Internati
o
nal
Journal of Ele
c
trical
and Computer
Engineering
(IJE
CE)
V
o
l.
4, N
o
. 4
,
A
ugu
st
2014
, pp
. 52
0
~
53
1
I
S
SN
: 208
8-8
7
0
8
5
20
Jo
urn
a
l
h
o
me
pa
ge
: h
ttp
://iaesjo
u
r
na
l.com/
o
n
lin
e/ind
e
x.ph
p
/
IJECE
Cognitive Sensor Platform
Mar
k
Mc Der
m
ot
t
Department o
f
Electrical and Co
mputer Engi
n
eer
ing, Univ
ers
i
t
y
of
T
e
xa
s at
Au
st
in
Article Info
A
B
STRAC
T
Article histo
r
y:
Received
Ja
n 21, 2014
Rev
i
sed
May 31
2
014
Accepted
Jun 17, 2014
This pap
e
r des
c
ribes
a platfor
m
that
is used
to build
embedded sensor
s
y
s
t
em
s
for low energ
y
im
plant
a
ble
appli
c
ations
. One of
the k
e
y
characteristics of the platform is the
ability
to reason a
bout the environment
and cogni
tiv
el
ym
odif
y
th
e oper
a
tion
a
l p
a
ram
e
ters of the s
y
s
t
em
.
Addition
a
l
l
y
the pl
atform
pr
ovides to
abil
it
y
to
compose application
specific sensor
s
y
stems using a novel
computation
a
l
element that d
i
rectly
supports a
s
y
nchronous-dataflow (SDF) programming parad
i
gm.
Keyword:
Co
gn
itiv
e Sen
s
o
r
Platform
C
o
m
posabl
e
S
y
st
em
s
Dataflow Proc
essor
Syn
c
hro
nou
s D
a
taf
l
ow
Copyright ©
201
4 Institut
e
o
f
Ad
vanced
Engin
eer
ing and S
c
i
e
nce.
All rights re
se
rve
d
.
Co
rresp
ond
i
ng
Autho
r
:
Mar
k
Mc D
e
rm
o
t
t
Depa
rt
m
e
nt
of
El
ect
ri
cal
and
C
o
m
put
er E
ngi
neeri
n
g
,
Un
i
v
ersity of
Tex
a
s at
Au
stin
1
U
n
i
v
er
sity St
atio
n
C
0
803
,
Au
stin
, Tex
a
s
7
8
7
12-
021
4
Em
a
il: mcderm
ot@ece.utexas.edu
1.
INTRODUCTION
Th
e
n
e
x
t
step
in
th
e evo
l
u
tion
o
f
i
n
tellig
en
t
sen
s
o
r
s is toward
s cogn
itiv
e
sen
s
o
r
s. Cogn
itiv
e sen
s
or
p
l
atform
s h
a
ve th
e cap
a
b
ility to
reason
ab
ou
t th
ei
r
in
tern
al/ex
t
ern
a
l en
v
i
ron
m
en
tal co
nd
itio
ns an
d th
en
m
odi
fy
t
h
ei
r sy
st
em
behavi
or.
The re
sul
t
i
s
a sens
or
pl
at
fo
r
m
t
h
at
dy
nam
i
cal
l
y
opt
im
i
zes i
t
s
operat
i
o
n t
o
m
eet
t
h
e sy
st
em
m
e
tri
c
s. T
h
e
hi
era
r
chy
of
a
dva
nc
ed se
ns
or
sy
st
em
s i
s
general
l
y
desc
ri
be
d as
h
a
vi
n
g
t
h
ree
di
s
t
i
n
ct
lev
e
ls of cap
a
bilit
ies. Th
is classificatio
n
is
grap
h
i
cally illu
strated
i
n
Fi
g
u
re
1
.
Fi
gu
re
1.
C
o
m
put
at
i
o
nal
hi
er
archy
of
ad
va
n
ced se
ns
or
sy
st
em
s [1]
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
I
J
ECE Vo
l. 4
,
N
o
. 4
,
Au
gu
st 2
014
:
52
0
–
53
1
52
1
The first level
is a “s
m
a
rt” s
e
ns
or system
whe
r
e th
e sen
s
o
r
is ab
le to
iden
tify its p
u
r
po
se and
can
com
m
uni
cat
e inf
o
rm
at
i
on t
o
and f
r
o
m
ot
her devi
ces. T
h
e seco
nd
lev
e
l is g
e
n
e
rally classified
as “in
t
elli
g
e
n
t
”
an
d
th
is is achiev
e
d
b
y
ad
d
i
n
g
th
e ab
ility
to
recogn
ize,
in
terp
ret and
un
d
e
rstand
sen
s
o
r
stim
u
li. Th
e th
ird
lev
e
l o
f
cap
ab
i
lity is th
e co
g
n
itiv
e sen
s
o
r
t
h
at ad
d
s
reason
in
g
and
co
gn
itio
n
t
o
th
e in
tellig
en
t sen
s
o
r
sy
ste
m
,
allo
wing
it to
mak
e
d
ecisions b
a
sed
o
n
th
e
sen
s
o
r
stim
u
li an
d to
b
e
aware o
f
th
e env
i
ron
m
en
t th
at th
e
sen
s
o
r
is op
erating
i
n
[1
].
Each
lev
e
l of h
i
erarch
ical cap
ab
ility req
u
i
res a co
rrespo
n
d
i
ng
i
m
p
r
ov
emen
t in
co
m
p
u
t
atio
n
a
l and
en
erg
y
p
e
rfo
r
man
ce.
Wh
ile tran
sistor scalin
g
h
a
s
p
r
o
v
i
d
e
d
so
m
e
o
f
th
e
ad
d
ition
a
l co
mp
u
t
ation
a
l effi
cien
cy,
t
h
e ener
gy
su
p
p
l
y
for em
bed
d
ed se
ns
ors i
s
st
ay
i
ng vi
rt
u
a
l
l
y
const
a
nt
si
nc
e M
oore
’
s La
w d
o
es n
o
t
ap
pl
y
t
o
b
a
ttery tech
no
log
y
. Th
at said
,
Ko
om
ey, et al, obse
rve t
h
at comput
ational-e
ffi
ciency (m
eas
ure
d
i
n
C
o
m
put
at
i
ons/
J
oul
e
)
i
s
im
pro
v
i
n
g at
a sim
i
lar rat
e
t
o
M
o
o
r
e’s La
w [2]
.
Thi
s
m
a
y
be adeq
uat
e
f
o
r w
o
rkl
o
ads
wh
ose com
put
at
i
onal
effi
ci
en
cy
requi
rem
e
nt
s rem
a
i
n
c
onst
a
nt
fr
om
one generat
i
o
n t
o
t
h
e next
h
o
we
ve
r fo
r
next
ge
nerat
i
o
n se
nso
r
wo
r
k
l
o
ad
scom
put
at
i
onal
-
ef
fi
ci
ency
i
s
n
o
t
im
pro
v
i
n
g
at
a fast
e
n
ou
g
h
rat
e
i
f
t
r
a
n
si
st
o
r
scal
i
ng i
s
t
h
e onl
y
sou
r
ce
of i
m
provem
e
nt
. Fr
om
a syst
em
desi
gn p
e
rspect
i
v
e
,
o
p
t
i
m
a
l co
m
put
at
i
onal
-
efficiency is achieve
d by “im
p
edance
m
a
tching” the four dom
ains that co
m
p
rise a se
nsor desi
gn. The four
dom
ains include t
h
e algorithmic dom
ain,
t
h
e so
ft
wa
re
pr
o
g
ram
dom
ai
n, t
h
e
har
d
ware
m
i
cro-a
r
c
h
i
t
ect
ure, a
n
d
lastly
th
e
silico
n
techno
log
y
. Th
is
is d
e
scribed
b
y
th
e fo
llowing
p
s
eu
do
-eq
u
a
tion
:
Comput
a
t
i
o
na
l
Efi
c
i
e
n
cy
∝
#
∗
#
∗
∗
∗
∗J
o
u
l
e
s
(1
)
whe
r
e:
#
I
n
s
t =
nu
m
b
er of ex
ecu
ted
instru
ction
s
for a Task
TP =
Instruction T
r
ace Pa
rallelis
m
PP =
Process
o
r Parallelism
LL = Le
vels
of Logic
ns =
na
n
o
seco
nds
o
r
1/
f
r
eq
ue
ncy
Th
e
nu
m
b
er of in
st
ru
ction
s
p
e
r task
is th
e m
a
p
p
i
n
g
of t
h
e algorith
m
i
c
do
m
a
in
b
y
the software
co
m
p
iler in
to
sin
g
l
e
or m
u
lti
p
l
e so
ft
ware thread
s.
Id
eally
t
h
e
nu
m
b
er of
so
ft
ware thread
s is m
a
tch
e
d
to
th
e
num
ber o
f
pr
o
cessi
ng el
em
ent
s
or har
d
ware
t
h
rea
d
s. T
h
e
pr
ocessi
n
g
el
em
ent
(
s
)
wo
ul
d
be desi
g
n
e
d
t
o
pr
ovi
de
the optim
a
l energy-p
e
r
form
ance in
order t
o
accom
p
lish the task as
dete
rmined by level
s
of l
ogic
needed pe
r
clock cycle. It
can
be see
n
t
h
at so
lv
i
n
g th
e
eq
u
a
tion
“as is” resu
lts in
a valu
e of “1
Jou
l
es-n
s/Task
”,
wh
ich
i
ndi
cat
es a
n
i
d
e
a
l
im
pedance
m
a
t
c
h bet
w
ee
n
al
l
com
pone
nt
s o
f
t
h
e e
q
u
a
t
i
o
n
.
If
t
h
e
r
e i
s
a m
i
sm
at
ch be
t
w
een
an
y of t
h
e
d
o
m
ain
s
th
e co
m
p
utatio
n
a
l efficien
cy of t
h
e m
a
c
h
in
e
will b
e
non
-op
t
i
m
al.
It is gene
rally accepted that general-purpos
e
com
puting
platform
s
do
not have the ene
r
gy-
perform
a
nce characteristics n
eeded
fo
r l
o
w
ener
gy
sens
or
sy
st
em
s. Hard
-
c
ode
d l
o
gi
c w
oul
d p
r
o
v
i
d
e t
h
e m
o
st
o
p
tim
al co
m
p
utatio
n
a
l-efficien
cy bu
t at t
h
e
ex
p
e
n
s
e
of
reprog
ramm
ab
ilit
y. Th
e co
m
p
u
t
atio
n
a
l elem
en
t u
s
ed
i
n
t
h
i
s
pl
at
fo
r
m
i
s
im
pl
em
ent
e
d u
s
i
n
g an
ener
gy
effi
ci
e
n
t
pr
o
g
ram
m
abl
e
m
i
croco
d
e
d
en
gi
ne
wi
t
h
a si
ngl
e
cycle d
a
tap
a
th. Th
e lev
e
ls of lo
g
i
c an
d
the ratio
o
f
co
n
t
ro
l log
i
c to
d
a
tap
a
th
log
i
c are o
p
tim
ized
fo
r low
ener
gy
a
ppl
i
cat
i
ons
.
The s
o
ft
ware
pr
o
g
ram
m
i
ng
para
di
gm
i
s
al
so
o
p
t
i
m
i
zed f
o
r
sen
s
o
r
a
ppl
i
cat
i
ons.
Se
nso
r
sy
st
em
s are
react
i
v
e an
d c
a
n be i
m
pl
em
ent
e
d
usi
n
g a
num
ber o
f
d
i
ffere
nt
react
i
v
e p
r
o
g
r
am
m
i
ng m
odel
s
i
n
c
l
udi
n
g
Dy
nam
i
c Dataflo
w
(
D
DF
),
S
y
nch
r
o
n
ous
Da
taflow
(S
DF
),
Discrete Ev
en
ts (DE), Petri
nets and
Kh
an Pro
cess
Net
w
or
ks (
K
P
N
) [
3
]
.
T
h
i
s
p
l
at
form
uses Sy
nch
r
o
n
ous
Dataflow
(SDF) as it is a special case of da
taflow
m
odel
i
ng w
h
e
r
e t
h
e
fl
o
w
o
f
cont
rol
i
s
pre
d
i
c
t
a
bl
e at
co
m
p
il
at
i
on t
i
m
e
[4]
.
T
h
e m
i
croco
d
e
d
com
p
u
t
at
i
onal
el
em
ent
used
o
n
t
h
i
s
pl
at
f
o
rm
i
s
desi
g
n
e
d
t
o
di
rect
l
y
su
pp
o
r
t
t
h
e S
D
F e
v
e
n
t
-
dri
v
en
p
r
og
ra
m
m
i
ng para
di
g
m
.
2.
REQU
IRE
M
ENTS F
O
R
A
CO
GNITI
V
E
SENS
OR
PL
ATFO
RM
(
C
SP)
Thi
s
pl
at
fo
rm
i
s
desi
gn
ed
be
use
d
i
n
de
epl
y
em
bedde
d a
ppl
i
cat
i
o
ns
suc
h
as
m
e
di
cal
im
pl
ant
s
,
st
ruct
u
r
al
i
m
plant
s
a
n
d
rem
o
t
e
sensi
n
g
.
T
h
e
key
fi
gu
re
of
m
e
ri
t
for
t
h
i
s
c
l
ass o
f
em
bedd
ed se
ns
ors
i
s
e
n
er
gy
-
p
e
rform
a
n
ce/v
o
l
u
m
e wh
ere
b
a
ttery vo
lu
me is th
e li
miti
ng
factor as it determ
ines the num
b
er of joules
available for
syste
m
operati
on. The a
d
dition
of c
o
gnitive ca
pabilities is necessary
for t
h
ese ty
pes
of
u
n
a
ttend
ed
app
licatio
n
s
as it is g
e
n
e
rally no
t feasi
b
le
to ro
u
tin
ely
rep
l
ace th
e
b
a
ttery
or sen
s
o
r
(s) in
t
h
ese
appl
i
cat
i
o
ns. C
o
g
n
i
t
i
on i
n
t
h
e
cont
ext
of a s
e
ns
or pl
at
f
o
rm
i
s
defi
ne
d as t
h
e “p
roces
s o
f
kn
owi
n
g
,
i
n
cl
udi
n
g
aspects of
a
w
a
r
enes
s, perce
p
t
i
on,
r
eas
oning, and
judgm
e
nt” [1]. Figure
2
shows
conce
p
tually how t
h
e
process
o
f
kno
wi
n
g
app
lies to
a cogn
itiv
e sen
s
or syst
e
m
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
C
o
g
n
i
t
i
ve Se
ns
or Pl
at
f
o
r
m
(
M
ark Mc
De
rm
o
t
t)
52
2
Fig
u
re
2
.
“Pro
cess of
Knowing
”
in a co
gn
itiv
e sen
s
o
r
system
Th
is “pro
cess o
f
k
nowing
”
driv
es th
e fo
llowing
b
a
selin
e
fun
c
tion
a
l requirem
e
n
t
s o
f
th
e CSP. Th
ese
in
clu
d
e
th
e ab
i
lity to
:
Perfo
rm
sel
f-di
a
gnost
i
c
s and s
e
l
f-cal
i
b
rat
i
on.
Co
m
p
ensate for sy
stem
atic errors, sy
stem
drift and ra
n
dom
erro
rs pr
od
uced
by
sy
st
em
param
e
t
r
i
c
changes
such as sens
or
agi
ng,
bat
t
e
ry
agi
ng.
Fuse dat
a
fr
om
hom
ogeneo
u
s
and het
e
roge
neous
sens
ors.
Detect and repair corrupted
data
Reason about the state of the syste
m
and pe
rform
n
eed
ed
serv
ices to
m
a
in
ta
i
n
op
ti
m
a
l sys
t
e
m
perf
orm
a
nce.
An
ticip
ate p
o
t
en
tial sys
t
e
m
at
i
c
ch
an
g
e
s a
nd
m
odi
fy
operat
i
onal
beha
vi
or.
Tran
smit/rec
ei
v
e
in
form
a
tio
n
to
/fro
m
o
t
h
e
r dev
i
ces.
Th
e CSP
fun
c
t
i
o
n
a
lity requ
iremen
ts d
e
scrib
e
d
ab
ov
e
driv
es
a nu
m
b
er of
k
e
y arch
itectural
featu
r
es in
cl
u
d
in
g
:
Debu
g an
d
di
agn
o
si
s
Tim
e
St
am
pi
ng
Ad
ap
tiv
e cap
ab
ilit
ies in
clu
d
i
n
g
:
Co
nfigu
r
ab
le
data lo
ok
up
capab
ility
R
econ
f
i
g
ura
b
l
e
eve
n
t
-
dri
v
en
pr
o
g
ram
m
i
ng
Dy
nam
i
c sam
p
l
i
ng a
n
d
f
r
eq
ue
ncy
scal
i
n
g
Dy
nam
i
c dat
a
preci
si
o
n
Fu
zzy lo
g
i
c cap
ab
ilit
y
Data fu
sio
n
cap
a
b
ilit
y
2
.
1
Debug
a
nd
Diag
no
stic
Ca
pa
bility
Th
e CSP
p
r
ovid
e
s a co
m
p
u
t
atio
n
a
l (d
ig
ital) d
i
agno
stic m
o
d
e
th
at u
til
izes au
x
iliary ch
ann
e
ls t
o
confirm
that the prim
ary ch
annels
a
r
e pe
rform
i
ng as expected.
Injecti
n
g
calibration
to
k
e
n
s
in
to
t
h
e SDF
network a
n
d a
n
alyzing t
h
e
response
to c
o
nfirm
com
putational acc
urac
y accom
p
lishes the
diagnostic. T
h
e
so
urce of t
h
e calib
ratio
n tok
e
n
s
is con
t
ro
lled
b
y
th
e Ch
ann
e
l
Nod
e
s i
n
respo
n
s
e to a co
n
t
ro
l
sign
al fro
m
th
e
Debug Unit. A wide range of
diagno
stics can be accom
p
lished
using “di
g
ital”
tokens. T
h
e discrete val
u
e and
te
m
p
o
r
al
respon
se o
f
th
e SDF
n
e
two
r
k
can
be
an
alyzed
b
y
t
h
e DPE u
tilizin
g
th
is d
i
agno
stic
m
o
d
e
. In
add
itio
n
to
th
e “d
ig
ital” d
i
ag
no
stics, the Debu
g
Un
it can
in
j
ect “an
alo
g
”
v
a
l
u
es in
to
th
e sen
s
o
r
read
ou
t circu
itry
.
The
val
u
es a
r
e ge
neral
l
y
l
i
m
i
t
e
d t
o
t
hose t
h
at
can be easi
l
y
im
pl
em
ent
e
d usi
ng
vol
t
a
ge re
fere
nces
t
h
at
are
in
sen
s
itiv
e
t
o
ag
ing
,
po
wer
sup
p
l
y v
a
riation
s
and
d
e
v
i
ce v
a
riatio
n
s
.
Calibration of the sensor readout circuit
r
y
is acco
m
p
lished usi
ng
a programma
ble conte
n
t-
ad
dressab
l
e loo
kup
tab
l
e
(C
LT) in th
e senso
r
d
a
ta-con
d
i
t
i
o
n
i
ng
elem
en
t. Th
e CLT is i
n
itialized
at reset with
the stored cali
b
ration
data from
an
ext
e
r
n
a
l
m
e
m
o
ry
. Du
ri
n
g
sy
st
em
operat
i
o
n, t
h
e
C
LT val
u
e
s
c
a
n b
e
up
dat
e
d t
o
ad
apt
t
o
en
vi
ro
n
m
ent
a
l
change
s i
n
t
h
e sens
or t
r
a
n
s
duce
r
.
The m
odi
fi
ed C
LT dat
a
can b
e
tran
sferred
to
th
e ex
tern
al
m
e
m
o
ry so
th
at s
u
b
s
equ
e
n
t
reset o
p
e
ration
s
load
th
e n
e
w calib
ration
d
a
ta in
to
th
e
CLT.
2.
2 T
i
me S
t
a
m
pi
ng
The t
i
m
e
st
amp fu
nct
i
o
n use
s
a t
i
m
i
ng refe
rence t
o
gene
r
a
t
e
a uni
que
v
a
l
u
e t
h
at
i
s
t
a
gge
d t
o
t
h
e
t
oke
n dat
a
pr
o
duce
d
by
A/
D
con
v
e
r
t
e
r. T
h
i
s
t
i
m
i
ng re
fer
e
nce can
be
g
e
nerat
e
d i
n
t
e
r
n
al
l
y
i
n
t
h
e C
SP
o
r
deri
ved
fr
om
an ext
e
r
n
al
net
w
o
r
k sy
nch
r
on
i
zat
i
on si
gnal
[5
]. If the ti
m
i
n
g
referen
ce is d
e
ri
v
e
d
in
tern
ally th
e
external receiver m
u
st synchronize to
the int
e
rnal tim
i
ng re
fere
nce by algo
rithm
i
c means
[6]
[7]. T
h
e num
b
er
o
f
sam
p
les is d
e
term
in
ed
b
y
th
e op
eratio
n
t
h
at th
e CSP
is in
tend
ed to
p
e
rfo
r
m
.
For ex
am
p
l
e if fiv
e
sam
p
le
s
are b
e
ing
av
erag
ed
to
a single d
a
tu
m th
en
th
e ti
min
g
wind
ow is v
a
lid
fo
r fi
v
e
sam
p
le
s an
d
a sing
le ti
me
sta
m
p
v
a
lu
e is
issu
ed. Th
e time sta
m
p
v
a
lu
e is in
crem
en
t
e
d
fo
r ev
ery sam
p
le an
d
o
n
e
o
f
the fiv
e
time sta
m
p
val
u
es i
s
use
d
t
o
t
a
g t
h
e dat
a
depe
ndi
ng o
n
t
h
e al
go
ri
t
h
m
bei
ng per
f
o
r
m
e
d. In t
h
i
s
exam
pl
e t
h
e t
h
ird t
i
m
e
sta
m
p
v
a
lu
e cou
l
d
b
e
u
s
ed
when
av
erag
i
n
g
fi
v
e
sam
p
les. Th
is tech
n
i
qu
e p
r
o
v
i
d
e
s a b
u
ilt in
ti
m
e
referen
c
e for
al
l
dat
a
t
h
at
i
s
bei
n
g p
r
oce
sse
d by
C
SP. T
h
e t
i
m
e
st
am
p val
u
e can be t
r
a
n
sm
i
t
t
e
d al
ong
t
h
e out
put
t
o
k
e
n dat
a
via the c
o
mm
unications
elemen
t fo
r ex
ter
n
al
pr
o
cessing
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
I
J
ECE Vo
l. 4
,
N
o
. 4
,
Au
gu
st 2
014
:
52
0
–
53
1
52
3
2
.
3
Ada
p
ti
v
e
Capabilities
The CSP dete
cts environm
ental changes by tracking f
act
ors s
u
ch as t
h
e rat
e
of cha
n
ge of se
ns
or
d
a
ta, co
m
p
u
t
atio
n
a
l d
a
ta errors, b
a
ttery vo
ltag
e
, tem
p
er
ature, etc. The CSP can ad
ap
t to
th
ese ch
an
g
e
s
u
s
ing
p
r
e-d
e
fin
e
d
ru
l
e
s. On
e of th
e
m
o
st u
s
efu
l
adap
tatio
n
techn
i
q
u
e
s u
tilizes fuzzy lo
g
i
c d
ecisio
n
-m
ak
in
g
[8
]. Th
is
i
s
descri
be
d b
e
l
o
w i
n
t
h
e n
e
xt
sect
i
on. T
h
e C
SP has a
cont
ent
ad
d
r
e
ss l
o
o
k
u
p
t
a
bl
e (C
LT) t
h
at
can be
dy
nam
i
cal
ly
repr
o
g
ram
m
ed
du
ri
n
g
sy
st
em
ope
rat
i
o
n. T
h
e C
LT i
s
use
d
t
o
cal
i
b
rat
e
sens
or
dat
a
,
p
r
o
v
i
d
e
com
p
lex logic
functions a
n
d for fuzzy logic
ope
rations s
u
c
h
as
the
de
fuz
z
ify step
where the a
n
tecede
n
ts are
m
a
pped
t
o
det
e
rm
i
n
i
s
ti
c out
pu
t
val
u
es
.
Th
e CSP can
d
y
n
a
m
i
call
y
reco
nfigu
r
e its prog
ra
m
sequ
en
ci
n
g
, wh
ich
is req
u
i
red for
co
gn
itive
sy
st
em
s t
h
at
need t
o
dy
nam
i
cal
l
y
m
odi
fy
t
h
ei
r al
g
o
r
i
t
h
m
s
base
d
on
o
p
e
rat
i
onal
c
o
ndi
t
i
ons
[9]
[
1
0]
.
Thi
s
cap
ab
ility in
th
e CSP is ach
iev
e
d
b
y
u
s
i
n
g
an
acto
r
/ev
e
n
t
qu
eu
e. The relativ
e o
r
d
e
r of the h
o
w th
e
o
p
e
ratio
ns
are
pr
ocesse
d
can
be
dy
nam
i
cal
l
y
chan
ged
by
ree
n
t
e
ri
ng
o
r
re
o
r
de
ri
n
g
t
h
e act
or
s i
n
t
h
e
que
ue.
T
h
e act
ors
ca
n
b
e
in
serted
in
t
o
th
e qu
eu
e in
eith
er FIFO
o
r
LIFO
orde
r. T
h
e actors are e
x
ecute
d
from
the “Top-of-Stack” as
sho
w
n
bel
o
w i
n
Fi
gu
re
3. Eac
h
o
p
e
r
at
i
o
n
i
s
m
a
pped
t
o
a m
i
croc
ode
routine that te
rm
inates usi
ng a
“
W
a
it-fo
r-
Eve
n
t
”
i
n
st
r
u
c
t
i
on. I
n
t
e
r
r
upt
s and
ot
he
r a
s
y
n
ch
ro
n
o
u
s
event
s
ca
n al
so
ent
e
r o
p
e
r
at
i
ons t
o
t
h
e
qu
eue by
in
serting
th
em
in
to
the ex
ecu
tio
n stream
. Th
ese ev
en
ts
are s
quas
h
e
d
from
the actor-que
ue a
f
ter t
h
ey are
ex
ecu
ted
.
Th
e acto
r
-qu
e
u
e
i
s
circu
l
ar wh
i
c
h
allo
ws it to
ex
ecu
t
e contin
u
o
u
s
ly un
til a b
r
eak
con
d
itio
n
is
en
coun
tered
.
Typ
i
cally a b
r
eak
co
nd
itio
n occu
rs wh
en
n
e
w to
k
e
n
s
d
a
ta are n
e
ed
ed.
Fi
gu
re
3.
Act
o
r
an
d E
v
e
n
t
Q
u
eue
The De
bu
g U
n
i
t
prel
oa
ds t
h
e act
or-
que
ue
wi
t
h
a presc
r
i
b
ed se
que
nce
of o
p
e
r
at
i
o
n
s
vi
a t
h
e JTA
G
scan c
h
ai
n
.
As
t
h
e C
S
P
bec
o
m
e
s ope
rat
i
o
n
a
l
l
y
and c
o
ndi
t
i
onal
l
y
awar
e
o
f
i
t
s
e
nvi
ro
n
m
ent
t
h
e
ope
r
a
t
i
on-
que
ue ca
n
be
s
a
ved
t
o
e
x
t
e
r
n
a
l
m
e
m
o
ry
so t
h
at
t
h
e ne
w
state can
be
reloa
d
ed
during t
h
e
next reset cycle.
The dat
a
pat
h
i
n
t
h
e C
SP i
s
d
e
si
gne
d t
o
u
s
e
si
gne
d sat
u
rat
i
ng ari
t
h
m
e
t
i
c
. The dat
a
p
r
ec
i
s
i
on can
be
dy
nam
i
cal
ly
modi
fi
ed t
o
sa
ve
po
wer
by
cha
ngi
ng t
h
e sat
u
r
a
t
i
on l
i
m
i
t
s
an
d scal
i
ng t
h
e d
a
t
a
t
okens as
n
eeded
.
Ad
di
t
i
onal
l
y
t
h
e C
S
P
ca
n m
odi
fy
t
h
e sam
p
l
i
ng
rat
e
o
f
t
h
e
sens
or
d
a
t
a
i
f
t
h
e rat
e
of
c
h
a
nge
o
f
t
h
e i
n
c
o
m
i
ng
d
a
ta is low.
2
.
4
Fuzzy
Lo
gic Capability
The C
SP c
o
ntains a
fuzzy logic e
ngi
ne to a
n
alyze
sensor
data a
n
d m
a
ke system
a
tic adjustm
e
nts to
t
h
e o
p
erat
i
o
n
of t
h
e
pl
at
fo
r
m
pl
us pr
ovi
d
e
speci
al
i
zed
fu
nct
i
o
ns l
i
k
e
dat
a
f
u
si
n
g
(
d
escri
b
ed
i
n
t
h
e
next
sect
i
on)
. T
h
i
s
engi
ne i
s
i
m
plem
ent
e
d usi
n
g
a com
b
i
n
a
t
i
o
n
of
speci
al
i
zed
har
d
ware
fu
n
c
t
i
ons a
n
d
m
i
croc
o
d
e
ro
ut
i
n
es.
The
speci
al
i
zed ha
rd
ware c
o
nsi
s
t
s
of l
ogi
c t
o
per
f
o
r
m
m
i
ni
m
u
m
,
m
a
xim
u
m
and t
a
bl
e l
o
o
k
u
p
fu
nct
i
o
ns. T
h
e
m
i
crocode e
n
gi
ne pe
rf
o
r
m
s
t
h
e
m
e
m
b
ershi
p
, r
u
l
e
eval
uat
i
on a
nd
wei
g
ht
ed ave
r
age f
u
n
c
t
i
ons
[1
1]
.
As m
e
n
tio
n
e
d
ab
ov
e t
h
e CSP can
co
n
t
ro
l energ
y
u
s
ag
e b
y
co
n
t
ro
lling
th
e
sam
p
lin
g
rate
o
f
t
h
e sen
s
o
r
dat
a
an
d c
o
nt
r
o
l
l
i
ng t
h
e cl
oc
k
fre
que
ncy
.
A f
u
zzy
l
ogi
c
base
d al
g
o
ri
t
h
m
i
s
used
t
o
d
e
t
e
rm
i
n
e t
h
e s
a
m
p
l
i
n
g
r
a
te b
y
an
alyzin
g th
e ch
ang
e
an
d th
e
r
a
te of
ch
ang
e
o
f
th
e i
n
pu
t
d
a
ta. Figur
e 4 b
e
l
o
w
show
s th
e f
l
ow
d
i
ag
r
a
m
for the
fuzzy logic e
n
gine. T
h
e fi
rst step
perform
s
a
m
e
m
b
ership e
v
aluation on
t
h
e inputs.
T
h
ere
are
two
i
n
p
u
t
s
;
o
n
e i
s
t
h
e a
b
sol
u
t
e
val
u
e
of t
h
e se
ns
or
dat
a
c
h
an
ge
an
d t
h
e sec
o
n
d
i
n
p
u
t
bei
n
g t
h
e rat
e
of
cha
n
ge
of
dat
a
cha
n
ge. T
h
e sec
o
n
d
st
e
p
per
f
o
r
m
s
t
h
e eval
uat
i
o
n
of
the ru
les t
h
at d
e
t
e
rm
in
e th
e en
erg
y
lev
e
ls. Th
e th
ird
st
ep co
n
v
ert
s
t
h
e en
er
gy
l
e
ve
l
s
t
o
co
nt
r
o
l
si
gnal
s
t
o
d
r
i
v
e t
h
e sam
p
l
i
ng ra
t
e
and/
or t
h
e cl
ock
fr
eq
ue
ncy
of t
h
e
CSP.
Que
u
e
Co
nt
roll
e
r
Ev
e
n
t
#1
Re
ci
r
c
u
l
a
t
e
d
Ac
to
rs
Ev
e
n
t
#2
Ev
en
t
#N
Rea
d
TO
S
Wri
t
e
BOS
AC
T
O
R
&
EVE
N
T
QUEUE
Wr
it
e
TO
S
Res
u
lt
Bu
s
μ
OP
AD
DR
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
C
o
g
n
i
t
i
ve Se
ns
or Pl
at
f
o
r
m
(
M
ark Mc
De
rm
o
t
t)
52
4
Fi
gu
re 4.
F
u
zz
y
l
ogi
c fl
o
w
di
agram
Fi
gu
re
5
bel
o
w s
h
o
w
s a
t
a
b
l
e of t
h
e
out
pu
t
s
fr
om
t
h
e rul
e
eval
uat
i
o
n
o
f
t
h
e e
n
e
r
gy
f
unct
i
o
n.
Th
e
l
o
we
r ene
r
gy
o
p
erat
i
o
ns
occ
u
r w
h
e
n
t
h
e i
n
p
u
t
dat
a
i
s
not
c
h
an
gi
n
g
very
r
a
pi
dl
y
an
d t
h
e
rat
e
o
f
cha
n
ge
of t
h
e
chan
ge i
s
al
so
not
c
h
a
ngi
ng
v
e
ry
ra
pi
dl
y
.
C
o
nve
rsel
y
i
f
t
h
e
i
n
p
u
t
dat
a
i
s
c
h
an
gi
n
g
ra
pi
dl
y
a hi
g
h
e
r
e
n
er
gy
i
s
requ
ired
to process th
e
d
a
ta.
Fi
gu
re
5.
Ene
r
gy
us
age
r
u
l
e
e
v
al
uat
i
o
n t
a
bl
e
2
.
5
Data Fusi
on Ca
pa
bility
Data fu
si
o
n
is im
p
o
r
tan
t
for d
a
ta
reliab
ility and
robu
stness,
d
a
ta co
mp
ressi
on
o
p
e
ratio
n
s
and
for
com
posi
n
g c
o
m
p
lim
ent
a
ry
or
spect
ral
da
t
a
[1
2]
.
The
C
SP s
u
p
p
o
rt
s
l
o
w
-
e
n
er
gy
dat
a
f
u
si
on
u
s
i
n
g
a
com
b
i
n
at
i
on
o
f
m
i
croco
d
e
r
out
i
n
e
s
a
n
d
t
h
e f
u
zzy
l
o
gi
c
engi
ne t
o
per
f
o
rm
t
h
e
dat
a
f
u
si
o
n
o
p
erat
i
o
ns
[1
3]
[14]. This low-ene
r
gy approach is
p
r
eferred
ov
er m
a
th
e
m
atical
ly
in
ten
s
iv
e alg
o
rith
ms u
s
ing
least s
q
u
a
re-
base
d estim
a
tion m
e
thods s
u
ch as Kalm
an
Filtering [15]
or
proba
bilistic
m
e
thods s
u
c
h
as Bayesian analysis
[16]. For those
sens
or applications
wh
e
r
e a
m
o
re accurate
data fusion al
gor
ithm
is needed a
h
ybrid of Kalm
an
filters an
d
fu
zzy lo
g
i
c can
b
e
i
m
p
l
e
m
en
ted
with
min
i
m
a
l
ad
d
ition
a
l lo
g
i
c [1
7
]
. Th
e limita
tio
n
o
f
th
i
s
h
y
b
r
i
d
app
r
oach
i
s
t
h
e
dat
a
preci
si
on
pr
o
v
i
d
e
d
by
the DPE a
n
d the
increase
d
e
n
ergy us
age
.
Data fro
m
eith
er sing
le
o
r
mu
ltip
le sen
s
ors can
b
e
fu
sed
in
to
a co
m
p
o
s
i
t
e d
a
ta stream
. Th
e fu
sed
d
a
ta co
n
t
ains m
o
re in
fo
rm
at
io
n
th
an
th
e o
r
i
g
in
al in
pu
ts an
d
is u
s
ed
eith
er lo
cally
in
th
e CSP an
d
/
or
transm
itted to a receiving
node for
furthe
r
processi
ng. The fu
zzy
data fusion al
gorith
m involves a
g
gregati
ng
d
a
ta fro
m
th
e in
pu
t sen
s
ors
an
d
u
tilizin
g
pred
ictiv
e
d
a
ta fro
m
p
a
st ag
g
r
eg
atio
n
t
o
g
e
nerate fused
d
a
t
a
an
d
opt
i
o
nal
si
deba
nd
dat
a
a
s
s
h
o
w
n
bel
o
w
i
n
Fi
gu
re
6.
DE
C
R
E
A
SIN
G
NONE
IN
CR
EA
S
I
N
G
CH
A
N
GE
RA
TE
of
CH
AN
GE
NE
G_B
I
G
NE
G_SM
A
L
L
PO
S
_
B
I
G
P
O
S_SMA
L
L
NO
N
E
HE:
Hi
gh
En
erg
y
ME
:
Med
i
u
m
en
e
r
g
y
LE
:
Lo
w
En
er
gy
HE
HE
ME
ME
ME
HE
HE
HE
ME
ME
LE
ME
ME
HE
HE
ME
LE
LE
LE
ME
HE
ME
LE
LE
LE
LE
LE
ME
HE
ME
LE
LE
LE
ME
HE
HE
ME
ME
LE
ME
ME
HE
HE
HE
ME
ME
ME
HE
HE
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
I
J
ECE Vo
l. 4
,
N
o
. 4
,
Au
gu
st 2
014
:
52
0
–
53
1
52
5
Fi
gu
re
6.
F
u
zz
y
dat
a
f
u
si
o
n
In th
is ex
am
p
l
e, th
e d
a
ta from
two asymmetrical sens
ors
are
fuse
d
tog
e
th
er to produ
ce a co
m
p
o
s
ite
signal that
has
the key c
h
aract
eristics from
e
ach se
ns
or
. The first
o
p
e
ration
invo
lv
es
d
e
term
in
in
g
th
e abso
lu
te
val
u
e
o
f
t
h
e c
h
an
ge
an
d
rat
e
o
f
c
h
a
nge
(
R
OC
)
fo
r t
h
e
t
w
o
se
ns
ors
.
The
sec
o
n
d
ope
rat
i
o
n
per
f
o
rm
s a
m
e
m
b
ershi
p
e
v
al
uat
i
o
n an
d
seri
es o
f
fu
zzy
rul
e
eval
uat
i
o
ns t
o
p
r
od
uce
a wei
g
ht
i
ng
fa
ct
or an
d a se
ns
or
dri
f
t
v
a
lu
e.
Th
e
th
i
r
d
o
p
e
ration
u
s
es
th
e weigh
tin
g
and
sen
s
or d
r
ift
i
n
fo
rm
ati
o
n
to
produ
ce th
e
fu
sed
d
a
ta an
d
a
q
u
a
lity tag
.
Th
e qu
ality tag
is sid
e
b
a
nd
data th
at can
b
e
u
s
ed
b
y
th
e CSP to
ad
ap
t
to
sen
s
o
r
drift, d
a
ta
sam
p
ling issue
s
, etc.
3.
C
S
P AR
CHITEC
TUR
E
Th
e CSP is an ev
en
t d
r
i
v
en
syn
c
hr
ono
us data f
l
o
w
arc
h
itecture. It is “co
m
p
o
s
ed
” b
y
in
stan
tiatin
g
functional elements that are conn
ect
ed
v
i
a chan
nel
s
. The f
unct
i
ona
l
el
em
ent
s
provi
de key
o
p
e
rat
i
onal
services
com
m
only called “actors”
in a
dataflow
syst
em
. In the
current im
ple
m
entation t
h
e c
h
annels a
r
e
m
odeled as bounded FIFOs
.
The
in
form
a
tion
datum
that is comm
unicat
ed via the
cha
nnel i
n
terface is
refe
rre
d t
o
as
a
“t
ok
en”
.
Fi
gu
r
e
7
bel
o
w
sh
o
w
s t
h
e
fi
ve
bas
i
c fu
nct
i
o
nal
el
em
ent
s
t
h
at
are
use
d
t
o
c
o
m
pose a
CSP:
1. Sens
or El
em
ent
- Trans
duc
er and R
ead
out
C
i
rcui
t
r
y
2. Sens
or
Dat
a
C
ondi
t
i
oni
ng
(
S
DC
) El
em
ent
3. Dat
a
fl
ow
Pr
ocessi
ng El
em
ent
(DPE)
4. C
o
m
m
uni
cati
ons El
em
ent
5. De
bu
g El
em
ent
Fi
gu
re 7.
C
SP B
l
ock Di
ag
ram
Th
e ou
tpu
t
from th
e read
ou
t
circu
itry in
th
e Sen
s
o
r
Elem
e
n
t will g
e
n
e
rally b
e
an
an
al
og
sign
al th
at
will requ
ire some ad
d
ition
a
l
an
alog
p
r
o
cessin
g
su
ch
as
filterin
g
, am
p
lificatio
n
and
co
nv
ersi
o
n
to a
d
i
g
ital
rep
r
ese
n
t
a
t
i
o
n
usi
ng a
n
an
al
og
-t
o-
di
gi
t
a
l
con
v
ert
e
r (
A
DC
). T
h
i
s
ad
di
t
i
onal
p
r
oce
ssi
ng i
s
d
one
i
n
a
pre
p
r
o
cessi
ng
uni
t
i
n
t
h
e S
D
C
el
em
ent
.
The C
SP
m
a
y
opt
i
onal
l
y
cont
ai
n o
n
e o
r
m
o
re dat
a
fl
o
w
-
p
r
o
c
e
ssi
n
g
ele
m
ents (DPE
) that
process
the
da
ta from
the SDC a
n
d c
o
mm
unicate the
output data
via the COM el
e
m
ent
to a receiving
device.
In addi
tion to
these four elem
ents,
an optional debug elem
ent c
a
n be
used to
debug
fu
nctio
nal failu
res a
n
d
rec
o
nfi
g
u
r
e t
h
e CSP
d
u
ri
ng
n
o
rm
al operatio
ns.
Thi
s
pl
at
f
o
rm
i
s
desi
gne
d t
o
sup
p
o
r
t
a rea
s
on
abl
y
wi
de
vari
et
y
of se
n
s
i
ng t
ech
ni
q
u
e
s
i
n
cl
udi
ng
,
voltage
, resistive, capacitive, induc
tive,
opti
cal,
m
a
gnetic, force and acce
leration. Ty
pical e
m
bedde
d sensors
would i
n
clude
strain gauges,
piezoel
ect
ri
c d
e
vi
ces, ph
ot
ot
r
a
nsi
s
t
o
rs, hal
l
-
effect
devi
ces,
t
h
erm
o
-co
upl
e
s
,
i
o
n-
sensitive tra
n
si
stors, ca
pacitiv
e displacem
ent
de
vices, a
n
d
bi
o-se
nsing
de
vices [18].
CO
M
Element
Se
ns
o
r
El
e
m
e
nt(s
)
Data
flo
w
P
r
o
cessin
g
El
em
ent(s
)
T
r
ansd
ucer(
s
)
Read
out
C
i
r
c
ui
tr
y
Se
nso
r
Data
Co
nditi
on
ing
El
e
m
en
t
Fe
ed
b
a
c
k
& Co
nt
r
o
l
D
e
bu
g
El
e
m
e
nt
(O
pti
o
n
a
l
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
C
o
g
n
i
t
i
ve Se
ns
or Pl
at
f
o
r
m
(
M
ark Mc
De
rm
o
t
t)
52
6
The se
ns
or
dat
a
-c
on
di
t
i
oni
ng
el
em
ent
pr
o
v
i
d
es a
b
r
oa
d
ran
g
e
o
f
dat
a
c
o
n
d
i
t
i
oni
n
g
a
n
d
tran
sform
a
t
i
o
n
serv
ices. Th
ese serv
ices in
clu
d
e
sign
al
con
d
ition
i
ng
, sign
al con
v
e
rsion, d
e
tectio
n
functio
n
s
,
dat
a
-re
d
u
ct
i
o
n
and
dat
a
-f
usi
o
n.T
h
e
hi
g
h
-l
e
v
el
bl
oc
k
di
ag
ra
m
for t
h
e
SDC
el
em
ent
i
s
sh
o
w
n
bel
o
w
i
n
F
i
gu
re
8.
The
SDC contains t
h
ree
ba
sic units:
1
.
Prep
ro
cessi
ng
u
n
it (PPU)
- in
cl
u
d
e
s filters, A/
D co
nv
erters, etc.
2.
Fu
nct
i
o
nal
s
e
rvi
ces
u
n
i
t
(F
SU)
-
pe
rf
orm
s
dat
a
c
o
n
d
i
t
i
o
n
i
ng
ser
v
i
ces.
3. C
h
a
nnel
n
o
d
e
(s)
Fi
gu
re
8.
Se
ns
or
Dat
a
C
o
n
d
i
t
i
oni
n
g
(S
DC
)
El
em
ent
B
l
ock
Di
ag
ram
A typical preprocessi
ng uni
t
contains som
e
co
m
b
ination
of the followi
ng c
o
m
ponents: filters
,
a
m
p
lifiers, an
alo
g
-to
-
d
i
g
ital co
nv
erters (ADC), sam
p
le
-h
o
l
d
circu
its an
d
an
alog
m
u
ltip
le
x
o
rs as sho
w
n
b
e
low
in
Figur
e
9
.
Fi
gu
re
9.
Ty
pi
cal
con
f
i
g
ur
at
i
o
n
o
f
a
PP
U
The F
S
U ca
n
be i
m
pl
em
ent
e
d u
s
i
n
g a sy
nt
hesi
zed
ha
rd
-c
ode
d l
o
gi
c
uni
t
or
wi
t
h
a m
i
croc
ode
d
engi
ne s
u
c
h
as
t
h
e
DPE.
T
h
e
sy
nt
he
sized implem
entation is prefe
rre
d
for
basic se
rvices i
n
cluding a
v
era
g
ing,
d
a
ta co
m
p
ressio
n
, tran
sition
co
un
ting
,
ev
ent trig
g
e
ring
and
t
h
resho
l
d
d
e
tectio
n
.
Th
e micro
c
o
d
e
d
eng
i
n
e
is
b
e
st su
ited
fo
r
th
e co
m
p
lex
serv
ices, as th
ey
typ
i
cally
require com
p
licated processing
o
f
t
e
m
poral
dat
a
.
These
i
n
cl
ude
ser
v
i
c
e
s
suc
h
as l
i
n
ea
ri
zat
i
on a
n
d s
m
oot
hi
ng,
ed
g
e
det
ect
i
o
n
,
da
t
a
sup
p
r
essi
o
n
,
dat
a
f
u
si
o
n
,
fi
l
t
e
ri
ng
and si
gn
al
rep
r
od
uct
i
o
n. C
h
an
nel
n
ode
s ha
n
d
l
e al
l
t
r
an
sm
i
s
si
on
a
nd b
u
f
f
eri
ng of dat
a
bet
w
een
t
h
e
F
unc
t
i
onal
Service Unit (FSU) and t
h
e
PPU. The FS
U
i
s
“fi
r
ed”
w
h
e
n
t
h
e c
h
an
nel
no
de
has
bu
ffe
red al
l
t
h
e t
o
ke
n dat
a
fro
m
th
e PPU an
d
is read
y
to
tran
sm
it i
t
. Ch
ann
e
l nod
es are also
used to
tran
sm
it d
a
ta b
e
tween
m
u
ltip
le
dat
a
fl
o
w
pr
oce
ssi
ng
el
em
ent
s
as desc
ri
be
d
be
l
o
w i
n
Sect
i
o
n
4.
The dat
a
fl
ow
pr
ocessi
ng el
e
m
ent
(DPE)
u
s
ed f
o
r t
h
is pl
atform
is
im
ple
m
ented as a
stack-base
d
micro
c
od
ed eng
i
n
e
with adv
a
n
ce
features such
as n
e
sted
l
o
o
p
i
n
g
, co
nd
ition
a
l execu
tion
,
rep
eat ex
ecu
tion
an
d
a p
r
o
g
ram
m
abl
e
l
o
o
k
u
p
t
a
bl
e
f
o
r
rec
o
n
f
i
g
ur
abl
e
f
u
nct
i
onal
o
p
erat
i
o
ns
[
1
9]
. Fi
gu
re
10
bel
o
w s
h
ows
a
bl
oc
k
di
ag
ram
of t
h
e DPE i
m
pl
em
ent
e
d
usi
n
g t
h
re
e Que
u
e
d
-St
a
c
k
(
Q
S) el
em
ent
s
and o
n
e
out
put
F
I
F
O
. T
h
e
i
nput
QS elem
ents are used to
rece
ive cha
n
nel da
ta from
two
so
urces
o
r
t
h
ey
c
a
n
be c
o
n
f
igured s
u
ch that
one
QS
ele
m
ent is receiving
data
while the ot
her is
processi
ng
da
ta from
a pre
v
ious tra
n
saction. T
h
e Res
u
lt-QS is
u
s
ed
to store the resu
lts of th
e
d
a
tap
a
th op
eratio
n
s
.
Funct
i
o
n
al
Se
r
v
i
c
e
s
Uni
t
(
F
SU)
Signal
Da
ta
T
o
k
e
ns
Sideba
nd
Da
t
a
T
o
ke
ns
P
r
e
p
roc
essi
n
g
Uni
t
(PPU
)
Condi
t
i
on
ing
Con
ver
sio
n
Det
ect
i
on
Redu
ct
ion
Inp
u
t
fro
m
Sen
s
or
Reado
u
t
Ci
rcui
try
Fus
i
o
n
Control
Ti
m
e
S
t
a
m
p
i
n
g
C
h
a
nnel
N
ode
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
I
J
ECE Vo
l. 4
,
N
o
. 4
,
Au
gu
st 2
014
:
52
0
–
53
1
52
7
The DP
E i
s
im
pl
em
ent
e
d usi
n
g a param
e
t
e
rized sy
nt
hesi
ze
d m
odel
where
t
h
e wi
dt
h an
d dept
h of t
h
e
stack
s,
fun
c
tion
a
l un
its, an
d data-p
ath
s
are determin
ed
du
ri
ng
al
g
o
ri
t
h
m
i
c devel
opm
ent time. For systems that
are com
posed
of m
u
ltiple DPEs, it is
feasible for each
DPE to be confi
g
ur
e
d
for a task or group
of
tasks
du
ri
n
g
t
h
e sy
nt
hesi
s
pr
ocess
b
y
sel
ect
i
ng t
h
e
opt
i
m
al
param
e
t
e
rs.
The m
i
crocode
st
orage i
s
i
m
pl
em
ent
e
d usi
ng a st
an
da
rd
si
ngl
e p
o
rt
R
O
M
or R
A
M
/
WC
S
m
e
m
o
ry
com
p
i
l
e
r. The
R
A
M
/
W
C
S c
o
nfi
g
u
r
at
i
o
n
i
s
usef
ul
f
o
r sy
st
em
s whe
r
e t
h
e
m
i
croco
d
e
nee
d
s t
o
be
u
pdat
e
d f
r
o
m
an e
x
ternal source s
u
ch as
flas
h m
e
m
o
ry [20].
Fi
gu
re
1
0
.
Dat
a
fl
o
w
P
r
oce
ssi
ng
El
em
ent
bl
o
c
k
di
ag
ram
4.
SYSTE
M
CO
MPO
S
ABILI
T
Y
Co
m
p
o
s
ab
ility p
r
o
v
i
d
e
s th
e
ab
ility to
sele
ct “co
m
p
o
s
ab
le” ele
m
en
ts a
n
d
assem
b
le th
em
in
to
a
topology nee
d
ed for a s
p
ecific algor
ith
m
.
Fo
r an
SDF el
e
m
en
t to
b
e
co
m
p
o
s
ab
le it
m
u
st b
e
m
o
d
u
lar (self
cont
ai
ne
d
)
a
n
d
can
be
de
pl
o
y
e
d i
n
de
pen
d
e
n
t
l
y
. It
m
u
st
also
be stateles
s which m
eans that it treats
eac
h
req
u
est
(o
r
firi
ng
) as
a
n
in
de
pen
d
e
n
t tra
n
sa
ction,
u
n
related
to an
y prev
i
o
u
s
requ
est
[21
]
. Th
e co
m
p
ositio
n
rules f
o
r
t
h
is platform
are:
Rule 1
:
All in
pu
ts to
an
elem
en
t (act
o
r
) will h
a
v
e
a FIFO
qu
eu
e.
Th
e ou
tpu
t
can h
a
v
e
a
q
u
e
u
e
to
satisfy th
e
n
eed of
R
u
le #3
b
e
low
.
Rule 2
:
All d
a
ta p
r
op
ag
ates th
ro
ugh
a d
a
taflow n
e
t
w
or
k
via channels. Note: channel nodes conve
r
t dat
a
str
eam
s as
th
ey p
a
ss th
ro
ugh th
e n
e
tw
ork
,
e.g
.
Ser
i
al-
P
arallel, Par
a
llel-
Ser
i
al, Str
eam
-
F
LI
T, FLI
T
-
Stream
, etc.
Rule 3
:
Fo
r
PUSH M
ODE
op
eration
,
Read
s to
the FI
FO will b
l
o
c
k, howev
er
Writes will not. For PULL
M
ODE
op
erat
i
on t
h
e i
nve
rse
i
s
t
r
ue. Thi
s
pl
at
fo
rm
i
s
desi
gne
d t
o
s
u
p
p
o
rt
b
o
t
h
PU
S
H
an
d P
U
L
L
m
ode ope
rat
i
o
n.
Rule 4
:
Th
e co
m
p
o
s
ed system
will b
e
d
e
term
in
ate wh
ich
re
qu
ires th
at each
act
or is
fun
c
tion
a
l
an
d th
at t
h
e
set of firi
ng
rules are sequent
i
al. “Functiona
l”
m
eans
that an actor
firing
lack
s side effe
cts and that
th
e ou
tpu
ttok
e
n
s
are
p
u
rely a
fun
c
tion
o
f
t
h
e
in
pu
t tok
e
ns con
s
u
m
ed
in th
at
firing
.
Rule 5
:
Ele
m
en
ts can
b
e
so
ft
ware rou
tin
es. Ru
le
#4
stat
es that these
routines
can
be m
ove
d
t
o
al
t
e
r
n
at
e
com
put
at
i
onal
engi
nes a
n
d e
x
ecut
e
wi
t
h
o
u
t
m
odi
fi
cat
i
on.
The com
posa
b
l
e
pl
at
form
i
s
im
pl
em
ent
e
d us
i
ng an e
v
ent
d
r
i
v
en sy
nc
hr
o
n
ous
dat
a
fl
o
w
a
r
chi
t
ect
u
r
e.
The system
is “com
posed”
by instantiating datafl
ow-p
rocessing elements (DPE
)
that are connect
ed via
PU
S
H
PO
P
/
I
N
S
E
RT
PU
SH
TO
S
BOS
TO
S
BOS
OUTP
UT
C
H
A
NNE
L
DAT
A
IM
M
E
DI
A
T
E
DAT
A
RE
S
U
LT_
B
U
S
I
nput
Queu
ed
‐
St
ac
k
Re
sul
t
Qu
eued
‐
St
a
c
k
RESULT
_
B
US
RE
S
U
LT_
B
US
PO
P
/
I
N
SE
RT
PU
S
H
TO
S
BOS
I
nput
Queu
ed
‐
St
ac
k
RE
SUL
T
_
B
U
S
INP
U
T
CHA
N
NE
L
DA
T
A
Datapath
Output
FIFO
(
C
hanne
l
Nod
e
)
RESULT_
B
U
S
PO
P/
I
N
SE
RT
Mi
cr
o
‐
Co
d
e
Eng
i
ne
S
e
que
ncer
GPI
O
IN
P
U
T
GP
I
O
OUT
P
U
T
RE
S
U
L
T
_
B
U
S
Act
o
r
/
E
v
e
n
t
Qu
e
u
e
Ev
e
n
t
s
Re
ci
rc
u
l
a
t
e
d
Acto
rs
INP
U
T
CHA
N
NE
L
DA
T
A
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
C
o
g
n
i
t
i
ve Se
ns
or Pl
at
f
o
r
m
(
M
ark Mc
De
rm
o
t
t)
52
8
chan
nel
s
.
Th
e
DPE
s
pr
ovi
d
e
key
fu
nct
i
o
nal
ser
v
i
ces
(
act
ors
)
i
n
t
h
e
dat
a
fl
ow
sy
s
t
em
. In t
h
e c
u
r
r
ent
i
m
p
l
e
m
en
tatio
n
th
e i
n
pu
ts to
th
e acto
r
s are
m
o
d
e
led
as bou
nd
ed
qu
eu
es
as it is p
o
ssi
b
l
e to
d
e
term
in
e
a-priori
what t
h
e
dept
h
of each que
u
e
needs to be
duri
ng al
gorithm
i
c
m
a
pping a
n
d s
y
ste
m
level
m
odeling.
Fi
gu
re
1
1
bel
o
w s
h
o
w
s
an
ex
am
pl
e of
a c
o
m
posed sy
st
e
m
wi
t
h
fu
nct
i
o
nal
an
d c
h
a
nne
l
no
des
.
I
n
t
h
i
s
exam
ple the channel nodes
are use
d
to
co
nv
ert d
a
ta fro
m
th
e sen
s
or ele
m
en
t in
to
d
a
ta to
k
e
n
s
t
h
at are
fo
rwa
r
ded t
o
t
h
e D
P
Es. T
h
e
C
o
m
m
uni
cat
ions
El
em
ent
i
s
an eve
n
t
d
r
i
v
en f
unct
i
onal
no
de a
nd
fol
l
o
ws t
h
e
co
m
p
o
s
ab
ility ru
les
d
e
scrib
e
d abo
v
e
.
Fi
gu
re
1
1
. C
o
m
posed sy
st
em
sh
owi
n
g
f
u
nct
i
onal
a
n
d
ch
an
nel
n
o
d
es
Fi
gu
re
1
2
b
e
l
o
w s
h
o
w
s
a se
n
s
or
sy
st
em
t
opol
o
g
y
w
h
e
r
e t
h
e cha
nnel
n
ode
s are c
o
nfi
g
u
r
e
d
as
r
out
e
r
s.
The ch
an
nel
r
o
ut
i
ng
n
odes
r
o
ut
e t
oke
ns t
h
r
o
ug
h t
h
e
net
w
or
k i
n
a p
r
e
d
efi
n
ed pat
t
e
r
n
. T
h
e
ro
ut
i
ng
pat
t
e
r
n
s are
lo
ad
ed
du
ring
syste
m
in
itial
i
zatio
n
and
are static. Th
es
e typ
e
s
of n
e
twork
top
o
l
o
g
i
es p
r
ov
id
e flex
i
b
ility
in
bui
l
d
i
n
g
a
wi
d
e
ra
nge
o
f
se
ns
or
pl
at
f
o
rm
s at
t
h
e ex
pe
nse
of
i
n
crease
d
e
n
er
gy
us
age
[
22]
.
Fi
gu
re
1
2
. C
o
m
posed sy
st
em
sh
owi
n
g
c
h
an
nel
n
o
d
es c
o
nfi
g
u
r
e
d
as
r
out
er
s
5.
M
O
D
ELING a
n
d
PROGRA
MM
IN
G
The C
SP i
s
m
odel
e
d, a
n
al
y
zed an
d
pr
o
g
ra
m
m
e
d usi
ng
M
A
TLAB
/
Si
m
u
l
i
nk
an
d Si
m
E
ve
nt
s fr
om
M
a
t
h
Wor
k
s
[2
3]
. Si
m
E
vent
s
i
s
an e
v
e
n
t
-
bas
e
d si
m
u
l
a
t
o
r t
h
at
w
o
rks
i
n
c
o
n
j
unct
i
o
n
wi
t
h
Si
m
u
l
i
nk t
o
m
odel
bot
h t
i
m
e based a
nd e
v
e
n
t
d
r
i
v
en
sy
st
em
s. The se
ns
or a
n
d
ADC
s
u
b-sy
st
e
m
can be desc
r
i
bed i
n
M
A
T
L
A
B
o
r
built from
Sim
u
link library m
odels
. The
out
put
of the
ADC is conve
rted
into
a signal-e
vent that is processe
d
b
y
th
e
Sim
E
v
e
n
t
s sim
u
lato
r. Sim
E
v
e
n
t
s does no
t do
co
mp
u
t
ation
a
l simu
latio
n
bu
t rat
h
er
sim
u
lates t
o
k
e
n
s
(en
tities) p
r
opag
a
tin
g
thro
ugh
th
e SDF
n
e
twork. Each
resou
r
ce in
th
e n
e
twork
can b
e
in
stru
m
e
n
t
ed
to
d
e
term
in
e if th
ere are an
y erro
rs as the to
k
e
n
s
prop
ag
ate through
th
e n
e
t
w
ork
.
Add
itio
n
a
ll
y th
e
in
stru
m
e
n
t
atio
n
en
ab
les d
e
bug
cap
a
b
ility b
y
prov
id
i
n
g v
i
si
b
ility to
v
a
ri
o
u
s p
a
ram
e
ters in
a n
e
t
w
ork
resou
r
ce.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
I
J
ECE Vo
l. 4
,
N
o
. 4
,
Au
gu
st 2
014
:
52
0
–
53
1
52
9
Fi
gu
re
1
3
. C
S
P
Si
m
E
vent
s M
odel
Fig
u
r
e
1
3
above sho
w
s a CSP Sim
E
v
e
n
t
s mo
d
e
l
w
ith
t
w
o
Fun
c
tio
n
a
l
Serv
ice Un
its (FSU
)
and
thr
e
e
Dataflow Proc
essing
Elem
en
ts. The
FSUs
are token ge
ne
ra
to
rs th
at laun
ch
t
o
k
e
n
s
i
n
t
o
th
e
n
e
two
r
k
.
Each
t
oke
n has t
w
o
at
t
r
i
but
es as sh
ow
n a
b
o
v
e i
n
Fi
gu
re 1
4
. T
h
e
DPE’
s p
r
oces
s
t
h
e t
oke
ns an
d
t
h
en o
u
t
p
ut
s t
h
em
t
o
a comm
unications
elem
ent that is m
odeled a
s
a toke
n sink.
DPE
_1 a
n
d D
P
E_
2 are m
odel
e
d wi
t
h
a si
ngl
e que
ue a
nd a
si
ngl
e act
o
r
as
sho
w
n bel
o
w i
n
Fi
g
u
re
1
4
.
Not
e
t
h
e i
n
st
r
u
m
e
nt
at
i
on po
rt
s on t
h
e actor. These are us
ed to determin
e optim
al resource allocation for the
single queue
DPE.
Fig
u
r
e
14
. Sing
le
qu
eu
e D
P
E
m
o
d
e
l
DPE
_3 i
s
m
o
d
e
l
e
d wi
t
h
t
w
o
que
ues
,
a t
o
ke
n com
b
i
n
er
, t
o
ken c
o
ns
um
er
and a si
n
g
l
e
act
or as s
h
o
w
n
b
e
low
in Figure 15
.
Fig
u
re
15
. M
u
ltip
le Qu
eu
e
DPE m
o
d
e
l
Th
e CSP is prog
ramm
ed
b
y
selectin
g
actors fro
m
a lib
rary an
d in
stan
tiating
th
em
o
n
a DPE.
A
b
r
o
a
d
selectio
n
o
f
acto
rs are av
ailab
l
e in
clud
i
n
g
:
d
i
g
ital filtering
fun
c
tion
s
,
deci
m
a
tio
n
,
lin
earizatio
n
,
av
erag
ing
,
Evaluation Warning : The document was created with Spire.PDF for Python.