Indoor Multiple-Antenna Channel Characterization from 2 to 8 GHz

Transcription

1 Indoor Multiple-Antenna Channel Characterization fro to 8 GHz Ada S Y Poon and Minnie Ho Counications and Interconnect Lab, Intel Corporation Abstract In ultiple-antenna channels, e optiality of a transission schee hinges on a detailed characterization of e channel. In is paper, we will present easureent results fro to 8 GHz in bo LOS and NLOS scenarios, at bo office and residential environents. Data processing eods are briefly outlined and a statistical characterization of e channel is presented as well. I. INTRODUCTION Recent indoor wireless channel easureents showed at received signals are not only clustered along e tie axis [1], but also in e directions of departure and arrival as well [] [3]. The clustering phenoenon is depicted in Figure 1, where e transitted signals bounce off two clusters of channel objects and arrive at e receiver wi different delay and directions. This physical description of e channel gives us useful insight on e ultiple-antenna channel beyond e coonly used independent fading odel [4]. In is paper, we will study is clustering phenoenon ore coprehensively. Based on data recorded fro an office and a townhouse over a 6 GHz bandwid, we will attept to address to e following questions: 1. What is e nuber of clusters and e angle spread? How do ey differ in e office versus residential environents and in e LOS versus NLOS scenarios?. How uch would e delay spread decrease when averaged over a cluster region? Also, how uch would e coherence bandwid increase when averaged over e cluster region? 3. How do ese channel paraeters vary wi frequency? 4. Which distribution functions would best fit e delay and directional statistics of e channel? cluster cluster 1 Rx array Figure 1: Clustering of received signals in a diffuse channel. The first question reflects e ultiplexing gain of a ultiple-antenna channel in different propagation scenarios and environents [5]. On e oer hand, e contrast of e delay spread and coherence bandwid wi eir clustered counterparts hints at a possibly lower coplexity transission schee. The changes of all ese paraeters wi frequency reflect how e ultiplexing gain and transceiver coplexity can be affected by e choice of operating frequency. Finally, a ore accurate statistical description of e physical channel is useful for bo analysis and siulation purposes. We begin wi a brief discussion of e easureent setup in Section II. Section III will present e techniques used to extract e necessary inforation fro e easureent data to tackle e above questions, and answers to e questions will be discussed in Section IV. II. MEASUREMENT SETUP A total of 11 data sets (6 in LOS and 5 in NLOS) were collected in a two-floor townhouse and anoer 9 data sets (3 in LOS and 6 in NLOS) were recorded in a typical office environent. The frequency of easureent is fro to 8 GHz wi resolution of 3.75 MHz. Each data set coprises a set of receive antenna positions lined up linearly and oriented eier towards or perpendicular to e transit antenna. The separation between e receive antenna positions is 1.5 c. We assue at e channel is static over e easuring interval. Each data set iics a easureent over a receiving array of eleents, naed a virtual array. III. DATA PROCESSING AND ANALYSIS First, e 6 GHz data is partitioned into 8 sub-bands, each of size 75 MHz. The processing techniques discussed below will be applied to each sub-band. A. Transforation fro Space to Angle Coordinates The easured channel response is in e frequency-space coordinate, H( f,n z) where z is e antenna separation and n denotes e n- receive antenna position. Assue at over a sufficiently sall local area, not ore an ten wavelengs wide, e channel paraeters reain constant. Then, e transforation is siply e inverse Fourier transfor as e receiving array is linear and unifor. Thus, e response in e frequency-angle coordinate is given by jπ f n z / H f, = H f,n z e c n The angle coordinate, in is paper, refers to e solid angle of e receiving signals wi reference to e center of e receiving array. Throughout e characterization, we use solid angle to easure e directional property of e channel due to its direct connection to e capacity of ultiple-antenna channels shown in [5] /3/$17. 3 IEEE 3519 Auorized licensed use liited to: Stanford University. Downloaded on October 7, 8 at :7 fro IEEE Xplore. Restrictions apply.

2 B. Deconvolution of e Angular Spectra Figure shows a saple channel response in e frequencyangle coordinate. Substantial sidelobes are observed along e angular axis due to e finite beapattern of e virtual array. The CLEAN algori [5], a deconvolution technique originally used in radio astronoy, is applied to reduce is sidelobe effect. The algori starts wi an assued beapattern, e Dirichlet function in is case. Then it locates e peak value of e response, subtracts a scaled version of e beapattern at e peak position and repeats e process on e residual response until a predeterined iteration liit is reached. The set of peak locations is en convolved wi a sooing function, e Gaussian function in is case, to synesis e angular response. Figure 3 shows e saple response after cleaning and sooing. Figure : Channel response in frequency-angle coordinate. Figure 3: Channel response after cleaning and sooing. The CLEAN algori is invoked again after e channel response is transfored to e delay-angle coordinate, h (,. This processing step deconvolves e sooing Gaussian function. Figure 4 shows e saple response after transforation and cleaning at e sub-band GHz. C. Cluster Identification The effect of clustering can be seen in Figure 4. Autoatic cluster identification algoris have been tried but e results were unsatisfactory. Instead, we visually identify e cluster angular interval denoted by, and en copute e cluster center and its delay boundary, as in []. Figure 4: Channel response in delay-angle coordinate at GHz. D. Clustering Characterization along e Angular Axis After identifying e clusters anually, e saple cluster center is given by h(, dd = h, dd and e saple cluster angle power profile is h, + d P = h, dd for and elsewhere. The saple cluster rs angle spread is given by rs = P ( d Averaging over all e clusters in all easureents gives e average cluster rs angle spread. E. Delay Spread and Cluster Delay Spread The delay power profile at a given angle is h(, P (, = h, d Averaging over gives e saple delay power profile. Siilarly, we can define e delay power profile at a given receive antenna position and average over all e receive antenna positions to get e saple delay power profile. However, we found out at e power profile is insensitive to wheer we use e angle or space coordinates. The saple rs delay spread is e average of 1/ rs ( = P (, d P (, d over e entire angular interval. The saple cluster rs delay spread is e average of rs ( over e cluster angular interval,. F. Coherence Bandwid and Cluster Coherence Bandwid The noralized frequency autocorrelation function in e angle coordinate is defined as H f + f, H f, df R f f, = H f, df 35 Auorized licensed use liited to: Stanford University. Downloaded on October 7, 8 at :7 fro IEEE Xplore. Restrictions apply.

3 The saple coherence bandwid is e average of B { θ } ( = inf ω > : R ( ω C f, over e entire angular interval for a pre-defined reshold value θ. The results presented in is paper are based on θ equal to.5. The saple cluster coherence bandwid is B over e cluster angular interval. e average of C G. Arrival Probability Density Functions (PDFs) For e sake of siplicity, we assue at e joint delay and angle arrival PDF is separable, at is, f, (, = f ( ) f ( Then e arrival delay and arrival angle distributions can be studied independently. One final processing stage at takes into account e tie and angular resolution liits of e easureent syste, is applied to avoid e over-counting of arrivals, known as binning [7]. In brevity, e peak position in e response is first located; a bin centered at e peak and of size equal to e teporal/angular resolution of e corresponding sub-band, is fored; all e signals inside e bin are cobined as a single arrival and are reoved fro furer consideration. The process repeats on e residue until no ore signal left. Arrival Delay PDF Suppose denotes e tie resolution. Then, e nuber of arrivals between n and ( n + over all e receive antenna positions is given by ( n+ N = ( ) θ n u h, z d n where u( ) is e step function. h(,n z) is e channel response in e delay-space coordinate, e Fourier transfor of H( f,n z) over f. The chosen detection reshold, θ, takes into account e noise floor of e channel. Finally, noralized N ( n ) gives e arrival delay PDF, N ( n ) f n =. N ) Arrival Angle PDF Siilarly, suppose denotes e angular resolution. Then, e nuber of arrivals between n and ( n + over all e resolvable delay intervals is written as ( n+ ( + N n = u h (, dd n θ Noralized is gives e arrival angle PDF. NLOS. In e townhouse, in contrast, ere is ostly 1 cluster in e LOS scenario and at ost clusters in e NLOS. Figure 5: Average nuber of clusters easured in e office. Figure 6: Average nuber of clusters easured in e townhouse. Figure 7: Average cluster rs angle spread in e office. IV. RESULTS AND INTERPRETATIONS In is section, we will address to e questions posed at e beginning of is paper based on e inforation extracted and presented in e last section. A. Nuber of Clusters and Cluster RMS Angle Spread Figure 5 6 show e average nuber of clusters. More clusters are recorded in e office. In e office, ere are 1 to clusters in e LOS scenario and up to 4 clusters in e Figure 8: Average cluster rs angle spread in e townhouse. Figure 7 8 show e average cluster rs angle spread. A larger angle spread is reported in e office, in general. The 351 Auorized licensed use liited to: Stanford University. Downloaded on October 7, 8 at :7 fro IEEE Xplore. Restrictions apply.

4 difference between LOS and NLOS scenarios is ore significant in e office as illustrated in Table 1. For reference, we also include e beawid in degree ( ) of a broadside array and an endfire array corresponding to e easured average cluster rs angle spread. angle spread (sr 1 ) broadside ( ) Endfire ( ) LOS Office NLOS LOS Townhouse NLOS Table 1: Average cluster rs angle spread in GHz sub-band. Fro e observation, we ay draw e conclusion at e office environent is ore diffuse an e residential environent. Thus, e office environent provides a better ultiplexing gain and is ore favorable for e deployent of ultiple-antenna systes. B. Delay Spread and Coherence Bandwid Figure 9 1 show e average rs delay spread. The delay spread ranges fro 1 ns to 3 ns. A larger delay spread is observed in e office. The average cluster rs delay spread is always saller an its non-clustered counterpart. The difference is ore apparent in e townhouse. Also, refer to Figure 11 1, e average cluster coherence bandwid is always larger an its non-clustered counterpart, which is consistent wi e observations fro e delay spread. First, e observation reflects at less intersybol interference (ISI) is attained in e angle coordinate provided at e transceiver is able to locate e cluster angular intervals. Second, e observed relative difference between e clustered and e non-clustered paraeters suggests at ere is ost likely ore an one bounce in e office environent and ostly a single bounce in e residential environent. However, a ore intensive easureent capaign needs to be carried out in order to confir is conjecture. Figure 1: Average (cluster) rs delay spread in e townhouse. Figure 11: Average (cluster) coherence bandwid in e office. Figure 1: Average (cluster) coherence bandwid in e townhouse. Figure 9: Average (cluster) rs delay spread in e office. 1 sr stands for steradian, which is e unit for solid angle. In is paper, e solid angle presented is noralized by π. C. Variation of Channel Paraeters wi Frequency Fro Figure 5 8, e average nuber of clusters and e average cluster rs angle spread decrease wi increasing operating frequency. The average rs delay spread, on e oer hand, increases wi frequency as shown in Figure 9 1. However, referring to Figure 11 1, we are not able to draw any conclusion between coherence bandwid and operating frequency. The decrease in e nuber of clusters and cluster rs angle spread can be due to e fact at an electroagnetic wave of higher frequency attenuates ore after passing rough or bouncing off objects in e environent. Most iportantly, e observation iplies at e channel is less diffuse at higher frequency. Hence, it counteracts e coon arguent at in oving to a higher operating frequency, one benefits fro packing ore antenna eleents on e wireless device. 35 Auorized licensed use liited to: Stanford University. Downloaded on October 7, 8 at :7 fro IEEE Xplore. Restrictions apply.

5 At is tie, we cannot explain e observation on rs delay spread. However, based on e easureent results, ore processing is required to itigate ISI at higher operating frequencies. e data, we observed at e distribution of e cluster centers along e angular axis is close to unifor. D. Statistical Characterization Figure show e average delay power profile and arrival delay PDF in NLOS respectively. Bo curves pursue a good fit to e exponential decay function, wi e decay constant being e easured rs delay spread. Figure 16: Arrival angle PDF at GHz (σ =.14 sr). Figure 13: Average delay power profile at GHz (σ = ns). Figure 14: Arrival delay PDF at GHz (σ = ns). V. SUMMARY An indoor channel easureent capaign is conducted, data processing eods are outlined, and a statistical characterization of e channel is presented as well. The results can be suarized as: 1. The nuber of clusters varies fro 1 up to 4 and e cluster rs angle spread ranges fro.6 to.18 sr. The office environent is ore diffuse an e residential environent.. The clustered rs delay spread and coherence bandwid are, in all cases, better an eir non-clustered counterpart, especially in e residential environent. 3. The nuber of clusters and cluster rs angle spread decrease wi increasing frequency while delay spread increases wi frequency. 4. Exponential decay functions are a good fit to bo e delay power profile and e arrival delay PDF. The Laplacian function fits well wi e angle power profile and Gaussian function provides a better fit to e arrival angle PDF. ACKNOWLEDGEMENT The auors would like to ank Professor Robert Brodersen and Ernest Tsui for e helpful discussions on post-processing e easureent data presented in Section III. REFERENCE Figure 15: Average angle power profile at GHz (σ =.14 sr). The average angle power profile in NLOS is shown in Figure 15. The sharpness of e profile akes e Laplacian distribution iediately apparent. However, for e arrival angle PDF shown in Figure 16, e flatness of e density function certainly rejects e Laplacian distribution. It turns out at Gaussian distribution provides a better fit. This is different fro e results reported in [] and [3]. Also, fro [1] A. A. M. Saleh and R. A. Valenzuela, A statistical odel for indoor ultipa propagation, IEEE JSAC, vol. 5, pp , Feb [] Q. H. Spencer, et al., Modeling e statistical tie and angle of arrival characteristics of an indoor ultipa channel, IEEE JSAC, vol. 18, pp , Mar.. [3] R. Heddergott and P. Truffer, Statistical characteristics of indoor radio propagation in NLOS scenarios, Tech. Rep. COST 59 TD() 4, Valencia, Spain, Jan.. [4] I. E. Telatar, Capacity of Multi-antenna Gaussian channels, European Trans. Tel., vol. 1, no. 6, pp , Nov./Dec [5] A. S. Y. Poon, D. N. C. Tse and R. W. Brodersen, Multiple-antenna channels fro a cobined physical and networking perspective, Asiloar Conference on Signals, Systes, and Coputers, Nov.. [6] J. A. Högbo, Aperture synesis wi a non-regular distribution of interferoeter baselines, Astron. Astrophys. Suppl., vol. 15, pp , [7] R. Jean-Merc Craer, An evaluation of ultra-wideband propagation channels, PhD esis, USC, Dec Auorized licensed use liited to: Stanford University. Downloaded on October 7, 8 at :7 fro IEEE Xplore. Restrictions apply.