Int
ern
at
i
onal
Journ
al of Ele
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
10
,
No.
4
,
A
ugus
t2
02
0
, p
p.
3715
~
3724
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v10
i
4
.
pp3715
-
37
24
3715
Journ
al h
om
e
page
:
http:
//
ij
ece.i
aesc
or
e.c
om/i
nd
ex
.ph
p/IJ
ECE
Enh
ancenig OLS
R routin
g proto
col usin
g
K
-
me
ans
clusteri
ng
in
MANETs
Y.
Ham
z
ao
ui
1
, M.
Amn
ai
2
, A.
Chou
kri
3
, Y.
Fakhri
4
1
, 2, 4
LARIT,
N
etw
orks a
nd
T
el
e
c
om
m
unic
at
ions
Te
am,
Facult
y
o
f
Scie
n
ce
s Ken
itra
,
IbnTof
a
il Uni
ver
sit
y
,
Morocc
o
3
Cadd
y
A
yy
ad
Univer
sit
y
,
S.
A.
R.
S Group, ENS
A Safi
,
Morocc
o
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
N
ov
15, 201
8
Re
vised
Jan
2
2
,
20
20
Accepte
d
Fe
b 2
, 2
020
The
design
of
robust
routi
ng
protoc
ol
sche
m
es
for
MAN
ET
s
is
quite
complex,
due
to
the
character
isti
cs
and
struct
ura
l
constra
int
s
of
th
is
net
work.
A
num
ero
us
var
ie
t
y
of
protoc
o
l
sche
m
es
have
bee
n
proposed
i
n
li
te
r
at
ure
.
Mos
t
of
the
m
a
re
base
d
on
traditi
ona
l
m
et
hod
of
routi
ng,
which
doesn’t
guar
antee
basi
c
l
eve
ls
of
Qos
,
w
hen
the
net
work
bec
om
es
la
rge
r,
dense
r
and
d
y
n
ami
c.
To
sol
ve
thi
s
p
roble
m
we
use
on
e
of
the
m
ost
popul
ar
m
et
hods
named
cl
usteri
n
g.
In
thi
s
work
we
tr
y
to
improve
the
Qos
in
MA
NETs.
W
e
propose
an
al
gorit
hm
of
c
lu
steri
ng
base
d
in
the
new
m
obil
i
t
y
m
et
r
ic
and
K
-
Mea
ns
m
et
hod
to
distr
ibute
the
nodes
i
nto
s
eve
r
al
cl
u
sters;
it
is
implemente
d
to
standa
rd
OLSR
protoc
ol
g
ivi
ng
birt
h
a
n
ew
prot
ocol
named
OLSR
K
m
ea
ns
-
SD
E.
The
sim
ula
ti
ons
show
ed
t
hat
the
r
esult
s
o
bta
in
ed
b
y
OLSR
Km
ea
ns
-
SD
E
exc
e
ed
tho
se
obta
in
ed
b
y
s
ta
ndar
d
OLSR
Km
ea
ns
and
OLSR Km
ed+
i
n
te
rm
s of, t
r
aff
i
c
Contro
l, de
l
a
y
and
pa
cket
de
li
v
er
y
ra
ti
o
.
Ke
yw
or
d
s
:
Cl
us
te
rin
g
K
-
Me
a
ns
MANET
s
OS
LR
QoS
Copyright
©
202
0
Instit
ute of
Ad
v
ance
d
Engi
ne
eri
ng
and
Sc
ie
n
ce
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
Y.
Ham
zaou
i,
Faculty
of S
ci
e
nces
Ke
nitra,
IbnT
of
ai
l
Un
i
ve
rsity
,
Av
e
nue
de
L'
U
niv
e
rsité
, K
é
ni
tra,
Mo
r
occo
.
Em
a
il
:
Yo
unes
_ieee@
ho
tm
ai
l
.co
m
1.
INTROD
U
CTION
The
a
d
hoc
ne
twork
is
a
ne
w
wireless
c
omm
un
ic
at
ion
syst
em
witho
ut
the
use
of
a
centrali
ze
d
m
anag
em
ent
i
nfrastr
uctu
re.
I
t
con
sist
s
of
a
set
of
node
s
(
m
ob
il
e
dev
ic
e)
that
ha
ve
a
c
om
m
un
ic
at
ion
wireless
dev
ic
e
f
or
c
omm
un
ic
at
ing
with
entit
ie
s
locat
ed
in
their
neig
hborh
ood.
Each
node
ca
n
therefore
di
rectl
y
reac
h
it
s
neighbors
direct
us
in
g
it
s
rad
io
inte
rf
a
ce,
or
c
omm
u
nicat
e
with
ot
her
nodes
insi
de
the
netw
or
k
us
in
g
interm
ediat
e
no
de
s
(l
ocated
betwee
n
the
s
ource
a
nd
the
recipie
nt).
Th
ese
are
res
ponsi
ble
f
or
c
om
m
ut
ing
the
m
essages
and
play
in
g
the
ro
le
of
t
he
r
oute
r,
s
o
it
off
e
rs
an
aut
onomou
s
net
work
a
nd
gi
ves
the
te
rm
inal
us
er
acce
ss
t
o
inf
or
m
at
ion
a
t
any
ti
m
e
and
from
any
pl
ace.
T
hese
m
ob
il
es
co
operat
e
with
e
ach
ot
her
to
ov
e
rc
om
e
the
Ad
hoc
net
work
c
onstrai
nts
su
c
h
as
dy
nam
ic
top
ology,
la
ck
of
c
entrali
zed
m
on
it
or
i
ng
po
i
nts,
li
m
i
te
d
ba
ndwidt
h,
et
c.
the
la
tt
er
r
equ
i
res
desig
ni
ng
a
r
obus
t
a
nd
no
n
-
tra
diti
on
al
rou
ti
ng
s
yst
e
m
to
bette
r
m
an
agem
ent
of
the
flow
of
inf
or
m
at
ion
and
ens
ur
e
the
qu
al
it
y
of
serv
ic
e
in
a
dynam
ic
and
decen
t
rali
zed
ne
twork
.
Ma
ny
routing
protoc
ols
ha
ve
been
pro
pose
d
[1
]
on
dif
fere
nt
ty
pes
of
ap
plica
ti
on
.
T
he
researc
h
has
no
t
cea
sed
to
ha
ve
e
ff
i
ci
ent
protoc
ols
that
a
da
pt
to
al
l
m
ob
i
li
ty
m
od
el
s
[2
]
.
T
he
r
outi
ng
of
the
inf
orm
ati
on
in
the
Ma
nets
can
be
cl
assifi
ed
into
two
ty
pes:
Flat
and
Hierarc
hical
.
In
the
first
ty
pe
al
l
netwo
r
k
no
de
s
play
the
sam
e
ro
le
,
this
can
over
load
the
netw
ork,
as
well
as
oth
e
r
pro
ble
m
s
cause
su
c
h
as
scal
abili
ty
and
com
plexity
,
w
hen
the
net
work
bec
om
es
wider
an
d
de
ns
er
.
T
he
sec
ond
ty
pe
of
routin
g
is
us
e
d
to
s
uppo
rt
netw
orks
that
are
wi
de
an
d
de
ns
e.
T
he
cl
us
te
ring
is
a
hier
arch
ic
al
str
uctur
e
t
hat
m
akes
it
po
ssible
to
gro
up
g
eo
grap
hical
ly
the
nodes
that
are
neig
hbors.
This
al
lows
to
each
node
to
s
tore
al
l
inform
at
ion
ab
ou
t
it
gro
up
and
on
ly
so
m
e
inform
at
ion
of
oth
er
gro
ups
(clu
ste
rs).
This
ap
proac
h
can
reduce
the
cost
of
r
outi
ng
of
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
& C
om
p
Eng
,
V
ol.
10
, N
o
.
4
,
A
ugust
2020
:
3
7
1
5
-
3
7
2
4
3716
inf
or
m
at
ion
in
la
rg
e
a
nd
de
nse
netw
orks.
Se
ve
ral
researc
he
s
ha
ve
bee
n
propose
d
to
im
prov
e
t
he
e
ff
ic
ie
ncy
of
the h
ie
rar
c
hica
l protoc
ols a
nd to
s
upport t
his
ty
pe
of
netw
ork
a
nd str
uctu
re
.
In
orde
r
to
bette
r
or
gan
iz
e
th
e
netw
ork
i
nto
s
ever
al
cl
us
te
rs
with
op
ti
m
al
st
ru
ct
ur
es
,
the
re
searche
rs
pro
po
se
d
var
io
us
crit
eria
f
or
bu
il
di
ng
a
nd
orga
nizin
g
the
nodes
in
gro
up
s.
This
hierar
c
hic
struct
ur
e
le
ads
t
o
op
ti
m
iz
ation
and
im
pr
ovem
e
nt
of
t
he
r
ou
ti
ng
protoc
ols.
Dep
e
ndin
g
on
the
ty
pe
of
perform
ance
that
can
be
ens
ur
e
d
in
the
cl
us
te
r
str
uctu
r
e,
we
us
e
the
s
uitable
m
e
tric
.
Sever
al
m
et
ri
cs
hav
e
bee
n
pro
posed
t
o
m
e
asure
the
ph
ysi
cal
and
lo
gical
pro
per
ti
es
of
the
nodes
(m
ob
il
it
y,
den
sit
y,
ene
rg
y,
et
c.)
.
A
good
m
et
ric
m
a
kes
it
po
s
sible
t
o
differentia
te
easi
ly
bet
ween
the
node
s
[2
]
,
to
refl
ect
the
real
be
hav
i
or
of
t
he
nod
e
s.
a
nd
facil
it
at
es
the p
e
rfo
rm
ance i
m
pr
ovem
ent of
protoc
ols [3
]
.
In
t
his ar
ti
cl
e
we
re
fine o
ur
m
et
ric o
f
m
ob
il
it
y, def
ine
d
in
[4
]
, ta
king int
o ac
co
un
t a
ny ty
pe of
m
otion
in
a
co
ve
rag
e
area
of
a
m
ob
i
le
node,
t
his
re
fine
d
m
et
ric
a
nd
ot
her
m
et
rics
(
den
sit
y,
e
ne
rg
y)
will
be
t
he
basis
of
a
cl
ust
er
buil
ding
al
gorithm
by
us
in
g
k
-
m
eans
m
et
ho
d.
T
o
c
reate
the
gr
oups
of
m
ob
il
e
and
to
el
ect
the
cl
us
te
r
hea
d,
this
ne
w
ap
proac
h
ca
n
ge
ne
rate
a
m
or
e
sta
ble
cl
us
te
r
a
nd
cl
us
te
r
head.
I
n
this
pap
e
r
w
e
sta
rt
with
a
prese
ntat
ion
of
s
om
e
relat
ed
wor
k.
I
n
the
thir
d
sec
ti
on
we
pr
e
sen
t
the
prob
le
m
form
ulati
on
,
a
nd
in
the
four
t
h
sect
ion
we
def
i
ne
the
cl
us
te
rin
g.
The
n
in
the
fif
th
sect
ion
we
pr
ese
nt
the
des
cripti
on
of
ap
proa
c
h
cl
us
te
rin
g
al
go
rithm
,
in
the
s
ixth
sect
ion
w
e
exp
la
in
o
ur
m
ob
il
i
ty
m
et
ri
c,
after
in
the
seven
t
h
sect
ion
we
pr
ese
nt
the
al
gorithm
bef
ore
we
giv
e
the
resu
lt
s
in
ei
gh
t
h
sect
ion.
In
the
ni
nth
s
ect
ion
we
fini
sh
by
a co
nclusi
on.
2.
RELATE
D
W
ORK
In
A
Se
veral
stud
ie
s
prov
i
de
diff
e
re
nt
cl
us
te
rin
g
te
ch
nique
s
to
im
pr
ov
e
t
he
net
wor
k
sca
la
bili
ty
and
si
m
plifyi
ng
the
com
plexity
of
r
ou
ti
ng
pro
ble
m
to
sm
aller
gro
ups
of
no
des
.
T
hey
usual
ly
diff
e
r
on
the
cr
it
eria
of
sel
ect
io
n
of
cl
us
te
r
hea
d.
I
n
this
pa
rt
we
pro
po
se
s
om
e
work
done
a
nd
that
fo
c
us
es
on
dif
fer
e
nt
crit
eria
for
the
g
e
ner
at
io
n
of
the
cl
us
te
r
head,
so
m
e
of
this
researc
h
a
re
i
m
ple
m
ented
in
the
OLS
R
[5]
(
Op
ti
m
iz
e
d
Lin
k
Stat
e Rou
ti
ng
Pr
ot
oc
ol)
to
im
pro
ve
it
Qos.
Mob
il
it
y
Ba
sed
Me
tric
for
Cl
us
te
rin
g
(MOB
IC):
is
t
he
or
i
gi
nal
pro
posal
of
Ba
s
u
et
al
[6
]
sug
gesting
that
cl
us
te
r
in
g,
especial
ly
cl
ust
er
hea
d
el
ect
ion
,
m
us
t
con
s
ider
m
ob
il
it
y
as
a
rele
van
t
crit
erion
in
order
t
o
ens
ur
e
a
certai
n
sta
bili
ty
of
gen
e
rated
cl
ust
ers.
T
his
al
go
rithm
is
based
on
a
l
ocal
m
ob
il
it
y
m
et
ric
cal
le
d
"relat
ive
m
ob
il
it
y
of
nodes",
it
is
rev
eal
ed
that
the
node
w
it
h
the
low
value
is
the
le
ast
m
ob
il
e,
ie
the
m
os
t
sta
ble. T
her
e
f
ore, it
is the
node
that is el
ect
ed
as
a cluste
r h
ead.
New
ap
pr
oac
h
nam
ed
SA
L
S
A
pr
ese
nted
in
[
7
]
,
it
is
a
distrib
uted
a
nd
sel
f
-
or
gan
isi
ng
cl
us
te
ri
ng
schem
e
assigni
ng
eq
ual
cl
us
te
r
m
anag
em
ent
ta
sk
s
to
al
l
no
de
s.
I
n
a
ddit
ion,
a
cl
us
te
r
balan
ci
ng
m
echan
is
m
is
introd
uced
al
lo
wing
no
des
to b
e
eve
nly
distr
ibu
te
d
am
ong
cl
us
te
rs.
Be
for
e
the
m
axi
m
u
m
capaci
ty
of
a
cl
us
te
r
is
reac
hed,
it
pro
gr
essi
vely
sta
rts
assig
ning
nodes
to
ne
ighbor
cl
us
te
r
s.
T
his
c
ontribu
ti
on
al
so
pr
opos
e
s
a
cl
us
te
r
qual
it
y
m
et
ric
in
order
to
assi
gn
node
s
to
the
m
os
t
su
it
able
cl
us
te
rs,
acco
rd
i
ng
to
connecti
vity
and
fr
ee
posit
ion
s
within
cl
us
te
rs
.
Re
su
lt
s
c
onfi
rm
ed
the
pe
rfo
rm
ance
eff
ic
ie
ncy
of
t
he
ne
w
sc
hem
e,
pro
vid
in
g
sta
bili
ty
an
d
lo
w
m
ai
ntenan
ce
over
hea
d,
e
ve
n
in
the la
rg
e
st netw
orks
.
An
optim
iz
ed
sta
ble
cl
us
te
ri
ng
al
gorithm
fo
r
Ad
H
oc
(
O
SCA)
pro
pose
d
by
[8
]
,
that
will
prov
i
de
m
or
e
sta
bili
t
y
to
the
netw
ork
by
m
ini
m
iz
ing
the
cl
us
te
r
head
c
ha
ng
es
and
reducin
g
cl
us
te
rin
g
ov
e
rh
ea
d.
In
this
al
go
rith
m
,
a
new
n
ode is
introd
uced wh
ic
h
act
s
as
a
bac
kup
node
i
n
the
cl
us
te
r.
S
uch
bac
kup
no
de
act
s
as
cl
us
te
r
head,
w
hen
act
ual
cl
us
te
r
hea
d
m
ov
e
s
out
(
or
died)
from
the
clu
ste
r.
This
pr
a
ct
ic
e
keep
s
net
w
or
k
avail
abili
ty
wit
hout
distu
rb
a
nc
e.
Fu
rt
her,
th
e
pr
io
rity
of
cl
us
te
r
hea
d
an
d
backup
no
de
is
cal
culat
ed
based
on
the
node
degre
e
and
the
rem
a
ining
batte
ry
li
fe
for
m
ob
il
e
nodes
.
Acc
ordi
ng
to
t
he
ex
pe
rim
ental
resu
lt
s
that
pro
po
se
d
a
n
optim
iz
ed
sta
ble
cl
us
te
rin
g
al
gorithm
fo
r
m
ob
il
e
ad
ho
c
net
works
(
OS
C
A)
al
gorithm
,
it
will
no
t
on
ly
be
able
t
o
m
ake
a
net
work
m
or
e
sta
ble
by
reducin
g
num
ber
of
c
luster
head
c
ha
ng
e
s
but
al
so
reduc
e
the clusteri
ng
ov
e
r
-
hea
d.
The
a
uthors
in
[9
]
pr
ese
nt
a
new
ap
proac
h
to
buil
d
cl
us
te
rs
f
or
W
i
reless
Senso
r
Net
wo
rk
s
(
WSN)
.
The
al
go
rithm
is
base
d
on
t
he
k
-
m
eans
m
et
hod
w
hic
h
is
well
know
n
a
s
a
cl
us
te
ri
ng
te
chn
iq
ue.
K
-
m
eans
cl
us
te
rin
g
te
nds
to
fi
nd
cl
ust
ers
of
com
par
able
s
patia
l
extent
(
de
ns
it
y
cl
us
te
ring).
They
try
to
e
nh
a
nce
the
cl
us
te
rin
g
process b
y
sel
ect
ing
no
des
as
cl
us
te
rs
that
ar
e
centric
an
d
ha
ve
a
high
le
ve
l
of
ene
rg
y.
T
hi
s
will
giv
e
the
sam
e
QoS
res
ults
as
giv
en
by
the
K
-
m
eans
appr
oach
with
a
re
du
ct
io
n
of
en
e
rg
y
co
nsum
pti
on
a
nd
a
prolo
ng
at
io
n
of
the
li
feti
m
e
of
the
sensor
net
wor
k.
Fo
r
the
sim
ulati
on
purpos
es,
the
auth
or
s
have
i
m
ple
m
ented
our
a
ppr
oac
h on the
OLS
R
rout
ing
prot
oco
l.
T
he
a
pproach
pr
opos
e
d
seem
s
to
giv
e
bette
r re
su
lt
s
than
t
he
Ma
xM
in approac
h.
In
[
10
]
an
i
m
pr
ovem
ent
of
protoc
ol
OLS
R
Ma
xMin2
c
wa
s
propose
d
by
the
introd
uctio
n
of
the
c
os
t
of
e
nergy.
T
he
m
ai
n
obj
ect
ive
of
t
his
ne
w
c
ontrib
utio
n,
cal
le
d
O
LSRM
ax
Mi
n2
C
/
Ene
r
gy
,
is
the
optim
i
zat
ion
of
e
ner
gy
c
on
s
um
ption
OL
S
RM
axMi
n2
C/
Energy.
It
co
nsi
sts
of
div
idi
ng
the
netw
ork
into
cl
us
te
rs.
Cl
us
te
r
heads
are
el
ect
ed
acco
rd
i
ng
t
o
their
I
Ds
a
nd
their
resid
ual
energies
(
batte
ry
le
vels);
this
al
gorithm
dete
rm
ines
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
& C
om
p
Eng
IS
S
N: 20
88
-
8708
En
hance
nig
O
LS
R
routi
ng
protoc
ol usi
ng K
-
me
ans clu
ste
ring i
n MAN
ETs
(
Y.
H
amz
aoui
)
3717
the
best
pat
h,
i
n
te
rm
s
of
ene
rg
y
co
st.
This
inv
olv
es
cal
c
ulati
ng
the
e
nerg
y
req
ui
red
for
each
avail
able
path.
It
is
an
al
go
rit
hm
that
op
tim
i
zes
the
nu
m
ber
of
li
ve
nodes
by
al
ways
choosin
g
the
ap
propriat
e
nod
es
f
or
eac
h
ta
sk
in
the
net
work
.
3.
PR
OBL
EM
FOR
M
ULAT
ION
The
dev
el
op
m
ent
of
te
ch
nolo
gy
an
d
the
r
e
voluti
on
of
wire
le
ss
te
chnolo
gy
hav
e
le
d
to
the
e
xistence
of
m
ob
il
e
de
vi
ces
in
al
l
areas
of
hum
an
act
ivit
y.
The
c
onne
ct
ion
of
t
hese
m
ob
il
es
in
ad
hoc
m
od
e
bec
om
es
a
necessit
y,
by
their
sim
pli
ci
ty
of
dep
l
oym
ent
and
i
n
the
a
bs
e
nce
of
a
pr
ee
xis
ti
ng
a
nd
ex
pe
n
sive
infr
a
struct
ur
e
.
Wh
e
n
the
nu
m
ber
of
m
ob
il
e
i
ncr
ease
s
an
d
the
topol
og
y
ch
ang
e
s
rap
i
dly
by
the
high
m
ob
il
it
y
of
nodes;
su
c
h
a
pr
oacti
ve
r
outi
ng
protoc
ol
cannot
sup
port
the
netw
ork
e
vo
l
ution,
due
to
the
ge
ne
rati
on
of
m
or
e
m
essage
con
t
ro
l
a
nd
routin
g
ta
bl
e,
et
c.
Co
ns
e
qu
e
ntly
the
ne
twork
bec
om
es
m
or
e
sen
sit
ive,
and
it
do
es
n'
t
ensure
a
m
ini
m
um
qu
al
it
y
of
s
erv
ic
e.
De
velo
ping
or
ref
ini
ng
m
et
ho
ds
that
will
ov
e
rco
m
e
these
ob
sta
cl
es
bec
om
es
a
necessit
y.
Ma
ny
stud
ie
s
on
r
ou
ti
ng
optim
iz
at
ion
ad
op
t
the
cl
us
te
ri
ng
m
et
ho
d
to
r
edu
ce
the
costs
pro
duct
by
de
ns
it
y
and
m
ob
il
it
y.
It
is
base
d
on
the
distrib
ut
io
n
of
m
ob
il
es
nodes
i
nto
gro
ups.
This
a
ppro
ac
h
is
pro
po
se
d
t
o
reduce
the
sto
r
age
a
nd
proces
sing
i
nfor
m
at
i
on
withi
n
a
cl
ust
er.
I
n
t
he
li
te
rature
we
can
hav
e
s
ever
al
cl
u
ste
ring
te
c
hn
i
qu
es;
the
K
-
m
eans
cl
us
te
rin
g
te
ch
nique
is
am
on
g
the
best
m
eth
od
s
known
[
1
1
,
1
2
]
us
ed
in
MAN
ETs.
This
one
will
al
low
cl
us
te
rs
to
be
m
or
e
sta
ble
and
the
center
will
be
bette
r
chosen
.
Choos
ing
a
le
ss
m
o
bile
and
den
se
cl
us
te
r
hea
d
without
ta
king
into
acco
un
t
their
resid
ual
energ
y
le
vel;
can
res
ult
in
t
he
is
olati
on
of
al
l
cl
us
te
r
m
e
m
ber
s
from
the
res
t
of
the
net
w
ork,
in
t
he
ca
se
of
the
ex
ha
us
ti
on
of
the
batte
r
y
[
1
3
]
.
T
he
i
m
pl
ic
at
ion
of
the
co
nce
pt
of
ene
rg
y
i
n
the
process
of
cl
us
te
rs
const
ru
ct
io
n
av
oid
s
the elect
io
n of a
node, we
ll
p
la
ced
but w
it
h
wea
k
e
nerg
y l
evel.
In
this
pap
e
r
we
pro
pose
th
e
us
e
of
K
-
m
eans
to
pr
oduce
op
ti
m
al
cl
us
te
rs.
T
his
m
e
thod
is
base
d
on
the
us
e
of
E
ucl
idean
distance
to
ge
ner
at
e
cl
ust
er
centers
.
I
n
our
pro
posit
io
n
we
will
introdu
ce
i
n
the
K
m
eans
al
gorithm
the
t
hr
ee
sta
bili
zat
ion
par
am
et
ers
(m
ob
il
it
y,
den
sit
y,
ener
gy)
as
a
distance
vect
or.
The
cl
us
te
r
hea
d
will
b
e ele
ct
ed
accor
ding to
th
ese three p
a
ra
m
et
ers.
Th
e center of
the clus
te
r
will
b
e the o
ne
that re
sp
ec
ts bo
t
h
the
t
hr
ee
sta
bil
it
y
par
am
et
ers
in
it
gro
up,
it
will
be
le
ss
m
ob
il
e,
de
ns
e
r
an
d
has
a
n
ene
r
gy
le
vel
that
inc
reases
the
li
fe
of
t
he
cl
us
te
r.
T
his
a
ppr
oach
al
lo
w
s
to
buil
d
the
cl
us
te
rs
that
ar
e
highly
resist
ant
to
the
c
onstrai
nt
structu
re
of
th
e
ad
hoc
net
w
ork.
It
is
i
m
pl
e
m
ented
to
sta
nd
a
r
d
OL
SR
prot
oco
l
gi
ving
birth
a
new
protoc
ol
nam
ed
OLS
R
-
Km
eans
-
SDE.
4.
CLUS
TE
RI
N
G
In
S
patia
l
cl
us
te
rin
g
al
gorith
m
s
can
be
cl
assifi
ed
into
fou
r
cat
ego
ries
.
Th
ey
are
the
par
t
it
ion
base
d,
the
hie
rar
c
hical
based,
the
de
ns
it
y
base
d
a
nd
t
he
gri
d
base
d
[
1
4
,
1
5
]
.
Ac
cordin
g
to
[
1
6
]
cl
us
te
rin
g
i
n
a
d
hoc
netw
orks
ca
n
be
de
fine
d
as
a
theo
reti
cal
arr
a
ng
em
ent
of
dy
nam
ic
no
des
c
orrespo
ndin
g
to
one
or
m
or
e
sp
eci
fic
pro
pe
r
ti
es
in
diff
ere
nt
su
bs
et
s
cal
le
d
"C
luster".
A
n
el
em
ent
of
a
cl
us
te
r
is
chara
ct
e
rized
by
a
strong
si
m
il
arity
of
the
c
om
po
ne
nts
of
it
s
gro
up,
an
d
a
str
ong
dissim
il
arit
y
w
it
h
res
pect
to
t
he
m
e
m
ber
s
of
ot
he
r
gro
up
s
[
1
7
]
.
E
ach
cl
us
te
r
is
identifie
d
by
a
par
ti
cula
r
node
cal
le
d
"C
luste
r
head
"
.
Cl
us
te
rin
g
al
lo
ws
a
node
t
o
store
only
par
t
of
rathe
r
tha
n
al
l
the
inform
at
ion
of
the
ne
twork
to
polo
gy
.
This
si
m
plifie
s
the
pr
oce
ss
ing
of
the
global
topolo
gy
[1
8
]
.
This
reduces
th
e
siz
e
of
routing
ta
bles
an
d
thereafte
r
the
reducti
on
of
c
on
t
ro
l
m
essages
gen
e
rated
by the
ro
uting sy
ste
m
[
1
9
].
The
us
e
of
c
l
ust
ering
i
n
M
A
NETs
has
seve
ral
adv
a
ntage
s
[
20
,
2
1
]
,
us
uall
y
a
cl
us
te
r
struc
ture
al
lows
the no
de
to
p
la
y on
e
of t
hr
ee
r
oles:
Cl
us
te
r
he
ad:
A
cl
us
te
r
head
is
el
ect
ed
in
the
cl
us
te
r
f
orm
at
ion
proces
s
f
or
eac
h
cl
ust
er.
Eac
h
cl
us
te
r
sh
oul
d hav
e
on
e an
d on
ly
on
e
cl
us
te
r head
.
Gateway
:
A
node
is
cal
le
d
a
gateway
node
of
a
cl
us
te
r
if
it
kn
ows
th
at
it
has
a
bidi
recti
on
al
or
uni
-
directi
onal
li
nk to
a
node
fro
m
anothe
r
cl
us
te
r.
Mem
ber
s: All
nodes
w
it
hi
n
a
cl
us
te
r
e
xce
pt the clu
ste
r hea
d are cal
le
d m
e
m
ber
s o
f
this c
lu
ste
r.
5.
DESCRIPTI
ON OF
APP
R
OCH CL
US
T
ERING
ALG
ORI
TH
M
In
the
abse
nce
of
a
ny
ass
umpti
on
a
bout
th
e
distri
bu
ti
on
of
no
des
i
n
a
m
ob
il
e
ad
ho
c
net
wor
k,
an
un
s
uper
vise
d
cl
assifi
cat
io
n
of
nodes
int
o
cl
asses
is
re
qu
i
red.
W
e
pr
opos
e
a
m
od
el
base
d
on
ge
om
et
ri
c
consi
d
erati
ons
(gr
ouping
ge
ogra
ph
ic
al
ly
cl
os
e
nodes
)
.
T
his
pr
opos
al
requests
def
i
ne
a
m
easur
e
of
the
pro
xim
i
ty
betwee
n
no
des
.
W
e
beg
i
n
by
introd
ucin
g
the
con
ce
pts
of
si
m
il
arity,
dissim
il
arity
and
di
sta
nce
betwee
n
node
s,
a
nd
the
n
we
will
us
e
t
he
ine
rtia
co
nc
ept
s
intr
a
-
cl
a
ss
an
d
i
nter
-
c
la
ss
to
dem
on
strat
e
m
at
he
m
at
ic
a
lly
how
a
cl
ass
ific
at
ion
m
et
ho
d
buil
t
hom
og
ene
ous
a
nd
di
sti
nc
t
gro
up
s
by
tra
ns
la
ti
ng
it
into
a sim
ple
m
inim
iz
at
ion
p
r
oblem
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
& C
om
p
Eng
,
V
ol.
10
, N
o
.
4
,
A
ugust
2020
:
3
7
1
5
-
3
7
2
4
3718
C i
s a p
a
rtit
ion o
f
X
if a
nd
on
l
y i
f
C
sat
isfie
s
the foll
owin
g p
roper
ti
es:
⊂
X
for al
l
X
⋃
=
1
= C
∩
=
∅
f
or
(a,
b) s
uc
h t
hat
a ≠
b
The
pro
pe
rty
(
3)
e
xpresses
that
the
cl
ust
er
s
f
or
m
ed
a
re
di
sj
oi
nt;
each
obj
ect
of
X
belo
ngs
only
to
a
sin
gle
cl
us
te
r
of
C
.
T
o
def
ine
the
ho
m
og
e
ne
it
y
of
ob
se
r
v
at
ion
s
set
,
it
is
ne
cessary
to
m
e
asur
e
a
sim
il
ar
it
y
betwee
n
tw
o
obser
vatio
ns
.
T
hen
we
int
rod
uce
the
c
on
c
e
pts
of
sim
i
la
rit
y
and
dissim
ilarity
.
Dissim
ilarity
is
a f
un
ct
io
n d w
hich
a
sso
ci
at
es
a v
al
ue
in
I
R
+ to eac
h pair
(
x
i
x
j
)
su
c
h
t
hat:
(
,
)
=
(
,
)
(
,
)
=
0
→
=
Conver
sel
y
,
a
no
t
her
possi
bili
ty
is
to
m
e
asur
e
the
res
e
m
blance
between
obse
rv
a
ti
on
s
us
i
ng
a sim
il
arit
y.
Si
m
il
arity is a functi
on s
t
o wh
i
ch
ass
ociat
es a
value
i
n IR* t
o ea
ch pair
(
x
i
x
j
)
su
c
h
that:
(
,
)
=
(
,
)
≥
0
(
,
)
≥
(
,
)
Thec
ho
ic
e
of
the
distance
is
a
key
issue
fo
r
cl
assifi
cat
ion
m
et
ho
ds.
To
offe
r
a
relevan
t
m
easur
e
of
si
m
il
arity
betw
een elem
ents, it i
s n
ecessa
ry to
well
us
e
the
avail
able in
for
m
at
ion
at the
node
s.
The
Mi
nk
owsk
i
distance is
th
e m
os
t used
t
o determ
ine the s
i
m
i
la
rity
b
et
ween
elem
ents:
(
,
)
=
(
∑
|
(
)
−
(
)
|
=
1
)
1
⁄
Wh
e
re
vk (xi
) i
s the
value o
f t
he
ob
j
e
ct
xi on
the
v
a
riable
vk.
Dep
e
ndin
g
on
the v
al
ues
ta
ke
n by the
pa
ram
et
er l, we tal
k
a
bout:
Eucli
dea
n dist
ance
(l =
2)
;
Ma
nh
at
ta
n dist
ance
(l =
1)
;
Chebyc
hev d
is
ta
nce (
l =
∞).
We
note
that
the
m
e
tric
s
com
m
on
ly
us
ed
to
analy
ze
the
ad
hoc
pe
rfo
rm
ances
su
c
h
as
den
sit
y,
m
ob
il
i
ty
and
e
nergy
can
be
use
d
to
e
xpres
s
distance.
O
ur
obj
ect
ive
is
t
o
div
ide
the
node
s
into
ho
m
ogeneous
and
disti
nct
cl
us
te
rs
.
To
do
this,
we
sta
rt
from
the
def
in
it
ion
of
net
work
i
ner
ti
a
that
can
be
m
od
el
ed
by
a cloud
of
po
i
nt
s.
Con
si
der
a
network o
f n
node
s (
x
1
,
…
x
n
)
an
d U
G de
sign
at
es t
he
ce
ntr
oid
of no
des
cloud
.
=
1
∑
=
1
(1)
The
cl
us
te
r
i
ne
rtia
is b
y
def
i
niti
on
the
s
um
o
f
square
s
of
dist
ances
from
the cen
te
r:
=
∑
2
(
,
)
=
1
(2)
=
∑
‖
−
‖
2
=
1
(3)
Assum
e
that
t
he
net
wor
k
co
ns
ist
s
of
P
disti
nct
cl
us
te
rs
C
1
,
,
,
C
P
.
Each
of
the
se
cl
us
te
rs
ha
vi
ng
as
centr
oid
U
ck
.
We c
an
the
n dec
ompo
s
e the
total
iner
ti
a
of
t
he
cl
oud o
f nodes a
s foll
ow
s:
=
∑
∑
‖
−
‖
2
=
1
=
1
=
∑
∑
‖
−
+
−
‖
2
=
1
=
1
Acco
r
ding to
the the
orem
o
f
Hu
y
gen
s
,
=
∑
∑
(
‖
−
‖
2
+
‖
−
‖
2
)
∈
=
1
=
(
∑
∑
2
(
,
)
∈
=
1
)
+
(
∑
2
(
,
)
=
1
)
(4)
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
& C
om
p
Eng
IS
S
N: 20
88
-
8708
En
hance
nig
O
LS
R
routi
ng
protoc
ol usi
ng K
-
me
ans clu
ste
ring i
n MAN
ETs
(
Y.
H
amz
aoui
)
3719
The
fi
rst
te
rm
cal
le
d
intra
-
cl
us
te
r
ine
rtia
cal
culat
es
the
su
m
of
distanc
es
betwee
n
no
des
an
d
thei
r
centr
oid
.
Lo
w
intra
-
cl
ust
er
in
erti
a
ind
ic
at
es
that
the
nodes
in
the
sam
e
cl
us
te
r
a
re
m
or
e
near
e
r
(Cluste
rs
are
ho
m
og
e
neous
).The
sec
ond
te
rm
called
the
inter
-
cl
ust
er
inerti
a
cal
culat
es
the
su
m
of
distances
betwee
n
centr
oid
s
of cl
us
te
rs
and
glob
al
centr
oid
,
tha
t i
s to
say
the
s
epar
at
io
n de
gree o
f
cl
us
te
rs.
=
+
(5)
Fr
om
a for
m
al
p
oi
nt of
view
, t
he op
ti
m
al
p
arti
ti
on
is
:
That m
ini
m
iz
e
s the i
ntra
cl
ass
iner
ti
a
Or
t
hat m
axi
m
i
zes the i
nter
cl
ass ine
rtia
,
Th
us
the
opti
m
al
p
arti
ti
on wo
uld
be de
fine
d as f
ollows:
=
∈
∑
∑
2
(
,
)
∈
=
1
(
6
)
(6)
The o
bj
ect
ive
of cla
ssific
at
io
n
al
go
rithm
s b
ased
on this
pri
nciple is t
he
se
arch o
f op
ti
m
al
pa
rtit
ion
.
In
pract
ic
e
it
i
s
i
m
po
ssible
to
ge
ner
at
e
al
l
patte
rn
s
of
cl
ust
ering
f
or
evi
den
t
reas
ons
of
com
plexity
.
We
the
n
see
k
a
schem
e
su
ff
ic
ie
ntly
cl
os
e
to
the
opti
m
u
m
.
T
his
opti
m
u
m
is
ob
ta
ine
d
in
a
n
it
erati
ve
m
anner
by
i
m
pr
ovin
g
an
init
ia
l
schem
e
rand
om
l
y
sel
e
ct
ed
by
reall
oc
at
ing
obj
ect
s
arou
nd
m
ob
il
e
centers
.
I
n
order
t
o
pa
rtit
ion t
he n
odes i
nto
cl
us
te
r
s,
we
used
this
te
chn
iq
ue (it
er
at
ive r
eal
locat
i
on) base
d on k
-
m
eans algorit
hm
.
5.1
.
K
-
me
an
s
met
ho
d
The
k
-
m
eans
al
gorithm
[1
5
]
is
a
par
ti
ti
on
i
ng
m
et
ho
d
wi
del
y
us
e
d
in
var
i
ous
ap
plica
ti
on
areas.
F
ro
m
P
sepa
rate
cl
us
te
rs,
the
k
-
m
ea
ns
al
gorithm
assigns
it
erati
ve
ly
obj
ect
s
(
x
1
,
…
x
n
)
at
P
ce
nters
of
cl
us
te
r
s
(
u
1
,
…
u
P
)
,
f
ollow
e
d
by
cal
culat
in
g
t
he
posit
io
ns
of
the
new
ce
nter
s.
T
he
sto
ppin
g
al
gorithm
is
a
c
rite
rio
n
fixe
d by the
use
r
a
nd can
b
e:
Ach
ie
ve
a li
m
i
te
d
num
ber
o
f
i
te
rati
on
s;
The
al
gorithm
co
nve
rg
es:
cl
ust
ers
f
orm
ed
rem
ai
n
the sam
e b
et
wee
n
tw
o s
uccessive
it
erati
on
s;
The
i
ner
ti
a
i
ntra
-
cl
ust
er
is
not
im
pr
ov
i
ng
betwee
n
t
wo
it
erati
on
s
(th
e
al
gorithm
is
su
f
fici
ently
cl
ose
t
o
conve
rg
e
nce).
To
j
ust
ify
the
K
-
m
eans
al
gor
i
th
m
in
view
of
ou
r
goal
ai
m
ed
to
m
ini
m
ize
the
intr
a
-
cl
us
te
r
inerti
a,
we
de
m
on
strat
e
that
the
re
de
finiti
on
of
ce
nters
of
cl
us
te
rs
a
nd
t
he
reall
ocati
on
of
no
des
(S
e
quence
of
tw
o
sub
-
sta
ges
of
K
-
m
eans
al
gorithm
)
resu
lt
s
in
a
decr
ease
in
in
tra
-
cl
ass
inerti
a.
W
e
sta
rt
fro
m
fo
ll
owin
g
c
ons
iderati
ons
at
t
he
end
of d
et
er
m
ined
ste
p j;
Ce
nters of
clus
te
rs
(
u
1
j
,
…
.
u
p
j
)
hav
e
b
ee
n ca
lc
ulate
d;
The
cl
asses
(
C
1
j
,
…
.
C
p
j
)
wer
e
ob
ta
ine
d
by
assi
gn
i
ng
at
the
cente
r
(
C
k
j
)
the
n
j
i
near
e
st
nodes.We
de
fi
ne
the
f
ollo
wi
ng
qual
it
y at
the end of
step
j
;
=
∑
∑
‖
−
‖
2
∈
=
1
The rede
finiti
on s
ub
-
ste
p of
ne
w
ce
nters
a
nd
r
eassi
gn
i
ng of
nodes
(next it
e
rati
on) req
uires
:
a.
Re
cal
culat
e
ce
nters
of
cl
us
te
r
s
(
u
1
j
+
1
,
…
.
u
p
j
+
1
)
base
d
o
n
po
i
nts
belo
ng
t
o
e
ach
of
cl
us
te
rs
(
C
1
j
+
1
,
…
.
C
p
j
+
1
)
po
s
sessin
g res
pecti
vely
n
k
j
el
e
m
ent.
We
have:
+
1
=
1
∑
∈
a
nd:
+
1
=
∑
∑
‖
−
+
1
‖
2
∈
=
1
w
he
re
u
k
j
+
1
the
ce
nter
of
t
he
c
luster
is
C
k
j
et
W
j
+
l
is
t
he
i
netria
i
ntra
-
cl
ust
ers
cl
ust
ers
ass
os
ia
te
d
t
o
cl
us
te
rs
(
C
1
j
,
…
.
C
p
j
)
. We
will
h
ave:
=
∑
∑
‖
−
‖
2
∈
=
1
=
∑
∑
‖
−
+
1
+
+
1
−
‖
2
∈
=
1
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
& C
om
p
Eng
,
V
ol.
10
, N
o
.
4
,
A
ugust
2020
:
3
7
1
5
-
3
7
2
4
3720
=
∑
∑
(
‖
−
+
1
‖
2
+
‖
+
1
−
‖
2
+
2
〈
−
+
1
,
+
1
−
〉
)
∈
=
1
=
+
1
+
∑
∑
‖
+
1
−
‖
2
∈
+
2
∑
〈
∑
(
−
+
1
)
∈
,
+
1
−
〉
=
1
=
1
≥
+
1
The vect
or
∑
(
x
i
−
u
k
j
+
1
)
i
∈
C
k
bei
ng the
null
v
ec
tor by
def
init
io
n of
u
k
j
+
1
b.
Re
assign n
ode
s to
t
he nearest
centers
. T
he
n we
ob
ta
in
n
e
w cl
us
te
r
s
(
C
1
j
+
1
,
…
.
C
p
j
+
1
)
an
d w
e d
e
fin
e
:
+
1
=
∑
∑
‖
−
+
1
‖
2
∈
=
1
Af
te
r
r
eassi
gn
m
ent
al
l
distances
dec
rease
because
each
node
xi
is
assigne
d
to
the
cl
us
te
r
center
m
ini
m
iz
ing
the
reb
y t
he gap:
‖
−
+
1
‖
2
We th
e
refo
re
ha
ve:
+
1
≤
+
1
Th
us
, f
or
eac
h j we
hav
e
pr
oved
the
foll
owin
g
ine
qual
it
y:
+
1
≤
+
1
≤
So
we ha
ve
in
par
ti
cula
r:
+
2
≤
+
1
≤
+
1
This
sho
ws
th
at
after
each
a
lgorit
hm
i
te
rati
on
the
im
pr
ov
e
m
ent
of
the
node
s
cl
assifi
cat
ion
is
sure
within
the
m
ea
ning
of
I
in
tra
crit
erio
n.
Be
cause
t
he
intra
-
cl
us
te
r
i
ner
ti
a
of
t
he
optim
al
par
ti
ti
on
is
the
sm
al
ler
,
the
m
arg
in
of
i
m
pr
ov
em
ent
is
finite
.
This
im
plies
that
the
a
lgorit
hm
con
ve
rg
es
i
nev
it
ably
.
The
disad
va
nt
age
of
the
k
-
m
eans
m
et
ho
d
is
that
the
nu
m
ber
of
cl
us
te
rs
is
a
par
am
et
er
of
the
al
go
rithm
.
This
is
no
t
obviou
s
fo
r
a
m
ob
il
e
ad
hoc
netw
ork
w
her
e
no
des
ca
n
j
oi
n
or
le
a
ve
the
netw
ork
rand
om
ly
.
Hen
ce
the
nee
d
t
o
m
ake
i
m
pr
ovem
ents to
the m
et
hod
t
o be
a
pp
li
ca
ble to our case:
In
it
ia
l
num
ber
of
Cl
us
te
rs:
At
the
be
ginning
each
no
de
represe
nts
it
s
own
cl
us
te
r.
T
her
ea
fter
we
m
ake
a
series
of
asc
end
i
ng
par
ti
ti
ons
by
c
om
bin
ing
t
he
no
des
be
longin
g
to
th
e
sa
m
e
neighb
orh
ood
into
sa
m
e
cl
us
te
r un
ti
l re
achin
g
a
st
able
num
ber
of clu
ste
rs.
Dista
nce:
We c
an use t
he dens
it
y,
m
ob
il
it
y or
en
e
rg
y
of no
de
s to
e
xpress
th
e d
ist
ance
m
et
r
ic
.
6.
METRI
C MO
BIL
ITY
The
m
et
ric
[2
2
]
pro
pose
d
al
lows
to
cal
c
ulate
the
sta
bili
ty
of
a
node
ba
sed
on
four
par
am
et
ers,
the
node
that
le
aves
the
co
ve
rag
e
area
of
node
stud
y
,
the
node
s
that
j
oi
ns
the
area
zo
ne,
the
node
ap
pro
aches
to
the
stud
ie
d
node
an
d
the
node
that
final
ly
m
ov
es
away
fr
om
the
exa
m
ined
node
a
nd
that
sta
ys
in
their
cov
e
ra
ge
a
rea.
The
first
t
wo
pa
ram
et
ers
will
be
retai
ned
by
the
c
ollec
te
d
c
on
t
ro
l
m
essage
s.
T
he
la
st
t
wo
will
be
cal
culat
ed
by
the
cal
c
ulati
ng
powe
r
f
or
two
s
uccessi
ve
receive
d
m
essages
(eg
Hell
o
m
essage
in
O
LSR
protoc
ol).
W
e
def
i
ne
the
foll
owin
g param
et
e
rs
that c
ha
racteri
ze o
ur m
et
ric
:
N
con
de
fines
the
num
ber
of
node
s
t
hat
co
nver
ge
on
the
stu
died
node.
N
div
de
fines
th
e num
ber
of
nodes wh
ic
h dive
rg
e
to
wards t
he
outsi
de of
the
nod
e
stu
died
.
N
in
de
fines
the
nu
m
ber
of
no
de
s
withi
n
the
ar
ea
of
the
stu
die
d
node.
N
out
de
fines
th
e num
ber
of
nodes o
ut of th
e
c
ov
e
ra
ge
are
a
of
the
stu
died n
od
e
Fo
ll
owin
g
the
s
e
f
our
ty
pes
of
m
ov
e
m
ent
we
ha
ve
c
reated
a
m
e
tric
that
will
rank
t
he
node
s
betwee
n
any
of
these f
our
m
et
r
ic
s o
f
stabil
it
y.
W
e
de
fine
eac
h
as
as s
how
n
i
n
Ta
ble
1
.
Table
1.
Cl
assi
fyi
ng of m
et
ric
s
Cas
es
Mou
v
e
m
en
ts p
ara
m
e
te
rs
Interval o
f
Cas
1
: Better
stab
i
lity
{
Ncon
>
N
in
>
∈
[
0
0
.25
[
Cas
2
: Stable
{
<
>
∈
[
0
.25
0.5
]
Cas
3
: less stab
le
{
>
<
∈
]0.5
0.7
5
[
Cas
4
: po
o
r
stab
ility
{
<
<
∈
[
0
.75
1]
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
& C
om
p
Eng
IS
S
N: 20
88
-
8708
En
hance
nig
O
LS
R
routi
ng
protoc
ol usi
ng K
-
me
ans clu
ste
ring i
n MAN
ETs
(
Y.
H
amz
aoui
)
3721
In
the
f
our
cl
assifi
cat
ion
s,
we
have
the
f
irst
case
t
hat
ref
le
ct
s
a
bett
er
sta
bili
ty
fo
r
the
node
in
qu
e
sti
on
a
nd
t
he
la
tt
er
wh
ic
h
rep
re
sents
a
poor
sta
bili
ty
of
the
stud
ie
d
node,
intuit
ively
the
seco
nd
is
bet
te
r
than
t
he
t
hir
d,
because
N
i
n
is
gr
eat
er
tha
n
N
out
and
s
eco
ndly
,
eve
n
with
N
div
>N
con
the
div
er
ging
node
s
ta
y
in
the
co
ver
a
ge
area
of
the
stud
ie
d
no
de.
We
de
te
rm
ine
subseq
ue
ntly
m
et
ric
deg
re
e
of
sta
bili
ty
that
will
cal
culat
e
fo
r
e
ach
cat
eg
or
y
the
best
sta
bili
ty
node,
we
us
e
in
the
form
ula
(7)
the
coe
ff
ic
ie
nt
γ
of
fl
ow
de
fine
d
in
[
2
3
]
,
we
divi
de
the
c
oe
f
fici
ent
of
4
inter
va
ls
as
show
n
i
n
Ta
ble
1,
an
d
m
et
ric
of
sta
bi
li
t
y
deg
ree
of
node
i
will
b
e as
foll
ows:
(
)
=
(
+
+
+
+
+
)
+
(
1
−
)
(
+
+
+
+
+
)
(7)
The
sta
bili
ty
m
et
ric
is
def
ine
d
i
n
our
pro
posi
ti
on
f
or
m
ula
(7)[2
3
]
.
T
he
disa
dv
a
ntage
of
t
his
approac
h
is
that
t
he
par
a
m
et
er
m
us
t
be
fix
ed
bet
wee
n
thr
ee
values
(
0.25,
0.5,
0.75)
,
acc
ordin
g
to
t
he
flo
w
of
m
otio
n
arou
nd
node
(
do
m
inant
i
nf
lo
w
=
0.75,
do
m
inant
ou
tfl
ow
=
0.25).
T
o
bette
r
ch
oos
e
the
par
am
et
e
r
,
we
hav
e
a
utom
at
ed
the
determ
inati
on
of
t
his
par
am
et
er
by
re
fine
d
t
he
pro
posed
e
quat
ion
.
So
the
f
or
m
ula
beco
m
es (
8):
(
)
=
(
+
+
+
+
+
)
+
(
+
+
+
+
+
)
(8)
=
(
+
+
+
+
)
=
(
+
+
+
+
)
(9)
7.
DESCRIPTI
ON OF
ALG
ORI
TH
M
We
pr
ese
nt b
el
ow
our
al
gorithm
in
detai
l (k
-
m
eans i
m
pr
o
ve
d
to m
ake it app
li
cable t
o
th
e p
arti
ti
on
of
nodes
int
o
cl
ust
ers)
:
The
i
ntrod
uction
of
c
hanges
to
t
he
sta
nd
a
rd
al
gori
thm
is
a
need
,
to
us
e
the
k
-
m
eans
m
et
ho
d
f
or
gr
ouping
node
s
of
a
Ma
nets
i
nto
cl
us
te
rs;
t
his
c
hange
ai
m
s
to
determ
i
ne
the
K
pa
ra
m
et
er
of
t
he
al
gorithm
.
In
fir
st
it
is
assum
ed
that
each
node
in
Ma
ne
ts
is
an
ow
n
cl
us
te
r;
then
f
ollow
e
d
by
a
seq
uen
ce
ascen
ding
pa
rt
it
ion
s,
in
t
he
e
nd
we
reunit
e
the
no
des
f
ro
m
the
sam
e
neigh
bo
rho
od
i
n
th
e
sam
e
cl
us
te
r,
un
ti
l
reachi
ng
a
fina
l
nu
m
ber
of
cl
us
te
rs
.
Th
e
reaf
te
r,
the
K
-
Me
a
ns
al
gorithm
will
be
us
ed
t
o
ge
ner
at
e
m
or
e
s
ta
ble
cl
us
te
rs
with
their
cl
us
te
r
he
ad.
The
par
am
et
ers
us
ed
in
t
he
K
-
m
eans
cal
culat
ion
al
gor
it
h
m
are
the
st
abili
ty
,
densi
ty
and
en
erg
y
resid
ual
of
a
m
ob
il
e,
s
o
each
no
de
is
identifie
d
by
a
vect
or
(sta
bili
ty
,
den
sit
y,
e
nergy
)
.
This
vecto
r wil
l be the
b
a
sis f
or ex
pr
e
ssin
g
t
he dist
ance
bet
ween n
odes.
We
pr
ese
nt
our
algorit
hm
b
as
ed on K
-
m
eans to
c
reate t
he p
arti
ti
on
of
t
he
c
luster
nodes
in M
anets.
Pr
op
os
e
d
al
go
r
it
h
m
Inp
ut: M
ob
il
e
ad ho
c
n
et
wor
k of
n n
odes.
Ou
t
pu
t:
Netw
ork
v
irt
ually
p
ar
ti
ti
on
ed
i
nto
P
cl
us
te
rs
Step
0
In
it
ia
li
zat
ion
wi
th n
ce
nters
(
1
0
…
0
)
eac
h n
od
e
is a cluste
r
Creat
ion
of a
n i
niti
al
p
arti
ti
on P0 = {
(
1
0
….
0
)}
In
it
ia
li
ze
to
1 (
)
.
Assign t
o
0
it
is two
-
ho
p neigbo
ur
s;
0
= {
∈
netw
ork
| d
(
,
0
)>=2hops};
Rem
ov
e f
r
om
l
ist
o
f
ce
nters
th
e
0
no
des
as
sig
ne
d
to
cent
ro
i
d
0
;
Mov
e
to
t
he
ce
nter
l+
0
;
Re
peati
ng step
s b
t
o
do
un
ti
l
al
l node a
re a
ffec
te
d;
Ca
lc
ulati
on
of
new
ce
nters
of
k
cl
us
te
r
obta
ined
(
1
1
…
1
):
no
de
s
ha
ving
the
optim
al
values
of
(stabili
ty
,
densi
ty
, en
er
gy
);
Step t
Creat
ion
of
ne
w partit
ion Pt
= {(
1
….
)} b
y as
sign
i
ng to
eac
h ce
n
te
rs
it
s tw
o
-
ho
p neig
hbors
;
The
ce
nters
af
f
ect
ed
to
ot
her
c
enters a
re
rem
ov
e
d from
the list
o
f
ce
nters;
The
is
oled n
odes are a
ssig
ned to the
li
st o
f
c
enters;
Ca
lc
ulate
the c
enters
of
k
cl
ust
ers
obta
ine
d (
1
0
…
0
);
Re
peat
ste
ps
3
to
6
unti
l
t
hat
a
sta
ble
pa
rtit
ion
is
ach
ie
ved
(
struct
ure
of
par
ti
ti
on
+
equ
al
s
tha
t
the
+
+
1
) or reac
h n i
te
rati
on
s.
We
use
d
t
he
NS2.
34
sim
ulator
to
a
naly
ze
the
qual
it
y
of
serv
ic
e
betwee
n
O
LSR
Kea
ns
-
S
DE
an
d
the
two
oth
e
r
protoc
ols
(T
he
cl
assic
al
K
m
e
an
s
im
ple
m
ent
ed
in
OLS
R
a
nd
OLS
R
Me
d+
[2
4
]
).
We
va
li
date
this
analy
sis
by
stu
dying
t
he
be
ha
vior
of
the
protoc
ol
in
te
rm
s
of
PD
R,
E
ED
and
O
verhea
d
[25]
.
The follo
wing
T
able
2
s
hows
the sim
ulati
on
p
aram
et
ers.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
& C
om
p
Eng
,
V
ol.
10
, N
o
.
4
,
A
ugust
2020
:
3
7
1
5
-
3
7
2
4
3722
T
able
2.
Sim
ul
at
ion
par
am
et
e
rs
Para
m
eter
Valu
e
Proto
co
l
OLSR,
OLSR
K
m
eans
-
SDE
,
O
LSR
Med+
Si
m
u
latio
n
ti
m
e
1200s
So
u
rce
an
d
ty
p
e of
tr
af
f
ic
CB
R/UDP
Netwo
rk size
1
0
0
0
m
x
1
0
0
0
m
Size paq
u
et
5
1
2
bytes
Sp
eed n
o
d
es
1
0
m
/s
Mob
ility
m
o
d
el
RWP
Nu
m
b
e
r
o
f
no
d
es
1
0
20
30
4
0
50
1
0
0
8.
RESU
LT
S
AND DES
C
USSION
8.1.
Packe
t
deli
ve
r
y
r
at
i
o
We
no
ti
ce
i
n
t
he
F
i
gure 1
,
th
at
wh
e
n
t
he
net
work b
ecom
es
dense
a
nd
la
r
ge
the num
ber
of
lost pac
ket
increases
,
an
d
this
is
becau
se
of
di
ff
e
ren
t
ph
eno
m
enon
(to
polo
gy
change,
batte
ry
exh
a
us
t
ion
...);
the
ret
ai
ned
structu
re
a
nd
ref
ine
d
a
lg
or
it
hm
in
the
ou
r
protoc
ol
propose
d
OL
SR
Km
eans
-
S
DE
al
lows
to
re
du
ce
s
the num
ber
of
pack
et
s
lost c
om
par
ed
to
our pr
e
vious
ve
rsion.
8.2.
End to
e
nd de
lay
To
day,
the
rea
l
-
tim
e
app
li
cation
s
are
in
va
di
ng
the
fiel
d
of
hum
an
act
ivi
ty
.
Fo
r
this
to
m
ini
m
iz
e
the
delive
ry
ti
m
e
of
the
pa
ck
et
s
betwee
n
th
e
transm
itter
and
t
he
recei
ver
beco
m
es
a
nec
essit
y.
The
m
eth
od
of
cl
us
te
rin
g
a
dopted
a
nd
ref
ine
d
le
d
us
to
ha
ve
a
sm
al
l
gain
of
t
ransfe
r
ti
m
e
(EE
D).
As
w
e
see
in
the
Figure
2
,
the
En
d
to
E
nd
Delay
(EE
D
)
of
al
l
ro
utin
g
protoc
ol
incre
ase
pro
portiona
ll
y
with
the
nu
m
ber
of
node
s;
an
d
the OLSR
Kme
ans
-
SDE re
sponde
d
the
b
e
st.
8.3
.
Ov
erhe
ad
The
over
head
of
th
ree
prot
oco
ls
are
ac
co
rd
i
ng
to
t
he
ne
twork
d
e
ns
it
y
in
F
i
gure
3
,
The
OLS
R
Km
eans
-
SDE
gen
e
rates
m
or
e
ov
e
rh
ea
d
f
ol
lowed
by
OL
SR
Km
eans
a
nd
OLS
R
Me
d
+.
It
is
note
d
that
the
div
isi
on
of
node
s
int
o
groups
al
low
a
s
ign
ific
a
nt
re
du
ct
ion
i
n
the
c
ontr
ols
m
essages
broa
dcaste
d
whe
n
the
netw
ork
be
com
es
den
ser
;
ho
we
ve
r
our
pr
ot
oc
ol
propose
d
OLS
R
K
m
eans
-
S
DE
ha
ve
m
or
e
infor
m
at
ion
to
exc
ha
ng
e
(
m
ob
il
i
ty
,
den
s
it
y,
energy).
Fo
r
this
we
no
ti
ce
th
e
sm
al
l
ov
e
rh
ea
d
com
par
ed
with
ot
he
r
protoc
ol
propo
sed.
F
igure
1
.
Pac
ke
t
delivery
rati
o
Figure
2
.
En
d
t
o
e
nd d
el
ay
F
igure
3
.
Co
ntr
ol traffi
c
ov
e
r
he
ad
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
& C
om
p
Eng
IS
S
N: 20
88
-
8708
En
hance
nig
O
LS
R
routi
ng
protoc
ol usi
ng K
-
me
ans clu
ste
ring i
n MAN
ETs
(
Y.
H
amz
aoui
)
3723
9.
CONCL
US
I
O
N
The
cl
us
te
rin
g
is
one
of
t
he
m
os
t
i
m
po
rtant
te
chn
iq
ues
use
d
to
orga
nize
the
netw
ork
i
nto
dif
fer
e
nt
gro
up
s
,
this
ca
n
reduce
the
c
om
plexity
in
m
anag
em
ent
of
nodes
an
d
t
he
refor
e
sim
plif
y
the
processes
r
ou
ti
ng
inf
or
m
at
ion
a
nd
inc
rease
t
he
Q
os
i
n
M
A
NETs
,
we
ha
ve
us
e
d
th
e
s
tro
ng
m
et
ho
d
(K
-
m
eans)
to
gro
up
the
node
s
into
sever
al
cl
us
te
rs.
We
ha
ve
c
onfirm
ed
that
k
-
m
eans
i
m
prov
e
s
the
cl
assi
ficat
ion
of
nodes
by
dem
on
strat
in
g
theo
reti
cal
ly
th
at
the
m
et
ho
d
reduces
the
i
nt
ra
-
cl
ust
er
ine
rtia
(ther
e
fore
it
increases
t
he
inter
-
cl
us
te
r
in
e
rtia
)
betwee
n
tw
o
su
ccessi
ve
it
er
at
ion
s.
In
t
his
pap
e
r
we
ha
ve
con
ti
nue
d
to
ref
ine
our
pro
po
s
ed
protoc
ol
by
add
i
ng
m
or
e
sta
bili
ty
par
a
m
e
te
rs.
The
r
outi
ng
protoc
ol
(
OLS
R
Km
eans
-
S
DE)
br
in
gs
so
m
e
i
m
pr
ovem
ent
t
o
the
le
vel
of
EED
a
nd
P
DR
,
ha
ving
a
pro
du
ct
that
m
eets
al
l
the
need
s
in
te
rm
s
of
qual
it
y
of
serv
ic
e
is
not
obvious
.
W
e
win
one
ty
pe
of
pe
rfor
m
ance
an
d
we
l
os
e
the
ot
her.
W
e
sta
rted
our
st
ud
y
by
i
m
ple
m
enting
bo
t
h
cl
us
te
ri
ng
te
chn
iq
ues
(
k
-
m
eans,
k
-
m
edo
id
),
an
d
we
add
e
d
m
et
rics
i
n
the
al
gorith
m
s
to
hav
e
a
r
obus
t
protoc
ol.
T
hat’
s
wh
at
we
achi
eved
duri
ng
ou
r
resea
rch.
I
n
f
uture
w
ork,
we
will
hav
e
a
plan
to
m
ini
m
i
ze the over
hea
d
in
O
L
SR Km
eans
-
S
DE.
REFERE
NCE
S
[1]
D.
T.
La
ros
e,
“
Discove
ring
Know
le
dge
in
Dat
a:
An
Introduc
t
i
on
to
Data
Min
ing
,”
John
W
il
e
y
&
Sons,
Inc
.
,
Hoboken,
New J
erse
y
.
2005
[2]
E.
Forg
y
.
W
.
,
“
Cluste
r
Anal
y
si
s
of
Multi
var
iat
e
Data
:
Efi
c
ienc
y
v
ersus
Inte
r
pre
ta
b
il
i
t
y
Mod
el
s
,
”
Bi
ometri
cs
,
vol.
61
,
no
.
3
,
pp
.
768
-
769
,
1965
.
[3]
J.
P.
B
enz
e
cri,
“
L'a
na
l
y
s
e
d
es
do
nnée
s :
l
a
t
axi
no
m
ie
,”
Tom
e
1.
Dunod
,
Paris
,
197
3.
[4]
G.
H.
Bal
l,
D.
J.
Hall
,
ISO
DAT
A,
“
A
n
Ite
rat
i
ve
Method
of
Multi
var
i
ate
Anal
y
s
is
and
Patt
e
rn
Rec
ogniti
on
,
”
Be
hav
ior Sc
i
ence
,
vo
l.
153,
1967
.
[5]
P.
Jac
que
t,
P.
Muhlet
aler,
A.
Qa
y
y
um
,
A.
Laoutit
i
,
T.
Cl
ause
n
and
L.
Vienn
ot,
“
Optimize
d
Li
nk
Sta
te
Rou
t
ing
Protocol
,
”
in
IE
EE
INMIC
,
(
Pakistan)
,
Dec
.
200
1.
[6]
S.
Dornbus
h,
A.
Jos
hi,
“
Stre
et
Sm
art
Tr
aff
i
c:
Discove
ring
a
n
d
Diss
eminat
i
ng
Autom
obil
e
Congesti
on
Us
ing
VA
NET’
s
,”
V
eh
ic
ular Te
chnol
o
gy
Conf
ere
nc
eVT
C2007
Spring.
IEE
E
65th
,
pp
1
1
-
15,
2007
.
[7]
E.
Yoneki
and
J
.
Bac
on
,
“
Distribut
ed
Multi
ca
st
Grouping
for
Pu
bli
sh/S
ubscribe
over
Mobile
Ad
Hoc
Networks
,”
Wirel
ess Comm
unic
at
ions
and
Ne
tworki
ng
Conf
er
enc
e
,
2005
IE
EE
,
vol
.
4
,
pp
.
2293
-
2299,
2005
.
[8]
R.
Khanna
and
H.
Li
u,
“
S
y
ste
m
Approac
h
to
Intrusion
Detect
ion
Us
ing
Hidden
Markov
Model
,”
IWCMC
'06
Proce
ed
ings
of
the
2006
international
con
f
ere
nce
on
Wir
el
ess
communic
ati
ons
and
mob
il
e
comput
ing
,
pp
.
349
-
354
,
20
06.
[9]
O.F
.
Ham
ad,
M.Y.
Kang,
J.
Jeo
n
and
J.
Nam
,
“
Neura
l
Network
's
k
-
m
ea
ns
Distanc
e
-
Based
Node
s
-
Cluste
ring
for
Enha
nc
ed
RDM
AR
Protocol
in
a
MA
NET
,”
Signal
Proc
essing
and
Information
Technol
ogy,
IE
EE
2
008
,
pp
192
-
197,
200
8.
[10]
R
.
Badonnel,
R
.
Stat
e
,
and
O
Fe
stor,
“
Self
-
conf
i
gura
ble
f
aul
t
m
onit
oring
in
ad
-
h
oc
net
works
,”
A
d
Hoc
Net
works
,
vol.
6
,
no
.
3
,
pp
458
-
473,
2008
.
[11]
A.
Choukri
,
A.
Habba
ni
,
M.
ElK
outbi
,
“
A
hie
r
arc
hi
ca
l
v
ersion
of
OLSR
for
MA
NET
,”
Worl
d
Appl
ie
d
S
cien
ce
s
Journal
,
vo
l.
21
,
no.
12,
pp.
1739
,
2013
.
[12]
A.
Choukri
,
A.
Habba
ni,
M.
El
.
Koutbi
,
“
Eff
i
cie
nt
heur
ist
ic
base
d
on
cl
ust
eri
ng
appr
oac
h
for
O
LSR
,”
Journal
of
Computer
Net
wo
rks and
Comm
unic
ati
ons
,
2013
.
[13]
A.
Sam
ee
r,
Z
.
Za
hri
la
dha
,
and
L.
Her
wans
y
a
h,
“
A
new
ene
rg
y
consum
pti
o
n
te
chni
qu
e
for
m
obil
e
Ad
-
Hoc
net
works
,”
Int
er
nati
onal
Journal
of
Elec
tri
cal
a
nd
Computer
Engi
nee
ring
(
IJE
C
E)
,
vol.
9
,
no
.
5
,
pp
.
4147
-
4153
,
2019.
[14]
A.
Choukri
,
A.
Habba
ni, M.
ElK
outbi
,
“
An E
n
e
rg
y
Eff
i
cient
C
lu
steri
ng
Algor
it
h
m
for
MA
NETs
,”
ICMCS
,
2014
.
[15]
J.
Mac
Que
en,
“
Som
e
m
et
hods
for
class
ifi
c
at
io
n
and
anal
y
s
is
of
m
ult
iva
r
ia
t
e
observa
ti
ons,
”
In
Proceedi
ngs
of
the
Fi
f
th
B
erk
e
ley
Symposium on
Mathe
mati
cal
st
ati
stic
s an
d
prob
abil
ity
,
pp.
281
-
297,
1967
.
[16]
R.
W
hit
ake
r
,
S.
Hurle
y
,
“
Evol
u
ti
on
of
pla
nning
for
wire
le
s
s
comm
unic
at
ion
s
y
stems
,”
Proce
ed
ings
of
the
36th
Annual
Hawai
i
I
nte
rnational
Co
nfe
renc
e
(
HICSS’03)
,
pp
.
296
,
20
03.
[17]
Angione,
G
.
,
B
el
l
avi
sta
,
P.
,
Co
rra
di,
A.
and
Magistre
tti,
E
.
,
“
A
k
-
hop
Cluste
ring
Proto
col
f
or
Dense
Mobi
l
e
Ad
-
Hoc
Networks
,”
26th
IEEE
Int
ernati
ona
l
Confe
renc
e
o
n
Distribute
d
Computing
Syst
ems
Workshops
(
ICDCSW’06
)
,
pp.
10
,
2006
.
[18]
Danie
l
T.
La
ros
e,
“
Discove
ring
Know
le
dge
in
Data
:
An
Intr
od
uct
ion
to
Data
Mining
,”
John
Wil
e
y
&
Sons
,
I
nc.
,
Hoboken,
New J
erse
y
,
2005.
[19]
R.
Anant
and
M
.
Mana
s,
“
Mobili
t
y
ad
apt
iv
e
d
en
sit
y
connect
ed
c
l
usteri
ng
appr
oach
in
veh
ic
ul
ar
a
d
hoc
ne
tworks
,”
Inte
rnational
Jo
urnal
of
Comm
u
nic
ati
on
Ne
tworks
and
Informati
on
Sec
uri
ty
,”
.
v
ol
.
9
,
p
p
.
222
-
229
,
2017
.
[20]
Inn,
E.
and
W
inston,
S.
,
“
Mobili
t
y
-
B
ase
d
D
-
Hop
Cluste
ring
A
lgori
thm
for
Mobile
Ad
Hoc
Networks
,”
IEEE
Wirel
ess Comm
unic
ati
ons
and
Ne
tworki
ng
Conf
er
enc
e
,
vo
l. 4,
pp.
2359
-
2364,
200
4.
[21]
M.
Dy
abi,
M.
Saadoune
,
A.
Haja
m
i,
H.
All
al
i
,
“
OLSR
Cl
usteri
ng
Algorithm
Based
on
Nodes
Mobili
t
y
,”
Inte
rnational
R
e
vi
ew
on
Comput
er
and
Sof
taware,
(
IRE
COS)
,
vo
l.
10
,
no
.
1
,
pp
.
3
6
-
43,
2015
.
[22]
Sanli
n
Xu
,
Kim
L.
Bl
ac
km
ore
,
H
al
e
y
M.
Jones
,
“
An
Anal
y
sis
Fra
m
ework
for
Mo
bil
ity
Me
t
ric
s
in
Mobile
Ad
Hoc
Networks,”
Published
in
EUR
ASI
P
J
.
W
irel
ess Co
mm
.
and
Ne
tworki
ng,
2007.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
& C
om
p
Eng
,
V
ol.
10
, N
o
.
4
,
A
ugust
2020
:
3
7
1
5
-
3
7
2
4
3724
[23]
Y.
Ham
za
oui
,
M.
Am
nai
,
A.
Choukri,
Y.
Fak
hri
,
“
Novel
C
lu
ste
ring
Method
Based
on
K
-
Me
doids
and
Mobil
ity
Me
tri
c
,”
Inte
rn
ati
onal
Journal
of
Int
erac
t
iv
e
Mult
imedi
a
and
Arti
f
ic
ia
l
Int
ell
ige
nc
e
(
IJI
MAI
)
,
vol
.
5
,
no.
1
,
pp.
29
-
33
,
2017
.
[24]
Y.
Ham
za
oui
,
M
.
Am
nai
,
A
.
Cho
ukri,
Y.
Fakhri
,
“
Enha
nci
ng
OL
SR
Routi
ng
Prot
ocol
Us
ing
K
-
m
edoi
ds
Cluste
r
in
g
in
MA
NETs
,”
In
te
rnational
Jour
nal
of
Applied
E
ngine
ering
Re
s
e
arch
, v
ol
.
12
,
no
.
2
,
pp
.
200
-
206
,
2017.
[25]
A.
Al
-
Maa
shri
and
M.
Ould
-
K
haoua
,
“
Perform
anc
e
An
aly
sis
o
f
MA
NET
Routi
ng
Protocol
s
in
the
Pr
ese
n
ce
o
f
Self
-
Sim
il
ar
Tr
aff
ic,”
Proc
ee
di
ngs
2006
31st
IEE
E
Conf
ere
n
ce
on
Local
Computer
Net
works
,
Ta
m
pa,
FL,
pp.
801
-
807
,
20
06.
BIOGR
AP
H
Y
O
F
AU
TH
ORS
Ali
Ch
ou
kri
is
an
As
sistant
pro
fessor
at
Nati
onal
School
of
A
p
pli
ed
Scie
n
ce
s.
He
was
awa
rde
d
a
Master
of
Com
pute
r
Sci
enc
e
an
d
Te
leco
m
from
the
Univer
sit
y
of
IbnTof
ail,
Kéni
t
ra
-
Morocc
o
in
2008.
He
got
E
NS
ET
(highe
r
N
orm
al
School
of
te
chn
ic
a
l
t
ea
ch
in
g)
degr
e
e
in
199
2.
He
has
a
PhD
from
the
School
of
Co
m
pute
r
S
ci
en
ce
and
S
y
st
ems
Analy
sis
(
ENSIA
S).
He
is
working
withi
n
the
MIS
te
am
o
f
the
La
bora
tor
y
SIM
E,
for
stud
y
ing
ad
hoc
m
obil
e
intelligent
c
om
m
unic
at
ion
s
y
stems
,
and
wire
l
ess
sensor
net
works
.
His
r
e
sea
rch
int
e
rests
are
in
the
ar
e
as
of
ubiqu
it
ou
s
computing,
Int
e
rne
t
of
Th
ings,
Delay
/Disrupt
io
n
Tol
er
ant
Ne
t
works
,
W
ire
le
ss
Networks,
QoS
rou
ti
ng,
Math
e
m
at
ic
a
l
m
odel
in
g
and
per
form
anc
e
an
aly
sis
of
net
works
,
Cont
rol
and
decisio
n
the
or
y
,
Gam
e
th
eor
y
,
trust
and
r
eput
a
ti
on
m
anag
ement,
Distr
ibuted
a
lgori
thms
,
Meta
heur
ti
sti
cs
and
opt
imiza
t
ion
,
Gen
et
i
c al
gori
t
hm
s
You
nes
Ham
z
a
oui
obta
ine
d
his DE
UG
degr
ee
on
phsy
si
cs
che
m
ic
a
l
in
2000
from
Abd
m
al
iksaa
di
Univer
sit
y
,
Ta
ng
er
cit
y
,
m
oroc
co.
The
n,
he
r
ec
e
iv
ed
his
degr
ee
on
IEE
A
(Com
puters
,
El
e
ct
ron
ic
s,
El
e
ct
ri
ca
l
and
Autom
at
ion)
in
20
03.
After
in
200
7,
he
obtained
h
is
DESA
degr
ee
on
informat
ion
s
y
stem
and
tele
comm
unic
at
ion.
The
aut
h
ers
is
a
profe
ss
or
of
c
om
pute
r
scie
n
ce
in
h
igh
schoo
l
.
He
is
a
PhD
stu
dent
of
m
ana
ge
m
ent
of
Qos
in
routi
ng
pro
toc
ol
,
a
lso,
h
e
do
ing
his
rese
arc
h
in
the
La
bor
at
or
y
i
n
Com
pute
r
Sci
e
nce
and Te
l
ec
o
m
unic
at
ions (
LaRIT).
Mohame
d
Amn
ai
rec
e
ive
d
h
is
bac
he
lor’s
deg
ree
in
2000,
in
IEE
A
(Com
pute
rs,
Elec
troni
cs
,
El
e
ct
ri
ca
l and
Autom
at
ion)
from
Mola
y
Ism
ai
l
’s
Univer
sit
y
,
Err
a
chi
di
a
ci
t
y
.
Th
en
,
he
obtained
his
m
aste
r’s de
gre
e in
2007,
from
Ibn
Tofa
il
Univ
ersity
,
Keni
tra
c
ity
.
In
2011,
h
e
received
his PhD on
Te
l
ec
om
m
unic
ation
and
computer
sci
enc
e
,
from
Ibn
Tof
ai
l
Univ
e
rsit
y
in
Kenit
r
a
ci
t
y
,
Moroc
co
.
Since
Marc
h
2014
he
is
a
Prof
essor,
he
has
bee
n
an
As
sistant
at
Nati
onal
Sc
hool
of
Applie
d
Scie
nc
es
Khouribga,
Settat
Uni
ver
sit
y
,
Moroc
c
o.
Th
e
au
thor
i
s
al
so
an
associa
t
e
m
ember
of
Resea
rch
La
bor
at
or
y
in
Com
pute
r
Scie
n
ce
and
Te
l
ec
om
m
unic
ations
(LARIT
),
Te
am
Networks
and
Te
l
ec
om
m
unic
a
ti
ons
Facul
t
y
of
Scie
n
ce
,
K
eni
tr
a,
Morocc
o
.
Also,
the
aut
ho
r
is
an
associa
te
m
ember
of
la
bora
tor
y
IPO
SI
N
at
ion
al
School
of
Applie
d
Sci
e
nce
s
Khouribga,
Sulta
n
Mou
l
a
y
Slim
ane
Univer
s
ity
,
Morocc
o
.
You
ss
ef
Fa
kh
ri
rec
ei
v
ed
his
Bac
he
lor’s
Degre
e
(B.
S)
in
El
ec
t
ronic
Ph
y
sics
in
2001
and
his
Master
’s
Degre
e
(DESA
)
in
C
om
pute
r
and
T
e
le
comm
unic
atio
n
from
the
Fac
ulty
of
S
cienc
es,
Univer
sit
y
Moh
amm
ed
V,
Rabat,
Moroc
co,
in
2
003
where
he
d
eve
lop
ed
his
Ma
ster’
s
Projec
t
at
the
ICI
Com
pan
y
,
Morocc
o.
He
rec
ei
v
ed
a
PhD
in
20
07
f
rom
the
Univer
sit
y
Moham
m
ed
V
-
Agdal,
Rabat,
Morocc
o
in
co
ll
abor
at
ion
wit
h
the
Pol
y
techni
c
Univer
sit
y
o
f
Ca
ta
lonia
(UP
C)
,
Spain.
He
joi
n
e
d
the
Facul
t
y
o
f
Scie
nce
s
of
Kénit
ra
,
Depa
rt
m
ent
of
Com
pu
te
r
Scie
n
ce
and
Mathe
m
at
i
cs,
Ib
nTofa
i
l
Univer
s
ity
,
Morocc
o
,
a
s
an
As
s
oci
ate
Profess
or
in
Ma
rch
2009,
he
is
the
La
bor
at
or
y
h
ea
d
at
L
aRIT
(L
abor
at
or
y
for
Re
sea
rch
in
Com
puti
ng
and
Telecom
m
unic
at
ions)
in
th
e
Fa
cul
t
y
of
Kénit
r
a, a
nd
Me
m
ber
of
Pole
of C
om
pet
enc
es
ST
IC
Morocc
o
.
Evaluation Warning : The document was created with Spire.PDF for Python.