Int
ern
at
i
onal
Journ
al of
El
e
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
8
, No
.
6
,
Decem
ber
201
8
, p
p.
5185
~
5205
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v8
i
6
.
pp.
51
85
-
5205
5185
Journ
al h
om
e
page
:
http:
//
ia
es
core
.c
om/
journa
ls
/i
ndex.
ph
p/IJECE
Sp
ec
t
rum Sensin
g w
ith
VSS
-
NL
MS Pr
ocess
in
Fem
to/Macro
-
ce
ll
Envi
ro
nm
ents
Sidi
Moh
amm
ed H
adj Irid
1
,
M
ohammed
Hicham
H
ache
mi
2
, H
ar
oun
Er
rachid
Ada
rdou
r
3
,
Moura
d H
adj
il
a
4
1,3,4
Depa
rte
m
ent
of
Telco
m
m
unicati
ons,
Facult
y
o
f
Technol
og
y
,
U
nive
rsit
y
of Tlem
ce
n
,
Alg
eri
a
2
Depa
rte
m
ent of
Elec
tron
ic
s,
Fac
ulty
of
Elec
tri
c
al E
ngin
ee
ring
,
U
nive
rsit
y
of
Sci
e
nce
and Technol
og
y
–
Oran
,
Alg
eri
a
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Ja
n 1
1
, 2
01
8
Re
vised
Ju
l
2
3
,
201
8
Accepte
d
J
ul
24
, 2
01
8
Handove
r
is
a
proc
ess
tha
t
a
ll
o
ws
a
m
obil
e
no
de
to
cha
ng
e
it
s
at
tachm
ent
point
.
A
m
obil
e
node
conn
ec
t
ed
to
a
n
et
work
c
an,
in
o
rde
r
to
i
m
prove
the
qual
ity
of
serv
ice,
h
ave
the
ne
ed
to
le
av
e
it
to
co
nnec
t
to
a
ce
l
l
either
of
the
sam
e
net
work
or
of
a
new
net
w
ork.
Th
e
pr
ese
nt
pap
er
int
r
oduce
thr
e
e
te
chn
ique
s
using
ada
pti
v
e
Vari
ab
le
Step
-
Siz
e
Le
a
st
Mea
n
Square
(VS
SLM
S)
fil
ter
combined
with
spec
trum
sensing
proba
bil
ity
m
et
hod
t
o
det
e
ct
th
e
tri
gger
ing
of
han
dover
in
heterog
ene
ous
LTE
netw
orks.
The
se
tec
hnique
s
are
Norm
al
iz
ed
LM
S
(NLMS
),
K
w
ong
-
NLMS
and
Li
-
NLMS
.
The
sim
ula
ti
on
envi
ronm
ent
is
compos
ed
of
tw
o
femtoce
ll
s
belonging
to
a
m
ac
roc
ell.
Fiv
e
Us
er
Equi
pements
(UEs)
are
p
ositi
oned
in
one
femtoce
ll
and
are
assum
ed
cl
osest
to
it
s
ci
r
cumfere
nc
e.
Sim
ula
ti
on
resul
ts
show
tha
t
sensing
proba
bi
lit
y
with
L
i
-
NLMS
al
gorit
hm
has
a
bet
t
er
per
form
anc
e
compare
d
wit
h
cl
assical
NLMS
and
Kw
ong
-
NLMS
Ke
yw
or
d:
Fem
tocel
l
Hand
ov
e
r
Lo
gar
it
hm
ic
p
r
op
a
gatio
n
m
odel
LTE
netw
ork
VS
SLM
S
Copyright
©
201
8
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
:
Sidi Mo
ham
m
ed Had
j
Ir
i
d
,
Dep
a
rtem
ent o
f
Tel
com
m
un
ic
at
ion
s
,
Faculty
of Tec
hnology,
Un
i
ver
sit
y o
f Tl
e
m
cen,
Al
ge
ria
.
Em
a
il
:
irid.tlm@gm
ai
l.co
m
1.
INTROD
U
CTION
The
capaci
ty
of
wireless
net
w
orks
ha
s
doubl
ed
eve
ry
30
m
on
t
hs
in
the
la
s
t
10
4
ye
ars
[1
]
.
Ov
e
r
tim
e,
dem
and
f
or
hig
h
tra
ns
m
issi
on
rates
c
on
ti
nu
es
to
rise.
F
or
exam
ple,
Ci
sco
a
ntici
pated
a
39
-
fo
l
d
i
ncr
e
ase
i
n
data
traf
fic
bet
ween
2009
a
nd
2014
[
2].
I
n
F
or
ecast
,
t
he
a
uth
ors
cl
ai
m
that
in
20
10
t
he
a
m
ou
nt
of
m
ob
i
le
data
traff
ic
near
ly
tr
ipled
f
or
the
t
hi
rd
c
onsecuti
ve
ye
ar
[
3]
.
Als
o,
they
f
or
ecast
t
hat
by
2015
a
bout
1
bill
ion
pe
op
le
sh
oul
d
acce
ss t
he
I
nter
net th
r
ough a
wireles
s m
ob
il
e d
evic
e. To co
pe wit
h
this t
rem
end
ou
s
gr
ow
t
h
in
dem
and,
sever
al
te
c
hnol
og
ie
s
a
nd
sta
ndar
ds
ha
ve
bee
n
de
vel
op
e
d.
The
m
os
t
adv
a
nced
cel
lular
ne
tworki
ng
sta
ndar
ds
include:
High
Sp
ee
d
Pac
ket
Access
(H
S
PA),
L
ong
Te
rm
Ev
olu
ti
on
(L
T
E),
a
nd
LTE
Adva
nced
(LT
E
-
A
)
of
3GPP,
the
nor
m
s
Evo
luti
on
-
Data
Op
ti
m
iz
e
d
(E
VDO)
a
nd
Ultra
W
i
de
Ba
nd
(
U
WB)
of
3GPP
2
an
d
fin
al
ly
the
World
wide
In
t
eropera
bili
ty
fo
r
Mi
cr
ow
a
ve
Access
(
WiM
AX)
sta
ndar
ds.
At
the
sam
e
t
i
m
e,
diff
e
ren
t
WL
AN
sta
nd
a
rds
hav
e
also
bee
n dev
e
lop
e
d.
Althou
gh
cell
ul
ar n
et
w
ork
sta
nd
a
r
ds
h
a
ve
se
ver
al
adva
ntag
es in term
s
o
f
m
ob
il
i
ty
an
d
cov
e
ra
ge
ove
r
WL
AN
sta
nda
rd
s
,
cel
lular
ne
tworks
suffe
r
from
lower
t
hro
ughput,
w
hi
ch
m
akes
them
le
ss
com
pet
it
ive
in
m
any
con
te
xts
.
Fo
r
cel
lular
netw
orks
to
of
fer
ser
vices
co
m
par
able
to
tho
se
of
W
L
ANs,
the
arch
it
ect
ur
e
of
cel
lular
netw
or
ks
need
s
to
underg
o
m
ajo
r
c
ha
ng
e
s
su
c
h
as
passa
ge
f
ro
m
ci
rcu
it
switc
hi
ng
to
pac
ket
swi
tc
hin
g
[4
]
.
Des
pite
va
rio
us
cha
ng
e
s,
cel
lular
netw
orks
can
not
provi
de
the
best
serv
ic
es
to
co
nsu
m
ers
throu
gh
t
hese
te
chn
iq
ues
oft
en
pr
ov
e
ver
y
c
os
tl
y
for
the
operat
or
s
of
t
he
cel
lular
netw
or
k
s
since
t
hey
r
equ
i
re
a
com
plete
or
par
ti
al
m
od
ific
at
ion
of
the
ex
ist
ing
inf
rastr
uc
ture.
Re
centl
y
,
m
or
e
and
m
or
e
researc
h
has
pu
s
hed
op
e
rat
or
s
to
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.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
5185
-
5194
5186
adopt
a
ne
w
s
olu
ti
on,
nam
ely
the
dep
l
oym
ent
of
a
fem
to
-
cel
l.
This
ne
w
te
ch
no
l
og
y
has
be
en
a
dopt
ed
by
op
e
rato
rs
as
it
d
ram
at
ic
ally redu
ce
s their
s
pe
nd
i
ng [1].
Fem
to
-
cel
ls
ar
e
s
m
al
l
qu
al
ifie
d
base
sta
ti
on
s
of
te
n
in
the
s
ta
nd
a
rd
s
of
ho
m
e
base
sta
ti
on
s.
T
hey
are
char
act
e
rized
by
their
ve
ry
s
m
al
l
siz
e
(o
f
the
orde
r
of
a
W
i
Fi
acce
ss
point)
,
their
lo
w
po
wer
a
nd
l
ow
c
os
t.
Fem
to
-
cel
ls
can
be
easi
ly
de
plo
ye
d
by
c
onsu
m
ers
and
bu
sinesses
in
a
c
om
plete
ly
arb
it
rar
y
m
ann
er.
Since
fem
to
-
cel
ls
are
instal
le
d
in
ex
ist
ing
cel
lular
netw
orks,
they
us
e
t
he
sam
e
com
m
ercial
sta
nd
a
r
ds
a
nd
tr
ansm
i
t
on
the
sam
e
r
adio
sp
ect
ru
m
.
The
connecti
on
b
et
wee
n
th
e
fe
m
to
-
cel
ls
and
the
base
netw
ork
of
cel
lular
netw
ork o
per
at
or
s
is usuall
y v
ia
a D
S
L c
onne
ct
ion
th
r
ough
the access
net
work li
nk [4].
In
it
ia
ll
y,
fem
t
o
-
cel
ls
wer
e
desig
ne
d
to
ha
ve
bette
r
vo
ic
e
co
ver
a
ge
in
ho
m
es.
I
ndeed
,
m
any
consum
ers
suf
fer
from
p
oor
sign
al
qual
it
y
inside
t
heir
ho
m
e
durin
g
handove
r.
N
ow
a
da
ys,
they
a
re
m
ai
nly
reg
a
rd
e
d
as
a
cost
-
e
ff
ect
ive
way
to
offl
oa
d
data
tra
ff
ic
from
cel
lular
netw
orks.
F
or
exam
ple,
2.3
m
illi
on
fem
to
-
cel
ls
we
re
dep
l
oyed
in
2011.
By
2014,
t
his
num
ber
has
qua
dru
pled
t
o
al
rea
dy
8.1
m
il
l
ion
.
Als
o,
al
l
global data
traf
fic will
b
e
sup
ported
b
y
fem
t
o
-
cel
ls i
n co
nj
un
ct
io
n wit
h W
i
Fi [
5].
In
this
pa
per,
the
trigg
e
rin
g
of
ha
ndove
r
[
6]
in
heteroge
neous
LTE
ne
tworks
will
be
based
on
sp
ect
r
um
sensing
p
r
obabili
ty
(S
SP
)
m
et
ho
d,
w
h
e
re,
we
will
be
intr
oduce
d
three
te
ch
niqu
es
us
in
g
the
a
da
ptive
Var
ia
ble Step
-
Size
Least M
e
an
S
quare
(VS
S
-
LMS
)
filt
er,
su
c
h
as:
NLM
S, Kwo
ng
-
N
L
MS an
d
Li
-
NL
MS.
The
s
pectru
m
sensing
te
ch
ni
qu
e
[7
]
,
[8
]
ha
s
been
widely
us
e
d
to
predict
the
pr
ese
nce
of
a
P
rim
ary
B
ase
Stat
ion
t
hro
ugh
the
est
i
m
ation
of
Re
cei
ved
Sig
nal
Stren
gth
I
nd
ic
at
ion
(R
SSI)
at
UEs.
A
nd
i
n
th
e
li
te
ratur
e,
the
RSSI
m
easur
e
m
ent
has
bee
n
al
so
ta
king
int
o
acco
unt
to
es
tim
a
te
the
handove
r
decisi
on
[9
]
.
A
s
a
resu
lt
,
the
trigg
e
rin
g
of
ha
ndove
r
in
the
pre
sent
pa
per
wil
l
be
pr
e
dicte
d
with
us
i
ng
SS
P
m
et
ho
d,
wh
i
ch
will
be
qua
ntifie
d
vi
a
the
detect
ion
pro
ba
bili
ty
.
In
Hac
hemi
et
al
.
,
sp
ect
ru
m
-
sen
sing
prob
a
bili
ty
of
the
Lin
k
Dow
n
of
the
cu
rr
e
nt
cel
l
dep
en
ds
on
the
co
nv
e
rgence
of
the
cl
assic
al
NLMS
al
go
rithm
[1
0]
.
Fo
r
fixe
d
ste
p
-
siz
e
LMS
al
gorith
m
,
the
Me
an
S
qu
a
re
E
rror
(
MSE)
is
direct
ly
pr
oport
io
nal
to
the
ada
ptati
on
ste
p
-
siz
e
w
hile
the
conve
rg
e
nce
r
at
e
increases
as
the
ste
p
-
siz
e
decr
eases
[
11
]
.
Ho
w
eve
r,
a
dap
ti
ve
filt
erin
g
resea
rch
has
sh
own
that a va
riable
ste
p
-
siz
e
offers
a b
et
te
r
co
m
prom
ise
b
et
ween t
he
c
onve
rg
e
nc
e rate a
nd a lo
w
est
im
ation
e
rror.
This
pa
pe
r
is
an
im
pr
ov
e
d
ver
si
on
of
Hac
hemi
et
al
[
10
]
wh
ic
h
is
it
sel
f
insp
ire
d
by
the
w
or
k
dev
el
op
e
d
in
[
12
]
,
[
13]
.
T
he
rem
ai
nd
er
of
this
pa
per
is
organ
iz
e
d
as
f
oll
ow
s
.
Sect
io
n
I
I
include
s
so
m
e
relat
ed
works.
Sect
io
n
II
I
descr
i
bes
bri
efly
var
ia
ble step
-
siz
e
LMS al
gorithm
s
us
ed
in
the
sim
ula
ti
on
pa
rt.
Sect
ion
I
V
pr
ese
nts sim
ulati
on
resu
lt
s a
nd
finall
y Sect
io
n V c
on
cl
ud
e
s
the p
a
pe
r
a
nd di
scusses fut
ur
e
works.
2.
RELATE
D
W
ORKS
In
this
sect
io
n
we
will
ci
te
so
m
e
wo
r
ks
on
f
e
m
tocel
l
acce
s
s
m
od
es
f
ollowe
d
by
a
bri
ef
de
scriptio
n
of
the
ha
ndove
r
de
ci
sion
al
gorithm
s.
Handove
r
or
inte
rcell
ul
ar
aut
om
at
ic
tr
ansf
e
r
is
a
f
undam
ental
m
ec
han
ism
in
cel
lular
c
omm
un
ic
at
ion
.
It
represe
nts
the
set
of
operati
ons
im
ple
m
ente
d
s
o
that
a
m
ob
il
e
sta
ti
on
ca
n
switc
h
cel
ls
without
interr
upti
on
of
serv
ic
e.
Th
e
proces
s
co
ns
ist
s
in
that
a
m
obil
e
te
r
m
inal
m
ai
ntains
the
c
urre
nt
com
m
un
ic
at
ion
duri
ng
a
m
ov
em
ent,
wh
i
ch
causes
th
e
m
ob
il
e
to
change
the
cel
l.
I
ndeed
,
w
he
n
the
transm
issi
on
sign
al
betwee
n
a
ha
nd
set
a
nd
a
base
sta
ti
on
is
weak
e
ne
d,
t
he
handset
syst
em
fin
ds
a
no
t
her
base
sta
ti
on
avail
abl
e
in
ano
t
her
ce
ll
,
wh
ic
h
is
abl
e
to
ensure
co
m
m
un
ic
at
ion
again
under
t
he
best
co
nd
it
io
ns.
Thi
s
m
echan
ism
allo
ws
roam
ing
bet
ween
cell
s
or
operat
or
s
.
The
num
ber
of
ha
ndovers
de
pends
sig
nifi
cantl
y
on
the
m
od
e
of
acce
s
s
for
the
fem
t
ocell
s.
The
fem
tocel
ls
can
be
de
plo
ye
d
in
ei
ther
cl
os
ed
or
ope
n
acce
ss
[1
4].
Se
ver
al
stud
ie
s
ha
ve
stud
ie
d
the
fem
t
o
-
cel
ls
acce
ss
m
od
es.
In
Xi
a
e
t
al.
,
a
uthors
we
re
st
ud
ie
d
fem
tocells
in
O
pe
n
Ac
cess
(
OA)
a
nd
Cl
os
ed
Acces
s
(CA
)
m
od
es
[15]
.
T
hey
cha
racteri
zed
the
diff
e
re
nce
bet
ween
t
he
tw
o
cat
eg
ori
es
and
s
howe
d
the
or
et
ic
al
ly
and
by
si
m
ulati
on
that
the
acce
ss
m
od
es
of
fem
tocel
ls
dep
e
nd
profoun
dly
on
t
he
m
ul
ti
ple
acce
ss
te
chnolo
gy
a
dopted
by
the
op
e
rato
rs
(
TDM
A,
O
FD
MA
,
or
C
D
MA).
They
sta
te
d
that
it
is
pr
efera
ble
to
us
e
CA
fem
to
-
cel
ls
in
a
TDMA
or
O
F
DMA
m
ulti
ple
acce
ss
cel
lular
netw
ork.
On
the
oth
e
r
ha
nd,
it
is
pr
eferab
l
e
to
us
e
OA
f
e
m
to
-
cel
ls
in
a
CDMA
m
ult
iple
acce
ss
cel
lular
netw
ork.
In
Y
un
et
al
.
t
he
auth
or
s
st
ud
ie
d
the
OA
a
nd
CA
fem
to
-
cel
ls from
an
econom
ic
p
oin
t
of
vie
w
[
16]
.
They
analy
zed
the
i
m
pact
of
us
er
ince
ntiv
es
on
the
tu
rnov
e
r
of
any
ne
twork
ope
rator.
Using
an
econom
ic
m
od
el
base
d
on
ga
m
e
t
heo
ry,
the
y
showe
d
t
hat
OA
fem
to
-
cel
ls
are
m
or
e
be
ne
fici
al
for
oper
at
or
s.
Jo
et
al.
hav
e
s
tud
ie
d an
d de
m
on
strat
ed
m
a
them
a
ti
cal
l
y, by
calc
ulati
ng
t
he dist
rib
ution
of
SINR i
n f
un
ct
ion
of
the
distance
be
tween
the
m
ac
ro
cel
l
an
d
the
fem
to
-
cel
ls,
that
there
is
a
co
nf
li
ct
betwee
n
the
con
s
um
ers
inside
and
t
ho
se
outsi
de
f
or
the
c
ho
ic
e
of
fem
t
o
-
cel
ls
acce
ss
m
od
e
in
the
dow
ns
tream
directi
on,
i.e.
indoor
s
consum
ers
pre
fer
C
A
fem
to
-
cel
ls
wh
il
e
ou
t
doors
c
onsu
m
ers
pr
e
fer
O
A
fem
to
-
cel
ls
[17].
I
n
t
his
cas
e,
they
sh
owe
d
t
hat
an
interm
ediat
e access m
od
e is
pr
e
fer
a
ble fo
r bo
t
h
ty
pes
of c
on
s
um
ers.
G.
God
or
et
al
.
pro
vid
e
a
n
over
view
a
bout
hand
ov
e
r
de
ci
sion
al
gorit
hm
s,
wh
ic
h
are
cl
a
ssifie
d
int
o
four
gro
ups
ba
sed
on
the
us
e
d
in
pu
t
par
am
et
ers
an
d
in
dep
e
nd
e
nt
pr
ocedure
s
su
c
h
as
posi
ti
on
in
g
ser
vice
[18]
.
The
locat
io
n
of
fem
tocel
ls
can
be
us
e
d
as
a
n
in
pu
t
par
am
et
er
to
em
end
th
e
hand
ov
e
r
e
ff
i
ci
ency
an
d
this
ty
pe
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
Sp
ect
r
um
Sen
si
ng wi
th V
SS
-
N
LMS
Proc
ess i
n
Fe
mto/M
acr
o
-
cel
l E
nviro
nments
(
Sid
i M
oham
med H
adj
Irid
)
5187
is known
as l
oc
at
ion
b
a
sed
al
gorithm
s.
The se
co
nd
gro
up
of alg
or
it
hm
s ta
kes
int
o
acco
unt UE
velocit
y durin
g
hand
ov
e
r
decisi
on
w
hile
the
third
cat
e
gory
of
al
gorith
m
s
us
es
so
m
e
pr
e
def
i
ne
d
po
li
cy
to
m
a
ke
th
e
appr
opriat
e
de
ci
sion
.
Fi
nally
,
the
la
st
gr
oup
of
al
gorithm
s
util
iz
es
so
m
e
l
earn
i
ng
te
ch
ni
qu
e
s
to
colle
ct
so
m
e
inf
or
m
at
ion
from
the surr
ound
ing
e
nvir
on
m
ent to im
pr
ove t
he deci
sio
n’
s
good
ness.
In
cas
e
of
loc
at
ion
base
d
al
gorithm
s,
auth
or
s
[19
]
,
[
20
]
us
e
any
ty
pe
of
sp
at
ia
l
infor
m
at
ion
about
UEs
or
He
NB
s
(Hom
e
evo
lved
N
od
eB
)
su
c
h
as
the
rou
gh
po
sit
io
n
of
th
e
po
ssi
ble
neig
hbors
of
HeNB
,
the
exact
locat
ion
of
He
NBs
usi
ng
the
co
ver
a
ge
area
of
each
f
e
m
toce
ll
s,
or
the
distance
of
a
UE
from
a
giv
en
HeN
B.
Wh
e
n
t
he
s
peed o
f UE
b
ec
om
es i
m
po
rtant, it c
ou
l
d
be
n
ecessa
ry to t
ake in
t
o
acc
ount this
par
am
eter
t
o
m
ake
handove
r
decisi
on.
A
ut
hors
Wu
et
al
.
pr
opos
e
a
n
al
gorithm
na
m
ed
a
per
iod
ic
sca
n
m
echan
is
m
wh
e
re
the
U
E
m
igh
t
be
forced
to
ha
ndover
into
fe
m
tocel
l
even
if
the
RSS
I
of
th
e
ser
ving
m
acro
cel
l
is
bette
r
than
a
giv
e
n
fem
tocel
l
[21]
.
A
n
oth
er
hand
ov
e
r
decisi
on
al
gori
thm
based
on
m
ob
il
it
y
pr
ed
ic
ti
on
of
t
he
UE
is
pro
po
se
d
i
n [22].
It u
ses
the c
urr
ent posit
ion an
d
the
velo
ci
ty
o
f
t
he UE to e
s
tim
a
te
the n
e
xt positi
on
w
here t
he
pr
ocess
of
ha
ndover
is
init
ia
te
d
ei
ther
by
He
NB
or
U
E.
I
n
S
hih
-
J
ung
Wu
et
al
.
pr
opos
e
a
ha
ndover
decisi
on
st
rategy
for
hy
br
id
fem
tocel
l
syst
e
m
s
[23]
.
To
m
ake
hand
ov
e
r
deci
sion
this
al
gor
it
h
m
ta
kes
into
acco
unt
the
RSSI
m
easur
em
ent,
the
vel
ocity
of
UE,
the
requi
red
Q
oS
(Qua
li
ty
of
Ser
vice
),
a
nd
the
ba
ndwi
dth
.
A
utho
rs
in
introd
uced
a
ha
ndover
m
ech
anism
na
m
ed
fem
tocel
l
collab
orat
io
n
ba
sed
ap
proach
in
wh
ic
h
the
UE
sen
ds
con
ti
nu
ously
m
easur
em
ent
re
ports
to
each
HeN
B,
w
hich
inclu
des
t
he
S
I
R
value
(F
ee
dback
I
nd
ic
at
or)
of
each
PRB
(P
hysic
al
Re
so
urce
Bl
oc
k)
[24]
.
Lea
rni
ng
base
d
deci
sion
al
gorithm
s
us
e
Q
-
Lea
r
nin
g,
w
hich
is
t
he
m
os
t
popula
r
rein
forcem
ent
-
le
arn
ing
al
gorithm
.
C.
D
ha
hr
i
et
al
.
[
25,
26
]
pro
po
s
e
a
cel
l
sel
ect
ion
te
c
hn
i
que
f
or
-
gr
ee
dy alg
ori
th
m
ex
te
nded
with
Q
-
le
ar
ning.
3.
DESCRIPTI
ON OF THE
US
ED
A
L
GO
RITH
MS
In the
fo
ll
owin
g,
we wil
l b
rief
ly
d
escribe
the
var
i
ou
s
alg
or
it
hm
s u
sed
i
n
si
m
ula
ti
on
secti
on.
3.1.
L
MS
Al
gori
th
m
In
1960,
Wi
drow
a
nd
Hoff
wer
e
de
vised
on
e
of
the
m
os
t
cel
ebr
at
ed
a
lgorit
hm
in
adap
ti
ve
sig
na
l
processi
ng
:
the
Least
Me
an
-
S
qu
a
re
(LMS
)
al
gorithm
,
wh
ic
h
is
a
m
e
m
ber
of
stoc
hastic
gradie
nt
al
go
rit
hm
s.
It
is
char
act
erize
d
by
it
s
robu
st
ness,
sim
ple
s
t
ru
ct
ur
e,
l
ow
c
om
pu
ta
ti
on
al
com
plexit
y
and
easy
i
m
ple
m
entat
ion
;
it
has
bee
n
use
d
in
a
wide
s
pectr
um
of
ap
plica
ti
on
s
s
uc
h
as
ada
ptive
c
on
t
ro
l,
ra
dar
,
s
yst
e
m
identific
at
ion
,
channel
e
qu
al
i
zat
ion
,
sp
ect
r
a
l
analy
sis,
sig
nal
detect
io
n,
no
ise
cancel
la
ti
on
a
nd
be
am
fo
rm
ing
[
27
]
,
[
28
]
.
I
n
LMS
al
gorith
m
,
the
Me
an
S
qu
a
re
E
rror
(
MSE)
is
direct
ly
pr
oport
io
nal
to
the
ada
ptati
on
ste
p
-
siz
e
w
hile
the
conve
rg
e
nce
r
at
e
increases
as
the
ste
p
-
siz
e
decr
ea
ses
[
11]
.
E
ns
uri
ng
t
he
sta
bili
ty
of
the
LMS
al
gorithm
requires
the
pe
rm
anen
t
adju
st
m
ent
of
the
ste
p
-
siz
e
so
that
it
is
m
a
i
ntained
i
n
the
appropr
ia
te
r
ang
e
.
A
si
m
ple
way
of
ob
ta
ini
ng
this
resu
lt
is
t
o
nor
m
al
iz
e
the
ste
p
by
the
va
rianc
e
of
the
excit
at
ion
si
gnal
,
ass
um
ed
to b
e
kn
own
a
pr
i
or
i
or esti
m
at
ed
on t
he
sa
m
ples o
f
the
sign
al
.
3.2.
N
L
MS Al
go
ri
th
m
The
Norm
al
ized
LMS
is
a
sp
eci
al
case
of
A
PA
al
gorithm
.
AP
A
sta
nd
s
f
or
A
ff
in
e
Proj
ect
io
n
Algorithm
and
belo
ng
s
t
o
the
data
reu
si
ng
f
a
m
ily.
It
was
pro
posed
or
i
gina
ll
y
by
T.
Hina
m
oto
et
al.
[
29]
and
la
te
r
by
K
.O
ze
ki
et
al.
[
30
]
.
Tw
o
pa
ram
et
e
rs
in
flue
nce
th
e
N
or
m
al
iz
ed
Least
Me
an
S
qu
a
re
al
gorith
m
:
the
norm
al
iz
ed
ste
p
-
siz
e
a
nd
regulariz
at
ion
te
r
m
s,
wh
ic
h
ca
n
be
c
ontr
olled
in
ord
er
t
o
a
ddress
the
co
ntra
dictory
requirem
ent
of
fast
co
nv
e
r
ge
nce
an
d
lo
w
m
isa
dju
stm
ent
[3
1].
T
he
im
ple
m
entat
ion
of
the
N
LMS
is
gove
r
ne
d
by
the
sam
e
ste
ps
an
d
t
he
s
a
m
e
equ
at
io
ns
as
the
LMS
.
The
diff
e
re
nce
li
es
in
the
le
vel
of
the
up
da
te
of
weig
hts.
T
he o
rigin
al
ly
NLM
S alg
or
it
hm
u
pdat
es the
w
ei
ghts
us
in
g
the
fo
ll
ow
in
g form
ul
as:
e
(
n
)
d
(
n
)
y
(
n
)
d
(
n
)
T
(
n
)
(
n
)
(1)
(
n
)
x
T
(
n
)
x
(
n
)
(2)
(
n
1
)
(
n
)
2
(
n
)
e
(
n
)
x
(
n
)
(3)
x(n),
w
(n)
an
d
e(n
)
de
no
te
r
especti
vely
the
inp
ut
vecto
r,
the
weig
htin
g
vecto
r
an
d
the
err
or
ve
ct
or.
(n
)
denotes
the
ste
p
siz
e
vecto
r
wh
e
re
0
<
<
2.
I
n
case
the
sign
al
po
wer
in
the
filt
er
shou
l
d
be
ze
ro,
a
s
m
all
nu
m
ber
is a
dded
to
it
. T
his
m
et
ho
d
is
kn
own
as
-
NLMS
algorit
hm
.
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.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
5185
-
5194
5188
-
NLMS
al
gor
it
h
m
is
a
var
ia
nt
al
gorithm
of
the
cl
assic
al
NLMS.
It
just
app
e
nds
a
sm
a
ll
po
sit
ive
nu
m
ber
in
the
de
nom
inato
r
of
t
he
form
ula
(4)
to
a
void
the
res
ult
eq
ua
ls
to
0.
W
ei
gh
t
s
in
-
N
LMS
a
lgorit
hm
are
up
date
d
us
in
g
t
he
f
ollo
wing
form
ula:
(
n
1
)
(
n
)
2
x
T
(
n
)
x
(
n
)
e
(
n
)
x
(
n
)
(4)
Wh
e
re
0
<
<
1.
3.3.
Kwon
g N
LMS
Algori
thm
Kwo
ng
et
al.
pro
pose
a
va
riable
ste
p
-
si
ze
LMS
al
gor
it
h
m
wh
e
re
t
he
ste
p
-
siz
e
a
dju
stm
ent
is
con
t
ro
ll
ed
by
t
he
s
quare
of
t
he
pr
e
dicti
on
e
rror
in
orde
r
t
o
reduce
t
he
tr
ade
off
betwee
n
m
isa
dj
us
tm
ent
an
d
the
trackin
g
abili
ty
of
the
fixed
ste
p
-
si
ze
LMS
al
gorithm
[32]
.
The
us
e
d
tim
e
var
yi
ng
ste
p
-
siz
e
is
giv
e
n by [
33]
:
(
n
1
)
(
n
)
e
2
(
n
)
(5)
wh
e
re
0
<
<
1
,
>
0
a
nd
(n +
1)
belo
ngs
to
the
inter
val.
3.4.
Li
-
N
L
MS Alg
orit
hm
Mi
nch
a
o
Li
a
nd
Xiaoli
Xi
[
34]
pro
pose
a
ne
w
NLMS
al
gorithm
base
d
on
gr
a
dient
vect
or
to
up
date
ste
p
-
siz
e
us
in
g t
he follo
wing
form
ulas:
g
(
n
1
)
g
(
n
)
(
1
)
e
(
n
)
x
(
n
)
x
T
(
n
)
x
(
n
)
(6)
g
(
n
)
p
.
g
(
n
)
2
(7)
(
n
1
)
(
n
)
g
(
n
)
e
(
n
)
x
(
n
)
g
(
n
)
x
T
(
n
)
x
(
n
)
(8)
Wh
e
re
g(n
)
is
the
s
m
oo
th
of
g
ra
dient
vect
or,
is
the
as
in
the
-
NL
MS
al
go
rithm
,
>
0
,
p
>
0,
is
cl
os
e to
1.
4.
RESU
LT
S
AND SI
MU
L
A
TION
In
orde
r
to
si
m
ula
te
the
al
go
rithm
s
ci
te
d
in
the
pr
e
vious
sect
ion
,
we
a
ssu
m
e
that
the
scenari
o
is
com
po
sed
of
t
wo
fem
to
-
cel
ls
(H
e
NB1
a
nd
HeN
B
2)
a
nd
one
m
acro
-
cel
l
(eN
B
).
Bot
h
H
eNBs
are
locat
ed
in
eNB
in
Fig
ur
e
1.
Five
UEs
ar
e
posit
ion
e
d
i
n
dif
fer
e
nt
locat
ion
s
i
n
He
NB1
an
d
a
re
cl
os
e
to
it
s
ci
rcu
m
fer
ence
in Figu
re
2.
T
he
U
Es
are
r
e
ferred
to
as
UE#
1, U
E#
2, UE
#3,
U
E
#4 a
nd U
E
#5.
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
Sp
ect
r
um
Sen
si
ng wi
th V
SS
-
N
LMS
Proc
ess i
n
Fe
mto/M
acr
o
-
cel
l E
nviro
nments
(
Sid
i M
oham
med H
adj
Irid
)
5189
Figure
1.
Pe
de
stri
ans’
m
ob
il
ity
Figure
2. Mo
vem
ent o
f
the
U
Es in fem
tocel
l
Pedestria
n
m
o
bili
ty
was
ta
ken
from
MET
IS
-
20
20
gro
up
[35]
and
is
i
m
plem
ented
in
our
topolo
gy
m
od
el
.
The
sim
ula
ti
on
tim
e
of
eac
h
UE
de
pends
on
it
s
m
ob
il
it
y.
Tables
1
an
d
2
res
pe
ct
ively
su
m
m
a
rize
the
par
am
et
ers
of t
he
a
dap
ti
ve
alg
or
it
hm
s an
d t
he
sim
ulatio
n pa
ram
et
ers.
Table
1.
Param
et
ers
of
A
dap
ti
ve Alg
or
it
hm
Alg
o
rith
m
s
Para
m
eters
NLM
S
=0
.08
Kwo
n
g
-
NLM
S
=0
.99
7
;
(0)=0
.01
;
=4
x
1
0
-
6
;
m
ax
=1
;
m
in
=0
.01
Li
-
N
LM
S
=0
.25
;
=0
.99
9
=0
.02
;
(0)
=0
.01
;
p
=1
; g(0
)
=0
Table
2.
Param
et
ers
Sim
ulati
on
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.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
5185
-
5194
5190
Para
m
eter
Valu
e
Si
m
u
latio
n
ar
ea
1
0
0
0
m
x 1
0
0
0
m
UEs’
n
u
m
b
e
rs
5
Macr
o
-
cell
radiu
s
5
0
0
m
Fe
m
to
-
ce
ll r
ad
iu
s
5
0
m
Frequ
en
cy
2
GHz
Path
los
s ex
p
o
n
en
t
3
Ref
erence dis
tan
ce
Starting
UE
’s po
in
t
Ou
td
o
o
r
p
en
etration
los
s
2
0
dB
Ind
o
o
r
p
en
etration
los
s
5
dB
Tr
an
s
m
it
po
wer
o
f
eNB
4
6
dB
m
Tr
an
s
m
it
po
wer
o
f
HeNB
2
3
dB
m
W
av
elen
g
th
of
the radio
sig
n
al
0
.12
4
m
#
of
walls sep
arati
n
g
apart
m
en
t betw
een
HeNB /
UE
3
UE’
s
m
o
b
ility
Ped
estrian
m
o
b
ilit
y
ME
T
IS
t
race
Ob
serv
atio
n
chan
n
el
AW
G
N
Predictio
n
ord
er
o
f
f
e
m
to
cell
3
Stan
d
ard d
ev
iatio
n
m
ac
ro/f
e
m
to
cell
8
/3
Ther
m
a
l no
ise
-
1
7
4
(
d
B
m
/
Hz)
No
ise f
ig
u
re
9
dB
Sen
sin
g
level f
o
r
fe
m
to
cell
-
7
5
dB
m
Prop
ag
atio
n
m
o
d
el
Log
-
n
o
r
m
al sh
ad
o
win
g
In
Fig
ur
e
s 3
to
7
,
t
he
red
li
ne
r
epr
ese
nts
t
he
t
hr
es
hold sen
si
ng
pr
ob
a
bilt
y
of H
eNB
1
Li
nk D
ow
n.
Th
e
two
cu
r
ves
dra
wn
by
em
pty
blu
e
and
bott
le
gr
ee
n
ci
rcles
r
epr
ese
nt
res
pe
ct
ively
the
pr
obabili
ti
es
of
de
te
ct
ion
of
He
NB2
a
nd
e
NB.
T
he
c
urve
dr
a
w
n
by
a
su
cces
s
io
n
of
t
wo
em
pty
ci
rcles
fo
ll
ow
e
d
by
a
fill
ed
ci
rcle
represe
nts
the
sensing
pro
babi
li
t
y
vector
of
HeN
B
1
sig
nal
at
each
in
pu
t
of
a
da
ptive
filt
er.
T
he
cy
an
,
gr
ee
n
and
blac
k
fill
ed
s
qu
a
res
re
pr
ese
nt
re
sp
ec
ti
vely
the
pr
e
dicti
on
s
ensi
ng
prob
a
bili
ty
of
He
NB1
by
NLMS
,
Kwo
ng
-
N
LMS
and Li
-
NLMS
algorit
hm
s (
ou
tpu
ts
of ada
pti
ve fil
te
rs)
.
Thro
ughout
th
e
si
m
ulati
on
,
we
note
that
t
he
seco
nd
fem
to
-
cel
l
sign
al
s
ensin
g
pro
ba
bili
ty
is
nu
ll
,
wh
ic
h
resu
lt
s
in
the
decisi
on
ph
a
se
of
[
10]
w
her
e
the
pro
bab
il
it
y
of
fin
ding
e
NB
sig
nal
is
im
m
inent.
Accor
ding
to
Figure
3,
sensi
ng
pro
bab
il
it
y
of
tri
gg
e
rin
g
ha
ndover
durin
g
the
m
ov
em
e
nt
of
UE
#1
by
NLMS
process
is
eq
ua
l
to
zero
.
O
n
th
e
oth
er
ha
nd,
both
Kwo
ng
-
N
LMS
an
d
Li
-
N
LMS
processe
s
trigg
e
r
res
pe
ct
ively
the
beg
i
nn
i
ng
of
ha
ndove
r
at
t
=
39
sec
and
t
=
33
sec
with
sensi
ng
pr
obabili
ty
of
eNB
equ
al
to
0.759
4
an
d
0.746
9.
Figure
4
il
lustr
at
es
diff
e
ren
t
ha
ndover
tri
gg
e
r
points
f
or
UE
#2
by
the
th
ree
propose
d
m
eth
ods.
First,
NLMS
al
gorithm
pr
ovides
s
witc
hing
po
i
nt
at
t
=
45
sec
w
it
h
a
detect
io
n
pro
bab
il
it
y
of
eNB
eq
ual
to
0.6
171.
Howe
ver,
K
w
ong
-
NLMS
al
gorithm
pr
edic
ts
cel
l
cha
ngin
g
to
e
NB
at
t
=
39
sec
with
a
detect
ion
pro
bab
il
it
y
equ
al
to 0.
6969. While
, th
e
pre dict
io
n
trig
ge
rin
g
ha
ndover
o
ccu
rs
at
t = 3
0
sec with a
de
te
ct
ion
proba
bili
ty
o
f
eNB e
qu
al
t
o 0
.7087 usi
ng Li
-
NLMS al
gorith
m
.
Figure
3. Pr
e
di
ct
ion
of sensi
ng
pro
b
abili
ty
wi
th thr
ee
m
et
ho
ds
for UE
#1
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
Sp
ect
r
um
Sen
si
ng wi
th V
SS
-
N
LMS
Proc
ess i
n
Fe
mto/M
acr
o
-
cel
l E
nviro
nments
(
Sid
i M
oham
med H
adj
Irid
)
5191
Figure
4. Pr
e
di
ct
ion
of sensi
ng
pro
bab
il
it
y w
it
h
three
m
et
ho
ds
for UE
#2
The
dif
fere
nt
predict
io
n
cu
rv
e
s
of
trig
ger
i
ng
hand
ov
e
r
f
or
UE#
3
are
sho
wn
in
Figure
5.
B
y
us
ing
th
e
NLMS
al
gorithm
,
the
hand
over
occ
ur
s
at
t
=
54
sec
with
a
sensi
ng
pro
bab
il
it
y
of
e
N
B
equ
al
t
o
0.7
240.
F
or
Kwo
ng
-
N
LMS
al
go
rithm
,
the
handove
r
happens
at
t
=
45
sec
with
a
sensing
pro
ba
bili
ty
of
eNB
equ
al
to
0.731
7.
Where
as
with
Li
-
N
L
MS
al
gorithm
,
ha
ndover
occ
ur
s
at
t
=
15
se
c
with
a
sensi
ng
prob
a
bili
ty
of
eNB
equ
al
t
o
0.748
0.
Sim
ulatio
n
r
esults
in
Fig
ure
6
i
nd
ic
at
e
th
at
hand
ov
e
r
for
U
E#
4
is
t
rig
ge
red
at
instants
t
=
48
sec,
t
=
39
sec
an
d
t
=2
7
sec
with
detect
ion
pro
bab
il
it
ie
s
of
eNB
eq
uals
t
o
0.7
671,
0.767
1
a
nd
0.802
8
usi
ng
resp
ect
ively
t
he
thr
ee
pre
ci
te
d
al
go
rithm
s.
Finall
y,
from
Figure
7,
we
obser
ve
t
hat
ha
ndover
f
or
UE
#5
is
trig
ger
e
d
at
t
=
42
sec
a
nd
t
=
39
se
c
with
sensi
ng
pro
ba
bili
ti
es
of
eNB
equ
al
s
to
0.
85
97
an
d
0.
8551
us
in
g
res
pecti
vely
Kwo
ng
-
N
LMS
an
d
Li
-
NLMS
al
go
rithm
s.
It
is
al
so
no
te
d
that
the
sensing
pro
babi
li
t
y
of
ot
her
s
sign
al
s
(eN
B
or
He
NB2)
by
NLM
S
process
is
null
.
The
NLM
S
sta
nd
a
r
d
al
gorithm
gen
erates
a
con
si
der
a
ble
loss
of
data
by
a
delay
in
handov
e
r
trigg
e
rin
g.
T
his
is
du
e
to
the
process
co
nver
gen
ce
s
pee
d.
T
he
si
m
ulati
on
r
esults
sh
o
w
th
at
the
pr
oba
bili
ty
of
detect
ion
b
y
th
e
Li
-
NLMS
m
et
hod
prese
nts
bette
r
pe
rfor
m
ances
in
te
rm
s
of
acc
ur
acy
;
co
nv
e
r
gen
ce
s
pe
ed
an
d
sta
bili
ty
co
m
par
ed
w
it
h t
he
t
w
o othe
rs
te
ch
ni
qu
e
s, Kw
ong
-
NLMS a
nd
NL
MS.
Figure
5. Pr
e
di
ct
ion
of sensi
ng
pro
bab
il
it
y w
it
h
three
m
et
ho
ds
for UE
#3
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.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
5185
-
5194
5192
Figure
6. Pr
e
di
ct
ion
of sensi
ng
pro
bab
il
it
y w
it
h
three
m
et
ho
ds
for UE
#4
Figure
7. Pr
e
di
ct
ion
of sensi
ng
pro
bab
il
it
y w
it
h
three
m
et
ho
ds
for UE
#5
5.
CONCL
US
I
O
N
The
I
n
this
pa
per,
we
ha
ve
us
ed
th
ree
va
riant
of
NL
MS
al
go
rithm
s
to
address
t
he
pro
blem
of
trigg
e
rin
g
ha
ndove
r
in
hetere
gen
e
ous
LTE
netw
ork
w
her
e
UEs
are
sit
uated
near
the
ci
r
cum
fer
ence
of
a
cel
l.
The
first
on
e
is
the
cl
assic
al
NLMS,
the
sec
ond
al
gorithm
i
s
Kwong
-
NL
MS
and
the
la
s
t
on
e
is
Li
-
NL
MS.
A
s
the
sim
ulati
on
res
ults
re
veal,
the
se
ns
i
ng
pro
bab
il
it
y
with
Li
-
NLMS
al
gorithm
has
a
bette
r
detect
ion
of
trigg
e
rin
g
ha
ndove
r
f
or
al
l
UEs
i
nclu
ded
i
niti
al
ly
in
fem
t
ocell
an
d
c
ons
equ
e
ntly
re
du
c
es
the
am
ount
of
lost
data
c
om
par
ed
with
cl
assic
al
NLMS
a
nd
K
won
g
-
NLMS
al
gorithm
s.
As
pa
rt
of
our
upcom
ing
w
ork
,
ot
her
der
i
vated NL
MS al
gorithm
s w
il
l be i
ntr
oduc
ed
to
im
pr
ov
e
the tri
gg
e
rin
g hand
ov
e
r proc
ess.
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
Sp
ect
r
um
Sen
si
ng wi
th V
SS
-
N
LMS
Proc
ess i
n
Fe
mto/M
acr
o
-
cel
l E
nviro
nments
(
Sid
i M
oham
med H
adj
Irid
)
5193
REFERE
NCE
S
[1]
Chandra
sekha
r
,
V.,
Andrews
,
J.
G.,
Gatherer,
A.
“
Fe
m
to
ce
ll
Networks:
A
Surve
y
”
.
IE
EE
Comm
unic
ati
ons
Magazine
,
2008
,
Vol.
46,
No.
9
,
P.
59
–
67.
[2]
Rea
rdon,
M.
Cisco
Predi
ct
s W
ireless Dat
a
Exp
losion.
2010
,
Press
Rel
e
ase
,
9th
Feb
.
[3]
Forec
ast,
C.
V.
“
Cisco
Visual
N
et
working
Ind
ex
:
Globa
l
Mobil
e
Data
Tr
aff
ic
Forec
ast
Updat
e
2
009
-
2014
”
.
201
0,
Cisco
Pub
li
c
Inf
orm
ati
on
,
Vol.
9
,
Februa
r
y
.
[4]
Za
hir
,
T.,
Ars
ha
d,
K.
,
Nak
at
a
,
A
.
,
Mo
essner,
K.
“
Inte
rfe
ren
ce
Mana
gement
in
Fe
m
toc
el
ls
”
.
I
EE
E
Comm
unic
ati
on
s
Surve
ys
&
Tutor
ial
s
,
2013
,
Vol
.
15,
No.
1,
P.
293
–
311.
[5]
Andrews
,
J.
G.,
Cla
uss
e
n,
H.
,
D
ohle
rr,
M.,
R
ang
an,
S.
,
Re
ed,
M.
C.
“
Fem
toc
ells:
Past,
Present
,
a
nd
Future
”
.
201
2,
IEE
E
Journal
on
Selec
t
ed Areas
in
Comm
unic
ati
o
ns
,
Vol.
30,
No. 3,
P.
497
–
508.
[6]
S.
Desw
al
and
A.
Singhrova,
“
A
Vert
ic
al
Han
dover
Algorit
h
m
in
Inte
gra
te
d
Mac
roc
ell
Fem
t
ocell
Networks
”
,
Inte
rnational
Jo
urnal
of
El
e
ct
ri
c
al
and
Comput
er
Engi
n
ee
ring
(
IJE
CE
)
,
Vol.
7,
No.
1
,
P.
299
-
308,
2017.
[7]
F.
Z.
E
l
Bahi,
H.
Ghennioui
,
And
M.
Zoua
k,
“
Perform
anc
e
Eva
lu
at
ion
Of
Ene
rg
y
Det
ec
tor
Based
Spect
ru
m
Sensing
For Cog
nit
ive
Rad
io
Us
ing
Ni
Us
rp
-
2930
”
,
Inte
rnat
ional
Journal
Of
El
ect
rical
And
Computer
Engi
ne
ering
(
IJE
CE
)
,
Vol.
7
,
No.
4
,
P.
1934
-
1940,
2017
.
[8]
F.
Zhou,
Y.
Su
And
X.
Fan,
“
Low
Com
ple
xit
y
Sparse
Chann
el
Esti
m
at
ion
B
ase
d
on
Com
pr
essed
Sensing
”
,
TEL
KOMNIKA
Tele
communic
a
t
ion,
Comput
i
ng,
El
e
ct
ronics
and
Control
,
Vol
.
14,
No.2, P.
538
-
54
7,
2016
.
[9]
L.
Ch
en,
Q
.
Gu
o,
Z
.
Na
And
K.
Jiang,
“
A
Thre
shold
Based
Handove
r
Tr
igg
eri
ng
Sch
eme
I
n
Hete
rog
ene
ou
s
W
ire
le
ss
Networks
”
,
TEL
KOMNIKA
Tele
comm
unic
ati
on
,
Computing,
E
lectroni
cs
and
Control
,
Vol.
12,
N
o.
1,
P.
163
-
172,
2014.
[10]
Hac
hemi,
M.
H.,
Feham,
M.,
Ad
ard
our,
H.
E.
“
Predicting
The
Pr
obabi
lit
y
Of
Spe
ct
rum
Sensing
W
it
h
Lms
Proce
ss
In
Hete
rog
ene
ou
s L
te
Networks
”
.
Radi
o
engi
ne
eri
ng
,
2016
,
Vol
.
2
5,
No.
4,
P.
809
–
822.
[11]
Aboulnasr,
T.,
Ma
y
y
as
,
K.
A
Robust
Vari
abl
e
Step
-
Size
Lms
-
T
y
p
e
Algorit
hm
:
Anal
y
sis
and
Sim
ula
ti
ons.
IE
EE
Tr
ansacti
ons
on
Signal
Proce
ss
in
g
,
1997
,
Vol
.
45
,
No.
3
,
P.
631
–
6
39.
[12]
Adardour,
H.
E
.
,
Meli
ani,
M.,
H
ac
hemi,
M.
H.
“
Esti
m
at
ion
Of
T
he
Spect
rum
Sensing
For
The
Cognit
ive
R
adi
os:
Te
st Ana
l
y
s
ing U
sing Ka
lman
Filt
er
”
.
Wire
le
ss
Pe
rs
onal
Comm
unic
ati
ons
,
2015
,
Vol
.
84
,
No.
2,
P.
1535
-
1549.
[13]
Adardour,
H.
E
.
,
Meliani,
M.
,
&
Hac
hemi,
M.
H.
“
Im
prove
d
Loc
a
l
Spect
rum
Sensing
In
Clut
t
ere
d
Envi
ronm
e
nt
Us
ing
A Sim
ple
Rec
ursive
Esti
m
at
or
”
.
Comput
ers
&
El
ec
tric
a
l
E
ngine
ering
,
Vol
.
61,
P.
208
-
222
,
2017.
[14]
Khali
d,
W
.
Kw
a
k,
K.
S.
“
Hando
ver
Optimizatio
n
in
Fem
toc
ell
Networks
”
.
201
3,
Inte
rnat
ional
Confe
renc
e
on
I
ct
Conve
rgenc
e
(
Ic
tc
)
,
IE
EE
,
2013
.
P.
122
–
127.
[15]
Xia,
P.
,
Chandras
ekha
r,
V.
,
Andr
ews,
J.
G.
“
Open
vs
.
Closed
Acc
e
ss
Fem
toc
el
ls
in
the
Uplink
”
.
I
EEE
Tr
ansacti
ons
on
Wireless
Comm
unic
ati
ons
,
2
010,
Vol
.
9
,
No.
12,
P.
3798
–
380
9.
[16]
Yun,
S.,
Yi,
Y
.
,
Cho,
D.
H.,
M
o,
J.
“
Open
Or
Close:
On
The
Sharing
Of
Fem
toc
e
ll
s
”
.
In
Info
com,
Proc
ee
din
gs
IEE
E
,
2011
,
Apr
il
,
P.
116
–
120
.
[17]
Jo,
H.
S.,
Xia
,
P
.
,
Andrews
,
J.
G
.
“
Dow
nli
nk
Fe
m
toc
el
l
N
et
work
s:
Open
Or
Clos
ed
?
”
In
2011
I
E
EE
Int
ernati
ona
l
Confe
renc
e
on
C
omm
unic
ati
ons (
ICC
)
,
2011,
Jun
e,
P
.
1
–
5.
[18]
Godor,
G.,
Jako
,
Z.
,
Knapp,
A.
,
Im
re,
S.
“
A
Sur
ve
y
of
Handove
r
Mana
gement
i
n
Lte
-
Based
Mu
lt
i
-
T
ie
r
Fem
toc
e
ll
Networks:
Requirem
ent
s,
Ch
al
l
en
ges
and
Solu
ti
on
s
”
.
Computer
Ne
tworks
,
2015,
V
ol.
76
,
P.
17
–
41.
[19]
Sung,
N.
W
.
,
Pham
,
N.
T.,
Yoon,
H.,
L
ee,
S.,
Hw
ang,
W
.
J.
“
Base
Stat
ion
As
socia
ti
on
Sch
emes
To
Reduce
Unnec
essar
y
Ha
ndover
s
Us
ing
Loc
a
ti
on
Aw
areness
”.
In
Fe
mto
ce
l
l
Net
works.
Wirel
ess
Net
wor
ks
,
2013,
Vol.
1
9,
No.
5,
P.
741
–
75
3.
[20]
W
ang,
Y.
H.,
Chang,
J.
L.
,
H
uang,
G.
R.
“
A
Handove
r
Predic
ti
on
Mec
h
ani
s
m
Based
On
Lt
e
-
A
Ue
Histor
y
Inform
at
ion
”
.
In
Net
work
-
Based
Information
Sy
stems
(
NBIS
)
,
2015
18th
Int
ernati
onal
Con
fe
re
nce
On
,
2015
,
Septe
m
ber
,
P.
1
67
–
172.
IE
EE.
[21]
W
u,
C.
S.,
Chu
,
Y.
S.,
Fang
,
C.
H.
“
The
Per
iodi
c
Scan
and
Veloc
ity
De
ci
s
ion
Handove
r
Scheme
for
Next
Gene
ration
Fem
toc
e
ll
/Ma
cro
c
el
l
Overl
a
y
Ne
tworks
”
,
In:
2013
Inte
rnationa
l
Confe
renc
e
o
n
ICT
Conve
rgen
c
e
(
ICTC
),
2013,
Octobe
r
,
P.
201
–
2
06.
[22]
Ulvan,
A.
,
Bestak,
R.
,
Ulv
an,
M.
“
Handove
r
Scena
rio
and
Proc
e
dure
in
Lte
-
Bas
ed
Fem
toc
ell
Ne
tworks
”
,
In:
The
Fourth
Int
ernational
Conf
ere
nce
on
Mobi
le Ubiq
uit
ous Computin
g,
Syst
ems,
S
erv
i
ce
s
and
Te
chnologies
,
2010
.
[23]
W
u,
S.
J.,
“
A
New
Handove
r
St
rat
eg
y
bet
we
en
Fem
toc
el
l
and
Mac
roc
e
ll
for
L
te
-
Based
Netwo
rk
”
,
In
:
2011
4t
h
Inte
rnational
Co
nfe
renc
e
on
Ubi
-
Me
dia
Com
put
in
g
(
U
-
Me
dia)
,
2011,
Jul
y
,
P.
203
–
208.
[24]
Shbat,
M.S.
Tuz
lukov,
V.
“
Handove
r
Te
chn
iq
ue
bet
wee
n
Fe
m
toc
el
ls
i
n
Lt
e
Network
U
sing
Coll
abor
ative
Approac
h
”
,
In
2
012
18th Asia
-
P
aci
fic
Con
fe
ren
c
e
On Comm
unications
(
AP
CC
),
2012,
P.
61
–
66.
IEE
E
.
[25]
Dhahri
,
C.
,
Ohts
uk
i,
T
.
“
Le
arn
in
g
-
Based
Cell
Sel
ec
t
ion
Method
f
or
Fem
toc
el
l
Ne
t
works
”
.
In
Ve
hi
c
ular
Technol
ogy
Confe
renc
e
(
VTC
Spring)
,
2012,
Ma
y
,
IEEE
75
th
,
P.
1
–
5.
IEEE.
[26]
Dhahri
,
C.
,
Ohts
uki,
T
.
“
Q
-
Le
ar
ning
Cel
l
Se
lect
ion
For
Fem
toc
el
l
Ne
tworks:
Single
-
And
Multi
-
Us
er
Case
”
,
In
:
2012
IEEE
Glob
al
Comm
unic
at
i
ons Confe
renc
e (
Globec
om)
,
2012,
De
ce
m
ber
,
P. 4975
–
4980.
[27]
Jingji
ng,
Z.
“
Va
ria
bl
e
Step
Si
ze
LMS
Algorit
hm
”
.
In
te
rnationa
l
Journal
of
Fut
ur
e
Computer
and
Comm
unic
ati
on
,
2012,
Vol
.
1
,
No
.
4
,
P.389
–
391
.
[28]
Dixit
,
Shubhra
;
Naga
ria
,
Dee
p
ak
.
“
Lms
Adapti
ve
Filt
ers
For
Noise
Cancel
l
at
ion
:
A
Revi
ew
”
,
Inter
nati
onal
Journa
l
of
E
le
c
tric
al
and
Computer
Eng
i
nee
ring (
IJE
C
E
)
,
Yog
y
ak
art
a
,
O
ct
2017
,
Vol
.
7
,
No.
5,
P.
2520
–
2
529.
[29]
Hinamoto,
T
.
,
Mae
kawa
,
S.
“
Extended
The
or
y
O
f
Learni
ng
Ide
nt
ifi
c
at
io
n
”
.
Elec
tr
ic
al
Engi
n
ee
ring
in
Japan
,
1975
,
Vol.
95
,
No.
5,
P.101
–
107.
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.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
5185
-
5194
5194
[30]
Oze
ki,
K.
,
Um
eda
,
T.
“
An Adapt
ive
Filt
eri
ng
Alg
orit
hm
Us
ing
an
Orthogonal
Projec
t
ion
to
an
Affine
Subs
pac
e
and
Its
Properti
es
”
.
El
e
ct
ronics
and
Comm
unic
ati
ons
in
Japan
(
Part
I:
Comm
unic
a
ti
ons)
,
1984,
Vol.
67,
No.
5,
P.
19
–
27.
[31]
Cioc
hina,
S.,
P
al
eo
logu,
C.
,
Bene
st
y
,
J.
“
An
Optimize
d
Nlm
s
Algorit
hm
for
Sy
st
em
Ide
nti
f
ic
a
ti
on
”
.
Signal
Proce
ss
ing
,
201
6,
Vol
.
118
,
P.
1
15
–
121.
[32]
Kw
ong,
R.
H.
,
J
ohnston,
E
.
W
.
“
A
Vari
able
Step
Size
Lms
Algorit
hm
”
.
IE
EE
Tr
ansacti
ons
On
Si
gnal
Proc
essing
,
1992,
Vol
.
40
,
N
o.
7
,
P.
1633
–
16
42.
[33]
Le
e
,
J.
,
Chen
,
J.
W
.
,
Huang
,
H.
C.
“
Perform
anc
e
Com
par
ison
of
Vari
abl
e
Step
-
Size
Nlm
s
Algorit
hm
s
”
.
I
n
Proce
ed
ings
of
t
he
World
Congr
ess
on
Engi
n
ee
ri
ng
and
Compute
r Sc
ie
n
ce
,
2009
,
O
ct
ober
,
Vol.
1,
P.
20
–
22.
[34]
Li
,
M.
,
Xi,
X.
“
A
New
Vari
abl
e
Step
-
Size
Nlm
s
Adapti
ve
Filt
ering
Algorit
hm
”
.
I
n
2013
Inte
rnati
onal
Confe
renc
e
on
Information T
ec
hnology
and
Appl
ic
a
ti
ons (
ITA
)
,
2013,
Nov
e
m
ber
,
P.
236
–
23
9.
[35]
W
ang,
Y
-
C.
,
C
huang,
C
-
A.
“
E
ffic
i
ent
Enb
De
plo
y
m
ent
Str
at
e
g
y
For
Hete
rog
ene
ous
Cel
ls
In
4g
Lt
e
S
y
st
ems
”
,
El
sev
ie
r Comput
er
Net
works
,
14
Marc
h
2015
,
Vo
l.
79
,
P.
297
-
312
,
Doi:
10.
1016
/J.
Com
net
.
2015.
01
.
013
BIOGR
AP
H
I
ES
OF
A
UTH
ORS
Sidi
Moham
m
ed
Hadj
Irid
re
ce
iv
ed
Eng
ine
e
r
and
Magist
er
degr
e
es
in
elec
tron
ic
and
comm
unic
at
ion
engi
ne
eri
ng
fro
m
Tl
emce
n
unive
rsit
y
,
Alger
ia,
i
n
1996
and
200
0
respe
ct
iv
ely
.
The
n
h
e
stud
ie
d
digi
tal
comm
unic
ation
in
Val
encie
nnes
Univ
ersity
,
Franc
e
in
200
1.
He
has
b
ee
n
a
net
work
sup
erv
isor
at
IBM
M
ontre
a
l,
Cana
d
a,
the
n
a
proje
ct
m
ana
ger
at
Ora
scom
Te
le
com
Alger
ia.
Sin
ce
2
008,
he
is
lectur
er
a
t
un
ive
rsit
y
o
f
Tlem
ce
n.
His
r
ese
arc
h
in
te
rests
are
in
the
ar
ea
of
digital a
nd
arr
a
y
signa
l
pro
ce
s
sing
Moham
m
ed
Hicha
m
Hac
hemi
rec
ei
v
ed
his
Engi
nee
r
degr
e
e
in
Telec
om
m
unic
atio
n
Engi
nee
ring
from
the
Univ
er
sit
y
of
Sa
ida,
Al
ger
ia i
n
2007,
th
e
subje
ct of
th
e
f
ina
l
y
ea
r
proj
ect w
as
on
Studie
s
&
Rea
lization
of
the
Dete
ction
of
a
Gas
Le
ak
by
SM
S.
Thi
s
proje
ct
was
pre
sente
d
in
the
ca
stin
g
of
the
emiss
ion
Stars
of
Scie
nce
in
Tuni
sia.
It
was
ran
ked
among
the
16
inve
ntions
in
the
Arab
W
orld.
From
2008
to
2011
,
h
e
re
ceive
s
th
e
Magiste
r
d
egr
e
e
in
the
do
ct
or
al
school
ent
i
tl
ed
Scie
nc
e,
Inform
at
ion
T
ec
hno
log
y
and
Te
l
ec
om
m
unic
at
ions
fro
m
the
Facul
t
y
o
f
Engi
ne
eri
ng
i
n
Sidi
Bel
Abbes
Univer
sit
y
,
Alge
ria
.
From
2012,
he
int
egr
at
e
d
STIC
la
bora
tor
y
as
a
m
ember
and
Ph.D.
student
at
the
Univ
ersity
of
Tl
emc
en,
Alg
eri
a
,
and
his
cur
ren
t
r
ese
arc
h
a
ctivit
i
es
invol
v
e
W
ire
le
ss
Comm
unic
at
ions
Networks,
Cogni
ti
ve
R
adi
o
,
Mac
ro
&
Fem
to
-
ce
l
l
net
works
,
Predic
ti
on
of
signal
s
Haroun
Err
a
chid
Adardour
rece
ive
d
his
Phd
de
gre
e
in
Te
l
ec
om
m
unic
at
ions
a
t
t
he
Facult
y
of
Te
chno
log
y
,
Uni
ver
sit
y
of
T
le
m
c
en
-
Alger
i
a,
in
2
016.
He
is
a
m
ember
of
STIC
laborat
or
y
in
the
sam
e
unive
rsit
y
.
After
recei
ving
his
Master
deg
ree
in
components
and
el
e
ct
ron
ic
s
s
y
stems
for
te
l
ec
om
m
unic
at
i
ons
from
the
Univer
sit
y
of
Tle
m
ce
n,
in
2012.
His
cur
ren
t
rese
arc
h
foc
uses
on
wire
le
ss
co
m
m
unic
a
ti
ons, cogni
t
ive
r
adi
o
net
wor
ks a
nd
spe
ct
rum
sensing.
Mourad
Hadji
la
rec
e
ive
d
his
eng
ine
er
degr
ee
s
in
1994,
his
M.S.
d
egr
ee
s
in
signal
s
and
sy
st
ems
in
1999,
and
his
Ph.
D.
In
T
el
e
co
m
m
unic
at
ions i
n
2014
from
th
e University
of
T
l
emce
n,
Alger
i
a.
Since
2002
h
e
h
as
bee
n
assistant
profe
ss
or
of
T
elec
om
m
unic
at
ion
Engi
ne
eri
ng
.
M
ember
of
STIC
la
bora
tor
y
in
the
Univer
sit
y
of
T
le
m
ce
n.
His
rese
arc
h
intere
st
is
i
n
te
lecom
m
unic
at
ion
s
y
s
te
m
s
and
m
obil
e
ne
tworks.
Evaluation Warning : The document was created with Spire.PDF for Python.