TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol. 14, No. 1, April 2015, pp. 154 ~ 1
6
2
DOI: 10.115
9
1
/telkomni
ka.
v
14i1.747
0
154
Re
cei
v
ed
Jan
uary 5, 2015;
Re
vised Ma
rch 13, 2015; A
c
cepted Ma
rch 25, 2015
Optimized Suitable Propagation Model for GSM 900
Path Loss Pred
iction
Sy
ahfrizal T
a
hcfulloh*, Eka Risk
a
y
adi
Dep
a
rtment of Electrical E
ngi
neer
i
ng, Un
iver
sitas Borne
o
T
a
raka
n,
Jl. Amal Lama
No.1 T
a
rakan, 771
23, Indo
ne
sia
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: rizalu
bt@gm
a
il.com
A
b
st
r
a
ct
T
h
is pa
per
pre
s
ent how
COS
T
-231 H
a
ta
mo
del
is
ch
osen
a
nd o
p
ti
mi
z
e
d
f
o
r
path l
o
ss pr
e
d
ictio
n
i
n
subur
ban
are
a
of Tarakan, In
don
esia
in th
e
GSM 900
MH
z
system.This
p
redicte
d
a
nd o
p
t
imi
z
e
d
path l
o
ss
m
o
del is based on the em
pirical
measur
em
ent coll
ected in the GSM
system
on Tarakan City. It
is
deve
l
op
ed
by
compari
n
g
the
calc
ulate
d
p
a
th l
o
ss fro
m
c
o
l
l
ected
meas
ur
ements w
i
th th
e w
e
ll-k
now
n
pa
t
h
loss mod
e
ls w
i
thin ap
plic
ab
le
frequency ra
n
ge of
GSM system, such as
COST-231 Hat
a
, Ericsson, SUI,
W
a
lfish, ECC-
33, and
Lee
Mode
l.
T
he COST
-231 Hata
mo
del w
a
s chose
n
bas
ed
on the cl
osest
an
d
sma
llest mea
n
error asco
mp
ared
to
th
e measur
ed pat
h
l
o
ss.
T
h
is opti
m
i
z
e
d
COST
-231 Hata mod
e
l
is
imple
m
ente
d
i
n
the path l
o
ss predict
i
ond
urin
g the vali
d
a
tion pr
ocess.
T
hus, this optimi
z
e
d
mode
l is
successfully im
pr
oved and
would be
m
o
re reliableto be
applied in t
he
TarakanG
SM900 MH
z
system
for
path loss pr
ed
i
c
tion.
Ke
y
w
ords
: pat
h loss pre
d
ictio
n
, opti
m
i
z
a
t
i
on,
COST
-231 Ha
ta,
propag
atio
n
mod
e
ls, GSM
Copy
right
©
2015 In
stitu
t
e o
f
Ad
van
ced
En
g
i
n
eerin
g and
Scien
ce. All
rig
h
t
s reser
ve
d
.
1. Introduc
tion
In mobile
rad
i
o sy
stem
s th
e ob
sta
c
le
s
bet
we
en th
e
base
station
(BS) a
nd th
e
mobile
station (MS)
signifi
cantly influen
ce
s the
stren
g
th of the mobil
e
si
gnal. The
attenuatio
n of the
radio
sig
nal i
s
refe
rred a
s
path loss. T
he path lo
ss predi
ction m
o
dels h
a
ve a
major
role in
the
radio frequ
e
n
cy cove
rag
e
optimizatio
n, interfer
e
n
ce analysi
s
a
nd efficient
utilization of
the
available
net
work
re
so
urces [1]. Th
e e
fficiency
of radio
netwo
rk plan
ning to
prod
uce a
Cost-
effective dep
loyment of GSM netwo
rk foroptim
al netwo
rk
cov
e
rag
e
largely depen
d on
the
degree
of accuracy
of the
pro
pag
ation
predi
ct
ion
mo
delempl
oyed
in ch
aracte
rizing the
uniq
u
e
feature
s
of th
e propa
gatio
n environme
n
t
whe
r
e th
e n
e
twork i
s
to
b
edepl
oyed. T
hus, th
e
choi
ce
of an a
dapta
b
le radio
pro
pagatio
n pat
h loss
model
plays
a pivo
tal role
in o
b
taining
ano
ptimal
netwo
rk pe
rfo
r
man
c
e
[2]. It is
req
u
ire
d
to
accurate
ly e
s
timate the
ch
annel
charact
e
risti
c
s in
ord
e
r
to maintain
the inte
rfere
n
c
e
at a mini
mum level. S
i
nce
the te
rrain
conditio
n
s
vary to
a l
a
rge
extent, the path loss p
r
edi
ction
model
s cannot be g
e
n
e
rali
zed. Th
i
s
drawba
ck
ca
n be overcom
e
by adjustin
g
the model p
a
rameters
to su
it the desire
d
environ
ment.
The
perfo
rma
n
ce
of a
n
y wi
rele
ss
comm
unication
syst
ems dep
end
s on th
e p
r
o
p
a
gation
cha
r
a
c
teri
stics of the ch
annel. Chan
nel ch
ara
c
te
ristics have
an impa
ct o
n
the desi
g
n of
thetran
s
mi
ssi
on strategy. Re
ceived
sig
nal and p
a
th
loss p
r
edi
cti
on model
s pl
ay an import
ant
role i
n
the
RF
cove
rag
e
o
p
timization
an
d
effici
ent use of
the
avail
a
b
l
e
re
so
urce
s. These
m
odel
s
can
differ in
their prope
rties
with lo
cations du
e t
o
different te
rrai
n
e
n
viron
m
ent. The
r
ef
ore,
extensive
study on th
e
effe
cts
ofra
dio p
r
op
agat
ion path
-
lo
ss had
dra
w
n
a con
s
ide
r
able
attention.
Suraju
deen
et a
l
.
[3] studie
d
com
p
a
r
atively three p
r
o
pagatio
n mod
e
ls, Hata, CO
ST 231,
and the
Lee
path lo
ss mo
del for
GSM
1800
and
WC
DMA System in Urba
n
are
a
of Ni
g
e
ria
based
re
ceiv
ed p
r
edi
cting
sign
al level
swing
s
fo
r va
rying site
s
an
d fre
quen
cy
and
wa
s fo
un
d to
be optimum f
o
rGSM 1
800
and very hi
gh for WCDMA.
Syahfrizal, in [4] also
pre
s
ente
d
u
s
ing
Oku
m
ura-Hat
a
propa
gatio
n mod
e
l to p
a
th lo
ss
dete
r
minatio
n at
900 M
H
z GS
M syste
m
s f
o
r
Tara
ka
n
city, whe
r
e Okum
ura
-
Hata
mo
del
was ado
p
t
ed
and modi
fied.
Similar work wa
s carried
out by Julie
et a
l
in [5] whereby they mo
dified Okumu
r
a-Hata a
nd
COST
-231
Hata model to path
loss p
r
edi
ctio
n at GSM
90
0 an
d 1
800
MHz for Po
rt
Ha
rcou
rt an
d Enug
u, Ni
g
e
ria,
whe
r
e
Root
Mean Squ
a
re
Error (RMSE
)
wa
s ag
ree
with the
acce
ptable Intern
a
t
ional ran
ge. Abrah
a
m De
me,
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Optim
i
zed Su
itable Prop
ag
ation Model f
o
r GS
M 90
0 Path Loss… (Syahfri
zal Ta
hcfullo
h)
155
in [6] the applicability of the COST
231
Hata Model to the metropo
lis of Maiduguri, Ni
geri
a
, is
tested by co
mputing vari
a
t
ions bet
wee
n
the CO
ST
231 Hata pre
d
iction
s an
d predi
ction
s
b
a
se
d
on the Lea
st
Square
s
fun
c
tion, bein
g
the be
st
fit curve thro
ugh
measu
r
e
d
d
a
ta points a
n
d
RMSE wa
s found to be
5.33dB, whi
c
h is acc
epta
b
le, the acce
ptable maxi
mum being
6dB.
Isabo
na, in
[7] wa
s
optimi
z
ed
path
lo
ss p
r
e
d
iction
usin
g O
k
u
m
u
r
a-Hata
mod
e
l for the
CDMA
800 MHz sy
st
em for urban
area in Be
nin
City, Nigeria.
In this pape
r, the data coll
e
c
tedexp
e
rim
e
nta
lly at 900 MHz b
and fo
r three ba
se st
ations,
locate
d in the
subu
rb
an re
gion of Ta
ra
kan city, Indon
esia a
r
e u
s
e
d
for path lo
ss analysi
s
. Dat
a
colle
cted a
r
e
use
d
to co
mpare the m
o
st wi
del
y used propa
gati
on mod
e
ls fo
r the pu
rpo
s
e of
comp
ari
s
o
n
and finding t
he most suitable mod
e
l
for the two tech
nolo
g
ies
that will assi
st
Enginee
rs in
carrying
out
e
ffective plan
n
i
ng for im
p
r
ov
ed
servi
c
e. T
h
is
wo
rk is dif
f
erent from th
e
aforem
ention
ed re
se
arch
work b
e
cau
s
e data colle
cted from d
r
ive tests a
r
e i
n
com
pari
ng
the
cho
s
e
n
propa
gation mo
del
s such u
s
CO
ST-231
Ha
ta,
Ericsson, SUI, Walfish, ECC-3
3
, and
Le
e
Model. Ba
se
d on the
sm
a
llest mea
n
e
r
ror
as
co
mpa
r
ed to th
e me
asu
r
ed
path l
o
ss, COST
-2
31
Hata mo
del i
s
found to b
e
the be
st suited path lo
ss p
r
e
d
ictio
n
model for o
p
timizing. Th
e
accuracy
of o
p
timized
path
loss thismod
e
l is e
nha
nce
d
by adju
s
tin
g
its pa
ram
e
ters, i
n
orde
r to
achi
eve mini
mum RMSE
betwe
en the
predi
cted a
n
d
the measu
r
e
d
values.
2. Empirical
Propaga
tion Path Los
s Models
In this pa
pe
r, we
studie
d
a
numbe
r of p
a
t
h loss mo
del
s for
pre
d
ictin
g
and
optimizing the
prop
agatio
n l
o
ss fo
r
GSM
900
M
H
z
system
s. In
all
model
s,
f
C
i
s
the carrie
r freque
ncy i
n
MHz
except for th
e ECC-33 m
odel inG
H
z,
d
is the dista
n
ce b
e
twe
e
n
the transmitt
er GSM Cell
BS
and the
re
cei
v
er MS user
in km for
all model
s exce
pt for Ericsso
n
model in
m
e
ters
(which i
s
fixed to 2.5 km in ou
r sim
u
lation), the
referen
c
e
dist
ance
d
0
is
10
0 m, the BS antenn
a hei
ght
h
BS
is e
qual
to 30m
for
all
model
s, the
MS anten
na
height
h
MS
is
equal
to 1.5
m for
all mo
d
e
ls,
G
BS
and
G
MS
are
BS and
MS
heig
h
t
an
tenna gain
fa
ctors cho
s
e
n
to
be 18dB,
a (h
MS
)
is the MS
antenn
a co
rrection fa
ctor.
The sh
ado
wi
ng margin
s
is ch
osen a
s
10 dB and a
dded in the
p
a
th
loss to all models in o
u
r si
mulation.
2.1. Simplifie
d Model
The sim
p
lifie
d Model is u
s
ed fo
r free
spa
c
e p
a
th loss. Free
sp
ace p
a
th loss (
PL
FS
) is
conve
n
iently expre
s
sed in
dB, as follows [3, 8, 11]:
PL
FS
= 32.45
+ 20 log
10
(d)
+ 20 log
10
(f
c
)
(1)
Whe
r
e
d
is in
km and
f
c
is in MHz.
2.2. COST-2
31 Ha
ta Mod
e
l
This
model
is derive
d
by m
odifying the
Hata
m
odel,
and i
s
u
s
ed
in
urb
an,
subu
rban, an
d
ruralenvironm
ents [1-2], [4], [7-9], [11]. Path
loss equ
a
t
ion for sub
u
rban area a
s
follows:
PL
COSTSU
= P
L
COSTU
- 2 (log
10
(f
c
/ 28))
2
- 5.4
(
2
)
PL
COSTU
= 46.3 + 33.9 log
10
(f
c
) - 13.82 lo
g
10
(h
BS
) +
(44.9 - 6.5
5
log
10
(h
BS
)) lo
g
10
(d)+
s – a
(h
MS
)
(3)
MS antenna
corre
c
tion fa
ctors
a(
h
MS
)
for all is:
a(h
MS
) =
(1.11
log
10
(f
c
) -0.7
)h
MS
- (1.56
lo
g
10
(f
c
) - 0.8)
(
4
)
2.3. Ericsso
n Model
The
planni
ng
network
eng
ineers u
s
e th
is m
odel
to
predi
ct th
e p
a
th lo
ss in
suburba
n
area, as
follows
[8]:
PL
Ericsson
= 43.2 + 68.93lo
g
10
(d) + 12lo
g
10
(h
BS
) + 0.1log
10
(h
BS
)log
10
(d)
– 3.2(log
10
(1
1.75 h
MS
)
2
) +
G(f
c
)
(5)
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 14, No. 1, April 2015 : 154 – 16
2
156
G(f
c
) = 4
4
.49l
og
10
(f
c
) – 4.78(log
10
(
f
c
))
(
6
)
2.4. Stanfor
d
Univ
ersit
y
Interim (SUI
) Model
Stanford
Uni
v
ersity Interi
m (S
UI)
ch
annel
model
is
developed for IEEE 802.16
broa
dba
nd
wi
rele
ss a
c
cess wo
rki
ng g
r
ou
p ba
sed
on
r
e
se
ar
ch
re
sul
t
s of
S
t
anf
or
d
Univ
e
r
sit
y
[
1
].
This m
odel
covers thre
e
comm
on te
rrain catego
rie
s
. Cate
go
ry A is the m
a
ximum path
-
loss
categ
o
ry, which
rep
r
e
s
ent
s a hilly te
rra
in with m
ode
rate to heavy
tree d
e
n
s
ities. Catego
ry B
is
the interme
d
iate path-lo
ss categ
o
ry suit
able for
flat terrai
n
s. The mi
nimum path
-
l
o
ss cate
gory
for
flat terrain
s with less tree
den
sities i
s
Category
C. The ba
sic
path
loss equ
atio
n for SUI mo
del
with c
o
rrec
tion fac
t
ors
is
given in [1-2], [7], [10-11] as
follows
:
PL
SUI
= S
+ 10
log
10
(d/d
0
) +
L
fc
+
L
hBS
+
s
(7)
W
h
er
e
d
i
s
i
n
m
,
d
0
=
1
00m,
i
s
th
e
path-lo
ss
expone
nt,
L
fc
is
the
c
o
rrec
t
i
on fac
t
or for the
freque
ncy,
L
hBS
is the co
rrection fa
ctor f
o
r
the
re
ceive
r
anten
na h
e
i
ght and
s
i
s
the log
norm
a
l
l
y
distrib
u
ted
sh
ado
w facto
r
d
ue to the
tree
s a
nd
oth
e
r
o
b
sta
c
le
s, havi
ng a val
ue
b
e
twee
n 8.2
dB
and 10.6
dB [1]. The term
S
, the path loss expon
en
t and the co
rrectio
n
facto
r
s in the ab
ove
equatio
n are
given as
S = 20 log
10
(4
d
0
/
)
(
8
)
= u - v
h
BS
+
(w/h
MS
)
(
9
)
L
fc
= 6 log
10
(f
c
/2000)
(
1
0
)
L
hBS
=
10.8
,
20
(
1
1
)
Whe
r
e
is th
e wavel
ength
(m),
f
c
is th
e
freque
ncy
(MHz),
h
BS
is t
he hei
ght of the ba
se
stati
on
(m),
h
MS
is th
e height of th
e re
ceiving
a
n
tenna
(m).
The pa
ram
e
ters
u
,
v,
a
nd
w
are sta
n
d
a
rd
values that d
epen
d on the type of terrain. Since
the terrain is
categ
o
ri
zed
as subu
rba
n
, an
interme
d
iate path loss (Ca
t
egoryB) is
chos
en for an
alysis. The v
a
lue
s
of con
s
tants
u
,
v
, and
w
are 4, 0.006
5
,
and 17.1 re
spectively.
2.5. Walfish
Model
The
Walfi
s
h
Model
is u
s
e
d
for fre
e
sp
ace, l
a
rge
a
nd m
edium
cities [8]. Pa
th loss
equatio
n for suburban a
r
e
a
as follows:
PL
W
a
lfish
= P
L
W
a
lfish_FS
+P
L
rts
+
P
L
ms
d
(
1
2
)
Scena
rio 1: F
r
ee Spa
c
e Pa
th loss.
PL
W
a
lfish_FS
= 32.4 + 20lo
g
10
(d)
+ 20lo
g
10
(f
c
)
+
s
(12a
)
Scena
rio 2: F
r
ee Spa
c
e in
a deep valley
Path loss.
PL
W
a
lfish_FS
= 42.6 + 26lo
g
10
(d)
+ 20lo
g
10
(f
c
)
+
s
(12b
)
Whe
r
e
PL
W
a
lfi
s
h_FS
is the free spa
c
e lo
ss,
PL
rts
is the ro
of-to-stre
et diffr
action an
d scatter loss,
and
PL
ms
d
is
the
multi s
c
reen diffrac
tion loss
.
PL
rts
= - 16.9 -10log
10
(w) +
10log
10
(f
c
) +
20log
10
(h
roof
– h
MS
) +
PL
ori
+ s
(12
c
)
h
roof
= 3n
floor
+ roof (m
)
(
1
2
d
)
PL
ori
=
4 – 0.114 (
-
55
)
(
1
2
e
)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Optim
i
zed Su
itable Prop
ag
ation Model f
o
r GS
M 90
0 Path Loss… (Syahfri
zal Ta
hcfullo
h)
157
Whe
r
e
w
is the width of the street in m and eq
ual
b/2
,
b
is the building
se
pa
ration eq
ual to 40
min ou
r
simul
a
tion,
i
s
th
e ro
ad o
r
ie
ntation with
re
spect to th
e di
rect radio
path
in de
gre
e
s a
nd
equal to 90
o
i
n
the simulati
on, roof heig
h
t is 3 m and
numbe
r of floors is 3. Th
e formula fo
r the
multi s
c
reen diffrac
tion loss
is
as
follows
:
PL
ms
d
= P
L
hBS
+k
a
+k
d
log
10
(d
) +
k
f
log1
0 (f
c
) – 9log
10
(b)
(
1
2
f
)
PL
hBS
= -18 lo
g
10
(1 +
∆
h)
(
1
2
g
)
k
f
= -
4
+ 0.
7
(
f
c
/925 - 1)
(
1
2
h
)
Whe
r
e
k
a
i
s
the acco
unt for the incre
a
s
e in pat
h lo
ss
whe
n
the BS antenna
s are belo
w
the
rooftop
s
of
a
d
jacent b
u
ildi
ngs,
and
eq
u
a
l to 5
4
, whil
e
k
d
whi
c
h
eq
ual
to 18 and
k
f
a
r
e
to
c
o
ntr
o
l
the depe
nde
n
c
y of the multi-screen diffra
c
ti
on lo
ss o
n
the dista
n
ce a
nd frequ
en
cy.
2.6. ECC-33
Model
The E
C
C-3
3
mod
e
l d
e
vel
oped
by the
Electro
n
icCo
mmuni
cation
Committee
(ECC) i
s
approp
riate for su
bu
rba
n
and sm
all urb
an are
a
s [8, 1
0
].
PL
ECC
= P
L
FS
+ P
L
bm
– G
BS_
ECC
– G
MS _
E
C
C
(
1
3
)
PL
FS
= 92.4 +
20 log
10
(d)
+ 20 log
10
(f
c
)
(
1
3
a
)
PL
bm
= 20.41+9.83l
og
10
(d
)+7.89
4log
10
(f
c
)+9.5
6
(l
og
10
(f
c
))
2
(13b
)
G
BS_ECC
=log
10
(h
BS
/200)(1
3.98 + 5.8 (lo
g
10
(d)
)
2
)
(
1
3
c
)
G
MS_ECC
= (42.57 + 13.7 lo
g
10
(f
c
))(log
10
(h
MS
) - 0.585)
(
1
3
d
)
Whe
r
e
PL
FS
and
PL
bm
are the free
spa
c
e
path loss an
d the basi
c
m
edian p
a
th lo
ss.
2.7. Lee Mod
e
l
Lee’
s Mod
e
l i
s
u
s
ed to
pre
d
ict the p
a
th
loss in u
r
ba
n
,
subu
rba
n
, rural a
nd free
spa
c
e
area
s [8]. Path loss equ
ation for subu
rb
an are
a
as fol
l
ows:
PL
Lee
= 99.86 +38.4 lo
g
10
(d
) +10
n
log
10
(f
c
) –(h
BS
/30.48
)
2
– (h
MS
/3)
2
– (P
t
/10)
2
– (G
BS
/4) – G
MS
(
1
4
)
Whe
r
e
n
i
s
a
n
expe
riment
value
cho
s
e
n
to be
3. In
our
sim
u
latio
n
, the p
a
ram
e
ter valu
es o
f
P
t
,
G
BS
, and G
MS
are 43 dBm,
18 dBm, and
18 dBm re
sp
ectively.
3. Materials
and Method
3.1. Descrip
tion of the
Ar
ea under Inv
estiga
t
ion
Tara
ka
n
cit
y
i
s
lo
cat
e
d
wit
h
in
tropi
cal
ra
iny re
gion. Its terrain
clutte
r i
s
cha
r
a
c
teri
zed
by
the availability of tropical rainy forest, h
ouses
mo
stly below 20 me
ters an
d an a
v
erage road
with
of about
20
meters. Attenuation i
s
caused by m
u
ltiple reflecti
ons, a
b
sorption an
d multi
p
le
diffraction
s o
ff roof tops, trees, cars e
t
c.
The con
c
rete ground
and tarred road
s have very
relative poo
r electri
c
al co
ndu
ctivity, an
d theref
o
r
e, cau
s
e attenu
ation by absorption. G
r
ou
nd
reflecte
d wav
e
s are blo
c
ke
d by building
s
and tree
s.
3.2. Measure
ment Proce
d
ure
Measurement
s were ta
ke
n from three
differ
ent Ba
se Station
s
of a mobile
netwo
rk
servi
c
e p
r
ovi
der in
GSM 9
00 MHz
Ne
t
w
ork, situ
ate
d
within the t
e
rr
ain. The te
sting tool u
s
e
d
in
the mea
s
u
r
e
m
ent was
GSM test ph
one
hand
set in th
e Net M
onitor mode
ca
pabl
e of mea
s
u
r
in
g
sign
al stre
ngt
h (
P
r
) in de
ci
bel milliwatts (dBm), in co
njun
ction wit
h
a Digital Gl
obal Positio
n
i
ng
System receiver ante
nna
to dete
r
min
e
distan
ce
(
d
) f
r
om th
e Ba
se
Station (BS
)
. Re
adin
g
s were
taken
within
the 900M
Hz
freque
ncy ba
nd at interv
al
s of 0.2 km, after an initia
l sepa
ration
of
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 14, No. 1, April 2015 : 154 – 16
2
158
0.1km a
w
ay from the Ba
se
Station up to 2.5 km
fixed length. Base
St
ation para
m
eters obtain
e
d
from GSM 9
00 MHz
Net
w
ork Provide
r
such
us
m
ean tra
n
smitter hei
ght (
h
BS
)
i
s
3
0
m
,
m
e
a
n
Effective Isotropi
cally
Radi
ated Po
wer (
EIRP
)is
43
d
B
m, transmitting fre
que
ncy
for BS1, BS
2,
and BS3
a
r
e
902.5,
903.
2, and
90
4.5
MHz
re
spe
c
tively. Receiv
ed p
o
wer (
P
r
) v
a
lue
s
w
e
r
e
recorded
at variou
s
dista
n
c
e
s
fro
m
ea
ch of the th
ree
BS named
BS1,BS2, and
BS3. For eve
r
y
received po
wer value, the corre
s
p
ondin
g
path lo
ss m
easure
d
wa
s
comp
uted u
s
i
ng the formul
a:
PL
Measure
d
= EI
RP
- P
r
(
1
5
)
3.3. Path Lo
ss Model Op
timization
Several exi
s
ting path l
o
ss
mod
e
ls
wa
s explai
n
ed in Se
ctio
n 2 a
r
e
ch
ose
n
for
comp
ari
s
o
n
with mea
s
u
r
e
m
ent data pa
th loss. Th
e
best existin
g
path loss mo
del with
smal
lest
mean e
r
ror to
the mea
s
u
r
e
d
path lo
ss d
a
ta will b
e
ch
ose
n
a
s
a
ref
e
ren
c
e fo
r th
e develo
p
me
nt of
the optimi
z
e
d
path
lo
ss
model. T
he
optimize
d
p
a
th loss
mod
e
l will
be
te
sted
du
ring
the
validation pro
c
e
ss by
com
parin
g the RMSE calc
ulat
ed path lo
ss
to the measu
r
ed p
a
th loss in
Tara
ka
n CityGSM 900 M
H
z syste
m
.
Path loss m
o
del optimization is a proce
ss in
whi
c
h a
theoretical p
r
opa
gation m
odel is
adju
s
ted
with
thehelp
of measure
d
valu
es o
b
taine
d
from te
st field
data. The ai
m is to get t
he
predi
cted
fiel
d st
ren
g
th a
s
clo
s
e
a
s
po
ssi
ble to
the
measured fiel
d st
ren
g
th. P
r
opa
gation
p
a
th
loss model
s o
p
timized fo
r d
i
fferent wirele
ss te
chn
o
logi
es an
d enviro
n
ments a
r
e a
bund
ant in [7].
In orde
r to optimize an
d validate the effe
ctivene
ss of the propo
sed mo
del, the Mean
error (
μ
e
), an
d RMSE (
e
) were
cal
c
ula
t
ed betwe
en the results of
the propo
se
d clo
s
e
s
t model
and the me
a
s
ured p
a
th lo
ss d
a
ta of ea
ch a
r
ea. Th
e
s
e me
an e
rro
r (
μ
e
), and
ro
ot mean squ
a
re
error, RMSE
(
e
) a
r
e defin
ed by the expressio
n
in (1
6
)
, and (1
7) re
spe
c
tively.
μ
e
= (1/N)
(PL
Measure
d
– PL
Predicted
)
(
1
6
)
e
=
(
1
7
)
Whe
r
e
PL
Measured
is measured Path lo
ss (dB),
PL
Predicted
ispredi
cted path loss (dB), and
N
is
numbe
r of measure
d
data
points.
4. Results a
nd Analy
s
is
Figure 1
sh
ow that BS
tran
smitter ant
enn
a h
e
ight in
cre
a
sed cau
s
ed
path lo
ss
decrea
s
e
d
. The path lo
ss of COST-23
1
Hata mo
de
l sho
w
s
de
creasi
ng tre
nd
with re
sp
ect
to
transmitter
a
n
tenna
hei
ght
and
in
crea
si
ng trend
with
respe
c
t to t
r
a
n
smi
ssi
on
di
stance. In
Fig
u
r
e
2, increa
sin
g
ly of freque
ncy in mobile
n
e
twork
syste
m
that ca
use
d
pat
h l
o
ss i
n
crea
sed. T
h
e
fluctuation
s
o
f
the sign
al levels a
s
the
results
of fadi
ng. As we drove from the
starting
point
of
the d
r
ive te
st, the receiver po
we
r
cha
n
ges si
gnifi
ca
ntly.
This
i
s
becau
se of multipath
fa
ding;
meanin
g
tran
smitted si
gna
l takes m
u
ltip
le paths
to th
e re
ceiver. T
he re
ceived
signal amplitu
de
at the m
obile
ch
ang
es wit
h
its
po
sition.
Both
d
e
fect
s the
sam
e
p
a
ttern a
s
we
moveaway fro
m
the transmitter, the received sign
al am
plitude deg
ra
de. Same re
sults
were ob
tained for all the
BTSs examin
ed throu
gho
u
t
the drive test. Except
for some ex
cepti
onal ca
se
s where, the sig
nal
amplitude d
e
c
re
ase with i
n
crea
se in di
stan
ce an
d la
ter increa
se f
o
r short di
sta
n
ce a
s
a resu
lt
o
f
line of
sight
b
e
twee
n the B
S
and
MS, po
ssi
bly due
to
valley in bet
ween
or
we
drove high
the
hill
and then d
e
scen
d.
The m
easure
d
sig
nal
stre
n
g
th (
P
r
)
res
u
lt
s
of three BS
(BS1, BS2, and BS3) in this
s
t
udy
were p
r
e
s
ent
ed graphi
call
y in Figure 3.
The value
s
o
f
the predi
cte
d
path lo
ss a
nd the me
asu
r
ed
path lo
ss
we
re
plotted
ag
ainst th
e di
stance
of
sep
a
ration
bet
ween th
e Ba
se Station
(B
S)
antenn
a an
d
the Mo
bile
Station (MS) anten
na. Fi
gure
4
pre
s
e
n
ts the
re
sult
of the
path
loss
predi
ction
s
a
s
co
mpa
r
ed
with the path
loss m
e
a
s
ured on BS1 p
r
opa
gation e
n
vironm
ent. The
result of the drive test carri
ed out within
t
he coverage
area of the cell repo
rted a
mean path lo
ss
of 120.23 dB
m. Mean pat
h loss value
s
of
115.80 d
B
m, 145.13 d
B
m, 91.38 dBm, 154.81 d
B
m,
126.60
dBm, 157.52
dBm, and 1
45.98
d
B
m we
re p
r
e
d
icted
by CO
ST-231
Hata
model, Eri
c
sson
model, Free
Space
mod
e
l, SUI mod
e
l, Wa
lfish
model, ECC-33 mod
e
l, a
nd Le
e mod
e
l
r
e
spec
tively.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Optim
i
zed Su
itable Prop
ag
ation Model f
o
r GS
M 90
0 Path Loss… (Syahfri
zal Ta
hcfullo
h)
159
Figure 1. Path loss of CO
ST-231
Hata
model
with re
spe
c
t to heig
h
t of BS1 transmitter
antenn
a
Figure 2. Vari
ation of COS
T
-23
1
Hata
pa
th
loss model
wi
th frequen
cy
Figure 3. Measu
r
ed
re
ceiv
ed po
wer
(
P
r
) in
BS1, BS2, an
d BS3
Figure 4. Co
mpari
s
o
n
of measured an
d
predi
cted p
a
th loss for BS1
Figure 5
sho
w
the
path
l
o
ss of th
etra
nsmitted
si
gn
al in th
e te
rrain
covered
by BS2.
COST
-231
Hata model, Ericsso
n
mod
e
l, Free Spa
c
e
model, SUI m
odel, Walfi
s
h
model, ECC-33
model, a
nd
L
ee mo
del
predicte
d
me
a
n
path l
o
ss
v
a
lue
s
of 1
15.
84 dBm, 1
4
5
.
15 dBm, 91.
41
dBm, 154.8
5
dBm, 1
26.6
5
dBm,
140.
76 dBm,
a
n
d
146.0
2
dBm
re
sp
ectively. The
me
an
p
a
th
loss obtain
e
d
from the p
r
opag
ation en
vironme
n
t wit
h
in the radio
coverage
of the BS2 wa
s
122.15 dBm.
The a
nalysi
s
of the d
r
ive te
st data
of the
BS3
as prese
n
ted in
Figu
re 6
rep
o
rted
that the
mean p
a
th loss cal
c
ul
ate
d
by the mo
bile us
ers in
that area
covered i
s
12
0.15 dBm. T
h
e
predi
ction
s
of
COS
T
-2
31 Hata mod
e
l, Ericsson
mod
e
l, Fre
e
Sp
ace mo
del
, S
U
I model, Walfish
model, ECC-33 mod
e
l, an
d Lee mo
del
gave mea
n
value
s
of 115.
86 dBm, 145.
16 dBm, 91.4
2
dBm, 154.86
dBm, 126.67
dBm, 140.78
dB
m, and 14
6.04 dBm re
spectively.
Figure 5. Co
mpari
s
o
n
of measured an
d
predi
cted p
a
th loss for BS2
Figure 6. Co
mpari
s
o
n
of measured an
d
predi
cted p
a
th loss for BS3
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 14, No. 1, April 2015 : 154 – 16
2
160
The findin
g
s
of this re
se
arch
sho
w
ed th
at
all the sev
en empi
rical
prop
agatio
n
predi
ction
model
s investigated unde
r the measu
r
ed path lo
ss of GSM 900 MHz de
ploy
ed in this area.
COST
-231
Hata mod
e
l h
a
s m
ean
error p
a
th lo
ss
values
with
smallest
and
clo
s
ets f
r
om
the
actual
path l
o
ss measured on the ‘liv
e’ GSM 90
0 network deployed in
the area. Path l
o
ss
estimated
by
Co
st-23
1
Hat
a
mo
del
matches with
the
measured
pa
th loss in
a m
u
ch
bette
r
wa
y,
comp
ared to
othe
r p
r
e
d
iction mo
del
s. Similar re
su
lts a
r
e
ob
se
rved for othe
r ba
se
statio
ns.
Selection of best path lo
ss pre
d
ictio
n
model for
opt
imizing that is req
u
ire
d
to identify the
best
predi
ction m
odel, so tha
t
this model
can b
e
adj
usted to a
c
hieve minim
u
m RMSE
with
themea
su
red
data. By using Equ
a
tio
n
(16),
the
mean erro
r of all the seven empiri
cal
prop
agatio
n path loss mo
del to measu
r
ed path lo
ss
has cal
c
ul
ated in Table
1. The small
e
st
mean erro
rs
for BS1, BS2
, and BS3 with the predi
cted COST
-23
1
Hata mod
e
l
are4.43
dBm
,
6.31dBm, an
d 4.29dBm re
spe
c
tively. From the re
sult
s
of Table 1, it is obse
r
ved
that the avera
ge
mean e
r
ror for COST-2
31
Hata mod
e
l is the lea
s
t, comp
ared to other mo
del
s. This indi
cat
e
s
that path l
o
ss isb
e
st
predicted by
COST
-231
Hata m
odel. Ba
sed
on thi
s
re
sult,COST
-
231
Hata
model is
sele
cted a
s
the suitable mod
e
l
for optimizin
g pro
c
e
ss.
Path loss mo
del optimi
z
ati
on is a
proce
ss
i
n
whi
c
h
a
n
empi
rical
p
r
opa
gation m
odel i
s
adju
s
ted
with
thehelp
of measure
d
valu
es o
b
taine
d
from te
st field
data. The ai
m is to get t
he
predi
cted
pat
h lo
ss as cl
o
s
e
as po
ssib
le to
the
me
asu
r
ed
path
loss. Pro
pag
ation p
a
th lo
ss
model
s optim
ized for diffe
rent wirel
e
ss tech
nolo
g
ies
and environm
ents are abu
n
dant in [7].
The cal
c
ul
ate
d
RMSE with
the field measur
ed data th
roug
hout the
three covera
ge are
a
investigate
d
by usin
g Eq
uation (17
)
. The RM
SE function
com
putation of t
h
is resi
dual
is
cal
c
ulate
d
b
a
se
d on the
least
squa
red algo
rithm
whi
c
h is
used to dete
r
mine the residual
minimum val
ues. Simila
r
to the wo
rk
by [4] and [7
], the RMSE is then
su
btracte
d
fro
m
the
Equation (2)
of COST-231
Hata mod
e
l to obtai
n opti
m
ized
path lo
ss m
odel
s for all BS sites in
the locatio
n
of study as given in Equatio
n (18
)
. Figure
7-9 belo
w
ill
ustrate
s
ho
w
measured pa
th
loss mo
del
s
have b
een
op
timized
with
COST
-231
Hata mod
e
l in
this
pap
er.
He
re, the
optimi
z
e
d
path loss mo
del for each o
perato
r
wa
s a
pplied for
pat
h loss
cal
c
ula
t
ion for other
base station
s
in
all the study
location, to
verify the accuracy
and t
he suitability of this opti
m
ized
path l
o
ss
model
s. In Table 2 shown that
all the base
station
s
fit into the optimized m
odel with lo
wer
averag
e RMS
E
was ab
out 4.87 dB and
still in the a
cceptable rang
e
is up to 6 dB [4]. From these
results
as de
picted i
n
Fig
u
re
7-9, it i
s
sh
o
w
n
that
the optimi
z
e
d
mod
e
l do
e
s
sho
w
a
go
od
agre
e
me
nt for the entire
studie
d
BS sites co
mpa
r
e
d
with COST
-231
Hata m
odel. Thu
s
the
optimize
d
mo
del is succe
s
sfully devel
op
ed with prope
r optimized proce
dure.
Table 1. Co
m
pari
s
on of me
an error
(
μ
e
)
Base Station
CO
STS
U
Ericsson
FS
SUI
Walfish
ECC-33
Lee
BS1 4.43
24.90
28.85
34.57
6.37
37.28
25.75
BS2 6.31
23.00
30.75
32.69
4.50
18.61
23.87
BS3 4.29
25.01
28.73
34.71
6.52
20.63
25.89
Average
5.01
24.30
29.44
33.99
5.80
25.51
25.17
Figure 7. Optimized a
nd p
r
edicte
d
path l
o
ss
for BS1
Figure 8. Optimized a
nd p
r
edicte
d
path l
o
ss
for BS2
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Optim
i
zed Su
itable Prop
ag
ation Model f
o
r GS
M 90
0 Path Loss… (Syahfri
zal Ta
hcfullo
h)
161
Figure 9. Optimized a
nd p
r
edicte
d
path l
o
ss for BS3
Table 2. Co
m
pari
s
on of RMSE (
e
) for
prop
osed mo
del perfo
rma
n
ce b
e
fore a
n
d
after
optimizatio
n
Base Station
Before
After
BS1 5.85
4.08
BS2 6.75
2.43
BS3 8.27
8.07
Average
6.96
4.87
4. Conclusio
n
In this
pap
er,
the m
e
a
s
ured p
a
th lo
sses i
n
th
ree
cells
are
com
pare
d
with th
eoreti
c
al
path loss mo
dels: COST-231 Hata m
odel, Ericsso
n
model, Fre
e
Space mo
del, SUI model,
Walfish mode
l, ECC-33 m
odel, and Le
e model. The
measu
r
ed p
a
th loss, whe
n
comp
are
d
with
theoreti
c
al va
lues from the
theoreti
c
alm
odel
s, sh
owe
d
the clo
s
e
s
t
agre
e
me
nt wi
th the path lo
ss
predi
cted
by the COST
-23
1
Hata m
ode
l in terms
of mean
squ
a
re error
analy
s
is. Ba
sed
o
n
this,an
optimi
z
ed
COST-2
31
Hata m
o
d
e
l for th
ep
re
diction
of p
a
th lo
ss expe
ri
enced
by GS
M
sign
als in the
900MHz ba
n
d
in sub
u
rb
a
n
environ
men
t
of Taraka
n City, Indonesi
a
is develo
p
e
d
.
The o
p
timize
d mod
e
l
sho
w
ed
hig
h
a
ccura
cy an
d
i
s
able to
predi
ct path lo
ss
withsmall
e
r average
RSMEwa
s ab
out 4.87 dBm
a
s compa
r
e
d
to the
COST-231 Hata mo
del wa
s ab
ou
t 6.98 dBm.
Referen
ces
[1]
A Bhuva
nesh
w
ari, R Hema
lat
ha, T
Saty
as
av
ithri.
Statistical
T
unin
g
of the
Best suited Pre
d
i
ction Mo
del
for Measure
m
e
n
ts ma
de in Hy
dera
bad C
i
ty o
f
Southern Ind
i
a.
Proceed
in
gs
of the W
o
rld Con
g
ress on
Engi
neer
in
g an
d Comp
uter Scienc
e. San F
r
ancisco. 20
13; 2
.
[2]
Segu
n IP, Ola
s
unka
n
mi F
O.
Empirica
l P
a
th
Loss
Mo
dels
for GSM N
e
t
w
o
r
k De
plo
y
m
ent
in
Makur
d
i
,
Niger
ia.
Intern
ation
a
l Ref
e
ree
d
Journ
a
l of En
gin
eeri
ng a
nd
Scienc
e (IRJES)
. 2014; 3: 85
-94.
[3]
NT
Surajude
en
Bakind
e, N F
a
ruk, AA A
y
e
n
i,
MY Muhamma
d, MI Gumel. Comp
ariso
n
of
Propag
atio
n
Mode
ls for GSM 1800
and W
CDMA S
y
st
ems in Select
ed U
r
ban Are
a
s of Niger
ia.
Intern
ation
a
l Jo
urna
l
of Appli
ed Infor
m
ati
on Syste
m
s (IJAIS).
2012
; 2(7): 6-13.
[4]
Sy
ahfrizal
T
.
U
s
ing
of the Ok
umura-H
a
ta Pr
opa
gati
on Mo
d
e
l for Path
loss
Deter
m
i
natio
n
in T
a
rak
a
n
.
Procee
din
g
s o
f
the 3
rd
Internatio
nal C
onfe
r
ence o
n
Rad
a
r,
Antenna,
Micro
w
av
e, Electronics a
n
d
T
e
lecommunic
a
tions (ICRAM
E
T
)
. 2014: 77-
79.
[5]
Julie
CO, MU
Onuu,
JO Ush
i
e, BE
Usi
be.
Pr
opa
gati
on M
ode
ls for
GSM 90
0
and
1
8
0
0
MHz for
Por
t
Harco
u
rt and E
nug
u, Nig
eria.
Netw
ork and C
o
mmunic
a
tio
n
T
e
chno
log
i
es
. 201
3;
2(2).
[6]
Abrah
a
m D, Danj
uma D, Bub
a
B, Mohammed M.
Hata-Okumura Model
Comput
er Anal
ysis for Pat
h
Loss D
e
termi
n
a
tion
at 9
00M
Hz for Ma
id
ug
uri, Ni
geri
a
.
Int
e
rnati
ona
l Insti
t
ute for Sci
enc
e, T
e
chn
o
lo
gy,
and Ed
ucati
on
(IIST
E
): Mathematic
al T
h
e
o
ry
and Mod
e
li
ng.
2013; 3(3).
[7]
Isabon
a Jose
p
h
, Kon
y
e
ha C
C
. Urban Ar
ea
Path loss Pro
pag
atio
n Predi
ction an
d Opti
mizatio
n
Usin
g
Hata Mod
e
l at 800M
Hz.
IOSR Journa
l of App
lied P
h
ysics (IOSR-JAP)
. 2013; 3: 8-18.
[8]
Mohamm
ed A,
T
ughrul A, J
o
hn T
,
Ahmet T
E
.
F
r
eque
ncy
Analys
is of P
a
th L
o
ss Mo
dels
on W
I
MAX
.
3rd Com
puter
Scienc
e an
d El
ectronic En
gin
eeri
ng Co
nfere
n
ce (CEEC). U
K
. 2011.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 14, No. 1, April 2015 : 154 – 16
2
162
[9]
Olasunk
anmi
F
O, Segun IP
, Rob
e
rt OA, Olu
w
ol
e
AA. Comp
arative
Anal
ys
is
of R
e
ceiv
ed
S
i
g
nal
Strength Pr
ed
i
c
tion M
ode
ls
for Ra
dio
Net
w
o
r
k Pl
an
nin
g
of GSM 9
0
0
MHz
in Il
o
r
in, Ni
geri
a
.
Internatio
na
l Journ
a
l of Innov
ative T
e
chn
o
l
o
gy and Exp
l
ori
ng Eng
i
n
eeri
n
g
(IJIT
EE).
2014
; 4(3): 45-50.
[10]
Sachi
n
SK, AN Jadh
av. An
Empirica
ll
y B
a
sed P
a
th Los
s Models for
LT
E Advanced
Net
w
ork a
n
d
Mode
lin
g for 4G W
i
reless
S
y
stems at
2.
4 GHz, 2.6 GHz and
3.5 GHz.
Internatio
n
a
l Jour
nal
of
Appl
icatio
n or Innov
atio
n in E
ngi
neer
in
g & Mana
ge
me
nt (IJAIEM)
. 2013; 2(9): 252-2
57.
[11]
Reka
w
t
SH, T
A
Rahma
n
, AY Abdulr
ahm
a
n
. LT
E Coverage Net
w
o
r
k Pl
ann
ing
and
Co
mpariso
n
w
i
t
h
Different Pro
p
a
gatio
n Mo
dels.
T
E
LKOMNIKA T
e
leco
mmunic
a
tion
Co
mp
utin
g Electro
n
ics
a
nd C
ontrol
.
201
4; 12(1): 15
3-16
2.
Evaluation Warning : The document was created with Spire.PDF for Python.