badwdth usage (c) Packet loss rate of a VoIP applcato wth a BER of 1e-4 Fg. 2. (a) Effcecy of the system (b) Overhead badwdth usage (c) Packet loss rate of a VoIP applcato wth a BER of 1e-6
B. Sample Scearo A sample scearo usg commo values for 802.16 wll be cosdered to demostrate the effects of t. bps for dfferet burst profles are gve Table I. We are cosderg the UL phase of the flow of a 256 sub carrer OFDM system. The burst profle used depeds o the sgal-to-ose rato whch depeds o the dstace from the BS. The VoIP applcato s assumed to produce a stream at a rate of 32kbps. The retrasmsso lmt s set at 2 ad the maxmum tolerated bt errors s 0. Fg. 1 gves the Effcecy, overhead badwdth ad packet loss rate for packetzato tervals ragg from 10ms to 150ms wth a BER of 10-4. The saw tooth effect s due to the trasmsso uts beg teger multples of OFDM symbols. The most effcet tervals are show Fg. 1(a). For 64 QAM 3/4 ths terval 41ms produces a packet loss rate of 0.4%. The ext best terval would be betwee 11ms~15ms whch has a packet loss rate of 0.05%. Whe the BER s lower as Fg. 2 the optmal packet sze s larger, whch s tutve. I the cotext of VoIP t s ot possble to select the largest possble packetzato terval satsfyg the QoS packet loss lmt. We also eed to stay wth the latecy bouds of the flow. By selectg a terval betwee 38ms ad 40ms, a effcecy of about 0% ca be acheved for the 64 QAM case. Wth a terval betwee 42ms ad 4ms the same effcecy ca be acheved for the 16 QAM case. It s also clear from both Fg. 1(b) ad Fg. 2(b) that a dfferece of a few mllsecods ca crease the overhead badwdth up to 10s of kbps whch ca be a few tmes the badwdth of the voce applcato. Fg. 3. Gves the percetage crease the umber of users for a fxed amout of UL resources. Two values of BER are compared. As gve [5] based o the SNR requremets of the dfferet modulato schemes (or burst profles) the cell area ca be classfed to aulus regos. The boudares mark the chage to a lower modulato. The area of these auluses as a percetage of the total cell area s gve by a, where for example a 3 represets the rego whch ca use ay oe of the three lowest modulato schemes. The 3 rd colum of Table I cotas these values. We assume that the total umber of SSs the cell s uformly spread out. b r, s the effectve bt rate of a SS the th aulus for a radomly chose t. b o, s the effectve bt rate of a SS the th aulus for a optmally chose t (whch s lower tha the radomly selected t ). r ad o are the umber of users the system for radom t ad optmal t, usg the same amout of resources measured OFDM symbols. The rato of o : r gve (9) s plotted as a percetage crease Fg. 3. r, r a = o r = Modulato Scheme r, a = = o, oa = TABLE I Bts per OFDM symbol ( bps ) o, a = The ga due to optmal selecto of t s more at hgher BERs due to the creased retrasmsso overhead. A bgger ga ca be acheved by usg a optmal t whch s larger tha the radom t but ths wll crease the latecy. The average latecy l avg s gve by (10). Ths cludes a addtoal compoet t the summato whch accouts for the lag due to packetzato. t ( ) = l avg = t + + T f 1 PER PER (10) 0 2 T f s the frame durato. For low BER values (<1e-4) latecy s roughly equal to 1.5t. Wthout latecy mmzg ehacemets we have cosdered the MAC servce delay to be half of the packetzato terval. At every talk spurt the startg pot could radomly fall Percetage of total area (a ) BPSK 1/2 96 39.40 QPSK 1/2 192 20.56 QPSK 3/4 288 2.95 16 QAM 1/2 384 4.10 16 QAM 3/4 56 5.15 64 QAM 1/2 68 0.92 64 QAM 3/4 864 1.92 Dfferet modulato schemes used ad the bts per OFDM for each of them. (9)
aywhere the rage (0, t ] from the curret grat posto. IV. Implemetato Scheme At the start of the servce flow the tatg SS wll sed a DSA_REQ message to the BS. If the recever s also a part of a IEEE 802.16 cell t too should follow the same procedure. To do ths the applcato layer of the SS must commucate wth the MAC layer to alert of the begg of the voce stream. A BS receves DSA_REQ If the BS agrees to all the parameters specfed t wll echo these back a DSA_RSP message. (It would seem logcal that ths step occurs after the SS has setup the sesso wth the recever usg H.323, SIP or aother setup protocol) The procedure for ths s show Fg. 4. If however the requested parameters are ot optmal ad ca be substtuted by more effcet oes the BS wll dcate these the DSA_RSP message. Oce a set of values s agreed upo the SS wll cofrm the use of the parameters by sedg a DSA_ACK to the BS. For applcatos whch caot chage t the SS should dcate ths to the BS. We propose usg oe of the uused Servce Flow Parameters [1] the DSA_REQ as a dcator. The BS wll ot attempt to optmze such parameters. Y UGS flow Use lookup tables for packet sze ad latecy N BS: Parameters OK Y BS seds DSA_RSP A N SS: Parameters OK Y SS seds DSA_ACK Complete rest of VoIP setup process Fg. 4. Flow dagram of the procedure to determe a optmal parameter set at the start of a UGS servce flow. I the frst decso box f the flow s ot a UGS type the treatmet wll be dfferet ad s ot show here. Fg. 5. Usablty of varous packetzato tervals. H ad L dcate Hgh ad Low usablty respectvely. A. Lookup Table Creato ad Usage The BS eeds to have lookup tables for a rage of BER/SNR values ad burst profles, so that t ca select the most sutable. Fg. 5 gves lookup table lke data for BER of 1e-4. H ad L dcate Hgh ad Low usablty. As a example f a SS usg 64 QAM 3/4 requests a 50ms terval the BS could respod wth 1 of 2 possble optos (lowest subplot of Fg. 5).
1) 6ms the more effcet opto. Has a hgher latecy ad P loss. 2) 41ms less effcet that 1) but has lower latecy ad P loss. Based o chael codtos whch the BS has kowledge of, ad estmated delay to the destato t ca select the best opto. It s mportat to remember that eve though for the aalyss we have cosdered t to be the ut of cocer Fg. 5, a lookup table at the BS would have to be based o packet sze Bytes. There s a lear relatoshp betwee them as gve (1). V. Cocluso 802.16 BWA systems," preseted at Vehcular Techology Coferece, 2004. VTC2004-Fall. 2004 IEEE 60th, 2004. [5] C. Hoyma, "Aalyss ad performace evaluato of the OFDM-based metropolta area etwork IEEE 802.16," Computer Networks, vol. 49, pp. 341-363, 2005. [6] G. Chu, D. Wag, ad S. Me, "A QoS archtecture for the MAC protocol of IEEE 802.16 BWA system," preseted at Commucatos, Crcuts ad Systems ad West So Expostos, IEEE 2002 Iteratoal Coferece o, 2002. The effcecy of badwdth usage s affected by the choce of packet sze IEEE 802.16. Ths s more proouced whe the packet sze s relatvely small such as VoIP applcatos. It has bee show that by careful selecto of packetzato tervals for VoIP the umber of users ca be creased ad badwdth wastage o overheads mmzed. A modfcato was troduced to the MAC operato to be able to chage the terval durg call setup. Ths modfcato ca be accommodated the exstg DSX had shakg process so o extra overhead s troduced. Creatg of a lookup table was proposed at the BS to make selectg a optmal terval fast ad smple. We are curretly lookg at the effect of packet sze o the other schedulg types 802.16. Refereces [1] "IEEE stadard for coformace to IEEE 802.16. Part 3: rado coformace tests (RCT) for 10-66 GHz WrelessMAN-SC ar terface," IEEE Std 802.16/Coformace03-2004 (Coformace to IEEE Std 802.16-2001 as ameded by IEEE Std 802.16a -2003 ad IEEE Std 802.16c-2002), 2004. [2] Y. Yao ad J. Su, "Study of UGS grat sychrozato for 802.16," preseted at Cosumer Electrocs, 2005. (ISCE 2005). Proceedgs of the Nth Iteratoal Symposum o, 2005. [3] H. Lee, T. Kwo, ad D.-H. Cho, "A ehaced uplk schedulg algorthm based o voce actvty for VoIP servces IEEE 802.16d/e system," Commucatos Letters, IEEE, vol. 9, pp. 691-693, 2005. [4] H. Lee, T. Kwo, ad D.-H. Cho, "A effcet uplk schedulg algorthm for VoIP servces IEEE