Cellular Network Planning

and Optimization
Part II: Fading
Jyri Hämäläinen,
Communications and Networking Department,
TKK, 17.1.2007
 Outline

Modeling approaches

Path loss models

Shadow fading

Fast fading

2
Modeling approaches

3
 Fading seen by moving terminal
Fast fading
Power
Modeling approach:
+20 dB

1. Distance between
TX and RX =>
path loss
Path loss
2. Shadowing by
large obstacles =>
shadow fading
3. Multi-path effects
=> fast fading - 20 dB Lognormal
fading
Path loss
Time

4
 Path Loss

Path loss is distance dependent mean

attenuation of the signal.

Once the allowed path loss of a certain system
is known we can solve the maximum distance
between transmitter and receiver and compute
the relative coverage area.

Suitable path loss model depends on the
environments (macro-cell, micro-cell, indoor)
 Outdoor to outdoor models
 Outdoor to indoor models
 Indoor models

5
 Shadow Fading

Shadow fading is used to model variations in

path loss due to large obstacles like buildings,
terrain conditions, trees.

Shadow fading is also called as log-normal
fading since it is modeled using log-normal
distribution

In cell dimensioning/link budget shadow fading
is taken into account through a certain margin
(=shadow fading margin)

6
 Path loss + shadow fading

Log-normal distribution
Signal strength in dB’s

Path loss

Standard deviation e.g. +/-8 dB

Distance between TX and RX in logarithmic scale

7
 Fast Fading

Fast fading is also called as multi-path fading since it is mainly
caused by multi-path reflections of a transmitted waves by local
scatterers such as human build structures or natural obstacles
 Fast fading occurs since MS and/or scatterers nearby MS are moving
 Signal strength in the receiver may change even tens of decibels
within a very short time frame
 Signal coherence distance = separation between locations where fast
fading correlation is negligible. Signal coherence distance is half of the
carrier wavelength
 f = 2GHz => coherence distance = c/(2*f) =7.5 cm
 Coherence time = time in which MS travels coherence distance
 Coherence time depends on MS speed.

In cell dimensioning/link budget fast fading is taken into account
through a certain margin (=fast fading margin)

8
 Fast Fading

Scatterers

Especially the changes in

component signal phases
a1 (t )e jφ1 ( t ) create rapid variations in
sum signal
a1 (t + t0 )e jφ1 ( t +t0 )

Sum signal at time t S (t ) = a1 (t )e jφ1 (t ) + ... + a5 (t )e jφ5 ( t )

Sum signal at time t+t0 S (t + t0 ) = a1 (t + t0 )e jφ1 ( t +t0 ) + ... + a5 (t + t0 )e jφ5 ( t +t0 )
9
Path loss models

10
 Content
 We recall first two important path loss models for macro- and micro-cell
environments
 I Model: Classical Okumura-Hata
 Okumura-Hata is based on only few parameters but it works well and is widely used
to predict path loss in macro-cell environments
 II Model: COST 231 or Walfisch – Ikegami
 This model is suitable for both macro- and micro-cell environments and it is mode
general than Okumura-Hata. Walfisch – Ikegami models propagation phenomena
more accurately but in cost of increased complexity.
 Then we consider path loss in urban environment when both transmitter
and receiver are below the rooftop (Berg model)
 Outdoor to outdoor model
 Path loss of RS – MS signal in street canyon II Model: BRT – BRT, NLOS
(Berg model)
 Finally, we discuss shortly on outdoor-to-indoor modeling
 Terminology
 ART= Above Roof Top
 BRT = Below Roof Top
 LOS = Line-of-Sight
 NLOS = Non Line-of-Sight

11
 General path loss model/outdoor

Outdoor path loss models are usually given in the form

(*) L= A+ 10 ⋅ n ⋅ log 10 ( R ) (in decibels)

Here
 R is the distance between TX and RX
 A and n are constants. Values of these constant are
depending on the various parameters such as carrier
frequency, antenna heights etc

An other form for formula (*)
~ ~ n
L = 10 = 10 ⋅ R = A ⋅ R
L / 10 A / 10 n

Note that n defines the exponential attenuation of the signal. Typically

its value is 3-4 in urban environment. In free space n=2.
12
 Okumura-Hata

Okumura-Hata propagation loss model
 Based on measurements in Tokyo
 May be the most widely used path loss model for
attenuation of cellular transmissions in built up areas.
 Most suitable for large macro-cells
L = A + B log10 f c − 13.82 log10 hb − a ( hm ) + ( C − 6.55log10 hb ) log10 d

A and B constants fc Carrier frequency (MHz)

150-1000 1500 – hb Base station antenna height
MHz 2000 MHz 30m ≤ hb ≤ 200m
A 69.55 46.3 hm Mobile station antenna height
B 26.16 33.9 hm ≈ 1.5m
a ( hm ) Mobile antenna gain function
C 44 – 47, default 44.9

C constant gives distance dependency and should be fitted to

local measurements 13
 Okumura-Hata

Mobile station antenna gain function
 Small/Medium size city

a ( hm ) = (1.1log10 f c − 0.7 ) hm − (1.56 log10 f c − 0.8 )

 Large city
 8.29 ( log (1.54h ) )2 − 1.1 f ≤ 200 MHz

a ( hm ) = 
10 m

3.2 ( log10 (11.75hm ) ) − 4.97 200

2
f ≥MHz < f c ≤ 1500 MHz
400 MHz

 Antenna gain function can be in most cases be ignored!

a ( h m ) = 0 , f c > 1500 MHz
a ( h m ) = 0 , h m = 1 .5 m

14
 Okumura-Hata: Example
170

Path loss according to 160

Okumura-Hata model in

Path Loss [dB] (Okumura-Hata Model)

150

large city when 140

 f = 450 MHz (□) 130

 f = 900 MHz (*) 120

110
 f = 1800 MHz (o)
100
 f = 1950 MHz (x) 90

Flash-OFDMA (NMT-450), 80

GSM900, GSM1800, 70
0 1 2 3 4 5 6 7 8
Range [km]
WCDMA

Note: There would be huge differences in coverage if

allowed path losses would be the same for different systems
15
BS height = 30m, MS height = 1.5m
 Okumura-Hata: Example
f c = 1950 Mhz

hb
hm = 1.5m
d

Impact of base station 165

antenna height: 160

Path Loss [dB] (Okumura-Hata Model)

155
 Distance = 3.0 km
150
 Distance = 2.0 km
145
 Distance = 1.0 km
140
 Distance = 0.5 km
135

Distance measured 130

between TX and RX 125

120
10 15 20 25 30 35 40 45 50
Base Station Antenna Height [m] 16
 COST231-Walfisch-Ikegami path loss model

In the following we
also use notations:

Note: we consider only 17

NLOS case
 COST231-Walfisch-Ikegami path loss model

The rooftop-to-street diffraction

loss term determines the loss
which occurs on the wave
coupling into the street where φ
the receiver is located

18
COST231-Walfisch-Ikegami path loss model

19
 COST231-Walfisch-Ikegami path loss model

Comparison with some measurements made by Nortel in 1996 for a base

antenna deployed in Central London well above the average rooftop
height revealed that the COST 231 W-I model did not correctly model the
variation of path loss with mobile height. This problem was solved by the
above correction factor.

20
 COST231-Walfisch-Ikegami path loss model:
Impact of rooftop height
190

180
Parameters:
170
BS antenna height = 30m
160
Carrier frequency = 1950MHz

Path Loss [dB]

150
Street width = 12m
140
Building spacing = 60m
130
Street orientation = 90 degrees
120
Roof top heights:
110
6m (□), 12m (*)
100
18m (o), 24m (x)
90
0 1 2 3 4 5 6 7 8
Range [km]

Remarks:
 W-I and Okumura-Hata give approximately the same path
loss curve when roof top height is 12m 21
 Impact for rooftop height is crucial for cell coverage
 COST231-Walfisch-Ikegami path loss model:
Impact of MS height

BS antenna height = 15m (o), 20m (*), 25m (x), 30m (+)
Parameters: BS - RS distance = 1 km
150
Roof top height = 12m
145
Carrier frequency = 1950MHz 140

Street width = 12m 135

Building spacing = 60m

Path Loss [dB]

130

Street orientation = 90 degrees 125

120

115

110

105
1 2 3 4 5 6 7 8 9 10 11
Relay Height [m]
MS height [m]

Notice:
- BS antenna 20m -> 30m => 10 dB gain 22
- MS antenna 1.5m -> 5m => 10 dB gain
 Berg model
Scenario:
Both BS and MS antennas are BS
below rooftop.
Model takes the minimum of an
over-the-rooftop signal
component and a round-the
streets component.
This scenario will be increasingly
important in future since density
of network elements is increasing
and macro-cell site costs are high

23
 Berg model

Check how to use this model 24

Note: Path loss depends heavily on corners (how many, how sharp)
 Berg model: Example
160
o
30 m 140

200 m o
120

Path Loss [dB]

100

x x
RS * * 80

60

∇∇
+ 40
+ 0 100 200 300 400
Distance from relay [m]
500 600

o
130

120
o
110
Red marks = range along the dashed route
100
Path Loss [dB]

Violet marks = range without penetration loss

90

Remarks: 80

- This model is quite pessimistic (high path 70

loss) 60

- Signal is dying soon round the corner 25

50
0 100 200 300 400 500 600
- BS location planning is important D is t a n c e fro m re la y [ m ]
Outdoor-to-indoor modeling: Example

Remark: 26
- Path loss depends on number of walls
Shadow fading

27
 General remarks
W-I with parameters:

In urban areas macro-cell ranges are BS antenna height = 25m
from few hundred meters up to few
kilometers Roof top height = 15m

Shadowing by big buildings etc can be Carrier frequency = 1950MHz
critical on cell edge. It may create Street width = 12m
large coverage holes Building spacing = 60m

Example: Allowed total signal fading in Street orientation = 90 degrees
system is 155dB and shadow fading
margin is 8dB. How much larger (in %) 170

would the coverage be without 160

shadow fading margin? Use figure of
the slide for range comparison. 150

Answer: Cell range would increase

Path Loss [dB]
140
from 1.35 km up to 2.2 km which leads
to 267% increase in coverage 130

Remark: the impact of shadow fading
can be really large 120

110

100 28
0 0.5 1 1.5 2 2.5 3
Range [km]
 Shadow fading model

Shadow fading is modeled by log-normal

distribution, i.e. signal strength in decibels is of
the form
(1) L =L + X
where first term is the mean path loss and latter
term follows the normal distribution,
x2
1 − 2
(2) X~ e 2σ
= f ( x)
2π σ
with zero mean and standard deviation σ.

29
 Cell edge coverage probability

In link budget shadow fading is taken into account

through a certain shadow fading margin (SFM). In cell
border we require that the signal strength plus SFM is
larger than mean signal level by a certain probability,
denoted by Pcov. Then we compute the corresponding
SFM. Hence, we require that

Pcov = P {L + SFM > L }

= P {L + X + SFM > L }
= P{X > − SFM }
-SFM

30
L= A+ 10 ⋅ n ⋅log10 ( R) L= A+ 10 ⋅ n ⋅log10 ( R)
 Cell edge coverage probability

Using the distribution (2) we find that

∞ x2
−
Pcov = P{X > − SFM } =
1
2π σ ∫e
− SFM
2σ 2
dx

∞ t2
1 −
= ∫ eσ 2
dt = Q ( − SFM / σ ) = 1 − Q ( SFM / σ )
2π − SFM /

From this equation we can solve SFM for given

Pcov and σ:
SFM = σ ⋅ Q −1 (1 − Pcov )

31
 Cell edge coverage probability

Function 1-Q is the cumulative density function

(CDF) of normal distribution with zero mean and
standard deviation 1. Moreover,
Q (∞ ) = 0, Q ( −∞ ) = 1,
∞ t2 ∞ t2 −x t2
1 − 1 − 1 −
Q (− x) =
2π ∫e
−x
2
dt =
2π ∫e
−∞
2
dt −
2π ∫e
−∞
2
dt

∞ t2 ∞ t2
1 − 1 −
=
2π ∫e
−∞
2
dt −
2π ∫e
x
2
dt = 1 − Q ( x )

32
 Cell edge coverage probability

We can estimate the value of SFM by using the

inversion curve of Q. 0
10

Example: Let Pcov=0.95 10

-1

and let σ=6dB. Then we

find from the curve that 10
-2
Q(x)

Q − 1 (1 − Prcov ) ≈ 1.6449 -3
10

and hence, SFM=11.6dB -4

10

-5
10
0 0.5 1 1.5 2 2.5 3 3.5 4
x

33
Outdoors: σ=5-8dB. Indoors: σ=10-12dB.
 Single cell coverage probability

Next we compute the cell coverage probability in

case of a single cell.

Analytical computation is pretty technical but
result shows the relation between path loss
exponent n, standard deviation σ of shadow
fading and the required cell coverage probability

34
 Single cell coverage probability

Let us compute the single cell coverage

probability. We use assumptions:
 Mean path loss follows the general formula, i.e.

L ( r )= A+ 10 ⋅ n ⋅ log 10 ( r )

 Cell radius is R
 Users are uniformly distributed in the cell, i.e.
r
p(r , ϕ ) = , 0 ≤ r ≤ R, 0 ≤ ϕ ≤ 2π
πR 2

35
 Single cell coverage probability

The cell coverage probability is obtained by averaging

the local coverage probability over all possible mobile
positions. Hence, we must compute the integral
2π R
Fu = ∫∫ P cov (r ) p(r,ϕ )drdϕ
0 0
First we need to find formula for coverage probability
within a certain distance r.

36
 Single cell coverage probability

The coverage probability at distance r is given

by
Pcov (r) = P{L (r) + X > L (R) − SFM | r}
where lower bound is defined by the maximum
allowed path loss. We use now the equations:
L ( R ) − L ( r ) = A + 10 n log 10 ( R ) − ( A + 10 n log 10 ( r ))
R
= 10 n log 10  
r
L(r) = L (r) + X

37
 Single cell coverage probability

We obtain a form

Pcov (r ) = P{X > − SFM − 10n log10 ( r R ) | r}

= Q(− (SFM + 10n log10 ( r R )) / σ )

and thus, there holds

R
Fu = 2 ∫ Q(− (SFM +10⋅ n log10 ( r R)) / σ )rdr
2
R 0
Next task is to compute this integral

38
 Single cell coverage probability

By using the substitution

SFM 1 r
a=− , b = − 10n log10 e x = a + b ln  
σ σ R
we obtain
 x−a
r = R exp  
 b  r = 0 ⇒ x → −∞
1  x−a r =R⇒a
dr = R exp   dx
b  b 

and
2
a
 2( x − a) 
Fu = ∫ Q ( x ) exp  dx
b −∞  b 
39
 Single cell coverage probability

We proceed using integration by parts

∫ uv 'dx = uv − ∫ u ' vdx

 2( x − a) 
now u = Q ( x) v ' = exp 
b

 
1  1   2( x − a) 
u'= exp  − x 2  b
v = exp 
2π  2  
2  b 
and we get
x=a
2  b  2( x − a)  b 1
a
 x2 2 ( x − a )  

Fu = Q( x) exp   + ∫−∞ exp  − 2 + b dx 
b 2  b  2 2π
 x =−∞ 
1
a
 x2 2 ( x − a ) 
= Q(a ) + ∫
2π −∞
exp  − +
 2 b
 dx

40
 Single cell coverage probability

We can still go forward by completing the

squares:
x2 2 ( x − a ) 1 2  2   2   2a
2 2
1  2 4  2a 2
− + = −  x − x− = − x −2 x+  −  −
2 b 2 b  b 2  b  b   b   b
2 2
1 2  1  2  2a
=− x−  +   −
2 b 2b b
2  2 (1 − ab )
2
1
=− x−  +
2 b b2

Then cell coverage probability admit the form

 2 (1 − ab )  1  1 2 
a 2

Fu = Q(a) + exp 
 b 2 
 2π
∫−∞ exp  − 2  x − b  dx
 

41
 Single cell coverage probability
2

We still need to substitute t = x − Then
a−
2 b
 2 (1 − ab )  1 b
 1 2  2 (1 − ab )    2 
Fu = Q(a ) + exp 
 b 2 
 2π
∫−∞  2 
exp − t dx = Q ( a ) + exp 
 b 2 
 
1 − Q 

a −
b

 

and finally we are able to write the

 2 (1 − ab )    2 
Fu = Q(a ) + exp  2  1 − Q  a − 
 b   b 

SFM 1
a=− , b = − 10n log10 e
σ σ

Coverage probability on the edge of the cell 42

 Single cell coverage probability
1

Path loss exponent n=3
0.95

Shadow fading margin is
0.9
 6dB (x)
0.85
9dB (o)

Cell coverage probability


0.8
 12dB (*)
0.75

Remark: The SFM
difference between 95% and 0.7

80% coverage requirements is 0.65

large 0.6

0.55

0.5
-5 0 5 10 15
Shadow Fading Margin [dB]

43
 Single cell coverage probability

Path loss exponent n=4 1

Shadow fading margin is 0.95

 6dB (x) 0.9

 9dB (o)

Cell coverage probability

0.85
 12dB (*)
0.8

Remark: The SFM
difference between 95% and 0.75

80% coverage requirements is 0.7

even larger than in case n=3. 0.65

0.6

0.55
-5 0 5 10 15
Shadow Fading Margin [dB]

44
Fast Fading

45
 Recall: Fast Fading

Scatterers

Especially the changes in

component signal phases
a1 (t )e jφ1 ( t ) create rapid variations in
sum signal
a1 (t + t0 )e jφ1 ( t +t0 )

Sum signal at time t S (t ) = a1 (t )e jφ1 (t ) + ... + a5 (t )e jφ5 ( t )

Sum signal at time t+t0 S (t + t0 ) = a1 (t + t0 )e jφ1 ( t +t0 ) + ... + a5 (t + t0 )e jφ5 ( t +t0 )
46
 Fast fading

In link budget a fast fading margin
is needed because
 If fast power control is applied, then
some headroom is needed
especially in uplink since MS power
reservations are limited. If power
control fails, the whole uplink may
beak down
 Although link adaptation
(=adaptation of channel coding and
modulation) would be used instead
of fast power control, there can be
need for fast fading margin; fast
fading can be crucial for slowly Fading time scale depends
moving users on the user speed.

47
 Ideal fast power control

Fast power control aims to convert the channel

so that mean power of the signal in the receiver
is constant
Transmitted power
from MS

Ideal case Response of the

physical channel

Channel seen by
BS receiver (AWGN
channel)
Time
48
 Limited power control dynamics

At the cell edge MS power control starts to hit its maximum
value.
=> Number of erroneous frames is increasing
=> Data rate is decreased when QoS degrades
=> in the worst case connection breaks down

Longer and longer times with ‘bad connection’ temporal

length of the fades depends on the user speed
Maximum TX
power value

MS travels towards cell edge/coverage hole =>

mean transmission power needs to be increased
49