Điện thoại di động vô tuyến điện - Tuyên truyền Channel P7 potx

It is essential for radio engineers to plan systemsthat encompass this need, and a knowledge of the path losses between base stationsand transceivers located inside buildings is a vital

Before dealing with such channels, it is worth pausing to clarifya few points and

to identifythe ways in which the characteristics of the various channels di€er Wewish to distinguish between di€erences which are merelythose of scale and morefundamental di€erences of statistical character relating to the signal or theinterference Di€erences of scale are exempli®ed bythe urban radio channel This

is characterised byRayleigh plus lognormal fading and is the same whether themobile is vehicle-borne or hand-portable The di€erences are apparent because thefading rate experienced bya moving vehicle is generallymuch greater than the fadingrate experienced bya hand-portable Although these di€erences do not represent afundamental change in the statistical nature of the channel, theymaynot be trivial asfar as system designers are concerned For vehicles moving at a reasonable speed, it

is often adequate to determine the system performance averaged over the (Rayleigh)fading For a hand-portable it maybe more meaningful to determine the maximumerror rate over a speci®ed large percentage of locations Changes of statisticalcharacter are exempli®ed byindoor radio channels where the interferenceenvironment di€ers markedlyin magnitude and nature from that outside, and therural channel where the signal statistics are not well described bythe Rayleighmodel

Copyright & 2000 John Wiley& Sons Ltd Print ISBN 0-471-98857-X Online ISBN 0-470-84152-4

7.2 RADIO PROPAGATION INTO BUILDINGS

During recent years there has been a marked increase in the use of hand-portableequipment, i.e transceivers carried bythe person rather than installed in a vehicle.Such equipment is particularlyuseful in cellular and personal radio systems and nowcompletelydominates the market It is essential for radio engineers to plan systemsthat encompass this need, and a knowledge of the path losses between base stationsand transceivers located inside buildings is a vital factor that needs to be evaluated.The problem of modelling radio wave penetration into buildings di€ers from themore familiar vehicular case in several respects In particular:

The problem is trulythree-dimensional because at a ®xed distance from the basestation the mobile can be at a number of heights depending on the ¯oor of thebuilding where it is located In an urban environment this mayresult in there being

an LOS path to the upper ¯oors of manybuildings, whereas this is a relativelyrareoccurrence in citystreets

The local environment within a building consists of a large number of obstructions.These are constructed of a varietyof materials, theyare in close proximityto themobile, and their nature and number can change over quite short distances.There have been several investigations of radio wave penetration into buildings,particularlyin the frequencybands used in cellular systems [1±7] Theycan bedivided into two main categories:

Those that consider base station antenna heights in the range 3.0±9.0 m andmobiles mainlyoperating in one- or two-storeysuburban houses

Those which consider the problem for base station antenna heights similar tothose used in cellular systems and mobiles operating in multi-storey oce buildings.Investigations in the ®rst categoryall originated in connection with the design of aproposed Universal Portable Radio Telephone System [8] Because such a systemwould need to cater for large numbers of verylow-power portables, it is based on averysmall cell size (<1.0 km radius) Moreover, in such a system it is considered thatcoverage within multi-storeyoce buildings will be provided bya number of cellswithin the building It is for these reasons that the studies have used low base stationantenna heights, base-to-mobile distances less than 1 km, and have concentrated ontaking measurements in buildings the size of suburban houses

In existing cellular systems, base stations for macrocells are typically located on theroof of a tall building which maybe 100 m or more above the local terrain, and base-to-mobile distances of 1 km or more are of interest Consequently, it is dicult to use theresults directlyin the design of current-generation systems However, these studieshave shown that the signal in small areas within buildings is approximatelyRayleighdistributed with the scatter of the medians being approximatelylognormallydistributed

In other words, the signal statistics within a building can be modelled as superimposedsmall-scale (Rayleigh) and large-scale (lognormal) processes ± the model used forradio propagation outside buildings in urban areas The variation of signal level withantenna height is consistent with the presence of a re¯ecting ground plane

Cox et al investigated the power±range law by®tting results to an equation of the form

where S is a constant and d is the distance between transmitter and receiver Theexperiments were conducted using a ®xed receiver and a hand-held transmitter whichwas moved around in areas of 4 ft2 (0.37 m2) throughout the building The values of

n were found to be 4.5, 3.9, 3.0 and 2.5 for measurements outside the building, on the

®rst ¯oor, on the second ¯oor and in the basement, respectively

With one exception [6], studies in the second categoryhave been concerned withthe statistical characterisation (median or mean, variance and CPD) of the `buildingloss', a term ®rst introduced byRice [9], to denote the di€erence between the mediansignal on a given ¯oor of a building and the median signal level outside, in the streetsimmediatelyadjacent to the building However, in reading the literature there is aneed for some care; this de®nition has been interpreted in di€erent ways There aretwo obvious possibilities, either to take a number of measurements in the streets thatsurround the building to produce an average external measurement as suggested byRice, or alternativelyto use the signal level at a point immediatelyoutside thebuilding in line with the centre of the building and the transmitter location [2].The second method has merit when an LOS path exists between the transmitterand the building concerned, but generallywhen this is not the case, and energyentersthe building via a number of scattered paths, the ®rst method seems more realistic.The method of data analysis also di€ers, although in almost all investigations thesignal has been sampled at ®xed intervals of time or distance In general thedi€erent methods of data analysis do not signi®cantly a€ect the measured value ofmean building penetration loss, but calculations of the signal variabilitycan bea€ected depending upon whether this is described in terms of a standard deviation or

as a statistical distribution function

For these reasons it is sometimes dicult to compare the results from the di€erentinvestigations The penetration loss depends on a number of factors, central amongthem being the carrier frequency, the propagation conditions along the path and theheight of the receiver within the building However, there are several otherin¯uencing factors which include the orientation of the building with respect to thebase station, the building construction (the construction materials and the numberand size of windows) and the internal building layout Their in¯uence and relativeimportance will become apparent later Almost all models for predicting signalstrength in buildings have used the technique proposed byRice, i.e ®rstlypredict themedian signal level in the neighbouring streets using one of the known methods andthen add the building penetration loss

An investigation byBarryand Williamson in New Zealand [10] concentrated originally

on buildings where the majorityof ¯oors had a line-of-sight path to the base station Byusing criteria similar to those for the vehicular environment, i.e that the best statisticaldescriptor was one which adequatelypredicted values near the tails, it was found that thesignal on any¯oor was best ®tted bySuzuki statistics and at 900 MHz the standarddeviation of the lognormal part of the distribution was 6.7 dB It was also suggested thatmirror-glass windows could introduce an additional loss of the order of 10 dB

A series of experiments in the UK at frequencies of 441, 896.5 and 1400 MHz [11]produced general conclusions about signal variabilitysimilar to those from previousinvestigations, and theyalso provided an insight into the e€ects of transmission conditionsand carrier frequency The transmission conditions appear to have a strong e€ect on thevalue of the standard deviation and on the departure of the distribution from lognormal

Table 7.1 shows the penetration loss for three di€erent frequencies (441, 896.5 and

1400 MHz) for a receiver located in a modern six-storeybuilding The penetrationloss decreases byaround 1.5 dB as the frequencyis increased from 441 to 896.5 MHzand bya further 4.3 dB when the frequencyis raised to 1400 MHz These results (thedecrease in penetration loss at higher frequencies) are consistent with the conclusionsdrawn byRice [9] and Mino [12]

A di€erent series of measurements using a number of large buildings has producedground-¯oor penetration loss values of 14.2, 13.4 and 12.8 dB at 900, 1800 and

2300 MHz respectively It can be argued that for system designers, the penetrationloss at ground-¯oor level is the most important because if a system is designed to giveadequate service to mobiles at ground-¯oor level, then service on higher ¯oors within

a building will almost certainlybe as good if not better

It is worth re-emphasising that the total loss between the base station and themobile has been split into two parts: the loss from the base station to points in thestreets surrounding the building concerned and the additional penetration loss fromthe street into the building itself This has the advantage that established methodscan be used to estimate the ®rst component, and the penetration loss then becomes

an additional factor Although the penetration loss, as de®ned, decreases withfrequencyin the range considered above, the path loss from the base station to thestreets outside will increase This factor dominates, so the total path loss betweentransmitter and receiver will always increase as the frequency is raised

The transmission conditions have a strong in¯uence on the value of the standarddeviation and also on the departure of the distribution from lognormal Figure 7.1shows that when no LOS path exists, the large-scale signal variations exactly®t alognormal distribution and that the standard deviation is about 4 dB In othercircumstances where there is an LOS path to the whole building or part of thebuilding, the large-scale signal variations depart somewhat from the lognormal andhave a higher standard deviation For complete LOS the standard deviation is 6±

7 dB These values are veryclose to those reported byCox [2]

Two building construction e€ects have been noted First, the standard deviation ofthe large-scale variations is related to the ¯oor area of the building concerned;smaller ¯oor areas lead to lower values of standard deviation and vice versa.Secondly, the penetration loss generally reduces as the receiver is moved higher

Table 7.1 Mean penetration loss on various ¯oors of a six-storeybuildinga

Floor level Penetration loss (dB)

441.0 MHz 896.5 MHz 1400.0 MHzGround 16.37 11.61 7.56

within a building; indeed there maybe an LOS path to the higher ¯oors of a buildingwhen no such path exists to the streets outside or to lower ¯oors of the building.Occasionally, however, it has been found that the penetration loss increases at highlevels within a building A result of this kind was reported without discussion byWalker [7], where the penetration loss increased from 1:4 dB at ¯oor 9 to 15.3 dB at

¯oor 12 of the same building It seems likelythat such increases result from thespeci®c propagation conditions existing between the transmitter and receiverlocations Figure 7.2 [11] shows a change of about 2 dB per ¯oor, and this agreesverycloselywith the ®ndings of other workers [4,7,13]

In summary, when the transmitter is outside, the signal within a building can becharacterised as follows:

The small-scale signal variation is Rayleigh distributed

The large-scale signal variation is lognormallydistributed with a standarddeviation related to the condition of transmission and the area of the ¯oor The building penetration loss, as de®ned, decreases at higher frequencies When no line-of-sight path exists between the transmitter and the buildingconcerned (i.e scattering is the predominant mechanism) the standard deviation ofthe local mean values is approximately4 dB When partial or complete line-of-sightconditions exist, the standard deviation rises to 6±9 dB

The rate of change of penetration loss with height within the building is about 2 dBper ¯oor

Finallywe comment brie¯yon the matter of modelling Most of the outdoorpropagation models in Chapter 4 were developed and optimised for macrocells, andwithout further validation theyare not necessarilyreliable for microcellularpropagation where the antenna height is low In addition, predicting ®rst the

Figure 7.1 Cumulative distribution of the large-scale variations of the signal at 900 MHzwithin a building when no line-of-sight path exists: (Ð) measured, (± ± ±) theoreticallognormal distribution with standard deviation 4 dB

average signal level in the streets surrounding a building using a method which haslimited accuracyand then adding a building penetration loss, itself subject tostatistical variation, inevitablyleads to a reduction in accuracy It seems clear thatthe prediction of path loss from an external transmitter to a receiver located within abuilding will be more accurate if it is undertaken directlyand not merelyas anextension of outdoor modelling Indeed, Barryand Williamson [14] suggestedcombining factors associated with propagation into buildings with factors associatedwith propagation inside buildings to produce a comprehensive model.

Toledo et al [15] undertook a multiple regression analysis of a large database andinvestigated the relationships between a number of variables The best results wereobtained byincluding three variables in the regression equations, the distance dbetween transmitter and receiver, the ¯oor area Af of the building concerned and afactor SQwhich represents the number of sides of the building which have an LOSpath to the receiver The models at 900 and 1800 MHz respectivelyare

L50 ˆ 37:7 ‡ 40 log10 d ‡ 17:6 log10 Af 27:5SQ

L50ˆ 27:9 ‡ 40 log10 d ‡ 23:3 log10Af 20:9SQ …7:2†The root mean square errors between these equations and the measurements fromwhich theywere derived are 2.4 and 2.2 dB respectively, slightlylower than thoseobtained byBarryand Williamson from their measurements in Auckland [14]

7.3 PROPAGATION INSIDE BUILDINGS

In cordless telephone systems the indoor portion of the subscriber line is replaced by

a radio link so that the telephone handset can be carried about freelywithin a limited

Figure 7.2 Building penetration loss as a function of height within the building:  areexperimental points

area, calls being initiated and received in the usual way The demand for such systemshas prompted research into the propagation characteristics of radio signals whereboth the transmitter and receiver are within a building The possibilityof cordlesstelephone exchanges and the general interest in indoor radio systems of various kindsare added factors that have given impetus to this topic There have been severalinvestigations over a wide range of frequencies; we will onlybe able to present a ratherbrief review However, let us begin bynoting that propagation within buildings is verystronglyin¯uenced bythe local features, i.e the layout of the particular buildingunder consideration and the building construction materials used for the walls, ¯oorsand ceilings It is conceivable that radio communication inside buildings could beaided by the use of leaky-feeder systems, but that topic will not be considered here.Indoor radio di€ers from normal mobile radio in two important respects: theinterference environment and the fading rate The interference environment is oftencaused byspurious emissions from electronic equipment such as computers, and thelevel can sometimes be much greater than that measured outside Moreover, there aresubstantial variations in signal strength from place to place within a building Thesignal can be highlyattenuated after propagating a few metres through walls, ceilingsand ¯oors or maystill be verystrong after propagating several hundred metres along acorridor The signal-to-interference ratio is unpredictable and highlyvariable.The slow fading rate makes it inappropriate to calculate system performance byaveraging over the fading; it is more appropriate to envisage two possibilities asfollows First if the user of, say, a cordless telephone is moving around slowly duringthe conversation then the antenna will pass through several fades, albeit ratherslowly This situation can best be described in terms of the percentage of time forwhich the signal-to-interference ratio falls below an acceptable threshold or, in adigital system, the percentage of time for which the error rate exceeds a given value.However, because of secondarye€ects (e.g motion of other people, doors beingopened and closed), these probabilities will change slowlywith time Surveypapersexist [16,17] which discuss the literature available at the time of writing.

intersymbol interference due to delay spread, and this limits the data rate Thus,

in narrowband systems, multipath and shadow fading limit the coverage, whereasinterference causes major problems even within the intended coverage area.Interference, discussed in Chapter 9, can be natural or man-made noise or it cancome from other users in a multi-user system It limits the number of users that can

be accommodated within the coverage area Techniques such as dynamic channelassignment, power control and diversity[18] can help used to reduce the problems.7.3.1 Propagation characteristics

Several investigations have been undertaken to determine radio propagationcharacteristics in houses [3,19±21], oce buildings [22±24] and factories [25] Oneearlyinvestigation, prompted bythe proposed introduction of a cordless telephonesystem in Japan, was concerned with the 250 MHz and 400 MHz bands [19] As aresult of measurements made using a low-power (10 mW) transmitter, it wasconcluded that the median path loss follows the free space law for veryshortdistances (up to 10 m), it then increases almost in proportion to distance If the

propagation path was blocked byfurniture of various kinds, the characteristics werea€ected in di€erent ways and no general statements were made The short-termvariations in signal about the median value were closelyrepresented bya Rayleighdistribution as a result of scattering from walls, ¯oors, ceilings and furniture.

A law relating path loss to distance from the transmitter can be used to predictsignal strength in a building of a given structure, but it is dicult to make generalstatements The best approximations to straight-line characteristics are most likelytooccur where rooms are of a similar size, uniformlyarranged, with walls of uniformattenuation between each room [20] The exponent n in the power law varies fromapproximately2 (free space) along hallways and corridors to nearly6 over highlycluttered paths

Motleyand Keenan [26] reported the results of experiments in a multi-storeyoceblock at 900 and 1700 MHz A portable transmitter was moved around selectedrooms in the building while a stationaryreceiver, located near the centre of the oceblock monitored the received signal levels The conventional power±distance law wasexpressed in the form of equation (7.1) as

P ˆ P0‡ kF ˆ S ‡ 10n log10dwhere F represents the attenuation provided byeach ¯oor of the building and k is thenumber of ¯oors traversed When P0was plotted against distance d, on a logarithmicscale, the experimental points layveryclose to a straight line Table 7.2 summarisesthe values of the measured parameters Notice that n is similar at both frequenciesbut F and S are respectively6 dB and 5 dB greater at 1700 MHz These results werecon®rmed bytests in another multi-storeybuilding with metal partitioning Overallthe measured path loss at 1700 MHz was 5.5 dB more than at 900 MHz, which agreeswell with theoretical predictions based on reduced e€ective antenna aperture.Other workers [27] have obtained a loss of 3±4 dB through a double plasterboardwall and a loss of 7±8 dB through a breeze block or brick wall These values are lessthan through a ¯oor, probablybecause ¯oors often have metal beams andreinforcing meshes which are not present in the walls It seems that at 1700 MHzthere is a greater tendencyfor RF energyto be channelled via stairwells and liftshafts than at 900 MHz It has been reported that the losses between ¯oors arein¯uenced bythe construction materials used for the external walls, the number andsize of windows and the type of glass [28]

The external surroundings also have to be considered since there is evidence[29,30] that energycan propagate outwards from a building, be re¯ected andscattered from adjacent buildings and re-enter the building at a higher and/or lowerlevel depending upon the location of the antenna and its polar pattern Experimentshave also shown that the attenuation between adjacent ¯oors is greater than the

Table 7.2 Propagation parameters within buildings

F (dB) S (dB) nFrequency ˆ 900 MHz 10 16 4

Frequency ˆ 1700 MHz 16 21 3.5

incremental attenuation caused byeach additional ¯oor and that after ®ve or six

¯oors there is little further attenuation Several workers [2,31] have publishedinformation about signal losses caused bypropagation through various buildingmaterials over a wide range of frequencies

It appears that propagation totallywithin buildings is more dependent on buildinglayout and construction in the 1700 MHz band than it is at 900 MHz The lowerband (860 MHz) is alreadyused for the Digital European Cordless Telephone(DECT) system, which is designed for domestic and business environments It o€ersgood qualityspeech and other services for voice and data applications, and itprovides local mobilityto users of portable equipment in conjunction with an in-building exchange Although propagation losses increase with frequency, the

1700 MHz band mayalso be viable for an in-building cordless telephone systemwhere, in anycase, the number of base stations is dictated bycapacityandperformance requirements rather than bythe limitations of signal coverage

Experiments reported byBultitude [24] give an indication of signal variabilitywithin buildings at 900 MHz Although it might be anticipated that for locationswhere there is no line-of-sight path, the data would be well represented bya Rayleighdistribution as reported at lower frequencies [19], this did not prove to be the case.Data representing such locations was generallyfound to be Rician distributed with aspecular/random power ratio K of approximately2 dB Exceptional locations werefound where Rayleigh statistics ®tted well For any ®xed location having theseRician statistics there is a 90% probabilitythat the signal is greater than 7 dB butless than ‡4 dB with respect to that determined bylosses along the transmitter±receiver path Temporal variations in the received signal envelope are also apparent

as a result of movement of people and equipment These variations are slow andhave characteristics that depend upon the ¯oor plan of the building

In buildings which are divided into individual rooms, fading is likelyto occur inbursts lasting several seconds with a dynamic range of about 30 dB In open oceenvironments fading is more continuous with a smaller dynamic range, typically

17 dB These temporal envelope variations are Rician with a value of K between 6and 12 dB The value of K is a function of the extent to which motion within thebuilding alters the multipath structure near the receiver location Terminal motionalso causes fading due to movement through the spatiallyvarying ®eld This isadequatelydescribed, as above, bya Rician distribution with K  2 dB

There have been several attempts to model indoor radio propagation using anextension of eqn (7.1):

where Xs is a lognormal variable (normallyin dB) with standard deviation s.Anderson et al [32] give typical values of n and s for a varietyof buildings over arange of frequencies, n lying in the range 1.6±3.3 and s being between 3.0 and 14 dB.Seidel [28] also gave values for a varietyof situations in di€erent buildings, derivedfrom measurements in a large number of locations These values were used to modelpropagation using an equation of the form

where nSFrepresents the value of the exponent for measurements on the same ¯oor.Assuming that a good estimate of nSFexists, the path loss on a di€erent ¯oor can befound byadding an appropriate value of the ¯oor attenuation factor F.Alternatively, in eqn (7.4) F can be removed byusing an exponent nMF whichalreadyincludes the e€ect of multiple-¯oor separation The propagation equationthen becomes

Devasirvatham [33] found that the in-building path loss could be modelled as the freespace loss plus an additional loss that increased exponentiallywith distance, thusimplying that the total loss could be expressed by a modi®cation of eqn (7.4):

L50 ˆ 18:8 ‡ 39:0 log10d ‡ 5:6kf‡ 13:0Swin 11:0G 0:024Af

L50 ˆ 24:5 ‡ 33:8 log10d ‡ 4:0kf‡ 16:6Swin 9:8G 0:017Af …7:7†

In these equations kf is the number of ¯oors separating the transmitter and receiver;

Swin is a factor representing the amount of energywhich leaves and re-enters thebuilding (it takes into account the position of the transmitter relative to the externalwalls of the building); G represents the observed tendencyfor the signal to bestronger on the lowest two ¯oor of the building; and Afis the ¯oor area of the roomcontaining the receiver Swin is given a value between 0 and 1 depending on therelative location of the radio terminals

For rooms on the same side of the building as the transmitter, Swinˆ 1; for rooms

on the opposite side Swinˆ 0:25; and for those on the two sides perpendicular to theside where the transmitter is located, Swinˆ 0:5 For internal rooms with no externalwindows Swin ˆ 0 Some judgement is needed to assign values to rooms close to thetransmitter, to corridors and to areas separated from the transmitter onlyby, say, asingle wooden door which mayor maynot be open at anytime

The factor G was set equal to 1 on the lower two ¯oors and it was 0 elsewhere.Although it maybe dicult to predict the path loss accuratelyfor receiver locationsclose to the transmitter, this is of academic interest onlysince the signal is likelyto behigh, providing good communication The best signal coverage of anybuilding isusuallyachieved bylocating the transmitter in a large room as near as possible to thecentre of the building [30]

7.3.2 Wideband measurements

In addition to narrowband measurements designed to determine how median signalstrength varies with distance and to evaluate signal variability, there have also beenseveral investigations of the wideband characteristics of propagation withinbuildings

Measurements of time-delayspread in oce buildings and residences have beenreported byDevasirvatham [37±39] using equipment operating at 850 MHz with atime-delayresolution capabilityof 25 ns (i.e paths di€ering in length by7.5 m ormore can be resolved) It appears that the detailed shape of the individual power±delaypro®les have little impact on the performance of a radio system [40,41], soe€ort was concentrated on evaluating the average delayand the RMS delayspread.

In general, the delays and delay spreads are smaller than corresponding valuesmeasured outside buildings The averaged time-delaypro®le in Figure 7.3 representsdata collected in a large, six-storeybuilding and has an RMS time-delayspread of

247 ns Figure 7.4 shows the cumulative distribution of time-delayspread for this ocebuilding and a smaller two-level building A portable communications system wouldhave to work under worst-case delayspread, which for both these oce buildings isabout 250 ns Larger delayspreads, in the range 300±420 ns, were measured atresidential locations, particularlyon inside-to-outside paths, but the limited number oflocations that were used makes general conclusions rather dicult to draw Note,however, that whenever a line-of-sight path exists between transmitter and receiver, theRMS delayspread is signi®cantlyreduced, typicallyto less than 100 ns

Bultitude et al [42] compared indoor characteristics at 900 MHz and 1.75 GHzusing equipment with parameters the same as Devasirvatham's Measurements weremade in a four-storeybrick building and in a modern building of reinforced concreteblocks, both in Ottawa, Canada There were perceivable di€erences in the measuredcharacteristics, but these seemed to be more a function of the location than afunction of the transmission frequency In one building, RMS delay spreads wereslightlygreater at 1.75 GHz for over 90% of locations (28 ns compared with 26 ns),whereas in the other building the reverse was true for about 70% of locations.Although the results indicated that coverage would be less uniform in both buildings

at 1.75 GHz, theyalso showed that coverage would be less uniform in one of thebuildings than in the other, regardless of the transmission frequency It seemsdicult, on the basis of this work, to conclude anything except that there is littledi€erence between the wideband frequencycorrelation statistics in the two frequencybands

A statistical model for indoor multipath propagation has been presented bySalahand Valenzuela [43] based on measurements at 1.5 GHz using 10 ns radar-like pulses

in a medium-sized oce building Their results showed that the indoor channel isquasi-static, i.e it varies veryslowly, principallyas a result of people moving around.The nature and statistics of the channel impulse response are sensiblyindependent ofthe polarisation of the transmitter and receiver provided that no line-of-sight pathexists The maximum delayspread observed was 100±200 ns within rooms, butoccasionallyvalues greater than 300 ns were measured in hallways It is veryinteresting to note that the measured RMS delayspread within rooms had a medianvalue of 25 ns and a worst-case value of 50 ns, ®ve times smaller thanDevasirvatham's results from a much larger building

A simple statistical model was proposed in which the rays that make up thereceived signal arrive in clusters The rayamplitudes are independent Rayleighrandom variables with variances that decayexponentiallywith cluster delayas well

as with raydelaywithin a cluster The corresponding phase angles are independentrandom variables uniformlydistributed in the range (0, 2 p) The clusters, and the

rays within a cluster, form Poisson arrival processes with di€erent but ®xed rates,and the clusters and the rays have exponentially distributed interarrival times Theformation of the clusters is determined bythe building structure and the rays within

a cluster are formed bymultiple re¯ections from objects in the vicinityof thetransmitter and receiver Both discrete and continuous versions of the model arepossible However, it has been suggested [44] that the discrepancies actuallyarise as aresult of the Poisson arrival assumption and that a modi®ed Poisson process is morerepresentative Furthermore, the path amplitudes have been shown to follow alognormal distribution rather than a Rayleigh distribution

Finally, Rappaport et al [25,45,46], again using similar equipment, have studiedmultipath propagation in factorybuildings at 1300 MHz Substantial physicaldi€erences exist between such buildings and oces or residential houses in respect ofconstruction techniques, contents and placement of walls and partitions It might be

Figure 7.3 Measured time-delaypro®le within a large six-storeybuilding (afterDevasirvatham)

Figure 7.4 Cumulative distribution of time-delayspread within two oce buildings

expected, therefore, that propagation characteristics would also be di€erent In fact,

it was found that the path loss exponent n was approximately2.2 and that Ricianfading was the norm The RMS delayspread ranged between 30 and 300 ns, themedian values being 96 ns for line-of-sight paths along aisles and 105 ns forobstructed paths across aisles The worst-case measured value was 300 ns Thesevalues are comparable with those measured in large oce buildings [38]

Table 7.3 brings together some of the in-building ®gures that have been reported.De®nitive conclusions are not easybecause the propagation conditions are so variable Itseems that where line-of-sight paths exist, the propagation law exponent is usuallynear 2,indicating that a free space mode is dominant, and this is accompanied byRician ratherthan Rayleigh fading For obstructed paths the exponent rises to 4 or more, and although

in manycases the fading is still characterised byRician statistics, Rayleigh characteristicshave also been reported It is likelythat Rician channels will support higher data rates.Wideband measurements have been made at frequencies in the range 850±1750 MHz butthere are no obvious e€ects that can be attributed to changes in the carrier frequency.There is no evidence to suggest that the scattering and re¯ecting properties of thematerials used for construction change appreciablyover this frequencyrange, as thedelayspreads do not exhibit anysigni®cant statistical di€erence

It might be expected that delayspread would decrease with frequencydue toincreased attenuation bythe structural materials, but this is certainlynot apparentbelow 2 GHz On the other hand, there is some evidence [21,47] that at 60 GHz thepropagation mechanism is di€erent since the radio waves are e€ectivelyscreened byanymetal partitions Although at this frequencythere is some leakage through doorsand windows, this is insucient to give room-to-room coupling except where a line-of-sight path exists At this frequencythe transmission, re¯ection and absorption

Table 7.3 Measured parameters from propagation experiments inside buildings

Investigators FrequencyEnviron- RMS delayspread (ns) Worst Propagation

ment

Medianvalue Standarddeviation

case(ns) exponent nlaw

Bultitude

et al 910 MHz1.75 GHz Within brickand concrete

ocebuildings

26±3028±29 17±228±11

Rappaport 1.3 GHz In factory

buildings 105 (NLOS)96 (LOS) 300 2.2

LOS ˆ line-of-sight.

NLOS ˆ non-line-of-sight.

properties of materials commonlyused for building construction varyverywidely.However, no wideband measurements have been reported.

7.4 RAY TRACING: A DETERMINISTIC APPROACH

In Chapter 3 we noted that the availabilityof high-resolution databases makes itmore attractive to move towards deterministic propagation methods We can neverhope for 100% accuracyof course because databases are rarelycompletelyup todate, and there are always factors such as moving vehicles, trees in or out of leaf and,inside buildings, changes of furniture location, which introduce uncertainties.Nevertheless, propagation methods based on raytheoryhave been the subject ofmanyinvestigations in recent years Theyhave been used for both indoor andoutdoor environments, and in theorytheyhave enormous potential If a number ofrays can be traced from a given transmitter location to a given receiver location, theelectrical lengths of the various raypaths give the amplitudes and phases of thecomponent waves and theycan be used to calculate the signal strength In thiscontext, due account must be taken of changes in amplitude and phase caused bypropagation through, or re¯ection from, obstacles along the raypath Moreover, thephysical lengths of the ray paths allow calculation of the propagation times alongthose paths, thus permitting evaluation of delayspread and other similar parameters.The characteristics of the antennas used at both ends of the link can be built intothe prediction algorithm, so methods based on raytheoryhave the potential toprovide a complete channel characterisation as far as propagation is concerned Thiscan be in two or three dimensions depending upon the nature of the availabledatabases In outdoor environments, sophisticated processing techniques can be used

to convert aerial or satellite photographs into 3D databases; in indoor environments,architectural drawings and other layout information can serve the same purpose.However, the extent to which anygiven raywill penetrate, be re¯ected from, or bedi€racted around a given obstacle depends cruciallyon the electrical properties ofthe material or materials from which the obstacle is constructed as well as on itsgeometrical shape

The equations in Section 2.3.1 show that the re¯ection coecient of a planesurface depends on the polarisation of the incident wave, the angle of incidence andveryimportantlyon the dielectric constant and conductivityof the material Precisevalues of conductivities and dielectric constants are needed if accurate predictionsare to be obtained Re¯ection from a curved surface, surface roughness anddi€raction were all discussed in Chapter 2 and have a part to playin predictionmethods based on raytheory

The propagation model normallyrecognises that when an obstacle exists in thepath of a ray, the ray can be specularly re¯ected, scattered, transmitted (and partiallyabsorbed in the process) or in some cases di€racted around the edge of the obstacle.Specular re¯ection is characterised bythe incident and re¯ected rays making equalangles with the normal to the surface, transmission obeys Snell's law of refraction,and di€raction e€ects can be estimated using anyof the methods discussed inChapter 3, e.g UTD Scattering is not so easyto deal with and is often neglected onthe basis that the vast majorityof the energyis contained in the specularlyre¯ectedcomponent Whether this is justi®ed or not, depends on the particular propagation

scenario Re¯ected and transmitted rays have an inverse square law powerdependence (cf free space propagation) depending on the total distance travelled.Care is necessaryin applying the re¯ection coecients given byeqns (2.9) and(2.10); for smooth surfaces, conservation of energydictates that the transmissioncoecient is (1 re¯ection coecient) The proper re¯ection coecient must be useddepending on the polarisation of the rayrelative to the obstacle concerned Forexample, in an indoor environment, when a verticallypolarised raylaunched from atransmitter meets the ¯oor or ceiling, the E-®eld is normal to the surface and eqn(2.10) applies On the other hand, the E-®eld is parallel to walls, so eqn (2.9) should

be used Oblique incidence can be treated byresolving the incident rayinto twoorthogonal components and proceeding appropriately

Two basic methods appear in the literature, the raylaunching or `brute force'method [48] and raytracing [49] Reciprocityapplies as far as each individualpropagation path is concerned, but it is customaryand more intuitive to trace raysassuming that theystart at the transmitter, since the single-transmitter/multiple-receiver scenario is byfar the most common This is particularlyrelevant in the raylaunching method which works as follows

A software program checks for an LOS between the speci®ed transmitter andreceiver locations Next it launches and traces a rayawayfrom the transmitter in aspeci®ed direction and detects whether it intersects an obstruction speci®ed on thedatabase If it does not, the process stops and a new source rayin a di€erentdirection is launched If an intersection is found, the program determines whether there¯ected rayfrom the intersection point has an unobstructed path to the receiver,and the re¯ected and transmitted rays are then traced to the receiver or to anotherobstruction This recursive process ± launching a rayat a given angle and tracing itspath ± continues for each rayuntil the rayreaches the receiver, until a speci®ednumber of intersections is exceeded, until the rayenergyfalls below a speci®edthreshold (e.g rays which pass through obstructions such as walls) or until nofurther intersections occur Of course, rays launched in certain directions will neverreach the receiver because the geometryis such that no path exists

To determine all possible rays that propagate between the transmitter andreceiver, it is necessaryto consider all possible angles of launch from the transmitterand arrival at the receiver One wayof doing this is to consider a large number ofrays, each separated from its neighbouring rays by a small but constant angle in 3Dspace It appears that an acceptable trade-o€ between coverage and computationtime is attained with an angular separation of about 18 [50] It is also necessarytodecide whether anyrayhas reached the receiver, byapplying a minimum distancetest Since it would be unrealistic to regard the receiving location as beingin®nitesimallysmall, an imaginarysphere of small radius is constructed around thereceiving point and anyraywhich intersects this sphere is considered to have beenreceived The signal strength calculated from the phasor addition of all received rays

is considered to be the mean signal over the area de®ned bythe sphere

Image-based raytracing di€ers from raylaunching and appears to have someadvantages Instead of using the `brute force' approach of launching manyrays(often up to 40 000) at verysimilar angles, the technique considers all obstructions aspotential re¯ectors and calculates their e€ect using the method of images This is astrictlyanalytical approach which does not require the use of a receiving sphere,