TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol. 12, No. 8, August 201
4, pp. 6281 ~ 6290
DOI: 10.115
9
1
/telkomni
ka.
v
12i8.618
7
6281
Re
cei
v
ed Ma
rch 1
5
, 2014;
Re
vised April
30, 2014; Accepte
d
May 1
4
, 2014
Load Balancin
g Based on the Specif
i
c Offset of
Handover
Zhanjun Liu*
1
, Qichao Ma
2
, Cong Ren
3
, Qianbin Ch
en
4
Cho
ngq
in
g Ke
y L
ab of Mobi
le
Communic
a
tio
n
, Chon
gq
ing
Univers
i
t
y
of Posts and T
e
lecommunic
a
tio
n
,
Cho
ngq
in
g 40
0
065, Ch
in
a
*Corres
p
o
ndi
n
g
author, em
ail
:
liuzj@cq
upt.e
du.cn
1
, hn
xcm
q
c@1
26.com
2
A
b
st
r
a
ct
Loa
d b
a
l
anci
n
g (LB)
techn
o
l
ogy
in
mobi
le
w
i
reless c
o
mmunic
a
tion
n
e
tw
orks h
a
s b
e
e
n
discuss
ed
larg
ely. T
he c
u
rrent LB
alg
o
rit
h
ms
hav
e
ma
i
n
ly a
d
j
u
sted th
e ha
nd
over
par
ameters w
i
tho
u
t consi
der
ing
the
inh
e
rent r
e
l
a
tio
n
shi
p
of
the
ha
ndov
er p
a
ra
me
ters. In the
pa
per, by
co
nsid
erin
g the
i
n
tern
al r
e
lati
ons
hip
of
specific
offset of han
dov
er, the constr
a
i
nt of
the spec
ific of
f
s
et of ha
ndov
e
r
w
a
s simp
lifi
e
d, so the
proce
ss
of mob
ility
loa
d
ba
lanc
in
g (ML
B) al
gorith
m
w
a
s i
m
pr
ove
d
.
With the
impro
v
ed M
L
B a
l
g
o
ri
thm, th
e n
u
mb
er of
han
dov
er p
a
ra
meters
w
a
s re
duce
d
a
n
d
the
sign
al
proc
ess
w
a
s simplifi
ed.
Si
mul
a
tio
n
res
u
lts sh
ow
ed th
at
the cong
estio
n
rate w
a
s reduc
ed, the reso
urc
e
utili
z
a
t
i
on rat
e
w
a
s impr
ove
d
and th
e Qos w
a
s improv
ed.
Ke
y
w
ords
:
ML
B, specific offset of hand
over,
resource uti
l
i
z
ation, opti
m
i
z
a
t
ion
Copy
right
©
2014 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 wirele
ss mobile com
m
unication n
e
twork, the arrival
s
of mobile u
s
ers and the
resulting traffic load a
r
e ra
ndom, time-v
arying,
and o
ften unbalan
ced, which ma
ke cell load o
u
t
of balan
ce [1]. With the third ge
neration mob
ile
co
mmuni
cation
techn
o
logy (3G) servin
g the
comm
erce, the netwo
rk offers mo
re
types of
service su
ch a
s
Voice, Da
te, multimedia.
Espe
cially, th
e multimedi
a
has wid
e
ly b
een
use
d
. Th
e u
s
er dem
a
nds highly fo
r the net
wo
rk.
As
the 3G cann
o
t
make a
corresp
ondi
ng ad
justment to
th
e dynami
c
tra
ffic distrib
u
tio
n
, SON (Self-
Orga
nizi
ng Networks) h
a
s
prop
osed the
framewor
k o
f
MLB Algorithm whi
c
h ma
ke
s a dynam
i
c
adju
s
tment to
the dynami
c
traffic di
strib
u
tion.
The
M
L
B algo
rithm
makes th
e n
e
twork
rea
c
h
a
highe
r re
sou
r
ce
rate, se
rves the use
r
wi
th a guaranteed Qo
s
and pr
events the overloa
d
.
More
over, SON ha
s intro
duced a sta
n
dard
sp
e
c
ification to MLB in literature [2].
The current
MLB Algorith
m
s almo
st ha
ve focused o
n
adju
s
ting th
e paramete
r
s [3, 4] of
hand
over, wh
ich h
a
ve adv
antage
s of
controllin
g
the
hand
over p
a
r
amete
r
s
sim
p
ly and flexib
ly.
More
over, these
solution
s result on the
load balan
ce
without affecting t
he other param
eters in
system i
n
lite
r
ature [5-7]. In additio
n
, th
e current M
L
B algorith
m
s
have p
r
op
ose
d
the p
r
o
c
e
ss o
f
adju
s
ting the
spe
c
ific
offset
of hand
over,
but they
hav
e not con
s
ide
r
the p
r
o
c
e
s
s of cal
c
ulatin
g
the optimal value of the specifi
c
offset
of handov
e
r
. The con
s
traint con
d
ition
s
of the sp
ecific
offset of h
a
ndover are
prop
osed to
avoid th
e
conflict b
e
twe
en MLB
and
Mobility ro
b
u
st
optimizatio
n (MRO
) in literature [8], but
unne
ce
ssary
param
eters
and compl
e
x conditio
n
s
a
r
e
brou
ght with
o
u
t con
s
id
erin
g the inh
e
re
n
t
relati
on
ship
of the han
d
o
ver offsets
betwe
en the
cell
and it
s n
e
igh
bors. Due to
those d
r
a
w
ba
cks, in
this
pa
per,
we
esta
b
lish th
e o
p
timization
mod
e
l
of
resou
r
ce utili
zation
rat
e
t
o
si
mplify th
e complexity of pa
ram
e
te
rs so a
s
to
obtain
an
op
timal
value a
nd
a
c
hieve
the
e
x
pected
pe
rforma
nce of
LB me
cha
n
ism compa
r
ed
to the
cu
rrent
algorith
m
s.
2.
The MLB T
h
e
or
y
MLB is an in
d
i
spe
n
sable te
chn
o
logy in S
O
N. The p
u
rp
ose of MLB i
s
to balan
ce th
e load
of cell-n
e
igh
b
o
r-pair by tra
n
sferrin
g
a pa
rt of
services
of overload
cell to its neigh
bor cell.
The
cu
rre
nt MLB alg
o
ri
thm ca
n fle
x
ibly and d
y
namically
a
d
just th
e h
andove
r
para
m
eters to balan
ce the
l
oad of cell-n
e
ighb
or-pai
r.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 8, August 2014: 628
1 –
6290
6282
In literature
[9], the events A1
-A6
can l
ead to
the han
dov
er. Spe
c
ifica
lly, this
spe
c
ification
has introdu
ce
d the con
d
it
io
n of handove
r
cau
s
ed by th
e event A3.
If receive
d
si
gnal p
o
wer from the
sou
r
ce cell
s
is lower tha
n
its
ne
ighbo
r
cell, th
e user
hand
overs fro
m
cell
s
to c
e
ll
n
. And the req
u
irem
ent of handove
r
ca
n be define
d
as below:
O
ff
Oc
Of
M
Hys
OC
Of
M
s
s
s
n
n
n
(
1
)
Inversely, the
requi
reme
nt of handove
r
from cell
n
to c
e
ll
s
is defined
as bel
ow:
Of
f
Ocn
Ofn
Mn
Hys
OC
Ofs
Ms
s
(
2
)
Whe
r
e
n
M
and
s
M
are
the re
ceived
sign
al po
we
r from cell
n
and cell
s
, res
p
ec
tively.
n
Of
and
s
Of
are
the
offset
of the
spe
c
i
a
l fre
quen
cy.
n
Oc
and
s
Oc
are
the
sp
ecifi
c
off
s
et of the
spe
c
ial
cell.
Hy
s
is the lag
param
eter.
Of
f
is the spe
c
ific offset of the event A3. All the param
e
t
ers
con
s
id
ere
d
are measured i
n
dB.
If we define
0
ns
Of
O
f
Of
f
, we could g
e
t simplified fo
rmula as b
e
lo
w
：
ns
s
n
M
MO
c
O
c
H
y
s
(
3
)
Similarly, the requi
rem
ent of the handov
er from
cell
n
to cell
s
is defin
ed as b
e
lo
w:
sn
n
n
M
MO
c
O
c
H
y
s
(
4
)
If we define
,
s
ns
n
Oc
Oc
Oc
,
,
ns
n
s
Oc
Oc
Oc
, the formula (3) is
conve
r
te
d as bel
ow:
,
ns
s
n
M
MO
c
H
y
s
(5)
Similarly, formula (3
) is co
nverted a
s
be
low:
,
sn
n
s
M
MO
c
H
y
s
(6)
Whe
r
e
,
s
n
Oc
rep
r
e
s
ents the off
s
et of handov
er from
cell
s
to cell
n
. Similarly,
,
ns
Oc
represent
s
the offset of handove
r
from
cell n to cell
s.
Whe
n
the loa
d
of cell s ha
s exce
eded
t
he thre
shol
d of overload, the cell
s
transf
e
rs a
part of use
r
s
to its neighbo
r cell to bala
n
ce the
loa
d
. The pro
c
e
s
s of MLB of
the cell-neig
h
b
o
r-
pair is
sho
w
n
in Figure 1.
MRO
is a
req
u
isite
optimization me
ch
an
ism in
SO
N.
The fun
c
tion
of MRO i
s
to
deal
with
optimizatio
n
probl
em
s of
hand
over
wh
ile MLB is to solve th
e
unbal
an
ced t
r
affic p
r
obl
e
m
.
Although M
L
B and M
R
O
operate ind
e
pend
ently, they
have an
i
nherent relati
onship
sin
c
e
they
optimize
the
spe
c
ific offse
t
of han
dover
Oc
in op
po
site di
rectio
n. Whe
n
MRO a
nd
MLB adj
ust
the pa
ram
e
te
r
simultan
eou
sly, they may
co
nflict
a
nd bring
ab
out some pro
b
lem
s
su
ch as
To
o-
Early HO, To
o-Late
HO a
n
d
Ping-P
ong
Effect. In
the optimizatio
n pro
c
e
ss, ML
B
decre
ase
s
the
spe
c
ific off
s
e
t
of handover to balan
ce t
he load
of ce
ll-neig
hbo
r-pa
ir whil
e MRO
increa
se
s the
spe
c
ific
offse
t
of handove
r
to optimize the net
wo
rk.
So the confli
ct lead
s to th
e endl
ess lo
op
betwe
en the
two o
p
timization
me
ch
anism
s. Th
e
re
sult is t
he de
clin
e
of the net
work
perfo
rman
ce.
Too-Ea
rly
HO
, Too-Late
HO de
grade
th
e Qo
s which
make
the
u
s
e
r
di
scon
ne
ct from th
e
servin
g cell. Ping-Pon
g
Effect
severely wa
stes
th
e n
e
twork
re
sou
r
ce fo
r the fre
quent h
and
over
betwe
en the two cells.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Load Bala
nci
ng Base
d on
the Specific
Offset of Han
dover (Zhanj
un Liu)
6283
Figure 1. The
MLB Proce
ss
In literature [8
], in order to avoid the co
n
f
lic
t, the const
r
aint co
nditio
n
s of the sp
e
c
ific
offset of hand
over are defin
ed as b
e
lo
w:
'
,,,
,
'
,,
,
,
''
,,
2
s
ns
ns
n
e
a
r
l
y
n
s
la
te
n
s
n
s
sn
n
s
Oc
O
c
Oc
Oc
O
c
Oc
Oc
Oc
H
y
s
(
7
)
Whe
r
e
,,
s
ne
a
r
l
y
Oc
is the thre
shol
d of Too-Ea
rly HO f
r
om cell s to cell n ,
,,
ns
l
a
t
e
Oc
is the thre
shol
d of
Too-Late HO from cell
n
to c
e
ll
s
.
3.
The Simplified MLB Model
As introdu
ced in
p
r
evi
ous sectio
n
2,
we
h
a
ve defin
ed
,
s
ns
n
Oc
Oc
Oc
and
,
ns
n
s
Oc
Oc
Oc
. The M
L
B pro
c
e
s
s in
[8] indepe
n
dently adju
s
t
n
s
Oc
,
and
s
n
Oc
,
without
considering
s
Oc
and
n
Oc
. The
result i
s
th
e
red
und
an
cy paramete
r
s and
the
complex
con
s
trai
nts. Accordi
ng to the formula (7),
we ca
n get the formul
a as follow:
''
,,
''
,,
02
s
ns
ns
n
e
a
r
l
y
n
s
la
te
n
s
n
s
Oc
Oc
Oc
Oc
Oc
Oc
Oc
Oc
Oc
Oc
Hy
s
(
8
)
Whe
r
e
0
Hys
, so the formula
(8) can be fu
rth
e
r sim
p
lif
ied j
u
st as th
e formula (9
) express a
s
belo
w
:
)
min(
,
,
,
,
late
s
n
early
n
s
n
s
n
s
Oc
Oc
Oc
Oc
Oc
Oc
(
9
)
Based
on th
e
analysi
s
a
b
o
v
e, the functi
on of the fo
rmula (9)
with
a con
c
ise ex
pre
ssi
on
is e
quivalent
to the formul
a 7
whi
c
h
ca
n be
re
garde
d a
s
the
co
n
s
traint
co
nditi
on to avoi
d T
oo-
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 8, August 2014: 628
1 –
6290
6284
Early HO, To
o-Late
HO, P
i
ng-Po
ng Eff
e
ct. On
e
han
d, the th
ree
con
s
trai
nts d
e
crea
se
to o
ne.
The a
c
tual
ca
lculatio
n co
ntains t
he
process that the form
with co
n
s
traint tran
sfo
r
ms to the fo
rm
without con
s
traint. The d
e
c
re
ase of the
numbe
r of
constraint con
d
ition re
du
ce
s the comple
xity
of the operati
on pro
c
e
s
s. On t
he other hand, before
the
simplification if the cell s is a reg
u
lar
hexago
n, it should
set 6 h
andove
r
pa
ra
meters
6
,
1
,
n
s
n
s
Oc
Oc
for its neighb
or
whi
l
e there a
r
e
six co
rrespon
ding h
andove
r
pa
ramete
rs
1,
6
,
ns
n
s
Oc
Oc
to the c
e
ll
s
. After the s
i
mplific
a
tion, the
cell s
set only one paramet
er
s
Oc
, so the number of pa
ra
meters of each cell and the
complexity
of the operati
on pro
c
e
s
s can be si
g
n
ificantly redu
ced
by using t
he simplified ML
B algorithm. The
sign
aling p
r
o
c
e
ss of the si
mplified ML
B algorith
m
is shown in Figu
re 2.
Figure 2. The
Simplified MLB Proce
ss
We a
s
sume t
hat the cell
s
has K neigh
bo
r cell
s, the de
tailed step
s o
f
LB algorith
m
are
illustrated as
followed:
(1) The
cell
s inqui
ry the h
andove
r
p
a
ra
meters
of its
neigh
bor, th
e
n
its n
e
igh
b
o
r
send
the relative parameters
—
nK
n
Oc
Oc
1
and
late
s
nK
late
s
n
Oc
Oc
,
,
,
,
1
.
(2) A
c
cording
to the formu
l
a (9
), the
ce
ll s
whi
c
h ta
ke
s
nK
n
Oc
Oc
1
a
s
a c
ons
ta
n
t
value ca
n cal
c
ulate the ran
ge of
s
Oc
and furt
her obtai
n the
optimal value of
s
Oc
.
(3)
The
c
e
ll
s
se
nds the
upd
ate value
of
s
Oc
to its n
e
igh
bors t
o
adj
ust th
e
specifi
c
offset
of handove
r
n
Oc
.
(4) A
c
cording
to the formu
l
a(9
)
, ea
ch cell cal
c
ulate t
he ra
nge
of
n
Oc
in condition
of
con
s
id
erin
g its sp
ecifi
c
offset of hand
o
v
er as
a
con
s
tant value a
nd furthe
r ob
tain the optimal
s
o
lution of
n
Oc
.
As the thresh
olds
of Too
-
E
a
rly HO an
d
Too-
Late
HO
are
only rel
a
ted to the tran
smitted
power
of the
BS, the thre
shold
s
a
r
e
rel
a
tively
stable
p
a
ram
e
ters. E
a
ch
cell
con
s
t
r
uct
s
a list
an
d
install
s
the p
a
ram
e
ters in
cludi
ng: Too
-
Early HO, To
o-Late
HO, t
he spe
c
ific of
fset of han
do
ver
and so on. When the
spe
c
i
f
ic offset of h
andove
r
ha
s
cha
nge
d, each cell
will info
rm its nei
ghb
ors
to update the
spe
c
ific offset of handov
er in thei
r list
.
So MLB algorithm
can
save the re
qu
est
and respon
se
step
s to si
m
p
lify the pro
c
ess of ML
B a
l
gorithm
de
scried a
s
the
Fi
gure
2. And t
he
upgrade p
r
o
c
ess ca
n be shown in Figu
re 3.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Load Bala
nci
ng Base
d on
the Specific
Offset of Han
dover (Zhanj
un Liu)
6285
Figure 3. The
Upgrade M
L
B Proce
ss
4.
The Resear
ch on the Resource
Utilization Model
4.1. The Opti
mization Mo
del
Acco
rdi
ng to
the optimi
z
ati
on mo
del
ba
sed
on th
e
re
sou
r
ce utili
za
tion rate,
we
can
get
the utility function of each
cell whi
c
h can
be expressed as below:
()
0
()
0
th
s
t
h
t
h
s
s
th
s
t
h
s
st
h
s
st
h
t
h
s
aa
u
b
a
bb
(
1
0
)
W
h
er
e
th
,
s
are
th
e threshold
of
overlo
ad
and
the
cu
rre
nt lo
ad of th
e
cell
s
, res
p
ec
tively.
s
u
is
the utility function of cell
s
.
b
a
,
are the pos
i
tive c
oeffic
i
ent. If
th
s
,
s
u
decrea
s
e
s
with the
value of
s
th
in
c
r
ea
s
i
ng
. An
d if
0
s
,
0
s
u
. If
th
s
,
s
u
is i
n
di
re
ct pro
portio
n
t
o
the
value of
s
th
, and
s
u
is
ne
gative. As we ca
n se
e
from
the
util
ity function,
e
a
ch
cell
prefe
r
s a
higher resource utiliz
ation rate rather than an
overflowing traffic.
The sy
stem utility
function
of the cell
s
an
d its neigh
bors ca
n be defi
ned bel
ow:
1
K
s
sn
i
i
u
tilit
y
u
u
(11)
Whe
r
e
s
u
,
ni
u
,K are the utility fu
nction of the
cell
s
, the utility function of its neighbor
cell
i
n
and the num
b
e
r of its neigh
bor. Fo
r the h
e
xagon
al cell
ular cell, K=6.
The optimi
z
at
ion model i
s
shown as b
e
lo
w:
:
s
s
OC
M
a
x
u
tility
(12)
Although the optimizatio
n model ha
s n
o
t the direct expre
ssi
on of
s
Oc
,
s
Oc
can chang
e
s
u
as the pa
ram
e
ter
s
of the utility function
s
u
vary with
s
Oc
.
The pa
ram
e
ter
s
of the utility function
s
u
cha
n
ges
as the va
lue of
s
Oc
ch
ang
es, so the
s
is co
nsi
dered
as an interm
ediate bet
we
en
s
Oc
and
s
u
.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 8, August 2014: 628
1 –
6290
6286
4.2. The Inhe
rent Relation
ship bet
w
e
e
n
s
and
s
Oc
We a
s
sume t
he load of cel
l
s
can be d
e
fin
ed as b
e
lo
w:
|(
)
u
uX
u
s
s
to
t
N
N
(
1
3
)
Whe
r
e
u
N
and
tot
N
are
the actu
al a
m
ount of re
source
s o
c
cup
i
ed by u
s
er
u
and total nu
m
ber of
r
e
sour
ces
,
r
e
s
p
ec
tively.
()
X
u
is
th
e
us
er
c
o
nne
c
t
io
n fu
nc
tion
. If
)
(
u
X
s
, it mean
s the
u
s
er u
has
con
n
e
c
te
d the cell s. A
nd the amou
n
t
of
require
d reso
urce
s ca
n
be written a
s
below:
()
u
u
u
D
N
R
SI
N
R
(14)
Whe
r
e
u
D
is
the UE required rate.
)
(
u
SINR
R
is the thro
ugh
p
u
t mappin
g
function
whi
c
h
expre
s
ses th
e data
rate p
e
r PRB
given
u
SINR
. So we
can
write th
e SINR for
UE u
of the cell s
as bel
ow:
()
,
(
)
,
()
(,
)
(,
)
sX
u
u
sX
u
u
s
us
cc
c
u
c
cX
u
PL
q
SI
N
R
NP
L
q
(
1
5
)
Whe
r
e
u
q
and
s
are
the coordin
a
te of Use
r
Equipme
n
t(UE)
u
and the
down tilt of cell
s
,
respe
c
tively. All use
r
s
are
locate
d on th
e gro
und
(h
ei
ght ze
ro
).
()
s
Xu
P
is t
he tra
n
smitte
d po
wer
of the
cell
s
serving UE
.
,(
)
us
X
u
L
is the
path lo
ss between
UE
u
and the
cell
s
. N is
the
additive Gau
s
sian
White Noise.
The
re
sult of
the h
and
over b
e
twe
en t
he
cell
and
i
t
s nei
ghb
or
can b
e
exp
r
e
s
sed
as
belo
w
：
)
)
,
(
)
,
(
max(
arg
)
(
s
s
u
s
s
n
n
u
n
n
T
q
L
P
Hys
T
q
L
P
u
X
r
r
(
1
6
)
Mean
while, we can exp
r
e
s
s the pa
ramet
e
rs of the formula (1
6) in d
B
unit as follow:
10
20
log
[
(
,
)]
nn
u
n
n
PL
q
M
r
10
20
l
o
g
[
(
,
)
]
s
su
s
s
PL
q
M
r
10
20
l
o
g
[
]
s
s
TO
c
10
20
l
o
g
[
]
nn
TO
c
So we ca
n ge
t the converte
d expre
ssi
on
whi
c
h is give
n as bel
ow:
,
()
a
r
g
m
a
x
(
,
)
nn
s
s
sn
X
u
M
Oc
Hy
s
M
O
c
(
1
7
)
4.3. Conv
ergence Analy
s
is
As introdu
ce
d in p
r
eviou
s
se
ction 4.
1,
with the
spe
c
ific offset of han
do
ver
s
Oc
increa
sing, th
e ed
ger u
s
e
r
who
satisfie
s
the A3
event
hand
overs to
its nei
ghb
or
cell. So we
ca
n
con
c
lu
de that
s
var
y
inver
s
ely w
i
th
s
Oc
, but
nK
n
1
are
on the co
ntra
ry.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Load Bala
nci
ng Base
d on
the Specific
Offset of Han
dover (Zhanj
un Liu)
6287
At the beginn
ing of adjusti
ng the param
eter
s
Oc
, if the neighbo
r cell
s’
load is un
de
r the
threshold, the system utility function of the cell
s
and its neigh
bors ca
n be expre
ssed as b
e
lo
w:
K
i
i
n
s
th
K
i
ni
s
s
a
b
u
u
utility
1
1
)
(
(
1
8
)
Whe
r
e
s
u
var
y
inver
s
ely w
i
th
s
and
nK
n
u
u
1
is in
prop
ortion to
nK
ni
, according to the
formula (18
)
,
s
utility
tend to
increa
se. A
s
s
decrea
s
e
s
and
nK
n
1
incr
ea
se,
t
h
e
cal
c
ulatio
n of
s
utility
can b
e
divide
d into the followin
g
co
nditions.
The conditio
n
1: The lo
ad of the ce
ll
s
and its ne
ighbo
r
)
1
(
k
i
S
i
excee
d
the
threshold of
overloa
d
, but
the load of the other neighb
or
)
1
(
1
K
i
k
S
i
are u
nder the
threshold, we can write th
e utility
function as bel
ow:
s
i
ni
th
K
s
j
j
n
s
th
K
i
ni
s
s
b
a
b
u
u
utility
1
1
1
)
(
)
(
(
1
9
)
As
s
decrea
s
e t
o
th
s
, the calcula
t
ion of
s
utility
in co
n
d
ition 1 i
s
co
nverted to
th
e
con
d
ition 3.
The conditio
n
2: The loa
d
of the cell
s
and
its neigh
bo
r
cell
s are bel
o
w
the thresho
l
d o
f
overload, the system utility f
unction
can
be defined as below:
K
i
ni
s
K
i
ni
s
s
a
a
u
u
utility
1
1
(
2
0
)
If the load
of
any nei
ghbo
r
of the
cell
s
exceed
s the
thre
shol
d of
overl
oad
with th
eir load
(
nK
n
1
) incre
a
si
ng,
the calculati
on of
s
utility
in cond
ition 1 is con
v
erted to the con
d
ition
3.
The
co
ndition
3: Th
e lo
ad
of the
cell
s
is
unde
r the
thresh
old, a
nd
a
pa
rt of its nei
ghbo
rs
)
1
(
k
i
S
i
exceed the threshold
whi
l
e anothe
r pa
rt cells
)
1
(
1
K
i
k
S
i
do not exceed
the threshold,
so we
can
write t
he system utility function as below:
K
s
j
nj
s
i
ni
th
s
K
i
ni
s
s
a
b
a
u
u
utility
1
1
1
)
(
(
2
1
)
Based
on th
e analysi
s
a
bove, we co
nclu
de that
s
u
and
nK
n
u
u
1
increa
se first
and
then de
cre
a
se with
s
Oc
increasing. We can prove that
the curve of the utility fu
nction
s
u
is
con
c
ave, so the optimal m
odel is
conv
e
r
gent an
d exists the optim
al value.
5.
Simulations Parameters
and Re
sults
The p
ape
r u
s
e
s
VC++6.0
and MAT
L
AB softwa
r
e
to simul
a
te. Table
1
sh
ows the
config
uratio
n
of the
sim
u
l
a
tion p
a
ra
me
ter. Unde
r th
e 9
ca
se
s
wi
th users
ran
g
i
ng fro
m
4
0
0
to
1200, the re
sults whi
c
h
co
ntain the co
n
gestio
n
ra
te,
the QoS and
the resource
utilization rat
e
cal
c
ulate
d
by the MLB algo
rithm are d
e
scrib
ed in Fig
u
r
e 4-6.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 8, August 2014: 628
1 –
6290
6288
Table
1. The
Simulation Param
e
ters
Description Values
Cell Number
23
Cell to Cell
Distance
500m
BandWidth 5MHz
Power of e
N
B
29dBm
Path loss
128.1+37.6log
10
(R) st. R(km)
shado
w
i
ng
standard
deviation
10dB
Correl
a
ti
on
distance of
the shado
w
10m
Antenna patte
rn
2
3
()
m
i
n
[
1
2
(
)
,
]
m
dB
A
A
st.
3
70
deg,
20
dB
m
A
dB
User Numbe
r
400-1200
H
y
steresis 3dB
Initial offset
of handover
0dB
Threshold of
Too-Ea
rl
y
H
O
10dB
Threshold of
Too-Lat
e H
O
-10dB
Threshold of
Overload
0.9
Figure 4. The
Cong
estio
n
Rate
It can be se
en from Figu
re 4, the con
gestio
n
rate tend
s to rise
with the increase of
use
r
s
wh
ethe
r the MLB alg
o
rithm is i
m
pl
emented
or
n
o
t. Apparentl
y
, the curve o
f
the cong
esti
on
rate with M
L
B algorithm i
s
und
er the
curve
witho
u
t MLB algorit
hm. It shows that the MLB
algorith
m
red
u
ce
s th
e
con
gestio
n
rate.
The m
a
ximu
m of de
crea
se is up
to 1.
1% co
mpa
r
e
d
to
the ca
se
without utilizin
g
MLB and th
e minimum o
f
decrea
s
e is also u
p
to 0
.
44%. It can be
con
c
lu
ded th
at: the MLB algorithm
can
effectivel
y improve the net
work cong
esti
on perfo
rma
n
ce.
Figure 5 shows the resource
utilization rate of
the
system
with the increasi
ng of the
use
r
s. Th
e cu
rve with a
s
terisks an
d solid
dots
re
prese
n
t the ca
se with and witho
u
t utilizing M
L
B
algorith
m
, re
spectively. Th
e syste
m
re
source utili
zati
on rates
of t
he
both ca
se
i
n
crea
se with the
increa
sing
of
the u
s
e
r
s.
Howeve
r, the
curve
of
the resou
r
ce
utili
zation rate
s with
the MLB
is
alway
s
abov
e the ca
se
without u
s
ing
MLB and re
sult sh
ows that the
MLB can imp
r
ove
the
system
re
so
u
r
ce
utili
zation
rate
with
the
maximum
in
cre
a
se
achi
e
v
ing 5.35%
a
nd the
minim
u
m
increa
sing
1.9%. It can be
con
c
lu
ded th
at: the ML
B
algorith
m
can
effectively improve
re
so
urce
utilization rate.
400
500
600
700
800
900
1000
1100
1200
0
5
10
15
20
25
30
T
he U
s
er
N
u
m
ber
T
he c
onges
t
i
on
r
a
t
e
(
%
)
ML
B
No M
L
B
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Load Bala
nci
ng Base
d on
the Specific
Offset of Han
dover (Zhanj
un Liu)
6289
Figure 5. The Resource
Utiliz
ation Rate
Figure
6.The Qos
It can b
e
se
en from
Figu
re 6, the
Qo
S tends
to
g
o
do
wn
with
the num
ber of user
increa
sing
re
gardl
ess
of th
e u
s
e
of MLB
algo
rith
m
or
not. Appa
rent
ly, the cu
rve
of the Q
o
S
with
MLB algo
rith
m is a
bove t
he curve
wit
hout ML
B al
gorithm. It shows that th
e MLB alg
o
ri
thm
improve
s
the
QoS. And the maximum of incre
a
se
is u
p
to 1.6% wh
ile the minim
u
m value is u
p
to 0.3%. It ca
n be
con
c
lud
ed that: MLB
algorith
m
can
effectively im
prove th
e p
e
rforman
c
e
of t
he
Qos.
6. Conclusio
n
This
pap
er fo
cu
s o
n
the
M
L
B tech
nolo
g
y
of the
wirel
e
ss mo
bile
co
mmuni
cation
netwo
rk.
Acco
rdi
ng to
the A3 event (the si
gnal
strength of nei
g
hbor
cell i
s
hi
gher th
an tha
t
of the source
cell),
we
can
get the inh
e
r
ent relation
ship of
the sp
ecific offset of
hand
over and simplify
the
con
s
trai
nt of handove
r
.
The alg
o
rith
m improve
s
the cu
rrent pro
c
e
s
ses
a
nd sim
p
lifies the
sign
aling p
r
o
c
ed
ure.
Duo to th
e
MLB algo
rith
ms d
o
n
o
t consi
der
the
o
p
timization
of
the spe
c
ific
offset of
hand
over, the
utility function wa
s establi
s
he
d to calc
u
l
ate the optimal value of the spe
c
ific offset
of handove
r
b
y
the optimization theo
ry in this pa
per.
The utility function ha
s the
optimal targ
et to
maximize the
reso
urce utili
zation rate an
d has b
een p
r
oved to be co
nverge
nt.
The sim
u
latio
n
re
sults
sho
w
that the ML
B algorithm
had re
duced
the cong
esti
on rat
e
about 1.1
p
e
rcentag
e p
o
ints a
nd h
ad imp
r
oved
the re
so
urce utilizatio
n
rate a
bout 5
.
35
percenta
ge p
o
ints an
d had
improved the
QoS
guarant
ee rate ab
out
1.6 percenta
ge point
s.
Ackn
o
w
l
e
dg
ements
Nation
al Nat
u
ral Sci
e
n
c
e
Foundatio
n
of China (No.61
171
111
) Foun
dation
Items:
Nation
al Sci
ence an
d T
e
ch
nolo
g
y
Major
Proj
ect
(2012ZX
030
0
3008
-00
4
);
Nation
al Natural
Scien
c
e
Fo
undatio
n of
Chi
n
a
(No
.
61171
111
)
the tra
n
sfo
r
mation
proj
e
c
t of
excell
ent
achi
evement
of Ch
ong
qi
ng
Muni
cip
a
l
Educat
ion Commi
ssion (Kjzh
112
06);
Natu
ral scie
nce
fund proj
ect
s
of CQUPT
Referen
ces
[1]
Hon
g
li
n Hu, Jian Z
h
a
ng, Xi
ao
yi
ng Z
h
e
n
g
, Y
ang Yan
g
, Ping W
u
. Self-confi
gurati
on an
d self-
optimiz
ation for
LT
E net
w
o
rks.
Communic
a
tio
n
s Maga
z
i
ne, IEEE
. 2010; 48(
2): 94-10
0.
[2]
3GPP T
R
36.902 V9.3.1. Ev
olve
d Un
ivers
a
l T
e
rrest
rial
Radi
o Access
Net
w
ork (E-
U
T
R
AN); Self-
config
urin
g an
d self-optim
izin
g
net
w
o
rk (SO
N
) use cases a
nd sol
u
tio
n
s (Rele
a
se 9). 20
1
1
; 3.
[3]
Viering I, Dottling M, Lobinger A.
A Mathematical Pers
pect
i
ve of Se
lf-Opti
m
i
z
i
ng W
i
re
les
s
Netw
orks
.
Commun
i
cati
o
n
s. ICC '09. Fullerto
n
, Califor
n
i
a, USA. 200
9: 1-6.
40
0
500
60
0
700
800
90
0
1
000
1100
120
0
45
50
55
60
65
70
75
80
85
90
95
T
h
e
U
s
er
N
u
m
ber
T
h
e r
e
s
our
c
e
ut
i
l
i
z
at
i
o
n(
%
)
ML
B
No
M
L
B
40
0
50
0
60
0
70
0
80
0
900
10
0
0
11
0
0
1
200
50
55
60
65
70
75
80
85
90
95
100
T
h
e U
s
er
N
u
m
b
er
T
he Q
o
s
(
%
)
Qo
s
M
L
B
No Q
o
s
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 8, August 2014: 628
1 –
6290
6290
[4]
Lob
ing
e
r A, S
t
efanski S,
Ja
nsen
T
,
Bala
n I.
L
oad
Ba
l
anci
ng
in
Do
w
n
li
nk
LT
E Self-Optimizin
g
Net
w
orks. Vehicular T
e
chnology
Confer
enc
e (VT
C
2010 Spring).
Ottaw
a
, Canada. 2010: 1-5.
[5]
A
w
a
d
a
A, W
egman
n B, Vier
ing I, Kle
i
n A.
A ga
me-th
eore
t
ic appr
oach t
o
lo
ad b
a
l
anci
ng i
n
cel
l
ul
a
r
radi
o n
e
tw
orks
. Person
al In
do
or an
d Mo
bil
e
Radi
o C
o
mmu
nicati
ons (PIM
RC). Istanbu
l,
T
u
rkey
.
201
0:
118
4-11
89.
[6]
K
w
a
n
R, Arn
o
tt R, Paterson R
,
T
r
ivisonno
R, Kubota M.
On
Mobil
i
ty Lo
ad
Bala
ncin
g for L
T
E Systems
.
Vehic
u
lar T
e
chnology
Conference Fall (VT
C
2010-Fall). Otta
w
a
, Ca
nada. 2010: 1-5.
[7]
Xi
aofa
n
g
Li
an
g, Ho
ng T
ang.
Res
earch
o
n
lo
ad
ba
lanc
in
g i
n
3GPP
sy
stem
. T
e
lecom
Eng
i
ne
eri
n
g
T
e
chnics and
Standar
diz
a
tio
n
. 2009; 1
1
: 81
-84
．
[8]
Z
h
iqi
ang
Li
u, Peili
n H
ong, K
a
ipi
ng
Xue, Mi
n Pen
g
.
Co
nfli
ct Avoida
nce
b
e
tw
een Mo
bilit
y Rob
u
stnes
s
Optimi
z
a
t
i
o
n
a
nd Mobi
lity Lo
ad Bal
anci
n
g
.
Global T
e
lecommunic
a
tio
n
s
Confere
n
ce
(
G
LOBECOM
201
0). Miami, F
L
, USA. 2010
: 1-5.
[9]
3GPP T
S
36.331 V1
0.5.0. E
v
olve
d Un
ivers
a
l T
e
rrest
rial Radi
o
Access
(E-UT
R
A);
Radio Reso
urce
Contro
l (RRC);
Protocol spec
i
f
ication (R
ele
a
s
e 10). 201
2; 3
.
Evaluation Warning : The document was created with Spire.PDF for Python.