Int
ern
at
i
onal
Journ
al of Ele
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
9
, No
.
5
,
Octo
ber
201
9
, pp.
4226
~4
237
IS
S
N
: 2
088
-
8708
,
DOI: 10
.11
591/
ijece
.
v9
i
5
.
pp
4226
-
42
37
4226
Journ
al h
om
e
page
:
http:
//
ia
es
core
.c
om/
journa
ls
/i
ndex.
ph
p/IJECE
Heterog
eneous n
etwork o
ptim
izat
i
on using
ro
bu
st
power
-
and
-
re
s
ource bas
ed
algorit
hm
Savit
a
P
at
il
1
, A.
M.
Bh
avik
atti
2
1
AM
C
Engi
nee
r
i
ng
Coll
ege
,
Indi
a
2
Com
pute
r
Scie
n
ce
and Engi
ne
ering
,
Bhe
emanna
Khandre
Insti
tute
of Te
chnol
og
y
(
BKIT
)
,
Ind
ia
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Sep
17
, 201
8
Re
vised
Ma
r
13
, 2
01
9
Accepte
d
Apr
24
, 201
9
In
orde
r
to
m
ee
t
the
inc
r
ea
sing
m
obil
e
dat
a
-
tra
f
fic
,
spat
ial
densifi
c
at
ion
o
f
net
work
with
seve
ra
l
low
-
power
nodes,
the
high
-
power
m
ac
ro
BS
and
HetNe
t
are
the
m
aj
or
ke
y
en
abl
ing
solu
ti
on
.
How
eve
r,
the
HetNe
t
is
unpla
nned
in
n
a
ture
,
ca
uses
irregula
rities
and
in
te
rfe
r
ences
tha
t
without
a
n
y
user
association
rule
s.
Th
e
ap
propria
t
e
dep
lo
y
m
ent
of
th
e
f
emto
-
ce
l
l
i
n
HetNe
t
ca
n
p
rov
ide
eff
e
ct
iv
e
tr
af
fic
off
loa
ding
,
w
her
e
the
alleviat
e
m
obbing
in
the
m
a
cro
-
c
el
l
s
ca
n
de
crease
th
e
power
consum
pti
on
th
ere
for
e
it
opti
m
iz
e
s
the
user
expe
r
ience
.
Moreov
er,
t
he
prote
c
ti
on
is
al
so
important
f
or
the
m
ac
ro
and
femto
ce
l
l
u
sers
in
a
n
et
wor
k
through
m
ai
nt
ai
ning
the
m
in
-
m
ax
le
ve
l
of
int
e
rf
ere
n
ce
s.
In
thi
s
pape
r
,
we
proposed
RP
RA
tha
t
comprises
two
robust
appr
oac
h
such
a
s
robust
power
-
cont
roller
and
t
he
robust
cha
nn
el
-
a
ll
oc
at
ion
appr
oac
h
,
which
ca
n
improve
th
e
spec
tr
al
eff
icie
nc
y
and
user
ex
per
ie
n
ce
s
a
t
lower
net
work
cove
rag
e
ar
ea
s
via
eli
m
ina
ti
ng
the
wee
k
cov
er
age
zon
es.
Also
provide
high
user
rat
e
c
onnec
t
ion
b
y
e
ffe
ctive
interfe
r
enc
e
in
a
n
eff
icient
spec
tru
m
,
lower
ing
in
tra
nsm
ission
power
and
cost
-
ef
fec
t
ive
ness
via
le
ss
ti
m
e
del
a
y
.
To
show
the
eff
ective
n
ess
of
ou
r
proposed
m
odel
we
h
ave
compare
d
wit
h
seve
ral
exi
st
i
ng
te
chn
ique
s
and
we
go
t
signifi
c
ant
improvem
ent
in
thr
oughput,
al
so
r
educ
t
ion
in
ti
m
e
del
a
y
and
tra
nsm
ission powe
r.
Ke
yw
or
d
s
:
Ba
se stat
ion
s
(
BS)
Heter
og
e
ne
ou
s
n
et
w
orks
(H
et
Nets)
Robust
powe
r
-
and
-
re
source
base
d
al
gorit
hm
(
R
PRA)
Copyright
©
201
9
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
:
Savita Pat
il
,
AMC En
gi
neeri
ng
C
ollege
,
18
t
h K.M,
Ban
nerg
hatta
Mai
n R
d, Be
ngal
uru
, K
a
r
nataka
56
0083, I
nd
ia
.
Em
a
il
:
sa
m
patil9
49@
gm
ai
l.c
om
1.
INTROD
U
CTION
The
prese
nt
cel
lular
netw
ork
faces
the
pro
bl
e
m
of
huge
inc
rem
ent
of
data
traff
ic
,
in
t
hat
the
sp
ect
ral
eff
ic
ie
ncy
al
lo
cat
ion
an
d
design
is ver
y
im
p
or
ta
nt.
I
n
ge
ne
ral,
the
co
nce
pt
of
f
reque
ncy
reu
se
is
the
m
ajor
key
for
4
-
Ge
ner
at
i
on
a
nd
th
e
fu
t
ure
cel
lular
net
works;
howe
ve
r,
the
Het
Nets
hav
e
at
tract
ed
lot
of
at
te
ntio
n
of
the
researc
hers
in
LTE
(L
ong
-
Te
rm
-
Evo
luti
on
)
wireless
netw
ork
[
1].
Mo
reover
,
the
pote
nt
ia
l
of
HetNets
has
us
e
d
to
opti
m
i
ze
the
sp
ect
ral
eff
ic
ie
ncy
an
d
us
er
e
xp
e
rien
ces
at
ver
y
de
ns
e
en
vir
onm
e
nt
‘or’
lo
we
r
ne
twork
cov
e
ra
ge
area
s
,
th
ough
the
de
plo
ym
ent
of
HetNets
to
sm
al
l
-
cov
e
ra
ge
ar
ea
(f
em
to
‘or’
Pico)
a
nd
lo
w
powe
r
cel
l
with
in
th
e
m
acro
-
cel
l
env
i
ronm
ent
is
gen
e
rall
y
el
i
m
inate
the
wee
k
co
ver
a
ge
z
ones,
al
so
pro
vide
the
higher
“
s
pectr
al
-
eff
ic
ie
ncy”
[2]
.
The
a
ppr
opriat
e
de
plo
y
m
ent
of
the
“
fem
to
-
cel
l”
in
HetN
et
can
pro
vid
e
eff
ect
ive
traf
fi
c
offl
oad
i
ng,
wh
e
re
t
he
al
l
eviat
e
m
ob
bing
in
the
m
ac
ro
-
cel
ls
can
de
crease
t
he
powe
r
consum
ption
there
fore
it
op
t
i
m
iz
es
the
us
e
r
ex
per
ie
nce.
I
n
or
de
r
to
de
ve
lop
m
acro
-
fe
m
to
ba
sed
H
et
Nets
,
the
m
ai
n
concern
is
to
m
a
nag
e
the
c
o
-
c
hannel
interfe
r
ences
that
oc
cu
rs
becau
se
of
s
pectru
m
sh
ari
ng,
the
interfe
re
nc
es
of
c
o
-
c
ha
nnel
m
ay
beco
m
e
worse
due
to
f
e
m
to
BSs
that
instal
le
d
thr
ough
e
nd
-
us
er
s
w
it
ho
ut
any
of
ce
ntrali
zed
pla
nn
i
ng
proce
dure.
Ge
ne
rall
y,
fem
to
BSs
are
de
ploy
ed
unde
r
m
ac
ro
-
cel
l
areas
w
it
ho
ut
any
warnin
g
due
to
sp
e
ci
al
req
ui
rem
ent
of
cl
ie
nt,
therefo
r
e
the
la
ck
cent
rali
zed
-
cel
l
pla
nn
i
ng
t
he
eff
e
ct
of
interfe
ren
ce
s i
n
tw
o
-
ti
er “m
a
cro
-
fem
to H
et
Nets”
va
ry c
onsidera
bly i
n
a
c
onve
ntion
al
cell
n
et
w
orks.
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
Hetero
ge
neous
n
et
work
opti
m
izati
on u
si
ng r
obus
t
powe
r …
(
Sa
vi
ta
Pa
ti
l
)
4227
A
m
a
in
diff
ic
ulty
occu
rs
i
n
dep
loym
ent
process
of
He
tNet
in
order
to
pro
vid
e
hi
gh
us
e
r
rate
connecti
on
by
eff
ect
ive
inte
rf
e
ren
ce
in
a
sp
ect
r
um
eff
ic
ie
nt
and
co
st
-
eff
ect
ive
m
ann
er
,
seve
ral
publis
he
d
works
ha
s
f
oc
use
d
on
“co
-
cha
nn
el
inter
fer
e
nc
e”
diff
ic
ulty
t
hat
arises
betw
een
fem
to
BSs
and
m
acro
BSs
[1
]
.
Thro
ugh
st
uddi
ng
se
ve
ral
w
orks
,
the
dem
on
strat
io
n
fem
to
BS
as
the
c
ogniti
ve
r
adi
os
per
m
it
s
fe
m
to
BS
to
sense
m
acro
BS
us
e
r
perform
ance
to
co
nt
ro
l
the
ty
pe
of
inter
fer
e
nce
at
the
avail
abl
e
sp
ect
ral
res
ources
;
howe
ver,
the
pro
blem
relat
ed
to
power
c
on
t
ro
ll
er
ha
s
disc
us
se
d
,
w
he
re
the
netw
ork
c
onfi
gurati
on
is
f
e
m
to
-
m
acro
two
-
ti
er
netw
orks
thr
ough
pro
posin
g
dow
nlin
k
ba
sed
powe
r
al
locat
ion
al
gori
thm
,
wh
ic
h
con
sist
i
nfor
m
at
ion
of
channel
f
or
al
l
inv
olv
e
d
li
nk
s
i
n
net
wor
k.
Wh
il
e
co
nsi
der
in
g
a
cent
rali
zed
base
d
fem
to
gateway
proce
ss
that
cal
le
d
a
s
f
us
io
n
ce
nter
,
is
basical
ly
reli
es
on
gate
w
ay
connecti
on
in
or
der
t
o
distrib
ute
and
colle
ct
the
necessa
ry
c
ha
nn
el
i
nter
fer
e
n
ce
inf
or
m
at
ion.
A
n
e
xten
sive
am
ou
nt
of
w
orks
has
bee
n
done
with
res
pect
t
o
use
r
as
so
ci
a
ti
on
in
t
he
H
e
tNet
s
co
ns
ide
r
ing
dif
fer
e
nt
ty
pe
of
sce
nar
i
os
,
li
ke
as
in
[3
]
ha
s
pro
po
se
d
a
joi
nt
opti
m
iz
at
ion
proce
dure
of
use
r
ass
ociat
ion
su
c
h
as
c
ha
nn
e
l
sel
ect
ion
a
nd
transm
issi
on
powe
r
has
co
ns
id
ere
d
to
m
ini
m
a
li
z
e
the
as
per
t
he
us
e
r
thr
oughput.
Wh
il
e
in
[
4],
the
aut
hor
pro
pose
d
for
the
adap
ti
ve
ass
oci
at
ion
rule i
n
order
to
e
nlarge t
he
co
ns
ist
ing
num
ber
o
f
use
rs
thr
ou
gh
decr
e
asi
ng
the total usa
ge
cost
of
r
adi
o
r
eso
ur
c
es
unde
r
a
HetNet
e
nv
iro
nm
ent
with
relay
nodes;
w
her
eas
in
[5
]
t
hey
us
e
d
i
nteg
rated
us
er
ass
ociat
ion
a
nd r
es
ource
al
locat
ion
i
n order
to
i
ncr
ease
the m
et
ric o
f p
rop
or
ti
onal
f
ai
r
(
i.e.
, log
-
rate s
um
).
The
decen
tr
al
iz
ed
ba
sed
i
nter
fer
e
nce
m
anage
m
ent
protoc
ol
for
the
c
ogniti
ve
fem
to
BSs,
wh
e
re
eac
h
fem
to
BSs
wil
l
al
locat
e
i
ts
sp
ect
ru
m
thro
ug
h
co
ns
ide
rin
g
the
pote
ntial
interfe
ren
ce
a
nd
i
nterf
e
re
nce
fro
m
the
neig
hborh
ood
f
em
to
-
cel
ls
is
discuss
e
d
i
n
[
6].
The
m
et
hodo
l
og
y
ne
cessary
f
or
c
omm
un
ic
at
ion
based
interfe
re
nce
ch
ann
el
a
re
ver
y
eff
ect
ive
am
on
g
the
neig
hborh
ood
cha
nnel
,
thou
gh
the
dir
ect
li
nk
in
bet
wee
n
m
acro
BS
a
nd
fem
to
BS
are
ver
y
im
po
rta
nt
in
a
n
ch
an
nel
inter
fer
e
nc
e,
in
[
7]
they
discusse
d
a
bo
ut
the
resou
rces
al
locat
ion
optim
izati
on
proces
s.
In
[
8],
they
va
luate
the
ch
ann
el
al
locat
ion
strat
e
gy
throu
gh
app
ly
in
g
t
he
gam
e
-
theor
y
ap
proac
h
f
or
c
ogniti
ve
fem
to
BS
ra
dio
m
anag
e
m
ent
resou
rce
s;
furthe
rm
or
e,
in
[9
]
they
stu
died
the
distrib
utio
n
ap
proac
h
for
the
sel
f
-
ori
gi
na
te
d
cel
l
netw
ork
i
n
orde
r
t
o
c
ontrol
the
t
ran
s
m
it
powe
rs
on
sev
eral
diff
e
re
nt
channels
that
to
m
ai
ntain
us
er
qu
al
it
y
and
th
e
channel
co
ndit
ion
s
.
A
th
re
sh
ol
d
base
d
sp
ect
rum
sensing
is
ve
ry
diff
ic
ult
to
desig
n,
wh
e
re
i
n
orde
r
to
get
the
hi
gh
desire
d
trad
e
-
off
in
be
tween
ou
ta
ge
pro
ba
bili
ty
and
sp
at
ia
l
fr
eq
ue
ncy
a
par
am
et
er
need
to
be
tun
e
d
car
efu
ll
y,
the
m
or
e
le
sser
the
thr
esh
old
of
sp
ect
ru
m
sensing
t
he
lo
we
r
the
ag
gr
e
gat
e
interfe
re
nce
and
la
r
ger
t
he
s
pectr
um
sensing
ra
nge.
H
owever,
the
reuse
of
sa
m
e
fr
eq
uen
cy
channel
can
be
done
afte
r
the
la
rg
er
‘s
patia
l
intervals
’
that
te
nd
s
in
l
ow
s
patia
l
fr
e
qu
e
ncy
re
use
,
the
analy
sis
and
desi
gn
i
ng
of
this
ty
pe
of
trade
-
off
in
two
-
ti
er
HetNet
with
fem
to
-
ce
ll
s
is
discusse
d i
n [10].
Howe
ver,
the
m
ai
n
aim
of
po
we
r
co
ntr
oller
is
to
decr
eas
e
the
transm
is
sion
po
wer
th
at
te
nd
s
to
decr
ease
the
hig
h
power
c
on
s
um
ption
,
al
so
r
edu
ce
s
the
interfe
ren
ce
at
inter
-
cel
l
that
is
ver
y
i
m
po
rtant
as
pe
r
the
ab
ov
e d
isc
us
sio
n;
the
refo
re,
the
al
te
ri
ng
the
powe
r
tra
nsm
issi
on
with respect
to
r
eso
urce
bl
ock
is
ess
entia
l
to
achieve
hi
gh
bit
-
rate
in
fem
to
-
cel
l
a
nd
pico
-
cel
l,
wh
e
re
the
oc
curred
inte
rf
e
r
ence
can
be
reduce
sign
ific
a
ntly
in
sm
a
ll
-
cel
l
network.
T
her
e
f
ore,
the
c
ha
nn
el
capacit
y
and
c
ov
e
ra
ge
of
poor
area
un
der
a
H
et
Net
scenari
o
is i
m
po
rta
nt for
e
nha
nced use
r
se
r
vices. Mo
reove
r, t
he pr
otect
ion
is al
so
im
po
rtant for
t
he
m
acro
a
nd
fem
to
cel
l
us
ers
in
a
netw
ork
thr
ough
m
ai
nta
ining
the
m
in
-
m
ax
le
vel
of
i
nterf
e
re
nces,
t
he
QoS
ef
fecti
ven
e
s
s
unde
r
a
traf
fic
is
m
ajo
r
fact
or
because
thi
s
can
dam
age
the
res
ource
al
locat
ion
pr
oc
edure
that
te
nd
s
t
o
degra
de
the Q
oS of
u
se
rs
in a sm
a
ll n
et
work
(i.e.,
Pico or
fe
m
to cell
s).
I
n
ge
ner
al
, th
e inc
r
e
m
ent o
f
nu
m
ber
of
m
ob
il
e
us
ers
c
an
ca
us
e
the
de
crem
ent
in
Q
oS
due
to
m
ore
ba
ndwidt
h
r
equ
i
rem
ent,
in
orde
r
to
ha
ndle
this
un
ce
rtai
nty,
th
e
sever
al
fem
t
o
BSs
are
re
quired,
the
refore
the
pro
po
se
d
a
lgorit
hm
sh
ou
ld
be
r
obus
t
in
natu
re
so
that
it
can
handle
un
ce
rtai
nties
eff
ect
ive
ly
.
In
this
pa
pe
r,
we
pr
opos
e
d
RPR
A
t
hat
c
om
pr
ise
s
two
r
obus
t
appr
oach suc
h as Ro
bu
sti
fie
d Po
wer
C
ontroll
er (
RPC
)
a
nd the Rob
us
t C
ha
nn
el
-
res
ource
Allo
cat
ion A
ppr
oa
c
h
(RCAA
),
w
hic
h
ca
n
im
pr
ov
e
the sp
ect
ral ef
f
ic
ie
ncy and u
s
er expe
riences
at
low
er
n
et
work
c
overa
ge
a
r
eas via
el
i
m
inati
ng
t
he
wee
k
c
ov
e
ra
ge
zo
nes.
Als
o
pro
vid
e
high
use
r
r
at
e
co
nnec
ti
on
by
ef
fecti
ve
inte
rf
e
ren
ce
in
a
n
eff
ic
ie
nt
s
pec
trum
,
lower
i
ng
in
tra
ns
m
i
ssion
po
wer
an
d
c
os
t
-
e
f
fecti
ven
es
s
via
le
ss
tim
e
delay
.
The
im
ple
m
ent
at
ion
of
po
wer
con
t
ro
ll
er
us
e
d
real
ti
m
e
based
sce
nar
i
o
(i.e.
,
He
NB)
f
or
m
acro
-
cel
l
an
d
f
e
m
to
-
cel
l,
al
so
the
s
ta
nd
a
rd
dy
nam
ic
represe
ntati
on
is
pe
rfo
rm
e
d
f
or
po
wer
c
on
t
ro
ll
er
[
11
]
,
wh
ic
h
is
e
ff
ect
ive
at
powe
r
switc
hi
ng
po
i
nts
in
HetNets.
In
res
ul
t
analy
sis
we
s
how
n
the
perform
ance
of
our
pro
po
se
d
m
od
el
with
resp
ect
t
o
e
xisti
ng
al
gorithm
u
nde
r diffe
ren
t
conside
red sce
nar
i
os
to
v
al
i
da
te
the m
od
el
perf
or
m
ance.
2.
LIT
ERATUR
E SU
RV
EY
The
m
ajo
r
c
halle
ng
e
is
to
e
nhance
the
in
door
area
c
ov
e
rage,
the
stu
dy
in
[12]
shows
m
or
e
than
70
%
of
th
e
traf
fic
da
ta
and
50%
of
al
l
vo
ic
e
cal
ls
are
m
ake
fr
om
ind
oors
e
nviro
nm
ent.
So
th
e
cov
e
ra
ge
of
in
do
or
env
i
ronm
ent
c
an
pro
vid
e
higher
data
-
rate
a
nd
Qo
S,
wh
ic
h
are
the
m
ajor
pro
blem
wh
ile
dev
el
op
i
ng
t
he
up
-
com
ing
gen
e
ra
ti
on
(
W
i
reless
Com
m
un
ic
at
io
n
Netw
orks
)
WSNs,
howe
ve
r
pro
vid
i
ng
m
acro
-
cel
l
BS
in
orde
r
to
handle
the
gro
wing
traf
fic
dem
and
is
m
uch
ex
pensi
ve.
T
her
e
fore,
a
no
ve
l
fu
tur
e
ge
nerat
ion
arc
hitec
t
ur
e
is
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.
9
, N
o.
5
,
Oct
ober
20
19
:
4226
-
4
2
3
7
4228
su
ggest
e
d
to
overc
om
e
the
on
goin
g
dem
and
of
data
tra
ff
i
c
and
it
co
ns
is
t
of
fem
to
-
cel
l
BSs
(F
BS)
[13
,
14
]
.
The
pr
ese
nce
of
pro
xim
it
y
at
receiver
a
nd
tra
ns
m
i
tt
er
causes
the
bet
te
r
us
e
rs
ex
pe
rience
a
nd
a
bl
e
to
com
m
un
ic
at
e
op
ti
m
iz
e
throu
ghput,
suppose
t
he
ind
oor
us
er
su
c
h
as
on
e
s
in
their
ho
m
es
or
apa
rtm
en
ts
are
li
nk
ed
t
o
a
n
F
BS,
an
d
s
om
e
of
i
ndoor
us
e
rs
com
m
un
ic
at
e
directl
y
thr
ough
the
m
acro
-
cel
l
that
te
nds
to
op
ti
m
iz
e
the
per
f
or
m
ance
of
netw
ork.
C
onsideri
ng
t
he
ope
r
at
or
s
si
de,
the
fem
to
-
cel
l
can
increa
se
t
he
re
us
e
of
sp
ect
r
um
in
or
der
to
pro
vid
e
the
high
sp
ect
ral
ef
fici
ency
and
the
net
work
capaci
ty
,
m
or
e
over,
the
F
BS
are
com
pen
sat
ed
a
nd
upheld
via
the
owners.
I
nspire
of
these
adv
a
ntage
s
it
com
es
with
se
ver
al
ne
w
chal
le
ng
es
su
c
h
as i
n
sy
nc
hro
nizat
ion
,
in
t
erf
e
ren
ce
m
anag
em
ent an
d net
work arc
hitec
ture
[14
]
.
In
gen
e
ral,
th
e
diff
ic
ulty
due
interfe
re
nc
es
hav
e
becom
e
m
ajo
r
pro
blem
wh
ic
h
r
equ
i
res
novel
so
luti
on
be
cau
se
of
m
or
e
co
m
plexit
y
in
com
par
ison
with
the
existi
ng
s
ta
nd
a
rd
cel
lula
r
netw
ork
.
Ba
s
ic
al
ly
there
are
se
veral
ty
pe
of
dep
l
oym
ent
arch
it
e
ct
ur
e
is
pr
ese
nt
fo
r
fem
to
-
cel
l
network
s
.
In
on
e
of
the
ty
pe
of
or
t
hogonal
de
plo
ym
ent,
the
sp
ect
r
um
is
s
liced
into
du
al
i
nd
e
pe
nd
e
nt
f
ra
gm
ents,
wh
e
re
on
e
is
use
d
vi
a
the
fem
tocel
ls
and
oth
e
r
is
us
ed
by
the
m
acro
-
c
el
ls
[15],
m
or
eov
e
r,
t
hey
rem
ov
e
inter
-
ti
er
le
vel
inter
fer
e
nc
es
due
to
that
re
sour
ces
of
f
re
qu
e
nc
y
are
not
fu
l
ly
util
iz
ed.
I
n
oth
e
r
ty
pe
of
c
ha
nn
el
de
plo
ym
ent
arch
it
ect
ur
e,
the
bo
t
h
fem
to
and
m
ic
ro
cel
ls
hav
in
g
di
rec
tl
y
acce
ss
to
a
vaila
ble
cha
nnel
s
[1
6],
this
t
ype
of
m
et
ho
dolo
gy
can
ge
ner
at
e
la
rg
e
inter
fer
e
nc
e
le
vels.
Howe
ver,
there
is
al
so
par
ti
al
‘co
-
c
hannel’
de
pl
oym
ent
app
r
oac
h,
su
ch
that
the
al
l
pr
esented
s
pectr
um
s
are
separ
at
ed
into
tw
o
pa
rts;
on
e
spe
ci
al
ly
ded
ic
at
ed
to
m
acro
-
cel
ls
and
the
seco
nd
on
e
s
ha
red
to
fem
to
and
m
acro
cel
ls
[17
,
18]
.
Th
ese
dep
l
oym
en
t
m
od
el
of
FBS
acce
ss
appr
oa
ch
is
us
e
d
to
restrict
‘o
r
’
al
low
the
us
age
of
s
pec
trum
by
us
ers.
Howev
e
r,
the
op
e
n
acce
ss
pro
vid
e
us
er
to
have
acce
ss,
w
hile
the
cl
os
e
d
acc
ess
only
al
lo
w
s
subsc
ribe
d
use
rs
for
sp
ect
r
um
acce
ss
(i.e
.,
cl
os
ed
s
ub
sc
ribe
r
gro
up)
,
m
or
eo
ver
t
he
us
e
of
hybri
d
acce
ss
al
lows
al
s
o
t
o
non
-
subsc
rib
er
to
us
e
pa
rtic
ular
am
ount
of
t
he
‘f
em
to
-
cel
l’
re
so
urces
.
I
n
pa
per
[19],
they
m
ulti
cast
ing
ty
pe
of
sc
he
du
li
ng
pro
blem
in
HetNets
via
a
“half
-
du
ple
x
re
la
y”
sta
ti
on
is discuss
ed
an
d
aim
is
to
m
ini
m
al
iz
e
the
tran
sm
issi
on
d
el
ay
of b
loc
k
(i.e., pack
et
s
to
use
rs
)
ov
e
r
t
he
ti
m
e
per
iod via
rate
-
le
ss
c
odes.
The
refo
re,
the operati
on o
f
half
-
duplex
is
perf
or
m
ed
at
each
tim
e
interval,
relay
sta
ti
on
is
sel
ect
a
m
ulti
cast
pack
et
to
us
e
rs
‘
or’
reali
ze
a
packet
fr
om
m
acro
BS
,
afterwa
r
ds
th
e
y
form
ulate
d
the
fluid
rela
xa
ti
on
to
op
ti
m
ize
the
decisi
on
diff
ic
ulty
an
d
pro
vid
e
t
he
optim
a
l
appr
oach f
or th
reshold
b
a
se
de
plo
ym
ent in ord
e
r
to
exploit
the
m
ulti
cast
s
pecu
la
ti
on c
ha
nn
el
.
Ther
e
f
or
e,
t
he
relay
sta
ti
on
only
m
ulti
-
cast
wh
e
ne
ver
t
he
c
hannel
cap
aci
ty
has
reac
hed
consi
der
a
bly
high,
t
hen
a
n
on
li
ne
poli
cy
-
base
d
rela
xation
t
hat
does
not
nee
de
d
the
inf
or
m
at
ion
of
cha
nn
el
distribu
ti
on
.
The
cha
nnel
di
stribu
ti
on
is
consi
de
re
d
to
be
sym
m
et
ric
to
al
l
us
ers
a
nd
t
he
cl
os
e
d
fo
rm
of
asy
m
pto
tic
expressi
on
is
perform
ed
by
their
poli
cy
.
Wh
il
e,
c
onside
rin
g
dual
use
r
m
od
el
,
they
use
d
a
n
asy
m
pt
otica
ll
y
op
ti
m
al
fu
ncti
on
that
base
d
on
the
“wate
r
fill
ing
”
rate
al
l
ocati
on
f
or
ap
pro
xim
a
ti
ng
the
bette
r
m
od
el
po
li
cy
.
The
use
r
sc
he
du
li
ng
(
‘US’
)
and
us
er
as
so
c
ia
ti
on
(‘U
A’)
has
co
ns
id
er
f
or
the
lo
ad
ba
la
ncing
at
do
wn
li
nk
op
e
rati
on
of
w
irel
ess
Het
Net
via
f
or
m
ulati
ng
the
m
axi
m
um
utilizat
ion
of
wide
netw
or
k
pro
blem
to
reso
lve
the
eff
ic
ie
ncy
pro
blem
[2
0].
First
of
al
l,
the
y
approxim
at
e
the
thr
ough
pu
t
of
non
-
c
onve
x
pro
blem
with
US
in
order
to
c
on
ca
ve
the
f
un
ct
io
nalit
y
and
this
can
dem
on
str
at
e
the
diff
ere
nce
of
ap
prox
i
m
at
ion
al
go
rit
hm
zero
wh
e
ne
ver
the
pr
ese
nt
us
e
rs
are
m
uch
la
rger.
A
fter
wards
,
distrib
u
te
d
c
onve
x
ap
proac
h
that
is
known
for
change
the
m
ulti
pliers
directi
on
m
et
ho
dolo
gy
is
us
ed
as
the
integrated
U
S
-
U
A
al
gorith
m
that
can
be
app
li
ed
on
BS
an
d
use
r
sides
i
nd
i
vid
ually
an
d,
propose
d
to
be
resou
rce
al
loc
at
ion
an
d
BS
associat
ion
so
l
utions.
The
pr
op
os
e
d
m
od
el
featur
e
enh
a
ncem
ent
can
be
us
ed
in
fu
tu
re
to
im
p
rove
pe
rfor
m
ance
that
apa
rt
from
m
ul
ti
-
us
er
div
e
rsity
an
d l
oa
d balanci
ng.
The
tra
de
-
off
in
bet
ween
th
e
energy
c
ons
um
ption
an
d
data
-
rate
pe
rfo
rm
ance
in
non
-
or
t
hogonal
m
ul
ti
ple
acce
ss
(‘
’
)
has
stu
di
ed
in
[
21
]
,
w
he
re
the
pro
ble
m
of
sche
du
li
ng
a
nd
t
he
pow
er
en
ha
ncem
ent
in
N
OMA
HetNets
is
perfor
m
ed;
thou
gh
the
powe
r
al
loc
at
ion
a
nd
e
nergy
ef
fici
ent
sc
hedulin
g
a
ppr
oa
ch
i
n
NO
M
A
HetN
e
ts
are
introdu
c
ed
f
or
bot
h
im
per
fect
and
perfect
CSI.
T
he
FQL
(Fuzz
y
-
Q
-
Lear
ning)
based
energy
eff
ic
ie
nt
app
r
oac
h
(i.e.
,
inclu
de
ei
ther
sle
ep
‘or’
wa
ke
-
up
m
od
el
)
f
or
the
BSs
in
a
HetNet
an
d
the
ai
m
is
to
m
ini
m
ali
ze
the
ene
rg
y
without
ef
fecti
ng
t
he
offe
re
d
QoS
thr
ough
perform
ing
sw
it
ch
off
ope
rati
on
at
redu
nd
a
n
t
BSs
in
accor
da
nce
to
native
tra
ff
i
c,
w
hich
de
pe
nd
up
on
the
c
ov
erag
e
a
rea
of
c
el
l.
The
sle
ep
-
m
od
e
act
ivit
y
fo
r
BSs
m
ay
te
nd
s
to
a
la
rg
e
cov
e
ra
ge
loss,
un
le
ss
so
m
e
par
ti
cular
optim
iz
e
so
l
ution
is
occ
urr
ed
at
sam
e
interval
[
22
]
.
T
he
HetN
et
s
con
sist
of
s
econ
da
ry
an
d
pr
im
ary
us
er,
wh
e
reas
the
ordina
ry
cel
lular
us
ers
are
co
ns
i
der
i
ng
to
be
pri
m
ar
y
us
ers
(PUs
),
wh
il
e
un
li
cen
s
ed,
IoT
a
nd
som
e
sensor
e
qu
i
pm
ent
are
co
nsi
der
in
g
to
be
seco
ndar
y
us
ers
(SUs
)
in
[23].
The
c
hannels
Het
Ne
ts
are
occupie
d
PU
s
wh
il
e
the
SU
s
c
a
n
on
ly
reu
se
the
P
Us
c
hann
el
s,
he
re
they
consi
der
e
d
dua
l
m
od
e
tra
ns
m
issi
on
for
th
e
SU
s
s
uc
h
as
S
U
on
ly
com
e
up
with
BS
directl
y
‘or’
via
the
help
of
ass
ociat
ive
relay
no
de
and
t
he
e
nergy
eff
ic
ie
ncy
enh
a
ncem
ent
of
SUs
prototype
is co
ns
ide
red.
In
ge
ne
ral,
the
y
m
ajo
rly
aim
on
as
so
ci
at
io
n
of
use
rs
s
uc
h
as
sel
ect
ion
of
BS,
m
od
e
s
el
ect
ion
a
nd
channel
al
locat
ion
in
orde
r
to
op
ti
m
iz
e
the
powe
r
co
ntr
ol
at
up
li
nk
com
m
un
ic
at
ion
in
betwee
n
the
B
Ss
an
d
SU
;
m
or
eov
e
r,
these
ty
pe
of
d
iffic
ulty
can
be
fo
rm
ulate
d
as
m
ixed
intege
r
optim
iz
e
and
non
-
co
nvex
prob
le
m
.
An
opportu
nist
ic
ty
pe
of
pow
er
con
t
ro
ll
er
a
ppr
oach
is
pro
po
s
ed
in
[
24]
t
o
m
it
igate
the
interfe
ren
ce
f
r
om
the
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
Hetero
ge
neous
n
et
work
opti
m
izati
on u
si
ng r
obus
t
powe
r …
(
Sa
vi
ta
Pa
ti
l
)
4229
act
ive
nu
m
ber
of
fem
tocel
ls
a
t
up
li
nk
tra
nsm
issi
on
,
the
m
acro
-
cel
l
BS
chooses
the
al
lo
wan
ce
inte
rf
e
r
ence
a
s
per
nu
m
ber
of
fem
to
-
cel
l
use
rs
an
d
fem
to
-
cel
ls,
after
wa
rd
s
al
locat
e
th
e
transm
it
po
wer
unde
r
cr
oss
-
ti
er
interfe
ren
ce
in
con
si
der
e
d
H
et
Nets.
The
a
ppr
oach
to
wards
the
po
wer
con
t
ro
ll
er
s
hould
ve
rify
the
act
ive
nu
m
ber
of
fem
tocel
ls
as
pe
r
t
he
e
ff
ect
ive
co
ntr
oller,
al
s
o
t
hey
pr
opos
e
d
t
wo
ki
nd
of
se
ns
in
g
a
ppr
oac
hes
t
o
pr
e
dict
the
act
ive
fem
tocel
ls
su
ch
as;
cent
rali
zed
an
d
th
e
distribu
te
d,
resp
ect
ively
.
The
dif
ficult
ou
ta
ge
pro
bab
il
it
y fu
nc
ti
on
und
e
r
the
Ray
le
igh
f
ading
e
ff
ect
h
as st
ud
ie
d
in [25
]
an
d
s
olv
e a con
verge
nce p
r
oblem
f
or
init
ia
ll
y
discov
ere
d
a
ppr
oac
h
es
i
n
t
he
cas
e
of
i
nterf
e
re
nce.
A
fter
wards,
they
ad
dr
e
ss
the
power
relat
ed
diff
ic
ulty
with
ho
l
ding
the
ou
ta
ge
co
ns
tr
ai
nts
al
so
co
nsi
der
in
g
thei
r
feasible
e
nv
i
ronm
ent;
m
or
eov
e
r,
a d
ynam
ic
ap
pr
oac
h has pr
opos
e
d
to pr
ov
i
de
o
uta
ge
pr
ob
a
bili
ty
co
nd
it
io
n i
n
a H
et
Net in
o
r
der
t
o
dec
re
ase the
ov
e
rall
ene
r
gy
consum
ption
,
a
lso
this
guara
nt
ee
to
al
l
fem
to
-
cel
l
us
e
rs
a
bo
und
of
m
in
an
d
m
ax
f
or
the
worst
le
vel of
ou
ta
ge
proba
bili
ty
.
3.
PROP
OSE
D MET
HO
DOL
OGY
In
ge
ne
ral,
th
e
interfer
e
nces
hav
e
beco
m
e
m
ajo
r
prob
le
m
wh
ic
h
require
s
novel
so
luti
on
beca
us
e
of
m
or
e
com
plexity
in
com
par
ison
with
the
exis
ti
ng
sta
nd
ard
cel
l
ular
ne
twork
.
A
ne
w
ty
pe
of
de
pl
oym
en
t
arch
it
ect
ure
is
present
su
c
h
as
f
em
to
-
cel
l
netw
orks.
I
n
on
e
of
t
he
ty
pe
o
f
ort
hogo
nal
de
pl
oym
ent,
the
sp
ect
r
um
is
sli
ced
into d
ua
l
ind
epe
ndent fragm
ents,
w
he
re
one
is
use
d
via
the
fem
tocel
ls
and
o
ther
i
s
us
e
d
by
the
m
acro
-
cel
ls
.
In
this
s
tud
y,
we
co
nsi
der
on
e
m
acro
-
cel
l
an
d
se
ve
ral
num
ber
of
fem
to
-
cel
ls,
wh
e
re
fem
to
-
cel
ls
has
able
to
dec
rea
se
the
interfe
re
nce
le
vel
with
var
i
ou
s
c
ha
nn
e
l
sign
al
s
that
will
help
to
op
tim
iz
e
the
s
pectral
e
f
fici
ency
as
per
the
c
overe
d
c
el
l
area.
T
he
s
ever
al
num
ber
of
fem
to
-
cel
ls
can
be
locat
e
d
in
a
m
acro
-
cel
l,
t
he
refor
e
the
B
Ss
of
fem
to
-
cel
ls
are
loc
at
e
d
at
ver
y
s
horte
d
distance
to
eac
h
oth
e
r,
w
hich
he
lp
t
o
cov
e
r
the
po
or
transm
issi
on
area
present
in
a
m
acro
-
cel
l;
m
or
eo
ver
,
t
he
equ
ipm
ent
be
com
es
m
or
e
e
nergy
eff
ic
ie
nt
due
to
the
shorte
d
distance
via
de
crem
ent
in
tr
ansm
issi
on
power
that
hel
p
to
increase
the
li
fe
of
batte
ry.
In
ge
ne
ral,
fem
to
-
cel
ls
are
us
e
d
i
n
indo
or
locat
i
on
s
o
that
the
m
acro
-
cel
l
can
pro
vid
e
t
he
rob
us
t
serv
ic
es
to
us
e
rs
du
e
to
it
s
re
du
ct
io
n
of
ov
e
rh
ea
d.
Bl
oc
k
di
agr
am
of
pro
pose
d
RPR
A
is
sh
ow
n
i
n
Fi
gure
1,
wh
e
re
the
pro
pose
d
ap
proac
h
sh
o
ws
c
onside
r
r
obus
t
ap
proa
ch
that
can
ha
ndle
va
riat
ion
in
BSs
(i.e.,
dif
fe
ren
t
scenari
os
)
, whi
ch wil
l be s
ho
wn in res
ult an
al
ysi
s secti
on
.
Figure
1
.
Bl
oc
k
diag
ram
o
f
pro
posed
RPR
A
In
t
his
sect
ion,
first
we
will
pro
vid
e
the
de
ta
il
ed
stud
y
an
d
num
erical
form
ulati
on
of
Robusti
fied
Power
Co
ntr
oller
(RPC
)
a
nd
afterwa
r
ds
we
will
go
f
or
the
Ro
bu
st
Cha
nnel
All
ocati
on
A
ppr
oach
(
R
CAA
).
In
co
ns
id
er
He
tNet
,
the
us
e
rs
of
fem
to
-
cel
ls
and
m
acro
-
cel
ls
m
akes
the
re
qu
e
st,
base
d
upon
t
heir
re
qu
i
rem
ent
the
c
hannel
a
nd
tra
ff
ic
detai
ls
are
acq
uire
d,
the
n
al
l
thes
e
thin
gs
are
f
orwarde
d
to
re
s
ource
al
lo
cat
io
n
a
nd
powe
r
co
ntr
ol
process
f
or
f
ur
t
her
optim
izati
on
process
,
so
we
ca
n
ge
t
m
axi
m
u
m
us
er
thr
ou
ghput
an
d
op
ti
m
iz
e p
ow
e
r
tra
ns
m
issi
on
.
3.1.
R
ob
ust
ifi
ed p
ower co
ntr
oller
Con
si
der
i
ng
t
he
cha
nn
el
unce
rtai
nties,
the
fe
m
to
-
cel
l
us
er
(
FCU)
re
qu
i
re
m
ents
an
d
m
a
cro
-
cel
l
us
e
r
(MCU)
perf
orm
ance
protect
ion
is
ver
y
nec
essary
to
co
nsi
der
e
d
sim
ultan
eo
us
ly
,
the
refor
e
we
co
ns
i
de
r
the
rob
us
t
power
c
on
t
ro
ll
er
an
d
r
eso
ur
ce
al
locat
ion
ap
proac
h
in
a
two
-
ti
er
H
et
Net
under
th
e
‘stoch
a
sti
c
m
od
el
’.
A
m
ulti
-
us
er
OFDM
ba
sed
HetNets
is
co
ns
ide
red
that
consi
st
num
ber
of
FCUs
,
c
om
m
un
ic
at
ing
with
associat
ed
fem
to
-
cel
l
BSs
ove
r
the
num
ber
o
f
subca
rr
ie
r
,
m
or
eo
ver
FC
Us
a
re u
se
d
t
o
op
e
r
at
e
the
MC
Us
via
FC
-
BSs,
w
here
and
are
c
ha
ng
e
s
acco
r
ding
to
t
he
num
ber
of
us
e
r’s
act
ive
an
d
av
ai
la
ble
su
bca
rri
er
channel,
t
hat
can
be
giv
e
n
as
∈
≜
{
1
,
2
,
3
.
.
.
.
.
,
}
an
d
∈
ℰ
≜
{
1
,
2
,
3
.
.
.
.
.
,
}
,
w
her
e
,
we
ass
um
ed
that
ℰ
≥
a
nd
the
s
ub
ca
rr
ie
r
ba
ndwidth
is
ex
pec
te
d
t
o
be
Hz
w
hich
is
m
uch
l
ess
in
c
om
parat
ive
to
th
e
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.
9
, N
o.
5
,
Oct
ober
20
19
:
4226
-
4
2
3
7
4230
wireless
cha
nnel
ban
dwidt
h.
Ther
e
f
or
e,
a
pply
ing
Sh
a
nnon
Har
tl
ey
Theo
r
e
m
(S
HT)
[
26
]
cor
res
pondin
g
FCU
data rate
at
sub
carrier
is w
ritt
en
as
.
,
=
ℎ
,
log
2
(
1
+
,
,
,
⁄
)
(
1
)
Wh
e
re,
ℎ
,
show
s
t
he
FCU
s
ub
ca
r
rier
ta
s
k
at
the
subca
rr
ie
r
,
,
sh
ows
t
he
FCU
ba
ckgr
ound
noise
a
t
su
bc
ar
rier
,
,
sh
ows
the
FC
U
s
trai
gh
t
cha
nne
l
gai
n
at
s
ub
c
arr
ie
r
a
nd,
,
sh
ow
s
FC
U
tra
ns
m
it
powe
r
at
s
ub
ca
rr
ie
r
.
The
re
is
0
‘
or’
1
s
ub
ca
r
rier
assig
nm
ent
w
hich
sig
nify
the
s
ub
ca
rr
ie
r
us
e
d
via
FC
U
or
no
t.
T
he
m
ai
n
pro
blem
with
this
is
batte
ry
ca
pacit
y
at
the
th
FC
U
tra
nsm
it
te
r,
there
f
or
e
t
he
c
on
st
ra
int
can
be writt
en a
s:
∑
ℎ
,
=
1
,
≤
,
∀
∈
(2)
In
ab
ove
e
quat
ion
,
shows
th
e
m
axi
m
a
l
power
tran
sm
i
t
of
the
FCU
,
wh
e
re
t
he
dat
a
-
rate
s
houl
d
accom
plish th
e
nom
inal requ
i
rem
ent o
f
FCU
Q
oS
s
o
t
hat it
can
be writt
en a
s:
∑
,
=
1
≥
,
∀
∈
(3)
Wh
e
re,
show
s
the
m
ini
m
u
m
requirem
ent
-
r
at
e
of
FCU,
al
so
inter
fer
e
nc
e
co
ns
trai
nt
of
t
he
total
cro
ss
-
ti
er
unde
r
fem
tocel
l HetNet t
o
t
he
MC
U
r
ecei
ver sect
ion
ca
n be
gi
ve
n
as:
∑
∑
ℎ
,
=
1
=
1
,
,
≤
(4)
The
i
nterf
e
re
nc
e
le
vel
at
rec
ei
ver
side
of
MC
U
is
denot
e
by
,
the
s
um
rate
m
axi
m
iz
at
ion
by
po
wer
con
t
ro
ll
er
at
th
e H
et
Nets
scen
ario
ca
n be
wr
i
tt
en
as:
ℎ
,
,
∑
∑
,
=
1
=
1
∑
ℎ
,
=
1
=
1
,
∀
∈
,
1
∑
ℎ
,
,
=
1
≤
,
∀
∈
,
2
(5)
Wh
e
re,
1
d
e
no
te
s
the
in
div
id
ua
l
su
bca
rr
ie
r
that
is
al
locat
ed
to
eac
h
FCU,
,
=
1
shows
t
he
th
-
su
bc
ar
rier
w
hich
is
use
d
via
FCU,
an
d
2
is
use
d
t
o
s
how
the
co
ns
trai
nt
of
powe
r
tra
ns
m
i
ssion
at
F
C
U
ov
e
r
t
he
s
ub
-
ca
rr
ie
r
.
∑
,
=
1
≥
,
∀
∈
,
3
∑
∑
h
d
,
e
I
d
,
e
N
d
,
e
E
e
=
1
D
d
=
1
≤
M
il
,
Z
4
(6)
ℎ
,
∈
{
0
,
1
}
,
∀
∈
,
∈
ℰ
,
5
The
(
6)
co
nfi
r
m
the
Qo
S
f
or
each
FC
Us,
w
her
e
,
4
denotes
the
interfe
ren
ce
of
t
otal
powe
r
at
MC
U
receiv
er
,
and
the
m
ai
n
pr
ob
le
m
is
ℎ
,
=
1
wh
ic
h
is
m
ixed
inte
ger
an
d
“
non
-
c
onve
x
program
m
ing
”
diff
ic
ulty
,
,
s
ign
i
fies
the
feedb
ack
of
c
ha
nn
e
l
gains
t
hat
pro
vid
e
d
th
r
ough
MC
U
to
FCU.
In
rece
nt
dev
el
op
m
ent,
m
ajo
rity
of
t
he
resea
rch
e
rs
has
ai
m
ed
towa
rd
s
po
wer
al
locat
ion
stra
te
gy
in
the
Het
N
et
s
[27]
wh
ic
h
f
oc
us
on
im
pr
ov
em
ent
power
with
pe
rf
ect
CSI
[
28]
.
Howe
ver
,
i
n
par
ti
cl
e
scena
rio,
t
he
qua
ntiza
ti
on
error
s
pr
ese
nc
e
reasons
the
channel
uncert
ai
nty
wh
ic
h
is
har
m
fu
l
fo
r
MC
Us,
in
ord
er
to
overc
ome
this
pro
blem
,
we
s
hould
c
on
si
de
r
so
m
e
i
m
pr
ov
e
te
ch
nique
s
o
that
ca
n
deal
with
these
ty
pe
of
unce
rtai
nties.
Ther
e
f
or
e,
w
e
us
e
R
obus
t
ifie
d
powe
r
c
on
tr
ol
le
r
an
d,
t
he
(6)
ca
n be r
e
w
ritt
en
in
a
pro
bab
i
li
ty
f
or
m
:
ℎ
,
,
∑
∑
,
=
1
=
1
.
.
1
,
2
,
5
P
{
∑
,
≤
=
1
}
≤
,
∀
∈
,
6
P
{
∑
∑
ℎ
,
,
,
=
1
=
1
>
}
≤
(7)
Wh
e
re,
both
(
7
)
guara
ntee
the
Qo
S
of
MC
U
and
FCU
th
rou
gh
us
i
ng
pro
bab
il
it
y
functi
on
,
and
shows
f
or
t
he
thres
hold
value
us
e
d
in
ou
ta
ge
pr
ob
a
bili
ty
.
OFDM
featu
re
m
et
ho
d
ha
s
consi
der,
the
re
fore
the
subcar
riers
a
re
in
depend
e
nt
f
r
om
e
ach
s
ub
-
car
rier
,
th
ough
each
of
FCU
i
nform
at
ion
are m
utu
al
ly
ind
epe
ndent
fro
m
su
bcar
rier
s a
nd
the
data
-
rat
e set i
s d
e
fine
d as:
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
Hetero
ge
neous
n
et
work
opti
m
izati
on u
si
ng r
obus
t
powe
r …
(
Sa
vi
ta
Pa
ti
l
)
4231
=
{
,
≤
}
=
{
∑
,
=
1
≤
}
(8)
Wh
e
re,
set
is a
n
inte
rsecti
on s
ub
s
et
of
su
ch
as
:
S
̅
⊂
S
=
S
1
⋂
S
2
…
S
e
(9)
A
pply
ing t
he p
roba
bili
ty
an
al
ysi
s
in
(
9)
, we
go
t
t
he
fo
ll
owi
ng r
el
at
io
ns
hi
p:
{
̅
}
≤
P
{
}
=
∏
P
=
1
{
}
(
10
)
Fu
rt
her
m
or
e
,
it
can be
giv
e
n
a
s:
P
{
∑
,
=
1
≤
}
≤
∏
P
=
1
{
,
≤
}
(1
1
)
Pr
oba
bili
s
ti
c
c
on
st
raint
rate
f
or
the
upper
boun
d
shou
l
d
f
ul
fils
the
essenti
al
ou
ta
ge
pr
obabili
ty
thro
ughout
the
worst sce
nar
i
o, there
fore t
he
c
onditi
on sce
nari
o
by
(
7)
ca
n b
e wri
tt
en
as:
Max
P
{
∑
,
=
1
≤
}
≤
∏
P
=
1
{
,
≤
}
≤
(12)
The
determ
inist
ic
o
utag
e
probabil
it
y
of
(1
2
)
can
be
giv
e
n
as
:
≤
ℎ
,
log
2
(
1
+
,
,
J
,
−
1
(
/
)
)
,
∀
∈
.
(13)
The
(
13)
c
onf
irm
the
‘po
we
r
tran
sm
issi
on
’
with
t
he
out
age
p
r
ob
a
bili
ty
,
so
the
pro
ba
bili
sti
c
interfer
ence
from
(7
)
ca
n b
e upd
at
e
d
as:
h
d
,
e
I
d
,
e
≤
M
il
EN
N
d
,
e
−
1
(
√
1
−
DE
)
,
,
∀
d
∈
,
∀
e
∈
ℰ
.
(14)
Ther
e
f
or
e,
t
he
(14)
is
sai
d
to
be
determ
inist
i
c
in
natur
e
a
nd
it
is
necessit
at
ing
to
keep
it
as
pr
ese
ntable
form
,
m
or
eov
er
, th
e
diff
ic
ulty
o
f
po
wer co
ntr
oller
without a
ny pr
i
or info
rm
ation
can
be give
n
as
:
ma
x
ℎ
,
,
∑
∑
,
=
1
=
1
.
.
1
,
2
,
5
(15)
ℎ
,
log
2
(
1
+
ℎ
,
,
J
,
−
1
(
/
)
)
≥
,
∈
.
(16)
Her
e
, th
e
in
verse f
un
ct
io
n
of
colle
ct
ive d
ist
r
ibu
ti
on
at
var
ia
ble
,
an
d
,
is app
li
e
d
a
nd can
b
e
writ
te
n
as
J
J
d
,
e
−
1
an
d
N
N
d
,
e
−
1
.
E
h
d
,
e
I
d
,
e
N
N
d
,
e
−
1
(
√
1
−
DE
)
≤
M
il
(17)
In
ge
ne
ral,
the
FCUs
can
obta
in
the
CSI
by
the
cha
nn
el
est
im
at
ion
procedure
bet
wee
n
the
FCUs
an
d
the
MC
Us
,
t
her
e
fore
t
hese
can
arise
som
e
trouble
at
CSI
acq
uisit
ion
.
T
her
e
fore
,
we
c
on
si
der
the
sel
f
-
determ
ining
prototype
of
Ga
us
sia
n
dist
rib
ution
in
ord
er
to
ha
nd
le
t
he
am
big
uity
par
am
et
ers.
Howev
e
r,
the
FCU
s
tra
nsm
it
te
r
chan
ne
l
gain
t
o
BS
i
s
obta
in
by
a
rob
us
t
us
e
r
-
aft
erw
a
rds
the
fe
edb
ac
k
is
sen
d back
to
t
he
corres
pondin
g
FC
Us
tr
ansm
it
te
r.
3.2.
R
obust
r
esou
rce
a
ll
oc
ati
on appr
oa
c
h
The
m
ajor
diff
ic
ulty
is
sti
ll
rob
us
ti
ficat
ion
at
ch
an
nel
al
locat
ion
pr
ocess
due
the
ℎ
,
integer
var
ia
ble
,
the
bo
t
h
inte
ger
a
nd
real
ti
m
e
var
ia
ble
su
c
h
as
,
a
nd
ℎ
,
are
unde
r
optim
iz
at
ion
prob
le
m
,
therefo
re
it
con
duct
s
di
ver
se
integer
progr
a
m
m
ing
dif
fic
ulty
,
so
the
re
la
xation
of
s
ubcar
rier
assig
nm
ent
featur
e
into
t
he
con
ti
nuous
on
e
t
hat
can
be
al
so
s
ho
w
n
as
var
ia
ble
,
=
,
ℎ
,
fo
r
a
fem
to
us
er
and
su
bc
ar
rier.
Wh
ere,
ℎ
,
∈
[
0
,
1
]
sign
ifie
s
a
tim
e
div
isi
on
strat
egy
f
or
t
he
var
i
ou
s
fem
t
o
use
rs
.
The
re
fore,
in
order
to
obta
in
the
,
a
nd
ℎ
,
co
ndit
ion
va
riables
we
app
ly
Karus
h
-
Kuh
n
-
T
uc
ker
(
KK
T
)
[
29
]
,
w
hich
i
s
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.
9
, N
o.
5
,
Oct
ober
20
19
:
4226
-
4
2
3
7
4232
nonlinea
r
pro
gram
m
ing
appr
oach
t
hat
al
lows
eq
ualit
y
con
st
raints
va
ri
ables
an
d
bas
ed
on
the
La
gr
a
nge
m
ul
ti
pliers
m
e
t
hod, th
us
t
he d
erivati
ve f
or
m
o
f
opti
m
a
l var
i
ables ca
n be
w
ritt
en
as:
0
≤
,
⊥
,
(
.
.
.
)
,
(18)
0
≤
ℎ
,
⊥
,
(
.
.
.
)
ℎ
,
(19)
Wh
e
re,
⊥
show
s t
he ort
hogonal
r
el
at
ion
to
corre
sp
on
ding
var
ia
bles a
nd
,
show
s
the La
gr
a
nge f
un
ct
io
n, in
add
it
io
n,
t
he
,
(
.
.
.
)
,
a
nd
,
(
.
.
.
)
ℎ
,
de
rivati
ves
c
an
be w
ritt
en
as:
,
(
.
.
.
)
ℎ
,
=
(
1
+
)
̃
,
(
,
,
ℎ
,
)
,
−
−
(
̂
,
+
−
1
(
1
−
)
,
)
(20)
,
(
.
.
.
)
ℎ
,
=
(
1
+
)
̃
,
(
∗
,
,
ℎ
,
)
ℎ
,
−
(21)
Wh
e
re
,
,
,
a
nd
are
nonne
gative
La
gr
a
ngia
n
m
ult
ipli
ers
f
or
c
orrespo
nd
i
ng
c
onstrai
nts
,
in
ve
rse
Gau
s
sia
n
Q
-
functi
on
is
giv
e
n
by
−
1
(
.
)
.
T
her
e
fore,
the
opti
m
a
l
tr
ansm
it
po
wer
with
the
c
ha
nn
el
al
locat
io
n
can
be writt
en a
s:
,
∗
=
,
ℎ
,
=
[
1
ln
2
(
1
+
)
+
(
̂
,
+
−
1
(
1
−
)
,
)
−
,
̂
,
+
2
,
]
+
(22)
Wh
e
re,
outa
ge
pro
bab
il
it
y
thres
hold
is
and
the
up
per
bo
und
reg
i
on
is
,
,
[
]
+
=
ma
x
(
0
,
)
and
t
he
su
bc
ar
rier
is al
locat
ed
t
o
the
∗
optim
al
u
ser
s
uc
h
as:
ℎ
,
∗
=
1
|
∗
=
ma
x
̂
,
(23
)
Wh
e
re
,
̂
,
is give
n by:
̂
,
=
(
1
+
)
̃
,
(
∗
,
,
ℎ
,
)
ℎ
,
=
(
1
+
)
(
log
2
(
1
+
̂
,
,
∗
,
)
−
̂
,
,
∗
ln
2
(
,
+
̂
,
,
∗
)
)
(24
)
Ther
e
f
or
e,
the
Lagr
a
ngia
n
m
ul
ti
pliers is upgr
ad
ed
as:
λ
d
t
+
1
=
[
λ
d
t
−
x
1
t
(
I
d
max
−
∑
I
d
,
e
t
e
∈
Z
d
)
]
+
(25)
μ
d
t
+
1
=
[
μ
d
t
−
x
2
t
(
∑
g
̃
d
,
e
t
e
∈
Z
d
−
G
d
min
)
]
+
(26)
u
t
+
1
=
[
u
t
−
x
3
t
(
M
th
−
∑
∑
I
d
,
e
t
e
∈
Z
d
(
N
̂
d
,
e
+
Q
−
1
(
1
−
)
u
d
,
e
)
D
d
=
1
)
]
+
(27)
W
he
re,
the
it
e
rati
on
nu
m
ber
is
gi
ven
by
,
t
he
1
,
2
and
3
sho
ws
t
he
sm
all
ste
p
siz
es,
wh
e
ne
ve
r
these
ste
p
siz
e
are
ap
pro
pr
ia
te
ly
s
m
all
the
“Lagr
a
nge
m
ulti
p
li
ers”
can
co
ng
reg
at
e
to
e
quil
ibri
um
po
ints
[
30
]
.
In
t
he
im
ple
m
entat
ion
pro
cess,
init
i
al
ly
m
axi
m
u
m
iterati
on
c
ount
is
giv
e
n
by
,
and
i
niti
al
iz
at
i
on
of
=
0
,
>
0
>
0
;
the
La
gra
ng
e
m
ulti
pliers
are
init
ia
li
z
es
as
0
<
(
0
)
,
0
<
(
0
)
and
0
<
(
0
)
,
though
t
he
th
re
sh
ol
d
value
of
ou
ta
ge
pro
bab
i
li
ty
is
∈
[
0
,
1
]
,
∈
[
0
,
1
]
.
The
up
per
cha
nnel
est
i
m
ation
boun
d
error
in
f
em
to
use
r
li
nk
is
(
,
∈
[
0
,
1
]
)
a
nd
t
he
e
sti
m
at
ed
er
ror
va
riance
in
fem
to
us
er
to
m
acro
us
er
is(
,
∈
[
0
,
1
]
).
T
he
m
axim
al
transm
i
t
powe
r
(
)
a
nd
in
it
ia
li
ze
po
we
r
(
,
)
is
set
t
o
gre
at
e
r
tha
n
zer
o
with
sam
e
values
at
al
l
s
ubc
arr
ie
rs
,
al
so
wi
th
t
he
ℎ
interfer
e
nce.
A
fter
ward
s
the
it
erati
on
is
sta
rt
f
ro
m
on
e,
al
so
f
or
eac
h
and
valu
e
it
com
pu
te
s
the
transm
it
po
we
r
(
,
∗
),
ℎ
,
∗
,
̂
,
and
t
he
updated
Lag
ra
ng
e
m
ul
ti
pliers, til
l
that the it
erati
on h
a
s
no
t
finis
hed.
4.
RESU
LT
A
N
ALYSIS
T
he
fem
to
-
cel
l
can
i
ncr
ease
the
s
pectr
um
reu
se
in
or
der
t
o
pro
vid
e
t
he
hig
h
sp
ect
ral
ef
fici
ency
a
nd
the
netw
ork
ca
pacit
y
.
Her
e,
our
ai
m
is
on
associat
ion
of
use
rs
su
c
h
as
sel
ect
ion
of
BS,
m
od
e
sel
ect
ion
and
channel
al
locat
ion
in
orde
r
to
op
ti
m
iz
e
the
powe
r
co
ntr
ol
at
up
li
nk
com
m
un
ic
at
ion
in
betwee
n
the
B
Ss
an
d
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
Hetero
ge
neous
n
et
work
opti
m
izati
on u
si
ng r
obus
t
powe
r …
(
Sa
vi
ta
Pa
ti
l
)
4233
FCUs.
T
he
ou
ta
ge
pro
bab
il
it
y
con
diti
on
in
a
HetNet
is
app
li
ed
in
ord
er
to
decr
e
ase
the
ov
e
rall
energy
consum
ption
,
a
lso
this
guara
nt
ee
to
al
l
fem
to
-
cel
l
us
e
rs
a
bo
und
of
m
in
and
m
ax
fo
r
the
w
or
st
le
vel
of
ou
ta
ge
pro
bab
il
it
y.
In
co
ns
ide
r
H
et
Net
en
vir
on
m
ent
,
the
FCUs
an
d
MC
Us
m
akes
the
request,
base
d
upon
the
i
r
requirem
ent
th
e
cha
nn
el
an
d
traff
ic
detai
ls
a
re
ac
qu
ire
d,
th
en
the
se
t
hi
ngs
f
orwarde
d
to
our
pro
posed
RPR
A
for
furthe
r op
ti
m
iz
at
ion
p
r
oce
ss
to
get m
axim
u
m
u
ser
thr
ough
pu
t a
nd
op
t
i
m
iz
e p
ow
e
r
tr
ansm
issi
on
.
4.1.
P
ara
me
te
r
ini
tializ
at
ion
In
order
t
o
sim
ula
te
the
resu
lt
s
in
Ma
tl
ab2
01
6b,
we
have
us
ed
syst
em
config
ur
at
io
n
s
uch
as
;
8G
b
RAM,
i5
I
ntel
Pr
oc
esso
r,
1T
b
Hard
disk
a
nd
W
in
dows
10
OS
.
T
he
se
ve
r
al
con
si
der
e
d
necessa
ry
par
a
m
et
ers
that
us
e
d
in
tra
ff
ic
c
onditi
on
s
cenari
os
is
pr
e
sented
i
n
ta
ble1.
T
he
gro
wth
of
m
ob
il
e
use
r
s
causes
ext
ra
sign
al
interfe
ren
ce
at
FCUs
a
nd
MC
Us
in
sm
a
ll
cells
HetNet
scen
ario;
ther
ef
or
e
,
the
fem
to
-
cel
l
input
is
ta
ken
from
real
-
ti
m
e
scenario
li
ke
as
vi
deo
data
[31
]
and
au
dio
data
[
32
]
in
order
to
get
reali
sti
c
m
ulti
-
m
edia
transm
issi
on
.
Unde
r
this c
on
sidere
d
tra
ff
ic
conditi
on, we a
re going t
o
ta
ke
f
ive
sc
ena
rio
s su
c
h
as;
Scen
ario
-
A
,
Scena
rio
-
B
,
Sc
enar
i
o
-
C,
Sce
na
rio
-
D
an
d
Sce
nar
i
o
-
E
.
In
eac
h
of
the
sce
nari
o
w
e
hav
e
ta
ke
n
1
m
acro
-
cel
l,
30
nu
m
ber
of
MC
Us,
12
0
nu
m
ber
of
FCUs
,
w
her
e
the
l
ocati
on
of
fem
to
-
cel
l,
FCUs
a
nd
MC
Us
are
ge
ne
rate
d
rand
om
l
y.
Table
1
.
Use
d param
et
er in
it
ia
li
zat
ion
Para
m
eter
Valu
e
Carrie
r
Ban
d
wid
th
2
0
0
0
Mhz
Su
b
carr
ier
Ban
d
wid
th
3
7
5
KHz
Sy
ste
m
ban
d
wid
th
1
0
MHz
Maxi
m
u
m
tr
an
s
m
i
t po
wer
2
0
d
B
m
Mini
m
u
m
trans
m
it
po
wer
0
d
B
m
An
ten
n
a gain
1
4
d
Bi
Cell radiu
s
1
K
m
Co
rr
elatio
n
d
istan
ce
4
0
m
BS trans
m
it
po
wer
4
3
d
B
m
User sp
eed
3
K
m
/h
Nu
m
b
e
r
o
f
M
a
cro
-
cell
1
Nu
m
b
e
r
o
f
FCU
30
Nu
m
b
e
r
o
f
M
CU
120
4.2.
Perf
or
man
ce
anal
ys
is
4.2.
1.
Scen
ari
o
-
A
In
sce
na
rio
-
A,
w
e
ha
ve
c
onsidere
d
1
m
acro
-
cel
l,
4
fem
to
-
cel
l,
30
nu
m
ber
of
MC
U
s
an
d
12
0
nu
m
be
r
of
FCUs
.
Her
e
,
F
ig
ur
e
2
s
hows
the
t
hro
ughput
i
n
Mb
ps
as
per
ti
m
e
inter
val
,
wh
e
re
our
pro
posed
m
od
el
ha
s
perform
ed
sign
ific
antly
bette
r
at
each
tim
e
interval.
On
l
y
in
110.9s
ti
m
e
interval
there
is
a
drop
in
ou
r
pro
po
se
d
ap
pr
oach,
ot
herwis
e
in
ta
kin
g
a
ve
rag
e
of
al
l
tim
e
interval
our
pro
posed
m
od
el
RPR
A
got
10
.
15
Mbp
s
thr
ou
ghput,
w
hich
is
13%
m
or
e
com
par
e
to
H
ARQ
-
CC
[33]
an
d
25.
5%
m
or
e
c
om
par
e
to
HA
R
Q
-
T1
[34].
Fig
ur
e
3
sh
ows
the
d
el
ay
in
s
(seco
nd)
as
pe
r
the
perform
ed
step
s
,
w
he
re
the
aver
a
ge
delay
co
ns
ide
ri
ng
al
l
the steps
is
0.1
5
s
f
r
om
our
RPR
A
m
od
el
, 0
.
24
s obta
ined
by HARQ
-
CC
[
33
]
a
nd 2.92
s
ob
ta
ine
d
by
HA
R
Q
-
T1
[
34]
,
the
refor
e
the
delay
in
pe
rfor
m
ance
th
rou
ghout
the
ste
ps
our
m
od
el
ha
s
consum
ed
ve
ry less t
i
m
e compare
to othe
rs.
Figure
2
.
Thr
ough
pu
t a
s
per
t
i
m
e interval
Figure
3
.
D
el
a
y as pe
r
the
p
e
r
form
ed
ste
ps
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.
9
, N
o.
5
,
Oct
ober
20
19
:
4226
-
4
2
3
7
4234
4.2.
2.
Scen
ari
o
-
B
In
Scen
ario
-
B,
w
e
hav
e
co
ns
i
der
e
d
1
m
acro
-
cel
l,
8
f
em
to
-
cel
l,
30
num
ber
of
MC
Us
a
nd
12
0
num
ber
of
FCUs
.
He
r
e
,
F
ig
ur
e
4
shows
the
t
hroughp
ut
in
M
bps
as
pe
r
ti
m
e
interval
,
w
here
our
pro
pose
d
m
od
el
RPR
A
got
21
.95
Mb
ps
ave
rag
e
th
r
oughput,
H
ARQ
-
CC
has
achiev
ed
17
.
33
Mb
ps
and
HA
R
Q
-
T1
has
achieve
d
15.4
Mbps
.
Fi
gur
e
5
s
hows
th
e
delay
as
pe
r
the
pe
rfor
m
ed
ste
ps,
w
he
re
th
e
a
ve
rage
delay
consi
der
i
ng
al
l
the
ste
ps
is
1.1
4
s
f
ro
m
our
RPR
A
m
od
el
,
0.5
2
s
is
obta
ined
by
H
ARQ
-
CC
a
nd
6.5
s
is
ob
ta
ine
d by
H
ARQ
-
T1
.
Figure
4
.
Th
r
ough
pu
t a
s
per
t
i
m
e interval
Figure
5
.
D
el
a
y as
pe
r
the
p
e
r
form
ed
ste
ps
4.2.
3.
Scen
ari
o
-
C
In
Sce
nar
i
o
-
C
,
w
e
ha
ve
co
nsi
der
e
d
1
m
ac
ro
-
cel
l,
12
fe
m
to
-
cel
l,
30
num
ber
of
MC
Us
an
d
12
0
nu
m
ber
of
FC
Us.
Fig
ure
6
s
hows
the
t
hro
ughput
as
pe
r
tim
e
interval
,
wh
e
re
dro
pout
of
thr
ough
pu
t
can
be
seen
in
194
s
,
24
9
s
an
d
263
s.
Howe
ver,
sti
ll
ou
r
pro
posed
m
od
el
aver
a
ge
throu
ghput
of
HetNet
is
28
Mbp
s
,
wh
ic
h
16
%
m
or
e
com
par
e
t
o
HA
R
Q
-
T
1
a
nd
5.8
%
m
or
e
co
m
par
e
to
H
AR
Q
-
CC.
Fig
ur
e
7
s
hows
t
he
de
la
y
as
per
the
perfor
m
ed
ste
ps
,
w
he
re
the
ave
rage
delay
con
sid
erin
g
al
l
the
st
eps
is
0.77
s
f
r
om
ou
r
RPR
A
m
od
el
,
wh
ic
h 67
% les
s co
m
par
e to
HARQ
-
T1
a
nd 44 %
less c
om
par
e to
HARQ
-
CC.
Figure
6
.
Th
r
ough
pu
t a
s
per
t
i
m
e interval
Figure
7
.
Dela
y as pe
r
the
p
e
r
form
ed
ste
ps
4.2.
4.
Scen
ari
o
-
D
In
Sce
nar
io
-
D,
w
e
ha
ve
c
on
sidere
d
1
m
acro
-
cel
l,
16
fe
m
to
-
cel
l,
30
num
ber
of
MC
Us
a
nd
12
0
nu
m
ber
of
FC
Us.
Fig
ur
e
8
sh
ows
the
t
hroug
hput
as
pe
r
ti
m
e
interva
l
,
w
her
e
propose
d
m
od
el
a
ver
a
ge
thr
oughput
of
HetNet
is
38
Mb
ps
,
w
hic
h
17
%
m
or
e
com
par
e
to
H
ARQ
-
T1
an
d
8
%
m
or
e
c
om
par
e
to
H
ARQ
-
CC.
Figure
9
s
how
s
the
delay
as
per
the
pe
rform
ed
ste
ps
,
w
he
re
the
ave
ra
ge
delay
c
on
si
de
rin
g
al
l
the steps
is
1.6
s f
r
om
o
ur RP
RA m
od
el
,
2.1
6
s
is
obta
ine
d by H
ARQ
-
CC
and
20
s
is
ob
t
ai
ned
by
HA
R
Q
-
T1
.
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
Hetero
ge
neous
n
et
work
opti
m
izati
on u
si
ng r
obus
t
powe
r …
(
Sa
vi
ta
Pa
ti
l
)
4235
Figure
8
.
Th
r
ough
pu
t a
s
per
t
i
m
e interval
Figure
9
.
Dela
y as pe
r
the
p
e
r
form
ed
ste
ps
4.2.
5.
Scen
ari
o
-
E
In
Scena
rio
-
E,
w
e
ha
ve
c
on
sidere
d
1
m
ac
ro
-
cel
l,
20
fe
m
to
-
cel
l,
30
num
ber
of
MC
Us
a
nd
12
0
nu
m
ber
of
FC
Us.
F
i
gure
10
shows
t
he
t
hro
ughp
ut
as
pe
r
tim
e
interv
al
,
wh
e
re
pro
po
s
ed
m
od
el
aver
a
ge
thr
oughput
of
HetNet
is
47
Mbps,
43
Mbp
s
is
obta
ined
by
HA
R
Q
-
CC
an
d
39
.6
Mb
ps
is
ob
ta
ine
d
by
H
ARQ
-
T1
.
Figure
11
s
hows
the
delay
as
per
the
pe
rform
ed
ste
ps
,
w
he
re
the
a
ve
rag
e
delay
co
ns
id
er
ing
al
l
the steps
is
3.6
s f
r
om
o
ur RP
RA m
od
el
,
4.7
s
is
obta
ine
d b
y HA
R
Q
-
CC
a
nd
13
s
is
ob
ta
i
ned b
y
H
ARQ
-
T1
.
Figure
10
.
T
hr
oughput as
p
e
r t
i
m
e interval
Figure
11
.
Del
ay
as p
e
r
the
pe
rfor
m
ed
ste
ps
Figure
12
.
A
ve
rag
e
po
wer
t
ra
ns
m
issi
on
Figure
13
.
O
btained
a
ve
rag
e
thro
ughput
Evaluation Warning : The document was created with Spire.PDF for Python.