TELKOM
NIKA
, Vol.11, No
.11, Novemb
er 201
3, pp. 6577
~6
583
e-ISSN: 2087
-278X
6577
Re
cei
v
ed Ap
ril 23, 2013; Revi
sed
Jun
e
20, 2013; Accepted July 2
1
,
2013
An Efficient Reverse C
onverter for The New High
Dynamic Range 5-Moduli Set
Xiaolan L
v
*,
Ming Xiao
Coll
eg
e of Co
mputer an
d Ele
c
tronic Informa
tion, G
uan
gdo
ng Un
iversit
y
o
f
Petrochemica
l
T
e
chnolog
y,
Maomin
g 52
50
00, Guan
gdo
n
g
,
Chin
a
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: 2909
01
37@
q
q
.com
A
b
st
r
a
ct
In this
pa
per,
an
efficie
n
t re
sidu
e to
bi
nar
y conv
erter
de
sign
for th
e
n
e
w
hig
h
dyn
a
m
ic
ra
ng
e
mo
du
li set {2
n
-1
,2
n
+1
,2
2n
,2
2n
+1,
2
2n-
1
-1} is pr
esente
d
. T
he r
e
verse c
onver
sion i
n
the fo
u
r-mo
dul
i set {2
2n
,
2
2n
+1
, 2
n
+1
, 2
n
-1} has b
een
p
r
opos
ed i
n
liter
ature. Henc
e, t
he conv
erters
are bas
ed o
n
the new
mod
u
li
set
{2
2n-
1
-1, (2
n
-1)(2
n
+1)(2
2n
+1
)2
2n
} and pr
op
ose i
t
s residue to
bi
nary conv
erter
usin
g New
Chi
nese R
e
ma
ind
e
r
T
heore
m
2 (
N
e
w
CRT
2). T
h
e n
e
w
modu
li
set are
pro
pos
ed w
i
th
a dy
na
mic
ran
ge
8n-
1
bits a
n
d
has
th
e
same fe
atures
of the
po
pul
ar
on
e. W
hen
c
o
mpar
ed to
th
e co
mmo
n
fiv
e
modu
li r
e
vers
e co
nverters, t
h
is
enh
anc
ed
mod
u
li set has
mor
e
dyna
mic ran
ge, and
it us
eful for hig
h
perf
o
rmanc
e co
mp
uting.
Ke
y
w
ords
: reverse conv
erte
r
,
residue n
u
m
ber system (R
NS), new
Chinese re
main
de
r theore
m
s (n
ew
CRT).
Copy
right
©
2013 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
The conve
n
tional weig
hte
d
num
be
r system
has
ca
rry p
r
op
agati
on am
ong
a
r
ithmetic
operation
s
, resultin
g in pe
rforma
nce de
grad
ati
on
in hard
w
a
r
e co
mputing syst
ems.
The re
sidue
numbe
r
sy
ste
m
(R
NS) i
s
a
n
effi
cient alt
e
rnative n
u
m
ber
system
which
ha
s be
e
n
an im
porta
nt
resea
r
ch field
in compute
r
arithmeti
c
for
many
de
ca
de
s. On
e of the
key adva
n
tag
e
s of
RNS is
a
carry-free nu
mber
sy
stem
,
whi
c
h ca
n repre
s
e
n
t
nu
mber in
a n
o
n
-weighte
d
f
o
rm.
High
sp
eed
and le
ss ha
rd
ware
compl
e
xity could be
achi
ev
ed by
decompo
sin
g
a large bi
nary numbe
r int
o
a
set of small
e
r resi
due
s [1-3]. RNS ha
s dra
w
n wi
de
sprea
d
attention for in
cre
a
s
ing p
e
rfo
r
m
ance
of com
m
uni
cation sy
stem,
Digital Sig
n
a
l Pro
c
e
ssi
n
g
such
a
s
F
ourie
r tran
sfo
r
m (FFT),
di
gita
l
filter, and ima
ge pro
c
e
s
sin
g
.
The b
a
si
c b
u
ilding bl
ocks
for a
RNS
ne
eded
re
sidu
e
arithmeti
c
u
n
it, binary-to
-redid
u
e
conve
r
ters, a
nd re
sid
ue-to
-bina
r
y co
nverters [4
, 5]. Choi
ce of t
he mod
u
li set that inclu
des
relatively pri
m
e intege
rs
and de
sig
n
s
of the
reverse co
nverte
r are imp
o
rta
n
t becau
se th
ese
issue
s
have
basic effe
cts on reve
rse conv
e
r
ter’
s perform
an
ce, they have long been
the
perfo
rman
ce
bottlene
ck f
o
r RNS [8-1
5]. Another
i
m
porta
nt arit
hmetic p
r
obl
em is
conve
r
ting
resi
due to bi
nary num
ber,
which is the
cru
c
ial fo
r
a RNS ap
plication. The tradit
i
onal alg
o
rith
ms
of reverse converte
r are
Chine
s
e Re
mainde
r The
o
rem (CRT) and Mixed-radix co
nversion
(MRC). Howe
ver, the use of CRT is un
profitable
sin
c
e it is involve a large mo
dulo M ope
ra
tion,
whe
r
e M is th
e prod
uct of all moduli [6]. The MRC i
s
a sequ
ential
pro
c
e
ss
whi
c
h requi
re
s times
[7]. Rece
ntly, Wan
g
ha
s p
r
opo
sed
Ne
w
Chin
ese Rem
a
inde
r Th
eorem 1 (Ne
w
CRT I) a
nd
Ne
w
Chin
ese Re
m
a
inde
r Th
eorem 2 (Ne
w
CRT II), whi
c
h
are b
a
sed o
n
CRT
and
M
RC [8]. The
Ne
w
CRT
s
algo
rithm elimin
ate
s
the d
r
a
w
b
a
cks
of tr
adit
i
onal
CRT
a
nd M
RC, resulting in n
o
ta
ble
improvem
ent in hard
w
a
r
e.
By far, many
different m
o
d
u
li sets
have
been
propo
se
d, whi
c
h
hav
e different p
r
opertie
s
,
whi
c
n acco
rdi
ng to dynami
c
ran
ge (DR),
arithmet
ic op
eration
s
an
d hard
w
a
r
e imp
l
ementation. In
some high
-p
erform
an
ce comp
utation system
s,
the
y
demand
more p
a
rall
e
lism with la
rger
dynamic
ran
g
e
. A few of five-mo
duli set
have bee
n re
ported, Sk
avantzo
s propo
sed a
fi
ve-m
o
duli
set
{2
n
-1,2
n
,2
n
+1,
2
n
+2
(n+1)/2
+1,
2
n
-2
(n+1)/2
+1} [9], Math
ew et al. al
so pro
p
o
s
ed
a
fi
ve-mod
uli
set
{2
3n
-1,2
3n
,2
n
+1,2
3n
+2
(3n+1)/2
+1,
2
3n
-2
(3n+1)/2
+1} [10], they have the sa
me
disa
dvant
age
s that all the
moduli a
r
e
no
t in the form
of 2
n
or 2
n
±1 ,
the multiplicative inverses
are in
poo
r fo
rmats,
re
sulti
n
g
in the sp
eed
of the arith
m
etic unit i
s
re
st
ricte
d
. Skavant
zo
s an
d Stouraiti
s prop
osed a
fi
ve-
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No
. 11, Novemb
er 201
3: 657
7 – 6583
6578
moduli set
{
2
n+1
, 2
n
-1, 2
n
+1, 2
n+1
-1,2
n+1
+1}
[11], Ca
o
et al. al
so p
r
opo
se
d a
rel
a
tively balan
ced
fi
ve-
m
oduli set { 2
n
-1,2
n
, 2
n
+1,
2
n+1
-1,2
n-1
-1} [12], though all the moduli are in the form of 2
n
or
2
n
±1 with thi
s
two mod
u
li sets, they are
only co
-pri
me
for even val
ues of
n
and dynamic ran
g
e
are only
5
n
bits , la
rg
er dyn
a
mic
rang
e mo
dul
i
set
s
are
necessa
ry fo
r large
num
ber
comp
utation
s
. A few of larger dyn
a
mic
rang
e mo
d
u
li
sets h
a
ve propo
sed, which com
p
ri
se
s
non
c
o
-prime moduli [13].
In this paper,
we propo
se a
resid
ue to binary co
nverte
r desi
gn ba
se
d on Ne
w CRT I and
New CRT II
with our new
moduli
s
e
t {2
n
-1,2
n
+1,
2
2n
,2
2n
+1,
2
2n-1
-1},
whi
c
h i
s
valid
any value
of n.
These involvi
ng mod
u
li of this mo
duli
se
t are in the fo
rm of 2
n
or
2
n
±1, this
makes
the
c
onverter
desi
gn i
n
le
ss h
a
rdware
complexity. Th
e p
r
op
osed
moduli
set is de
rived f
r
om
the m
oduli
set
{2
2n
, 2
2n
+1,
2
n
+1,
2
n
-1
}
[14],
while
the dynamic ran
ge has rai
s
ed
to 8
n
-1
bits. A two-l
e
vel de
si
gn
of reverse converte
r for the prop
osed
m
oduli set based on co
mbination of
New Chine
s
e
Remai
nde
r T
heorem 1
(New
CRT I)
an
d Ne
w
Ch
i
n
e
s
e
Rem
a
ind
e
r
Th
eorem 2
(Ne
w
CRT II) is
pre
s
ente
d
.
The rest of
the pap
er i
s
org
anized
as fo
llo
ws: In se
ction
2, we p
r
ovid
e
a bri
e
f
introdu
ction
f
o
r th
e
RNS
s
and
Ne
w CRTs. In
sectio
n
3, the
ne
w 5
-
mod
u
li
set
is
propo
se
d,
and
an
efficient re
verse co
nvert
e
r
alg
o
rithm
s
for
the
ne
w fi
ve-mod
uli set
that base
d
o
n
Ne
w CRT
s
is
provide
d
. Ha
rdwa
re impl
e
m
entation is
pre
s
ente
d
in se
ction 4. Evaluating the p
r
opo
se
d reve
rse
conve
r
ter a
n
d
com
pare th
e re
sult in te
rms of
conve
r
sio
n
del
ay and ha
rd
ware co
st with oth
e
r
simila
r reve
rse conve
r
ters i
n
se
ction 5. T
he pap
er is
concl
ude
d in section 6
2. Backg
rou
nd
In a re
sid
u
e
numb
e
r
system (RNS),
an integ
e
r
X
can
be
defi
ned by
a m
oduli
set
{
m
1
,m
2
,
...
,m
n
}, which is
co
nsi
s
ted of a
set of pairwise
relatively pri
m
e intege
r n
u
mbe
r
s, i.e., gcd
(
m
i
, m
j
)=1 for
i
≠
j. The dyn
a
mic
rang
e (DR) M is th
e
prod
uct of the mod
u
li,
1
n
i
i
M
m
, A
weig
hted re
prese
n
tation
X
can b
e
rep
r
e
s
ented a
s
X
=(
x
1
,
x
2
,...,
x
n
), where:
mo
d
1
,
2
,
0
i
ii
i
i
m
x
XX
m
i
n
x
m
(1)
Then integ
e
r
X
has a uniq
u
e
n-tuple
rep
r
ese
n
tion:
12
,,
0
RN
S
n
Xx
x
x
X
M
.
Ne
w Chin
ese
Rem
a
ind
e
r
Theo
rem 1 (
N
e
w
CRT
I)
[
8
]:
For a
m
o
duli set
{
m
1
,
m
2
,
...
,
m
n
},
the weighte
d
number X can be co
nverted
from its a set of resi
due num
ber
(
x
1
,
x
2
,...,
x
n
), it is
pre
s
ente
d
usi
ng Ne
w CRT
I as:
23
1
12
1
2
2
3
2
11
12
3
1
1
()
(
)
()
nn
nn
n
n
mm
m
m
kx
x
k
m
x
x
Xx
m
km
m
m
x
x
(
2
)
Whe
r
e
23
1
3
1
3
1
11
2
1
2
2
1
2
1,
1
,
,
1
nn
nn
nn
m
m
mm
m
m
m
m
mm
km
k
m
m
k
mm
Ne
w
Chin
es
e R
e
mai
nde
r
The
o
re
m 2
(
Ne
w
CR
T II)
[8]: For
a
2
-
mod
u
li
set
{
m
1
,m
2
},
whe
r
e
m
1
<
m
2
. the weig
hte
d
numb
e
r
X
can be
co
nvert
ed from it
s re
sidu
e re
prese
n
tation (
x
1
,x
2
),
it can be re
prese
n
ted by New CRT II as
follows:
1
22
0
1
2
()
m
Xx
m
k
x
x
(3)
1
02
1
m
km
(4)
Whe
r
e,
k
i
is t
he multiplicative inverse.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
An Efficient Reve
rse Co
nverter fo
r The
Ne
w
Hig
h
Dynam
ic Ran
g
e
5-Mod
u
li Set (Xiaolan L
v
)
6579
3. Rev
e
rse Conv
erter to 5
-
Moduli Set {
2
n
-1,2
n
+1,2
2n
,2
2n
+1,
2
2n-1
-1}
In this sectio
n, we apple t
he Ne
w CRT-I and the Ne
w CRT-
Ⅱ
to the prop
osed
supe
rset
{2
n
-1,2
n
+1,
2
2n
,2
2n
+1,
2
2n-1
-1}
to achieve a high-perfo
rmance re
si
d
ue to binary
converte
r, its
dynamic
ra
ng
e is 8
n
-1
bits. We de
note
the re
sidu
es
corre
s
p
ondin
g
to this mo
d
u
li set (
m
1
,
m
2
,
m
3
,
m
4
,
m
5
) as (
x
1
,
x
2
,
x
3
,
x
4
,
x
5
).
A two-level
conversion
alg
o
rithm for th
e
resi
due to
bi
nary conve
r
si
on is
pre
s
e
n
ted. The
moduli
set
S={2
n
-1,2
n
+1,2
2n
,2
2n
+1,
2
2n-1
-1} i
s
de
comp
ose
d
into
two
su
bsets. i.e.,
S
1
={2
2n
, 2
2n
+1,
2
n
+1,
2
n
-1} a
nd
S
2
={
2
2n-1
-1,2
2n
(2
n
-1
)(
2
n
+1)
(
2
2n
+1)}.
From [14], the moduli set S
1
={2
2n
, 2
2n
+1,
2
n
+1,
2
n
-1} a
r
e pai
rwise p
r
ime
whe
n
n
is valid for
any value. T
her
efo
r
e, it is necessa
ry that
prove the all
moduli of S
1
are all pai
rwise prime to the
moduli of 2
2n-1
-1 for any value of
n
.
Theo
rem 1:
The num
be
r of 2
n
-1,2
n
+1,
2
2n
,2
2n
+1,and
2
2n-1
-1 are pairwi
se
relativ
e
ly prime
for any value of
n
.
Proof: Acco
rding to the Euclid’
s
The
o
rem gcd
(
a
,
b
)=
gc
d(
b
,
a
mod
b
), it is
easy to find.
21
1
1
21
1
1
21
2
2
22
1
2
1
g
c
d
2
1,
2
1
g
c
d
2
1,
2
1
g
c
d
2
1,
1
1
2
1
,
2
1
2
1
,
(
2
1)
(
2
1)
,
2
1
21
,
2
2
,
1
1
21
,
2
1
,
2
1
,
3
1
nn
n
n
n
nn
n
n
n
nn
n
nn
n
gc
d
g
c
d
g
c
d
gc
d
g
c
d
gc
d
g
c
d
(5)
It can
be
see
n
that
all the
greate
s
t
com
m
on
di
viso
rs
are
eq
ual to
1, therefore
t
he fou
r
numbe
rs 2
n
-1
,2
n
+1,
2
2n
,and 2
2n
+1 are rela
tively prime to the modulo
2
2n-1
-1.
At the beginn
ing, we assu
me that an integer
X
1
is re
pre
s
ente
d
as (
x
1
,
x
2
,
x
3
,
x
4
) based on th
e
four-m
oduli set
S
1
={2
2n
, 2
2n
+1,
2
n
+1,
2
n
-1}. The
fi
nal w
e
ig
h
t
ed
X
can b
e
calculated fro
m
the
resi
due
s (
X
1
, x
5
) corre
s
po
n
d
to the seco
nd moduli
set
S
2
={2
2n-1
-1, (
2
n
-1)
(
2
n
+1)
(
2
2n
+1)
2
2n
}.
In the firs
t level of the converter, A
c
c
o
rding
to th
e
conve
r
si
on
al
gorithm
of [1
4], the
weigh
t
ed
X
1
=(
x
1
,
x
2
,
x
3
,
x
4
) ca
n be o
b
tained by:
2
11
2
n
X
xZ
(6)
Z
is th
e o
u
tpu
t
of modu
o 2
4n
-1 ad
de
r, the
r
efore Z i
s
a
4
n
-bit i
n
tege
r, then X
1
ha
s
6
n
bit
s
.
In the se
co
n
d
level of the
reverse
co
n
v
erter, by ap
plying (3
) to
the moduli
set S
2
, S
2
can be
rewritten as
:
2
2
21
2
4
21
2
2
(
2
1
)
(
2
1
)(
2
1
),
2
1
(
2
(
2
1
)
,
2
1
)
n
n
nn
n
n
nn
S
(7)
X
can be
calculated by Ne
w CRT-
Ⅱ
.
21
24
10
5
1
21
2(
2
1
)
(
)
n
nn
XX
k
x
X
(8)
L
e
mma
1
: Th
e multiplicativ
e inverse of the numb
e
r 2
2n
(2
4n
-1) m
odu
lo 2
2n-1
-1 is
k
0.
21
2
2
0
1
2(
2
1
)
1
2
3
nn
k
(9)
Proof
: Ac
cording to (4), we have
21
24
0
21
2(
2
1
)
1
n
nn
k
。
Then, (8
) can
be rewritten
as:
21
24
2
1
2
2
1
51
21
1
2(
2
1
)
2
(
2
1
)
1
2
(
)
3
n
nn
n
n
XX
x
X
(10
)
Its origin
al form of
k
0
is n
o
t amena
ble to
hard
w
a
r
e d
e
s
ign, the
n
it is expa
nded i
n
to geom
etri
cal
seri
es.
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No
. 11, Novemb
er 201
3: 657
7 – 6583
6580
2
1
0
246
2
2
1
[2(2
1
)
1
)
]
2
2
2
2
2
3
nn
(11)
Finally, the weighted X ca
n
be rep
r
e
s
ent
ed as follo
w:
21
2
4
0
246
2
2
2
2
1
51
21
2
(
2
1
)
(
2
222
2
)
2
(
)
n
nn
n
n
XX
x
X
(12
)
4. Hard
w
a
re
Realization
of Rev
e
rse Conv
erter
Hardware fra
m
e of the propo
sed 5
-
mo
duli set S
=
{2
n
-1,2
n
+1,
2
2n
,2
2n
+1,
2
2n-1
-1} r
e
v
e
rs
e
conve
r
ter i
s
shown Figu
re
1.
Figure 1. Fra
m
e of the Pro
posed
5-mod
u
li Set Reverse Converte
r
Hardware implementatio
n for intege
r
X
1
from the re
si
due
s (
x
1
,
x
2
,
x
3
,
x
4
) correspond to
the four-mod
uli set S
1
={2
2n
, 2
2n
+1,
2
n
+1
,
2
n
-1} h
a
s
co
mpleted by
[1
4]. The aim o
f
this se
ction
is
to implement
the hardware
architectu
re o
f
X
from the resid
u
e
s
(
X
1
, x
5
) corre
s
po
n
d
to the se
co
nd
moduli set S
2
={2
2n-1
-1, (2
n
-1)
(
2
n
+1
)(
2
2n
+1)
2
2n
}. Figure
2 shows the architectu
re o
f
the residue
to
binary conve
r
ter for the 2-moduli set S
2
.
Figure 2. Architecture of Reverse Co
nverter fo
r the 2
-
mod
u
li set S
2
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
An Efficient Reve
rse Co
nverter fo
r The
Ne
w
Hig
h
Dynam
ic Ran
g
e
5-Mod
u
li Set (Xiaolan L
v
)
6581
The (1
2)
can
be re
written a
s
:
24
1
2(
2
1
)
nn
X
XR
(13)
Whe
r
e:
21
21
21
21
0
2
22
22
0
2
22
51
1
2
21
2
1
22
22
15
2
1
21
2
1
(2
2
2
)
2
(
)
(2
2
2
)
(
)
2,
2
(
)
n
n
nn
nn
n
nn
Rx
X
r
r
rx
r
X
(14
)
No
w, we
co
nsid
er Equ
a
tion (1
3) a
nd
(14
)
, and
si
mplify them for efficie
n
t hard
w
a
r
e
implement. Firs
t, we
s
h
ow
two
L
e
mma
s
,
they ca
n b
e
use
d
to impl
e
m
ent mod
u
lo
2
n
-1 add
er by
s
i
mply c
i
rcular s
h
ift.
L
e
mma
2
:
21
2
m
n
x
is equivalent to
circula
r
lef shi
ft a
m
-bits bin
a
ry
namb
e
r
x
by
n
bits
[12].
L
e
mma
3
:
21
2
m
n
x
is equivalent
to circula
r
lef
sh
ift
a
m
-bits binary na
mbe
r
x
by
n
bits , t
hen
compl
e
me
nt of result [12].
L
e
mma
4
:
21
2
m
ms
i
is equivalent to
21
2
m
i
Whe
r
e,
n
and
m
are any natural num
bers.
Since
X
1
is a 6
n
-bits n
u
m
ber an
d
x
5
is a (2
n
-1
)-bits numbe
r, they are rep
r
e
s
ented in
their binary forms
as
follows
:
1
1
,6
1
1
,6
2
1
,
1
1
,
0
6
nn
n
X
XX
X
X
5
5
,2
2
5
,2
3
5
,
1
5
,
0
21
nn
n
x
xx
x
x
Whe
r
e,
x
i,j
de
note the
jth
of
x
i
.
Th
en,
r
1
an
d
r
2
can
be
furthe
rmo
r
e
si
mply by lem
m
a 2
an
d
lemma 3.
21
22
15
5
,
0
5
,
2
2
5
,
2
3
5
,
1
21
21
2
n
n
nn
n
rx
x
x
x
x
(15)
21
21
22
21
21
6
3
6
1
62
42
64
6
5
4
1
24
22
1
2
1
4
3
44
2
0
22
2
3
1
21
2
1
21
2(
)
11
11
n
n
n
n
n
n
nnn
n
n
n
nn
n
n
n
n
nn
rX
XX
X
X
X
X
X
XX
X
X
X
X
X
X
21
22
23
2
()
+
()
r
rrr
()
()
21
21
n
4
(16
)
Whe
r
e,
i
x
is the complem
e
nt of
i
x
, and
r
21
,
r
22
,
r
23
,
r
24
can
be rep
r
esent
ed in th
eir bin
a
ry form
s .
By substitutin
g
(15
)
and (1
6) in
to (1
4), R can be
cal
c
ul
ated by:
21
024
2
2
12
12
2
2
3
2
4
21
(2
2
2
2
)
(
)
n
n
Rr
r
r
r
r
(17)
By applying
Le
mma
2, Le
mm
a
3
and
Le
mm
a
4
in Equation (15), (16) a
nd (1
7),
they can
be sim
p
lified
to decrea
s
e
the hardwa
r
e comp
lexit
y
. The Ope
r
and Prepa
rat
i
on Unit
(OP
U
)
inclu
d
e
s
simp
ly manipul
ating the
routin
g of th
e
bit
s
and i
n
verte
r
s of the
re
sid
u
e
s th
at p
r
ep
a
r
es
the ope
ran
d
s for mod
u
lo
adde
r. The i
m
pleme
n
tatio
n
of R
requi
res a
5
n
-ope
rand carry
sa
ve
adde
r (CSA)
with en
d-a
r
o
und
carry (E
AC) [15] tre
e
followe
d by a modul
o 2
2n-1
-1 add
er th
a
t
is
one (5
n
, 2
2n-1
-1) M
u
lti-O
p
e
r
and
Modul
ar Adder
(MO
M
A). In the p
aper,
we
con
s
ide
r
ed
mod
u
lar
adde
r that i
s
the ca
rry p
r
o
pagate
add
er (CPA)
wi
th e
nd aroun
d ca
rry (EA
C
) [16
]. Furtherm
o
re,
(13
)
ca
n be rewritten a
s
:
24
6
2
2
11
2(
2
1
)
X
2
2
2
nn
n
n
n
XX
R
R
R
Y
R
(
1
8
)
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No
. 11, Novemb
er 201
3: 657
7 – 6583
6582
Whe
r
e:
6
1
2
2
2
3
1
0
1
,6
1
1
,6
2
1
,
1
1
,
0
21
6
2
n
nn
n
n
n
n
YX
R
R
R
R
R
X
X
X
X
(19)
It should
be
noted, b
e
cause
X
1
is a
6
n
-bits nu
mber, (19
)
can be reali
z
ed by
sim
p
le
con
c
ate
natio
n without ad
d
i
tional
hardware an
d co
mp
utation co
st.
Finally, by s
ubs
tituting (19) in (18), we have:
2
11
2
2
2
3
1
0
42
21
21
1
1
1
1
1
1
1
1
n
nn
nn
n
XY
V
Y
V
V
V
V
(20)
Reali
z
ation
o
f
X
relie
s on a
(8
n
-1)-bits binary su
btra
cter, whi
c
h can
b
e
impl
e
m
ented rely on
a
(8n
-
1)-bit
s re
gular
CPA with ‘1’ carry-in
and (2
n
-1 ) NOT gate
s
.
Example 1
:
Fo
r
n
=8
,
the pro
p
o
s
e
d
5-m
oduli
set is S
=
(6
5536,6
553
7,2
55,257,
3276
7
),
th
e weig
hted
numbe
r
X
can be
calcul
ated fro
m
its RNS
representat
ion
(121,1
65,27
4
1
,1056,3
012
5
)
, the re
si
d
u
e
s
have
bina
ry rep
r
e
s
entati
on
x
1
=(0000
1
0101
0110
101
),
x
2
=(00
0000
1
0000
1000
00
),
x
3
=(0
111
10
01),
x
4
=(010
1001
01) and
x
5
=(111
010
1101
0110
1).
For
the first level
of the propo
sed reverse
co
nverter, the
4
-
mod
u
li set S
1
=(6
5536,6
5
5
37,255,2
57), i
t
s
equivalent weighted
n
u
m
ber
X
1
=(11
9
4449
7348
884
)
10
that we
can obtain
ed
by (6). Fo
r t
h
e
se
con
d
l
e
vel of
the
prop
osed
reve
rse
co
nverte
r, the
2-m
oduli
S
2
=(32
767,2
5
5
×2
57
×65
5
3
6
×6
553
7), a
c
cordi
ng to
(14
)
-(19
),
we h
a
ve R=(3
080
4)
10
and
Y
1
=(86
706
74
6275
6853
624
5)
10
, finally,
X
=(8
670
674
6
2554
9765
301
)
10
can be o
b
tained by (2
0).
5. Performan
ce Ev
aluation
In this sectio
n
,
the area an
d delay of
pro
posed reve
rse conve
r
ter fo
r the new
fi
ve
-mod
uli
sup
e
rset a
r
e
evaluated. According to
Fi
gure
1
an
d Fi
gure
2, the p
r
opo
sed
5-m
o
duli re
sid
ue-t
o
-
binary
co
nve
r
ter co
nsi
s
ts of
one
fou
r
-moduli set
converte
r,
on
e
(5
n
, 2
2n-1
-1) MOMA for t
he
cal
c
ulatio
n of R, one (8
n
-1)-bit bina
ry subtracte
r
an
d som
e
inverters. Th
e esti
mated ha
rd
ware
co
sts a
c
cordi
ng to full add
er (FA
)
, and
modula
r
ad
d
e
r (m
od ad
d), subtra
ctor
(sub
) an
d so
me
logic gate
s
. F
u
ll ad
ders (F
A’s)
with
co
n
s
tant bit
s
of 1
’
s o
r
0’s a
r
e
redu
ced
to p
a
i
r
s of XO
R/AND
or XNOR/O
R gates. T
he t
o
tal co
nversi
on del
ay
and
hard
w
a
r
e
co
sts are the
su
m of the resi
due-
to-binary converter for the 4-moduli
s
e
t
S
1
and 2-m
oduli
set S
2
. Table 1
sh
o
w
s th
e ha
rd
ware
co
sts a
nd co
nversi
on d
e
la
ys of the pro
pos
ed 5
-
mod
u
li set reverse co
nverte
rs,
whe
r
e
t
FA
den
ote
the delay of an FA and L
d
enote the nu
mber of level
s
in an CSA tree, re
sp
ectiv
e
ly.
Table 1. Ha
rd
ware Ch
aract
e
rization of Each
Pa
rt of the Propo
se
d Reverse Co
nvere
r
Mod set
part
FA
NOT
XOR/AND
XNOR/
O
R
Dela
y
S
1
OP
U
-
6
n
+3 -
-
t
not
CSA1 2
n
+1 -
2
n
-1
-
t
FA
CSA2 2
n
+2 -
2
n
-2
-
t
FA
CSA3 2
n
+3 -
-
2
n
-3
t
FA
CPA 4
n
-
-
-
(8
n
)t
FA
Area
(10
n
+6)A
FA
+(4
n
-3
)
A
xo
r
+(4
n
-3
)A
an
d
+(2
n
-3
)A
x
nor
+(2
n
-3
)
A
or
+(6n+3)A
no
t
Dela
y (8
n
+3)t
F
A
+t
not
S
2
OP
U1
-
6
n
-
-
t
NOT
CSA 10
n
2
-9
n
+
2
-
-
--
L
1
CPA 2
n
-1
-
-
-
(4
n
-2
)t
FA
OP
U2
2
n
-1
-
-
t
NOT
8n-1 bit sub
8
n
-1
-
-
-
(8
n
-1
)t
FA
Area
(10
n
2
+
n
)A
F
A
+(8
n
-1
)
A
not
Dela
y
L
1
+2 t
NOT
+(12
n
-3
)t
FA
S
Area
(10
n
2
+11
n
+6) t
FA
+(4
n
-3
) t
xo
r
+(4
n
-3
) t
an
d
+(2
n
-2
)
t
x
nor
+(
2
n
-2
) t
or
+(14
n
+2
)t
not
Dela
y 3t
NOT
+20
n
t
F
A
+
L
1
The dynami
c
rang
e of the
prop
osed 5
-
moduli
set ha
s 8
n
-1 bits
.
Up to now, it is dif
fi
c
u
lt to find
a
existing m
o
d
u
li set
s
that
has th
e
sam
e
dynami
c
ra
nge a
s
th
e p
r
opo
se
d 5
-
m
oduli
set. Mo
duli
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
An Efficient Reve
rse Co
nverter fo
r The
Ne
w
Hig
h
Dynam
ic Ran
g
e
5-Mod
u
li Set (Xiaolan L
v
)
6583
sets of [10] a
nd [12] have t
he same
nu
mbers of ch
a
nnel
s as o
u
r
prop
osed mo
duli set, but h
a
ve
different dyn
a
m
ic
ran
ge.
We compa
r
e
o
u
r m
oduli
set
with [1
0] an
d [12]. Fo
r a
fair comp
ari
s
on
,
n
-bit CPA
s
with EAC are
con
s
id
ere
d
for the im
ple
m
entation of
the modul
o 2
n
-1 ad
der fo
r all
conve
r
ters. Table 2 sh
ows the
hardware co
sts and
conv
ersion d
e
l
a
ys of the propo
sed 5
-
mo
duli
set
reverse
conve
r
ters, [
10] an
d [12].
It is
cl
ea
r from the
table
that p
r
opo
sed 5
-
mo
duli
set
reverse co
nverters
h
a
s a
signifi
cant su
perio
rity
in hi
gh dynami
c
range m
oduli
set with
effici
ent
reverse
conv
erter.
L
1
a
nd
L
2
a
r
e
num
ber
of the l
e
vels of th
e
CSA tree
wit
h
(5
n) inp
u
t
and
((n/2)+1) input res
p
ec
tively.
Table 2. Ha
rd
ware Co
mple
xity and Conversi
on Delay Comp
ari
s
o
n
converter
DR
n
area
Dela
y
[10]
15n
odd
>(
36n
2
+144
n
)
A
F
A
>(27
n
+9)
t
FA
[12](
n
=6k)
5n
even
1
6
(5
n
2
+145
n
-
12)
A
FA
n=6k
(18
n
+
L
2
+7
)
t
FA
Proposed
8n-1
an
y
(10
n
2
+11
n
+6)
A
FA
+(4
n
-3)A
xo
r
+
(4
n
-3
)A
an
d
+(2
n
-2)
A
x
nor
+(2
n
-2
)A
or
+
(
14
n
+2
)
A
not
(3t
NOT
+20
n
t
FA
+
L
1
)
t
FA
6. Conclusio
n
This pa
per i
n
trodu
ce
d th
e ne
w high
dynami
c
ra
nge
5-m
odul
i set
{2
n
-1,2
n
+1,
2
2n
,
2
2n
+1,
2
2n-1
-1},
whe
r
e
n
i
s
a
n
y positive in
teger. Th
e efficient
reverse
conve
r
ter fo
r the moduli
set
is presented t
hat is ba
sed
on the NE
W CRT
s a
nd
th
e conve
r
se al
gorithm fo
r the four-moduli
set
{2
n
-1,2
n
+1,
2
2n
,2
2n
+1}. The
overall
spe
e
d
of the arith
m
et
ic unit of
RNS
syste
m
s ba
se
d on
the
moduli set S
1
and S is re
stricted to the
2
2n
+1 chann
e
l
, but the pro
posed 5-mod
u
li set reve
rse
conve
r
ter that
its dynamic range h
a
s raised to 8
n
-1 bit
s
.
Referen
ces
[1]
HL
Garn
er
.
T
h
e Resi
due N
u
mber S
y
stem.
IRE T
r
ansactio
n
s on Electro
n
i
c Co
mput
ers
. 195
9; 8(6):
140-
147.
[2]
NS Szab
o, RJ
T
anaka. R
e
sid
ue
Arithmetic
a
nd
its
Ap
plic
ati
ons to C
o
mp
uter
T
e
chnolo
g
y
.
Ne
w
Y
o
rk
:
McGra
w
-
Hil
l.1
967; 1-2
0
.
[3]
MA
Soderstran
d
. Resid
ue nu
mber s
y
stem a
r
ithmetic:
mod
e
rn ap
plic
ation
s
in digita
l sign
al process
i
n
g
.
Califor
ni
a: IEEE Press. 1986:
1-32.
[4]
B Parhami. Co
mputer
Arithme
t
ic:
Algorithms and H
a
rd
w
a
re
Desig
n
.
Oxford, U.K.:
Oxford
Univ
.
20
00.
[5]
PV
A
Mohan. R
e
sid
ue Num
ber
S
y
stems:
Algo
rith
ms and
Arc
h
itectures. Nor
w
e
ll, MA: Klu
w
er
, 2002.
[6]
G Dimauro, S Impdedovo, G Pirlo.
A
new
t
e
chn
i
q
ue for fast number comparison i
n
the residu
e
number s
y
stems.
IEEE T
r
ans.
Com
p
ut.
, 1993; 42: 60
8–6
1
2
.
[7]
M Lu, J Chia
n
g
.
A
novel div
i
sion a
l
g
o
rithm
for the resid
u
e
numb
e
r s
y
stem.
IEEE T
r
ans. Comput.,
199
2; 41: 102
6
–10
32.
[8] Y
W
ang.
New
Chin
ese re
ma
i
nder theor
e
m
s.
Proc. 32th
Asil
omar Conf. Si
gnals, S
y
stems, Comp
uters,
199
8; 1: 165–
1
71.
[9] A
Skavantzos.
An ef
fi
cient res
i
due to weighted conv
er
ter for
a new residue
num
b
er system
. Proc. 8th
Great Lakes S
y
mp. VLSI, Ne
w
Orle
ans, LA.
1998; 9: 18
5–
191.
[10]
J Mathe
w
, D R
adh
akrish
na
n,
T
Srikanthan.
F
a
st residue-to
-bin
ary convert
e
r architectur
e
s
. Proc. 4nd
Mid
w
est S
y
mp.
Circuits S
y
st. 199
9; 2: 1090
–
109
3.
[11]
A
Skavantzos,
T
Stouraitis.
Groupe
d-
mod
u
li resid
ue nu
mber systems for fast signal process
i
ng
.
Proc. Int. Sy
m
p
. Circuits S
y
st
. 1999; 3: 478
–
483.
[12]
B Cao, CH C
h
ang,
T
Srikant
han.
A
resid
u
e
-
to-bin
ar
y
c
onv
erter for ne
w
fi
ve-moduli set.
IEEE
T
r
ans.
Circuits System
s I.
2007; 54(
5): 1041
–1
049.
[13]
A
Skavantzos, M
Abda
lla
h. Impl
em
entatio
n is
sues of t
he t
w
o-lev
e
l resid
u
e
number s
y
st
e
m
w
i
th pa
irs
of conju
gate m
odu
li.
IEEE T
r
ans. Signal Proc
ess
. 1999; 4
7
(
3
): 826–
83
8.
[14]
AS Mola
hoss
e
i
n
i, K N
a
vi, O H
a
shem
ipo
u
r
,
et al. Ef
ficie
n
t re
verse co
nverte
r for the N
e
w
4
-modul
i set
s
{2
n
−
1, 2
n
, 2
n
+1,
2
2n+
1
−
1} and {2
n
−
1, 2
n
+1
, 2
2n
, 2
2n
+
1
} b
a
sed on the
Ne
w
CR
T
s
.
IEEE T
r
ans. on
Circuits and Sy
stem
s I
. 2010;
57(4): 82
3-8
3
5
[15]
SJ Piestrak.
A
hig
h
- spee
d re
alisati
on
of res
i
du
e to bin
a
r
y
s
y
stem conv
er
sion.
IEEE T
r
ans. Circuits
Syst. II,
Analog
Digit. Sign
al P
r
ocess
. 199
5; 42(6): 66
1–
663
,
[16]
Patel RA, Ben
a
issa M, Bouss
a
kta S. F
a
st
paralle
l-prefi
x
arc
h
itectures for modu
lo 2
n
−
1 A
dditi
on
w
i
th
a
singl
e repr
ese
n
tation of zer
o
.
IEEE T
r
ansactions on Com
puters.
2007; 56(
1
1
): 148
4-14
92
.
Evaluation Warning : The document was created with Spire.PDF for Python.