A Survey of Various Propagation Models

for Mobile Communication

Tapan K. Sarkar', Zhong Ji', Kyungjung Kim', Abdellatif Medour?, and
Magdalena Salazar-Palma3
'Department of Electrical Engineering and Computer Science
Syracuse University
121 Link Hall, Syracuse, New York 13244-1240, USA
Tel: 1 (315) 443-3775; Fax: 1 (315) 443-4441; E-mail: [email protected]

'Ecole Nationale des Sciences Appliquees

Abdelmalek Essadi University
Tangier, Morocco
E-mail: [email protected]
3
Grupo de Microondas y Radar, Dpto. Senales, Sistemas y Radiocommunicacion,ETSl Telecommunication
Universidad Politecnica de Madrid
Ciudad Universitaria, 28040 Madrid, Spain
E-mail: [email protected]

Abstract
In order to estimate the signal parameters accurately for mobile systems, it is necessary to estimate a system's propagation
characteristics through a medium. Propagation analysis provides a good initial estimate of the signal characteristics. The abil-
ity to accurately predict radio-propagation behavior for wireless personal communication systems, such as cellular mobile
radio, is becoming crucial to system design. Since site measurements are costly, propagation models have been developed
as a suitable, low-cost, and convenient alternative. Channel modeling is required to predict path loss and to characterize the
impulse response of the propagating channel. The path loss is associated with the design of base stations, as this tells us how
much a transmitter needs to radiate to service a given region. Channel characterization, on the other hand, deals with the
fidelity of the received signals, and has to do with the nature of the waveform received at a receiver. The objective here is to
design a suitable receiver that will receive the transmitted signal, distorted.due to the multipath and dispersion effects of the
channel, and that will decode the transmitted signal. An understanding of the various propagation models can. actually
address both problems. This paper begins with a review of the information available on the various propagation models for
both indoor and outdoor environments. .The existing models can be classified into two major classes: statistical models and
site-specific models. The main characteristics of the radio channel - such as path loss, fading, and time-delay spread - are
discussed. Currently, a third alternative, which includes many new numerical methods, is being introduced to propagation
prediction. The advantages and disadvantages of some of these methods are summarized. In'addition, an impulse-response
characterization for the propagation path is also presented, including models for small-scale fading. Finally, it is shown that
when two-way communication ports can be defined for a mobile system, it is possible to use reciprocity to focus the energy
along the direction of an intended user without any explicit knowledge of the electromagnetic environment in which the system
is operating, or knowledge of the spatial locations of the transmitter and the receiver.

Keywords: Land mobile radio cellular systems; land mobile radio propagation factors; communication channels; multipath
channels; fading channels; transient response; reciprocity

1. Introduction

T. he. commercial
. success of cellular communications, since its
initial implementation in the early 1980s, has led to an intense
capability of determining optimum base-station locations, obtain-
ing suitable data rates, and estimating their coverage, without con-
interest among wireless engineers in understanding and predicting ducting a series of propagation measurements, which are very
radio-propagation characteristics in various urban and suburban expensive and time consuming.It is therefore important to develop
areas, and even within buildings..As the explosive gowtb of effective propagation models for mobile communications,in order
mobile communications continues, it is very valuable to have the to provide design guidelines for mobile systems.

/€€€Antennas and Propagation Magazine, Vol. 45, No. 3,June 2003 51

 2. Definitions and Terminology used for Multipath propagation causes severe dispersion of the trans-
mitted signal. The expected degree of dispersion is determined
Characterizing Various Parameters of a through the measuement of the power-delay profile of the chan-
Propagation Channel nel. The power-delay profile provides an indication of the disper-
sion or distribution of transmitted power over various paths in a
In order to understand the nature ofthe models that are going multipath model for propagation. The power-delay profile of the
2
to be presented, several definitions and the terminology for both channel is calculated by taking the spatial average of Ih(t)l over a
narrowband and wideband wave propagation over a radio channel
2
are first described, in order to familiarize the reader with the termi- local area. By making several local-area measurements of Ih(f)l
nology and parameters of the problem. for different locations, it is possible to build an ensemble ofpower-
delay profiles, each one representing a possible small-scale multi-
path channel state [3, IO]. A typical plot ofthe power-delay profile
is shown in Figure 1.
2.1 Path Loss
Many multipath-channel parameters are derived from the
Path loss (PL)is a measure of the average RF attenuation power-delay profile. Power-delay profiles are measured using
suffered by a transmitted signal when it arrives at the receiver, wideband channel-sounding techniques, and are presented in the
after haversing a path of several wavelengths. It is defined by [I] form of plots of the received power as a function of an additional
or excess delay with respect to a fixed time-delay reference. There
is a delay between the time of transmission of the signal and when
(1) it is received, due to the finite velocity of propagation of the elec-
tromagnetic signal. However, additional delay may be introduced
by the propagation medium, as well. A mobile channel exhibits a
where pr and P, are the transmitted and received power, respec-
continuous multipath structure; hence, the power-delay profile can
tively. In free space, the power reaching the receiving antenna - he thought of as a density function, ofthe form
which is separated from the transmitting antenna by a distance d -
is given by the Friis free-space equation:
(4)
Pp=(d)
PGG
' L~
(4x)2 d 2 L '

where G, and C, are the gain of the transmitting and the receiving
antenna, respectively. L is the system loss factor, not related to
2.3 Time-Delay Spread
propagation. Z is the wavelength in meters. It is clear that Equa-
tion (2) does not hold for d = 0 . Hence, many propagation models Time dispersion varies widely in a mobile radio channel, due
use a different representation for a close-in distance, d o , known as to the fact that reflections and scattering occur at seemingly ran-
dom locations, and the resulting multipath channel response
the received-power reference point. This is typically chosen to be
appears random, as well. Because time dispersion is dependent on
1 m. In realistic mobile radio channels, free space is not the appro-
the geometrical position relationships among the transmitter, the
priate medium. A general PL model uses,a parameter, y , to denote
receiver, and the surrounding physical environment, some
the power-law relationshipbetween the separation distance and the parameters that can grossly quantify the multipath channel are
received power. So, path loss (in decibels) can'be expressed as [Z] used. They are described in the following subparagraphs.

PL (d) = PL (do) + 1Oy log(d/do) + X , , (3)

where y = 2 characterizes free space. However, y is generally

higher for wireless channels.:'A denotes a zero-mean Gaussian
random variable of standard deviation 0 , which reflects the varia-
tion, on average, of the received power that naturally occurs when
a PL model of this type is used. Path loss is the main ingredient of
a propagation model. It is related to the area of coverage of mobile
systems.

2.2 Power-Delay Profile

Random and complicated radio-propagation channels can be

characterized using the impulse-responseapproach. For each point
in the three-dimensional environment, the channel is a linear filter
with impulse response h ( t ) . The impulse response provides a
wideband characterizationof the propagating channel, and contains
all of +e information necessary to simulate or analyze any type of Figure 1. An illustration of a typical power-delay profile and
radjo transmission through that channel. the definition of the delay parameters.

52 IEEEAnfennas and Propagaton Magazine, Vol. 45. No 3, June 2003

 2.3.1 First-Arrival Delay ( z A ) (7)
This is a time delay corresponding to the amval of the first
transmitted signal at the receiver. It is usually measured at the It is important to note that an exact relationship between the coher-
receiver. This delay is set hy the minimum possible propagation- ence bandwidth and the RMS delay spread does not exist. The real
path delay from the transmitter to the receiver. It serves as a refer- coherence bandwidth depends on the actual impulse response of
ence, and all delay measurements are made relative to it. How the the channel.
origin is defined is seen in Figure I. Any measured delay longer
than this reference delay is called an excess delay.
2.5 Types of Fading
2.3.2 Mean Excess Delay ( z,) The type of fading experienced by a signal propagating
through a mobile radio channel depends on the nature of the
This is the first moment of the power-delay profile (as shown transmitted signal, as well as on the characteristics of the channel.
Different transmitted signals will undergo different types of fading,
in Figure 1) with respect to the first delay. It is expressed as
according to the relationship among the signal parameters, such as
the path loss, the bandwidth (BW), the symbol period, etc., and the
channel parameters (such as the RMS delay spread and the
Doppler spread). Figure 2 describes the different types of fading
and the different relationships that exist among them [ 5 ] .

2.3.3 RMS Delay (zms) The phenomenon of large-scale fading is affected primarily
by the presence of hills, forests, and buildings between the trans-
mitter and the receiver. The statistics of large-scale fading provide
This is the square root of the second central moment of a a way of computing an estimate of the path loss as a function of
power-delay profile, as seen in Figure 1. It is the standard devia- distance and other factors.
tion about the mean excess delay, and is expressed as
A channel is said to exhibit frequency-selective fading when
the delay spread is greater than the symbol period. This condition
occurs whenever the received multipath components of a symbol
extend beyond the time duration of the symhols. Such multipath
The RMS delay is a good measure of the multipath spread. It gives dispersion of the signal yields a kind of inter-symbol interference
an indication of the nature of the inter-symbol interference (ISI). (ISI) called channel-induced ISI. When the delay spread is less
Strong echoes (relative to the shortest path) with long delays con- than the symbol period, a channel is said to exhibit flat fading, and
tribute significantly to ‘-8. The effects of dispersion on the per- there is no channel-induced IS1 distortion. But there can still be
formance of a digital receiver can he reliably related only to it, performance degradation, due to the irresolvable pbasor compo-
independently of the shape of the power-delay profile, so long as it nents that add up destructively to yield a substantial reduction in
is small compared to the symbol period (n of the digital modula- signal-to-noise ratio ( S N R ) at the receiver.
tion It is also used to give an estimate of the maximum data rate
for transmission. Fast fading and slow fading are classified on the hasis of how
rapidly the transmitted baseband signal changes, compared to the
rate of the electrical-parameter changes of the channel. If the chan-
nel impulse response changes at a rate much faster than the trans-
2.3.4 Maximum Excess Delay (7,) mitted signal, the channel may he assumed to he a fast-fading
channel. Otherwise, it is assumed to be a slow-fading channel. It is
important to note that the velocity of the mobile unit or the velocity
This is measured with respect to a specific power level, of objects using the channel through a baseband signal determines
which is characterized as the threshold of the signal. When the sig- whether a signal undergoes fast fading or slow fading.
nal level is lower than the threshold, it is processed as noise. For
example, the maximum excess delay spread can be specified as the
excess delay (r,,,) for which P(r) falls helow -30 dB with respect
2.6 Adaptive Antennas
to its peak value, as shown in Figure 1.
An application of antenna arrays has been suggested in recent
years for mobile communications systems, to overcome the prob-
2.4 Coherence Bandwidth lems of single-antenna systems. The use of adaptive antenna arrays
helps improve the system’s performance by increasing channel
capacity and spectrum efficiency, extending range coverage, tai-
While the delay spread is a natural phenomenon caused by loring beam shape, steering multiple beams to track many mobiles,
the reflection and scattering of the transmitted signal in a radio and electronically compensating for the aperture distortion. It also
channel, the coherence bandwidth, Bc , is defined in terms of the reduces multipath fading, co-channel interference, system com-
RMS delay spread. It is a statistical measure of the range of &e- plexity and cost, hit-error rate (BER), and outage probability [6].
quencies over which the channel can be considered “flat.” It is
defined as the bandwidth over which the variation of the signal is A phased-array antenna uses an array of simple antennas, and
about IO%, and is approximated by [4, IO] combines the signal induced on the elements to form the output.

IEEE Antennas and Propagation Magazine, Vol. 45, No. 3, June 2003 53
 1 Pading channel
manifestations

Large-scale fading due to Small-scale fading due to Small

of the signal of t
k channel
Vs distance

1 Timedelay
domain Fourier 4'zZy-I I 1 Timedomain I I Doppler-shift I
description UanSfOnnS description

Frequency- Flat / Fast Slow

selective wing

!' Duals :

The term "adaptive antenna" is used for a phased array when the The channel impulse response can thus he characterizedby N time-
gains and the phases of the signals induced in the various elements delayed impulses, each represented by an attenuated and phase-
are weighted before combining to adjust the gain of the array in a shifted version of the original transmitted impulse. Here, A., ra ,
dynamic fashion along a particular look direction, while simulta- and qn are the attenuation, delay in time of arrival, and phase,
neously placing nulls along undesired directions.
corresponding to path n, respectively.
The propagation models used for an adaptive antenna are dif-
Although multipath interference seriously degrades the per-
ferent from those for a single antenna. Details of a phased-array
formance of communication systems, little can he done to elimi-
antenna can he obtained in [6,7].
nate it. However, if we characterize the multipath medium well,
and have sound knowledge of the propagation mechanisms and
their influence on the system, the best design for the system can be
selected to achieve good propagation performance and, hence, to
3. Multipath Propagation achieve a better quality of service.

In a typical mobile-radio application, the base station is fixed

in position, while the mobile unit is moving. This is usually subject
to a condition such that the propagation between them is largely 3.1 Three Basic Propagation Mechanisms
through scattering, either by reflection or diffraction from build-
ings and terrain or objects within buildings, because of obstruction Reflection, diffraction, and scattering are the three basic
of the line-of-sight (LOS) path. Radio waves therefore amve at the propagation mechanisms [lo] that impact propagation in mobile
mobile receiver from different directions with different amplitudes, communication systems. They are briefly explained below.
phases, and time delays, resulting in a phenomenon known as mul-
tipath propagation. The radio channel is then obtained as the sum
of the contributions from all of the paths. If the input signal is a
unit impulse, 6 ( t ) , the output will be the channel impulse
3.1.1 Reflection
response, which can be written as [9]
Reflection occurs when a propagating electromagnetic wave
impinges upon an object that has very large dimensions compared
h(t)= cN

"=I
A d ( r -rn )..P(-iPn). (8) to the wavelength of the propagating wave. Reflection occurs from
the surface of the ground, from walls, and from furniture. When

54 IEEEAntennas and propagation Magazine, Vol. 45, No. 3, June 2003

reflection occurs, the wave may also be partially refracted. The 3.2 Propagation in Outdoor and
coefficients of reflection and refraction are functions of the mate- Indoor Environments
rial properties of the mrdium, and generally depend on the wave
polarization, the angle of incidence, and the frequency of the
propagating wave. With the growth in the capacity of mobile communications,
the size of a cell is becoming smaller and smaller: from macrocell
to microcell, and then to picocell. The service environments
3.1.2 Diffraction include both outdoor and indoor areas.

When propagation is considered in an outdoor environment,

Diffraction occurs when the radio path between the transmit- one is primarily interested in three types of areas: urban, suburban,
ter and receiver is obstructed by a surface that has sharp edges. The and nual areas. The temain profile of a particular area also needs to
waves produced by the obstructing surface are present throughout be taken into account. The terrain profile may vary from a simple,
space and even behind the obstacle, giving rise to the bending of curved Earth to a highly mountainous region. The presence of
waves around the obstacle, even when a line of sight (LOS) path trees, buildings, moving cars, and other obstacles must also be
does not exist between the transmitter and receiver. At high fre- considered. The direct path, reflections from the ground and
quencies, dimaction - like reflection - depends on the geometry of buildings, and dimaction from the comers and roofs of buildings
the object, as well as on the amplitude, phase, and polarization of are the main contributions to the total field generated at a receiver,
the incident wave at the point of diffraction. due to radio-wave propagation.

With the advent of personal communication systems (PCS),

there is also a great deal of interest in characterizing radio propa-
3.1.3 Scattering gation inside buildings. The indoor radio channel differs from the
traditional outdoor mobile radio channel in two aspects: the dis-
Scattering occurs when the medium through which the wave tance covered is much smaller, and the variability of the environ-
propagates consists of objects with dimensions that are small com- ment is much greater for a much smaller range of transmitter and
pared to the wavelength, and where the number of obstacles per receiver separation distances [ 101. Propagation into and inside
unit volume is large. Scattered waves are produced by rough sur- buildings has, to some extent, a more complex multipath struchue
faces, small objects, or by other irreguleties in the channel. In than an outdoor propagation environment. This is mainly because
practice, foliage, street signs, lampposts, and stairs within build- of the nature of the structures used for the buildings, the layout of
ings can induce scattering in mobile-communication systems. A rooms and, most importantly, the type of construction materials
sound knowledge of the physical details of the objects can be used used. Table 1 represents the categories of buildings where propa-
to accurately predict the scattered signal strength. gation measurements have been made [13].

In most cases, the scattering can be neglected [ 1I], and the

complex field received from the various incident paths is given by
DZl 3.3 Summary of Propagation Models

In mobile communications, signals from the mobile unit

(9)
amve at a base station via multiple paths, each with its own angle
of arrival (AOA), path delay, and attenuation. When the communi-
where cation system uses an adaptive-processing methodology, it is also
important to estimate the joint angle and delay for various signals.

There are two main models for characterizing path loss:

i P empirical (or statistical) models, and site-specific [or deterministic)
models. The former are based on the statistical characterization of
and the received signal. They are easier to implement, require less
computational effort, and are less sensitive to the environment’s
Eo is the amplitude of the reference incident field (Vim); 631, and
pti are the transmitting- and receiving-antenna field radiation pat-
terns along the direction of interest for the ith multipath compo-
nent; Li ( d ) is the path-loss Characterization for the ith multipath Table 1. Classification ofbuildings.

component; T(4ji) is the reflection coefficient for the jth reflec- I Proposed Categories I
tion of the ith multipath component; T(&,) is the transmission
coefficient for thepth transmission of the ith multipath component;
e-W is the propagation phase factor due to the path length d
( k = 2 n / A , with A representing the wavelength); d is the path
length (m); and E, is the field strength of the ith multipath compo-
nent.

For dimaction, the product of the complex reflection and I centers

transmission coefficients is replaced by the complex diffraction 7 I Open environment, e.g. railway stations, airports, etc.
coefficient. 8 I Underground, e.g. subways, underground streets, etc.
IEEE Antennas and Propagalion Magazine, Vol. 45, NO.3, June 2003 55
geometry. The latter have a certain physical basis, and require a among the simplest and best in terms of accuracy in predicting
vast amount of data regarding geometry, terrain profile, locations path loss for early cellular systems. It is very practical, and has
of building and of furniture in buildings, and so on. These determi- become a standard for system planning in Japan. The major disad-
nistic models require more computations, and are more accurate. vantage of this model is its slow response to rapid changes in ter-
rain profile.
Most models of fading apply stochastic processes to describe
the distribution of the received signal. It is useful to use these
models to simulate propagation channels and to estimate the per-
formance of the system in a homogeneous environment. Models of 4.1.I.2 Hata Model
time-delay spread, both for outdoor and indoor environments, are
generally derived from a lot of measurements. Some of the propa- The Hata model is an empirical formulation [15] of the
gation models have been discussed in [142], and here we have graphical path-loss data provided by Okumura’s model. The for-
included additional, new models. mula for the median path loss in urban areas is given by

There is another parameter that is sometimes seen in the vari- Lso(urban)( dB) = 69.55 + 26.16logfC - 13.82log hte
ous models: this is the effect of the Doppler frequency on the
channel characterization. The thinking is that since the mobile -a(h,) + (44.9 - 6.5510ghle)logd,
wireless receiver is moving, the Doppler shift is relevant to the (13)
model. If a mobile unit is operating at, say, 900 MHz, and is trav-
eling at 150 km/hr,then we h o w that in this case the Doppler shift where f, is the frequency (in MHz), which varies from 150 MHz
will he given by
to 1500MHz. heand h,eare the effective heights of the base-
station and the mobile antennas (in meters), respectively. d is the
2 x 150000 x 900 x lo6 = 250 H z ,
fdoppfer = (11) distance from the base station to the mobile antenna, a(hre) is the
3600 x 3 x IO8
correction factor for the effective antenna height of the mobile
unit, which is a function of the size of the area of coverage. For
It would be useful to study why such a small value of the Doppler small- to medium-sized cities, the mobile-antenna correction factor
sluft will have an impact on the processing of a modulated
is given by
900 MHz signal.
a (h,e) = (I. 1logf, - 0.7) h, - (1.5610gf~ - 0.8) dB . (14)

4. Models for Path Loss For a large city, it is given by

4.1 Empirical or Statistical Models a (h,) = 8.29(log 1.54h,$ - 1.1dB for f, < 300 MHz (1 5.a)
for Path Loss
a (h, ) = 3.2 (log1 1.75h,)’ - 4.97 dB for f, L 300 MHz (l5.b)
4.1.1 Outdoor Case
To obtain the path loss in a suburban area, the standard Hata for-
mula is modified as follows:
There are a number of empirical or statistical models suitable
for both macrocell and microcell scenarios for the outdoor envi-
ronment. Some of these are described below. Lso (dB)= Lso(urban- 2[10g(fc/28)]’ - 5.4. (16)

The path loss in open ~ r aareas

l is expressed through

4.1.1.1 Okumura et al. Model Lso(dB)= Go(urban) -4.78(l0g/,)~ - 18.331ogfC-40.98. (17)

This is one of the most widely used models for propagation
in urban areas [14]. The model can he expressed as This model is quite suitable for large-cell mobile systems, but not
for personal communications systems that cover a circular area of
approximately 1 km in radius.
L w ( ~ B ) = L F+ 4m (f,d)-G(hi~)-C(h,)-GA,Ru 3 (12)

where Lso is the median value of the propagation path loss, L, is 4.1.1.3 COST-231-Walfisch-lkegami Model
the free-space propagation loss, A,, is the median attenuation in
the medium relative to free space at frequencyf; and d corresponds This model utilizes the theoretical Walfisch-Bertoni model
to the distance between the base and the mobile unit. G(hte) and [16], and is composed of three terms [17]:

G(h,) are the gain factors for the base-station antenna and the
mobile antenna, respectively. h,, and h,e are the effective heights
of the base-station and the mobile antennas (in meters), respec-
tively. C A , is the gain generated by the environment in which where I,, represents the free-space loss. Lrlr is the “roof-top-to-
the system is operating. Both A,, (/,d) and GAmAcan be found street diffraction and scattering loss.” Lmd is the “multi-screen
from empirical curves. Okumura et al.’s model is considered to he diffraction loss.” The free-space loss is given by

56 IEEE Antennas and Propagation Magazine. Vol. 45, No. 3, June 2003
 = 32.4+ 201ogd + 201ogf, (19) for metropolitan centers

where d is the radio-path length (in km), f is the radio frequency (in This model is being considered for use by the International
Telecommunication Union Radiocommunication Sector (ITU-R)
MHz), and in the International Mobile Telecommunications - 2000 (IMT-
2000) standards activities. Some improved solutions for diffraction
L,, = - I 6.9-10log w+lOlog f + 20log AhMobile + Lon. (20) by multiple absorbing half-planes have also been developed [18,
191. Other solutions, based on the Uniform Theory of Diffraction
Here, wis the street width (in m), and (UTD), are also available [20, 211. The performance of the differ-
ent methods in estimating the multiple-diffraction loss term and the
AhMobile hRoo/ - hMobiie (21) final diffraction loss term is given in [22]. Recently, a correction to
the COST-231-Walfisch-IkegamiModel has also been reported
is the difference between the height of the building on which the ~231.
base-station antenna is located, hRoof, and the height of the mobile
antenna, hMobire.Lori is given by
4.1.I
.4 Dual-Slope Model

I -IO+ 0.3544
Lo,i= 2.5+0.075()-35)
4.0-0.1 14(( - 5 5 )
for
0" -<(< 35"
3S"<(<5S0, (22)
5S0-<4 < 90"

where ( is the angle of incidence relative to the direction of the

street. Lmsd is given by
This model is based on a two-ray model [24, 251, which is
used commonly when the transmitting antenna is several wave-
lengths or more above the horizontal ground plane. It is suitable for
the line-of-sight (LOS) propagation regions. The propagation loss,
L(d), in this case is descrihed'bya dual-slope model. This can he
representdd LS a function of d, the distance between the base sta-
tion and the receiver. It is given by [26]
Lmsd = Lb>h+ k, + kd log d + k, log f -910gb , (23)
10n, log d + F j I < d < dbrk
L(d) = Lb +
{lO(nI -n2)logdbrk+10nzlogd+Fj d2dbrk '
where 6 is the distance between the buildings along the signal path.
Lbsh and k, represent the increase of path loss due ta a reduced (28)
base-station antenna height. Using the abbreviation
where P, = PL(do), the path loss in dB at the reference point, do.
Ab,, = hB, - hRoof > (24) dbrk represents the breakpoint or the tuming-point distance. The
'point" where this transition occurs is often called the Fresnel
where hB,, is the base-station antenna height, we observe that breakpoint. Lb is a basic transmission-loss parameter that depends
LbXhand k, are given through on frequency and the antenna heights, and n1 and nz represent the
slopes of the bestlfit line before and after the breakpoint. If the
transmitter and receiver antenna heights are known, along with the
distance hitween them, then the path loss can he computed, based
on the two parameters n, and n 2 . It is very reasonable to let
n, = 2 for the region prior to the Fresnel breakpoint. There is much
more variability in the path loss and the exponent for the region
beyond the Fresnel breakpoint, with values of nz ranging from two
to seven.

The terms kd and k/ control the dependence of the multi-screen

diffraction loss as a function of distance, and the radio frequency 4.1.1.5 Other Models
of operation, respectively. They are given hy
0 t h models,
~~ including the use of wideband measurements
for different situations, have been discussed in recent times [27,
28, 1411. These models have been developed from measurements,
and use different parameters for different situations.

and
4.1.2 Indoor Case

Indoor radio propagation is not influenced by the terrain pro-

for medium-sized cities and suburban centers with moderate tree file, as is ourdoor propagation, but it can be affected by the layout
densities; and by in a building, especially if there are Vkious building materials. The
transmitted signal often reaches the receiver through more than one
k, = -4 +I.S(&- 11, path, due to reflection, refraction, and diffraction of the radio wave
by objects such as walls, windows, and doors inside a building.

IEEEAntennas and Propagation Magazine, Vol. 45, No. 3, June 2003 57

 Table 2. The path loss measured in different buildings.

I
Building
, ,_-,,
Frequency
, 1, Ref. I
I

Suburban office building soft partition I 2.8 I 14.2 I 915

Suburban ofice building soft partition I 3.8 I 12.7 1 1900 I [391
Suburban office building hard partition I 3 I - 850 I [40]

Table 3. The average floor attenuation factor (FAF). fashion, less power will be transmitted through the wall than would
occur at normal incidence. Reference [41] modifies the W A F ( p )
term to W A F ( ~ ) / C O S # ~ and the FAF(q) term to
FAF(q)/cosbq, where W A F ( p ) and FAF(q) are the values of
the attenuation factors at normal incidence, and bp and @q are the
angles of incidence of the signal on the walls and floors, respec-
tively. A difiaction term was also added to the formula in [41].
When the base station is out of the building, the path loss in the
buiIding was given in reference [43].

The empirical or statistical models described in this section

are simple to implement, and are widely used when the accuracy of
the data is not a critical requirement.

The distancdpower model is the main propagation model for

path loss. As shown in Equation (3), many researchers estimate the 4.2 Site-Specific Models for Path Loss
rate of decay of a transmitter signal through this relation [29-341.
In an enclosed environment, the value of y in Equation (3) may be
Site-specific propagation models, also called deterministic
1.5 to 1.8 when the transmitter and receiver are placed in the same models, are based on the theory of electromagnetic-wave propaga-
hallway and are in sight of each other. When the receiver is located tion. Unlike statistical models, site-specific propagation models do
within a room off the hallway, y ranges from three to four [30-331. not rely on extensive measurements, but on knowledge of greater
y also varies with frequency [33], and is dependent on the build- detail of the environment, and they provide accurate predictions of
ing materials used in a particular environment [34]. References [SI the signal propagation.
and [32] provide reviews of early work on these models. Table 2
[29] shows the parameters of Equation (3), determined by meas- In theory, the propagation characteristics of electromagnetic
urements for different buildings [35-40]. waves could he computed exactly by solving Maxwell’s equations.
Unfortunately, this approach requires very complex mathematical
In order to take into account the attenuation due to walls and operations and requires considerable computing power. In refer-
floors, two additional terms are added to Equation (3), resulting in ence [44], this method has been applied to simplified environ-
[35,411 ments.

P L ( d ) = PL (do)+ 10n log(d/do) + 2

q=l
FAF (4)+
P
2 WAF (p),
p 4 4.2.1 Ray-Tracing Technique
(29)
Ray tracing is a techniqne based on Geometrical Optics (GO)
where FAF(q) and WAF(p) are the floor and the wall attenua- that can easily he applied as an approximate method for estimating
tion factors, respectively. Table 3 lists the FAF values for two the levels of high-frequency electromagnetic fields. GO assumes
buildings [29,30]. that energy can be considered to be radiated through infinitesi-
mally small tubes, often called rays. These rays are normal to the
It is bas been observed that the propagation path loss as a surface of equal signal power. They lie along the direction of
function of the distance also has two distinct regions for indoor propagation and travel in straight lines, provided that the relative
environments [42], as described in Equation (29). When electro- refractive index of the medium is constant. Therefore, signal
magnetic radiation is incident on a wall or a floar in an oblique propagation can be modeled via ray propagation. By using the

58 IEEE Antennas and Propagation Magazine, Vol. 45, No, 3, June 2003
concept of ray tracing, rays can be launched &om a transmitter building building
location, and the interaction of the rays can be described using the
well-known theory of refraction and reflection and interactions
with the neighboring environment. In GO, only direct, reflected,
and refracted rays are considered. Consequently, abrupt hansition
areas may occur, corresponding to the boundaries of the regions
where these rays exist. The Geometrical Theory of Difiaction
(GTD)and its uniform extension, the Uniform GTD (UTD) [45],
complement the GO theory by introducing a new type of rays,
h o w n as the diffracted rays. The purpose of these rays is to
remove the field discontinuities and to introduce proper field cor-
rections, especially in the zero-field regions predicted by GO.

The Fermat principle and the principle of the local field are
two basic concents extensivelvused bv the rav models. The Fermat
principle states that a ray follows the shortest path from a source
Figure 4. The reflection by a wall in a building is modeled by
point to a field point, while the principle of the local field states a lit region in front oflt,
an image Source placed it
that the high-frequency rays produce reflection, refraction, and dif-
fraction when hittine a surface. This denends onlv on the electrical
I

and geometrical properties of the scatterer in the immediate

neighborhood of the point of interaction. two-reflection images, N ( N - I ) ( N - I ) three-reflection images,
and so on [48]. To determine whether an image of the source is
The ray-tracing method is widely used in propagation-model visible at the destination is to trace the intersection of the reflected
and system design [12, 42, 46-72, 1451. It is most accurate when ray at all the necessary planes of interest. Thus, the energy reaches
the point of observation is many wavelengths away from the near- the destination through multiple reflections and contributes to the
est scatterer. All scatterers are assumed to he large when compared received power. Once a ray bas been traced through all its reflec-
to a wavelength. Two types of ray-tracing methods the image
~

tions to the source, the attenuations associated with all the reflec-
method [IZ, 48, 5 5 , 581 and the brute-force ray-tracing method - tion terms are calculated.
are generally used. These are now explained.
The image method is efficient, but it can only handle simple
environments.Many environments with which we are concemed in
our daily life are complicated, and the conventional image method
4.2.1.1 image Method is not adequate. Figure 3 shows a source and its images, corre-
sponding to two reflections [48]. The concept of a lit region has
This method generates the images of a source at all planes. heen introduced in [58] and is illustrated in Figure 4. An image is
These images then serve as secondary sources for the subsequent formed behind the cube representing a structure. This region where
points of reflections. If there are N reflecting planes, then there are the image is formed is termed the unlit domain. For the two-
N first-order (Le., one-reflection) images of a source, N ( N - 1 ) dimensional case, only reflections from walls, and diffraction from
comers in buildings, are taken into account. Ground reflections and
rays over rooftops are neglected [58]. In references [53-55], a
modified shooting-and-houncing-raytechnique, combined with the
image method, has heen used to deal with radio-wave propagation
in iiunished rooms. The effects of diffraction have also been con-
sidered. A threshold must be set with respect to the number and
order of reflection and diffraction rays that can be considered.

/- I 4.2.1.2 Brute-Force Ray-Tracing Method

This method considers a bundle of transmitted rays that may

or may not reach the receiver. The number of rays considered and
the distance from the transmitter to the receiver location deter-

'..---'* ' -.s

SBA
e?----
.**+
?--
mines the available spatial resolution and, hence, the accuracy of
the model. This method requires more computing power than the
image method.

':,-. .' ,, A finite sample of the possible directions of the propagation

-* Sa ,/ from the transmitter is chosen. A ray is launched for each such

'\
. '
U
'
'
#'
I direction. If a ray hits an object, then a reflecting ray and a
refracting ray are generated. If a ray hits a wedge, then a family of
diffracting rays is generated. A reception sphere with the correct
SO radius can describe a region that will receive exactly one ray. If the
radius is too large, two rays could be received, and the same
Figure 3. Images due to a source placed between two mirrors, specular ray might be counted twice. If the radius is too small, it is
A and B. possible that none of the rays will reach the reception sphere, and

/€€€Antennas and Propagation Magazine, Vol. 45, No. 3, June 2003 59

 tions with the same sector angle, 4, in a plane. How the angle 4 is
chosen depends on the accuracy required and the computation
time. If the angle is small, it will provide high accuracy and it will
take a lot of time to compute. For example, if the angle ( = l o ,
then there will be 360 rays to be traced.

Each ray is launched from the source, and can be traced

through a binary tree. An intersection with a surface of an object is
represented by a node in the tree. The incident ray is decomposed
into an object-reflected ray and an object-penetrated ray. It is
TX assumed that the reflected ray propagates along the specular direc-
tion (the incidence angle equals the aspect angle), and that the ray
that penetrates the object keeps the original direction of the inci-
I I dent ray. Both rays then propagate to the next intersection. An
Reception Radius Reception Radius intersection with a wedge is also represented by ,a node, and the
too large too small diffraction point is processed as a source, and a large number of
rays must be launched. The decomposition process is repeated as a
Figure 5. The reception sphere for a two-dimensional ray recursion process. This procedure is continued until the rays are
tracing. weaker than an assumed threshold, or until the ray leaves a prede-
fined propagation area, or until the ray is received. The saength of
the field at the receiver is then calculated according to Equa-
tion (9).
4
A two-dimensional diffraction model was introduced in [62,
641. The authors considered the various buildings to be vertical
knife edges, neglecting the over-rooftop diffraction and ground
reflection. Because the buildings were much higher than the base-
station (BS) and mobile-station (MS) antennas in an urban micro-
cellular environment, the weak contribution from the signals from
the over-rooftop rays could be neglected. No rays due to a single
ground reflection from the transmitter to the receiver existed in the
shadow regions. For ranges less than 1 + from the transmitter
(primary source), the received power could have a range ( R )
dependence given by a power law of 1 / R 2 . However, for LOS
regions;the ground reflections appeared to be less important.

L -...... 1
As reported in [64], the two-dimensional ray-tracing algo-
rithm is quite accurate when the transmitting and receiving antenna
heights are significantly helow the rooftops of the surrounding
buildings. This model of propagation between a transmitter and
receiver located close to the ground is usually called the canyon
model.
Figure 6. Rays generated from a source in two dimensions.
When using a two-dimensional model, the inputs are: a) the
two-dimensional geometry described by means of vectors speci-
the specular ray will be excluded [42, 591. Figure 5 shows &e fying $e location of the building walls; h) the estimated electrical
proper size of the reception sphere that may receive a ray. For each characteristics of the building walls (the permittivity and conduc-
receiver location, the perpendicular distance, d, from the receiver tivity, or the scalar reflection coeficient); c) the base-station loca-
to the ray is computed, together with the total (unfolded) ray-path tion; d) the antenna pattems; and e) the frequency of operation.
length, L, from the source to the perpendicular projection point. If
d is greater than or equal to ( ( L ) / 2 for the two-dimensional case, Three-dimensional ray-tracing model [12, 64-72].
The transmitter and receiver are modeled as point sources when
or ( & L ) / h for the three-dimensional case, the ray is treated as using this ray-tracing technique. In order to determine all possible
not having reached the receiver location. Here, ( is the angle rays that may leave the transmitter and arrive at the receiver in
between two rays. Otherwise, the ray is considered to contribute to three dimensions, it is necessary to consider all possible angles of
the received signal. There is no reception sphere associated with departure and arrival at the transmitter and receiver. Rays are
the usage of the image method. launched from the transmitter at an elevation angle 0 and with an
azimuth angle 4 , as defined in the usual coordinate system.
The key part of the ray-Acing method is the generation and Antenna pattems are incorporated to include the effects of antenna
description of the fays. There are two kinds of methods to obtain beamwidth in both azimuth and elevation.
the rays at the source point. One is a two-dimensional (2D)
approach; the other is a three-dimensional (3D) method, To keep all the raymanipulation routines general, it is desir-
able that each ray ?be occupy the same solid angle, dQ , and that
Two-dimensional ray-tracing model [42]. In two each wavefront have an identical shape and size at a distance r
dimensions, all the rays or ray tubes are ray sectors, as shown in from the transmitter. Additionally, these wavefronts must he such
Figure 6 . At the source, rays are launched along different direc- that they can be subdivided, so that an increased ray resolution can ,

60 IEEEAntennas and Propagation Magazine, Vol. 45. No. 3, June 2003

be easily handled. For example, let r = 1, and let the total wave- 4.2.2 FDTD Models
front be the surface of a unit sphere. The problem then becomes
one of subdividing the sphere’s surface into areas of equal
“patches,” so that all are of the same size and shape and so that, Based on Geometrical Optics (GO) and usually supple-
collectively, they cover the surface of interest without gaps. Hex- mented by UTD, a ray-tracing algorithm provides a relatively sim-
agonal [I21 and triangular [53, 541 ray wavefronts have also been ple solution for radio propagation. However, it is well known that
used. GO provides good results for electrically large objects, and UTD is
rigorous only for perfectly conducting wedges. For complex lossy
The procedure of ray tracing in three dimensions is similar to structures with finite dimensions, ray-tracing fails to correctly pre-
a two-dimensional model, but more computational time is needed. dict the scattered fields. In a complicated communication environ-
ment, transmitting and receiving antennas are often inevitably
Some sectors of the walls in a conidor can be made of differ- installed close to structures with complex material properties for
ent materials, for example, wood, metal, concrete, or even glass, which no asymptotic solutions are available. Such problems can be
which may have different reflectivities for the incident wave. solved by numerical solution of Maxwell’s equations. In particular,
Neglecting the differences among the reflectivities of the various the Finite-Difference Time-Domain (FDTD) method is an altema-
materials will degrade the prediction accuracy of the propagation tive. The advantages of the FDTD method are its accuracy, and
model. Therefore, in [I21 the concept of an “effective building that it simultaneously provides a complete solution for all the
material” was proposed to represent the complicated physical con- points in the map, which can give signal-coverage information
stitutive materials used in the walls of a building. However, the throughout a given area.
permittivity of this effective material is not easy to determine,
since it depends on experimental data as well as on the propagation In a simple outdoor environment, a two-dimensional FDTD
model. To simplify this problem, patches of different dielectric is generally applied [73]. A simple approach for introducing the
constants and physical sizes were introduced in [65]. It is noted correct spherical-wave spreading has been developed. A compari-
that the size and dielectric constant of each patch were chosen son with FDTD predictions could be used to evaluate and refine
according to the physical dimensions and the material of which it the GTD-based methods.
was made.
A reduced formulation for the standard FDTD technique [74]
The key to a propagation model based on ray-tracing is to - requiring four scalar field components instead of the usual six -
find a computationally fast way to determine the dominant ray was used to predict channel statistics inside a residential building.
paths so as to provide accurate path-loss predictions. It is well Measurements were also conducted, and the results were compared
hown that for outdoor propagation prediction, diffraction from with simulated data. In order to introduce arbitrarily shaped anten-
edges must be accounted for in addition to specular reflections, nas into the simulations, a two-and-one-half dimension (2.5D) or a
especially in non-LOS regions. Unfortunately, diffraction is very multi-mode FDTD method was established for indoor radio-propa-
time consuming to model, since a single incident ray encountering gation calculations [75].
an edge will generate a whole family of new rays. The generation
of a large number of diffracted rays limits the number of diffrac- A hybrid technique [76], based on combining a ray-tracing
tions that can be considered. For any given path, we choose at most method with an FDTD method, for more accurate modeling of
two, unless an approximation can be made to find the important radio-wave propagation was also suggested. The basic idea was to
contributing rays. In order to find the contributing rays in an urban use ray tracing to analyze wide areas, and the FDTD to study areas
environment, where the building walls are nearly always vertical close to a structure with complex material properties, where ray-
planar polygons, a vertical-plane-launch (WL) method has been based solutions are not sufficiently accurate.
developed [66]. The vertical-plane-launch approach accounts for
specular reflections from vertical surfaces and for diffraction at the As a numerical method, the FDTD requires large amounts of
vertical edges, and approximates diffraction along horizontal edges computer memory to keep track of the solution at all locations, and
by restricting the diffracted rays to lie in the plane of incidence or extensive calculations to update the solution at successive instants
in the plane of reflection. The vertical-plane-launch approach can of time. The application of an accurate numerical-analysis method
treat many multiple forward diffractions at horizontal edges. It can to model an entire area is neither practical (because of the compu-
also be used for rooftop antennas, and for areas where buildings of tational resources required) nor is it necessary for open areas with-
different heights exist. out many objects.

To improve the efficiency of ray-tracing models, many

researchers have developed a large number of methods [68,72]. In
[68], a hybrid technique was presented, where the object database 4.2.3 Moment-Method Models
was held in two dimensions, but a ray-tracing engine operated in
three dimensions. The three-dimensional rays were produced by Ray-tracing models can be used with sufficient precision to
combining the results of two two-dimensional ray tracers, one on a predict radio coverage for large buildings having a large number of
horizontal plane and the other on a vertical plane. Moreover, by walls between the transmitter and the receiver, while the Method
significantly enhancing the concept of “illumination zones,” the of Moments (MOM) model is better when higher precision is
performance of the algorithm could be dramatically improved. required and when the size of the buildings is smaller. A combina-
tion of these two models is also possible, using the advantages of
A comparison between a two-dimensional model and a three- each of them. For cases where a lot of small but dominant obsta-
dimensional model was made in [64]. cles are present, or where there are paths that cannot be taken into

IEEEAnfennasand Pmpagafion Magazine, Vol. 45, NO. 3, June 2003 61

account by a ray-tracing model, the MOM model can be used [77, In [82], theoretical investigations into the suitability of a neu-
781. ral-network simulator for the prediction of field strength, based on
topographical and morphographical data, were presented. Effective
The solutions determined by the MOM are numerically exact input and output data processing was developed using determinis-
as long as the spatial segmentation used for the objects is small tic and heuristic formulas for the training of a neural-network
enough. Due to limitations of computer memory and CPU time, the simulator. The network used was similar to that described in [81].
MOM is usually applied for analyzing objects that are tens of The inputs were the frequency, the heights of the antenna for the
wavelengths in size. However, by choosing structures with dimen- base and mobile stations, respectively, and the distances between
sions around a few wavelengths, the MOM can be used to check them. The output was the field strength.
and verify a ray-tracing program. A two-dimensional problem that
included stair-shaped walls above a lossy ground was simulated by Although the multilayer neural network [82] is a useful
using both the MOMand the ray-tracing method [77]. method for approximating the propagation loss, it however suffers
from the drawbacks of slow convergence and unpredictable solu-
The transmission of a UHF wave through a window in a wall tions during learning. To overcome these difficulties, radial-basis
is critical when integrating systems that include both indoor and function (RBF) neural networks that have a ”linear-in-the-
outdoor areas. A novel simulation approach, based on the Moment parameters” representation were proposed to enhance the real-time
Method, was presented in [79]. In the simulations, the walls were leaming capabilities and to achieve rapid convergence (831. The
modeled as two long dielectric slabs, long enough so that any dif- RBF neural network is a two-layer localized receptive field net-
fraction or reflection at the outer edges would not influence the work, the output nodes of which form a combination of radial acti-
results. No distinction was made between the concrete and the vation functions computed by the hidden-layer nodes. Appropriate
brick parts of the walls. The glass plates were also assumed to he centers and connection weights in the RBF network lead to a net-
homogeneous, and the aluminum frame was modeled as a perfectly work that is capable of forming the best approximation to any con-
conducting material. The resulting simulations were then compared tinuous nonlinear mapping, up to an arbitrary resolution. Such an
to a set of measurements, where good agreement was achieved. approximation introduces a best-nonlinear-approximation capabil-
ity into the prediction model, in order to accurately estimate the
A hybrid approach, combining the ray-tracing method and propagation loss over an arbitrary environment, based on adaptive
the Periodic Moment Method (PMM) for material objects, was learning from measurement data. Okumura’s data are oflen
developed to study indoor wave propagation, penetrations, and also included to demonstrate the effectiveness of the RBF neural-
scattering due to periodic structures in buildings [SO]. The PMM network approach.
was applied to evaluate the specular and grating transmission and
reflection coefficients of the periodic structures. Those data were
then used in a ray-tracing program to find the specular and gating
rays for each ray tube illuminating one of the periodic structures. 4.2.5 Other Models
Those excited ray tubes were continuously traced to determine
their contributions to the receiving antennas. Recently, many new methods have been introduced to predict
propagation for mobile communications. Some of these are
described in the following subparagraphs.
4.2.4 Artificial Neural-Network Models

The main problem with the statistical models is usually the 4.2.5.1 Vector Parabolic-Equation Model
accuracy, while the site-specific models lack computational effi-
ciency. The use of artificial neural networks (ANNs) bas shown This model has been applied to the modeling of radiowave
very good performance in solving problems with mild nonlinear- propagation in an urban environment. As a parabolic version of
ties on a set of noisy data. This case corresponds to a problem of Maxwell’s equations, it allows a full treatment [84, 138, 1391 of
field-level prediction, as the data obtained from measurements is three-dimensional electromagnetic scattering, which is not possible
always noisy. Another key feature of the neural network is the with scalar versions of the algorithm. It is particularly useful for
intrinsic parallelism, allowing for fast evaluation ofthe solutions. accurate modeling of scattering by a single building or by a group
of buildings at microwave frequencies. Examples include scatter-
The ANN model [81] that bad the form of a multilayer ing by a building with a hemispherical roof, and scattering by a
perception was generally used with 12 inputs and one output. It group of buildings with sloping roofs.
was developed to predict the propagation in an indoor environ-
ment. In this case, a two-dimensional floor plan was used for a
database with a resolution of lOxl0 cm. All particular locations
were classified into 11 distinct categories, such as wall, comdor, 4.2.5.2 Fast Far-Field
outdoor area, laboratory, and so on. One input of the network rep-
resented the normalized distance from the transmitter to the
Approximation Model
receiver. In addition, there was an input for each defined environ-
mental category. Other inputs represented either a normalized This model is substantially faster than conventional integral-
number of occurrences (doors and windows), or an appropriate equation- (IE-) based techniques. The technique is improved by
percentage (wall, comdor, and so on) of that category along the incorporating the Green’s-function perturbation method. The
straight line drawn from the transmitter to the receiver. The proc- method has been applied to gently undulating terrains, and com-
ess of learning may have lasted for a couple of hours, but the proc- pared with published experimental results in the 900 MHz band. It
ess for field-level prediction was fast. The accuracy of such a pre- has also been successfully applied to more hilly terrain, and to sur-
diction model depends significantly on the accuracy of the envi- faces with added buildings [85]. An improved version of the
ronmental databases. “shifting function” has been introduced, which can improve the

62 lEEE Antennas and Pmpagstion Magazine, Vol. 45, No. 3, June 2003
performance of the technique for more-challenging problems, such 4.3 Summary of Models for Path Loss
as scattering from a wedge. The issues of profile inncation and
small-scale-roughness effects have been addressed, and the
A summary of the various propagation models dealing with
numerical results presented showed excellent agreement with pub-
path loss has been given. A brief comparison of some of the main
lished measured data.
models is presented in Table 4. Propagation models dealing with
path loss for mobile communication have been emphasized, using
two very different approaches. First, a simple empirical or statisti-
4.2.5.3 Waveguide Model cal model of the path loss was considered, where some of the
parameters used were determined empirically from measurements.
In large metropolitan areas that have tall buildings, the trans- The second approach used site-specific methods. Ray tracing was
mitting and receiving antennas are both located below the rooflops, the main method. Some other numerical methods used in electro-
and the city streets act as a type of wave-guiding smcture for the magnetic-fieldcomputation have also been applied.
propagating signal [86, 87, 1401. In this case, there is a need to
develop efticient algorithm for the computation and mapping of Each of these two kinds of approaches makes a very different
the field distribution in such structures. Theoretical analysis of trade-off of accuracy versus complexity. The empirical (statistical)
propagation in a city street modeled as a three-dimensional multi- models are extremely simple (no environmental information is
slit waveguide was proposed. Assuming the screens and slits are used, other than in the choice of the parameters), but the predic-
distributed by a Poisson law, the statistical propagation character- tions are not very accurate. On the other hand, site-specific models
istics in such a waveguide are expressed in tenns of multiple ray are considerably more accurate than the empirical models, but they
fields approaching the observer. Algorithms for the path-loss pre- require a great deal of specific information about the area of inter-
diction were presented and compared with experimental data in the est (at a minimum, the locations of all the objects, and possibly the
references cited above. locations of large objects).

4.2.5.4 Boltzmann Model 4.4 Efficient Computational Methods For

Propagation Prediction for
This model was initially developed for simulated flnid flows.
It describes a physical system in terms of the motion of fictitious Indoor Wireless Communication
microscopic particles on a lattice [88, 1431. This technique can
take into account complicated boundary conditions. Two-dimen- This section presents two different deterministic methods for
sional simulations were performed, starting from a city map, and a efficient characterization of an indoor channel. First, an improved
renormalization scheme was proposed. The method, which is sim- version of a ray-tracing method is presented. Next, the FDTD
ple and easy to implement, provided good path-loss predictions method is used to calculate the effects of walls in an indoor wave-
when compared with site measurements. propagation environment.

Table 4. A comparison of various models for path loss.

Model
Name
I Suitable
Environment
1 Complexity I Experimental
Data
Details of
Environment
Accuracy Time Other

Model I I experiments I I path-ioss data

IEEEAnlennas and Pmpagafion Magazine. Vol. 45, No. 3, June 2003 63

 method of homogenization can he used to determine the effective
material properties of walls, but they did not consider non-specular
reflection.

The FDTD method was used to predict propagation proper-

ties for indoor environments [137]. The periodic structure of walls
was also considered. Numerical results for the path loss were cal-
culated by an FDTD method, and were compared to those obtained
by the ray-tracing method. It was proved that the inner structure of
the walls had a considerable influence on the path loss when pre-
dicting propagation, and that the FDTD method can give more
accurate results.

4.4.2.1 Numerical Results

I The fields of the environment were calculated using a two-
dimensional FDTD method when the wall was first assumed to he
Figure 7.A typical periodic inner structure of walls. a periodic structure, and then uniform in nature, with an effective
dielectric constant. For the FDTD method, 5000 time steps were
calculated, and it took more than six hours of CPU time on a PC. A
two-dimensional ray-tracing method was used to analyze the uni-
4.4.1 Efficient Ray-Tracing Methods form structure, and it took only 10 minutes to compute the path
loss for 250 sample points.
The applicniioii u i w e r a l ray-tracing techniques in combina-
tion with the Uniform 'I'hcor) of Dilfraction has been an cffcient Figure 8 shows the simulation results of the path loss for the
method for the prediction ofpropagation i n the KIIF (communica- sample points. This indicates that the effects of the inner shucture
tion) band in an indoor cnvironnient. This i s discussed in detail of the walls cannot be neglected for propagation prediction. The
[ 1361, whcrc the computational cfiicicncy o i the two-dimcnsional maximum difference in the field strength when the wave was
my-tracing method was improvcd by rcorgmimy thc objccis in an assumed to propagate through a periodic structure and through a
indoor envirmment intu irregular sclls. In adjitiun, by making usc uniform structure was more than l O d B when using the FDTD
o i the two-dimensional ray-tracing results, a ne\\' thrce-dimm- method. As a full-wave analysis tool, the FDTD is more accurate
sional prupagarion-prediction mudel wils developed that ciln save than the ray-tracing method. All the propagation phenomena, such
99% of the compulntional Iimc of 3 rrsditiunal three-dimensional as reflection, transmission, and diffraction, are included. A ray-
model. This ne\%'hybrid model i. more accurate t h i n ruo-dtmen- tracing method considers only reflection and transmission. The
sional modcl>and more efficient than trdditiunal three-dimensionnl maximum difference in the field strength when propagating
models in computing the pslli Ius> t o any poinl in n huilding. In
this model, reflection 2nd refraction by layered matcnals and dif-
fraction from the corncrs of walls \vas coiisidcrcd. This theoretical
model had a gond correlation with experimental h t a . r
-30

4.4.2 Analysis of the Effects of Walls on

Indoor Wave Propagation Using FDTD

The ray-tracing technique does not take the effects of the

inner structure of walls into account for indoor propagation pre-
diction. A numerical approach to treat this problem, using the
FDTD method, is now described. Numerical results for the path
loss calculated hy an FDTD method were compared with those
obtained by the ray-tracing technique. The application of ray-trac-
ing techniques in propagation prediction has been under develop-
ment for more than a decade. They are powerful and easy to use.
There have been a number of recent investigations into indoor
radio-propagation modeling using the ray-tracing method [57, 65,
1261, but few have paid attention to the inner structure of the walls
[134, 1351. When using a ray-tracing method, it is assumed that
reflection from walls has a substantial specular component. Typi-
cal concrete blocks used in wall construction are shown in Fig-
0 20 40 60 so 100120140160180u)om240
ure 7. The air hole in the block forms a periodicity in the blocks. It Sample Points
has been found that non-specular reflection occurs when the oper-
ating frequency is above 1.2GHz [134]. In this approach, the Figure 8. Simulation results using three different FDTD
reflection and transmission characteristics of the walls can be methods: -FDTD (periodic); -FDTD (uniform);
derived by solving for higher-order Floquet modes. Also, the ____ Ray tracing (uniform).

64 IEEEAnIennas and Propagallon Magazine, Vol 45, No 3,June 2003

through a periodic structure using the FDTD method and the ray- 5.2 Rayleigh Distribution
tracing method was around 18 dB.
As the dominant signal in a Rician distribution becomes
weaker, the composite signal resembles a noise signal that has an
envelope of a Rayleigh distribution. For mobile-radio channels, the
5. Models For
. .
Small-Scale Fading Rayleigh distribution is widely used to describe the statistical time-
varying nature of the received envelope of a flat-fading signal, or
Small-scale fading refers to the dramatic changes in the sig- of an individual multipath component. The Rayleigh distribution
nal amplitude and phase that can be experienced as a result of has a probability density function (PDF) given by
small changes (as small as a half wavelength) in the spatial separa-
tion between a receiver and a transmitter. These dramatic changes
in the envelope of the received signal are described statistically by
a stochastic process.
r<O
In order to get a sound understanding of the channel, it is
important to study the distribution of the envelope of the received
The probability that the envelope of the received signal does not
signal. A few possible choices of the statistical distributions for
exceed a specified value R is given by the corresponding cumula-
modeling the envelope are explained below. As explained and
tive distribution function (CDF):
emphasized in [160], we have to he extremely careful when using a
statistical model, as a complete lack of knowledge of the system R
cannot he supplemented by a stochastic distribution. P ( R ) = P , ( r < R ) = Ip(r)dr=l-exp
0

5.1 Ricean Distribution where r is the envelope amplitude of the received signal, and 2a2
is the predicted mean power of the multipath signal.
This is also called a Rice distribution or a Rician distribution.
When there is a dominant stationary (non-fading) signal compo- Since the fading data are usually measured in terms of the
nent present, such as P line-of-sight propagation path, the fading fields, quantities for a particular distribution cannot be assumed.
distribution is Ricean. The Ricean distribution is given by The median value is oAen used rather than the mean values, and it
is easy to compare different fading distributions, which may have
widely varying means. A typical Rayleigh fading envelope for a
moving mobile at 900 MHz is shown in Figure 9 [90].

0 r<0
5.3 Log-Normal Fading Model
where r is the amplitude of the envelope of the received signal, and
2a2 is the predicted mean power of the multipath signal. A This model is used to quantify the distribution of rays that
denotes the peak amplitude of the dominant signal, and Io (.) is experience multiple reflections and dimactions between a trans-
the modified Bessel function of the first kind and zero order. The mitter and a receiver. The log-normal PDF can he expressed as
Ricean distribution is oAen described in terms of a parameter, K,
which is defined as the ratio between the deterministic signal
power &d the variance of the multipath. It is given by [IO]

AZ
K (dB) = 10log- (31)
2 2 '

10
K is known as the Ricean factor, and completely specifies the dis-
tribution.

It is possible to estimate the Ricean K factor of a signal from

8g :
I

-3;
-5
measurements of received power versus time. One approach is to
compute the distributions of the measured data and to then com- -10
pare the result to a set of hypothesis distributions, using a suitable --I5
0
goodness-of-fit test. Another is to compute a maiimun-likelihood -20
estimate using ab expectatiodmaximization (EM) algorithm.
However, both of these approaches are relatively cumbersome and ';d -25
c
time consuming. A simple and rapid method has been developed, .M
v)
-30
based on calculating the first and second moments of the time- -35
series data. The method is exact when perfect mnment estimates of
the Ricean envelopes are available. In that case, the factor K can -40
0 50 100 150 200 250
only be obtained implicitly by equating a ratio of the measured Elapsed Time (ms)
moments to a complicated function of K. By contrast, the method
described in [89] yields an explicit and quite simple expression for Figure 9. A typical Rayleigh fading envelope at 900MKg
Kin terms of the measured moments. received by a mobile unit traveling at 120 W h r .

IEEEAntennas and Propagation Magazine. Vol. 45, No. 3. June 2003 65

 where m is the median value, and U is the standard deviation of
m = ( r 2 ) 2 / ( ( i - - ( r ) z ) 2 ) is the inverse of the normalized vari-
the corresponding normal distribution, obtained by using the trans-
formation y = l n ( r ) [91]. Techniques such as the Monte Carlo
ance of r z . r(*)is the Gamma function. The Nakagami PDF may
method and the Schwartz and Yeh method have been developed to
simulate the power sum for the log-normal components [92]. he shown to he a more-general expression of other well-hown
density functions. The Rayleigh Probability density function is
obtained for m = I . The Nakagami PDF can also he approximated
by both Rice and lognormal distributions over certain domains,
5.4 Suzuki Model given appropriate bounds on the parameters.

A Nakagami model, parameterized by the fading severity

The Suzuki model combines two distributions, comprised of parameter, m, has been shown to fit well to some urban multipath
the log-normal and Rayleigh distributions. It provides a more- propagation data. A simulation of the Nakagami model has been
accnrate approximation of the sum of correlated complex log- presented in [97].
normals for a wider variety of channel behaviors. Usually, the
Rayleigh distribution is obtained from two statistically independent
normal processes, pl ( t ) and pz ( t ) , with zero means and identical
autocorrelation functions, according to the relation 5.6 Weibull Model

(35) This model arises when results from mobile-radio propaga-

tion measurements are plotted on graph paper that is scaled such
that a Rayleigh distribution appears as a straight line with a slope
where <(f) can he regarded as the envelope of one complex- of -1. The Weibull PDF can be written as [91]
valued normal random process, A ( 1 ) . The requirement of statisti-
cal independence between p l ( t ) and p 2 ( t ) is identical to the
demand for a symmetrical power spectrum for L(t). The received
power, averaged over a period of few seconds, can vary considera-
bly, due to various shadowing effects. In order to adapt the model where a is a shape parameter that is chosen so as to yield a best fit
to this behavior, the process 5(t) is substituted for by the product to the measured results. Here, ro is the RMS value for r, and
V ( t ) = < ( t ) . c ( r ) , where the log-normal process T ( t ) = exp[p3(t)]
b=[(Z/a)r(Z/a)]1’*is a normalization factor. For the special
is defmed by a normal process with variance sz and mean
case in which a = 1/2, the Weibull PDF becomes a Rayleigh PDF.
m. The product process with this particular amplitude-density dis-
bihution is called the Suzuki process, and is given by [93] The Weibull distribution provides flexibility to model any shape
qffered by a Nakagami distribution, but it lacks a theoretical basis.
It was shown that the Weibull distribution characterizes indoor
propagation-path losses quite well in [146].

where U is the standard deviation and r is the amplitude. The 5.7 Other Fading Models
assumption of statistical independence does not always meet the
real conditions in multipath wave propagation, so it has been modi- Many other models for fading have been developed. Some of
tied [94], and was simulated in [95]. these are the Rice Log-Normal Model [98], the Nakagam-Rice
Model [99], the Nakagami Log-Normal Model [IOO], and the K-
This is a widely accepted statistical model for the received- distribution, which is a substitute for the Rayleigh log-normal dis-
signal envelope in macrocellular mobile radio channels where tribution [loll. These are mixtures of two kinds of distributions,
there is no direct LOS path. and are now widely used. In [102], propagation models that
include both the effects of shadowing and multipath fading were
developed and used in studying terrestrial and satellite channels. In
[IO31 and [104], a new theoretical model for the prediction of fast
5.5 Nakagami Model fading in an indoor environment was developed. This model made
the assumption that the number of dominant propagation paths that
The Nakagami model was developed in the early 1940s by contribute to the signal in the receiver in a multipath environment
Nakagami. The corresponding probability density function is wit- is not infinite, but rather small (e.g., 15). This assumption led to
ten as [96] the development of a new theoretical model, called “POCA,” that
is more general and more efficient than the Rayleigh Model, espe-
cially for indoor environments. It is generally fit for environments
in which there exists a small number of dominant propagation
(37) paths.

In this section, some of the statistical models for fading have

Here, r is the envelope amplitude of the received signal. been introduced. Only their PDF distributions have been described.
Details of the implementations of these models can he obtained
( 2 ) .is the time-averaged power of the received signal, and
Cl = r
from the various references cited

66 IEEE Antennas and Propagation Magazine, Vol. 45. No 3, June 2003

 6. Impulse-Response Models sight (OLOS) chainels are subject to higher variations% RMS
delay spread, caused by movement of the terminal around a small
area, because of the presence of a larger number of scatters in the
The narrowband or continuous-wave (CW) path loss is a obstructed line-of-sight environment than in the LOS environment.
parameter that can predict the power level of the system, and the On the other hand, the short-time variations in the RMS delay
space coverage of a base station. In modem mobile-communica- spread in obstructed line-of-sight environments were found to he
tion systems, with high data rates and small cell sizes, it is neces- less than those in LOS environments [117]. It is concluded that in
sary to model the effects of multipath delay, as well as modeling an empty indoor environment, the RMS delay spread is a function
fading. The impulse response is a useful characterization of the of the transmitter-receiver separation. When there is furniture and
system, since the output of the system can he computed through mazes of semi-permanent partitions, the RMS delay spread
convolution of the input with the impulse response, if the system is becomes constant for all ranges [105]. It was shown in [I211 that
linear. A multipath propagation channel is modeled as a linear fil- in many cities where the terrain is flat; the RMS delay spreads do
ter, and has a complex baseband impulse response [105-125, 144, not exceed 7 or 8 ps, and that in urban areas with surroundinghills,
147-1511. the RMS delay spreads do not exceed 13.5 ps. In [26, 1231, the
RMS delay spread was analyzed as a function of the path loss and
In digital wireless communications, one of the main reasons
for the occurrence of hit errors is inter-symbol interference (ISI)
caused by multipath propagation. If the symbol rate is much lower
than the coherent bandwidth, the time-delay spread can be
neglected. In this case, multipath propagation only causes fading of
the signal level, and Gaussian noise is the dominant factor that
causes bit errors. If the symbol rate is relatively high, the time-
delay spread cannot be neglected.

A large number of measurements have been made for the

impulse response, both for outdoor and indoor environments. Some
models have been derived based on the measured data. The results
of measurements, along with some mica1 models, are described
here, followed by a description of some deterministic models.

Figure loa. A block diagram of the direct RF channel impulse-

6.1 Models Based on response measurement system.
Measurement Results

Because of the importance of determining the time-delay 1

spread for wireless communications, a number of wideband chan-
nel-sounding techniques have been developed. These techniques
may be classified as

A. Direct pulse measurements;

B. Spread-spect” sliding correlator measurements; and
C. Swept-frequency measurements.

The block diagram for each of these techniques is shown in I

Figure 10 [IO]. Almost the same results can he obtained from the
last two methods [105]. Some measurements for both indoor and WidebaodFih NmWAllsr
outdoor environments, and the calculated RMS time delays have
been summarized in Table 5 . In this table, “Med” indicates median Figure lob. A block diagram of the spread-spectrum channel
. values, ‘T-R”indicates the distance between the transmitter and impulse measurement system.
receiver, and “Ave” implies the mean average values. In Table 5 , it
is found that the RMS delay spread varies from several nanosec-
onds to several microseconds, corresponding to different environ-
ments and frequencies.

There are many factors affecting the impulse response, such I, I

as the frequency, the height of the transmitting and receiving

antennas, fxed objects (mountains, buildings, or indoor furniture),
and moving humans or objects close to the transmitter and/or the S-pouamctcrfeJfast
receiver. It is shown in Table 5 that the RMS time delay vanes lit-
tle for different frequencies for indoor environments [106], but a
different conclusion was obtained in [107]. In [108], it was found
that at 11.5 GHz, the RMS time delay in most cases was signifi-
cantly smaller than at 2.4 GHz and 4.75 GHz, for which the values
were, in general, about the same. The reason that two different
conclusions were reached is that all of these various conjectures Figure 1Oc. A block diagram of the frequency-domain channel
have been derived from specific measurements. Obstmcted line-of- impulse measurement.

IEEEAntennas and Propagation Magazine, Vol. 45, No. 3. June 2003 67

 Table 5. Measurements and RMS delay spreads.

Frequency I Ref.

I 1 19-37 (Area E)
Indoor sports 7- I20
arena 40.130
Open-plan factory 15-125
Textile plant
Office buildin
Within a room
4 rooms
Urban area I 98-1270

480 (Ave.)
Building 1 About 12-72
C
Building 2 About 4-25
Sidewalk of LOS 0.5d +40
B
street (d is T-R)
Engineering bldg. 12.85-84.60
A
Retail store 20.74-102.44
8.3 LOS (Med.)
Laboratories at
- 14.1 OLOSl Med.
same floor
22.3 OLOS2 Med.
Outdoor Site 1 60-250. 130
Site 2 wed)
B Site 3 40-130,70 (Med)
60-250, 120
(Med)
10-50 with 900-1300
C In two buildings
mean 20-30
10-20 (Med )
Four types Of
C 10.20 (Med)
indoor locations
5-15 (Med)
Hamburg 1300 (typical) 942.225
Dusseldorf 3100 (typical)
A
Frankfurt(bank) 8100 (largest)
Frankfurt(apart.) 19600 (largest)
Washington 2500-7500
Greenbelt 2000-7000
A
Oakland 2500- 13500
San Francisco 1000-25500
- LOS (100-400 m) 140.6-325.4

Table 6. The RMS delay spread as a function of the height of

an antenna.

I I I 1 I I
I I I
Mean Standard

I Antenna
Height 1 I Deviation of
RMS Delay
Location Ref.

68 IEEEAnlennas and Propagation Magazine, Vol. 45, No. 3, June 2003

the height of the antenna. The predicted results can be hounded by 6.2.4 SlRClM Model
an exponential model of the form rWS(ns) = exp(0.065PL),
where PL is the path loss in decibels. Table 6 shows the measure- Based on measurements at 1300 MHz in five factoly and
ments of rWs with relation to the height of the antennas. It indi- other types of buildings, the piecewise functions of excess delay
cateS that the delay spread increases as the antenna height for the probabilities of multipath amvals were derived as [3]
increases, because as the antenna is raised, it becomes visible to
more objects, scattering the electromagnetic radiation at greater
TK < 110 ns
distances.

I
PR(TK,SI)= 0.65-
(TK - 1 10)
360
110 ns < TK < 200 ns, (41)

6.2 Statistical Models of

Time-Delay Spread I0.22 -
(TK - 200)
1360
200 ns < TK < 500 ns

6.2.1 Two-Ray Rayleigh Fading Model 0.55+K TK < 100 n s

667
PR (TKS’Z)= (TK -100) , (42)
A commonly used multipath model is an independent -~
0.08 + 0.62e 75 100 ns < TK < 500 ns
Rayleigh fading two-ray model [IO]. The impulse response of the
model is represented by
where S, and S2 correspond to LOS and obstructed LOS environ-
h ( 1 ) = a1 e x p [ n ( I ) ] & (1) + a2 exp[jo2 ( l ) ] ~ (-t r ) , (39) ments, respectively. TK ,in units of ns, is the excess delay at which
a multipath component will arrive at the receiver, and takes on val-
where a, and a2 are independent random variables and have a ues that are integer multiples of 7.8 ns.
Rayleigh PDF. S, and 0, are also two independent random vari-
ables, and their density functions are uniformly distributed over
[0,2n].r is the time delay between the two rays.
6.2.5 A - K Model

6.2.2 Saleh and Valenzuela Model This model [9, 1191 takes into account the clustering property
of paths caused by the grouping property of scatters. The process is
described by transitions between two states, representing different
Saleh and Valenzuela [30] reported the results of indoor mean arrival rates. Initially, the process starts in State 1 , with a
measurements between two vertically polarized omnidirectional mean arrival rate /1, ( t ) , If a path arrives at time 1, a transition is
antennas, located on the same floor of a medium-sized ofiice
building. The measurements indicated that the statistics of the mad6 to State 2, with a mean arrival rate K& ( f ) . If no further
channel impulse response were independent of the polarizations of paths arrive in the interval [ t , t + A ] , a transition is made back to
the transmitting and the receiving antennas if there was no line-of-
sight path between them. The model assumed that the multipath State 1 at the end ofthe interval.
components arrived in clusters. The clusters and components
within a cluster formed a Poisson amval process, with different
rates.

6.2.6 Discrete-Time Model

6.2.3 Log-Normal at Any Distance In this model [119, 1251, the time axis is divided into small
time intervals, called “bins.” Each bin is assumed to contain either
In [124], Cardoso and Mouliness presented three conjectures one multipath component or no multipath component. The possi-
about delay spread. The first was that rRMswas log-normal at any bility of more than one path in a bin is excluded. A reasonable bin
distance. This was derived from measurement data that were clas- size is the resolution of the specific measurement.
sified into urban, suburhan, rural, and mountainous areas. The sec-
ond conjecture was that the median cRMs increased with distance,
and the third conjecture was that rRMs tended to increase with
shadow fading. rRMS can he given by 6.3 Deterministic Models of
Time-Delay Spread
~ R M S= Td“Y, (40)
The above statistical models are based on measurements
where 4 is the median value of rWs at d = 1h,E is an expo- made in a specific environment. They may not he suitable for pre-
nent that lies between 0.5 and 1.0, andy is the log-normal variable. diction in other environments.

IEEEAntennas and Propagation Magazine. Vol. 45. No. 3, June 2003 69

 6.3.1 Ray Tracing parameters are intimately related through the distance between the
source and the field point, and the velocity of light. Hence, a basic
physical understanding of the problem is absolutely essential in
In theory, the ray-tracing technique [U, 105, 118, 123, 1271 trying to obtain a meaningful solution. There are several variations
can determine almost all multipath components, including their of this technique, which is often abbreviated as JADE.
amplitudes, time delays, and phases, and the technique is effective
in predicting the time-delay spread. A detailed howledge of the
environment is required when applying a ray-tracing method.
Another advantage of ray-tracing models over other propagation 7.1 JADE Maximum-LikelihoodModel
models are their ability to incorporate antenna radiation pattems
and, in particular, to consider the effect of the radiation pattem 18,1531
individually on each ray.
In the conditional-maximum method, the vector representing
Since the phase of each ray arriving at a receiver varies the complex envelope of the fading path is assumed to he non-ran-
significantly with distance and cannot he accurately predicted, it dom. Under this assumption, the maximum-likelihood-method
will he impossible to achieve agreement with either the instantane- estimates of the angle of arrival and the time delay are obtained by
ous measured shape or the value of the RMS delay spread for a minimizing the cost function. In the unconditional-maximum
single individual measurement. The comparison hetween measured method, the envelope of the fading path is assumed to be a station-
and modeled results should be for average values in a small area ary Gaussian random process, and a different cost function is
around the actual mobile position [127]. A seven-ray and a 25-ray obtained. It involves a 2I-dimensional optimization over both the
model consider only single or double reflections, respectively, and angles of arrivals and the time delays, where L is the number of
were used to simulate propagation in a room [105]. The RMS multipath components. Hence, it is computationally prohibitive,
delay spread was predicted, and agreed well with that calculated even for moderate values of L. In [153], an efficient method in
from measurements. which 0 and 'I were estimated in parallel was developed.

6.3.2 VRP Model 7.2 JADE-MUSIC Model [I541

This is similar to a seven-ray model, but in an outdoor envi- A faster though suboptimal - approach is based on the
~

ronment. It assumed some virtual reflection points (VRF's), located MUSIC algorithm, and involves only a two-dimensional search.
at the intersection points along the LOS on streets and at building The MUSIC method estimates the angles of arrival and the delay
walls [122]. It did not consider the effects of traffic and moving parameters by finding the L minima of the pseudo-spectrum.
humans. The predicted results for the R M S delay spreads were
verified by measurements.

7.3 SIJADE Model [155,156]

7. Joint Angle-Delay Estimation (JADE)
By assuming a uniform time sampling and a uniform linear
array, this algorithm transforms the data using a discrete Fourier
In mobile communications systems, the multipath channel is transform (DFT), and performs a deconvolution with the lmown
characterized not only by the various time delays of the different pulse-shape function. It then stacks the result into a Hankel matrix.
propagation paths, but also by their directions of arrival @OA) or This reduces the problem to one that can he solved using a two-
angles of arrival (AOA). Estimating the time delay by ignoring the dimensional ESPRIT algorithm. The algorithm can thus be written
spatial characteristics of the multipath channel - as most systems in closed form and is computationally attractive.
now do is definitely not optimum. Indeed, exploiting the spatial
~

structure by using adaptive antenna systems and properly estimat-

ing the delays can improve the characterization of the channel and,
consequently, can lead to better system service.
7.4 JADE-ESPRIT Model [I571
The joint angle- and delay-estimation technique is, in fact,
often employed in radars using space-time concepts. In comparison If the array is a uniform linear antenna array, or if it has the
to disjoint techniques - which first estimate delays and, subse- ESPRIT doublet-structure, then the angles and delays can be esti-
quently, the angle corresponding to each delay -joint estimation mated jointly. The JADE-ESPRIT algorithm requires stricter iden-
has a number of advantages. First, the relative delay estimates tifiably conditions, and can deal with the number of resolvable
between two or more synchronized base stations can he used, in paths that are determined by the number of samples available. This
conjunction with the DOA information, to estimate the location of model is quite similar to the SI-JADE algorithm. SI-JADE, how-
the mobile unit. Second, it is theoretically possible to exploit the ever, only acts on a single-channel estimate, and may also intro-
difference in path delays to improve angle-estimation accuracy, duce an undesired coupling between angle and delay estimates in
and vice-versa [8, 1521. Third, when several angles are detected, the case where the model is not perfectly valid.
the most-probable portable direction is the one that is associated
with the smallest delay, because the direct path has the shortest There are other techniques in addition to the JADE models,
propagation time. Unfortunately, the problem with this model - such as the subspace-based methods [I%], and a Bayesian-based
even though the formulation is quite esoteric from a signal-proc- model [159]. The Cramer-Rao bound (CRB) has been derived for
essing point of view - is that the directions of arrival and the time each of these models. However, only simulation results have been
delays of the signals are not independent quantities. The two used to verify all these models.

70 IEEE Anlennas and Propagation Magazine, Vol. 45, No. 3, June 2003
 To conclude this section, it is important to point out that all knowledge about the spatial locations of the transmitters and the
these models treat each antenna as an isotropic radiator. An receivers. This can he accomplished using the principle of recip-
admirer of James Clark Maxwell will immediatelyrealize that such rocity. The details are available in [160]. A brief description of this
an antenna does not exist in the electromagnetics world, which is novel technique is provided here.
the domain of wireless, but does exist in the acoustic arena. The
smallest antenna is a Hertzian dipole, and even that has a direc- In this novel procedure, we simultaneously employ the con-
tional pattem; however, it may he omnidirectional in certain cept of reciprocity and use the principle of adaptivity on transmis-
planes. Finally, we point out that an electrically small antenna dif- sion. In this way, it is possible to direct the energy transmitted
ferentiates the pulse on transmission, and integrates or samples the from a base station to a pre-selected mobile station, without wor-
signal on reception. However, for finite-length antennas, the radi- rying about either the presence of other near-field scatterers or the
ated fields are proportional to the double derivative of the current existence of a multipath environment. For example, consider the
induced on the structure. On reception, the open-circuit voltages system represented in Figure 11. Here, let us assume, for example,
are proportional to the single derivative of the incident fields. that the transmitter is sending signals using spatial diversity, i.e.,
Hence, unless the impulse responses of the transmitting and using a number of transmitting antenna spatially separated from
receiving antennas are included in a broadband channel characteri- each other. Let us assume that transmitting antenna #1 is transmit-
zation, it is difficult to see how one can obtain a meaningful physi- ting a voltage Vq at a frequency fo.The intemal impedance of
cal model of the wireless propagation channel! this voltage source is R I ; . At the second transmitting antenna, we
have a voltage source V., transmitting at the same frequency, with
an intemal impedance of RT2 . In addition, there are two receivers,
8. Epilogue
represented by RI and Rz . The goal ofthis transmitting system is
However, in a mobile communication environment where that we want to maximize the received power at receiver 1 by
there is an established two-way commuuication, it is not necessary exciting transmitting systems 1 and 2 with voltages V i and T i 2 .
to know these path propagation models in order to direct the Our goal is to minimize the received power at receiver 2 while
energy from the transmitter to a pre-specified mobile. It is not even directing the energy to receiver 1. Similarly, we want to find
necessary to h o w the particular electromagnetic environment in
which they are transmitting. It is not even necessary to have any another set of voltages, V$ and V 2 , that will excite the two trans-

RspiwR
Figure 11. A multiple-user transmiffreceive scenario.

IEEE Antennas and Propagation Magazine, Vol. 45, No. 3, June 2003 71
 mitting antennas so that they will induce zero fields on receiver #I then it is going to induce currents 1; in transmitting antenna #1,
(designated by R I ,and maximum fields at receiver #2 (designated
and I), in transmitting antenna #2. From these induced currents,
by Rz). The characteristics of these two voltage sets with a super-
script of 2 are such that when they are used to excite the transmit- we know the voltage induced across the loads of the transmitting
antennas, since we know the impedance values of the loads. Now,
ting antennas, the electromagnetic signal will he directed to
receiver 2 and will induce practically no energy at receiver 1. In what the reciprocity principle tells us is that if we apply 1 V at the
this way, we can essentially do adaptivity on transmission. This is feed of the transmitting antenna #I, so that V., = I and
ideally suited in a CDMA (code-division multiple access) envi- VT, = VRb= VR, = 0 , then it will induce a current I;, into receiver
ronment, for example, where a unique code is assigned to each
receiver, so that when we receive any signal we know from which 1, so that
receiver it originated. When we are going to transmit that particu-
larly weighted signal, broadcast simultaneously over all the trans-
mitting antennas, we can spatially direct the energy to a pre-speci-
fied receiver only, and simultaneously minimize the received sig- Similarly, if we excite receiver 2 only with a voltage of 1 V, so that
nals at the other receivers for which that particular transmission VR2= 1V and Vq = VT2= VR, = 0 , then this will induce currents
was not intended. In this way, it is possible to direct the energy to a
pre-selected receiver. I: and I:, at transmitters 1 and 2, respectively. From reciprocity,
we have the result that if we excite transmitter 2 with 1 V at the
The problem now is, how to carry this out? We propose to
achieve this objective using the principle of reciprocity. Reciproc- same frequency fo,then Vr2 = I and V1; = VRl = VR, = 0. Under
ity tells us that if we excite transmitting antenna #I with a voltage this situation, there will he an induced current in receiver 2.
V., and with the other voltages set to zero, such
From the principles of reciprocity in the frequency domain, one
that Vr2 = VR, = VR2= 0 , then the transmitted signal will induce a obtains
current, IR,, at the feed point of the receiving antenna marked #I
(i.e., RI).It will also induce currents at receiver #2 and transmitter (45)
#2. However, let us ignore that for the moment. Under the same
environment, if we excite receiver I with a voltage VR, with It is important to ohserve that reciprocity only links the two trans-
mitting and receiving ports under consideration, namely, those
VT = VT2= VR2= 0, then it will induce a current - let us say, I., - related to the transmitter and the receiver. The other portions of the
at transmitter # I . It will also induce currents at receiver #2 and electromagnetic network do not come into the picture, as long as it
transmitter #2. However, let us also ignore these for the moment. remains the same when we switch the voltages on the transmitters
Then, from the principle of reciprocity, we know that and receivers to satisfy the relationships of Equations (44) and
(45). Nor is reciprocity dictated by the material medium influenc-
ing the propagation path. This implies that the propagation that
resulted in the induced currents described by the variables in
Equations (44) and (45) can occur even when one is communicat-
Equation (43) represents the reciprocity theorem in the frequency ing in non-isotropic and heterogeneous media.
domain. The relationship is much more involved in the time
domain, where the reciprocity theorem performs an integration From the above observations, we can carry out the following
over time. That is why it is possible - without violating the princi- procedure:
ple of reciprocity - for a transmitting antenna to radiate essentially
a differentiated version of the temporal input waveform applied to Step 1: Excite receiver #I with 1 V, with
it, whereas a receiving antenna integrates the incident signal, both
spatially and in the temporal domain. Vq = VTz = VRz = 0 . This will induce currents I i and in trans-
mitting antennas 1 and 2, respectively. Let us assume that we can
Here, we consider the currents that are induced at the measure these currents.
transmitting and receiving antennas to be at the loads that are asso-
ciated with the feed point of each antenna. The relationship pro- Step 2: Excite receiver #2 with 1 V, with
vided by Equation (43) is valid irrespective of the shape of the Vq = VT2= V., = 0 . This will induce currents 1; and I;> in trans-
transmitting or receiving antennas. It also does not depend on the
presence of various near-field scatterers, like trees or buildings that mitting antennas 1 and 2, respectively. Let us assume that w e can
are in the immediate neighborhood of the transmitting or the measure these two currents. The measurement is easy to cany out
receiving antennas, as illustrated in Figure 11. One does not even in a CDMA environment, where each receiver carries a unique
require any knowledge about the spatial locations of the various code.
antennas. The principle of reciprocity can help one to treat such a
complex situation without any knowledge of either the electro- Step 3: If we now excite transmitter 1 with a voltage W,'
magnetic environment in which the antennas are communicating,
and transmitter 2 with a voltage W: , then the current induced in
and it is not necessary to know the spatial locations of objects that
influence the electromagnetic coupling mechanisms. Based on the receiver 1 will he the sum of the following two terms:
principle of reciprocity, we now implement the system of adaptiv- W,'Ii + W i I i 2 .Of course, this is under the assumption that all the
ity on transmission. loads terminating the two transmitting and the receiving antennas
are all equal and numerically equal to
Let us assume that each receiver is given a particular code, as R., = RT2 = RR,= Rr2 = 1 0.The induced current at receiver 2
in a CDMA environment. Let us assume that we transmit the code
with amplitude 1 V. So, when VRt = 1V with VR2= V., = V., = 0, under these conditions will he 4'1; + WiI$2.If the signal is to be

72 IEEE Anlennas and Propagalion Magazine, Vol. 45, No. 3,June 2003
directed to receiver 1 and not to receiver 2, then the weights W,’ 8.1 Numerical Simulations
and W i should he such that the first sum should he unity and the
AS in example, consider three transmitting antennas, A,,
second sum should he zero. Equivalently, if we now excite trans-
I
mitter 1 with a voltage W,’ and transmitter 2 with a voltage U’;, A ? , and A,, located at a base station. The three antennas are con-
then the current induced in receiver I will be the sum of these two sidered to be three dipoles of radius 5 mm, and their lengths are
15 cm. If the operating frequency is 1 GHz, the three antennas are
terms: 4’1;+ K‘;Ii2 . In receiver 2, the induced current wili he a half-wavelength long. Next, we consider three receiving anten-
+ W;I;% . If we were to direct the signal to receiver 2 and nas, marked as A4, A,, and A,, as shown in Figure 12. They are
not to receiver 1, then.the first sum should be zero and the second spaced with a half wavelength between themselves, and they are
sum should be unity. Therefore, by appropriate choice of the volt- separated from the transmitters by a distance of 2 m. In this simu-
lation, we want to demonstrate that by choosing a proper set of
ages exciting the two msmitting antennas, it is possible to direct
the signal to a receiver chosen a priori, irrespective of the near- weighted excitations on the transmitters A , , A 2 , and Aj it is
field electromagnetic environment in which they may he operating. possible to direct the electromagnetic energy in such a way, that it
In summary, the four equations descnhed above that can direct the will add vectorially at only one receiver (say, A4), and that the
transmitted energy to a pre-selected receiver can he written in a voltages received from these three transmitting antennas will com-
compact matrix form as bine vectorially in such a way that they will produce practically
zero currents at the terminals of the remaining two receivers, A,
and A6 . In a similar fashion, it is possibie to select &other set of
weights that will result in receiver A, developing a large signal at
its terminals, while the signals received at the terminals of the
other two receivers, A, and A,, will he practically zero, and so
No loss of generality will occur with the choice of a value different
on.
that 1n for the impedance terminating both the transmitting and
the receiving antennas, since it will only affect the scale factor for
all four of the weights. We now illustrate how this can be accomplished based on the
principles of reciprocity. First, we excite receiver 1 with 1 V, and
Step 4: Solution of Equation(46) provides the set of that will induce a current in both receivers A, and A,, and also in
weights that, when used in conjunction with the signals received transmitters A , , A 2 , and A,, comprising the base station. These
from the two receivers producing currents at the transmitters, will induced currents will generate voltages across the loads of the
direct the transmitted energy to receiver 1 or to receiver 2, remaining five centrally loaded dipole antennas. Let us call the
depending on the values of the weights chosen. Hence, voltages received at transmitters A , , A 2 , and A,, due to an
excitation voltage of I v at receiver , ,
A ~ vi, vi2, and v13,
(47) respectively. Similarly, if we excite the antenna at the mobile, A,,
with 1 V, then the voltages received at the three hase-station
antennas will he V j , , Vj2 , and V j 3 . Even though voltages are
The caveat here is that if we apply a voltage equal to W,’ on also induced in the other two receivers, they are of no conse-
transmitting antenna #I and Wi on transmitting antenna #2 then quence. Finally, if we excite receiver A, with 1 V, then it will
the electromagnetic energy received from these two transmitting generate the following voltages at the three transmitters: V j , , V j 2 ,
antennas will add vectorially at the terminals of receiver #I, and
and V j 3, Now, the claim is that, based on the available information’
will he destructive at the terminals of receiver #2, producing zero
current at its loads. In contrast, if we apply the voltages W,’ on from the above three experiments, we can choose a set of complex

transmitting antenna #I and W: on transmitting antenna #2, then

the received electromagnetic energy will he vectorially destructive
at the terminals of receiver # I , resulting in practically zero current
flowing through its load, and will add vectorially at the terminals /AS /A*
of receiver #2, producing a large value for the induced current.

In summruy, by knowing the voltages that are induced at

each of the transmitting antennas due to each one of the receivers,
it is possible to select a set of weights based on reciprocity. This
relationship, based on reciprocity, can be applied only at the termi-
nals of the transmitting and receiving antennas. It is independent of
the size and shapes of the receiving antennas, nor does it depend
on the near-field environment.

Now, if there are more receivers thin transmitting antennas,

then it may not he possible to achieve a perfect match (namely,
simultaneously directing the received energy to a pre-selected
receiver and having practically zero energy induced on the rest of
the receivers), but some intermediate solution will still he possible.
We now illustrate these statements through examples. Figure 12. A six-antenna system.

iEEEAntennag and Propagation Magazine, Vol. 45, No. 3, June 2003 73

. ' voltages W,' , Wj , and Wj that, when fed to the three transmitting Similarly, using the above procedure, one can show that if
the excitation voltages are chosen for the three transmitting anten-
antennas, will result in an vectorial additive combination of the nas as follows:
electromagnetic energy at receiver I, while simultaneously induc-
ing practically no currents in the other two receivers. From the
principle of reciprocity, we are going to evaluate the induced cur- W,' = -0.679- j0.446,
rents at the loads located at the feed points of the three receivers,
when all of the three transmitting antennas A,, A,, and A, are W
: =0.137+ jO.011,
simultaneously excited with the voltages W,' , Wj , and Wi . The
receiver currents at the three locations will then he W: = -0.679 - j0.446,

At Receiver 1: @I;, + WjIi2 + WiIL, then the currents induced at the terminals of the three receivers will
he given by

At Receiver 2: W,'I;, + WiIi2 + WiI;, I., = -0.003 - j0.0003,

At Receiver 3: W,'II, + WdI12 + WiIi3 IT, =1.0- j0.01,

The objective is now to select the weights Win such a fashion that IT, = -0.003 + j0.0003
the voltage received will he finite at receiver 1, and zero at receiv-
ers 2 and 3. This is enforced through the following equation:
(all the currents are in d),

which clearly shows that the induced energy can he vectorially

additive at only receiver 2, while producing no appreciable CU-I
rents at the terminals of the other two receivers, 1 and 3.

Finally, if the excitation voltages at the three transmitting

antemas are chosen to be
Now, if the three transmitting antennas are excited with the fol-
lowing three complex voltages, obtained ffom the solution of
Equation (48), F3= 0.845 + j0.062,
W: =-0.415+ j0.366,
W; = -0.679 - j0.446,
Wi = -0.679 - j0.446,
W
: =-0.415+ j0.366,
Wi = 0.845 + j0.062,
then one can show that the currents induced at the terminals of the
three receivers will be
then the electromagnetic energy will add vectorially at the termi-
nals of receiver 1 , and the received voltages will cancel each other
resulting in practically zero currents at the terminals of the other IT =-0.0012- j0.0025
two receivers. Based on these three voltages, we carried out an
electromagnetic simulation using WIPL-D [161], and obtained the IT, = 4.003 + j0.001,
following three induced currents in the three receivers (with
Rq = RT2 = RT, = 50 C2 ):
Ir,=1.0- jO.001
IT = 1.0- j0.007,
(all the currents are in mA).
IT2 = 4.003+ jO.001,
This clearly demonstrates that by appropriately choosing the three
complex amplitudes of the excitations at different transmitting
IT, = 0.001 - j0.002 antennas simultaneously it is possible to create an electromagnetic
environment such that the induced electromagnetic energy will he
(all the currents are in d). vectorially additive at any one of the terminals of the pre-selected
receivers, and that the received voltages will try to cancel each
This clearly shows that for this particular choice of the other at the terminals of the other receiving antennas, resulting in
excitation voltages, the electromagnetic environment will be such practically zero current. Such a procedure can be carried out with-
that it will induce a large current at the load of receiver 1, and out knowing precisely the near-field environment in which the
practically zero currents at~theterminals of the other two receivers. transmitter and the receiver are operating, and without knowing the
This in no way implies that the current distihution on the receiv- precise spatial locations of the transmitting or the receiving anten-
ing antenna is zero. It is not. It is practically zero only at the feed nas.
terminals and nowhere else! This is because the principle of recip-
rocity only holds at the pre-specified terminals or ports of a net- In summary, the steps involved in this new method are as fol-
work. lows:

74 IEEEAntennas and Pmpagahon Magazine, Val. 45, No. 3. June 2003

 Step 1: Excite each of the mobile receiving antennas and Building Radio Communicate System Design,” IEEE Transactions
measure the currents that are induced in the base-station transmit- an Communications, COM-39,5, May 1991, pp. 794-807.
ting antennas.
4. W. C. Y. Lee, Mobile Cellular Telecommunications Systems,
Step 2: Using these induced currents, generate a maiix, the New York, McGraw Hill Publications, 1989.
inversion of which will contain the information necessary to excite
the transmitting antennas. 5 . B. Sklar, “Rayleigh Fading Channels in Mobile Digital Commu-
nication Systems. Part I,” IEEE Communications Magazine, 35, 9,
Step 3: Then, depending on the choice of which of the 1997, pp. 136-146.
mobile receiving antennas the signal is meant for, a set of weights
can be solved for that will result in a vectorial addition of all the 6. L. C. Godara, “Applications of Antenna Arrays to Mobile
incident electromagnetic energy at the terminals, maximizing the Communications. Part I. Performance Improvement, Feasibility,
induced currents. This will produce the maximum voltage at that and System Considerations,” Proceedings of the IEEE, 85, 7, July
pre-determined receiver, and practically zero induced currents at 1997,pp. 1031-1060.
the terminals ofthe other receivers.
7. M. Chryssomallis, “Smart Antennas,” IEEE Antennas and
By following the above procedure, one can carry out adaptiv- Propagation Magazine, 42, 3, June 2000, pp. 129-136.
ity on transmission based on the principle of reciprocity. This
methodology works even when there are near-field scatterers. 8. G. G. Raleigh and T. Boros, “Joint Space-Time Parameter Esti-
From OUT numerical simulations [160], we observed that this prin- mation for Wireless Communication Channels,” IEEE Transac-
ciple can be applied over a relatively wide band of frequencies. tions on Signal Processing, 46,5, May 1998, pp. 1333-1343.
Hence, when the ports of a communication system are well
defined, the propagation-path modeling can be avoided through the 9. G. L. Turin, “A Statistical Model for Urban Multipath Propaga-
use of the principle of reciprocity, at least in this case. tion,” IEEE Transactions on Vehicular Technology, VT-21, 1,
1972,pp. 1-9.

10. T. S. Rappaport, Wireless Communications: Principles &

9. Conclusions Practice, Upper Saddle River, NJ, Prentice Hall PTR, 1996.

Propagation models are needed not only for installation 11. M. F. Catedra, J. Perez, F. Saez de Adana, and 0. Gutieerez,
guidelines, but they also play a key part in any analysis or design “Efficient Ray-Tracing Techniques for Three Dimensional Analy-
that strives to mitigate interference. In this paper, we have SUI- ses of Propagation in Mobile Communications: Application to
veyed some of the typical propagation models that provide good Picocell and Microcell Scenarios,” IEEE Antennas and Propaga-
estimates for both large-scale and small-scale fading channels. IionMagazine, 40,2, April 1998, pp. 15-27,
Despite the enormous efforts to date, much work remains in under-
standing and predicting the characteristics of mobile communica- 12. S. Y. Seidel and T. S. Rappaport, “Site-Specific Propagation
tions channels. In addition, an eficient ray-tracing method has Prediction for Wireless In-Building Personal Communication Sys-
been presented for tracing rays in an indoor propagation system. tem Design,” IEEE Transactions on Vehicular Technology, VT-43,
An FDTD method has been described to analyze wave propagation 4, 1994, pp. 879-891.
through the walls in a building. Finally, it has been seen that the
principle of reciprocity can he applied to direct the transmitted 13. D. Molkdar, “Review on Radio Propagation into and within
energy to a pre-specified mobile unit without having any a priori Buildings,”IEE Proceedings-If, 138, 1, Feb. 1991, pp. 61-73.
howledge of the electromagnetic environment.
14. T. Okumura, E. Ohmori, and K. Fukuda, “Field Strength and
its Variability in VHF and UHF Land Mobile Service,” Review
Electrical Communication Laboratory, 16, 9-10, 1968, pp. 825-
I O . Acknowledgement 873.

15. M. Hata, “Empirical Formula for Propagation Loss in Land

Part of this work was supported by a grant from the National Mobile Radio Service,” IEEE Transactions on Vehicular Technol-
Science Foundation, ECS-9901361, and by the AFRL Sensors ogy, VT-29,3, 1980, pp. 317-325.
Directorate under grant F30602-95-1-0014.
16. J. Waltisch and H. L. Bertoni, “A Theoretical Model of UHF
Propagation in Urban Environments,” IEEE Transactions on
Antennas and Propagation, AP-36, 12, 1988, pp. 1788-1796.
11. References
17. K. Low, “Comparison of Urban Propagation Models with CW-
1. H. L. Bertoni, Radio Propagationfor Modern Wireless Systems, Measurements,” IEEE Vehicular Technology Sociep 42nd VTS
Upper Saddle River, NJ, Prentice Hall PTR, 2000, pp. 90-92. Conference. Frontiers of Technology. From Pioneers to the 2Ist
Century, 2, 1992, pp. 936-942.
2. 1. B. Andersen,.T. S. Rappaport, and S.Youshida, “Propagation
Measurement and Models for Wireless Communications Chan- 18. S. R. Saunders and F. R. Bonar, “Explicit Multiple Building
nels,’’ IEEE Communications Magazine, 33, 1, January 1995, pp. Diffraction Attenuation Function for Mobile Radio Wave Propa-
42-49. gation,” Electronics Letters, 21, 14, July 1991, pp. 1276-1277.

3. T. S. Rappaport, S. Y. Seidel, and K. Takamizawa, “Statistical 19. S. R. Saunders and F. R. Bonar, “Prediction of Mobile Radio
Channel Impulse Response Models for Factory and Open Plan Wave Propagation over Buildings of Irregular Heights and Spac-

IEEE Antennas and Propagation Magazine. Vol. 45, No. 3, June 2003 75
 ings,” IEEE TransactiZFonAntennas and Propagation, AP-42,2, 33. P. Valch, B. Segal, J. Lebel, and T. Pavlasek, “Cross-Floor
1994, pp. 137-144. Signal Propagation Inside a Contemporary Ferro-Concrete Build-
ing at 434, 864, and 1705 MHz,“ IEEE Transactions on Antennas
20. M. J. Neve and G. B. Rowe, “Assessment of GTD for Mobile and Propagation, AP-47,7, 1999, pp. 1230-1232.
Radio Propagation Prediction,” Electronics Leliers, 29, 7, April
1993, pp. 618-620. 34. S. E. Alexander, “Characterising Buildings for Propagation at
900 MHz,” Electronics Letlers, 19,20, September 1983, p. 860.
21. L. Juan-Llacer and N. Cardona, “UTD Solution for the Multi-
ple Building Diffraction Attenuation Function for Mobile Radio- 35. S. Y . Seidel and T. S. Rappaport, “914 MHz Path Loss Predic-
wave Propagation,” Electronics Letters, 33, 1, April 1997, pp. 92- tion Models for Indoor Wireless Communications in Multifloored
93. Buildings,” IEEE Transactions on Antennas and Propagation, AP-
40, 1992, pp. 207-217.
22. L. Juan-Llacer, L: Ramos, and N. Cardoua, “Application of
Some Theoretical Models for Coverage Prediction in Macrocell 36. T. S. Rappaport, “Characterization of UHF Multipath Radio
Urban Environments,” IEEE Transactions on Vehicular Technol- Channels in Factory Buildings,” IEEE Transactions on Antennas
ogv, VT-48,5, 1999, pp. 1463-1468. and Propagation, AP-37,8, August 1989, pp. 1058-1069.

23. D. Har, A. M. Watson, and A. G . Chadney, “Comment on Dif- 37. K. Pahlavan, R. Ganesb, and T. Hotaling, “Multipath Propaga-
kaction Loss of Rooftop-to-Street in Cost 231-Walfich-Ikegami tion Measurements on Manufacturing Floors at 910 MHz,” Elec-
Model,” IEEE Transactions on Vehicular Technology, VT-48, 5 , tronics Letters, 25,3, February 1989, pp. 225-227.
1999, pp. 1451-1452.
38. D.A. Haind, “Indoor Wide Band Radio Wave Propagation and
24. T. S. Rappaport and L. B. Milstein, “Effect of Radio Propaga- Models at 1.3 GHz and 4.0 GHz,” Electronics Cetters, 26, 21,
tion Path Loss on DS-CDMA Cellular Frequency Reuse Efficiency October 1990, pp. 1800-1802.
for the Reverse Channel,’’ IEEE Transactions on Vehicular Tech-
nologv, VT41, 3, August 1992, pp. 231-242. 39. Telesis Technologies Laboratory, “Experimental License
Report to FCC,” August 1991.
25. H. Xia, H. L. Bertoni, L. R Maciel, A. Lindsay-Stewart, and R.
Rowe, “Radio Propagation Characteristics for Line-of-sight 40. D.M. J. Davasirvatham, “A Comparison of Time Delay Spread
Microcellular and Personal Communications,” IEEE Transactions and Signal Level Measurements within Two Dissimilar Ofice
on Antennas and Propagation, AP-41, 10, October 1993, pp. 1439- Buildings,” IEEE Transactions on Antennas and Propagation, AP-
1447. 35, 3 , March 1987, pp. 319-324.

26. M. J. Feuerstein, K. L Blackard, T. S. Rappaport, S. Y. Seidel; 41. K. W. Cheung, J. H. M. Sau, and R. D. Murch, “A New
and H. H. Xia, “Path Loss, Delay Spread, and Outage Models as Empirical Model for Indoor Propagation Prediction,” IEEE Trans-
Functions of Antenna Height for Macrocellular System Design,” actions on Vehicular Technology,VT-47, 3, August 1998, pp. 996-
IEEE Transactions on Vehicular Technology, VT-43, 3, August 1001.
1994, pp. 487-498.
42. W. Honcbarenko, H. L. Bertoni, and J. Dailing, “Mechanisms
27. N. Beaunstein, “Prediction of Cellular characteristics for Var- Goveming UHF Propagation on Single Floors in Modem Office
ous Urban Environments,” IEEE Antennas and Propagation Buildings,” IEEE Transactions on Antennas and Propagation, AP-
Magazine, 41,6, 1999, pp. 135-145. 41, 4, 1992, pp. 496-504.

28. S. Ichitsubo, T. Furuno, T. Taga, and R. Kawasaki, “Multipath 43. S. Ruiz, Y . Samper, J. Perez, R. Agusti, and J. Olmos, “Soft-
Propagation Model for Line-of-Sight Street Microcells in Urban ware Tool for Optimising IndooriOutdoor Coverage in a Construc-
Area,” IEEE Transactions on Vehicular Technologv, VT-49, 2, tion Site,” Electronics Letters, 34, 22, October ,1998, pp. 2104-
2000, pp. 422-427. 2101.

29. T. S. Rappaport and S. Sandbu, “Radio-Wave Propagation for 44. G. M. Whitman, K. S. Kim, and E. Niver, “A Theoretical
Emerging Wireless Personal-Communication Systems,” IEEE Model for Radio Signal Attenuation Inside Buildings,” IEEE
Antennas and Propagation Magazine, 36, 5, October 1994, pp. 14- Transactions on Vehicular Technology, VT-44, 3 , August 1995,
24. pp. 621-629.
30. A. A. M. Saleh and R. L. Valemela, “A Statistical Model for 45. R. J. Luebbers, “Finite Conductivity Uniform GTD versus
Indoor Multipath Propagation,” IEEE Journal on Selected Areas in Knife Edge Diffraction in Prediction of Propagation Path Loss,”
Communications, 5,2, 1987, pp. 128-137. IEEE Transactions on Antennas and Propagation, AP-32, 1, Janu-
ary 1984, pp. 70-76.
31. R. J. C. Bultitude, “Measurement Characterization and Mod-
eling of Indoor 800/900 MHz Radio Channels for Digital Co&U- 46. F. Ikegami, T. Takeucbi, and S. Yoshida, “Theoretical Predic-
nications,” IEEE Communications Magazine, 5,6, 1987, pp. 5-12. tion of Mean Field Strength for Urban Mobile Radio,” IEEE
Transactions on Antennas and Propagation, AP-39, 3, March
1991, pp. 299-302.
32. F. C. Owen and C. D. Pundey, “In-Building Propagation at 900
MHz and 1650 MHz for Digital Cordless Telephone,” 6th Int. 47. W. K. Tam and V. N. Tran, “Propagation Modeling for Indoor
Conf. on Antennas and Propagation, ICAP’89, Part 2: Propagation, Wireless Communication,” Electronics & Communication Engi-
1989, pp. 276-281. neering Journal, October 1995, pp, 221-228.

76 IEEEAnlennas and Propagation Magazine, Vol. 45, No. 3, June 2003

48. J. W. Mckown and R. L. Hamilton, Jr., “Ray Tracing as a 62. W. Zhang, “Fast Two-Dimensional Diffraction Modeling for
Design Tool for Radio Networks,” IEEE Network Magazine, 5, Site-Specific Propagation Prediction in Urban Microcellular Envi-
Nov. 1991, pp. 27-30. ronments,” IEEE Tronsoctions on Vehicular Technology, VT-49,
2, March 2000, pp. 4281136.
49. S. Y . Tan and H. S. Tan, “Improved Three-Dimension Ray
Tracing Technique for Microcellular Propagation,” Electronics 63. H. Mokhtari and P. Lazaridis, “Comparative Study of Lateral
Letters, 31, 17,August 1995, pp. 1503-1505. Profile Knife-Edge Diffraction and Ray Tracing Technique Using
GTD in Urban Environment,” IEEE Transactions on Vehicular
50. S. Y . Tan and H. S. Tan, “Propagation Model for Microcellular Technology, VT-48, 1, January 1999,pp. 255-261.
Communications Applied to Path Loss-Measurements in Ottawa
City Streets,” IEEE Transaclions an Vehicular Technology, VT- 64. S. C. Kim, B. J. Guarino Jr., T. M. Willis 111, V. Erceg, S. J.
44,2, May 1995,pp. 313-317. Fortune and R. A. Valenzuela, L. W. Thomas, J. Ling, and J . D.
Moore, “Radio Propagation Measurements and Prediction Using
51. S.Y. Tan and H.S. Tan, “A Theory for Propagation Path-Loss Three-Dimensional Ray Tracing in Urban Environments at 908
Characteristics in a City-Street Grid,” IEEE Transactions on Elec- MHz and 1.9 GHz,” IEEE Transactions on Vehicular Technology,
fromagneficCompatibility, EMC-37,3, August 1995, pp. 333-342. VT-48, 3, 1999, pp, 931-944.

52. M. G. Sanchez, L. de Haro, A. G. Pino, and M. Calvo, 65. J. H. Tamg, W. R. Chang, and B. J. Hsu, “Three-Dimensional
“Exhaustive Ray Tracing Algorithm for Microcellular Propagation Modeling of 900-MHz and 2.44-GHz Radio Propagation in Con%
Prediction Models,” Electronics Letfers, 32, 7, March 1996, pp. dors,” IEEE Transactions on Vehicular Technology, VT-46, 2,
624-625. May 1997, pp. 519-527.

53. S. H. Chen and S. K. Jeng, “SBR Image Approach for Radio 66. G. Liang and H. L. Bertoni, ”A New Approach to 3-D Ray
Wave Propagation in Tunnels with and without Traffic,” IEEE Tracing for Propagation Prediction in Cities,” IEEE Transactions
Transactions on Vehicular Technology, VT-45, 3, August 1996, on Antennas and Propagation, AP-46,6, June 1998, pp. 853-863.
pp. 570-578.
67. G. Durgin, N. Patwari, and T. S. Rappaport, “Improved 3D
54. S. H. Chen and S. K. J a g , “An SBMmage Approach for Ray Launching Method for Wireless Propagation Prediction,”
Radio Wave Propagation in Indoor Environments with Metallic ElectronicsLetters, 33, 16, July 1997, pp. 1412-1413.
Furniture,” IEEE Transactions on Antennas and Propagation, AP-
45, 1, January 1997, pp. 98-106. 68. G. E. Athanasiadou, A. R. Nix, and J. P. McGeehan, “A
Microcellular Ray-tracing Propagation Model and Evaluation of Its
55. F. Villanese, W. G. Scanlon, N. E. Evans, and E. Gamhi, Narrow-Band and Wide-Band Predictions,” IEEE Journal on
“Hybrid ImageRay-Shooting UHF Radio Propagation Predictor Selected Areas in Communications, 18, 3, March 2000, pp. 322-
for Populated Indoor Environments,” Electronics Leffers 35, 21, 335.
October 1999, pp. 1804-1805.
69. J. P. Rossi, J. C. Bic, A. J. Levy, Y. Gahillet, and M. Rosen, “A
56. M. C. Lawton and .I. P. McGeehan, “The Application of a Ray Launching Method for Radio-mobile Propagation in Urban
Deterministic Ray Launching Algorithm for the Prediction of Area,” IEEE Intemational Symposium on Antennas and Propaga-
Radio Channel Characteristics in Small-Cell Environments,” IEEE tion Digest, London, Ontario, Canada, 3, 1991, pp. 1540-1543.
Transaclions on Vehicular Techriology, VT-43, 4, November
1994, pp. 955-969. 70. W. M. O’Brien, E. M. KeMy, and P. J. Cullen, “An Efficient
Implementation of a Three-Dimensional Microcell Propagation
57. U. Dersch and E. Zollinger, “Propagation Mechanisms in Tool for Indoor and Outdoor Urban Environments,” IEEE Trans-
Microcell and Indoor Environments,” IEEE Transactions on actions on Vehicular Technology, VT-49, 2, March 2000, pp. 622-
Vehicular Technology, VT-43,4, November 1994, pp, 1058-1066. 630.

58. K. Rizk, J. F. Wagen, and F. Gardiol, “Two-Dimensional Ray- 71. M. F. Catedra, J. Perez-Saez, F. de Adana, and 0. Gutierrez,
tracing Modeling for Propagation Prediction in Microcellular Envi- “Efficient Ray-Tracing Techniques for Three-Dimensional Analy-
ronments,” IEEE Transactions on Vehicular Technology, VT-46, ses of Propagation in Mobile Communications: Application to
2, May 1997,pp. 508-518. Picocell and Microcell Scenarios,” IEEE Antennas and Propaga-
fionMagazine, 40,2, April 1998, pp. 15-28.
59. G. M. Whitman, K. S. Kim, and E. A. Niver, “Theoretical
Model for Radio Signal Attenuation Inside Buildings,” IEEE 72. V. Erceg, S. J. Fortune, J. Ling, A. J. Rustako Jr., and R. A.
Transactions on Vehicular Technology, VT-44, 3, August 1995, Valenzuela, “Comparisons of a Computer-Based Propagation Pre-
pp. 621-629. diction Tool with Experimental Data Collected in Urban Micro-
cellular Environments,” IEEE Journal on Selected Areas in Com-
60. J. H. Tamg and T. R. Liu, “Effective Models in Evaluating munications, 15,4, May 1997, pp. 677-684.
Radio Coverage on Single Floors of Multifloor Buildings,” IEEE
Transactions on Vehicular Technology, VT-48, 3, May 1999, pp. 73. J. W. Schuster and R. I. Luebhers, “Comparison of GTD and
782-789. FDTD Predictions for UHF Radio Wave Propagation in a Simple
Outdoor Urban Environment,” IEEE International Symposium on
61. S. C. Jan and S. K. Jeng, “A Novel Propagation Modeling for Antennas and Propagation Digesl, 3,1997, pp. 2022-2025.
Microcellular Communications in Urban Environments,” IEEE
Transactions on Vehicular Technology, VT-46, 4, November 74. G. D. Kondylis, F. DeFlaviis, G. J. Pottie, and Y. Rahmat-
1997, pp. 1021- 1026. Samii, “Indoor Channel Characterization for Wireless Communi-

IEEEAntennasand Propagation Magazine, Vol. 45, No. 3, June 2003 77

 cations Using Reduced Finite Difference Time Domain,” IEEE 88. B. Chopard, P. 0. Luthi, and I. F. Wagen, “Lattice BoltzmaM
Vehicular Technologv Conference, VT-3, 1999, pp. 1402-1406. Method for Wave Propagation in Urban Microcells,” IEE Pro-
ceedings - Antennas and Propagation, 144, 4, August 1997, pp.
75. A. Lauer, 1. Woff, A. Bahr, J. Pamp, J. Kunisch, “Multi-Mode 251-255.
FDTD Simulations of Indoor Propagation Including Antenna
Properties,” in Proceedings of the I995 IEEE 45th Vehicular 89. L. J. Greenstein, D. G. Michelson, and V. Erceg, “Moment-
Technology Conference, Chicago, IL,pp. 454-458. Method Estimation of the Ricean K-factor,” IEEE Communica-
tions Letters, 3 , 6 , June 1999, pp. 175-176.
76. W. Ying, S. Safavi-Naeini, and S. K. Chaudhuri, :’A Hybrid
Technique Based on Combining Ray Tracing and FDTD Methods 90. V. Fung, T. S. Rappaport, and B. Thomas, “Bit Error Simula-
for Site-Specific Modeling of Indoor Radio Wave Propagation,” tion for pi 14 DQPSK Mobile Radio Communications Using Two-
IEEE Transacrions on Antennas and Propagation, VT-48, 5 , May ray and Measurement-based Impulse Response Models,” IEEE
2000, pp. 743-754. Journal on Selected Areas in Communications, 11, 3, April 1993,
pp. 393-405.
77. C. Yang, B. Wu, and C. KO,“A Ray-Tracing Method for Mod-
eling Indoor Wave Propagation and Penetration,” IEEE Transac- 91. C. C. Hess, Handbook of Land-Mobile Radio System Coverage,
tions on Antennas andPropagation, VT-46,6, June 1998, pp. 907- Norwood, MA, Anech House, Inc.
919.
92. A. Safak, “Statistical Analysis of the Power Sum of Multiple
78. B. De Backer, H. Bojeson, F. Olyslager, and D. De Zutter, Correlated Lognormal Components,” ZEEE Transactions on
“The Study of Wave-Propagation through a Windowed Wall at 1.8 Vehicular Technology,VT-42, 1, February 1993, pp. 58-61.
GHz,” IEEE 46th Vehicular Technology Conference, 1, 1996, pp.
165-169. 93. H. Suzuki, “A Statistical Model for Urban Radio Propagation,”
IEEE Transactions on Communications, COM-25, 1977, pp. 673-
79. C. Yang and B. Wu, “Simulations and Measurements for 680.
Indoor Wave Propagation Through Periodic Structures,” IEEE
International Symposium on Antennas and Propagation Digest, 1, 94. A. Krantzik and D. Wolf, “Analysis of a Modified Suzuki
1999, pp. 384-387. Fading Channel Model,” 1989 Intemational Conference on Acous-
tics, Speech, and Signal Processing, 1989 ICASSP-89,4, pp. 2250-
80. Z. Sandor, L. Nagy, 2. Szabo, and T. Csaba, “3D Ray 2253.
Launching and Moment Method for Indoor Propagation Purposes,”
The 8th IEEE International Symposium on Personal, Indoor and 95. M. Patzold, U. Killat, and F. Laue, “A Deterministic Digital
MobileRadio Communicafions, 1997, PIMRC ‘97, 1, pp. 130-134. Simulation Model for Suzuki Processes with Application to a
Shadowed Rayleigh Land Mobile Radio Channel,” IEEE Transac-
81. A. Neskovic and D. Paumovic, “Indoor Electric Field Level tions on Vehicular Technology, VT-452, May 1996, pp. 318-331.
Prediction Model Based on the Artificial Neural Networks,” ZEEE
Communicalions Letters, 4, 6,2000, pp. 190-192. 96. IEEE Vehicular Technology Society Committee on Radio
Propagation, “Coverage Prediction for Mobile Radio Systems
82. K. E. Stocker, B. E. Gschwendtner, and F. M. Landstorfer, Operating in the 8001900 MHz Frequency Range,” IEEE Transac-
‘‘Neural Network Approach to Prediction of Terrestrial Wave tions on Vehicular Technology,VT-37, 1, February 1988, pp. 3-72.
Propagation for Mobile Radio,” ZEE Proceedings H Microwaves,
Antennas andt‘ropagation, 140,4, August 1993, pp. 315-320. 97. K. Yip and T. Ng, “A Simulation Model for Nakagami-m
Fading Channels, m > l”, IEEE Transactions on Communications,
83. P. Chang and W. Yang, “Environment-Adaptation Mobile COM-48,2, Fehruaq 2000, pp. 214-221.
Radio Propagation Prediction Using Radial Basis Function Neural
Networks,” IEEE Transactions on Vehicular Technologv, VT-46, 98. F. Vatalaro, “Generalized Rice-Lognormal Channel Model for
I, February 1997, pp, 155-160. Wireless Communications,” Electronics Letters, 31, 22, October
1995, pp. 1899-1900.
84. A. A. Zaporozhets, “Application of Vector Parabolic Equation
Method to Urban Radiowave Propagation Problems,” ZEE Pro- 99. Y. Karasawa and H. Iwai, “Modeling of Signal Envelope Cor-
ceedings - Microwaves, Antennas and Propagation, 146,4, August relation of Line-of-sight Fading with Applications to Frequency
1999, pp. 253-256. Correlation Analysis,” IEEE Transactions on Communications,
COM-42,6, June 1994, pp. 2201-2203.
85. C. Brennan and P. J. Cullen, “Application of the Fast Far-field
Approximation to the Computation of UHF Path Loss Over 100. T. T. Tjbung and C. C. Chai, “Fade Statistics in Nakagami-
Irregular Terrain,” IEEE Transactions on Antennas and Propaga- Lognormal Channels,” IEEE Transactions on Communications,
tion, AP-46,6, June 1998, pp. 881-890. COM-47,12, December 1999, pp. 1769-1772.

86. R. Mazar and A. Bronshtein, “Propagation Model of a City 101. A. Abdi and M. Kaveh, “K Distribution: An Appropriate Sub-
Slreet for Personal and Microcellular Communications,” Elec- stitute for Rayleigb-lognormal Distribution in Fading-Shadowing
tronics Letters, 33; 1, January 1991, pp. 91-92. Wireless Channels,” Electronics Letters, 34, 9, April 1998, pp.
851-852.
87. N. Blaunstein, “Average Field Attenuation in the Street
Waveguide,” IEEE Transactions on Antennas and Propagation, 102. G. E. Corazza and F. Vatalaro, “A Statistical Model for Land
AP-46,12, December 1998, pp. 1782-1789. Mobile, Satellite Channels and its Application to Nongeostationary

78 IEEE Antennas and Propagation Magazine, Vol. 45, No. 3, June 2003
Orbit Systems,” IEEE Transactions on Vehicular Technology, VT- 116. T. R Liu and J. H. Tamg, “Modeling and Measurement of
43, August 1994, pp. 738-741. 2.44 GHz Radio Out-Of-Sight Propagation on Single Floors”,
Microwave and Optical Technology Letters, 14, 1, January 1997,
103. P. G. Babalis and C. N. Capsalis, “Impact of the Combined pp. 56-59.
Slow and Fast Fading Channel Characteristics on the Symbol Error
Probability for Multipath Dispersionless Channel Characterized by 117. R. Ganesh and K. Pahlavan, “Statistics of Short Time and
a Small Number of Dominant Paths,” IEEE Transactions on Com- Spatial Variations Measured in Wideband Indoor Radio Channels”,
munications, COM-47, 5 , May 1999, pp. 653-657. IEE Proceedings-H Antennas and Propagation, 140, 4, August
1993, pp. 297-302.
104. D. S. Polydorou and C. N. Capsalis, “A New Theoretical
Model for the Prediction of Rapid Fading Variations in Indoor 118. M. C. Lawton and J. P. McGeehan, “The Application of a
Environment,” IEEE Transactions on Vehicular Technology, VT- Deterministic Ray Launching Algorithm for the Prediction of
46, August 1997, pp. 748-755. Radio Channel Characteristics in Small-Cell Environments,” IEEE
Transactions on Vehicular Technology, VT-43, 4, November
105. R. J. C. Bultitude, P. Melancon, H. Zaghloul, G. Morrison, 1994, pp. 955-969.
and M. Prokki, “The Dependence of Indoor Radio Channel Multi-
path Characteristics of Transmimeceiver Ranges,” IEEE Journal 119. H. Hashemi, “Impulse Response Modeling of Indoor Radio

979-990.
.
on Selected Areas in Communications., 11., 7.. SeDtember ,..
1993. DD. Propagation Channels,” IEEE Journal on Selected Areas in Com-
munications, 11, 7, September 1993, pp. 967-978.

106. D. M. J. Devasirvatham, R. R. Murray, and C. Banejee, 120. S. Y. Seidel, T. S. Rappaport, S. Jain, M. L. Lord, and R.
“Time Delay Spread Measurements at 850 MHz and 1.7 GHz Singh, “Path Loss, Scattering and Multipath Delay Statistics in
Inside a Metropolitan Office Building,” Electronics Letters, 25, 3, Four European Cities for Digital Cellular and Microcellular
February 1989, pp. 194-196. Radiotelephone,” IEEE Transactions on Vehicular Technolow,
VT-40,4, November 1991, pp. 721-730.
107. H. Zaghbul, M. Fattouche, G. Morrison, and D. Tholl, “Com-
parison of Indoor Propagation Channel Characteristics at Different 121. T. S. Rappaport and S. Y. Seidel, “900 MHz Multipath
Frequencies,’’ Electronics Letters, 27, 22, October 1991, pp. 2077- Propagation Measurements in Four States Cities,” Electronics
2079. Letters, 25, 15, July 1989, pp. 956-958.

108. G. J. M. Janssen, P. A. Stigter, and R. Prasad, “Wideband 122. T. Taga, T. Furuno, and K. Suwa, “Channel Modeling for 2-
Indoor Channel Measurements and BER Analysis of Frequency GHz-band Urban Line-of-sight Street Microcells,” IEEE Transac-
Selective Multipath Channels at 2.4, 4.75, and 11.5 GHz,” IEEE tions on Vehicular Technology, VT-48, 1, January 1999, pp. 262-
Transactions on Communications, COM-44, 10, October 1996, pp. 272.
1272-1288.
123. H. Son and N. Myung, “A Deterministic Ray Tube Method
109. K. Pahlavan, R. Ganesh, and T. Hotaling, “Multipath Propa- for Microcellular Wave Propagation Prediction Model,” IEEE
gation Measurements on Manufacturing Floors at 910 MHz,” Transactions on Antennas and Propagation, AP-47, 8, August
Electronics Letters, 25,3, February 1989, pp. 225-227. 1999,pp. 1344-1350.

110. D. A. Hawbaker and T. S. Rappaport, “Indoor Wideband 124. L. J. Greenstein, V. Erceg, Y. S. Yeh, and M. V. Clark, “A
Radiowave Propagation Measurements at 1.3 GHz and 4.0 GHz,” New Path-Gain/Delay-Spread Propagation Model for Digital Cel-
Electronics Letters, 26,21, October 1990, pp. 1800-1802. lular Channels,” IEEE Transactions on Vehicular Technology,VT-
46,2, May 1997, pp. 477-485.
111. L. Talbi and G. Y. Delisle, “Experimental Characterization of
EHF Multipath Indoor Radio Channels,” IEEE Journal on Selected 125. R. Ganesb and K. Pahlavan, “Statistical Modelling and Com-
Areas in Communications, 14,3, April 1996, pp. 431-440. puter Simulation of Indoor Radio Channel,” IEE Proceedings I
Communications, Speech and Vision, 138, 3, June 1991, pp. 153-
112. R. J. C. Bultitude and G. K. Bedal, “Propagation Characteris- 161.
tics on Microcellular Urban Mobile Radio Channels at 910 MHz,”
IEEE Journal on Selected Areas in Communications, 7 , 1, January 126. R. A. Valenzula, 0. Lankon, and D. L. Jacobs, “Estimating
1989, pp. 31-39. Local Mean Signal Strength of Indoor Multipath Propagation,”
IEEE Transactions on Vehicular Technology, VT-46, 1, February
113. R. J. C. Bultitude, R. F. Hahn, and R. J. Davies, “Propagation 1997, pp. 203-212.
Considerations for the Design of an Indoor Broad-Band Cornmuni-
cations System at EHF,” IEEE Transactions on Vehicular Tech-
nologv, VT-47, 1, February 1998, pp. 235-245. 127. G. E. Athanasiadou, A. R. Nix, and J. P. McGeehan, “A
Microcellular Ray-tracing Propagation Model and Evaluation of its
114. S. Icbitsubo, T. Furuno, T. Taga, and R. Kawasaki, “Multi- Narrow-Band and Wide-Band predictions,” IEEE Journal on
path Propagation Model for Line-of-sight Street Microcells in Selected Areas in Communications, 18, 3 , March 2000 pp. 322-
Urban Area,” IEEE Transactions on Vehicular Technology, VT- 335.
49, March 2000, pp. 422-427.
128. Z. Ji, B. H. Li, H. X. Wang, H. Y. Chen, and Y. G. Zhou, “An
115. S. C. Kim, H. L. Bertoni, and M. Stem, “Pulse Propagation Improved Ray-Tracing Propagation Model for Predicting Path
Characteristics at 2.4 GHz lnside Buildings,” IEEE Transactions Loss on Single Floors,” Microwave and Optical Technology Let-
on Vehicular Technology,VT-45,3, August 1996, pp. 579-592. ters, 22, 1, 1999, pp. 39-41.

IEEE Antennas and Propagation Magazine, Vol. 45, No. 3, June 2003 79
 129. W. K. Tam and V. N.Tran, “Multi-Ray Propagation Model 143. D. Ullmo and H. U. Baranger, “Wireless Propagation in
for Indoor Wireless Communications,” Electronics Letters, 32, 2, Buildings: A Statistical Scattering Approach,” IEEE Transactions
1996, pp. 135-137. on Vehicular Technology, VT-48,3, 1999, pp. 947-955.

130. K. R. Chang and H. T. Kim, “Improvement of the Computa- 144. S. Kozono and A. Taguchi, “Mobile Propagation Loss and
tion Efficiency for a Ray-Launching model,” IEE Proceedings - Delay Spread Characteristicswith a Low Base Station Antenna on
Microwaves, Antennas, and Propagation, 145, 4, 1997, pp. 303- an Urban Road,” IEEE Transactions on Vehicular Technology,
308. VT-42, 1, 1993,pp. 103-109.

131. T. B. Gibson and D. C. Jenn, “Prediction and Measurement of 145. S. Ohayashi and J. Zander, “A Body-Shadowing Model for
Wall Insertion Loss,” IEEE Transactions on Antenna and Propa- Indoor Radio Communication Environments,” IEEE Transactions
gation, AP-47, 1, 1999, pp. 55-57. on Antennas and Propagation, AP-46,6, 1998, pp. 920-927.

132. M. F. Iskander and 2. Yun, “Propagation Prediction Models 146. F. Bahich and G. Lombardi, “Statistical Analysis and Char-
for Wireless Communication Systems,” IEEE Transactions on acterization of the Indoor Propagation Channel,’’ IEEE Transac-
Microwave Theory and Techniques, MTT-50, 3, 2002, pp. 662- tions on Communications, COM-48,3,2000, pp. 455-464.
673.
147. H. Hashemi and D. Tholl, “Statistical Modeling and Simula-
133. G. Ghobadi, P. R. Shepherd, and S . R. Pennock, “2D Ray- tion of the RMS Delay Spread of Indoor Radio Propagation Chan-
Tracing Model for Indoor Radio Propagation at Millimeter Fre- nels,” IEEE Transactions on Vehicular Technology, VT-43, 1,
quencies, and the Study of Diversity Techniques”, IEE Proceed- 1993, pp. 110-120.
ings - Microwaves, Antennas, and Propagation, 145, 4, 1998, pp.
349-353. 148. J. A. Wepman, J. R. Hoffman, and L. H. Loew, “Analysis of
Impulse Response Measurements for PCS Channel Modeling
134. W. Honcharenko and H. L. Bertoni, “Transmission and Applications,” IEEE Transactions on Vehicular Technology, VT-
Reflection Characteristics at Concrete Block Walls in the UHF 44,3,1995, pp. 613-620.
Bands Proposal for Future PCs,” IEEE Transactions on Antennas
andPropagation, AP-42,2, February 1994, pp. 232-239. 149. C. L. Holloway, M. G. Cotton, and P. McKenna, “A Model
for Predicting the Power Delay Profile Characteristics Inside a
135. C. L. Holloway and P. L. Perini, “Analysis of Composite Room,” IEEE Transactions on Vehicular Technology, VT-48, 4,
Walls and Their Effects on Short-Path Propagation Modeling,” 1999, pp. 1110-1120.
IEEE Transactions on Vehicular Technology, VT-46, 3, August
1997, pp. 730-738. 150. P.E. Driessen, “Prediction of Multipath Delay Profiles in
Mountainous Terrain,” IEEE Joumal on Selected Areas in Com-
136. Z. Ji, B. H. Li, H. X. Wang, El. Y. Chen, and T. K Sarkar, munications, 18,3,2000, pp. 336-346.
“Efficient Ray-Tracing Methods for Propagation Prediction for
Indoor Wireless Communications,” IEEE Antennas and Propaga- 151. Y. Li, “A Theoretical Formulation for the Distribution Den-
tion Magazine, 43, 2, April 2001, pp. 41-49. sity of Multipath Delay Spread in a Land Mobile Radio Environ-
ment,” IEEE Transactions on Vehicular Technology, VT-43, 2,
137. Z. Ji, T. K. Sarkar, and B. H. Li, “Analysis of the Effects of 1994, pp. 379-388.
Walls on Indoor Wave Propagation Using the FDTD Method,”
Microwave and Optical Technology Letters, 29, 1, 2001, pp. 19- 152. M. C. Vanderveen, C. B. Papadias, and A. Paulraj, “Joint
21. Angle and Delay Estimation (JADE) for Multipath Signals Aniv-
ing at an Antenna Array,” IEEE Communicatiow Letters, 1, 1,
138. C. A. Zelley and C. C. Constantinon, “A Three-Dimensional January 1997,pp. 12 -14.
Parabolic Equation Applied to VHFiUHF Propagation over
Irregular Terrain,” IEEE Transactions on Antennas and Propaga- 153. M. Wax and A. Leshem, “Joint Estimation of Time Delays
tion, AP-47, 10, 1999, pp. 1586-1596. and Directions of Anival of Multiple Reflections of a Known Sig-
nal,” IEEE Transactions on Signal Processing, SP-45, IO, October
139. D. J. Donohue and J. R. Kuttler, “Propagation Modeling over 1997, pp. 2477-2484.
Terrain Using the Parabolic Wave Equation,” IEEE Transactions
on Antennas and Propagation, AP-48,2,2000, pp. 260-277. 154. M. C. Vanderveen, C. B. Papadias, and A. Paulraj, “Joint
Angle and Delay Estimation (JADE) for Multipath Signals Arriv-
140. Y. P. Zhang, Y. Hwang, and J. D. Parsons, “UHF Radio ing at an Antenna Array,” IEEE Communicatiow Letters, 1, 1,
Propagation Characteristics in Straight Open-groove Structures,’’ January 1997, pp. 12 -14.
IEEE Tramactions on Vehicular Technology, VT-48, 1, 1999, pp.
249-254. 155. A. J. van der Veen, M. C. Vanderveen, and A. J. Paulraj,
“Joint Angle and Delay Estimation Using Shift-invariance Proper-
141. D. Har, H. H. Xia, and H. L. Bertoni, “Path-Loss Prediction ties,” IEEE Signal Processing Letters, 4, 5 , May 1997, pp. 142-
Model for Microcells,” IEEE Transactions on Vehicular Technol- 145.
ogy, VT-48,5, 1999, pp. 1453-1462.
156. A. J. van der Veen, M. C. Vanderveen, and A. J. Paulraj,
142. H. L. Bertoni, W. Honcharenko, L. R. Macel, and H. H. Xia, “Joint Angle and Delay Estimation Using Shift-Invariance Tech-
“UHF Propagation Prediction for Wireless Personal Communica- niques,” IEEE Transactions on Signal Processing, SP-46, 2, Feb-
tions,” Proceedings of the IEEE, 82, 9, 1994, pp. 1333-1359. ruary 1998, pp. 405-418.

80 IEEE Antennas and Pmpagation Magazine. Val 45. No. 3. June 2003
157. M. C. Vanderveen, A. J. Van der Veen, and A. Paulraj, Transactions on Electromagnetic Compatibility in 1979, and for
“Estimation of Multipath Parameters in Wireless Communica- the 1997 National Radar Conference. He received the College of
tions,” IEEE Transactions on Signal Processing, SP-46, 3, March Engineering Research Award in 1996, and the Chancellor’s Cita-
1998, pp. 682-690. tion for Excellence in Research in 1998 at Syracuse University. He
was elected Fellow of the IEEE in 1991. He was an Associate
158. M. C. Vanderveen, B. C. Ng, C. B. Papadias, and A. Paulraj, Editor for Feature Articles of the IEEE Antennas and Propagation
“Joint Angle and Delay Estimation (JADE) for Signals in Multi- Society Newsletter, and he was the Technical Program Chairman
path Environments,” Conference Record of the Thirtieth Asilomar for the 1988 IEEE International Symposium on qtennas and
Conference on Signals, Systems and Computers, 2 , 1996, pp. 250- Propagation and USNC/URSI National Radio Science Meeting.
1254. Currently, he is a Distinguished Lecturer for the IEEE Antennas
and Propagation Society. He is on the editorial boards of Journal
159. B. m a n and W. I.,Fitzgerald, “Joint Angle and Delay of Electromagnetic Waves and Applications and Microwave and
Estimation (JADE) for Fading Multipath Signals in a -stable Optical Technology Letters. He bas been a US National Committee
noise,” in Proceedtngs of Signal Processing Advances in Wireless representative to many URSI General Assemblies. He was the
Communications, 1999, pp. 333-336. Chair of the URSl Inter-Commission Working Group on Time-
Domain Metrology (1990-1996). Dr. Sarkar is a member of Sigma
160. T. K. Sarkar et. al., Smart Antennas, New York, John Wiley Xi and USNCiLiRSI Commissions A and B. He received the title
and Sons, March 2003. Docteur Honoris Causa from Universite Blaise Pascal, Clermont
Ferrand, France, in 1998, and the Friends of Clermont Ferrand
161. B. M. Kolundzija, J. S . Ognjanovic, and T. K. Sarkar, WIPL- award in 2000.
D: Electromagnetic Modeling of Composite Metallic and Dielec-
tric Structures, Norwood, MA, Artech House, 2000.

Introducing the Feature Article Authors

Zhong Ji received both the BS and MS degrees in Electronic

Engineering from Shandong University, China, in 1988 and 1991,
respectively. He received the PhD degree from Shanghai Jiao Tong
University, China, in 2000.

Tapau Kumar Sarkar received the BTech degree from the He w& a teacher of Shandong University from 1991 to 1997.
Indian Institute of Technology, ‘Kharagpnr,.India, in 1969; the He is currently a Research Associate with Syracuse University,
MScE degree from the University of New Brunswick, Fredericton, Syracuse, NY. His research interests are in the areas of propagation
Canada, in 1971; and the MS and PhD degrees from Syracuse Uni- models for wireless communications, and time-domain analysis of
versity, Syracuse, New York in 1975. electromagnetic fields.

From 1975 to 1976, he was with the TACO Division ofGen-

era1 Instruments Corporation. He was with the Rochester Institute Kyuugjuug Kim was horn in Seoul, Korea. He received the
of Technology, Rochester, NY, from 1976 to 1985. He was a BS degree from Inha University, Korea, and the MS degree from
Research Fellow at the Gordon McKay Laboratory, Harvard Uni- Syracuse University, Syracuse, NY. He is currently working
versity, Cambridge, MA, from 1977 to 1978. He is now a Profes- towards the PhD degree in the Department of Electrical Engineer-
sor in the Department of Electrical and Computer Engineering, ing at Syracuse University.
Syracuse University, Syracuse, NY. His current research interests
deal with numerical solutions of operator equations arising in He was a Research Assistant at Syracuse University from
electromagnetics and signal processing with application to system 1998 to 2001, and has been a Graduate Fellow since 2001. His cur-
design. He obtained one of the ”best solution” awakds in May, rent research interests include digital signal processing related to
1977, at the Rome Air Development Center (RADC) Spectral adaptive antenna problems and wavelet transforms.
Estimation Workshop. He has authored or coauthored ten hooks,
including most recently Iterative and Self Adaptive
Finite-Elements in Elecfromagnetic Modeling and Wavelet Appli- AbdeUatif Medouri was horn in Tetouan, Morocco, on
cations in Engineering Electromagnetics, which were published in December 1 1 , 1965. He received the BS degree in Physics from
1998 and 2002, respectively by Artech House. He has published University Abdelmalek Essaadi, Tetnuan, Morocco, in 1989, and
more than 210 journal articles and numerous conference papers, the PhD degree from the University of Granada, Spain, in 1995. He
and has written chapters in 28 hooks. His hook on Smart Antennas was a research student from 1989 to 1995 in the Department of
will be published by John Wiley in March, 2003. Applied Physics, University of Granada, Spain. He was a Research
Scientist at the National Research Center of Morocco from 1996 to
Dr. Sarkar is a registered Professional Engineer in the State 2001. He is currently a professor at the National School of Applied
of New York. He received the Best Paper Award of the IEEE Physics, Tangier, Morocco. His research interests include sensor

IEEEAntennas and Propagation Magazine, Vol. 45, No. 3, June 2003 ai

array signal processing, identification of radar targets, and high- Fditor’sComments Com’nuedfrompage 40
order statistics. implementation and further discussion with IEEE IT staff, Dan
Senese told me that the use of this spam-labeling software would
he offered to IEEE alias-account users only on an “opt-in” basis. I
think that’s an excellent solution. I salute and thank Dan Senese
and the IEEE staff for their willingness to listen, and for making
the decision they did.

You’ll have to decide for yourself whether or not you want

the IEEE to implement such filtering on your e-mail, if you use an
IEEE e-mail alias. For now, at least, I plan not to. Because my
opposition to this is not jnst a matter of practicality, but also of
principle, I hope I never have to. If I were ever to do so, I would
never turn the final decision about the disposition of an e-mail
message so labeled over to software: the potential consequence of
Magdalena Salazar-Palma was horn in Granada, Spain. She a “false positive” is too great. At least, now you’ll he~ahleto make
received the PhD degree in Ingeniero’de Telecomunicaciirn from the decision for yourself.
the Universidad Politknica de Madrid (Spain), where she is a Pro-
fessor Titular of the Departamento de Sefiales, Sistemas y Radio- A final note: As this is going to press, I have yet another
comunicaciones (Signals, Systems and Radiocommunications situation in which my e-mails to an author are being rejected as
Department) at the Escuela Tecnica Superior de Ingenieros de spam, even though I’ve tried sending them via two e-mail accounts
Telecomunicaci6n of the same university. She has taught courses with toti$ly different ISPs. This is beginning to he as serious a
on electromagnetic field theory, microwave and antenna theory, problem as the spam, itself!
circuit networks and filter theory, analog and digital c o v u n i c a -
tion systems theory, and numerical methods for electromagnetic
field problems, as well as related laboratories. Her research within Thank You!
the Grupo de Microondas y Radar (Microwave and Radar Group)
is in the areas of electromagnetic field theory; and computational At the Columbus Symposium, I was very honored and grate-
and numerical methods for microwave structures, passive compo- ful to receive the M - S Outstanding Senice Award Cor being Edi-
nents, and antenna analysis; design, simulation, optimization, tor of the Newsletter and Editor-in-Chief of the Magazine, Actu-
implementation, and measurements of hybrid and monolithic ally, the wording of the citation was very nice and somewhat more
microwave integrated circuits; and network and filter theory and detailed; you can read it, along with information about the other
design. AP-S and IEEE awards presented there to M - S members, in Allan
Schell’s report of the Awards and Fellow Committee in this issue.
She bas authored a total of 10 contributions for chapters and I’m going to take the liberty of repeating most of the comments I
articles in hooks, 15 papers in international journals, and 75 papers made in receiving the award, because it’s important that all of
in international symposiums and workshops, plus ‘a number of those who have been such an important part of creating and sus-
national publications and reports. She has lectured in several short taining these publications he recognized.
courses, some of them in the framework of European Community
Programs. She has participated in 19 projects and contracts, One of the most rewarding aspects of editing the Magazine is
financed by international, European, and national institutions and that I get to work with a large number of very wonderful people.
companies. She has been a member of the Technical Program This includes the many AP-S members who contribute what we
Committee of several international symposiums, and has acted as publish. Without them, we have no Magazine. Thank you! It also
reviewer Cor international scientific journals. She has assisted the includes our Staff, a dedicated group of professionals who have
Comision Interministerial de Ciencia y Tecnologia (National given selflessly of their time and talents. Without them there also
Board of Research) in the evaluation of projects. She has also would be no Magazine, and they must share this award I accepted
served in several evaluation panels of the Commission of the it, in part, on their behalf. Thanks, too, to all of you, our readers.
European Communities. She bas acted in the past as topical Editor Your support is what has made the Magazine possible. Finally, I
for the disk of references of the Review ofRadio Science. She is a wouldn’t be able to edit the Magazine without the constant love,
member of the editorial board of two scientific journals. She has support, and giving up of time I would otherwise spend with them
served as Vice Chair and Chair of the IEEE MTTIAP-S Spanish by my wife and daughter. My thanks to them - and to all of you,
Joint Chapter, and is currently serving as Chair of the Spain Sec- again.
tion of the IEEE. She has received two individual research awards
from national institutions. @ One of the hest things about this award is that I get to keep
on editing the Magazine. Thank you for the award - and for the
wonderful support of our AdComs and Officers and members,
through the years and now, in letting me continue to do this job!

82 IEEEAntennas and Propagalion Magazine, Vol. 45, No. 3, June 2003