34 1 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 44, NO. 3, MARCH 1996 A Comparison of Theoretical and Empirical Reflection Coefficients for Typical Exterior Wall Surfaces in a Mobile Radio Environment Orlando Landron, Member, IEEE, Martin J. Feuerstein, Member, IEEE, and Theodore S. Rappaport, Senior Member, IEEE Abstract-This paper presents microwave reflection coefficient the electric fields of each multipath component (or ray path) measurements at 1.9 GHz and 4.0 GHz for a variety of typical that illuminates the receiver. smooth and rough exterior building surfaces. The measured As with any radio propagation model, ray-tracing techniques test surfaces include walls composed of limestone blocks, glass, need to be verified and enhanced with actual RF measurements and brick. Reflection coefficients were measured by resolving individual reflected signal components temporally and spatially, which are representative of the possible installation scenarios. using a spread-spectrum sliding correlation system with direc- Propagation studies in microcellular environments have shown tional antennas. Measured reflection coefficients are compared that significant multipath components arise from reflections off to theoretical Fresnel reflection coefficients, applying Gaussian rough surface scattering corrections where applicable. Compar- of building surfaces [ 131-[ 161, [23]. Hence, ray-tracing techisons of theoretical calculations and measured test cases reveal niques must reliably predict the influence of these buildings that Fresnel reflection coefficients adequately predict the reflec- and other obstructions. For specularly reflected ray paths (i.e., tive properties of the glass and brick wall surfaces. The rough reflection for which parallel incident rays remain parallel after limestone block wall reflection measurements are shown to be reflection), the Fresnel reflection coefficients can be used to bounded by the predictions using the Fresnel reflection coefficients for a smooth surface and the modified reflection coefficients predict the reflection loss of a building surface, provided its using the Gaussian rough surface correction factors. A simple, but dielectric properties are known. While there exists a classic effective, reflection model for rough surfaces is proposed, which body of literature for the electromagnetic properties of various is in good agreement with propagation measurements at 1.9 GHz materials [17]-[ 181, the actual reflective properties of typical and 4 GHz for both vertical and horizontal antenna polarizations. These reflection coefficient models can be directly applied to the external building structures at microwave frequency bands are estimation of multipath signal strength in ray tracing algorithms just beginning to be explored [19]-[25]. for propagation prediction. To provide enhanced reflection coefficient models for buildings, measurements at 1.9 GHz and 4.0 GHz have been made for a variety of typical smooth and rough exterior I. INTRODUCTION building surfaces. The measured test surfaces include walls ITH THE eminent arrival of commercial personal com- composed of limestone blocks, glass, and brick. Reflection munication services (PCS), wireless service providers coefficients were measured by resolving individual reflected will need to rapidly deploy their networks to quickly gain signal components temporally and spatially, using a spreadmarketshare. To install and maintain reliable PCS systems, spectrum sliding correlation system with directional antennas. a thorough understanding of the RF propagation channel The measured reflection coefficients are compared with theis essential. Recent studies on ray tracing techniques for oretical Fresnel reflection coefficients using Gaussian rough propagation prediction [ 11-[ 121 have shown promising results surface scattering corrections when applicable. The dielectric in predicting channel parameters such as path loss and delay material properties of the tested building surfaces were taken spread in complex environments. Ray-tracing approximates from either published data [17], [18], or measurements [191. electromagnetic waves as discrete propagating rays that un- These comparisons are used to develop simple, and empirically dergo attenuation, reflection, and diffuse scattering phenomena accurate, reflection models for building surfaces that can be due to the presence of buildings, walls, and other obstructions. directly applied to ray-tracing algorithms. The total received electric field at a point is the summation of Manuscript received March 22, 1995. This work was supported by DARPAESTO and the MPRG Industrial Affiliates Program at Virginia Tech. 0. Landron is with AT&T Bell Laboratones, Wireless Communications Systems Engineering, Holmdel, NJ 07733 USA. M Feuerstein is with AT&T Bell Laboratories, Radio Technology Performance Group, Whlppany, NJ 07981 USA. T. S . Rappaport is with the Mobile and Portable Radio Research Group (MPRG), Bradley Department of Electrical Engineering, Virginia Polytechnic Institute and State University, Blacksburg, VA 24061 USA. Publisher Item Identifier S 0018-926X(96)01818-2. 11. REFLECTION AND SCATTERING MODELS The Fresnel reflection coefficients (r)relate the field reflected from an infinite dielectric slab to the incident field, and are described in [26], [27] as 0018-926X/96$05.00 0 1996 IEEE 342 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 44, NO. 3, MARCH 1996 h I\ Fig. 1. Geometry for fresnel reflection coefficient calculation. A i where E , and ET are the incident and reflected fields, respectively. The parallel (perpendicular) subscript refers to the E-field component that is parallel (perpendicular) to the plane of incidence, as shown in Fig. 1. The reflection coefficients, determined by material properties, angle of incidence (Q%), and frequency, are given by 8, 771 Qt rl = 77 22 COS COS 8,+ 7 1 COS Qt 8t r l l = 7 2 cos 8t 7 2 cos - COS - 171 cos 0, (2) + 111 cos 8, where the wave impedances (qm) and the transmitted wave angle (8,) are expressed as where w is the radian frequency and the wavenumbers (k,) are (4) The properties of each of the dielectric materials at the interface are characterized by their permittivity (E,), magnetic permeability ( p m ) ,and conductance ( c T ~ ) . The Fresnel reflection coefficients in (2) account only for specular reflection, which occurs for smooth surfaces. When the surface is rough, impinging energy will be scattered in angles other than the specular angle of reflection (i.e., diffuse reflection), thereby reducing energy in the specularly reflected component [26]-[30]. The Rayleigh criterion is commonly used as a test for surface roughness, giving the critical height (h,) of surface protuberances as h, = x ~ (5) 8 cos 8, where X is the RF wavelength. The height ( h ) of a given rough surface is defined as the minimum to maximum surface protuberance, as shown in Fig. 2. A surface is considered smooth if h < h, and rough if h > h,. Equation (5) shows that as the incident angle approaches grazing (i.e., 8, go"), --f the critical height becomes larger, effectively reducing the scattering effect of the surface protuberances. For the case of rough surfaces, a scattering loss factor (ps) was derived in [28] to account for diminished energy in the specular direction of reflection, given by where 5 h is the standard deviation of the surface height about the mean surface height in the first Fresnel zone of the illuminating antenna. The assumption in (6) is that the surface heights are Gaussian distributed. In general, incident radiation on a surface will induce a current density ( J ) which is a function of both 2 and y directions shown in Fig. 2. Equation (6) assumes that this surface current density at height y is always J(y), regardless of whether the surface element at a particular z is shadowed or illuminated by another part of the surface. This approximation was made to simplify the integral equations used to derive (6). When ps is used to modify the reflection coefficients, we refer to this model as the Gaussian rough surface scattering model, which is written as (rl)rough (7) (rll)rough = P S r l I . PSrl It was reported in 1291 that the scattering loss factor of (6) gives better agreement with measured results when modified as ps = e-[ -8 ( " ) 1, [y ( ") 2] (8) where l o ( z ) is the modified Bessel function of zeroth order. When the Bessel function argument is small, (6) and (8) are approximately equal, since 10 ( 2 ) approaches unity. The Fresnel models presented here assume the dielectric slab is infinite in extent. Although this is not true for real buildings, the dimensions of the tested building reflecting surfaces are sufficiently large that they may be approximated as infinite, such that these models can be applied. 111. 1.9 GHz AND 4.0 GHz MEASUREMENT SYSTEMS Various techniques for measuring UHF and microwave reflection coefficients of materials have been reported in the literature [20]-[25] and [31)-[33]. Of these, wideband measurement systems have the ability to temporally resolve desired reflections from spurious, unwanted signals. For this research, a wideband spread-spectrum sliding correlation system was ~ 343 LANDRON et al.: COMPARISON OF REFLECTION COEFFICIENTS IN A MOBILE RADIO ENVIRONMENT reference - .. .. , ~ length 32,767 chips -.: : m#e= LOS Measurement % ~ IOMHz- 4GHz Tnggercable to RX Transmitter System 239.960MHz SGA 40 4 GHz hom m cable from Tx Exthlgger \59-- RefleeemMesurement Fig. 4. Two-step technique for reflection coefficient measurement measured E-plane and H-plane half-power beamwidths of 17.2’ and 14.3’, respectively. The 1.9-GHz system used trapezoidal log-periodic antennas with 60’ half-power beamwidths. Further specifications of these systems are given in [20]. Due to the superior spatial resolution of the 4.0-GHz system, most of the measurements were performed at this band, where the surface roughness effects are more pronounced. IV. REFLECTION COEFFICIENT MEASUREMENT The reflection coefficients were measured using a two-step technique [20], [21] ?hewn in Fig. 4. First, 10 line-of-sight Receiver System (LOS) channel-impulse responses were measured with the two antennas facing each other. Then, the antennas were aimed at Fig. 3. 4.0-GHz wide band spread spect” sliding correlation system. the reflection point on the test surface and 10 reflected channelimpulse responses were collected. The LOS measurements developed, similar to that in [34]. The temporal resolution of provided a reference for comparison with the received channelthese systems is determined by the transmitted RF bandwidth. impulse responses from the reflection measurements. PerformThe 1.9 GHz and 4.0 GHz measurement systems used ing the LOS and reflected path measurements sequentially slightly different hardware, but were conceptually the same. ensured that variances due to the measurement equipment (e.g., Fig. 3 shows the system block diagram of the 4.0 GHz temperature variations, cable losses, connector losses, etc.) measurement system. At the transmitter, a CW source was were minimized. This two-step method was repeated for each modulated by the maximal length pseudonoise (PN) sequence, desired incident angle, as well as for different test surfaces. giving the characteristic (sin(z)/z)’ power spectrum. This Examples of measured LOS and reflected path impulse signal was then amplified up to a maximum of 10 W and responses are shown in Fig. 5. In many instances, the reflected transmitted via the high gain horn antenna. The transmitted path impulse response contained both an LOS and reflected power at the antenna terminal was monitored using the power peak, where the LOS peak was attenuated by the directional meter and accounting for the RF coupler and cable loss. antenna patterns. These multipath components were resolved At the receiver, the spread-spectrum signal was correlated to temporally by comparing the measured differential delay (i.e., the identical PN sequence with a clock rate which was slightly the difference in arrival times of the reflected path with respect offset from the transmitter. This causes the two sequences to the LOS path) to the calculated differential delay using the to slide past each other, giving maximal correlation when geometry in Fig. 4. The result of this comparison is shown in aligned. When multiple incoming signal paths arrive at the Fig. 5(b). receiver with different time delays, each will correlate with The transmitter and receiver locations were constrained the receiver PN sequence at the corresponding time. The by a number of factors, including desired specular refleccorrelation process effectively converts the wideband signal tion angles, temporal resojution of the sliding correlator, into the RF channel impulse response, where it was filtered, far-field criteria, and environment topography. For example, down-converted, and detected using a spectrum analyzer set to the temporal resolution of the measurement system dictates zero span. The baseband output was displayed in real time on the minimum resolvable differential delay, and hence, the the digital storage oscilloscope (DSO) and stored on a portable allowable transmitter and receiver positions. In addition, the computer for later processing. transmitter distance ( d l in Fig. 4) was selected to ensure that The temporal resolution of these systems allowed easy iden- incident waves striking the test surface could be approximated tification of desired reflected multipaths from other spurious as plane waves. For these reasons, the dimensions shown in signals. In particular, the 240-MHz PN sequence clock gave Fig. 4 varied over the following ranges: 3 m < d l < 27 m, 3 an RMS system resolution of approximately 4.2 ns. For spatial m < dz < 40 m, and 3 m < dLos < 65 m. For a particular resolution, the 4.0-GHz measurement system used 18.6-dB incident angle, d l remained fixed while the receiver distance was selectively varied. standard gain horns (at the transmitter and receiver) with (d’), and hence, IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 44, NO 3, MARCH 1996 344 0.3 - 0.2 "0.1 limestone wall 0 4 GHz LOS component <--> 0.25 V Q t = 1.0 ns 3 B 0.0 n 2a -0.1 8 -0.2 v) -0.3 -0.4 0 -50 100 50 150 200 Time (ns) effect, the receiver distance (d2 in Fig. 4) was selectively varied for most measured incident angles (Q%). These results are discussed in Section VII. The post-processing task began with the conversion of the individual impulse responses into corresponding measured power delay profiles, using the calibration characteristics of the receiving system. Then for each transmitter and receiver location, the ensemble average of the 10 raw power delay profiles was calculated. The averaged LOS power from the averaged LOS delay profile was taken to be the free space value, i.e., (a) m > limestone wall Q 4 GHz LOS component <--> 0.22 V Q t = -19.7 ns Reflected component <--> 0.24 V Q t = 1.O ns . -0.4l -50 calculated differentialdelay = 20.3 ns measured differentialdelay = 20.7 ns . ' 0 . ' ' 50 ' 100 . ' 150 200 Time (ns) (b) Fig. 5. Examples of (a) LOS channel-impulse response and (b) reflected path channel impulse response at 4.0 GHz (time axis has arbitrary zero reference). During the measurements, the antennas were kept stationary for two reasons. The first was that moving the directional antenna over a local area would induce artificial multipath fluctuations caused by directional antenna patterns, pointing errors, etc. The second was that it was observed in the field that the magnitudes of individual multipaths faded only slightly with small linear movements (i.e., a few wavelengths) of the receiving antenna. This observation agrees with the reasoning that signal fading in local neighborhoods is mainly due to the constructive and destructive interference of multipaths, rather than variations on the individual multipath amplitudes. This phenomena was also shown to be true for indoor factory channels containing an LOS path [35]. Individual multipath amplitudes should vary significantly within a small area only when a path is composed of a number of subpaths or when a path suddenly becomes shadowed by some obstacle in the environment. Since there w5re no obstacles in the test environments, the main cause of fades on any individual multipath is due to clusters of subpaths forming that multipath. Since the sliding correlation system can resolve individual multipath components 4.2 nanoseconds apart, the subpath clusters can possibly affect the wideband measurements in two ways. The first is in the LOS measurement where the ground reflection generally exceeds the system temporal resolution due to the physical geometry (discussed further in Section V). The second is in the reflected path measurements, where wall surface protuberances can induce interference effects that are also irresolvable by the wideband system. To observe this The averaged power delay profile for the reflected path measurements was calculated in an identical manner. The averaged reflected power was taken to be the free space value for the unfolded path length multiplied by the square of the voltage reflection coefficient (I?), or The use of the Fresnel reflection coefficient in (10) is an approximation since rigorous electromagnetic image theory requires perfect conductivity [36]. Equation (10) also assumes that the reflecting boundary is infinitely large, which is approximately true, when the wall dimensions are much larger than distances dl and dz in Fig. 4 and the wall surface area is much larger than the illuminated area. Assuming these conditions are satisfied, taking the ratio of (9) and (10) and solving for the reflection coefficient yields Therefore, the empirical reflection coefficient is the square root of the ratio of the reflected and LOS power measurements, weighted by the difference in measurement distances (i.e., accounting for the path loss difference in the two measurements). v. EFFECTSOF GROUNDREFLECTION In the LOS measurements, a ground-reflected multipath may interfere with the direct LOS path. These ground reflected multipaths are irresolvable from the LOS path because they generally exceed the temporal and spatial resolution capabilities of the measurement system. The severity of this interference depends on the antenna patterns and the distance between the transmitter and receiver. To determine whether ground reflection would have a significant effect on the reflection coefficient measurements, a series of LOS measurements were made. The transmitter and receiver separation was varied along a linear path while the wideband delay profiles were measured. These impulse responses were converted into their equivalent powers and plotted versus predictions with a two-ray model [20]. The two-ray model computes the received power as the magnitude squared of 345 LANDRON et al.: COMPARISON OF REFLECTION COEFFICIENTS IN A MOBILE RADIO ENVIRONMENT - -15 -20 - -20 -15 ! %-25' - - -25 . zl ," -30 -30 0 2 g -35 , ,$l -40 G,=G,= 18.6dB perfect gnd . r = -1 soil gnd er= 30, a = 0 02 -45 -35 ' G t = G r = 1 8 6 d B -40 -45' perfect gnd r = -1 Soil gnd : Er = 30,D = 0 02 ' ' I Fig. 7. Measured LOS power versus two-ray model at 4 GHz (horizontal antenna polarization). the vector sum of the LOS and ground reflected electric field components, whose magnitudes and phases were computed theoretically. The ground reflected component was multiplied by the reflection coefficient of the ground and the elevation patterns of the antennas. The ground was modeled as either being moist soil or a perfect reflector. The moist soil reflection coefficients were calculated using (2) with tabulated dielectric parameters [26] (E, = 30, c = 0.02 S/m). The perfect reflector was simply modeled as having a unity reflection coefficient (I? = -l), regardless of incident angle. The elevation patterns of the directional antennas were modeled as (sin(z)/z) functions. The two-ray model assumed antenna 3-dB beamwidths of 20" and two meter antenna heights (above ground). Figs. 6 and 7 show the results of the wideband LOS measurements and the various ground reflection models. The measurements show that there are minor variations around the free space LOS power, which loosely agrees with the predictions of the two-ray model. In general, the measurements agree well with the free space prediction, with the exception of one measurement point near 16 meters. Let the LOS error (ELOS) be defined as expressed as Er = Ir[(PR)MEAS]I- lr[(pR)LOS]I - For the measurements presented in this paper, the distance ratio in (13) is less than two for more than 85% of the transmitter and receiver positions used. Therefore, a conservative error estimate can be written as where E L ~ isS now the linear power ratio (not in dB). As the reflected to LOS power ratio decreases, the effect of LOS errors decreases. Typical reflected to LOS power ratios for the measurements reported here were less than -15 dB, although occasionally were above -10 dB. Therefore, if 94% of the measured LOS variations were less than 4 dB, the typical reflection coefficient errors should be no worse than 0.2 (in magnitude of field ratio). In general, the errors will be much smaller than that, so these measurements should give adequate results for the reflection characteristics of the test surfaces. VI. MEASUREMENT SITES where (PR)LOS is the ideal free space LOS power (in mW) and (PR)MEAS is the measured LOS power (in mW) including ground reflections. For each measurement point in Figs. 6 and 7, the LOS error was computed. From the ensemble of LOS errors, the mean error and standard deviation are less than 1.3 dB and 1.9 dB, respectively. The positive mean error of 1.3 dB could be easily caused by measurement system losses (such as cables, connectors, aperture efficiencies, etc.) which are not included in the free space model. Therefore, the significant parameter is the 1.9-dB standard deviation, which gives an indication of the variability of the measured LOS power. Assuming these errors are Gaussian distributed, 94% (i.e., two standard deviations) of the variations due to ground reflection are expected to be within 4 dB. To quantify the effects of these LOS errors on the computed reflection coefficients, the reflection coefficient error ( E r ) is Two buildings were carefully selected for the reflection coefficient measurements. These buildings had exterior walls made of rough limestone, smooth metallized glass, and brick. These walls were chosen since they were representative of typical building surfaces in a mobile radio environment. The chosen sites had homogenous surface characteristics (i.e., free of clutter, windows, doors, etc.) and were relatively isolated from objects which could induce additional reflection and scattering. The variation of surface roughness at these sites ranged from very rough stone to smooth glass. The rough stone wall was located toward the rear of a sixstory office building constructed of large rectangular limestone blocks. The sizes of the limestone blocks varied, with the average block being approximately 0.5 meters wide and 0.3 meters high. The outer surfaces of these blocks exhibited random roughness features, with a roughness height up to 12.7 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 44, NO. 3, MARCH 1996 346 TABLE I ROUGHNESS AND DIELECTRIC PARAMETERS FOR THE LIMESTONE AND BRICKWALLSURFACES AT 4.0 GHZ Wall h(cm) oh (cm) E, P? Limestone 12.7 2.5 7.51 0.95 0.03 Brick 1.3 0.5 4.44 0.99 0.01 Limestone wall freq = 1.9 GHz, Perpendicular (1) Polarization , 0 ? cm. The metallized glass wall was located on the same building and separated an interior lounge area from a parking lot. The glass wall measurements were conducted at 4.0 GHz only, while both frequencies were measured at the limestone wall. The brick wall was located in the rear of a multilevel recreational facility constructed of concrete and red masonry brick common in residential homes, as well as older urban buildings. Table I lists the estimated surface roughness parameters and the measured dielectric properties [19] of the limestone and brick surfaces at 4.0 GHz. The roughness parameters for these surfaces are the same for both frequencies, and the dielectric properties are approximately the same. The dielectric properties for the glass wall were not measured in [19]. 5 0.4 . % 0.3 - 0 0 0.0 1 0 .+ .!!",..-.-.p..' -.-, ,e. -.-. . c ,..'...<.' . .,.' 0 0 15 30 45 60 75 90 8, (degrees) Fig. S. Measured reflection coefficients for limestone wall at 1.9 GHz (perpendicular polarization). Limestone wall freq = 1.9 GHz, Parallel (11) Polanzation 1.0 S .g 0.9 f 0.8 E Gaussianrough surface Moddiedgaussian rough surface Measureddata 0.7 0.6 VII. MEASUREMENT RESULTS Measurement and prediction comparisons were performed for perpendicular and parallel polarizations for each test wall surface. The predicted reflection coefficients are simply the Fresnel formulas of (2) versus incident angle, using the rough surface scattering corrections when appropriate. Each measured reflection coefficient was obtained by averaging 10 LOS and 10 reflected path power delay profiles, then using (11) with the corresponding transmitter and receiver distances. For the receiver distance (d2 in Fig. 4) is most incident angles selectively varied to observe overall measurement variations due to surface protuberances. (e,), c?. Be 0.5 0.4 0.3 3 c '2 0.2 9 0.1 0.0 0 15 30 45 60 75 90 8, (degrees) Fig. 9. Measured reflection coefficients for limestone wall at 1.9 GHz (parallel polarization). Figs. 10 and 11 show the 4.0 GHz comparison results for the limestone wall for perpendicular and parallel polarizations, respectively. Again, the measurements are compared with the theoretical Fresnel reflection formulas for smooth surfaces A. Limestone Wall Results at 1.9 GHz and 4.0 GHz and for rough surfaces using the scattering corrections of Figs. 8 and 9 show the 1.9 GHz comparison results for (6) and (8). The theoretical models show that the smooth the rough limestone wall for perpendicular and parallel polar- surface reflection coefficients at 1.9 GHz and 4.0 GHz are izations, respectively. The measurements are compared with nearly equivalent. This is an expected result since the reflection the theoretical Fresnel reflection formulas for smooth surfaces coefficients in (2) become frequency independent when cr -+0. and for rough surfaces using the scattering correction of However, the scattering corrections are a strong function of (6) and (8). The Fresnel model using (6) is referred to as frequency. From the 4.0-GHz results in Figs. 10 and 11, over 92% the Gaussian rough surface, while using (8) is referred to as the modified Gaussian rough surface. For this surface, of the measured reflection coefficients are bounded by the the measured data shows significant variability as a function Fresnel formulas for smooth and rough surfaces. Similar to of the receiver distance (dz). This variability can be the the 1.9-GHz measurements, the 4.0-GHz measured data shows result of a number of factors, including ground reflection significant variability when varying the receiver distance (dz), effects and interference of the reflected multipath due to although the variations tend to be more clustered. In addition, surface protuberances. Regardless, over 90% of the measured the Gaussian rough-surface model (with either scattering loss reflection coefficients are bounded by the Fresnel formulas for factor) tends to give pessimistic predictions of the measured smooth and rough surfaces. In almost all cases, Figs. 8 and 9 reflection coefficient values, although (8) does provide some show that using the Gaussian rough surface model alone gives improvement over (6). The observation that the scattering loss factor overestimates pessimistic predictions to the measured reflection coefficient values. The comparison between the scattering loss factors the scattering loss for many of the measurements may be in (6) and (8) shows that (8) gives better agreement with attributed to a number of simplifying assumptions (e.g., Gaussian distribution of surface heights, sharp edge effects, etc.). measured values. LANDRON et al.: COMPARISON OF REFLECTION COEFFICIENTS IN A MOBILE RADIO ENVIRONMENT TABLE 11 ERRORSTATISTICS FOR COMPARISONS OF MEASURED REFLECTION COEFFICIENTS AT LIMESTONE WALL WITH MODELOF (15) Limestone wall Polarization freq = 4 GHz, Perpendicular (I) 1.0 C ~ .$ 0.9 5 0.8 8 0.7 2 0.6 d $ 3 .F m .$ 347 Freq.(GHz) Gaussianrough surface Modifiedgaussian rough surface Measureddata 0 0.5 0.4 " 0.3 .^ 0.2 ,.,.'.:.. I . 0 0 15 0 30 90rh % S I 1.9 Parallel 4.03 0.06 4.0 Perpendicular -0.07 0.16 0.29 0.10 4.0 Parallel -0.07 0.09 0.15 .e' . e . . . . _ . 0 Eays 0 0 0.1 0.0 ~ Polanzauon 45 . 60 75 90 0, (degrees) TABLE I11 ERRORSTATISTICS FOR COMPARISONS OF MEASURED 1.9 GHZAND 4.0 GHZ REFLECTION COEFFICIENTS AT LIMESTONE WALL WITH VARIOUS REFLECTION MODELS Reflection model Fig. 10. Measured reflection coefficients for limestone wall at 4.0 GHz (perpendicular polarization). Symbolic notation E, smooth surface rL 11 -0.23 Gaussian rough surface prrL 4.18 averagedmodel of (15) r,,, -0.03 s 1 Wrh% 0.16 0.42 0.13 0.31 0.13 0.19 Limestone wall freq = 4 GHz, Parallel(11) Polarization 0 15 30 45 60 75 90 0, (degrees) Fig. 11. Measured reflection coefficients for limestone wall at 4.0 GHz (parallel polarization). This was then repeated using the modified Fresnel formulas in (7), accounting for roughness effects. Table 111 details the error statistics comparing all of the 1.9 GHz and 4.0 GHz measured data with each of the three model combinations (i.e., r A v G , r l , l l , and psrl,ll). This table shows that using the smooth surface Fresnel coefficient alone gives overly optimistic results, while the rough surface model gives overly pessimistic results. The average error improvement of r A V G over the smooth and rough surface models is approximately 0.2 in both cases. The impact of this difference will be discussed in the next section. B. Glass Wall Results at 4.0 GHz Figs. 12 and 13 show the comparisons between the measured glass wall reflection coefficients and the Fresnel formulas for smooth surfaces for perpendicular and parallel polarizations, respectively. Since the glass wall was electromagnetically smooth, the rough surface scattering corrections were not applied. For this particular glass wall, the dielectric properties were unknown. The dielectric properties of clear, nondoped glass are typically quoted as: E, = 5 , U , , = 1 and U = lo-" where rl,li is either the perpendicular or parallel polarization S/m [26]. Since the glass wall was metallized, the conductivity reflection coefficient given in (2) and ps is the scattering loss was higher than lop1' S/m. Because the conductivity was factor given in (6). To assess the accuracy of the simple model unknown, several Fresnel reflection coefficient curves with in (15), r A V G was compared to each measured reflection various conductivities (10-l' to 10 S/m) are plotted in Figs. 12 coefficient in Figs. 8-11. From the resulting ensemble of and 13. A material having a conductivity of 10.0 S/m would errors, the error statistics, including average error (Eavg), be considered a semiconductor, similar to seawater or intrinsic standard deviation (s), and the 90th percentile are given in germanium. A good conductor would typically exhibit n > Table 11. From this table, r A V G tends to underestimate the lo6 S/m. reflection coefficient at 1.9 GHz and overestimate the reflection Figs. 12 and 13 show that the measured reflection coefcoefficient at 4.0 GHz (on average). The overall fit of r A V G ficients exhibit a dependence on incident angle similar to to the measured reflection coefficients was slightly better for that predicted by the Fresnel formulas. In addition, these parallel polarization, as seen by the smaller standard deviations measurements exhibit significantly less variability than that and 90th percentiles. Over the ensemble of measured results seen in the rough limestone wall for a particular incident angle at both frequencies and polarizations, the model of (15) gives ( B i ) , indicating much less interference effects from surface a 90th percentile of error of 0.19. protuberances. To quantify the improvement of r A V G over using only The measured reflection coefficients of Fig. 12 closely agree rl,ll for smooth surfaces, the error statistics comparing the with the theoretical Fresnel reflection coefficients using cr w 5 measured data with the Fresnel formulas in (2) were computed. S/m. For many of the measurements where 0, 2 45", the Regardless, Figs. 8-1 1 suggest that in the absence of measured data, an adequate model for rough building surfaces can be computed by averaging the Fresnel reflection coefficients for smooth and rough surfaces, as a function of incident angle, i.e., 348 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 44, NO. 3, MARCH 1996 0 : E 0.8 .-5 0.9 o.7 ._ 0 o,6 c $ 0.5 2 0 % 0.4 0.3 - ..... -.-.-----8 ........ ' 8c 0 .................. .......... . . 90th % .. 0.W /..*' ,.e' ,,_...<...-'" <..._... ....... -...-."-"-' .g 0.2 Y.V TABLE IV ERRORSTATISTICS FOR COMPARISONS OF MEASURED 4.0 GHZ REFL.ECTION COEFFICIENTS WITH IDEAL SMOOTH SURFACE MODELS Glass wall freq = 4 GHz, Perpendicular (I) Polarization 1.0 - a = 2.5 Slm a = 5.0 wrn G = 10.0 S/m Measureddata 0 15 30 45 60 75 Brick Perpendicular Brick Parallel _----;-- g 0.9 1.o 0 8 ..... ....... 0.8 E 0.7 0.6 $ 0.5 2 0.4 0 XI 0.3 o=io.OS/m .................... .......................... -,.. .... 5 0.1 0.0 -.- ....1)......-.-.2'. ,,<,;:I ........ '.'. ::! ............-3 ......a....... .< 0.2 B a = 2.5 Slm _---. o = 5.0 Sim 0 .... .... 15 30 a.01 1 0.10 0.16 I 0.18 8 06 e? e $ 0.5 - 04- 03/8-----+f !;I , 0 ..f 15 30 45 60 75 90 6, (degrees) .. .. %.. '.. Glass : E~ = 5.0 0 0.10 0 + ' -.., .... 0.32 -0.05 Ideal smoolh surface Gaussianrough sutface Modifiedgaussianrough sutiace Measureddata Glass wall freq = 4 GHz, Parallel (11) Polarization s 0.38 0.08 Brick wall Polarization freq = 4 GHz, Perpendicular (1) Fig. 12. Measured reflection coefficients for glass wall at 4.0 GHz (perpendicular polarization). ._ 0.9 0.06 4.26 90 6, (degrees) 4- 1 0.07 45 '. . '.. I.,: 60 .* 75 90 Fig. 14. Measured reflection coefficients for brick wall at 4.0 GHz (perpendicular polarization). 8, (degrees) Fig. 13. Measured reflection coefficients for glass wall at 4.0 GHz (parallel polarization). measured reflection coefficients are even greater than that predicted using F = 10 S/m. In Fig. 13, the comparisons indicate that the measured reflection coefficients are well approximated by the Fresnel reflection coefficients using 0 M 2.5 S/m. For modeling the glass wall, the conductivities of 5 S/m and 2.5 S/m were used for perpendicular and parallel polarizations, respectively. For each measured reflection coefficient, the modeling error was computed. From the resulting ensemble of errors, the error statistics, including average error (Eavg),standard deviation (s), and the 90th percentile are given in Table IV. The overall fit of the Fresnel reflection formulas to the measured data shows excellent agreement when using the appropriate dielectric properties. Over the ensemble of all measured results at the glass wall, the theoretical models give a 90th percentile of error of 0.13. These error calculations were then repeated using the tabulated conductivity for clear, nondoped glass (i.e., F = S/m) for comparison purposes. Table IV clearly shows that modeling this glass wall using the clear glass conductivity was pessimistic (as seen by the large positive Eavg).The average error improvement of using a conductivity which fits the measured data over using the published value for clear glass is approximately 0.31 and 0.26 for perpendicular and parallel polarizations, respectively. The impact of these differences will be discussed further in the next section. These results demonstrate that metallization and other impurities can have significant effects on the reflectivity of glass surfaces commonly used in building construction. C. Brick Wall Results ut 4.0 GHz Figs. 14 and 15 show the brick wall reflection coefficient results for perpendicular and parallel polarizations, respectively. For these comparisons, the Fresnel reflection formulas for smooth surfaces and rough surfaces were used. These formulas used the measured dielectric properties for brick [19] and the estimated roughness parameters from Table I. For this brick wall surface, there is little difference between the scattering loss factors of (6) and (8), because the small standard deviation of height (oh)causes the Bessel function in (8) to approach unity. Similar to the glass wall measurements, the brick wall data points are clustered together for particular incident angles (e,). This indicates little variability in the measured coefficients as the receiver distance ( d 2 ) is varied and, therefore, are relatively insensitive to interference effects from surface protuberances. The differences between the theoretical models of the smooth versus rough surface for the brick wall are much smaller than in the rough limestone case. Hence, the measured reflection coefficients of the brick wall are not well bounded by the smooth and rough surface predictions, in general. The roughness of the brick surface is such that the critical heights (h,)are greater than the maximum protuberances for many of the incident angles. Because of this, the Rayleigh criterion predicts that the surface appears smooth to the incident radiation and the reflection will be predominantly ~ 349 LANDRON et al.: COMPARISON OF REFLECTION COEFFICIENTS IN A MORILE RADIO ENVIRONMENT Brick wall Power difference due to reflectioncoefficient error (on single specular ray path) freq = 4 GHz, Parallel (11) Polanzation 10- 40 6i- S 30 Gaussian rough surface -U- 8 truer =0.5 20 $ t B 10 O -10 0 15 30 60 45 75 90 9, (degrees) Fig. 15. Measured reflection coefficients for brick wall at 4.0 GHz (parallel polarization). specular. To some degree, this is seen in Figs. 14 and 15 since the measurements are distributed about the predicted Fresnel reflection coefficients for smooth surfaces. To quantify the differences between the predicted reflection coefficients and the measured data, the modeling error is calculated for each measured data point. From the resulting ensemble of errors, the error statistics, including average error (Eavg),standard deviation (s), and the 90th percentile are given in Table IV. The overall fit of the smooth surface reflection formulas to the measured data shows excellent agreement when using the known dielectric properties. Over the ensemble of all measured results at the brick wall, the theoretical models give a 90th percentile of error of 0.18. In comparing the glass and brick wall measurements, both are accurately described using the smooth surface Fresnel formulas of (2).These formulas give slightly better agreement with the glass wall (using 0 = 2.5 5 S/m) than the brick wall. This is noted by the larger standard deviations and 90th percentiles in Table IV for the brick surface. Over the ensemble of measured results at the glass and brick walls, the smooth surface model of (2) gives a 90th percentile of error of 0.14. These error statistics indicate that (2) gives an empirically accurate model for surfaces with roughness features smaller than the critical height of (5). N VIII. IMPLICATIONS FOR PROPAGATION PREDICTION When discussing ray-tracing in urban environments and reflection properties of building surfaces, the following question typically arises, “How accurate must reflection coefficients be to obtain an accurate prediction?“ While the question is straightforward,the answer is, unfortunately, that it depends on the situation. On one hand, there is potentially a large number of variables which can have a significant impact on accuracy, in which reflection properties are but one (e.g., LOSIOBS topography, reflection properties, diffraction modeling, terrain effects, accuracy of building database, local scattering effects, etc.). On the other hand, ray-tracing algorithms are fairly sensitive to the reflective properties of the environmental objects, and if not adequately modeled, can give large prediction errors. As an example, let the true reflection coefficient of a building surface be represented by I‘. Now, assume the reflection -1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 reflection coefficient error (Er) Fig. 16. Effects of reflection coefficient error on a first-order, specularly reflected-ray path. + coefficient was actually modeled as I’ E r , where Er is the reflection coefficient error. Using (10) to calculate the reflected ray powers using the true reflection coefficient versus the modeled coefficient, the power difference (A) in dB of the two rays is A = 2010g (----) r r+Er [dB]. Fig. 16 shows the magnitudes of A for various reflection coefficient values. A positive power difference (A > 0) implies that the true reflected power is greater than the predicted power, and vice versa. As an example, the glass wall measurements in Section VI1 showed that average error improvement of the fitted reflection models over the predictions using published conductivities was approximately 0.3 1 for perpendicular polarization. Assuming the true reflection coefficient for this case was 0.75 for all incident angles, the predicted power error on a single reflected path would be approximately 4.6 dB. IX. CONCLUSION Microwave reflection coefficient measurements were presented at 1.9 GHz and 4.0 GHz for a variety of typical smooth and rough exterior building surfaces. The measurement test cases included walls made of limestone blocks, glass, and brick. The reflection coefficients were measured by resolving individual reflected signal multipaths temporally and spatially, using a wideband spread spectrum system and directional antennas. The measurement results were compared to the theoretical Fresnel reflection coefficients using Gaussian rough surface scattering corrections when applicable. The measured reflection coefficients at the limestone wall showed significant variability, which can be attributed to several factors, including ground reflection effects and interference of the reflected multipath due to surface protuberances. In any case, over 92% of the measured reflection coefficients at 1.9 GHz and 4.0 GHz were bounded by the Fresnel formulas for smooth and rough surfaces. An averaged reflection coefficient (FAVG) was proposed to improve the prediction accuracy. Over the ensemble of measured results at both IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 44, NO. 3, MARCH 1996 350 frequencies and polarizations, r A V G was shown to give a 90th percentile of error of 0.19. This was an average error improvement of approximately 0.2 over the smooth and rough surface models alone. The measured reflection coefficients at the metallized glass wall exhibited very little variability for particular incident angles This indicated significantly less interference effects (from surface protuberances) as that seen in the limestone wall. The measurements were compared with smooth surface Fresnel reflection models with various conductivities. Over the ensemble of all measured results at the glass wall, the theoretical models give a 90th percentile of error of 0.13 when using o = 5 S/m and 2.5 S/m for perpendicular and parallel polarizations, respectively. The average error improvement of using a conductivity which fits the measured data over using S/m) is the published value for clear glass (i.e., o = approximately 0.31 and 0.26 for perpendicular and parallel polarizations, respectively. Similar to the glass wall measurements, the measured reflection coefficients for the brick wall showed little variability as the receiver distance was varied and, therefore, were also insensitive to interference effects from surface protuberances. The differences between the theoretical models of the smooth versus rough surface for the brick wall were small due to the small standard deviation of surface protuberances (oh).Because of this, the measured reflection coefficients of the brick wall were not well bounded by the smooth and rough surface predictions, in general. Using the Fresnel reflection formulas for smooth surfaces, the comparison gave a 90th percentile of error of 0.18 over the ensemble of all measurements at the brick wall. These comparisons show that the actual reflective properties of various external building surfaces can be adequately modeled using the Fresnel reflection coefficients and rough surface scattering corrections when applicable. These results can be directly applied to ray-tracing algorithms for the purpose of propagation prediction in microcellular environments. The simple models presented here are defined by physical roughness features and material dielectric properties of the reflecting wall surface in question. (e,). ACKNOWLEDGMENT The authors would like to thank J. Don Moore, V. Erceg, A. Rustako, R . Valenzuela, M. Keitz, and D. Sweeney for many valuable discussions on this topic. For their help with the data collection and processing they would also like to thank C K Sou, A . M. Landron, H. Meng, B. Rele and P. Koushik. REFERENCES [ l ] T. S. Rappaporl, S. Y. Seidel, and K. R. Schaubach, “Site-specific propagation prediction for PCS system design,” in Virginia Tech. 2nd Symp. Wireless Personal Communicat., Blacksburg, VA, June 17-19, 1992, pp. 16.1-16.27. [2] R. A. Valenzuela, “A ray tracing approach to predicting indoor wireless transmission,” in IEEE Veh. Technol. Con$, Secaucus, NJ, May 18-20, 1993, pp. 214-218. [3] V. Erceg, A. J. Rustako, and R. S. Roman, “Diffraction around comers and its effects on the microcell coverage area in urban and suburban environments at 900 MHz, 2 GHz, and 6 GHz,” IEEE Trans. Veh. Technol., vol. 43, pp. 762-766, Aug. 1994. [4] S. J. Fortune, D. M. Gay, B. W. Kemighan, 0. Landron, R. A. Valenzuela, and M. H. Wright, ‘‘WISE design of indoor wireless systems: Practical computation and optimization,” IEEE Computat. Scz. Enn., vol. ‘ 2, no. 1, pp.-58-68, Spring 1995. r51 J. W. McKown and R. L. Hamilton, “Ray tracing as a design tool for .~ radio networks,” IEEE Network Mag., vol. 5 , no. 6, pp. 27-30, Nov. 1991. [6] N. Amitay, “Modeling and computer simulation of wave propagation in lineal line-of-sight microcells,” IEEE Trans. Veh. Technol., vol. 41, pp. 337-342, Nov. 1992. [7] S. Y. Seidel and T. S. Rappaport, “Site-specific propagation prediction for wireless in-building communication system design,” ZEEE Trans. Veh. Technol., vol. 43, pp. 879-891, Nov. 1994. [8] P. F. Driessen, M. Gimersky, and T. Rhodes, “Ray model of indoor propagation,’’ in Wrginia Tech. 2nd Symp. Wireless Personal Communicat., Blacksburg VA, June 17-19, 1992, pp. 17.1-17.12. [9] H. L. Bertoni, W. Honcharenko, L. R. Maciel, and H. H. Xia, “UHF propagation prediction for wireless personal communications,” Proc. IEEE, vol. 82, no. 9, pp. 1333-1359, Sept. 1994. [lo] W. Honcharenko, H. L. Bertoni, J. Dailing, J. Qian, and H. D. Lee, “Mechanisms governing UHF propagation on single floors in modern office buildings,” IEEE Trans. Veh. Technol., vol. 41, pp. 496-504, Nov. 1992. [l 11 M. C. Lawton and J. P. McGeehan, “The application of a deterministic ray launching algorithm for the prediction of radio channel characteristics in small-cell environments,” ZEEE Trans. Veh. Technol., vol. 43, pp. 955-969, Nov. 1994. [12] F. Ikegami, T. Takeuchi, and S. Yoshida, “Theoretical prediction of mean field strength for urban mobile radio,” IEEE Trans. Antennas Propagat., vol. 39, pp. 299-302, Mar. 1991. [13] A. R. Noerpel, M. J. Krain, and A. Ranade, “Measured scattered signals at 4 GHz confirm strong specular reflections off buildings,’’ Electron. Lett., vol. 27, no. 10, pp. 869-871, 1991. [14] P. A. Matthews, D. Molkdar, and B. Mohebbi, “Direction of arrival and frequency response measurements at UHF,” presented at IEE 5th Int. Con$ Mobile Radio Personal Communicat., Dec. 1989, pp. 4347. [15] S. Y. Seidel, T. S. Rappaport, M. J. Feuerstein, and K. L. Blackard, ‘%e impact of surrounding buildings on propagation for wireless inbuilding personal communications system design,” presented at IEEE Veh. Technol. Con$, Denver, CO, May 1992, pp. 814-818. [I61 A. J. Rustako, N. Amitay, G. J. Owens, and R. S. Roman, “Propagation measurements at microwave frequencies for microcellular mobile and personal communications,” in IEEE Veh. Technol. Con$, San Francisco, CA, May 1989, pp. 316-320. [17] A. R. Von Hipple, Dielectric Materials and Applications. New’ York M.I.T. Press and Wiley, 1954. [18] American Institute of Physics Handbook, 3rd ed. New York McGrawHill, 1972. E191 C. K. Sou, 0. Landron, and M. J. Feuerstein, “Characterization of electromagnetic properties of building materials for use in site-specific propagation prediction,” Mobile Portable Radio Res. Grp. (MPRG) Tech. Rep. 92-12, Virginia Tech., 1992. [20] 0. Landron, “Microwave Multipath Resolution in Microcellular Environments,” Virginia Tech. masters thesis, Aug. 1992. [21] 0. Landron, M. J. Feuerstein, and T. S. Rappaport, “In situ microwave reflection coefficient measurements for smooth and rough exterior wall surfaces,” in IEEE Veh. Technol. Con$, Secaucus, NJ, May 18-20, 1993, pp. 77-80. [22] W. Honcharenko and H. L. Bertoni, “Transmission and reflection characteristics at concrete block walls in the UHF bands proposed for future PCS,” IEEE Trans. Antennas Propagat., vol. 43, no. 2, pp. 232-239, 1994. 1231 M. 0. AI-Nuaim and M. S. Ding, “Prediction models and measurements of microwave signals scattered from buildings,” IEEE Trans. Antennas Propugat., vol. 42, pp. 1126-1137, Aug. 1994. [24] E. J. Violette, R. H. Espeland, and K. C. Allen, “Millimeter-wave propagation characteristics and channel performance for urban-suburban environments,” Natl. Telecommunicat. Inform. Administrat., NTIA Rep. pp. 88-239, Dec. 1988. [25] N. Papadakis, T. Sofos, P. Hatzopoulos, and P. Constantinou, “Material characterization,” in ZEEE Veh. Technol. Con$, Stockholm Sweden, June 7-10, 1994, pp. 1771-1774. [26] C. A. Balanis, Advanced Engineering Electromagnetics. New York Wiley, 1989. [27] E’. Rohan, Introduction to Electromagnetic Wave Propagation. Boston, MA: Artech, pp. 57-61, 1991. [28] W. S. ,Ament, “Toward a theory of reflection by a rough surface,” Proc. IRE, vol. 41, no. 1, pp. 142-146, Jan. 1953. LANDRON et al.: COMPARISON OF REFLECTION COEFFICIENTS IN A MOBILE RADIO ENVIRONMENT 1311 [34] [35] [36] L. Boithias, Radio Wave Propagation. New York McGraw-Hill, 1987. P. Beckmann and A. Spizzichino, The Scattering of Electromagnetic Waves from Rough Surfaces. Oxford: Pergamon Press, 1963. A. R. Ondrejka and M. Kanda, “A time-domain method for measuring the reflection coefficient of microwave absorbers at frequencies below 1 GHz,” in IEEE Antennas Propagat. Symp. Dig., Ontario, Canada, June 1991, pp. 16561659. J. Brune, “Polarimetric submillimeterwave reflectometry using a realtime quasioptical system,” in IEEE Antennas Propagat. Symp. Dig., Albuquerque, NM, June 1992, pp. 357-360. D. Kralj and L. Carin, “Wideband dispersion measurements of water in reflection and transmission,” IEEE Trans. Microwave Theory Tech., vol. 42, pp. 553-557, Apr. 1994. D. M. J. Devasirvatham, “Time delay spread and signal level measurements of 850 MHz radio waves in building environments,” IEEE Trans. Antennas Propagat., vol. AP-34, pp. 1300-1305, Nov. 1986. T. S. Rappaport, “Characterization of UHF multipath radio channels in factory buildings,” IEEE Trans. Antennas Propagut., vol. 37, pp. 1058-1069, Aug. 1989. W. L. Stutzman and G. A. Thiele, Antenna Theory and Design. New York: Wiley, 1981, pp. 229-235. 351 Martin J. Feuerstein (S’83-M’90) received the B.E. degree from Vanderbilt University, Nashville, TN, in 1984, the M.S. degree from Northwestern University, Evanston, IL, in 1987, and the Ph.D. from Virginia Tech, Blacksburg, in 1990, all in electrical engineering. He worked for Northern Telecom developing hardware and software for testing of telecommunications terminals from 1984-1985. During the 1991-1992 academic year, he was a Visiting Assistant Professor with the Mobile and Portable Radio Research Group at Virginia Tech, where his research focused on propagation modeling. From 1992-1995, he was employed by US WEST on the design and modeling of digital cellular and PCS radio systems. He is currently with the Radio Technology Performance Group at AT&T Bell Labs, Whippany, NJ, where he is working on simulation of CDMA systems. Dr. Feuerstein is a member of Tau Beta Pi, Eta Kappa Nu, and Phi Kappa phi. Theodore S. Rappaport (S’83-M’ 84-S ’85-M’ 87- Orlando Landron (S’90-M’92) was born in Washington, DC, in 1967. He received the B.S. and M.S. degrees in electrical engineering from Viginia Tech, Blacksburg, VA, in 1990 and 1992, respectively. From 1990-1992, he was a member of the Mobile and Portable Radio Research Group (MPRG) at Virginia Tech where he studied microwave reflection and scattering characteristics from typical building surfaces found in urban environments. As part of this work. he helued develou an RF channel imuulse response measurement system using spread-spectrum techniques. In 1992, he joined the Wireless Communications Systems Engineering Group at AT&T Bell Laboratories, Holmdel, NJ. He has been involved with a number of projects, including the AT&T PCN trial and the development of installation techniques for indoor wireless PBX systems. His work has primarily involved the characterization and modeling of radiowave propagation for personal communication systems design. SM’90) received the B.S.E.E., M.S.E.E., and Ph.D. degrees from Purdue University, West Lafayette, IN, in 1982, 1984, and 1987, respectively. Since 1988, he has been on the faculty of Virginia Tech, Blacksburg, where he is Professor of Electrical Engineering. In 1990, he formed the Mobile and Portable Radio Research Group (MPRG) of Virginia Tech, and founded TSR Technologies, Inc., a cellular radio/PCS manufacturing company that was purchased by Grayson Electronics (a subsidiary of Allen Telecom) in 1993. He has authored and edited technical papers and books in the field of wireless communications, including the textbook Wireless Communications, Priciples & Practice (Prentice-Hall), and serves on the ediON SELECTED AREAS IN COMMUNICATIONS, torial boards of the IEEE JOURNAL IEEE Personal Communications Magazine, and the Intemational Journal on Wireless Information Networks (Plenum: NY). Dr. Rappaport is CEO of Wireless Valley Communications, Inc., a technical training and consulting firm.He was awarded the Marconi Young Scientist Award in 1990 and the NSF Presidential Faculty Fellowship in 1992. He is a registered Professional Engineer in the state of Virginia and is a Fellow and Member of the Board of Directors of the Radio Club of America.