~-~~511ii'MZ'W't"fii&1!f!?Wi'tW:'t'Wt 'X'wtflfid@W w••&t · ·"r:fv -·wli!e· re•s·-wtefe" ,;,w ,,,_,.. ...... ~~·--··''·~~'":..,,,_or;-.-·-~ .... ""'"""-~~·""""'..., ... . , ... V'=•'"''-""<'"."'"'' .. , -m··~.------···-- ..... ,,,.,.""'•

1 Proc. of 2nd Annual Symp. on WPC., Virginia Tech, Blacksburg,VA June 17-19, 1992 !

SITE-SPECIFIC PROPAGATION PREDICTION FOR PCS SYSTEM DESIGN

Theodore S. Rappaport, Scott Y. Seidel, and ~urt R. Schaubach Mobile & Portable Radio Research Group

Virginia Polytechnic Institute_ and State University Blacksburg, VA

1.0 INTRODUCTION

Wireless communications has been plagued by our inability to fully understand how radio

waves behave in complicated channels. In the past, anti-fading and anti-multipath techniques such

as diversity and equalization have been used to combat the random small-scale fluctuations

induced by radio channels, and propagation models were developed to help the communication

engineer design radio links in a statistical sense, without much hope for complete understanding

of the channel conditions at any particular time or location. Given the difficulty of predicting

instantaneous channel conditions, a common thread in radio communication system design has

histo:-: :ally been that first there must be good first-order models to successfully deploy radio sys­

tem links, and then some form of diversity or equalization must be used to ensure a particular out­

age criterion is met during rapid changes in the channel.

Economies of scale and new RF technology will soon allow ubiquitous, mass pe~onal

communications to be provided by radio. These new systems will in some ways be similar to

existing radio systems, but in many ways will be quite different. Like existing radio systems, new

wireless personal communication networks (PCNs) will operate in channels plagued by complex

radio propagation. However, the "high density" aspect of emerging personal communication sys­

tems will make PCNs drastically different from all radio system predecessors. PCNs will operate

in confined environments with high user densities, such as city blocks and inside large buildings,

and, using the concept of cellular frequency reuse, will rely on an immense infrastructure of low­

power base stations which connect to the public switched network. These base stations will be

distributed throughout a city or within a building, and will represent an enormous infrastructure

cost.

This research is sponsored by DARPA ESTO, NSF, and the .MPRG INDUS1RIAL AFFIILATES PROGRAM

16-1

The results of propagation experiments throughout the past decade have converged to

some basic propagation models which give insight into the statistical nature of signal strength and

multipath propagation in and around buildings. Models such as those presented in [1]-[8] have

indicated that depending on the physical surroundings of the· communication path, signal

strengths and multipath delay spreads can vary by orders of magnitude within a relatively small

area of a few hundred meters. As an example of the large variation, consider a typical office build­

ing. Radio waves propagating along a hallway or in a large room will typically have signal

strengths stronger than signals propagating in free space, whereas transmissions between floors or

walls will experience signal loss which is several orders of magnitude giea:ter than free space

propagation [8] [9]. If a wireless system operates near the top of a high rise building, scattering

from neighboring buildings can cause significant multipath delays [10]. Over the paSt decade,

researchers have found that the predictability of channel behavior can be improved by incorporat­

ing a greater quantity of site-specific data in the propagation model. With the.recent technological

advances in computing, graphical imaging, and information management, powerful moqelsl inte­

grating site-specific data are now realizable.

While current channel models are useful for statistical predictions and theoretical systetp

designs, they cannot select the base station position that provides "optimum" capacity in a specific

physical environment. An example of the strong impact of propagation on system capacity was

shown in [11] where a simple parameter, the propagation path loss exponent, was varied for an

ideal spread-spectrum system using total frequency reuse. By considering widely known values of

path loss exponents and considering a few surrounding rings of co-channel cells, capacity was

shown· to decrease by almost an order of magnitude depending on the propagation conditions.

Some sample spatial distributions of the mobile terminals created additional capacity sensitivities

of several dB, thus illustrating the need to model the propagation environment and user locations

accurately. Since channels which offer higher path loss will provide more capacity due to the

attenuation of co-channel interlerence from neighboring cells, judicious placement of base sta­

tions should exploit the natural attenuation of the physical surroundings. However, in light of the

fact that such wide variations of path loss occur in microcell and in-building channels, proper base

station site selection is not obvious. Based on the work in [11] it is not difficult to see that picking

base station sites in the blind, without exploiting the physical attenuation in a wireless channel,

could lead to capacity losses of ten or twenty times the "best" system design. When ~ne considers

that the wireless system infr~tructure may become as pervasive as electrical outlets in the coming

16-2

··- filW!J'iH?fl ... ;,

decades, methods that can automatically and easily detennine the best system installations for any

specific location could save billions of dollars.

While many in the wireless community view site-specific propagation prediction and

installation tools as impossible, impractical, or unusable, we believe there are three compelling

reasons why they offer promise and are likely to be in widespread use well before the turn of the

century. First, the fact that emerging wireless systems are using spread spectrum and other wide

bandwidth modulation techniques implies that the instantaneous signal level will become -much

less important. For conventional narrow band transmissions, signals fade violently over small dis­

placements or observation times, which is known as Rayleigh-like fading. It is impossible to pre­

dict the exact signal level of such a fading signal at a particular point in time and space. However, _

wide band transmissions offer frequency diversity such that the composite signal level does not

change rapidly over the bandwidth of interest [12][13]. The inherent frequency diversity of wide

band transmissions allows system designers and installers to concentrate on local average signal

levels, which are subject to variations induced by the physical dimensions of objects in the chan­

nel, as opposed to sub-wavelength variations. Conventional systems virtually always employ

some sort of diversity (e.g. space or polarization) to mitigate rapid fading, so that again it is the

average signal level in a local area that will dominate coverage, and thus capacity. The ability to

concentrate on local averages rather than instantaneous signal levels allows site-specific propaga­

tion models to operate on a segmented model of the coverage area, where each segment of the

coverage area represents a specific local area typically spanning several wavelengths in any

dimension.

Secondly, physical databases such as terrain maps, building blueprints, aerial and satellite

images of building locations, and corresponding graphical database management 'systems such as

AutoCAD ™ and Geographic Resources Analysis Support System (GRASS) are widely avail­

able, inexpensive, or available free in the public domain. Attribute tags for database entities will

allow system researchers, designers, and installers to specify additional features about the physi­

cal environment to any degree necessary for the desired predication accuracy. Computational con­

siderations will diminish as the increasing computing power offered by workstations with parallel

processing engines will let system installers integrate detailed physical descriptions of the wire­

less environment with new site-specific type propagation models.

The third reason why site-specific system design and installation techniques show promise

is the subject of this paper. Basically, it appears that by integrating site-specific blueprint or map

information of the physical environment with innovative modeling techniques based on ray trac­

ing, excellent agreement between measured and predicted path loss and multipath profiles can be

obtained. The use of site-specific data to enhance mobile radio propagation prediction is not at all

new or surprising. In [14], it was shown that the local average signal level can be predicted with

good accuracy by considering the sum of only a few dominant propagation paths (rays) in a non­

coherent fashion. In [4], the applicability of a geometrically-based channel modeling technique

was postulated, and a series of wide band measurements were made to experimentally verify that

the local average signal strength of a wide band transmission is virtually identical to, but fades

much less than, a narrow band signal. Work in [15] in urban canyons showed excellent agreement

between theoretical and measured narrow band signal strength, and [16] showed how a transfor­

mation could be applied to a building database in order to evaluate three dimensional diffraction

·.due· to the shadowing of buildings. Reference [8] showed that by drawing a ray between a trans­

mitter and receiver and counting the number of partitions in an office building, it is possible to

predict the local average signal level with a standard deviation of 4 dB.

In this paper, we outline our approach to site-specific propagation prediction in microcell

and indoor environments. Section 2 describes the important propagation mechanisms in each of

these two environments, and highlights some of our early attempts to characterize propagation

using site-specific techniques. Section 3 details how maps and building blueprints are used to

characterize microcell and indoor channels, and how they are integrated with site-specific propa­

gation models. The ray tracing approach is described in Sections 4 and 5. Using the theory of geo­

desic domes, a novel technique for generating equal volume ray tubes is used to ensure uniform

illumination of rays from the transmitter to the entire coverage area. A reception sphere at any

received point is used to ensure all reflected, scattered, and diffracted energy is recovered. Section

5 also demonstrates acceleration techniques that provide huge computational savings by exploit­

ing knowledge of the ray tracing paths through the coverage area. Comparisons between mea­

sured and predicted multipath power del~y profiles in microcell and indoor channels are given in

Section 6.

' I" I

nw·em ., ··rttfMtm"hi''MP#t y· ·seetr· ""V ?er·'riht6 r · r~dl+h ··wac··~" ... ~., ·v·w~.,_., • ....,aa..._-....-.--........_-.._

2.0 PROPAGATION ENVIRONMENT

2.1 Microcellular

Although propagation in urban microcells is a complicated process, a few obsetVations

can be made to simplify the required propagation prediction. It is well known that a narrow band

signal is comprised of both fast and slow fading components. The fast fading is attributed to the

interaction of the multipath components and is influenced by the small scale features of the envi­

ronment (e.g. the summation of many multipath components with changing phases). On the other

hand, slow fading is attributed to large scale environmental factors, such as building location and

geometry. If the fast fading is neglected, then the average path loss over a local area can be deter­

mined via geometrical considerations using a ray optics model .[14]. Wide band signals (RF band-

. widths much greater than the reciprocal of the rms delay spread of the channel), however, do not

exhibit the same fast fading characteristics due to the inherent frequency diversity within the sig­

nal. . Given that temporal fading can be. neglected, it is assumed here -that large scale. building

geometry (building height, location, and orientation) will dominate the average path loss of wide­

band signals. This is a valid assumption, as was shown in [4]. Furthermore, the time dispersion of

a wideband channel is due to the location of the scatters, and can easily be determined with a geo­

metric optics. model. Hence, geometrical optics is a viable technique for predicting both wide

·band and narrow band channel characteristics.

The ray tracing concept was applied in a preliminary study conducted on the Vrrginia Tech

campus. A modified version of the model proposed in [14] for conventional urban cellular sys­

tems was applied to predict propagation in microcells with elevated antennas. For the study, nar­

row band (CW) measurements were made on the Vrrginia Tech campus at 900 MHz. Average path

loss values were computed at 20A. intervals and only a few primary rays were considered. These

primary rays were reflected rays (less than three hops per path), direct diffracted rays, or rays

which were both diffracted and reflected. The average path loss was predicted through a non­

coherent sum of the power in each ray arriving at the receiver. Predicted path loss values were

generally within 5 dB of the measured path loss [17]. The worst case error was 17 dB and was

obseiVed at only one location. At this particular location, the receiver was heavily shadowed by a

relatively tall building, such that additional effects such as scattering became significant. Overall,

the results indicate that ray tracing is a tractable method for predicting coverage in microcells

when measured propagation data is unavailable.

16-5

2.2 Indoor

Buildings vary greatly in size, shape, and type of construction materials. This· can make

propagation prediction difficult. The statistics of propagation measurements vary greatly from

building to building and only broad conclusions can be made using statistical models. Work in

[10] showed that radiation can exit a building and return after being scattered by neighboring

buildings.

Ray tracing represents the high-frequency limit of the more rigorous field problem and can

give quick approximate solutions when more elaborate methods are unworkable. Here, a modified

geometrical optics model is proposed to predict the propagation of radio waves in buildings. Like

for microcell channels, ray tracing is a physically tractable method of predicting the delay spread

and path loss of in-building radio signals. The time delays of individual multipath components

can be linked to specific radio paths. This concept is similar to radar systems that use the time

delay to estimate the distance to a target. Preliminary results from coherent ray tracing in build­

ings using image theory at 18 G Hz were presented in [ 18].

It is assumed here that large scale building geometry such as the placement and orienta­

tion of major walls and partitions will dominate the average path loss. Measurements show the

individual resolvable multipath components fade only slightly (-3 dB) as a receiver is moved in

indoor and outdoor channels over several wavelengths without major topography changes, or as

the terminals remain stationary over several seconds [19]. The time dispersion of a wide band

channel is due to the location and orientation of the scatterers, and can easily be determined with

a geometrical optics model. The principle propagation model uses geometrical optics to trace the

propagation of direct, reflected, transmitted and scattered fields. The rays, which represent a local

plane wave of the total field, originate from point sources and propagate in three dimensional

space. The lack of significant difference in propagation characteristics throughout the low micro­

wave band [20] indicates that a geometrical optics model where the actual frequency is ignored,

except when required as in the case of diffraction, could provide accurate prediction for a variety

of frequencies.

3.0 SITE-SPECIFIC DATABASE

In order to incorporate site-specific information into a propagation prediction tool, it is

necessary to have a method of handling large amounts of building and terrain data. AutoCAD TM:

is used as the building database manager and as the graphical interface for the prediction tool.

This program was selected because it is considered an industry standard CAD package, and build­

ing data is often readily available from city planners and architects in an AutoCAD fo:nnat.

AutoCAD supports a standard drawing exchange format permitting data to be exported in ASCII.

This exchange format is used to export the building database directly from AutoCAD to the prop­

agation prediction tool. Using special macro commands developed for extruding three dimen­

sional objects from two dimensional maps, a building database can be constructed and exported to

the prediction tool in a straightforward manner.

Within the next few years, computer databases of terrain and building locations for micro-

. cellular environments, and wall locations for buildings will be available. Today a significant

amount of environment data is physically accessible. We have found some considerations are nec­

essary regarding the use of such data. In a static microcellular channel, the received signal is.com­

posed of energy which is reflected, transmitted, or scattered by buildings in the microcell.

Additional scattered signals from trees and lamp posts also contribute to the received signal.

However, as was indicated in Section 2.1, these effects will not usually dominate the local mean

signal strength and can thus be neglected. Therefore, the data required for a propagation model

would consist of the geometrical and the electrical characteristics of buildings in the microcell.

A building database structure was developed to store the required building data. The

buildings in the database are modeled as solids bounded by polygons (i.e., each building is com­

prised of a series of individual planar panels). A planiform environment is assumed where terrain

is flat and buildings have an effective height above the terrain level. This assumption will be valid

in most cases due to the small geographic areas required for analysis. Exact building heights are

used where available, otherwise the building is given a height according to 4.2N (meters) where N

is the number of floors in the building. Each building within the database can be assigned reflec­

tion and transmission losses according to the building material or by user specification.

Although terrain data is not currently included in the database, there is ongoing work at

MPRG to integrate a comprehensive building and terrain database with the three dimensional ray

16-7

tracing model. In the future it will be possible to overlay three dimensional building data, derived

from an AutoCAD drawing, on top of a terrain map (which can be obtained in a digitized format

from US Geological Survey) stored in GRASS. Using both building and terrain data will enhance

the accuracy of the prediction in urban areas and will permit predi~tions in less urbanized areas

where hilly terrain is a dominant effect.

Inside buildings, walls and floors are represented as fiat parallelograms that are specified

by their location, size, and orientation. Objects may have arbitrary orientation. In most buildings,

the propagation environment can be completely modeled using these parallelograms. The electro­

magnetic properties of the objects such as reflection coefficient, or relative permittivity are stored

with each object. The locations of transmitters and receivers can be placed directly in the

AutoCAD representation of the building. A utility conversion file program has been written to

convert the data from a standard ASCII CAD format to a format recognized by the ray tracing

computer code.

Thus, we have developed the framework to incorporate readily available environment

information, handle large amounts of that information within the propagation prediction tool, and

utilize the information in more accurately predicting the propagation.

4.0 PROPAGATION MODELS

We choose to use a three-dimensional "pin cushion" method of launching rays instead of

image theory for several reasons. Image theory works well for simple environments, but can

become quite complicated as the complexity of the site-specific database increases. Most impor­

tantly, by using ray tracing "tubes" of energy from a point source, the physical process of a radio

wave being launched from a point source is closely modeled. Although it appears at first that this

method is much less computationally efficient than using image theory, it is much more flexible,

and different propagation modes can be handled easily. In addition, ray tracing acceleration tech­

niques can be employed to drastically reduce computational requirements. Although this method

is computationally intensive, faster computers are continually being developed. As these comput­

ers become available, it is critical that the models for accurate propagation prediction tools are in

pla·ce to take advantage of these technological advances.

lfl-R

.. ·,.)

'l'itW'

4.1 Reflection, Transmission, and Scattering

The propagation of energy from the transmitter to the receiver occurs in various modes ·

such as by direct, reflected, transmitted, and diffusely scattered paths. In considering these modes

for the propagation model, it is important to recognize the path loss dependence with distance (d)

of each mode. Direct (line-of-sight) rays exhibit a l/d2 power dependence according to Friis free

space transmission. Specularly reflected and/or transmitted rays follow a l/d2 dependence, where

d represents the total ray path length. For example, the specularly reflected ray shown in Figure

4.1-1, whose path segments are labeled r1 and r2, has a path loss proportional to 1/(r1 + r2)2. Alter-

Building Interfaces

Tx

FIGURE 4.1-1. Transmitted, reflected, and scattered rays at a planar interface.

natively, the diffusely scattered field, comprised of the path segments s1 and s2, exhibits a multi­

plicative dependence given by l/(s1 s2)2. The multiplicative dependence is due to the additional

spreading loss the ray experiences after re-radiation from the scattering surlace.

The diffuse scattered component is commonly neglected in propagation models because

its magnitude decreases more rapidly with distance than the specular component. In computing

average path loss, this is generally a valid assumption since the specular component contains a

greater portion of the total energy and thus contributes more to the received power. However, the

diffuse component will influence the local average at locations close to the scattering point (i.e.,

when s2 is small). This accounts for scattering objects in a local area of the receiving point which

may have a significant quantity of incident energy but do not satisfy the geometric requirements to

specularly reflect this energy toward the receiver. Measurements in [ 16] in fact showed that when

the specular components are blocked, scattering becomes the dominant propagation mechanism in

microcell channels. For flat surfaces, such as building faces, we use a simple rough surface scat­

tering model that depends on the surface roughness, smface area illuminated by the ray tube, and

object dist~ce from the transmitter and receiver. The scattered energy at the receiver due to a spe­

cific object in the channel is modeled as the superposition of the scattered energy carried by all ray

tubes that intersect the object and scatter energy toward the receiver. This superposition can be

seen in Figure 4.1-2. Scattered rays, although weak, can have long time delays which would influ­

ence the channel time dispersion. For these reasons, the diffuse component is included in this

model. A ray tracing model based solely on image theory cannot account for diffusely scattered

rays.

Rectangular Area ex

Rx Tx

Area Illuminated by Ray Tube

FIGURE 4.1-2. Scattering model for flat surfaces. The total scattering surface is subdivided by the rays incident on the rough surface.

In summary, the model includes direct, reflected, transmitted, and scattered fields . repre­

sented by the rays. Table 4.1-1 summarizes the variables used to describe the model. Each propa­

gation mechanism is treated separately, and the total field is determined via non-coherent

superposition of the individual contributions of each ray. Non-coherent energy reception is used

because the program predicts mean field strength (path loss) as influenced by the local variations

in site-specific geometry over several wavelengths. The constructive and destructive interlerence

of the individual multipath components still exist, but are not modeled here. This program pre­

dicts the channel impulse response which is a wide band characte~zation. Recall, wide bandwidth

signals do not vary significantly over small distances due to the inherent frequency diversity in the

signal [4]. The field amplitude of the ith ray at the receiver is given.by

E; == Eofti f,;L; (d)Ls ('V) rrr (8ji) f1T(8ki) j k

(4.1-1)

TABLE 4.1-1: Summary of the variables used to describe the ray tracing propagation model.

, Variable : Description [units] , . : ·! "; : : : ..

fti Field amplitude radiation pattern of the transmitter antenna

fri Field amplitude radiation pattern of the receiver antenna

~(d) Path loss for the ith multi path component

d Path length [meters]

L,('l'i) Path loss for scattered rays

'Vi Angle of ith scattered ray with the specular direction [radians]

r (eii) Reflection coefficient

T (eki) Trarismission coefficient

Ei Field strength of the ith multipath component [V/m]

ER Mean received field strength [V /m]

Eo Reference field strength [V /m]

In (4.1-1), the reflection, r(ej;), and transmission, T(8ki), losses for the ith ray are given as

functions of incidence angle, e. This is to illustrate the dependence of these specular losses on

incidence angle. The mean field strength, ER, is the sum of the magnitudes of the individual mul­

tipath components as given by

(4.1-2)

Path loss is computed by referencing the result of (4.1-2) to a 1-meter free space path loss.

16-11

4.2 Diffraction

At microwave frequencies, it is likely that significant diffracted energy will reach the

receiver at some locations. Although wavelengths are short enough to utilize geometrical optics

assumptions, it is expected that there will be some discrepancy betWeen measured and predicted

propagation impulse responses. In microcellular and in-b~i1ding radio channels, there are many

ninety degree corners either around a city block or around the corner in a hallway. In these heavily

shadowed environments, diffraction can be a significant contributor to the received signal strength

[16]. Hence, we also incorporate wedge diffraction to make our propagation model complete.

5.0 COMPUTER RAY TRACING

5.1 General Ray Tracing

A computer program has been written to implement an automated propagation modeling

tool. The program uses "brute force" ray tracing to account for all possible propagation paths.

This approach allows randomly oriented objects to be incorporated in a general fashion. Image

theory cannot handle such random surfaces easily. Ray tracing acceleration techniques [21] are

employed to combat the computational requirements of brute force ray tracing.

Code is written in C++ for use on a UNIX-based Sun SparcStation 2 [26]. In order to

facilitate modular ray tracing code it is desirable to have the capability to manipulate vectors,

objects and the functions that match them. As an object-oriented language, C++ is a logical solu­

tion for .maintaining code modularity. For each ray, a test must be performed on every object in

the database to determine if an intersection occurs. This is a computationally intensive process·

due to the number of calculations required to perlorm intersection tests. Memory requirements are

also quite extensive because usage grows exponentially with the number of levels in the tree.

Although particular care is taken to use object bounding volumes to reduce intersection tests and

dispose of intermediate data as soon as possible, there are still large computation and memory

requirements. As such, a workstation is necessary at a minimum.

The computer program uses ray tracing to find each ray path by which significant levels of

energy radiated from the transmitting location reaches the receiving point. Multiple receiving

locations can be defined, so the procedure described here can be applied to each receiving point.

•• r ·n

()

The ray tracing is accomplished by an exhaustive search of a ray tree accounting for the decompo­

sition of the ray at each planar intersection. After the direct path has been checked, the program

traces a ray from the source in a predetermined direction and detects if an object intersection

occurs. H no intersection is found, the process stops and a new source ray is initiated. Once the

program determines that an intersection has occurred, it then checks to see if a reflected or trans­

mitted ray has an unobstructed path to one of the receiving locations. H both of these rays are

blocked, the program proceeds to determine if a scattered ray has an unobstructed path to one of

the receiving locations. After checking for reception, the program divides the source ray into a

transmitted and reflected ray, which are then treated in a similar fashion to the source ray. This

recursion continues until a maximum number of tree levels is exceeded or the energy in the ray

falls below a user-specified threshold. The ray tree which shows how rays are traced is shown in

Figure 5.1-1. Note that scattered rays are not traced recursively. Each intersection can produce an

infinite number of scattered rays, however, only the scattered ray passing directly through a

receiving point is considered. This is justified by recognizing that amplitude of these rays

decreases rapidly with distance; multiple scattering of a ray will not contribute significantly to the

received power and thus is not considered.

R

Source Ray

2

\ \ . . I

I I

I

T , 2

/ FIGURE 5.1-1. Schematic form of the ray tree. Reflected and transmitted rays are traced recursively. Scattered rays are traced directly to each receiver.

5.2 Source Ray Directions

The transmitter and receiver are modeled as point sources in a building. In order to deter­

mine all possible rays that may leave the transmitter and ani.ve at the receiver, it is necessary to

consider all possible angles of departure and arri~al at the transmitter and receiver. Rays are

launched from the transmitter at angles 8 and~ relative to a fixed coordinate system.

To keep all ray manipulation routines general, it is desirable that each ray tube occupy the

same solid angle dQ, and each wavefront be an identical shape and size at a distance r from the

transmitter. An ideal wavefront is shown in Figure 5.2-1. Additionally, these wavefronts must be

-----ideal wavefront

FIGURE 5.2-1. Ideal wavefront represented by each source ray.

subdividable so that an increased ray resolution can be handled easily. For reference, let r= 1 and

the total wavefront is the surlace of a unit sphere. The problem then becomes one of subdividing

the sphere surface into equal area "patches" that are all the same size and shape and completely

cover the surface. For a two-dimensional (flat) surface, this problem is easy. Regular triangles,

squares, and hexagons can completely cover an area with equal size and shape objects without

leaving gaps. Constant angular separation of the rays may be easy to conceptualize, but the

mechanics of the problem are rather complicated. Common methods of sub-dividing the total

wavefront by using spherical coordinates are insufficient. This is due to a decrease in the angular

separation of the rays as they are launched in directions near the poles of the sphere.

The solution is adapted from the theory of geodesic domes [22]. An icosahedron is

inscribed inside the unit sphere. A regular icosahedron is a twenty sided polygon with triangular

faces and twelve vertices. Each vertex joins five faces. Figures 5.2-2a and b show two views of a

regular icosahedron. In Figure 5.2-2a, the "top" faces are highlighted, and in Figure 5.2-2b, the

"middle" faces are highlighted. If rays are launched at each of the twelve vertices, each ray wave-

lli-1 4

,.,

7 -cnp!Sf@'f'. $ C:''err?W• ftltAW¥oltW -rt1Htfebat·r... .. ;-• t'hte!¥MM'. WMittitdd' ·wt· .. ·ttztti'< ;re·: H MkaNre~· ... ~ ~.t·• .. aew t·#'e"'i+t,. · · · -·-.,;:, .. -~~·.-..aw..;.s~..-.;t.4h»w '* .te·~

a.) b.)

FIGURE 5.2-2. Two views of a regular icosahedron that is inscribed inside a unit sphere. Rays are launched at icosahedron vertices.

front is an identically shaped pentagon separated by 63 degrees from each of its five nearest

neighbors. A wavefronts 's nearest neighbors are the rays whose wavefront "patches" are adjacent

to the ray wavefront.

Since 63 degrees is not enough ray resolution for accurate propagation prediction, each tri­

angular face of the tetrahedron is subdivided as follows. Each triangular face is tessellated into N

equal segments where N is the tessellation frequency [22]. Figure 5.2-3 shows an example with

N=4. Lines parallel to one of the three sides are drawn that subdivide the triangle into smaller

equilateral triangles. Rays are launched at angles that pass through the vertices of the triangles.

Waverronts are hexagonal for rays that pass through interior and edge vertices. Rays that pass

through the twelve original tetrahedron vertices are pentagonal. Ray wavefronts are hexagonal

and pentagonal on the surface of the tetrahedron. As the tessellation frequency increases, the

wavefronts decrease in size, but keep their shape and relation to their nearest neighbors. Hexago­

nal rays have six nearest neighbors, and pentagonal rays have five. The angular separation

between a ray and its nearest neighbors is nearly identical for all nearest neighbors. However, the

angular separations are different for rays that originate from the three different types of vertices.

The total number of source rays traced from a point source is 10N2+2 [22].

Figure 5.2-4 shows a plot of minimum and maximum angular ray separation between

adjacent rays for the source rays as a function of the number of rays traced (tessellation fre-

1 f)-1 _I)

~ Original Vertex

Edge Vertex

1

Wavefront Interior Vertex

FIGURE 5.2-3. Tessellation of tetrahedron face. Ray wavefronts are hexagonal for edge and interior vertices.

quency). Notice that the angular separation decreases as the number of rays increases. For a max-

U) 100 4.) 4.) I... ~ 4.)

"0 Maximum ,_..

(I)

>--0

0:::: 10 c:: 4.) ll)

3: _. 4.) 3

CD c:

,Q _. 0 I... Minimum 0 0.. 4.)

Vl I... 3 0

"S ~ c:: 0.1 <(

101 102 2

103 2

104 2

1~

Number of Rays

FIGURE 5.2-4. Maximum and minimum ray separation in degrees between nearest neighbors as a function of the number of rays traced. The difference between the maximum and minimum is caused by the difference between the pentagonal and hexagonal shaped wavefronts.

imum angular separation of one degree between rays, 40,000 source rays must be traced over a

spherical region. This requires a tessellation frequency of 63. This method of launching the source

rays provides "tubes" of wavefronts that completely subdivide the surface of the unit sphere with

virtually identical shapes and areas.

1{..-lf..

I I ·:) I

I • ?'''?' urrmsttrr · Tl' ·a r awer"NStif ffiii'&pr.. ·' -ewsvn *' ·

5.3 Identification of Specular Rays

Consider a transmitter and receiver in the presence of a single planar scattering surface as

shown in Figure 5 .3-1. To determine the ray traced impulse response by a brute force method, it is

Tx

Rx

specularly reflected ray

FIGURE 5.3-1. Transmitter and receiver in the presence of a specularly reflecting surface. A ray tracing solution must include only one specularly reflected ray for a given surface.

necessary that the solution include only one specular ray regardless of how many rays are traced.

The test to determine if the ray reaches a receiving point is accomplished in the following fashion.

A perpendicular projection from the receiving location to the ray path is computed and the total

(unfolded) path length, d, that the ray travels from the transmitter to the projection point is deter­

mined. A reception sphere (from [23], extended to three dimensions [24]) is constructed about the

receiving location with a radius proportional to the unfolded path length and the angular spacing

between neighboring rays at the source. If the ray intersects the sphere, it is received and contrib­

utes to the total received signaL Otherwise, the ray is not received and the recurSion proceeds as

previously described. The reception sphere effectively accounts for the divergence of the rays

from the source. For ray separation a. sufficiently small, the ray intercepting the sphere will be an

accurate measure of the ray that would pass directly through the receiving point. The physical

interpretation of the reception sphere can be justified with the aid of Figure 5 .3-2. This figure is a

two-dimensional representation of a ray being traced. Two adjacent rays, launched at ±a. relative

to the test ray, are also shown. Note that in three dimensions, any ray will have more than two

adjacent rays and the angular separation of the adjacent rays will not necessarily coincide with the

coordinate axes. As shown in the figtire, a reception sphere with the correct radius can receive

only one of the rays. If the radius is greater than a.d/ J3, two of the rays could be received and

would, in effect, count the same specular ray path twice resulting in a 3 dB overestimate in power.

Likewise, if the radius is too small, none of the rays will intercept the sphere and the specular

16-17

su ..

;·,

\ Tx

Undersized reception sphere

Oversized reception sphere

sphere

ray

Adjacent ray

FIGURE 5.3-2. 1\vo dimensional view of the reception sphere. The total ray path length is d producing a reception sphere radius of ad/ )3.

energy will be excluded. It is important to emphasize that the reception sphere radius is propor­

tional to the unfolded path length from the source to receiver since the reception sphere will have

different radius for each received ray.

In order to make sure that the specular point in Figure 5.3-1 is not missed, the ray wave­

front is considered as a circle circumscribed about the hexagonal (or pentagonal) wavefront

shape. Figure 5.3-3 shows two adjacent rays and their respective wavefronts. The angular radius

FIGURE 5.3-3. Illustration of assumed wavefront for the purposes of the reception sphere. There is slight overlap between adjacent rays.

of the circumscribed circle is a/ fj. Although it is possible, but rare, to receive . two specular

points with this approach because of the slight overlap of adjacent rays, a missed specular point

will lead to a much larger error than if a ray is counted twice. Multiple specular rays are easily

eliminated in processing the raw ray traced impulse response data. For the reception sphere to be

r- ---•vs rwmm ··rrtWXr ts'f wr ,.,., - -US' . · ·rr m ez«n···wtt'ti'· ff't · ·- Srtn1W

valid in three dimensions, the rays must be launched such that each ray is separated from neigh-·

boring rays by a constant angle, a. If uniform separation is not maintained, the test r~y will not be

separated from adjacent rays by a constant angle and the reception sphere loses its physical signif­

icance.

5.4 Acceleration Techniques

. We use several techniques to decrease the amount of computation time required to predict

the propagation. The primary function of the ray tracer that must be computed over and over is an

intersection test to determine if a ray hits an object. As the number of objects increases linearly,

the number of intersection tests increases exponentially. To combat this inefficiency, objects are

placed in a hierarchy of bounding volumes. The intersection test is peiformed on the bounding

volume. If an intersection occurs, then intersection tests are performed on each object within the

bounding volume. If there is no intersection, then no subsequent intersection tests for these

objects need to be performed. By logically grouping closely spaced objects (and closely spaced

bounding volumes), the total number of intersection tests can be significantly reduced.

Another technique used to decrease computation time is thresholding. If the intensity of a

ray traced drops· below a certain threshold level, then it is dropped from the ray tree and is no

longer traced. Selecting a threshold is similar to having a receiver noise floor below which noth­

ing can be received. By selecting an appropriate threshold, accurate prediction can be accom­

plished without wasting time tracing many rays that do not contribute to the receive~ signal.

An input parameter is the tessellation frequency of the icosahedron. This determines the

number of rays traced, and hence the ray separation (resolution). One goal of this work is to deter­

mine the minimum resolution required to accurately predict the propagation, and determine how

the resolution affects the accuracy.

16-19

6.0 PREDICTED vs. MEASURED RESULTS

6.1 Microcellular Results

The following figures are examples of ongoing propagation prediction work at MPRG.

The figures show comparisons of predicted and measured power delay profiles taken throughout

/

Hancock Hall

Rx

FIGURE 6.1-1. Three dimensional building database for the academic quad on the Virginia Tech campus. The circles represent locations where 1900 MHz microcell measurements were made

I~-. ~-· ....

the academic quad of the Virginia Tech campus.

Figure 6.1-1 illustrates the locations where sample measurements (and predictions) were

made. These measurements were made with a 1900 MHz 230 Mcps spread spectrum system

developed within the MPRG. The spread spectrum system was used for these measurements pro­

vides fine time resolution (less than 9 nanoseconds for a complete· chip, and about a 4 ns rms chip

width) and has a dynamic measurement range comparable to that of a narrow band receiver (can

measure up to 120 dB of path loss). For the measurements, the transmitter was placed behind Bur­

russ Hall in a location typical of a microcell base station. An omnidirectional discone antenna was

placed at a height of 7.6 meters and had an ERP of 1 Watt. The receiver was moved to several

locations as shown on the figure. These locations were selected to provide specific test cases (e.g.,

line-of-sight and obstructed topographies) for the prediction program. The receiver used an omni­

directional antenna mounted at a height of 1. 7 meters.

(I)

> 0

..c 0

CD "'0 "-"

Cl) Cl)

0 _J

..c ..... c

a..

20

30

40

50

60

70

80 -so

Measured Impulse Response

Predicted Impulse Response

50 150 250 350 450 550

Excess Delay (ns)

FIGURE 6.1-2. Measured and predicted power delay profiles for the Hancock measurement site.

Figures 6.1-2 and 6.1-3 illustrate some of the measured and predicted results. These two

profiles are representative of the results generated at other locations. The propagation model

described in Section 4 was applied here, and reflected, diffracted, and scattered rays were consid:

ered. For the predictions, a 12 dB reflection loss, regardless of the incidence angle or building

material, was assumed. Additionally, a 9 ns base-width Gaussian pulse was convolved with the

predicted impulse response for comparison purposes. The profile in Figure 6.1-2 was generated

for the measurement location adjacent to Hancock Hall. As can be seen in the figure, the line-of­

sight component is accurately predicted and the amplitudes and time delays of several specular

reflections are also identified. The measured path loss and delay spread was 30.8 dB above 1

meter FSPL and 23.2 nanoseconds respectively. The predicted results were 33.1 dB and 44.8 ns

respectively. The profile in Figure 6.1-3 was generated for the measurement location in front of

C1) > 0 .a 0

CD -o '-"

(/) (I)

0 ...J .c ..... 0

a..

40

50

60

70

80

90

100 -50

Predicted Impulse Response

Measured Impulse Response

/

50 150 250 350 450 550

Excess Delay (ns)

FIGURE 6.1-3. Measured and predicted power delay profiles for the Norris measurement site.

Norris Hall. Notice that this is a shadowed geometry. Again, the first arriving component (in this

I I -I

I I

r : , I ,~,

· ·rre .. ft'W" . t IS

case a diffracted component) is accurately predicted and some specular reflections as well as scat- ·

tering is identified. The measured path loss and delay spread for this case is 45.7 dB and 69.2 ns

whereas for the prediction, the path loss and delay spread are 51.2 dB and 144.3 ns. For both mea­

surement sites it was possible to obtain an accurate path loss prediction (within 5 dB) while the

delay spread prediction had considerable error. This is due to components arriving at long delays

which heavily influence the delay spread calculation, ·even if these components are relatively low

in amplitude.

6.2 In-Building Results

The ray tracing program has been run with building blueprint information from the second

floor of Whittemore Hall on the Virginia Tech campus. Only the hallways, floor, ceiling, and drop

ceiling were included. Walls that separate individual offices were not included. In [25], extensive

measurements were made in Whittemore Hall with various antenna patterns and polarizations

with a 4 ns baseband pulse width. Figure 6.2-1 shows the floor plan where the measurements were

taken. Figure 6.2-2 and Figure 6.2-3 show measured data and the ray traced impulse response pre-

300 Whittfl'TTtore Ha.J.l I 8'- I rt4' i

rto' rto' MPRG

1 Cla.ss Rooms 14'-{Rx)

and Ma.ch:inny

Television 72' Studios 6·3' 58' .--...

48' (Rx 58' 68' 94' .__,

EE Shop

• (Tz

Offices 178'

10'J I 14 ,j

FIGURE 6.2-1. Floor plan of the second floor of Whittemore Hall. The transmitter and two receiver locations are circled to indicate where measurements and ray traced impulses are compared.

diction using the form of the models stated here. The transmitter and receiver locations ar~ indi­

cated in Figure 6.2-1. In Figure 6.2-2, the path loss of the first arriving component in a line-of­

sight environment is very nearly predicted and other multipath components can be seen. In Figure

lfl-23

en PM

20

..-. 30 _j

0..

~ 40

E ~ 50 0

] 60 Ill "'0 --;; 70

(/')

0 _j 80 ..c. -' 0

0.. 90

rm mrmr· . tlrWW

Measured Impulse Response

Predicted Impulse Response

100~~--------~~~~~~~--~~~~~

0 1 5 JO 45 60 75 90 1 05 1 20 1 J5 1 50 Total Delay (nanoseconds)

FIGURE 6.2-2. Measured data and a ray traced impulse response for a line-of­sight location in Whittemore Hall.

20

..-. 30 ...J Cl. l/) u.. 40 E ';:; 50 > 0 .g 60 Cil "0 ._. 70

(I) (I)

0 ...J eo :5 0

0.. 90

/ Predicted Impulse Response

Measured Impulse Response

too~~~~~~~~~--~~~~~~~

0 30 60 90 1 20 1 50 1 80 21 0 240 270 JOO Total Delay (nanoseconds)

FIGURE 6.2-3. Measured data and a ray traced impulse response for an obstructed location in Whittemore Hall.

6.2-3, an obstructed location is presented. Notice that the ray traced impulse response predicts less

path loss than the measured data. However,. the attenuation for transmitted rays has not yet been

implemented. Although the path loss of individual multipath components in Figure 6.2-3 are

incorrect, the time delays appear to be very close to those measured. The results of Figure 6.2-2

and Figure 6.2-3 show that with refinement, more detailed building information, and the incorpo-

, , ~ J.

ration of diffraction into the in-building simulation tool, that the geometrical optics ray tracing

model presented here holds promise for multipath impulse response prediction in buildings.

7.0 CONCLUSIONS

This paper has summarized our work in site-specific propagation prediction. As wireless

systems become broad band, the local average channel characteristics will be most important for

system design and installation, making site-specific techniques viable and necessary. We have

demonstrated how ray tracing can be implemented using a novel ''ray ~be" method. Ray tubes

with virtually identical angular separation and illumination area can be generated using an icosa­

hedron. Scattering can be accounted for using a discretized RCS technique. Through recursive

programming, the paths of all radio energy can be systematically tracked and used for system

design. While in its infancy, the field of site-specific propagation appears to hold promise for pro­

viding significant improvements in predicting coverage, as shown here by measurements on the

Virginia Tech campus which are consistently within a few dB of those measured.

8.0 REFERENCES

[1]

[2]

[3] .

[4]

[5]

[6]

D.C. Cox, R.R. Murray, and A.W. Nonis, "800 MHz Attenuation Measured In and Around Suburban Houses," AT&T Bell Laboratories Technical Journal, Vol. 63, July/ August, 1984, pp. 921-954.

S.E. Alexander, "900 MHz Propagation Within Buildings," lEE 2nd Inti. Conf on Radio Spectrum Conservation Techniques, Binningham, UK, 6-8 Sept. 1983, pp. 51-55.

D.M.J. Devasirvatham, "Multipath Time Delay Jitter Measured at 850 MHz in the Porta­ble Radio Environment," IEEE Journal on Selected Areas in Communications, Vol. SAC-5, No. 5, Jun. 1987, pp. 855-861.

T.S. Rappaport, "Characterization of UHF Multi path Radio Channels in Factory Build­ings," IEEE Transactions on Antennas and Propagation, Vol. 37, No. 8, August 1989, PP· 1058-1069.

T.S. Rappaport, S.Y. Seidel, and K. Takamizawa, "Statistical C~~el Impulse Resp.ons~ Models for Factory and Open. Plan Building Radio Commurucatlon System Design, IEEE Transactions on Commuriications, Vol. 39, No.6, May 1991, PP· 794-897.

R. J. C. Bultitude and G. K. Bedal, "Propagation characteristics on rnicrocellular urban mobile radio channels at 910 MHz," IEEE J. Select. Areas Comm., Vol. 7, PP· 31-39, Jan. 1989

16-25

-·, ,.

[7]

[8]

[9]

[10]

[11]

[12]

[13]

[14]

[15]

[16]

[17]

[18]

[19]

a . ,. W't" t). i't "S •. , 1 ·= .•. w

J.-F. Lafortune, and M. Lecours, "Measurement and Modeling of Propagation Losses in a Building at 900 MHz,'' IEEE Transactions on Vehicular Technology, Vol. 39, No. 2, May 1990, pp. 101-108.

S.Y. Seidel and T.S. Rappaport, "914 MHz Path Loss Prediction Models for Wrreless Communications in Multifloored Buildings," IEEE Transactions on Antennas and Propa­gation, Vol. 40, No.2, Feb. 1992, pp. 207-217.

A.J. Motley, and J.M. Keenan, "Radio coverage in buildings," British Telecom Technology Journal, Special Issue on Mobile Communications, Vol. 8, No. 1, Jan. 1990, pp. 19-24.

S. Y. Seidel, et. al., "The Impact of Surrounding Buildings on Propagation for Wrreless In­Building Personal Communications Systems Design", 42nd IEEE Vehicular Technology Conference, Denver, May 1992, pp. 814-818.

T. S. Rappaport, and L.B. Milstein, "Effects of Path Loss and Fringe User Distribution on CDMA Cellular Frequency Reuse Efficiency," IEEE Global Communications Conference, San Diego, CA, December 1990, pp. 500-506, also accepted to IEEE Transactions on Vehicular Technology.

J. M. Holtzman and L. M. A. Jalloul, "Rayleigh Fading Effect Reduction with Wideband DS/CDMA Signals," in IEEE Globecom Conference, Phoenix, AZ, pp. 16.7.1 - 16.7.5, Dec. 1991.

R. L. Pickholtz, D. L. Schilling, and L. B. Milstein, "Spread Spectrum for Mobile Com­munications," IEEE Transactions on Vehicular Technology, Vol. 40, No.2, May 1991.

F. Ikegami, T. Takeuchi, and S. Yoshida., "Theoretical prediction of mean field strength for urban mobile radio," IEEE Transactions on Antennas and Propagation, Vol. 39, pp. 299-302, Mar. 1991.

A.J. Rustako, Jr., N. Amitay, G. J. Owens, and R. S. Roman, "Radio Propagation at Micro­wave Frequencies for Line-of-Sight Microcellular Mobile and Personal Communica­tions," IEEE Transactions on Vehicular Technology, Vol. 40, pp. 203-210, Feb 1991.

T. A. Russell, T.S. Rappaport, and C. W. Bostian, "Use of a Building Database in Predic­tion of Three Dimensional Diffraction," 42nd IEEE Vehicular Technology Conference, Denver, May 1992, pp. 943-946.

K. R. Schaubach, "Microcellular Radio Channel Prediction Using Ray Tracing," Master's Thesis in Electrical Engineering, Virginia Polytechnic Institute and State University, forthcoming.

J. W. McKown and R.L. Hamilton, "Ray Tracing as a Design Tool for Radio Networks," IEEE Network Magazine, Vol. 5, No.6, Nov. 1991, pp. 27-30.

Telesis Technologies Laboratory, Experimental License Report to the FCC, Aug. 1991.

lh-?h

I ~.~ 1 ,·",

\)

[20]

[21]

[22]

[23]

[24]

[25]

[26]

'7$ n , a

T.S. Rappaport, "The Wireless Revolution," IEEE Communications Magazine, Vol. 29, No. 11, Nov. 1991, pp. 52-71.

A.S. Glassner, Ed., An Introduction to Ray Tracing, Academic Press, New York, 19~9.

H. Kenner, Geodesic Math and How to Use It, University of California Press, Berkeley, 1976.

W. Honcharenko, H.L. Bertoni, J. Dailing, J. Qian, and H.D. Yee, "Mechanisms Govern­ing UHF Propagation on Single Floors in Modern Office Buildings," submitted to IEEE Transactions on Vehicular Technology, Sep. 1991.

K.R. Schaubach, N.J. Davis IV, and T.S. Rappaport, "A Ray Tracing Method for Predict­ing Path Loss and Delay Spread in Microcellular Environments," 42nd IEEE Vehicular Technology Conference, Denver, May 1992, pp. 932-935.

T.S. Rappaport and D.A. Hawbaker, "Wide-Band Microwave Propagation Parameters Using Circular and Linear Antennas for Indoor Wireless Channels," IEEE Transactions on Communications, Vol. 40, No. 2, Feb. 1992, pp. 240-245.

R.T. Stevens, Fractal Programming and Ray Tracing with C++, M&T Books, Redwood City, CA, 1990.

-Mt-