Technical University Berlin

Telecommunication Networks Group

Wireless Channel Models

Ana Aguiar, James Gross
{aaguiar,gross}@ee.tu-berlin.de

Berlin, April 2003

TKN Technical Report TKN-03-007

TKN Technical Reports Series

Editor: Prof. Dr.–Ing. Adam Wolisz
 Abstract

In this technical report analytical models of wireless channels are presented. The report ad-
dresses the reader interested in the various effects, which lead to the well known, unreliable
and stochastic nature of wireless channels. The report is composed from various other books,
reports and so, due to the reason that a comprehensive, but still easy understandable dis-
cussion of the matter for engineers working on protocols is hard to find. Instead many other
presentations are quite specific, deal only with a certain amount of the topic, or do not give
a certain required level of detail on the other hand.
The report gives an insight into the matter of what influences the performance of wireless
channels. Two major effects are distinguished in the report: multiplicative effects, attenuat-
ing the transmitted signal, and additive effects, distorting the received signal at the receiver.
While the attenuating effects not always have to be considered as stochastic processes, the
distroting effects are always of that kind. However, for all effects good mathematical rep-
resentations are found, which might be utilized for considering the system perfomance by
simulation or analysis. Furthermore, at the end this report, we also briefly give an example
analysis of the performance of some digital wireless modulation channels.
TU Berlin

Contents

1 Introduction 3

2 Radio Channel 7
2.1 Path Loss . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.1 Antennas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.1.2 Free-Space Propagation . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.1.3 Two-ray Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.1.4 Empirical and Semi-empirical Models . . . . . . . . . . . . . . . . . . 13
2.1.5 Other Models and Parameters . . . . . . . . . . . . . . . . . . . . . . 16
2.2 Shadowing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2.1 Shadowing Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2.2 Measurement Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2.3 Shadowing Correlation . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.3 Fading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3.1 Physical Basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3.2 Mathematical Model of Fading . . . . . . . . . . . . . . . . . . . . . . 22
2.3.3 Characterization in Time and Frequency . . . . . . . . . . . . . . . . . 24
2.3.4 First Order Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.3.5 Second Order Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.3.6 Fading Rate and Duration . . . . . . . . . . . . . . . . . . . . . . . . . 36

3 Modulation Channel 40
3.1 Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.1.1 Thermal Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.1.2 Filtered White Gaussian Noise . . . . . . . . . . . . . . . . . . . . . . 41
3.1.3 Man-made Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.1.4 Some Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.2 Interference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

4 Digital Channel 45
4.1 Structure of the Digital Channel . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.2 Calculation of the Bit Error Probability as a function of SNIR for Binary PAM
over an AWGN Channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.3 Calculation of the Bit Error Probability as a function of SNIR for BPSK over
a Rayleigh Fading Channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 1
TU Berlin

4.4 Results for Other Digital Modulation Schemes over an AWGN Channel . . . 48

5 Conclusions 50

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 2
TU Berlin

Chapter 1

Introduction

The performance of any communication system is ultimately determined by the medium

utilized. This medium, may it be an optical fiber, a hard disk drive of a computer or a wireless
link, is referred to as communication channel. There exists a large variety of channels,
which may be divided into two groups. If a solid connection exists between transmitter and
receiver, the channel is called a wired channel. If this solid connection is missing, this
connection is called a wireless channel. Wireless channels differ a lot from wired channels,
due to their unreliable behavior compared to wired channels. In wireless channels the state
of the channel may change within a very short time span. This random and drastic behavior
of wireless channels turns communication over such channels into a difficult task.
Wireless channels may be further distinguished by the propagation environment en-
countered. Many different propagation environments have been identified, such as urban,
suburban, indoor, underwater or orbital propagation environments, which differ in various
ways.
In this report we focus on the factors which influence the performance of wireless channels.
We consider analytic models of basic propagation effects encountered in wireless channels
and show how they translate into the performance of different communication systems. For
example this knowledge is crucial in order to design and parameterize simulation models
of wireless channels. A different area where this knowledge is important is the design of
communication protocols. In general we address the reader who is interested in the analysis
behind wireless channel.
There are multiple books which include various aspects of wireless channels, such as
[4, 12, 13, 8, 17, 9, 3]. However, some lack a certain degree of precision when someone is
interested in modelling correlated fading on a wireless channel for example. On the other
hand some books deal only with some effects, while leaving out other aspects. Also some
sources have a degree of detail which is much to high for engineers working on protocol
design. Therefore this report has been compiled from various sources, giving a detailed
introduction to the main effects one has to consider when dealing with wireless channels
in order to evaluate communication protocols. Also we present numerical values for many
different situations, such that someone interested in simulating a certain scenario has an easy
and complete access to typical values fitting his scenario.
It is not always clear what is referred to as wireless channel in a communication system
since the are multiple instances in the transmission and reception process of a signal. Figure

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 3
TU Berlin

Transmitter Receiver

Packets

Bits 0100100100111010011010 0100100100111010011010

Digital Channel

Digital/ Analog/
Analog Digital

Baseband Baseband
Symbols Symbols

Modulator Demodulator

Modulation Channel

IF/RF IF/RF
Stages Stages

Propagation Channel

Radio Channel
Antenna Antenna

Figure 1.1: Channel classification: propagation channel, radio channel, modulation channel,
and digital channel.

1.1 represents the most commonly referenced channels (as referred in [18]) to clarify different
notions related to the concept of wireless channels in digital communication systems.

• The propagation channel: lies between the transmitter and receiver antennas and
is influenced only by the phenomena that influence the propagation of electromagnetic
waves. It is almost always linear and reciprocal so that these characteristics will be
assumed. All phenomena of this channel only effect the attenuation of the transmitted
signal and therefore this channel has an multiplicative effect on the signal. The signal
transmitted consists of the information modulated on top of the carrier frequency.

• The radio channel: consists of the propagation channel and both the transmitter
and receiver antennas. As long as the antennas are considered to be linear, bilateral
and passive, the channel is also linear and reciprocal. Still the signal is only effected by
attenuation, however the attenuation of the propagation channel might be different since
it might be modified by the used antennas, where the antenna influence is strictly linear.
The signal transmitted here is the same as with the propagation channel, however it
might be scaled by the use of antennas.

• The modulation channel: consists of the radio channel plus all system components
(like amplifiers and different stages of radio frequency circuits) up to the output of the
modulator on the transmitter side and the input of the demodulator on the receiver side.
Whether the system is linear depends on the transfer characteristics of the components

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 4
TU Berlin

between de- or modulator and the antennas. The channel is also non-reciprocal, because
amplifiers (the system component added to the radio channel) are non-reciprocal. Due
to the process of amplifying the received signal at this point, additive effects damaging
the signal come here into play. These are noise and interference. They might already be
present at the radio channel, however especially noise from electric circuits is added at
this channel level, therefore a complete characterization of the additive effects can not
be done at the radio channel level. The signal consists here of the baseband symbols,
which are modulated on top of the carrier frequency (refer to next paragraph).
• The digital channel: consists of the modulation channel plus the modulator and
demodulator. It relates the digital baseband signal at the transmitter to the digital
signal at the receiver, and describes the bit error patterns. The channel is non-linear
and non-reciprocal. At this channel level no further effects come into play, instead the
corrupted signal is interpreted at this level as bit sequence and if the signal has been
corrupted too heavily, the interpreted bit sequence differs from the true bit sequence
intended to convey. The input to this channel are bits, which might stem from packets.
The bits are grouped then and turned into analog representations, so called symbols.
These symbols belong to the baseband. This analog signal is then passed to a modulator,
which modulates these baseband signals on top of the carrier frequency.
Assuming ideal antennas, the propagation channel becomes identical to the radio channel.
The radio channel attenuates the received signal by a time varying factor, denoted by a(t).
This attenuation might be compensated by the modulation channel, since amplifiers are
employed here to boost the received signal. However, at the modulation channel random,
time varying noise n(t) also enters the system, which adds a distorting element to the signal.
If the attenuated signal is largely amplified, the noise will also be amplified strongly. Therefore
it is up to reliable detection methods at the digital channel to extract the transmitted signal
from the noise. In addition to noise, which is always present at this stage of a communication
system, it is also possible that electromagnetic waves from other communicating devices
interfere with the received signal. This is called interference and has a significant impact
on the performance, similar to noise. The interfering signal is also time variant and is denoted
by j(t). The resulting mathematical model of the received signal y(t), depending on the sent
signal s(t) and all influencing factors is given in Figure 1.2 (see also [4, 12, 17]).
For a running digital communication system a couple of performance metrics are of com-
mon use, such as the symbol error probability (SEP) or the bit error probability
(BEP) . Both performance metrics relate to the digital channel, the BEP relates to the in-
terpreted bit stream, while the SEP relates to the stream of symbols, not being interpreted
yet. Both metrics depend on the instantaneous power ratio between the received signal
power y 2 (t) and the noise and interference powers n 2 (t) and j 2 (t). This instantaneous power
ratio is given by the Signal-to-Noise-and-Interference Ratio (SNIR) . Note that the
attenuating influence of the radio channel is already included in the received signal power
y 2 (t). A varying SNIR might result in a varying SEP or BEP still this is not necessarily so.
If the average SNIR of a link is available, also the average error rates like symbol error
rate (SER) or bit error rate (BER) can be obtained. In general the relationship between
SNIR and error rates or error probabilities is not linear, instead it is highly complex and
depends on a lot of details. Therefore a deep understanding of the influence of all effects

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 5
TU Berlin

a(t) j(t) n(t)

s(t)
* + + y(t)

Radio Channel

Modulation Channel

Figure 1.2: Mathematical model of the modulation channel

regarding the receiver SNIR is required for investigations of the performance of any wireless
communication systems.
This report is structured as following: we first discuss the effects of the radio channel
influencing the signal, therefore at first the attenuation a(t) is discussed in Chapther 2.
Next the noise n(t) is characterised as well as the interference j(t), hence the effects of the
modulation channel influencing the signal are discussed in Chapter 3. Finally, we show how
these effects influence the received bit stream in the digital channel and therefore have an
impact on any of the mentioned performance metrics. This is shown in Chapter 4.

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 6
TU Berlin

Chapter 2

Radio Channel

The radio channel influences the received signal only by a multiplicative factor, the atten-
uation a(t), as given in Figure 1.2. Analytically it is useful to distinguish between three
different effects which result in an overall attenuation of the transmitted signal. The first
effect is called path loss. It is a deterministic effect depending only on the distance be-
tween the transmitter and the receiver. It plays an important role on larger time scales like
seconds or minutes, since the distance between transmitter and receiver in most situations
does not change significantly on smaller time scales. The second effect is called shadowing.
Shadowing is not deterministic. It varies on the same time scale as the path loss does and
causes fluctuations of the received signal strength at points with the same distance to the
transmitter. However, the mean over all these points yields the signal strength given by path
loss only. The third effect is called fading. Fading has also a stochastic nature, but leads
to significant attenuation changes within smaller time scales such as milliseconds or even mi-
croseconds. Fading is always caused by a multipath propagation environment, therefore by
an environment reflecting the transmitted electromagnetic waves such that multiple copies of
this wave interfere at the receiving antenna.
All three attenuating effects combined result in the actual experienced attenuation of the
radio channel. Therefore this attenuation might be decomposed as given in Equation 2.1 .

a(t) = aP L (t) · aSH (t) · aF A (t) (2.1)

In the literature shadowing is sometimes also referred to as slow fading while fading
is referred to as fast fading in these cases. However, since it is usefull to distinguish the
phenomenom of fading also into two groups, fast fading and slow fading (Section 2.3.3), we
will use throughout this report the terms shadowing and fading to clearly distinguish between
them.

2.1 Path Loss

In this section the phenomena that influences the propagation of electromagnetic waves will
be treated and expressions will be presented for the loss in power that a transmitted signal
experiences depending on the distance and frequency. This phenomenom is called path loss.

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 7
TU Berlin

In general it was said in Chapter 1, that the received signal y(t) is a stochastic process
with average power given by y 2 (t). Since we only deal with the radio channel at this point,
the signal of interest at the receiver, will be the signal at the antenna connector without
noise, and thus it is given by y(t) = a(t) · s(t), while the power of the received signal is
given by P0 = y 2 (t) = a2 (t) · s2 (t), where Pt = s2 (t) is the transmitted signal power. We
consider as attenuation a(t) only the component which represents the path loss, a P L (t),
which is constant in time for a specific non-changing environment, distance and frequency: it
is deterministic. It follows that P 0 = a2P L (t) · Pt = a2P L · Pt . This is valid when no movement
is assumed: neither the environment is changing nor is the receiver moving. This assumption
is equivalent to averaging the smaller scale phenomena in time (shadowing and fading) which
will be described in the next Sections.
We focus further on deriving path loss formulas and give some examples on how to param-
terise these formulas in realistic environments. Before going into path loss details, some
considerations about antennas should be made.

2.1.1 Antennas
The function of an antenna is to transforms electric energy in electromagnetic waves (trans-
mission) and to transform electromagnetic waves back into electromagnetic energy (recep-
tion). In the following, we will be talking about conventional antennas: passive and reciprocal
(similar characteristics both as transmitter and receiver antenna).
Antennas are characterised by two properties: the gain and the radiation pattern. The
antenna gain is a measure of the signal amplification introduced by a specific antenna
compared to a reference antenna (a dipole). The antenna pattern describes the variations
of the antenna gain with direction, with reference to the antenna itself. The antenna pattern
is usually represented by an attenuation with respect to the maximum antenna gain. The
antenna gain and pattern are the same both for transmission and reception.
Antennas can be classified as omnidirectional or directional, depending on whether the
gain remains constant in every direction or not, respectively.
When antennas are directional, the antenna pattern has to be taken into account when
doing path loss calculations. The value of the gain to be used is the one of the direction of
the straight line which connects transmitter and receiver. The antenna gain is then
[T x Rx]
g = Gmax [dB] − Gatt [dB]
[T x Rx]
, where Gmax is the antenna gain and Gatt is the value from the radiation pattern in the
direction of the straight line connecting transmitter and receiver (both in dB); the negative
sign is due to the antenna pattern being given as an attenuation with respect to the maximum
antenna gain.

2.1.2 Free-Space Propagation

The attenuation a signal suffers from propagating in free space over a distance d between
two antennas assuming line of sight—LOS 1 —, usually named free space path loss, can be
1
LOS: no objects obstructing the path between transmitter and receiver

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 8
TU Berlin

exactly calculated using the Maxwell equations to calculate the far field of an antenna and
is given by [8]
λ 2
 
P0
= gT x gRx ,
Pt 4πd
or in dB  
P0 P0 λ
[dB] = 10 log = 20 log + 10 log(gT x ) + 10 log(gRx ),
Pt Pt 4πd
where P0 is the received power, Pt is the transmitted power, λ is the wavelength, g T is the
gain of the transmitter antenna and g R is the gain of the receiver antenna (both gains in
the direction of the straight line that connects the two antennas in space, see Section 2.1.1).
The received power is inversely proportional to the square of the distance and the square of
the frequency. The physical explanation for the first is quite straightforward: in free space
(no obstacles or reflecting surfaces) the radiated energy propagates equally in every direction
and the wave can be seen as a sphere of increasing radius. Since energy cannot be destroyed,
it will be the same whatever the distance from the radiating point is. So that the total
energy over the sphere is the same independent of the radius, the energy per unit surface
must decrease. As the surface increases with the square of the radius, so does energy per
unit surface decrease at the inverse rate.

2.1.3 Two-ray Model

Since most communications happen close to the earth surface, the scenario for free-space loss
is unrealistic. The two-ray model, also known as plane earth, is a simple model based on
physical-optics theory which takes the reflection on the earth surface into account. It also
assumes LOS and no influence on propagation besides the earth surface. It is a useful starting
point for the study of propagation for personal communications. It is often used to describe
propagation over water or over open fields.
For its derivation, three waves should be considered: the direct one, one reflected on the
earth, and a surface wave. The latter becomes insignificant a few wavelengths above the
earth surface and is not significant for mobile communications. It will therefore be neglected.
The following expressions can easily be calculated from the geometry of the model shown
in Figure 2.1.
p
d1 = (h − hRx )2 + d2
p Tx
d2 = (hT x + hRx )2 + d2
hT x − hRx
α = arctan
d

From the free space propagation expression, the power of the direct wave at the receiver is
also easily calculated:
 2
λ
PR1 = Pt gT x gRx
4πd1
The power of the wave reflected on the earth at the receiver is calculated using the laws
of reflection of plane waves [9, 4]. An approximation is needed, since the reflection factors

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 9
TU Berlin

h

α
d21 d22
θ θ hm

Figure 2.1: Geometry for the calculation of propagation over plane earth (two-ray model).

used are actually calculated for plane waves although the wave is spherical. The reflection
factor for an incident wave from vacuum into a surface of electromagnetic properties  ef f (and
magnetic permeability 1 — no magnetic properties), which sets together the conductivity and
permitivity of the material is given by
q
cos θ − ef f − sin2 θ
R(θ, ef f ) = q .
2
cos θ + ef f − sin θ

Now the power from the reflected wave at the receiver can be written as a function of known
parameters:
 2
λ
PR2 = Pt R(θ, ef f ) gT x gRx
4πd2
The total power received is then given by
 2
λ 1 ej∆Φ
PR = P t gT x gRx + R(θ, ef f )
4π d1 d2

applying the superposition principle to the arriving electric field strengths, where ∆Φ =
2π(d2 − d1)λ is the phase difference between the two waves. An approximation can be
made using the Taylor series since (d2 − d1) is very small in the general case (see [9]):
d2 − d1 = 2hT xdhRx .
Since usually the vertical distance between transmitter and receiver is much smaller than
the horizontal one, the approximation d = d1 ≈ d2 can be used in the fraction terms. The
expression can be simplified if the earth surface is considered a metal surface 2 (a very usual
2
Average ground characteristics are assumed. Electromagnetic characteristics of other types of surface can
be found on page 83 of [8].

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 10
TU Berlin

assumption in electromagnetic calculations) and incidence angles θ very near 90 degrees

(distance to the reflection point much bigger than height of the transmitter) 3 . In this case,
the reflection factor R(θ, ef f ) takes values very near -1 and a simplified expression for the
received power can be written:
 2
λ
P0 = P t gT x · gRx · 2 · (1 − cos ∆Φ)
4πd
 2
λ 2πhT x hRx
= Pt gT x · gRx · 4 · sin2
4πd λd
 2
λ
= Pt gT x · gRx · 4 · sin2 (∆Φ/2), for horizontal polarisation
4πd

The approximation for horizontal polarisation in the previous equation is due to the
fact that in the real world, most of the time, waves behave like horizontal polarised waves
(independently of their actual polarisation).
For values of ∆Φ smaller than 0,6 rad, sin ∆Φ/2 ≈ ∆Φ/2 and the expression can be
further simplified to the known 4th-power-law form:
 2
hT x hRx
P0 = Pt · gT x · gRx · ,
d2

where the dependence on frequency vanishes.

We calculated the received power P0 = a2P L Pt as a function of distance according to
the three different models seen thus far: the free-space path loss, the two-ray model (with
the assumption that the incidence angle is very near 90 degrees), and the 4th power law
approximation to the two-ray model. The curves can be seen in Figure 2.2. It can be seen
that for the two-ray model there are clearly two different areas: near the transmitter (before
the breakpoint), where the received power decreases according to the squared sinus function,
with the peak value following the square of the distance; and after the breakpoint, when
the phase difference between direct and reflected rays is smaller than 0.6 rad, the second
approximation becomes valid, and the received power decreases with the 4th power of the
distance (so that the difference between the curves can no longer be seen). The breakpoint
can be calculated according to dBreakpoint = d|∆Φ<0.6 = 2πh0.6
T x hRx
λ .
The theoretical results presented have been confirmed by measurements campaigns: re-
sults have shown that the 2-ray model fits quite well the actual path loss in line-of-sight
(LOS) environments with few or no reflectors and scatterers, e. g. highways (see Section
1.2.6 of [18]).
As was mentioned, the simplified expressions are valid only for very low incidence angles.
For steeper incidence angles (transmitter antenna height in the magnitude of the distance)
the parameters of the reflecting surface play a determinative role on the propagation and the
more accurate expression should be used.
3
Notice also that for frequencies above 100 MHz, changes in the ground parameters do not cause big
variations in the path loss [8].

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 11
TU Berlin

-50

-100

-150
Received Power (dBm)

-200

-250

-300

-350

-400

-450
100 1000 10000 100000
Distance d (m)
Free Space 4th-power Law 2-ray Model

Figure 2.2: Value of the received power P 0 as a function of the distance between transmitter
and receiver according to the free space path loss, the two ray model and the 4th-power law.
The parameters used are: antennas with unit gain, h T x =25 m, hRx =1.5 m, transmitted
power 0 dBm (1 mW) and 2.4 GHz frequency.

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 12
TU Berlin

2.1.4 Empirical and Semi-empirical Models

For the exact calculations in the previous sections LOS was assumed, as well as no other
objects surrounding the path or the transmitter and receiver. These assumptions are not
valid in many realistic environments like for urban, suburban and indoor environments, where
non-LOS (NLOS) is as common as LOS, a multitude of physical phenomena influence the
propagation of electromagnetic waves and the number of possible propagation paths is very
high. Ray-tracing models are accurate path loss models, which calculate ever possible path
between transmitter and receiver, then the attenuation suffered in each path, and finally add
all the signal components which arrive at the receiver. These methods not only require exact
data about the terrain, the buildings and vegetation but are also very demanding in terms
of computing capacity to process all the data and therefore extremely time consuming.
For these reasons, empirical and semi-empirical models have been developed to calculate
the path loss between a transmitter and a receiver a certain distance from each other in
specific environments for different frequencies. The first are based on extensive measurement
campaigns in different environments and the latter on a mix of empirical and theoretical data.
For every new area, calibration measurements are required to calculate correction factors for
the general models. The models are usually of the form:
P0 1
=K· α
Pt d
or, in dB,
P0
[dB] = 10 log(K) − 10 · α · log(d)
Pt
, where the constants K and α are fitted to measurement results according to the areas
under consideration. The factor K usually depends on the frequency used, as well as height
of the base station and wireless terminal. The distance d is in units referenced to a reference
distance, and has to be defined along with the exponent. Notice that no difference is made
between LOS and NLOS anymore, since the models are obtained averaging measurements
obtained under both conditions, and te different propagation effects are approximatively
condensed in a single parameter — α.
Some well-known and widely used models are explained next. An overview of other path
loss models can be found in [10].

The Okumura-Hata Model

The Okumura-Hata model is the most popular of the empirical models. It is based on
extensive measurement campaigns made by Okumura in Japan and on a formula developed
by Hata which approximates the measured statistics. It is valid for the following values of
the parameters:

• frequency: 150. . . 1000 MHz

• distance: 1. . . 20 Km

• Transmitter height over ground: 30. . . 200 m

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 13
TU Berlin

• Receiver height over ground: 1. . . 10 m

The formula for the path loss is

A[dB] = 69.55 + 26.16 · log(f [M Hz]) − 13.82 · log(h T x[m] )

+ (44.9 − 6.55 · log(hT x [m])) · log(d[m]) − β ,

where β is a correction factor which depends on the environment and takes the following
values:
[1.1 · log(f [M Hz]) − 0.7] · hRx [m] − [1.56 · log(f [M Hz]) − 0.8]





 , for small cities
8.29 · [log(1.54 · hRx [m])]2 − 1.1


β=
 , for urban areas and f ≤ 300 MHz
· hRx [m])]2 − 4.97

3.2 · [log(11.75





, for urban areas and f ≥ 300 MHz

For three further environments, further correction factor exist form the corresponding urban
formula:

βurban − 2 · [log(f [M Hz]/28)] 2 − 5.4






 , for suburban areas
βurban − 4.78 · [log(f [M Hz])] 2 + 18.33 · log(f [M Hz]) − 35.94


β2 =

 , for almost open rural areas
β − 4.78 · [log(f [M Hz])]2 + 18.33 · log(f [M HZ]) − 40.94


 urban


, for open rural areas

This model has been extended by COST (European Cooperation in the Field of Scientific
an Technical Research) to COST 231-Hata for frequencies between 1500 MHz and 2000 MHz:

A[dB] = 46.3 + 33.9 · log(f [M Hz]) − 13.82 · log(h T x [m])


3 for big city centre
+ [44.9 − 6.55 · log(hT x [m])] · log(d[Km]) − β +
0 other cases
β = [1.1 · log(f [M Hz]) − 0.7] · hRx [m]
− [1.56 · log(f [M Hz]) − 0.8]

The Lee Model

This model [9] is quite popular because its parameters can easily be adapted to a new envi-
ronment using measurement results. The model consists of two parts:

• the point-to-point model which takes the terrain into account;

• the area-to-area model based on the previous one, which reflects the effects of construc-
tions.

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 14
TU Berlin

The area-to-area model is similar to the Hata model (Section 2.1.4), where the parameters
can easily be found by measurements in the area of interest. Since the area-to-area model
reflects the effects of the man-made constructions, if the terrain is hilly, for every distance,
measurements have to be made at different terrain heights. The parameters are always the
calculated average values from the measurements. Since the measurements are made for
certain specific values of antenna height, transmitted power and antenna gain, the model
formula has to be corrected to refer the actual used parameters to the ones used in the
measurements.
The point-to-point part of Lee’s model takes the terrain effects into account, using the
effective antenna height4 and diffraction losses, depending on the situation. The dependance
on the effective antenna height is taken from the two-ray model to be of 20 db/dec. The
calculation of the effective antenna height used in the expression for the path loss depends
on the path under consideration, specifically, on whether it is obstructed or not. In the first
case it will be calculated using knife-edge diffraction rules, on the latter using image rule
from a two-ray model. The calculation of the effective antenna height is explained in detail
in Section 4.7 of [9] and depends on whether a line of sight path is available or not.

h0e
P0 = Pr0 − 10 · α · log(d) + 20 · log + CF
h1
, for a path without obstructions
h00
P0 = Pr0 − 10 · α · log(d) + 20 · log e + L + CF
h1
, for an obstructed path and h00e ≥ 1
P0 = Pr0 − 10 · α · log(d) + L + CF
, for an obstructed path and h00e ≈ 1
 
λ
P0 = P t − 20 · log + 10 · log(gT x ) + 10 · log(gRx )
4πd
, for propagation over a water surface,

where the first terms Pr0 − 10 · α · log d account for the loss due to human-made structures
(parameters taken from measurements) and d refers to the distance used to calculate the
exponent of the path loss, CF is the correction factor which has the constants due to referring
actual parameters to the ones used fr model calibration, h 0e and h00e are effective heights for
the LOS and non-LOS situations, and L is the loss due to knife-edge diffraction. Vegetation
loss can be added to any of the expressions depending on the situation.
According to Lee [9], the use of the point-to-point models reduces the standard deviation
of the predicted values to 3 dB, from the value of 8 dB usually obtained with the area-to-
area model. This model is only valid for big distances (above 1 Km) since for lower distances
street orientation and buildings can cause very big deviations on actual values and calibration
measurement are no longer statistically representative.
4
Effective antenna height is the height of the transmitter antenna referred to the height of the receievr
antenna

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 15
TU Berlin

City Used antenna height (m) α

Hamburg 40 2.5
Stuttgart 23 2.8
Dusseldorf 88 2.1
Frankfurt 20 3.8
Frankfurt 93 2.4
Kronberg 50 2.4

Table 2.1: Path loss exponent measured in 4 european cities at 900 MHz for a reference
distance of 100 m (source [15, 14]).

Environment Model Type Used antenna height (m) α 10 · log K [dB]

Urban LOS 4 1.4 58.6
Urban NLOS 4 2.8 50.6
Urban LOS 12 2.5 35.8
Urban NLOS 12 4.5 20.0
Urban LOS 45 3.5 16.7
Urban NLOS 45 5.8 -16.9
Suburban LOS 12 2.5 38.0
Suburban NLOS 5 3.4 25.6
Rural LOS 55 3.3 21.8
Rural NLOS 55 5.9 -27.8

Table 2.2: Parameters of the general path loss model from measurements in Finnland for
5.3 GHz and a reference distance of 1 m(source [21]).

2.1.5 Other Models and Parameters

Besides the models described previously, measurement campaigns have been made in different
environments in several countries to better adapt the parameters of the general model form
to the real propagation environment, since building materials, urban planning and vegetation
differ from country to country.
In [11] measurement done in Tokio for frequencies ranging from 457 MHz to 15.45 GHz
have shown that the frequency dependency of the path loss in the UHF band ([300;3000] MHz)
follows the free-space trend of 20 · log(f ).
Best-fit parameters for measurements made in european cities for 900 MHz and a reference
distance of 100 m can be found in [15, 14] (see Table 2.1.
In [16] the value of α for 3.5 GHz in a dense urban environment is found from measure-
ments to be 4.7-4.8.
Results of path loss model parameters obtained from a least square analysis of mea-
surement results in several environments at 5,3 GHz in outdoor urban, suburban and rural
environments are presented in [21] — see Table 2.2.
The models presented and refered in this short survey are for the frequency ranges in which

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 16
TU Berlin

wireless LANs and HiperLAN operate (2,4-5 GHz). It is not pretended to list all existing
models, but to give a short overview of the different possibilities and of the parameters which
influence both propagation and model development and choice.
The models presented are from the author’s point of view the most meaningful in the
UHF and VHF frequency bands, either because they are widely accepted, or because they
are representative of other models. Several other models exist for other frequencies and
environments, which are certainly not less important, but are out of the scope of this short
survey.

2.2 Shadowing
The path loss model presented in the previous section aims at a deterministic calculation of
the path loss for a determined position of transmitter and receiver. In reality the position of
a receiver involves also the objects surrounding the transmission path as well as the terrain.
Measurements have been made under several different conditions and statistical variations
have been observed. For a fixed frequency and distance, different values of the received signal
power were measured. Thus, for a given fixed distance, frequency and transmission power,
the received signal power is not deterministic, but varies due to the objects in and around
the signal path. These stochastic, location dependent variations are called shadowing and
were denoted in Equation 2.1 by aSH (t). Note that these stochastic variations are constant in
time, as long as the receiver and his complete environment do not move. Shadowing reflects
the differences in the measured received signal power with relation to the theoretical value
calculated by path loss formulas. Averaging over many received signal power values for the
same distance, however, yields the exact value given by path loss.
Note that shadowing is an abstraction which reflects the result of a sum of several propa-
gation phenomena which occurr when an electromagnetic wave propagates in an environment:
reflections (e. g. on buildings and ground), diffraction (e. g. around buildings), refraction
(e. g. through walls or windows), scattering (e. g. on buildings, trees or ground) and absorp-
tion (e. g. on forest or parks)5 . The calculation of the effects of every of these phenomena
for each location is not feasible (sometimes even impossible) both due to complexity and
time limitations. Therefore, shadowing is used to describe the aggregated effects of all these
phenomena.
The objects causing these variations are of such dimensions that a receiver moving along
a line at constant distance from the transmitter will take several hundreds of milliseconds
(ms) to move to an area with different characteristics.

2.2.1 Shadowing Model

From measurements of path loss for a variety of environments and distances, the variations
of the measured signal level relative to the average predicted path loss were calculated (see
figures 2.37 to 2.41 from [18]). Its distribution is normal with 0 mean in dB, which implies
a log-normal distribution of the received power around the mean value corresponding to
5
Note that, although the term shadowing comes from diffraction around buildings (a non LOS situation),
it is also present in LOS situations (see the measurements results shown in ...

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 17
TU Berlin

City Used antenna height (m) σSH [dB]

Hamburg 40 8.3
Stuttgart 23 9.6
Dusseldorf 88 10.8
Frankfurt 20 7.1
Frankfurt 93 13.1
Kronberg 50 8.5

Table 2.3: Standard deviation of the path loss measured in european cities at 900 MHz
(source [15, 14])

the path loss. This hypothesis has been verified with the χ 2 and Kolmogrov-Smirnov test
and found to be valid with high confidence intervals. The theoretical basis to the log-normal
distribution is that in an environment with surrounding objects different signals suffer random
reflections and diffractions as they traverse the propagation medium. Expressed in dB, the
extra loss in each path corresponds to subtracting a random loss from the path loss value.
As the different propagation paths are independent, the sum of all the dB losses for a large
number of propagations path converges to a normally distributed random variable (central
limit theorem). In natural units, that becomes a log-normal distribution.
The shadowing variations of the path loss can therefore be calculated from the distribution

1√ a2
p(aSH ) = σSH 2π
exp − 2σSH
2 ,
SH

where σSH is the variability of the signal and all variables are expressed in dB.
The value of the variation due to shadowing is then added to the path loss value to obtain
the variations.
P0
a[dB] = 10 · log Pt = aP L [dB] + aSH [dB]

2.2.2 Measurement Results

The standard deviation of aSH , σSH , usually takes values between 5 dB and 12 dB [20],
depending on the communications system used (value is chosen based on extensive measure-
ment campaigns). For cellular communication networks a value of 7-8 dB [9, 19] taken from
several measurements is usually used.
Measurement campaigns have been made in several different environments and some
results for the standard deviation σ SH are presented here.
In [16] σP LSH is found to be around 6.12 dB for a dense urban environment at 3.5 GHz.
Some values of the standard deviation of the path-loss from measurement results at
5,3 GHz in outdoor urban, suburban and rural environments are presented in Table 2.5.

2.2.3 Shadowing Correlation

The autocorrelation of the shadowing process in space also needs to be modelled, since values
at close locations are expected to be correlated. A simple negative exponential correlation

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 18
TU Berlin

Frequency (MHz) Average building height(m)

30 20 20
457.2 8.62 6.34 6.89
813 — 5.10 —
2200 7.29 5.59 7.56
8450 7.98 5.48 7.21
15450 7.45 5.2 6.82

Table 2.4: Standard deviation of the path loss measured in Tokio at several UHF frequencies
for different environments, here for BS-WT distance lower than 1 Km and BS antenna height
above rooftops (see source [11] for further results)

Environment Model Type Used antenna height (m) σ P LSF [dB]

Urban LOS 4 3.7
Urban NLOS 4 4.4
Urban LOS 12 2.9
Urban NLOS 12 1.7
Urban LOS 45 4.6
Urban NLOS 45 2.8
Suburban LOS 12 4.9
Suburban NLOS 5 2.8
Rural LOS 55 3.7
Rural NLOS 55 1.9

Table 2.5: Standard deviation of the path-loss from measurements in Finnland for 5.3 GHz
(source [21]).

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 19
TU Berlin

model is proposed in [7] to work on an Okumura path loss model:

vT
2
RP L(k) = σSH · a|k| a = DD ,
where k is the distance between two points in m,  D is the correlation between two points
at distance D, T is the interval between samples and v is the speed of the movement. The
parameters of the model — a and k — should be fitted to measured values after removal of
fast fading variations.
The proposed model has been fitted to data for urban and suburban environments in two
European cities. The data for suburban environments was measured at 900 MHz, and σ aSH
was estimated to be 7.5 dB and D 0.82 for a distance of 100 m. The urban data was collected
for 1700 MHz and the estimated model parameters were 4.3 dB for σ SH and 0.3 for D for a
distance D of 10 m. The fit in suburban environments is good for distances k up to 500 m
but for urban environments the model departs from the measured values for distances above
15 m. The measured values of autocorrelation in an urban environment are also very low
(below 20%) for distances above 13 m (see Figure 2 in [7]). It is suggested that this is due to
the fact that fast fading (see next section) was not completely removed of the measurement
traces used.

2.3 Fading
Fading is the interference of many scattered signals arriving at an antenna. It is responsible
for the most rapid and violent changes of the signal strength itself as well as its phase. These
signal variations are experienced on a small time scale, mostly a fraction of a second or shorter,
depending on the velocity of the receiver. In this Section we will discuss the physical reasons
causing fading, present a mathematical model for fading and characterize it as a stochastic
process. Fading might have a time varying or frequency varying attenuating impact on the
transmitted signal, denoted by aF A (t) in Equation 2.1. Due to the frequency varying and
time varying (complex valued) nature of fading, we will denote the attenuating impact in
this Section by Ḡ(t, f ). The relationship to the notation used in Equation 2.1 is given by
aF A (t) = |Ḡ(t, f )| for the observed carrier frequency. In some cases the fading might be only
time varying or frequency varying, in these cases we denote the fading by ḡ(t) = Ḡ(t, 0) in
the case of time varying fading only and by Ḡ(f ) = Ḡ(0, f ) in the case of frequency varying
fading only. The relationship to the notation used in Equation 2.1 is in these cases still given
by aF A (t) = |Ḡ(t, f )| for the observed carrier frequency.
In the following we will first introduce a mathematical model, which allows the an an-
alytical study of multipath propagation and fading. Then we will discuss first and second
order statistics of fading as well as methods in order to characterize fading channels and
certain parameters describing the severeness of the fading channel. Most of this is following
the chapther abpout fading in [4].

2.3.1 Physical Basis

The physical basis of fading is given by the reception of multiple copies of the transmitted
signal, each having followed a different path. Depending on the environment of transmitter

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 20
TU Berlin

and receiver, there can be many or only few objects reflecting the transmitted radio signal.
In general these objects are known as scatterers and the transmission of a signal leads to a
situation shown in Figure 2.3, which is called a multipath signal propagation.

Receiver

Transmitter

Scatterer

Figure 2.3: Multipath propagation scenario with a transmitter, a receiver and seven scatterers

In a typical environment each path i has a different length l i . Due to this difference in
length, each signal travelling along a path arrives with a different delay τ i = lci , where c is
the speed of light. Some signal copies travelling along short paths will arrive quite fast, while
other copies travelling along longer paths will arrive later. Physically this equals an echo,
encountered in a canyon. The channel is said to have a memory, since it is able to store signal
copies for a certain time span.
Beside this multipath propagation, each signal copy is attenuated differently, since the
signal paths have to pass different obstacles like windows, building walls of different materials,
trees of different sizes and so on. Denote the attenuation factor of path i by a i .
Taking all this into account, the multipath propagation of a transmitted radio wave results
in an interference pattern, where at certain points the wave interfere constructively while at
other points they interfere destructively. If each element within the propagation environment
(transmitter, scatterers, receiver) do not move, the receiving signal will only suffer from the
delay spread and the different attenuation. In this case, the interference situation of the
channel stays constant and therefore the channel is said to be time invariant. In contrast,
if any kind of movement is encountered in the propagation environment, all or some paths
change in time, such that all ai and τi change in time. As a consequence the wireless channel
become time variant. Here, along with a constant changing delay spread, the receiver also

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 21
TU Berlin

experiences a varying signal strength due to its movement through the interference pattern,
therefore the received signal fades.

2.3.2 Mathematical Model of Fading

Consider the transmission of a bandpass signal at carrier frequency f c with complex envelope
s̄(t). This transmitted bandpass signal is given by Equation 2.2.
 
s(t) = Re s̄(t) · e2πj·fc t (2.2)
The received bandpass signal is given by Equation 2.3.
 
y(t) = Re ȳ(t) · e2πj·fc t (2.3)
We look for a mathematical model of the received bandpass signal taking into account
the effect of multipath propagation. At first we consider the case where we do not encounter
motion in the environment. As described in Section 2.3.1, each path is associated with a
different length li and a different attenuation ai . Therefore the received signal y(t) is the
superposition of all copies, given in Equation 2.4.
    !
li li

l
X X 2πj·fc · t− ci
y(t) = ai · s t − = Re ai · s̄ t − ·e (2.4)
c c
∀i ∀i
c
Considering the relationship between wavelength and frequency λ = fc , we obtain a
f c li
complex envelope representation in Equation 2.5. Denote by ϕ i = 2π · = 2π · lλi the phase
c
shift of the carrier frequency caused by the different length of each path. Also recall the
introduction of the path delay τi = lci from Section 2.3.1.
  X
X l
−2πj· λi li
ȳ(t) = ai · e · s̄ t − = ai · e−j·ϕi · s̄ (t − τi ) (2.5)
c
∀i ∀i
Without motion a multipath environment leads to the interference of multiple copies with
a different attenuation of the envelope, respectively the carrier, (a i ), a different phase shift
of the carrier (ϕi ) and a different delay of the envelope (τ i ).
Now let us consider the effect of motion in this model. Does the motion of the receiver
have a great impact on the behavior of the received signal, as described in Section 2.3.1 ?
Denote by γi the angle of arrival of path i with respect to the direction of motion of the
receiver, as shown in Figure 2.4.
The path length change, as a function of speed v and time t is given by ∆l i = −v·cos (γi )·t.
From this we obtain a different function for the complex envelope, which depends now on the
time t, as given in Equation 2.6 .

 
X
−2πj·
li +∆li li + ∆li
ȳ(t) = ai · e · s̄ t −
λ
c
∀i
 
X v v · cos (γi ) · t
= ai · e−jϕi ·e −2πj·cos (γi )·t λ
· s̄ t − τi + (2.6)
c
∀i

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 22
TU Berlin

Angle of Signal
Arrival (Gamma)

Signal Direction
Copies of Travel

Receiver

Figure 2.4: Detailed sketch of signal arrival

Equation 2.6 can be simplified. First we include the phase e −jϕi in ai , but indicate this
by writing Āi instead. Second, when comparing the delay caused by the term v · cos(γ i )t/c
with the overall signal length of the complex envelope s̄(t), then the delay variation is very
short such that it can be ignored. Another simplification is done through introducing the
Doppler frequency fd = fcc · v = λv and the Doppler shift νi = cos(γi ) · fd . With this we
obtain Equation 2.7.
X X
ȳ(t) = Āi · e2πj·cos(γi )·t·fd · s̄ (t − τi ) = Āi · e2πj·νi ·t · s̄ (t − τi ) (2.7)
∀i ∀i

Interpreting Equation 2.7 yields the following. The motion of the receiver in combination
with the i-th scatterer influences the received signal in the amplitude of the carrier, respec-
tively the envelope and in the phase ( Āi ), in the carrier frequency (νi ) and in the delay of the
envelope (τi ). We have not mentioned the delay change of the envelope, which is relatively
small. Therefore motion of the receiver or a scatterer in the model introduces a frequency
offset of the carrier in addition to the signal changes that are already present if no motion is
involved.
If the number of scatterers is very high, the discrete scatterer model has to be turned into
a continuous scatterer model, where each specific scenario is represented by a gain density,
given by the delay-Doppler spread function in Equation 2.8.
X
%̄ (ν, τ ) dνdτ = Āî (2.8)
∀î

Here, î indexes all scatterers with delay in dτ and Doppler shift in dν. With this the
received signal ȳ(t) is given in Equation 2.9.
Z ∞ Z fd
ȳ(t) = %̄ (ν, τ ) · e2πj·νi ·t · s̄ (t − τi ) · dνdτ (2.9)
0 −fd

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 23
TU Berlin

2.3.3 Characterization in Time and Frequency

Out of the different impacts on the signal received in a multipath environment including
motion, the frequency offset (Doppler shift) of the carrier and the time delay of the envelope
damage the signal most. This is because these shifted and delayed waves might interfere
destructive and therefore cause severe attenuation. In practice a wireless transmission in a
certain environment including a certain velocity of objects is described by two values, the
Doppler spread ∆fd and the delay spread ∆τ . Both spreads result from multipath reception
(and in the case of the Doppler spread also from the velocity involved), where each path
may be characterized by a different Doppler shift (due to a different receive angle) and time
delay. While the Doppler spread is caused by the motion of objects within the environment
(which might be the transmitter, the receiver or scatterers), the delay spread is caused by
the topology of the environment itself. As we will see, although the Doppler spread is a
phenomenon in frequency (generating Doppler shift, a shift in frequency), the overall result
on the received signal (which is the result of multiple Doppler shifted signal copies interfering)
is a time selective behavior. For the delay spread this is exactly the opposite. While the delay
spread is a phenomenon in time, the resulting impact on the received signal is a frequency
selective behavior. This is now derived from the mathematical model in Section 2.3.2. Here
we first start with the discussion of the Doppler spread impact, then the discussion of the
delay spread impact follows.
Consider a receiver, which moves through a multipath environment with a certain fixed
speed. Further consider all path delays in this environment to be negligible small, such that
s̄(t − τi ) ≈ s(t). Then the received complex envelope, given by Equation 2.7, is simplified
and turns into Equation 2.10.
X
ȳ(t) = s̄(t) · Āi · e2πj·cos(γi )·t·fd = s̄(t) · ḡ(t) (2.10)
∀i

Here ḡ(t) is called the complex gain of the channel. In this case, the input s̄(t) and
the output ȳ(t) of the channel are connected by a simple multiplicative relationship. Be-
cause the phase angles 2πj · cos(γi ) · t · fd change in time, the channel complex gain is time
varying. Considering the transmission of a pure tone (setting the amplitude constant, thus
s̄(t) = U ), the received signal would be spread out in frequency, shifted within the interval
[−fd , fd ]. Due to this spreading, the received signal ȳ(t), which consists now of several tones
at different frequencies interfering at the receiving antenna, would vary in time. The wireless
channel therefore becomes time selective. At some time instances the received signal is not
attenuated and could appear even enhanced, at other time instances the signal is severely
attenuated. Thus, ḡ(t) varies in time.
For maximum values of the Doppler shift f d at different velocities and different carrier
frequencies refer to table 2.6.
The severity of the time selective behavior caused by the Doppler spread depends on
the time span the receiver needs to process the incoming envelope at the digital channel
(as explained in the introduction). If coherent detection is assumed (where each symbol
is processed independently, since transmitter and receiver are perfectly synchronized), the
processing time is the length of the envelope itself, which is the symbol length T s . If the
receiver employs for example differential detection (as with DPSK), the processing time

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 24
TU Berlin

Carrier Frequency fd @ 1 ms fd @ 10 ms fd @ 20 ms fd @ 100 m

s
1 Mhz 0.003 Hz 0.033 Hz 0.066 Hz 0.33 Hz
100 Mhz 0.33 Hz 3.33 Hz 6.6 Hz 33 Hz
1 Ghz 3.33 Hz 33.33 Hz 66 Hz 333 Hz
2.4 Ghz 8 Hz 80 Hz 160 Hz 800 Hz
5.4 Ghz 18 Hz 180 Hz 360 Hz 1.8 kHz
10 Ghz 33 Hz 333 Hz 666 Hz 3.33 kHz
60 Ghz 200 Hz 2 kHz 4 kHz 20 kHz

Table 2.6: Doppler shift frequencies for different carrier frequencies at velocities comparable
to walking (1 m m m m
s ), a slow car (10 s ), a fast car (20 s ) and traveling per train (ICE 100 s )

equals twice the time span of a symbol, since the receiver always has to take two symbols
into account in order to obtain the information. If the receiver also employs an equalizer,
a device which tries to revert the attenuating influence of th channel by means of adaptive
digital filtering over many symbols, this time span is further increased. In general we represent
the processing time span by N · Ts . If the fade rate of the time selective process given by
the Doppler frequency fd is larger than the processing rate (given by N 1·Ts ), then the fading
is called time selective6 . In contrast, if the fade rate is much lower than the processing
rate, therefore if fd · N · Ts  1, then the fading is called not time selective. These two
conditions are also named fast and slow fading, respectively.
Now let us focus on the impact of delay spread, without the presence of Doppler spread.
For a stationary receiver, we can assume the phases of the reflected signal copies to be
constant. Therefore we can consider Equation 2.11 as mathematical model.
X
ȳ(t) = Āi · s̄ (t − τi ) = ḡ(t) ∗ s̄(t) (2.11)
∀i

The input is in this case related to the output by convolution with the complex gain ḡ(t).
Since the phases are constant (but still random), the channel might be modelled in this case
as linear, time invariant filter with an impulse response given in Equation 2.12.
X X
ḡ(t) = ai · e−j·ϕi · δ (t − τi ) = Āi · δ (t − τi ) (2.12)
∀i ∀i

In the frequency domain the relationship between input S̄(f ) and output Ȳ (f ) is given
by multiplication with the frequency response of the filter, the complex tranfer function
Ḡ(f ). The transfer function is given in Equation 2.13.
X
Ḡ(f ) = Āi · e−2πj·f ·τi (2.13)
∀i
Since the delays τi are different for several paths, some frequencies are attenuated while
others are not. If the delay difference between the paths is very small or even not existing,
6
Note that a channel is always time selective if motion is present, while time selctive fading depends on
the relationship between symbol time and Doppler frequency

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 25
TU Berlin

Environment τrms
Urban 1 − 25µ s
Suburban 0.2 − 2µ s
Indoor 25 − 250 ns

Table 2.7: Standard deviation of delay spread values for three often referred transmission
environments

Criteria Category
∆τ
Ts  1, fd· Ts  1 not frequency selective (flat), not time selective (slow)
¬ ∆τTs  1 , f d · T s  1 frequency selective, not time selective (slow)
∆τ
Ts  1, ¬ (fd · Ts  1)

not frequency selective (flat), time selective (fast)
¬ ∆τTs  1 , ¬ (fd · Ts  1) frequency selective, time selective (fast)

Table 2.8: Categories in order to characterize the fading of a wireless channel depending on
the Doppler and delay spread

then no frequency is attenuated by the delay spread. The severity of the delay spread can be
estimated by the product of the required baseband bandwidth of the signal (denoted by W ,
and related to the symbol time Ts ) and the delay spread. If the delay spread is very small
compared to symbol time Ts , then it has almost no impact on the reception of the signal (if
∆τ · W  1). In this case the transfer function of the channel has no significant attenuating
behavior within the bandwidth of the signal W . The fading is called to be flat or frequency
non selective, since for the utilized frequencies of the channel no signal attenuation is present
due to delay spread. On the other hand if the delay spread is significant compared to the
symbol time Ts , then the channel has a frequency selective behavior. That is, at some
frequencies of the baseband signal the received signal is attenuated while at other frequencies
the signal might be enhanced. In this case the receiver suffers in the time domain from
intersymbol interference (ISI ) . If the delay spread is for example half of the symbol
time, then signal copies of two consecutively sent symbols interfere at the receiver, such that
the ’fast’ signal copy of the latter sent symbol interferes with the ’slow’ signal copy of the
previous sent symbol.
Table 2.7 gives typical ranges for the standard deviation of the delay spread. Note that
mean values of the mentioned environments are of no interest, since the variation of delays
damages the signal, not a longer or shorter mean delay.
In practical situations most of the time both, Doppler and delay spread, are present.
However, both effects can be of harm or not, depending on the ratio between the symbol
time (baseband bandwidth) and the characteristic values of the effects f d and ∆τ . Therefore,
a channel might be categorized into four different types, always depending on the ratios
mentioned. The four categories are listed in Table 2.8, assuming coherent detection.
If both kinds of spread are present the channel has to be modelled as linear time variant
filter (the filter model is necessary due to the delay spread, the time variant behavior is due

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 26
TU Berlin

to the Doppler spread). In this case input and output of the channel are related by Equation
2.14.

X Z ∆τ
ȳ(t) = Āi · e2πj·νi ·t · s̄ (t − τi ) = ḡ (t, τ ) ∗ s̄(t) = g (t, τ ) · s̄(t − τ )dτ (2.14)
∀i 0

The impulse response is given by Equation 2.15.

X
ḡ(t, τ ) = Āi · e2πj·νi ·t · δ (τ − τi ) (2.15)
∀i
In the frequency domain, Equation 2.14 turns into Equation 2.16.
X
Ȳ (f ) = Ḡ (t, f ) · S̄(f ) = Āi · e2πj·νi ·t · e−2πj·f ·τi · S̄(f ) (2.16)
∀i

Here Ḡ(t, f ) is called the time variant transfer function and determines the gain ex-
perienced at time t to a frequency component at frequency f . From this function Ḡ(t, f ), the
complex gain ḡ(t) in the presence of Doppler spread only is obtained by setting the frequency
component to zero, hence ḡ(t) = Ḡ(t, 0). Accordingly, the complex transfer function in the
presence of delay spread only is given by Ḡ(f ) = Ḡ(0, f ).

2.3.4 First Order Statistics

A received signal consists in general of a large number of signal copies which interfere at the
receiver’s antenna. The channel could be seen in principle as a deterministic channel, if all
reflection coefficients were known at each time instance. Due to the large number of reflection
paths this is not possible in practice. Therefore a statistical description is the only way to
characterize at least some properties of the channel.
Due to the high number of signal paths existing in a usual transmission environment,
the central limit theorem may be applied to the statistical behavior of the interfering signal
copies at the receiver. Therefore the complex gain ḡ(t) from Equation 2.10 can be modeled
as Gaussian random process in time if Doppler spread is present. If delay spread is present,
the complex transfer function Ḡ(f ) from Equation 2.13 can be modelled as Gaussian random
process in frequency. If both kinds of spread are present, the time variant transfer function
Ḡ(t, f ), the Fourier transform of the time variant impulse response of Equation 2.15, can be
modelled as Gaussian random process in both time and frequency.
Assuming the complex gain to be Gaussian (in the case of flat fading and the absence
of a line of sight component), the probability density of the complex gain ḡ(t) is given by
Equation 2.17 .
(|ḡ|) 2
1 −
σg 2
p(ḡ) = · e (2.17)
2πσg 2
The variance σg 2 is given by Equation 2.18 .
1
1
σḡ 2 = · E |ḡ(t)|2 = · E gr (t)2 + E ḡi (t)2




(2.18)
2 2
Copyright at Technical University
Berlin. All Rights reserved.
TKN-03-007 Page 27
TU Berlin

Note that the real and imaginary components are individually Gaussian processes. Also
the Gaussian process of the complex gain has a mean equal to zero. Figure 2.5 shows a plot
of the probability density function.

0.16
0.14
0.12
0.1
0.08
0.06
0.04
p(g) 0.02
0
4
2
−4 0
−2 −2
0 Im(g)
2 −4
4
Re(g)

Figure 2.5: Probability density function of the complex gain ḡ(t)

Changing from Cartesian coordinates to polar coordinates (ḡ = g r + ḡi = r · ejθ ) by

standard transformation yields the following joint probability density function in Equation
2.19.
2
r −r
2σ2 2
p (r, θ) = · e (2.19)
2π · σg 2
r and θ are independent, where θ has a uniform distribution, while the distribution of r
is given in Equation 2.20 and is called the Rayleigh distribution.
2
r −r
p (|ḡ(t)|) = p (r) = 2
· e 2σ2 2 (2.20)
σg
If such a type of fading is present with no path dominating among all received paths due
to the absence of a line of sight, the amplitude of the received signal will vary according to
a Rayleigh distribution. This form of fading is characterized by the absence of a line of sight
component, which is a very strong and ’fast’ path compared to all other paths. A plot of the
distribution is given in Figure 2.6.
For determining the actual SNR at the receiver the instantaneous power has to be obtained
rather than the instantaneous amplitude. This is given by the squared amplitude z = r 2 = |ḡ|2
and the distribution of z is given in Equation 2.21. In fact z has a χ 2 distribution with two
degrees of freedom due to the two independent jointly Gaussian processes of the real and
imaginary parts of the signal combining.
1 −z
p ḡ(t)2 = p (z) =


2σ2 2
· e (2.21)
2σg 2
A plot of the distribution is given in Figure 2.7.

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 28
TU Berlin

0.84

0.72

0.6

0.48

p(|g|) 0.36
0.24

0.12

0
0 1 2 3 4 5
|g|=r

Figure 2.6: Probability density function of the magnitude of the complex gain |ḡ(t)|, which
is Rayleigh distributed

0.6

0.5

0.4

0.3
p(r2)
0.2

0.1

0
0 2 4 6 8 10
2 2
|g| = r

Figure 2.7: Probability density function of the squared magnitude of the complex gain |ḡ(t)| 2 ,
which is χ2 - distributed

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 29
TU Berlin

From the cumulative distribution function and its asymptotic behavior (both not given
here, refer to [4], page 46) two rules of thumb can be derived. First, the probability that the
received instantaneous power will be 10 dB or more below the mean power level is 0.1. A
20 dB fade or more has the probability of 0.01 and so on. Second, the probability that the
instantaneous power drops below a given level decreases only inversely with increasing the
average power. Therefore doubling the power only reduces a level dropping probability by
half.
In the case of an existing line of sight component, the distribution of r is no longer
Rayleigh but Rice, since now one distinct path dominates, which means that it is received
much stronger (in terms of power) than the remaining paths. Accordingly, the instantaneous
power z resulting from the fading has a non central χ 2 distribution with two degrees of
freedom. The Rice distribution depends on the ratio between power of the strong path
and power of the remaining paths and can characterize therefore many different line-of-sight
scenarios, while the Rayleigh distribution (only) characterizes the situation, when no path
dominates, therefore a strict non-line-of-sight scenario. For a detailed derivation refer to [4].
For a small number of paths (< 5), the assumption of the Gaussian process as a result
of the central limit theorem does not fit adequately any more. In this case the amplitude
r of the received signal can be assumed to have a Nakagami distribution density. The
instantaneous power z of the signal has a Γ distribution. By varying a variable m the
Nakagami distribution can take the absence or presence of a line of sight into account. In
addition to this, the Nakagami distribution is more convenient for analytical work. For a
detailed derivation and discussion refer to [4].
As already mentioned does the Rayleigh distribution show up in most non-line-of-sight
settings, which are encountered mostly with indoor scenarios as well as with macrocells in
urban areas. Rice distributions are encountered on the opposite with urban microcells and
suburban macrocells (refer to [4]). However, by considering Rayleigh fading, one is working
with the worst possible scenario, since in settings where the Rice distribution applies, fading
is less destructive and the performance of communication systems is better.
The Nakagami distribution is an alternative distribution to the Rice distribution, it models
line-of-sight scenarios, where a certain dominating path exists. Note that both distributions,
the Nakagami as well as the Rice one, can be reduced to the Rayleigh distribution by setting
the ratio between power of the dominating path and remaining power to zero

2.3.5 Second Order Statistics

In the previous Section the complex gain ḡ(t) as well as the complex transfer function Ḡ(f )
were shown to be Gaussian random processes in time, respectively in frequency. In order to
describe a Gaussian random process it is sufficient to know its mean and its autocorrelation
function or the transform of it, the power spectrum. The mean has already been shown
implicitly in the previous Section. In this Section we desire to obtain the second order
description of the process in the case of Doppler spread as well as in the case of delay spread.
At first the second order statistics in the case of Doppler spread are derived, afterwards we
derive them for the case of delay spread.
In the presence of Doppler spread only, thus s(t − τ i ) ≈ s(t), the received signal ȳ(t) is
given by the product of the transmitted signal s̄(t) and the complex gain of the channel ḡ(t).

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 30
TU Berlin

ḡ(t) is time variant and is given by Equation 2.10. If the input signal is a tone at the carrier
frequency, therefore s̄(t) = 1 for example, the received signal consists of multiple tones at
frequencies in the vicinity of the carrier with a maximum shift of the Doppler frequency f d .
Hence the tone is spread.
We desire the power spectrum of such a reception. Considering a mobile receiver moving
with the speed v in a multipath environment as shown in Figure 2.4 each Doppler shift
frequency ν is given by Equation 2.22 .

ν = fd · cos (γ) (2.22)

In general ν varies from +fd , resulting from reflected paths in front of the receiver, to
−fd , resulting from reflected paths behind the receiver (behind and in front relate to the
direction of movement of the receiver). Since cos () is an even function, the Doppler shift
frequency ν might result from a scatterer at the left or the right side of the receiver, therefore
from a scattering angle of +γ or −γ. Differentiating Equation 2.22 yields us the relationship
between (small) ranges of ν and of the angle γ in Equation 2.23.
s  2
dν ν
q
2
= fd · sin (γ) = fd · 1 − cos (γ) = fd · 1 − (2.23)
dγ fd
Assume that the amount of scatterers is high, therefore the power received from differential
angle dγ is given by the product of power density P (γ) and differential angle dγ. With this
we can relate the received power Sḡ to the Doppler shift frequency ν in Equation 2.24 and
therefore receive the power spectrum S ḡ (ν).

P (γ) + P (−γ)
Sḡ (ν) = r (2.24)
  2
ν
fd · 1− fd

Now consider the special case of isotropic scattering, which implies that the power
(σḡ )2
received from different angles is equivalent (P (γ) = 2π ). In this special case Equation 2.24
turns into Equation 2.25.

(σḡ )2 1
Sḡ (ν) = ·r (2.25)
π · fd  2
1 − fνd

Equation 2.25 describes an U shaped spectrum, shown in Figure 2.8. It is often referred
to as Jakes power spectrum [8].
From the power spectrum, the autocorrelation function of the complex gain is easy to
derive by inverse Fourier transform, which yields Equation 2.26.

π
σḡ 2
Z
R fd
rḡ (τ ) = −fd Sḡ (ν) · e
2πj·ν·τ dν = · e2πj·fd ·cos (γ)·τ dγ
2π
−π x 
2
= σḡ 2 J0 (2πfd τ ) = σḡ J0 2π (2.26)
λ

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 31
TU Berlin

Power 2
spectrum
S(v)
1

0
−fD 0 fD

Doppler shift v

Figure 2.8: Power spectrum for the isotropic scattering case

Equation 2.26 relates the autocorrelation function depending on a time difference τ with
a space difference x. The fact that the autocorrelation function only depends on differences,
either in time or in space, result from the assumption that ḡ(t) is a wide sense stationary
Gaussian random process. For wide sense stationary processes Equation 2.27 is valid.

E [ḡ (t) ḡ ∗ (t − τ )] = rḡ (τ ) (2.27)

Under the assumption, that the scatterers at different Doppler shifts ν are uncorrelated,
ḡ (t) is in fact uncorrelated (shown by setting ḡ(t) from Equation 2.10 in the left side of
Equation 2.27). This assumption is called the wide sense stationary (WSS) assumption.
From the autocorrelation function of the complex gain in Equation 2.26 a measure in time
can be derived characterizing the channel encountered. This is called the coherence time
and indicates the time span the channel roughly stays constant. One mathematical definition
of the coherence time is given in Equation 2.28, which equals an autocorrelation value of 0.98
[4]. However, this definition is somewhat subjective and other definitions might be found in
[12, 13, 17].
1 1
Tc = =√ (2.28)
2π · νrms 2πfd
In Equation 2.28 νrms denotes the standard deviation of the Doppler power spectrum
fd
Sḡ (ν) given by Equation 2.24. For the isotropic scattering case the variance is ν rms = √ 2
.
In Table 2.9 typical values for the coherence time are shown. Note that these values only
depend on the carrier frequency used. Therefore if a wireless local area network according to
IEEE 802.11b is installed, the system can asssume a coherence time of 28.1 ms if all objects
within the relevant area do not move faster than 1 m s . However, if the system is installed

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 32
TU Berlin

Carrier Frequency Tc @ 1 ms Tc @ 10 m
s Tc @ 20 m
s Tc @ 100 m
s
1 Mhz 68.2 s 6.82 s 3.41 s 0.68 s
100 Mhz 0.68 s 68.2 ms 34.1 ms 6.82 ms
1 Ghz 68.2 ms 6.82 ms 3.41 ms 0.68 ms
2.4 Ghz 28.1 ms 2.81 ms 1.4 ms 0.28 ms
5.4 Ghz 12.5 ms 1.25 ms 0.62 ms 0.12 ms
10 Ghz 6.82 ms 0.68 ms 0.34 ms 68.2µs
60 Ghz 1.12 ms 0.11 ms 56.2µ s 11.2µ s

Table 2.9: Coherence time values for different carrier frequencies at speeds comparable to
walking (1 m m m m
s ), a slow car (10 s ), a fast car (20 s ) and traveling per train (ICE 100 s )

near a highway, the coherence time is significantly smaller. The required bandwidth has no
impact, as long as it is much smaller than the carrier frequency. If the required bandwidth is
not significantly smaller compared to the carrier frequency, the coherence time to be expected
equals the coherence time of the highest frequency involved in the communication scheme.
At next we are interested in the second order statistics stemming from the delay spread
only. As we have seen, it was possible to obtain a power density distribution in the fre-
quency domain, characterizing the Doppler spread in an isotropic environment. From this
the autocorrelation in the time domain was derived, indicating the second order statistical
behavior in time of the complex gain function ḡ(t). As it was already shown in Section 2.3.3
the Doppler spread gives a wireless channel a time selective behavior, however the severity of
this behavior depends on the time structure of the transmitted signal. In contrast does the
delay spread give a wireless channel a frequency selective behavior. Therefore we desire now
an autocorrelation function in frequency, which yields us the second order statistical behavior
of the complex transfer function Ḡ(f ) from Equation 2.13.
In the presence of delay spread only the wireless channel can be modelled as linear time
invariant filter. Thus the received signal ȳ(t) is given by the convolution of the transmitted
signal s̄(t) and the channel impulse response ḡ(t), given by Equation 2.12. Accordingly, the
received signal in the frequency domain is given by the product of the frequency domain
representation of the transmitted signal and the transfer function Ḡ(f ) of the channel, given
by Equation 2.13. We are interested in a spaced frequency correlation function of Ḡ(f ),
that is a function which gives us the correlation between the transfer function at different
frequencies. If two tones at different frequencies are transmitted, the desired correlation
function gives us the correlation between the received signals of the two tones. In general
this function is given by Equation 2.29.
1
· E Ḡ (f ) · Ḡ∗ (f − ∆f )
 
r̄Ḡ (f, f − ∆f ) = (2.29)
2
Substitution of Equation 2.13 into Equation 2.29 yields Equation 2.30.
" #
1 XX
∗ −2πj·f ·(τi −τk ) −2πj∆f ·τk
r̄Ḡ (f, f − ∆f ) = · E Āi Āk · e ·e (2.30)
2
∀i ∀k

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 33
TU Berlin

If scatterers at different delays are uncorrelated, this autocorrelation function depends

only on the frequency difference ∆f , due to the then existing relationship given by Equation
2.31. This assumption is called the uncorrelated scatterers (US) assumption of wireless
channels.
" #  
1 XX 1 X
E Āi Ā∗k = E  Āi Ā∗ki  = σi (2.31)
2 2
∀i ∀k ∀ki

Here ki indicates all scatterers with the same delay τ i . With this Equation 2.30 turns into
Equation 2.32.

(σk )2 · e−2πj∆f ·τk

X
r̄Ḡ (∆f ) = (2.32)
∀k

Now consider many scatterers to be present. In this case the summation in Equation 2.32
becomes a density depending on the delay τ . This density is called the power delay profile
P (τ ) and with this we obtain Equation 2.33.
Z ∞
r̄Ḡ (∆f ) = P (τ ) · e−2πj∆f ·τ · dτ (2.33)
0
For the power delay profile one idealized but often used function is the exponential profile
given by Equation 2.34.

(σḡ )2 − τ τ
P (τ ) = · e rms (2.34)
τrms
(τrms )2 is the delay variance (mean squared delay) given by Equation 2.35. The square
root of this, the standard deviation of the delay, is an often used measure for the delay spread
of a propagation environment (also used in Table 2.7).
Z ∞
1
2
τrms = 2 · (τ − τm )2 · P (τ ) dτ (2.35)
σḡ 0
τm denotes the mean delay and is given by Equation 2.36.
Z ∞
1
τm = 2 · τ · P (τ ) dτ (2.36)
σḡ 0
For the exponential power delay profile τ rms equals τm .
From the autocorrelation function of the transfer function of the wireless channel with
uncorrelated scatterers a measure in frequency can be derived characterizing the channel
encountered. The meaning of this measure is related to the coherence time and is called the
coherence bandwidth. The coherence bandwidth measures roughly the frequency spacing
for which the channel does not change significantly. Again the exact mathematical definition
is to some extend subjective. One definition of the coherence bandwidth is given by Equation
2.37, following [4]. Other definition might be found in [12, 13, 17].
1
Wc = (2.37)
2πτrms

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 34
TU Berlin

Environment Wc
Urban 6.4 kHz - 160 kHz
Suburban 80 kHz - 800 kHz
Indoor 0.64 MHz - 6.4 MHz

Table 2.10: Coherence bandwidth ranges for three typical environments

In Table 2.10 ranges of the coherence bandwidth are given for different environments.
Note that the carrier frequency has no influence on the coherence bandwidth, therefore if one
happens to be in an indoor environment, one will always encounter a coherence bandwidth
of some MHz, no matter if for example a wireless local area network working at 2.4 GHz is
used or at a carrier frequency of 5.4 GHz. If the coherence bandwidth is much smaller than
the required bandwidth for transmission, the system will suffer from ISI, no matter at which
carrier frequency the system is working. Therefore, as the Table highlights, it is much easier
to communicate at high data rates in indoor scenarios due to the large coherence bandwidth
than in urban environments, here ISI degradates the performance severly.
Up to now, we have only investigated the second order statistics in the presence of one
phenomenon. In the case of the time selective behavior of the complex gain function ḡ(t)
caused by Doppler spread we obtained a gain autocorrelation function in time and as Fourier
transform the Doppler power spectrum in the frequency domain. In the case of the frequency
selective behavior of the transfer function Ḡ(f ) caused by delay spread we obtained a gain
autocorrelation function in the frequency domain and as Fourier transform the power delay
profile in the time domain.
At next we focus on the presence of both effects at the same time and will obtain some
results regarding the correlation behavior in both domains.
First let us recall the input output relationship in the case of many scatterers given by
Equation 2.9.
Z ∞ Z fd
ȳ(t) = %̄ (ν, τ ) · e2πj·ν·t · s̄ (t − τ ) · dνdτ
0 −fd

%̄ (ν, τ ) is the delay-Doppler spread function from Equation 2.8. Observing the output in
the case of a single carrier input at frequency f , where s(t) = e 2πj·f ·t , we obtain the output
given in Equation 2.38.

Z ∞ Z fd
ȳ(t) = %̄ (ν, τ ) · e2πj·ν·t · ē2πj·f ·t · e−2πj·f ·τ · dνdτ = e2πj·f ·t · Ḡ (t, f ) (2.38)
0 −fd

Here Ḡ(t, f ) is the time variant transfer function and determines the complex gain expe-
rienced at frequency f at time t. It was already introduced in Equation 2.16. Since both,
delay and Doppler spread are now present, we desire a time-frequency correlation function of
Ḡ(t, f ). That is, a function representing the correlation between the complex gain at time t
and at frequency f , compared to the complex gain at time t + ∆t and at frequency f + ∆f .

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 35
TU Berlin

Considering the WSS assumption (scatterers at different Doppler shifts are uncorrelated)
and the US assumption (scatterers at different delays are uncorrelated), this desired function
becomes the one in Equation 2.39, only depending on the delay and frequency differences
present.
Z ∞Z ∞
rḠ (∆t, ∆f ) = S%̄ (ν, τ ) · e2πj·ν·∆t · e2πj·∆f ·τ dνdτ (2.39)
−∞ −∞

In this equation S%̄ (ν, τ ) is called the delay-Doppler power density function or also
the scattering function. It represents the power density of the environment at Doppler
shift ν and delay τ . This function is related to the delay power profile introduced in Equation
2.33 by Equation 2.40.
Z ∞
P (τ ) = S%̄ (ν, τ ) dν (2.40)
−∞
It is related to the Doppler spectrum in a similar way by integrating the scattering function
in the delay domain.
In order to determine the functional behavior of the scattering function let us assume
that the Doppler spectrum is not linked to the delay profile. Then the scattering function is
called to be separable. Therefore the scattering function is given by Equation 2.41.

Sḡ (ν) · P (τ )
S%̄ (ν, τ ) = (2.41)
σḡ2
With this assumption the autocorrelation function from Equation 2.39 becomes simpler
and is given in Equation 2.42.

rḡ (∆t) · rḠ (∆f )

rḠ (∆t, ∆f ) = (2.42)
σḡ2
A plot of the scattering function is given in Figure 2.9, where the power spectrum is Jakes
like and the power delay profile has an exponential behavior.

2.3.6 Fading Rate and Duration

In Section 2.3.4 a rule of thumb was given, indicating the probability of a fade on a wireless
channel by 10 dB, 20 dB and so on. As we have seen from the discussion of the complex
gain of a time invariant channel, at some instances the channels attenuation might exceed
a certain threshold. Clearly, this depends on the time variant behavior of the channel and
therefore ultimately on the velocity assumed to be present in the propagation environment.
Given a certain attenuation threshold X it is of great interest for system design to know
the average rate with which the complex gains amplitude r(t) = |ḡ(t)| will cross this rate.
This rate is called the fading rate. In addition it is of interest to know the average duration
the complex gains amplitude will be below the threshold X, which is called the fading
duration.
Here we will derive both averages and will start with the fading rate. For the derivation
consider first the sketch of a level crossing given in Figure 2.10. Here the complex gains
amplitude is a time ti below the given threshold, a time ti + dt it is above the threshold.

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 36
TU Berlin

S(f,t)

−fD 0 t
fD 0
f

Figure 2.9: Scattering function in the case of seperability

r(t)

r(t)i + r’ dt
X
r(t)i

ti ti + dt t
Figure 2.10: Illustration of a level crossing at a threshold of X

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 37
TU Berlin

A level crossing will take place, if the differential increase of the complex gains amplitude
during the duration dt is larger or equal to X − r(t), therefore if X − r(t) ≤ dr = r 0 · dt which
equals X − r 0 · dt ≤ r(t) < X (we only focus on upward crossings, since these will yield the
desired overall crossing rate, therefore assume in general here r 0 > 0. The probability of such
an increase of r(t) at time t during the interval dt at a slope r 0 with a slope increase of dr 0 is
given by Equation 2.43.
Z X
0
pr,r0 r, r 0 dr


dr (2.43)
X−r 0 ·dt
With this we easily obtain the overall probability of an upward crossing in Equation 2.44.
Z ∞Z X Z ∞
0 0
r 0 pX,r0 r, r 0 dr 0



Pup = pr,r0 r, r drdr = dt · (2.44)
0 X−r 0 ·dt 0
The second integrand in this equation is the expected number of upward level crossings
in any dt, thus this yields the fading rate as explicitly shown in Equation 2.45.
Z ∞
r 0 pX,r0 r, r 0 dr 0


Rf (X) = (2.45)
0
For Rayleigh fading channels, the joint probability of the complex gains amplitude and
derivative of it is given by Equation 2.46.
!
2
r2 r0
r − 12 · 2 + σ2
σg
pr,r0 r, r 0 = √


e g0 (2.46)
2π · σg · σg0
Substitution of Equation 2.46 into Equation 2.45 yields, under the assumption of isotropic
scatterers, Equation 2.47 (for more refer to [4]).
2
√ X −√ X
2
2σg
Rf (X) = 2π · fd · q ·e (2.47)
2σg2

In Table 2.11 we give some values of the fading rates for a 10 dB fade for various carrier
frequencies and different velocities of the receiver.
From Equation 2.45 it is easy to derive the fading duration. Consider the relationship
given by Equation 2.48.

P (r ≤ X) = Rf (X) · Tf (X) (2.48)

Once the fading rate is given, the fading duration denoted by T f (X) can be derived
by considering the overall probability that the complex gains amplitude is below the given
threshold. For a Rayleigh fading channel we therefore obtain together with the CDF of the
complex gains amplitude (not given in this report, refer to [4]) Equation 2.49 for the fading
duration. Note that this again assumes isotropic scattering.

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 38
TU Berlin

Carrier Frequency Rf (0.1) @ 1 m

s Rf (0.1) @ 10 m
s Rf (0.1) @ 20 m
s Rf (0.1) @ 100 m
s
1 Mhz 0.0005 Hz 0.0054 Hz 0.0119 Hz 0.0594 Hz
100 Mhz 0.059 Hz 0.594 Hz 1.19 Hz 5.94 Hz
1 Ghz 0.59 Hz 5.94 Hz 11.9 Hz 59.4 Hz
2.4 Ghz 1.41 Hz 14.1 Hz 28.2 Hz 141 Hz
5.4 Ghz 3.17 Hz 31.6 Hz 63.3 Hz 317 Hz
10 Ghz 5.94 Hz 59.4 Hz 119 Hz 594 Hz
60 Ghz 35.2 Hz 352 Hz 704 Hz 3.52 kHz

Table 2.11: Fading rates for a fading threshold of 10 dB equaling X = 0.1 for different carrier
frequencies at velocities comparable to walking (1 m m m
s ), a slow car (10 s ), a fast car (20 s ) and
m
traveling per train (ICE 100 s )

Carrier Frequency Tf (0.1) @ 1 m

s Tf (0.1) @ 10 m
s Tf (0.1) @ 20 m
s Tf (0.1) @ 100 ms
1 Mhz 8.4 s 800 ms 400 ms 80 ms
100 Mhz 84 ms 8 ms 4 ms 800µ s
1 Ghz 8.4 ms 0.8 ms 0.4 ms 80µ s
2.4 Ghz 3.5 ms 0.3 ms 0.17 ms 30µ s
5.4 Ghz 1.6 ms 0.16 ms 80µ s 16µ s
10 Ghz 0.84 ms 84µ s 42µ s 8.4µ s
60 Ghz 0.142 ms 14.2µ s 7.6µ s 1.42µ s

Table 2.12: Fading duration for a fading threshold of 10 dB equaling X = 0.1 for different
carrier frequencies at velocities comparable to walking (1 m m
s ), a slow car (10 s ), a fast car
m m
(20 s ) and traveling per train (ICE 100 s )

2 2
− √X − √X
2
2σg 2
2σg
1−e 1−e
Tf (X) = 2 = (2.49)
√ −√ X
Rf (X)
2
2π · fd · √X ·e 2σg
2σg2

In Table 2.12 we give the corresponding fading durations for a threshold of X = 0.1
From the asymptotic behavior of the fading rate as well as the fading duration with
respect to a varying fading threshold, one further rule of thumb can be derived. If the fading
threshold is reduced by a factor of 4, which equals a decrease of 6 dB, then the fades are on the
average half as long while they occur half as frequent. Therefore increasing the transmission
power by a factor of 4 yields exactly this result on the fading rate and duration.

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 39
TU Berlin

Chapter 3

Modulation Channel

While the effects of the radio channel have an attenuating impact on the relationship between
transmitted and received signal, the modulation channel has an additive impact on this
relationship. Two major sources of effects are modelled in general. The first one is noise.
Noise is always of stochastic nature and varies with time. It is denoted by n(t).
The second effect corrupting the received signal in an additive manner is interference.
Interference is caused by other RF transmitting electronic devices. As it is with the noise,
interference has a stochastic nature and varies with time. It is denoted by j(t). Interference
is due to other systems operating in the same frequency band in the case of unlicensed bands,
or to co- and adjacent-channel interference in licensed bands. Co-channel interference
happens due to frequency re-utilisation in a cellular environment and is unavoidable. Adja-
cent channel interference is due to realistic filters letting some power be transmitted in
the sidebands. Interference is always the results of other wireless systems operating in the
same or nearby frequencies.
The quality of demodulation and decoding of a signal depends on the difference between
the power of the received signal and the power of other signals with power in the same
frequency band — interference — added to the power of the noise.

3.1 Noise
Noise is always present and comes from several sources, e. g. atmospheric disturbances, elec-
tronic circuitry, human-made machinery. The first two belong to the group of thermal noise
sources described in 3.1.1. Noise produced by human-made sources is described in more detail
in Section 3.1.3.

3.1.1 Thermal Noise

Thermal noise is due to the movement of charged particles inside electronic components
existent in every receiver system and is therefore unavoidable. The characteristics of thermal
noise were firstly studied for a resistor. Since many other noise sources besides resistors are
Gaussian and have a spectrum which is constant over a wide frequency range, the calculation
of the noise power produced by a thermal noise source is shown here on the example of a
resistor.

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 40
TU Berlin

The voltage at the terminals of a resistor is Gaussian-distributed and with variance pro-
portional to the temperature at which the resistor operates. A simplified expression for the
spectral density of thermal noise from a resistor of resistance R is given by
 
h · |f | kT
Gvn (f ) ≈ 2 · k · T 1 − |f |  ,
2·h·T h

where f is the frequency, h = 6.62 · 10 −34 J · s the Planck constant, k = 1.37 · 10 −23 J/deg
the Boltzman constant and T the absolute temperature ([3]). This expression can be further
h·|f |
simplified since for usual working temperatures, 2·h·T  1 and the spectrum of the noise can
be considered constant over the frequencies used: G v (f ) ≈ 2·k·T . Take for example a working
temperature of 0o , and the condition for the approximation becomes f  22 · 10 −12 , and the
VHF and UHF frequency bands are way within the limits for which the approximation is
valid.
The power of thermal noise can be obtained using the maximum power transfer theorem:

Gn = 1/2 · k · T [W/Hz] (3.1)

Notice that the power delivered by a resistor at temperature T is independent of the value
of the resistance. Usually a temperature of T 0 = 290o is used as a reference for measurements
and calculations.
This kind of noise is called white because it contains all frequencies, in analogy to white
light, which contains all light frequencies. White noise is uncorrelated.
Thermal noise sources have a power spectral density which depends on the working tem-
perature. There are also noise sources whose produced noise does not depend on the tem-
perature. Nevertheless, an effective noise temperature is defined for every gaussian noise
source, e. g. amplifiers, and is used in equation (3.1) to calculated the noise power introduced
by the component on the system (see Appendix A of [3] for more details). This noise is always
referred to the input of the component, so that it will also be filtered, as explained in the
next section.

3.1.2 Filtered White Gaussian Noise

In practical systems of limited bandwidth, the noise is also filtered and is at the output
no longer white, taking the shape of the filter’s transfer function and no longer having all
frequencies in the same proportion. This means that the noise becomes correlated when it is
low-pass filtered. If an ideal low-pass filter of bandwidth B is assumed, the autocorrelation
function of the coloured noise will be R n (τ ) = 2 · Gn · sinc (2 · B · τ ), where Gn is calculated
as stated at the end of last section.
The noise equivalent bandwidth of a filter B n is defined as the bandwidth of the ideal
bandpass filter which would let the same gaussian white noise power through as the filter
in question1 . The average noise power at the output of the filter can then be expressed as
N = g · 2 · Gn · Bn , where g is the power gain of the filter at the centre frequency. If the noise
has zero mean value, then σn = N .
1
Note that every electronic component is, in this context, a filter.

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 41
TU Berlin

The effective noise temperature and noise equivalent bandwidth can be found in the
datasheets of electronic components. In mobile communication systems this values are usually
not relevant since the limitations to system performance are due to interference and man-
made noise (see Section 3.2), which takes values several magnitudes higher.

3.1.3 Man-made Noise

Man-made noise is noise produced by machines operated by humans. It is noise because it
is energy radiated by machines which are not supposed to generate that energy. Their main
purpose is some other function, but due to turning on and off of electrical and electronic
components or to wiring, they radiate electromagnetic energy which can disturb wireless
communications nearby.
The characterisation of man-made noise is very complicated, since many parameters have
to be measured and they are always specific to the measurement environment. The parame-
ters to be measured include average total power, power spectrum, probability distribution of
the noise voltage, the pulse heights, widths and rates, as well as system specific parameters
like dependance on antenna polarisation, height and directivity, and long-time dependance on
time and location. In [5] three methods for measuring man-made noise are explained. Also
some measurement results are presented, which lead to the conclusion that for frequencies
above 4 GHz the power of man-made noise is negligible compared to the noise produced by
a typical receiver. Nevertheless, these results were obtained in the 70’s, and it must be taken
in consideration that due to the technological advances in the last decades typical receivers
nowadays probably have much lower noise figures than 30 years ago.

3.1.4 Some Results

Quite few results exist about man-made noise, but some are presented in the following para-
graphs.
In [2] some results of measurements in indoor environment are documented, which were
made with the objective of developing empirical impulsive noise models. The measurements
were made with a 40 MHz bandwidth receiver at 918 MHz, 2.44 GHz and 4 GHz in different
indoor environments: a department store, a grocery store, soft-partitioned office buildings
and a hard-partitioned office building. It could be concluded that the main noise sources
of impulsive noise are microwave ovens, photocopiers, printers, elevator door switched and
ignition systems. Measurement results show that the noise from a microwave oven (1st order
statistics) can be modelled by a 2.45 GHz carrier modulated by a pulse train at the AC power
frequency. Some conclusions can be taken from the measurement results:

• amplitude of power of noise pulses is greater in the 918 MHz band than in the others
(see Table 3.1);

• amplitudes of the power of the impulses are not gaussian distributed;

• noise pulse durations were similar in all the measured frequency bands and the average
values lie between 120 ns and 150 ns;

• for statistics on the rate of the pulses, refer to [2].

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 42
TU Berlin

Frequency Average Amplitude σ 1% Probability 0.001% Probability

914 MHz 18.87 6.86 40 63
2.44 GHz 16.14 6.99 29 45
4 GHz 11.75 5.87 23 43

Table 3.1: Statistics of the peak power of the noise pulses measured at the 2 first locations;
amplitude power was measured above thermal noise k · T 0 . All values in dB. (source [2]).

• data relative to the impulsive noise generated by the photocopier, the microwave oven
and the elevator door switched can also be found in [2].

In [1] some are presented for man-made noise due mainly to cars (ignition and motors).
The distribution of the power is also not gaussian, but follows a combination of a gaussian
with a Weibull distribution.
Most man-made noise measurement results available are 3 decades old, as well as the
models derived from them. Measurements reported in [6] indicate that those results are
no longer valid, due to the technological advances on the one hand, which have improved
the isolation of noise sources, and to the changes in the way of life on the other, where a
personal computer, a television and cordless phones belong to almost every home. The new
environments have new noise sources, whose characteristics need to be measured to develop
new models. Nevertheless, the characteristics presented here provide a more realistic scenario
than simply assuming uncorrelated or non-existing man-made noise.

3.2 Interference
In general the source of noise is a source which primarily does not intend to produce electro-
magnetic disturbance patterns, for example microwaves ovens or other electrical or electronic
equipment. Another source of noise is given by the thermal effects existing for example in
any electric circuit as in amplifiers.
Beside these sources of signal distortion, other communication system might be active in
the environment. Such sources, which have the primary goal to produce electromagnetic radi-
ation for communication purposes are not represented by noise, instead they are represented
by interference. Like noise, interference has a additive distorting impact on the signal. For
example, interference occurs in cellular systems, due to the fact that bandwidth is limited and
system operaters have to reuse certain spectra of the overall bandwidth. Frequency planning
is one method to control interference in cellular systems. In unlicensed bands, interference
may stem from local wireless networks, which just happen to be deployed quite close to each
other. In general there are different kinds of interference with a different impact on the
received signal.
First their is co-channel interference [4]. Co-channel interference occurs if two trans-
mission devices operating within the same radio frequency band are active and a receiver,
originally trying to receive the signal from one transmitter also receives a (weak) signal from
the second transmitter. In cellular systems co-channel interference is an important factor
limiting the systems performance, more important than noise (see [4], page 8). If more than
two or three interference sources are active, interference may be modelled as white Gaussian

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 43
TU Berlin

process, adopting many of the characteristics discussed in the previous Section about noise.
The average interference power is then derived from the path loss equations of Section 2.1.
In unlicensed bands co-channel interference is also an important issue to take into con-
sideration. Here, the same assumption may be maid as with a cellular system: if multiple
interference sources exist, the interference can be modelled as white Gaussian noise, which
changes the consideration of interference much easier. The interference power level can be
determined by path loss equations then.
Beside co-channel interference their is also the possibility that transmissions conveyed on
a different but close frequency band produce a significant interference power in a receiver.
This is mainly due to imperfect pass band filters and is called adjacent channel interference
([13]). Adjacent channel interference is encountered in cellular systems as well as in unlicensed
bands.
In the following we give a rough, example analysis for the severity of co-channel interfer-
ence in cellular systems. Note that in principle the same derivation is valid for ’ any situation
where multiple transmitters operate within the same frequency band, just separated by a
certain distance from each other and from the receiver, trying to detect the signal originally
transmitted to him (as it is with transmissions in unlicensed bands).
Consider a cellular system where the cells radius is denoted by R and the separation of
cells operating at the same frequency band is denoted by D. We consider a receiver which
is located at the fringe of a cell, therefore the distance between the desired base station
(transmitter) and it (receiver) is equal to R. If we denote by P 0 the overall used transmission
power of each base station, then following from the path loss equation the power level of the
desired signal is given by Equation 3.2.

P0
Py = (3.2)
Rk
k depends on the environment of this scenario, values where given in Section 2.1. Fol-
lowing Equation 3.2 the interfering power level received from each base station (interfering
transmitters) of the neighboring cells is given by Equation 3.3.

P0
Pj = (3.3)
(D − R)k
Considering i equally strong interfering transmitters, each one at one of the neighboring
cells (which is a pessimistic assumption) in this scenario (note that for a hexagonal cell layout,
i = 6), the ratio between desired signal power and interfering power level is given by Equation
3.4.
 k
Py 1 D−R
CIR = = · (3.4)
i · Pi i R
One important fact about co-channel interference in a cellular environment is that increas-
ing the transmission power at each base station (so at each transmitter) provides no system
improvement regarding the interference power level, since Equation 3.4 is independent of the
transmitted power. For a further discussion on this refer to [4].

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 44
TU Berlin

Chapter 4

Digital Channel

According to the mathematical model shown in Figure 1.2 in Chapter 1 the transmitted
signal is influenced by three effects, which have been discussed in Chapter 2 and 3. For a
digital communication system the question arises, how these signal distortions translate into
noticeable service degradations of running applications. Common measures, indicating al-
ready the impact on higher communication layers, are for example the bit error rate (BER).
It depends on the power ratio between received signal power and disturbing noise and in-
terference power, as well as on the modulation used. Recalling the mathematical model of
Chapter 1, the received signal is given by Equation 4.1.

y(t) = a(t) · s(t) + n(t) + j(t) (4.1)

A very common measure in order to characterize the instantaneous power ratio is the
Signal-to-Noise-and-Interference Ratio (SNIR) , usually measured in dB. The SNIR
is easily obtained from Equation 4.1 and is given in Equation 4.2.

a2 (t) · s2 (t)
SNIR(t) = (4.2)
2 · (n2 (t) + j 2 (t))
In this Chapter the relationship between SNIR and SER is shown for an example modu-
lation type. Note that there exist many different modulation types for which the derivations
might significantly change. However, in principle the performance of all modulation types
depends on the actual SNIR and therefore an example derivation will give a good insight to
the nonlinear relationship between performance metrics of the modulation channel (SNIR)
and performance metrics of the digital channel (SER or BER).

4.1 Structure of the Digital Channel

The received analog signal, which is subject to the phenomena described in the previous
Chapters, is a sequence of analog pulses and has to be converted back into the digital bit
sequence which originated it; this process is called demodulation. The demodulation can be
separated into two phases: the demodulation itself, and the detection.

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 45
TU Berlin

Demodulator The digitally modulated signal can be seen as a linear combination of func-
tions which form an orthonormal base of a vector space. The function of the demodu-
lator during each symbol period is to decompose the received waveform into a vector in
that space. This vector contains the value of the projections of the received waveform
into the functions in the orthonormal base of the vector space. This can be achieved
either by a matched filter or a correlator [12], depending on the type of signal being de-
modulated. The vector is of dimension N , the same as the dimension of the transmitted
signal.

Detector The detector’s function is to compare the vector output from the demodulator
with the values of the M possible waveforms. The distances in the vector space are
compared, and the ”closest” possible waveform is transmitted. The detection is made
such that the probability of making a wrong decision is minimised.

The received signal is, thus, converted into a sequence of symbols (demodulation) and
then into a sequence of bits (detection). This sequence of bits should be the same as the
transmitted sequence of bits. This is not always the case, since errors occur. This error
process can be described by an indicator sequence — a stream of ’0’ and ’1’ —, where a ’1’
represents wrongly received bits and a ’0’ correctly received ones. This conversion is possible
provided that the amount of noise and interference at the receiver is known. In the following
considerations the interference will not be taken into account, and SNR will be used instead
of SNIR.
The steps for the conversion of the analog in the digital channel are illustrated in Figure
4.1. They consist of taking, for every bit time, the value of the received signal, dividing it
by the sum of noise and interference in the same time period, and substituting them into
the expression for the used modulation. The resulting bit error probability is then used to
randomly generate the fate of a bit. The digital channel can thus be derived from the analog
channel, by calculating with which probability a pulse received with a certain power can be
correctly demodulated. Notice that, for a given analog channel, different digital channels can
be generated, depending on the modulation used (second step in Figure 4.1).
The calculation of the probability of correctly demodulating a symbol is shown using
the examples of binary pulse amplitude modulation (PAM) over AWGN channel and over a
Rayleigh fading channel in Sections 4.2 and 4.3, respectively. In Section 4.4 results for most
commonly used modulations are presented in a table (they can be obtained following the
same steps as for BPSK [12]).

4.2 Calculation of the Bit Error Probability as a function of

SNIR for Binary PAM over an AWGN Channel
The following calculation follows the one in [12]. Consider a binary bipolar PAM signal,
where the waveforms used to transmit information are f 1 = g(t) and f2 = −g(t), where g(t)
is the pulse used for the transmission, with energy E b = a2 g 2 (t) and g(t) 6= 0 in the interval
0 ≤ t ≤ T , where T is the duration of a bit. Since PAM signals are unidimensional (only the
amplitude varies for different waveforms, so there is only one base function), the output of
the demodulator will be a single value: the energy of the received pulse (signal plus noise)

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 46
TU Berlin

Figure 4.1: Steps for the conversion of an analog into a digital channel.

during the bit time. Notice that the binary bipolar PAM modulation under consideration
here is similar to the binary PSK modulation.
Then, assuming√f1 has been transmitted over an AWGN channel, the output of the
demodulator is r = Eb +n, where n is a Gaussian random variable with variance σ n2 = N0 /2
(see Section 3.1.4 for the calculation of N ). The detector, in this case, makes a decision based
on the comparison of r with the threshold 0; the probability of wrongly detecting f 1 can be
calculated by calculating the probability that r will have a value lower than 0 (in which case
the detector would choose f2 and not f1 ).
√
1 −
r− Eb
p(r|f1 ) = √ e N0
πN0
Z 0
P (r ≤ 0|f1 ) = p(r|f1 )dr
−∞
√
0 2Eb
1 r−
Z
−
=√ e N0
dr
2π −∞
Z √
− 2Eb /N
1 2
=√ e−x dx
2π −∞
r !
Eb
=Q ,
N
R∞ 2 /2
where Q(x) = √1 e−t dt.
2π x
√
If f2 has been transmitted, the vector r would have the value r = − Eb +n, and the error
probability P (e|f 2) = P (r ≥ 0|f2 ) will have the same value as P (e|f 1). Since f 1 and f2 are
equiprobable, the probability that the received waveform will produce a wrong demodulated
bit is
1 1
Pe = P (e|f1 ) + P (e|f2 )
2 !2
r
2Eb
=Q .
N0

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 47
TU Berlin

4.3 Calculation of the Bit Error Probability as a function of

SNIR for BPSK over a Rayleigh Fading Channel
In the previous section is was assumed that the channel over which the transmission occurred
was AWGN, so that the noise was modelled as a stochastic process and the signal power as
a constant value. It happens often, though, that also the received signal power varies and
can only be characterised as a stochastic process. To obtain the bit error probability for
the communication over a channel where the received signal level is described as a random
stochastic process, the probability of error over the AWGN channel has to be integrated over
all the possible values of the signal, that is
Z ∞
Pe |Rayleigh = Pe |AWGN (γb )p(γb )dγb , γb = Eb /N0 .
0

This is the case, for example, of a Rayleigh fading channel, described previously in Section
2.3. The energy of a pulse is in this case no longer constant, but is described by a stochastic
process Eb = E[aF A ]2 ] Eb |AW GN . For a Rayleigh fading channel, aF A is Rayleigh distributed,
and a2F A therefore chi-square distributed, with two degrees of freedom.

1 −γb /γb
p(γb ) = e
γb
Eb
γb = E(α2 )
N0
 r 
1 γb
Pe |Rayleigh = 1−
2 1 + γb
The bit error probability for the communication over a Rayleigh fading channel is much
higher than over an AWGN channel.
For other statistical description of the fading process, the bit error probability can be
obtained by substituting the chi-square distribution by the corresponding one and integrating.

4.4 Results for Other Digital Modulation Schemes over an

AWGN Channel
Expressions for the bit error probabilities of several binary modulations are given in Table
4.1.
Table 4.2 gives expressions for the calculation of the symbol error probabilities of multi-
level modulations. Since a modulation with k = log 2 M levels transmits M bits with a single
pulse, blocks of M bits are jointly demodulated. The expressions are nevertheless in function
of the energy to noise per bit relationship E b which is calculated as Eb = ES /k, where ES is
the energy per symbol pulse.

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 48
TU Berlin

Modulation Coherent Demodulation Non-coherent Demodulation

q 
Eb
BPSK Q —
qN0
Eb 1 −Eb /N0
DBPSK 2 Q( N 0
) 2e
Eb
1 − 2N0
FSK — 2e

Table 4.1: Bit error probabilities in dependance of the per-bit energy-to-noise ratio for dif-
ferent binary digital modulations.

Modulation Symbol
q Error   Probability q P M  Pb = f (PM )
2Eb 1 2Eb 1
QPSK 2Q N 1 − 2Q N k PM
q   
Eb
DQPSK ≈ 4Q N 2 Q 0.59 ENb
q 
Eb π
M-PSK ≈ 2Q 2 log 2 M N sin ≈ k1 PM
  0 hM i
PM −1 n+1 M −1
1 nkEb 2k−1
M-FSK n=1 (−1) n n+1 exp − P
2k −1 M
q (n+1)2N
2(M −1) (6 log 2 M )Eb avg 1
M-PAM M Q (M 2 −1)N0
—
   q 2
3 Eb av
M-QAM 1 − 1 − 2 1 − √1m Q M −1 2N
2 —

Table 4.2: Symbol error probabilities in dependance of the per-bit energy-to-noise ratio for
different M-ary digital modulations.

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 49
TU Berlin

Chapter 5

Conclusions

In this report analytical models were discussed, explaining the behavior of the wireless channel
as it is observed in reality. The wireless channel is known for its unreliable and stochastic
nature. Still, for many application scenarios it is the first choise of medium, due to the lack
of alternatives.
The behavior of the wireless channel may be divided into two parts: attenuating fac-
tors and additive factors. Both contribute to the stochastic behavior of the channel. The
attenuating factors, residing in the radio channel, can be analytically be decomposed into
three components. One of them is deterministic and depends solely on the distance between
transmitter and receiver. The other two are stochastic, but might be modeled according to
their primary and secondary statistics. They depend on the propagation environment as well
as on the time scale and sampling rate with which the wireless channel is observed.
The additive factors, residing in the modulation channel, may be modeled by two compo-
nents, which are both stochastic. The first one, noise, is a omnipresent effect and can not be
exluded from system analysis. It is not charateristic for wireless channels, but is a limiting
factor in any communication system. The second source, interference, depends again on the
considered scenario. Since interference is caused by devices transmitting radio frequency sig-
nals, it is important to clarify first, if these signals are transmitted within the same frequency
bands or not and further how far such disturbing sources are away from the actual receiver
considered. There exist multiple scenarios, where interference has no significant impact.
The analysis in this report provides a framework in order to classify different, considered
transmission scenarios and determine in advance, which effects might be of interest for the
specific scenario and which one not. Thus, it builds a fundamental block for simulating a
certain scenario. However, turning these analytical models into running simulations requires
a further step. This will be treated in a next technical report, which will use the results of
this one in order to guide the reader to implementable simulation models of wireless channels.

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 50
TU Berlin

Bibliography

[1] D. Apostolakis and P. Constantinou. Man made noise measurements and modelling. In
2nd International Conference on Universal Personal Communications, Personal Com-
munications: Gateway to the 21st Century, volume 2, pages 585–589, 1993.

[2] K.L Blackard, T.S. Rappaport, and C. W. Bostian. Measurements and models of radio
frequency impulsive noise for indoor wireless communications. IEEE Journal on Selected
Areas in Communications, 11(7):991–1001, Sep 1993.

[3] A. Bruce Carlson. Communication Systems. McGraw-Hill Inernational Editions, 3rd

edition, 1986.

[4] J. K. Cavers. Mobile Channel Characteristics. Kluver Academic Publishers, 2000.

[5] W. C.Jakes. Microwave Mobile Communications, chapter 4, pages 295–298. IEEE Press,
Wiley Interscience, 1994.

[6] R. Dalke, R. Achatz, Y. Lo, P. Papazian, and G. Hufford. Measurement and analysis
of man-made noise in vhf and uhf bands. In Proceedings of Wireless Communications
Conference, pages 229–233, 1997.

[7] M. Gudmundson. Correlation model for shadow fading in mobile radio systems. In
Electronics Letters, volume 27, pages 2145–2146. IEEE, November 1991.

[8] W. C. Jakes. Microwave Mobile Communications. IEEE Press, Wiley Interscience, 1994.

[9] W.C.Y. Lee. Mobile Cellular Telecommunications. McGraw-Hill International Editions,

1995.

[10] A. Neskovic, N. Neskovic, and G. Paunovic. Modern approaches in modeling of mobile

radio systems propagation environment. IEEE Communications Surveys and Tutorials,
Third Quarter 2000.

[11] Y. Oda, R. Tsuchihashi, K. Tsunekawa, and M. Hata. Measured path loss and multipath
propagation characteristics in uhf and microwave frequency bands for urban mobile
communications. In IEEE, editor, Proceedings of the 53rd IEEE Vehicular Technology
Conference — VTC 2001, volume 1, pages 337–341, 2001.

[12] J. G. Proakis. Digital Communications. McGraw-Hill International, 4 edition, 2001.

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 51
TU Berlin

[13] Theodore S. Rappaport. Wirelsess Communications. Prentice Hall, 1999.

[14] S. Y. Seidel and T. S. Rappaport. 900 mhz path loss measurements and prediction
techniques for in-building communication system design. In Proceedings of the 41st
IEEE Vehicular Technology Conference, pages 613–618, 1991.

[15] S. Y. Seidel, T. S. Rappaport, and R. Singh. Path loss and multipath delay statis-
tics in four european cities for 900 mhz cellular and microcellular communications. In
Electronics Letters, volume 26, pages 1713–1715. IEEE, Sep 1990.

[16] G.L. Siqueira, G.L. Ramos, and R.D. Vieira. Propagation measurements of a 3.5 ghz
signal: Path-loss and variability studies. In Proceedings of the SBMO/IEEE MTT-S
International Microwave and Optoelectronics Conference — IMOC 2001, pages 209–212,
2001.

[17] R. Steele. Mobile Radio Communications. Pentech Press, 1992.

[18] R. Steele and L. Hanzo, editors. Mobile Radio Communications. J. Wiley & Sons Ltd,
2000.

[19] S. C. Yang. CDMA RF System Engineering. Mobile Communications Series. Artech

House Publishers, 1998. page 19.

[20] J. Zander and S.-L. Kim. Radio Resource Managements for Wireless Networks. Mobile
Communications Series. Artech House Publishers, 2001. page 330.

[21] X. Zhao, J. Kivinen, P. Vainikainen, and K. Skog. Propagation characteristics for wide-
band outdoor mobile communnications at 5.3 ghz. IEEE Journal on Selected Areas in
Communications, 20(3):507–514, Apr 2002.

Copyright at Technical University

Berlin. All Rights reserved.
TKN-03-007 Page 52