I
A
E
S
I
n
t
e
r
n
at
io
n
al
Jou
r
n
al
of
A
r
t
if
ic
ia
l
I
n
t
e
ll
ig
e
n
c
e
(
I
J
-
A
I
)
V
ol
.
10
, N
o.
2
,
J
une
2021
, pp.
355
~
364
I
S
S
N
:
2252
-
8938
,
D
O
I
:
10.11591/
ij
a
i.
v
10
.i
2
.pp
355
-
364
355
Jou
r
n
al
h
om
e
page
:
ht
tp
:
//
ij
ai
.
ia
e
s
c
or
e
.c
om
B
i
gr
ad
i
e
n
t
n
e
u
r
al
n
e
t
w
or
k
-
b
ase
d
q
u
an
t
u
m
p
ar
t
i
c
l
e
swar
m
op
t
i
m
i
z
at
i
on
f
or
b
l
i
n
d
so
u
r
c
e
se
p
ar
at
i
on
H
u
s
s
e
in
M
. S
al
m
an
1
, A
li
K
ad
h
u
m
M
. A
l
-
Q
u
r
ab
at
2
, A
b
d
al
n
as
ir
R
iy
ad
h
F
in
j
an
3
1
College o
f Mater
ial Engine
ering, U
niversity
of Babylo
n,
Babylon, I
raq
2
Department of Compu
ter Science, College o
f Science for Women, Universi
ty of Babylon,
Babylon, Iraq
3
Supreme
Commissio
n for
Hajj a
nd Umr
ah, Ba
hgdad,
Iraq
A
r
t
ic
le
I
n
f
o
A
B
S
T
R
A
C
T
A
r
ti
c
le
h
is
to
r
y
:
R
e
c
e
iv
e
d
S
e
p
1
4
, 20
20
R
e
vi
s
e
d
N
ov
2
3
, 20
20
A
c
c
e
pt
e
d
M
a
r
2
1
, 20
21
An
independent
component
analysis
(ICA)
is
one
of
the
solutions
of
a
blind
source
separation
problem.
ICA
is
a
statistical
approach
that
depends
on
the
statistical
properties
of
the
mixed
signals.
The
purpose
of
the
ICA
m
e
thod
is
to
demix
the
mixed
source
signals
(observation
signals)
and
rcoverin
g
those
signals.
The
abbreviation
of
the
problem
is
that
the
ICA
needs
for
opti
mizing
by
using
one
of
the
optimization
approaches
as
swarm
intelligent,
neural
neworks,
and
genetic
algorithms.
This
paper
presents
a
hybrid
met
hod
to
optimize
the
ICA
method
by
using
the
quantum
particle
swarm
optim
ization
method
(QPSO)
to
optimize
the
B
igradient
neural
network
meth
od
that
applies
to
separate
mixed
signals
and
recover
sources
signals
.
The
re
sults
of
an
implement
this
wo
rk
prove
that
this
method
gave
good
results
comparing
with
other
methods
such
as
the
B
igradient
neural
network
and
the
QPSO
method, base
d on sever
al evalua
tion measure
s as signal
-
to
-
noise ratio, signal
-
to
-
distortion ratio, absolute value correlation coef
fi
cient, an
d the com
p
utation
time.
K
e
y
w
o
r
d
s
:
B
ig
r
a
di
e
nt
ne
ur
a
l
ne
twor
k
B
S
S
I
C
A
Q
P
S
O
This is an
open
acce
ss artic
le unde
r the
CC BY
-
SA
license.
C
or
r
e
s
pon
di
n
g A
u
th
or
:
A
li
K
a
dhum
M
. A
l
-
Q
ur
a
ba
t
D
e
p
ar
t
m
e
nt
of
C
om
put
e
r
S
c
ie
nc
e
U
ni
ve
r
s
it
y of
B
a
byl
on
B
a
byl
on, I
r
a
q
E
m
a
il
:
a
li
k.m
.a
lq
ur
a
ba
t@uoba
byl
on.e
du.i
q
1.
I
N
T
R
O
D
U
C
T
I
O
N
B
li
nd
s
our
c
e
s
e
pa
r
a
ti
on
(
B
S
S
)
is
a
po
w
e
r
f
ul
s
ig
na
l
pr
oc
e
s
s
in
g
m
e
th
od
pr
opos
e
d
in
th
e
la
te
1980s
.
A
s
th
e
pr
oduc
t
of
a
r
ti
f
ic
ia
l
ne
ur
a
l
ne
twor
ks
,
s
ta
ti
s
ti
c
a
l
s
ig
na
l
p
r
oc
e
s
s
in
g,
a
nd
in
f
or
m
a
ti
on
th
e
or
y.
A
f
te
r
th
e
n
B
S
S
be
c
om
e
s
a
n i
m
por
ta
nt
t
opi
c
i
n r
e
s
e
a
r
c
h
a
nd de
ve
lo
pm
e
nt
i
n m
a
ny a
r
e
a
s
[
1]
.
T
he
m
a
in
ta
s
k
of
th
e
B
S
S
is
e
xt
r
a
c
ti
ng
a
nd
r
e
c
ov
e
r
in
g
th
e
unde
r
ly
in
g
s
our
c
e
s
ig
na
ls
f
r
om
m
ul
ti
va
r
ia
bl
e
s
ta
ti
s
ti
c
a
l
da
ta
(
obs
e
r
va
ti
on
s
ig
na
ls
)
.
T
he
ob
s
e
r
va
ti
on
s
ig
na
ls
c
a
n
be
m
a
ni
pul
a
te
d
a
s
th
e
m
ix
in
g
of
s
our
c
e
s
ig
na
ls
,
th
a
t
is
,
th
e
obs
e
r
ve
d
m
ix
e
d
s
ig
na
l
is
a
s
e
r
ie
s
of
s
e
ns
or
out
put
s
.
T
he
m
ix
in
g
pr
oc
e
s
s
is
done
un
de
r
s
om
e
c
ondi
ti
ons
a
s
t
he
w
e
ll
-
c
ondi
ti
on of
t
he
m
ix
in
g m
a
tr
ix
a
nd t
he
g
a
us
s
ia
ni
ty
of
t
he
s
our
c
e
s
ig
na
l
s
, a
s
in
th
e
c
oc
kt
a
il
pa
r
ty
pr
obl
e
m
,
th
a
t
r
e
pr
e
s
e
nt
s
th
e
ty
pi
c
a
l
e
xa
m
pl
e
of
th
e
B
S
S
[
1
]
-
[
4]
.
F
ig
ur
e
1
s
ke
tc
hi
ng
th
e
c
oc
kt
a
il
-
pa
r
ty
pr
obl
e
m
.
I
n
th
is
pa
pe
r
,
w
e
pr
opos
e
d
a
ne
w
hybr
id
m
e
th
od
of
th
e
I
C
A
b
a
s
e
d
on
th
e
qua
nt
um
pa
r
ti
c
le
s
w
a
r
m
opt
im
iz
a
ti
on
(
Q
P
S
O
)
a
nd
B
ig
r
a
di
e
nt
ne
ur
a
l
ne
twor
k
m
e
th
od.
T
he
pr
opos
e
d
m
e
th
od
in
c
lu
d
e
s
e
nha
nc
in
g
th
e
pe
r
f
or
m
a
nc
e
of
th
e
B
ig
r
a
di
e
nt
-
ba
s
e
d
I
C
A
m
e
th
od
by
us
in
g
th
e
Q
P
S
O
opt
im
iz
a
t
io
n
m
e
th
od.
T
he
B
ig
r
a
di
e
nt
ne
ur
a
l
ne
twor
k
m
e
th
od
c
ha
r
a
c
te
r
iz
e
s
w
it
h
th
e
s
pe
e
d
c
onve
r
ge
nc
e
but
not
a
c
c
ur
a
te
in
th
e
s
e
pa
r
a
ti
on
pr
oc
e
s
s
.
B
y us
in
g t
he
Q
P
S
O
m
e
th
od t
o opti
m
iz
e
t
he
I
C
A
m
e
th
od ba
s
e
d on the
B
ig
r
a
di
e
nt
ne
ur
a
l
ne
twor
k m
e
th
od. T
he
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
.
10
, N
o.
2, J
une
20
21
:
355
–
364
356
Q
P
S
O
is
us
in
g
th
e
B
ig
r
a
di
e
nt
f
unc
ti
on
a
s
a
n
obj
e
c
ti
ve
f
unc
ti
on
w
it
h
two
le
a
r
ni
ng
pa
r
a
m
e
te
r
s
,
to
a
dj
u
s
t
th
e
c
onve
r
ge
nc
e
of
t
he
I
C
A
a
lg
or
it
hm
a
nd t
o a
c
c
ur
a
te
of
t
he
s
e
pa
r
a
ti
on pr
oc
e
s
s
.
T
he
r
e
s
ul
ts
of
th
e
pr
opos
e
d
m
e
th
od
c
om
pa
r
e
d
w
it
h
ot
he
r
m
e
th
ods
a
s
th
e
F
a
s
tI
C
A
[
5
]
,
[
6]
,
a
nd
“
th
e
I
C
A
ba
s
e
d
on
qua
nt
um
p
a
r
ti
c
le
s
w
a
r
m
opt
im
iz
a
ti
on”
a
s
in
[
7]
.
I
n
a
ddi
ti
on,
e
va
lu
a
te
th
e
pr
opos
e
d
m
e
th
od
by
num
be
r
of
m
e
a
s
ur
e
m
e
nt
s
a
s
th
e
s
ig
na
l
-
to
-
noi
s
e
r
a
ti
o
(
S
N
R
)
[
8]
,
th
e
s
ig
na
l
-
to
-
di
s
to
r
ti
on
r
a
ti
o
(
S
D
R
)
[
9
]
,
th
e
a
bs
ol
ut
e
va
lu
e
of
c
or
r
e
la
ti
on c
oe
f
f
i
c
ie
nt
(
A
V
C
C
)
[
10]
, a
nd t
he
c
om
put
a
ti
on t
im
e
.
F
ig
ur
e
1
.
C
oc
kt
a
il
pa
r
ty
pr
obl
e
m
T
he
r
e
s
t
of
th
is
pa
pe
r
is
or
ga
ni
z
e
d
a
s
:
s
e
c
ti
on
2
de
s
c
r
ib
e
d
I
C
A
in
de
ta
il
.
T
he
Q
P
S
O
a
nd
B
ig
r
a
di
e
nt
ne
ur
a
l
ne
twor
ks
a
r
e
in
tr
oduc
e
d
in
s
e
c
ti
ons
3
a
nd
4
r
e
s
p
e
c
ti
ve
ly
.
S
e
c
ti
on
5
s
ta
te
s
th
e
r
e
la
te
d
w
or
k
s
.
S
e
c
ti
on
6
de
s
c
r
ib
e
d
th
e
pr
opos
e
d
m
e
th
od.
T
he
e
xpe
r
im
e
nt
a
l
r
e
s
ul
ts
d
e
s
c
r
ib
e
d
in
s
e
c
ti
on
7.
T
he
c
on
c
lu
s
io
n
c
a
m
e
in
s
e
c
ti
on 8.
1.1.
I
n
d
e
p
e
n
d
e
n
t
c
om
p
on
e
n
t
an
al
ys
is
(
I
C
A
)
I
t
is
a
s
ta
ti
s
ti
c
a
l
c
om
put
a
ti
on me
th
od a
nd ba
s
e
on t
he
s
ta
ti
s
ti
c
a
l
pr
ope
r
ti
e
s
of
t
he
obs
e
r
va
ti
on s
ig
na
ls
.
M
a
in
ta
s
k
of
I
C
A
is
r
e
c
ove
r
in
g
a
nd
f
in
di
ng
th
e
or
ig
in
a
l
s
our
c
e
s
f
r
om
th
e
obs
e
r
va
ti
on
s
ig
na
l.
T
h
e
m
a
th
e
m
a
ti
c
a
l
r
e
pr
e
s
e
nt
a
ti
on of
t
he
obs
e
r
va
ti
on
s
ig
na
ls
c
a
n don
e
a
s
i
n t
he
(
1)
:
(
)
=
(
)
(
1)
W
he
r
e
(
)
=
[
1
,
2
,
…
,
]
r
e
pr
e
s
e
nt
s
×
1
obs
e
r
va
ti
ons
ve
c
to
r
,
(
)
=
[
1
,
2
,
…
,
]
is
a
×
1
unknown
s
our
c
e
v
e
c
to
r
a
nd
z
e
r
o
-
m
e
a
n
non
-
G
a
us
s
ia
n
e
le
m
e
nt
s
,
a
nd
is
a
n
unknown
×
non
-
s
in
gul
a
r
m
ix
in
g m
a
tr
ix
. A
bove
m
ode
l
is
t
he
ge
ne
r
a
l
li
ne
a
r
m
ode
l
of
t
he
I
C
A
m
e
th
ods
[
2
]
,
[
3
]
,
[
7]
.
I
n
th
e
li
ne
a
r
s
c
h
e
m
a
,
th
e
pr
oc
e
s
s
c
on
s
is
ts
of
f
in
di
ng
th
e
in
ve
r
s
e
of
th
e
m
ix
in
g
m
a
tr
ix
.
A
ls
o,
to
s
ol
ve
th
e
obs
e
r
va
ti
on
m
ode
l
–
a
s
in
(
1)
,
to
r
e
c
ove
r
th
e
s
our
c
e
s
a
nd
s
e
pa
r
a
te
th
e
m
,
m
us
t
a
s
s
um
e
f
ound
m
a
tr
ix
,
so
-
c
a
ll
e
d a
s
e
pa
r
a
ti
on ma
tr
ix
-
to
be
i
n ne
w
f
or
m
ul
a
i
n m
a
th
e
m
a
ti
c
a
l
r
e
pr
e
s
e
nt
a
ti
on a
s
:
y
(
)
=
Wx
(
)
≈
s
(
)
(
2)
W
he
r
e
(
)
=
[
1
,
2
,
…
,
]
r
e
pr
e
s
e
nt
×
1
r
e
c
ove
r
e
d
s
ig
na
l,
a
nd
is
×
s
e
pa
r
a
te
d
m
a
tr
ix
.
T
he
m
ode
l
in
(
2)
r
e
pr
e
s
e
nt
th
e
s
e
pa
r
a
ti
on
pr
oc
e
s
s
in
th
e
I
C
A
m
e
th
ods
[
1]
.
T
he
r
e
a
r
e
s
om
e
pr
e
pr
oc
e
s
s
e
s
m
us
t
do on the
obs
e
r
va
ti
on s
ig
na
ls
a
s
t
he
c
e
nt
e
r
in
g a
nd w
hi
te
ni
ng [
1
]
,
[
2
]
,
[
5]
.
A
ny
a
lg
or
it
hm
of
th
e
I
C
A
m
e
th
ods
de
p
e
nds
on
two
e
xt
r
e
m
e
ly
de
pe
nde
nc
e
a
xi
om
s
a
r
e
th
e
opt
im
iz
a
ti
on
m
e
th
od
a
nd
th
e
obj
e
c
ti
ve
(
c
ont
r
a
s
t)
f
unc
ti
on.
T
he
opt
im
iz
a
ti
on
m
e
th
od
a
f
f
e
c
ts
w
it
h
th
e
a
lg
or
it
hm
ic
pr
ope
r
ti
e
s
of
th
e
I
C
A
m
e
th
od;
a
nd
th
e
obj
e
c
ti
ve
(
c
ont
r
a
s
t)
f
unc
ti
on
a
f
f
e
c
t
w
it
h
th
e
s
ta
ti
s
ti
c
a
l
pr
ope
r
ti
e
s
of
th
e
I
C
A
m
e
th
od.
I
n
a
ddi
ti
on,
th
e
pow
e
r
f
ul
of
th
e
I
C
A
de
pe
nds
on
th
e
c
hoos
in
g
of
th
e
obj
e
c
ti
ve
f
unc
ti
on, whic
h m
us
t
be
s
im
pl
e
a
nd f
a
s
t
c
om
put
a
ti
on [
3]
.
F
ir
s
tl
y
,
th
e
I
C
A
us
e
d
s
om
e
of
th
e
c
la
s
s
ic
a
l
ne
ur
a
l
ne
twor
k
a
s
a
n
opt
im
iz
a
ti
on
m
e
th
od
f
or
e
xa
m
pl
e
,
gr
a
di
e
nt
m
e
th
ods
,
N
e
w
to
n
-
li
ke
m
e
th
ods
,
a
nd
ot
he
r
s
[
1
]
,
[
11
]
,
[
12]
th
e
n
th
e
ge
ne
ti
c
a
lg
or
it
hm
s
a
nd
e
vol
ut
io
na
r
y
a
lg
or
it
hm
s
a
s
th
e
s
w
a
r
m
in
te
ll
ig
e
nc
e
opt
im
iz
a
ti
on
m
e
th
ods
[
3
]
,
[
13]
.
L
in
e
a
r
ly
,
F
a
s
tI
C
A
m
e
th
od
[5
]
,
[
6]
is
a
m
os
t
popula
r
li
ne
a
r
I
C
A
m
e
th
ods
w
hi
c
h
d
e
pe
nds
o
n
th
e
f
ix
e
d
-
poi
nt
it
e
r
a
ti
on
m
e
th
od
a
ls
o
c
a
n
b
e
c
ons
id
e
r
a
s
a
n
a
ppr
oxi
m
a
ti
ve
N
e
w
to
n
it
e
r
a
ti
on
m
e
th
od,
a
ls
o
th
e
r
e
a
r
e
num
be
r
of
a
da
pt
e
d
a
nd
pr
opos
e
d
m
e
th
ods
t
ha
t
de
pe
nds
on numb
e
r
of
l
in
e
a
r
f
unc
ti
ons
[
1]
.
I
n
or
de
r
,
th
e
s
e
c
ond
pa
r
t
of
th
e
I
C
A
m
e
th
ods
-
obj
e
c
ti
ve
f
unc
ti
o
n
-
done
by
us
in
g
one
of
th
e
g
a
us
s
ia
n
m
e
a
s
ur
in
g
f
unc
ti
ons
a
s
K
ur
to
s
i
s
f
unc
ti
on,
N
e
ge
nt
r
opy
f
unc
ti
on,
m
ut
ua
l
in
f
or
m
a
ti
on
(
M
I
)
f
unc
ti
on
a
nd
m
a
xi
m
um
li
ke
li
hood
(
M
L
)
f
unc
ti
on
[
1]
.
M
os
t
r
e
s
e
a
r
c
he
r
s
a
da
pt
th
e
kur
to
s
is
f
unc
ti
on
a
s
in
(
3)
a
nd
th
e
n
e
ge
nt
r
opy ba
s
e
d on kur
to
s
is
f
unc
ti
on a
s
i
n (
4)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
B
ig
r
adi
e
nt
ne
ur
al
ne
tw
or
k
-
bas
e
d quantum par
ti
c
l
e
s
w
a
r
m
opt
i
m
iz
at
io
n f
or
…
(
H
us
s
e
in
M
. Sal
m
an
)
357
=
(
4
)
−
3
[
(
4
)
]
2
(
3)
(
)
≈
1
12
3
(
)
2
+
1
48
4
(
)
2
(
4)
W
he
r
e
r
e
pr
e
s
e
nt
s
th
e
-
th
c
um
ul
a
nt
,
is
a
n
e
xp
e
c
ta
ti
on
ope
r
a
t
io
n,
a
nd
is
da
ta
ve
c
to
r
of
th
e
s
ig
na
ls
[1
]
,
[
7
]
,
[
14
]
,
[
15
]
.
1.2.
Q
u
an
t
u
m
p
ar
t
ic
le
s
w
ar
m
op
t
im
iz
at
io
n
(
Q
P
S
O
)
Q
P
S
O
is
one
of
m
o
s
t
popula
r
m
e
ta
-
he
ur
is
ti
c
opt
im
iz
a
ti
on
m
e
t
hods
ba
s
e
d
on
th
e
qua
nt
um
pr
in
c
ip
le
of
th
e
a
ni
m
a
l
na
tu
r
e
a
s
f
is
he
s
a
nd
bi
r
ds
.
T
o
f
in
d
th
e
e
f
f
ic
ie
nt
s
ol
ut
io
n,
th
e
m
e
ta
-
he
ur
is
ti
c
a
lg
or
it
hm
s
us
e
th
e
le
a
r
ni
ng a
lg
or
it
hm
s
f
or
a
n i
nf
or
m
a
ti
on s
tr
uc
tu
r
in
g [
14
]
,
[
16]
.
T
hi
s
m
e
th
od
a
s
s
um
e
s
th
a
t
e
a
c
h
pa
r
ti
c
le
lo
ok
s
in
th
e
s
e
a
r
c
h
a
r
e
a
w
it
h
a
δ
pot
e
nt
ia
l
on
a
c
e
r
ta
in
di
m
e
ns
io
n,
ne
a
r
by
th
e
poi
nt
pi
j.
G
e
ne
r
a
ll
y,
th
e
pa
r
ti
c
le
s
w
a
r
m
c
a
n
be
r
e
pr
e
s
e
nt
e
d
in
a
c
e
r
ta
in
di
m
e
ns
io
na
l
a
r
e
a
,
w
it
h
a
c
e
nt
e
r
p.
T
o
s
ol
ve
th
e
di
m
e
ns
io
na
l
δ
pot
e
nt
ia
l,
th
e
S
c
hr
ödi
nge
r
f
or
m
ul
a
us
e
d
f
or
th
is
pur
pos
e
.
B
a
s
e
d on thi
s
f
or
m
ul
a
, t
he
pdf
Q
a
nd t
he
di
s
tr
ib
ut
io
n f
unc
ti
on F
c
a
n be
de
f
in
e
d a
s
i
n (
5)
a
nd (
6)
r
e
s
pe
c
ti
ve
ly
.
(
(
+
1
)
)
=
1
(
)
−
2
|
(
)
−
(
+
1
)
|
/
(
)
(
5)
(
(
+
1
)
)
=
−
2
|
(
)
−
(
+
1
)
|
/
(
)
(
6)
W
he
r
e
(
)
c
a
lc
ul
a
te
d
u
s
in
g
M
ont
e
C
a
r
lo
e
s
ti
m
a
ti
on
a
ppr
oa
c
h, w
h
e
r
e
de
not
e
a
s
ta
nda
r
d
d
e
vi
a
ti
on,
a
ls
o t
he
pa
r
ti
c
le
pos
it
io
n c
a
n b
e
c
a
lc
ul
a
te
d
a
s
i
n (
7)
.
(
+
1
)
=
(
)
±
(
)
2
(
1
⁄
)
,
=
(
0
,
1
)
(
7)
F
or
e
va
lu
a
ti
ng t
he
(
)
, t
he
a
lg
or
it
hm
us
e
s
t
he
m
e
a
n be
s
t
pos
it
io
n
m
, w
hi
c
h i
s
a
gl
oba
l
poi
nt
o
f
t
he
popula
ti
on, i
s
pbe
s
t
of
a
ll
pa
r
ti
c
le
s
, a
s
gi
ve
n i
n t
he
(
8)
.
(
)
=
(
1
(
)
,
2
(
)
,
.
.
.
,
(
)
)
=
(
1
∑
,
1
(
)
,
1
∑
,
2
(
)
,
.
.
.
,
1
∑
,
(
)
=
1
=
1
=
1
)
(
8)
M
de
not
e
th
e
s
i
z
e
of
popula
ti
on a
nd P
i
r
e
pr
e
s
e
nt
t
he
pbe
s
t
of
t
h
e
pa
r
ti
c
le
i
. T
he
(
)
is
gi
ve
n i
n (
9)
,
(
)
=
2
∗
|
(
)
−
(
)
|
(
9)
A
ls
o, t
he
pos
it
io
n of
t
he
pa
r
ti
c
le
i
i
s
gi
ve
n i
n (
10)
(
+
1
)
=
(
)
±
∗
|
(
)
−
(
)
|
∗
(
1
⁄
)
(
10)
W
he
r
e
r
e
pr
e
s
e
nt
s
th
e
c
ont
r
a
c
ti
on
–
e
xpa
ns
io
n
f
a
c
to
r
,
is
th
e
c
o
nt
r
ol
pa
r
a
m
e
te
r
of
th
e
a
lg
or
it
h
m
c
onve
r
ge
nt
[
17
]
,
[
18]
.
1.3.
B
ig
r
ad
ie
n
t
n
e
u
r
al
n
e
t
w
or
k
al
gor
it
h
m
T
he
r
e
a
r
e
va
r
io
us
m
e
th
ods
d
e
pe
nd
on
th
e
ne
ur
a
l
ne
twor
ks
to
s
ol
ve
th
e
I
C
A
a
lg
or
it
hm
in
bot
h
li
ne
a
r
m
ix
tu
r
e
a
nd
nonl
in
e
a
r
m
ix
tu
r
e
.
N
e
ur
a
l
P
C
A
a
nd
I
C
A
a
r
c
hi
te
c
tu
r
e
s
a
nd
le
a
r
ni
ng
a
lg
or
it
hm
s
c
a
n
be
di
vi
de
d
in
to
two
m
a
in
gr
oups
:
hi
e
r
a
r
c
hi
c
a
ppr
oa
c
h
e
s
,
w
hi
c
h
e
s
ti
m
a
t
e
th
e
pr
in
c
ip
a
l
c
om
pone
nt
s
or
e
ig
e
nve
c
to
r
s
th
e
m
s
e
lv
e
s
;
a
nd
s
ym
m
e
tr
ic
s
ubs
p
a
c
e
ty
pe
a
ppr
oa
c
he
s
,
w
hi
c
h
e
s
ti
m
a
te
th
e
I
C
A
s
ubs
pa
c
e
onl
y
[
19
]
,
[
20]
.
T
he
B
ig
r
a
di
e
nt
a
lg
or
it
hm
i
s
l
e
a
r
ni
ng a
lg
or
it
hm
f
or
s
e
pa
r
a
ti
ng ma
tr
ix
W
a
f
te
r
pr
e
-
w
hi
te
ni
ng [
21
]
,
[
22
]
, a
s
:
+
=
+
µ
(
)
−
(
−
)
(
11)
I
n
(
11)
,
th
e
le
a
r
ni
ng
pa
r
a
m
e
te
r
µk
de
c
r
e
a
s
e
d
li
ne
a
r
ly
f
r
om
0.01
to
0.00001
w
it
h
th
e
num
be
r
o
f
it
e
r
a
ti
on
s
te
ps
k,
a
nd
γ
k
is
a
not
he
r
pos
it
iv
e
le
a
r
ni
ng
pa
r
a
m
e
te
r
,
us
ua
ll
y
a
bout
0.5
[
23]
.
T
he
f
ir
s
t
upda
te
te
r
m
µ
(
)
is
e
s
s
e
nt
ia
ll
y
a
nonl
in
e
a
r
H
e
bbi
a
n
te
r
m
,
a
nd
th
e
s
e
c
ond
te
r
m
(
−
)
ke
e
ps
th
e
w
e
ig
ht
m
a
tr
ix
r
oughly
or
th
ogona
l.
O
ne
of
it
s
b
e
s
t
f
e
a
tu
r
e
s
i
s
f
le
xi
bi
li
ty
.
T
he
(
11)
c
a
n
be
a
ppl
ie
d
w
it
h
s
li
ght
ly
di
f
f
e
r
e
nt
f
or
m
s
a
nd
c
hoi
c
e
s
to
s
e
pa
r
a
ti
n
g
e
it
he
r
s
ub
-
G
a
us
s
ia
n
or
s
upe
r
-
G
a
us
s
ia
n
s
our
c
e
s
.
I
t
is
a
ls
o
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
.
10
, N
o.
2, J
une
20
21
:
355
–
364
358
e
a
s
y
to
m
odi
f
y
th
e
(
11)
s
o
th
a
t
th
e
w
e
ig
ht
ve
c
to
r
s
of
th
e
ne
ur
ons
a
r
e
c
om
put
e
d
s
e
que
nt
ia
ll
y
in
a
hi
e
r
a
r
c
hi
c
or
de
r
[
12]
.
2.
L
I
T
E
R
A
T
U
R
E
R
E
V
I
E
W
I
n
th
is
s
e
c
ti
on,
w
e
w
il
l
r
e
vi
e
w
s
om
e
r
e
c
e
nt
ly
r
e
la
te
d
w
o
r
k
s
a
bout
th
e
bl
in
d
s
our
c
e
s
e
pa
r
a
ti
on
pr
obl
e
m
a
nd t
he
ne
ur
a
l
ne
twor
ks
a
lg
or
it
hm
s
.
P
e
hl
e
va
n
e
t
al
.
[
23]
in
tr
oduc
e
d
a
m
e
th
od
f
or
th
e
bl
in
d
s
ig
na
l
s
e
pa
r
a
ti
on
pr
obl
e
m
to
s
ol
ve
th
e
ne
nne
ga
ti
ve
s
im
il
a
r
ty
m
a
tc
hi
ng
pobl
e
m
de
pe
ndi
ng
on
de
e
p
le
a
r
ni
ng
of
ne
ur
ons
in
t
he
n
e
ur
a
l
ne
twor
ks
,
th
r
ough
de
s
ig
ni
ng
th
r
e
e
-
la
ye
r
s
ne
ur
a
l
ne
twor
k
und
e
r
f
e
e
df
or
w
a
r
d
a
r
c
hi
te
c
tu
r
e
.
I
n
th
e
s
e
c
ond
la
ye
r
,
a
ll
ne
ur
ons
w
e
r
e
l
e
a
r
ne
d
w
it
h
d
e
e
p
le
a
r
ni
ng
us
in
g
th
e
ba
c
kpr
op
a
ga
ti
on.
T
he
la
s
t
la
ye
r
r
e
c
ov
e
r
e
d
th
e
hi
dde
n
s
our
c
e
s
.
I
n
th
is
w
or
k,
obj
e
c
ti
ve
f
unc
ti
on
w
a
s
us
e
d
to
de
r
iv
e
th
e
le
a
r
ni
ng
r
ul
e
s
a
nd
th
e
a
r
c
hi
te
c
tu
r
e
of
th
e
de
s
ig
ne
d
ne
twor
k.
T
he
a
ut
hor
s
c
om
pa
r
e
d
th
os
e
w
or
k
w
it
h
f
iv
e
I
C
A
m
e
th
ods
a
r
e
pr
oj
e
c
te
d
gr
a
di
e
nt
de
s
e
nt
a
lg
or
it
hm
,
F
a
s
tI
C
A
,
I
nf
om
a
x
I
C
A
,
L
in
s
ke
r
’
N
e
twor
k,
a
nd
N
onne
ga
ti
ve
P
C
A
.
S
o
i
m
pl
e
m
e
nt
a
ll
th
e
s
e
m
e
th
ods
w
it
h na
tu
r
a
l
im
a
ge
s
.
S
a
lm
a
n
a
nd
A
bba
s
[
7]
in
tr
oduc
e
d
ne
w
m
e
th
od
to
opt
im
iz
e
th
e
I
C
A
m
e
th
od
by
us
in
g
qua
nt
um
pa
r
ti
c
le
s
w
a
r
m
opt
im
iz
a
ti
on
m
e
th
od.
T
hi
s
m
e
th
od
us
e
d a
N
e
ge
nt
r
opy
f
unc
ti
on
a
s
a
n
obj
e
c
ti
ve
f
unc
ti
on
in
th
e
I
C
A
. T
he
m
e
th
o
d yi
e
ld
s
good r
e
s
ul
ts
i
n t
he
s
e
p
a
r
a
ti
on pr
oc
e
s
s
,
but
i
t
s
om
e
th
in
g s
lo
w
c
om
pa
r
e
d w
it
h s
ta
nda
r
d
F
a
s
tI
C
A
m
e
th
od.
T
he
a
ut
hor
s
e
va
lu
a
te
d
th
e
pe
r
f
or
m
a
nc
e
of
th
is
m
e
th
od
us
in
g
a
num
be
r
of
m
e
tr
ic
s
a
s
s
ig
n
a
l
-
to
-
noi
s
e
r
a
ti
o i
nde
x a
nd s
ig
na
l
-
to
-
di
s
to
r
ti
on r
a
ti
o i
nde
x
.
I
s
om
ur
a
a
nd
T
oyoi
z
um
i
[
24]
p
r
opos
e
d
a
m
e
th
od
in
ne
ur
a
l
ne
t
w
or
ks
de
pe
nds
on
e
r
r
or
ga
te
d
he
bbi
a
n
r
ul
e
(
E
G
H
R
)
t
o
e
xt
r
a
c
t
th
e
m
ix
e
d s
ounds
i
n t
he
B
S
S
. T
he
E
G
H
R
l
e
a
r
ni
ng r
ul
e
be
ne
f
it
s
i
n
r
e
duc
in
g t
he
s
e
ns
or
in
put
s
e
s
pe
c
ia
ll
y
in
r
e
c
or
di
ng
a
ni
m
a
l
s
ound
s
.
I
n
a
dd
it
io
n,
th
e
E
G
H
R
c
a
n
ope
r
a
te
w
it
h
m
ul
ti
c
ont
e
xt
of
th
e
B
S
S
.
O
th
e
r
be
ne
f
it
s
of
th
e
E
G
H
R
is
e
xt
r
a
c
t
s
our
c
e
s
w
it
h
lo
w
di
m
e
ns
io
na
l
c
ont
e
xt
.
T
he
a
ut
hor
s
a
ppl
ie
d
th
e
pr
opos
e
d m
e
th
od t
o e
xt
r
a
c
t
th
e
a
ni
m
a
l
s
ound
s
.
A
bba
s
a
nd
S
a
lm
a
n
[
15]
in
tr
oduc
e
d
s
om
e
m
e
th
od
s
to
e
n
h
a
nc
e
th
e
pe
r
f
or
m
a
nc
e
of
th
e
l
in
e
a
r
I
C
A
de
pe
nde
ni
ng
on
th
e
qua
nt
um
pa
r
ti
c
le
s
w
a
r
m
opt
im
iz
a
ti
on
(
Q
P
S
O
)
a
nd
th
e
gl
ow
or
m
s
w
a
r
m
opt
im
iz
a
ti
on
(
G
S
O
)
w
it
h
th
r
e
e
obj
e
c
ti
ve
f
unc
ti
ons
a
r
e
E
nt
r
opy,
N
e
ge
nt
r
o
py,
a
nd
M
ut
ua
l
I
nf
o
r
m
a
ti
on.
S
o,
th
e
a
ut
ho
r
pr
opos
e
d
ne
w
N
onl
i
ne
a
r
I
C
A
m
e
th
od
d
e
pe
nds
on
s
om
e
nonl
in
e
a
r
m
e
th
ods
.
T
he
pr
opos
e
d
nonl
in
e
a
r
I
C
A
m
e
th
od
c
om
pa
r
e
d
w
it
h
c
om
m
onl
y
s
ta
nd
a
r
d
nonl
in
e
a
r
I
C
A
m
e
t
hods
a
s
S
O
M
ba
s
e
d
I
C
A
a
nd
R
B
F
ba
s
e
d
I
C
A
m
e
th
ods
.
T
he
r
e
s
ul
ts
pr
ove
d
th
a
t
th
e
pr
opos
e
d
m
e
th
od
ga
v
e
good
r
e
s
ul
t
s
a
c
c
or
di
ng
t
o
s
om
e
e
va
lu
a
ti
on
m
e
a
s
ur
e
m
e
nt
s
a
s
S
N
R
,
S
I
R
,
lo
g
-
L
ik
li
hood
r
a
ti
o,
a
nd
pe
r
c
e
pt
u
a
l
e
va
lu
a
ti
on
s
pe
e
c
h
qua
li
ty
(
P
E
S
Q
)
.
A
ll
th
e
pr
opos
e
d
m
e
th
ods
(
li
ne
a
r
I
C
A
a
nd
n
onl
in
e
a
r
I
C
A
)
im
pl
e
m
e
nt
e
d
w
it
h
da
ta
s
e
t
of
r
e
a
l
s
pe
e
c
he
s
ta
ke
n
f
r
om
th
e
in
te
r
na
ti
ona
l
te
le
c
om
m
uni
c
a
ti
on union (
I
T
U
)
, unde
r
8 K
H
z
f
r
e
que
nc
ie
s
.
B
r
e
nde
l,
a
nd
K
e
ll
e
r
m
a
nn
[
25]
,
in
tr
oduc
e
d
a
n
a
lg
or
it
hm
to
e
nha
nc
e
th
e
in
de
p
e
nde
nt
ve
c
to
r
a
n
a
ly
s
is
(
I
V
A
)
,
w
hi
c
h
is
one
of
th
e
B
S
S
m
e
th
ods
de
pe
nde
d
on
th
e
da
t
a
-
dr
iv
e
n
s
c
he
m
e
to
th
e
a
c
ous
ti
c
m
e
c
ha
ni
s
m
s
.
T
he
a
ut
hor
s
pr
op
os
e
d
f
a
s
t
c
onve
r
ge
n
c
e
r
ul
e
s
ba
s
e
d
on
e
ig
e
nva
lu
e
e
xt
r
a
c
ti
on
a
nd
th
e
m
a
jo
r
iz
e
-
m
in
im
iz
e
(
M
M
)
c
onc
e
pt
s
w
it
h
th
e
N
e
g
e
nt
r
opy
obj
e
c
ti
ve
f
unc
ti
on.
T
h
e
upda
te
d
r
ul
e
s
c
oul
d
b
e
e
f
f
ic
ie
nt
opt
im
iz
a
ti
on
a
ppr
oa
c
h
of
in
de
pe
nde
nt
lo
w
r
a
nk
m
a
tr
ix
a
na
ly
s
is
(
I
L
R
M
A
)
m
e
th
ods
.
T
he
a
ut
hor
s
a
ppl
ie
d
th
e
ir
pr
opos
e
d
m
e
th
od w
it
h da
ta
r
e
c
or
de
d i
n r
e
a
l
w
or
ld
s
ounds
.
3.
R
E
S
E
A
R
C
H
M
E
T
H
O
D
O
L
O
G
Y
A
s
m
e
nt
io
ne
d
in
th
e
pr
e
vi
ous
s
e
c
ti
on
s
,
one
of
s
ti
ll
pr
obl
e
m
s
in
t
he
di
gi
ta
l
s
ig
na
l
pr
oc
e
s
s
in
g
(
D
S
P
)
is
bl
in
d
s
our
c
e
(
S
ig
na
l)
s
e
pa
r
a
ti
on
(
B
S
S
)
.
T
h
e
B
S
S
pr
obl
e
m
e
m
e
r
gi
ng
in
m
a
ny
r
e
a
l
-
w
or
ld
f
ie
ld
s
a
s
s
ound
(
s
pe
e
c
h)
s
ig
na
l
pr
oc
e
s
s
in
g,
na
tu
r
a
l
im
a
ge
pr
oc
e
s
s
in
g,
M
R
I
,
f
M
R
I
,
E
E
G
a
nd
M
E
G
.
T
he
I
C
A
a
ppr
oa
c
h
is
m
os
t
e
f
f
ic
ie
nt
m
e
th
od t
o
s
ol
ve
t
he
B
S
S
pr
obl
e
m
.
T
he
I
C
A
ne
e
ds
t
o u
s
e
a
nd i
m
pl
e
m
e
nt
s
om
e
opt
im
iz
a
ti
on me
th
od
s
a
s
a
p
a
r
t
of
it
s
w
or
k.
T
h
e
r
e
f
or
e
,
in
m
a
ny
s
ta
nd
a
r
d
a
nd
pr
opos
e
d
m
e
th
ods
of
th
e
I
C
A
,
it
us
e
d
n
e
ur
a
l
ne
twor
ks
,
ge
ne
ti
c
a
lg
or
it
hm
s
,
a
nd/
or
s
w
a
r
m
in
te
ll
ig
e
nc
e
m
e
th
ods
.
M
o
s
t
I
C
A
m
e
th
ods
c
onf
r
ont
s
om
e
di
f
f
ic
ul
ts
in
e
f
f
ic
ie
nt
, a
c
c
ur
a
c
y, a
nd s
p
e
e
d.
T
hi
s
s
e
c
ti
on
c
onc
e
nt
r
a
te
s
on
th
e
pr
opos
e
d
m
e
th
od
th
a
t
c
ont
a
in
two
pa
r
ts
,
f
ir
s
tl
y
w
a
lk
a
bout
th
e
m
e
th
od
in
li
ne
s
te
ps
a
nd
it
s
e
qua
ti
ons
a
nd
s
ta
ge
s
.
S
e
c
ond
pa
r
t
c
ont
a
in
s
th
e
f
lo
w
c
ha
r
t
a
nd
th
e
a
lg
or
it
hm
of
th
e
pr
opos
e
d m
e
th
od.
3.1. P
r
op
os
e
d
m
e
t
h
od
I
n
th
e
pr
opos
e
d
m
e
th
od,
w
e
us
e
d
one
of
th
e
ne
ur
a
l
ne
twor
k
m
e
th
ods
is
th
e
B
ig
r
a
di
e
nt
m
e
th
od
a
s
a
n
I
C
A
s
tr
a
te
gy
to
s
ol
ve
B
S
S
pr
obl
e
m
.
A
t
s
a
m
e
ti
m
e
,
w
e
us
e
d
th
e
qua
nt
um
pa
r
ti
c
le
s
w
a
r
m
opt
im
iz
a
ti
on
a
s
a
n
opt
im
iz
a
ti
on me
th
od f
or
t
he
I
C
A
i
n a
hybr
id
m
a
nne
r
. T
he
B
S
S
m
e
t
hod will
be
a
s
:
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
B
ig
r
adi
e
nt
ne
ur
al
ne
tw
or
k
-
bas
e
d quantum par
ti
c
l
e
s
w
a
r
m
opt
i
m
iz
at
io
n f
or
…
(
H
us
s
e
in
M
. Sal
m
an
)
359
F
ir
s
tl
y,
th
e
m
e
th
od
a
s
s
um
e
s
th
a
t
th
e
r
e
a
r
e
,
a
t
le
a
s
t,
two
m
ix
e
d m
ono
-
s
pe
e
c
h
s
ig
na
l
s
to
f
or
m
ul
a
te
s
o
-
c
a
ll
e
d
s
upe
r
ve
c
to
r
w
it
h
two
ve
c
to
r
s
[
1]
,
th
is
s
upe
r
ve
c
to
r
r
e
pr
e
s
e
nt
th
e
ob
s
e
r
va
ti
on
s
ig
na
ls
.
B
e
f
or
e
e
xe
c
ut
e
th
e
I
C
A
s
te
ps
, m
u
s
t
pe
r
f
or
m
in
g
m
a
in
t
w
o pr
e
-
pr
oc
e
s
s
e
s
[
1
]
,
[
2]
:
C
e
nt
e
r
in
g:
in
c
lu
de
c
om
put
e
th
e
m
e
a
n
of
th
e
obs
e
r
va
ti
on
s
ig
na
l
a
nd
th
e
n
s
ubt
r
a
c
t
th
is
m
e
a
n
f
r
om
th
e
obs
e
r
va
ti
on
s
our
c
e
it
s
e
lf
,
(
’
=
−
[
]
)
a
nd
th
e
n
a
dd
th
e
m
e
a
n
ve
c
to
r
t
o
th
e
e
s
ti
m
a
te
d
s
our
c
e
ve
c
to
r
,
(
=
’
+
−
1
[
]
).
W
hi
te
ni
ng:
w
hi
te
n
th
e
m
ix
e
d
s
ig
na
l
x.
T
o
obt
a
in
th
e
ob
s
e
r
va
ti
on
s
ig
na
ls
un
c
or
r
e
la
te
d
a
nd
h
a
ve
uni
t
va
r
ia
nc
e
,
a
ppl
yi
ng
th
e
li
ne
a
r
m
ode
l
tr
a
ns
f
or
m
a
ti
on
(
~
=
1
2
),
w
he
r
e
r
e
pr
e
s
e
nt
e
ig
e
nve
c
to
r
of
[
]
,
a
nd
de
not
e
th
e
e
ig
e
nva
lu
e
s
of
[
]
.
T
he
a
im
of
w
hi
te
ni
ng
pr
oc
e
s
s
is
to
or
th
ogona
l
th
e
m
ix
in
g
m
a
tr
ix
.
A
f
te
r
th
e
n,
s
e
pa
r
a
te
th
e
s
e
w
hi
te
ne
d
s
ig
na
ls
b
a
s
e
d
on
th
e
obj
e
c
ti
ve
f
unc
ti
on.
T
he
pr
opos
e
d
m
e
th
od
us
e
d t
he
a
ppr
oxi
m
a
ti
on ne
ge
nt
r
opy f
unc
ti
on ba
s
e
d on Kur
to
s
i
s
(
3)
a
nd (
4)
a
s
a
n obje
c
ti
ve
f
unc
ti
on.
T
o
opt
im
iz
e
th
e
r
e
s
ul
ts
of
th
e
I
C
A
,
w
e
us
e
d
th
e
Q
P
S
O
opt
im
iz
a
ti
on
a
lg
or
it
hm
:
in
th
is
a
lg
or
it
hm
,
w
e
us
e
d t
he
f
our
th
-
or
de
r
s
ta
ti
s
ti
c
de
gr
e
e
e
qua
ti
on (
K
ur
to
s
is
)
t
o f
in
d t
he
i
ni
ti
a
l
va
lu
e
of
t
he
f
it
ne
s
s
f
unc
ti
on. T
he
n,
in
tr
om
is
s
io
n
in
to
m
a
in
lo
op
of
th
e
a
lg
or
it
hm
;
in
s
id
e
th
e
a
lg
or
it
hm
a
nd
unde
r
pr
e
de
f
in
e
d
it
e
r
a
ti
ons
,
f
in
d
m
e
a
n
be
s
t
s
ta
te
of
t
he
gl
oba
l
s
ta
te
i
n
t
he
s
e
a
r
c
h s
pa
c
e
of
t
he
pr
obl
e
m
.
T
o f
in
d t
he
va
lu
e
of
t
h
e
f
it
ne
s
s
va
lu
e
f
o
r
e
a
c
h
it
e
r
a
ti
on i
ns
id
e
t
he
Q
P
S
O
a
lg
or
it
hm
, w
e
us
e
d B
ig
r
a
di
e
nt
ne
ur
a
l
l
e
a
r
ni
ng f
or
t
hi
s
pur
pos
e
.
S
e
c
ondl
y,
w
hi
le
th
e
B
ig
r
a
di
e
nt
ne
ur
a
l
ne
twor
k
c
ha
r
a
c
te
r
iz
e
d
w
it
h
hi
gh
s
pe
e
d
c
onve
r
ge
n
c
e
–
a
s
m
e
nt
io
ne
d i
n s
e
c
ti
on 1.3 in t
hi
s
pa
pe
r
-
it
us
e
d
in
t
he
pr
opos
e
d
m
e
th
od a
s
i
n (
11)
;
to
c
om
put
e
t
he
f
it
ne
s
s
va
lu
e
of
t
he
Q
P
S
O
a
lg
or
it
hm
;
w
he
r
e
t
he
l
e
a
r
ni
ng f
unc
ti
on g is
de
f
in
e
d a
s
i
n (
12)
.
=
∗
(
12)
W
he
r
e
x r
e
pr
e
s
e
nt
s
t
he
ob
s
e
r
va
ti
on s
ig
na
l
ve
c
to
r
.
N
e
ve
r
th
e
le
s
s
,
th
is
a
lg
or
it
hm
ha
s
im
por
ta
nt
li
m
it
a
ti
on
is
una
bl
e
to
ge
t
good
r
e
s
ul
t
in
th
e
s
e
pa
r
a
ti
on
pr
oc
e
s
s
.
T
he
r
e
f
or
e
, i
n t
he
t
hi
r
d pa
r
t,
t
he
pr
opos
e
d m
e
th
od t
e
nds
t
o opti
m
iz
e
t
he
B
ig
r
a
di
e
nt
a
lg
or
it
hm
by us
in
g
th
e
qua
nt
um
pa
r
ti
c
le
s
w
a
r
m
opt
im
iz
a
ti
on
m
e
th
od
be
c
a
us
e
th
i
s
m
e
th
od
ha
ve
s
om
e
f
e
a
tu
r
e
s
a
s
a
n
a
c
c
ur
a
te
r
e
s
ul
ts
in
th
e
s
e
pa
r
a
ti
on
pr
oc
e
s
s
,
f
e
w
pa
r
a
m
e
te
r
s
,
a
nd
lo
w
e
r
c
om
put
a
ti
on
r
e
qui
r
e
m
e
nt
s
,
but
th
is
m
e
th
od
s
lo
w
e
r
th
a
n
B
ig
r
a
di
e
nt
m
e
th
od.
F
or
s
pe
c
if
ic
it
e
r
a
ti
ons
,
th
e
pr
o
pos
e
d
hybr
id
m
e
th
od
ge
t
good
r
e
s
ul
ts
c
ol
le
c
t
be
twe
e
n t
he
B
ig
r
a
di
e
n
t
m
e
th
od a
nd t
he
Q
P
S
O
m
e
th
od.
3.2. E
xt
r
e
m
e
ly
s
t
e
p
s
o
f
t
h
e
p
r
op
os
e
d
m
e
t
h
od
T
he
pr
opos
e
d
m
e
th
od
hybr
id
be
twe
e
n
Q
P
S
O
a
nd
B
ig
r
a
di
e
nt
to
s
e
pa
r
a
te
th
e
m
ono
-
s
pe
e
c
h
m
ix
e
d
s
ig
na
ls
.
T
h
e
f
lo
w
c
ha
r
t
or
de
r
a
s
f
lo
w
:
f
ir
s
t
s
te
p,
r
e
c
e
iv
in
g
obs
e
r
va
ti
on
(
m
ix
tu
r
e
s
ig
na
ls
)
,
th
e
s
ig
na
ls
m
ix
e
d
in
in
s
ta
nt
unous
m
a
nne
r
a
nd
in
c
lu
de
a
t
le
a
s
t
two
s
pe
e
c
he
s
.
S
e
c
ond
s
te
p
pe
r
f
or
m
s
th
e
e
xt
r
e
m
e
pr
e
pr
oc
e
s
s
e
s
(
c
e
nt
e
r
in
g
a
nd
w
hi
te
ni
ng)
,
s
te
p
th
r
e
e
in
c
lu
de
s
c
a
lc
ul
a
ti
ng
in
it
ia
l
f
it
ne
s
s
va
lu
e
by
u
s
in
g
th
e
obj
e
c
ti
ve
f
unc
ti
o
“
K
ur
to
s
is
f
unc
ti
on”
a
s
in
(
3
)
.
F
r
om
s
te
p
4
th
e
c
or
e
of
th
e
pr
o
pos
e
d
m
e
th
od
w
il
l
s
ta
r
te
d,
w
he
r
e
in
pr
e
de
f
in
e
it
e
r
a
ti
on,
th
e
Q
P
S
O
im
pl
e
m
e
nt
e
d
a
nd
opt
im
iz
e
d
us
in
g
th
e
B
ig
r
a
di
e
nt
f
unc
ti
on
unt
il
te
r
m
in
a
te
th
e
it
e
r
a
ti
on.
T
he
B
ig
r
a
di
e
nt
f
unc
ti
on
c
ons
id
e
r
a
s
th
e
obj
e
c
ti
ve
f
unc
ti
on
of
t
he
pr
opos
e
d
I
C
A
m
e
th
od.
A
t
th
e
e
nd
of
th
e
f
lo
w
c
ha
r
t,
r
e
c
ove
r
in
g
th
e
s
our
c
e
s
ig
na
ls
a
nd
e
va
lu
a
ti
ng
of
th
e
pe
r
f
or
m
a
nc
e
of
th
e
pr
opos
e
d
m
e
th
od.
T
he
F
ig
ur
e
2 i
ll
us
ta
r
e
t
he
f
lo
w
c
ha
r
t
of
t
he
pr
opos
e
d m
e
th
od.
I
n
a
ddi
ti
on,
th
e
a
lg
or
it
hm
of
th
e
pr
opos
e
d
m
e
th
od
c
ont
a
in
s
ni
n
e
m
a
in
s
te
ps
.
T
h
e
s
e
s
t
e
ps
or
de
r
e
d
a
s
f
lo
w
:
s
te
ps
(
1,
2)
in
c
lu
de
pe
r
f
or
m
in
g
th
e
pr
e
pr
oc
e
s
s
e
s
of
th
e
I
C
A
(
c
e
nt
e
r
in
g
a
nd
w
hi
te
ni
ng)
,
s
te
p
3,
f
in
d
th
e
in
it
ia
l
m
a
xi
m
um
va
lu
e
of
th
e
f
it
ne
s
s
f
unc
ti
on.
F
r
om
s
te
p
4
to
s
te
p
7,
th
e
Q
P
S
O
pe
r
f
or
m
e
d
a
nd
in
s
id
e
it
th
e
B
ig
r
a
di
e
nt
f
unc
ti
on
is
us
e
d.
T
h
e
s
te
p
8
in
c
lu
de
s
s
e
p
a
r
a
ti
ng
th
e
m
ix
e
d
s
ig
n
a
ls
a
nd
r
e
c
ov
e
r
in
g
th
e
s
our
c
e
s
.
I
n
s
te
p
9,
pe
r
f
or
m
in
g
th
e
e
va
lu
a
ti
on
pr
oc
e
s
s
of
th
e
pe
r
f
or
m
a
n
c
e
of
th
e
pr
opos
e
d
m
e
th
od.
T
hi
s
a
lg
or
it
hm
il
lu
s
tr
a
te
d f
ol
lo
w
in
g i
n A
lg
or
it
hm
1.
A
l
gor
i
t
h
m
(
1)
:
B
i
gr
ad
i
e
n
t
b
as
e
d
on
Q
P
S
O
I
C
A
al
gor
i
t
h
m
I
n
p
u
t
:
w
hi
t
e
-
x
%
w
hi
t
e
ne
d ve
c
t
or
s
O
u
t
p
u
t
:
z
;
%
s
e
pa
r
a
t
e
d ve
c
t
or
s
(
r
e
c
ove
r
e
d s
our
c
e
s
)
A
l
gor
i
t
h
m
S
t
e
p
s
1.
I
ni
t
i
a
l
i
z
i
ng s
e
t
of
di
a
gona
l
s
e
pa
r
a
t
e
d m
a
t
r
i
c
e
s
.
x
1
=
r
andom
l
y
(
K
, K
, popul
at
i
on
);
2.
C
a
l
c
ul
a
t
e
i
ni
t
i
a
l
l
y f
i
t
ne
s
s
va
l
ue
s
of
t
he
c
ur
r
e
nt
pos
i
t
i
ons
of
pa
r
t
i
c
l
e
s
us
i
ng t
he
o
bj
e
c
t
i
ve
f
unc
t
i
on
3.
F
or
i
=
1 t
o popul
at
i
on
y
=
x
1
* w
hi
t
e
-
x;
C
e
nt
e
r
i
ng and W
hi
t
e
ni
ng y
;
P
e
r
f
or
m
t
he
obj
e
c
t
i
ve
f
unc
t
i
on ba
s
e
d on t
he
s
ys
t
e
m
i
n (
3 a
nd 4)
fit
(
i
)
=
s
um
(
pr
opos
e
d f
un.)
;
%
c
om
put
e
c
ur
r
e
nt
f
i
t
ne
s
s
va
l
ue
E
nd_ f
or
4.
F
i
nd t
he
i
ni
t
i
a
l
m
a
xi
m
um
va
l
ue
of
t
he
f
i
t
ne
s
s
va
l
ue
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
.
10
, N
o.
2, J
une
20
21
:
355
–
364
360
pgm
ax
= ma
xi
m
um
(
fit
)
5.
M
a
i
n l
oop i
t
e
r
a
t
i
on of
t
he
Q
P
S
O
a
l
gor
i
t
hm
d=
1;
%
i
t
e
r
a
t
i
on
i
nde
x
D
o
m
be
s
t
=s
um
(
fit
)
/
popul
at
i
on
;
%
m
e
a
n of
t
he
be
s
t
l
oc
a
l
pos
i
t
i
ons
f
or
i
=
1 t
o popul
at
i
on
f
or
j
=
1 t
o K
f
or
s
=
1 t
o K
phi
=
r
andom
()
;
p=
phi
* pi
m
ax
j
,s
,i
+
(
1
-
phi
)
* pgm
ax
j
,s
;
u=
r
andom
(
)
;
x
j
,s
,i
=
p
±
(
al
pha * |
|
m
be
s
t
-
x
j
,s
,i
|
|
* l
n
(
1/
u
))
;
e
nd_ f
or
(
s
)
e
nd_ f
or
(
j
)
e
nd_ f
or
(
i
)
6.
C
a
l
c
ul
a
t
e
ne
w
va
l
ue
s
of
t
he
pos
i
t
i
ons
of
pa
r
t
i
c
l
e
s
F
or
m
=
1 t
o popul
at
i
on
y
=
x
1
* x
;
C
e
nt
e
r
i
ng and W
hi
t
e
ni
ng y
;
P
e
r
f
or
m
B
i
gr
a
di
e
nt
r
u
l
e
-
a
s
a
n obj
e
c
t
i
ve
f
unc
t
i
on
-
w
i
t
h t
w
o pa
r
a
m
e
t
e
r
s
(
µ
k
,
γ
k
)
t
o f
i
nd ne
w
f
i
t
ne
s
s
va
l
ue
a
s
i
n (
11)
f
i
t
ne
w
(
m
)
= W
k
+
µ
k
v
k
g
(
y
k
T
)
-
γ
k
W
k
(I
-
W
k
T
W
k
)
%
f
i
nd ne
w
f
i
t
ne
s
s
va
l
ue
E
nd_ f
or
(
m
)
7.
F
i
nd
t
he
ne
w
m
a
xi
m
um
va
l
ue
of
t
he
f
i
t
ne
s
s
va
l
ue
s
.
pgm
ax
= ma
xi
m
um
(
f
i
t
ne
w
)
8.
I
nc
r
e
m
e
nt
t
he
i
t
e
r
a
t
i
on, a
nd s
t
oppe
d
d=
d
+1
;
%
i
t
e
r
a
t
i
on
i
nde
x.
U
nt
i
l
d=
m
ax
i
t
e
r
;
%
t
e
r
m
i
na
t
e
t
he
l
oop of
Q
P
S
O
a
l
gor
i
t
hm
.
9.
z
=
y
;
%
r
e
c
ove
r
e
d
s
our
c
e
s
10.
SN
R
,
SD
R
,
A
V
C
C
;
%
E
va
l
ua
t
i
on pr
oc
e
s
s
.
E
n
d
A
l
gor
i
t
h
m
.
F
ig
ur
e
2. F
lo
w
c
ha
r
t
of
t
he
pr
opos
e
d m
e
th
od
4.
R
E
S
U
L
T
S
A
N
D
D
I
S
C
U
S
S
I
O
N
A
f
te
r
im
pl
e
m
e
nt
th
e
pr
opos
e
d
m
e
th
od
w
it
h
a
s
pe
c
if
ie
d
s
pe
e
c
he
s
,
th
e
r
e
s
ul
ts
of
th
e
s
e
p
a
r
a
ti
on
pr
oc
e
s
s
w
a
s
nove
l
a
nd
m
or
e
s
pe
e
d
c
onve
r
ge
nc
e
th
a
n
ot
he
r
c
o
m
pa
r
e
d
m
e
th
ods
.
A
ddi
ti
ona
ll
y,
to
il
lu
s
tr
a
te
th
e
pe
r
f
or
m
a
nc
e
of
th
e
p
r
opos
e
d
H
ybr
id
I
C
A
m
e
th
od,
w
e
r
e
por
t
th
e
e
xpe
r
im
e
nt
a
l
r
e
s
ul
ts
on
th
e
s
e
pa
r
a
ti
on
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
B
ig
r
adi
e
nt
ne
ur
al
ne
tw
or
k
-
bas
e
d quantum par
ti
c
l
e
s
w
a
r
m
opt
i
m
iz
at
io
n f
or
…
(
H
us
s
e
in
M
. Sal
m
an
)
361
s
ig
na
ls
.
W
e
c
om
p
a
r
e
d
th
e
r
e
s
ul
ts
w
it
h
t
hr
e
e
I
C
A
m
e
th
ods
,
Q
P
S
O
-
ba
s
e
d
I
C
A
m
e
th
od
[
7]
in
tr
oduc
e
d
in
s
e
c
ti
on 1.2, B
ig
r
a
di
e
nt
-
ba
s
e
d I
C
A
m
e
th
od [
21]
i
ll
us
tr
a
te
d i
n s
e
c
ti
on 1.3, a
nd s
ta
nda
r
d F
a
s
tI
C
A
m
e
th
od [
5]
.
T
he
pr
opos
e
d
m
e
th
od
(
hybr
id
m
e
th
od)
c
ol
le
c
t
s
om
e
of
th
e
Q
P
S
O
a
dva
nt
a
ge
s
in
th
e
a
c
c
ur
a
c
y
a
nd
s
om
e
of
th
e
B
ig
r
a
di
e
nt
ne
ur
a
l
ne
twor
k
a
dva
nt
a
ge
s
in
th
e
s
p
e
e
d
c
onve
r
ge
nc
e
.
T
he
s
e
pr
ope
r
ti
e
s
m
a
de
th
e
pr
opos
e
d
m
e
th
od
ga
ve
good
r
e
s
ul
ts
be
tt
e
r
th
a
n
Q
P
S
O
a
nd
B
ig
r
a
di
e
nt
,
a
ls
o
be
tt
e
r
th
a
n
s
ta
nda
r
d
F
a
s
tI
C
A
m
e
th
od.
4.1.
S
ou
r
c
e
s
ig
n
al
s
an
d
p
ar
am
e
t
e
r
s
s
e
t
t
in
g
th
is
pa
pe
r
a
da
pt
s
th
e
c
oc
kt
a
il
-
pa
r
ty
is
s
u
e
to
il
lu
s
tr
a
te
th
e
pr
opo
s
e
d
m
e
th
od
in
th
e
s
e
pa
r
a
ti
on
pr
oc
e
s
s
.
F
ir
s
tl
y,
th
e
s
pe
e
c
h
s
ig
na
l
s
a
r
e
s
e
le
c
t
e
d
f
r
om
th
e
w
e
b
s
it
e
of
th
e
da
ta
ba
s
e
of
th
e
in
te
r
na
ti
ona
l
te
le
c
om
m
uni
c
a
ti
on
uni
on
(
I
T
U
)
(
h
tt
ps
:/
/g
it
hub.c
om
/d
e
nni
s
gus
e
/I
T
U
-
T
_pe
s
q/
tr
e
e
/m
a
s
te
r
/c
om
f
or
m
)
,
a
nd
th
e
da
ta
ba
s
e
of
th
e
U
ni
v
e
r
s
it
y
of
D
a
ll
a
s
(
ht
tp
:/
/ww
w
.ut
da
ll
a
s
.
e
du/
~
lo
iz
ou/
s
pe
e
c
h/
noi
z
e
us
/
)
.
D
if
f
e
r
e
nt
e
ig
ht
s
pe
e
c
h
s
ig
na
l
s
s
e
le
c
t
e
d
f
r
om
th
os
e
w
e
bs
it
e
s
a
nd
r
e
m
a
ne
d
by
th
e
a
ut
hor
s
a
r
e
s
our
c
e
11,
s
our
c
e
22,
s
our
c
e
4,
s
our
c
e
7,
ju
li
a
8,
22m
,
r
a
y8,
a
r
e
r
ic
h8.
A
ll
th
e
s
e
s
ig
na
ls
a
r
e
noi
s
e
le
s
s
a
nd
8K
H
z
f
r
e
que
nc
y.
S
e
c
ondl
y,
a
ll
th
e
s
e
s
ig
na
ls
a
c
hi
e
ve
th
e
s
up
e
r
-
ga
us
s
ia
nt
y
pr
in
c
ip
le
,
a
nd
in
de
pe
nde
nt
id
e
nt
ic
a
l
di
s
tr
ib
ut
io
n
(
i.
i.
d)
c
ondi
ti
ons
a
c
c
or
di
ng K
ur
to
s
is
m
e
a
s
ur
e
m
e
nt
i
n
(
3)
, a
s
t
he
I
C
A
c
ondi
ti
ons
[
1
]
,
[
6]
.
T
he
p
a
pe
r
a
s
s
um
e
s
a
c
oc
kt
a
il
pa
r
ty
pr
obl
e
m
w
it
h
two
s
our
c
e
s
a
nd
two
s
e
ns
or
s
a
s
a
n
a
pp
li
c
a
ti
on
of
th
e
B
S
S
.
T
he
s
our
c
e
s
m
ix
e
d
in
in
s
ta
nt
a
ne
ous
li
ne
a
r
m
e
th
od
in
(
1)
,
a
nd
th
e
2×
2
m
ix
in
g
m
a
tr
ix
A
is
ge
ne
r
a
te
d
in
r
a
ndoml
y di
s
tr
ib
ut
e
d nor
m
a
l,
i
n t
he
c
lo
s
e
d r
a
nge
[
-
20,20]
, a
s
i
n (
13)
:
=
+
(
−
)
∗
(
2
,
2
)
(
13)
W
he
r
e
a
,
a
nd
b
a
r
e
th
e
m
in
im
um
a
nd
m
a
xi
m
um
r
a
nge
of
s
i
gna
ls
di
s
tr
ib
ut
io
n
r
e
s
pe
c
ti
ve
ly
.
T
hi
s
m
a
tr
ix
a
c
hi
e
ve
s
th
e
w
e
ll
-
c
ondi
ti
on
num
be
r
.
A
s
r
e
s
ul
t,
f
r
om
th
e
s
e
le
c
te
d
s
pe
e
c
h
s
ig
na
ls
f
or
m
ul
a
te
f
our
m
ix
tu
r
e
c
a
s
e
s
.
T
he
m
ix
tu
r
e
c
a
s
e
s
a
r
e
c
a
s
e
1
(
s
our
c
e
11,
s
our
c
e
22)
,
c
a
s
e
2
(
s
our
c
e
4,
s
ou
r
c
e
7)
,
c
a
s
e
3
(
ju
li
a
8,
22m
)
,
a
nd
c
a
s
e
4
(
r
a
y8,
r
ic
h8)
unde
r
th
e
s
a
m
e
m
ix
in
g
c
ondi
ti
on
s
.
T
he
T
a
bl
e
1
vi
e
w
s
a
ll
th
e
in
it
ia
l
pa
r
a
m
e
te
r
s
of
t
he
m
ix
tu
r
e
pr
oc
e
s
s
.
T
a
bl
e
1
.
S
pe
e
c
h
e
s
f
il
e
s
a
nd mi
xi
ng ma
tr
ic
e
s
M
i
xe
d
c
a
s
e
S
our
c
e
s
n
a
m
e
s
K
ur
t
os
i
s
of
s
our
c
e
s
i
gna
l
s
L
e
ngt
h
(
s
a
m
pl
e
s
)
A
b
M
i
xe
d
m
a
t
r
i
x
C
ondi
t
i
on
n
um
be
r
1
s
our
c
e
11
4.2686
50000
-
20
20
-
43.3496
-
35.8895
-
26.8210 19.2581
1.5189
s
our
c
e
22
6.1309
2
S
our
c
e
4
5.4201
50000
-
12
12
8.8056
-
13.3578
-
14.3396
-
21.1326
1.4703
S
our
c
e
7
3.8568
3
J
ul
i
a
8
6.3970
21582
-
1
1
0.7338
-
1.1131
-
1.1950
-
1.7611
1.2703
22m
7.9978
4
R
a
y8
7.4982
61038
-
3
2
-
1.2118 1.6230
-
1.2055
-
1.1720
1.2558
R
i
c
h8
6.5054
I
n
a
ddi
ti
ona
l,
m
a
in
pa
r
a
m
e
te
r
s
of
th
e
pr
opos
e
d
m
e
th
od
s
e
t
a
s
popula
ti
on=
10,
m
a
xi
m
um
it
e
r
a
ti
on
i
s
60,
a
nd
c
ont
r
a
c
ti
on
–
e
xpa
n
s
io
n
f
a
c
to
r
(
β
in
(
10)
)
is
0.75,
(
pa
r
a
m
e
te
r
s
of
B
ig
r
a
di
e
nt
in
(
11
)
a
r
e
µ=
0.00001,
γ
=
0.5)
.
T
he
Q
P
S
O
-
ba
s
e
d
I
C
A
a
s
in
[
7]
pa
r
a
m
e
te
r
s
a
r
e
m
a
xi
m
um
it
e
r
a
ti
on
is
35,
popul
a
ti
on=
10,
a
nd
c
ont
r
a
c
ti
on
–
e
xpa
ns
io
n
f
a
c
to
r
(
β
in
10)
is
0.75.
T
he
B
ig
r
a
di
e
nt
-
ba
s
e
d
I
C
A
a
lg
or
it
hm
pa
r
a
m
e
te
r
s
a
r
e
µ=
0.00001, γ
=
0.5 a
s
i
n
(
11
)
.
4.2. P
e
r
f
or
m
an
c
e
e
val
u
at
io
n
I
n
or
de
r
to
m
e
a
s
ur
e
th
e
a
c
c
ur
a
c
y
of
th
e
pr
opos
e
d
a
lg
or
it
hm
,
w
e
e
va
lu
a
te
it
us
in
g
th
r
e
e
pe
r
f
or
m
a
nc
e
in
de
xe
s
:
s
ig
na
l
-
to
-
noi
s
e
r
a
ti
o
(
S
N
R
)
,
s
ig
na
l
-
to
-
di
s
to
r
ti
on
r
a
ti
o
(
S
D
R
)
,
a
nd
a
bs
ol
ut
e
va
lu
e
of
c
or
r
e
la
ti
on
c
oe
f
f
ic
ie
nt
(
A
V
C
C
)
. T
he
y a
r
e
, r
e
s
pe
c
ti
ve
ly
, de
f
in
e
d a
s
f
ol
lo
w
s
.
T
he
r
e
c
ons
tr
uc
ti
on me
a
s
ur
e
i
s
s
t
a
te
d a
s
a
s
ig
na
l
-
to
-
noi
s
e
r
a
ti
o i
nde
x of
t
he
e
r
r
or
[
8]
, t
ha
t
is
:
S
N
R
=
10 l
og
(
∑
(
(
)
)
2
∑
(
(
)
−
(
)
)
2
=
1
)
(
)
(
14)
W
he
r
e
(
)
is
th
e
s
our
c
e
s
ig
na
ls
,
(
)
de
not
e
th
e
r
e
c
ove
r
e
d
s
ig
na
ls
,
N
is
le
ngt
h
of
th
e
s
ig
na
ls
(
num
be
r
of
s
a
m
pl
e
s
)
,
t
is
ti
m
e
in
de
x,
a
nd
i
s
ig
na
l
in
de
x.
T
h
e
S
N
R
m
e
a
s
ur
e
m
e
nt
pl
a
c
e
in
t
he
r
a
nge
[
0,1]
be
twe
e
n
two
s
ig
na
ls
.
I
t
ne
a
r
by
to
0,
w
he
n
bot
h
s
ig
na
l
s
ne
a
r
by
to
ha
ve
s
a
m
e
e
n
e
r
gy
le
ve
l.
B
a
s
e
d
on
th
e
S
N
R
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
.
10
, N
o.
2, J
une
20
21
:
355
–
364
362
in
de
x,
th
e
r
e
c
ove
r
e
d
s
ig
na
ls
s
houl
d
be
r
e
s
c
a
le
d
to
th
e
s
a
m
e
e
ne
r
gy
le
ve
l
a
s
th
e
ir
c
or
r
e
s
ponding
or
ig
in
a
l
s
ig
na
ls
. A
ls
o,
s
ig
na
l
-
to
-
di
s
to
r
ti
on r
a
ti
o (
S
D
R
)
[
9
]
, i
s
de
f
in
e
d a
s
:
S
D
R
=
10 l
og
(
∑
(
(
)
−
(
)
)
2
=
1
∑
(
(
)
)
2
)
(
dB
)
(
15)
I
n
a
ddi
ti
ona
l,
th
e
a
bs
ol
ut
e
va
lu
e
of
c
or
r
e
la
ti
on
c
oe
f
f
ic
ie
nt
(
A
V
C
C
)
[
10]
,
is
e
xpl
oi
t
to
de
te
r
m
in
e
th
e
s
im
il
a
r
it
y de
gr
e
e
be
twe
e
n or
ig
in
a
l
s
ig
na
ls
a
nd r
e
c
ove
r
e
d s
ig
na
l
s
. T
he
A
V
C
C
de
s
c
r
ib
e
d
in
(
16)
:
A
V
C
C
=
|
∑
(
)
(
)
=
1
√
∑
2
(
)
∑
2
(
)
=
1
=
1
|
(
16)
L
ow
e
r
S
N
R
,
a
nd
hi
ghe
r
S
D
R
a
nd
A
V
C
C
r
e
pr
e
s
e
nt
th
a
t
th
e
s
e
p
a
r
a
te
d
a
nd
r
e
c
ove
r
e
d s
ig
na
ls
a
r
e
ne
a
r
s
im
il
a
r
to
th
e
s
our
c
e
s
ig
na
ls
.
F
ur
th
e
r
m
or
e
,
th
e
c
om
put
a
ti
on
ti
m
e
is
u
s
e
d
a
s
e
va
lu
a
ti
on
in
de
x
f
or
a
ll
m
e
th
ods
unde
r
s
a
m
e
de
vi
c
e
a
nd e
qui
pm
e
nt
c
ondi
ti
ons
.
4.3. P
e
r
f
or
m
an
c
e
an
al
ys
is
o
f
s
e
p
ar
at
io
n
r
e
s
u
lt
s
T
he
pr
opos
e
d
m
e
th
od
a
nd
ot
he
r
m
e
th
ods
a
r
e
s
im
ul
a
te
d
a
nd
p
r
ogr
a
m
m
e
d
w
it
h
M
A
T
L
A
B
R
201
7b.
T
he
y a
r
e
e
xe
c
ut
e
d on P
C
unde
r
I
nt
e
l
C
or
e
i
5, C
P
U
2.5 G
H
z
, a
n
d R
A
M
12 G
B
.
T
o
e
va
lu
a
te
a
nd
a
na
ly
s
is
th
e
p
e
r
f
or
m
a
nc
e
of
th
e
pr
opos
e
d
m
e
th
od
by
us
in
g
th
e
pe
r
f
or
m
a
nc
e
m
e
a
s
ur
e
m
e
nt
s
:
S
N
R
, S
D
R
, A
V
C
C
a
nd c
om
put
a
ti
on t
im
e
. T
he
T
a
bl
e
s
2
-
5 de
s
c
r
ib
e
t
he
r
e
s
ul
ts
of
th
e
pr
opos
e
d
m
e
th
od
in
s
e
pa
r
a
ti
on
pr
oc
e
s
s
unde
r
th
e
s
e
e
va
lu
a
ti
on
in
de
x
e
s
,
a
nd
s
a
m
e
s
e
pa
r
a
ti
on
c
ondi
ti
ons
f
or
a
ll
s
e
pa
r
a
ti
on c
a
s
e
s
.
T
a
bl
e
2
il
lu
s
tr
a
te
s
th
e
a
c
c
ur
a
c
y
of
th
e
pr
opos
e
d
m
e
th
od
ve
r
s
us
ot
he
r
m
e
th
ods
unde
r
th
e
S
N
R
m
e
a
s
ur
e
m
e
nt
,
w
h
e
r
e
lo
w
e
r
r
e
s
ul
ts
a
r
e
e
vi
d
e
nc
e
on
hi
ghe
r
a
c
c
ur
a
c
y
s
e
pa
r
a
ti
on.
I
n
th
is
ta
bl
e
,
th
e
pr
opos
e
d
m
e
th
od
(
H
ybr
id
)
a
ppe
a
r
m
or
e
a
c
c
ur
a
c
y
th
a
n
ot
he
r
m
e
th
od
in
t
w
o
s
e
pa
r
a
ti
on
c
a
s
e
s
a
s
in
di
c
a
te
d
in
r
e
d
c
ol
or
.
S
o,
s
a
m
e
in
de
x
obs
e
r
ve
d
in
T
a
bl
e
3,
unde
r
th
e
S
D
R
m
e
a
s
ur
e
m
e
nt
,
th
e
pr
opos
e
d
m
e
th
od
(
H
ybr
id
)
ga
ve
be
tt
e
r
a
c
c
ur
a
c
y
r
e
s
ul
t
in
one
s
e
pa
r
a
ti
on
c
a
s
e
in
di
c
a
te
d
in
r
e
d
c
ol
or
.
A
ls
o,
in
th
e
T
a
bl
e
4,
th
e
pr
opos
e
d
m
e
th
od
(
H
y
b
r
id
)
ga
ve
be
s
t
r
e
s
ul
ts
in
one
c
a
s
e
a
s
in
di
c
a
te
d
in
r
e
d
c
ol
o
r
,
unde
r
th
e
A
V
C
C
m
e
a
s
ur
e
m
e
nt
.
I
n
th
e
S
D
R
a
nd
A
V
C
C
m
e
a
s
ur
m
e
nt
s
,
be
tt
e
r
r
e
s
ul
ts
(
hi
ghe
r
a
c
c
ur
a
c
y)
a
r
e
hi
ghe
r
va
lu
e
s
.
T
a
bl
e
5,
th
e
c
om
put
a
ti
on
ti
m
e
m
e
a
s
ur
e
m
e
nt
,
a
pp
e
a
r
s
th
a
t
th
e
B
ig
r
a
di
e
nt
m
e
th
od
is
f
a
s
te
r
t
ha
n
ot
he
r
m
e
th
ods
but
th
e
pr
opos
e
d
m
e
th
od
(
hybr
id
)
w
a
s
f
a
s
te
r
t
ha
n Q
P
S
O
m
e
th
od a
s
i
ndi
c
a
te
d i
n gr
e
e
n c
ol
or
.
A
s
a
r
e
s
ul
ts
,
th
e
Q
P
S
O
m
e
th
od
w
a
s
be
tt
e
r
th
a
n
B
ig
r
a
di
e
nt
m
e
th
od
in
th
e
a
c
c
ur
a
c
y
m
e
a
s
ur
e
m
e
nt
s
(
S
N
R
,
S
D
R
,
a
nd
A
V
C
C
)
,
but
th
e
B
ig
r
a
di
e
nt
m
e
th
od
w
a
s
be
tt
e
r
th
a
n
th
e
Q
P
S
O
m
e
th
od
in
c
om
put
a
ti
on
ti
m
e
m
e
a
s
ur
e
m
e
nt
.
W
he
r
e
a
s
th
e
pr
opo
s
e
d
m
e
th
od
c
ol
le
c
ts
be
s
t
a
c
c
u
r
a
c
y
pr
ope
r
ti
e
s
of
th
e
Q
P
S
O
m
e
th
od
w
it
h
th
e
s
pe
e
d of
t
he
B
ig
r
a
di
e
nt
m
e
th
od
.
T
a
bl
e
2
.
S
N
R
m
e
a
s
ur
e
m
e
nt
(
dB
)
Q
P
S
O
H
ybr
i
d
F
a
s
t
I
C
A
B
i
G
r
a
di
e
nt
1
0.1916
0.2013
0.2003
0.2395
2
0.2234
0.0813
0.2517
0.1413
3
0.1032
0.1032
0.1125
0.0416
4
0.1182
0.102
0.2748
0.3032
T
a
bl
e
3
.
S
D
R
m
e
a
s
ur
e
m
e
nt
(
dB
)
Q
P
S
O
H
ybr
i
d
F
a
s
t
I
C
A
B
i
G
r
a
di
e
nt
1
14.0401
13.9585
13.9598
13.087
2
19.2094
18.0378
18.6442
18.6747
3
20.1899
20.1919
19.8358
20.1905
4
16.964
16.649
18.1005
17.0802
T
a
bl
e
4
.
A
V
C
C
m
e
a
s
ur
e
m
e
nt
Q
P
S
O
H
ybr
i
d
F
a
s
t
I
C
A
B
i
G
r
a
di
e
nt
1
0.761
0.8741
0.6567
0.6532
2
12.4493
1.0372
12.6557
5.9844
3
3.011
3.0069
3.1057
0.0684
4
1.5409
0.4259
11.9029
11.2008
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
B
ig
r
adi
e
nt
ne
ur
al
ne
tw
or
k
-
bas
e
d quantum par
ti
c
l
e
s
w
a
r
m
opt
i
m
iz
at
io
n f
or
…
(
H
us
s
e
in
M
. Sal
m
an
)
363
T
a
bl
e
5
.
C
om
put
a
ti
on
t
im
e
(
s
e
c
ond)
Q
P
S
O
H
ybr
i
d
F
a
s
t
I
C
A
B
i
G
r
a
di
e
nt
1
8.5599
4.497
0.3988
0.1078
2
8.545
4.3368
0.1582
0.0831
3
3.8416
1.9454
0.1603
0.0565
4
10.4473
5.4693
0.2105
0.0876
F
r
om
th
e
s
e
ta
bl
e
s
, w
e
c
a
n
c
le
a
r
ly
ob
s
e
r
ve
th
a
t
th
e
pr
opos
e
d
m
e
th
od
be
ha
ve
w
e
ll
th
a
n
ot
he
r
m
e
th
ods
in
m
os
t
c
a
s
e
s
a
c
c
or
di
ng
to
a
ll
m
e
a
s
ur
e
m
e
nt
s
.
B
a
s
e
d
on
S
N
R
in
de
x
a
nd
A
V
C
C
in
de
x,
th
e
pr
opos
e
d
(
hybr
id
)
m
e
th
od
be
tt
e
r
th
a
n
th
e
B
ig
r
a
di
e
nt
a
nd
Q
P
S
O
m
e
th
ods
in
m
o
s
t
c
a
s
e
s
.
A
ls
o,
in
th
e
S
D
R
in
de
x,
th
e
hybr
id
(
pr
opos
e
d)
m
e
th
od
be
tt
e
r
th
a
n B
ig
r
a
di
e
nt
m
e
th
od. W
he
r
e
a
s
,
th
e
ti
m
e
c
om
put
a
ti
on
in
d
e
x
c
le
a
r
ly
de
m
ons
tr
a
te
s
th
a
t
th
e
hybr
id
m
e
th
od
f
a
s
te
r
th
a
n
Q
P
S
O
m
e
th
od.
T
he
r
e
f
or
e
,
th
e
hybr
id
m
e
th
od
(
Q
P
S
O
a
nd
B
ig
r
a
di
e
nt
)
c
ons
id
e
r
a
s
a
nov
e
l
m
e
th
od of
t
he
B
S
S
pr
obl
e
m
.
5.
C
O
N
C
L
U
S
I
O
N
T
he
I
C
A
a
ppr
oa
c
he
s
a
r
e
one
of
th
e
s
ol
ut
io
n
s
of
th
e
B
S
S
p
r
obl
e
m
.
I
t
de
pe
nds
on
th
e
obj
e
c
ti
ve
f
unc
ti
on
a
nd
th
e
opt
im
iz
a
ti
on
m
e
th
od.
O
ne
of
th
e
I
C
A
m
e
th
o
ds
is
th
e
Q
P
S
O
-
ba
s
e
d
I
C
A
,
w
hi
c
h
ga
ve
good
a
c
c
ur
a
c
y
r
e
s
ul
ts
but
it
s
uf
f
e
r
th
e
lo
w
s
pe
e
d
c
onve
r
ge
nc
e
in
t
he
s
e
pa
r
a
ti
on
pr
oc
e
s
s
.
A
not
he
r
m
e
th
od
of
th
e
I
C
A
is
B
ig
r
a
di
e
nt
ne
ur
a
l
ne
twor
k
m
e
th
od,
w
hi
c
h
w
a
s
f
a
s
te
r
th
a
n
Q
P
S
O
a
nd
F
a
s
tI
C
A
m
e
th
ods
in
th
e
s
e
pa
r
a
ti
on
pr
oc
e
s
s
,
but
it
lo
w
e
r
a
c
c
ur
a
c
y
th
a
n
Q
P
S
O
-
ba
s
e
d
I
C
A
m
e
th
od.
I
n
th
is
pa
pe
r
,
th
e
a
ut
hor
pr
opos
e
d
ne
w
hybr
id
m
e
th
od
c
ol
le
c
t
th
e
a
dva
nt
a
ge
s
of
bot
h
Q
P
S
O
a
nd
B
ig
r
a
di
e
nt
m
e
th
ods
.
T
he
pr
opos
e
d
h
ybr
id
m
e
th
od
ga
ve
good
a
c
c
ur
a
c
y
s
e
p
a
r
a
ti
on
r
e
s
ul
ts
,
a
t
s
a
m
e
ti
m
e
it
c
ons
um
e
d
lo
w
c
om
put
a
ti
on
ti
m
e
und
e
r
s
a
m
e
s
e
pa
r
a
ti
on
c
ondi
ti
ons
.
A
ll
m
e
th
ods
(
pr
opos
e
d
m
e
th
od
a
nd
ot
he
r
m
e
th
od)
e
va
lu
a
te
d
unde
r
s
om
e
obj
e
c
ti
ve
m
e
a
s
ur
e
m
e
nt
s
a
s
S
N
R
,
S
D
R
,
A
V
C
C
,
a
nd
c
om
put
a
ti
on
ti
m
e
.
A
ls
o,
th
e
pr
opos
e
d
hybr
id
m
e
th
od
c
om
pa
r
e
d
w
it
h
ot
he
r
m
e
th
ods
a
s
Q
P
S
O
-
ba
s
e
d
I
C
A
,
s
ta
nd
a
r
d
F
a
s
tI
C
A
,
a
nd
B
ig
r
a
di
e
nt
ne
ur
a
l
ne
th
w
or
k.
A
ll
th
e
s
e
m
e
th
ods
ope
r
a
te
w
it
h
li
ne
a
r
in
s
ta
nt
a
ne
ous
m
ix
tu
r
e
o
f
m
ono
-
s
pe
e
c
h
s
ig
na
ls
,
a
nd
e
xe
c
ut
e
s
w
it
h
e
ig
ht
s
ig
na
ls
unde
r
g
a
us
s
ia
n di
s
tr
ib
ut
io
n a
nd 8
-
K
H
z
f
r
e
que
nc
y.
R
E
F
E
R
E
N
C
E
S
[1]
A.
Hyvärinen
and
E.
Oja,
“Independent
component
analysis:
Algorit
hms
and
applications,”
Neural
Networks
,
vol
.
13, no. 4
–
5, 2000, doi:
10.1016/S0893
-
6080(00)00026
-
5.
[2]
M.
R.
Mohebian,
H.
R.
Marateb,
S.
Karimimehr,
M.
A.
Mañanas,
J.
Kranjec,
and
A.
Holobar,
“Non
-
invasive
decoding
of
the
motoneurons:
A
guided
source
separation
method
ba
sed
on
convolution
kernel
compensation
with
clustered
ini
tial points,”
Front. Comput. Neurosci.
, vol. 13, 2019, doi: 10.3389/fncom.2019.00014.
[3]
D.
G.
Fantinato,
L.
T.
Duarte,
Y.
Deville,
R.
Attux,
C.
Jutten,
and
A.
Neves,
“A
second
-
order
statistics
method
for
blind
source
separation
in
post
-
nonlinear
mixtures,”
S
ign
al
Processing
,
vol.
155,
2019,
doi
:
10.1016/j.sigpro.2018.09.031.
[4]
J.
G.
Webster
and
Y.
Deville,
“Blind
Source
Separation
and
Blin
d
Mixture
Identification
Methods,”
in
Wiley
Encyclopedia of Electrical and Electronics Engi
neering
, 2016
,
doi
:
10.1002/047134
608X.W8300
.
[5]
A.
Hyvärinen,
“Fast
and
robust
fixed
-
point
algorithms
for
independe
nt
component
analysis,”
IEEE
Trans.
Neural
Networks
, vol. 10, no. 3, 1999, doi: 10.1109/72.761722.
[6]
A.
Tharwat,
“Independent
component
analysis:
An
introduction
,”
Appl.
Comput.
I
nformatics
,
2018,
doi:
10.1016/j.aci.2018.08.006.
[7]
N.
A.
M.
Abbas
and
H.
M.
Salman,
“Independent
component
analysis
based
on
quantum
particle
swarm
optimization,”
Egypt. Informatics J.
, vol. 19, no. 2, 2018, doi: 10.1016/j.eij.2017.11.001.
[8]
P.
Bofill
and
M.
Zibulevsky,
“Underdetermined
blind
source
sep
aration
using
sparse
representations,”
Signal
Processing
, vol. 81, no. 11, 2001, doi: 10.1016/S0165
-
1684(01)00120
-
7.
[9]
R.
Gribonval,
L.
Benaroya,
E.
Vincent,
and
C.
Févotte,
“Propos
als
for
performance
measurement
in
sourc
e
separation,”
Proc. Int. Symp. Indep. Compon. Anal. Bli
nd Signal Sep.
, 2003
,
https://hal.inria.fr/inria
-
00570123
.
[10]
D.
H.
Kaye
and
D.
a
Freedman,
“Reference
Guide
on
Statistics,”
Ref.
Man.
Sci.
Evid.
,
2011
,
https://ssrn.com/abstract=
2705655
.
[11]
J.
Ka
rhunen,
E.
Oja,
L.
Wang,
R.
Vigário,
and
J.
Joutsensal
o,
“
A
class
of
neural
networks
for
independen
t
component
analysi
s,”
IEEE Trans
. Neural
Networks
, vol. 8, no. 3, 1997, doi: 10.1109/72.572090.
[12]
J.
Karhunen,
“Neural
Appro
aches
to
Indep
endent
Component
Ana
lysis
and
Source
Separation,”
1996
,
https://ci.nii.ac.jp/naid/10014609286
.
[13]
L.
T.
Le,
H.
Nguyen,
J.
Dou,
and
J.
Zhou,
“A
comparative
study
of
PSO
-
ANN,
GA
-
ANN,
ICA
-
ANN,
and
ABC
-
ANN
in
estimating
the
heating
load
of
buildings’
energy
efficiency
f
or
smart
city
planning,”
Appl.
Sci.
,
vol.
9,
no.
13, 2019, doi: 10.3390/app9132630.
[14]
H.
M
.
Salman
and
N.
A.
M.
Abbas,
“Compar
ative
Study
of
QPSO
an
d
other
methods
in
Blind
Source
Separation,”
in
Journal
of Phys
ics: Con
ference Ser
ies
, 202
1
, vol.
1804
, no.
1
, doi: 10.1088/1742
-
6596/1804/1/012097.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
.
10
, N
o.
2, J
une
20
21
:
355
–
364
364
[15]
N.
Abbas
and
H.
Salman,
“Enhancing
Linear
Indepen
dent
Component
Analysis:
Comparison
of
Various
Metaheuristic Me
thods,”
Iraqi J. El
ectr. Electr
on. Eng.
, vol. 16, no. 1, 2020, doi: 10.37917/ijeee.16.1.14.
[16]
J. Sun, C. H. Lai, and
X. J. Wu,
Particle
swarm optimisati
on: Classica
l and quantum perspectives
. 201
6
,
ISBN:978
-
1
-
4398
-
3576
-
0
.
[17]
H.
Li,
Z.
Li,
and
H.
Li,
“A
blind
source
separation
algorithm
based
on
dynamic
niching
particle
swarm
optimization,” in
MATEC Web o
f Confer
ences
, 2016, vol. 61, doi: 10.1051/matecconf/20166103008.
[18]
W.
Fang,
J.
Sun,
Y.
Ding,
X.
Wu,
and
W.
Xu,
“A
review
of
quantum
-
behaved
particle
swarm
optimization,”
IETE
Tech.
Rev.
(Instit
ution El
ectron
. Telec
ommun.
Eng. I
ndia)
, vol. 27, no. 4, 2010, doi: 10.4103/0256
-
4602.64601.
[19]
A.
K.
M.
Al
-
Qurabat
and
A.
K.
Idrees,
“Two
level
data
aggregatio
n
pro
tocol
for
prolonging
lifetime
of
periodic
sensor networks,”
Wirel. Net
works
, vol. 25, no. 6, 2019, doi: 10.1007/s11276
-
019
-
01957
-
0.
[20]
T.
Halperin,
A.
Ephrat,
and
Y.
Hoshen,
“Neural
separation
of
obs
erved
and
unobserved
distributions,”
in
36th
Internati
onal
C
onference
on
Machine
Learning,
ICML
2019
,
2019,
vol.
2019
,
http://proceedings.mlr.press/v97/halperin19a.html
.
[21]
L.
Wang,
J.
Karhunen,
and
E.
Oja,
“Bigradient
optimization
ap
proach
for
robust
PCA,
MCA,
and
source
separation,”
in
IEEE
International
Conference
on
Neural
Networks
-
Conferenc
e
Procee
dings
,
1995,
vol.
4,
doi:
10.1109/icnn.1995.488872.
[22]
L.
Wang
and
J.
Karhunen,
“A
unified
neural
bigradient
algorithm
for
robust
PCA
and
MCA.,”
Int.
J.
Neural
Syst.
,
vol. 7, no. 1, 1996, doi: 10.1142/S0129065796000063.
[23]
C
.
Pehlevan,
S.
Mohan,
and
D.
B.
Chklovskii,
“Blind
nonnegativ
e
source
separation
using
biological
neura
l
networks,”
Neural Comp
utation
, vol. 29, no. 11. 2017, doi: 10.1162/NECO_a_01007.
[24]
T. Isomura and T. Toyoizumi,
“Multi
-
context
blind
source separ
ation
by error
-
gated Hebbian rule,”
Sci. Rep.
,
vol. 9,
no. 1, 2019, doi: 10.1038/s41598
-
019
-
43423
-
z.
[25]
A.
Brendel
and
W.
Kellermann,
“Faster
IVA:
Update
rules
for
indepe
ndent
vector
analysis
based
on
negentropy
and
the Majoriz
e
-
Minimize principle,”
arXiv
. 2020
,
https://arxiv.org/abs/2003.09531
.
Evaluation Warning : The document was created with Spire.PDF for Python.