Clustering Characteristics of Millimeter Wave Indoor Channels

Transcription

1 This full ex paper was peer reviewed a he direcion of IEEE Communicaions Sociey subjec maer expers for publicaion in he WCNC 8 proceedings. Clusering Characerisics of Millimeer Wave Indoor Channels Behnam Neekzad, Kamran Sayrafian-Pour*, John S. Baras Elecrical & Compuer Engineering Deparmen *Informaion Technology Laboraory Universiy of Maryland Naional Insiue of Sandard & Technology College Park, MD, 742 Gaihersburg, MD, 899 Absrac Temporal-Angular channel sounding measuremens of an indoor millimeer wave channel ( GHz is analyzed o deermine wheher ray arrivals a he receiver form clusers in he wo-dimensional ime-angle space. The channel behavior and cluser energy disribuion for direcional anennas (wih various 3dB beamwidhs a he ransmier are discussed. Geomeric inerpreaion of he resuls is also provided. Index Terms Channel characerizaion, Millimeer wave propagaion. Temporal-Angular clusering I. INTRODUCTION Millimeer wave echnology is becoming increasingly imporan in many miliary and commercial applicaions. Remoe sensing, radio asronomy, passive imaging, radar and high daa rae communicaion are among hese applicaions. To esablish a high daa rae communicaion link beween wo nodes, accepable Signal o Noise Raio (SNR a he receiver is required. However, due o he high propagaion pah loss and signal aenuaion ha is associaed wih he millimeer wave signal, i is exremely valuable o have knowledge of how millimeer wave signals propagae in he environmen. Such informaion along wih appropriae smar anenna echnology could enable a sysem o maximize he qualiy of he received signal and herefore, achieve higher hroughpu. Indoor environmens creae challenging mulipah propagaion scenarios wih significan ime and angular spread. Join emporal-angular clusering phenomenon have been observed a frequencies commonly used in wireless neworks and 2-D saisical channel models have been proposed based on such phenomenon [1], [2]. However, a millimeer wave frequencies, and o he bes of our knowledge, no repors have been made for such clusering behavior. The clusering of arrival rays can have a significan impac on channel capaciy. Un-clusered models end o overesimae he capaciy if he mulipah componens are indeed clusered [14]. Considering angular (i.e. spaial aspecs of indoor channel is exremely imporan for sysems employing muliple anennas. Millimeer wave wireless communicaion sysems can grealy benefi from his knowledge as he physical size of array anennas makes hem amenable for pracical implemenaion [15]. Knowledge of cluser locaions is especially useful for such sysems as i could be exploied by various spaial diversiy combining or beamforming algorihms o enhance he sysem s performance [18]. The objecive of his paper is o sudy he 2-D clusering phenomenon of millimeer wave propagaion ( GHz a indoor environmens and paricularly observe he effec of direcional anennas wih various beamwidhs a he ransmier on he formaion and energy disribuion of such clusers. The res of his paper is organized as follows. Measuremen seup and he corresponding environmens will be described in secion 2. In secion 3, we will ouline daa analysis and processing mehodology. Resuls are provided in secion 4 and finally, conclusions will be discussed in Secion 5. II. MEASUREMENT SETUP AND ENVIRONMENTS Field measuremens have been conduced in [3] for boh residenial and office environmens. The residenial environmen is a Line of Sigh (LOS scenario as shown in Fig. 1. The room where he measuremens ook place has no furniure and a LOS pah exiss beween he ransmier and receiver. The floor, ceiling and he door are made of wood. The surface of he walls and ceiling is covered wih wallpaper. There are also 3 large windows (plane glass on he wo inersecing walls. The heigh of he ceiling and he window is 2.47m and 2.11m respecively. The ransmier and receiver are locaed 1.1m above he floor..48m Door (Widh:.75m, Heigh: 2.12m 1.925m RX Window #1 Wall side 3m 1.785m 1.48m 1.64m 1.12m 1.64m 6.85m TX Window #2 Figure 1: Layou of he residenial room 1.64m.965m.97m The office environmen is a Non-Line of Sigh (NLOS scenario as shown in Fig. 2. The room is made of seel wall, seel ceiling and seel floor. The floor and he ceiling are covered wih carpe and plaser board, respecively. Glass windows exis on one side of he office. To obain he angular componen of he received millimeer wave signal, a Window #3 3.57m /8/$25. 8 IEEE

2 This full ex paper was peer reviewed a he direcion of IEEE Communicaions Sociey subjec maer expers for publicaion in he WCNC 8 proceedings. direcional anenna wih a narrow angular resoluion of 15 degrees was used a he receiver. The anenna was posiioned o face he ransmier iniially and hen mechanically roaed in seps of 5 unil full 3 was obained m 4.25 m 4 m Meeing room.5 m 3.5 m 5 m 6 m 1 m 7 m 3.5 m 7.5 m 7.2 m 12.5 m III. DATA ANALYSIS AND PROCESSING A. CLEAN As menioned in he previous secion, he process of measuring angular daa involves roaing he direcional anenna a he receiver in seps of 5 afer each measuremen. This is equivalen o circularly sliding he anenna beam paern around he 3 field of view in he azimuh dimension. Noing ha he 3 db beamwidh of he receiver s direcional anenna is 15, he acual channel daa will be affeced by his process in a way ha is similar o circular convoluion. 2 m Locker room Wall (22 m 7.5 m 1.5 m Figure 2: Layou of he office A Vecor Nework Analyzer (VNA was used as he signal generaor and receiver. All measuremens were conduced in he frequency domain using he VNA. The receiver anenna used in [3] is a pyramid horn anenna wih 22 dbi gain (see Fig. 3. The 3dB beamwidh of he anenna in horizonal direcion is 15. A he ransmier, in addiion o an omni-direcional anenna, various pyramid horn anennas wih 1, 16 and 22 dbi gains were used for he measuremens. These are equivalen o, 3, and 15 3-dB beamwidhs. Calibraion was performed in an anechoic chamber wih one meer reference disance o remove he anenna effecs. (a (b Figure 3: Pyramidal horn anenna paern a he receiver (a H-plane (b E-plane During he measuremens, he VNA was se o sequenially ransmi 1 (1 for he office coninuous waves ones uniformly disribued over he frequency range GHz which resuls in frequency seps of 7.5 MHz (3.75 MHz for he office. This frequency resoluion gives maximum excess delay of abou ns (266.7 ns for he office. The 3 GHz bandwidh resuls in a emporal resoluion of.25 ns. An example of he raw daa produced by his seup is shown in Fig. 4. This is a 3-D plo of he received power wih respec o ime and azimuh angle. The frequency band for he experimen was GHz. In order o reduce he effec of non-saionariy in he channel and improve he signal-o-noise raio, 128 measuremens were aken for each angle and he resuls were averaged. Figure 4: Example of raw daa produced by he measuremen Also, he windowing and pulse shaping funcions a he receiver impacs he emporal response of he acual channel daa. Considering he effecs of he measuring process and insrumens, he resuling 2-D angular-emporal daa can be modeled as he rue channel response blurred by a Poin Spread Funcion (PSF. PSF is he measuremen sysem s 2-D angular-emporal impulse response when here is no mulipah. Here, he underlying assumpion is ha specular componens are he main significan componens of he mulipah channel. This is a valid assumpion since, as will be seen laer, scaered mulipah componens are negligible and mos of he propagaed energy is highly localized in ime-angle space. The raw measuremen daa, as a resul, represens arrivals rays ha have been blurred by he PSF and corruped by noise a he receiver. The PSF for he measuremen experimen discussed in he previous secion is displayed in Figure 5. Figure 5: Normalized 2-D PSF The side-lobes along he ime and angle dimensions are due o he effecs of windowing, pulse shaping and beam-paern of he anenna respecively. PSF basically represens how a single 1218

3 This full ex paper was peer reviewed a he direcion of IEEE Communicaions Sociey subjec maer expers for publicaion in he WCNC 8 proceedings. arrival a he receiver is displayed by he measuremen sysem. Removing he effec of he PSF and herefore idenifying he exac imes and angles of arrival is consequenly equivalen o a wo-dimensional (2-D de-convoluion. To solve his problem, we ried differen echniques such as Lucy- Richardson, CLEAN and blind deconvoluion. Blind deconvoluion [4] did no perform well and i is usually used in siuaions when complee knowledge of he PSF is no available. The Lucy-Richardson [5], [6] algorihm assumes ha he observed daa is he resul of a perfec convoluion wih a PSF added o a random noise wih Poisson disribuion. The CLEAN algorihm is commonly used for processing asronomical images [7] where i is assumed ha he radio sky can be represened by a number of poin sources in an oherwise empy field of view. Therefore, he resuling image is modeled as groups of poin sources convolved wih a blurring funcion ha is due o he limiaion of he imaging apparaus. In comparison o he oher wo mehodologies, CLEAN displayed a beer performance; herefore, we used his echnique o remove he angular-emporal effec of he measuring insrumen from he raw daa. Implemenaion-wise, he CLEAN algorihm essenially urns ino a recursive subracion of he shifed PSF from he raw daa [8, 9]. We processed he daa for each scenario o remove he effec of he anenna paern and oher processing a he receiver. For cases where here is a direcional anenna a he ransmier, i is imporan o noe ha he CLEAN-ed daa sill has he effec of he ransmier s anenna paern. Therefore, he observed channel characerisics will be dependen on ha paricular beam paern. B Cluser Idenificaion Mehodology Kernel Densiy Esimaion (KDE is a nonparameric Probabiliy Densiy Funcion (PDF esimaion approach ha can be applied o a given se of measured daa [1], [11]. KDE is a robus, simple and convenien mehod o esimae he PDF of a random variable given is sample realizaion. We have used he wo dimensional version of his echnique in order o esimae he emporal-angular disribuion funcion of he ray arrivals a he receiver. Having his disribuion funcion and using an appropriae hreshold, all 2-D clusers (i.e. emporalangular can be easily idenified. Le (, θ denoe he random variable ha represens he ime and angle of arrival of a ray a he receiver. If f (, θ is he 2-D PDF of hese arrivals, hen he mulivariae kernel densiy esimae of his funcion (i.e. f ˆ(, θ can be expressed by: N 1 (, θ ( i, θ i fˆ(, θ = K( Nh h h h θ i= 1 Where h and h θ are he emporal and angular bandwidhs respecively; K( x, y is he kernel funcion, ( i, θ i i = 1,2,..., N are sample realizaions of he random variable (, θ (i.e. arrivals wih he unknown densiy funcion f (, θ and N is he number of samples. Here, we θ have wo ses of samples which have been obained hrough measuremen and simulaion. The kernel funcion is ofen seleced o be a PDF ha is symmeric boh in ime and angle [12]. In our sudy, we have used he 2-D Normal disribuion o be he kernel funcion as wrien below: K ( x, y = exp( ( x, y 2 π 2 Therefore, he kernel densiy esimae of f (, a will be: 1 fˆ (, θ = Nh h N θ i = 1 1 (, θ ( i, θ i exp( 2π 2h h Using his approach, we have processed he CLEAN-ed ime-angle impulse response of he channel obained hrough measuremen. In he nex secion, we provide he resuls of his analysis and discuss heir geomeric relevance. IV. RESULTS One dimensional emporal clusering phenomenon has been observed in indoor channels and wideband saisical channel models have been proposed accordingly. Here, we also observe ha arrivals come in few groups (i.e. clusers scaered in differen coordinaes hroughou he ime-angle space. These 2-D clusers are idenified by using he KDE and an appropriae hreshold represening he noise floor of he receiver. A. LOS Residenial Environmen Figure 6a displays he clusers obained by processing he measuremen daa when an omni-direcional anenna is used in he residenial environmen. A oal of 6 clusers are observed in he channel daa. The coordinae of hese clusers are lised in Table 1. A cluser ime (angle coordinae is defined o be he median of ime (angle of all arrival rays in ha cluser. Table 1: Cluser coordinaes for he residenial environmens Cluser # Cluser Arrival Angel (Degrees θ Cluser Arrival Time (nsec A comprehensive geomeric inerpreaion of hese resuls can be provided by considering he main pahs ha he RF signal ravels beween he ransmier and receiver [19]. Each cluser formaion is he resul of a unique pah beween he ransmier and receiver. These pahs which include direc LOS and various single-refleced and double-refleced signals have been graphically idenified in Figure 7. There are no clusers due o reflecions from he walls behind he ransmier or receiver. This is because ha he ransmier and receiver are posiioned a he same heigh and he measuring equipmen 1219

4 This full ex paper was peer reviewed a he direcion of IEEE Communicaions Sociey subjec maer expers for publicaion in he WCNC 8 proceedings (a (e.g. covered by elecromagneic absorbers in case of he ransmier blocks he pah of he refleced signal from eiher wall. Cluser Cluser 1 5 Cluser (a Cluser (b Cluser 5 (b (c (d Figure 6: 2-D clusers (Residenial LOS environmen (a Omni (b α= (c α=3 (d α=15 Cluser (c Figure 7: Radio pahs resuling o clusers (a 1,2,3 (b 4,5 (c 6 The plos indicaing he cluser locaions for he case when direcional anennas wih differen beamwidhs were used a he ransmier are shown in Figure 6. α represens he 3-dB beamwidh of he ransmiing anenna. I is ineresing o observe ha as he beamwidh reduces, all clusers ha occur due o pahs ouside he beamwidh of he anenna disappear. In oher words, if he angle of deparure of a pah leading up o a cluser falls ouside he beamwidh of he ransmier anenna, hen he corresponding cluser will no be formed a he receiver. Also, looking a he case where α= and compared o he omni-direcional scenario, a new cluser (i.e. 7 is seen ha is approximaely locaed a (231, 24.25ns. Following he geomery of he layou, we conjecure ha his cluser forms as a resul of he reflecion from he door as seen in Fig. 8. In he omni-direcional case, he arrivals from his pah are oo weak and herefore no idenifiable by he KDE. For α=15, only he LOS cluser wih minimal delay spread is observed. This is 12

5 This full ex paper was peer reviewed a he direcion of IEEE Communicaions Sociey subjec maer expers for publicaion in he WCNC 8 proceedings. consisen wih our expecaion ha as α ges smaller, channel behavior approaches an AWGN channel. B. NLOS Office Environmen The daa from he NLOS office environmen was also processed for cluser idenificaion. Figure 9 demonsraes he clusers ha have been idenified from he measured daa. The scaering effec of he furniure and oher office equipmen in his environmen causes some spreading in he shape of he clusers. For he omni-direcional case, a oal of 3 clusers are observed in channel daa ha are he resuls of he LOS and single-refleced pahs. receiver when a direcional anenna wih 3 3-dB bemawidh is used a he ransmier. The direcional anenna is facing he receiver in his experimen Cluser Figure 9: 2-D clusers (Office NLOS environmen, Omni TX 1 1 Figure 8: Radio pah resuling o cluser 7 The coordinae of hese clusers are lised in Table 2. The direc pah beween he ransmier and receiver, cluser 1, and he arrivals due o he reflecion from he meal wall, cluser2, have merged and formed a bigger cluser. The cluser 3, occurs because of he single reflecion from he meal wall behind he receiver. Table 2: Cluser coordinaes for he residenial environmens Cluser # Cluser Arrival Angle (Degrees Cluser Arrival Time (nsec Looking a he ime coordinae of some of hese clusers, a mismach is observed beween he resuls and geomeric calculaion based on he RF pahs. A physical explanaion for his discrepancy can be given by noicing he velociy change of RF waves as i goes hrough differen maerials ha exis on is pah beween he ransmier and receiver [13]. In general, he exac coordinae, shape and size of he clusers depends on he deails of he propagaion environmens; however, as seen here, he locaions of he main clusers (i.e. high densiy can be reasonably approximaed by using informaion abou he layou and locaions of he ransmier-receiver pair. The clusers in figure 9 are due o he use of an omnidirecional anenna a he ransmier. If a direcional anenna is used insead, arrivals become more localized in ime-angle space and he cluser s sizes are reduced. The RF energy ransmission in ha case is more focused oward cerain direcion and herefore, he impac of he scaering on he ray arrivals is reduced. Figure 1 displays he clusers a he Figure 1: 2-D clusers (Office NLOS environmen, direcional TX C. Cluser Energy We have also invesigaed energy disribuion among he clusers in boh residenial and office environmens. The normalized percenage of he received energy in each cluser is calculaed as in [16], [17] and is referred o as he normalized relaive energy. Figures 11 and 12 illusrae hese resuls for omni and direcional anennas in residenial LOS and office NLOS environmens respecively. I is seen ha high percenage of he energy is concenraed in he firs cluser. This is especially expeced in he LOS case where refleced rays ravel longer pah and herefore conribue less energy o heir corresponding cluser a he receiver. This migh no be he case for NLOS scenarios in general. Also as observed, using direcional anennas wih smaller beamwidhs increases he energy of he 1 s cluser. This means ha for such scenarios, a single cluser saisical model is probably sufficien o describe he channel behavior. V. CONCLUSION In his paper, we have discussed he clusering characerisics of a millimeer wave indoor channel, he influence of geomery on his phenomenon, and energy 1221

6 This full ex paper was peer reviewed a he direcion of IEEE Communicaions Sociey subjec maer expers for publicaion in he WCNC 8 proceedings. disribuion among he clusers for direcional anennas wih various beamwidhs. Knowledge of such spaial disribuion of RF energy around he receiver in conjuncion wih array anennas and appropriae signal processing algorihms can lead o sysems ha provide superior performance by aking advanage of clusers locaion and heir corresponding characerisics. Also, saisical models ha conain boh angular and emporal componens of he millimeer wave indoor channel are needed o ensure more efficien design of millimeer wave sysems. According o he energy disribuion among he clusers, a single cluser saisical model is probably sufficien o describe he channel behavior in LOS scenarios. Furher experimenal resuls are required o verify and invesigae his observaion. Normalized Relaive Energy Cluser # Omni 3 15 Figure 11: Normalized percenage of clusers energy for differen ransmier 3-dB beamwidhs (Residenial LOS environmen Normalized Relaive Energy 1 Omni 3 resuls. Also, auhors would like o hank Dr. Quenin H. Spencer from Disribued Conrol Sysems, Inc. for his valuable commens and insighs regarding he CLEAN algorihm. REFERENCES [1] Q. H. Spencer, B. D. Jeffs, M. A. Jensen, A. L. Swindlehurs, Modeling he saisical ime and angle of arrival characerisics of an indoor mulipah channel, IEEE Journal on Seleced Areas in Communicaions, Vol. 18, No 3, Pages: 347 3, March [2] C. Chong, C. Tan, D. Laurenson, S. McLaughlin, M. A. Beach, A. R. Nix, A New Saisical Wideband Spaio-Temporal Channel Model for 5- GHz Band WLAN Sysems, IEEE Journal on Seleced Areas in Communicaions, Vol. 21, No. 2, February 3 [3] H. Sawada, Y. Shoji, H. Ogawa, Angle of Arrival Measuremen in Home and Office Environmens, Naional Insiue of Informaion and Communicaions Technology (NICT, Japan, doc# IEEE c, IEEE c [4] S.M. Jefferies, J.C. Chrisou, "Resoraion of Asronomical Images by Ieraive Blind Deconvoluion", Journal of Asrophy, No. 415, 1993 [5] L. B. Lucy, "An ieraive echnique for he recificaion of observed disribuions", The Asronomical Journal, Vol. 79, No. 6, 1974 [6] W. H. Richardson, "Bayesian-Based Ieraive Mehod of Image Resoraion", Journal of he Opical Sociey of America, Vol. 62, 1972 [7] J.A. Högbom, "Aperure synhesis wih non-regular disribuion of inerferomeer baselines" Asronomy and Asrophysics, Vol. 15, 1974 [8] A. Segaloviz, B. R. Frieden, A CLEAN-ype deconvoluion algorihm, Asron. Asrophys., No. 7, Pages: , 1978 [9] D. G. Seer, P. E. Dewdney, M. R. Io, Enhancemens o he deconvoluion algorihm CLEAN, Asron. Asrophys., No. 137, 1984 [1] D. W. Sco, Mulivariae Densiy Esimaion: Theory, Pracice and Visualizaion New York: Wiley, [11] A. W. Bowman, A. Azzalini, Applied Smoohing Techniques for Daa Analysis, Oxford Univ. Press, 1997 [12] L. A. Thombs, S. J. Sheaher, Local Bandwidh Selecion for Densiy Esimaion in he proceeding of he 22 nd Symposium on he Inerface, page Springer-Verlag, New York 199 [13] M. A. Hussein, Characerizaion of Ulra Wideband Communicaion Channels, Ph.D. Disseraion, Virginia Polyechnic Insiue and Sae Universiy, March 3 [14] K. H. Li, M. A. Ingram, A. V. Nguyen, Impac of clusering in saisical indoor propagaion models on link capaciy, IEEE Trans. on Communicaions, Vol. 5, Page(s: , April [15] K. Huang, Z. Wang, Millimeer Wave Circular Polarized Beam- Seering Anenna Array for Gigabi Wireless Communicaions, IEEE Trans. on Anennas and Propagaion, Vol. 54, February 6. [16] Q. Spencer, M. Rice, B. Jeffs, and M. Jensen, A saisical model for he angle-of-arrival in indoor mulipah propagaion, in Proc. IEEE Veh. Technol. Conf., 1997, pp [17] R. J. Cramer, R. A. Scholz, and M. Z. Win, Evaluaion of an Ulra- Wide-Band Propagaion Channel, IEEE Transacion on Anennas and Propagaion, Vol. 5, No. 5, Pages: , May 2 [18] B. Neekzad, K. Sayrafian-Pour, J. S. Baras, Energy Efficien Millimeer Wave Radio Link Esablishmen for Low Probabiliy of Inercep wih Smar Anenna, 25 h Army Science Conference, Orlando, FL, November 6 [19] B. Neekzad, K. Sayrafian-Pour, J. Perez, J. S. Baras, Comparison of Ray Tracing Simulaion and Millimeer Wave Channel Sounding Measuremen, 18 h IEEE PIMRC, Ahens, Greece, Sepember Cluser # Figure 12: Normalized percenage of clusers energy for differen 3-dB ransmier beamwidhs (Office NLOS environmen ACKNOWLEDGMENT The auhors would like o express heir graiude o Dr. Hirokazu Sawada, Dr. Yozo Shoji and Dr. Hiroyo Ogawa from he Naional Insiue of Informaion and Communicaions Technology (NICT in Japan for providing he measuremen 1222