transmission techniques for wireless lans

8/3/2019 Transmission Techniques for Wireless LANs

1/5

A Comparison of Different Wireless LAN Transmission Techniques Using Ray TracingArmFah&*, Kaveh Pahlavan, Ganing Yang**

*Center for Wireless Information Network Studies, WPI, Worcester, MA* * Rockwell Telecommunications, Newport Beach, CAAbstractThis paper presents a comparison among the performances of varioustransmissiontechnologies in a typical indoorradio environment. The performanceof fresuencyhopping and direct sequence spreadspectrum are compared with the performanceof multicarrier, decision feedbackequalizer, and sectored antenna systems. Givena maximum tolerable outage probability, the m a x i achievabledata rates areobtained for the different modulation techniques. Outage probabilities areestimated through simulation, using channel impulse responses obtained from rayuacing. The bandwidth efficiencyof spread spectrummodulation is shownbe verylow. For a fixed bandwidth, theminimumpower requirementof each modulationtechnique is also presented, showing significantly lower power xqukments forspread spectrum. Overall, multicarrier modulation is found to present the besttradeoff between bandwidth efficiency and power consumption.1. IntroductionIn the US, the Industrial, Scientific, Medical (ISM) bands werereleased in May 1985 as unregulated secondary bands forwireless data communications using spread spectrummodulation. The IEEE 802.11 committee was formed in 1990.Its goal is to develop interoperability standards for wirelessnetworks that use either spread spectrum modulation in the 2.4GHz ISM band or diffused infrared (DFIR) transmission[IEEE94]. In 1993, the EC announced allocations in the 5.2and 17.1 GHz bands for the development of HIPERLANnetworks. RES 10 is chartered to define a radio transmissiontechnique that includes type of modulation, coding, channelaccess, and MAC-layer protocols (Bou941. The physical-layertransmission techniques considered for the two standards arequite different. This situation has created the need for aquantitative study of the performance of different transmissiontechniques in a typical indoor environment.Given an estimate of the channel impulse response, it ispossible to predict the performance of different modulationschemes in that channel. Three methods have been used forgenerating channel impulse responses; statistical channelmodeling, direct measurements, and ray tracing. Earlierperformance evaluation studies such as [Kav87] have usedstatistical channel models and estimated the performance ofdifferent systems using either derivations or Monte-Carlosimulations. Realizing the need for more accurate site-specificpropagation estimates, other researchers have directly used theresults of wideband measurements, as their channel impulseresponses [Cha93]. However, wideband measurements at everysite of interestare not cost-effective. Ray tracing has beenproposed as a more economical method of predicting thepropagation of radio waves in indoor, metropolitan, and ruralareas -- it requires only the floorplan of the site and a high-performance general purpose computer [Yan94b]. Ray tracingalso provides direction-of-arrival information, which isnecessary for analyzing sectored-antenna systems.Ray tracing has been used to obtain impulse response profiles atdifferent locations in a building where extensive wideband

propagation measurements have been carried out. The softwarewas developed at the Center for Wireless Information NetworkStudies (CWINS) at Worcester Polytechnic Institute, and isdescribed in [Yan94b]. Use of ray tracing to generate theimpulse responses is validated by obtaining channel impulseresponses for a set of locations using both measurements andray tracing, and comparing the resulting two sets of BERestimates. The channel profiles obtained through ray tracingare then used to compare the performance of differenttransmission techniques. Given a fixed bandwidth, the noisefloor error rate of direct sequence spread spectrum (DSSS)with one-, two-, and four-tap RAKE receivers and frequencyhopping spread spectrum (FHSS) with and without channelcoding areplotted as a function of data rate. Similar results formulticarrier modulation (MCM) with and without channelcoding areplotted as a function of the number of carriers, sincein the case of MCM the data rate is not a function of thenumber of carriers.The above systems are compared with systems employingdecision feedback equalization ( D E ) and sectored antenna(SA) selective diversity, for which performance results werepreviously reported [Yan94b]. The maximum achievable datarate for each system is obtained, given a reasonable outagerequirement and no bandwidth restrictions. For a fixedbandwidth, minimum power requirements are also presented.

t2. Method for AnalysisIn order for code division multiple access to be successful, irequires global power control [Kav87]. As a result, mosexisting systems, as well as the 802.11 standard, specify somform of contention-based time division access with collisioavoidance [Tuc91, IEEE941. Therefore, this paper does noaddress CDMA, assuming that the channel is used by at mosone transmitter at a time.2.1 ChannelModelThe most popular model for describing multipath channels idIthe so-called discrete multipath time-domain model. It w adeveloped by Turin to describe propagation in the urban radidchannel, and was later applied to the indoor radio channeltKav87, Yan94bl. It describes the channel as a linear time]invariant filter whose impulse response is the sum of L path1with the Zth path having amplitude P I ,delay q, nd phase 91; 1

h(t)= xPl 6 ( t - z l ) e@, IIThis model also lends itself well to ray-tracing, which generate1the gain, delay, and phase of the surviving rays. These ca(simply be plugged into the above equation to yield the channelimpulse response. In this paper we draw PI, and zI from rajtracing. During the simulations, path phases (pI are treated 4

el11

L- 1I= O

0-7803-3002-1/954.00 1995 IEEE 1062


2/5

random variables uniformly distributed in [0,2n], simulatingsmall-scale movements of the transmitter and/or receiver, ormotion of objects close to either unit.2.2 System DescriptionsWe examine five transmission techniques in this paper: DSSS,FHSS, MCM, DFE, and SA . QPSK modulation and coherentdetection are assumed. At the transmitter, data is modulatedonto the I and Q channels of a carrier. The signal is thenpassed through the spread spectrum or multicarrier modulator,and transmitted over the channel, where it is distorted by thechannel multipath characteristics and AWGN. At the receiver,the signal is demodulated, despread (if applicable), andsampled at the time determined by the timing recovery circuit.The resulting decision variable is phase-compensated and usedto determine the bits transmitted over the I and Q channels.Spread SpectrumThe frequency hopping and direct sequence systems weredescribed in detail in [Fa195]. In FHSS, he transmitter andreceiver pulse-shaping filters, g,(t) and gt(t), are identical witha square root of raised cosine frequency response. The receivedsignal is obtained by convolving the transmitter output with thechannel impulse response and adding noise. This is then de-hopped and passed through the receivers matched filter,yielding;

L-1i 1=0

z(tln)= x b i . P I .g(t - iT -z I ) . j(f-*t+4@n) +n1( t )where f, is the nth carrier frequency, (bj} is the complex datasequence, 8, is the phase offset between the FH modulator anddemodulator, nl(t) is the front-end filters response to AWGN,and g(t)=g,(t)*g,(t) is the impulse response of a raised-cosinepulse shaping filter. Finally, z(t) is sampled and phase-compensated, yielding a decision variable conditioned on hopcarrier n. The sampling time for a raised cosine pulse isobtained following the practical method derived in [Val87].Perfect phase recovery was also assumed.The DSSS system is assumed to have a spreading gain of N andchip period T, , such that T=N.T, is the bit period. Thereceiver filter is matched to the PN sequence, such that(without multipath), the cross-correlation of the transmittedsignal and the periodic despreading signal has a peak of widthT, centered at zero delay. The despreaders output is thereforegiven by

L-1i 1=0z ( f ) = xx b i . P I . R , ( t - i T - z l ) . e i @ r +nl ( f )where Rx(z) is the cross-correlation of the PN sequence. TheRAKE receiver used in the direct sequence system samples z(t)

at the highest peaks, which are detected using a slidingwindow. It then phase-compensates the complex sample toyield the decision variable.Multicarrier ModulationThe multicarrier system is shown in figure 1. Since a raisedcosine pulse is assumed, error probability expressions -- againconditioned on carrier number n -- are similar to those for

frequency hopping. However, depending on the frequencyseparation between carriers, interference from adjacent carriersmay also need to be taken into account. If the total systembandwidth and number of carriers are fixed, increasing thedistance between carriers will reduce the amount of interferencefrom adjacent carriers, at the expense of data rate. Defining thetime-bandwidth product as the product of symbol period T andfrequency separationAa, time-bandwidth product of 1.5 wasfound to provide sufficient separation between carriers for araised-cosine rolloff factor of 0.5. This was used throughoutthe simulations cited in this paper. The carriers are assumed tobe equally spaced with the nth carrier given by o, = Q +(Ao)n. Given reasonable pulse shaping filters, every carrierwill only be interfered with by its two adjacent neighbors. Thereceived signal on the nth carrier can therefore be written as

z( t l t t )= C xbin * g , ( t - i T ) *h(t)+tt,(t){ 1whereca

g,(t) J g,(TI. j(Am)mqg , ( t - ) .dz-=a

and bin is the information bit transmitted on carrier n during theith symbol interval. Exact expressions for g+l(t) and g.l(t) for araised cosine pulse can be found in [Yan94a]; go(t) isequivalent to g(t) in the frequency hopping system describedabove.DFE and SA SystemsPerformance results for decision-feedback equalizer andsectored antenna systems at this location were first reported in[Yan94b]. In the case of DFE modems, the overall sampledimpulse response after equalization is used for z(t) [Pah95].When sectored antennas are used, the channel impulse responsefor each antenna sector is modified to take into account thedirection of arrival of each path. This analysis assumed thatthe receiver is equipped with a six-sector directional antennawhose polarizations are vertical. The ith antenna pattern isdefined by the function

2.78 (v- i 3)xiSyf


3/5

same equations used for spread spectrum in [Fa1951 arerepeated for MCM, DFE, and SA systems. The receiver'sfront-end noise was computed following the derivation in[Lee891 at an operating temperature of 290"K, thermal noisewill be -174 dBm/Hz. In the case of direct sequence spreadspectrum, the bandwidth of the front-end filter is 25 M H z ,resulting in noise variance of about -100 a m . In frequencyhopping and multicarrier modulation, the front-end filter'sbandwidth is dependent on the data rate, resulting in a highernoise variance at higher data rates (lower number of carriers).In this paper we also study the effect of Reed-Solomon coding,when used in conjunction with frequency hopping ormulticarrier modulation. If the encoder is an (n, k) coderdefined over G(2*), it is capable of correcting up to (n-k)/2errors [Pah95]. The number of bits per channel symbol is setequal to the number of carriers, and each bit of the resultingcodeword is transmitted via a different carrier. The I and Qchannels also carry bits from different codewords. Theprobability of codeword error is equal to the probability ofhaving more than t bits in error. In this study, since the raytracing program predicted the error probability of each channelseparately, it was possible to use a more advanced technique toarrive at more accurate results. As described previously[Fa195], we follow the procedure presented in Appendix A of[Pur89]. We also assume that if a codeword is received inerror, half of the decoded information bits will be in error.2.4 Validation of Ray Tracing MethodFrom the perspective of system design -- and deployment -- therequirement is that impulse responses generated via ray tracingand those obtained from wideband measurements result insimilar performance estimates for the transmission techniquesunder study. In [Fa195], channel impulse responses obtainedfrom measurements and ray tracing were used to obtain errorand outage probabilities for the same set of transmitterheceiverlocations. It was shown that similar performance results areobtained from either set of impulse responses.3. Results & DiscussionIt is often desirable to view a data channel as odoff, whereoutage is defined as the bit error rate exceeding a certainthreshold [Yan94b]. Outage probability can then be defined asthe percent of time that a given location is in outage or thepercent of locations that are in outage at a given time. Theformer is often used to describe time-varying channels such asurban mobile, while the latter is used to describe fixedinstallations such as indoor wireless LANs. In this study, anoutage threshold of was chosen. In this building, 100 mWof transmitted power (100 mW per carrier in the case ofmulticarrier systems) was found to eliminate the effects ofadditive noise for all three modulation techniques [Fa195]. Itwas therefore chosen as the noise floor power level.3.I Spread SpectrumWhen combined with a RAKE receiver [Pah95], directsequence spread spectrum provides sufficient resolution toallow individual paths to be extracted from the received signal,

providing a form of time diversity. Frequency hopping spreadspectrum, on the other hand, takes advantage of the fact thatdifferent portions of the spectrum tend to fade independently.Significant improvements can be achieved with a smallreduction in data rate through the use of channel coding andinterleaving, as described in the previous section.Assuming QPSK modulation in a fixed bandwidth of 25 MHz,processing gains of 7, 15, 31, and 63 result in uncoded datarates of 7.1, 3.3, 1.6, and 0.79 MBPS, respectively. Whenchannel coding is used, Reed-Solomon codes of ( 7 3 , (15,11),(31,23), and (63,45) are used, so that in all cases the reductionin data rate is approximately a factor of 5/7. The resulting datarates are 5.1,2.4, 1.2, and 0.57 MBPS for the FHSS system.Figure 3 is a plot of outage probability vs. data rate forfrequency hopping spread spectrum with and without Reed-Solomon coding and direct sequence spread spectrum with 1-,2-, and 4-tap RAKE receivers employing maximal ratiocombining. It can be seen that with a four-tap RAKE receiver,DSSS performs better than FHSS without coding. However,when coding is added to frequency hopping spread spectrum,its performance improves significantly; the error rate falls topractically zero with only 15 carriers.3.2 Multicarrier ModulationUsing multicarrier modulation with a time-bandwidth productof 1.5, a system bandwidth of 25 MHz, and QPSK modulationresulted in a data rate of 33.3 MBPS. With coding, the datarate was reduced to 24 MBPS. Figure 4 shows the outage rateof multicarrier systems with and without channel coding, as afunction of number of carriers (since the data rate wasessentially constant).Comparing these results with those of the previous section, it isobvious that in a bandlimited system that does not implementcode division multiple access, for a given bit error rate,multicarrier modulation allows a higher data rate than eitherform of spread spectrum modulation. Furthermore, since it isobvious from figure 4 that the outage probability can be reducedby increasing the number of carriers, multicarrier modulationoffers the potential of lowering the outage probability without acorresponding reduction in the data rate.3.3 A Comparison Among Transmission TechniquesTable II of [Yan94b] presents the maximum data rates that canbe achieved with DFE, SA , and DFE/SA systems, given anoutage probability of .01 and assuming unlimited availablebandwidth. Since these results were for BPSK modulation, thedata rates were doubled to reflect the increased data rateachievable with QPSK modulation and perfect phase recovery.These numbers assumed no limit on the available bandwidth.Therefore, comparable results were obtained for the spreadspectrum systems by repeating the simulations at differentsystem bandwidths with the processing gain fixed. The resultshave been summarized in figure 5 . Results for multicarriermodulation have not been presented, because (given nobandwidth limitations) the data rate of multicarrier systems is

1064 I


4/5

infinite; the restriction in this case is the complexity of thesystem. To have outage rates of .01 or less, spread spectrumappears to only allow data rates of betwetn 1 and 10 MBPS,even with no limits on the available bandwidth.Based on the above discussion, it may appear that spreadspectrum modulation is not a suitable transmission techniquefor wireless data networks. However, bandwidth efficiency isonly one side of the equation; the processing gain that spreadspectrum provides can serve to increase the fade margin,reducing the transmitters power requirement. This could bean important consideration if the wireless network is being usedto connect portable, battery-operated computers. Using thesame outage criterion as above, the minimum powerrequirement of spread spectrum modulation was compared tothat of MCM, DFE, and sectored antenna systems. Figure 6shows the minimum power required to achieve an outage rateof .01 for DSSS with 4 taps, as well as FHSS and MCM withand without Reed-Solomon coding. Also presented are thepower requirements to achieve the same outage probabilityusing DFE, SA , and DFE/SA systems. A total systembandwidth of 10 MHz is assumed. According to this figure,spread spectrum can add between 30 and 35 dB to the fademargin. Furthermore, equalization itself is a power-intensiveoperation, a significant factor for portable, battery-operatedcomputers. On the other hand, it can be expected that wirelessnetworks that serve mostly portable users will operate at lowerpower levels with much smaller cells -- especially if significantprocessing gain is available. This will result in betterfrequency reuse, partially canceling the bandwidth inefficiencyof spread spectrum modulation.The power savings possible with spread spectrum modulationcan only be achieved at the expense of data rate. On the otherhand, multicarrier modulation can provide an improvement ofabout 20-25 dB,with only a small reduction in the data rate.As a result, in systems where both bandwidth efficiency andpower consumption are important, multicarrier modulation canprovide the best tradeoff.4. Summary and ConclusionsDifferent transmission techniques have been presented, andcompared using channel impulse responses generated by a raytracing algorithm. Outage probability was plotted as a functionof data rate for spread spectrum systems with different levels ofsystem complexity and as a function of number of carriers formulticarrier modulation. These were compared with DFE, SA,and DFEISA systems.It was shown that spread spectrum is not an efficient way ofcombating multipath fading in an indoor environment; the lossin bandwidth efficiency is too high in a large LANenvironment where throughput is a critical factor. On the otherhand, in networks where users have battery-operated portable:omputers, due to smaller cell size and better frequency reuse,bandwidth efficiency requirements may not be as strict, makingspread spectrum modulation more attractive due to itsxocessing gain (which further reduces the power

requirements). If both bandwidth efficiency and powerconsumption are important, multicarrier modulation appears topresent the best tradeoff.References[Bou94] B. Bourin, HIPERLAN - Markets and ApplicationsStandardisation Issues, IEEE Intl. Conf. on Personal, Indoor,and Mobile Radio Communications, Hague, Netherlands, Pp.863-868, Sept. 1994.[Cha93] M. Chase, K. Pahlavan, Performance of DS-CDMAOver Measured Indoor Radio Channels Using RandomOrthogonal Codes, IEEE Trans. on Vehicular Technology,Vol. 42, Pp. 617-624, NOV.1993.[Fa1951 A. Falsafi, K. Pahlavan, A Comparison Between thePerformance of FHSS and DSSS for Wireless LANs Using a3D Ray Tracing Program, IEEE Vehicular TechnologyConference, Chicago, IL, July 1995.[IEEE94] IEEE Std. 802.1 1-93/20b3, Wireless LAN MediumAccess Control (MAC) and Physical Layer (PHY)Specifications, draft standard, September 1994.[Kav87] M. Kavehrad, B . Ramamurthi, Direct SequenceSpread Spectrum with DPSK Modulation and Diversity forIndoor Wireless Communications, IEEE Transactions onCommunications, Vol. 35, Pp. 224-236, February 1987.[Lee891 W. C. Y. Lee, Mobile Cellular Telecomm. Svstems,New York, McGraw-Hill, 1989.[Pah95] K. Pahlavan, A. Levesque, Wireless InformationNetworks, John Wiley, 1995.[Pur891 M. Pursley, S . Sandberg, Variable-rate Coding forMeteor-burst Communications, IEEE Transactions onCommunications, Vol. 37, Pp. 1105-1112, Nov. 1989[Tuc91] B . Tuch, An ISM Band Spread Spectrum Locan AreaNetwork: WaveLAN, IEEE Workshop on Wireless Local AreaNetworks, Worcester, MA, Pp. 103-111, May 1991.[Val871 R. Valenzuela, Performance of Quadrature AmplitudeModulation for Indoor Radio Communications, IEEE Trans.on Comm., Vol. 35, Pp. 1236-1238, Nov. 1987.[Yan94a] G. Yang, K. Pahlavan, Performance Analysis ofMulticarrier Modems in Office Environment Using 3D RayTracing, IEEE GLOBECOM 94 San Francisco, CA, Pp. 42-46, November 1994.[Yan94b] G . Yang, K. Pahlavan, T. Holt, Sector Antenna andDFE Modems for High Speed Indoor Radio Communications,IEEE Trans. on Vehicular Technology, Vol. 43, Pp. 925-933,Nov. 1994.

1065


5/5

Fii 1 MnlticanierModulafjonSystemTiming-bp+,e .O'=.

DATA

Fig. -Antenna patternfor ofie sectorof &-sector a n m a .

Fig. 5 - Max. data rate, unlimited bandwidth. . . . . . .

= : :Fig. 6 - Min. power, lOMHz Total Bandwidth1066

transmission techniques for wireless lans

Documents