Copyright ©2019

Thomas
Schwengler. A
significantly
updated and
completed 2019
Edition is available.

Chapter 3
Radio
Propagation
Modeling
This chapter
introduces
propagation
characteristics and
models for cellular
systems. It
summarizes
important notions,
and expands on
aspects of fixed-
versus-mobile and
indoor-versus-
outdoor
propagation
modeling. 1

Before studying
details of
propagation we
need to define a
few notation
conventions, which
will require the
reader 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 | | = η0| |,
and the power
density of the
electromagnetic
wave is
proportional to the
modulus squared of
the electric field: Pd
= |E|2∕η0, where η0 ≈
377 Ω 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(j2πft), 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(2πft) 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. 3
Similarly, we
identify the
received signal
with the received
electric field; we
denote the received
signal: R(t) = r(t) ⋅
exp(j2πft).
Given the above,
received power
densities are given
by the expression
Pd(t) = |R(t)|2∕η0.
Actual received
power Pr also
depends on the
effective area of the
receiving antenna
Pr(t) = AePd(t) =
Ae|R(t)|2∕η0 (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, they
are often separated
in three types [1]
[3]:

Distance
Dependence
of path loss
(measured in dB) is
approximated by
L(d) = L0
+10n×log(d∕d0),
where n is the path
loss exponent,
which varies with
terrain and
environment, and
L0 and d0 are
parameters
described further
in §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,σ),
with mean m,
standard deviation
σ, and which
probability density
function 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,
and is detailed
further in §??.
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.

Copyright ©2019
Thomas
Schwengler.
Significantly
updated and
completed in 2019
Edition available
here.

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

A simple approach
to propagation
modeling 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∕(4πd2). The
effective area of an
antenna is related
to antenna gain by
Ae∕λ2 = Gr∕4π, which
is used here for the
receiving antenna
(of gain Gr), and
thus yields:

(3.2)

(Pt and Pr are the

transmitted and
received power, Gt
and Gr are the
transmitter and
receiver antenna
gain, λ is the
wavelength of the
signal, and d is the
separation
distance).

Figure 3.1: Spherical
free-space
propagation: power
transmitted over
the air radiates
spherically, with
different gain in
different directions.
A portion of that
power is received
on an effective
aperture area Ae of
the receiving
antenna.

This equation
shows a free-space
dependence in 1∕d2,
and is often
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).
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 its value varies
with 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 32.45
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) = 32.45 +
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 (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
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 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), we define the
ratio p0(t) = R(t)∕S(t).
Then the power
link budget
between
transmitter and
receiver can be
written Pr∕Pt = |p0|2.

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.

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: a
direct ray links
transmitter and
receiver, and a
second ray bounces
off the ground.

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). 4

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 λ = c∕f is the
wavelength, t0 = l0∕c
(respectively t′0 -
l′0∕c) is the time
needed for the
wave to propagate
over distance l0
(respectively l′0),
and Γ is the ground
reflection
coefficient. For now
let us simply
assume perfect
reflection and use Γ
= -1; will elaborate
further on Γ in
§3.3.3. 5

Another important
assumption must
be made here to
simplify the model:
we will assume that
propagation time
differences are
small compared to
the symbol length
of the useful
information, that is
s(t-t0) = s(t-t′0). 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 is
small compared to
transmitted symbol
rates.

So we obtain:

(3.7)

which simplifies for

distances much
greater than
heights (d0 ≈ l0 ≈ l′0):

(3.8)

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
Pr∕Pt = |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 Γ as given
later in §3.3.3. The
direct path, using
the first term only
of (3.7), leads to the
simple one-slope
free-space model;
the complete
expression leads to
the two-ray model.
The figure 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 1∕d2.
After a certain
cutoff distance
(4hbhm∕λ) the model
approaches power
decay in 1∕d4.

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
θ, the ground
reflection
coefficient depends
on the polarization
and is given by
equation (3.9):

(3.9)

Γ is the ground
reflection
coefficient; Z is the
characteristic
impedance of the
media, as obtained
by transmission
line theory [76]
[19]; θ is the ray
angle of incidence
(as shown on
fig. 3.4); ϵr is the
complex relative
permittivity of the
medium: ϵr = ϵr -j
≈ ϵr -j60σλ where ϵr
is the lossless
relative
permittivity and σ
is the conductivity
(in Ω-1m-1).

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. The sign
of Γ depends on the
convention of the
direction of
incident and
reflected electric
fields. In our
convention a
perfect conducting
plane has a
reflection Γ = +1.

Figure 3.5: Horizontal
polarization ground
reflection
coefficient. With
this convention a
perfect conducting
plane has a
reflection Γ = -1.

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 ϵr σ Comments
(Fm-1) (Ω-1m-1)
Vacuum 1 By definition
Air 1.00054 Usually approximated to 1.0
Glass 3.8-8 Varies with glass types
Wood 1.5-2.1
Drywall 2.8
Polystyrene 2.4-2.7
Dry brick 4
Concrete 4.5 Varies 4-6
Limestone 7.5 0.03
Marble 11.6
Fresh water 80.2 0.01
Sea water 80.2 5
 Snow 1.3
Ice 3.2
Ground 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. 6

3.3.4 Multiple Rays

Figure 3.6: Six-ray
model (R0,R1,R2) and
ten-ray model (R0 to
R4) geometry: each
line hides two rays:
one direct, the
other bouncing off
the ground.

The above 2-ray

approach can easily
be extended to add
more rays  [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:

(3.10)

where l1 =
, and l1′ =
. Γ1 is the refection
coefficient off the
nearest wall, and is
computed from
(3.9), but with
angles with respect
to the walls.

Additional rays (R2

and more) can be
calculated in
expressions
resembling (3.10),
and added to others
in order to produce
a multiple-ray
model.

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}for
a typical suburban
case 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 [42].
 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.

In our previous
urban corridor
model, the angles of
incidences should
be restricted to rays
illuminating walls
(as in figure 3.11). 7
8

Figure 3.11: Angles
of incidence
illuminating homes
in an urban
corridor.

(3.11)

(3.12)

(3.13)

(3.14)

Angles of
incidences between
these values should
be used to calculate
penetration losses
such as:

(3.15)

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 and
their resulting
estimated
penetration losses.

Pole position θ3 (deg) θ4 (deg) L′ge from (3.15)

Across street 19.4 14.6 0.5 Lge
Same side 6.7 5.0 8.0 Lge

3.4 Classic
Empirical
Models

Empirical models
are simple models
that provide a first
order estimate 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 a set of field
data, and deriving a
model for path loss
(in dB) like:

(3.16)

(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 defined
and reviewed later.
They generally
provide a first
estimate used by
service provider in
wireless systems’
design phase.9

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 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).

(3.17)

with the following

values:

Table 3.3: Values
for COST 231 Hata
and Modified Hata
models.

Frequency c0 cf b(hB)
(MHz) (dB) (dB) (dB)
150-1500 69.55 26.16 13.82log(hB∕1m)
1500-2000 46.3 33.9 13.82log(hB∕1m)

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( ) - 0.7)
- 1.56log( ) + 0.8
150-300 Large 8.29(log(1.54hM∕1m))2
- 1.1
300-2000 Large 3.2(log(11.75hM∕1m))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 CM

(MHz) (dB)
150-1500 Urban 0
150-1500 Suburban -2(log( ))2 - 5.4
150-1500 Open rural -4.78(log( ))2 +
18.33log( )-
40.94
1500-2000 Medium city, 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:

: 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 [30]
[31], 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:

(3.18)

The model for non

line of sight takes
into account
various scattering
and diffraction
properties of the
surrounding
buildings:

(3.19)

where L0
represents free
space loss, L

rtsisacorrectionfactorrepresentingdiffractionandscatterfromrooftoptostreet,andL˙msdrepresentsmultiscreendiffractionduetourbanrowsofbuilding

Table 3.6: Values
for COST 231-
Walfish-Ikegami
model.

Parameter Value (dB)

L0 32.4 + 20log(d∕1km) + 20log(f∕1MHz)
Lrts -16.9 - 10log(w∕1m) + 10log(f∕1MHz) + 20log(ΔhM∕1m) + LOri
w Average street width
ΔhM hRoof - hM

LOri

ϕ Road orientation with respect to direct radio path (see figure(3.12))

Lmsd Lbsh + ka + kd log(d∕1km) + kf log(f∕1MHz) - 9log(b∕1m)
b Average building separation
ΔhB hB - hRoof
Lbsh

ka

kd
 kf

Figure 3.12: Definition
of street orientation
angle ϕ for use in
COST-231 Walfish-
Ikegami model: in
the best case (ϕ = 0∘)
the direction of
propagation follows
the street; in the
worst case (ϕ = 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. [32]
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 [33] and is
popular with
WiMAX suppliers
for 2.5 GHz
products, and even
3.5 GHz fixed
WiMAX.

(3.20)

where free space

approximation is
used for d < d0.
Values for L0, γ, and
s are defined in
tables 3.7 and 3.8:
 Table 3.7: Values
for Erceg model.

Parameter Value (dB)

L0 20log(4πd0∕λ) as in free space
d0 100 m
γ (a - bhB + c∕hB) + xσγ
s yσ
σ μσ + zσσ
x,y,z Gaussian random variables N(0,1)

Table 3.8: Values
for Erceg model
parameters in
various terrain
categories.

Parameter Terrain Category

A B C
(Hilly / moderate to (Hilly / light tree (Flat / light tree
heavy tree density) density or flat / density)
moderate to heavy
tree density)
a 4.6 4.0 3.6
b(m-1) 0.0075 0.0065 0.0050
c (m) 12.6 17.1 20.0
σγ 0.57 0.75 0.59
μσ 10.6 9.6 8.2
σσ 2.3 3.0 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
we’ve seen that our
2-ray model could
be approximated
by a 2-slope model:
n1 = 2 and n2 = 4 for
distances greater
than 4hthr∕λ.

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).

(3.21)

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
[37], 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
[40], and used for
instance in the
Okumura-Hata
model [22] and the
802.16 contribution
[33].

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 ([41]
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 [43].
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
[44]. Propagation
losses and path loss
exponents vary
strongly with the
position of
transmitter with
respect to the tree
canopy; they also
vary with the types
and density of
foliage, and with
seasons. [45][46]
[47][48][49]

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; 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 frequencies
around 1 ot 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.)

Table 3.9:
Vegetation loss
caused by tree
foliage, reported
for various
frequencies: single-
tree model loss in
dB, and dB/m loss.

Source Frequency single tree loss per meter loss Comments

(GHz) (dB) (dB/m)
Benzair [45] 2.0 20.0 1.05 Summer
4.0 27.5 1.40
2.0 9.5 0.70 Winter
4.0 10.7 0.85
Dalley [46] 3.5 11.2 1.9 With leaves
5.8 12.0 2.0
Wang [48] 1.0 10.0 - Single tree
2.0 14.0 -
4.0 18.0 -
Torrico [49] 1.0 - 0.7 With leaves
2.0 - 1.0
Approximation 12.01+7.46 logfGHz 0.54+1.40 logfGHz

Another common
models for
vegetation proposes
an empirical
exponent both for
distance and
frequency; thus
approximating
vegetation loss by
L(dB) = Afαdβ, where
A,α, and β
parameters vary
with the type and
density of
vegetation. See
figure 3.13 taken
from [50].

Figure 3.13: Summary
of tree foliage
attenuation models
published - table
copied from [50].

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 LAN’s,
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
europénne 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
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]. That
work was later
continued in the
COST2100
group[27] in
particular to
provide further
MIMO advances for
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 [28][29].

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 [33],
the 802.11n and ac
models [34] for
various indoor
models (from the
802.11n task group
on channel
modeling), or [38]
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 [33].
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 [34] 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 (στ in
nanoseconds – see
table 4.2 in §??).

Model στ(ns) Environment Example

A 0 Direct Cabled
B 15 Residential In room or room-to-
room
C 30 Res. or small office Conference rooms,
classrooms
D 50 Typical office Cubicles in open
office space
E 100 Large office Large office space,
multi-floor
F 150 Large space 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
σ1 = 3dB near
transmitter, and
higher values σ2
beyond d0.

Model d0(m) n1 n2 σ1 σ2
A 5 2 3.5 3 4
B 5 2 3.5 3 4
C 5 2 3.5 3 5
D 10 2 3.5 3 5
E 20 2 3.5 3 6
F 30 2 3.5 3 6

3GPP SCM
The 3GPP spatial
channel models
(SCM) [38] 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 [38] 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
means additional
building
penetration loss in
the link budget.
Indoor penetration
measurements are
difficult to perform,
and difficult to
compare from one
experiment to
another. Difficulties
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 μi and
standard deviation
σi 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
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 σo) and
the in-building loss
(with standard
deviation σi). 10 And
the aggregate
random variate is
also log-normal
distributed, and has
a standard

deviation σ =
.

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:

(3.22)

where S is the
outdoor path, d is
the indoor path,
and

(3.23)

where Le is the
normal incidence
first wall
penetration; the
next term
represents the
added loss due to
angle of incidence θ
and is sometimes
measured over an
average of
empirical values of
incidence, in which
case it may be
noted L′ge = Lge(1 -
sinθ)2; and the last
term max(Γ1,Γ2)
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 L′ge = Lge(1-
sinθ)2 is sometimes
more convenient.
As a rough estimate
for angles of
incidence between
-π∕2 and π∕2 lead to
the following:

(3.24)

Empirical values of
L′ge are reported to
be ≈ 5.7 - 6.4 [54]
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: Γ1 =
niLi. In a corridor:
Γ2 = α(d′- 2)(1 -
sinθ)2, with an
empirical
propagation loss α =
0.6dB/m.

Figure 3.14: COST-
231 indoor
penetration loss
model.

Typical values for

the model reported
in [24] and [54] are
summarized in
table 3.12.

Table 3.12: Penetration
loss into buildings,
from COST-231
model.

Material Frequency Le Lge L′ge Li

Wood, plaster 900MHz 4 4 4
Concrete w/windows 1.8GHz 7 ≈20 6 10
Residential 2.5GHz 6.2 ≈10 6.1 3

Figure 3.15: Penetration
loss into residential
buildings,
cumulative density
distribution for
700 MHz, 900 MHz,
1.9 GHz, and
5.8 GHz.

3.7.2 Residential
Homes

In most residential
and suburban
environments,
surfaces involved
are mostly made of
glass, bricks, wood,
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 μi
around 10 to 15 dB
and a standard
deviation σi 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 (μi) and
standard deviation
(σi) from
experimental
results reported at
various
frequencies.

Source Frequency μi σi Comments

(GHz) (dB) (dB)
Aguirre [51][55] 0.9 6.4 6.8 7 Boulder residences
1.9 11.6 7.0
5.9 16.1 9.0
Wells [52] 0.86 6.3 6 Sat. meas. into 5 homes
1.55 6.7 6
2.57 6.7 6
Durgin [72] 5.8 14.9 5.6 [72]Table 5 average
Martijn [53] 1.8 12.0 4.0 [53]Table 1
Oestges [54] 2.5 12.3 — [54]Table 6 (avg. Le + L′ge)
Schwengler 1.9 12.0 6.0 Personal measurements
Schwengler [75] 5.8 14.7 5.5 [75]Table 2
Average 0.9 6.4 6.4
≈2 10.3 6.3
5.8 13.8 6.7

Figure 3.16: In-
building loss for
residential
buildings:
 measurements
campaigns
published for
different
frequencies, in
different
residential areas.

Figure 3.17: 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 [51][55];
some claim loss are
independent of
frequency [64][57];
others show a
decrease with
frequency [61][60]
[59].

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.
 Table 3.14: Penetration
loss into vehicles:
median loss (μv)
and standard
deviation (σv) from
experimental
results reported at
various
frequencies. (Some
references have
additional
measurements or
simulated values.)

Source Frequency μv σv Comments

(GHz) (dB) (dB)
Hill [66] 0.15 5.3 – Head level
0.45 6.9 –
0.8 3.8 –
0.9 3.9 –
Kostanic [67] 0.8 8.8 3.0 In minivan
0.8 8.4 3.1 In full size car
0.8 12.0 2.9 In sports car
Tanghe [68] 0.6 16.8 3.2 V pol. Tx rear of van
0.9 7.5 3.0
1.8 9.5 3.4
2.4 13.8 4.1
0.6 4.9 4.7 V pol. Tx front of van
0.9 3.2 3.3
1.8 3.8 4.4
2.4 5.3 4.1
Average 0.6 10.9 4.0
0.8-0.9 6.5 3.1
1.8 6.7 3.9
2.4 9.6 4.1

3.7.4 In-Vehicle
Loss

Given the mobile

nature of wireless
communications,
penetration loss
into vehicle is
important as well.
Precise
characterization of
in-vehicle
penetration is
difficult as well,
and varies with
type of vehicles,
frequency,
polarization,
antenna placement
in the vehicle, and
direction of
incidence. [66][67]
[68] An average
penetration loss μv
around 8 dB and a
standard deviation
σv around 3 dB
seems to be the
norm in published
studies – see table
3.14.

Just like in-building

penetration, in-
vehicle penetration
is a log-normally
distributed random
variate (with
standard deviation
σv), it is
independent of the
outdoor large scale
shadowing statistic
(σo), 11 and the
aggregate loss is
log-normal
distributed with
standard deviation
σ= .

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 [32]
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;
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 Γ = -1
b. Then use the actual
Γ 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.
7. Find and read the
three papers
referenced for in-
vehicle penetration
[66][67][68], as well
as any other such
 reports or paper if
you can.
a. Produce a table like
3.14 with clear
references, but
with more details.
b. Define average
values (for mean
and standard
deviation) that you
would recommend
using for 700MHz,
800MHz, 1.9GHz,
2.4GHz propagation
into vehicles.
Justify.
8. Find and read reference [33].
a. Implement the
software
emulations from
appendix B (e.g.
using Matlab) and
produce a
simulation of the
fading for SUI-3
with
omnidirectional
antenna at in figure
1 of that appendix
B.
b. Play with some
parameters of the
simulation, and
submit one more,
and analyze some
noticeable
differences.
9. This problem
studies the main
differences in link
budget to be
expected between
cellular and PCS
systems. This
problem is inspired
from a 1999 paper
by Chu and
Greenstein [40] —
feel free to refer to
it for further
details, but some
analyses have been
simplified, and
some values were
changed, so do not
use any numerical
values from the
paper or this
problem.

Cellular and PCS

systems are usually
FDD, thus operating
at different
frequencies for
uplink and
downlink. Still in
this problem, we
neglect that small
difference, and we
assume cellular
frequency to be
800MHz, and PCS
frequency to be
1900MHz.
a. Typical path loss
models like one
slope models do
take into account
the variation due to
frequency. For
simplicity we use
here the frequency
variation seen in
the free space
model. Evaluate
that variation (in
dB).
b. Calculate the link
budget difference
between cellular
and PCS. Give your
answer as
LB(cellular)-
LB(PCS) in dB.
We’ll try in the
following questions
to make up for that
difference.
c. Antenna gain can
make up some of
that difference.
Let’s assume that
we use antennas of
the same size,
meaning that they
have the same
effective area Ae =
Gaλ2∕(4π). What
antenna gain
difference ΔGa is
there between
cellular and PCS?
d. Another
improvement is
that of diversity. We
consider receive
diversity at the
base station only;
handset receive
diversity will be
considered
negligible. Diversity
gain depends
(inversely) on the
correlation ρ
between the
signals:
(3.25)

(3.26)

where λ is the
wavelength, D is
the antenna
separation, and σ is
the standard
deviation of the
angle of arrival,
which is difficult to
estimate, and we’ll
use an empirical
estimate of 1
degree, which
means σ = 0.0175
rad in the formula
above.

Calculate ρ at both
frequencies for 6-
feet antenna
 separation.
Calculate ΔGd, the
diversity gain
difference between
cellular and PCS.
e. Another difference
may be the impact
of vegetation in the
cell: a simple
vegetation
attenuation
formula is
sometimes:
Lv(dB∕km) = 1.33f0.284
with f in MHz. We’ll
consider that a cell
is on average 1km
radius, therefore
we’ll use Lv for our
link budget.
Calculate the
difference in
vegetation loss
between cellular
and PCS.
f. Noise at the
receiver is another
important
consideration.
Handset noise is
considered
identical, but base
station noise
parameters vary.
The main
difference seems to
be that ambient
noises (mostly man-
made) is higher at
800MHz, that
impact on the
receiver can be
approximated (in
dB) by Pmm = 2.4 -
2.3log(f∕100) with f
in MHz. Calculate
that difference
between cellular
and PCS.
g. In-building
penetration is also
likely to vary: from
the relevant figure
in the class notes,
estimate the
difference in
building
penetration loss for
the 90th percentile
of the CDF.
(Meaning the
additional loss we
should consider to
account for 90% of
residential
buildings.)
h. Summarize and
calculate the link
budget difference
seen in all
questions above.
That is the link
budget difference
between cellular
and PCS for the
 uplink (reverse
link).
i. Conclude and
comment about the
value of cellular
vs. PCS spectrum —
e.g. quantify
coverage.
10. Repeat the previous
problem to
compare 700MHz
and AWS
(1700MHz)
spectrum.

Copyright ©2019
Thomas
Schwengler. A
significantly
updated and
completed 2019
Edition is available.