Propagation Measurements and Models for Wireless Communications Channels

To achieve ubiquitous PCS, new and novel ways of classifying wireless environments will be needed that are bothwidely encompassing and reasonably compact.
Jargen Bach Andersen, Theodore 5. Rappaport, and Susumu Yoshida

JGRGEN BACHANDERSEN is u professor at Aalborh. Universip and head of thc Centerfor Personkommnnikation.

THEODORE S. RAPPAPORT 0 un associareprafessor of electrical rngineenng at Viwnia Tech. SUSUMU YOSHIDA

u electrical engineering at Kyoto Universip.

ireless personal communicae1shasbeenintroduced.Thesemodelsareofamore deterministicnaturc,characterizingaspecificstreet, tionscouldinprincipleusesevera1 physical media, ranging building, etc. They are necessarily more time consuming to use, but are more revealing concerning also from sound to radio to light. Since we want to overcome the physical details and hopefully morc accurate. Firstsome keyparametersandthemeasurement limitations of acoustical comofthemwill be discussed and then the differentwiremunications, we shall concentrate on propagation of electromagnetic wavcs in the frequency range less environments will be treated. The latter topic is divided here into outdoor environments, indoor from some hundreds of MHz to a few G H z . Although thereisconsiderable interest atthe momentenvironments, and radio penetration from outin millimeter wave communications in indoor door to indoor environments. environments, they be mentioned brieflyin will only this survey of propagation of signals. The Physics of Propagation It is interesting to observe that propagation results influence personal communications systems in hemechanismswhich governradio propagation several ways. First there obviously the distribution is are complex and diverse, and can generalthey basic propagation mechofmeanpoweroveracertainareaorvolumeofinter- ly be attributed to three anisms: reflection, diliraction, and scattering. est, which is the basic requirement for reliable Reflection occurs when a propagating electrocommunications. The energy should be sufficient magnetic wave impinges upon an obstruction for the link in question, but not too strong, in order not to create cochannel interfcrcnce at a with dimensions very large compared to the wavelength of thc radio wave. Reflections from distance in another cell. Also, since the radio link is highly variable over short distances, not only the surface of the earth and from buildings produce rcflccted waves that may interfere constructhe mean poweris significant; the statistical distively or destructively ata receiver. tribution is also important. This is especially true Diffraction occurs when the radio path between when the fading distribution is dependent on thc if thetransmitterandreceivcrisobstructedbyanimpenbandwidth of the signal. Secondly. even there is etrable body. BasedonHuygensprinciple,secondary sufficient power availablefor communications, the quality the signal may be such that large errorswaves are formed behind the obstructing body of occur anyway. This results from rapid movement even though there is no line-of-sight (LOS) through thescatteringenvironment,or impairments between the transmitter and receiver. Diffraction due to long echoes leading to inter-symbol-inter- explains how radio frcquency(RF) energy can ference. A basic understanding of the channel is travel in urban and rural environments without a is important for finding modulation andcodingschemes LOS path. This phenomenon also called shadthat improve thc channel, for designing equaliz- owing,becausethediffractedfieldcanreachareccivers or, if this is not possible, for deploying basc er even when itis shadowed by an obstruction. Scatteringoccurswhenthcradiochannelcontains station antcnnas in such a way that the detrimental effects are lesslikely to occur. objects with dimensions that are on the order of In this article we will describe the type of sigthewavelengthor1cssofthepropagatingwave.Scatnals that occur invarious cnvironments and the modtering, which follows the same physical principles eling of the propagation parameters. Models are as diffraction, causes energy from a transmitterto be essentially of two classes. The first class consists reradiatedin many different directions.Ithasproven on of parametric statistical models that average to be the most difficult of the three propagationmechdescribcthephenomenonwithinagivenerror.They anisms to predictin emerging wireless personal communication systems. For example, in urban are simple to use, but relativcly coarse. the last In few years a second class ofenvironment-specific modmicroccllular systems, lamp posts and street signs

42

IEEE Communications Magazine

January 1995
-

~~

scatter energy in many directions, thereby prowding RF coverage to locations which might not rcccive energyviareflectionordiffnction. The three mechanisms are illustrated in Fig. 1. As a mobile radio moves throughout a coverage area, the three propagation mechanisms have an impact on the instantaneous received signal in differentways. Forexample, ifthe mobile hasaclear LOS path to the base station, diffraction and scattering are not likely to dominate the propagation. Likewise, if the mobile is at street level in a large metropolitan areawithout LOS to the base statlon. a diffraction and scattering are most likely to dominate the propagati0n.k the mobilemovesoversmall distances. thc instantaneous received narrowband signal strength will fluctuate rapidly giving rise to small-scale fading.The reasonfor this is that the field isasumofmanycontributionscomingfromdifferent directionsand since the phases are random, the sum bchaves as a noise signal,Le.. Rayleigh fading. In A small-scale fading, the received signal power may W Figure 1. Sketch of three lmpottantpropagation mechanisms: rejlection (R). vary by as much as threeor four orders o f magniscattering (S), diffraction (Dl. tude (30 or 30 dB) when the receiver IS moved by only a fraction of a wavelength. As the mobile moves away from the transmitter over largerdistances, the local average received signal gradually will decrease. Typically, the local average signal is computed over receiver movementsof 5 to 40 wavelengths 11-31, Figure 2 demonstrates the effects of small-scale fading and large scale signal variation for an indoor radiocommunication system. Notice in the figure that the signal fades rapidly as thereceivermoves,butthelocalave.eragesignalchanges much mure slowly with distance. In mobile radio systems, communications engineers are generally concerned with two main radio channelissues: link budget and time dispersion. The link budget is determined by the amount of received power that may expected at a particular distance orlocation from a transmitter, and it determines fundamental quantities such as transmitterpowerrequirements, coverage areas, andbatI tery life. Time dispersion arises due to multipath W Figure 2. Tvpical received simal levels for anindoor radio communication propagationwherebyreplicasof the transmittcd sigsy'tem. ,\'otic> that small .scak fading produces changes of 20 db or more, level nal reach thereceiver with different propagation where the local average signal level changes much moreslowly with distance. delays due to the propagation mechanisms described above. The time-dispersive nature of the channel determines the maximum data rate that may be transmitted without requiring equalthananinstantaneousvalue.Inrealisticmob~leradio ization and also determines the accuracyof navichannels, free space does not apply. A general PL gational services such as vehicle location. model that has been demonstrated through measurements [ l , 4) uses a parameter, n , to denote Propagation parameters the power law relationship between distance and Path loss - Link budget calculations require an received power. As a function of distance, d , PL estimate of the power level so that a signal-to-noise (in decibels) is expressed as ratio (SNR) or, similarly, a carrier-to-intcrference (QI) ratio may be computed. Because mobile P L ( d ) = PL(d0) + 10 n log(d/do)+Xo (1) radio systems tend to be interference limited (due tootheruserssharing thesamechanne1)rather than where n = 2 for free space, and is generally highnoise limited, the thermal and man-made effects er for wireless channels. noise are often insignificant compared to thesignal levels ThetermPL(do)simplygivesPLataknownclose of cochannel users. Thus, understanding the propain reference distance do which is in the farfield of gation mechanisms in wireless systems becomes the transmitting antenna (typically 1 km for large important for not only predicting awerage to a particurban mobile systems, 100 m for microcell sysular mobile user, also for predicting the interfering but tems, and 1 m for indoor systems) and X , denotes signalsthatuserwillexperiencefromotherRFsources. a zero mean Gaussian random variable (with Weusepathloss(PL) heretodenote thelocalaverunits of dB) that reflects the variation average in age received signal power relative to the transmit receivedpower that naturallyoccurswhenaPL model power. This auseful quantity, since receivedpower is of this type is used. thePL model Since only accounts is usuallymeasured as a local spatial average rather for the distancewhich separates the transmitter Magazme Communications IEEE

January 1995

43

All measurement locations

n=A

..

.
n=3

130 120
1

0 110 I

n=2

loo

90
88
70 0.1

n=l

agation delay. This corresponds to an excess travel distance of over 30 km! When the receiverwas moved by just a few meters, the observed multipath characteristics became much more tame, with an observed excess delay of no more than 1 0 ~ sIn typ. ical urban cellular systems, the worst case excess delays (at 99 percent probability levels) echoes for which are within 10 dB of the maximum signal are less than 25 ~s [6]. A collection of power delay profiles, such as is showninFig.4,canbedoneinmanyways.Typically, a channel sounderis used to transmit a wideband signal which is received at many locations within a desired coverage area. This channel sounder may operate in the time or frequency domain. The time domain response appear as in Fig. 4, and a swept will frequency response will reveal that the channelis frequency-selective, causing some portionsof the spectrum to have much greater signal levels than oth-

measurement techniques are equivalent and can be related by the Fourier transform. 2 3 4 Power delay profiles may be averagedover time 1 10 oroverspaceasthereceivermovesabout.Theaveraging interval is key to determining the applicaT- R separation tkm) bility of the time dispersion statistics, derived H Figure 3. Scatterplot ofpath loss vs. distance for measurements in five Gerfrom power delay profiles [6-8]. man cities. The path loss is referenced to a free space reference measurement at Some important time statisticsmay be derived do = I O 0 m. Note that CY has a value of 11.8 dB dure to the large spreud ofPL from power delay profiles, and can be used to values relativeto the straight-linefit. quantify time dispersionin mobile channels. They are: maximum excess delay atX dB down from maximum (MED), mean excess delay(T), and and receiver, and notany of the physical features a. rms delay spread, The MED is simply computed by inspecting a power delay profile and noting it of the propagation environment,is natural for sevthe value of excess time delay at which the profile eral measurements to have the same T-R separation, but to have widely varying PL values. Thisis monotonically dips below the X d B level. For due to the fact that shadowing occur at some may example, in Fig. 4, the MED at 10 dB down would of be l o p s . It should be noted that these paramelocations and not others, etc. The precision a PL model is thus measured by the standard deviters are also interest when they are derived of ation 6 of the random variable&, with a smaller from instantaneous impulse responses. The mean excess delay (7)is the first central PL value of CY reflecting a more accurate prediction model. moment of the power delay profile and indicates the Figure 3 demonstrates measured values PL of average excess delay offered by the channel. Therms measure of the spread of power about the value asafunctionofdistanceinfivecitieswithinGermany and shows the above model superimposed the on of 7, aT, the most commonly used parameter to is measured values. Thistype of plot is called a scat- describe multipath channels. Chuang[9] and ter plot, and it is useful for quickly assessing the more recent simulation studies have confirmed tendencies and variationsof path loss throughout that a good rule-of-thumb is as follows: if a digital a radio system. To achieve lower values 6 , signal has a symbol duration which more than of is more site-specific information about the environten times the rms delay spread, then an equala, ment is required or smaller coverage distances izer is not required for bit error rates better than are used when applying the PL model above [5]. For such low values of spread, the shapeof the delay profile is not important. On the other hand, Mulfipafhdelayspread-Timedispersionvaries if values of approach or exceed 1/10 the durawidely in a mobile radio channel dueto the fact tionofa symbol, irreducibleerrorsdue the frequency that reflections and scattering occur at seemingly selectivity of the channel will occur, and the randomlocations,andtheresultingmultipathchan- shape of the delay profilewill be a factor in deterne1 response appears random, aswell. Because mining performance. time dispersion is dependent upon the geometric relationships between transmitter, receiver, and the engineers often are concerned with statistical models of time dispersion parameters such as the average rms or worst-case values. To demonstrate how rapidly the multipath characteristics may change, consider Fig. 4,which illustrates oneof the worst cases of multipath time dispersion ever reported in aU.S. cellular radio system. Figure 4 indicates that multipath energy 18dSdown from the 100-psexcess propfirst arriving signal incurred a

erswithinthemeasurementband.Ofcourse,thetwo

surroundingphysicalenvironment,communications

Outdoor Propagation

F requencyreuseisobtainedusingacellularstructureofcoveragezonesfrombasestations.1ntextbooks, one usually sees a honeycomb structure of

hdvepocketsinsideeachother,perhapsduetoheight
IEEE Communications Magazine January 1995

44

~~

variations from natural terrain o r from man-made structures. Often interference condition\ prevent the optimal use of the cellular system because reality does not correspond to the simple mathematical models. Increasing the power to cover dead spotswouldcreateotherproblems.Thenetworkplanner hassome possibilities though, ouyh placingof thl the base station antenna and to sonic extent choosingitsradiationpattcrn,toshapethec~ilextent.Tilting the antenna avertical plane isa in normal mcasure and by choosing the height above o r helow rooftops, maybe even down tu street level, v m e thing can be done to control the ot the cell. hire Cells are typically classified roughly according I tosize as macrocells andmicrocells. Table presents some distinctive cell features.

-85 -90 -95

csf47
I I I , , /

Displa threshold = -111.5 dBm per 40 ns RMS &lay spread = 22.85 ps Worst caw San Francisco measurement Mansell St.

-1 05 -1 l a
-115

Macrocells

The average path loss IS what remains after averExcess delay time, Cs) aging the pathloss over the fast fading due to 2 multipath.The macrocellswere the basis for thc first Figure 4. One u the wor.sr cases of multipath time disperslon observedin a f generationsystcmsmeant for mobile users, and they cellular rudlo s).srern. Meawement mademSan Francisco, California. generally have base stationsat high points like broadcasting systems, with a coverage of several kilometers. The key problem is t o make some order of the seemingly noise-like path lossexistingover an undulatingterrainwithavarietyoflandcover,mked land and sea, etc. The Rayleigh fading correfast spondstoatruenoisedistribution,buttheslowshadowingtypevariationisremarkablywelldescribedby a log-normal distribution, i,e., the average path loss in dB has a normal distribution.A good reason for this could be that the average path is loss the result of forward scattering overa great many obstacles, each contributinga random multiplicaa tive factor. When converted to dB this gives .Table 1. Typica/parameters for macrocells and sum of random numbers,which leads toa normal microcells. distribution in the centrallimit. In the carly days, planning was mainly based on empirical formulas with an experimental back- metrical Theory of Diffraction (GTD) nor Uniform Theory of Diffraction (UTD) are really successful ground. Okumurd[ 101did thefirst comprehensive in macrocellssince we are dealingwith forward mulmeasurements in 1968 for Japanese environments, and the derived curves have later been trans- tiplescatteringin the transitionzoneswhere nogood theories exist. The resulting prediction power formed into parametric formulas Hata [4]. by It normally leads toa standard deviation of errors was noted that a good model for the path loss (or the field strength)was a simple power law where between 6 and 10 dB, which is not really impressive. Even the availability of modern computers and the theexponentncouldvaryasafunctionoffrequency, use of topographical data bases havc not resulted antenna heights etc. Expressed simply.on a log-login a real break-through in better predictions. scale, this is a linear dependency(Eq. The rela1). The time domain alsobe modeledusinggeomay tionship is strictly empirical, but it has proven exgraphical data bases, andw m e success has been tremely robust overthe years not only in Japanese surroundings, but other surroundings well. The in as achieved in mountainous regions and large cities reason for this general validity, also for indoor [12].Ingeneral,theraytracingroutineslackthecontrhutions from random scatterers. environments, is not completely clear; Walfisch and physical situations where the model exact. One is is ofcourse freespace corresponding ton =2, another is asymptotic propagation overa flat, finitely conducting surface(n=4), and finally, there is forof ward scattering over a numberabsorbing screens of equal height and distance (n = 4). The real world in outdoor macrocells leads tons between and 2 4,withvaluescloserto4inurbansituations. Weshall meet the same modelin the indoor case as well. The models have been refined with a number of correction factors dependingon the land of forestsand cover, different degrees built-up areas, rural areas, andin some cases supplemented with one or two knife-edge diffractions. Neither Geo-

Bertoni[ll]getgoodagreementwithdiffractionover a multiplicity of absorbing screens. There are simple

Microcells
Microce//u/arSystems - Microcells are attracting

much attention simply because they can accommodate more subscribers per unit service area than macrocells. Also, they permit access by low-power portables. Propagation in microcells differs significantly from that in macrocells As suggested in antenna Table 1, smaller cell coverage with lower height and lower transmissionpower results milder in propagationcharacteristicswhencomparedtomacrocellular systems. The smaller multipath delay spread and shallow fading imply the feasibility of broadband signal transmission without excessive countermeasure techniques against multipath fading. Microcells are most frequently plannedin an

IEEE CommunicationsMagazine * January 1995

45

0
I

o"0awn
A

log (distance)

(C)
W Figure 5 . Ideal urban street layour: a ) typicalpath loss cuwes along an LOS
street b) and

along a NLOS vtreet after turning a corner.

urban area where the heaviest tclctraffic is expected. If a transmitter antenna height is lower than the surrounding building helght. then most ot the signal power propagates along the street. The coverageareaextendsalongthestreet. thusthename "street microcell" is often used. I n suburban to areas, microcells might be used realize alast quarter mile connection to a wired network. I n this case, cell coverage might be circular hut strongly affected by the buildings and obstacles.

Modeling and Prediction- Various propagation models for the street microcells based ona r w o p t i c theory have been proposed. Prediction of microcell coverage basedon the model Israther ray accurate comparedto the case of macrocells. Afourray model consisting o f direct ray, ground-reflected ray. and the two rays reflected by the building walls along the street is often as+umed..4 six-ray model that takes doubly reflected rays by the building wall$ or larger-number-ol-ray models are reported t o give more accurate prediction at the cost of Increased computation time [14]. Corner diffraction is another problem attracting much attention. Various modelshave been inbestigated StZlrting from a simple knife-edge or wedge diffractlon t() GTD or UTD with a multiplicity ut rays. An analytical path loss model describing mlcrocellular propagation including reflected rays and corner diffractionhas been prehented in [ 151. Due to the irregularityofthe bulldingstructures, excessive overlap or non-overlapping of the microcell+ are sometimes found. Thus, measurementbased methods are also adopted to enhance the accuracy of the prediction [16]. Coverage area is strongly affected by the location of the transmitter antenna. When i t is installed in the intenection of the streets in Manhattan. for instance, the waves propagate along the streets four direcin tions with subsequent corner diffractionsat each encounteredcorner.Thisgeneratesacoveragearea similar to the shape of a diamond. When a transmitter is located on a street between intersections, this makes the cell more elongated the in direction along that street [17j.

Indoor Propagation

1ndoor radio communication systems are becom.. . . mg mcreasmgly Important for extending and voice
IesssystemsinterconnectwiththePSTNorIocalPBX to provide a seamless extension of the convention-

StreetMicrocells-Anumberofstudieshavebeen done regarding street microccll propagation where mostofthesignalpowerpropagatesalongthestr~et. Let us assume an ideal rectangular street grid with building blocks of relatively uniform height as portrayed in Fig. 5a. In this case. typical path loss curves along the line-of-sight (LOS) path are shown to be characterized by two slopes and a single breakpoint. The path loss exponent is around two, as in free space propagation, from the transmitter up to the breakpoint.Beyond the breakpoint the loss exponent isaround four, implyingasteeper decrease of the signal strength as illustrated in Fig. 5b. This breakpointis approximatclygiven byZrdz,Jt,,ih where hb is the base antenna height,h, is the if mobile antenna height. However,the receiver turns the corner from theLOS street into a non-lineof-sight (NLOS) street, then receiver experiences the adramaticdecreaseofthesignalstrengthbyaround ZOdBasshowninFig.Sc,thoughthisvariesdcpending on the street width and the distance between [ the transmitter and the corner131. O n a NLOS street, the signal envelope tends to follow a Rayleigh distribution. On a LOS path. the signal envelope follows not a Rayleigh but a Nakagami-Rice distribution, Multipath delay spread is usually small due to the existenceof a dominating, direct signal component.

data communication services within the workplace. Recent trials by major telephone companies show a huge demand and customer acceptance of indoor wireless communications. Some wiredlofficetelephonesystem,whileothersystemsoffer services separate from the PSTN. Indoorsystems can be broken down into three main classes: cordless telephone systems; in-building cellular systems: and local area networks (LANs). Each of thesetypesofemergingcommunicationsystemsmust he designed with the indoor channel in mind. For indoor communicationsystems,many design issues such as the distance between servers, expectedportable batterylife,customerperformance expectations, and the appropriate radio link budget are directly linked to the propagation environment. The amount RF interference that canhe of

expectedfromcochannelusersisanequallyimportant parameter, which is a direct functlon of the propagation characteristics inside buildings.

Propagation inside Single-story Buildings Agreat manypropagationmeasurementshave been conducted by telecommunication companies, research laboratories, and universities in order to determine reasonable design guidelines and propagation parameters for indoor systems. These measurements show that the particular type of

46

IEEE Communications Magazine * Januar?. IYY5

Indoor radio communica1 Office, hard partition I1500 13.0 I 7.0 1 tion systems I Office, soft partition I1900 I 2.6 I 14.1 1 are becoming increasingly important Multipath Delay Spread Buildings that have fewer metal and hard partitions typically have small o rms delay spreads, on the order f 30 to 60 ns. for extending Such buildings can support dald ratesin excess of several Mb/s without the need for equalization. voice and However, larger buildings with a great deal of metal and openaisles can have rms delay spreads as largc as 300 ns. Such buildings are limited to data comdata ratesof a few hundred kilobits per second without equalization. munication 1 Metalworking 1 1300 Path Loss - Path loss is a measure of the averservices age RF attenuation inside a building, and is it H Table 2.Path loss evponent and standard dummeasured by averaging the received signal over lion nleu\ured in diffrrenr huildings. several wavelengths at the receiver. Equation(1) within the describes the path loss situation also for indoor building has a direct impact on thc obscrvcd situations whcre the standard deviation is easily workplace. propagation characterictics. This has motivated more derived from the scatter plotof measured data. A
Retail stores
I914
-

by people throughout the building causes Ricean fading. with the ratio of specular signal powerto multipath signal having a valueof about 10 dB. This results in a typical variation of less than 15 dB for 99.9 percent of the time. a Ingeneral. a portable receiver moving inbuilding experienccs Raylcigh fading for OBS propagation paths and Ricean fading for LOS paths, regardless of the type of building. The Ricean K factor can vary from2 to 1 0 dB for portable Lerminals depending on the structureof the building and the number ofmultipathcomponentsarrive that at thereceiver.

smaller value of cs i m p l i e s t h a t t h e p a t h loss recent research that predicts indoor propagation using either statistical or deterministic models. model provides a better prediction for actual loss In order to differentiate propagation phenom- within a building. Equation ( I ) is well suited for system analysisor Monte Carlo simulationwhen the ena, researchers often classify buildings the by followingcategorics: rcsidential homesinsuburban impact of a large number of users within a building area5, residential homes in urban areas, traditionmust be considcrcd. Typical values path-loss of a l older office buildings with fixed walls (hard cxponents and standard deviations for different partitioni), open plan buildings with movable classes of buildings are given in the literature. wall panels (soft partitions), factory buildings, Table 2 provides a range of typical values that grocery stores, retail stores, and sports arenas. Hard been measured in the past. have partitions describe obstructions within the buildMolkdar's survey paper on the subject of indoor ing which cannot be easily moved such as existing propagationcontainsasummaryofnumerouspropwalls and aislcs. Soft partitions describe movable agationmeasurementsandmcdelsthat havebeendeveloped [18].The early works by Devasirvatham [19] obstructions such as office furniture panels,which have a height less than the cciling height. Inside a and by Salch and Valenzuela [20] have provided a building, propagation geometry be classified may as foundation for propagation work inindoor channels. lineofsight (L0S)where the transmitter andreceiver are visible to one another or obstructed (OBS), Propagation Between Floors where objects in the channel block a visible propPredicting radiocoverage between floors of buildphysicalsimilaritiesexistbetween agationpath. Often, ings has proven to be difficult, but measurements diffcrcnt typcs of buildings. For example, factory have shown that there are general rules-of-thumb buildings and grocery stores both contain a large that apply. Quantifying this propagation condiamount of metal inventory and oftcn havc fcw tion isimportant for in-buildingwirelesssystemsfor hard partitic~nswithin building. Traditionaloffice the multifloored buildings that need to share frebuildings,withmanywallsmadeofplasterandmetal queneieswithinthebuilding.1nordertoavoidcochanlathe,often hal.esimilarpropagationcharacteristics ne1 interference, frequencies must be reused on to large residential homes or retails stores that of different floors. The type building material contain many partitions. used between floors has been shown to impact Asdiscussedinthesectiononpropagationphysics, the RFattenuation betweenfloors. Poured concrete there arc number of important propagationevents a overmetallathingisapopularmodernconstruction techniquc which provides less RF attenuation that must he measured or modclcd before a reliwhich were used to ahlzwirelesssystemean bedesigned. Measurements than does solid steel planks separate floors in older buildings. The aspect by many researchers have more or lesc provided similar results, w,hich are summari& below. ratio of the building sides also makes a difference. Buildings that have a square footprint have higher attenuation levelsbetweenfloors than buildings that Temporal Fadingfor Fixed and Moving Terare rectangular. Also, windows with mctallic tint minals - For fixed terminals within buildings, measurements have shown that ambicnt motion impedc RF transmission, thus causing greater

IEEE CommunicationsMagazine * January I995

47

.~

60

60

70

80

90
4

During the
90
80

past few years, indoor propagation prediction techniques based on ray tracing have been used to reconstruct the many possible reflections porn wall surfaces within a building.

Tran
59 meters

[Path loss error]

59 meters
I
1

Figure 6. Error contour whch compares measuredand predicted path loss using a wtnple sire specificpurlition model described itt(51. Note that error is less fhun 3 dl3 over 80percenr of the coverage area.

attenuation between floors of a building. Measurements have shownthat the loss between floors does not increase linearly in dB with increasing separation distance. Rather, the greatesttloorattenuationfactor(F4F)indBoccurswhen the transmitter and receiver are separated by a single floor. The overall path loss increases at a smaller rate as the number of floors increase. This phenomenon is thought to be caused by diffraction o f radio energy along the sidesof a huilding, as well a\ scattercd energy from neighboring buildings, which can be reccived on different floors of the same building [Zl]. Typical values of attenuation between floors is 15 dB for one floor o f separation and an additional t o 10 d B p e r 6 tloor of separation up tofour floors of separation. For five or morefloors of separation, pathloss will increase by only a few dB foreach additional floor.

Computer-aided Design for In-building Propagation Prediction Statisticalpropagationmodelsdonotexplolt knowledge of the physical surroundings to the degree that building drawings can provide. State-of-the art propagation prediction uses computer-aided designdrawingsof the indoor environment as ameans of representing the physical locations of base stations, receivers, and partitions. For economic reasons, transceiversmustbe placed strategicallyso that only desired portionsof the building are provided with radio coverage. C A D tools enable a system designer to interactively model the performance of indoor systems. Under this interactive framework, RF Penetration into Buildings to propagation models can be modified provide a he signal strength received inside a buildof wide range of scenarios and base stations can be mg due to an external transmitter imporis added, repositioned, or deletedinteractively. Furtant f u r wireless systems that share frequencies

thermore, communication parameters such as received carrler-to-noise thresholds can adjustbe ed by the user as needed. Computer-aided design touls.such as AutoCAD. can display both single-floor and multifloor buildings. Building partitions can be identified in AutoCAD and assigned an attribute,so that appropriate propagation models may bc applied. Figure h shows an example of the modules providedinatvpicalsitemodelingtool.Workin[22,23] demonstrated that asimple and surprisingly accurate way tc) predict RF path loss inside a building is to count partitions between a line drawn between transmitter andrecewer. Ernpiricallyderived attenuation factors (AFs) may lhen he applied to each partition which intersects the line. Through an interactive proccss. parameters such FAFand AF as can be tunedfor a particular class ofbuildings. Measurements made within the environment can then be added to the CAD modeling tool a means as of updating the modelsin order t o provide more accurate mcasurernents at future sites. During the past fcw years, indoor propagation prediction techniques based on ray tracing have beenusedtoreconstructthemanypossiblcreflections from wall surfaces wlthin a building. This approach can easily be accommodated in computer aided design tools and shows promise for reducing the standard deviation o to within 4 dB over a50 or bo dB dynamic range within buildings[23-251.

T.

48

IEEE Communications Magarme January 1995

withneighboringbuildingsorwithoutdoorsystems.

Like KFpenetration measurementsbetweenfloors, it is difficult to determineexact models for penetration as only a limited number of experiments havc bccn published, and they are sometimes difficult to comparc. However, some generalizations can be made from the literature. In measurements reported date, to signal strength received inside a building increases with height. At thc lower floorsofabuilding,theurbanclutterinducesgreater attenuationandreducesthelevelofpenetratic~n.At higher floors, a LOS path may exist. thus causing a stronger incident signal at the exterior wall of the building. RF penetration has bccn found be a to function o f Crequency as well as height within thc building. Most measurements haveconsideredoutdoortransless than the mitters with antenna heights far maximum height of the building under test. MeasurcmcntsinLiverpool[26] showedthatpenetration loss decreascs with increasing frequency. Specifically, penetration attcnuation valuesof 16.4, 11.6, and 7.6 dB were measured on the ground floor of a buildiug at frequenciesof 441 MHz, 896.5 MHz, and 1400 MHz, respectively. Measureloss of 14.2, ments in [27] showed penetration 13.3. and 12.8 dB for 900 MHz, 1800 MHz, and 2300 MHz, respectively. Measurements made in front of windows indicated 6dB less penetration loss on average than did measurements made in parts of the building without windows. Walker [ZX] measured radio signals into 14 different buildings in Chicago from seven external cellular transmitters. Results showed that of building penetration loss decreased at a rate 1.9 dB per floor from the ground level to the up 15th floor and then began increasing above the 15th floor. The increase in penetration loss at the higher floors was attributed to shadowing effects ol' adjacent huildings. Similarly, [26] reported of penetration loss decreased at a rate 2 dB per floor from the ground levelup to the 9th floor and then increasedabove Ythfloor. Similarresults the were also reported in [29]. Measurements have shown that the percentage of windows, when compared with the building face surface area, impacts the level RI; of penetration loss. asdoes the presencetinted metal of in the windows. Metallic tints can provide from3 to 30 dBof RF attenuation in a single pane of glass.Theang1eofincidenceofthetransmittedwave upon the of the building also hasa strong impact face on the penetrationloss as shown in [30].

References
[ l ] D Cox. R Murray, and A Norrls, "800 MHz Atlenuatlon Measured in and Around Suburban Houses," AT&T Bell Laboratories Tech. 1.. "01. 673. no. 6 , July-Aug, 1984. pp 921-954 I21 W. C.Y. Lee, Mobile Communfcatlonr Deslgn Fundamentals. (Wiley. Second Edltion, 1993). I31 T 5 . Rappaport, "Indoor Radio Comrnunicatlons for Factories of the Future," IEEE Commun. Mag., May 1989. pp. 1 5 2 4 L41M Hata."Emplrlcalformulaeforpropagat~onloss~nlandmob~lerad~o services," IEEE Trans.Veh.Tech., L7-29, 1980, pp 31 7-325. I51 5. Y. Seldel and T. 5. Rappaport, "914 MHz Path Loss Predlctlon Models for Indoor Wireless C o m m u n t r a i l o n s ~n Multlfloored Bulldmgs." IEEE Trans. Ant. Prop, vol. 40. no. 2 . Feb 1992. 161T 5 Rappapon. et al , "900 MHz Multlpath Propagation Measure mentsforU.S.DigltalCellularRad~otelephone."iEEETrans. Tech , Veh. "01.39, no. 2, May 1990. pp 132-139. I71 D. C. Cox and R. P. Leck, "Distnbutjons of Multipaih Delay Spread and Average Excess Delay for 910 MHz Urban Mobile Radlo Paths." IEEE Trans Ant Prop, "01. AP 23. March 1975, pp. 206-213 1811. Bach Andersen. "Distribution of phase derivatlves In moblle communlcatlons."lEEProc,RH. "01. 137, no 4, Aug 1990, pp 197.201 191J. C-I Chuang. "The Effects of Time Delay Spread on Portable Radlo Communicatlons Channels wtth Digital Modulation:' IEEEJSAC. YOI 5,no 5, June 1987. pp 879 - 889 [ l o ] Y.Okumura et. a/., "Field strength and its variablllty in VHF and UHF land moblle rad10 m c e , " R of the ECL MI 16. 1968, pp. 825-873. w I l l ] 1. Walflsch and H. Bertani, "ATheoret~calModel of UHF Propagatlon In Urban Envlronments;' IEEE Trans. Antennas and Propagation. voI 36. no 12. Dec 1988. pp 1788.1796 1121 T Kurner. D. J Cichon. and W. Wiesbeck. "Concepts and Results far 30 Dlgntal Terraln-Based Wave Propagatlon Models." IEEE JSAC. vol. 11, no. 7, Sept. 1993, pp 1002 -1012 I131 Y. Nagata. et. ai.. "Measurement and modehng o f 2 GHr-band out-of-saghi radlo propagainon charactemtics under mlcrocellular environments," Proc. PIMRC'91. Sept. 1991. pp. 341-346. I14lA J Rustako,Jr et.al."Radiopropagationatmtcrowavefrequenc~es forl~ne-of-sightmicrocellularmobileandpersonalcommunicat~ons," IEEE Trans Veh. Techn, "01. 40, no. 1,Feb 1991, pp 203-210 I1511 Wiart."M~cro-cellularModel~ngwhenBaseStationAniennairbelow Roof Tops," Proc. IEEE Vehicular Technology Conference, June 1994, Stockholm, pp. 200-204. I161 W. T Webb, "Sizing up the microcell for rnoblle rad10 communicatlons. "Electronrcs and Commun. Eng J., pp. 133-140. June 1993 I171 A. J. Goldsmith and L. 1. Greenstem. "A Measurement-Based ModelforPredlctlngCoverageAreasofUrbanMicrocells."lEEEJSAC. vol 1 l . m 7,Sept1993,pp1013-1023 1181 D. Molkdar. "Review on Radlo Propagatlon Into and Within B u W ingr,"IEFProc voI 138. no 1. Feb 1991 1191 D. Oevasirvatham, '71me Delay Spread and Slgnal Level MeasureAnt. Prop.. YOI. AP-34, no. 11, NO". 1986. pp. 1300 1308. 1201 A. Saleh and R Valenzuela, "A statlstlcal model for Indoor multlpath propagabm,"IEEEJC4C. MI. SAC 5 , no 2, F b 1987, pp 138.146 e [21]W.Honcharenko,H.L.BeRon~,andJ.Da~ling,"MechanlsmsGoverning Propagatton Between Floors ~nBulld!ngs." IEEE Tram Antennas andPropagation, "01. 41, no. 6. June 1993, pp. 787-790. i221 A. Motley and 1. Keenan, "Rad10 Coverage In Bulldlngs." Brrttsh Telecom Tech.1.. "01.8, no. 1,Jan 1990, pp 19-24. 1231 5. Y. Seidel and T. S Rappapon. "A Ray Tracing Tcchnlque io Predlct Path tors and Delay Spread mslde Buildings." IEEE GLOBECOM, Dec. 1992, pp. 1825-1829. 1241 R. Valenzuela. "A Ray Traclng Approach t o Predictlng Indoor W~relers Transmlsrlan." IEEE Veh. Tech. Conf. 1993. pp. 214-218. 1251C. M. P. Ho, et. ai.. "Antenna Effects on Indoor Obstructed wire^ les~ Channels and a Determmlsttc Image-Bared Wldeband PropagationMod~forIn-BuildingPersonalCommunicat~onSystems,"lnf'l I Wirelesslnfo. Networks, "01. 1, no. 1, 1994. pp. 61-76. 1261A. M. D. Turkmanl, 1. D. Parson, D. G. Lewis, "Radio Propagatton Into Buildings at 441, 900. and 1400 MHz." Proc 4th Intl. Conf. on Land Moblle Radlo, Dec 1987. 1271A M 0. Turkmani. A. F. Toledo. "Propagation Into and Wlthln build^ fngr at 900, 1800, and 2300 MHz." 1992 IEEEVehTech C o d 1281 E H. Walker. "Penetration of Radio Stgnals Into Bulldmgs ~nCellular Radlo Envlronments:' BellSys. Tech. 1.. vol. 62.no 9. 1983.
[29]J.M.Durante,"BuildingPenetrationLossai900MHr."lEEEVeh.Tech.

mentsof850MH~rad~owaver1nbu~ld~ngenv~ranments."lEEFTram

The percentage of windows, compared with the building face sueace area, impacts the level of RF penetration loss, as does the presence of tinted metal in the windows.

Conclusion

Conf., 1993. I301 J Hortklshl. e t a / , "1 2 GHz Band Wave Propagatlon Measure-

TransVehTech.. L7-35. no. 4, 1986. espite theenonnouseffortsandprogrcsstodate, much work remains in the understanding and Biographies characterization of wireless communications IBRGEN BACH ANOERSEN IF '921 il a professor at Aalborg UnlverSlty. channels. Atrend in3Dnumerical modeling is seen, Aalborg. Denmark and head of Center for Personkommunikatlan He from which time delay statistics as well as coverIS actlve in International research concerning mobile communlcatlons bemg chalrman of COST 231 Warklng Group on UHF prapagatlon He age and interfercncc might be inferred.This isespe- IS also actwe ~nthe area of bloelectromagnetlrs and is presently "Icecially important with the extension to ever-higher president for URSI, the lnternatlonal Sclentlflc Radio Union datarates. Angle-of-arrival statisticsindifferentenviTHEODORE RAPPAPORT ISM '91 I I S an asroclate professor of electrical S ronmentswill also need to be better modeled for use englneermg and founder of the Moblle& Portable Rad10 Rerearch Group (MPRG) at Virginia Tech, Blacksburg, Virginla. with adaptiveantennas. In general, we may conclude PCS new a n d novel that to achieve ubiquitous Susuvu YOSHIDA15 a professor of electrical engineering at Kyoto Unlversity. Kyoto, Japan, leading a research group on wlreless personal ways of classifying wireless environn~ents be will communtcationr He I S Involved ~n multlpath propagatmn modellng nccdcd that are both widely encompassing and and predlrtlon, rector-antenna dsverslty, anti-multlpath modulation reasonably compact. schemes, and u-servlce multlpath delay spread monitoring, eic

mentsinConcreteBuildingsforlndoorRadioCommunications:'lEEE

IEEF Cummunications Magazine

January 1995

49