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.
2
,
A
pr
il
2020, p
p. 13
52
~
1358
IS
S
N: 20
88
-
870
8
,
DOI: 10
.11
591/
ijece
.
v10
i
2
.
pp1352
-
13
58
1352
Journ
al h
om
e
page
:
http:
//
ij
ece.i
aesc
or
e.c
om/i
nd
ex
.ph
p/IJ
ECE
Chaotic
si
gnals denoisin
g
usin
g e
m
pirical
mode d
ecomp
osition
insp
ired b
y multivari
ate denoising
Fad
hil
Sahi
b
Ha
s
an
Depa
rtment
o
f
E
le
c
tri
c
al E
ngin
eering,
Mus
ta
nsir
i
y
ah
Univer
si
t
y
,
I
raq
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
J
un
4
, 2019
Re
vis
ed
Oct
17
,
2019
Accepte
d
Oct
27, 201
9
Empiric
al
m
od
e
dec
om
positi
o
n
(EMD)
is
an
eff
ectiv
e
noi
se
red
uction
m
et
hod
to
enh
an
ce
th
e
nois
y
chaotic
sign
al
ov
er
addi
ti
v
e
noise
.
I
n
thi
s
pap
er,
the
in
tri
nsic
m
ode
func
t
ions
(IMF
s)
gene
ra
te
d
b
y
EMD
are
thr
esholde
d
usin
g
m
ult
iva
ri
at
e
d
e
noising.
Multi
v
ari
a
te
deno
ising
is
m
ult
iva
ri
able
denosing
al
gorit
hm
that
is
combined
wave
l
et
tr
ansform
and
princ
ipal
component
ana
l
y
sis
to
den
oise
m
ult
iva
ri
ate
signal
s
in
ada
pti
v
e
wa
y
.
Th
e
proposed
m
et
hod
is
comp
are
d
at
a
v
ari
ous
signa
l
to
nois
e
rat
ios
(SN
Rs)
with
diff
ere
n
t
te
chn
ique
s
and
diffe
ren
t
t
y
pes
of
noise.
Also,
sca
le
d
epe
nd
ent
L
y
apunov
expone
nt
(SD
L
E)
is
used
to
te
s
t
the
b
eha
v
ior
o
f
the
d
enoi
sed
c
haot
i
c
signal
compari
ng
with
cl
ea
n
sign
al
.
T
he
result
s
show
tha
t
EMD
-
MD
m
et
hod
ha
s
the
best
root
m
ea
n
square
err
o
r
(RMS
E)
and
signal
to
noise
rat
io
gain
(SN
RG) c
om
par
ing
with
the c
on
vent
ion
al
m
et
ho
ds.
Ke
yw
or
d
s
:
Chaotic
sig
nal
Em
pirical
m
od
e d
ec
om
po
sit
ion
Mult
ivariat
e d
e
no
isi
ng
Wav
el
et
de
no
i
sing
Copyright
©
202
0
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
Fadhil
Sah
i
b H
asan,
Dep
a
rtm
ent o
f El
ect
rical
En
gi
neer
i
ng,
Mustansiri
ryah
Unive
rsity
, I
ra
q.
Em
a
il
:
fad
el
_sa
hib@u
om
us
ta
ns
iriy
ah.
e
du.iq
1.
INTROD
U
CTION
Chaotic
sign
al
s
hav
e
m
any
pro
per
ti
es
su
c
h
as
aper
i
od
ic
it
y,
sensiti
vity
to
init
ia
l
con
diti
on
s
,
an
d
wide
band
s
pe
ct
ru
m
that
m
ake
them
su
it
able
to
us
e
in
m
any
app
li
cat
io
ns
ar
eas
su
c
h
as
secu
r
e
com
m
un
ic
at
ion
[
1],
i
m
age
encr
y
ption
[2
]
,
sp
eec
h
enc
ryp
ti
on
[
3]
and
ot
her
ap
plica
ti
ons
.
Howe
ver
,
w
h
en
the
chao
ti
c
sign
al
s
co
rru
pte
d
with
a
no
is
e
beco
m
e
ver
y
har
d
to
fin
d
Ly
apun
ove
expo
nen
t,
c
orr
el
at
ion
dim
ension
,
K
olm
og
oro
v
e
ntr
oy
an
d
oth
e
r
c
hao
ti
c
syst
em
par
am
et
ers
[4
]
.
The
refor
e
,
re
m
ov
e
s
the
nois
e
from
the ch
a
otic si
gnal
s
in
af
fecti
ve
w
ay
a
re th
e
m
ai
n
chall
en
ge
s and t
he great
sign
ific
a
nt in
this r
e
searc
h.
In
t
he
la
st
ye
ars,
diff
e
re
nt
te
chn
i
qu
e
s
are
i
ntr
oduce
d
to
r
e
m
ov
e
the
no
i
se
from
the
chao
ti
c
sign
al
s
.
The
m
os
t
fa
m
ou
s
e
m
e
tho
d
us
in
g
wa
velet
transfor
m
[
5
-
7
]
in
wh
ic
h
the
cha
otic
sign
al
is
denoise
d
by
d
ecom
po
s
i
ng
it
into
detai
l
a
nd
ap
pro
xim
ate
com
ponen
t
s
and
the
n
t
he
de
ta
il
s
are
sm
oo
the
d
us
in
g
a
da
ptive
thres
ho
l
ds
.
T
o
find
op
ti
m
u
m
threshol
d
for
each
scal
e
in
wav
el
e
t
dom
ain
,
ge
netic
al
gorithm
is
su
ggeset
ed
in
[8
]
.
In
[
9
-
11]
du
al
wa
velet
trans
form
are
us
e
d
as
an
e
xtensi
on
to
wav
el
et
trsform
to
re
m
ov
e
the
no
ise
from
chao
ti
c
si
gn
al
wh
e
re
op
ti
m
al
deco
m
po
sit
io
n
scal
e
an
d
a
dapt
ive
sel
ect
ing
wav
el
et
c
oeffici
ents
are
deter
m
ined
.
Wh
il
e
wa
velet
transfo
rm
con
ta
ins
only
ti
m
e
do
m
ai
n
local
it
y,
synchr
osque
ezed
wa
ve
le
t
tra
ns
f
or
m
(SWT)
con
at
in
s
cot
h
tim
e
and
f
requen
cy
pro
per
ti
es
is
us
e
d
with
hierar
c
hical
thres
hold
to
enh
a
nce
t
he
c
hao
ti
c
sign
al
[12].
A
no
t
her
m
os
t
fa
m
ou
se
denoisi
ng
te
ch
nique
i
s
em
pirical
m
od
e
dec
om
po
s
it
ion
(EM
D)
[
13
]
in
wh
ic
h
the
sig
na
l
is
deco
m
po
se
d
into
m
any
sign
al
s
of
am
plit
ud
e
a
nd
fr
e
qu
e
ncy
m
od
ul
at
ed
with
zer
o
m
ean
value
that
are
cal
le
d
intrinsic
m
od
e
functi
ons
(I
MFs)
a
nd
t
hen
at
certai
n
thres
hold
sel
ect
wh
ic
h
m
od
e
is
us
e
d
to
reconstr
uct
the
denoised
sign
al
.
EMD
is
i
m
pr
ov
ed
in
[
14,
15
]
by
us
ing
e
ns
e
m
ble
e
m
p
iric
a
l
m
od
e
deco
m
po
sit
io
n
(EEMD
)
a
nd
EEM
D
an
d
singular
val
ue
deco
m
po
si
ti
on
(
SVD
)
r
especti
vely
.
A
no
t
her
i
m
pv
em
nt
to
E
MD
is
dep
ic
te
d
in
[
4]
in
w
hi
ch
zer
o
-
c
r
os
sin
g
scal
e
thrsh
old
in
g
en
ha
ncem
ent
al
gorithm
i
s
us
e
d
to
en
ha
nce
noisy
cha
otic
sign
al
.
Othe
r
de
no
isi
n
g
c
ha
otg
ic
te
ch
niques
wh
ic
h
are
c
om
bin
ed
EM
D
a
nd
ind
e
pende
nt c
om
po
nen
t a
naly
sis (I
C
A)
are
depict
ed
i
n
[
16, 17]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Chaoti
c sig
na
l
s d
e
noisi
ng usi
ng empiri
cal
m
od
e
d
ec
omp
os
i
ti
on
in
sp
ire
d b
y … (
Fa
dh
il
Sahib
H
as
an
1353
In
this
pap
e
r,
a
ne
w
denoisi
ng
te
ch
nique
t
hat
is
c
om
bin
ed
bo
t
h
EM
D
a
nd
m
ulti
var
ia
t
e
de
no
isi
ng
us
in
g
wa
velet
and
pr
inci
pal
com
po
ne
nt
an
al
ysi
s
(MD
-
WPCA
)
to
de
noise
the
chao
ti
c
sign
al
corrupt
ed
by
certai
n
a
dd
it
iv
e
noise
is
pro
pose
d
a
nd
nam
ed
EM
D
-
M
D
al
gorithm
.
MD
-
WPCA
is
an
extensi
on
of
w
avelet
denoisin
g
t
o
m
ulti
var
ia
te
sig
na
ls
that
is
pro
pose
d
in
[
18
]
to
de
no
ise
m
ultiv
ariat
e
si
gn
al
s
instea
d
of
un
i
va
riat
e
sign
al
an
d
it
is
com
bin
ed
wa
velet
transfor
m
and
pri
ncipal
com
po
ne
nt
ana
ly
sis
(P
CA)
.
I
n
this
pap
er
,
ins
pire
d
by
MD
-
WPCA
,
an
intri
ns
ic
m
od
e
f
unct
ion
s
(
IMFs)
ge
ner
at
ed
by
EMD
ar
e
pro
per
ly
a
da
pted
a
nd
th
res
ho
l
de
d
to
de
no
ise
t
he
chao
ti
c
sig
nal.
Fu
rt
her
m
or
e,
s
cal
e
d
epe
nde
nt
Ly
ap
unov
e
xponent
(SDL
E)
f
unct
ion
is
us
e
d
as
a
m
easur
e
to
fin
d
the
am
ou
nt
of
e
nh
a
nce
m
ent
factor
f
or
the
pro
po
s
ed
syst
e
m
com
par
ing
to
th
e
cl
ean
chao
ti
c si
gn
al
.
The
rest
of
t
hi
s
pap
e
r
is
fo
l
lowe
d
as:
S
ec
ti
on
2
prov
i
de
s
the
blo
c
k
diagr
am
of
t
he
su
ggest
e
d
al
gorithm
.
Sect
ion
3
pro
vid
e
s
perform
ance
evaluati
on
of
no
ise
reducti
on
m
et
ho
d.
T
he
si
m
ulati
on
res
ults
of
noi
se
r
e
du
ct
i
on are
s
umm
erize
d
in
Secti
on
4.
Fu
rt
her
m
or
e,
S
ect
ion
5
c
on
ta
i
ns
t
he
c
on
cl
us
ion
.
2.
EMD
B
AS
E
D
C
HAOT
I
C
DE
NOISI
NG
I
NS
P
IR
ED
BY
M
U
LT
IVARI
ATE
DENOISI
NG
(EM
D
-
M
D)
Figure
1
s
how
the
bl
ock
diag
ram
of
E
MD
ba
sed
c
ha
otic
de
no
isi
ng
ins
pire
d
by
m
ult
ivariat
e
denoisin
g
(EM
D
-
M
D)
.
I
n
thi
s
syst
e
m
,
the
c
le
an
cha
otic
si
gn
al
x
(
n
)
is
c
orr
up
te
d
by
a
noise
w
(
n
)
with
le
ng
t
h
N,
t
hen the
noisy
ch
a
ot
ic
signal
r
(
n
)
is
giv
e
n by:
r
(
n
)
=x
(
n
)
+ w
(
n
)
,
n=1,
…,N
.
(
1)
The
obj
ect
ive
is
to
se
par
at
e
the
cl
ean
c
ha
ot
ic
sign
al
f
r
om
the
noise
si
gn
al
a
nd
re
co
ver
t
he
inte
res
t
cl
ean
chao
ti
c
sig
nal.
In
t
he
first
st
ep
,
the
si
gnal
x
(
n
)
is
dec
om
po
se
d
into
a
set
of
L
basis
f
un
ct
ion
cal
le
d
int
rinsic
m
od
e fu
nctio
ns (I
MFs
),
c
i
(
n
),i
=1,
…,
L,
us
i
ng EMD algo
rith
m
[4, 13
,
14
]
. Two
conditi
on
s ar
e require
d
in eac
h
IMFs
[
1
3,
14
]
:
First,
the
extr
e
m
a
nu
m
ber
a
nd
zer
o
cr
os
si
ng
nu
m
ber
m
u
st
be
equ
al
or
diff
e
r
at
m
os
t
by
on
e
.
Seco
nd,
the
av
erag
e
val
ue
of
the
up
per
an
d
lowe
r
e
nv
el
opes
de
fine
d
by
t
he
local
m
axi
m
a
and
m
ini
m
a
m
us
t
be
ze
ro.
O
ne
of
the
m
os
t
f
a
m
ou
s
al
gorith
m
s
to
find
eac
h
IMFs
is
cal
le
d
sifti
ng
pr
oc
ess
that
is
it
er
at
ive
process
. T
he pr
ocedu
re
of
sifti
ng alg
or
it
hm
can be s
umm
eri
zed in b
riefly
a
s
[
4,
13
,
14
]:
1)
Com
pu
te
local
m
axi
m
a,
m
ax
j
, j
=1
, 2,…
and l
ocal m
ini
m
a,
m
in
k
, k
=
1,2,…. in
r
(
n).
2)
Using
c
ub
ic
s
pline
inter
pola
ti
on
to
c
onstr
uc
t
the
uppe
r
a
nd
l
ow
e
r
e
nv
e
lop
e,
m
ax(n)
=
(
,
)
an
d
m
in(n
)=
(
,
)
res
pec
ti
vely
.
3)
Find the e
nvel
op
e
m
ean,
(
)
=
[
(
)
+
(
)
]
/
2
.
4)
If
e
(
n)
sat
isfie
s the
IMF c
ondi
ti
on
s,
assig
n
(
)
=
(
)
f
or
i
th I
M
F a
nd
update
r
(n)
as
(
)
=
(
)
−
(
)
.
5)
If
r(n
)
rem
ai
ns
appr
ox
im
at
ely u
nc
ha
ng
e
d
t
he
n back
to st
ep
(
1) an
d
st
op.
6)
Af
te
r
obta
inin
g
an
IMFs,
(
)
,
s
ub
t
ract
(
)
from
t
he
sig
nal
r
(n)=
r(n)
-
(
)
and
back
to
ste
p
(
1)
i
f
r
(n)
is n
ot constant
or tren
d
t
he res
idu
al
si
g
nal,
(n)
.
Con
se
quently
, t
he
ori
gi
nal sig
nal,
r(n), is
rec
ov
e
re
d by the
f
ollow
i
ng equat
ion
:
(
)
=
∑
(
)
=
1
+
(
)
(2)
In
the
ne
xt
ste
p,
the
IMFs
si
gn
al
s
are
passi
ng
th
rou
gh
M
D
-
WPC
A
al
gorithm
to
get
the
denoised
ver
si
on
of
the
IMFs
s
ig
nals.
MD
-
WPCA
is
pro
po
se
d
by
A
m
ing
ha
far
i
[
18
]
to
rem
ov
e
noise
from
m
ul
tiv
ariat
e
no
isy
sig
nals
by
com
bin
ed
pr
i
ncipal
com
pone
nt
analy
sis
(P
CA
)
an
d
u
niv
a
riat
e
w
avelet
thres
holding.
Give
n
t
he
IMF
s
sig
nals
f
ro
m
the
pr
e
vious
st
ep,
c
i
(
n
)
,
a
nd
t
he
resid
ual
(
)
an
d
de
no
te
d
by
C(i)
w
her
e
C(i
)
i
s
the m
at
rix
for
m
o
f
c
i
(
n
)
,
C(i
)
∈
×
(
+
1
).
T
he
MD
-
WPCA al
gorit
hm
is o
utli
ned
in the f
ollow
i
ng ste
ps
:
1)
Apply
the
D
WT
at
a
le
vel
J
for
eac
h
c
olu
m
n
of
C
t
o
obta
in
t
he
(
J+1)
detai
l
co
eff
ic
ie
nts
m
at
rices
D
j
,
j
=
1,…,J
at
le
ve
l
1
to
J
an
d
the
approxim
at
e
coeffic
ie
nts
A
J
of
L+
1
channels,
w
her
e
D
j
∈
2
−
×
(
+
1
)
, j=
1,
.
.,J
m
at
rices an
d A
J
∈
2
−
×
(
+
1
)
m
at
rix.
2)
Fin
d
t
he
nois
e
co
va
riance
e
stim
at
e
∑
by
a
pp
ly
in
g
t
he
m
ini
m
u
m
covari
ance
determ
inant
(MCD)
to
D
1
(
∑
=M
CD(
D
1
))
.
The
n
fin
d
a
n
ort
ho
gonal
m
at
rix
V
by
com
pu
ti
ng
t
he
singular
value
deco
m
po
sit
i
on
(S
V
D
)
of
∑
(
∑
=
Λ
)
,
wh
e
re
=
diag
(
,
i
=
1
,
.
.
,
+
1
)
a
nd
,
i
=1,
.
.,L+
1
a
re
t
he
ei
genvalue
s
for
eac
h
c
hann
el
.
3)
Nex
t,
c
ha
ng
e
t
he
basis
u
sin
g
V
f
or
eac
h
det
ai
l
D
j
by
us
in
g
the
f
ollo
wing
m
ulti
plication,
E
j
=D
j
V,
j
=
1,..,J,
and ap
ply t
he
unive
rsal th
reshold
=
√
2
log
(
)
, i=1,..,
L+
1 for
t
he
i
th
co
lum
n
of
E
j
to
obta
in
̂
.
4)
Find the
PCA
of the m
at
rix
A
J
an
d
sel
ect
the
su
it
able
num
ber
L
J+1
of
us
e
ful
p
ri
ncipal c
ompone
nt.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
,
No.
2
,
A
pr
i
l 202
0
:
1352
-
1358
1354
5)
Chan
ge
the
ba
sis
of
̂
us
in
g
and
t
hen
m
ake
an
inv
e
rse
D
WT
to
ob
ta
in
the
en
han
ce
d
m
ul
ti
var
ia
te
sign
al
s
̃
(
)
.
6)
Apply PCA
to
̃
(
)
an
d retur
n
t
he m
os
t si
gn
ific
an
t pr
i
ncipal c
om
pone
nts.
The final
ste
p
,
the d
e
noise
d c
hao
ti
c si
gn
al
̃
(
)
is rec
ov
e
re
d
f
r
om
̃
(
)
accor
ding t
o:
̃
(
)
=
∑
̃
(
)
+
1
=
1
,
n=
1,
…,N
.
(3)
Figure
1. The
pr
opos
e
d
c
ha
otic denoisi
ng sy
stem
3.
PERFO
R
MANC
E E
V
ALU
ATIO
N
O
F
N
OISE
RED
U
CTIO
N MET
HOD
Let
us
def
ine
d
(
)
an
d
̃
(
)
as
the
cl
ean
an
d
de
noise
d
ch
aotic
sign
al
res
pecti
vely
.
I
n
or
der
t
o
com
par
e
bet
w
een
the
dif
fere
nt
noise
reducti
on
m
et
ho
ds,
the
re
are
di
ff
ere
nt
f
orm
ulas
that
are
use
d
as
a
perform
ance
evaluati
on
m
e
asur
em
ent
s
uc
h
as
sig
nal
to
no
ise
rati
o
(
S
NR)
[
9
,
10]
,
r
oo
t
m
ean
squa
re
er
ror
(RMSE)
[
6
,
8
,
10]
an
d
si
gn
al
to
noise
rat
io
gai
n
(SNRG)
[
8
,
10]
.
T
he
form
ulas
of
these
m
easur
es
a
re
def
i
ned as:
=
10
×
log
10
[
(
(
)
)
(
̃
(
)
−
(
)
)
]
(
4
)
=
√
1
2
∑
(
(
)
−
̃
(
)
)
2
=
1
(
5
)
=
−
(
6
)
wh
e
re
(
(
)
)
is
the
var
ia
nce
of
cl
ean
cha
otic
sign
al
,
(
̃
(
)
−
(
)
)
is
the
va
riance
of
th
e
error
betwee
n
cl
ean
and
de
noise
d
c
hao
ti
c
si
gn
al
th
at
is
equ
i
valent
to
the
noise
a
nd
S
NR
i
is
the
i
nput
sig
nal
to
no
ise
rati
o
that i
s
conside
red in t
he ran
ge (0
-
30) d
B wit
h
ste
p
a
bout
5 dB.
Othe
r
m
easur
e
that
help
us
to
know
w
heth
er
the
no
isy
chao
ti
c
sig
nal
i
s
per
fectl
y
de
no
ise
d
or
no
t
is
the
scal
e
depend
e
nt
Ly
a
punov
e
xpone
nt
(SDLE
)
[
19,
20]
.
T
he
al
gorithm
of
SDLE
is
s
umm
a
rized
i
n
al
gorithm
1.
Algori
th
m
1
:
Scal
e
de
pende
nt ly
apun
ov expon
e
nt (SDL
E)
Inp
ut: The
sig
na
l
x
(
n
).
Oup
ut: The
S
D
LE
Λ
(
)
.
1.
Create
th
e sui
table v
ectors
V
i
fro
m
a ti
me
serie
s sign
al
x
(
n
),
n
=1,.
.,
N
usin
g
=
[
(
)
,
(
+
)
,
.
.
,
(
+
(
−
1
)
)
]
,
=
1
,
.
.
,
wh
e
re
=
−
(
−
1
)
is
the
rec
onstr
ucted
vect
or
s
nu
m
ber,
m
is
the
em
bed
di
ng
dim
ension
a
nd
is t
he d
el
ay
ti
m
e.
2.
Check
whet
her
pairs
of
vecto
rs
(
F
i
,
F
i
)
satisfy
th
e
hi
gh
dim
ensio
nal
s
hell
in
equ
alit
y
,
≤
‖
−
‖
≤
+
Δ
,
k=
1,
2,
3,
..
wh
e
re
an
d
Δ
ar
e
the
rad
i
us
a
nd
the
wi
dth
of
the
s
hell
res
pe
ct
ively
that
are
ar
bitrar
ily
chosen
sm
al
l
distances
a
nd
‖
.
‖
is
the
norm
fu
nctio
n.
Als
o,
the
f
ollo
wing
co
ndit
ion
is
need
e
d:
|
−
|
≥
(
−
1
)
3.
The SDL
E in
term
o
f
tim
e t,
Λ
(
)
is give
n by
:
Λ
(
)
=
〈
‖
+
+
∆
−
+
+
∆
‖
−
‖
+
−
+
‖
〉
∆
,
w
he
re
∆
is
the
sam
pling
tim
e.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Chaoti
c sig
na
l
s d
e
noisi
ng usi
ng empiri
cal
m
od
e
d
ec
omp
os
i
ti
on
in
sp
ire
d b
y … (
Fa
dh
il
Sahib
H
as
an
1355
4.
SIMULATI
O
N RESULTS
In
this
sim
ulati
on
,
L
or
e
nz
[
5
]
,
Chen
[
21
]
and
Ros
sle
r
[
22
]
are
us
e
d
as
cha
otic
sys
tem
s
to
te
st
the
pro
po
se
d
m
et
ho
d.
T
he
chao
ti
c
syst
em
equ
at
ion
s
of
L
or
e
nz,
C
hen
a
nd
R
os
s
le
r
with
their
set
ti
ng
par
am
et
ers
are
descr
i
b
ed
in (
7),
(8)
a
nd (9
)
re
sp
ect
ively
:
Lor
e
nz sy
stem
[5]
:
⁄
=
(
−
)
⁄
=
(
−
)
−
⁄
=
−
,
=
10
,
=
28
,
=
8
/
3
.
(7)
Chen
syst
em
[
21
]
:
⁄
=
(
−
)
⁄
=
(
−
)
−
⁄
=
−
,a=3
5,
b=3 an
d
c=
28
(8)
Rossler syst
em
[
22
]
:
⁄
=
−
−
⁄
=
+
⁄
=
−
+
,
a=0
.38, b
=0.3 a
nd c=4
.82
(
9)
The
dif
fer
e
ntial
equ
at
ion
s
of
these
sy
stem
s
a
re
so
lve
d
us
i
ng
a
4
th
or
der
Ru
ng
e
-
Ku
tt
a
m
eth
od
with
a
ste
p
siz
e
of 0.0
01 sec
wi
th
50
000
num
ber
s
of sam
ples. Th
e
d
if
fer
e
nt s
i
m
ulati
on
scen
arios
a
re
de
picte
d belo
w.
Figure
2
an
d
Figure
3
s
how
SN
RG
an
d
R
MSE
te
sts
of
EMD
-
M
D
m
e
t
hod
res
pecti
ve
ly
to
re
m
ov
e
A
WGN
in
L
or
ez,
Che
n
an
d
Rossler
c
hao
ti
c
syst
e
m
.
The
perform
ance
evaluati
on
are
app
li
e
d
to
only
x(n
)
sign
al
of
these
cha
otic
syst
em
s.
The
ra
ng
e
of
S
NRi
is
(
0
-
30
dB)
with
s
te
p
5
dB
.
F
rom
these
tw
o
figure
s
,
it
can
be
noti
ced
that
SN
R
G
for
al
l
ty
pes
of
chao
ti
c
syst
em
s
has
at
le
ast
17
dB
gai
n
ov
er
unen
ha
nced
syst
e
m
.
Also
,
L
or
e
nz
s
yst
e
m
h
as the l
ow
est
RM
SE
va
lues c
om
par
ed wit
h
C
hen an
d
R
os
sle
r f
or
di
ff
ere
nt S
NRi.
Figure
2.
The
SN
RG
m
easure
s for dif
fer
e
nt
ty
pes of
chao
ti
c syst
em
s when EM
D
-
MD
al
gorithm
and AWG
N
ar
e u
se
d
Figure
3.
The
RM
SE m
easur
e f
or
dif
fer
e
nt t
ypes
of
chao
ti
c syst
em
s when EM
D
-
MD
al
gorithm
and AWG
N
ar
e u
se
d
In
t
his
sim
ulatio
n,
dif
fer
e
nt
t
ype
s
of
no
ise
are
use
d
t
o
te
st
the
abili
ty
of
t
he
propose
d
syst
em
to
rem
ov
e
no
ise
.
The
sel
ect
ed
a
dd
it
ive
noise
s
are
A
WGN,
F
act
or
y,
Ba
bble
,
Pink
an
d
H
Fc
hannel
noise
that
are
extracte
d
from
No
ise
x
-
92
da
ta
base
[
23
].
T
he
ra
ng
e
of
S
NRi
is
(0
-
30
dB)
with
ste
p
5
dB.
Fig
ure
4
an
d
Figure
5
sho
w
SN
RG
an
d
R
MSE
te
sts
of
EMD
-
M
D
m
eth
od
resp
ect
i
vely
to
rem
ov
e
nois
e
in
L
or
e
nz
chao
ti
c
syst
e
m
with
diff
e
ren
t
ty
pes
of
no
ise
(
A
WGN
,
Fact
or
y,
Ba
bble
,
Pin
k
a
nd
H
Fc
hannel)
.
It can
be
see
n
tha
t
fr
om
these
fig
ur
es
,
Fact
or
y
noise
has
the
w
or
st
SN
RG
perfor
m
ance
about
4
dB
and
the
w
or
st
RM
SE
pe
r
form
ance
com
par
ed
wit
h
oth
e
r
noise
s.
A
WGN
an
d
Pin
k
no
ise
app
r
oxim
at
e
l
y
hav
e
the
sam
e
per
form
ance.
Also
,
HF
c
hannel
no
ise
has
the
best
S
NR
G
perform
ance
ab
out
26
d
B
an
d
t
he
bes
t
RM
SE
perform
ance
com
par
ed wit
h othe
r n
oises.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
,
No.
2
,
A
pr
i
l 202
0
:
1352
-
1358
1356
Figure
4. The
SN
RG
m
easure
for
dif
fer
e
nt
no
ise
ty
pes
w
he
n
E
MD
-
M
D
alg
or
i
thm
an
d
L
or
e
nz
ch
a
otic
syst
e
m
are
us
e
d
Figure
5. The
RM
SE
m
easur
e f
or
dif
fer
e
nt
no
ise
ty
pes
w
he
n
E
MD
-
MD
alg
or
i
thm
is app
li
ed t
o
Lor
e
nz
c
ha
otic sy
stem
Table
1
sho
ws
the
com
par
iso
n
of
the
prpo
se
d
m
eth
od
with
different
de
noise
d
te
chn
i
ques.
The
par
am
et
er
of
sim
ulati
on
is
set
ti
ng
a
s:
the
scal
e
J
=
4,
wa
velet
fam
i
ly
=db
10
a
nd
t
he
th
res
ho
l
d
i
s
soft
thres
ho
l
d,
the
s
a
m
pling
tim
e=
0.01.
From
this
ta
ble
it
can
be
seen
that
the
pro
po
se
d
m
et
h
od
has
the
best
SN
R
and
RM
SE
va
lues
com
par
in
g
with
oth
e
r
m
et
ho
ds.
The
S
DLE
c
urve
f
or
the
noise
free
Lor
e
nz
sig
nal
and
denoise
d
Lo
re
nz
sig
nal
us
i
ng
EMD
-
MD
t
echn
i
qu
e
f
or
SNR
i
=
(
0,
5,
15,
20)
dB
is
sh
ow
n
in
Fi
gure
6.
Her
e
m
=5
an
d
=4.
From
these
fig
ures,
it
ca
n
no
ti
ce
t
hat
the
curve
of
denoi
sed
si
gnal
S
D
LE
is
go
a
way
fro
m
the
curve
of
cl
ean
sig
nal
wh
e
n
SN
R
i
is
decre
ased
or
noise
le
vel
is
increased.
T
her
e
fore,
the
SD
LE
m
ea
sur
e
giv
es
good
est
i
m
a
ti
on
ab
ou
t
the
le
vel
of
no
ise
in
the
n
oisy
chao
ti
c
sign
al
an
d
dist
i
nguish
no
ise
from
chaos sig
nal.
Table
1.
T
he
c
om
par
ison o
f
t
he pr
posed
m
eth
od
with
dif
fere
nt d
e
noise
d t
echn
i
qu
e
s
Metho
d
Ch
ao
tic sig
n
al
SNRi
[
d
B]
SNR [
d
B]
RMSE
W
av
elet so
f
t thres
h
o
ld
(
Han et a
l
.
2
0
0
7
)
[
5
]
Lorenz
14
2
3
.18
0
.3
840
Du
al wavelet
an
d
sp
atial corr
el
atio
n
(
Han
et
al
.
2
0
0
9
)
[
9
]
Lorenz
14
2
4
.60
3
9
0
.32
1
7
Ad
ap
tiv
e du
al
-
lif
tin
g
wavelet (
Y
.
Liu
and
X.
Liao 2
0
1
1
)
[
1
0
]
Lorenz
14
2
4
.66
3
1
0
.31
9
Prop
o
sed
m
e
th
o
d
Lorenz
14
2
5
.03
6
1
0
.28
0
9
I
m
p
rov
ed
E
EM
D
(
X.
W
ei
et al.
20
1
6
)
[
1
5]
Lorenz
15
2
4
.73
2
Prop
o
sed
m
e
th
o
d
Lorenz
15
2
5
.11
9
0
.27
4
0
I
m
p
rov
ed
E
MD
(
M.
W
an
g
et
al.
2
0
1
8
)
[
1
6
]
Ch
en
15
2
3
.37
2
6
0
.57
7
9
Prop
o
sed
m
e
th
o
d
Ch
en
15
2
5
.90
1
0
.26
8
0
Figure
6
.
Th
e
SD
LE
cur
ve
f
or the
noise
fr
ee
Lorenz
sig
nal
and d
e
noise
d L
or
e
nz
si
gn
al
for vari
ou
s
v
al
ues of S
NR
i
(
0,5,15,2
0)
dB. H
ere m
=5,
=4
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Chaoti
c sig
na
l
s d
e
noisi
ng usi
ng empiri
cal
m
od
e
d
ec
omp
os
i
ti
on
in
sp
ire
d b
y … (
Fa
dh
il
Sahib
H
as
an
1357
5.
CONCL
US
I
O
N
I
n
this
pa
per
,
the
pro
posed
Mult
ivariat
e
Denoisi
ng
(M
D)
dep
e
nds
on
w
a
velet
an
d
p
rinci
pal
c
om
po
ne
nt
(M
D
-
WPC)
t
hr
es
ho
l
ded
em
pirical
m
od
e
dec
om
po
sit
ion
(EM
D)
base
d
cha
ot
ic
sign
al
de
noisi
ng
is
inv
est
igate
d
a
nd
nam
ed
(EM
D
-
M
D)
.
In EM
D
-
M
D,
t
he
M
D
-
WPC is
sug
gested
to
t
hr
es
ho
l
d
the
intri
nsi
c
m
od
e
functi
ons
(I
M
Fs)
of
the
noi
sy
cha
otic
sig
nal.
T
he
pro
po
sed
syst
em
is
te
ste
d
f
or
di
fferent
ty
pe
s
of
chao
ti
c
sign
al
s,
Lo
re
nz
,
Che
n
an
d
R
os
sle
r
syst
em
,
and
dif
fer
e
nt
t
ypes
of
no
ise
,
A
WGN,
Fact
ory
,
Ba
b
ble,
Pi
nk
a
nd
HF
c
hannel.
Th
e
prp
os
e
d
m
eth
od
is
c
om
pari
ng
with
c
onve
ntion
al
c
ha
otic
de
no
isi
ng
te
chn
i
qu
e
s.
T
he
resu
lt
s
sh
ow
that
E
MD
-
MD
has
the
best
S
NR
G
an
d
RM
SE
values
.
Furth
erm
or
e,
scal
e
dep
e
ndent
L
ya
puno
v
expo
nen
t
(SDL
E)
is
us
e
d
to
d
i
sti
nguish
t
he
e
vel of
noise
c
om
par
ing
to
the
cl
ean chaoti
c s
ign
al
.
ACKN
OWLE
DGE
MENTS
This
w
ork
is
su
pp
or
te
d
by
the
F
acult
y
of
E
ng
i
neer
i
ng
/
Mustansiriy
ah
Un
i
ver
sit
y
(h
tt
ps
:/
/we
bm
a
il
.u
om
us
ta
ns
iri
ya
h.
ed
u.
i
q).
REFERE
NCE
S
[1]
A.
N.
Mili
ou,
I
.
P.
Antonia
des,
S.
G.
Stavri
nid
e
s,
A.
N.
Anag
nostopoulos
,
“
Secur
e
C
om
m
unic
at
ion
b
y
C
hao
tic
S
y
nchr
oni
zation
:
Robustness
under
nois
y
condit
ions
,”
Nonl
ine
ar
Anal
ysis:
R
eal
World
Appl
i
cat
i
ons
,
vol.
8
,
no
.
3
,
pp.
1003
-
1012
,
J
ul
y
2007.
[2]
Chong
Fu,
Gao
-
y
uan
Zh
ang,
Ma
i
Zhu
,
Zhe
Ch
en
,
and
W
ei
-
m
in
L
ei
,
“
A
New
Ch
a
os
-
Based
Color
I
m
age
Encr
y
pti
o
n
Scheme
with
an
Eff
icien
t
Subs
ti
tut
ion
Ke
y
s
tre
a
m
Gene
rat
ion
S
tra
t
eg
y
,”
Hinda
wi
Sec
uri
ty
and
Comm
unic
ati
o
n
Net
works
,
vo
l.
2
018
,
pp
.
1
-
13
,
2
018.
[3]
Fadhil
S.
Hasan
,
“
Speec
h
E
n
cr
y
pt
ion
using
Fi
xed
Point
Chaos
base
d
Strea
m
Ciphe
r
(
FPC
-
SC
)
,”
Eng
.
&
T
ec
h.
Journal
,
vo
l.
34
,
no.
11
,
pp.
2152
-
2166,
2016
.
[4]
M.
W
ang,
Z
.
Zh
ou,
Z.
Li
and
Y.
Ze
ng
,
“
An
a
dap
ti
ve
Algor
iht
m
for
Chaot
i
c
Signa
ls
Based
on
Im
prove
d
Empirica
l
Mode
Dec
om
positi
on
,”
Circu
it
s,
Syste
ms
and
Sig
nal
Proc
essing
,
vol.
38
,
no
.
6
, p
p
.
2471
-
2488
,
20
18.
[5]
M.
Han,
Y.
Li
u,
J.
Xi
,
and
W
.
Guo
,
“
Noise
s
moot
hing
fo
r
nonl
ine
ar
t
imes
series
using
W
ave
le
t
Soft
Thre
shold,
"
IEE
E
Signal P
ro
ce
ss
ing
Let
te
rs
,
vol.
14
,
no
.
1
,
pp
.
62
-
65
,
2007
.
[6]
J.
Gao,
H.
Sulta
n,
J.
Hu
and
W
e
n
-
wen
Tung,
“
D
enoi
sing
Nonlin
ea
r
Ti
m
e
Serie
s
b
y
Adapti
v
e
Filt
eri
ng
and
W
ave
l
et
Shrinkage
:
A Co
m
par
ison
,”
I
EEE
Signa
l
Proc
es
sing Le
tters
,
vol
.
17
,
no.
3
,
pp.
23
7
-
240,
2010
.
[7]
Deng
Ke,
Zha
ng
Lu
and
Luo
Mao
-
Kang
,
“
A
Denoising
Algorit
h
m
for
noisy
Cha
oti
c
Signa
ls
base
d
on
the
higher
orde
r
Thr
eshold F
unct
ion
i
n
W
av
el
e
t
-
Packet
,
”
Ch
ine
se
Phy
sics
Le
tt
ers
,
vo
l.
28
,
no.
2,
pp.
1
-
4,
2011
.
[8]
Xiao
Hong
Han
and
Xiao
Ming
Chang
,
“
Gene
tic
Algorit
hm
As
si
sted
W
ave
let
noi
se
Reduc
ti
on
Sc
heme
f
or
Chaot
i
c
Signal
s
,
”
Journa
l
of
Optimizati
on
Theory
and
App
li
cations
,
vo
l.
15
1
,
no
.
3
,
pp
.
646
-
653
,
2011
.
[9]
M.
Han
and
Y
.
Li
u
,
“
Noise
Reduc
ti
on
Meth
od
f
or
Chaotic
Signal
s
Based
On
Dual
-
W
ave
let
a
nd
Spa
ti
a
l
Corre
lation,”
Expert
Syst
ems
w
ith
Applications
,
v
ol.
36
,
no
.
6
,
pp
.
10060
-
10067,
2
009.
[10]
Y.
Li
u
and
X
.
Liao
,
“
Adapti
ve
C
haot
i
c
Noise
Re
duct
ion
Method
Based
o
n
Dual
-
Li
fti
ng
W
av
el
et
,
”
Ex
pert
S
yste
m
s
wit
h
App
licati
on
s
,
vol.
38
,
no.
3
,
pp.
1346
-
1355
,
2011.
[11]
E.
Ercel
eb
i
,
“
E
l
ec
tro
ca
rdiog
ram
Signal
s
de
-
noising
using
L
ift
in
g
-
b
ase
d
Discre
te
W
ave
let
T
ran
sf
orm
,”
Computer
s
in
B
iol
ogy
and
Me
dicine
,
vol
.
3
4,
no
. 6
,
pp
.
479
-
493
,
2004
.
[12]
W
en
-
Bo
W
ang,
Yun
-
y
u
Jing
,
Y
an
-
cha
o
Zha
o
,
L
ia
n
-
Hua
Zha
ng
and
Xian
g
-
Li
W
ang.
“
Chaotic
S
igna
l
D
enoi
sing
Based
o
n
Hier
ar
chi
c
al
Thre
shold
S
y
nchr
osque
ezed
W
ave
l
et
Trans
for
m
,
”
1st
Int
ernati
onal
Glob
al
on
R
ene
wabl
e
Ene
rgy
and
Dev
el
opment
(
IGRE
D)
,
vol.
100
,
pp
.
1
-
6
,
2017.
[13]
D.
Siwal,
V.
Suy
a
l,
A.
Prasad
,
S.
Manda
l
and
R.
Singh
,
“
A
new
Approac
h
of
Denoising
the
R
egul
ar
and
Ch
ao
ti
c
S
igna
ls
using
Empiric
al
Mode
Dec
om
positi
on:
Com
par
ison
a
nd
Appl
ic
at
ion
,”
Re
v
ie
w
of
Scien
ti
fic
Instrum
ent
s
,
vol.
84
,
no
. 7
,
pp
1
-
10,
2013.
[14]
M
.
W
an
g,
Z.
W
u,
Y.
Chen
a
nd
J.
Feng,
“
Denoising
of
Cha
oti
c
Signa
ls
Ba
sed
on
Ensemble
Empirical
Mo
de
Dec
om
positi
on
,”
IEE
E
Inte
rna
ti
onal
Confe
ren
ce
on
Signal
Proce
ss
ing,
Comm
u
nic
ati
ons
and
Computing
(
ICSPCC
)
,
pp.
14
-
17,
2014
.
[15]
X.
W
ei
,
R.
L
i
n,
S.
Li
u
and
C.
Zha
ng
,
“
I
m
prove
d
EE
MD
Denoising
Method
Based
on
Singula
r
Value
Dec
om
positi
on
f
or
the
Chao
ti
c
S
igna
l
,”
Hindawi
Publ
ishing
Corp
oration
Shock
a
nd
Vi
bration
,
vo
l.
2016
,
no
.
12
,
pp.
1
-
14
,
2016
.
[16]
X.
Li
and
W
.
Wa
ng
,
Stud
y
ing
on
Denoising
of
Chaot
ic
S
ignal
using
ICA
and
E
MD
,”
Geo
-
Infor
m
at
ic
s
in
Resourc
e
Mana
gement an
d
Sus
ta
ina
b
le
Ecos
y
stem
,
pp
.
564
-
572,
2013
.
[17]
W
.
W
ang,
X.
Zha
ng,
Y.
Chang,
X.
W
ang,
Z.
W
ang,
X.
Chen
an
d
L.
Zhe
ng
,
“
Denoising
of
Chaot
ic
Signal
u
sin
g
Inde
pende
n
t
Co
m
ponent
Anal
y
s
is
a
nd
Empirical
Mode
Dec
om
positi
on
W
it
h
Cir
c
ula
t
e
Tra
nsl
at
in
g
,”
Chin
.
Ph
ys.
B
.
vol.
25
,
no
.
1
,
pp
.
1
-
7
,
2016
.
[18]
M.
Am
ingha
far
i
,
N.
Che
ze
and
J.
Pogg
I,
“
Multi
var
iate
Deno
i
sing
using
W
ave
lets
and
Princ
i
pal
Com
ponent
A
naly
sis
,”
Com
p
utat
ional Statistic
s
&
Data
Anal
ysis
,
vol
.
50
,
no
.
9
,
pp
.
2381
-
239
8,
2006
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
,
No.
2
,
A
pr
i
l 202
0
:
1352
-
1358
1358
[19]
J.
Gao,
J
.
Hu,
W
.
W
.
Tun
g,
Y
.
H.
Cao
,
“
Disti
ngui
shing
cha
os
fro
m
noise
b
y
Sca
l
e
-
depe
nd
ent
L
y
a
punov
E
xponen
t
,”
Phy
s. Re
v. E St
a
t.
Non
li
n.
Sof
t
M
att
er
Phy
s
,
vol.
74,
no
.
6
,
pp
.
1
-
9
,
2006
.
[20]
J.
Gao
,
J.
Hu
,
W
.
W
.
Tung
and
E
,
Bla
sch
,
“
Multi
sca
le
Ana
l
y
s
is
of
Biol
ogical
Da
t
a
b
y
Sc
ale
-
dep
e
ndent
L
y
a
punov
E
xponent
,”
Fron
t
Ph
ysiol
ogy
,
vo
l
.
2
,
pp
.
1
-
13
,
20
11.
[21]
G.
Chen
,
The
Chen
s
y
stem
rev
isit
ed,
D
ynamics
of
Conti
nuous,
Discr
et
e
and
Im
pulsive
S
y
stems
,
Serie
s B
:
Appl
i
ca
t
ions &
Algo
ri
thms
20
,
pp.
691
-
696,
2013
.
[22]
P.
Gaspard
,
Ross
le
r
s
y
s
te
m
,
Encyc
lop
edi
a
o
f
No
nli
near
S
ci
en
ce
,
pp.
808
-
811
,
20
05.
[23]
A
.
Varga
and
H.
J.
M.
Stee
nek
en
,
“
As
ses
sm
ent
for
aut
om
at
i
c
spe
ec
h
re
cogni
t
ion
i
i:
Noisex
-
92:
A
dat
ab
ase
and
an
expe
riment
to
st
ud
y
th
e
ef
fect
o
f
addi
t
ive
nois
e
on
spee
ch
r
ec
og
nit
ion
s
y
s
te
m
s
”
Spee
ch
Comm
un
,
vol.
12
,
no
.
3
,
pp.
247
-
251
,
19
93.
BIOGR
AP
H
Y
O
F
AU
TH
ORS
Fadh
il
S.
Hasa
n
was
born
in
Baghda
d,
Ira
q
i
n
1978.
He
r
ec
e
ive
d
his
B
.
Sc.
d
egr
ee
in
E
lectr
i
c
al
Engi
ne
eri
ng
in
2
000
and
his
M.S
c.
degr
ee
in
Elec
troni
cs
and
Com
m
unic
at
ion
Eng
i
nee
ring
in
2003,
both
from
the
M
ustansiriy
ah
Uni
ve
rsit
y
,
I
raq
.
He
rec
e
ive
d
Ph.D
.
d
egr
ee
in
2013
in
El
e
ct
roni
cs
and
Com
m
unic
at
ion
Engi
ne
eri
ng
fro
m
the
Basra
h
Univer
sit
y
,
Ira
q
.
I
n
2005,
he
jo
ined
the
fa
cul
t
y
of
Engi
ne
eri
ng
a
t
the
Mus
ta
nsiri
yah
Univer
sit
y
i
n
Baghda
d.
Hi
s
rec
ent
rese
ar
ch
ac
t
ivi
t
ie
s
ar
e
W
ire
le
ss
Com
muni
cation
S
y
st
e
m
s,
Multi
ca
rri
er
Sy
st
em,
W
ave
l
et
base
d
OF
DM
,
MIM
O
Sy
stem,
Speec
h
Signa
l
Proce
ss
ing,
Cha
oti
c
Modul
at
ion
,
FP
GA
and
Xili
nx
S
y
s
te
m
Gene
ra
tor
base
d
Com
m
unic
at
ion
S
y
stem.
Now
he
has
bee
n
an
Assist.
Prof.
at
the
Mus
ta
nsiri
y
ah
Univer
sit
y
,
Ira
q.
Em
ai
l:
fad
el
_sah
ib@uom
ustansiriy
ah.edu.iq
Evaluation Warning : The document was created with Spire.PDF for Python.