1

Log-normal Shadowing
Instructor: M.A. Ingram
ECE4823
Shadowing
„ Also called slow-fading
„ Accounts for random variations in
received power observed over distances
comparable to the widths of buildings
„ Extra transmit power (a fading margin)
must be provided to compensate for
these fades
Local Average Power
Measurements
„ Take power measurements in Watts as the antenna is moved in
a on the order of a few wavelengths
„ Average these measurements to give a local average power
measurement
Velocity of antenna
Points where power measurements are made
> λ/2
Scanning in 3D
„ After the cart is stabilized,
the XYZ actuators move an
antenna in 3 D to make a set
local power measurements
„ One local average power is
computed for each cart
location
0 5 10 15 20 25 30 35 40 45 50
0
10
20
30
400
5
Scan volume
X Y
Z
X
Y
Z
The antenna
Same-Distance Measurements
„ Local averages are made for many
different locations, keeping the same
transmitter-receiver distance
„ These local averages will vary randomly
with location
Example Tx-Rx
locations within
a floor of a building
Repeat for Multiple Distances
„ Similar collections of average powers
are made for other Tx-Rx distances
2
Likelihood of Coverage
„ At a certain distance, d, what is the
probability that the local average
received power is below a certain
threshold γ?
) ) ( ( γ < d P P
r
r t
r t t
r
L d L L
G G P
d P
) (
) ( =
distance dependence shown explicitly
Likelihood of Coverage, cont’d
„ Since only the path loss L(d) is random,
the probability can be expressed as a
probability involving path loss:
) ) ( ( ) ) ( ( β γ > = < d L P d P P
r
maximum tolerable path loss
Typical Macrocell
Characteristics
T
o
t
a
l

P
a
t
h

L
o
s
s

[
-
d
B
]
10log
10
(Distance from Base Station [km])
0 2 3 4 5 6 7 9 10 1
-60
-70
-80
-90
-100
-110
-120
.
.
.
. .
.
.
.
.
. .
.
. . .
.
. .
.
.
. .
..
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
. . .
.
. .
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
. .
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
. .
.
.
.
.
.. .
.
.
.
.
.
.
.
.
.
.
. .
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
.
. .
[not real data]
The set of average
losses measured
at about 4km
[10log
10
(4)=6]
Path Loss Assumptions
„ The mean loss in dB follows the power law :
„ The measured loss in dB varies about this
mean according to a zero-mean Gaussian RV,
X
σ
, with standard deviation σ
|
|
.
|

\
|
+ =
o
o
d
d
n d L d L
10
log 10 ) ( ) (
σ
X
d
d
n d L d L
o
o
+
|
|
.
|

\
|
+ =
10
log 10 ) ( ) (
Typical Data Characteristics
T
o
t
a
l

P
a
t
h

L
o
s
s

[
-
d
B
]
10log
10
(Distance from Base Station [km])
0 2 3 4 5 6 7 9 10 1
-60
-70
-80
-90
-100
-110
-120
.
.
.
. .
.
.
.
.
. .
.
. . .
.
. .
.
.
. .
..
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
. . .
.
. .
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
. .
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
. .
.
.
.
.
.. .
.
.
.
.
.
.
.
.
.
.
. .
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
.
. .
Best Fit
Path Loss
Exponent:
n=4
[not real data]
σ is usually
5-12 dB for
mobile comm
Probability Calculation
„ Since L(d) is Gaussian, we need to
know how to calculate probability
involving Gaussian RVs
) ) ( ( β > d L P
3
Q Function
„ If X is a Gaussian RV with mean α and
standard deviation σ, then
where Q is a function defined as
|
.
|

\
| −
= >
σ
α b
Q b X P ) (
dx
x
z Q
z

+∞
|
|
.
|

\
|
− =
2
exp
2
1
) (
2
π
The Problem with Q
„ The integrand of Q has no
antiderivative
„ Q is found tabulated in books
„ Q can be calculated using numerical
integration
What is Log-normal
Shadowing?
„ If Y is a Gaussian RV and Z is defined
such that Y=logZ, then Z is a log-
normal RV
„ Shadowing is log-normal shadowing
when the path loss in dB is Gaussian;
this means that the path loss expressed
as a ratio is log-normal
Inverse Q Problems
„ Sometimes, the probability is specified and
we must find one of the parameters in the
argument of Q
„ Suppose the value of is given, along
with values of b and α. Solve for σ
„ Must look up the argument of Q that gives
the specified value.
|
.
|

\
| −
= >
σ
α b
Q b X P ) (
) ( b X P >
Example Inverse Q Problem
„ Suppose the mean of the local average received powers at a
certain distance is -30dBm, that the standard deviation of
shadow fading is 9 dB, and that the observed received power is
above the threshold 95% of the time. What is the threshold
power?
„ Q is usually tabulated for arguments of 0.5 and less, so we must
use the fact that
„ The argument of Q that yields 0.05 is about 1.65
95 . 0
9
) 30 (
) ( = |
.
|

\
| − −
= >
b
Q b P P
r
dBm 85 . 44 and , 65 . 1
9
30
− = =
+
− b
b
( ) ) ( 1 z Q z Q − − =
Boundary Coverage
„ Suppose that a cell has radius R and γ is the
minimum acceptable received power level
„ Then is the “likelihood of
coverage” at the boundary of the cell
„ is also the “fraction of time” that a
mobile’s signal is acceptable at a distance R
from the transmitter, assuming the car moves
around that circle
) ) ( ( γ > R P P
r
) ) ( ( γ > R P P
r
4
Percentage of Useful Service
Area
„ By integrating these probabilities over
all the circles within a disk, one can
compute the fraction of the area within
the cell that will have acceptable power
levels
θ γ
π
γ
π
d rdr r P P
R
U
R
r
∫ ∫
> =
2
0 0
2
) ) ( (
1
) (
Integral Evaluated
„ Assuming log-normal shadowing and the
power path loss model, the fraction of useful
service area is
where
|
|
.
|

\
|
(
¸
(

¸

|
.
|

\
| −
− |
.
|

\
| −
+ − =
b
ab
b
ab
a U
1
erf 1
2 1
exp ) ( erf 1
2
1
) (
2
γ
2
) (
σ
γ R P
a
r

=
2
log 10
10
σ
e n
b = and
The Error Function
„ erf(x) is another form of the Gaussian integral
(like Q(x))
„ erf(x) has odd symmetry, with extreme values
±1.
„ Note that some authors may define erf
differently


=
x
t
dt e x
0
2 2
) ( erf
π
1
erf
1
(x)
x
-1
erf and Q
„ The erf function and Q are related:
( ) z Q z 2 2 1 ) ( erf − =
When the Average Boundary
Power is Acceptable
„ Suppose
.
Then we may
use this graph
from [Rappaport
’96] to figure
the percent
useful service
area
γ = ) (R P
r
Summary
„ Local averages in dB of received power (or
path loss) tend to be Gaussian when the
ensemble is all Tx-Rx locations with the same
distance in the same type of environment
„ The mean local average path loss follows the
standard power model (proportional to
10logd
n
)
„ Can use Q or erf to calculate the likelihood of
boundary coverage or the percent of useful
service area
5
References
„ [Rapp, ’96] T.S. Rappaport, Wireless
Communications, Prentice Hall, 1996
„ [Saunders,`99] Simon R. Saunders,
Antennas and Propagation for Wireless
Communication Systems, John Wiley
and Sons, LTD, 1999.

. Total Path Loss [-dB] -80 -90 -100 -110 -120 0 1 2 10log10(Distance from Base Station [km]) . .. . .. ...... . ....... . . . . .. Total Path Loss [-dB] ....... . .... .. .. . . . -60 . ...... . . .... . .. . ...... . .. . . ... .. . ...... . .. .... . .... . . . . . . .. we need to know how to calculate probability involving Gaussian RVs [not real data] Best Fit Path Loss Exponent: n=4 σ is usually 5-12 dB for mobile comm P ( L( d ) > β ) .. .. ... .. . .... .. . .. . .. .. . . . . .. . .. . .. .. .. . . . . . .. .... ... . . . . .. .... ... .... . . . .. . Xσ. . . . ..... . . . ... . with standard deviation σ d  L(d ) = L(d o ) + 10n log10   + X σ d   o Probability Calculation Since L(d) is Gaussian. . . ... .. .. .. what is the probability that the local average received power is below a certain threshold γ? Likelihood of Coverage. . .. . . . . . . ... . .. .... . . . . 3 4 5 6 7 9 ... .. . . .. ... .. ..... ..... . . . .. .. .. .. d. .. .. . .... .. . . 3 4 5 6 7 9 10 10 . . . . . . . . . . .... .. . . . . . . . . .. . . . ... . . .. .. . ... . .. .. .... . . . . ... . .. . . . . .. .. .. .. .. .. . . . .. .... . ... .. . . . .. . .. . ... . .. . .. .... . . .. ...... . . ...... . .. . . . . . .Likelihood of Coverage At a certain distance. . -60 The mean loss in dB follows the power law : d  L(d ) = L(d o ) + 10n log10   d   o .... .. ... ... .. . . . . . .. .. .... ..... ... .... .. ...... . . .. . . . . . .. . ... ... . . .. .. ..... . .. .. ... .. . . The set of average losses measured at about 4km [10log10(4)=6] The measured loss in dB varies about this mean according to a zero-mean Gaussian RV....... .... ... .. .... . . . .... . . . . .. . .... . . . . ... . . . . . .. ... .. cont’d Since only the path loss L(d) is random. . . .. .. .. .. . . ... . the probability can be expressed as a probability involving path loss: P( Pr (d ) < γ ) P( Pr (d ) < γ ) = P( L(d ) > β ) maximum tolerable path loss Pr (d ) = Pt Gt Gr Lt L(d ) Lr distance dependence shown explicitly Typical Macrocell Characteristics -70 Path Loss Assumptions [not real data] -80 -90 -100 -110 -120 0 1 2 10log10(Distance from Base Station [km]) Typical Data Characteristics -70 . .. . . ... .

05 is about 1. the probability is specified and we must find one of the parameters in the argument of Q  b −α  P ( X > b) = Q   σ  Suppose the value of P( X > b) is given. then  b −α  P ( X > b) = Q   σ  The Problem with Q The integrand of Q has no antiderivative Q is found tabulated in books Q can be calculated using numerical integration where Q is a function defined as Q( z ) = 1 2π +∞ ∫ exp −   z  x2  dx 2  What is Log-normal Shadowing? If Y is a Gaussian RV and Z is defined such that Y=logZ. that the standard deviation of shadow fading is 9 dB. assuming the car moves around that circle P( Pr > b) = Q  9  = 0. this means that the path loss expressed as a ratio is log-normal Inverse Q Problems Sometimes. then Z is a lognormal RV Shadowing is log-normal shadowing when the path loss in dB is Gaussian. and b = −44.5 and less. and that the observed received power is above the threshold 95% of the time. Solve for σ Must look up the argument of Q that gives the specified value.85 dBm 9 . along with values of b and α. Example Inverse Q Problem Suppose the mean of the local average received powers at a certain distance is -30dBm.Q Function If X is a Gaussian RV with mean α and standard deviation σ. so we must use the fact that Boundary Coverage Suppose that a cell has radius R and γ is the minimum acceptable received power level Then P( Pr ( R) > γ ) is the “likelihood of coverage” at the boundary of the cell P( Pr ( R) > γ ) is also the “fraction of time” that a mobile’s signal is acceptable at a distance R from the transmitter.95  Q(z ) = 1 − Q(− z ) The argument of Q that yields 0.65 − b + 30 = 1. What is the threshold power?  b − (−30)  Q is usually tabulated for arguments of 0.65.

. erf ( x ) = 2 erf and Q The erf function and Q are related: erf ( z ) = 1 − 2Q ( 2z) π ∫e 0 x 1 −t 2 dt -1 x Note that some authors may define erf differently When the Average Boundary Power is Acceptable Suppose Summary Local averages in dB of received power (or path loss) tend to be Gaussian when the ensemble is all Tx-Rx locations with the same distance in the same type of environment The mean local average path loss follows the standard power model (proportional to 10logdn) Can use Q or erf to calculate the likelihood of boundary coverage or the percent of useful service area Then we may use this graph from [Rappaport ’96] to figure the percent useful service area Pr (R ) = γ .Percentage of Useful Service Area By integrating these probabilities over all the circles within a disk. one can compute the fraction of the area within the cell that will have acceptable power levels U (γ ) = 1 πR 2 2π R Integral Evaluated Assuming log-normal shadowing and the power path loss model. the fraction of useful service area is 1  1 − 2ab    1 − ab   U (γ ) = 1 − erf (a) + exp  1 − erf    2 b2    b      ∫ ∫ P( P (r ) > γ ) rdrdθ r 0 0 where a= γ − Pr ( R) σ 2 and b= 10n log10 e σ 2 The Error Function erf(x) is another form of the Gaussian integral (like Q(x)) erf(x) has odd symmetry. with extreme values erf1(x) ±1.

1999. LTD.References [Rapp. Prentice Hall. Wireless Communications. John Wiley and Sons.S. Saunders. Rappaport. ’96] T. . [Saunders. 1996 Antennas and Propagation for Wireless Communication Systems.`99] Simon R.

Sign up to vote on this title
UsefulNot useful