3 Radio Propagation Modeling

1 of 24

http://morse.colorado.edu/~tlen5510/text/classwebch3.html

Copyright 2016 Thomas Schwengler

Chapter 3
Radio Propagation Modeling
This chapter introduces propagation characteristics and models for cellular systems. It summarizes important
notions, and expands a bit on some aspects of fixed-versus-mobile and indoor-versus-outdoor propagation
modeling. 1
Before studying details of propagation, a few notation conventions are necessary. Although no further details
are derived here, the reader is assumed to be familiar with the following general concepts of electromagnetic
field and wave theory.
Wireless communications signals of interest are electromagnetic waves, and may be derived in free
space from the electric field

The power density of the electromagnetic wave may be written in the form of the Poynting vector:
=

. The power density is the modulus of the Poynting vector Pd = |

|.

In free space the electric and magnetic field strength are related by | | = 0| |, and the power density
of the electromagnetic wave is proportional to the modulus squared of the electric field: Pd = |E|20,
where 0 377 is the impedance of the vacuum (and by approximation of air). 2
The electric field may be identified with the transmitted signal S(t) = s(t)exp(j2ft), where s(t) is the
(real) user encoded information to transmit, and f is the carrier frequency. S(t) is a complex function
which real part Re{S(t)} = s(t)cos(2ft) is the physical quantity of interest; although the complex
function S(t) is usually used for simpler mathematical treatment, one should remember that its real part
is the meaningful quantity.
Similarly, we identify the received signal with the received electric field; we denote the received signal:
R(t) = r(t)exp(j2ft).
Given the above, received power densities are given by the expression Pd(t) = |R(t)|20.
Actual received power Pr also depends on the effective area of the receiving antenna Pr(t) = AePd(t) =
Ae|R(t)|20 (see further details in 3.2).
Some details of E-field propagation will be studied later with ray tracing; but most of the remainder of the
section deals with very simple expressions of power levels for paths loss modeling.

3.1 Propagation Characteristics

Between transmitter and receiver, the wireless channel is modeled by several key parameters. These
parameters vary significantly with the environment, rural versus urban, or flat versus mountainous. Different
kinds of fading occur; they are often separated in three types [1] [3]:
Distance Dependence
of path loss (measured in dB) is approximated by L(d) = L0 +10nlog(dd0), where n is the path loss
exponent, which varies with terrain and environment, and L0 and d0 are parameters described further in

31-Oct-16 9:31 PM

3 Radio Propagation Modeling

2 of 24

http://morse.colorado.edu/~tlen5510/text/classwebch3.html

3.3 and 3.4.

Large-scale Shadowing
causes variations over larger areas, and is caused by terrain, building, and foliage obstructions; its
impact on link budgets is detailed further in 4.1.
The large-scale fading due to various obstacles is commonly accepted to follow a log-normal
distribution ([17], [18], [19] ch. 7). This means that its attenuation x measured in dB is normally
distributed N(m,), with mean m and standard deviation . The probability density function of x is
given by the usual Gaussian formula:
(3.1)
Small-scale fading
causes great variation within a half wavelength. It is caused by multipath and moving scatterers.
Resulting fades are usually approximated by Rayleigh, Ricean, or similar fading statistics
measurements also show good fit to Nakagami-m and Weibull distributions.
Radio systems rely on diversity, equalizing, channel coding, and interleaving schemes to mitigate its
impact.
Different spectrum bands have very different propagation characteristics and require different prediction
models. Some propagation models are well suited for computer simulation in presence of detailed terrain and
building data; others aim at providing simpler general path loss estimates [20].

3.2 Free-Space Propagation

The simplest approach is to estimate the power ratio between transmitter and receiver as a function of the
separation distance d, that ratio is referred to as path loss. A physical argument of conservation of energy
leads to the Friis power transmission formula in free space. A transmitted power source Pt radiates
spherically, with an antenna gain Gt; the portion of that power impinging an effective area Ae at a distance d is
Pr = PtGtAe(4d2). The effective area of an antenna is related to antenna gain by Ae2 = Gt4, which is used
for the receiving antenna, and thus yields:
(3.2)
(Pt and Pr are the transmitted and received power, Gt and Gr are the transmitter and receiver antenna gain, is
the wavelength of the signal, and d is the separation distance).

31-Oct-16 9:31 PM

3 Radio Propagation Modeling

3 of 24

http://morse.colorado.edu/~tlen5510/text/classwebch3.html

Figure 3.1: Spherical free-space propagation.

This equation shows a free-space dependence in 1d2, and is sometimes expressed in decibels (dB): L(dB)
= 10 log
.
In many cases, antenna gains are considered separately, and one choses to focus on the path loss between
the two antennas. The path loss reflects how much power is dissipated between transceiver and receiver
antennas (without counting any antenna gain). Of course the path loss variation with distance is d2, or 20
log(d) in dB, which is characteristic of a free space model. The exponent (here n = 2) is called the path loss
exponent, and may vary in other models. Path loss is often expressed as a function of frequency (f), distance
(d), and a scaling constant that contains all other factors of the formula. For instance:
(3.3)
where f0 = 1 MHz, and d0 = 1 km. These reference values are arbitrary and chosen to be convenient to use
values in MHz and km in the formula. Note that the constant 32.44 changes if the reference frequency f0 and
the reference distance d0 are chosen to be different. Many presentations will refer to the formula as L(dB) =
32.44 + 20 log(f) + 20 log(d), adding a statement such as f is in MHz and d is in km. That shorter
notation is equivalent to equation (3.3), but should be treated carefully: with the loss of f0 and d0 in the
equation, mistakes can be introduced by further manipulating the expression.

3.3 Ray Tracing

Ray tracing is a method that uses a geometric approach, and examines what paths the wireless radio signal
takes from transmitter to receiver as if each path was a ray of light (possibly reflecting off surfaces).
Ray-tracing predictions are good when detailed information of the area is available. But the predicted results
may not be applicable to other locations, thus making these models site-specific.
Nevertheless, fairly general models may be devised from ray tracing concepts. The well-known two-ray
model uses the fact that for most wireless propagation cases, two paths exist from transmitter to receiver: a
direct path and a bounce off the ground. That model alone shows some important variations of the received
signal with distance. [1] [3]
Ray tracing models are important for a good understanding of radio propagation; they are extensively
used in software propagation prediction packages, which justifies a closer look at them in this section.
3.3.1 Ray Tracing Notations
Rays are an optical approximation for the propagation of the electromagnetic wave; as seen earlier, in free
space it is convenient to focus on the propagation of the electric field. The (complex) transmitted electric
field is noted S(t), the received field is R(t), therefore the link budget between transmitter and receiver can be
written PrPt = |p0|2 in terms of a parameter p0(t) = R(t)S(t).
In the next few subsections, we will consider how the electric field propagates, and we will compute
various expressions for p0(t) in various conditions; each time the final step will be to take its modulus squared
in order to derive the path loss.

31-Oct-16 9:31 PM

3 Radio Propagation Modeling

4 of 24

http://morse.colorado.edu/~tlen5510/text/classwebch3.html

3.3.2 Two-Ray Model

With these few notation conventions we can derive a simple but useful model with two rays.

Figure 3.2: 2-ray model geometry.

Copyright 2016 Thomas Schwengler
Figure 3.2 shows a fixed tower (e.g. in a cellular system) at a height hb, and a client device at a distance
d0, and at a height hm (usually lower). The figure shows a direct ray and an indirect ray bouncing off the
ground, assumed to be a perfect plane (this assumption is referred to as the flat-earth model). 3
It is easy to see from this figure that the two path lengths are:
(3.4)
(3.5)
The received signal at a distance d0 is therefore:
(3.6)
where = cf is the wavelength, is the time difference between the two paths, and is the ground reflection
coefficient. For now let us simply assume perfect reflection and use = -1; will elaborate further on in
3.3.3. 4
Another important assumption must be made here to simplify the model: we will assume that is small
compared to the symbol length of the useful information, that is s(t) = s(t - ). For a bounce off the ground,
that assumption is fairly safe, but in general we will have to recall that such an assumption means that we
assume that the delay spread (spread of values of ) is small compared to transmitted symbol rates.
So finally we obtain:
(3.7)
the last factor of which can be easily plotted and examined for variations of the received signal strength.

31-Oct-16 9:31 PM

3 Radio Propagation Modeling

5 of 24

http://morse.colorado.edu/~tlen5510/text/classwebch3.html

Figure 3.3: Simple propagation models: free-space one-slope direct line of sight, and two-ray with
direct ray and ground reflected ray. In some places signal add constructively, in others phase
differences cause deep fades
Figure 3.3 represents the path loss attenuation PrPt = |p0|2 (in dB) as a function of logarithm of distance; it
uses hb = 8 m, hm = 2 m, f = 2.4 GHz, Gt = Gr = 0dBi, and as given later in 3.3.3. The direct path (using the
first term only of (3.7)) leads to the simple on-slope free-space model; the complete expression leads to the
two-ray model, which shows interesting characteristics:
The presence of a second ray causes great variations: signals can add up or nearly cancel each other,
causing deep fades over small distances.
In close proximity, the overall envelope of power decay varies in 1d2.
After a certain cutoff distance (approximately 4hbhm) the model approaches power decay in 1d4.
3.3.3 Reflection and Refraction
Before moving ahead, we need to take a closer look at reflection coefficients used for indirect rays. The
details of this analysis come from boundary conditions for electromagnetic waves traveling between two
media. In general these boundary conditions vary with the polarization of the wave and the media
permittivities. For a wave impinging on the ground with an angle of incidence , the ground reflection
coefficient depends on the polarization and is given by equation (3.8):
(3.8)
is the ground reflection coefficient; Z is the characteristic impedance of the media, as obtained by
transmission line theory [73] [19]; is the ray angle of incidence (as shown on fig. 3.4); r is the complex
relative permittivity of the medium: r = r -j
r -j60 where r is the lossless relative permittivity and
is the conductivity (in -1m-1).
31-Oct-16 9:31 PM

3 Radio Propagation Modeling

6 of 24

http://morse.colorado.edu/~tlen5510/text/classwebch3.html

The wave typically has many polarized component to it; even when a transmitter uses vertically polarized
antennas, different scatterers in the path may depolarize the wave. Nevertheless, the majority of cellular
systems use vertical polarization, which is shown empirically to propagate slightly better in most practical
cellular environments. In these cases, the electrical field is near vertical, and the reflection (and refraction) on
a surface is shown on figure 3.4.

Figure 3.4: Vertical polarization ground reflection coefficient.

Figure 3.5: Horizontal polarization ground reflection coefficient.

Similarly, rays bouncing off walls have a reflection coefficient (of course the vertically polarized waves
now needs to be considered as impinging the surface with electric field near the surface plane as a
horizontally polarized wave does on the ground).
Values for complex permittivities may be used approximately from table 3.1 (from [19] p. 55, and a few
other references); www.fcc.gov/mb/audio/m3/ gives ground conductivity maps for the US.
Table 3.1: Relative permittivities for various materials.
Material

r
(Fm-1)
Vacuum
1
Air
1.00054
Glass
3.8-8
Wood
1.5-2.1
Drywall
2.8
Polystyrene 2.4-2.7
Dry brick
4
Concrete
4.5
Limestone 7.5
Marble
11.6
Fresh water 80.2
Sea water
80.2

(-1m-1)

Comments
By definition
Usually approximated to 1.0
Varies with glass types

Varies 4-6
0.03
0.01
5

31-Oct-16 9:31 PM

3 Radio Propagation Modeling

7 of 24

Snow
Ice
Ground

http://morse.colorado.edu/~tlen5510/text/classwebch3.html

1.3
3.2
15 (7-30) 0.005 (0.001-0.03) Varies with type and humidity

Further refinements may be thought of regarding the thickness of walls: the ground may easily be
considered as an infinite semi-plane, but walls are usually thin enough to make that approximation
questionable. 5
3.3.4 Multiple Rays

Figure 3.6: Six- and ten-ray model geometry.

The above 2-ray approach can easily be extended to add as many rays as required [3]. We may add rays
bouncing off each side of a street in an urban corridor, leading to a 6-ray model (with rays R0,R1,R2 each
having a direct and a ground bouncing ray). Adding four more rays (bouncing on both sides: R3,R4 dashed
line in figure 3.6) lead to a 10-ray model.
The direct two rays were computed earlier, the additional rays may easily be obtained from geometry of
figure 3.6. Let us assume for instance a street corridor of width ws, with a transmitter on a light pole wt from
the walls. For simplicity, let us move the receiving point down the street at constant wt from the wall, in that
case distance Tx to Rx represented by R1 is d1 =

, so, much like equation (3.7), we have:

(3.9)

where l1 =
, and l1 =
. 1 is the refection
coefficient off the nearest wall, and is computed from (3.8), but with angles with respect to the walls.
Additional rays (R2 and more) can be calculated in expressions resembling (3.9), and added to others in
order to produce a multiple-ray model.

31-Oct-16 9:31 PM

3 Radio Propagation Modeling

8 of 24

http://morse.colorado.edu/~tlen5510/text/classwebch3.html

Figure 3.7: Ray tracing plots of received signal power indicator 20log | i=0i=Npi| as a function of
log d0 for N {0,2,3,5}. A typical suburban case is taken with street width of 20 feet, and
average distance from street to home of wt = 10 feet (so ws = 40 feet).
Figure 3.7 shows the increased fading statistic when more rays are taken into account. The figure simply
represents the received signal power indicator 20log | i=0i=Npi| as a function of log d0 for N {0,2,3,5}. For
that plot a typical suburban case is taken with street width of 20 feet, and average distance from street to
home of wt = 10 feet (so ws = 40 feet).
3.3.5 Residential Model
As previously mentioned, that approach is interesting for urban and suburban corridors. We further assume
that property lengths and home lengths along the street are approximately identical (say 100 feet and 80 feet
respectively). In that case, some rays escape the corridor and never reach the receiver as illustrated in figure
3.8, R3 rays escape the urban canyon and never reach the receiver. Taking into account these gaps show a
slightly modified model (figure 3.9). Alternatively, instead of examining where rays may escape the corridor,
a simplified model may be used that takes into account a power loss proportional to the gaps [40].

31-Oct-16 9:31 PM

3 Radio Propagation Modeling

9 of 24

http://morse.colorado.edu/~tlen5510/text/classwebch3.html

Figure 3.8: Ray tracing geometry for a street corridor: some rays escape the corridor through gaps
between homes.

Figure 3.9: Ray tracing power levels down a street, with gaps between homes.
3.3.6 Indoor Penetration
Most cellular towers are placed outdoors, while eighty percent of phone calls are placed indoors. Therefore
the problem of how much of the signal strength propagating down the street might be available indoor is of
great interest. Grazing angles of incidence are somewhat concerning in urban and suburban corridors. Figure
3.10 shows a typical case where wireless systems (base stations or access points) may be placed on opposite
side of the street to provide coverage to residences.

Figure 3.10: Ray tracing impinging on home walls.

An indoor system may detect the optimal signal between outside sources. Received power levels on the
home front wall (inside and outside) is compared on figure 3.11.

31-Oct-16 9:31 PM

3 Radio Propagation Modeling

10 of 24

http://morse.colorado.edu/~tlen5510/text/classwebch3.html

Figure 3.11: Received power levels from four rays outside and inside home front wall.
In our previous urban corridor model, the angles of incidences should be restricted to rays illuminating
walls (as in figure 3.12). 6 7

Figure 3.12: Angles of incidence illuminating homes in an urban corridor.

(3.10)

(3.11)

(3.12)

(3.13)
Angles of incidences between these values should be used to calculate penetration losses such as: Copyright
2016 Thomas Schwengler

31-Oct-16 9:31 PM

3 Radio Propagation Modeling

11 of 24

http://morse.colorado.edu/~tlen5510/text/classwebch3.html

(3.14)
For instance in a Lakewood neighborhood a light pole is placed every three homes on opposite street sides
(i.e. a pole every 6 homes); we get the values in table 3.2 for the furthest home (n = 3, 100-feet properties,
80-feet long homes, 40-feet wide streets, and wt = 10 feet). And the value Lge 10dB is typical for residential
areas. (More details in 3.7).
Table 3.2: Angles of incidence in a suburban area in Lakewood, CO.
Pole position 3 (deg) 4 (deg) Lge from (3.14)
Across street 19.4
14.6
0.5 Lge
Same side
6.7
5.0
8.0 Lge
3.3.7 Indoor Propagation
Propagation within a building is yet another problem of interest, and is different when signal comes from the
outside, or has a source within the building. Indoor propagation varies greatly with the type of buildings, and
the position of access points within the building how far from wall, how high compare to obstructions and
furniture. 3D ray models are sometimes used to better predict these situations. Other generic models are
detailed in 3.4.5.

3.4 Classic Empirical Models

Empirical models are simpler and less site specific, and provide a first order modeling for a wide range of
locations. A handful of empirical models are widely accepted for cellular communications; these models
usually simply consist of computing a path loss exponent n from some linear regression argument on a set of
field data, and deriving a model like:
(3.15)
(where the intercept L0 is the path loss at an arbitrary reference distance d0). These models are referred to as
empirical one-slope models; their applications and domains of validity are well described and analyzed for
instance in [3] ch. 2, [1] ch. 4, [6], [19] ch. 6-7. They provide a first estimate used by service provider in
wireless systems design phase.8
A couple of important points should be kept in mind about most propagation models. The first is that
large amounts of empirical data are collected usually at cellular or PCS frequencies (800 MHz or 1900 MHz),
and extensions to other frequencies are derived as discussed in 3.4.6. The second is that these data points are
collected while driving and may not accurately reflect fixed wireless links, which is discussed in more details
in 2.10.
3.4.1 COST 231-Hata Model
A one-slope empirical model was derived by Okumura [21] from extensive measurements in urban and
suburban areas. It was later put into equations by Hata [22]. This Okumura-Hata model, valid from 150 MHz

31-Oct-16 9:31 PM

3 Radio Propagation Modeling

12 of 24

http://morse.colorado.edu/~tlen5510/text/classwebch3.html

to 1.5 GHz, was later extended to PCS frequencies, 1.5 GHz to 2 GHz, by the COST project ([23], [24]
ch. 4), and is referred to as the COST 231-Hata model; it is still widely used by cellular operators. The model
provides good path loss estimates for large urban cells (1 to 20 km), and a wide range of parameters like
frequency, base station height (30 to 200 m), and environment (rural, suburban or dense urban).
(3.16)
with the following values:
Table 3.3: Values for COST 231 Hata and Modified Hata model.
Frequency
(MHz)
150-1500
1500-2000

c0
(dB)
69.55
46.3

cf
(dB)
26.16
33.9

b(hB)
(dB)
13.82log(hB1m)
13.82log(hB1m)

The parameter a(hM) is strongly impacted by surrounding buildings, and is sometimes refined according to
city sizes:
Table 3.4: Values of a(hM) for COST 231-Hata model according to city size.
Frequency City size
a(hM)
(MHz)
(dB)
150-2000 Small-medium (1.1log(
150-300
300-2000

Large
Large

) - 0.7)

- 1.56log(

) + 0.8

8.29(log(1.54hM1m))2 - 1.1
3.2(log(11.75hM1m))2 - 4.97

And an additional parameter CM is added to take into account city size, and can be summarized for both
models as:
Table 3.5: Values of CM for COST 231 Hata model according to city size.
Frequency City size
(MHz)
150-1500 Urban
150-1500 Suburban

CM
(dB)
0
-2(log(

150-1500

-4.78(log(

Open rural

))2 - 5.4
))2 + 18.33log(

) - 40.94

1500-2000 Medium city and suburban 0

1500-2000 Metropolitan center
3
Empirical values of the model are limited to distances and tower heights that were used to derive the
model; consequently the model is usually restricted to:

31-Oct-16 9:31 PM

3 Radio Propagation Modeling

13 of 24

http://morse.colorado.edu/~tlen5510/text/classwebch3.html

: Base station antenna height: 30 to 200 m

: Mobile height: 1 to 10 m
: Cell range: 1 to 20 km
3.4.2 COST 231-Walfish-Ikegami Model
Another popular model is the Walfisch-Ikegami-Bertoni model [28] [29], also revised the COST project
([23], [24] ch. 4), into a COST 231-Walfisch-Ikegami model. It is based on considerations of reflection and
scattering above and between buildings in urban environments. It considers both line of sight (LOS) and non
line of sight (NLOS) situations. It is designed for 800 MHz to 2 GHz, base station heights of 4 to 50 m, and
cell sizes up to 5 km, and is especially convenient for predictions in urban corridors.
The case of line of sight is approximated by a model using free-space approximation up to 20 m and the
following beyond:
(3.17)
The model for non line of sight takes into account various scattering and diffraction properties of the
surrounding buildings:
(3.18)
where L0 represents free space loss, Lrts is a correction factor representing diffraction and scatter from rooftop
to street, and Lmsd represents multiscreen diffraction due to urban rows of buildings. These terms vary with
street width, building height and separation, angle of incidence, and are detailed in table 3.6.
Table 3.6: Values for COST 231-Walfish-Ikegami model.
Parameter
L0
Lrts
w
hM

Value (dB)
32.4 + 20log(d1km) + 20log(f1MHz)
-16.9 - 10log(w1m) + 10log(f1MHz) + 20log(hM1m) + LOri
Average street width
hRoof - hM

LOri

Lmsd
b
hB

Road orientation with respect to direct radio path (see figure(3.13))

Lbsh + ka + kd log(d1km) + kf log(f1MHz) - 9log(b1m)
Average building separation
hB - hRoof

Lbsh
ka

31-Oct-16 9:31 PM

3 Radio Propagation Modeling

14 of 24

http://morse.colorado.edu/~tlen5510/text/classwebch3.html

kd
kf

Figure 3.13: Definition of street orientation angle for use in COST-231 Walfish-Ikegami model:
in the best case ( = 0) the direction of propagation follows the street; in the worst case ( =
90) the main radio wave is perpendicular to the street.
The model is usually restricted to:
: Frequency: 800 to 2000 MHz
: Base station antenna height: 4 to 50 m
: Mobile height: 1 to 3 m
: Cell range: 0.2 to 5 km
3.4.3 Erceg Model
More recently Erceg et al. [30] proposed a model derived from a vast amount of data at 1.9 GHz, which
makes it a preferred model for PCS and higher frequencies. The model was in particular adopted in the
802.16 study group [31] and is popular with WiMAX suppliers for 2.5 GHz products, and even 3.5 GHz
fixed WiMAX.
(3.19)
where free space approximation is used for d < d0. Values for L0, , and s are defined in tables 3.7 and 3.8:
Table 3.7: Values for Erceg model.
Parameter
L0
d0

x,y,z

Value (dB)
20log(4d0) as in free space
100 m
(a - bhB + chB) + x
y
+ z
Gaussian random variables N(0,1)

31-Oct-16 9:31 PM

3 Radio Propagation Modeling

15 of 24

http://morse.colorado.edu/~tlen5510/text/classwebch3.html

Table 3.8: Values for Erceg model parameters in various terrain categories.
Parameter Terrain Category
A
(Hilly / moderate to heavy
tree density)
a
4.6
b(m-1)
0.0075
c (m)
12.6

0.57

10.6

2.3

B
(Hilly / light tree density or flat /
moderate to heavy tree density)
4.0
0.0065
17.1
0.75
9.6
3.0

C
(Flat / light tree
density)
3.6
0.0050
20.0
0.59
8.2
1.6

The model is usually restricted to:

: Frequency: 800 to 3700 MHz
: Base station antenna height: 10 to 80 m
: Mobile height: around 2 m
: Cell range: 0.1 to 8 km
The model is particularly interesting as it provides more than a median estimate for path loss: it also gives
a measure of its variation about that median value in terms of three zero-mean Gaussian random variables of
variance 1 (x,y, and z = N(0,1)).
3.4.4 Multiple Slope Models
Further refinements to these models in which multiple path loss exponents (n1,n2) are used at different ranges
provide some improvements, especially in heavy multipath indoors environments. For outdoor propagation,
two slopes are sometimes used: one near free-space for close points, and another empirically determined. In
fact weve seen that our 2-ray model could be approximated by a 2-slope model: n1 = 2 and n2 = 4 for
distances greater than 4hthr.
It seems however that variations from site to site generally are such that these multiple slope
improvements are fairly small, and simple one-slope models are generaly a good enough first approximation.
More detailed site specific models are required for better results; but they require additional efforts and site
specific terrain or building data.
3.4.5 In-building
Indoor propagation often has to be estimated by site-specific models with features specific to a particular
building: construction material, wall thickness, floor and ceiling material, all have a strong impact on wave
guiding within the building. Some models simply approximate the number of walls and floors, with an
average loss for each. See in particular the COST 231 approach in 3.7.
A similar model for indoor environment is the Motley-Keenan model ([2],7.2), which estimates path
loss between transmitter and receiver by a free space component (L0) and additive loss in terms of wall
attenuation factors (Fwall) and floor attenuation factors (Ffloor).
31-Oct-16 9:31 PM

3 Radio Propagation Modeling

16 of 24

http://morse.colorado.edu/~tlen5510/text/classwebch3.html

(3.20)
Wall attenuation factors vary greatly, typically 10 to 20dB (see table 3.12 in 3.7); and floor attenuation
factors are reported to vary between 10 and 40dB depending on buildings. [1]
This model is very site specific, yet sometimes imprecise as it does not take into account proximity of
windows external walls, etc; but it can be useful as a guideline to estimate signal strength to different rooms,
suites, and floors in buildings.
3.4.6 Frequency Variations
Frequency of operations impacts propagation and path loss estimates. As many models are built on cellular or
PCS data measurements, one must be careful about extending them to other frequency ranges.
As seen in equation (3.3) in 3.2, the impact of frequency on free-space propagation is 20logf. Some
empirical measurements confirm the trend [35], and the extension is used for instance in the COST-231
Walfish-Ikegami model.
Empirical evidence also shows however that frequency extensions are obtained by adding a frequency
dependence in f2.6 (or a 26log f term in dB) as suggested by [38], and used for instance in the Okumura-Hata
model [22] and the 802.16 contribution [31].
Finally other important aspects have a significant impact as frequency changes. Spatial diversity gain
typically improves with frequency since spatial separation increases when related to wavelength ([39] shows
a 2dB diversity gain from cellular 850 MHz to PCS 1.9 GHz). Doppler spread and impact on symbol duration
should also be studied separately and may have a significant impact on a change of frequency [41]. Impact on
in-building penetration is examined further in 3.7.
3.4.7 Foliage
Foliage attenuates radio waves and may cause additional variations in high wind conditions [42]. Propagation
losses vary for instance with the position of transmitter with respect to the tree canopy; they also vary with
the types and density of foliage, and with seasons. [43][44][45][46]
We will report in a later chapter on the impact of foliage for fixed wireless links at 3.5 GHz, in a suburban
area as foliage grows from the winter months into the spring (see figure 10.9).
Studies have been published at different frequencies, and impact of foliage is given in a number of ways:
some identify empirical attenuation statistic with Raleigh, Ricean, or Gaussian variables, others derive excess
path loss, or attenuation per meter of vegetation.
As a rule of thumb, at our frequencies of interest (2-6 GHz) single tree causes approximately 10-12 dB
attenuation, and typical estimates are 1-2 dB/m attenuation. Deciduous trees in winter cause less attenuation:
0.7-0.9 dB/m. See table 3.9 and figure 3.14.
Practically, the height of the antenna with respect to tree height (or canopy height) strongly impact
propagation characteristics; different path loss estimates and path loss exponents may be empirically derived
depending on relative height with the canopy.[47]

31-Oct-16 9:31 PM

3 Radio Propagation Modeling

17 of 24

http://morse.colorado.edu/~tlen5510/text/classwebch3.html

Table 3.9: Vegetation loss caused by tree Foliage: single-tree model loss in dB, and estimates for dB/m
loss) summary of values for various frequencies reported.
Source
Benzair [43]

Dalley [44]
Wang [46]

Torrico [47]

Frequency
(GHz)
2.0
4.0
2.0
4.0
3.5
5.8
1.0
2.0
4.0
1.0
2.0

Approximation

single tree
(dB)
20.0
27.5
9.5
10.7
11.2
12.0
10.0
14.0
18.0
12.01+7.46 logfGHz

per meter loss

(dB/m)
1.05
1.40
0.70
0.85
1.9
2.0
0.7
1.0
0.54+1.40 logfGHz

Comments
Summer
Winter
With leaves
Single tree

With leaves

Figure 3.14: Tree foliage attenuation as a function of frequency.

3.5 Further Modeling Work

The above models are in a sense simplistic as they focus on path loss as a function of distance. Although
these models work well in large cellular coverage prediction, they are often deemed insufficient for smaller
cells such as wireless LANs, especially where multipath is dominant, as in a heavy urban environment or
indoor environment.
An interesting and important activity around propagation modeling is the COST project (COperation
europnne dans le domaine de la recherche Scientifique et Technique), a European Union Forum for
cooperative scientific research that has been useful in focusing efforts and publishing valuable summary
reports for wireless communications needs.
COST 207
Digital Land Mobile Radio Communications, March 1984 - September 1988, developed channel

31-Oct-16 9:31 PM

3 Radio Propagation Modeling

18 of 24

http://morse.colorado.edu/~tlen5510/text/classwebch3.html

model used for GSM,

COST 231
Evolution of Land Mobile Radio (Including Personal) Communications, April 1989 - April 1996,
contributed to the deployment of GSM1800, DECT, HIPERLAN 1 and UMTS, and defined
propagation models for IMT-2000 frequency bands [24]
COST 259
Wireless Flexible Personalised Communications, December 1996 - April 2000, contributed to
wireless LAN modeling, and 3GPP channel model [25]
COST 273
Towards Mobile Broadband Multimedia Communications, May 2001 - June 2005, which contributed
to standardisation efforts in 3GPP, UMTS networks, provided channel models for MIMO systems [26]
The COST2100 effort has now started [27] to continue the COST 273 work, and is an important effort around
the current MIMO advances in the wireless industry.
Finally, new interesting activities of research and modeling are taking place at higher frequencies, in
millimeter-wave for mobile use with 5G [87][88].

3.6 Dispersive Models

Modern radio systems now make extensive use of multiple paths between transmitter and receiver, even
deploying multiple antenna systems (such as MIMO). These systems require more than path loss estimates,
as path loss is sensibly the same between transmitting and receiving antenna systems. MIMO channel models
are therefore much more complex; several approaches have been used, such as different groups of delayed
paths. Of particular interest are the 802.16 model [31], the 802.11n and ac models [32] for various indoor
models (from the 802.11n task group on channel modeling), or [36] for mobile cellular models.
IEEE 802.16e and SUI
The work of 802.16 and Wimax provided the first widespread models with interesting multipath
considerations and convenient matlab simulations [31]. The work starts with six typical environments
modeled by six Stanford University Interim (SUI) channel models. These models are described in
terms of terrain types (and propagation from 3.4.3), amount of Doppler, delay spread, and fading
statistics.
One way of modeling transmission delay (beyond a simple delay spread value) is to consider a series of
successive impluses, each delayed and attenuated. This is referred to as the tapped delay line model.
The SUI models define 6 different types of environments, with different tap delays, Doppler effect, and
fading statistics: in that manner the model represent different scenarios: pedestrian/vehicular,
urban/suburban/rural, indoor/outdoor, etc.
IEEE 802.11n
The IEEE task group TGn produced a series of models [32] for 802.11n (and ac) for LAN applications
at 2.4GHz and 5GHz. Consequently these models focus on pedestrian mobility; the group presents six
different models (named A to F) aim at describing typical LAN environments, with generally much
more multipath than many outdoor cellular environments.
Table 3.10: TGn channel models A to F are used to model MIMO systems in different
environment, with different RMS delay spreads ( in nanoseconds see table 4.5 in 4.3).

31-Oct-16 9:31 PM

3 Radio Propagation Modeling

19 of 24

Model
A
B
C
D
E
F

http://morse.colorado.edu/~tlen5510/text/classwebch3.html

(ns)
0
15
30
50
100
150

Environment
Direct
Residential
Res. or small office
Typical office
Large office
Large space

Example
Cabled
In room or room-to-room
Conference rooms, classrooms
Cubicles in open office space
Large office space, multi-floor
Indoor large hangars, outdoor campus / urban

Table 3.11: TGn channel models A to F use two slopes n1 = 2 near transmitter, and different
values of n2 beyond a critical distance d0. They also estimate different log-normal shadowing
standard deviations 1 = 3dB near transmitter, and higher values 2 beyond d0.
Model
A
B
C
D
E
F

d0(m)
5
5
5
10
20
30

n1
2
2
2
2
2
2

n2
3.5
3.5
3.5
3.5
3.5
3.5

1
3
3
3
3
3
3

2
4
4
5
5
6
6

3GPP SCM
The 3GPP spatial channel models (SCM) [36] focus on 3G and 4G applications such as UMTS and
LTE, in 5MHz channels around 2GHz. There are again different models, based on environment, speed
etc, typically modeling N delayed multipaths, each comprised of M subpaths (in typical urban and
suburban environments, the model uses N = 6,M = 20). Different parameters are also given for different
environments (suburban macrocell, urban macrocell, and urban microcell): pathloss (LOS and NLOS),
antenna beamwidth, delay statistics, log-normal shadowing, angles of arrival distribution, etc.
Important work on correlation between these multiple path is also presented, as it is crucial to
estimating the MIMO rank important for system capacity (see 9.1.3). The 3GPP spatial channel
models (SCM) report [36] is a wonderful source of many other parameters and typical values very
useful for any propagation aspects of propagation for mobile communication systems.

3.7 In-Building Penetration

Sending RF signal into buildings comes at an additional cost, which can be quantified by an additional
building penetration loss in the link budget. Unfortunately indoor penetration measurements are difficult to
perform, and difficult to compare from one experiment to another. The difficulty arises mostly from the fact
that indoor and outdoor environments are so different that the method of data collection may cause large
variations between the two environments; the following parameters have an influence: antenna beamwidth,
angle of incidence, outside multipath, indoor multipath, distance from the walls, etc.
Measurement campaigns show that the distribution of building penetration loss is close to log-normal
[17], a Gaussian function is a good approximation of the cumulative distribution function (CDF) of indoor
measurements. The mean i and standard deviation i of indoor penetration loss vary with frequency, types of
homes, and environment around the homes. Variations also depend on the location within the building (near
an outside wall, a window, or further inside). Finally the angle of incidence with the outside wall also has a
31-Oct-16 9:31 PM

3 Radio Propagation Modeling

20 of 24

http://morse.colorado.edu/~tlen5510/text/classwebch3.html

significant impact.
With that in mind, we can consider that wireless systems with in-building penetration have a shadowing
statistic with a log-normal random variate which combines two independent log-normal variates: the outdoor
shadowing (detailed further in 4.1 with standard deviation o) and the in-building loss (with standard
deviation i). 9 And the aggregate random variate is also log-normal distributed, and has a standard deviation
=

3.7.1 In-Building Models

The COST project proposes models for indoor penetration ([24] 4.6) with variations of angle of incidence.
The COST 231 indoor model simply uses a line-of-sight path loss with an indoor component:
(3.21)
where S is the outdoor path, d is the indoor path, and
(3.22)
where Le is the normal incidence first wall penetration; the next term represents the added loss due to angle of
incidence and is sometimes measured over an average of empirical values of incidence, in which case it
may be noted Lge = Lge(1 - sin)2; and the last term max(1,2) aims at estimating loss within the building,
whether going through walls or in a corridor.
Since angles of incidence are not always known the estimate Lge = Lge(1-sin)2 is sometimes more
convenient. As a rough estimate for angles of incidence between -2 and 2 lead to the following:
(3.23)
Empirical values of Lge are reported to be 5.7 - 6.4 [51] for residential areas, therefore we may use Lge
10dB. For urban environments, COST-231 reports Lge 20dB.
As for further interior loss, the COST model distinguishes between propagation through walls and
propagation down coridors. Through ni interior walls of loss Li each: 1 = niLi. In a corridor: 2 = (d- 2)(1 sin)2, with an empirical propagation loss = 0.6dB/m.

31-Oct-16 9:31 PM

3 Radio Propagation Modeling

21 of 24

http://morse.colorado.edu/~tlen5510/text/classwebch3.html

Figure 3.15: COST-231 indoor penetration loss model.

Typical values for the model reported in [24] and [51] are summarized in table 3.12.
Table 3.12: Penetration Loss into buildings, from COST-231 model.
Material
Wood, plaster
Concrete w/ windows
Residential

Frequency
900MHz
1.8GHz
2.5GHz

Le Lge
4
7
20
6.2 10

Lge
4
6
6.1

Li
4
10
3

Figure 3.16: Penetration loss into residential buildings, cumulative density distribution for 700 MHz,
900 MHz, 1.9 GHz, and 5.8 GHz.
Measurement campaigns show that the distribution of building penetration loss is close to log-normal
[17], a Gaussian function is a good approximation of the cumulative distribution function (CDF) of indoor
measurements. The mean and standard deviation of indoor penetration loss vary with frequency, types of
homes, and environment around the homes. Variations also depend on the location within the building (near
an outside wall, a window, or further inside). Copyright 2016 Thomas Schwengler. Finally the angle of
incidence with the outside wall also has a significant impact.
With that in mind, we consider that in-building penetration is a log-normal random variate independent of
the large-scale shadowing. Therefore, the log-normal fading used for indoor propagation should be the
normal random variable N
. Both median penetration loss and modified excess margin
should be taken into account for a new indoor link budget.
This has a significant impact on the total link budget. Consequently indoor radio units need to somehow
increase their link budgets with a plurality of antennas making use of diversity schemes or MIMO.
3.7.2 Residential Homes
In most residential and suburban environments, surfaces involved are mostly made of glass, bricks, wood,
31-Oct-16 9:31 PM

3 Radio Propagation Modeling

22 of 24

http://morse.colorado.edu/~tlen5510/text/classwebch3.html

and drywall. Penetration is often dominated by paths through windows and roofs, loss are relatively low and
go up with frequency.
Precise characterization of in-building penetration is difficult, a rough approximation of an average
penetration loss i around 10 to 15 dB and a standard deviation i around 6 dB seems to be the norm in
published studies. Table 3.13 and figure 3.7.2 summarize some published results for residential homes.
Table 3.13: Penetration Loss into residential buildings: median loss (i) and standard deviation (i) from
experimental results reported at various frequencies.
Source

Frequency
(GHz)
Aguirre [48][52] 0.9
1.9
5.9
Wells [49]
0.86
1.55
2.57
Durgin [69]
5.8
Martijn [50]
1.8
Oestges [51]
2.5
Schwengler
1.9
Schwengler [72] 5.8
Average
0.9
2
5.8

i
(dB)
6.4
11.6
16.1
6.3
6.7
6.7
14.9
12.0
12.3
12.0
14.7
6.4
10.3
13.8

i
(dB)
6.8
7.0
9.0
6
6
6
5.6
4.0

6.0
5.5
6.4
6.3
6.7

Comments
7 Boulder residences

Sat. meas. into 5 homes

[69]Table 5 average
[50]Table 1
[51]Table 6 (avg. Le + Lge)
Personal measurements
[72]Table 2

Figure 3.17: In-building loss for residential buildings: measurements campaigns published for
different frequencies, in different residential areas.

31-Oct-16 9:31 PM

3 Radio Propagation Modeling

23 of 24

http://morse.colorado.edu/~tlen5510/text/classwebch3.html

Figure 3.18: In-building loss for urban office buildings and high-rises: measurements campaigns
published for different frequencies, in different urban areas.
3.7.3 Urban Environments
In dense urban areas experiments show different trends as illustrated in figure 3.7.2: some papers show
penetration loss increasing with frequency [48][52]; some claim loss are independent of frequency [61][54];
others show a decrease with frequency [58][57][56].
Furthermore the variations between buildings and types of environments nearly always exceed the
frequency variations. These environments are dominated by reflections off metal reinforced concrete and
heavily reflective glass. In case of high-rises, penetration also depends on the floor and height of neighboring
buildings or clutter.

3.8 Homework
1. At the beginning of section 3.2, we start to derive a free-space model from Friis equation. (a) Rederive
in details the Friis formula (3.2). (b) Assume in the above that Gt = Gr = 1 (=0 dBi), derive (3.3).
2. Find the paper [30] by V. Erceg & al. An Empirically Based Path Loss Model for Wireless Channels
in Suburban Environments, in IEEE Journal on Selected Areas in Communications, Vol. 17, No 7,
July 1999. This popular paper for PCS propagation modelling and design deserves some attention.
Read it and answer the following questions:
a. Summarize data collection campaign methods and size.
b. Summarize key findings.
c. A key finding is that path loss exponent variations are Gaussian, how is that proven in the paper?
3. Plot path loss prediction versus distance and log distance for a cellular system you are designing with
the following assumptions: PCS frequency (1900 MHz), base height 20 m, mobile 2 m, suburban area,
flat terrain with moderate tree density.
a. Use and compare the 3 following models: Free space, COST 231-Hata, & Erceg (use a median
path loss: i.e. x=y=z=0)
b. Using typical 140 dB maximum allowable path loss for a CDMA voice system, what is the range
(cell radius) according to these models?
4. Repeat the above problem with unlicensed frequency 5.8 GHz and a link budget of 120 dB. Compare.
(Use the same models, including COST 231-Hata and Erceg models even though the frequency
exceeds their domain of validity.)
5. Compare the received power level of free-space (n=2), and 2-ray models for a PCS signal (1900 MHz;

31-Oct-16 9:31 PM

3 Radio Propagation Modeling

24 of 24

http://morse.colorado.edu/~tlen5510/text/classwebch3.html

use hb = 8 m, hm = 2 m). Write a program (in any language of your choice) to plot a graph of power
level versus log of distance (from 10m to 10km). Submit code with comments and explanations, and a
resulting figure.
a. First assume a simple perfect reflection = -1
b. Then use the actual given for a vertical polarized wave. Is the difference significant?
c. What cell site radius would be ideal for a system design? Why?
6. Plot and compare on a same graph the propagation estimate for a radio system at 2.4 GHz and another
at 5.8 GHz (all other parameters being equal); use hb = 8 m, hm = 2 m; use a) the two-ray model from
3.3.2, b) the 6-ray model from 3.3.4. Point out the main differences.

Copyright 2016 Thomas Schwengler

31-Oct-16 9:31 PM