Internati
o
nal
Journal of Ele
c
trical
and Computer
Engineering
(IJE
CE)
V
o
l.
6, N
o
. 1
,
Febr
u
a
r
y
201
6,
pp
. 25
7
~
26
7
I
S
SN
: 208
8-8
7
0
8
,
D
O
I
:
10.115
91
/ij
ece.v6
i
1.9
107
2
57
Jo
urn
a
l
h
o
me
pa
ge
: h
ttp
://iaesjo
u
r
na
l.com/
o
n
lin
e/ind
e
x.ph
p
/
IJECE
Improved Learning Scheme for C
ogniti
v
e Radio us
ing Artif
i
cial
Neural Network
s
R
i
ta
M
a
haj
a
n, D
e
epak B
a
gai
Electronics
and
Communication Engine
ering D
e
p
a
rtment, PEC U
n
iversity
of Technolog
y
,
Ch
andigarh, India
Article Info
A
B
STRAC
T
Article histo
r
y:
Received Sep 30, 2015
Rev
i
sed
No
v 1, 201
5
Accepted Nov 18, 2015
The futur
e
of w
i
rel
e
s
s
s
y
s
t
em
i
s
facing
the pro
b
lem
of s
p
ectru
m
s
carci
t
y
.
Num
b
er of us
ers is
increas
ing r
a
pidl
y but av
ail
a
ble s
p
ectrum
is
l
i
m
ited. Th
e
Cognitive Radio
(CR) network technolog
y
can
en
able the unli
cens
e
d users t
o
s
h
are the freque
nc
y
s
p
ec
trum
with the li
cens
e
d
us
ers
on a dy
n
a
m
i
c bas
i
s
without creating
an
y
interfer
e
nce to pr
imar
y
user. Whenever secondar
y
user
finds that prim
a
r
y
user is not t
r
ansm
itting and
channe
l is free
t
h
en it uses
channel opport
unisticall
y
. In t
h
is paper cogn
itive
radio wi
t
h
predictiv
e
capab
ility
using
artif
icial neur
al
netw
ork has been proposed. The advantag
e
of such cognitiv
e user is saving of tim
e and energ
y
for spe
c
tru
m
sensing.
Proposed radio will sense onl
y
t
h
at chann
e
l whi
c
h is predicted t
o
be free an
d
channe
l is
s
e
lec
t
ed on the bas
i
s
of m
a
xim
u
m vacant t
i
m
e
. P
e
rfo
rm
ance has
been ev
alua
ted i
n
the term
of m
ean s
quare
error
.
The r
e
s
u
lts
s
how that this
learn
i
ng cap
ability
can be embedde
d in secondar
y
users
for better
performance of f
u
ture
wir
e
less technologies.
Keyword:
Artificial n
e
u
r
al n
e
two
r
k
s
Co
gn
itiv
e rad
i
o
n
e
two
r
k
s
Dynam
i
c spectrum
access
W
i
reless co
mm
u
n
i
catio
n
Copyright ©
201
6 Institut
e
o
f
Ad
vanced
Engin
eer
ing and S
c
i
e
nce.
All rights re
se
rve
d
.
Co
rresp
ond
i
ng
Autho
r
:
R
i
t
a
M
a
haja
n,
Depa
rt
em
ent
of El
ect
r
oni
cs
a
n
d
C
o
m
m
uni
cat
i
on E
n
gi
nee
r
i
n
g
,
PEC Un
iv
er
sity o
f
Techno
logy,
Sect
or
1
2
,
C
h
a
ndi
gar
h
,
I
ndi
a
Em
a
il: rita
mah
a
j
a
n@p
ec.ac.i
n
1.
INTRODUCTION
M
ode
rn a
dva
n
c
em
ent
s
i
n
wi
rel
e
ss equi
pm
ent
s
l
ead t
o
dat
a
t
r
ansfe
r
capa
c
i
t
y
of 1-1
0
M
bps
. I
n
t
h
e
upc
om
i
ng t
i
m
e
s, t
h
i
s
ca
n be
up
g
r
ade
d
t
o
nearl
y
1
0
0
M
b
ps a
nd t
h
en t
o
1G
b
p
s i
n
t
h
e
next
t
e
n y
ear
s. Thi
s
cap
ab
ility to
han
d
l
e
d
a
ta at very h
i
gh
sp
eed will en
ab
le
u
s
ers to
easily han
d
l
e
h
i
gh
reso
lu
tion
im
ag
es, h
i
gh
q
u
a
lity aud
i
o an
d v
i
d
e
o.
Ho
wev
e
r, it’s
often ov
erl
o
ok
ed
that as h
i
g
h
p
e
rfo
rm
an
ce wi
reless d
a
ta serv
ices are
wid
e
ly d
e
p
l
o
y
ed
,
d
e
ficien
cy
o
f
ad
d
ition
a
l freq
u
e
n
c
y sp
ectru
m
will b
eco
m
e
a v
e
ry seriou
s li
m
ita
tio
n
.
Th
e
pr
o
b
l
e
m
i
s
, fre
que
ncy
al
l
o
cat
i
on i
s
fi
xe
d a
n
d
i
s
do
ne
by
co
or
di
nat
i
o
n
bet
w
een
co
u
n
t
r
i
e
s
[1]
.
The
p
r
o
b
l
e
m
of
deficiency
of
spectrum
can be obse
rve
d
from
US
Fre
q
uency Allocation
c
h
art
[2].
Recent researc
h
on
spectrum
usage shows t
h
at the
m
a
xim
u
m
usage i
s
o
n
l
y
6%
as sh
ow
n i
n
F
i
gu
re 1
[3]. T
h
e inefficie
n
cy in the
spectrum
usage leads t
o
a
ne
w technique t
o
access wi
re
less
fre
quency
spe
c
trum
opportunistically [4].
Th
e
u
n
d
e
ru
tilizatio
n
o
f
elect
ro
m
a
g
n
e
tic spectru
m
lead
s us to
th
i
n
k in
term
s o
f
spectrum
h
o
l
es, for
wh
ich
th
e fo
llowing
d
e
fi
n
itio
n h
a
s b
e
en
o
f
fered
:
“A s
p
ect
r
u
m
hol
e i
s
a
ba
nd
o
f
f
r
e
que
nci
e
s a
ssi
gne
d t
o
a
p
r
im
ary
user,
b
u
t
, at
a pa
rt
i
c
ul
ar t
i
m
e and
sp
ecific
g
e
ograp
h
i
c l
o
cation
,
th
e
b
a
nd
is
no
t b
e
i
n
g u
tilized
b
y
th
at
u
s
er”. Th
e i
n
efficien
cy in
th
e spectru
m
usa
g
e neces
sitates a new c
o
mmunication paradi
gm
to expl
oit the existing wireless
spec
trum
opport
uni
stically
[2-3
]. Cogn
itiv
e rad
i
o
op
eratio
n
can
b
e
d
i
v
i
d
e
d
in
to
t
h
ree ph
ases: rad
i
o-scen
e analysis, ch
an
n
e
l-state
esti
m
a
t
i
o
n
an
d con
f
i
g
uration
selectio
n
.
It can
b
e
con
s
id
ered
as on
e
of th
e in
tellig
en
t
freq
u
e
n
c
y
reu
s
e sch
e
m
e
[5]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
6, No
. 1, Feb
r
uar
y
20
1
6
:
25
7 – 26
7
25
8
Fig
u
re
1
.
View of th
e actu
a
l av
ailab
ility o
f
sp
ectru
m
.
A f
r
eq
ue
ncy
s
p
ect
r
u
m
i
s
assigne
d
on
pay
m
ent
basi
s t
o
a n
u
m
b
er of
Pri
m
ary
Use
r
s. Sec
o
n
d
a
r
y
Use
r
s
can
scan for unocc
u
pied
c
h
a
nnels (spect
rum
holes) w
ith
in
th
e sam
e
sp
ectru
m
b
a
nd a
n
d comm
unicate in that
band. The
way
to detect
hole
s
in a s
p
ectrum
is channe
l
-
b
y
-
cha
nnel
sca
n
ni
n
g
.
[6]
.
CR
can also be
termed as
soft
ware
-controlled ra
dio tha
t
can
sense the
environm
ent
and a
d
just its
param
e
ters accordingly on run tim
e
[7
]. Th
ese add
itio
n
a
l cog
n
i
t
i
v
e
cap
a
b
ilities in
software
d
e
fin
e
d
rad
i
o
s
are prov
id
ed b
y
th
e co
gn
itiv
e en
g
i
n
e
(CE) a
s
s
h
o
w
n
in Fi
gu
re
2.
T
h
e CE
hel
p
s th
e so
ftwa
re
-d
ef
i
n
ed r
a
d
i
o to
adj
u
st th
e
p
a
r
a
meter
s
b
a
sed upo
n th
e
k
now
ledg
e b
a
se.
Fig
u
re
2
.
CR as software
d
e
fi
n
e
d rad
i
o with
in
tellig
en
ce
In
t
h
is p
a
p
e
r
we pro
p
o
s
e a tech
n
i
q
u
e
fo
r in
cu
lcating
th
e in
tellig
en
ce in
cogn
itiv
e
rad
i
o
u
s
i
ng
Artificial Ne
ural Network. The ANN
will pre
d
ict the cha
nnels
which
will
be vacant i
n
future ba
sed on t
h
e
p
r
ev
iou
s
h
i
story o
f
ch
an
n
e
l st
ates. It h
e
l
p
s t
h
e co
gn
itiv
e
user to
sen
s
e only th
o
s
e ch
annels an
d will sav
e
its
t
i
m
e
and e
n
er
g
y
.
The structure
of the
pape
r is as
fo
llo
ws. In
sectio
n
II a rev
i
ew
o
f
Artificial Neu
r
al Network
is
prese
n
ted and
related work is
projected i
n
s
ection III
and
th
e system
m
o
d
e
l is d
i
scu
ssed
in
Sectio
n IV, and
the architecture and traini
ng
proce
d
ure
of
th
e p
r
op
osed
netw
or
k
fo
r
is pr
es
en
ted
in
Sectio
n
V. Th
e resu
lts o
f
sim
u
l
a
t
i
on i
n
t
e
rm
s of t
h
e
ac
curacy
of
t
h
e
p
r
o
p
o
sed
l
ear
ni
ng
sc
hem
e
are di
scuss
e
d
i
n
S
ect
i
on
VI
. T
h
e
n
i
n
t
h
e
last sectio
n
con
c
lu
si
o
n
is presen
ted.
2.
ARTIF
ICI
A
L
NEU
R
A
L
NETWOR
K
An
art
i
f
i
c
i
a
l
n
e
ural
net
w
or
k
(A
N
N
) c
o
nsi
s
t
s
of
seve
ral
pr
ocessi
ng
u
n
i
t
s
, cal
l
e
d ne
u
r
ons
.
W
i
t
h
i
n
neu
r
al
net
w
o
r
k t
h
ree t
y
pe
s of
neu
r
ons a
r
e
prese
n
t
(s
ho
w
n
i
n
Fi
g
u
r
e 3
)
:
i
n
p
u
t
,
hi
dde
n
and
out
put
ne
ur
o
n
s.
Each c
o
n
n
ect
i
on i
s
defi
ned
by
a wei
g
ht
,
w
jk
, which
det
e
rm
ine the effect that
th
e sign
al of
n
e
u
r
on
j
h
a
s on
neu
r
on
k.
Eac
h
ne
ur
on
ha
s a s
t
at
e of act
i
v
at
i
o
n
,
be i
t
y
k
, called output
of t
h
e
neur
on. During
proces
sing, each
neuron k
receives input s
k
fr
om
(a) nei
g
h
b
o
rs
bel
o
ngi
ng
t
o
di
ffe
re
nt
lay
e
rs, an
d fr
om
(b) e
x
tern
al sou
r
ces
cal
l
e
d bi
as of
fs
et
b
k
, and
uses
t
h
em
t
o
com
put
e an up
dat
e
d l
e
vel
of act
i
v
at
i
on y
k
. T
h
i
s
i
s
done
by
t
h
e use
of a
n
act
i
v
at
i
on fu
nc
t
i
on
F
k
as follo
ws (sh
o
w
n
in F
i
gu
re 4)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Imp
r
o
ved Lea
r
n
i
ng
S
c
h
e
me f
o
r Cog
n
itive Rad
i
o
u
s
i
n
g Artificia
l Neu
r
a
l
Netwo
r
ks
(Rita
Ma
ha
jan
)
25
9
Fi
gu
re
3.
Ty
pi
cal
neu
r
al
net
w
or
k a
r
chi
t
ect
u
r
e
Fi
gu
re
4.
Ne
ur
on
st
r
u
ct
u
r
e
Weigh
t
ed
su
m
o
f
th
e i
n
pu
ts co
m
i
n
g
fro
m
th
e ou
tpu
t
s
o
f
n
e
u
r
on
s in th
e
prev
iou
s
layer is
S
k
an
d i
s
gi
ve
n
by
_
.
Expressi
on for
these tra
n
sfe
r
fu
n
c
tion
s
is as
fo
llo
ws:
1.
y
k
= F
k
(
_
) =
_
--
--
-- linea
r tr
a
n
sfe
r
fu
nctio
n.
2.
y
k
= F
k
(
_
) =
_
--
--
--
l
ogistic-s
igm
o
id.
3.
y
k
= F
k
(
_
) =
_
1
--
--
--
hy
pe
r
bol
i
c
t
a
nge
nt
si
gm
oi
d.
In
th
is p
a
p
e
r, we u
s
e Mu
ltilayer Percep
t
r
on
(MLP),
a feed
fo
rward
structu
r
e, wh
ich
has b
een
used
wi
del
y
i
n
t
i
m
e
seri
es predi
c
t
i
on a
nd bi
nary
pre
d
i
c
t
i
o
n
.
It
i
s
appa
rent
t
h
at
NN has t
o
be
t
r
ai
ned s
u
ch t
h
at
t
h
e
appl
i
cat
i
o
n o
f
a set
of
i
n
put
s
pr
o
duce
s
t
h
e
desi
re
d set
of
out
put
s
.
T
h
i
s
p
r
oces
s i
s
cal
l
e
d l
ear
ni
n
g
or t
r
ai
ni
n
g
and ca
n
be ac
hieved by
prop
erly adj
u
stin
g th
e
weigh
t
s
w
jk
o
f
t
h
e c
o
n
n
ect
i
ons
am
ong
al
l
(
j
,
k
)
ne
ur
on
pai
r
s.
Trai
ni
n
g
a
d
ju
st
s co
nnect
i
on
w
e
i
ght
s t
o
p
r
od
u
ce t
h
e
desi
re
d
out
put
w
h
i
c
h c
a
n
be ac
hi
eve
d
by
wei
g
ht
-cha
ngi
n
g
proce
d
ure
of the connecti
o
n is term
ed
as the
backpropa
g
a
tio
n learn
i
ng
ru
l
e
.
3.
RELATED WORK
In
literature [8-9
] au
t
h
ors
d
i
scu
sses th
e
ro
le o
f
learn
i
ng
i
n
CRs and
emp
h
a
sizes ho
w
cru
c
ial th
e
in
d
e
p
e
nd
en
t learn
i
n
g
ab
ility in
realizin
g a
real CR d
e
v
i
ce.
Th
ey
h
a
v
e
p
r
esen
ted a su
rv
ey
o
f
th
e
state-of t
h
e art
achi
e
vem
e
nt
s i
n
appl
y
i
n
g
m
achi
n
e l
earni
ng t
ech
ni
que
s. T
h
ey
have al
s
o
re
vi
ewe
d
seve
r
a
l
C
R
i
m
p
l
e
m
en
tatio
n
s
th
at are u
s
ed
as th
e v
a
riou
s artificial in
tellig
en
ce tech
niq
u
e
s. It h
a
s been
con
c
lud
e
d in
[8
]
t
h
at
t
h
ey
do n
o
t
pr
ovi
de any
m
eans of l
earni
n
g
fr
om
pa
st experi
e
n
ces, t
hus
fai
l
i
ng t
o
exhi
bi
t
one o
f
t
h
e key
pr
o
p
ert
i
e
s of
C
R
.
I
n
[
10-
12
]
au
t
h
or
s h
a
v
e
pr
op
o
s
ed
Elm
a
n
Recu
rr
en
t
Neu
r
al
Netwo
r
k
s
to
pred
ict the sp
ectru
m
occu
pa
ncy
by
m
odel
i
ng R
F
t
i
m
e
seri
es. In t
h
i
s
pa
per
[1
3]
,
aut
h
ors
ha
ve
desi
g
n
t
h
e c
h
a
nnel
st
at
us
p
r
e
d
i
c
t
o
r
u
s
ing
th
e n
e
ural n
e
twork
m
o
d
e
l
m
u
ltilayer
p
e
rcep
tro
n
(M
LP).
In
co
gn
itiv
e rad
i
o
n
e
t
w
ork
s
, it is n
o
t
p
o
ssib
l
e
t
o
obt
ai
n t
h
e d
a
t
a
of cha
n
nel
usa
g
e by
t
h
e p
r
i
m
ary
users. The ad
va
nt
age o
f
ne
ural
net
w
o
r
ks i
s
t
h
at
i
t
do
es n
o
t
req
u
i
r
e a p
r
i
o
r
i
kno
wl
e
dge
o
f
t
h
e u
nde
rl
y
i
n
g
di
st
ri
but
i
o
ns
of t
h
e
obse
r
v
e
d p
r
oce
ss. S
o
, t
h
e ne
ural
ne
t
w
o
r
ks
are the
good choice for m
odeling th
e c
h
a
nnel
st
at
us p
r
e
d
i
c
t
o
r
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
6, No
. 1, Feb
r
uar
y
20
1
6
:
25
7 – 26
7
26
0
In [1
4]
aut
h
or
s
have
a
n
al
y
zed
vari
o
u
s bui
l
d
i
ng bl
oc
ks of D
S
A
o
p
e
r
at
i
on whi
c
h
can be
i
n
co
r
p
o
r
at
ed
i
n
t
o
f
u
t
u
re L
TE st
an
dar
d
s.
They
have
al
so di
sc
u
ssed th
e op
eration
a
l sign
alin
g
scen
ari
o
s to su
ppo
rt
coope
r
ative se
nsing techniques, coor
dinati
on and m
onitori
ng
of freque
nc
y access rules
.
The a
u
thors i
n
[15]
utilize the fra
m
ework
pre
s
e
n
ted i
n
[14] and int
r
oduce
the
opport
unistic spectrum
access in LTE-A
network.
Their
work illustrates the a
d
option
of a
geo-location
database withi
n
LTE-A net
w
ork that gat
h
e
r
s the
su
ppo
rtiv
e inform
at
io
n
fro
m
t
h
e CR
u
s
ers
with
its n
e
i
g
hbo
ri
n
g
env
i
ron
m
en
t.
Man
y
researchers h
a
v
e
app
lied
v
a
riou
s artificial
n
e
u
r
al network
s
in
cog
n
itiv
e rad
i
o
an
d
ach
i
ev
ed
good results because of
their
sp
ecial c
h
aracteristics.
Baldo a
n
d Z
o
rzi [17] propos
ed
a
m
u
ltilayered
feedforward
neu
r
al n
e
t
w
ork
for p
e
rfo
r
m
a
n
ce o
f
real ti
m
e
co
mm
u
n
i
catio
n
for cogn
itiv
e rad
i
o
n
e
two
r
k. Th
e
fun
c
tion
ap
prox
im
a
tio
n
cap
a
b
ility o
f
MFNN is
u
s
ed
t
o
perfo
r
m
a
n
ce characterizatio
n
o
f
th
e cogn
itiv
e rad
i
o
syste
m
. Th
e co
gn
itiv
e
u
s
er
sh
ou
l
d
b
e
ab
l
e
to
g
a
th
er the relev
a
n
t
en
viron
m
en
t p
a
rameters wh
ich
h
a
v
e
si
gni
fi
ca
nt
i
m
pact
on
i
t
s
pe
rf
o
r
m
a
nce.
In
pap
e
r [1
8
]
au
tho
r
s
h
a
v
e
desig
n
e
d
a n
e
u
r
al n
e
twork
for d
ecisio
n
m
a
k
i
n
g
o
f
Cogn
itiv
e Eng
i
n
e
,
wh
ich
is b
a
sed o
n
ev
al
u
a
tio
n, an
d
learn
i
n
g
.
Th
is m
o
d
e
l
o
f
co
gn
itiv
e en
g
i
n
e
is co
m
p
ared with
Reiser’s
m
o
d
e
l
whi
c
h i
s
base
d
on
o
n
l
y
u
n
c
han
g
ea
bl
e i
n
f
o
rm
at
i
on [
19]
.
A
n
A
d
a
p
t
i
v
e
R
e
so
nance
T
h
eo
ry
AR
T
2
neu
r
al
n
e
two
r
k
h
a
s been
pro
p
o
s
ed
fo
r sp
ectru
m
se
n
s
ing
in
p
a
p
e
r [20
]
. Th
is
n
e
u
r
al
n
e
two
r
k
satisfies th
e cog
n
itive
W
i
reless Mesh Network stru
ctu
r
e
wh
ich
com
b
in
es with
si
g
n
a
l
bro
a
d
cast
syste
m
.
4.
SYSTE
M
MO
DEL
Our propose
d
syste
m
m
odel
fo
r future
wireless com
m
unication
net
w
ork (LTE- Adva
nced (5G))
co
nsists o
f
Dy
nam
i
c spectrum access with
additio
nal capability of prediction for
secondary users as shown in
Figure
5. T
h
es
e secondary
us
ers a
r
e called
cognitive
use
r
s with pre
d
ic
tion (C
UP
).
Norm
al cognitive
use
r
s
(CU) se
nse all
cha
nnels
and
find the
spect
rum
hole for
their use
.
T
h
e a
dva
ntage
of C
U
P
over CU i
s
that
form
er senses
only those c
h
a
nnels
which a
r
e pre
d
icted
t
o
be
vacant a
n
d
saves tim
e and energy
for s
p
e
c
trum
sensing. CUP will select
the vacant c
h
annel
which
has be
e
n
free
for longer ti
m
e
. Th
is is
done by additiona
l
equi
pm
ent in norm
al CU called
Artific
ial Neural
Network. It
will pre
d
ict
the vacant c
h
a
nnel
and its i
d
le ti
m
e
base
d o
n
t
h
e
p
r
evi
ous
hi
st
ory
.
T
h
ere
f
o
r
e ea
ch C
U
P
need
s
t
o
m
a
i
n
t
a
i
n
t
h
e dat
a
ba
se
of st
at
us o
f
t
h
e c
h
a
nnel
s
an
d
th
is informatio
n
can
b
e
u
s
ed
to
train th
e ANN in
co
n
t
ro
l un
it. Th
e p
e
rfo
r
m
a
n
ce o
f
th
is m
o
d
e
l will
in
crease
sign
ifi
can
tly o
v
e
r t
h
e ex
isting
m
o
d
e
l.
Fi
gu
re 5.
Sce
n
ari
o
of
f
u
t
u
re
L
TE-
A wi
rel
e
ss net
w
or
k
In LT
E-
A any
user e
qui
pm
ent
can se
nse t
h
e spect
rum
i
n
al
l
o
t
t
e
d sense
t
i
m
e
sl
ot
out
of a
v
ai
l
a
bl
e
ti
m
e
slo
t
[1
6
]
. Acco
rd
ing
to th
e stan
d
a
rd
,
measu
r
em
en
t
g
a
p
leng
th
(M
GL) is reserv
ed
for ex
tracting
th
e
sam
p
les fro
m
a certain
b
a
ndwid
th
and
is fix
e
d
for 6
m
i
l
liseco
nd
s.
Durin
g
th
e m
easurem
en
t g
a
p
s
th
e u
s
er
equi
pm
ent shall not transm
it any data and is not exp
ected to tune its receiver
on any carrier frequency.
Ano
t
h
e
r term
measu
r
em
en
t g
a
p
rep
e
titio
n
period
(M
GRP)
d
e
fi
n
e
d
i
n
th
e
sam
e
s
t
an
d
a
rd
as th
e tim
e a
llo
tted
to
sense a
n
d tra
n
s
m
ission, a
n
d is
fixe
d
for
40 m
illiseconds
shown in
Figure
6.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Imp
r
o
ved Lea
r
n
i
ng
S
c
h
e
me f
o
r Cog
n
itive Rad
i
o
u
s
i
n
g Artificia
l Neu
r
a
l
Netwo
r
ks
(Rita
Ma
ha
jan
)
26
1
Fig
u
re
6
.
Measu
r
em
en
t g
a
p
rep
e
titio
n
p
e
ri
o
d
5.
AR
CHITE
C
T
URE
A
N
D
T
RAI
NI
NG
OF
PRE
D
ICT
O
R
Th
e
n
e
twork used
h
e
re is a
m
u
l
tilayered
stru
ct
u
r
e wh
ich
co
nsists of an
i
n
pu
t layer, an
o
u
t
p
u
t
layer,
and
o
n
e
or
m
o
re
hi
d
d
e
n
l
a
y
e
rs s
h
o
w
n i
n
Fi
gu
re
7.
Th
e in
pu
t layer
has lin
ear t
r
ans
f
er function where as
hi
d
d
en
an
d
o
u
t
put
l
a
y
e
rs
ha
ve
hy
pe
rb
ol
i
c
t
a
n
g
ent
si
gm
oi
d f
unct
i
o
n
(
d
i
s
cus
s
ed i
n
sect
i
o
n I
I).
The
t
o
t
a
l
n
u
m
ber
o
f
inp
u
t
s i
n
th
e in
pu
t layer is
called
th
e
o
r
d
e
r
o
f
t
h
e MLP network and
is
den
o
t
ed
b
y
τ
.
Fi
gu
re 7.
A
r
chi
t
ect
ure of ne
ur
al
net
w
or
k
Du
rin
g
first p
h
ase of
fee
d
f
o
r
w
ar
d,
eac
h
i
n
p
u
t unit (x
i
)
receives
the i
n
put
signal and fee
d
s t
o
all
neu
r
ons i
n
t
h
e
next
hi
dde
n l
a
y
e
r t
h
ro
u
gh
we
i
ght
m
a
t
r
i
x
‘U’
and ‘
u
ih
’ is th
e weig
h
t
fro
m
i
th
uni
t
of i
n
put
l
a
y
e
r
to
h
th
ne
ur
o
n
o
f
fi
rst
hi
dde
n l
a
y
e
r. T
o
t
a
l
i
n
pu
t
fo
r
neu
r
on
Z
h
is calcu
lated
as
_
∑
.
for
all h (
1
t
o
q)
Th
en
activ
ation
s
‘z
h
’
of fi
rst
h
i
dd
en layer are calcu
lated
u
s
in
g
its activ
ati
o
n fun
c
tio
n. Si
milarly th
ese
act
i
v
at
i
ons are
fed t
o
ne
xt
hi
dde
n l
a
y
e
r t
h
r
o
u
g
h
wei
g
ht
m
a
t
r
i
x
‘V
’ an
d ‘
v
hj
’ is th
e weigh
t
fro
m
h
th
n
e
ur
on
of
fi
rst
hi
dde
n l
a
y
e
r t
o
j
th
ne
ur
on
of sec
o
nd
hi
d
d
en l
a
y
e
r
.
Act
i
vat
i
o
n
s
‘zz
j
’
(j=1
to
p
)
are calcu
lated
.
Fi
n
a
lly th
ese
activ
atio
n
s
are
fed to
ou
tpu
t
layer th
rou
g
h
weig
h
t
m
a
trix
‘W’ an
d ‘w
jk
’ is t
h
e weigh
t
fro
m
j
th
n
e
ur
on
o
f
seco
nd
h
i
dd
en
layer
to k
th
neu
r
on
of
out
put
l
a
y
e
r.
The
n
o
u
t
p
ut
s ‘
y
k
’ (k = 1 to m
)
are com
puted. Mean square error is
cal
cul
a
t
e
d usi
n
g t
h
e desi
re
d o
u
t
p
ut
s t
k
and fed bac
k
to the lower layers
.
Weights are modi
fied accordi
ng to
t
h
e g
r
a
d
i
e
nt
de
scent
m
e
t
hod.
Mean
Squ
a
r
e
Er
ro
r is
1
2
∗
= -
(
∗
′
_
∗
_
= -
∗
′
_
∗
Z
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
6, No
. 1, Feb
r
uar
y
20
1
6
:
25
7 – 26
7
26
2
Er
ro
r ter
m
f
o
r
w
e
igh
t
s ‘w’
is
= -(
∗
′
_
C
h
an
ge i
n
wei
ght
s
i
s
∆
(1
)
N
o
w
f
o
r
er
ro
r ter
m
an
d
ch
ange of
w
e
igh
t
s for
‘v’
′
_
_
′
_
∗
′
Er
ro
r ter
m
f
o
r
w
e
igh
t
s ‘v
is
∗
′
C
h
an
ge i
n
wei
ght
s
i
s
∆
(2
)
N
o
w
f
o
r
er
ro
r ter
m
an
d
ch
ange of
w
e
igh
t
s for
‘u’
′
_
_
′
_
′
′
′
′
Er
ro
r ter
m
f
o
r
w
e
igh
t
s ‘u’
is
′
C
h
an
ge i
n
wei
ght
s
i
s
∆
(3
)
New
wei
ght
s
c
a
n
be
obt
ai
n
e
d
usi
n
g e
q
uat
i
o
n
s
(
1
),
(
2
)
a
n
d
(
3
)
f
o
r
al
l
wei
g
ht
s:
weigh
t
(n
ew) =
weigh
t
(o
ld
) +
ch
ang
e
i
n
weigh
t
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Imp
r
o
ved Lea
r
n
i
ng
S
c
h
e
me f
o
r Cog
n
itive Rad
i
o
u
s
i
n
g Artificia
l Neu
r
a
l
Netwo
r
ks
(Rita
Ma
ha
jan
)
26
3
Th
is m
a
th
e
m
a
t
ical
m
o
d
e
l fo
r
b
ackprop
ag
atio
n
trai
n
i
ng
algo
rith
m
is u
s
ed
for ou
r
p
r
op
osed
pred
ict
o
r.
To
pre
d
i
c
t
t
h
e
st
at
us o
f
t
h
e
chan
nel
i
s
de
p
e
nde
nt
o
n
pre
v
i
o
us a
nd c
u
r
r
e
nt
sense
d
ob
s
e
rvat
i
o
ns.
A t
r
ai
ne
d
artificial n
e
ural n
e
twor
k (ANN) is em
b
e
d
d
e
d
in its con
t
ro
l
u
n
it.
In
th
is wo
rk
data is co
n
s
id
ered
as p
a
st ob
serv
ation
s
p
e
r chan
n
e
l i.e. sp
ectru
m
u
n
it. Th
e slo
t
o
f
eight
v
a
lu
es
o
f
ch
ann
e
l statu
s
an
d
th
eir v
acan
t
time is fe
d
to
B
ack
pro
p
a
g
a
tion
Neural n
e
two
r
k
wh
ich
will p
r
ed
ict
t
h
e next
st
at
us
of cha
n
nel
.
St
at
us i
s
ei
t
h
er 1
or -
1
,
1 m
ean
s the channel is busy
or -1 means the cha
nnel is
vacant.
And
va
lue of tim
e varies from
1 to 6 for vaca
nt
cha
nnels
only othe
rwise it is zero. So the net
w
ork has
si
xt
een
ne
ur
o
n
s
at
i
n
put
l
a
y
e
r
an
d t
w
o
ne
ur
o
n
s at
o
u
t
p
ut
l
a
y
e
r. B
e
st
num
ber
of
hi
dde
n l
a
y
e
rs an
d
ne
ur
o
n
s a
r
e
deci
des
usi
n
g
hi
t
an
d t
r
i
a
l
m
e
t
h
o
d
f
o
r t
h
e
best
pe
rf
o
r
m
a
nce. T
o
o sm
all
num
ber
of
n
e
ur
o
n
s i
n
t
h
e
hi
d
d
e
n
l
a
y
e
rs m
a
y
resul
t
i
n
p
o
o
r acc
uracy
of t
h
e re
sul
t
s
.
W
i
t
h
t
o
o
l
a
rge
num
ber
of
ne
ur
on
s i
n
h
i
dde
n l
a
y
e
r,
ne
t
w
o
r
k
will n
o
t
be ab
l
e
to
g
e
n
e
ralize th
e efficien
cy. On
ce th
e arch
itectu
r
e is
d
e
cid
e
d
,
it can
be u
s
ed
in
CUP an
d
p
e
rf
or
m
a
n
ce o
f
n
e
twor
k im
p
r
ov
es si
gn
if
ican
tly.
6.
SIMULATIONS
AN
D R
E
SU
LTS
In th
is section
we
will d
i
scu
ss th
e sim
u
latio
n
s
do
ne an
d an
alyze th
e resu
lts fo
r accuracy.
Fol
l
o
wi
n
g
st
e
p
s ha
ve
bee
n
f
o
l
l
owe
d
:
1.
Sel
ect
i
on
of
t
o
ol
s t
o
ge
nerat
e
t
h
e dat
a
base.
2.
Gen
e
rate th
e datab
a
se
u
s
ing
Po
isso
n d
i
stri
bu
tio
n.
3.
Selectio
n
o
f
too
l
to
sim
u
late
n
e
ural n
e
two
r
k.
4.
Desi
g
n
t
h
e neu
r
al
net
w
o
r
k
ba
sed o
n
num
ber of
i
n
put
,
n
u
m
b
er of o
u
t
p
ut
s.
5.
D
ecid
e
t
h
e
n
u
m
b
er
o
f
h
i
dd
en layer
s
and
t
h
ei
r
n
u
m
b
e
r
o
f
n
e
u
r
on
s at
r
a
ndom
.
6.
An
alyze th
e resu
lt and
red
e
sig
n
th
e
n
e
twork till
m
i
n
i
m
u
m
MSE is ach
i
eved
.
Sim
u
l
a
t
i
ons ha
ve bee
n
carri
e
d
o
u
t
i
n
Neu
r
al
Net
w
o
r
k T
ool
B
ox of M
A
T
L
AB
R
2
0
1
0
of
M
a
t
h
wo
rks
,
Nat
i
c
k, M
A
,
US
A. T
h
e
dat
a
has
been
ge
n
e
rat
e
d i
n
st
at
i
s
t
i
cal
dom
ai
n. The st
oc
hast
i
c
pr
ocesses
,
P
o
i
s
on
an
d
Paret
o
ra
nd
om
are
use
d
t
o
ge
nerat
e
PU
t
r
af
f
i
c and
f
r
ee sl
ot
s for cha
n
nel respectivel
y. Poisso
n pro
cess
i
s
th
e
trad
itio
n
a
l traffic g
e
n
e
rati
o
n
m
o
d
e
l fo
r
ci
rcu
it-switch
e
d
d
a
ta
as well as packet data and num
b
er of
packets pe
r
uni
t
t
i
m
e
sl
ot
fol
l
o
w
s
t
h
e
poi
s
on
di
st
ri
b
u
t
i
o
n.
B
u
t
fo
r m
o
re r
eal
i
s
t
i
c
m
odel
,
Paret
o
di
st
ri
b
u
t
i
on
ca
n be
use
d
as
a
l
o
n
g
-
t
a
i
l
t
r
af
fi
c
m
odel
.
T
r
a
i
ni
ng
of
t
h
e
n
e
ural
net
w
o
r
k
i
s
d
o
n
e
wi
t
h
st
oc
hast
i
c
dat
a
. T
r
ai
ni
n
g
i
n
ne
ural
net
w
or
ks m
eans t
h
e
n
onl
i
n
ea
r
m
a
ppi
n
g
of
i
n
p
u
t
an
d
desi
re
d
out
put
s
.
T
h
i
s
i
s
d
one
by
a
d
ju
st
i
ng
wei
g
ht
s s
o
t
h
at
m
ean squa
re err
o
r i
s
m
i
nim
u
m
.
As num
ber
of i
n
p
u
t
u
n
i
t
s
and n
u
m
b
er of o
u
t
p
ut
ne
ur
ons a
r
e fi
xe
d.
The
num
ber o
f
hi
dde
n l
a
y
e
rs and t
h
ei
r n
u
m
b
er
of ne
u
r
o
n
s
are deci
de
d
by
hi
t
and t
r
i
a
l
m
e
t
hod.
Thr
o
ug
h
si
m
u
latio
n
s
it
h
a
s b
e
en
ob
serv
ed
th
at n
e
t
w
ork
with
two
hi
dde
n l
a
y
e
rs an
d wi
t
h
1
5
ne
ur
ons i
n
fi
rst
l
a
yer an
d
10
i
n
seco
n
d
l
a
y
e
r pe
rf
orm
s
t
h
e
best
i
n
t
e
rm
s o
f
M
S
E
.
Pa
ra
m
e
t
e
rs use
d
f
o
r si
m
u
l
a
t
i
ons a
r
e
gi
ve
n i
n
Ta
b
l
e 1
Tabl
e 1. Param
e
t
e
rs
f
o
r
si
m
u
lat
i
ons
Net T
y
pe
Multilaye
red feedforward network
T
r
aining algor
ith
m
Backpr
opagation
T
r
aining function
T
r
ainlm
T
r
aining data set size
8000sl
o
t (
1000 dat
a
points)
validation data set size
4000sl
o
t (
500 data
points)
No.
of hidden lay
e
r
s
2
No.
of neur
ons in f
i
r
s
t hidden lay
e
r
s
15
No.
of neur
ons in s
econd hidden lay
e
r
s
10
E
pochs
1000
L
ear
ning r
a
te
.
001
Testing a
nd
validation are a
l
so done
using offlin
e
data to validate the
accu
racy of the network.
Trai
ni
n
g
o
f
p
r
edi
c
t
o
r i
s
d
o
n
e
wi
t
h
t
r
ai
ni
n
g
dat
a
set
of 8,
00
0 sl
ot
s i
.
e.
10
00 dat
a
p
o
i
nt
s. Here ei
g
h
t
sl
ot
s
rep
r
ese
n
t
one
dat
a
p
o
i
n
t
(ei
g
ht
cha
nnel
s
i
n
one r
e
so
u
r
ce bl
oc
k)
. Thi
s
m
eans 10
0
0
p
r
evi
ous
val
u
es
of
o
n
e
reso
u
r
ce bl
ock
are fe
d t
o
ne
u
r
al
net
w
or
k f
o
r t
r
ai
ni
n
g
. T
h
e
resul
t
i
s
sh
ow
n aft
e
r
o
n
e t
h
o
u
sa
nd e
poc
hs
.
It
has
b
een ob
ser
v
ed
af
ter
tr
ai
n
i
ng
t
h
e m
ean
squ
a
re er
ro
r is ex
t
r
emely lo
w
.
Fo
r
clar
ity statu
s
an
d fr
ee tim
e f
o
r
300
dat
a
p
o
i
n
t
s
ha
v
e
bee
n
pl
ot
t
e
d
i
n
gra
p
hs s
h
ow
n i
n
Fi
g
u
r
e
8 a
n
d
Fi
g
u
r
e
9.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
6, No
. 1, Feb
r
uar
y
20
1
6
:
25
7 – 26
7
26
4
Fi
gu
re
8.
Trai
n
i
ng
o
f
c
h
an
nel
st
at
us f
o
r
3
0
0
dat
a
p
o
i
n
t
s
Figure 9.
Trai
ning of num
be
r
of tim
e
slots for
vacant
c
h
a
n
nel
for 300 data poi
nts
Aft
e
r
t
r
ai
ni
ng
and
t
e
st
i
ng
o
f
t
h
e net
w
o
r
k w
e
nee
d
to app
l
y u
n
s
een
d
a
ta
to
v
a
lid
ate it.
Valid
atio
n of
p
r
ed
icto
r is don
e with
train
i
ng
d
a
ta set o
f
32
00
slo
t
s i.
e. 40
0 dat
a
p
o
i
n
t
s
. R
e
sul
t
s
fo
r st
at
us of cha
n
n
e
l
and
num
ber
of
i
d
l
e
sl
ot
s are
sh
o
w
n i
n
Fi
g
u
r
e
10
and
Fi
g
u
r
e
11
.
Fig
u
re
10
.
v
a
lid
atio
n of ch
ann
e
l statu
s
0
50
10
0
15
0
20
0
25
0
30
0
-1.
5
-1
-0.
5
0
0.
5
1
1.
5
D
a
t
a
se
t f
o
r
tr
a
i
n
i
n
g
s
t
at
us
of
c
h
an
ne
l
t
a
r
g
et
ou
t
p
ut
pred
i
c
t
ed o
u
t
p
u
t
0
50
10
0
15
0
20
0
25
0
30
0
-1
0
1
2
3
4
5
6
7
D
a
t
a
se
t f
o
r
tr
a
i
n
i
n
g
n
u
m
b
e
r o
f ti
m
e
sl
o
ts
fo
r c
h
a
n
n
e
l t
o
b
e
va
ca
n
t
t
a
r
g
et
ou
t
p
ut
p
r
e
d
i
c
t
e
d ou
t
p
ut
0
50
100
150
200
250
300
350
400
-1
.
5
-1
-0
.
5
0
0.
5
1
1.
5
dat
a
s
e
t
f
o
r
v
a
l
i
dat
i
o
n
c
h
ann
el
s
t
at
us
S
t
at
e V
a
l
i
d
a
t
i
on of
pr
edi
c
t
or
out
put
v
a
l
u
e
t
a
r
get
v
a
l
u
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Imp
r
o
ved Lea
r
n
i
ng
S
c
h
e
me f
o
r Cog
n
itive Rad
i
o
u
s
i
n
g Artificia
l Neu
r
a
l
Netwo
r
ks
(Rita
Ma
ha
jan
)
26
5
Fi
gu
re 1
1
. Val
i
d
at
i
o
n
o
f
n
u
m
b
er of
t
i
m
e
sl
ot
s
f
o
r vaca
nt
ch
a
nnel
Figu
re 1
2
.
M
ean
S
q
uare
er
ro
r of netw
o
r
k
The
per
f
o
rm
ance in term
s of
m
ean sq
uare
e
r
r
o
r
(M
SE
) is
sho
w
n in
Fig
u
r
e 1
2
.
It ca
n
b
e
o
b
ser
v
e
d
t
h
at
aft
e
r 1
0
0
0
epoc
hes M
S
E
i
s
of t
h
e o
r
de
r
of 1
0
-12
. G
r
a
d
i
e
nt
dece
nt
i
s
of t
h
e o
r
de
r 1
0
-7
and zer
o
val
i
d
at
i
o
n
failu
re. M
o
m
e
n
t
u
m
co
efficien
t (m
u
)
is th
e fraction
of
prev
iou
s
wei
g
h
t
up
d
a
te to
cu
rren
t weigh
t
, it’s
v
a
lu
e
also very
low (10
-10
)
as s
h
ow
n i
n
Fi
g
u
r
e
13
.
Fi
gu
re 1
3
. Gra
d
i
e
nt
,
m
u
a
n
d
val
i
d
at
i
o
n
fai
l
u
re of net
w
o
r
k
d
u
ri
ng
t
r
ai
ni
ng
0
50
10
0
15
0
200
25
0
30
0
350
40
0
45
0
500
-1
0
1
2
3
4
5
6
7
d
a
ta
s
e
t f
o
r
v
a
l
i
d
a
t
i
o
n
n
u
m
b
e
r o
f t
i
m
e
s
lo
ts
fo
r c
h
a
n
n
e
l t
o
b
e
v
a
c
a
n
t
V
a
l
i
d
at
i
o
n f
o
r
i
d
l
e
t
i
m
e
s
l
ot
s
f
o
r
a c
h
a
nne
l
o
u
tp
u
t
v
a
l
u
e
t
a
r
get
v
a
l
u
e
0
100
200
30
0
400
500
60
0
700
800
900
1000
10
-1
0
10
-8
10
-6
10
-4
10
-2
10
0
B
e
st
T
r
ai
n
i
ng
P
e
r
f
o
r
m
a
n
c
e
i
s
2
.
17
73
e-
1
1
a
t
e
p
och 10
00
M
e
a
n
S
q
u
a
re
d
E
r
ro
r
(
m
se)
10
00
E
p
o
c
h
s
Tr
a
i
n
Be
s
t
10
-10
10
0
10
10
gr
adi
ent
G
r
a
d
i
e
nt
=
2.
450
9e
-
0
7
,
at
epo
c
h
10
00
10
-10
10
-5
10
0
mu
M
u
=
1e
-
1
0
,
a
t
ep
oc
h 1
000
0
100
20
0
30
0
400
50
0
60
0
700
80
0
90
0
10
00
-1
0
1
v
a
l f
a
il
1
0
0
0
E
poc
hs
V
a
l
i
da
t
i
on
C
h
ec
k
s
=
0,
at
epo
c
h
10
00
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
6, No
. 1, Feb
r
uar
y
20
1
6
:
25
7 – 26
7
26
6
Now the
propose
d CUP
will sense only those cha
nnels whic
h are
predicted to
be vacant for
max
i
m
u
m
ti
me
. Th
is will fu
rt
h
e
r m
i
n
i
mize t
h
e switch
i
ng
ov
er th
e ch
an
n
e
l
s
.
7.
CO
NCL
USI
O
N
In t
h
i
s
wo
r
k
p
r
obl
em
of f
u
t
u
r
e
wi
rel
e
ss net
w
o
r
k
has
been
exam
i
n
ed and
new l
ear
ni
ng s
c
hem
e
usi
n
g
art
i
f
i
c
i
a
l
neu
r
a
l
net
w
or
k i
s
p
r
op
ose
d
.
T
h
i
s
net
w
or
k
has
b
een t
r
ai
ned
usi
n
g
st
at
i
s
t
i
cal
pri
m
ary
user da
t
a
fo
r
m
obi
l
e
com
m
u
n
i
cat
i
o
n
an
d
ca
n
be em
bed
d
ed
i
n
c
o
nt
r
o
l
u
n
i
t
o
f
sec
o
nda
ry
users
.
N
o
w
sec
o
n
d
a
r
y
u
s
er
ne
ed t
o
sense
o
n
l
y
t
h
o
s
e cha
nnel
s
w
h
i
c
h a
r
e
pre
d
i
c
t
e
d t
o
be
f
r
ee a
n
d select a
p
propriate cha
nnel
for transm
issio
n
.
This
p
r
op
o
s
ed
techn
o
l
o
g
y
will n
o
t o
n
l
y sav
e
time an
d
en
erg
y
for sp
ectru
m
sen
s
ing
b
u
t
also
im
p
r
o
v
e
s spectru
m
u
tilizatio
n
.
Fu
rth
e
r
on
ce th
e artificial n
e
ural
n
e
two
r
k
is
t
r
ain
e
d, its co
m
p
u
t
atio
n
a
l co
m
p
le
x
ity is min
i
m
u
m
.
REFERE
NC
ES
[1]
J. Mitola
, “
C
ognitive R
a
dio: A
n
Integra
t
ed Ag
ent Archit
ec
ture
for Software Defined R
a
dio”
,
Ph.D. thesis, Ro
ya
l
Institute of
Tech
nolog
y
,
2000
.
[2]
S. Ha
y
k
in
, “
C
ognitive r
a
dio: B
r
ain-em
powered
wireless com
m
unica
tions”,
I
E
EE Journal on Selected Ar
eas in
Communications
, Vol. 23
, No. 2,
pp. 210–230
, Feb. 2005
.
[3]
Maziar Nekov
ee
, “
D
y
n
am
ic spec
trum
access wit
h
cognitiv
e ra
di
os: future arch
it
ectur
es and rese
arch ch
all
e
nges”
,
IEEE 1st Inter
national Confer
ence on Cognit
i
ve Rad
i
o Oriented Wirel
e
ss Networks and Communications
(
C
ROWNCOM
)
,
Greece, pp 1-5
,
June 8-10, 2006.
[4]
Ashish Bagwari , Geetam Singh Tomar
,
“Adaptive double-
thresh
old based en
erg
y
detector for spectrum sensing in
cognitiv
e r
a
dio
n
e
tworks”,
Intern
ational Journal
of Electronics L
e
tters,
Vol. 1, Issue 1, pp 24-32
,
2013
[5]
Tsung-Hui Chu
ang, Guan-Hong Chen, Meng-
Hsun Tsai, Chun-Lung Lin,
“Alleviating Interference through
Cognitive R
a
dio
for LTE-Adva
nced Network”
,
International
Journal of Electr
ical and Computer Engineering
(I
J
E
C
E
)
Vol. 5,
No. 3, pp. 539-5
47, June 2015
[6]
Gamal Abdel
Fadeel Khalaf
, “An
Optimal Sinsing Algorithm for Multi
band Cognitive Radio Networ
k”,
International Jo
urnal of In
forma
tion
&
Network Security (
I
JINS)
Vol. 2
,
No. 1, pp
. 60-67
, Febru
a
r
y
2013
[7]
Moham
m
a
dreza Am
ini , Asra
Mirzavand
i
, “
P
hase-T
y
p
e
Model Spe
c
trum
Se
nsing for Cogni
tive R
a
dios”
,
IETE
Journal of Resea
r
ch
, published o
n
line 25
th
Mar
2
015, pp
1-7.Mar
c
h 2015.
[8]
Tsagkaris K, Ka
tidiot
i
s A, Dem
e
sticha
s P., “
N
eu
ral network-b
a
se
d learning sch
e
m
e
s for cognitiv
e radio s
y
st
em
s”,
Computer Communications
, 20
08; 31(14), pp 3
394–3404, Vol 3
1
, 2008
[9]
M. Bkassin
y
, Y. Li, S.K. Jay
a
w
eera, "A survey
on machine learn
i
ng techniques in cognitive radios",
IE
EE
Communications
Sur
veys
&
T
u
tor
i
als
,
vol. 15
, no
. 3
,
pp
. 1136-11
59, Mar
c
h. 2013
.
[10]
VK Tum
u
luru,
W
Ping, D Niyat
o
, “
A
neural network ba
sed spectrum
prediction
schem
e
for cognitive r
a
dio”
,
IE
EE
International Co
nference on
Co
mmunications
(I
EEE ICC), 2010
, Cap
e
Town, pp
. 1–5
. 23-27
May
2010.
[11]
Zhen
yu Zh
ang,
and Xiao
yao Xi
e, “
I
ntel
ligen
t co
gnitive r
a
dio: r
e
search on l
earni
ng and evalu
a
t
i
o
n
of CR based on
neural n
e
twork”,
5th Internatio
nal Conferen
ce
on Informati
on
and Communications Technolog
y, ICICT 2007
,
Cairo, pp 33-37
,
16-18 Dec. 2007
.
[12]
MI Taj
,
M Akil
,
“
C
ognitive r
a
di
o spectrum
evol
ution pred
ict
i
on
using artif
ici
a
l
n
e
ural
network b
a
sed m
u
ltiva
r
ia
t
e
tim
e series m
odeling
”
,
11th Eur
opean Wireless Conference 2011 - Susta
inable Wireless Techno
logies (
E
uropean
Wire
le
ss),
(Vien
n
a), pp
. 1–6
. 27-
29 April 2011
[13]
A.
He ,
K.
K.
Bae,
T.
R.
Newman,
J.
Gaeddert,
K.
Kim,
R.
Me
non
,
L.
Mora
le
s-Tira
do,
J.
J.
Ne
e
l
,
Y.
Zha
o
,
J.
H.
Reed
and W
.
H. Tr
ant
e
r,
"A surve
y
of artif
ic
ial
int
e
lligen
ce
for cog
n
itive
radios",
I
EEE T
r
ansactio
n on
V
e
hi
cular
Technology
, Vol. 59
,
No. 4
,
pp.1578 -1592, 20
10
[14]
J
D
Deaton, RE I
r
win, LA DaS
ilv
a, “
T
he
effe
ct of
a d
y
nam
i
c s
p
e
c
t
r
um
acces
s
overl
a
y
in L
TE-Adva
nced ne
tworks
”
,
IEEE Symposiu
m on New Front
iers in Dynamic Spectrum Acc
e
ss Networks
(
D
ySPAN)
, Aachen,
pp. 488–497. 3-6
May
2011
[15]
O Vucenta, H C
a
rlos, JF Monserrat, G Xavi
er, “
I
m
p
lem
e
nting o
pportunistic spe
c
trum
acc
ess in LTE-Advan
ced
”
,
EURASIP Journ
a
l of Wireless C
o
mmunication a
nd Networking,
Vol. 99
, 2012
[16]
3GPP TS 36133, Evo
l
ved
Universal
Terr
estrial Radio
Access (E-UTRA);
requireme
nts for
support of rad
i
o
resource management, April 201
0.
http://www.etsi.org/deliver/ets
i_ts/136100_ 136199/136133/ 9.03.00_60/ts _13
6133v090300p.p
d
f
[17]
N. Baldo and M. Zorzi, “Learni
ng and Adaptation in Cognitive Ra
dios using Neural Networks”,
5th IEEE
Consumer Communications and
Networking Co
nference (
CCNC 2008)
, pp. 998-
1003, Januar
y
2
008
[18]
Z. Zh
ang and X. Xie, “
Intelligen
t Cognitive Rad
i
o: Research on
Le
arning and Evaluation o
f
CR
Based on Neural
Network
”, Proceedings ITI 5th I
n
ternational Con
f
erence
on Infor
m
ation and Communications Technolog
y
(ICIC
T
2007), pp
. 33-37
, December 200
7.
[19]
C.J. Ri
eser,
T
.
W
.
Rondeau
, C
.
W
.
Bostian
,
and
T.M.
Ga
l
l
aghe
r
.
“
C
ognitiv
e R
a
dioTest
bed:
Further De
tai
l
s an
d
Testing of a Di
stributed Gene
ti
c Algorithm
Base
d Cognitive E
ngine for Program
m
a
ble Radio
s
”,
IEE
E
Mili
ta
ry
Communications Conferen
ce,
MI
LCOM, Octob
e
r
2004.
[20]
X. Zhu, Y.
Liu, W.
Weng,
and D.
Yuan,
“
Channel Sensing
Al
gorithm based on Neural Netwo
r
k for Cognitive
Wire
le
ss Me
sh Ne
twork
”
,
P
r
oce
e
dings of IEEE
Interna
tiona
l Co
nferenc
e
on W
i
reless Com
m
unications (W
iCom
),
pp. 1-4
,
2008
.
Evaluation Warning : The document was created with Spire.PDF for Python.