Wdeband ped Delay Lne Channel Model at 3.5GHz for Broadband Fxed Wreless Access system as functon of Subscrber Antenna heght n Suburban Envronment Cha Leong Hong, Ian J. Wassell, Georga E. Athanasadou, Steve Greaves 2, Malcolm Sellars 2 Laboratory for Communcaton Engneerng, Unversty of Cambrdge 2 Cambrdge Broadband Lmted, Unted Kngdom Abstract Ths paper presents the results of measurements made to characterse the 3.5GHz Broadband Fxed Wreless Access channel n a suburban envronment, usng a sectored base staton antenna and a drectonal subscrber antenna. A tapped delay lne channel mpulse response model of the Sngle-Input Sngle-Output (SISO) channel s derved from the measurements. Prevously, t has been reported that the delay spread decreases wth an ncrease n the heght of the Subscrber Unt (SU) antenna [9]. In ths paper, t s reported that the multpath tap also decreases wth an ncrease n the SU antenna heght. Furthermore, t s reported that a 3-tap channel model wth tap spacng of 2ns, and maxmum tap delay at 4ns can adequately descrbe the rado channel under nvestgaton. The narrowband Rcean and wdeband root mean squared delay spread are observed to correlate strongly wth excess path loss.. Introducton The fundamental lmt that s mposed on any wreless system s due to the rado channel. Propagaton and channel models are essental for the analyss and smulaton of wreless systems. Knowledge of the rado channel s essental to the development and deployment of a wreless system. Rado channel models are used to support research nto methods for mtgatng channel mparments, such as the desgn of the equalser and the choce of sngle-carrer or mult-carrer systems []. In addton, they also form an ntegral part of wreless system plannng and deployment. There s a large body of lterature concernng the moble rado channel, but publcatons concernng the Fxed Wreless Access (FWA) channel are stll very lmted [4],[5]. Prevous works have nvestgated the narrowband channel characterstcs of FWA systems, e.g. Crosby et al [2]. As the data rate of FWA systems ncreases, the effect of the wdeband channel plays an ncreasngly mportant role on system performance. It s therefore necessary to characterse the wdeband channel effects. Investgaton nto the effects of wdeband channels on FWA systems have been conducted n the 2.5 GHz frequency band by Porter et al [3] and Gans et al [4], n the.9 GHz frequency band by Erceg et al [5]; and at 3.5 GHz by Saud et al [6]. Models for path loss, Rcean and tapped delay lne channel mpulse response for Sngle-Input Sngle-Output (SISO) fxed wreless channel were reported n [7]. Models for Rcean, Cross Polarsaton Dscrmnaton and path loss at 2.5GHz for a 2x2 Multple-Input Multple- Output (MIMO) fxed wreless system were reported n [8]. Prevous lterature has reported that the drectonal subscrber antennas usually employed n FWA systems sgnfcantly reduce the delay spread of the channel compared wth that when an omn-drectonal antenna s used [3]. However the nfluence of SU antenna heght has not prevously been thoroughly nvestgated. Ths paper presents the results of propagaton measurements at 3.5 GHz for a SISO Broadband Fxed Wreless Access (BFWA) system wth a drectonal subscrber antenna and a sectored base staton antenna. In the prevous paper [9], we have reported the statstcs of the path loss and Root Mean Squared (RMS) delay spread wth respect to SU antenna heght. In ths paper, we report on Rcean, excess path loss and the wdeband tapped delay lne channel mpulse response model wth respect to SU antenna heght. Frst we descrbe the equpment and data processng methods used for extractng the channel mpulse response. Ths s followed by the results of wdeband channel measurements, presented n the form of the wdeband tapped delay lne channel models, wth respect to SU antenna heght. 2. Measurement Methods The measurements were conducted n the northern suburbs of Cambrdge, England, durng the summer months from June to September 22 wth trees n full folage. Ths area has a relatvely flat terran and few hgh rse buldngs, and covers an area of 9 km 2. The base staton (BS) s located at a heght of 5 m above ground level. The SU antenna s postoned on top of a retractable vehcle-mounted mast. Two sets of readngs are taken at each locaton separated by a horzontal dstance of about 2m. At each locaton, the bearng of the antenna s adjusted to pont n the drecton wth hghest receved power. Once the bearng s fxed, the heght of the SU antenna s subsequently vared between 4 to m n steps of m. At each heght, a total of delay profles and path loss measurement were collected over a perod of 3 seconds. A total of 54 sets of measurement have been collected, from 65 subscrber locatons wth BS to SU dstances rangng from 73 m to 3. km. The precse
locatons of the measurement stes are dentfed wth a GPS recever. The BS antenna and the SU antenna have half power beamwdths of 9 and 2 respectvely, and are both vertcally polarsed. The s of the BS and SU antennas are 2.5dB and 5.6dB respectvely. The bandwdth of the system s 5 MHz, gvng rse to a delay spread resoluton of 2ns. The length of the pseudo-random soundng sequence used s 28 symbols and the maxmum excess delay that can be measured wth the system s 2.8 µs. The soundng sequence s modulated usng QPSK. At the subscrber, the n-phase and quadrature components of the receved soundng burst are captured on the hard dsk of a computer for later processng. A post-processng method based on correlaton processng extracts the power delay profles and the RMS delay spread from the stored data [9]. 3. Data Processng Methods The processng method nvolves Fourer transformaton of the correlaton between soundng sequence and receved sgnal. The rado channel s often modelled as a lnear tmevarant flter wth mpulse response h(t,τ) or equvalently by ts frequency response H(f,t), where h(t,τ) and H(f,t) are a Fourer transform par, []. Wthout loss of generalty, consder a lnear tme-nvarant system that s charactersed by ts mpulse response h(τ), as shown n Fgure. nput x(t) Channel h( τ ) output y(t) Fgure. Lnear sngle nput/sngle output system The complex envelope of the channel output y(t) s the convoluton of the mpulse response h(τ) wth the complex envelope of the channel nput x(t),.e. y ( t) = x( t) h( τ ) = h( τ ) x( t τ ) dτ () The tme-nvarant mpulse response h(τ) s a specal case of the tme-varant mpulse response h(t,τ) f the unt mpulse response functon s ndependent of the tme an nput s appled,.e., h( t, τ ) = h( τ ) for - < t < (2) To estmate the channel mpulse response h(τ), the frst step s to cross-correlate the nput of the channel wth ts output, assumng that the nput to the channel,.e. the transmtted sgnal, s known. For jontly statonary stochastc processes x(t) and y(t), t can be shown [] that ther crosscorrelaton φ xy (τ) s related to the autocorrelaton of the nput φ xx (τ) as xy = xx ) φ ( τ ) h( α) φ ( τ α dα (3) whch s a convoluton ntegral. Snce convoluton n tme doman s equvalent to multplcaton n frequency doman, the relaton (3) becomes Φ ( f ) = Φ ( f ) H ( f ) (4) xy xx where Φ xy (f) denotes the Fourer transform of φ xy (τ), Φ xx (f) denotes the Fourer transform of φ xx (τ) and H(f) s the frequency response of the channel. Hence, the channel mpulse response h(τ) s found va the nverse Fourer transform of ts frequency response H(f). Ths forms the bass of the technque used n the post-processng algorthm. The power delay profle s the expected value of the magntude squared of h(τ),.e., 2 E[ h( τ ) ] P ( τ ) = (5) 2 The RMS delay spread of the channel s calculated accordng to where n 2 2 τ = Pτ τ (6) RMS PT = PT = n τ = Pτ (7) A Blackman wndow s appled to the sgnals before Dscrete Fourer transformaton. Ths enhances the sgnal to nose rato of the power delay profle to more than 3dB. Snce RMS delay spread s known to be senstve to nose components on the power delay profle havng large excess delays [2], a nose excluson threshold s appled on the power delay profle so that any components more than 3dB below the peak response are excluded before calculatng the RMS delay spread. Usng ths crtera, 33 sets of measurements out of a total of 54 sets were processed and the results are presented as follows. The measurements are separated nto groups havng a range of m accordng to the SU antenna heght. Thereafter, the tap s of the tap delay lne channel models for each heght group are computed by takng the average of tap s at the same tap poston for all power delay profles n each group. The Rcean s computed usng the moment-method [3]. The Rcean for each tap of the wdeband tapped delay lne n each heght group s the average of all s of the power delay profles at the same tap poston.
4. Results The narrowband Rcean s plotted ast the excess path loss (path loss n excess of free space loss) n Fgure 2. The result shows that Rcean decreases, (approachng Raylegh fadng) as excess path loss ncreases (due to heavy shadow fadng). Rcean 5 4 3 2 y = -.49*x + 33 2 3 4 5 Excess Path Loss Fgure 2 Rcean vs. Excess Path loss Fgure 3 shows the RMS delay spread as a functon of excess path loss. The delay spread s observed to be hghly correlated to excess path loss. However, Fgure 4 shows that RMS delay spread does not sgnfcantly depend on dstance between SU and BS. The wdeband tapped delay lne channel model s summarsed n Table. The tapped delay lne channel model at a 5m SU heght s shown n Fgure 5. The results show that the tap at delays of 2ns (2 nd tap) and 4ns (3 rd tap) dmnsh as the SU antenna heght s ncreased. The Rcean s of the second tap are very small, whlst the value for the thrd taps are very close to zero, showng that the ampltude of the multpath echoes s close to Raylegh dstrbuted. Trms(ns) Trms(ns) 3 25 2 5 5 2 3 4 5 excess loss 3 25 2 5 5 Fgure 3 RMS delay spread vs. Excess Path loss.5.5 2 2.5 3 3.5 dstance (km) Fgure 4 RMS delay spread vs. dstance between subscrber antenna and base staton SU Heght (m) delay (ns) 4.5<h<5.5 3 samples 5.5<h<6.5 48 samples 6.5<h<7.5 65 samples 7.5<h<8.5 43 samples 8.5<h<9.5 78 samples 9.5<h<.5 57 samples 43.8 44. 44.6 45.6 46. 46.7 2-9.8 5.5-2. 8.6-2.6 7.5-22.5 9. -23.2 7.7-24.3 9.6 4-25.5 -.34-26..54-26.9.9-25.5 2.8-28.4 3.6-28.5 3.5 Table Summary of Wdeband ped delay lne channel model
Magntude -5 - -5-2 -25-3 2 4 6 8 Excess Delay (ns) Fgure 5 Wdeband tapped delay lne channel model at 5m SU antenna heght 5. Conclusons Ths paper has shown the nfluence of SU antenna heght on a BWA system employng drectonal SU antennas and sectored BS antennas at 3.5 GHz. It has been prevously reported that the average RMS delay spread s observed to decrease wth ncreasng SU antenna heght [9]. Ths paper further elaborates on the wdeband characterstcs of outdoor SISO BFWA rado channel at 3.5GHz, by showng the average tapped delay lne channel mpulse response. Smlarly to [7], t s confrmed that a 3-tap model can adequately descrbe the channel. However, due to the relatvely few hgh rse buldng and flat terran of the envronment and the narrower SU antenna beamwdth (2 o cf. 3 o n [7]), the maxmum average channel tap delay observed s 4ns. The ndvdual tap s of the multpath components are observed to decrease wth an ncrease n the SU antenna heght. Ths s consstent wth [9] whch reported that the RMS delay spread decreases wth an ncrease n the SU antenna heght. The narrowband Rcean s observed to correlate strongly wth excess path loss, whch s smlar to the fndng n [8]. In summary, the nfluence of dstance between SU and BS on the delay spread s less sgnfcant than s the heght of the SU antenna. References [] Cambrdge Broadband Lmted, Sngle carrer and OFDM modulaton: Ther sutablty for broadband fxed wreless access systems, 2 http://www.cambrdgebroadband.com/ [2] Crosby D., Greaves S. and Hopper A., A Theoretcal Analyss of Multple Dffracton n Urban Envronments for Wreless Local Loop Systems, Proceedngs 9th Vrgna Tech/MPRG Symposum on Wreless Personal Comms, 999 Blacksburg, Vrgna. [3] Porter J. W., Thweatt J. A., Mcrowave propagaton characterstcs n the MMDS frequency band, ICC2 (Internatonal Conference on Communcaton) Conference Proceedngs, 2, pp.578-582. [4] Gans M. J., Amtay N., Yeh Y. S., Damen T. C., Valenzuela R. A., Cheon C., Lee J., Propagaton Measurements for Fxed Wreless Lopps (FWL) n a Suburban Regon Wth Folage and Terran Blockages, IEEE Trans. on Wreless Communcatons, 22 Vol., No. 2, pp. 32-3. [5] Erceg V., Mchelson D. G., Ghassemzadeh S. S., Greensten L. J., Rustako A. J., Guerlan P. B., Dennson M. K., Roman R. S., Barnckel D. J., Wang S. C., Mller R. R., A Model for the Multpath Delay Profle of Fxed Wreless Channels, IEEE JSAC, 999 Vol. 7, No. 3, pp.399-4. [6] Saud I., Morn B., Investgaton on rado propagaton channel measurements at 2.2GHz and 3.5GHz for the fxed wreless access n an urban area Ann. Telecommun., 999 Vol 54, No. 9-, pp.464-478. [7] V. Erceg, K.V.S. Har, M.S. Smth, D.S. Baum, K.P. Shekh, C. penden, J.M. Costa, C. Bushue, A. Sarajedn, R. Schwartz, D. Branlund. Channel models for fxed wreless Applcatons (fnal IEEE 82.6 TG ad hoc verson), IEEE 82.6 Broadband Wreless Access Workng Group, <http://eee82.org/6/tg3/ndex.html> report number IEEE 82.6.3c-/29r4, 7 July 2. [8] P. Soma, D.S. Baum, V. Erceg, R. Krshnamoorthy, A.J. Pauraj, Analyss and Modelng of multple-nput multple-output (MIMO) rado channel based on outdoor measurements conducted at 2.5GHz for Fxed BWA Applcatons, Internatonal Conference on Communcatons, (ICC 22), Vol, pp 272-276, May 22. [9] C. L. Hong, I. J. Wassell, G. E. Athanasadou, S. Greaves, M. Sellars, Wdeband channel measurements and charactersaton for broadband wreless access, Twelfth Internatonal Conference on Antennas and Propagaton, 23 (ICAP 23), Volume, pp 429-432, Aprl 23. [] Bello P. A., 963, Characterzaton of Randomly Tme-Varant Lnear Channels, IEEE Trans on Communcaton Systems, Vol. CS-, No., 36-393 [] Proaks, J. G., 995, Dgtal Communcatons, Thrd Edton, McGraw-Hll Book Company, New York, 68-72 [2] Cullen P. J., Fannn P. C., Molna A., Wde-band Measurements and Analyss Technques for the Moble
Rado Channel, IEEE Trans. On Vehcular Technology, 993 Vol. 42, No. 4, 589-62. [3] L.J. Greensten, D.G. Mchelson, V. Erceg, Momentmethod estmaton of the Rcean. IEEE Comms Letter, Vol 3 No 6 June 999. [4] D. Crosby, S. Greaves, and A. Hopper. The Effect of Buldng Heght Varaton on the Multple Dffracton Loss Component of the Walfsch-Berton Model. IEEE Personal, Indoor andmoble Rado Communcatons Conference (PIMRC), September 23 (Bejng, Chna). [5] M.P. Sellars, G.E. Athanasadou, B. Zolko, S.D. Greaves, and A. Hopper. Smulaton of Broadband FWA Networks n Hgh-rse Ctes wth Lnear Antenna Polarsaton. IEEE Personal, Indoor andmoble Rado Communcatons Conference (PIMRC), September 23 (Bejng, Chna).