You are on page 1of 40

3-1

3. The Mobile Radio Channels


One of the distinguishing features of mobile communications is the channel
behavior, of which most is unwelcome. Well look at these :
1. path loss analogous to inverse square law, but worst, much worse.
Inverse fourth power ??
2. Shadowing variation due to obstacles blocking the path (makes
handoff difficult problem).
3. fading tens of dB and phase reversal in a fraction of a second. Very
challenging, very destructive.
We will elaborate on fading :
- two phenomena; fading in time, dispersion (delay spread, frequency
selective fading). Keep them separate in your mind !
- differences between mobile and base
- fading spectra, fade rate, delay profile.
3.1 Path Loss
References :
1. Steele, R, Mobile Radio Communications, Pentech Press, 1992 ;
TK6570; [1.2.5,2.7]
3-2
2. Lee, William C. Y., Mobile Communications Engineering : Theory and
Applications, McGraw Hill, 1998, 2
nd
Edition; TK6570; [3.1-3.4]
Free space inverse square law is optimistic. Even in the absence of local
scatterers and obstacles, there are large geometry objects with reflections.
Consider the following simple propagation model. Let s(t) be the
transmitted complex envelop of the transmitted bandpass signal
{ }
2
( ) Re ( )
c
j f t
s t s t e

= ! . The received bandpass signal (ignoring free space path
loss for the time being) can be written as :
( )
{ }
( )
{ }
( )
{ }
2 ( / )
( )
( , ) Re /
Re /
Re /
c
c
c
j f t x c
j t x
j t jx
r t x s t x c e
s t x c e
s t x c e e

=
=
=
!
(plane wave)
where x is the distance from the transmitter to the receiver (in the direction
of propagation), c is the speed of light,
2
c c
f =
is the carrier frequency in
rad/s, 2 / = , and
/
c
c f =
is the wavelength.
Let
a
h
and
m
h
be the height of the transmitting and receiving antennae
respectively, and d the horizontal distance between them.
m
h
a
h
d
3-3
Any phase difference between the direct and reflected rays leads to partial
cancellation in addition to the
2
1/ d free space loss.
Direct path length :
2
2 2 2 2
1
2
( ) 1 ( ) / 1
a m
d a m a m
h h
x d h h d h h d d
d
! "

# $
= + = + +
% &
' (
% &
) *
+ ,
Reflected path length :
2
2 2
1
2
( ) 1
a m
r a m
h h
x d h h d
d
! "
+
# $
= + + +
% &
' (
% &
) *
+ ,
Differential path length :
2 /
r d a m
x x x h h d

= =
Sum of arrivals (note inverse square and a reflection coefficient of 1) :
( ) ( )
( )
( )
1 1
( ) / /
1
/ 1
d r
d
jx jx
d r
jx j x
d
sum t s t x c e s t x c e
d d
s t x c e e
d





=

since ( ) ( )
/ /
d r
s t x c s t x c
(narrowband processes, time scale >>
/ x c

)
Received power is proportional to :
( )
2 2
1
2 2
2 2
2
2
2
1
( ) ( / ) 1 cos( )
2 for 1 (far away)
2
4

d
o
a m
o
E sum t E s t x c x
d
P x
x
d
h h
P
d

# $ # $
=
) * ) *
=
! "
=
% &
+ ,
"
3-4
that is the inverse fourth power of the distance.
Example : Consider the case
30 , 5 , ' /
a m
h h d d = = =
. The plot of
the normalized power as a function of ' d is as shown below. There are
2 /
m
h
maxima and the farthest one is at
' 4( / )( / )
a m
d h h =
. The inverse
fourth power phenomenon kicks in after the last peak.
Reality is not a plane earth obstacles, terrain variations, etc introduce
diffraction, reflections. However, the exponent is usually between 3 and 4
when averaged over a large ensemble of real configurations [M. Hata,
Empirical Forms for Propagation Loss in Land Mobile Radio, IEEE
Trans. On Vehicular Technology, Aug 1980]
Consequences of faster than inverse square
bad news
- need a powerful transmitter for reasonable range
- accentuated near-far problem
3-5
- n increase in transmit power doesnt give a proportional
increase in coverage area.
Edge of coverage defined by specific received power level, so
1 1 2 2
/ /
k k
P r P r =
. This means
2 2/
2 1 2 1 2 1
/ ( / ) ( / )
k
A A r r P P = =
Good news
- coverage areas are relatively sharply defined which is good for
cellular layout
3.2 Shadowing
Hills, large buildings cause variations in the received power indicated by
path loss. These obstacles prevent the existence of a direct line of sight path
between the transmitter and the receiver. When the receiver is in a shadow,
the signal strength will be weak.
Shadowing can also be caused by foliage attenuation and precipitation.
This is especially true when transmitting at super high frequency (tens of
GHz).
Shadowing is particular problematic for inbound (mobile to base) signals
where the transmitted power is low.
/
k
r t
P P r =
3-6
One remedy is to use auxiliary (or satellite) antennas, usually for inbound
transmission. Why not outbound (i.e. simulcast)? Simultaneous
transmission on same frequency from different sites gives destructive
interference in some places if exactly the same frequency; beats if not
exactly the same; garbling or ISI if envelopes not matched in time.
Shadowing variation is usually log-normal. That means if you take the dB
equivalent of the received signal strength, it will be a Gaussian random
variable. A typical variation is 6-8 dB.
3.3 Slow Fading
Here we examine the fine structure of the received signal for mobile
channels. Discussion is limited to the pdf (time variations in a later
section), so it applies to very slow fading (large , slow or motionless
mobile) so that the fade rate
d
f "
modulation bandwidth.
3.3.1 Fading Mechanism
3-7
Receive superposition of several reflections, each with own path length
from transmitter
i
x and complex reflection coefficient
i
a
(magnitude and
phase change).
Average path length :
1
1
N
i
i
x x
N
=
=
-
Differential path length :
i i
x x x =
Received signal :

{ }
{ }
( )
( )
( , ) Re ( / )
Re ( / )
Re ( [ ] / )
c i
i c
i c
j t x
i i
i
j x j t
i i
i
j x j t x
i i
i
r t x a s t x c e
a s t x c e e
a s t x x c e e



# $ =
) *
# $
=
) *
# $
= +
) *
-
-
-
!
Assume that modulation s(t) changes slowly enough to be unaffected by
differential delays
/
i
x c
, i.e.
/ 1
i
B x c "
. Then
( )
( )
( , ) Re ( / )
Re ( / )
i c
c
j x j t x
i
i
j t x
r t x a e s t x c e
g s t x c e


# $
! "

' ( % &
+ ,
) *
# $ =
) *
-
!
where g is a complex gain, constant at that point in space. As vehicle
moves, set of different path lengths change, get time varying g(t).
The last equation depicts the phenomenon of flat fading. We examine next
the first order statistics of g, the complex gain.
3-8
3.3.2 Gaussian Model Flat Fading
Examine the first order statistics of g, the complex gain.
Apply central limit theorem to real and imaginary part of
i
j x
i I Q
i
g a e g jg

= = +
- and it becomes Gaussian :
2
2 2
1 | |
( ) exp
2 2
g
g
p g

! "
=
% &
+ ,
where
( )
2
2 2 2 2
1 1
2 2 g I Q
E g E g E g
# $
# $ # $ = = +
) * ) *
) *
If you plot the pdf
( )
g
p g
as a function of
I
g
and
Q
g
, you see circular
symmetry.
In polar co-ordinate,
j
I Q
g g jg re

= + =
, where
| | r g =
, 0 r < and
( )
arctan /
Q I
g g =
,

. Can show that
I
g
Q
g
constant prob.
contour
3-9
2 2
/ 2
,
2
1
( , )
2
r
r
r
p r e

! "
=
% &
+ ,
Consequently
2 2
/ 2
2
( ) 0
r
r
r
p r e r

=

(Rayleigh)
1
( )
2
p

=
Clearly r,

are independent.
The received power is proportional to
2
z r = , which is exponentially
distributed. Follows from change of variables in Rayleigh, or from
2 2 2
I Q
r g g = +
(sum of independent squared Gaussian is
2
and
2
with 2
degrees of freedom is exponential with mean
2
2 ). In other word
2
/ 2
2
1
( ) 0
2
z
z
p z e z

=
How good is the Gaussian approximation ? As shown in the diagram
below, even 6 to 10 components is a close approximation for
/ 2 r r = , except for the tails.
3-10
The distribution of the resultant of N unit vectors with random relative
phase. The x-component of each point represents
( )
Pr / a N y >
where a is
the magnitude of the sum and y is the corresponding value on the y-axis.
Mobile satellite with a line of sight component (LOS) is better modeled by
Rice fading. In this case, the complex gain g becomes
s d
g g g = +
2 2
1
2
| | 2 | |
s s s
g K P g K = . = =
2
2
1
2 d d
E g P
# $
= =
) *
s
g
d
g
g
3-11
where
s
g
is the specular (LOS) component and
d
g
is the diffuse
component.
K represents the power ratio in the specular and diffuse components. The
total power is
2 2 2 2
1 1 1
2 2 2
| | | | | | (1 )
s d
P E g g E g K # $ # $ = = + = +
) * ) *
.
Since g has a non zero mean of
s
g
, it means
2
| | z g =
is non central
2

with
2 degree of freedom, or the pdf of
2
| | r z g = = is Ricean :
2
2 2
2
( ) exp
2
r o
r r r K
p r K I

! "
! "
=
% &
% &
% &
+ ,
+ ,
and phase [see also Eqn 5-2-55 of Proakis]
( )
2
cos ( )
1
( ) 1 4 cos( ) [1 ( 2 cos )]
2
K K
p e K e Q K

= +
3-12
Note that
2
itself varies with the lognormal distribution determined by
shadowing. The K factor depends on terrain.
3.4 Fading Autocorrelation Function and Power Spectrum
We need second order statistics of the complex gain process if we are to
describe its time variation. Its a Gaussian process, so thats all we need. If
second order statistics are constant in time, than it is WSS.
3.4.1 Power Spectrum and Autocorrelation at Mobile
References :
1. Jakes, William C. Jr., Microwave Mobile Communications, Wiley,
1974; TK6570; [Chapter 1].
2. Lee [7.1]
Consider the case we transmit an unmodulated carrier and a moving
receiver. Consequences :
1. The received complex envelop is the fading gain g. Since the
receiver is moving, g changes with time, i.e. a random process g(t).
2. When there is a relative motion between the transmitter and
receiver, there exists a Doppler frequency.
How do we relate the two ?
3-13
Assume the mobile is traveling through a set of scatterers at a velocity of V.
(Top View)
The scatterer at an angle of

to the direction of motion introduces a


Doppler shift of
cos( ) cos( )
d
V
f f

= =
to the unmodulated carrier, where
/
d
f V =
is the maximum Doppler
frequency.
The Doppler shift is the largest (i.e.
d
f
) when the mobile is traveling
towards the scatterer and the smallest (i.e.
d
f
) when away from the
scatterer
Conclude :
- the transmission of an unmodulated tone produces multiple tones with
different frequencies because of the scatterers and the relative motion.

3-14
- The power distribution, as a function of the Doppler frequency shift, is
the power spectral density of the received complex envelop, which as
mentioned earlier, is the channel complex gain process g(t).
- The inverse Fourier transform of the PSD gives the autocorrelation
function.
Note :
1. Both

and

contribute to the same Doppler shift.
2. Since
cos( )
d
f f =
,
2
1 ( / )
d d
df f f f d =
3. Let
( ) p
be the (raw) received power density coming from the
direction

. For isotropic scattering, this is a constant.


4. ( ) G be the antenna gain pattern; assume the antenna is pointing at
an angle of

with respect to the direction of motion. For a vertical


polarized antenna, ( ) G is a constant.


V
3-15
Based on the above, we can conclude that the power arriving in
d
at

is
proportional to
2
( ) ( ) G p d
. So power in
df
at f is
{ }
2 2
2
( ) ( ) ( ) ( ) ( )
1 ( / )
g
d d
df
S f df G p G p
f f f
+

#
For isotropic scattering and vertical polarized antenna,
2
2 2
| |
( )
0 otherwise
d
g
d
f f
S f
f f

/
<
0
=

1
0
2
This is known as the Jakes spectrum in the literature.
Shape of the fading spectrum (solid line) with the maximum Doppler
frequency normalized to 1.
From the power spectrum, we can obtain the autocorrelation function :
3-16
*
1
2
2 2
2
( ) ( ) ( ) IFT ( )
(2 ) 2
2
g g
o d o
o
R E g t g t S f
V
J f J
x
J

# $ # $ = =
) * ) *
! "
= =
% &
+ ,
! "
=
% &
+ ,
A plot of the autocorrelation function versus the normalized
(to the wavelength) distance of travel
From the autocorrelation function plot, we see that points separated by 0.4
are uncorrelated; however oscillations die slowly, so, in principle,
correlation lasts for many wavelengths
Exercise: Write a computer program to simulate Rayleigh fading with a
Jakes spectrum.
As expected from the autocorrelation function, signal magnitude is quasi-
periodic in space, with a dip every / 2 to or so.
From the complex envelope trajectory, we see that fades are associated
with rapid phase change and rapid fractional amplitude change.
3-17
0 2 4 6 8 10 12 14 16 18 20
-60
-50
-40
-30
-20
-10
0
10
20
x/lambda
|
g
(
x
)
|

i
n

d
B
Magnitude of g(x) in dB scale. The horizontal line denotes the
root-mean value in dB.
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
Real[g(x)]
I
m
a
g
[
g
(
x
)
]
Complex envelop trajectory. The point near (0,0) corresponds to
the deep fade at 6 x = in the amplitude plot. Separation in space
between successive circles is /10 .
3-18
In summary, the received signal in a flat fading channel is
( ) ( ) ( ) r t g t s t =
where ( ) g t is a zero mean complex Gaussian process with an
autocorrelation function of
2
( ) (2 )
g o d
R J f =
.
If the transmitted complex envelop is ( )
( ) exp 2
o
s t j f t =
, the received
complex envelop is ( )
( ) ( ) exp 2
o
r t g t j f t =
. In other word, ( ) g t is the
frequency response of the channel at time t and frequency
o
f . Since ( ) g t is
actually independent of
o
f , the instantaneous frequency response at any
time t is flat, hence the term flat fading; see the figure below.
3-19
3.4.2 Autocorrelation at the Base
The channel is reciprocal in an electromagnetic sense. This means the
autocorrelation function of the complex channel gain is the same at the
base as it is at the mobile.
The spatial correlation at the base, however, is considerably different - far
less decorrelation when base is moved than when mobile moves. Base
station antennas need much wider spacing for diversity (at mobile, only
half a wavelength or so is enough for decorrelation).
Source of difference is geometry : mobile is surrounded by scatterers, base
has signal from narrow spatial angle.
Jakes derives a detailed model based on a ring of scatterers about the
mobile, none near the base. It agrees with experiment only in a general
sense, and it is immediately thrown off by scatterers near the base.
Use of antenna arrays at the base is currently a very active research topic,
and better models for correlation among the antennas are needed.
3-20
3.5 Multipath Spread and Correlation Bandwidth
References : Lee [1.5, 1.6], Steele [Chapter 2].
Returning to Section 3.3, we now examine the case of large differential
delays amongst the different paths. By large, we mean the differential
delays are large with respect to signal variations in time.
Rewrite the received signal as
{ }
( )
( ) Re ( / )
Re ( )
Re ( )
c i
i c
c
j t x
i i
i
j x j t
i i
i
j t
i i
i
r t a s t x c e
a s t e e
h s t e

# $ =
) *
# $
=
) *
# $
! "
=
' ( % &
+ ,
) *
-
-
-
!
where
/
i i
x c =
is the delay associated with the i-th path, and
i c i
j x j
i i i
h a e a e

= =
.
The complex envelop of the received signal is
( ) ( )
i i
i
r t h s t =
-
implying a (baseband equivalent) channel with an impulse response of
( ) ( )
i i
i
h t h t =
-
and a frequency response of
3-21
2
( )
i
j f
i
i
H f he

=
-
Since we assume the differential delays are large with respect to signal
variations in time, this means
( ) H f
has significant variation across the
signal band (frequency selective).
Alternative description for the channel :
( ) ( )
i i
i
h h =
-
is defined as the response of the channel at the present moment due to an
impulse applied

second earlier. This means when the input to the channel


is ( ) s t , the output is
( ) ( ) ( ) ( )
i i
i
r t h s t d h s t = =
-
3
Multipath spread (or delay spread),
d

, is the range of
i

over which there


is significant energy, that is significant values for
2
| |
i
h
.
Correlation bandwidth (coherence bandwidth) is the frequency separation
over which there is a significant change in ( ) H f . It is the width of
*
1
2
( ) ( ) H f H f f df
3
or
*
1
2
( ) ( ) E H f H f f # $
) *
(loose definition).
A rough approximation is: coherence bandwidth
1/
d
#
, since fine structure
width in one domain (frequency) is roughly the reciprocal of the width in
the other (time). Little point in being more precise.
3-22
Example : Two path model :
1 2
( ) ( ) ( )
d
h h h = +
. The transfer
function of the channel is
2
1 2
( )
d
j f
H f h h e

= +
. Assume
1
h
and
2
h
are
drawn from an ensemble, the frequency correlation function becomes
*
1
2
2
2 2
2
1
2
1
( ) ( ) ( )
1
d
h
j f
S f E H f H f f
e

# $ =
) *
! "
= +
% &
+ ,
Caution :
( )
h
S f
is NOT the transform of autocorrelation of gain.
Now a general model is
2
( ) ( )
j f
H f h e d

=
3
and
( )( )
2 * 2 ( )
* 2 2 ( )
* 2 2 ( )
1
( ) ( ) ( )
2
1
( ) ( )
2
1
[ ( ) ( )]
2
j f j f f s
h
j f j f f s
j f j f f s
S f E h e d h s e ds
E h h s e e d ds
E h h s e e d ds









# $
=
) *
# $
=
) *
=
3 3
3 3
3 3
If we assume uncorrelated scattering gains at different delays are
uncorrelated (wide-sense stationary uncorrelated scatterers WSSUS),
*
1
2
( ) ( ) ( ) ( ) E h h s G s # $ =
) *
where
( ) G
is the delay power profile. Then
2
( ) ( )
j f
h
S f G e d



=
3
: frequency correlation function
3-23
Root mean square (RMS) delay spread is a common measure
2
2
( ) ( )
; = ( )
( )
rms
G d
G d
G d


=
3
3
3
If the mobile moves, then the path length, hence phases change, so
coefficients
i
h
in impulse response are time varying. Now it is
( , ) ( ) ( )
i i
i
h t h t =
-
It describes the response, seen at time t, to an impulse

seconds earlier.
The tap-delay line model for the channel is shown below.
( ) ( , ) ( ) ( ) ( )
i i
i
r t h t s t d h t s t = =
-
3
Finally, lets take the transform in the input/output relation
( ) r t
( ) s t
( ) r t
3
( ) h t
2
( ) h t
1
( ) h t
1
( ) h t
1

3 2

2 1

3-24
( )
2 2
2
( ) ( , ) ( )
( ) ( , )
j ft j f
j ft
r t h t S f e df e d
S f H f t e df

=
=
3 3
3
where
2
( , ) ( , )
j f
H f t h t e d

=
3
is called the time variant transfer function of the channel.
Example : If
( ) ( )
o
S f f f =
, then
( )
2
( ) ,
o
j f t
o
r t H f t e

=
. So ( ) ,
o
H f t must
be the frequency response of the channel at time t and frequency
o
f . A
sample ( , ) H f t is shown below.
3-25
At this point, want to find out if
( , ) h t
is still Gaussian for any t and

.
Look at the complex envelop of the received signal again.
( ) ( ) ( )
c i
j
i i i i
i i
r t h s t a e s t

= =
- -
(any given point in space)
If the time variation of s(t) is comparable to or less than the individual
delays (i.e very wideband), then a collection of separable paths, no
opportunity for central limit theorem. With lower resolving power, can
aggregate paths into bins { }
: ( 1)
k
T k k = +
and rewrite r(t) as :
( ) ( )
k
k
r t h s t k =
-
where
c i
i k
j
k i
T
h a e

=
-
are still more or less Gaussian, but with less plausibility than flat fading. So
channel becomes less Gaussian as the resolving power (bandwidth) of the
signal increases.
In general, the gain
k
h
is time dependent because of motion of mobile, just
like in the flat fading case.
Given that
( )
k
h t
are plausibly Gaussian for any t, the time variant transfer
function
2
( , ) ( )
j kf
k
k
H f t h t e

=
-
can be viewed as a Gaussian process in f, with autocorrelation function
3-26
*
1
2
* 2 ( )
1
2
( ) ( , ) ( , )
( ) ( )
h
j kf f f
k
k
S f E H f t H f f t
E h t h t e
+
# $ =
) *
# $ =
) *
--
$ $
$
$
It is reasonable to assume that complex gains in paths with resolvable
delays are independent, i.e.
*
1
2
( ) ( )
0 otherwise
k
k
G k
E h t h t
=
/
# $ =
1
) *
2
$
$
with
( ) ( )
k
k
G G k =
-
being the delay power profile of the channel. Consequently
2 ( )
( )
j k f
h k
k
S f G e

=
-
is simply the Fourier transform of the delay power profile, evaluated at
f
A more general correlation function is :
*
1
2
* 2 ( )
1
2
2 ( )( )
( , ) ( , ) ( , )
( ) ( )
( )
( ) ( )
h
j kf f f
k
k
j k f
k G
k
G h
S f t E H f t H f f t t
E h t h t t e
G t e
t S f

+

# $ =
) *
# $ =
) *
=
=
--
-
$ $
$
$
where
3-27
*
1
2
2
1
2
( ) ( )
( )
| ( ) |
k k
G
k
E h t h t t
t
E h t

# $
) *
=
# $
) *
is the normalized autocorrelation function for any delay.
Conclusion : uncorrelated scattering is a convenient model since the joint
time-frequency correlation function is separable.
3.6 More on Flat Fading
As shown earlier, in the absence of a LOS component, fading in a mobile
radio channel can be modeled as a zero mean complex Gaussian process
( ) g t
with an autocorrelation of
2
( ) (2 )
g o d
R J f =
(the corresponding
fading spectrum is
( )
2 2 2
( )
g d
S f f f = ,
d
f f ). We now examine the
characteristics of this fading model.
3.6.1 Some important probability density functions
References :
1. S.O. Rice Mathematical Analysis of Random Noise, BSTJ Vol. 23, pp. 282-332,
July 44, and BSTJ Vol. 24, pp. 46-156, Jan 45.
2. S.O. Rice, Statistical Properties of s Sine Wave Plus Random Noise, BSTJ, Vol.
27, pp. 109-157, Jan 48.
3. W.B. Davenport, W.L. Root, An Introduction to the Theory of Random Signals
and Noise, McGraw Hill, 58.
3-28
The probability densities functions given below are general in the sense
that they are applicable to any fading spectrum ( )
g
S f with even symmetry.
Exercise : Consider the fading process
2
( ) ( )
o
j f t
h t e g t

=
, where g(t) is a
fading process with a Jakes spectrum. Show that the PSD of h(t) is not
symmetrical about f = 0.
Note : the above represents the scenario where there is a frequency offset
between the oscillators in the transmitter and receiver, in addition to fading.
First order
( )
( ) ( ) ( ) ( )
j t
I Q
g t g t jg t a t e

= + =
:
2
2 2
1 | |
( ) exp
2 2
g
g
p g

! "
=
% &
+ ,
2 2
/ 2
,
2
1
( , )
2
a
a
a
p a e

! "
=
% &
+ ,
Derivative fading process
( )
( ) ( ) ( ) ( ) ( )
j t
d
I Q dt
g t g t g t jg t b t e

= = + =
% % %
:
2
2 2
1 | |
( ) exp
2 2
g
g
p g

! "
=
% &
+ ,
%
%
%
where

( )
2
2 2
2
rms
f =
(variance of derivative process)
2
2
( )
( )
g
rms
g
f S f df
f
S f df
=
3
3
(mean square Doppler spread)
3-29
Proof :
The derivative process
( ) g t
%
can be obtained by passing g(t) through a
linear filter with frequency response
2 j f
. The output process is
Gaussian and has a PSD of ( )
2
( ) 2 ( )
g g
S f f S f = %
Exercise : Show that
2 2
/ 2
rms d
f f = for the Jakes spectrum.
Joint pdf of g(t) and
( ) g t
%
:
( )( )
2 2
,
2 2 2 2
1 | | | |
( , ) exp
2 2 2 2
g g
g g
p g g

! "
# $
= +
% &
' (
) *
+ ,
%
%
%
Proof :
The processes g(t) and
( ) g t
%
are jointly Gaussian. Their cross-correlation
can be written as :
* *
*
1 1
2 2
0
* *
0
0
( ) ( )
( ) ( ) ( ) lim ( )
1 ( ) ( ) ( ) ( )
lim
2
( ) ( )
lim ( )
gg
t
t
g g
g
t
g t g t t
R E g t g t E g t
t
g t g t g t g t t
E
t
R R t
R
t






# $
# $ = =
' (
) *

) *
# $
=
' (

) *
+
/ 4
= =
1 5

2 6
%
%
%
Therefore [ ]
1
2
( ) ( ) (0)
g
E g t g t R =
%
%
. For Jakes spectrum, the derivative of
the autocorrelation function at 0 = is 0.
Conclude : g(t) and ( ) g t
%
are independent when evaluated at the same
time instants. Therefore
,
( , ) ( ) ( )
g g g g
p g g p g p g =
% %
% %
, which is basically
what the equations says.
3-30
Exercise : Show that [ ]
1
2
( ) ( ) 0 E g t g t =
%
is true for any ( )
g
S f with even
symmetry.
Joint PDF, polar co-ordinate :
Since
,
j j j j
g ae g be ae j ae

= = = +
% % %
, we have
2
2 2 2 2
g b a a = = +
% % %
This information can be used (see [Rice]) to show that
2 2 2 2 2
, , ,
2 2 2 2 2
1
( , , , ) exp
4 2
a a
a a a a
p a a



/ 4
! " + 0 0
= +
1 5
% &
0 0
+ ,
2 6
% %
% %
% %
where a range from 0 to , range from

to

, and both a
%
and
%
range from to .
Because of the form of this joint pdf, the marginal pdfs can be easily
obtained.
Why are these probability density functions of importance? Well, they are
directly related to performance. Will examine two here first :
1. random FM
2. fade rate and fade duration
More to come later.
3.6.2 Random FM
3-31
Refer to [Lee 7.3] and [Jakes, Chapter 1].
In digital FM system, a popular demodulator is the limiter/discriminator.
Let
( )
( )
j t
s t e

=
and
( )
( ) ( )
j t
g t a t e

=
. Then in the absence of noise, the
signal at the input of the LPF is
[ ( ) ( )]
( ) ( ) ( ) ( )
j t t
r t s t g t a t e
+
= =
. The
output of the limiter/discriminator is approximately
( ) ( ) t t +
%
%
where the first term is the signal component and the second term is called
random FM.
From the joint pdf
, , ,
( , , , )
a a
p a a


% %
% %
, we can obtain the marginal pdf for
the random FM

%
:
, , ,
0
3/ 2
2
2
2
3/ 2
2
( ) ( , , , )
1
2
1
1
4 2
a a
a a
rms rms
p p a a d da da
f f



= = =

=
! "
= +
% &
+ ,
! "
# $
% & = +
' (
% &
) *
+ ,
3 3 3
% % %
%
% % % %
%
%
( ) s t
LPF
Limiter/
Discriminator
( ) g t ( ) n t
3-32
This is a long-tailed pdf where
2 2 3
( ) 2 /
rms
p f


%
% %
for large

%
. So even
though
( ) t
is bounded, its derivative is not.
It is quite non-Gaussian
Example: A
4800
b
R =
bps FSK modem, modulation index h=1/2, detected
with discriminator and no post-filter, transmit at 800 MHz, vehicle at 100
km/hr, what is the irreducible error rate ?
- A modulation index of means
( )
( ) exp 2 s t j f t

=
where
/ 2 1200
b
f hR

= =
.
- Doppler frequency :
/ / 74
d c
f v vf c = = =
Hz.
- Make an error if random FM exceeds signal. If the negative frequency
was sent, then the irreducible error probability is
[ ]
( )
( )
2
2
2
1
2
Pr 2 ( )
1 1
1
2
/ 1
1 1
1 for Jakes spectrum
2
/ 1
eo
f
rms
d
P f p d
f f
f f

= > =
! "
% &
=
% &
+
+ ,
! "
% &
=
% &
+
+ ,
3
%
% % %
When
/ 1
d
f f

<<
, ( )
2
8
eo d
P f f

. Variation as
2
d
f
is a characteristic
of the error floor. You will see more error floors in Chapter 4.
3-33
3.6.3 Frequency and Duration of Fades
The big questions in relating fading to system designs :
- what type of burst error correcting code ?
- and is burst error correction useful, anyway ?
- length of forward error correcting code (FEC) ?
- depth of interleaving to break up bursts ?
- in antenna switched diversity, how much delay is tolerable ?
Answers to some can be provided by analysis. Others through
measurements and/or simulations.
Preliminary observations
- Let
2
( ) ( ) z t g t = . As shown in Section 3.3.2,
2
/ 2 2
( ) (2 )
z
z
p z e

= . Thus
[ ]
2
/(2 )
Pr 1
o
z
o
z z e

=
- fades are of varying depths, and instantaneous system performance
(like bit error probability, output signal-to-noise ratio) may be
relatively soft function of depth.
- But we will think in terms of threshold, since useful for coded
systems, antenna switching etc., and some applicability to uncoded
systems
- Define threshold level with respect to the mean (rms) level. Then at
lower levels :
1. less time spent in fade state
2. fewer fades
3. shorter fades
3-34
0 2 4 6 8 1 0 1 2 1 4 1 6 1 8 2 0
- 6 0
- 5 0
- 4 0
- 3 0
- 2 0
- 1 0
0
1 0
2 0
x / l a m b d a
|
g
(
x
)
|

i
n

d
B
Fade Rate :
- Refer to [Lee 2.7, 6.4], [Rice 1948]
- This is a level crossing rate (LCR). For a given amplitude threshold A,
the number of upward crossings per second is
,
( ) ( , )
a a
o
n A a p A a da

=
3
%
% % %
where
,
( , )
a a
p a a
%
%
is the joint pdf of the fading envelop
( ) a g z = =
and
its derivative. The proof below is due to [Rice] :
Amplitude of
fading gain
a adt +
%
a
A
t dt +
t
Time
3-35
- In any time range t, t+dt, there is an upward crossing if
A a a dt
%
and
0 a >
%
Equivalently, we need
A a dt a A
%
Therefore
[ ]
,
0
,
0
Pr upward crossing in ( , )
( , )
A
a a
a a A adt
a a
a
dt da p a a da
dt da a p A a

= =

=
=
=
3 3
3
%
% %
%
%
% %
% % %
This means the expected number of crossings in any dt, hence the
crossing rate is
,
( ) ( , )
a a
o
n A a p A a da

=
3
%
% % %
Note that the lower limit of integration is zero because we are interested
in positive crossings.
- The joint pdf
,
( , )
a a
p a a
%
%
is
, , , ,
2 2
2 2
2 2
( , ) ( , , , )
1
exp
2
2
a a a a
p a a p a a d d
a a a

= =
=
/ 4
! "
0 0
= +
1 5
% &
0 0
+ ,
2 6
3 3
% % %
%
% % % %
%
Substitution gives
3-36
2
2
2
2
( ) exp
2
2
2 for Jakes spectrum
R
d
A A
n A
f R e


# $
=
' (
) *
=
where
2
2
A
R

=
is the crossing threshold normalized to the RMS envelop level.
- The figure below shows the normalized crossing rate (with respect to the
Doppler frequency) as a function of the normalized threshold level R.
-40 -35 -30 -25 -20 -15 -10 -5 0 5 10
10
-4
10
-3
10
-2
10
-1
10
0
10
1
20*log10(R)
n
(
A
)
/
f
d
Normalized zero crossing (fade) rate versus normalized threshold level.
3-37
It is observed that
a. a 6 dB decrease in level (R cut in half) means cutting the fade
(zero crossing) rate by half,
b. fade rate proportional to
d
f
,
c. maximum rate at 3 R = dB
Fade Duration
- Refer to [Lee 6.5], [Jakes 1.3.5]
- Easy, now that we have the other relations
2
Pr / 2 ( ) ( ) a R T R n R
# $
=
) *
Hence
2 2
2
1 1
( )
2 2 2
R R
R
d d d
e e R
T R
f R f f R e


= =
- Plot of normalized fade duration (
( )
d
f T R
) vs R is shown in the plot
below.
1. fade duration is proportional to amplitude level for R equal 5 dB and
below
2. a 6 dB decrease in level means fades are half as long (and half as
frequent, as we saw earlier)
3. fade duration inversely proportional to
d
f
3-38
-40 -35 -30 -25 -20 -15 -10 -5 0 5 10
10
-3
10
-2
10
-1
10
0
10
1
10
2
10
3
10
4
20*log10(R)
T
(
R
)
*
f
d
Normalized fade duration versus normalized threshold
- It is shown in [Arnold and Boltman, Interfade Interval Statistics of a
Rayleigh distributed wave, IEEE Trans. Communications, Sept. 83]
that intervals between fades are approximately exponentially distributed
for 15 R < dB or so.
The same two authors in [Boltman and Arnold, Fade distribution
statistics of Rayleigh distributed wave, IEEE Trans. Communications,
March 82] show that the fade duration is not exponential
Example : Consider a mobile data system. What does a 3 dB increase in
transmit power buy in terms of error burst statistics ?
Solutions :
3-39
Since
2
/ 2 R A = , increase
2
by 3 dB (i.e doubling it) effectively reduces
R to 0.71 times its old value. Since for small R, the fade duration is
approximately proportional to R, so bursts are now 71% as long.
Example : Have a mobile data system operating at 1 GHz, vehicle speed
108 km/hr, data rate 9600 bps using FSK with discriminator detection,
average error probability (BEP) is
3
10

. Estimate (roughly) burst duration


and frequency, both in bits. Use the following expressions for the
instantaneous and the average BEPs :
2
/ 4
1
( )
2
a
e
P a e

=
(instantaneous BEP)
2
0
1
( ) ( )
2
e e a
a
P P a p a da

=
= =
+
3 (average BEP)
where
2
is the variance of the underlying complex fading process. For this
question, it also respresents the average SNR in the system.
Solutions :
- From the BEP expressions, we can deduce that
2
1000 or 30 dB.
- Major approximation : bursts defined by instantaneous BEP > 0.1. This
means 2.5373 A = is the threshold.
- The normalized threshold is
2
/ 2 0.0032 R A = =
or approximately 25
dB.
3-40
- From equations (or graphs), the normalized fade rate and normal fade
duration are approximately 0.14 and 0.022 respectively.
- Since / 100
d
f V = = Hz, we have
( ) 14 n A =
deep fades/s and
( ) 224 T A s =
- Bit duration is
1/ 9600 104 T s = =
. So average fade duration is 2 bits
- Inter-fade interval is 9600/n(A) = 686 bits.
Exercise : Repeat last exercise but with a velocity of 50 km/hr.

You might also like