TELKOM
NIKA
, Vol. 13, No. 4, Dece
mb
er 201
5, pp. 1251
~1
262
ISSN: 1693-6
930,
accredited
A
by DIKTI, De
cree No: 58/DIK
T
I/Kep/2013
DOI
:
10.12928/TELKOMNIKA.v13i4.1847
1251
Re
cei
v
ed Ap
ril 11, 2015; Revi
sed Septe
m
ber
20, 201
5; Acce
pted
Octob
e
r 2, 20
15
Energy-efficient Routing Model Based on Vector Field
Theory for Large-scale Wireless Senso
r
Networks
Ming Li
1,2,a*
, Huany
a
n
Qia
n
1,b
, Min
Xu
3,
c
1
School of Co
mputer Scie
nc
e and T
e
chno
l
o
g
y
, Nan
jin
g
U
n
iversit
y
of Sci
ence a
nd T
e
chnol
og
y, Na
nji
n
g,
210
09
4, Chin
a
2
School of Busi
ness, Hoh
a
i U
n
iversit
y
, Na
nji
ng, 211
10
0, Chin
a
3
Departme
n
t of Accountin
g & Informatio
n
S
y
s
t
ems, Calif
orn
i
a State Univer
sit
y
No
rthr
id
ge,
Northrid
ge, C
A
913
30-8
3
7
2
, USA
e-mail: lm@
h
h
u
.edu.cn
a
, h
y
q
i
an@
njust.ed
u
.
c
n
b
, xu-mi
n@
1
63.com
c
A
b
st
r
a
ct
Routi
ng
desi
g
n is a
key is
sue for l
a
rg
e-
scale w
i
re
less
sensor
netw
o
rks (W
SNs). Energ
y
consu
m
ption
associ
ated w
i
th all
o
cate
d re
sources
sh
ou
l
d
be co
nsi
der
ed. T
h
is pa
p
e
r prop
oses t
h
e
integr
ation
of an ener
gy-effi
cient mo
de
l,
w
h
ich is
bas
e
d
o
n
vector
fi
eld t
heory,
in
larg
e-scal
e
W
S
Ns.
Source
no
des
in W
S
Ns
hav
e
the c
haracter
i
stics of so
urce
poi
nts i
n
a
ve
ctor fiel
d, w
her
eas si
nk
nod
e
s
coul
d b
e
c
har
a
c
teri
z
e
d
as
gat
heri
ng
po
ints.
Our sche
m
e
d
e
monstrates
th
at w
e
ca
n s
o
lv
e a
set
of p
a
rti
a
l
differenti
a
l
eq
u
a
tions
in
e
l
ectr
ostatic
the
o
ry t
o
d
e
ter
m
in
e th
e ro
utes th
at r
e
sult
in
en
ergy
efficie
n
cy. T
h
us,
the routin
g pro
b
le
m in W
S
N for ener
gy effici
ency bec
o
m
es
a typical PDE
solutio
n
. Our simulati
on res
u
lts
show
sign
ifica
n
t improve
m
e
n
t in en
ergy
consu
m
pt
ion.
Co
mp
ared w
i
t
h
the traditi
on
al shortest p
a
t
h
appr
oach, the
prop
osed
mod
e
l show
s consi
dera
b
le i
m
prov
ement in the l
i
fetime of the ne
tw
ork.
Ke
y
w
ords
: en
ergy efficie
n
t, routin
g mod
e
l, vector field the
o
r
y, w
i
reless sensor netw
o
rk
Copy
right
©
2015 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
Wirel
e
ss sen
s
or networks
(WSNs)
[1]–[3]
con
s
ist
of several h
undred
s to several
thousand
s of
sen
s
ors p
r
o
pagate
d
in a geog
rap
h
ic
al area. Th
e se
nso
r
s
can
co
mmuni
cate with
each oth
e
r th
roug
h
wirel
e
ss lin
ks an
d of
ten u
s
e
radi
o
frequ
en
cy ch
annel
s to
co
mmuni
cate. T
h
e
s
e
ns
or
s me
as
u
r
e
s
p
ec
ific me
tr
ic
s lik
e
,
s
u
c
h
a
s
te
mperature,
p
r
essu
re
, m
o
vements,
or o
t
her
physi
cal values, in a peri
odic
or non-periodic m
a
nner. Howeve
r, the sensors utilize battery
power, a
nd
e
fficient en
erg
y
use
of sen
s
ors i
s
cr
uci
a
l to in
crea
se the lifetime
of the n
e
twork.
Thus, d
e
termi
n
ing optimal t
r
an
smi
ssi
on
paths from
ea
ch sen
s
or to t
he de
stinatio
n is impo
rtant
.
The ro
uting
probl
em in sensor net
wo
rks h
a
s b
een
studied by
many re
sea
r
che
r
s. S.
Toumpi
s an
d L. Tassiula
s first intro
d
u
c
e
d
electros
tatic
field theory into
the WSN [4]. Their works
focu
s on th
e
sen
s
o
r
d
epl
oyment probl
em in
WSNs. The re
se
arche
r
s ab
stra
cted the
nod
e
optimal di
stri
bution p
r
obl
e
m
into a cha
r
ge di
stri
but
ion problem i
n
the ele
c
tro
s
tatic field, t
hus
providin
g an
effective con
c
ept of bu
ildin
g up the routi
ng model fo
r WSNs.
Mehdi K
a
lant
ari a
nd
Mark Shayman
u
s
ed
ele
c
tro
s
t
a
tic field
the
o
ry to
study
ad h
o
c
networks and verified the f
easi
b
ility
of analyzing
work rout
ing
with t
he el
ect
r
ostatic field.
On the
basi
s
of a p
r
eviou
s
wo
rk [7], they also em
ploy
ed
vector an
al
ysis in the li
terature [8] and
descri
bed th
e
routing
probl
em in
WS
Ns with the diff
erential
equ
a
t
ion in the el
ectro
s
tati
c fie
l
d
.
They de
rive t
he pa
rtial diff
erential
equ
at
ion a
c
cordin
g
to the p
r
op
erties of the
ele
c
tro
s
tatic fiel
d,
applie
d the e
quation to
ro
ute WSNs, a
nd empl
oy
ed
the mathem
atical phy
sics method to solve
the energy ro
uting pro
b
lem
of WSNs.
Yeling Zhan
g
et al [9] visualized the trans
mi
ssion o
f
messa
ge p
a
ckets in WSNs into
abstract i
n
formation flo
w
and d
e
si
gne
d two
algo
rithms, n
a
mely, maximum in
formation
ro
u
t
ing
and
con
d
ition
a
l maximum
informatio
n routing; t
hey
attributed the
routing
solution proble
m
of
WSNs to th
e extremal p
r
oble
m
of transfe
rri
ng
inf
o
rmatio
n flows in th
e n
e
twork life
cycle.
Ho
wever,
the
maximum
tra
n
smi
ssi
on
of
the info
rm
atio
n flow comp
romise
s the
n
e
twork life
cycle
and i
s
therefore
a
que
stio
nable
meth
od
. The
pe
rspe
ctive
from
wh
ich we
st
udy the
net
wo
rk with
the informatio
n flow indi
cat
e
s th
at we
ca
n al
so
study t
he routing
of
WSNs b
a
sed
on th
e ve
cto
r
field.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Energ
y
-efficie
n
t Routing M
odel Based o
n
Vector Fi
eld Theo
ry fo
r Larg
e
-scale
…
(Mi
ng Li)
1252
A previou
s
work in the lit
eratu
r
e [10–
1
2
] adopted t
he theory an
alysis meth
o
d
of th
e
Euler e
quatio
n, whi
c
h is
b
a
se
d on the
flow field
to
simulate th
e
mobility of sensor n
ode
s
in
WSNs to e
n
s
ure b
a
lan
c
e
d
dist
ributio
n
of nod
es
, ef
fective utiliza
t
ion of en
erg
y
, and coverage
rate of high
netwo
rks. Alireza Saleh
a
n
et al [
13] modified the framework
of the simul
a
tion
of
WSNs an
d simulated the
commu
nication pro
c
e
s
s
of the ad ho
c network in
a flow mod
e
l.
Acco
rdi
ng to
simulatio
n
re
sults
of the al
gorit
hm
s, such as DS
DV, DSR, AO
DV, and T
O
RA, t
h
e
resea
r
chers verified the correctn
ess of th
e flow model.
By integratin
g the
de
scrip
t
ions i
n
the
p
r
ev
iou
s
wo
rks a
bove,
we
determi
ne th
at all th
e
said
problem
s can
be transfo
rme
d
in
to definit
e
solution p
r
obl
ems
of typical mathem
atical
physi
cs ba
se
d on
static
ch
arge
move
m
ent in
the
el
ec
tros
tatic field or flui
d part
icle
movement in
the flow field
and the el
ectrostatic field
di
fferenc
e eq
ua
tions o
r
Euler equation
s
. In
this pap
er, we
analyze the
feature
s
of th
e field i
n
WSNs by u
s
in
g
vector field th
eory to
dete
r
mine the
effe
ctive
energy route
for WSNs.
2. WSN Mod
e
l Based o
n
Vector Field
Theor
y
2.1. Introduc
tion of Vec
t
o
r
Field Theor
y
Model
Con
s
id
erin
g
that a WSN con
s
i
s
ts of
N nod
es,
we a
s
sume
that each n
ode
can
comm
uni
cate
throug
h wi
re
less chann
els and the n
o
d
e
s a
r
e pla
c
e
d
in a pla
nar
regio
n
called
A.
Whe
n
events occur
som
e
whe
r
e in the
netwo
rk, the
clo
s
e
s
t node
will trigg
e
r a
messag
e. Each
node attem
p
ts to obtain th
e messa
ge in
the sen
s
o
r
n
e
twork a
nd re
spo
n
d
s
to the
messag
e of the
neigh
bori
ng
node. Fin
a
lly, all the message
s are tran
smitted to th
e sin
k
no
de.
Con
s
id
erin
g the
compl
e
tene
ss of a model,
we ma
ke the
followin
g
assumption
s:
A1: Every node ca
rri
es lim
ited energy, and
the re
sidu
e energy time is noted;
A2: In sen
s
or net
wo
rks,
every geo
g
r
aphi
c
regio
n
that gene
ra
tes trigg
e
r
e
v
ents is
accomp
anie
d
by a given regio
nal lo
ad den
sity
distrib
u
tion l
a
w. The lo
a
d
den
sity in
rep
r
e
s
ent
s th
e loa
d
g
ene
rated in
by a
unit time. If we a
s
sume
tha
t
the coo
r
dina
te in
is
(x,
y), then we reco
rd that th
e load d
e
n
s
ity in
is
(,
)
x
y
. The value of
()
ws
reflec
ts
the load
gene
rated
in
place a
(
a
A
)
by unit time. That is to
say, the rate
gen
erated
by trig
ger
events
i
s
as
follows
:
()
(
,
)
a
ws
x
y
d
(
1
)
In the formul
a above,
()
ws
is the integ
r
al in
flat area a,
d
is the differentia
l unit in the
plana
r re
gion
inclu
d
ing
;
A3: The positi
on of the nod
e is fixed and
kno
w
abl
e;
A4: Whe
n
th
e nod
e is
de
ployed, ea
ch
node i
n
the n
e
twork
can
a
rrive at the
si
nk n
ode
throug
h a mul
t
i-hop tra
n
smi
ssi
on sequ
en
ce;
A5: To find the ro
ute, we
provide
a di
rectio
n to ea
ch p
o
int in sensor n
e
two
r
ks,
whi
c
h
points the n
e
x
t nearest no
de, and a
s
su
me that it is a continu
o
u
s
functio
n
abo
ut
.
On the ba
sis
of assumptio
n
A2, the rate of
each
se
nso
r
that gen
erate
s
a me
ssag
e ca
n
be derive
d
according to its resp
on
se tri
gger eve
n
ts
within the re
g
i
on. Wi is used as the loa
d
of
sen
s
o
r
n
ode
i
,
whi
c
h
sh
ows the
rate of
node
i p
r
od
u
c
ing
me
ssag
es. If
i
t
rep
r
e
s
ents th
e
regi
on
whe
r
e the se
nso
r
nod
e i is resp
on
sible f
o
r re
sp
ondi
n
g
to trigger e
v
ents, then Wi ca
n be de
rived
by the a
r
ea i
n
tegral
in
i
t
b
a
se
d on
Fo
rmula (1). T
h
e value
of W
i
is the
weigh
t
of node
i. We
assume
that
each trig
ge
r
event is
re
cei
v
ed by o
n
ly o
ne
sen
s
o
r
. If
several
sen
s
or n
ode
s
pro
duce
messag
es b
e
c
au
se of a trigger eve
n
t in the r
egion, then we
can
assume that
only one of the
node
s send
s
the messag
e to repo
rt the event.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No
. 4, Decem
b
e
r
2015 : 125
1 – 1262
1253
At the same
time, we assi
gn a weight to the sin
k
no
de usi
ng
0
w
to expre
ss th
e sin
k
node. We use Form
ula (2
) to define the sin
k
nod
e
0i
1
w
N
i
w
(
2
)
Formul
a (2
) should
sho
w
t
hat the rate o
f
the
sink
nod
e that re
ceive
s
the me
ssag
e is the
total rate of a
ll the sou
r
ce
node
s that send
me
ssag
e
s
to the si
nk
node a
nd the
negative val
ue
indicates the
role of the sin
k
nod
e.
We d
e
fine th
e path a
s
a d
i
rectin
g curve
deriv
ed from
the initial no
de to the si
n
k
nod
e.
Thus, N
corresp
ondi
ng pa
ths exist in the sen
s
o
r
ne
twork co
ntain
i
ng N so
urce
nodes. We use
i
p
to express t
he path of
se
nso
r
no
de i. The loa
d
of e
a
ch p
a
th is t
he rate of the
origin
al nod
e
prod
uci
ng me
ssage
s in the
path. Thus, the wei
ght of path
i
p
is
i
w
.
Acco
rdi
ng to
the ab
stra
ct
path
set from
ea
ch
se
nsor
nod
e
to
th
e sin
k
n
od, we define a
vector field
o
n
regi
on A a
s
a lo
ad vect
or field, de
no
ted by
D
, w
h
ich
r
e
pr
es
en
ts
th
e
me
ss
ag
e
spe
ed
distri
b
u
tion field
in
sen
s
o
r
n
e
two
r
ks a
nd th
e
d
i
rectio
ns of al
l the p
o
ints to the
sin
k
no
de.
Given any po
int
in region
A, we sele
ct the minimal
re
gion unit a
r
e
a
a in
. When
area S o
f
a infinitely tends to be 0, we can d
e
rive the l
oad ve
cto
r
of every poi
nt, as sho
w
n i
n
Form
ula (3
)
0
1
()
l
i
m
i
ii
S
pS
Dw
l
S
(
3
)
In the formul
a above,
i
l
is the unit tang
ent of path
i
p
in regi
on S, whi
c
h di
re
cts the
destin
a
tion n
ode alo
ng th
e path. Figu
re 1 illust
rate
s the ab
ove
definition. On
the basi
s
of
the
descri
p
tion of
Hypothe
si
s 5
,
all the path
s
throu
gh S
wil
l
have the
sa
me unit tan
g
e
n
t
i
l
wh
en
S
tends to be
zero. T
h
u
s
, re
sult
()
D
adde
d b
y
the vecto
r
s in Equ
a
tion
(3) will
have
the sam
e
dire
ction. In other wo
rd
s,
()
D
is the sum of all the path loads
throug
h region S, and it reflects
the a
c
tual
co
mmuni
cation
activity of pl
ace
in
th
e
s
e
ns
or
n
e
t
wor
k
,
w
h
ic
h is th
e
me
ss
ag
e
transfe
r condi
tion in
. On th
e basi
s
of field theory, loa
d
vector
D
can also be calle
d the load
dens
i
ty.
Mathemati
c
al
ly, if the num
ber of
se
n
s
o
r
s i
s
limite
d
, then th
e valu
e of
D
may
be zero
only when
th
e set i
s
e
m
pt
y. We
ca
n di
vide region
S
into
seve
ral
recta
ngul
ar a
r
ea
s
usi
ng th
e
vertical a
nd h
o
rizontal g
r
id
line. Wh
en th
ese
re
ctang
ul
ar ele
m
ent
s a
r
e sufficiently
small, we ca
n
treat
D
as a co
ntinuou
s vari
able for p
r
o
c
e
ssi
ng.
Figure 1. Illustration of the load de
nsity vector field b
a
s
ed o
n
path
s
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Energ
y
-efficie
n
t Routing M
odel Based o
n
Vector Fi
eld Theo
ry fo
r Larg
e
-scale
…
(Mi
ng Li)
1254
We
define th
e integ
r
al
wh
en loa
d
ve
cto
r
field
D
point
s to the di
re
ction of the
curve as
follows
:
Dd
n
(
4
)
It is the load
flux when
D
p
a
sse
s
throug
h the curve
, whi
c
h d
enot
es the
me
ssage
volume pe
r u
n
it time throu
gh the cu
rve
.
If the defined
in Equation (4) i
s
a clo
s
e
d
curve, then
the defined
D
in Equation (3
)
must meet th
e followin
g
eq
uation:
Dd
n
w
(
5
)
In the p
r
evio
us fo
rmula,
is
a
c
l
os
ed
c
u
r
v
e
.
dn
is
a no
rmal differe
ntial vecto
r
of e
a
ch
point o
n
the
clo
s
ed
curve,
and
w
is th
e
load
sum
of nod
es withi
n
the
cl
ose
d
area. In
the
electrostati
c field, Gau
ss’
s
law exhibit
s
a
similar
fo
rm, whi
c
h sho
w
s
that the electric flux throug
h
a cl
osed
cu
rv
e is p
r
op
ortio
nal to
the
am
ount of
charg
e
s
su
rroun
de
d by
a
clo
s
ed
cu
rve. F
o
r th
e
steady incom
p
re
ssi
ble flui
d in the flow field,
the above formula al
so sh
ows the
unique phy
si
cal
signifi
can
c
e.
The flo
w
p
r
o
d
u
ce
d by
a
so
urce
within fl
uid bl
ocks in
a unit tim
e
i
s
equal
to the
fluid
quality from the su
rface of the fluid blocks in the sa
me
time period.
In se
nsor
net
works, if a
m
e
ssag
e i
s
tra
n
sf
erre
d fro
m
the n
ode
out
side
the
clo
s
ed
curve
to a node insi
de it, then we can say that the mess
ag
e enters the clo
s
ed a
r
ea. Oth
e
rwi
s
e, we ca
n
assume th
at the me
ssage
exits t
he clo
s
ed area. Accordin
g to the
definition ab
o
v
e, Equation (5)
indicates that
the g
ene
rate
d me
ssage
a
m
ount i
n
the
unit time
of a
clo
s
e
d
regi
o
n
is the
sum
of
the real
-time messag
e tran
smissio
n
rate
s between int
e
rnal n
ode
s.
We d
e
fine ch
ara
c
teri
stic fu
nction
, which
sho
w
s that the
me
ssage
reache
s the pl
anar
regio
n
, an
d its value
is
ab
out the p
o
siti
on fun
c
tion. If
()
(
,
)
x
y
, then, exce
pt for the
sin
k
node, the val
ue of
()
is that of
()
.
00
()
(
)
(
)
w
(
6
)
On the
ba
sis
of the d
e
finition of
stated
above, Equ
a
tion (5)
ca
n b
e
re
written int
o
the
differential eq
uation a
c
cord
ing to the formula of Gre
e
n
.
()
()
D
(
7
)
In the formula
above,
is de
fined as
ij
x
y
(
8
)
In the form
ul
a, x and y
repre
s
e
n
t the
hori
z
o
n
tal a
nd verti
c
al a
x
es in th
e
Descarte
s
Carte
s
ia
n re
ctangula
r
coordinate sy
ste
m
, respe
c
tively.
i
and
j
deno
te the unit no
rmal vecto
r
s
along axial
co
ordin
a
te x, y.
The value of
vector
D
will change
with the sel
e
cti
on of the path set. However,
whateve
r
the sele
cted p
a
th
is, vector
D
m
u
s
t
s
a
tis
f
y the following equation:
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No
. 4, Decem
b
e
r
2015 : 125
1 – 1262
1255
(
)0
,
n
D
th
e
e
dge
of
the
a
rea
A
D
(
9
)
In the formul
a above, A is the are
a
set in the sen
s
or
netwo
rk.
()
n
D
s
h
ow
s
the
sc
alar
along the b
o
unda
ry of reg
i
on A by
D
. The first equ
ation of Equatio
n (9) expl
ain
s
that all the
message flows produced
in the net
work will finally point to the
sink node. The
second equation
derive
s
from t
he fact that no messag
e flow co
me
s ou
t of the senso
r
netwo
rk are
a
boun
dary.
2.2. Establis
hing the Loa
d Vector Fiel
d
On the basi
s
of the vector field theory model,
whi
c
h
is illustrated
in the former
se
ction,
combi
ned
with electrostati
c field theo
ry and flui
d me
cha
n
ics theo
ry, we determi
ne that the load
vector field
of the se
nsor i
s
simila
r to the
electr
ostati
c
field in theo
ry of electroma
gnetic fiel
d a
nd
the flow field in fluid mech
a
n
ics. Similar to the el
ectri
c
field lines in the ele
c
tro
s
tatic field and th
e
strea
m
line
fie
l
d line
s
in th
e
flow field,
we
defin
e
the
co
nce
p
t of th
e l
oad li
ne i
n
se
nso
r
networks.
Load li
ne is
an in
stantan
eou
s sm
ooth
cu
rve in th
e
load ve
ctor
field wh
ere t
he tran
smi
s
si
on
dire
ction of t
he me
ssage
unit coi
n
cid
e
s
with it
s
tan
gent directio
n. Obviou
sly, the load lin
e is
simila
r to the electri
c
field
line and st
reamline
s
fiel
d line, excep
t
for some speci
a
l point
s that
neither turn nor inters
ec
t.
By establishi
ng the loa
d
vector field
a
nd
definin
g the con
c
ept
of load line,
we
can
determi
ne th
e
rout
e fro
m
th
e source
no
d
e
to the
de
sti
nation
nod
e a
c
cordi
ng to
F
o
rmul
a (5).
T
h
e
load line can
be approximately con
s
ide
r
ed the spa
c
e–tran
s
fe
r cu
rve of the messag
e unit in a
time interval.
By identifyin
g
optimal
sol
u
tion
D
that
sa
tisfies th
e m
odel,
we
ca
n
de
scribe
o
u
r
load vecto
r
field acco
rding
to the gradie
n
t func
tion a
nd determi
ne
the route of the WSN ba
sed
on the load li
ne.
Figure 2. Sample of load l
i
ne routin
g
The ci
rcl
e
s i
n
Figure 2 repre
s
e
n
t the
sen
s
o
r
nod
es in the
WSN regi
on.
Whe
n
load
vector field
D
in the regi
on i
s
dete
r
mine
d, we ca
n de
scribe the lo
ad
lines b
e
twe
e
n
the sou
r
ce
node
s an
d si
nk no
de
s accordin
g to the gradi
ent funct
i
on and ide
n
tify the relay node
s aro
und t
he
load lin
es to
determi
ne th
e ro
ute for th
e se
nsor
net
work. In the fi
gure,
solid
lin
es
rep
r
e
s
ent l
oad
lines,
and
dot
ted line
s
. It is the lo
ad flux
den
otes
the
actual
tran
sm
issi
on lin
es of
the p
a
cket
s i
n
the
net
wo
rk. One of
the
se
lected
line
s
st
arts
from
the
sou
r
ce node,
throug
h node
s
1
N
an
d
2
N
,
and
so
on
to
the
sin
k
no
de in
the
pro
per seque
nce. The
a
c
tual
tran
smi
ssi
on
line
of p
a
ckets
coin
cid
e
s rou
ghly with the load line.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Energ
y
-efficie
n
t Routing M
odel Based o
n
Vector Fi
eld Theo
ry fo
r Larg
e
-scale
…
(Mi
ng Li)
1256
3. WSN Ener
g
y
Efficienc
y
Routing Based
on a Stable Field Model and Solution
The
routin
g
mech
ani
sm i
n
the traditio
nal
n
e
two
r
k usu
a
lly
sel
e
cts
the path with
the
minimum
cost betwe
en th
e sou
r
ce n
o
d
e
an
d the
si
nk
nod
e. In f
a
ct, we
sh
ou
ld con
s
ide
r
t
he
balan
ce
of th
e wh
ole
network loa
d
an
d t
he net
wo
rk
survival time in
sele
cting
the
network routi
ng
becau
se of t
he limited n
o
de en
ergy
co
nsum
ption.
T
he core
prob
lem of the e
nergy effi
cien
cy
routing i
s
sel
e
cting the ro
ute according
to t
he energ
y
consumptio
n and the re
sidu
al ene
rgy
of
the node
s in the path.
In esta
blishi
n
g
the lo
ad v
e
ctor field,
we
define
a
space vecto
r
f
o
r e
a
ch poi
n
t
in the
spa
c
e
and
al
so
set up
an
energy field
for ea
ch p
o
in
t in the sp
ace, whi
c
h i
s
a
re
sidual
ene
rgy
distrib
u
tion fu
nction o
n
poi
nt
.
0
node i
1
(,
)
l
i
m
(
)
i
S
S
wt
e
t
S
(
1
0
)
In the previo
us form
ula,
i
e
()
t
rep
r
e
s
ent
s the resi
dual en
ergy
of the sensor no
de i at
time t. Evidently,
(,
)
wt
will
proba
bly be
ze
ro
when th
e e
n
e
r
gy set
is emp
t
y only in th
e
place of
.
3.1. Energ
y
Rou
t
ing Und
e
r a Stable F
i
eld Model
We a
s
sume t
hat the initial
energy of
the
WSNs is
unif
i
ed allo
cated,
i.e.,
(,
0
)
wc
(c
is a con
s
tant). To determi
n
e
the ene
rgy
ef
f
i
cien
cy
rou
t
e,
we mu
st
max
i
mize
()
D
in each point
of
unde
r th
e co
mmuni
cation conditi
on of the
sensor n
e
two
r
k.
()
D
r
e
pr
es
en
ts
th
e
instanta
neo
u
s
m
e
ssa
ge transmi
ssion
rate in
of th
e
sen
s
o
r
netwo
rk. T
h
u
s
, the
attenuation
of
energy per u
n
it time in
can be app
roxi
mately prop
ortional to
()
D
.
A unified
D
is made a
s
fa
r as
po
ssi
ble
to determi
n
e
the en
ergy
efficien
cy ro
uting
evenly sp
rea
d
s the tra
n
sm
issi
on of me
ssag
es in
the netwo
rk. Hen
c
e,
we can a
v
oid
som
e
no
des
in the network that are no
t fully utilized and overused.
A unified loa
d
distrib
u
tion
can be expressed
a
c
cord
ing to the followin
g
minim
u
m co
st
function.
2
()
av
A
J
DD
D
d
s
(
1
1
)
In the formul
a ab
ove,
av
D
is
a mea
n
valu
e
of vecto
r
fiel
d
D
in the
s
e
t
of A. It c
a
n be
defined simpl
y
as
1
av
A
DD
d
s
A
(
1
2
)
In Equation
(11), the
cost
functio
n
in
a squa
re fo
rm en
su
res e
v
en dispe
r
sal
of the
distrib
u
ted lo
ad, thereby p
r
event
ing ex
ceedin
g
ly high
load in the local a
r
ea of the netwo
rk a
n
d
unde
rutilization of re
sou
r
ces in t
he oth
e
r area. The
form of the co
st function
is simila
r to the
energy definit
ion in th
e el
e
c
tro
s
tatic fiel
d.
The
optimi
z
ation
proble
m
above
can
be
summ
ari
z
ed
as
follows
:
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No
. 4, Decem
b
e
r
2015 : 125
1 – 1262
1257
2
()
()
0
av
A
n
M
inimi
z
e
J
D
D
D
d
s
D
D
the
e
dge
o
f
A
s
u
b
j
e
c
t
to
:
,
(
1
3
)
The followi
ng
lemmas p
r
ov
ide key ide
a
s
in the optimization pro
b
lem
for Equation
(13
)
.
Lemma
1: If
D
re
pre
s
e
n
ts the o
p
timal
solution
of Eq
uation
(1
4), t
hen it
mu
st
satisfy
the followin
g
equatio
n:
D0
(
1
4
)
In the equ
a
t
ion above,
is a 2
D
d
i
rect
io
n vect
or op
erator.
In vector f
i
eld
xy
FF
F
, we define th
is ope
ration
with the form
ula
y
x
F
F
F
k
yx
(
1
5
)
whe
r
e
is a un
it complex vector, which i
s
comp
osed of
i
and
j
, i.e.,
ij
.
On the
ba
si
s
of the a
bove l
e
mma,
we
ca
n writ
e
a
set
of sp
atial diff
eren
ce
eq
uati
ons for
optimal sol
u
tion
D
.
D
,
D0
(
1
6
)
The ab
ove eq
uation
s
are si
milar to the M
a
xwell
eq
uati
ons of el
ectro
s
tatic field the
o
ry. In
spatial
differe
nce
equ
ation
theory, we verified
th
e bo
unda
ry condit
i
ons
of the a
bove eq
uatio
ns,
whi
c
h
can
b
e
given p
a
rti
c
ula
r
ly thro
u
gh
D
. In the model of m
a
thematical
fie
l
d theo
ry, th
e
vector field t
hat sati
sfies
D0
is al
so
calle
d a con
s
erva
tive vector field, whi
c
h
can b
e
expre
s
sed a
s
the gradie
n
t of
the scal
a
r f
i
eld, i.e.,
D
U
(
1
7
)
U i
s
a
pote
n
tial fun
c
tion, a
nd it
can
be
dra
w
n
acco
rd
ing to
dire
ctio
nal d
e
rivative
s, which
we call the scalar fun
c
tion. Equation (16) can be
simpli
fied to
2
U
.
Operator
2
is d
e
fined a
s
22
2
22
x
y
(
1
8
)
The b
oun
da
ry con
d
ition
o
f
D
implie
s tha
t
the tang
ent
di
re
ction
of
the gradie
n
t
of U
along the b
o
u
ndary curve i
s
always 0, which h
a
s the f
o
llowin
g
form
using the formula:
ˆ
()
()
0
,
U
n
the
e
dge
of
A
(
1
9
)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Energ
y
-efficie
n
t Routing M
odel Based o
n
Vector Fi
eld Theo
ry fo
r Larg
e
-scale
…
(Mi
ng Li)
1258
ˆ
()
n
expresse
s th
e unit tangent
vector of poi
nt
along the dire
ction of the boun
dary
curv
e.
3.2. Definite Conditio
n
s u
nder
th
e Mo
del of the Stable Field
In Section
4.1, we
pre
s
e
n
ted key idea
s
about
the
loa
d
co
st fun
c
tio
n
in a
WS
N u
nder a
n
ideal state
an
d the optimu
m
solutio
n
of the diffe
rentia
l equation
s
. In this sectio
n
,
we sum
m
ari
z
e
the load vect
or field pro
p
e
r
ty of the field model
that
was unifie
d
and dist
ribut
ed by the initial
energy of the sen
s
o
r
nod
e.
D0
D
(
2
0
)
The above di
fferential equ
ations fro
m
the micr
o
s
cop
i
c point of view explain th
e active
and irrotation
al ch
ara
c
te
ristics of the lo
a
d
vector.
D0
. Thus, we can find a scal
ar fu
nctio
n
U to establi
s
h
D
U
. We define
U as the loa
d
potential in the vector fiel
d. The cha
n
g
e
rate o
f
the load pote
n
tial in any directio
n of
i
l
is equal to the nume
r
ical value of the load vector in this
dire
ction (
i
l
is the unit vector in the directi
on).
D
li
u
Dl
l
(
2
1
)
Thus, alo
ng d
i
rectio
n
i
l
, the increa
sing a
m
ount of the load potential i
s
as follo
ws:
D
li
du
D
d
l
d
l
(
2
2
)
The load p
o
tential differen
c
e bet
ween t
w
o point
s, na
mely, arbitra
r
y P and Q, along the
dire
ction
i
l
is
as
follows
:
D
Q
QP
i
P
uu
u
d
l
(
2
3
)
Equation (23) reflect
s
the co
st paid by the vector fie
l
d whe
n
the messag
e unit
move
s
from point Q to point P.
On the basi
s
of the differential equation
sets,
the differential e
qua
tions that sati
sfy the
load p
o
tential
of a lo
ad ve
ctor fiel
d in
s
ensor
networks can b
e
d
e
r
ived. We int
egrate
D
U
into the
se
co
nd e
quation
of Form
ula
(2
0), the
r
eby fo
rming
2
D(
)
UU
wh
ere
2
U
is a Poisso
n’
s equ
ation th
at is co
rrect f
o
r any
poi
nt of the vector
field in the m
odel. In
the regi
on
where th
e tra
n
s
missio
n of the me
ss
age
unit doe
s not
occur, the
a
bove form
ula
is
transfo
rme
d
into Lapla
c
e’
s equation.
2
0
U
(
2
4
)
Thus,
the
La
place’s eq
uat
ion of th
e lo
a
d
pote
n
tial i
s
a
spe
c
ial
ca
se
only
whe
n
0
.
Ho
wever, the
Poisson’
s eq
uation i
s
univ
e
rsally
appli
c
able. In solving the Poi
sson’s
equ
ation
or
the Lapl
ace’
s equatio
n, so
me und
eterm
i
ned
con
s
t
ant
s in the
gene
ral solution
will emerge. We
can o
b
tain a
fixed solution
that corresp
ond
s to
a ce
rtain field sou
r
ce di
stributio
n only whe
n
the
equatio
ns are dete
r
min
e
d
.
Usi
ng
boun
dary
con
d
itio
ns
of the l
o
a
d
vecto
r
field
ca
n d
e
termi
n
e
these co
nsta
nts.
The bou
ndary con
d
itions usually
in
dicate th
e co
nne
ction
con
d
itions
of the
load
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No
. 4, Decem
b
e
r
2015 : 125
1 – 1262
1259
intensity and
the load pote
n
tial on both side
s of
the specifi
c
interfa
c
e wh
en me
ssag
e units are
being tra
n
smi
tted.
4. Simulation and Analy
s
is
4.1. Routing
A total of 3
00
sen
s
o
r
n
e
twork
nod
e
s
a
r
e
ra
ndo
mly distri
but
ed in
a pl
an
e area
of
500*5
00. Assume all n
ode
s bel
ong to t
he ro
uti
ng n
o
de except th
e sin
k
n
ode
and the
so
urce
node. In this
ca
se, the loa
d
potential of
any point
in the load vect
or field is the
supe
rpo
s
itio
n of
potential
s
of all so
urce n
o
des i
n
the po
int.
Figure
3 simulate
s the
singl
e-sou
r
ce nod
e and t
h
e
singl
e-sin
k
no
de; the source nod
e
is in t
he up
per l
e
ft corne
r
in the
scene,
whe
r
e
a
s the
sin
k
n
ode
is in the lower right corner.
Figure 3. Single-sou
r
ce si
ngle-sin
k
sce
ne: velocity
vector a
nd pot
ential field line (left), poten
tial
field line and
strea
m
lin. It is the load flu
x
e (right)
Paper also
e
x
plore
s
the si
mulation of
the
multi-sou
r
ce nod
e a
nd
multi-si
nk
no
de sce
n
e
(Fig. 4
)
whe
r
e the
sou
r
ce
node
s a
nd
si
nk n
ode
s
are
marked
in th
e figure. The
figure i
ndi
cat
e
s
the load vect
or field obtain
ed acco
rdin
g to the supe
rp
osition p
r
in
cip
l
e.
Figure 4. Multi-so
urce, mult
i-sin
k
sce
ne: velocity
vecto
r
and pote
n
tia
l
field line (the left picture
)
,
potential line
and streamli
n
e
(the rig
h
t picture
)
-5
00
0
50
0
-
500
-
400
-
300
-
200
-
100
0
100
200
300
400
500
-5
00
0
50
0
-
500
-
400
-
300
-
200
-
100
0
100
200
300
400
500
-500
0
50
0
-5
0
0
0
50
0
si
n
k
1
so
u
r
c
e
1
so
u
r
c
e
2
so
u
r
c
e
3
s
our
c
e
4
si
n
k
2
-
500
0
500
-50
0
0
50
0
si
n
k
1
s
our
c
e
1
s
our
c
e
2
s
our
c
e
3
so
u
r
c
e
4
si
n
k
2
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Energ
y
-efficie
n
t Routing M
odel Based o
n
Vector Fi
eld Theo
ry fo
r Larg
e
-scale
…
(Mi
ng Li)
1260
4.2.
Result Anal
y
s
is and
Pe
rf
ormance C
o
mparison
The pap
er u
s
e
s
the MAT
L
AB platform
tool to carry
out routing
perfo
rman
ce
analysi
s
and
ch
oo
se
s four i
ndi
ce
s, namely, n
e
twork
delay
, t
h
rou
ghp
ut, succe
ssful
del
ivery rate,
a
nd
energy con
s
u
m
ption, to evaluate the
pe
rforman
c
e of a
routing p
r
oto
c
ol.
The pa
per
co
mpares ve
cto
r
field ro
uting
(VF)
, the rou
t
ing proto
c
ol
prop
osed
with many
classi
cal
routing protocol
s,
such as energy-effici
ent
ant-bas
ed routing
prot
ocol
(EEAB
R),
sen
s
o
r
-driven
and
co
st-a
ware a
n
t ro
uting (S
C), f
l
oode
d forwa
r
d a
n
t routin
g (AF
)
, floo
ded
piggybacked
ant routing (FP), basi
c
ant-bas
ed rout
ing (BABR),
ad hoc on-demand di
stance
vector (AO
D
V), and
ad
a
p
tive sp
anni
ng tre
e
(M
CBR-AST).
Ta
ble 1
provid
es th
e
simul
a
tion
para
m
eters.
Table 1. Simulation pa
ram
e
ters
Parameter
Value
Routing It is the load flux prot
ocol
VF
, EEABR, SC,
FF, FP, BAB
R,
AODV, MCBR
-A
ST
Number of
nodes
300
Maximum hop co
unt
30
Data transmission
Constant bit rate
(CBR)
Transmission rate
500 kbps
Simulation time
100 s
Node ene
rg
y
80 J.
The follo
win
g
perfo
rma
n
ce
analysi
s
ch
o
o
se
s F
F
, SC,
and A
O
DV f
o
r the
compa
r
iso
n
. In
the left pi
cture in Fi
gure 5
,
the
net
work delay of VF
is
signifi
cant
ly lowe
r tha
n
that of FF
b
u
t
slightly highe
r than those of SC and AODV. VF co
n
t
ains multiple
paths, inclu
d
ing the sho
r
t
e
st
path, the
pat
h from
the
so
urce n
ode
to
the de
st
inatio
n no
de, a
nd t
he ave
r
ag
e h
op
cou
n
t of a
l
l
the path
s
, which i
s
hi
ghe
r than th
at of the shorte
st
path. Thu
s
,
the network
delay is
slig
h
t
ly
highe
r. The si
mulation re
su
lt indicates th
at t
he network delays of VF and AODV are 0.04
21 a
n
d
0.036 s, resp
ectively. The delay of FV is higher by 25
% or so.
Figure 5. Net
w
ork del
ay (l
eft picture
)
an
d netwo
rk th
rough
put (ri
gh
t picture
)
The right pict
ure
i
n
Fi
gure 5
dete
r
min
e
s that
t
he th
rou
ghput
of FV i
s
hi
ghe
r tha
n
those
of
the three rout
ing proto
c
ol
s
becau
se VF can tran
sm
it d
a
ta throug
h multiple path
s
simulta
neo
usly,
but AODV ha
s only one tra
n
smi
ssi
on pa
th. The simu
l
a
tion re
sult shows that the number of d
a
ta
packet
s
sent
in AO
DV
re
ach
e
s 10
24 t
he n
u
mbe
r
i
n
VF i
s
3
321,
whi
c
h i
s
app
roximately trip
le
that of the fo
rmer.
Moreov
er, the fig
u
re
indi
ca
te
s tha
t
the thro
ugh
put of SC an
d FF i
n
crea
ses
with time, wh
erea
s VF doe
s not exhibit this ph
enom
e
non but re
mai
n
s sta
b
le at 3
.
8 or so.
The left pictu
r
e in Figu
re 6
sho
w
s that th
e su
cce
ssful
delivery rate
of VF remai
n
s ab
ove
90%, in the entire simul
a
tion pro
c
e
s
s, there
b
y i
ndica
ting that more data packe
ts can rea
c
h the
destin
a
tion n
ode
corre
c
tly. From the
30th se
cond,
the index
become
s
cl
o
s
e to 9
0
.3%.
Evaluation Warning : The document was created with Spire.PDF for Python.