Measurement of multipath delay statistics over a 72-to 90-MHz bandwidth at 1.8 GHz in two European cities using a chirp sounder

Transcription

1 Rado Scence, Volume 34, Number 4, Pages , July-August 1999 Measurement of multpath delay statstcs over a 72-to 90-MHz bandwdth at 1.8 GHz n two European ctes usng a chrp sounder S. Salous Department of Electrcal Engneerng and Electroncs, Unversty of Manchester nsttute of Scence and Technology Manchester, England, Unted Kngdom. AbstracL Wdeband measurements at 1.8 GHz across a 72- to 90-MHz bandwdth were camed out n the urban envronments of Dubln (reland) cty center and Manchester (England) cty cen- ter. The measurements were conducted usng a programable chrp sounder, whch employed a heterodyne detector at the recever. The dgtzed raw data were subsequently analyzed off-lne by dvdng the 72- or 90-MHz bandwdth nto sectons of 18 MHz. The resultng multpath delay structure as a functon of frequency gave a panoramc vew of the frequency selectve fadng of the ndvdual multpath components. The effects of the frequency selectvty on the channel tme delay descrptor such as average delay, rms delay spread, and the-10-, -15-, and-20-db wdths of the power delay profle were subsequently analyzed. These parameters were estmated both at the small-scale and large-scale levels as a functon of frequency. For the 18-MHz sectons, sgmfcant devatons of the order of % and % were detected n the small-scale channel characterzaton n Dubln cty center and Manchester cty center, respectvely. n the large-scale characterzaton, devatons were more pronounced n the Dubln measurements, whch were of the order of %, whereas the measurements from Manchester had devatons on the order of %. The results also show the effect of the bandwdth of analyss of the computed parameters. These effects are for the frst tme presented from the same set of measurements. 1. ntroducton t s now accepted that the measurement of the mpulse response of the moble rado channel and the subsequent estmaton of the tme delay descrptorsuch as average delay, delay spread, and varous wdths of the power delay profle provde the necessary nformaton to assess the desgn of varous commumcaton systems. To ths end, varous wdeband channel sounders were developed and used for the characterzaton of the moble rado channel [De Lange, 1952; Cox, 1972; Bajwa, 1979 ; Devarsrvatham, 1986 ; Fannn et al., 1991; Herman et al, 1990; Nche et al., 1993]. Soundng measurements were carred out at varous frequences and Copyrght 1999 by the Amercan Geophyscal Unon. Paper number 1999RS /99/1999RS wth dfferent bandwdths. Most of the outdoor soun- dng measurements reported to date used bandwdths whch vared from a few klohertz to a few tens of me- gahertz. These measurements were manly camed out usng pseudo random bnary sequence (PRBS) sounders wth fxed bandwdths [Cox, 1972; Bajwa, 1979; Devarsrvatham, 1986; Fannn et al., 1991; Herman et al., 1990; Nche et al., 1993] or chrp sounders whch employed a matche detector at the recever [Baley and Summers, 1990; Schwarz and Schussler, 1993; Lovnes et al., 1993]. Although n prncple the PRBS sounder and the chrp sounder should provde smlar nformaton, the chrp waveform s characterzed by a lnear frequency relatonshp wth tme whch can be utlzed to provde the channel descrptors as a functon of frequency over the swept bandwdth. Ths can be easly acheved by employng a heterodyne detector at the recever, whch preserves ths lnear relatonshp between tme and frequency. The advantages of such a technque have long 797

2 798 SALOUS: WDEBAND 1.8-GHZ MULTPATH STATSTCS been recognzed m HF mdars and HF sounders [Salous, of resolvng multpath components whch are separated 1986]. However, ts applcaton to moble rado soun- by (l/b) s. Practcally, ths resoluton s reduced by the ders has not, to the author's knowledge, been reported wndow functon, whch s used to reduce the sdelobes prevously. of the compressed pulse. Snce m a tme-varyng envronment t s not always Alternatvely, the resoluton of multpath can be possble to repeat the measurements for comparatve acheved by frst mxng the receved chrp sgnal wth a purposes, t s desrable to employ a soundng technque delayed replca at the recever (.e., heterodyne detecwhch provdes smultaneous measurements of dfferent ton) and low-pass flterng the mxed sgnal. Ths probandwdths and tme delay profles versus frequency. duces a beat note for each multpath component whch To study these effects, a fully programable wdeband s proportonal both to the tme delay between the recechrp channel sounder whch employs a heterodyne de- ved sgnal and the stored replca and the sweep rate, tector was developed at the Umversty of Manchester whch s equal to (B/T) Hz/s, where T s the duraton of nsttute of Scence and Technology (UMST). The the chrp pulse. The resoluton of multpath s subsesounder was used to conduct measurements at 1.8 GHz quently acheved by spectral analyss. Snce the output n the cty center of Dubln (reland) usng a 72-MHz of the detector s not compressed m tme, t can be lbandwdth and m the cty center of Manchester (En- nearly mapped to frequency. Hence, by dvdng T (or gland) usng a 90-MHz bandwdth. Subsequently, the B) nto sectons whose length s determned by the redgtzed data were analyzed to provde the channel des- qured frequency and tme resoluton, the multpath crptors for dfferent bandwdths of analyss and to ob- structure as a functon of frequency can be obtaned stan the mpulse response of the wdeband channel ver- multaneously. The maxmum unambguous Doppler sus frequency. frequency measured by a chrp sounder s equal to Ths paper brefly descrbes the chrp sounder and ds- +(1/2T) Hz. cusses the processng stages necessary to obtan the po- To utlze the advantages of the chrp technque, a wwer delay profles versus frequency. These profles de- deband fully programable chrp sounder was developed monstrate the frequency selectve fadng experenced by at UMST [Salous et al., 1998]]. The sounder uses stateeach resolvable multpath component. The effect of ths of-the-an dgtal frequency synthess, whch enables the selectvty on the estmated channel parameters whch generaton of chrp pulses wth hgh sweep rates. t s nclude the average delay, rms delay spread, and the- capable of producng bandwdths of up to 90 MHz 10-, -15-, and-20-db wdths of the profles s studed (deal spatal resoluton s 3.3 m) between 1755 and for dfferent bandwdths whch vared from 18 MHz to 1845 MHz wth programable repetton rates, whch vathe full bandwdth of the soundng waveform. The ry from tens to thousands of hertz. Ths permts chanparameters were calculated to detennne both the small- gng the waveform parameters to sut the envronment. scale and large-scale characterzaton of the channel. For outdoor measurements, hgh vehcle speeds are ex- Ths paper presents the results and enumerates the pected, and hence the tme varablty of the channel s varous channel descrptors as a functon of frequency consderably hgher than the ndoor channel. Therefore and bandwdth for the two measurements of the urban hgh waveform repetton rates are requred for outdoor envronments of Dubln and Manchester. measurements to accommodate the hgh Doppler frequences. Assumng 100 km/h vehcle speeds at The Chrp Sounder GHz career frequency, a waveform repetton rate of 500 Hz would be adequate to produce three samples per A chrp sounder transmts a contnuous waveform wavelength. Smlarly, for ndoor measurements, repetwhch s characterzed by a lnear frequency modul- ton rates of about 45 Hz would produce a smlar numaton. The frequency versus tme functon can be ether ber of samples for a 2.5 m/s walkng speed. ncreasng or decreasng. At the recever, the chrp pulse The recever of the sounder s based on the heterodyne s compressed by the use of ether a matched flter or a detector and employs two low-pass flters wth 75-kHz heterodyne detector. The matched flter produces a and 150-kHz bandwdths. Dependng on the waveform snx/x functon whch has a 4-dB wdth equal to 1/B s parameters, the recever can accommodate dfferent (B s the swept bandwdth) and hgh sdelobes (the frst tme delay wndows. For example, the tme delay wnsdelobe level s only 13.5 db down from the man dows of the sounder correspond to 8.3 gs or 16.6 gs for lobe). Therefore a chrp sounder s theoretcally capable a 90-MHz soundng bandwdth at 100-Hz repetton

3 SALOUS: WDEBAND 1.8-GHZ MULTPA STATSTCS 799 rate. To study the varous parameters and ther varatons wth tme and across the swept bandwdth, the fltered output s dgtzed and stored by transferrng t to a laptop for off-lne processng. The data are acqured n blocks of ether one sweep or 62 sweeps per rm. The sngle-sweep mode s used to montor the sgnal onlne and to adjust the receved sgnal level. The gan s set manually and s kept constant per stored rm. A block dagram of the sounder s shown n Fgure la. 'T T-X BLOCK DAGRAM LOCALLY GENERATED CHRP L_ R-X BLOCK DAGRAM The transmtter uses a 1-W power amplfer, and the transmt and receve antennas are omndrectonal wth vertcal polarzaton. The theoretcal senstvty of the recever can be calculated as follows: Assumng a nose temperature Tn of 300øK, a 90-MHz sgnal bandwdth, and an RF nose factor of 10 db, the thermal nose Nr s gven by Nr = 101og(kT, B)+10= dbm. For a 90-MHz sgnal repeated at 250 Hz, the processng gan (101og(BT)) s equal to db. Hence, for mpulse responses wth a sgnal-to-nose rato of the order of 30 db, the receved power should be of the order of dbm. The sounder's performance was verfed by carryng out back-to-back tests wth and wthout nterference. The nterference-free back-to-back test gave a sgnal-to-nose rato of the order of 25, 30, and 40 db for RF sgnal nput levels of - 115, - 110, and dbm, respectvely. For a 40-dB sgnal-to-nose rato the back-to-back test [Salous et al., 1998] shows that the sounder has a spurous free dynamc range of 38 db. To characterze the sounder's performance n the presence of nterference, the back-to-back test was also repeated by addng a CW sgnal n the 90-MHz band of the sounder. The CW nterferng level was db hgher than the level of the wanted sgnal, whch was at about -86 dbm. Apart from a very short.duraton recovery nterval, the tme doman sgnal dd not reveal any blockng of the recever, as can be seen from Fgure lb. To verfy the effect of nterference on the processed data, the back-to-back test n the presence of nterference was processed as outlned n secton 4, where b ' T o ? f -, ' ; o f equency MHz ffequency/mhz delay/mcsec Fgure 1. (a) Block dagram of sounder. (b) Output of sounder n the presence of nterferece for a 90-MHz sweep. (c) Response across the 90 MHz bandwdth n the presence of nterference.

4 800 SALOUS: WDEBAND 1.8-GHZ MULTPATH STATSTCS the 90-MHz sweep was dvded nto 18-MHz sectons buldng. The recever was drven around the cty center and 75% overlap; the resultng specmun s shown n Fgure l c. nterference s seen to rase the nose floor and to cause a slght dp, of the order of 2 db (relatve to the neghborng sectons), only n the frequency secton where t occurred. n a van whose route extended to a radus of about 900 m, as shown n Fgure 2b. The measurement area n Dubln was characterzed by open felds on the campus of the college and buldngs of a few storeys on the campus and around the college; 3. Measurement Envronment the measurement area n Manchester s manly characterzed by hgh-rse buldngs, a nearby tran staton, and a metro lnk. The cty cemer streets n both ctes vary between narrow and wde streets, whch are heavly The measurements were carred out n two ctes, ntally n Dubln and subsequently n Manchester. The measurements n Dubln were carred out n the vcnty of Trnty College Dubln, where the transmtter was placed on the roof of a three-storey buldng and the recever was n a van drven around the campus and the permeter of the college, as shown n Fgure 2a. The measurements n Manchester were carred out by congested. The measurements n Dubln were camed out wth a 72-MHz chrp sgnal repeated at 250 Hz, and the measurements n Manchester cty center were camed out wth a 90-MHz bandwdth at 250-Hz repetton rate. The use of the 150-kHz flter gave a tme delay wndow equal to 8.7 and 6.6 [rs, respectvely. Note that n both sets of measurements, the area covered dd not extend to placng the transmtter on the roof of an eght-storey the sounder's full range due to operatonal lmtatons. a 1330 m 1400 m Fgure 2. Maps of measurements areas of (a) Dubln and (b) Manchester.

5 SALOUS: WDEBAND 1.8-GHZ MULTPATH STATSTCS 801 b CCADLLY 1720 m 2200 m Fgure 2. (contnued) 4. Data Processng: General Consderatons Wdeband soundng s used to generate the tmevaryng mpulse response of the channel. n ths study ths s obtaned by processng the dgtzed data off-lne usng the fast Fourer transform (FFT). Pror to the FFT, the data were weghted by the Hammng wndow to reduce the effects of spectraleakage. The Hammng wndow has a -43-dB frst sdelobe level, whch s adequate for the 38 db dynamc range of the sounder. t also has a 6 db wdth whch s equal to 1.8 dscrete Fourer transform (DFT) bns. Ths has the adverse effect of reducng the resoluton of the sounder. For example, a 90- MHz sgnal deally has an 11-ns tme delay resoluton whch s reduced to 19.8 ns. The FFT analyss provdes an estmate of the nstantaneous mpulse response of the channel. The overall processng flow graph s shown n Fgure 3. The followng dscusson outlnes the dfferera stages. Apart from the choce of the wndow functon used n the FFT processng, several other aspects need to be consdered n the processng of the acqured data. These nclude (1) the fdelty of the data, (2) the requred parameters and ther estmaton, and (3) presentaton of the results The Fdelty of the Data There are two aspects n checkng the fdelty of the data. The frst s to check the mw data pror to any processng to dentfy whether the data are corrupted by the presence of nterference and/or mpulsve nose. The effect of the varous sources of nterference s to rase the

6 802 SALOUS: WDEBAND 1.8-GHZ MULTPATH STATSTCS Read data Clp nterference FFT and check SNR Compute channel parameters Calculate CDF's Fgure 3. Flow graph of processng. nose floor of the spectrally analyzed data and hence to reduce the overall dynamc range of the measure-ment. The data obtaned n Manchester cty center were characterzed by hgh nterference caused by other users of the channel and by mpulsve nterference caused by the metro lnk. To reduce these effects, a technque smlar to that used by $alous [1997] was employed, whereby the level of nterference was clpped to a level hgher than or equal to the fadng envelope of the receved sgnal. Ths s carred out n stage 2 of Fgure 3. The data were also verfed to ensure that the presence of nterference dd not block the recever. Fgure 4 shows four examples of four dgtzed sweeps n Manchester cty center, where nterference was detected n the data at about 1810 and 1820 MHz. Note that the upper two sweeps and the lower two sweeps shown n Fgure 4 are from dfferent runs and that the sweeps are noncontguous and are only dsplayed to demonstrate the dfferent fadng envelopes detected n the measured data. t can been seen from the Fgure that the nterference appears at a level hgher than the fadng envelope of the receved sgnal and that the deep fades detected n Fgures 4a and 4b (upper two sweeps) occurred at lower frequences than the nterference frequences. n addton, Fgures 4c and 4d (lower two sweeps) show that such fades were not detected n the data despte the presence of nterference. The hgh level of nterference would affect the sgnal-to-nose rato (SNR) n the frequency secton where t occurred (Fgure l c). Thus the deep fades whch were observed n Fgure 4 (upper two sweeps) cannot be attrbuted to nterference snce the nterference occurred at a hgher frequency. Clppng the level of nterference to a value hgher than or equal to the fadng envelope of the receved sgnal mproves the SNR [Salous, 1997]. The other aspect that needs to be determned s the valdty of the mpulse response. Usually, a number of consecutve mpulse responses, N, are averaged to produce the average power delay profle, whch s an estmate of the receved power versus tme delay. n general, averagng assumes wde-sense statonarty over N and s usually used to mprove the sgnal-to-nose rato of the power delay profle. Snce the tme of Cox [1972] t has been accepted to average over a 5-to 30-m secton of the road. However, snce the measurements carred out by Cox [1972] used a 10-M-z sldng tme cross correlator whch was nherently slow and were wth only a 30-m resoluton, the number of mpulse responses that are averaged can be chosen accordng to the soundng parameters. n the present study, the data were acqured n blocks of 62 sweeps. Assumng a drvng speed of 10 kn/h, averagng over the 62 sweeps corresponds to about 1.1 rn of the road, whch, for the 90- MHz bandwdth, s smaller than the 3.3-m deal resoluton of the sounder. Ths restfts n a dynamc range of about 38 db, whch s adequate to dentfy hghly attenuated echoes. CCR [1990] recommends a 3-dB nose/spur safety margn and a mnmum value of peak-to-nose/spur rato before a profle s accepted. n the present study, the back-to-back test carred out on

7 SALOUS: WDEBAND 1.8-GHZ MULTPATH STATSTCS 803 (a) (b) ½) (d) frequency/mhz frequency/mhz Fgure 4 Four dfferent sweeps demonstratng the fadng of the receved sgnal envelope as a functon of frequency. the sounder gave at least 38 db peak4o-spur level [Salous et al., 1998]. Hence the crteron for acceptance of the profle was chosen to be between 25 and 30 db. Ths s camed out n stage 3 of Fgure 3, where a FFr over the whole sweep s camed out at ths stage to dentfy the sutable runs The Requred Parameters and Ther Estmaton The parameters that are usually determned from the average power delay profle are the average delay rz) and the rms delay spread S, whch are the frst and second moments of the power delay profle, respectvely. These parameters are obtaned from the followng equatons: t 1 ( - %)p( )d to 'to = t, -% (1) p( )d to tl (x - D ) 2 p(x)dx tl p( )d to where P( ) s the average power as a functon of tme delay,, and to s chosen before the arrval tme of the frst receved echo, ra such that the magntude of the power profle s nsgnfcantly small. Here t s the last tme that the average power delay profle crosses a partcular threshold level, whch s chosen above the nose floor of the system. Other parameters that are measured from the power delay profle are the wdths of the profle whch exceed a partcular threshold level below the peak. n ths study the -10-, -15-, and -20-dB wdths of the profles are calculated. To determne the above parameters as a functon of frequency and bandwdth, power delay profles can be (2)

8 ,,, 804 SALOUS: WlDEB 1.8-GHZ MULTPATH STATSTCS obtaned ether over the whole swept bandwdth or by dvdng each sweep nto a number of sectons and performng the FFT over these sectons. By dvdng each sweep nto M sectons, the resoluton of the sounder s reduced by M. Hence t s necessary to choose a sutable length for the secton such that adequate multpath resoluton s stll acheved and a sutable number of sectons across B s obtaned to study the frequency selecwdths, 13 and 17 mpulse responses per sweep, respectvely, are obtaned. Subsequently, the chosen power delay profles are processed to estmate the channel descrptors outlned above. n the data collected n Manchester, fade depths of the order of 27 db were observed n certan frequency ranges. To account for such deep fades, two dfferent sets of threshold levels were used. The frst set s used tve fadng. To reconcle these contrary requrements, on a run bass where all the 17 power delay profles n a the sectons can be overlapped. The use of overlap pro- partcular nm are normalzed wth respecto the same cessng provdes the followng advantages: (1) t prov- level whch s set to the maxmum value detected des more mpulse responses over B, (2) t reveals an event whch mght occur on the edges of the secton and whch mght get masked by the use of the wndow among the 17 profles (see Fgure 5a). Ths ensures that when the sgnalevel fades due to propagaton, the power delay profle n the partcular nm s not rejected acfuncton, and (3) t provdes adequate sgnal-to-nose cordng to the mramum sgnal-to-nose crteron (see rato (SNR), whch s determned by the length of the secton. n ths study, to compare the results from the two sets of measurements, each sweep s dvded nto 18-MHz sectons wth 75% overlap. Ths provdes a tme delay resoluton equal to 100 ns, whch s com- Fgure 5b). To accommodate the deep fades encountered n the Manchester data, threshold levels were set at - 25, -30, and-35 db wth respecto the normalzed run. When ths set s used, only the average delay and the rms delay spreads are estmated snce the wdth paramensmate wth the 18-MHz bandwdth wth the Ham- meters assume a locally normalzed value,.e., per secmng weghtng and provdes an mpulse responsevery 4.5 MHz. Therefore, for the 72- and 90-MHz bandton. A second set of thresholds s set locally wth respect to each secton, and the parameters are computed o ø MHz ' ' o ø -2o MHz - / r ' / / O MHz 1836 MHz m -10 f o ø -2o f - o excess delay/mcsec excess delay/mcsec Fgure 5. (a) Four frequency sectons normalzed wth respect to the peak of the run. (b) Four frequency sectons of Fgure 5a each normalzed wth respecto ts peak.

9 SALOUS' WDEBAND 1.8-GHZ MULTPATH STATSTCS 805 o m MHz ß MHz ' ; MHz -20 1' '1 '1-3O ' r' ' - r 1836 MHz b ),' ' ' m -10-2o, _ / excess delay/mcsec excess delay/mcsec Fgure 5. (contnued) agan ncludng the wdth parameters. For the data collected n Dubln, no sgnfcant nterference or deep fades (fades were less than 10 db) were detected. Therefore the parameter estmatons were carred out only wth local thresholds, whch were set at -25 and -30 db. To compare rte values obtaned wth an 18-MHz bandwdth and the full swept bandwdth, the varous parameters are also calculated for the full swept bandwdth. Also, the threshold levels were set at -25 and -30 db for both sets of data Presentaton of the Results n the estmate of the tme descrptors, the results are usually presemed n two forms, whch are referred to as small-scale and large-scale characterzaton. The former s determned over a few wavelengths (.e., one run), whereas the latter s found by calculatng the cumulatve dstrbutons over a number of locatons n the same en- vronment (.e., all the nms n a measurement) [Cox, 1972; Bajwa, 1979]. Subsequently, varous values of the dstrbuton are calculated, such as the medan (the 50%) value, the 90% value, and the 100% value of the dstrbuton. n the present study these parameters are determned for each frequency secton. The varatons across the swept bandwdth are then determned by computng the average of each estmated value over all the sectons and subsequently the standard devaton. The resultant mean s then compared wth the value obtaned over the whole sweep. These parameters are computed n the fnal stage of Fgure Results Both sets of data were processed as outlned n the prevous secton. Fgures 6 and 7 dsplay typcal power delay profles from Dubln cty center and Manchester cty center, respectvely. The profles were obtaned by dvdng each sweep nto 18-MHz sectons wth 75% overlap, whch gves an mpulse response every 4.5 MHz, and each power delay profle was normalzed wth respecto the maxmum of the nm. Fgure 7a s the profle of the nm of Fgure 5. The fgures show the frequency selectve fadng of the dfferent multpath corn-

10 806 SALOUS: WDEBAND 1.8-GHZ MLE, TPATH STATSTCS o ,-'- (a) : -.. :',.?..., : '.L ':' ".. ;.', "' ' -" -o ' (b)...-.n- '-. m.-' "-- -! O[ j., _-, ' _,,-r ;...:,... &'.7L;,.:;,.;- ':;' ',' ' td '.0 ",[ :: ' - --.;:':' ;:':' %': -b:. :--.'.'d-?/: -3O o -20-3O freq/mhz 1850 delaytmcsec freq/mhz 1850 del@y/mcsec (c) "" '1- --' T ' F; ---:--, T ', - -:- / : ---" : '- ' f - ;'-' :',?.½ m - '-,' '. :-'.'.:..-" :..:...':;:::':. :-..': * ß,.., /,,-. ß :.:. ".-.:- :Y'-z.¾.._.. ;. +-": '.'.?...:, ;M' ¾,' ¾-.'t.'.. o-20 _..... '--- ' --- m, ,,,,_L_,_.,,. _ 5 freq/mhz 1850 delaytmcsec freq/mhz 1850 delay/mcsec ( ) 0 "" : 1: >!:.' ß.:-. /._ a'.l_-.",', ' ': -:. :': '.,...-..':..'2:'.' ".:::.5,.:. / : _-;:-::-?t':z,: %.-":,.,_..'-;;;:.+:4,. : a:..,.-,,:.,,,.:.;., : :.½.' '!g{.,_ , ::', :-:!"::::: :..? _..:::.,:'4->;:.:..':.::_-z '-'-.-:' Fgure 6. Average power delay profles as a functon of frequency for four dfferent runs n Dubln. ponems. Comparng the runs dsplayed n Fgures 6 and 7, t can be seen that fadng moves across frequency at both a shallow and a deep level. The runs of Fgures 6a- 6c are separated by approxmately 4.4 m, whereas the runs of Fgure 7 are spread around an area of 500 m. The fgures thus demonstrate the extent of frequency selectve fadng n both a small area and a relatvely larger area. n ths secton, the results for the small-scale and large-scale characterzaton for both sets of data are presented Small-Scale Characterzaton Tables 1 and 2 gve the estmated average tme delay and rms delay spread as a functon of frequency for the four rms of Fgures 6 and 7, respectvely. The frequency columns n the tables gve the center frequency of the analyzed 18-MHz sectons, whch are separated by 4.5 MHz. The Tables show that the devatons for the rms delay spreads are generally smaller than the correspondng average tme delay devatons. However, sgnf- -3O '." : --"' 7, ::k:k,,.' (a) (b),- r' -', ' "'-- :-,, : ---' o [ ß., _- ½.. '-:'..>.'-- 9./;.::,:.:.. < ',, :e, >.:...,. - -: _:.-..;:,.,, }, m -10 -"...,::.,,'; -.,,.,.,' e:.'.:,.'.', ' ' ; :,2. '-.5,½..:':.:.::: x '--.,.:.''?;.,: ' --,':-'--'.,--/.--'-...,:..--'.. <½ ,:...,.',, -: ,-:,.,.::!. 4½ ,- --:--...: ' ':' t,,- -;:*:-:.' 4,:2 ½..½' ":';:' -':' - ',- o O _ ',_-'--_'-::-::!:::':!::!::.:;; : :: ---'-' '- fre q/m H z 1850 (c) dela y/mcs e c fre q/m H z 1850 (d) delay/mcsec rn A -2o -3O Fgure 7. Average power delay profles as a functon of frequency for four dfferent runs n Manchester.

11 SALOUS: WDEBAND 1.8-GHZ MULTPATH STATSTCS 807 Table 1. Average and RMS Delay Spreads for Runs A-D of Fgure 5 Wth -25-dB Threshold Where the Power Profles Each Secton Are Normalzed Wth Respect to ts Own Peak Delay, Run A Run B Run C Run D Frequency, MHz RMS Average RMS Average RMS Average RMS Average Mean, rn Standard devaton, ( /m) x 100% 18.1% 31.7% 30.5% 51.4% 22.7% 26.4% 14.2% 11.9% Table 2. Average and RMS Delay Spreads for Runs A-D of Fgure 6 wth -25-dB Threshold Where All the Power Profles n Each Run Are Normalzed Wth Respect to the Peak of the Run Delay, Run A Run B Run C Run D RMS Average RMS Average RMS Average RMS Average Frequency, MHz Mean, rn Standard Devaton, ( /m) x 100% 29.7% 42.7% 7.9% 18.3% 14.6% 23.8% 15.% 23.3%

12 808 SALOUS: WDEBAND 1.8-GHZ MULTPATH STATSTCS cant devatons are observed, where run A of Table 2 and nm B of Table 1, for example, have devatons between % and %, respectvely. The varatons over dstance can be observed from consecutve runs A-C of Table 1, whch are separated by 4.4 m assumng a 10 km/h vehcle speed. The mean of the average delay and the mean of the rms delay spread vared by 166 and 90 ns from run A to run C, respectvely. The effect of the dfferent threshold levels s shown n Fgure 8, whch dsplays the estmated parameters for run A of Fgure 7 for the-25-db threshold levels, where the sold lnes correspond to the case when all the sectons are normalzed wth respecto the peak of the run (as n Fgure 5a). The lnes wth crcles correspond to the locally normalzed profles as shown n Fgure 5b. All the other threshold levels of-30 and -35 db gave smlar values to the lnes wth crcles and hence are not reproduced here. From Fgure 8, the varatons across the swept bandwdth are evdent regardless of the threshold crteron. Dfferences n the estmate appear around the frequency sectons where deep fades occur. Ths can be seen from Fgures 5a and 5b, whch show the two dfferent threshold crtera. The effect s pronounced n the estmate of the -15- and -20-dB wdths shown n Fgure 8c where both estmates are the same. Ths ndcates that the SNR n these frequency sectons s between 10 and 15 db as shown n Fgures 5a and 5b. cant. The table also contans the values for the 72-MHz sweep and the mean and standard devaton for fve sectons obtaned by dvdng the 72-MHz sweeps nto 36-MHz sectons wth 75% overlap. The table therefore also shows the effect of the processng bandwdth on the estmated parameters. From the table, t can be observed that the values for the 72-MHz bandwdth are generally smaller than those obtaned from the 18-or 36-MHz sectons. The-30-dB threshold gave smlar results for the mean of the estmated parameters and hence s not produced here Manchester cty center. The results for the Manchester data were smlarly analyzed, and the results for the -25-dB threshold for the dfferent parameters are presented n Table 4. The restfts for the-30- and-35- db thresholds were smlarly analyzed, and no sgnfcam dfference n the percentage devatons was detec- ted. Table 4 shows that, on average, there s no sgnfcant dfference between the dfferent parameters for the 17 frequency sectons, the largest devaton beng for the - 20-dB wdths, whch s only about 14.2%. Ths s n contrast to Table 3, whch shows sgnfcant devaton for all the parameters and a consstent dfference m the frst two frequency sectons. Comments can be made regardng the estmates obtaned wth the wder bandwdth that are smlar to those made for the measure- ments n Dubln Large-Scale Characterzaton 5.3. Dscusson The cumulatve dstrbutons across the 13 sectons for the Dubln data and the 17 sectons for the Manchester The results presemed n ths paper are a prelmnary data were also computed and compared aganst the 72- study of the frequency varatons of the mpulse resand the 90-MHz values, respectvely. The valdated ponse and ts tme delay descrptors across a 72- to 90- number of nms was 33 and 54 for the Dubln and Man- MHz bandwdth. Analyss of the measurements obtachester measurements, respectvely. These correspond ned n two European cty centers revealed two dfferent to 2046 and 3348 mpulse responses, respectvely. types of frequency selectve fadng. These wll be clas Dubln cty center. The results for the Dubln sfted here as ntermodal frequency fadng and ntramodata for the -25-dB threshold are presented n Table 3 dal frequency fadng. n onospherc prop-agaton, nfor the medan and the 100% value of the CDFs. Fgure tramodal frequency fadng refers to the merference of 9 dsplays the CDFs for the rms delay spread, average the ordnary and the extraordnary waves, and ntermodelay, and -10- and -20-dB wdths for the frst four fre- dal fadng usually refers to the fadng envelope of the quency sectons. Fgure 9 and Table 3 show that the frst receved sgnal, whch results from the nterference two frequency sectons have values whch are sgnf- between all the modes. cantly greater than the rest of the sectons and whch n the present study the above types of fadng can be converge to comparable values. The mean and the stan- related to the receved sgnal n the followng manner. dard devatons are also ndcated n Table 3. The table Assumng a transmtted sgnal s(o, the receved sgnal shows that varatons for the medan value are hgher Sr(t) Can be represented by the sum of delayed and attethan those for the 100% value, but these are stll sgnf- nuated replcas of s(t ).e.

13 SALOUS: WDEBAND 1.8-GHZ MULTPATH STATSTCS ' 0.6 O frequency/mhz b 0.8,,,,,, ' ; ' frequency/mhz 3.5,,, 3 :: ,,, ' ', r... -r r..., --, frequency/mhz Fgure 8. (a) Average delay versus frequency for run of Fgure 7a. Sold lne ndcates the -25-dB threshold, and sold lne wth crcles ndcates the -30-dB threshold. (b) The rms delay spread versus frequency for run of Fgure 7a. Sold lne ndcates the -25-dB threshold, and sold lne wth crcles ndcates the -30-dB threshold. (c) Sold lne ndcates the -10 db, sold lne wth crcles ndcates the -15 db, sold lne wth astersks ndcates the -20-dB wdths as a functon of frequency for run of Fgure 7a.

14 810 SALOUS' WDEBAND 1.8-GHZ MULTPATH STATSTCS Table 3. Tme Doman Descrptors for Data from Dubln Cty Center as a Functon of Frequency for -25-dB Threshold RMS Delay, gs Average Delay, ts -10 db Wdth, gs -15 db Wdth, gs -20 db Wdth, ps 50% 100% 50% 100% 50% 100% 50% 100% 50% 100% Frequency, MHz MHz Mean, m Standard devaton, o (o/m) x 100%, 50.8% 31.4% 56.5% 27.9% 58.5% 35.5% 65% 33.4% 47.6% 19.2% 18 MHz 36 MHz m o (o/m) x 100% 18.6% 6.78% 11.3% 6.35% 10.6% 15.8% 7.5% 10.6% 17.8% 10% 36 MHz 72 MHz The mean, standardevaton, and percentage of varaton are ndcated for analyss bandwdths of 18 and 36 MHz. The bottom row gves the correspondng values for the 72-MHz bandwdth. Sr(O = S(t-q ) <<N (3) between the two rays would produce maxma and mnma as a functon of frequency and tme. At a partcu- The receved sgnal can be broken down nto several sums, where each sum term s a "mode", and t represents the multpath components wthn a tme delay nterval whch corresponds to the tme delay resoluton of the sounder. Hence the receved sgnal s lar frequency and nstant n tme, the two rays would produce a resultant envelope, whch depends on ther relatve ampltude and ther dfferental tme delay. As the frequency of the sgnal s changed, the relatve phase shft between the two rays would change producng Sr(t) = zll s(t- 'l) + 2 s(t- '2) + zls s(t- s) + (4) maxma and mnma. The frequency separaton Af between consecutve maxma and mmma occurs when the phase shft 2xAfA s equal to x, where A s the dfwhere < <N, 1_<2 <N2, 1_<3_<N3,..., and N = N + N2 + N3 + N n (4), the components n each sum term ferental tme delay between the two rays. The depth of fade depends on the relatve ampltude of the two rays would nterfere ether constmctvely or destructvely and hence can vary from shallow fades to complete candependng on ther dfferental tme delay. Hence ths type of fadng corresponds to the ntramodal fadng. Characterzaton of ntramodal frequency selectve facelaton when they are of equal ampltude. n the present study, Axm x s equal to 55.5 ns (equvalent spatal resoluton of 16.6 m). Ths s the deal resoluton of the dng was obtaned by analyss of the mpulse response sounder, whch corresponds to the 18-MHz analyss wth the 18-MHz bandwdth and wth an mpulse response every 4.5 MHz. Assumng a smple two-ray mobandwdth chosen n ths study to characterze ntramodal frequency fadng. Thus the mnmum separaton del to characterze ths type of fadng, the nterference between consecutve maxma and mmma, Afrore, s

15 SALOUS: WDEBAND 1.8-GHZ MULTPATH STATSTCS 811 loo 100 ß ' 6o '... loo 6o 1 rms delay/mcsec,! 0 ' ' average delay/mcsec 100 ;, 4!! so...,..., 1.6 o o db wdth/mcsec db wdth/mcsec Fgure 9. The rms delay spread and average delay for the frst four frequency sectons of the Dubln measurements. Sold lnes ndcate 1773 MHz, sold lnes wth crosses ndcate MHz, sold lnes wth astersks ndcate 1782 MHz, sold lnes wth crcles ndcate MHz. equal to 9 MHz. For a partcular mode not to experence ronment of the measurement. n Dubln, the measureany fadng wthn the 90-MHz bandwdth of the soun- meres were taken nsde and around the permeter of der, the frequency separaton between mnma should Trnty College, whch had two- to four-storey buldngs exceed 180 MHz, whch corresponds to a dfferental tme delay A- of 5.5 ns. ntramodal fadng can be observed from Fgures 6 and 7, and ts effect on the tme deand open felds. These measuremeres were manly characterzed by the recepton of the lne of sght. n contrast, Manchester cty center has manly hgh-rse bullay descrptors was presented n Tables 1-4. For certan dngs, and the tme delay spreads were generally smaller locatons the fadng of some of the modes was slow, and for other locatons, fadng of the order of db was detected. n addton, whle some of the modes experenced one mnmum across the bandwdth, others experenced two or more. The effect of ntramodal fadng gave sgnfcant dfferences n the small-scale descrptors n both cty centers. However, the large-scale varatons showed sgnfcant devatons wth frequency of than those for Dubln cty center and manly were nonlne-of-sght. Furthermore, the range of the measurements n Manchester cty center extended up to 900 m, whereas n Dubln cty center t was lmted to m. The other type of fadng that was observed n the data s the fadng of the overall envelope of the receved sgnal when the multpath componems cannot be resolthe order of % for Dubln cty center, whereas ved. Ths type of fadng can be related ether to (3) or these devatons were only of the order of % n (4). n (3) t refers to the fadng that results from the n- Manchester cty center. Closer nspecton of the mpulse terrefence between the dfferent componems, whch are responses showed that for the data obtaned n Dubln, not resolved by a CW sgnal; n (4), t refers to the nterfar-away echoes (4-7 ps) faded at hgher frequences, ference between the dfferent modes (hence ntermodal whch resulted n hgher values for the channel descrp- nterference). ntermodal frequency selectve fadng tors for the lowest two frequency sectons. However, for lmts the coherent bandwdth of the channel and affects the Manchester data, the fadng of the multpath components moved n frequency for dfferent rms, whch, the performance of communcaton systems, whch are not capable of dscrmnatng aganst the presence of on average, resulted n smlar values for the large-scale multpath. Ths type of fadng can be characterze dtme delay descrptors. Ths could be due to the env- rectly by envelope detecton of the demodulated sgnal.

16 ._ 812 SALOUS' WDEBAND 1.8-GttZ MULTPATH STATSTCS Table 4. Tme Doman Descrptors for Data from Manchester Cty Center as a Functon of Frequency for -25-dB Threshold RMS Delay, s Average Delay, gs -10 db Wdth, s -15 db Wdth, s -20 db Wdth, s 50% 100% 50% 100% 50% 100% 50% 100% 50% 100% Frequency, MHz Mean, m Standard devaton,c MHz c /m x 100% 5.9% 5.2% 9.3% 9.9% 10.5% 10.6% 6.8% 12.2% 9.3% 14.2% 18 MHz 90 MHz The mean, standar devaton, and percentage of varaton are ndcated for analyss bandwdths of 18 MHz. The bottom row gves the correspondng values for the 90-MHz bandwdth. Although ths type of fadng s not analyzed n ths study, t can be observed n Fgure 4, whch dsplays the overall receved sgnal as a functon of frequency for four sweeps. Fgure 4 shows that the overall frequency selectve fadng of the envelope vared from slow varatons wth dstnct mnma (Fgure 4, lower two sweeps) at certan frequences to flat fadng over certan frequency sectons (Fgure 4, upper two sweeps). Because of the tnfamlar nature of the flat fadng observed n Fgure 4 (upper two sweeps), the two sweeps were analyzed as prevously descrbed to observe the ntramodal fadng. The results for the sweep of Fgure 4a are shown n Fgure 10 for the frst 16 frequency sectons. Fgure 10 shows that the fade observed n Fgure 4a at about 1787 MHz occurred when the dfferent modes, n the frequency secton centered at 1791 MHz, faded n tandem. A smlar effect was observed for the sweep n Fgure 4b. To ensure that the flat fadng observed n the upper two sweeps of Fgure 4 was not due to the recever beng "blocked" by nterference from other users of the channel, an ndoor test was set up n the absence of nterference, wth a 90-MHz sgnal and a 250-Hz repetton rate. n the test t was observed, both on the osclloscope and from the dgtzed data, that such flat fadng occurred over certan frequency sectons when an obstructon was n the path between the transmtter and the recever. As the obstructon moved (both transmtter and recever are statonary), dfferent frequency sectons experenced dfferent fadng. The results of the test are shown n Fgure 11, where t s observed that flat fadng occurred over frequency sectons of the order of MHz and that ths type of fadng moved across the spectrum. Fgure 12 shows the correspondng ntramodal fadng for one of the acqured sweeps. Snce the test was carred out ndoors, when the data were analyzed wth 18-MHz sectons, the resultant was manly composed of a sngle mode. Hence both ntermodal fadng and ntramodal fadng should follow a smlar trend as can be observed from Fgure 12. The man dfference arses due to the method of analyss of ntramodal frequency fadng. Observng the envelope drectly gave more de-

17 SALOUS: WDEBAND 1.8-GttZ MT TPATH STATSTCS 813 (a) (b) o..? 'r f--- '._ o (c) (d) O ' ' ' -- ß T frequency/mhz o frequenc¾/mhz Fgure 10. Output of sounder for four 90-MHz sweeps n an ndoor envronment llustratng the occurrence of flat fadng due to the movement of an obstructon n the path between the transmtter and recever. tal, snce t comams 2000 samples, whle the curve obtaned from the ntramodal fadng had only 17 ponts across the 90 MHz. Although t s very dffcult to emulate the exact condtons of the measurements n Manchester cty cemer, Fgures 11 and 12 show the feasblty of the observed fadng phenomenon of Fgures 4-7. Ths can be confrmed by notng that the frequency sectons affected by mefference n Fgure 10 occurred at MHz,.e., after the 1791-MHz frequency secton where the frst deep fade was detected. ntramodal fadng s sgnfcant for broadband systems such as code dvson multple access (CDMA), whch rely on the resoluton of multpath to obtan the mproved performance of the Rake recever. Such deep fades would result n a reduced performance. Although no attempt s made here to evaluate any partcular communcaton system, the analyss technque of the present study can be appled to broadband CDMA whch s expected to operate wth a 5-MHz bandwdth and a 15- MHz allocaton per operator. 6. Conclusons Ths paper presems the results of wdeband measuremeres obtaned n the cty centers of two European ctes usng a chrp sounder. The measurements were analyzed wth dfferera tme delay resoluton to deter- mne the mpulse response of the channel as a functon of frequency. The dfferent tme delay descrptor such as the average tme delay, rms delay spread, and the - 10-,-15-, and-20-db wdths of the average power delay profle were subsequently estmated. These were computed ether for the whole swept bandwdth, whch was 72 or 90 MHz, or for 18-MHz sectons of the sweep 72 or 90 MHz, or for 18-MHz sectons of the sweep wth 75% overlap, whch gave an mpulse responsevery 4.5 MHz. The parameters were calculated for threshold levels between-35 and -25 db. Comparson of the results showed that, n general, the wder bandwdth gave a smaller estmate of the parameters. However, the 18- MHz bandwdth revealed the frequency dependence of these parameters and showed that sgnfcant devatons can occur both at the small-scale and the large-scale levels. Ths comparson obtaned for the frst tme from the same set of measurements. The results also showed that severe flat fadng occurs across a wde secton of the swept bandwdth. Although ths severe fadng was only detected n a few runs wthn a 500-m area, t s nterestng to see the movement of ths type of fadng across the spectrum. An attempto explan ths flat fadng phenomenon that was observed n the measurements was made by settng an ndoor test and movng an obstructon n the path between the

18 SALOUS' WDEBAND 1.8-GtZ MULTPATH STATSTCS 1764 o , o o : l [' 11 'l " ' o 0 'l ll q ' --F l _ ' , O -30 ', :, ' _ O -lo -4.; 'r -10-2O B27-3O,', tme delay/mcrosec, -:30 Fgure 11. Multpath structure for the frst 16-MHz sectons for sweep shown n Fgure 4.

19 _ SALOUS' WDEBAND 1.8-GHZ MULTPATH STATSTCS O frequency/mhz -'"' _ ', ' L - 'r :1- r... m m /, - -r' "- - ;E t,e ',,-... " ,.L-.,- -,' -' '-,.,,, : e '..f.:',.'..-..4, ', rn. 10 -,,. " w -- -z -....,.,.,, -2...,,.-... :...,...,[ :'... :' :,,' _... ;.-: :-_,,. -:..,::,..:.,= =:..;f::..:.,.. - _, ', -'. ":',.'.' E?:':";' :".'- : '.' '...,...'- _, ß _EE_-_. -: '& ;' delay/mcrosec frequency/mhz Fgure!2. A sweep obtaned the ndoor test and ts co espondj 8 mu]qpath stnzcture. transmtter and the recever. n the outdoor envron- ment, a large vehcle movng n the vcnty of the moble can act as such an obstructon. Evdently, more measurements need to be taken and a more comprehensve model to analyze the dfferent types of fadng detected n the measurements needs to be developed. Despte the fact that the measurements presented here were carred out n urban envronments and are lmted to two small areas of the respectve ctes, the results ndcate the need for further studes of the frequency fadng phenomenon n dfferent envronments and ts effect on communcaton systems. Prevous studes of frequency effects were carred out at frequences whch were sgnfcantly separated, such as 488 versus 1280, versus 2920 MHz [Turn et al., 1972]. The other reported smultaneous measurements were over a 10- MHz bandwdth, where 195-kHz sectons of the bandwdth were analyzed to determne the frequency correlaton functon of the channel for Group Specal Moble (GSM) [Bc and Pajusco, 1995]. Comparatve bandwdth effects were camed out by Nelson [1978] wth a PRBS sounder usng 10-MHz and then 20-MHz bandwdths at 1370 MHz. n contrast, the presem paper presents the effects of frequency varatons of the mpulse response over a 72-to 90-MHz bandwdth and studes, from the same set of data, the effect of the bandwdth of the measurements. The results therefore can be drectly compared n both bandwdth and frequency effects. The processng and analyss of both sets of data showed that large devatons are possble. These conclusons are mportant snce most systems' parameters are usually estmated from MHz bandwdth soundngs usng a PRBS sounder whch does not reveal the effects of the type of frequency fadng reported n the present study. n addton, the present study shows that measurements wth conventonal sounders whch do not resolve the frequency effect on the measured parameters can gve an estmate of the channel parameters whch cannot be extrapolated across the allocated bandwdth of the partcular servce. Fnally, the type of results presented here demonstrate that chrp sounders whch employ the heterodyne detector can be a powerful tool for studyng the varous narrowband and wdeband propagaton effects. Acknowledgment. The author thanks S. Swords and the department of Electronc and Electrcal Engneerng at Trmty College Dubln for provdng the facltes for the measuremerts n Dubln. The assstance of R. Lewenz of BT Labs, Marresham, n obtanng the measurements n Manchester s gratefully acknowledged. Thanks are also due to Chu-Po Peng for clppng the nterference for the Manchester measure- ments. References Baley, R., and G. Summers, Rado channel charactersaton for the dgtal European cordless telecommuncatons systems, Br. Telecommun. Technol., 8 (1), 25-30, 1990.

20 816 SALOUS: WDEBAND 1.8-GHZ MULTPATH STATSTCS Bajwa, A.S., Wdeband characterzaton of UHF moble rado channels n urban areas, Ph.D. thess, Dep. of Electron. and Electr. Eng., Unv. of Brmngham, Brmngham, England, Bc, J.-C., and P. Pajusco, Restfts on frequency cross correlaton from wdeband measurements at 1.8 GHz, COST Rep. 231 TD(95), 6 pp., France Telecom CNET, Belfort, France, CCR, Propagaton data and predcton methods for the terrestral land moble servces usng the frequency range 30 M3lz to 3 GHz, Rep , nt. Telecommun. Unon, Geneva, Cox, D.C., Delay Doppler characterstcs of multpath propagaton at 910 MHz n a suburban moble rado envronment, EEE Trans. Antennas Propag., AP-20(5), , De Lange, O.E., Propagaton studes at mcrowave frequences by means of very short pulses, Bell Syst. Tech. d., 31, , January Devasrvatham, D.M.J., Tme delay spread and sgnal level measurements of 850 MHz rado waves n buldng envronments, EEE Trans. Antennas Propag., AP-34(11), , Fannn, P.C., A. Molna, S. Swords, and P. Ctflen, Dgtal sgnal processng technques apphed to moble rado channel soundng, EE Proc. Part F,! 38(5), , Hennmm, S., U. Martn, R. Reng, H. W. Schussler, and K. Schwartz, A devce for hgh precson real-tme channel measurement, Cost 231 TD (90), 7 pp., Unv. of Erlangen- Nurberg, Erlangen, Cmmnany, Lovnes, G., S. E. Pattsen, and R. H. Raekken, A mllmetre wave channel sounder based on the chrp/correlaton technque, lee Colloq., Dg.!993/233, 8/1-8/6, Nche, C., A.M. D. Turkman, and A. A. Arowojolu, Channel sounder for PCN networks, lee Colloq., Dg. 1993/233, 5/1-5/6, Nelson, D., Mcrowave propagaton measurements for moble dgtal rado applcaton, EEE Trans. Veh. Technol., FT- 27(3), , Salous, S., FMCW channel sounder wth dgtal processng for measurng the coherence of wdeband -F rado hnks, EE Proc. Part. F, 5( 133), , Salous, S., On the potental applcablty of autoregressve spectral estmaton to HF chrp sounders, d. Atmos. Sol. Terr. Phys., 59(15), , Salous, S., N. Nkandrou, and N. Bajj, Dgtal technques for moble rado chrp sounders, EE Proc. Cornrnun, 145(3), , Schwarz, U.M., and H. W. Schussler, Devces for propagaton measurement n moble rado channels, n Fourth nternatonal Symposum on Personal, ndoor and Moble Rado Cornrnuncat 'ons, pp , Yokohama, Japan, Sept. 8-11, Turn, G. L., F. D. Clapp, T. L. Johnston, S. B. Fne, and D. Lavry, A statstcal model for urban mtftpath propagaton, EEE Trans. Veh. Technol., VT-21(1), 1-9, S. Salous, Department of Electrcal Engneerng and Electroncs, Unversty of Manchester nsttute of Scence and Technology, P.O. Box 88, Manchester M60 1QD, England. (salous@umst. ac.uk) (Receved March 3, 1998; revsed January 26, 1999; accepted February 5, 1999.)