/
IEEE Communications Magazine  September   IEEE IEEE Communications Magazine  September   IEEE

IEEE Communications Magazine September IEEE - PDF document

faustina-dinatale
faustina-dinatale . @faustina-dinatale
Follow
465 views
Uploaded On 2014-12-17

IEEE Communications Magazine September IEEE - PPT Presentation

11 Wireless Local Area Networks 01636804971000 1997 IEEE ireless computing is a rapidly emerging technology providing users with network connectivity without being tethered off of a wired network Wireless local area net works WLANs like their wired ID: 25409

Wireless Local Area Networks

Share:

Link:

Embed:

Download Presentation from below link

Download Pdf The PPT/PDF document "IEEE Communications Magazine September ..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.


Presentation Transcript

IEEE Communications Magazine ¥ September 1997 being tethered off of a wired network. Wireless local area net- BSTRACT The draft IEEE 802.11 Wireless Local Area Network (WLAN) specification is approaching completion. In this article, the IEEE 802.11 The views and opinions expressed in this article are those of the authorsand do not reflect MITREÕs or Fujitsu Network CommunicationsÕ current IEEE Communications Magazine ¥ September 1997cient, resulting in ÒsleepÓ modes andlow-power displays, causing users tomake cost versus performance andcost versus capability trade-offs.Human Safety Ñ Research is ongo-ing to determine whether radio fre-quency (RF) transmissions from radioand cellular phones are linked tohuman illness. Networks should bedesigned to minimize the powertransmitted by network devices. Forinfrared (IR) WLAN systems, optical transmitters must bedesigned to prevent vision impairment.Mobility Ñ Unlike wired terminals, which are static whenoperating on the network, one of the primary advantages ofwireless terminals is freedom of mobility. Therefore, systemdesigns must accommodate handoff between transmissionboundaries and route traffic to mobile users.Throughput Ñ The capacity of WLANs should ideallyapproach that of their wired counterparts. However, due tophysical limitations and limited available bandwidth, WLANsare currently targeted to operate at data rates between 1Ð20Mb/s. To support multiple transmissions simultaneously,spread spectrum techniques are frequently employed.Currently, there are two emerging WLAN standards: theEuropean Telecommunications Standards Institute (ETSI)High-Performance European Radio LAN (HIPERLAN) andthe IEEE 802.11 WLAN. Both draft standards cover the phys-ical layer and medium access control (MAC) sublayer of theopen systems interconnection (OSI) seven-layer referencemodel. The HIPERLAN committee has identified the5.15Ð5.30 GHz and 17.1Ð17.2 GHz bands for transmission.The 5 GHz band has been ratified for HIPERLAN use by theConference of European Postal and TelecommunicationsAdministrations (CEPT). Data rates up to 23.529 Mb/s areprojected, and multihop routing, time-bounded services, andpower-saving features are expected. For further informationregarding HIPERLAN, see the article by LaMaire et al.[1] orthe HIPERLAN specification [2].The IEEE is developing an international WLAN standardidentified as IEEE 802.11 [3]. This project was initiated in 1990,and several draft standards have been published for review.The scope of the standard is Òto develop a Medium AccessControl (MAC) and Physical Layer (PHY) specification forwireless connectivity for fixed, portable and moving stationswithin a local area.Ó The purpose of the standard is twofold:ÒTo provide wireless connectivity to automatic machin-ery, equipment, or stations that require rapid deploy-ment, which may be portable, or hand-held or which maybe mounted on moving vehicles within a local areaÓ¥ÒTo offer a standard for use by regulatory bodies to stan-dardize access to one or more frequency bands for thepurpose of local area communicationÓ [3].The IEEE 802.11 draft standard describes mandatory sup-port for a 1 Mb/s WLAN with optional support for a 2 Mb/sdata transmission rate. Mandatory support for asynchronousdata transfer is specified as well as optional support for dis-tributed time-bounded services (DTBS). Asynchronous datatransfer refers to traffic that is relatively insensitive to timetraffic like electronic mail and file transfers. Time-boundedtraffic, on the other hand, is traffic that is bounded by speci-fied time delays to achieve an acceptable quality of service(QoS) (e.g., packetized voice and video).Of particular interest in the speci-fication is the support for two funda-mentally different MAC schemes totransport asynchronous and time-bounded services. The first scheme,distributed coordination function(DCF), is similar to traditional legacypacket networks supporting best-effort delivery of the data. The DCFis designed for asynchronous datatransport, where all users with datato transmit have an equally fairchance of accessing the network. The point coordination func-tion (PCF) is the second MAC scheme. The PCF is based onpolling that is controlled by an access point (AP). The PCF isprimarily designed for the transmission of delay-sensitive traf-fic. While the DCF has been studied by several researchers[4Ð7], the combined performance of the DCF and PCF oper-ating in a common repetition interval is much less understood.In this article, the performance of an ad hoc network (DCF-only) and an infrastructure network (DCF and PCF) areinvestigated by means of simulation. We also investigate theeffect of channel errors on the performances of PCF andDCF, which is absent in all previous studies. Channel degra-dation, in terms of burst errors due to multipath fading, willbe factored into the simulations, and the effects on through-put and delay will be determined. We also develop an effi-cient polling scheme used during the PCF to drop inactivestations from the polling list for a polling cycle, thereby pro-viding more bandwidth to currently active stations.In the remainder of the article, we will summarize theIEEE 802.11 WLAN specification (emphasis on the MACsublayer), briefly describe the simulation model which sup-ports asynchronous data and packetized voice traffic, and pro-vide performance results from the simulation.DESCRIPTIONOFTHEIEEE 802.11 DRAFTSTANDARDARCHITECTUREbasic service set(BSS) is the fundamental building blockof the IEEE 802.11 architecture. A BSS is defined as a groupof stations that are under the direct control of a single coordi-nation function (i.e., a DCF or PCF) which is defined below.The geographical area covered by the BSS is known as thebasic service area(BSA), which is analogous to a cell in a cel-BSS can communicate directly with all other stations in a BSS.fading, or interference from nearby BSSs reusing the samephysical-layer characteristics (e.g., frequency and spreadingcode, or hopping pattern), can cause some stations to appearÒhiddenÓ from other stations.An ad hoc network is a deliberate grouping of stations intoa single BSS for the purposes of internetworked communica-tions without the aid of an infrastructure network. Figure 1 isan illustration of an -mal name of an ad hoc network in the IEEE 802.11 standard.Any station can establish a direct communications sessionwith any other station in the BSS, without the requirement ofchanneling all traffic through a centralized access point (AP).In contrast to the ad hoc network, infrastructure networksare established to provide wireless users with specific servicesand range extension. Infrastructure networks in the context ofIEEE 802.11 are established using APs. The AP is analogousto the base station in a cellular communications network. The Figure 1.Sketch of an ad hoc network. Independent BSS Mb/s of instantaneous bandwidth. Three different hoppingMb/s of instantaneous bandwidth. Three different hoppingis 2.5 hops/s. The basic access rate of 1Mb/s uses two-level Gaussian frequencyshift keying (GFSK), where a logical 1 isencoded using frequency Fc+ fand a log-ical 0 using frequency FcÐ f. The enhancedaccess rate of 2 Mb/s uses four-levelGFSK, where 2 bits are encoded at a timeusing four frequencies.The DSSS also uses the 2.4 GHz ISMfrequency band, where the 1 Mb/sbasic rate is encoded using differentialbinary phase shift keying (DBPSK),and a 2 Mb/s enhanced rate uses dif-ferential quadrature phase shiftkeying (DQPSK). The spreading isdone by dividing the available band-width into 11 subchannels, each 11MHz wide, and using an 11-chipBarker sequence to spread eachdata symbol. The maximum channelcapacity is therefore (11 chips/sym-bol)/(11 MHz) = 1 Mb/s if DBPSKis used [8]. Overlapping and adja-cent BSSs can be accommodated byensuring that the center frequenciesof each BSS are separated by atleast 30 MHz [3]. This rigid require-ment will enable only two overlap-ping or adjacent BSSs to operate without interference.The IR specification identifies a wavelength range from850 to 950 nm. The IR band is designed for indoor use onlyand operates with nondirected transmissions. The IR specifi-cation was designed to enable stations to receive line-of-siteand reflected transmissions. Encoding of the basic access rateof 1 Mb/s is performed using 16-pulse position modulation(PPM), where 4 data bits are mapped to 16 coded bits fortransmission. The enhanced access rate (2 Mb/s) is performedusing 4-PPM modulation, where 2 data bits are mapped to 4coded bits for transmission.MEDIUMACCESSCONTROLSUBLAYERThe MAC sublayer is responsible for the channel allocationprocedures, protocol data unit (PDU) addressing, frameformatting, error checking, and fragmentation and reassembly.The transmission medium can operate in the contention modeexclusively, requiring all stations to contend for access to thechannel for each packet transmitted. The medium can also Figure 2.Sketch of an infrastructure network. BSSBSSPortalIEEE 802.XDS: Distribution system Figure 3.Standard IEEE 802.11 frame format. control2 conn. ID2 6 6 6Address6 body 4 control2 version2 2 4 1From DS1 fragment1 1 mgt1 data1 1 1 whether the packet was received correctly. Upon receipt of a Figure 4.MAC architecture. Point coordinationfunction (PCF)MACextent Used for contention servicesand basis for PCFRequired for contention-freeservices Figure 5.Transmission of an MPDU without RTS/CTS. DataSourceDestinationOther DIFSCW ACKNAVBackoff after deferDefer access IEEE Communications Magazine ¥ September 1997updating their NAVs based on the RTS from the source sta-tion and CTS from the destination station, which helps tocombat the Òhidden terminalÓ problem. Figure 6 illustratesthe transmission of an MPDU using the RTS/CTS mecha-nism. Stations can choose to never use RTS/CTS, useRTS/CTS whenever the MSDU exceeds the value ofRTS_Threshold (manageable parameter), or always useRTS/CTS. If a collision occurs with an RTS or CTS MPDU,far less bandwidth is wasted when compared to a large dataMPDU. However, for a lightly loaded medium, additionaldelay is imposed by the overhead of the RTS/CTS frames.Large MSDUs handed down from the LLC to the MAC mayrequire fragmentation to increase transmission reliability. Todetermine whether to perform fragmentation, MPDUs are com-pared to the manageable parameter Fragmentation_Threshold.-old, the MSDU is broken into multiple fragments. The resultingthe last MPDU, which is of variable size not to exceed Frag-When an MSDU is fragmented, all frag-released until the complete MSDU has been transmitted suc-cessfully, or the source station fails to receive an acknowledg-ment for a transmitted fragment. The destination stationpositively acknowledges each successfully received fragment bysending a DCF ACKback to the source station. The sourcestation maintains control of the channel throughout the -mission of the MSDU by waiting only an SIFS period afterreceiving an ACK and transmitting the next fragment. When anACK is not received for a previously transmitted frame, thesource station halts transmission and recontends for the chan-Upon gaining access to thechannel, the source starts transmit-ting with the last unacknowledgedIf RTS and CTS are used, onlythe first fragment is sent using thehandshaking mechanism. The dura-tion value of RTS and CTS onlyaccounts for the transmission ofthe first fragment through thereceipt of its ACK. Stations in theBSS thereafter maintain their NAVby extracting the duration informa-tion from all subsequent fragments.The collision avoidance portionof CSMA/CA is performed througha random backoff procedure. If ation with a frame to transmitinitially senses the channel to be busy;then the station waits until the channelbecomes idle for a DIFS period, andthen computes a random backoff time.IEEE 802.11, time is slotted intime periods that correspond to aSlot_Time. Unlike slotted Aloha,where the slot time is equal to thetransmission time of one packet, theSlot_Time used in IEEE 802.11 ismuch smaller than an MPDU and isused to define the IFS intervals anddetermine the backoff time for stationsin the CP. The Slot_Time is differentfor each physical layer implementation.The random backoff time is an integervalue that corresponds to a number oftime slots. Initially, the station computesa backoff time in the range 0Ð7. After the medium becomesidle after a DIFS period, stations decrement their backofftimer until the medium becomes busy again or the timerreaches zero. -um becomes busy, the station is finally decremented to zero, the station transmits its frame. If-lision will occur, and each backoff time in the range 0Ð15. For each retransmission attempt,the backoff time grows as 22 + iá ranf()ûá Slot_Time, where iisthe number of consecutive times a station attempts to send anranf() is a uniform random variate in (0,1), and ëxûrepresents the largest integer less than or equal to x. The idleperiod after a DIFS period is referred to as the contention(CW). The advantage of this channel access method isthat it promotes fairness among stations, but its weakness isthat it probably could not support DTBS. Fairness is main-tained because each station must recontend for the channelafter every transmission of an MSDU. All stations have equalprobability of gaining access to the channel after each DIFSinterval. Time-bounded services typically support applicationssuch as packetized voice or video that must be maintainedwith a specified minimum delay. With DCF, there is nomechanism to guarantee minimum delay to stations support-ing time-bounded services.POINTCOORDINATIONFUNCTION(PCF)The PCF is an optional capability, which is connection-orient-ed, and provides contention-free (CF) frame transfer. ThePCF relies on the point coordinator (PC) to perform polling,enabling polled stations to transmit without contending for Figure 6.Transmission of an MPDUusing RTS/CTS. RTSSource DestinationOther NAV (RTS)NAV (CTS)NAV (data) Defer accessBackoff started CW DataCTSACK Figure 7.Transmission of a fragmented MPDU. SIFSSIFSFragment burstSIFSDest.OtherOther ACK 2ACK 1ACK 0NAV (CTS)NAV (fragment 1)NAV (fragment 0)NAV (ACK 0)NAV (ACK 11)NAV (frag 2)SIFSDIFS Fragment 1Fragment 2 The PCF is required to coexist with the DCF andof its primary functions is synchronization and timing. The8 is a sketch of the CFP repetition interval, illustrating the Figure 8.Coexistence of the PCF and DCF. CPCFPPCFDCF NAV CPCFP repetition intervalCFP repetition interval PCFDCFB Figure 9.PC-to-station transmission. PIFSB Contention free period NAV CP Figure 10.Station-to-station transmissions. PIFSB CFP repetition interval Contention period Contention free period NAV IEEE Communications Magazine ¥ September 1997STIONMODELTwo different simulation models are presented in this arti-cle. The first model represents an ad hoc network, whereall stations in the BSS are capable of directly communicatingwith all other stations in the BSS. All stations in the ad hocnetwork are assumed to be asynchronous data users. The sec-ond model represents an infrastructure network which charac-terizes a single BSS with an AP. The infrastructure networkoperates with asynchronous data users in the CP and packe-tized voice terminals operating in the CFP. Both simulationmodels are implemented using the physical-layer parametersspecified in the standard for the DSSS implementation. Moredetailed explanation of the simulation model is found in [9].Several assumptions have been made to reduce the com-plexity of the model. A short description of each of theassumptions is provided below:The effects of propagation delay on the model are neglect-ed. This is a fairly realistic assumption if transmission -tances are on the order of 100 ft between stations.¥The Òhidden terminalÓ problem is not addressed in thesimulation models.¥The basic rate of 1 Mb/s was simulated for the DSSS. Thisdecision was made because the enhanced rate, 2 Mb/s,would add additional complexity since control, multicast,and broadcast frames are required to be transmitted at thebasic rate (to ensure that all stations in the BSS can beproperly received), while management and data framesare transmitted at any available rate (1 Mb/s or 2 Mb/s).¥No stations operate in the Òpower-savingÓ mode (PS-Mode). By requiring all stations to be ÒawakeÓ at alltimes, transmitted MPDUs can be received immediatelyby the destination station without buffering at the AP.¥No interference is considered from nearby BSSs reusingthe same DSSS spreading sequence.When the PCF and DCF coexist together in the infra-structure network, all stations operating during the CP areasynchronous data users, and all users operating during theCFP are packetized voice users.A finite transmit buffer is maintained for each station. Ifthe finite buffer fills, all newly generated MSDUs will be con-sidered dropped without returning.For the ad hoc and infrastructure network simulations, aburst error model is introduced to characterize fading in thecommunications channel [10]. A two-state continuous-timeMarkov chain is used to represent the burst error model.Grepresents the channel in a ÒgoodÓ state. This indi-cates that the channel is operating with a very low bit errorrate (denoted by BERBindicates the channel isoperating in a fading condition with a higher error rate,denoted by BERbad. The transition rate from state Gto stateBis denoted by a, while the transition rate from state BtoGis denoted by b. A frame is considered to be corrupt ifit contains one or more bit errors.The simulation model uses the error model above to deter--ted successfully. When the frame is transmitted, a portion ofthe frame can be sent over the communications medium whenthe channel is in state G, and a portion can be transmittedwhen the channel is in state B. The number of bits transmit-ted in the frame during state Bis denoted by n1, and the num-ber transmitted during state Gn2frame is transmitted successfully is then calculated asPr{success} = (1 Ð BERbad)n1á (1 Ð BERgood)n2.ADHNMODELWith the ad hoc network model, all users are assumed to beasynchronous data users, and they shall operate in a self-contained BSS. The arrival of frames from a stationÕs higher-layer protocol to the MAC sublayer is modeled with exponentialinterarrival times and a truncated geometric distribution forthe frame lengths. The truncated geometric distribution isused to ensure that the MSDU does not exceed the maximumlength established by the specification (i.e., 2312 octets). How-ever, the simulation model can easily accommodate otherarrival processes and frame length distributions.During the simulation, if collisions or bit errors affect thetransmission of a frame, retransmission will occur according tothe backoff procedure described previously. The number ofretransmissions is limited before the frame is dropped from thestationÕs transmit queue. In the case of MSDUs smaller thanRTS_Threshold, the number of retransmissions is limited toShort_Retry_Limit. For MSDUs larger than RTS_Threshold, themaximum number of retransmissions is set by Long_Retry_Limit.The number of retransmissions is extended short RTS frames are not as wasteful of bandwidth as largerdata payloads. Typical default values used in the simulation ofthe ad hoc network are illustrated in Table 1.INFRASTRUCTURENETWORKMODELThe effect of a single BSS with an AP is simulated, whereasynchronous data users transmit during the CP and pack-etized voice users transmit during the CFP. The coexistenceof the DCF and PCF is illustrated in Fig. 8, where, for thepurposes of this simulation, the value of CFP_Max_Durationis provided in Table 2. The duration of theCFP_Repetition_Interval is approximately 0.4096 s; -fore, approximately 94 percent of the repetition intervalcan be allocated by the AP for contention-free services.During the CFP, if a station is polled by the AP to transmit,the station can transmit directly to another station in the BSS(Fig. 10) or to a station in another BSS. When the transmis-sion is directed to a station in another BSS, the source stationtransmits the frame to the AP, who is responsible for forwardingthe frame through the DS to the remote AP servicing the desti-nation station. Since the size of the BSS is relatively small, packetized voice activity is assumed to occur between stationsin different BSSs. Therefore, the simulation model directs allvoice traffic from a station through the AP. All voice trafficdestined for a mobile station is also delivered via the AP.The polling scheme during the CFP uses a cyclical schedul-ing algorithm, where each station is polled sequentially in theorder in which it is placed in the polling list. When the CFPends, the AP keeps track of the location in the polling list Figure 11.Burst error effects on data throughput. Data offered load0.2000.1Data throughput 0.2 0.3 0.4 0.5 0.6 0.70.8 0.4 0.6 0.81 bad 10-3BERbad 10-4BERbad 10-5BERbad 10-6 the station responds each time withoutthe station responds each time withoutThe length of the voice payloadshould be chosen so that voice packeti-zation delay is minimized and headeroverhead is not large, which is a conflict-ing goal. No retransmissions will be per-formed for voice frames since this trafficis delay-sensitive. QoS parameters for voice typically limit maxi-mum delay to 25 ms without echo canceling, and 500 ms usingecho canceling [12]. Asynchronous data frames are transmit-ted in the CP portion of the repetition interval using the DCFprotocol described above. Table 2 lists the additional defaultvalues used for simulation of the infrastructure network.SIMULATIONRESULTSSimulation results are shown for an ad hoc network and aninfrastructure network. The results below are presented inthe form of plots and, where applicable, with 95 percent confi-dence intervals. The throughput plots shown below representaggregate throughput. Approximate throughput per stationcan be calculated by dividing the aggregate throughput by thetotal number of data stations in the BSS.ADHOCNETWORKFor the ad hoc network, we assume allmobile stations generate asynchronousdata traffic with the same intensity.Figure 11 shows the aggregate datathroughput in megabits per second ver-sus the offered load in megabits persecond for several BERs (i.e., theBERbad). The offered load is definedto be the average number of bits persecond passed down to the MAC sub-layer at the source. The throughput isthe average number of bits per secondpassed up from the MAC sublayer atthe destination.Note that the burst error transitionrates for this model indicate that moretime will be spent in the ÒbadÓ statethan in the ÒgoodÓ state. When themedium is relatively clean, BERmum throughput can drop tospecified value. A bursty channel error model is used with the Table 1.Default attribute values for thead hoc network unless otherwise specified.Data stations10Average MSDU length1000 octetsChannel rate1 Mb/sBERgood10-10a30 s--1b10 s-1RTS_Threshold250 octetsFragmentation_Threshold800 octetsShort_Retry_Limit5Long_Retry_Limit7DSSS preamble144 bitsDSSS header48 bitsStation buffer size300 framesSlot_Time20 msSIFS_Time10 msDIFS_Time50 ms Typical value Table 2.Default attribute values for theinfrastructure network unless otherwiseBER-5Number of voice stations10Voice transmission rate64 kb/sVoice station buffer size100 framesCFP_Max_Duration0.39 sCFP_Repetition_Interval0.41 sPIFS_Time30 ms Typical value delay between an AP and a mobile station. Here the delay is Figure 12.RTS_Threshold effects on data throughput. 50000.30.35Maximum data throughput 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.750.8 1000 1500 20002500 MSDU length 500MSDU length 1000 Figure 13. 50000.10.2Maximum data throughput 0.3 0.4 0.5 0.6 0.7 0.80.9 1000 1500 20002500 bad 10-3BERbad 10-4BERbad 10-5BERbad 10-6 Figure 14.Average MSDU length effects on data throughput. Data offered load0.2000.1Data throughput 0.2 0.3 0.4 0.5 0.6 0.70.8 0.4 0.6 0.81 MSDU length 300MSDU length 1000 IEEE Communications Magazine ¥ September 1997must be satisfied. It is obvious from the figure that an echocanceler must be used since a large proportion of the voicetraffic exceeds the 25-ms requirement in delay. Thus, it isassumed that an echo canceler is employed, and that voicepackets delayed by more than 0.5 s at the receiver becomeuseless and have to be discarded. Thus, the performance mea-sure of interest for voice traffic is the probability that a voice1 percent should be maintained [13]. Shorter voice payloadsincur larger overheads, translating into longer delays. At theother extreme, longer payloads imply longer packetizationdelays. Thus, these two parameters must be traded off. As isseen from Fig. 15, the best operating points appear to bearound 100Ð400 octets long for voice payload. When the CFPrepetition interval is 5, the recommended voice payloadlengths have been shown to be 100Ð200 octets long [9]. Notethat the average delay calculated when the voice payload is50, 100, 200, 400, and 800 octets is 200, 186, 205, 233, and 284ms, respectively.Figure 16 shows the impact of voice payload length on datathroughput over a range of offered loads. It is shown that datatraffic will suffer more as the voice payload length isdecreased. Given a fixed amount of voice information to betransmitted during the CFP, shortening the voice payloadlength will result in more frames (i.e., overhead) transmitted.Shortening the payload length will therefore lengthen theduration of CFP operation, leaving less available bandwidthfor the transmission of data if the CFP is foreshortened. Thus,from the point of view of data traffic, the voice payloadshould be made relatively long. However, beyond 200 octets,the data throughput improvement is marginal.THEEFFECTOFPOLLINGSCHEMEONPERFORMANCEAs mentioned previously, an AP drops a station from thepolling list if the station does not transmit and receive anydata for kconsecutive polls in the current CFP interval. Tosee the appropriate values of k, the effect of kon datathroughput and voice delay is plotted, as illustrated in Figs. 17and 18. Figure 17 shows throughput plotted against offeredload. For the PCF, five voice station pairs are used with voicepayload fixed at 200 octets.The curves indicate that a higher value of ktends to reducethe aggregate data throughput. When kincreases, there is ahigher probability that a voice station will receive or have traf-fic to transmit, which tends to prolong the duration of theCFP. Prolonging the CFP corresponds to a reduction in theamount of time that data stations have access to the channel.In Fig. 18, the value of khas very little impact on the voicepacket loss rate, mainly due to the fact that voice stationsoperate on an ON/OFF basis. That is, when a voice stationdoes not have any data to send during an OFF period, it islikely that it will not have any data to send in the near future.Thus, when a communicating pair of voice buffers are empty,the best policy is to drop the stations from the polling listimmediately (k= 1). If the CFP is foreshortened due to lighttraffic at that particular instant in time, the wait until the nextpolling cycle is still well under the acceptable delay specifica-tions levied by the echo canceler. Therefore, from a datathroughput perspective, it is best to select k= 1 and have aforeshortened CFP period. Figure 15.Complementary cumulative distribution for voicedelay. (s)0.100.00010.001Probability{X � x} 0.01 0.1 1 0.2 0.3 0.4 0.5 0.607 Voice payload=100 Figure 16.Effect of 1voiceon data throughput. Data offered load0.200.050.1Data throughput 0.15 0.2 0.25 0.3 0.35 0.40.45 0.5 0.6 0.81 Voice payload = 90 Figure 17.k. Data offered load0.200.30.32Data throughput 0.34 0.36 0.38 0.4 0.42 0.44 0.4 0.6 0.81 = 1k = 2k = 3k = 4k = 5 Figure 18.Effect of kon voice delay. (s)0.100.00010.001Probability {X� x} 0.01 0.1 1 0.2 0.3 0.4 0.5 0.607 = 1k = 2k = 3k = 4k = 5 structure network. The PCF is simulated using a fixed voiceasynchronous data being transmitted over the DCF, which et al.,ÒWireless LANs and Mobile Networking: Standardsand Future Directions,Ó IEEE Commun. Mag., vol. 34, no. 8, Aug. 1996,pp. 86Ð94.[2] ETSI TC-RES, ÒRadio Equipment and Systems (RES); High PerformanceRadio Local Area Network (HIPERLAN); Functional Specification,Ó ETSI,06921 Sophia Antipolis Cedex, France, draft prETS 300 652, July 1995.[3] Wireless Medium Access Control and Physical Layer WG, IEEE DraftStandard P802.11, ÒWireless L[4] K. C. Chen, ÒMedium Access Control of Wireless LANs for Mobile Com-IEEE Network, vol. 8, no. 5, Sept. 1994, pp. 50Ð63.[5] H. S. Chhaya and S. Gupta, ÒThroughput and Fairness Properties ofAsynchronous Data Transfer Methods in the IEEE 802.11 MAC Proto-PIMRC Õ95, 1995, pp. 613Ð17.[7] J. Weinmiller, H. Woesner, and A. Wolisz, ÒAnalyzing and Improving theIEEE 802.11-MAC Protocol for Wireless LANs,Ó Proc. MASCOTS Õ96, SanJose, CA, Feb. 1996, pp. 200Ð6.[8] D. Bantz and F. Bauchot, ÒWireless LAN Design Alternatives,Ó IEEE Net-, vol. 8, no. 2, Apr. 1994, pp. 43Ð53.[9] B. Crow et al., ÒInvestigation of the IEEE 802.11 Medium Access Control(MAC) Sublayer Functions,Ó Proc. INFOCOM 97, Kobe, Japan, Apr. 1997,[10] E. Gilbert, ÒCapacity of a Burst Noise Channel,Ó Bell Sys. Tech. J., vol.39, Sept. 1960, pp. 1253Ð66.[11] P. Brady, ÒA Model for Generating On-Off Speech Patterns in Two-WayBell Sys. Tech. J., vol. 48, no. 7, Sept. 1969, pp. 2445Ð72.[12] M. de Prycker, Asynchronous Transfer Mode: Solution for Broadband, 3rd ed., Englewood Cliffs, NJ: Prentice Hall, 1995.[13] L. Hanzo et al., ÒA Packet Reservation Multiple Access Assisted Cord-less Telecommunications Scheme,Ó IEEE Trans. Vehic. Tech., vol. 43, no.2, May 1994, pp. 234Ð44.BIOGRAPHIESBRIANP. CROW(bcrow@mitre.org) received a B.S. from Arizona State Univer-sity in 1987 and an M.S. from the University of Arizona in 1996. From1988 to 1992, he was a signal officer in the U.S. Army. He is currently alead engineer with the MITRE Corporation. His research interests includewireless and broadband networks, and network and systems management.IWIDJAJAreceived a B.A.Sc. degree from the University of BritishColumbia, an M.S. from Columbia University, and a Ph.D. from the Univer-sity of Toronto, all in electrical engineering. From 1994 to 1997, he wasassistant professor of ECE at the University of Arizona. Since July 1997, hehas been with Fujitsu Network Communications. His research interestsinclude mobile and wireless networks, switching architectures, and trafficJGEUNKIMreceived B.S. and M.S. degrees, both in electrical engineer-ing, from Yonsei University, Seoul, Korea, in 1990 and 1992, respectively.He is pursing a Ph.D. at the University of Arizona, studying issues of wire-less ATM and performance evaluation of broadband networks.PT. SAKAIreceived an M.S.E.E. from the University of Arizona in Tuc-son. He is currently a new product planning and applications engineer atCypress Semiconductor, where he is involved with defining next-generationdata communication products. Figure 19.Effect of voice stations on voice delay. (s)0.100.00010.001Probability {X� x} 0.01 0.1 1 0.2 0.3 0.4 0.5 0.60.7 Voice stations = 6