Outdoor IEEE Cellular Networks: MAC Protocol Design and Performance

Transcription

1 Outoor IEEE Cellulr Networks: MAC Protocol Design n Performnce Kin K. eung, Bruce McNir, eonr J. Cimini, Jr., n Jck H. Winters AT&T bs - Reserch Miletown, NJ Abstrct We explore the fesibility of esigning n outoor cellulr network bse on the IEEE specifiction. Since the stnr is intene for wireless locl-re networks (WAN), there re mny technicl chllenges when pplying the ir interfce to the outoor environment. We stuy here how the meium ccess control (MAC) protocol cn be pplie n how it performs in the outoor network. By exploiting the fct tht timeout intervls re not explicitly specifie, without moifying the stnr, we propose new timing structure for the istribution coorintion function (DCF) n the hnshke of request-to-sen (RTS) n cler-to-sen (CTS) to hnle increse signl propgtion ely in the outoor network. We fin tht the DCF n RTS/CTS protocols s specifie in the stnr continue to work properly for link istnce up to 6 km. Our nlysis revels tht the DCF performnce egres slightly in the network with cell size of 6 km when compre with the 600 m WAN. Thus, s fr s the MAC protocol is concerne, the outoor, cellulr network with 6 km cell size is fesible. I. INTRODUCTION While the wireless inustry is ctively eveloping, testing n eploying thir genertion (3G) wireless networks, customers re expecting services with t rte higher thn tht to be provie by 3G networks. To meet such emn for better services, mny compnies hve strte to provie highspee t services using wireless locl-re-networks (WAN) in plces such s irports n hotels. Such n pproch is prticulrly ttrctive ue to the mturity n low cost of the IEEE b technology [I99b, VAM99]. The b network provies t rtes up to 11 Mbps, fr exceeing tht to be offere by, for exmple, EDGE [SAE98, CQW99] n W-CDMA networks [HT00]. Besies high t rtes, b networks offer severl vntges over 3G networks. First, the cost of b equipment is much lower thn tht for 3G equipment becuse of the simple esign of the former networks, couple with competition mong WAN venors. Secon, b networks operte in the 2.4 GHz ISM bn, which is free spectrum. In contrst, the 3G spectrum is license n very expensive. Thus, both resons mke the operting cost of the 3G network higher thn tht for the WAN. On the other hn, ech WAN cn serve only smll re, up to few hunre meters, where cell rius of ten kilometers is supporte in the 3G networks. In ition, future 3G networks re expecte to provie ubiquitous coverge n vilbility. In contrst, public WAN service is vilble only in isolte plces such s irports n hotels. Users will use both types of networks, one for excellent coverge while the other for enhnce t rtes. In this reserch, we explore the following question: Is it possible to esign n outoor, cellulr network bse on the existing ir-interfce stnr for wireless t services? If the nswer is ffirmtive, then users cn use the sme ir interfce mechnism to obtin wireless services from inoor WAN n outoor networks. There re mny technicl issues pertinent to the esign of n cellulr network. Recll tht s well s its extension b [I99b] n [I99] stnrs were evelope specificlly for WAN with the trnsmission rnge up to few hunre meters in inoor environment. First, the signl propgtion ely increses when pplying the to outoor networks reltive to the inoor WAN, which in turn my ffect the pplicbility of the meium ccess control (MAC) protocol. Secon, the outoor environment hs increse ely spre tht cuses intersymbol interference. Further, Doppler effects ue to mobility my require sophisticte processing for chnnel estimtion. The focus of this pper is on the MAC protocol esign n performnce when using the specifiction for outoor, cellulr networks, while rio issues will be resse in our subsequent ppers. Much work relte to the MAC protocol hs been publishe; see e.g., [B00], [CCG00] n [VCM01]. The orgniztion of the rest of this pper is s follows. We provie brief escription of the MAC protocols in Section II. In Section III, we iscuss how the protocols my or my not work properly in the outoor networks. In ition, we estimte the mximum cell rius in outoor networks ue to the consiertion of MAC protocols. Section IV nlyzes the MAC performnce for outoor networks. Finlly, our conclusion is in Section V. II. IEEE MAC PROTOCOS The IEEE specifiction [I97] llows three kins of physicl lyer: irect sequence spre spectrum (DSSS), frequency hopping spre spectrum (FHSS) n infrre (IR). In prticulr, the DSSS esign supports t rtes of 1 n 2 Mbps. Subsequently, while mintining bckwr comptibility to the DSSS specifiction, the b ws opte to support t rtes of 5.5 n 11 Mbps, operting in the 2.4 GHz bn (the ISM bn). As result, the b network cn support 1, 2, 5.5 n 11 Mbps, epening on rio conitions. Another extension is , which uses n entirely ifferent physicl lyer known s orthogonl frequency ivision multiplexing (OFDM)

2 cn support t rtes rnging from 6 to 54 Mbps, operting in the 5.5 GHz bn (the U-NII bn). It is importnt to note tht it is the b networks tht hve been wiely use recently. For this reson, we focus on b networks here. We lso note tht lthough t rtes hve been increse, b networks continue to use the originl MAC protocol in the specifiction. Furthermore, the MAC protocol supports the inepenent bsic service set (BSS), which hs no connection to wire networks (i.e., n -hoc wireless network), s well s n infrstructure BSS, which inclues n ccess point (AP) connecting to wire network. The ltter is similr to cellulr networks with bse sttions replce by AP s. We consier only the infrstructure BSS in this pper. We provie brief escription of the MAC protocol here [I97, OP99]. The specifiction efines five timing intervls for the MAC protocol. Two of them re consiere to be bsic ones tht re etermine by the physicl lyer: the short interfrme spce (SIPS) n the slot time. The other three intervls re efine bse on the two bsic intervls: the priority interfrme spce (PIFS) n the istribute interfrme spce (DIFS), n the extene interfrme spce (EIFS). The is the shortest intervl, followe by the slot time. The ltter cn be viewe s time unit for the MAC protocol opertions, lthough the chnnel s whole oes not operte on slotte-time bsis. For b networks (i.e., with DSSS physicl lyer), the n slot time re 10 n 20 µs, respectively. The slot time of 20 µs is chosen to ccount for the signl propgtion n processing elys. The PIFS is equl to plus one slot time, while the DIFS is the plus two slot times. The EIFS is much longer thn the other four intervls, n is use if t frme is receive in error. The MAC supports two moes of opertion: the Point Coorintion Function (PCF) n the Distribute Coorintion Function (DCF). The PCF provies contentionfree ccess, while the DCF uses the crrier sense multiple ccess with collision voince (CSMA/CA) mechnism for contention bse ccess. The two moes re use lterntely in time. Tht is, contention-free perio by the PCF is followe by contention perio of the DCF. A. The PCF Protocol In the PCF protocol, n AP polls its ssocite mobile sttions one fter nother by sening polling messges. If the AP hs t to sen to mobile sttion being polle, the t cn be inclue in the polling messge. If the polle sttion hs t for the AP, it is sent in the response messge. When pplicble, n cknowlegment (which cknowleges receipt of previous t frme from the AP) cn lso be inclue in the response messge. As n illustrtive exmple in Figure 1, the AP first sens the polling messge n t, if ny, to mobile sttion 1 (enote by S1). Sttion 1 shoul immeitely sen n cknowlegment or t frme, if ny, to the AP within the intervl. After receiving n ACK or t from sttion 1, the AP polls mobile sttion 2 within the intervl. In this illustrtion, sttion 2 oes not respon, either becuse the polling messge is lost or sttion 2 hs no t to sen to the AP. In this cse, s response is not receive from sttion 2 before the expires, the AP moves on to poll sttion 3 within the PIFS intervl, which strts from the en of the lst polling messge for sttion 2. AP Sttion 1 AP Dt: S1 Figure 1. The PCF of the MAC Protocol B. The DCF Protocol The DCF employs the CSMA/CA mechnism n works s follows. A sttion (incluing the AP) with new pcket rey for trnsmission senses whether or not the chnnel is busy. If the chnnel is etecte ile for DIFS intervl (i.e., 50 µs for b networks), the sttion strts pcket trnsmission. Otherwise, the sttion continues to monitor the chnnel busy or ile sttus. After fining the chnnel ile for DIFS intervl, the sttion: ) strts to tret chnnel time in units of slot time, b) genertes rnom bckoff intervl in units of slot time, n c) continues to monitor whether the chnnel is busy or ile. In the ltter step, for ech slot time where the chnnel remins ile, the bckoff intervl is ecremente by one. When the intervl vlue reches zero, the sttion strts pcket trnsmission. During this bckoff perio, if the chnnel is sense busy in slot time, the ecrement of the bckoff intervl stops (i.e., is frozen) n resumes only fter the chnnel is etecte ile continuously for the DIFS intervl n the following one slot time. Agin, pcket trnsmission is strte when the bckoff intervl reches zero. The bckoff mechnism helps voi collision since the chnnel hs been etecte to be busy recently. Further, to voi chnnel cpture, sttion must wit for bckoff intervl between two consecutive new pcket trnsmissions, even if the chnnel is sense ile in the DIFS intervl. This is epicte in Figure 2. Tx strts sensing DIFS Pcket 1 Ack or Dt Dt: S2 Ack PIFS (30 µs) DIFS + Bckoff Figure 2. The DCF of the MAC Protocol The bckoff mechnism for the DCF is n exponentil one. For ech pcket trnsmission, the bckoff time in units of slot time (i.e., n integer) is uniformly chosen from 0 to n-1, where n epens on the number of file trnsmissions for the pcket. At the first trnsmission ttempt, n is set to vlue of CW min =32, the so-clle minimum contention winow. After AP Dt: S3 Pcket 2

3 ech unsuccessful trnsmission, n is ouble, up to mximum vlue of CW mx =1024. The specifiction requires receiver to sen n ACK for ech pcket tht is successfully receive. Furthermore, to simplify the protocol heer, n ACK contins no sequence number, n is use to cknowlege receipt of the immeitely previous pcket sent. Tht is, sttions exchnge t bse on stop-n-go protocol. As shown in Figure 2, the sening sttion is expecte to receive the ACK within the 10 µs intervl fter the pcket trnsmission is complete. If the ACK oes not rrive t the sening sttion within specifie ACK_timeout perio, or it etects trnsmission of ifferent pcket on the chnnel, the originl trnsmission is consiere to hve file n is subject to retrnsmission by the bckoff mechnism. In ition to the physicl chnnel sensing, the MAC protocol implements network lloction vector (NAV), whose vlue inictes to ech sttion the mount of time tht remins before the chnnel will become ile. All pckets contin urtion fiel n the NAV is upte ccoring to the fiel vlue in ech pcket trnsmitte. The NAV is thus referre to s virtul crrier sensing mechnism. The MAC uses the combine physicl n virtul sensing to voi collision. The protocol escribe bove is clle the two-wy hnshking mechnism. In ition, the MAC lso contins four-wy frme exchnge protocol. Essentilly, the four-wy protocol requires tht sttion sen to the AP specil, Request-to-Sen (RTS) messge, inste of the ctul t pcket, fter gining chnnel ccess through the contention process escribe bove. In response, if the AP sees tht it is pproprite, it sens Cler-to-Sen (CTS) messge within the intervl to instruct the requesting sttion to strt the pcket trnsmission immeitely. The min purpose of the RTS/CTS hnshke is to resolve the so-clle hien terminl problem. III. MAC PROTOCOS IN OUTDOOR NETWORKS A. The PCF Protocol Infesible It is importnt to emphsize tht the n PIFS timing requirements for the PCF in Figure 1 re clerly efine in the stnr. In prticulr, the most stringent requirement is tht the ACK hs to be receive from the polle sttion to the AP within the intervl, which is 10 µs for b networks. When the stnr is use for outoor, cellulr networks, the istnce between mobile sttion n its AP is expecte to be longer thn tht in the WAN. Consier link istnce of 1.5 km s n exmple. The roun-trip signl propgtion ely for the 1.5 km istnce requires 10 µs. Since t lest severl µs re neee for signl processing t the receiver, the link istnce is likely to be limite to hunres of meters, s in WAN environments. In fct, this is the intention of the specifiction. Thus, it is unrelistic to expect tht the PCF cn be supporte for outoor networks with cell rius of severl km. B. Applicbility of the DCF Protocol et us consier the DCF in the outoor networks. It is worth noting tht s fr s the MAC protocol is concerne, the mjor ifference between outoor networks n their WAN counterprts is increse signl propgtion ely. As shown in Figure 2, the mjor constrint for the pplicbility of the DCF in outoor networks is tht the ACK is expecte to be receive within the intervl fter pcket trnsmission. Tht is, the 10 µs inclues the roun-trip signl propgtion n processing t the receiver. However, in orer to be useful, we im t hving n outoor cell size of severl km. Thus, the one-wy signl propgtion ely cn be more thn 10 µs, even neglecting the return propgtion n processing time. Eviently, this woul not be prcticl without violting the protocol specifiction. Our solution is bse on the following key observtion: Typiclly, there is no consequence if the ACK is receive lter thn the intervl. This is becuse, fter sttion trnsmits pcket, it strts n ACK_timeout perio, which is not specifie in the stnr n is usully chosen to be vlue much lrger thn 10 µs by venors. Thus, s long s the ACK is receive before the timeout expires, the MAC protocol respons properly. As in typicl implementtions, we ssume tht the ACK_timeout perio is longer thn the DIFS intervl of 50 µs. Then, we rgue tht s long s the ACK rrives t the sening sttion within the DIFS intervl following pcket trnsmission, the DCF opertes properly in the outoor network environment where the link istnce cn rech s much s severl km. The resoning is s follows. First, becuse the ACK is receive within the DIFS intervl, the ACK_timeout hs not expire so tht the protocol cn respon upon receipt of the ACK s if it were receive within the intervl, s originlly specifie in the protocol stnr. Secon, since the DCF protocol requires ny sttion to sense the chnnel being ile for t lest the DIFS intervl before trnsmitting, the return of the ACK within the DIFS intervl following the previous pcket trnsmission by the sening sttion prevents ny sttions other thn the receiving one from gining ccess to the chnnel. Consequently, the chnnel is implicitly reserve for the receiving sttion to sen the ACK. In ition, the piring of pcket trnsmission n its ACK trnsmitte in sequence for ny pir of sening n receiving sttions remins intct, s require by the specifiction. Extening the rrivl ely of ACK from the to the DIFS intervl comes with penlty. Tht is, the computtion of the NAV ssumes tht the ACK returns within the intervl. So, the ely extension cuses n erroneous etermintion of the NAV, thus incorrect virtul sensing. Nevertheless, since protocol opertions re bse on both physicl n virtul chnnel sensing, s long s the former works properly, the mlfunctioning of the virtul sensing ue to incorrect NAV vlue cuses no pprent, negtive impcts. Actully, the extension of the ACK rrivl ely from the intervl to the DIFS intervl cn lso be pplie to the RTS n CTS hnshke for resolving the hien terminl

4 problem. Specificlly, sening sttion strts CTS_timeout perio fter sening n RTS. The MAC protocol specifies tht the CTS, if ny, is suppose to rrive from the receiving sttion within the intervl. However, similr to the ACK_timeout, the CTS_timeout perio is typiclly chosen to be much longer thn 10 µs by equipment mnufcturers. Therefore, by the sme rguments iscusse bove, the rrivl ely for the CTS cn be extene to the DIFS intervl. trnsmitting. The pcket trnsmission time is ssume to be constnt µs. Consier the chnnel ctivity for successful pcket trnsmission. The chnnel is ile for µs n followe by pcket trnsmission of µs. As Figure 3 shows, the trnsmitter wits for µs (DIFS intervl) for the ACK. et the ACK trnsmission time be c µs. The chnnel is sense ile gin by ll sttions µs fter the ACK trnsmission. A. Mximum Cell Size for the DCF Protocol With the rrivl ely for the ACK n CTS extene to the DIFS intervl, let us consier its limit on the mximum cell size (i.e., link istnce) in outoor networks. Tx strts sensing DIFS (50 µs) Pkt/RTS DIFS (50 µs) 20µs 10µs 20µs Ack/CTS Figure 3. Alloction of ACK/CTS ely Recll tht the ACK n CTS rrivl ely consists of roun-trip signl propgtion ely n signl processing time. As shown in Figure 3, one resonble lloction of the 50 µs DIFS ely is: one-wy signl propgtion ely of 20 µs n processing time of 10 µs t the receiving sttion. The ltter shoul not cuse processing buren for the receiver becuse the originl ely of the intervl is 10 µs. For the 20 µs propgtion ely, the mximum cell size is bout 6 km. In other wors, with the cell size of 6 km or less, the DCF protocol opertes properly in cellulr networks. IV. DCF PERFORMANCE IN OUTDOOR NETWORKS We present n pproximte nlysis of the DCF throughput for outoor networks n WAN. As shown in Figure 3, if sttion with pcket for trnsmission senses the chnnel ile for the DIFS intervl (enote by in µs in the following), it strts to trnsmit. Following the pcket trnsmission, the chnnel remins ile for the DIFS intervl n then the ACK is trnsmitte by the receiver. If the sening sttion senses the chnnel busy, it goes through the bckoff mechnism iscusse bove. For simplicity, we o not moel the etils of the bckoff lgorithm. Inste, it is ssume tht the ggregte trffic, which inclues new pckets n trnsmission rettempts, from ll sttions forms Poisson process with n intensity of G pckets/µs. This ssumption is resonble if the bckoff perio is sufficiently long so tht new trnsmission n rettempts become inepenent sources. For simplicity, ssume tht the signl propgtion ely in µs is ienticl between ny pir of sttions. Thus, the vulnerble perio is lso given by, uring which new pcket trnsmission cnnot be sense by other sttions. As result, these sttions uner the CSMA protocol cn possibly strt their own trnsmissions n cuse collisions. Ech sttion senses the chnnel ile for µs (DIFS intervl) before Y Figure 4. Busy perio with Collie Trnsmissions Figure 4 shows typicl busy perio with collie trnsmissions ue to the vulnerble perio for the CSMA protocol, where Y enotes the time spn between the first n the lst pcket trnsmissions in the busy perio. Using the result in [K76], the verge urtion of Y is given by e G 1 Y = (1) G. The verge length of busy perio (which contins successful trnsmission or collisions) is given by G B = + Y ( + c) e (2) where the lst term ccounts for the witing n trnsmission time of the ACK for successful trnsmission with probbility G e, bse on the Poisson ssumption of ggregte trffic. By the sme ssumption, the verge cycle time, consisting of busy perio n the following ile perio, is given by G T = + Y ( + c) e + 1 G The chnnel throughput S is efine s the frction of time t which t is successfully trnsmitte. Thus, we hve G e S = (4) T where the numertor is the verge mount of time when t is trnsmitte without collision n T is obtine from (3). Three common pcket sizes of 60 bytes (e.g., TCP ACK), 576 bytes (typicl size for web browsing) n 1500 bytes (the mximum size for Ethernet) plus 34 byte MAC heer re consiere. For n network with 1 Mbps t rte, the corresponing trnsmission time is 0.75, 4.88 n msec, respectively. The sensing ile time of the DIFS intervl of 50 µs n the trnsmission time c for the 112- bit ACK is µs. Bse on our iscussions bove, the link istnce is ssume to be 6 km, n thus the one-wy (3)

5 propgtion ely is 20 µs. For comprison, we lso consier WAN with service rius of 600 m with signl propgtion ely of 2 µs. In this WAN, fter pcket trnsmission, sttion wits for the intervl of 10 µs s in the stnr, inste of the DIFS intervl s shown in Figure 3, for the rrivl of the ssocite ACK. Applying these prmeters to (1) to (4), we obtin in Figure 5 the MAC throughput s function of the ggregte trffic lo for selecte pcket lengths. As expecte, when the link istnce increses from 600 m to 6 km for given pcket length, the mximum throughput ecreses becuse of the increse signl propgtion ely n thus the vulnerble perio. For the 576-byte pcket size, the mximum throughput rops from 92.9% to 84.8%, when the link istnce increses from 600 m to 6 km. Nevertheless, since 576-byte size is typicl for populr web pplictions, the throughput of 84.8% is still stisfctory. For 1500-byte pckets, the chnnel throughput for the 6 km cell cn rech mximum of 90.8%. Even for the short TCP ACKs of 60 bytes long, the chnnel throughput is bout 60%. In summry, the MAC throughput is still stisfctory espite the increse of cell size to 6 km. Figure 5. MAC Throughput Comprison. V. CONCUSION AND FUTURE WORK We hve stuie how the MAC cn be pplie n how it performs in outoor networks. By exploiting the fct tht timeout intervls re not explicitly specifie, without moifying the stnr, we hve propose new timing structure for the istribution coorintion function (DCF) n the hnshke of request-to-sen (RTS) n cler-to-sen (CTS) to hnle increse signl propgtion ely in the outoor network. It ws foun tht the DCF n RTS/CTS protocols s specifie in the stnr continues to work properly if the cell rius is less thn 6 km. Our nlysis revels tht the DCF performnce egres slightly for cell size of 6 km when compre with the 600 m WAN. Thus, s fr s the MAC protocol is concerne, the cellulr network with cell size of 6 km is fesible cellulr network. We lso pln to investigte techniques such s vnce equlizers, smrt ntenns n cll mission control to further improve the performnce of the outoor cellulr networks. ACKNOWEDGMENTS We woul like to thnk Mrtin Clrk, Peter Driessen, Zorn Kostic n Stefn Muller-Weinfurtner for their iscussion. REFERENCES [B00] G. Binchi, Performnce Anlysis of the IEEE Distribute Coorintion Function, IEEE J. Selecte Ares in Commun., Vol. 18, Mrch 2000, pp [CCG00] F. Cli, M. Conti n E. Gregori, Dynmic Tuning of the IEEE Protocol to Achieve Theoreticl Throughput imit, IEEE/ACM Trns. on Networking, Vol. 8, Dec. 2000, pp [CMK01] M.V. Clrk, K.K. eung, B. McNir n Z. Kostic, Outoor IEEE Cellulr Networks: Rio ink Performnce, Proc. IEEE ICC [CQW99] J. Chung, X. Qiu n J. Whitehe, Dt Throughput Enhncement in Wireless Pcket Systems by Improve ink Apttion with Appliction to the EDGE System, Proc. of IEEE VTC 99, Sept [HT00] H. Holm n A. Toskl (E.), WCDMA for UMTS, John Wiley & Sons, New York, [I97] IEEE , Wireless AN Meium Access Control (MAC) n Physicl yer (PHY) Specifiction, [I99] IEEE , Prt 11: Wireless AN Meium Access Control (MAC) n Physicl yer (PHY) Specifiction: High-Spee Physicl yer Extension in the 5 GHz Bn, [I99b] IEEE b, Prt 11: Wireless AN Meium Access Control (MAC) n Physicl yer (PHY) Specifiction: High-Spee Physicl yer Extension in the 2.4 GHz Bn, [K76]. Kleinrock, Queueing Systems, Vol. 2: Computer Applictions, John Wiley & Sons, 1976, p [OP99] B. O Hr n A. Petrick, IEEE Hnbook, IEEE Press, New York, [SAE98] P. Schrmm, et. l., Rio Interference Performnce of EDGE, Proposl for Enhnce Dt Rtes in Existing Digitl Cellulr Systems, Proc. of IEEE VTC, Ottw, Cn, My 1998, pp [VAM99] R. vn Nee, et. l., New High-Rte Wireless AN Stnr, IEEE Commun. Mg., Dec. 1999, pp [VCM01] M. Veerrghvn, N. Cocker n T. Moors, Support of Voice Services in IEEE Wireless ANs, Proc. of IEEE Infocom 2001, Anchorge, Alsk, April 2001, pp In terms of future work, mjor issue is to exmine n enhnce the rio esign so tht it performs properly in the cellulr environment. In compnion pper [CMK01], we shll ress the issue of rio link performnce in the