Int
ern
at
i
onal
Journ
al of Ele
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
V
ol.
10,
No.
3,
June
2020,
pp. 3
095
~
3
107
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v10
i
3
.
pp3095
-
31
07
3095
Journ
al h
om
e
page
:
http:
//
ij
ece.i
aesc
or
e.c
om/i
nd
ex
.ph
p/IJ
ECE
Bio
-
insp
ired r
out
e estim
ation in
cogni
tive
radio n
etworks
Migu
el
T
uber
quia, H
an
s L
opez
-
Ch
avez
,
Cesar Her
nan
dez
T
ec
hno
logi
c
al
a
nd
Engi
n
ee
ring
F
ac
ulty
,
Univ
ersida
d
Distr
it
a
l
Fr
anc
isco
Jos
é
de
Cal
das
,
Colom
bi
a
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Ma
r 11
, 201
9
Re
vised
N
ov 30
,
2019
Accepte
d
Dec
10, 201
9
Cognit
ive
r
adi
o
is
a
te
chn
ique
th
at
was
origi
na
lly
cre
a
te
d
for
the
prope
r
use
of
the
r
adi
o
elec
tri
c
spe
ct
rum
du
e
it
s
und
eru
se.
A
few
m
et
hods
were
used
to
pre
dict
th
e
ne
twork
tra
ff
ic
to
d
e
te
rm
ine
the
o
ccupancy
of
the
spec
trum
an
d
the
n
use
the
‘hole
s’
bet
wee
n
th
e
tra
nsm
issions
o
f
primar
y
users.
The
goal
is
to
guar
an
te
e
a
complet
e
tra
n
sm
ission
for
t
he
sec
ond
user
while
no
t
int
err
up
ti
ng
th
e
tr
ans
-
m
ission
of
primar
y
users.
Thi
s
st
ud
y
se
eks
the
m
ult
if
racta
l
gene
ra
ti
on
of
tr
a
ffic
for
a
spec
if
i
c
rad
io
e
lectr
i
c
s
pe
ct
rum
as
well
as
a
b
io
-
inspire
d
rout
e
esti
m
at
ion
fo
r
sec
ondar
y
user
s.
It
uses
the
MF
HW
al
g
orit
hm
to
g
ene
r
at
e
m
ult
ifr
ac
t
al
tra
c
es
and
two
bio
-
inspired
al
go
-
ri
thms
:
Ant
Colon
y
Opti
m
iz
at
ion
and
Max
Feedi
ng
t
o
ca
l
cul
a
te
the
sec
ond
ar
y
u
ser’s
pat
h.
Mul
t
ifra
c
ta
l
cha
r
acte
risti
cs
offe
r
a
p
red
ic
-
ti
on
,
which
is
10%
lower
in
compa
rison
with
the
origi
nal
tra
ff
ic
val
ues
and
a
complete
tr
an
sm
ission
for
se
conda
r
y
users.
In
fac
t
,
a
h
y
br
i
d
strat
e
g
y
combining
both
bio
-
inspire
d
algorithms
prom
i
se
a
red
uct
ion
in
handof
f.
Th
e
purpose
of
thi
s
rese
ar
ch
c
onsists
on
der
iv
ing
future
inve
s
ti
gation
in
the
gen
era
t
ion
of
m
ult
ifra
c
tal
tra
ff
ic
an
d
a
m
obil
ity
spec
trum
using
bio
-
inspire
d
al
go
rit
hm
s.
Ke
yw
or
d
s
:
Bio
-
ins
pire
d
al
gorithm
Cognit
ive
r
adi
o
Mult
ifracta
l t
ra
ff
ic
P
re
dicti
on
Sp
ect
r
um
h
an
dof
f
Copyright
©
202
0
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
Ce
sar Her
nánd
ez,
Dep
a
rtm
ent o
f Te
ch
no
l
og
y,
Un
i
ver
si
da
d Di
strit
al
Fr
ancisc
o
J
os
é
de
Ca
l
da
s,
Cl
. 68D Bi
s A
Su
r
N 4
9F
-
70
,
Bo
gota
, Co
l
om
bia
Em
a
il
:
cahern
a
nd
ez
s@
ud
ist
rit
al
.ed
u.co
1.
INTROD
U
CTION
The
im
ple
m
en
ta
ti
on
of
ne
w
te
chnolo
gies
in
Cognit
ive
Ra
dio
N
et
w
orks
(CRN)
s
houl
d
require
li
tt
le
com
pu
ta
ti
on
al
com
plexity
since
the
est
im
ati
on
of
detect
ion,
decisi
on,
div
i
sion,
a
nd
m
ob
il
it
y
sh
ou
l
d
no
t
ta
ke
m
or
e
than
f
rac
ti
on
s
of
a
se
co
nd.
A
sp
ect
rum
han
do
ff
occ
ur
s
w
hen
a
Pr
i
m
ary
User
(PU)
requests
se
rv
ic
e
i
n
a
channel
that
is
al
read
y
occ
upie
d
by
a
Sec
onda
ry
Use
r
(SU).
Mo
reover
,
the
SU
m
us
t
le
ave
this
c
hann
el
and
look
f
or
a
n
av
ai
la
ble
on
e.
T
his
proces
s
go
es
on
unti
l
the
SU
fi
nish
es
hi
s
transm
issi
on
.
A
s
pectru
m
hand
off
has
a
neg
at
iv
e
i
m
pact
on
the
perform
ance
of
seco
nd
a
ry
use
rs
in
te
rm
s
of
delay
and
li
nk
m
ai
ntenan
ce.
Hen
ce
,
the
pri
ori
ty
is
to
re
du
c
e
ha
ndovers
i
n
the
syst
e
m
[1
]
.
A
CR
N
is
a
s
yst
e
m
that
al
l
ow
s
the
e
valu
at
ion
of
the
tran
sm
issio
n
m
edium
,
analy
ze
th
e
tra
ns
m
issi
on
para
m
et
ers
and
m
ake
decisi
on
s
in
a
dy
nam
i
c
tim
e
-
fr
e
qu
e
ncy
sp
a
ce.
Ba
sed
on
the
al
locat
ion
and
m
anag
em
ent
of
resou
rc
es,
it
aim
s
to
i
m
pr
ov
e
t
he
us
e
of
the
el
ect
ro
m
agn
et
ic
rad
i
o
spe
ct
ru
m
[2
]
.
Th
eref
or
e
,
a
CR
N
sho
uld
be
s
m
art
and
be
a
ble
to
le
ar
n
from
i
ts
interact
ion
e
xp
erience
with
th
e
RF
en
vir
on
m
ent.
Ac
co
rd
i
ng
to
this
sta
te
m
e
nt,
the
le
ar
ni
ng
process
is
a
c
r
ucial
com
po
ne
nt
tha
t
can
be
ta
c
kl
ed
from
var
io
us
a
reas
of
know
le
dg
e
s
uch
as
arti
fici
al
in
te
ll
igence,
m
a
chin
e
le
arn
in
g, ev
olut
ion
ary al
gorith
m
s o
r rob
us
t
con
t
ro
l m
et
ho
ds [
2].
The
m
anag
em
ent
of
the
s
pe
ct
ru
m
un
der
a
CR
N
inv
ol
ve
s
four
sta
ge
s
that
exp
la
i
n
th
e
interact
ion
betwee
n
P
U
a
nd S
U, in term
s o
f occ
upyi
ng
a
nd sh
a
rin
g
t
he spect
r
um
[
3]. Spect
r
um
d
et
ection
is t
he pr
oce
ss in
wh
ic
h
the
SUs
look
f
or
av
ai
la
ble
bands,
captur
e
thei
r
inform
at
ion
and
detect
ga
ps
in
the
s
pe
ct
ru
m
.
The
s
pectr
um
decisi
on
-
m
aking
process
is
encou
rag
e
d
t
o
assig
n
a
c
ha
nn
el
by
co
nsi
der
in
g
the
s
pe
ct
ru
m
avail
abili
ty
and
al
locat
ion
po
li
ci
es.
The
sp
ect
ru
m
div
isi
on
coo
r
din
at
es
th
e
a
ll
ocati
on
of
sp
ect
ral
sp
ace
s
and
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
, No
.
3
,
J
une
2020
:
30
95
-
3107
3096
pr
e
ve
nts
use
rs
from
crash
in
g
into
sect
io
ns
of
the
s
pectru
m
or
e
ven
ove
rlapp
i
ng
w
he
n
m
ul
ti
ple
SU
s
wish
t
o
acce
ss
the
s
pec
trum
.
Sp
ect
rum
m
ob
il
ity
is
t
he
pro
cess
of
m
ob
il
iz
ing
seconda
ry
us
e
rs
towa
r
ds
avail
ab
le
areas
of
the
sp
e
ct
r
um
wh
en
a
P
U
re
qu
est
s
acce
ss
[
4].
T
he
di
ff
e
ren
t
pro
ble
m
s
hav
e
bee
n
div
i
de
d
acc
ordin
g
t
o
researc
h
m
et
ho
dolo
gies
for
t
he
stu
dy
of
C
RN:
sp
ect
r
um
div
isi
on,
sp
ect
ru
m
decisi
on,
sp
ect
r
um
m
ob
il
it
y
and
sp
ect
r
um
detec
ti
on
.
S
pectr
um
detect
ion
is
a
crit
ic
al
aspect
of
s
pectru
m
infer
e
nce
ap
plica
ti
on
s
since
the
y
aim
to
ex
plore
ina
ct
ive
‘
ho
le
s
’.
The
em
erg
in
g
par
a
dig
m
s
of
sp
ect
r
um
detect
ion
li
e
in
t
he
ti
m
e,
fr
eq
ue
ncy,
and
sp
at
ia
l
do
m
ai
ns
.
I
nf
e
rence
te
chn
i
qu
e
s
are
wi
dely
us
e
d
to
determ
ine
as
m
any
e
m
pt
y
channels
as
po
s
sibl
e
and
im
pr
ove
detect
ion
pe
rfor
m
ance.
T
he
y
al
so
re
du
ce
energy
co
nsu
m
pt
ion
a
nd
t
he
tim
e
that
it
ta
kes
to
change
bet
wee
n
cha
nn
el
s
.
Other
as
pects
are
stud
ie
d
s
uch
a
s
centrali
zed
al
locat
ion
of
the
sp
ect
ru
m
,
sel
e
ct
ion
of d
ece
ntrali
ze
d
c
hannels,
ad
a
ptati
on
of the
physi
cal
lay
er a
nd d
y
nam
ic
ac
cess to t
he
s
pe
ct
ru
m
.
Sp
ect
r
um
m
ob
il
ity
in
CR
Ns
hav
e
a
n
am
big
uous
inte
rpreta
ti
on
.
On
one
s
ide,
the
c
on
ce
pt
ref
e
rs
t
o
the
sp
ect
ral
tra
ns
fe
r
f
r
om
on
e
band
to
an
othe
r,
du
e
t
o
the
app
ea
ra
nce
of
PU
s
or
i
nterf
e
ren
ce
e
vasi
on,
fiel
d
widely
stu
died
in
pr
e
dicti
on.
On
the
ot
her
s
ide,
t
he
m
ob
il
it
y
of
C
ogniti
ve
Ra
dio
(CR
)
and
P
Us
(as
se
en
for
instance
in
ve
hi
cular
CR
Ns)
c
an
al
so
af
fect
the
surr
oundin
g
sp
ect
ru
m
env
iro
nm
ent
reg
ar
ding
the
i
m
po
s
it
ion
of
ad
diti
on
al
interfe
re
nce
or
cha
ngin
g
the
conditi
ons
of
the
c
ha
nn
el
a
s
well
as
t
he
s
pe
ct
ru
m
avail
ab
il
ity.
A
fiel
d
that
is
no
t
hea
vily
stud
ie
d
an
d
has
f
ew
co
ntributi
ons
is
the
sp
ect
ru
m
exch
an
ge,
wh
ic
h
has
dif
fer
e
nt
unde
rstan
dings
in
the
li
te
rat
ur
e
.
The
co
nc
ept
of
s
pectrum
sh
aring
is
exch
a
ngeable
w
it
h
dynam
ic
acce
ss
theo
ries,
c
on
si
sti
ng
of
t
hr
ee
par
a
dig
m
s
of
sp
ect
ru
m
use
:
under
ly
in
g
m
od
e,
s
uperi
m
po
sed
m
od
e
and
interco
nnect
io
n
m
od
e.
T
his
per
ce
ptio
n
of
s
pectr
um
exch
a
ng
e
gi
ves
a
m
eanin
g
that
is
t
oo
broa
d
t
o
c
over
al
l
aspects
of
CR
Ns.
F
urt
he
rm
or
e,
s
pect
r
um
e
xch
a
nge
f
oc
use
s
on
the
unde
rly
ing
m
od
e,
wh
ic
h
al
lows
CR
Ns
to
op
e
rate
sim
ult
aneously
in
th
e
sa
m
e
ban
d.
In
this
m
et
ho
d,
flexible
th
res
ho
l
ds
can
be
est
ablished
bet
ween
bands
.
The
inf
eren
ce
of
sp
ec
trum
in
the
di
visio
n
ta
sk
is
m
ai
nly
associat
ed
with
t
he
s
u
pp
or
t
a
nd
m
ediat
ion
betwee
n
the
C
RN
an
d
the
P
us
[5
]
.
Fi
gure
1
disp
la
ys
a
dist
rib
ution
of
i
nference
-
relat
ed
stud
ie
s
in
eac
h
CR
N
arch
it
ect
ure.
Fig
ure
1
.
Di
vision
of
pr
e
di
ct
ion o
pe
n pro
blem
s in
the CR
N
s
In
Ma
rc
h
2012
,
data
f
ro
m
the
rad
i
o
el
ect
ric
sp
ect
r
um
was
colle
ct
ed
in
th
e
ci
ty
of
Bogot
á,
Colom
bia,
wh
e
re
a
s
pectr
um
analy
zer
pro
vid
es
da
ta
tr
aff
ic
detect
ion,
ba
sed
on
the
powe
r
of
si
gnal
s.
C
onseq
ue
ntly
,
the
gathered
in
form
ation
will
ind
ic
at
e
wh
et
he
r
the
sign
al
s
are
pr
ese
nt
or
abse
nt
durin
g
the
sam
pling
per
io
d.
The
captu
re
d
da
ta
is
locat
ed
in
the
GS
M,
Wi
-
Fi
and
18
50
MHz
to
2000
MHz
bands
[
6].
The
obj
ect
ive
of
this
pap
e
r
is
to
est
ablish
a
predic
ti
on
of
the
ra
di
o
-
el
ect
ric
sp
e
ct
ru
m
us
ing
a
m
ul
ti
fr
act
al
alg
ori
thm
.
Af
te
r
wards,
base
d
on
the
pr
e
dicte
d
tra
ffi
c,
a
path
f
or
seco
nd
a
ry
us
e
rs
is
est
i
m
at
ed
so
that
they
can
tran
sm
i
t
without
interr
up
ti
on
a
s
well
as
not
interl
ope
wit
h
pri
m
ary
us
ers.
T
he
est
i
m
at
ed
routes
are
cal
culat
ed
us
in
g
bio
-
ins
pire
d
al
gorithm
s.
The
rem
ai
nin
g
se
ct
ion
s
of
this
arti
cl
e
are
div
i
ded
as
f
ollows
.
Sect
io
n
2
ex
plains
the
m
at
he
m
at
i
cal
fo
un
dation
need
e
d
f
or
th
e
con
str
uctio
n
of
the
pr
e
dicti
ve
m
od
el
and
the
route
est
im
at
ion
.
In
Sect
io
n
3,
t
he
r
esults
ob
ta
ined
f
or
t
he
predict
ion
of
W
i
-
Fi
tra
ff
ic
a
n
d
the
r
oute
est
i
m
at
ion
a
re
discu
ssed
.
Last
ly
, th
e c
oncl
us
io
ns
on the
overall
work a
re
pr
ese
nted
in Sec
ti
on
4.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Bio
-
in
sp
ire
d
r
oute
esti
m
ation
in cog
niti
ve ra
dio
network
s
(
Mi
gu
el
Tuber
quia
)
3097
2.
MA
TE
RIA
L
S
AND MET
H
ODS
2.1.
Multifr
ac
ta
l
predic
ti
on
The
t
rad
it
io
nal
to
ol
f
or
the
m
od
el
in
g
an
d
a
na
ly
sis
of
net
w
ork
t
raffic
is
t
he
cl
assic
al
P
oi
sso
n
tra
ff
ic
m
od
el
,
wh
ic
h
was
la
te
r
m
odifie
d
an
d
a
da
pt
ed
f
or
t
he
stu
dy
of
que
uing
syst
e
m
s
[7
]
.
The
m
easur
ed
traff
ic
beg
a
n
to
e
xhibit
beh
a
viors
t
hat
wer
e
dif
fe
ren
t
f
ro
m
wh
a
t
was
ex
pected
by
the
P
ois
so
n/Ma
rko
v
m
od
el
s
.
The
data
colle
ct
ed
in
the
Be
ll
cor
e
la
bs
pa
ve
d
the
way
for
the
stud
y
of
s
cal
e
-
inv
a
riant
char
act
e
risti
cs
from
traff
ic
i
n
L
A
N
[
8].
T
rad
it
io
na
l
tim
e
series
m
od
el
s
wer
e
pro
ve
n
to
be
ins
uffici
ent
to
m
od
el
the
sel
f
-
sim
il
arity
pr
ese
nt
in
the
traff
ic
an
d
the
analy
sis
of
s
uc
h
proce
sses
c
al
le
d
for
ne
w
te
chn
i
qu
e
s.
S
om
e
m
od
el
s
based
on
m
on
of
ractal
procedu
res
we
re
pro
po
se
d
to
ou
tl
ine
this
traff
i
c
[9
]
.
A
rec
ent
analy
sis
of
th
e
m
easur
e
d
dat
a
has
rev
eal
e
d
the
e
xistence
of
m
ulti
fr
act
al
scal
ing
be
ha
vior
[
10,
11]
.
A
proces
s
X(
t)
is
sai
d
to
ha
ve
local
sc
al
in
g
pro
per
ti
es
with
a
local
scal
in
g
ex
pone
nt
na
m
ed
(t)
if
the
pr
oc
ess
be
ha
ves
li
ke
X(
t
)
~
(
t)
(t)
as
(
t)
0.
Fo
r
a
m
on
ofra
ct
al
process,
th
e
scal
ing
ex
po
nen
t
(t)
=
H
for
al
l
tim
e
s
w
hile
the
m
ulti
fr
act
al
te
r
m
is
us
ed
t
o
denote
t
he
pro
cesses
that
sho
w
a
non
-
c
onsta
nt
scal
ing
par
a
m
et
er
(t).
Th
e
local
Ho
l
der
expo
nen
t
is
giv
en
by
(t)
[5]
.
I
f
th
e
analy
sis
is
c
arr
ie
d
ou
t
it
i
n
the
scal
e
dom
ai
n
us
ing
a
W
avelet
tran
sf
or
m
,
the
coef
f
ic
ie
nts
con
ta
ini
ng
the
m
a
in
inf
or
m
at
ion
of
si
gn
a
l
X(
t
)
c
an
be
est
i
m
at
ed
us
in
g
the
Discrete
W
a
velet
Tra
nsfo
rm
(DWT
) dx(j,
k).
S
ee
(
1)
a
nd
(2)
[
12]
.
(1)
(2)
A
f
undam
ental
featur
e
of
the
c
on
ti
nu
ou
s
wavel
et
trans
form
Tx
(a,
b)
is
it
s
redund
ancy,
sin
c
e
neig
hbori
ng
coeffic
ie
nts
sha
re
so
m
e
com
m
on
inform
at
ion
re
gardin
g
X.
In
fact,
in
order
to
reduc
e
the
re
dunda
nc
y
from
the
inn
er
pro
duct
be
tween
t
he
si
gnal
an
d
t
he
se
t
of
dilat
ions
(a)
a
nd
s
hifts
(b)
of
a
m
oth
er
W
a
ve
le
t
Ψ(∙)
,
the
D
WT
is
introdu
ce
d
f
or
j
sc
al
es.
The
va
ri
ance
of
the
pr
ocess
dx
(j,k)
can
be
est
i
m
at
ed
as
j
,
si
nce
t
he
m
os
t
im
po
rtant
char
act
e
risti
cs
fo
ll
ow
t
he
scal
ing
be
hav
i
or
of
th
e
ori
gin
al
s
ign
al
.
See
(
3) an
d (
4) [13
]
.
(
3
)
(4)
The
Hurst
para
m
et
er
H
,
ca
n
be
est
im
at
ed
by
cal
culat
ing
t
he
li
nea
r
re
gre
ssion
slo
pe
of
yj
agai
ns
t
j
.
This
represe
ntati
on
is cal
le
d t
he
Log
-
scal
e
Di
agr
am
(
LD
),
a
s
shown i
n
t
he
(
5).
(5)
The
est
i
m
at
io
n
of
H
is
us
efu
l
to
stu
dy
second
ord
er
s
ta
ti
sti
cs
(
q=2
)
in
stochastic
pr
oce
ss
es
.
H
ow
e
ve
r
,
the
W
a
velet
trans
f
or
m
can
be
us
e
d
f
or
both
higher
an
d
lo
wer
orde
r
sta
ti
sti
cs
in
the
real
do
m
ai
n
(
q
R
).
At t
his
point, t
he
e
xte
ns
io
n of
j
to
j
q
an
d
t
he
est
i
m
at
or
o
f
j
q a
re c
on
si
der
e
d.
See
(
6) an
d (
7) [14
]
.
(6)
(
7
)
A
m
on
ofracta
l
process
c
ould
be
descr
i
be
d
as
H(q
)=H
q
,
i
nd
ic
at
in
g
t
hat
the
Hurst
pa
ram
et
er
i
s
the
sam
e
fo
r
al
l
sta
ti
s
ti
cs
o
rd
e
rs.
I
n
co
ntrast,
wh
e
n
H
(q)
dec
rease
s
as
the
sta
ti
sti
c
al
or
de
rs
rise
this
m
eans
that
the
process
is
m
ulti
fr
act
al
.
The
li
near
M
ulti
scal
e
D
ia
gra
m
(MD)
pl
ots
the
singulari
ty
exp
one
nt
H
(q)
of
qth
order
ver
s
us
q
,
there
by
il
lustrati
ng
the
be
ha
vio
r
of
H
f
or
dif
fere
nt
values
of
q.
Th
e
Mult
ifracta
l
Sp
ect
r
um
(MS)
plo
ts
H(q)
ve
rsu
s
t
he
sin
gula
rity
di
m
ension
D
(
q)
.
T
hes
e
two
va
riable
s
represent
the
li
near
R
a
b
t
b
a,
d
t
,
X
(
t
)
b)
(
a
,
T
x
R
k
j,
k
)
,
,2
(2
T
=
k)
(
j
,
d
j
j
x
x
)
,
(
d
1
1
2
x
j
n
k
j
j
k
j
n
)
1
2
(
2
x
2
)
,
(
d
H
j
j
k
j
C
j
H
y
j
j
2
2
l
o
g
)
1
2
(
l
o
g
)
,
(
d
1
1
x
q
j
n
k
q
j
j
k
j
n
)
2
)
(
(
x
q
2
)
,
(
d
q
q
j
q
q
j
C
k
j
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
, No
.
3
,
J
une
2020
:
30
95
-
3107
3098
trans
form
ation
from
scal
es
to
sta
ti
sti
cal
m
o
m
ents.
The
refo
re,
t
he
f
unct
ion
that
m
aps
the
sam
pling
scal
es
wit
h
the
co
rr
es
pond
ing
sta
ti
sti
cal
m
o
m
ents
is
non
-
li
nea
r
[
15]
.
D
(
q)
ca
n
be
es
tim
a
te
d
by
us
ing
the
m
ass
ex
pone
nt
of ord
e
r q
desi
gn
at
e
d
as
(
q)
.
See
(
8) an
d (
9) [16
]
.
(8)
(9)
The
MS
f
or
a
pure
m
onofrac
ta
l
process
is
a
sp
eci
fic
point
in
s
pace,
in
c
on
t
rast
wit
h
a
m
ul
ti
fr
act
al
tim
e
series
where
it
is
a
con
cave
curve
poi
n
ti
ng
to
wards
the
x
-
a
xis.
Th
e
MS
fo
rm
ca
n
be
ap
pro
xim
at
ed
to
a
second
orde
r
po
ly
nom
ia
l
f
un
ct
io
n
an
d
it
s
width
can
be
m
easur
ed
by
zero
-
c
ro
s
sin
g
sai
d
f
un
ct
io
n
wit
h
D(q)
=
0.
T
his
width
is
re
ferr
ed
to
as
the
Mult
ifracta
l
Sp
e
ct
ru
m
W
idth
(
MS
W
)
.
In
[17],
the
auth
o
rs
pro
posed
a
m
od
el
f
or
t
raffic
ge
ner
at
i
on
us
i
ng
a
C
on
s
er
vative
Bi
no
m
ia
l
Ca
sca
de
(CBC
)
.
Ca
ll
ed
the
Mult
ifracta
l
Wav
el
et
M
od
e
l (M
W
M)
, it i
s
base
d on the
H
aar
Wav
el
et
tra
n
sf
or
m
, u
si
ng the str
uctu
re
of
(10),
(10)
w
he
re
A(j,k
)
i
s
a
r
an
dom
var
ia
ble
whose
va
lues
r
em
ai
n
i
n
the
[
-
1,1]
int
erv
al
.
T
o
ass
ure
that
|
Wj
,
k|
Uj,k.,
the
scal
in
g
c
oe
ff
ic
ie
nts (j,k)
ha
ve
t
o
be
posit
ive
a
nd
sym
m
etr
ic
to
ze
r
o.
M
oreo
ver,
t
he
m
ulti
pliers
A
j
,
k=
2B
j
,
k,
are
identic
al
ly
distrib
uted
ra
ndom
par
am
et
er
s
within
the
[
0,1]
interval
an
d
sy
m
m
et
ric
to
0.5.
Th
e
relat
ion
s
hi
p
betwee
n
the
M
WM
m
od
el
and
the
casca
des
is
born
f
r
om
us
ing
the H
aar
tr
ansfo
rm
as
a
mu
lt
ipli
cat
ive
casca
de
coeffic
ie
nt.
Said
c
oeffici
ent
is
est
ablishe
d
a
s
a
ra
ndom
var
ia
ble,
with
a
n
av
erag
e
val
ue
of
0.5
a
nd
[0,1
]
r
ang
e
.
Hen
ce
,
posit
ive
data
ge
nerat
ion
with
m
ulti
fr
act
al
cha
racteri
sti
cs
is
assu
red.
T
he
M
W
M
sta
rts
with
the
it
erati
on
va
lue
U0,0,
wh
i
ch
is
distri
bu
te
d
for
t
wo
inter
vals
base
d
on
Bj,
k
a
nd
Bj
,k
-
1.
B
j
,
k
is
a
ra
ndom
nu
m
ber
wit
h
a
beta
(β
)
distri
bu
ti
on.
The
n,
t
hese
va
l
ues
a
r
e
sp
li
t
into
tw
o
on
the
t
hird
scal
e
with
a
di
ff
e
ren
t
Bj,
k
f
or
eac
h
c
ouple.
This
pr
ocess
is
re
peat
ed
un
ti
l
the
Nt
h
scal
e
is
reac
hed,
wh
e
re
t
he
re
will
be
2n
i
nter
vals
with
U
0,0
init
ia
l
fr
act
ions,
re
su
lt
ing
in
a
C
BC
.
Ther
e
fore,
accor
ding
to
(
10),
the
c
o
ns
tr
uction
of
t
he
CB
C
is
giv
e
n by
(
11).
(11)
The
a
uthors
in
[
12]
pro
pos
ed
the
M
ulti
frac
ta
l
-
Hurst
(
MFH)
al
gorith
m
to
gen
e
rate
traces
with
po
sit
ive
data
a
nd
lo
ng
-
range
dep
e
ndency
(L
RD).
T
he
MF
W
obey
s
to
a
powe
r
la
w,
he
nc
e
it
s
f
ractal
it
y,
wh
ic
h
i
m
plies
a
dj
ust
ing
t
he
Hurst
pa
ram
et
er
and
it
s
avera
ge
valu
e.
The
Wa
velet
coeffic
ie
nt
m
ulti
pliers
are
gi
ve
n
by
the Bet
a d
ist
ri
buti
on as
pro
pos
ed
in
[1
7] for
a
ll
scal
es o
f
the
CB
C. See
(
12)
.
(12)
By
adjustin
g
kh
with
t
he
desire
d
value
of
H
a
nd
e
qu
at
in
g
it
s
va
lue
to
th
e
pa
ram
et
er
P
in
the
m
ulti
plica
t
i
ve
casca
de,
a
m
ul
ti
fr
act
al
trace
of
2
n
le
ngth
is
obta
ine
d
f
or
a
gi
ven
H
urst
par
am
et
er
H
a
nd
it
s
corres
pondin
g
aver
a
ge.
Fin
al
ly
,
the
MFH
valid
at
es
the
H
pa
ram
e
te
r
from
the
trace
us
i
ng
t
he
L
D.
I
f
the
e
stim
at
e
d
H
is
no
t
sm
al
l
eno
ug
h
to
c
om
pl
y
with
the
confide
nce
values,
t
he
trace
is
thu
s
dism
issed
a
nd
a
ne
w
one
is
create
d.
T
his
process
is
re
pe
at
ed
un
ti
l
H
rem
ai
ns
within
the
c
onfide
nc
e
val
ues.
The
refor
e
,
the constr
uctio
n of t
he
C
onser
vative Bi
nom
i
a
l C
ascade is
gi
ven b
y
(
11)
a
nd
(13).
(13)
In
[
18]
,
the
aut
hors
pr
opos
e
d
the
Mult
ifracta
l
Hu
rst
S
pectr
um
W
idth (
MH
S
W)
al
gorithm
to
gen
e
rate
po
sit
ive
m
ulti
fr
act
al
traff
ic
w
it
h
the
values
of
the
m
ean,
H
an
d
S
W
bei
ng
receive
d
f
r
om
the
us
er.
T
he
ne
w
m
od
el
distribut
es
two
kh
a
round
the
scal
es
of
the
CB
C,
not
just
one
in
com
par
ison
wi
th
the
MFH
m
et
hod.
The
m
ai
n
go
al
was
to
a
dju
st
t
he
wi
dth
of
the
m
ulti
fr
act
al
sp
ect
ru
m
at
the
l
ast
sta
ge
of
t
he
CB
C
with
kh
=
khw
that
can
m
od
if
y
the
m
ulti
fr
act
al
sp
ect
r
um
.
T
he
khw
is
c
om
pu
te
d
with
a
series
of
e
xp
e
rim
ental
cur
ve
s
that
associat
e the
d
i
stribu
ti
on
of
H i
n
te
r
m
s o
f
t
he sc
al
es as stat
ed
in
. (
14).
1
)
(
)
(
q
H
q
q
1
1
)
(
1
)
(
)
(
q
q
H
q
q
q
q
D
k
j
k
j
k
j
U
A
W
,
,
,
k
j
n
j
k
j
k
j
U
B
W
,
0
,
,
2
1
,
5
.
0
2
2
1
2
1
2
1
2
,
H
k
B
H
H
h
k
j
ot
he
r
w
i
s
e
f
or
x
x
k
k
B
P
D
F
h
h
k
k
h
h
k
j
0
)
1
,
0
(
)
1
(
)
(
)
2
(
)
(
1
1
2
,
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
Bio
-
in
sp
ire
d
r
oute
esti
m
ation
in cog
niti
ve ra
dio
network
s
(
Mi
gu
el
Tuber
quia
)
3099
(14)
The
est
im
at
io
n
of
kh
is
de
te
rm
ined
by
Hh,
wh
ic
h
is
the
Hurst
pa
ram
et
er
of
th
e
final
trace
.
The
e
valuati
on
of
kh
w
is
pro
vid
e
d
by
H
hw
(js,
Ws
),
a
li
ne
ar
f
unct
ion
re
la
ti
ng
the
scal
e
w
her
e
t
he
ca
scade
sta
rts
to
cha
nge
(j
s
)
an
d
the
desire
d
m
ulti
fr
act
al
SW
(
Ws).
The
refor
e
,
th
e
con
st
ru
ct
i
on
of
the
CB
C
is
giv
e
n
by the
(
15).
(15)
wh
e
re
(1
6
)
(1
7
)
(1
8
)
B1j,k
an
d
B2
j
,
k
are
ra
ndom
nu
m
ber
s
with
a
Be
ta
distribu
ti
on.
T
he
al
go
rithm
t
hat
descr
i
bes
the
MFH
W
is
sh
ow
n
in
Fig
ure
2.
D
uri
ng
each
transm
issio
n,
so
m
e
chan
nels
have
gaps
,
ind
ic
at
ing
the
end
of
the co
m
m
un
ic
at
ion
un
ti
l a
ne
w on
e is
b
e
gun. The
se ti
m
e
-
fr
equ
e
ncy s
pace
s can be
us
e
d
t
o
acce
ss
the s
pe
ct
ru
m
dynam
ic
al
l
y.
This
DSA
(
dy
nam
ic
sp
ect
ru
m
acce
ss)
is
t
hen
us
e
d
by
s
econda
ry
us
ers
who
nee
d
to
init
ia
te
a
transm
issi
on.
Each
ti
m
e
t
hat
a
us
e
r
j
um
ps
,
the
handoff
i
ncr
ea
ses
as
well
as
the
ene
r
gy
use
d
f
or
each
hop.
Co
nse
quently
,
the
op
ti
m
iz
ation
proces
s
in
vo
l
ve
s
m
ini
m
iz
ing
the
hand
off
f
unc
ti
on
ex
press
ed
in
the form
o
f
. (1
9).
(19)
Figure
2
.
A
l
gorithm
f
or
t
he
c
on
st
ru
ct
io
n o
f
t
raffic
traces
wi
th m
ulti
fr
act
al
ch
aracte
risti
cs si
m
il
ar
to
the
ra
dio
s
pe
ct
ru
m
2
1
5
.
0
hw
h
H
H
k
j
n
js
j
k
j
k
j
js
j
k
j
k
j
U
B
U
B
W
,
,
2
,
0
,
1
,
2
2
1
,
5
.
0
2
2
1
2
1
2
1
2
,
1
h
H
H
h
k
j
H
k
B
h
h
1
,
5
.
0
2
2
1
2
1
2
1
2
,
2
hw
H
H
hw
k
j
H
k
B
hw
hw
)
,
(
s
s
hw
W
j
f
H
)
(
m
i
n
:
t
ho
f
o
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
, No
.
3
,
J
une
2020
:
30
95
-
3107
3100
Wh
e
re
ho(t),
r
epr
ese
nts
the
nu
m
ber
of
handoff
s
detect
ed
in
a
route
loca
te
d
by
the
E
voluti
on
a
ry
Algo
rithm
(EA).
Th
e
tra
nsm
issi
on
will
not s
how disco
nt
inu
it
ie
s,
an
d w
il
l gu
ara
ntee a
total
tran
s
fer re
gardin
g
t
he
se
r
vice.
2
.
2.
A
nt
c
ol
ony
op
timi
z
at
i
on
Thro
ugh
t
he
us
e o
f p
herom
one trai
ls, ants es
ta
blish a
c
ommun
ic
at
io
n
strat
e
gy w
it
h
eac
h ot
her
t
o
fi
nd
the
path
t
hat
le
ads
to
wards
food
[
19
]
.
Hen
ce
,
as
m
or
e
ants
fo
ll
ow
a
certai
n
pat
h,
the
gre
at
er
is
the
am
ou
nt
of
ph
e
r
om
on
es
de
posit
ed
on
th
e
tr
ai
l.
In
othe
r
wor
ds
,
the
pro
bab
il
it
y
of
an
ant
choosi
ng
a
sp
eci
fic
path
is
increase
d
by
the
nu
m
ber
of
ants
that
ha
ve
crossed
the
path
.
T
he
c
ol
le
ct
ive
beh
a
vi
our
of
a
nts
c
an
be
char
act
e
rized
a
s
a
posit
ive
fe
edb
ac
k
proces
s,
i.e.,
it
is
a
r
ei
nfor
cem
ent
-
ba
sed
p
r
ocess
t
hat
co
nver
ges
fast
as
long
as
there
is
no
lim
it
a
ti
on
i
n
the
env
i
ronm
ent
in
te
r
m
s
of
seeking
a
so
l
ut
ion
[15].
Each
ant
is
identifie
d
a
s
an
a
gen
t
t
hat
l
eaves
a
sig
nal
in
the
wal
ked
path
,
in
flue
nc
ing
f
utu
re
dec
isi
on
s
f
or
ne
xt
age
nts.
T
he
re
fore
,
the
s
et
of
a
nts
do
e
s
not
c
onve
rg
e
t
o
a
sin
gle
so
luti
on
a
nd
in
ste
ad
c
onverge
to
a
s
ubs
pace
of
so
l
utions
w
here
the
best
one
is
chosen
.
T
he
A
nt
Colo
ny
O
pti
m
iz
at
ion
(A
C
O)
m
et
ho
d
is
ba
sed
on
stoc
ha
sti
c
processes
and
is
insp
ire
d
by
the
so
ci
al
be
ha
vi
our
of
ants.
It
ca
n
s
olv
e
c
om
plex
optim
iz
at
i
on
pro
blem
s
[2
0]
wh
ic
h
is
s
uitable
for
CR
N
.
S
om
e
char
act
eris
ti
cs
su
ch
as
pa
rall
el
com
pu
ti
ng
,
sel
f
-
orga
ni
zat
ion
,
a
nd
posit
ive
fee
dba
ck
are
inh
e
ren
t
t
o
th
e
ant
c
olony
m
et
ho
d,
al
lowi
ng
m
ulti
-
agent
op
ti
m
iz
ation
to
ob
ta
in
a
global
s
olu
ti
on,
t
hus
reducin
g
c
om
pu
ta
ti
on
ti
m
es
a
nd
com
plexity
[21]
.
T
he
fi
rst
bio
-
ins
pire
d
al
gorithm
to
ta
ckle
is
the
A
nt
Colo
ny
Syst
e
m
(A
CS),
wh
ic
h
is
base
d
on
t
he
beh
a
vi
our
of
a
nts
f
ol
lowing
pher
omon
e
traces
that
le
ad
to
foo
d
.
I
n
this
researc
h,
t
he
ACS
ca
n
fi
nd
routes
for
the
transm
issi
on
of
c
onti
nuous
data.
F
or
this
m
od
el
,
the
ob
je
ct
ive
functi
on
(
fo)
is
r
e
pr
ese
nted
as
show
n
i
n
(20
),
(
20
)
w
he
re
p(
i
j
|s
p
k
)
is
the
pro
bab
il
it
y
of
c
ha
nn
el
transiti
on.
T
he
refor
e
,
it
is
i
nferr
e
d
t
hat
the
pro
bab
il
it
y
bet
ween
ro
a
ds
is
m
axi
m
iz
ed
w
he
n
the
nu
m
ber
of ants
that ha
ve wal
ke
d on a
path
is
m
axi
m
u
m
.
The
fir
st
ste
p
in
the
im
ple
m
e
ntati
on
process
is
the
descr
ipt
ion
of
the
ph
e
r
om
on
es
spread
by
the
ants
wh
e
n
th
e
entir
e
pac
ket
of
le
ngth
(l
p
)
has
been
s
uccessfull
y
transm
itted.
T
he
pher
om
on
e
va
lues
are
updat
ed
by
al
l
ants
on
ce
a
sp
eci
fic
route
is
com
plete
d.
Up
datin
g
pher
om
on
e
P
u
(
c,t
)
f
or
cha
nnel
c
and
ti
m
e
t
is
ex
pr
ess
e
d
as sho
wn in
(
21)
for an
m
nu
m
ber
o
f
a
nts
w
it
h
an
e
va
porat
ion
rate
∈
(0,1
)
.
(
21
)
Δτ
k
ij
re
pr
ese
nts
the
ants
’
at
tra
ct
ion
feeli
ng
t
o
co
ntinu
e
o
n
th
e
sam
e
chan
nel
or
hop
to
a
nother
one
a
nd
is ex
pr
esse
d
i
n t
he
(
22).
(22)
w
he
re
w
ho
repr
esents
the
assi
gn
e
d
weig
ht
w
hen
the
a
nt
ju
m
p
s
,
i
is
the
c
hannel
posit
ion
at
ti
m
e
t
,
and
j
is
the
cha
nnel
posit
ion
at
ti
m
e
t
+
1
.
I
n
t
he
c
onstr
uction
of
possible
so
l
utio
ns
,
the
a
nts
i
n
the
AC
S
cr
os
s
from
tim
e
t
to tim
e
t
+
lp
, m
aking
pro
ba
bili
sti
c d
eci
sion
s i
n
eac
h
t
i
m
e j
um
p.
The
seco
nd
ste
p
co
ns
ist
s
on
i
den
ti
fy
ing
the
transiti
on
pro
ba
bili
ty
p
(
ij
)
of
t
he
k
th
ant
that
m
ov
es
fr
om
channel
i
at
ti
m
e
t
to ch
a
nnel
j
at
tim
e
t + 1
.
It i
s
giv
e
n by (
23)
[22]
.
(23)
wh
e
re
(
ij)
is
the
set
of
com
ponen
ts
that
do
not
belo
ng
ye
t
t
o
the
pa
rtia
l
sol
ution
s
of
the
ant
k
,
an
d
α
a
nd
β
are
par
am
et
ers
that
con
trol
the
re
la
ti
ve
i
m
po
rtan
ce
of
the
pher
om
on
e
against
heu
risti
c
inf
or
m
at
ion
,
η
ij
=
1
/
d
ij
.
d
ij
is
the
le
ng
th
be
tween
c
hannel
s
i
and
j
at
time
s
t
and
t+1
re
sp
ect
ively
.
The
le
ng
th
is
dete
r
m
ined
by
the
c
hange
in
cha
nnel
.
I
f
the
destinat
i
on
cha
nn
el
wh
e
r
e
the
ant
jum
ps
to
is
the
sa
m
e,
the
n
the
le
ng
t
h
is
D
.
I
f
the
ant
j
um
ps
to
a
different c
ha
nn
el
,
the
n t
he
le
ng
t
h
has
a
v
al
ue of
10D,
as s
how
n
in
(24).
(24)
)
(
m
a
x
:
p
k
o
s
ij
p
f
m
k
ij
k
t
c
t
c
Pu
Pu
1
)
,
(
)
,
(
)
1
(
o
t
h
e
r
w
i
s
e
0
t
o
u
r
i
t
s
e
d
g
e
(
i
j
)
t
h
e
u
s
e
a
n
t
k
if
1
ho
ij
k
w
o
t
h
e
r
w
i
s
e
Pu
Pu
s
ij
p
il
t
l
N
l
ij
t
c
p
k
0
t
o
u
r
i
t
s
e
d
g
e
(
i
j
)
t
h
e
u
s
e
a
n
t
k
if
)
(
)
1
,
(
)
,
(
o
t
h
e
r
w
i
s
e
D
D
ij
10
c
h
a
n
n
e
l
s
a
m
e
i
n
s
t
a
y
a
n
t
k
if
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
Bio
-
in
sp
ire
d
r
oute
esti
m
ation
in cog
niti
ve ra
dio
network
s
(
Mi
gu
el
Tuber
quia
)
3101
Finall
y,
a
gr
ee
dy
sel
ect
ion
of
the
ne
xt
cha
nnel
is
pro
po
se
d
base
d
on
a
ps
e
udo
-
ra
ndom
var
ia
ble
q
.
q
is
c
om
par
ed
to
(
0
<
q
0
<
1)
and
is
a
c
on
sta
nt
cha
nge
us
ed
to
est
ablis
h
th
e
relat
ion
s
hip
betwee
n
e
xp
l
oitat
io
n
and
e
xplo
rati
on.
As
q
0
decr
e
ases,
the
a
nt
chooses
t
he
ne
xt
channel
de
pe
nd
i
ng
on
the
l
evels
of
phe
rom
on
es
wh
ic
h
is
the
e
xp
l
oitat
ion
al
te
rn
at
ive
.
A
s
q
0
increases
,
the
ant
ra
ndom
ly
ta
kes
an
al
te
r
native
r
oute
w
hich
i
s
seen
as
a
n
ad
ve
nturo
us
e
xp
l
orat
ion.
The
ex
plorat
ion
ta
ct
ic
al
so
ta
kes
pla
ce
wh
e
n
the
a
nts
go
beyo
nd
local
m
ini
m
u
m
s,
as sh
o
w
n
in
(
25)
.
(25)
Ther
e
f
or
e,
q
0
is
ass
ociat
ed
with
the
t
oleran
ce
ris
k
a
nd
val
ues
cl
ose
to
1
sug
ge
st
a
hi
gh
e
r
unde
rstan
ding
of
the
i
m
plied
risk.
More
over
,
th
e
exp
l
orat
ion
w
ould
be
ap
propriat
e
at
the
beg
inni
ng
of
the
searc
h
an
d
the
exp
l
oitat
i
on
would
be
a
ppr
opriat
e
at
t
he
en
d
[
22
]
.
It
is
no
te
w
or
t
hy
to
m
ention
th
at
q
is
a
rand
om
nu
m
ber
betwee
n
0
and
1
.
I
f
the
a
nt
s
are
instr
ucte
d
to
ti
lt
the
sea
rch,
the
pro
babi
li
t
y
fu
nctio
n
c
an
be
m
od
ifie
d
f
or th
e sele
ct
ion
of ra
ndom
n
um
bers
.
2
.
3
.
M
ax
feed
ing
op
timi
z
at
i
on
The
sec
ond
bi
o
-
i
ns
pi
red
al
go
rithm
that
is
con
si
der
e
d
is
ca
ll
ed
Ma
x
Feed
ing
(MaF
)
a
nd
is
base
d
on
the
am
ou
nt
of
energy
that
a
n
insect
re
qu
i
re
s
for
it
s
ow
n
nutrit
ion.
Alth
ough
t
he
al
gorit
hm
do
es
not
i
nclu
de
a
rando
m
co
m
pone
nt
seen
in
conven
ti
on
al
evo
l
ution
a
ry
al
gorithm
s,
it
do
es
hav
e
a
n
ada
ptati
on
el
em
ent
fo
r
fam
i
li
ar
scenar
ios.
T
he
ba
sic
idea
of
this
a
lgorit
hm
is
to
achieve
m
axim
u
m
exp
loit
at
ion
of
t
he
sim
ulati
on
scenari
o.
The
obj
ect
iv
e f
unct
ion (
fo) of
this
m
od
el
is repres
ented
i
n (26).
(
26
)
wh
e
re
re
pr
es
ents the
en
e
rg
y
u
se
d by the
k
t
h
a
nt.
The
first
sta
ge
is
the
tra
ns
f
or
m
at
ion
of
the
s
cene
int
o
t
he
pher
om
on
e
trac
es
withi
n
the
ne
st.
The
ne
st
is
com
po
sed
by
al
l
channels
for
al
l
the
tim
es.
The
traces
of
ph
e
r
omon
e
s
are
assig
ned
i
n
pro
port
ion
t
o
the num
ber
o
f free s
paces
within the nest.
T
her
e
fore,
as t
he
n
um
ber
of
ti
m
e instants du
rin
g
w
hich
a c
ha
nnel
is
avail
able
gro
w
s
higher,
t
he
great
er
will
be
the
trai
l
of
ph
ero
m
on
es
le
ft
from
the
nest.
As
a
conseq
ue
nce,
the
insect
is
le
d
by
the
at
tract
ion
of
a
s
pecifi
c
trai
l.
The
al
locat
ion
of
wei
gh
ts
for
ph
e
rom
on
es
is
sta
ti
c
at
first
,
du
e
to
t
he
co
unti
ng
ta
sk
of
bu
sy
a
nd
idle
channels.
Sec
ondly,
a
dyna
m
ic
weig
ht
al
lo
cat
ion
ta
ke
s
pl
ace
in
wh
ic
h
the
ph
e
r
om
on
es
will
evapo
rate
as
the
avail
able
tim
e
slots
decre
a
se.
In
it
ia
ll
y,
t
he
pher
om
on
e
s
ha
ve
high
weig
ht
value
s
accor
ding
to
the
num
ber
of
avail
able
tim
e
slots
an
d
sai
d
val
ues
dec
ay
un
ti
l
they
r
each
0.
Fo
r
the
cal
cul
at
ion
of
ph
e
r
om
on
e
s
,
it
is
e
ssentia
l
to
hav
e
com
plete
kn
owle
dge
of
the
stud
ie
d
sce
nar
i
o.
Algo
rit
hm
1
de
scribe
s
the calc
ulati
on
of the
pher
om
on
e
path
m
a
trix
.
Al
go
rit
hm
1:
Esti
ma
ti
on
of pher
omone trail
s
1
Inpu
t:
Avail
ab
il
it
y_Ma
tri
x
2
Out
p
ut:
P
her
omo
ne
_M
atrix
3
for
ch
annel
1
t
o
the
last
cha
nn
el
of a
vaila
bili
ty
m
at
rix
4
pa
ck
ages
0;
5
counts
0;
6
for
tim
e
1 t
o
the
last
tim
e o
f
av
ai
la
bili
ty
m
at
rix
7
if
Avail
abil
it
y_
Matrix
(
ti
me
,
c
hannel
)
=
1
the
n
8
counts
c
ount
s
+ 1
;
9
el
se
10
counts
0;
11
end
12
if
Avail
abil
it
y_
Matrix
(
ti
me
+
1,
c
hannel
)
=
0 t
hen
13
pa
ck
ages
packa
ges
+
1;
14
Pher
omo
ne_M
atrix(
ti
me
-
c
ounts
+
1
to
time
,
ch
annel
)
=
{
co
un
ts
,
c
ount
s
-
1,
…
,1
};
15
end
16
end
17
end
18
Ret
u
rn
P
her
omo
ne
_M
atrix
ot
he
r
w
i
s
e
s
ij
p
q
q
r
and
om
ne
x
t
p
k
t
c
)
(
m
a
x
a
r
g
0
)
,
(
)
(
m
i
n
:
k
f
o
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
, No
.
3
,
J
une
2020
:
30
95
-
3107
3102
The
ne
xt
sta
ge
involves
fi
nd
i
ng
t
he
best
r
ou
te
based
on
th
e
ph
e
ro
m
on
e
-
insti
ll
ed
paths
.
An
a
ge
nt
is
create
d
f
or
suc
h
la
bor,
w
hich
in
this
case
would
be
a
n
insect
in
cha
r
ge
of
locat
ing
t
he
str
ongest
tr
ace
of
ph
e
r
om
on
e
in
channel
m
at
ti
m
e
t
1
.
The
inse
ct
will
then
m
ov
e
to
the
c
ha
nnel
with
the
l
on
gest
phe
ro
m
one
trai
l
and
will
c
on
si
der
the
le
ng
t
h
of
the
t
ran
sm
issi
on
pac
ket
(
l
p
).
T
he
t
ransm
i
t
te
d
pa
cket
(
Pt
)
an
d
t
he
am
ount
of
el
apsed
ti
m
e
(
ti
)
in
the
c
ha
nn
el
will
be
s
ub
t
racted.
G
uid
ed
by
the
ph
ero
m
on
e
,
t
he
insect
will
transm
i
t
the
pac
ket
unti
l
al
l
traces
disap
pear
an
d
then
i
ntell
igently
sel
ect
a
new
cha
nnel
on
ce
the
num
ber
of
ph
e
r
om
on
e
s
re
aches
1
.
The
r
efore,
it
will
hav
e
e
nough
t
i
m
e
to
change
fr
om
chan
nel
m
to
channel
n
and
gu
a
ra
ntee
the
con
ti
nuit
y
of
the
ser
vice.
Algo
rit
hm
2
des
cribes
the
ste
ps
fo
ll
owed
by
the
bug
to
det
erm
ine
the
r
oute
to e
xplo
re
.
Al
go
rit
hm
2:
Ro
ute
e
xpl
orati
on
by b
ugs
1
Inpu
t:
Pher
om
on
e
_M
atrix, A
vailabili
ty
_M
at
rix
2
Out
p
ut:
Ti
mes
_Array,
Ch
anne
ls_Arr
ay
3
In
it
ia
li
zat
ion
t
i
= 0
,
P
t
=
0;
4
w
hil
e
(
lp
-
pt
)
≥ 0
5
In
c
rease
ha
ndoff
i
n on
e;
6
t
i
U
pd
at
e
Tim
e ;
7
Times_Arr
ay
(i
)
m
ax
(
P
he
ro
m
one_
Matri
x
(
t
i
, all col
um
n)
);
8
if
Times
_Array
(i)
=
0
th
e
n
9
No Mo
re ro
ute
s;
10
Break
;
11
el
se
12
Chan
nels_Ar
r
ay
(i)
Fin
d P
os
it
ion (
Ti
mes
_Array
(i
))
i
n
Pher
omo
ne_M
atrix
(
t
i
, all col
um
n)
;
13
pt
t
i
+
Times
_Array
(i
);
14
Ne
w_Availa
bili
ty
_Ma
tri
x
Elim
inate
trail
routed;
15
Pher
omo
ne_M
atrix
update
u
si
ng
Al
go
rit
hm
_1
(
Ne
w_Availa
bili
ty
_Ma
tri
x
);
16
end
17
end
18
Ret
u
rn
Ti
mes
_Array,
Ch
anne
ls_Arr
ay
The
i
ns
ect
will
re
peat
the
sa
m
e
process
unti
l
the
cu
rr
e
nt
tr
ansm
issi
on
is
com
plete
d
or
a
n
e
qu
i
valent
sta
te
of
fi
nd
i
ng
a
ne
w
t
rans
m
issi
on
r
oute
is
entere
d.
A
ft
erw
a
rds,
t
he
a
vaila
bili
ty
m
atr
ix
is
up
dated
with
Algorithm
1
and
a
nothe
r
ins
ect
can
fin
d
a
diff
e
re
nt
r
ou
t
e
(so
l
utio
n)
.
This
pr
ocess
i
s
rep
eat
e
d
k
tim
es,
by
k
inse
ct
s,
t
o
fin
d
k
r
ou
te
s
unti
l
there
is
discon
ti
nuit
y
in
t
he
tran
sm
issi
on
.
Wh
e
n
on
e
of
t
he
i
ns
ect
s
m
eets
it
s
pur
po
se
(i.e.,
t
ran
sm
it
s
a
set
a
m
ou
nt
of
data),
the
nest
is
updated
,
th
us
el
i
m
inati
ng
th
e
route
f
or
the
foo
d
searchi
ng
ta
sk
carried
out
by
the
kt
h
insect
.
Since
the
i
ns
ec
t
travels
the
sa
m
e
distance
f
r
om
beg
in
ning
to
e
nd
,
the
on
ly
discr
i
m
inati
ng
par
a
m
et
er
is
the
nu
m
ber
of
j
um
ps
m
ade
by
the
insect
to
avo
id
inter
r
upti
ng
the
transm
issi
on
.
E
ve
ry
ti
m
e
that
the
insect
j
um
ps
,
it
s
ene
r
gy
co
nsum
ption
(
)
increa
ses
wh
ic
h
m
eans
that
the insect t
hat
consum
es the least
am
ou
nt of
energy i
s
un
e
quiv
ocall
y t
he o
ne
that
fin
ds
t
he
b
est
route.
3.
RESU
LT
S
AND DI
SCUS
S
ION
Af
te
r
im
ple
m
e
nting
the
MF
H
W
al
gorithm
,
the
gen
e
rate
d
tra
ff
ic
is
a
ppr
oac
hed
with
the
sam
e
char
act
e
risti
cs
of
the
or
i
gina
l
traff
ic
.
T
he
refor
e
,
the
fir
st
ste
p
befor
e
m
ov
in
g
on
to
sai
d
gen
e
rat
ion
is
the
est
i
m
at
ion
of
the
m
ean,
th
e
Hurst
pa
ram
et
er
an
d
the
wi
dt
h
of
the
m
ulti
fr
act
al
sp
ect
r
um
.
The
LD,
MD
an
d
MS
are
est
ablishe
d
for
each
channel
in
the
sp
ect
ru
m
and
the
input
valu
es
are
est
i
m
ated
f
or
the
generate
d
traff
ic
.
The
461
ti
m
e
series
ob
ta
ine
d
for
each
cha
nn
el
hav
e
a
m
ean
value
of
8.1
262
within
the
[5.85
29,
18.
4342]
range
and
with
a
sta
nd
a
r
d
de
viati
on
of
1.634
6.
O
ut
of
the
461
c
hannels
of
the
rad
i
o
-
el
ect
ric
sp
ect
r
um
,
25
wer
e
f
ound
for
H
<
0.
5,
tw
o
c
ha
nn
el
s
for
H
>
1
an
d
43
4
c
hannels
in
the
[0.5,
1]
range.
The
a
ve
rag
e
va
lue
of
t
he
est
i
m
at
ed
H
is
0.6
508
with
a
sta
ndar
d
de
viati
on
of
0.1
023.
T
he
sp
ect
ral
wi
dths
ha
ve
an
aver
a
ge
val
ue
of
1.264
5
and
a
sta
ndar
d
dev
ia
ti
on
of
0.364
7
within
th
e
[0
.
7285,
4.7
910]
ra
ng
e
.
W
it
h
thes
e
est
i
m
ation
s,
it
is
po
ssible
t
o
ge
ner
at
e
ne
w
tim
e
series
with
sim
il
ar
char
act
erist
ic
s.
The
ref
or
e
,
after
the
predict
io
n
process
is
c
oncl
ud
e
d,
trace
s
wer
e
obta
ined
with
the
fo
ll
ow
i
ng
c
ha
racteri
sti
cs:
m
ean
traces
within
the
[
3.893
5,
18.
4175
]
range,
an
a
ver
a
ge
valu
e
of
7.9
151
a
nd
a
sta
nd
a
rd
de
viati
on
of
1.812
9.
H
is
within
t
he
[0
.
5904,
0.9
292]
ra
nge
with
an
ave
ra
ge
va
lue
of
0.6
686
and
a
sta
ndar
d
dev
ia
ti
on
of
0.0
805.
The
wi
dth
of
the
m
ulti
fr
act
al
sp
ect
ru
m
has
a
[0
.
7387,
4.6
036]
range,
an
ave
rag
e
va
lue
of
1.3
180
and
a stan
dard
dev
i
at
ion
of 0.3
645.
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
Bio
-
in
sp
ire
d
r
oute
esti
m
ation
in cog
niti
ve ra
dio
network
s
(
Mi
gu
el
Tuber
quia
)
3103
Ba
sed
on
the
m
ul
ti
fr
act
al
ser
ie
s
ge
ner
at
ed
f
or
each
ch
a
nne
l,
three
avail
ab
il
ity
m
at
rixes
wer
e
create
d
with
100
ti
m
e
s
and
10
0
co
nse
cutive
cha
nn
el
s.
In
Fig
ur
e
3,
the
three
pr
e
dicte
d
avail
abili
ty
m
at
rixes
are
m
ark
ed
in
re
d,
the
or
iginal
a
vaila
bili
ty
m
atr
ix
f
or
10
0
tim
es
fr
om
chan
ne
l
1
up
to
cha
nn
el
30
0
is
m
a
rk
e
d
in
blu
e
an
d
t
he
f
ai
le
d
predict
io
n
ti
m
e
is
m
ark
ed
i
n
black.
T
he
pr
e
dicti
on
e
rror
s
do
not
s
urpass
34
%
.
Althou
gh
the
pe
rce
ntage
error
f
or
the
a
va
il
abili
ty
m
at
ri
xes
is hig
h,
thi
s
is
com
pen
sat
ed
with
acc
ur
a
ci
es
in
the
pre
di
ct
ion
of the
H
ur
st
pa
ram
et
er o
f 4.
054
7%
a
nd 8.85
83%
for
t
he wid
th of t
he
m
ulti
fr
act
al
sp
ect
r
um.
Figure
3
.
Pr
e
di
ct
ed
avail
a
bili
t
y
m
at
rices using the
synt
hesis
of m
ulti
fr
act
al
traces
Unde
r
the
pre
dicte
d
scena
ri
os
,
it
is
pr
op
ose
d
to
use
bioi
ns
pire
d
al
gori
thm
s
with
the
purpose
of
cal
culat
ing
routes
the
re
by
m
ini
m
iz
ing
ha
ndoff.
I
n
a
si
m
ulati
on
sce
na
ri
o
c
on
sist
i
ng
of
461
ch
ann
el
s
,
it
is
pr
opos
e
d
to
fin
d
r
ou
te
s
for
a
transm
issi
on
of
70
-
ti
m
e
instants.
For
the
Ma
x
Fee
din
g
(MaF)
al
go
rithm
,
the
res
ults
ob
t
ai
ned
by
the
m
et
h
od
offer
an
ad
va
ntage
since
it
has
co
nf
i
gurati
on
vari
ables.
The
al
gorithm
receives
t
he
s
i
m
ulati
on
scenari
o
as
a
n
input
pa
ram
et
e
r
an
d
the
n
find
s
the
best
routes
th
us
re
du
ci
ng
the
num
ber
of
possible
jum
ps
i
n
eac
h
it
er
at
ion
.
The
al
gorithm
boos
ts
perform
ance
by
div
idi
ng
the
tota
l
nu
m
ber
of
ch
ann
el
s
t
o
op
ti
m
iz
e
bo
th
t
he
searc
h
a
nd
the
j
um
ps
car
r
ie
d
out
in
rel
at
ively
con
ti
nuous
fr
e
qu
e
ncies.
Fi
gure
4
s
hows
a
fr
act
io
n
of
t
he
res
ults.
In
this
exam
ple,
the
first
div
isi
on
of
the
c
ha
nn
el
s
delivere
d
19
routes,
f
r
om
wh
i
ch
the
m
ini
m
um
han
do
ff
was
5,
the
m
axim
u
m
han
do
ff
was
19
a
nd
the
a
ve
rag
e
hand
off
for
th
e
first
div
isi
on
was
8.94.
When
run
ning
the
Ma
F
al
go
rith
m
ov
er
al
l
chan
nels
,
266
di
fferent
routes
wer
e
ob
ta
ined
a
nd 10
of them
are
s
hown in Fi
gure
4.
Fo
r
t
he
exec
ut
ion
o
f
the
A
CO
(
A
nt
C
ol
on
y
O
ptim
iz
a
t
ion)
al
gorithm
ov
e
r
the
461
predict
e
d
channels,
the
ant
s’
f
ollo
w
-
up
pa
ram
et
ers
m
us
t
be
set
.
P
aram
et
er
config
ur
at
io
n
i
s
base
d
on
previ
ou
s
si
m
ulati
on
s
where
the
ef
fect
s
of
va
r
ying
α
,
β
an
d
q
o
w
ere
c
on
si
der
e
d
re
ga
rd
i
ng
ha
ndoff
r
ed
uctio
n.
The
final
config
ur
at
io
n
i
s
sta
te
d
as:
300
it
erati
on
s,
3
ants
per
div
isi
on,
30
co
ntinuou
s
sim
ulati
on
channels,
D=
2,
=6
0,
α
=0.
05,
β=1
and
q
o
=0
.05.
In
Fi
gure
5,
t
he
first
five
ants
are
visu
a
li
zed
with
the
final
pat
h
re
su
lt
s.
This
e
xp
e
rim
e
nt
hel
ps
in
f
ollow
in
g
t
he
path
of
t
he
fi
rst
f
ive
ants.
T
he
div
isi
on
of
ch
ann
el
s
i
nto
fr
a
gm
ents
seeks
to
el
im
i
nate
local
m
ini
m
u
m
s.
26
6
r
oute
s
wer
e
fou
nd
with
the
Ma
F
al
gorithm
a
nd
46
for
the
ACO
al
gorithm
.
It
is
w
or
th
m
entionin
g
t
hat
the n
um
ber
of
r
oute
s
determ
ined
by the
AC
O
m
et
ho
d
is
pro
portio
nal
t
o
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
, No
.
3
,
J
une
2020
:
30
95
-
3107
3104
the
nu
m
ber
of
ants
locat
e
d
i
n
each
cha
nnel
div
isi
on.
Gi
ve
n
t
he
nu
m
ber
of
it
erati
on
s
of
the
ACO
al
go
rithm
,
the
ant
op
ti
m
izes
pr
e
vious
r
oute
s
in
each
it
erati
on
wh
ic
h
m
eans
that
there
are
n
r
oute
s
at
the
end
of
t
he
n
th
ite
rati
on
.
This
m
eans
that
t
he
num
ber
of
r
oute
s
is
eq
ual
t
o
the
nu
m
ber
of
it
erati
on
s
ti
m
e
s
the
num
ber
of
a
nts
tim
es
the
num
ber
of
c
ha
nn
el
di
visions.
I
n
the
c
urren
t
sc
enar
i
o,
t
hr
ee
a
nts
wer
e
place
d
i
n
eac
h
c
ha
nn
el
div
isi
on,
30
0
i
te
rati
on
s
we
re
m
ade,
an
d
461
ch
a
nn
el
s
we
re
div
i
ded
into
15
sect
io
ns
le
a
di
ng
t
o
13500
r
ou
te
s
of
wh
ic
h
46
r
ou
te
s
pr
e
sent
the
lowe
st
ha
ndoff
s
.
To
c
om
par
e
the
res
ults
of
t
he
r
oute
s
com
pu
te
d
for
al
l
channels,
they
are
dis
play
ed
in
a
norm
al
ized
histo
gr
am
in
Fi
gure
6.
T
he
sta
ti
sti
cs
ar
e
sho
wn
in
re
d
for
the MaF al
gorithm
an
d
in
bl
ue
for
t
he
A
nt
C
olony m
et
ho
d.
Fig
ure
4
.
First
channel
div
isi
on
disp
la
yi
ng
10
routes
fou
nd b
y t
he M
aF alg
or
it
hm
Fig
ure
5
.
A
C
O
algorit
hm
d
isp
la
yi
ng
the
first
five
routes
fou
nd in
a synt
hetic
pre
dicti
on
proce
ss
of
the av
ai
la
bili
ty m
a
trix
Fig
ure
6
.
Pro
ba
bili
ty
d
ist
ribu
t
ion
of ro
utes
ha
ndoff f
ound
by
EA
Figure
6
gat
he
rs
the
per
ce
nt
ages
of
the
num
ber
of
i
nce
sts
for
a
spe
c
ific
nu
m
ber
of
ha
ndoffs
.
The
ACO
al
gorithm
rev
eal
e
d
the
highest
nu
m
ber
of
rou
te
s
with
a
ha
ndoff
betwee
n
10
a
nd
12
whic
h
is
equ
i
valent
to
39%
of
the
cal
culat
ed
r
ou
te
s
.
I
n
co
ntrast,
the
Ma
F
al
go
rit
hm
rev
ea
le
d
m
or
e
handoffs
betw
een
7
and
8
w
hich
i
s
equ
i
valent
to
21
%
of
t
he
c
al
culat
ed
r
ou
te
s.
The
routes
f
ound
by
the
A
CO
m
e
tho
d
ca
n
be
sp
rea
d
within
the
5
-
17
ha
nd
off
range
a
nd
tho
se
cal
culat
ed
by
the
Ma
F
al
gorithm
rem
ai
n
within
th
e
5
-
40
Evaluation Warning : The document was created with Spire.PDF for Python.