TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol. 12, No. 9, September
2014, pp. 68
7
4
~ 688
3
DOI: 10.115
9
1
/telkomni
ka.
v
12i9.499
0
6874
Re
cei
v
ed O
c
t
ober 2
9
, 201
3; Revi
se
d Ju
ne 28, 201
4; Acce
pted Jul
y
21, 201
4
A Cooperative Time Synchronization Protocol for
Wireless Sensor Networks
Min Li*
1
, Gu
oq
iang
Z
h
eng
1
, J
i
s
hun
Li
2
, Le
i
Fu
1
1
School of Infor
m
ation En
gi
ne
erin
g, Hena
n
U
n
iversit
y
of Sci
ence a
nd T
e
chnol
og
y,
H
e
na
n
Lu
oy
ang
, C
h
i
na
2
Hena
n Ke
y
La
borator
y for Ma
chin
er
y
Desi
gn
and T
r
ansmission S
y
stem,
Hen
an Un
ivers
i
t
y
of Scie
nce a
nd T
e
chnol
og
y, Henan L
u
o
y
a
ng, Chi
n
a
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: limin0
9
1
9
@1
63.com
A
b
st
r
a
ct
T
he sync
h
ron
o
u
s pr
ecisi
on
of
synchr
oni
z
a
t
i
o
n
pr
otocol
is
lo
w
and th
e sc
al
abil
i
ty is
li
mit
e
d strictly
i
n
la
rg
e sca
le wi
re
l
e
ss se
nso
r
netw
o
rks
(W
SNs).Consi
deri
ng th
e iss
ue, a
nov
el
coop
erative
ti
me
synchro
ni
z
a
ti
o
n
protoco
l
bas
ed on p
u
lse-c
o
upl
ed osci
llator
s
and distri
but
ed diffusi
on (C
T
SP) is propos
ed
in this
pap
er. T
he pr
otocol w
o
rks in tw
o phas
es
clock tick s
y
nchro
n
i
z
a
t
i
on
phas
e a
nd ti
me synchr
oni
z
a
t
i
o
n
phas
e
In the
clock tick sy
nc
hroni
z
ation
ph
ase, the
ph
ase
s
of the
pu
lse-
coup
led
osci
ll
a
t
ors is ch
an
ge
d b
y
the puls
e
cou
p
l
ed sig
nal of th
e SINK node.
T
he tick syn
chroni
z
a
t
i
on of al
l nod
es is achi
e
v
ed by distri
but
ed
diffusio
n
. In time synchro
ni
z
a
ti
on ph
ase, the
avera
ge ti
me
of re
fe
re
n
c
e
no
de
s i
s
sp
read to a lim
i
ted num
ber
of hops bas
e
d
on tw
o-w
a
y messa
ge e
xchan
ge
mec
han
is
m. More
over, in ord
e
r
to achieve
the
synchro
ni
z
a
ti
o
n
of entire
net
w
o
rk, it adopts the metho
d
of mutu
al d
i
ffu
sion to fin
i
sh t
he a
pprox
i
m
at
ely
synchro
ni
z
a
ti
o
n
b
e
tw
een th
e
no
de ti
me
an
d the
av
erag
e
time
of al
l n
o
des. By c
o
mp
arin
g CT
SP w
i
t
h
TPSN, we show that, CTSP
can syn
chr
oni
z
e
the network quickly with
good prec
ision
conver
genc
e speed
and sca
la
bility,
w
h
ich appro
p
ri
ates for large-s
c
ale W
S
Ns
.
Ke
y
w
ords
: w
i
reless se
nsor n
e
tw
orks, pulse
-coup
led osc
ill
ators, cooper
ative synchr
oni
zation, distri
but
e
d
diffusion
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
Wirel
e
s
s
s
e
n
s
or n
e
two
r
ks
(WSN
s
) nor
mally
con
s
ist
of a large numbe
r of sensors
distrib
u
ted ov
er a given
area.
The
s
e lo
w-co
st se
nso
r
s h
a
ve limite
d
com
puting,
commu
nication,
and
se
nsi
n
g
ca
pa
city [1]. WSNs
ca
n
be
used fo
r monito
ring
[2-3], obj
ect
locali
zatio
n
a
nd
tracking [4
-5]
,
etc. Most of these ap
p
licatio
n
s
re
q
u
ire the op
e
r
ation of dat
a fusion, po
wer
manag
eme
n
t, and
tra
n
smi
ssi
on
sche
du
ling am
ong
a
large
set
of
sen
s
o
r
node
s, whi
c
h, in
tu
rn,
requi
re all the
node
s run
n
in
g on a com
m
on time frame
.
Ho
wever,
every individ
ual
se
nsor i
n
th
e WS
N h
a
s
i
t
s o
w
n
clo
c
k.
Dif
f
e
re
nt
cl
o
c
k
s
drif
t
from ea
ch other over tim
e
due to many factors,
such a
s
impe
rfection of the oscillato
rs and
environ
menta
l
cha
nge
s.
This ma
kes
clo
c
k synchroni
zatio
n
bet
ween
different
node
s
indispen
sabl
e
.
In ad
dition,
sen
s
o
r
node
s in a
WSN are too
ene
rgy
con
s
trai
ned
a
nd
com
putati
on
limited to use any compl
e
x synchro
n
i
z
ation
sc
hem
es. Due to al
l aforem
entio
ned
chall
eng
es,
several time
synchro
n
ization sch
e
me
s for WS
Ns hav
e bee
n propo
sed
sin
c
e El
son an
d Ro
m
e
r
first di
scusse
d this p
r
obl
e
m
in 200
2 [6], incl
udi
ng
the refe
ren
c
e b
r
oa
dcast
synchro
n
iza
t
ion
(RBS) p
r
oto
c
ol [7], time s
y
nchroni
zatio
n
proto
c
ol fo
r sen
s
o
r
net
works (TPSN) [8], the delay
measurement
time syn
c
hronization p
r
o
t
ocol
(DM
T
S) [9], the flooding time
synchroni
zatio
n
proto
c
ol
(FT
SP) [10], etc. Most of th
e pro
p
o
s
ed synchro
n
ization
metho
d
s just
focus on
the
minimizi
ng of synch
r
oni
za
tion error an
d energy
co
nsum
ption, ignori
ng the
requi
rem
ent for
scalability. At present, the ti
me synchronization protocol
s for si
ngle-hop net
works are very
mature
an
d t
he
synchro
n
i
z
ation
erro
r
can a
c
hi
ev
e a
bout u
p
to
a
dozen mi
cro
s
eco
n
d
s
. Th
e
co
st
is lower
whi
c
h can
satisfy
most appli
c
ations. As
to
multi-hop
synch
r
oni
zatio
n
,
it results in
the
accumul
a
tion
of syn
c
hroni
zation
error o
v
er ho
p
di
sta
n
ce i
n
la
rge
-
scale
WSNs.
The the
o
reti
cal
analysi
s
a
n
d
nume
r
ical e
x
perime
n
t sh
ow th
e syn
c
hroni
zatio
n
e
rro
rs a
r
e p
r
o
portion
al to t
h
e
distan
ce
of
hop
s b
e
twe
e
n
no
de
s a
n
d
refere
nce
nod
es [11].
So
there must be so
me
synchro
n
ization error a
c
cu
mulation an
d can not
satisf
y synchroni
zation accu
ra
cy in large-sca
l
e
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
A Coope
rati
ve Tim
e
Synch
r
oni
zation Protocol
for
Wireless Sen
s
o
r
Networks
(Mi
n
Li)
6875
WSNs. In addition, the compli
cate
d cal
c
ulat
io
n and frequ
ent data pa
cket
s
exchang
e will
gene
rally b
u
rden fo
r the
n
o
rmal
ru
nnin
g
of WS
Ns. Fu
rtherm
o
re, wit
h
the
gra
dual
increa
se
of t
he
netwo
rk scal
e and
the n
e
t
work-ba
s
ed
appli
c
ation,
it
become
s
m
o
re
and m
o
re time to a
c
h
i
eve
synchro
n
ization [12].
In this pape
r,
a coop
erativ
e time synch
r
oni
zation p
r
otocol (CTSP
)
is propo
se
d
to dea
l
with the issue of low precisi
on and scalability in
sensor nodes.
Accordi
ng to
the distri
buti
on
cha
r
a
c
teri
stics and
ene
rg
y information
of node
s,
maste
r
nod
e
s
are sel
e
cte
d
. Based o
n
the
partition, the
algorith
m
works in t
w
o pha
se
s
cl
ock tick syn
c
hroni
zation
pha
se a
nd t
i
me
synchro
n
ization ph
ase
Si
mulation re
sults sho
w
th
at
the
unifie
d
clo
c
k-ti
ck can ultimatel
y
be
reali
z
ed for n
e
twork no
de
s. In addition,
a salient
feat
ure of the propo
sed meth
od is that, in
the
regim
e
of
a
symptotically
den
se
netwo
rks,
it
can
m
a
intain
glob
al
syn
c
h
r
oni
zat
i
on in
the
se
nse
that all multi-hop net
wo
rk node
s
can su
ccessfully
achi
eve
co
nv
ersi
on fro
m
synchro
n
icity
to
synchro
n
ization. And
be
si
de that
it ca
n ext
end
ed holdin
g
synchroni
zatio
n
ti
me
a
nd ha
s
the
higher robust
ness and scal
ability.
The remain
d
e
r of this p
a
p
e
r is
org
ani
ze
d as follo
ws: The related
works are reviewe
d
in
Section 2. Section 3 fo
rmulates th
e time-syn
ch
ro
nizatio
n
pro
b
l
em co
nsid
ered in this pa
per.
Section
4 p
r
opo
se
s a
coo
perative time syn
c
hronization p
r
otocol to
a
nalyze th
e
time
synchro
n
ization i
s
sue
s
. Th
e re
sult
s
of e
x
perime
n
ts
an
d
s
i
mu
la
tions
ar
e d
i
s
c
u
sse
d
in Se
c
t
ion 5
,
followe
d by the con
c
lu
sio
n
s of this paper
given in Secti
on 6.
2. Related Works
In re
cent
yea
r
s,
nume
r
ou
s syn
c
hroni
zat
i
on p
r
oto
c
ol
s
have b
een
propo
sed, fo
cu
sing
on
different
asp
e
cts of th
e
synchroni
zatio
n
p
r
oble
m
in
WS
Ns [13]
. For the
tra
d
itional p
r
oto
c
ols,
they nee
d a
root no
de a
n
d
are
tre
e
-b
ased, an
d they
are
not fully
distrib
u
ted,
which
mea
n
s that
they are fra
g
ile to link or
node failu
re
s. Thus,
these traditional
proto
c
ol
s are
not optimal for
handli
ng clo
c
k syn
c
hroni
zation in rand
om mobile
se
nso
r
network. Existing distributed protocols
[14-16], the
s
e proto
c
ol
s a
nd the a
s
soci
ated theo
reti
cal results a
r
e obtaine
d a
s
suming th
at the
topology of t
he net
work i
s
conn
ecte
d
or join
t
con
n
e
cted,
whi
c
h
hold
s
no lo
nger i
n
rand
om
mobile sen
s
o
r
netwo
rk. Th
ese p
r
oto
c
ol
s also have
slo
w
co
nverg
e
n
c
e spee
d.
There are some wor
ks
which investigate Pulse-Coupl
ed
Oscillator (PCO) for sensor
netwo
rks. Th
e nonli
nea
r d
y
namics of l
a
rge
pop
ulat
i
ons
of PCO
were stu
d
ied
to describe
the
synchro
nou
s
fireflies fla
s
hi
ng, observed
in the s
outh
east of Asia
sin
c
e the pa
st two centu
r
ie
s.
The PCO al
gorithm [1
7] makes
mu
ch m
o
re
lib
eral
use of
the phy
sical co
mmuni
cation
con
s
trai
nts th
at are a
c
kno
w
led
ged
po
ssible
in
t
r
aditi
onal pa
cket-switch
ed point
-to-p
o
int
n
e
twork
model
s. Fro
m
the the
o
re
tical p
o
int of
view, t
he protocol
[14]
i
s
ba
sed on g
o
ssip avera
g
i
n
g
algorith
m
s, a
nd the proto
c
ols GTSP in [16]
and ATS in [15] are base
d
on average co
nsen
sus
algorith
m
,
wh
ich have slo
w
conve
r
ge
n
c
e sp
eed
a
s
pointed
in
[1
8].
The study
of con
s
e
n
su
s
fo
r
spa
r
se, mobi
le Ad Ho
c n
e
tworks is
p
r
opo
se
d in[1
9]. Recent result
s in [20
]
sho
w
that
the
approximate
model u
s
ed i
n
[21] to prove conve
r
ge
nce doe
s not, in fact, warrant conve
r
ge
nce for
all conn
ecte
d
networks. Note that the
convergi
ng
sp
eed
of the ti
me
synchro
n
i
z
ation
is a
cri
t
ical
probl
em in p
r
actice, while
most of existi
ng co
nsen
su
s ba
sed p
r
ot
ocol
s, whi
c
h
aim to rea
c
h
an
averag
e valu
e within the network, are ti
me-con
su
mi
n
g
. And besid
es the PCO
algorith
m
is o
n
ly
provide
s
a
unified ti
cki
n
g
rhythm a
c
ro
ss
se
nso
r
n
ode
s, n
a
m
ely syn
c
h
r
onicity n
o
t the
synchro
n
ization of time. In orde
r to real
ize the
time
synchro
n
ization, the time of each
wirel
e
ss
sen
s
o
r
nod
es nee
d to
be
synchroni
zed.
Hen
c
e
it i
s
of
great inte
re
st to devel
op
a
protocol
whi
c
h
own
s
mu
ch faster
conve
r
g
i
ng time while
maintaining t
he advanta
g
e
s
of con
s
e
n
sus.
3. Summarize the Protoc
ol Algorithm
The pe
riodi
ca
l process of CTSP as sho
w
n in Figu
re
1 is com
p
o
s
e
d
of two majo
r part
s
:
clo
ck tick sy
nch
r
oni
zatio
n
based o
n
p
u
lse
-
coupl
ed
oscillators an
d time synch
r
oni
zation. Each
cycle of syn
c
hron
ou
s execution pro
c
e
ss is as follo
ws:
1) Th
e cl
ock t
i
ck
syn
c
h
r
oni
zation
ba
sed
on pul
se
-cou
pled o
s
cillators: Firstly, SINK node
emits
m
pulse
s with
eq
uidi
stant zero-crossing.
T
he surro
undi
ng node
s re
ceiv
e
this pul
se
seq
uen
ce, an
d based on t
he location
s of the
obse
r
v
ed ze
ro
-cro
ssing
s
, the su
rro
undi
ng no
des
predi
ct
when
the next pul
se will
be
transmitted. Then, t
hese nodes emit pul
se at
their
predict
ed
times a
nd a
n
agg
regate
p
u
lse
se
que
nce is g
ene
ra
te
d. Although t
he p
r
edi
ction
at an individ
ual
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 9, September 20
14: 68
74 – 688
3
6876
node may no
t be perfect, unde
r ce
rtain
condition
s o
n
the pulse
and in asym
ptotically den
se
netwo
rks [2
2-23], the zero
-cro
ssing
s
of the ag
gr
e
gate
wavefo
rm
se
quen
ce
will b
e
at the
sam
e
positio
ns
as t
he zero
-cro
ssing
s
of the
origin
al wave
form sequ
en
ce emitted
b
y
the SINK node
due
to sp
atia
l
avera
g
ing. This agg
reg
a
t
e
pul
se se
q
uen
ce i
s
h
e
a
r
d by th
e no
des lying fu
rther
away from SI
NK nod
e and
these n
ode
s perform
pre
d
iction a
s
de
scribe
d ab
ove and emit th
eir
pulses to
the
i
r predi
cted ti
me [3]. Perip
heral
no
des being add
ed to
the
ra
nks of
the
sendi
n
g
synchro
n
ization pul
se by
outwa
rd reca
stion form
ula.
After
m
pulse
s, the new ag
ent to join the
ran
ks a
nd th
e more n
ode
s pul
se can
gene
rate sufficient en
ergy
through
cou
p
ling. At last,
synchro
n
ization pul
se
s ar
e sent by all
the node
s i
n
the net
works
at the sa
me time, na
mely,
achi
eving syn
c
hroni
zation
state
[24-25].
2) The time
synchroni
zatio
n
pha
se: Firstly,
with two-way message
excha
nge mo
dels, all
neigh
bor
nod
es time in th
e maste
r
-nod
e domai
n is
obtaine
d, and
the avera
ge
time of node
s i
n
broa
dcast d
o
m
ain is
cal
c
ul
ated. Seco
ndl
y, the ma
ster-nod
e is
synchroni
ze
d to the avera
ge tim
e
of all node
s
in the ma
ste
r-n
ode
bro
a
d
c
a
s
t domai
n, and then
d
e
fines th
e m
a
ster's time
as
referen
c
e time. The diffusi
on-n
ode
s a
r
e
cho
s
en o
n
the basi
s
of averag
e tran
sm
issi
on del
ay and
energy, which just begin
s
from t
he mast
er-nod
e and i
t
s spread to,
hop
s distan
ce node
s from
maste
r-n
ode.
The
nod
es within
hop
s di
stan
ce f
r
om t
he ma
ste
r
n
o
des will
re
cei
v
e more
clo
c
k
synchro
n
ization informatio
n from
the sa
me maste
r
n
ode or differe
nt master no
des. By usin
g of
the informatio
n to update local
cycle th
e clo
ck an
d o
peratin
g
cycl
e according t
o
the pro
c
e
ss
of implement
ation, the network will
com
p
le
te synchronization process at a time.
A
ssu
me t
hat
N
se
nsor n
ode
s of la
rg
e-scal
e
WSNs
have hi
gh de
nsity i
n
the
recta
ngul
ar a
r
ea
of
W
W
a
c
cording to
a
uni
form p
r
o
babil
i
ty distrib
u
tio
n
. In a
ddition
to the
SINK, the hardwa
r
e facilitie
s of
any other network no
d
e
are si
milar
and the com
m
unication ra
ng
o
f
s
e
ns
o
r
no
de
s
is
R
. Using
straightforwa
r
d
broad
ca
sting
or
flooding, SINK can re
a
lize the initial
informatio
n o
f
startup
syn
c
hroni
zation
and the
refe
rence nod
es
electio
n
tran
sfer ope
ration
to
sensor nodes. The implem
entation
of two stages will
be
described in the fo
llowing discussi
on.
1
T
t
Figure 1. Cycle Implement
ation of Synchroni
zatio
n
Protocol
4. The Coop
erativ
e Time
Sy
nchroniza
tion Protoc
o
l
Scheme
4.1. Clock Tick Sy
nchronization
Ba
s
e
d
on Pulse-Coupled Oscil
l
ators
In the phase, each n
o
d
e
is co
nsi
d
e
r
ed a
s
a co
ntrollabl
e oscillator. Th
e cou
p
ling
intera
ction
be
tween
nod
es
in the n
e
two
r
k i
s
mai
n
ly finish
ed th
rou
gh tra
n
smitting and
receiving
the pe
riodi
c
narro
w pul
se
sign
al, and
then real
i
z
e
s
the node
pha
se synch
r
ono
us.
In
t
h
is
scheme,
ea
ch nod
e (say
node
i
) in
th
e sensor net
work i
s
asso
ciated with an
in
crea
sin
g
monotoni
c ph
ase fun
c
tion
)
(
t
i
taking valu
es
from 0 to 1, defined a
s
:
)
0
(
)
(
i
i
T
t
t
(1)
If a nod
e i
s
i
s
olated, the
st
ate fun
c
tion
j
x
in
c
r
e
a
s
e
s
linea
r
l
y fr
om 0 to 1
s
m
oo
th
ly as
a
func
tion of time as
follows:
N
j
f
x
j
j
j
j
,...,
2
,
1
1
,
0
)
(
,
,
(2)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
A Coope
rati
ve Tim
e
Synch
r
oni
zation Protocol
for
Wireless Sen
s
o
r
Networks
(Mi
n
Li)
6877
)
(
t
X
j
)
(
t
X
j
re
f
)
(
t
j
1
1
)
(
)
(
b
t
bx
j
e
e
t
j
X
)
(
f
X
Figure 2. Gen
e
ral Mo
del of Pulse
-
Couple
d
Upd
a
ting Dynamics
The no
de em
its a pul
se wh
en the state f
unc
tio
n
achie
v
es the threshold value
(
1
j
x
). After the
pulse titillated, the node
will immediat
ely reset its
state to ze
ro
so that emi
s
sion pul
se can
be
sent in the pe
riodi
c
T
. If a node is not isol
ated, it can rece
ive p
u
lse
s
from other n
ode
s and the
n
cha
nge the st
atus varia
b
le.
The phase functio
n
)
(
t
i
repla
c
e
s
the state
function
j
x
ch
ange
s
accordingly a
s
:
()
()
(
(
)
)
1
()
1
j
jj
bx
t
j
b
xt
B
x
t
e
t
e
(3)
And
)
(
1
)
0
(
)
1
0
(
1
0
)
(
x
x
x
x
x
B
. This mean
s that a node re
ceivin
g
a
pulse eithe
r
emits the p
u
lse at
the same tim
e
or sho
r
ten
s
the waiting time for
the next cycle of e
m
issi
on
s. It can be sh
own that
only when th
e node
s emit
the pulse si
multaneo
usly
will they be insen
s
itive to coupli
ng, a
n
d
therefo
r
e a
c
hi
eve syn
c
hron
iz
ation. If there is time
0
t
su
ch that meet
s
(4),
we
con
s
i
der all
nod
es
achi
eve clo
ck tick syn
c
h
r
on
ization:
N
n
t
t
t
t
t
t
n
i
,
),
(
...
)
(
...
)
(
)
(
0
2
1
(4)
It may trigge
r infinity a
n
d
ci
rculation
firing
problem,
be
cau
s
e
of
cou
p
ling
dela
y
is n
o
t
con
s
id
ere
d
,
namely infinite feedba
ck [2
3].
Node
i
firing can result in node
j
and other no
de
s
firing,
cou
p
lin
g data
pa
cket
whi
c
h
sendi
n
g
ba
ck from
j
and
othe
r n
o
des firin
g
ma
y cau
s
e
s
nod
e
i
firing a
gain.
If
i
firing breaks out in the network, a new
cycl
e of the fi
ring nodes will
be
happ
en ag
ai
n and
cau
s
e
infinity and circulatio
n firi
ng. In ord
e
r t
o
avoid infini
te feedba
ck, we
sup
p
o
s
ed th
at after a no
de fire
s a pu
lse, it ent
ers a sho
r
t refra
c
tory pe
riod,
durin
g whi
c
h
no
sign
al can b
e
received from other no
des. That
me
ans the no
de
cann
ot resp
onse to the new
firing sig
nal. It can better
solve infinite feedba
ck pro
b
l
e
m of node
s.
4.2. The Time Sy
nchronizatio
n
Phase
The
pulse-co
upled
algo
rit
h
m
only p
r
ov
ides a u
n
ifie
d ticking
rhy
t
hm acro
ss
sen
s
o
r
node
s, nam
e
l
y synch
r
oni
city not the synchroni
zati
o
n
of time. In orde
r to
re
alize th
e tim
e
synchro
n
ization, ea
ch
sen
s
or n
ode
time ne
ed
s
to
b
e
syn
c
h
r
o
n
ized. So
we
sh
ould
ma
ke
use of
the co
ncept
of distri
but
ed
diffusin
g
and e
m
pl
oy
maste
r-n
ode
s
and diffusi
o
n
-no
d
e
s
dyn
a
mic
electio
n
me
chani
sm, the
maste
r-n
ode
s
and diffu
sio
n
-no
d
e
s
a
r
e
ch
osen o
n
the ba
sis of
the
energie
s
an
d
averag
e tra
n
smi
ssi
on d
e
l
ay. The net
work time
sy
nch
r
oni
zatio
n
is a
c
hieved
b
y
usin
g the inte
r-diffu
sion m
e
thod, the averag
e time
of the master-n
ode dom
ain i
s
diffuse
d finite
hop
s ba
se
d o
n
the two-wa
y messag
e e
x
chan
ge m
e
chani
sm. Th
e
entire
network n
ode
s time
ha
s
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 9, September 20
14: 68
74 – 688
3
6878
been
app
roxi
mately synch
r
oni
zed to
averag
e time.
T
he
ma
ster-no
des ele
c
tion must satisfy
the
following rules
Rule 1.
A
s
su
me that ea
ch
node
maintai
n
s th
re
shol
d
value
. During
the ele
c
tion
of the
maste
r-n
ode
s, each n
ode
gene
rate
s a
rand
om
num
ber
,
1
,
0
.
r
e
pr
es
en
ts
th
e ra
tio
o
f
the current
resid
ual
ene
rgy and
the
first
maximum
ene
rgy
of th
e no
de,
is
expre
s
sed
a
s
follow:
1
-
(5)
Provided
that
implie
s th
at the n
ode
ca
n be
de
cla
r
e
d
a
s
a
ma
ste
r-n
ode. In
(5
),
the thresh
old
value
of
de
termine
s
th
e
numbe
r
of n
ode
s all
o
ws
decl
a
rin
g
, th
e ratio of th
e
maste
r-n
ode
s for netwo
rk n
ode
s is
1
.
Rule 2.
All
node
s, satisfying Rule
1, wait
ce
rtai
n time
and
sen
d
a m
a
ster-n
ode
statement
m
e
ssag
e in it
s bro
a
d
c
a
s
t d
o
main
s
stoch
a
stic. If othe
r node
s
sati
sfy Rule
1 in i
t
s
broa
dcast
ra
nge, the
s
e
n
ode
s
will exit
the
as
the m
a
ster-n
ode
s comp
etit
ive; If neigh
bor n
o
des
receive different statement mess
ages
or packet collisions in th
e scope of their broadcasts
domain
s
, the
s
e no
de
s wi
ll immediatel
y send a
re
spo
nd for
conflicting inf
o
rmatio
n. Up
on
receiving the
resp
on
se m
e
ssag
e, the node that st
a
t
ements issu
es pa
cket
in accordan
ce
with
prob
ability 1/2 determi
ne
s whethe
r to continue
se
n
d
i
ng the state
m
ent messa
g
e
as the p
r
im
ary
node
until n
o
t
existing the
neigh
bo
r n
o
des ,whi
ch
receive
de
cla
r
ation p
a
cket
s from
differe
n
t
node
s in bro
adcast
s
dom
ains of sendi
ng nod
e
stat
ement me
ssage; If its neighbo
r nod
es in
broa
dcast ra
nge re
ceive only
the stat
ement
m
e
ssage, the
n
th
e no
de that
se
nt statem
ent
messag
e
can
begin
to exe
c
ute
synchro
n
izatio
n afte
r waiting
for
a
ce
rtain p
e
ri
o
d
of time. Th
e
maste
r-n
ode
s ele
c
tion i
s
multi-cy
cle; e
a
ch
second
will be re-electi
on to the
mast
er-nodes in
synchro
nou
s time
.
4.2.1. The Av
erage Time
of Nod
e
in Maste
r
-Node
Broad
cas
t Domain
A
ssu
ming t
h
a
t
S
nod
es a
r
e
electe
d to be
the maste
r
-n
ode, and the
numbe
r of no
des in
each ma
ster-node
bro
a
d
c
ast dom
ain i
s
(
1
,
2
,
...,
)
j
nj
s
,
)
(
l
n
l
l
l
j
j
c
c
c
c
,...,
,
2
1
is the tim
e
value
of
j
n
neighb
or
node
s i
n
the
maste
r-n
ode
electio
n
b
r
oa
dca
s
t d
o
mai
n
at time
l
,
(
1
,
2
,
...,
)
l
kj
ck
n
is the
time v
a
lue of
nod
e
at time
l
, where
l
c
1
is th
e ma
ster-no
de'
s ti
me value. T
h
e a
c
qui
sition
pro
c
e
ss of av
erag
e time in maste
r-n
ode
broa
dcast do
main as follo
ws
1) T
he m
a
ste
r-n
ode
broad
ca
sts
a
ch-qu
e
st p
a
ck
et
(in
c
ludi
ng the
sync-start
of st
art time
synchro
n
ization, the
ma
ster-no
de ID, the local
time value
)
to sta
r
t a
new cy
cle
of
synchro
n
ization;
2) A
c
cordi
n
g
to the
ch
-q
u
e
st p
a
cket, t
he n
e
igh
bor
node
s
se
nd
ACK re
sp
on
se pa
cket
contai
ning
a t
i
mestam
p (i
n
c
ludi
ng the
lo
cal time
value
wh
en n
ode
receivin
g ch-q
uest
pa
cket, the
local time val
ue wh
en no
d
e
tran
smitting
ACK packet, the neighb
or node
s ID) af
ter ce
rtain time
of rando
m wa
iting
3)
Whe
n
re
ceived t
he
resp
on
se p
a
cket the
ma
ster-n
ode
s
st
art to
cal
c
ul
ate the
prop
agatio
n
delay b
e
twe
e
n
no
de
s, an
d
then
se
nd
a
syn
c
-co
n
tinu
e pa
cket(i
ncl
uding
syn
c
-fl
ag,
the delay
k
d
between m
a
ste
r
-nod
e an
d re
spo
n
sive n
e
i
ghbo
r no
de
k
, the node
k
ID and the
transmissio
n time of the cu
rre
nt broa
dca
s
t inform
atio
n
)
to the neigh
bor no
de. After the neig
h
b
o
r
node
k
re
ceivin
g the syn
c
-continue p
a
cket, we ca
n o
b
tain the del
ay
k
d
,and othe
r neig
hbo
r
node
s i
s
n
o
t rep
eated
inf
o
rmatio
n ex
chang
e by
re
ceiving
the
sync-contin
ue
pa
cket until
the
maste
r-n
ode
reissu
e
a
syn
c
-contin
ue
pa
cket
an
d the
n
wait f
o
r
a m
a
ximum d
e
la
y time
max
D
but
failure to
re
ce
ive the timest
amp info
rmati
on from
neig
hbor
nod
es.
All of which show th
e ma
ster-
node o
b
tain
s the time information of all the neig
hbo
r n
ode
s.
Figure 3 sh
o
w
s the m
a
ste
r-n
ode
A
bro
a
d
c
asts
the start time
synchro
n
izatio
n pa
ckets at
time
l
, accordi
ng to the describ
ed ste
p
s t
o
impl
eme
n
t the inform
atio
n excha
nge p
r
ocess.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
A Coope
rati
ve Tim
e
Synch
r
oni
zation Protocol
for
Wireless Sen
s
o
r
Networks
(Mi
n
Li)
6879
Figure 3. A Sende
r
A
-re
ceiv
e
r
B
Two-way Ti
ming Excha
n
ge Model
k
and
k
d
denote
the cl
ock
drift and
propa
ga
tion del
ay bet
wee
n
n
ode
k
an
d nei
ghbo
r
n
ode
s,
then
k
and
k
d
can
be expre
s
se
d as:
21
kk
tt
d
(6)
21
4
3
21
4
3
T-
2
T
2
k
k
TT
T
TT
T
d
()
(
)
()
(
)
(7)
At time
l
, the maste
r-n
ode
clo
c
k time
1
1
T
c
l
was m
a
in time
of the n
e
igh
bor
nod
es which
can
be written a
s
:
1
ll
kk
cc
(8)
4) From (7)
and (8), all n
ode
s averag
e time
l
j
c
and
averag
e p
r
op
agation
delay
j
d
within the ma
ster-no
de bro
adcast dom
ai
n at time
l
tak
e
the form:
1
11
1
/
/
jj
j
nn
ll
l
jk
j
k
j
kk
n
jk
j
k
cc
n
c
n
dd
n
(9)
4.2.2. The Selection of
Diffusion
-
No
de
s and the
Diffusion of
Av
erage Time
Usi
ng a
reference time of
the avera
ge t
i
me
of the m
a
ster-n
ode, t
he con
c
rete
sele
ction
rule
s of diffusion-n
ode
s wh
ich imple
m
en
t the average
time diffusion
as follows
Rule 3.
After
receiving the ch-que
st pa
cket of
first-o
r
der ma
ster-n
ode or diffusi
on-n
ode,
node p
r
od
uces ra
ndom n
u
mbe
r
, where
)
(
1
,
0
.
1
-
is calculat
ed acco
rdin
g
to
the method
of electio
n
the
maste
r-n
ode
and n
ode e
n
e
r
gy. If
1
, the node can b
e
a d
i
ffusion-
node,
otherwise
ca
nnot
b
e
com
e
the
d
i
ffusion-nod
e. The
thre
sh
o
l
d value
1
determin
e
s the
numbe
r
of el
ection
diffusi
on-n
ode
s
an
d relates to
n
ode
den
sity, comm
uni
cati
on
radiu
s
an
d so
on. Sin
c
e
is va
riation
with e
nergy, the th
re
sho
l
d value
must adj
ust
according
to
1
1
after each
cycle of
syn
c
hroni
zation,
whe
r
e
can
set
f
o
r any
small po
sit
i
ve
according to
appli
c
ation.
Rule 4.
The node re
ceive
d
the ch
-qu
e
s
t
pa
cket
fo
r
the upp
er ma
ster-no
de o
r
diffusion
-
node, an
d then if
L
kj
dd
(whe
r
e
j
d
corre
s
po
n
d
s to the av
erag
e sin
g
le
-hop del
ay,
L
stand
s for
h
op co
mmuni
cation a
nd
k
d
denote
s
the
messag
e del
ay betwee
n
node a
nd th
e
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 9, September 20
14: 68
74 – 688
3
6880
maste
r-n
ode
or
th
e
u
ppe
r diffusion
-no
d
e
.),
the nod
e
ca
n be diffusion
-no
de,
ot
herwise can
n
o
t
become the
diffusion
-no
d
e
. Time spre
ads o
u
t fr
om the maste
r-n
ode
s and the
diffusion pro
c
e
ss
of average ti
me is a
s
follows
0-Diffu
sion: the maste
r
-no
de se
nd
s synch
r
on
ou
s p
a
cket whi
c
h
contai
ns the
followin
g
informatio
n: the ma
ster-n
ode ID; the
averag
e time
l
j
c
; the averag
e time tran
smissi
on h
o
p
numbe
r
, each
diffusion time the value minus 1; the av
erag
e delay
j
d
.
1-Diffu
sion:
whe
n
the
0-diffusion
carri
e
s
out, the
1
-
diffusio
n
n
o
des an
d the
neigh
bor
node
s which
cann
ot obtain avera
g
e
time info
rm
ation of the same ma
st
er-nod
e nee
d to
comm
uni
cate
informatio
n i
n
the b
r
o
adcast d
o
main,
and the
n
co
mputes the
a
v
erage
si
ngle
-
hop
delay
j
d
for all
the neigh
bor
node
s an
d the 1-diffu
sion
node
s up
dat
es the receiving ma
ster-
node
averag
e time fo
r:
0,
ll
jj
k
cc
d
,where
0,
k
d
sta
n
d
s fo
r info
rm
ation p
r
op
ag
ation d
e
lay
betwe
en the
cu
rrent dif
f
usion
-
no
de
and it
s ma
ster-no
de; t
he receiving
avera
ge ti
me
transmissio
n hop numb
e
r
in the
diffusi
on p
a
cket mi
nus 1; the
averag
e d
e
lay
j
d
betwe
en
the
ma
ster-n
ode and neig
hbor
no
de
s i
s
repl
aced
by
1
d
,k
,which i
s
the
averag
e d
e
la
y betwe
en
curre
n
t diffusi
on-n
ode
and
its neig
hbo
r
node
s; the u
pdated
diffusi
on pa
cket was b
r
o
a
d
c
ast
to
the next hop of the neighb
or nod
es.
f
-Diffus
i
on: the
f
-Diffu
sion proce
s
s
is simil
a
r
with
1-Diffusio
n
.The diff
usio
n process is
repe
ated until
hop distan
ces from the m
a
ster-n
ode.
Whe
r
e
de
pe
nds on
the
preci
s
ion
and
t
he
spe
ed
of the
synchro
n
i
z
ation
an
d m
u
st b
e
able to su
re t
he adja
c
e
n
t maste
r-n
ode
obtain the
av
erag
e time in
two maste
r
-node d
o
main
s at
least a
nd is
use
d
a
s
a time refe
ren
c
e
to synchro
n
i
s
tically u
pdat
e, so
sat
i
sf
i
e
s f
o
llo
wing
conditions:
22
(
)
2
2
WR
W
R
N
SS
W
R
(10)
The ma
ster-node ID i
s
set to avoid
repe
atedly
receiving the
time synchronization
informatio
n from the sam
e
master-no
d
e
domai
n. Du
ring a ce
rtain
diffusion no
des p
e
rfo
r
mi
ng
diffusion, the
neigh
bor
no
des
have received the av
erag
e time d
i
ffusion info
rmation fro
m
the
same
ma
ste
r
-nod
e,
which
ha
s
wa
s
be
en
confi
r
med
ba
sed
on
th
e recordi
ng t
he m
a
ste
r
-no
d
e
numbe
r, and
it is no longer involved
in t
he diffusion node inf
o
rmatio
n exchang
e; If all the
neigh
bor n
o
d
e
s of diffusi
on-n
ode
s ha
ve been al
re
ady re
ceived
the averag
e
time diffusion
in
fo
r
m
a
t
ion
fro
m
th
e sa
me mas
t
e
r
-
n
od
e, th
e
d
i
ffus
i
on
-
n
od
e w
ill en
d
th
e d
i
ffu
s
i
o
n
pr
oc
ess
afte
r
having be
en
waited o
n
for a certai
n time
.
Acco
rdi
ng to the re
ceived a
v
erage time d
i
ffusi
on pa
cke
t
s of the master-nod
e in the each
cycle of syn
c
hroni
zatio
n
time, the node
cal
c
ulate
s
the
new time
new
T
[(
)
]
(
1
)
n
e
w
M
k
l
oc
al
Tc
l
d
L
T
L
(11)
Whe
r
e
)
(
l
c
M
is th
e avera
ge ti
me of ma
ste
r-n
ode
s d
o
m
a
in,
k
d
is n
ode
prop
agatio
n
delay b
e
twee
n
k
node
and
the ma
ster-n
ode
and
L
is the received
synchro
n
ization diffu
sio
n
packet
s
n
u
m
ber which
mi
nus 1. If
ne
w
l
oca
l
TT
, the
update
of n
o
d
e
is such th
at
new
T
, otherwi
se
to maintain local time
local
T
at consta
nt value
.
5. Experiment and Simulation Results
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
A Coope
rati
ve Tim
e
Synch
r
oni
zation Protocol
for
Wireless Sen
s
o
r
Networks
(Mi
n
Li)
6881
In orde
r to validate the synchrono
us e
ffect
of the propo
se
d syn
c
hroni
zatio
n
al
gorithm
simulate
s th
e
experi
m
ent
on the
Mica
Z platform.
G
i
ving the
co
rrelative pa
ram
e
ter valu
e fo
r the
simulatio
n
: monitori
ng area
is
2
m
500
500
, 1
000 n
ode
s are
ra
nd
omly depl
o
y
ed
,
comm
uni
cati
on ra
diu
s
of sen
s
o
r
no
de
s are
m
R
10
, the cou
p
ling st
rengt
h
is
02
.
0
,
cy
cle
time is
10
Ts
, the simulatio
n
time is 50min
, and each
experim
ent is the mean
of 100
simulat
i
o
n
s.
5.1.
Analy
s
in
g the Co
nv
er
gence o
f
CT
SP
In the previo
us
se
ction,
we analy
z
ed
a
sy
nchrono
us versi
on of
CTSP alg
o
rit
h
m an
d
given its im
p
l
ement p
r
o
c
e
dure.
We
wil
l
analyze
the
conve
r
g
e
n
c
e of CTSP i
n
this
se
ctio
n.
Suppo
se that
the deviation
s for all th
e n
ode
s time
an
d stan
dard time are unifo
rm distri
buted
in
[L
t
, Ht]
at time
l
, CTSP satisfi
e
s convergen
ce theo
rem a
s
follow.
Theorem 1
.
For la
rge
-
sca
l
e se
nsor n
e
twork, CTSP can
be g
r
ad
u
a
l co
nverg
e
n
c
e in
C
,
whi
c
h eq
uals
to the averag
e clo
ck of all the nod
es in t
he network.
Pro
v
e
Assu
ming that the
numbe
r of main no
des i
s
S
ea
ch sele
cted roun
d,
j
t
H
and
j
t
L
denote
the m
a
ximum a
nd
minimum val
ue of
devia
ti
ons for th
e a
v
erage
time
of S ma
ster-
node
s dom
ai
ns after
j
s
e
t;
)
(
l
c
j
n
and
)
(
l
c
std
stand f
o
r the arbitra
r
y nodes time
and sta
nda
r
d
time after
j
se
t o
f
s
y
nc
hr
on
iz
a
t
ion
,
then
after
synchronizi
ng
set,
the arbitra
r
y node tim
e
)
(
l
c
n
sat
i
sf
ie
s:
11
1
1
...
(
)
(
)
...
tt
t
t
n
s
t
d
t
t
t
t
L
LL
L
c
l
c
l
H
H
H
H
(12)
We kn
ow
j
t
H
C
an
d
j
t
H
is
noni
ncre
asin
g. Letting
the infimum
of the
seri
e
s
j
t
H
be
M
,we have
li
m
j
t
t
HM
C
.Su
ppo
se
M
C
We will derive a contradi
ction.
Con
s
id
er th
e
function
1
()
i
i
x
nx
. Ch
oose
x
su
ch t
h
at
1
1
()
1
n
n
M
xM
x
n
C
n
wh
ere
n
is the
num
b
e
r
of sen
s
o
r
s. For any
(
,
1
,
...,
1
)
nn
de
fin
e
1
to be th
e set of
se
nsors
who
s
e value
s
are gre
a
ter than
()
M
xn
and
2
to be the set of the re
st of the nodes. Fo
r
x
,
there mu
st ex
ist a time
t
su
ch t
hat
t
HM
x
; also, there mu
st b
e
som
e
nod
e
who
s
e valu
e
is
less than
()
CM
x
n
becau
se
C
is th
e averag
e value. Starting from sets
1
and
2
at time
t
, we have
2
1
n
. After the first
averag
e op
eration for
nod
es that
are
in
1
and
2
, we have
2
1
2
n
. After the first averag
e o
p
e
ration
on
no
des in
1
1
n
and
2
1
n
, we
have
2
2
3
n
. So,
this contradicts that the in
fimum of
j
t
H
is
M
.Therefo
r
e, we
have
li
m
j
t
t
HC
. In the s
a
me
way,
we can
prove
t
hat
li
m
j
t
t
LC
.Combinin
g
the
s
e t
w
o
re
sult
s, we
have th
a
t
all the val
u
e
s
o
n
the sen
s
o
r
s
converg
e
to
C.
Provide a
sta
t
ement that what
i
s
expe
cted, a
s
stat
ed in t
he
"Int
rodu
ction" ch
apter can
ultimately result in "Results and
Discu
s
sion"
chapt
e
r
, so there i
s
comp
atibility.
More
over, it ca
n
also
be
ad
de
d the
prospe
ct of
the
devel
opment
of re
sea
r
ch
re
sult
s a
nd
appli
c
a
t
ion p
r
o
s
pe
cts of
further
studie
s
into the nex
t (base
d
on result an
d discussion
)
.
5.2.
CTSP Versus TPSN
Assu
ming th
e average
synchroni
za
tio
n
erro
r of e
a
ch
hop i
s
, the time to realize
synchro
n
ization b
e
twe
en
one
hop
neig
hbors i
s
alm
o
st
id
entical results whi
c
h
are
. In the
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 9, September 20
14: 68
74 – 688
3
6882
con
d
ition
of the
same
net
work
pa
ramet
e
r, the
syn
c
h
r
oni
zation
e
r
ror of
CTSP
a
nd TPS
N
can
be
expre
s
sed a
s
:
2
2
CT
S
P
TP
SN
nS
S
W
S
S
R
N
WR
(13)
From
comp
arison, it ca
n b
e
found
that synchro
nou
s error
of
CTSP is greatly redu
ced.
The
com
pute
r
si
mulatio
n
result
s of th
e
TPSN an
d
CTSP algo
rith
m are give
n f
o
r
com
pari
s
o
n
, by
mean
s of the
Figure 4
we
can find that,
with t
he wire
less hop
s in
crea
sing, the
synchro
n
ization
conve
r
ge
nt ra
te incre
a
se a
s
loga
rithm m
ode ap
proxim
ately.
Figure 4. Synchroni
zation
Conve
r
ge
nt Rate
The syn
c
h
r
o
n
izatio
n pro
p
o
rtion valve
and k
eepi
ng
the synchroni
zation time o
f
the two
algorith
m
s wi
th
different
n
e
twork scale con
d
it
ion
s
are given in Fi
gure
5 and F
i
gure
6 Thro
ugh
the
contrast a
nd com
pari
s
o
n
of
syn
c
hron
ization
p
r
opo
rtion valve
of
different network scale
s
,
we
can fin
d
the
CTSP algo
rit
h
m always p
r
ovides
an
eff
e
ctive techni
que to
synchronize for all
kinds
of network
u
nder vari
ou
s scen
ario
s.
As
well
as the
synchro
n
i
z
ation
pe
rformance of
T
PSN
algorith
m
is relatively high
er by the infl
uen
ce
of the
variety of the netwo
rk scale. Therefore,
CTSP algo
rithm can a
dapt
to the chang
e of network
scale very we
ll.
Figure 5. Synchroni
zation
Propo
rtion
Figur
e 6. Hol
d
ing Synch
r
o
n
izatio
n Time
6. Conclusio
n
In this pap
er,
a coo
p
e
r
ative
synch
r
o
n
izat
ion protocol, named
CTSP, has b
een p
r
o
posed
to solve
the
probl
em
s a
ssociate
d
with l
o
w
accu
ra
cy and poo
r sca
l
ability,
whi
c
h
wid
e
ly
exist
i
n
mo
s
t
c
l
oc
ks
in
W
S
N
s
. T
he p
u
l
se
-c
o
u
p
l
ed
is
us
e
d
for
clo
ck ti
ck
syn
c
hroni
zing, a
nd the n
e
two
r
k
node
s at the
same tim
e
wi
th distrib
u
ted
coo
per
ation diffusion are pre
s
ente
d
.
And
be
sid
e
s,
t
h
e
conve
r
ge
nce
of algo
rithm
i
s
a
nalyzed
in
theo
reti
cal, a
nd the
synchronizati
on
erro
r exp
r
e
ssi
on
is
given. That
dire
ctly proves
th
e b
e
tter
synchro
n
ization e
r
ror
perfo
rman
ce
of CTSP. The
5
10
15
20
25
30
35
40
45
50
0
5
10
15
T
he num
ber
of
hop
C
o
nv
erg
e
n
c
e ra
t
e
of
s
y
nc
hro
n
i
z
a
t
i
o
n
T
PSN
CT
S
P
2
4
6
8
10
12
14
16
18
20
30
40
50
60
70
80
90
10
0
L
eng
t
h
o
f
ed
ges
f
o
r
gr
i
d
t
o
p
o
P
e
rc
en
t
age
of
S
y
nc
h
r
oni
z
a
t
i
on(%
)
TP
S
N
CT
S
P
2
4
6
8
10
12
14
16
18
20
20
40
60
80
10
0
12
0
14
0
16
0
T
h
e
ne
t
w
or
k
di
am
et
e
r
T
i
m
e
t
o
S
y
nc
hr
oni
z
a
t
i
on(N
u
m
ber of
P
e
ri
ods
)
TP
S
N
CT
S
P
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
A Coope
rati
ve Tim
e
Synch
r
oni
zation Protocol
for
Wireless Sen
s
o
r
Networks
(Mi
n
Li)
6883
simulatio
n
re
sults
demo
n
strate the correctne
ss of
theoreti
c
al a
nal
ysis an
d CTSP can effe
ctively
hold l
ong
er
synchro
n
ization time, im
p
r
ove th
e
rate
of conve
r
ge
nce
of synch
r
oni
zation, a
nd
signifi
cantly redu
ce th
e
synch
r
oni
zatio
n
error.
As
our
future
wo
rk, i
t
is inte
re
stin
g to eval
uate
the
perfo
rman
ce
of the propo
sed CTSP in a
netwo
rk-wi
d
e scena
rio wit
h
a dynamic t
opolo
g
y
Referen
ces
[1]
Daoj
in
g He,
Li
n Cu
i, Hej
i
a
o
H
uan
g. Desi
gn
a
nd ver
i
ficatio
n
of enh
anc
ed s
e
cure
loca
liz
ati
on sch
eme
in
w
i
reless sensor net
w
o
rks.
IEEE Transactio
n
s on P
a
ral
l
el
and
Distrib
uted
Systems.
20
0
9
; 20(7): 1
050
-
105
8.
[2]
Yik-Chu
ng W
u
, Qasim C
h
aud
hari, Erc
h
i
n
Serp
ed
in.
Clock s
y
n
c
hro
n
izati
on
of
w
i
reless s
ens
or
net
w
o
rks.
IEEE Signal Processing Maga
z
i
n
e
.
2011; 2
8
(1): 124-
138.
[3]
Duzu
n Do
ng,
Xi
an
gke L
i
a
o
, Yunh
ao
Liu. E
dge s
e
lf-mon
i
torin
g
for
w
i
r
e
le
ss sensor
net
w
o
rks [J].
IEEE
T
r
ansactio
n
on
Parall
el an
d Di
stribed Syste
m
s.
2012; 23(
3): 514-
527.
[4]
W
a
i-Le
ong
Ye
o
w
,
C
hen-K
h
o
ng T
ham, W
a
i-
Cho
ong
W
o
n
g
. Ener
g
y
effici
e
n
t multi
p
le
targ
et trackin
g
i
n
w
i
rel
e
ss sens
o
r
net
w
o
rks
. IEEE Transactions
on Vehic
u
lar Technology
. 200
7; 56(2): 91
8–9
28.
[5]
Jiming
Che
n
, Keji
e Cao, Ke
yo
ng L
i
. Distri
buted s
ens
or activatio
n
al
go
rithm
for target tracking
w
i
t
h
bin
a
r
y
s
ensor
net
w
o
rks.
Cl
us
ter Computi
n
g
.
2011; 1
4
(1): 5
5
-64.
[6]
Jerem
y
E
l
son,
Ka
y
Romer.
W
i
reless s
ens
or n
e
t
w
orks:
A ne
w
re
gim
e
for time
s
y
n
c
hron
izatio
n.
Co
mp
uter Co
mmu
n
ic
ation R
e
view
. 2003; 33(
1): 149–
15
4.
[7]
Jerem
y
E
l
son,
Le
w
i
s Gir
od,
De
bora
h
Est
r
in.
F
i
n
e
-gra
in
ed
netw
o
rk ti
me
sync
h
ron
i
z
a
ti
on
usi
n
g
referenc
e bro
a
d
casts. 5th Sy
mp
osi
u
m
on O
perati
ng Syste
m
s
Des
i
g
n
an
d
Impl
e
m
entati
o
n.
200
2: 147-
163.
[8]
Saura
bh Ga
ne
ri
w
a
l,
Ram K
u
mar, Mani
B
Srivastava. T
i
ming-s
y
nc
pr
o
t
ocol for s
ens
or net
w
o
rks.
Procee
din
g
of F
i
rst Internatio
nal C
onfere
n
c
e
on E
m
be
dd
e
d
Netw
orked S
ensor Syste
m
s
. 2003: 13
8-
149.
[9]
Ping S
u
. De
la
y m
easur
eme
n
t time s
y
nc
hr
oniz
a
tion
for
w
i
rel
e
ss se
nso
r
net
w
o
rks.
Int
e
l R
e
searc
h
.
200
3.
[10]
Shib
o H
e
, Jimi
ng C
h
e
n
, Yo
u
x
ian
Sun. On
o
p
tima
l
inform
ation c
aptur
e b
y
en
erg
y
-co
n
strain
ed m
obi
l
e
sensor.
IEEE Transactions on Vehic
u
lar Technology
. 20
10; 59(5): 39
–4
9.
[11]
Hu An-S
w
o
l,
Sergi
o
D S
e
rv
etto. On the s
c
ala
b
il
it
y
of co
oper
ative tim
e
s
y
nchro
n
iz
ati
on i
n
p
u
ls
e-
conn
ected n
e
tw
o
r
ks.
IEEE Tr
ansactions on
Information Theory
. 2006; 5
2
(6
): 2725-2
7
4
8
.
[12]
B
y
o
u
n
g
-Ku
g
Ki
m, Sung-
H
w
a
Hon
g
, K
y
e
ong
Hur. E
nerg
y
-
e
fficient a
n
d
rap
i
d tim
e
s
y
nc
hr
oniz
a
tion
fo
r
w
i
reless sens
or net
w
o
rks.
IEEE Transactions
on Cons
um
er
Electronics
. 20
10; 56(4): 2
258
-226
6.
[13]
Ill-Keu
n
Rh
ee,
Jaeh
an
Lee;
J
angs
ub Kim.
Clock s
y
nc
hro
n
izati
on i
n
w
i
r
e
less s
ensor
net
w
o
rks:
A
n
overvi
e
w
.
Sens
ors
. 2009; 9(
1): 56–8
5.
[14]
Nicol
as Mar
e
c
hal, Je
an-B
e
n
o
it Pierr
o
t, Jea
n
-Mari
e
Gorce
.
F
i
ne s
y
nc
hro
n
izati
on for
w
i
r
e
less s
ens
or
net
w
o
rks usi
n
g
goss
i
p aver
agi
ng alg
o
rith
ms.
EEE International Confer
ence on Comm
unication
s.
200
8: 496
3–
49
67.
[15]
Luca Sc
he
nat
o, F
ederic
o F
i
or
e
n
tin. Aver
age tim
e
s
y
nc
h: a co
nsens
us-bas
ed
prot
ocol for tim
e
s
y
nc
hro
n
izati
o
n in
w
i
r
e
less s
ensor n
e
t
w
orks
.
Autom
a
tica.
2011; 47(
9): 187
8-18
86.
[16]
Phili
pp S
o
mm
er, Rog
e
r W
a
t
t
enhofer.
Gradient clock synchroni
z
a
tion
i
n
w
i
reless se
ns
or netw
o
rks
.
Procee
din
g
s of
IPSN. 2009: 3
7
-48.
[17]
Rob
e
rto Pagl
ia
ri, Anna Scag
li
one. Scal
ab
le
net
w
o
rk s
y
nc
h
r
oniz
a
tion
w
i
t
h
pulse-c
oup
le
d
oscillat
o
rs.
IEEE Transactions on Mobile
Com
p
uting
. 2
0
11; 10(3): 3
92-
405.
[18]
He Ji
an
pin
g
, C
hen
g Pe
ng, S
h
i Li
ng, C
h
e
n
Ji
mi
ng. T
i
me s
y
nchro
n
izati
on
i
n
W
S
Ns: A ma
xim
u
m va
lu
e
base
d
co
nsen
sus ap
pro
a
ch.
Proceedings of
the
IEEE
Conference on De
cision and Control
. 20
11:
788
2–
788
7.
[19]
Aleke
i
sh Kh
al
ed, Ezhi
lche
lv
an Pa
ul. C
o
n
s
ensus
in Sp
arse, Mob
ile
Ad Hoc N
e
t
w
orks
. IEEE
T
r
ansactio
n
s o
n
Paral
l
el a
nd
Distribut
ed Sys
t
ems
. 20
12; 23
(3): 467–
47
4.
[20]
F
u
mito Mori, T
a
kashi Odag
aki. S
y
nchro
n
i
z
ati
on of co
u
p
le
d oscil
l
ator
s on small-
w
o
rld net
w
o
rks
.
Physica D: No
nlin
ear Ph
en
o
m
e
na.
20
09; 2
38(1
4
): 118
0-1
185.
[21]
Lucar
ell
i
D
e
n
n
is, W
ang
IJeng.
D
e
ce
ntra
li
z
e
d
sync
h
ro
ni
z
a
ti
on
proto
c
ols w
i
th n
e
a
r
est nei
gh
bo
r
communic
a
tio
n
. Proc. Second
Int’l Conf. Embed
de
d Net
w
o
r
ked Se
nsor S
y
stems (Se
n
S
ys ’04). 20
04:
62-6
8
.
[22]
W
u
Jiansh
e
, Jiao Lic
h
e
ng, Di
ng Ra
nran. Av
erag
e ti
me s
y
n
c
hron
izatio
n in
w
i
rel
e
ss sens
o
r
net
w
o
rks b
y
pair
w
i
s
e mess
ages.
Co
mpute
r
Communic
a
ti
ons
. 201
2; 35(
2): 221–
23
3.
[23]
Xi
ao Y W
ang,
Raje
ev K Dok
a
nia,
Al
yssa Ap
sel. PCO-base
d
s
y
nchro
n
izati
on for cog
n
itiv
e dut
y-c
y
cle
d
impuls
e
rad
i
o s
ensor n
e
t
w
orks
.
IEEE Sensors Journal.
201
1; 11(3): 555-
56
4.
[24]
Mengfa
n
Ch
e
ng, Han
p
i
ng
Hu. Impulsiv
e
s
y
nc
hr
on
izati
on of cha
o
tic
s
y
stems
w
i
th
time-var
yin
g
param
eters.
Journa
l of Co
mp
utation
a
l Infor
m
ati
on Syste
m
s.
2012; 8(1
1
): 475
9-47
68.
[25]
Jiming
Ch
en,
Qing Yu, Y
a
n Z
han
g. F
e
e
dback-
base
d
clock s
y
nc
hro
n
izati
on i
n
w
i
r
e
less s
enso
r
net
w
o
rks: A control the
o
retic
appro
a
ch.
IEEE Transactions on Vehicular Technology.
2010; 59(
6):
296
3-29
73.
Evaluation Warning : The document was created with Spire.PDF for Python.