TELKOM
NIKA
, Vol.11, No
.11, Novemb
er 201
3, pp. 6309
~6
314
e-ISSN: 2087
-278X
6309
Re
cei
v
ed Ap
ril 1, 2013; Re
vised June
9,
2013; Accept
ed Jul
y
1, 20
13
Hypersonic Vehicle Tracking based o
n
Improved
Current Statistical Model
He Gu
angju
n
*, L
v
Hang, Li Baoqua
n, Li Yanbin
Schoo
l of Air and Missil
e
Def
ense, Air F
o
rce
Engin
eeri
ng U
n
iversit
y
,
Xi’
a
n
,
71005
1, Chi
n
a
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: guan
gj
unh
e
@
sina.c
o
m
A
b
st
r
a
ct
A new
meth
o
d
of track
i
ng
the n
ear
spac
e hy
per
so
nic
vehicl
e
is p
u
t
forw
ard. Acc
o
rdi
ng to
hypers
onic
veh
i
cles
’
char
acter
i
stics, w
e
impr
oved c
u
rre
nt
statistical mod
e
l throug
h
o
n
li
ne ide
n
tificatio
n
of
the ma
ne
uveri
ng frequ
ency. A Monte
Carl
o
simu
latio
n
is u
s
ed to ana
ly
z
e
the performan
c
e of the meth
od
.
T
he results sh
ow
that the improve
d
metho
d
exhib
i
ts
very goo
d trackin
g
perfor
m
a
n
ce i
n
comparis
on w
i
t
h
the old
meth
od
.
Ke
y
w
ords
:
cur
r
ent statistical
mo
de
l, near
sp
ace, hypers
o
n
i
c, target
tracking, mane
uveri
ng frequ
ency
Copy
right
©
2013 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
In recent yea
r
s, ne
ar
spa
c
e
hyperso
nic vehicle
h
a
s al
ways
b
een
a hot are
a
of re
sea
r
ch.
X-37B
and
X-51A te
st vehi
cle
s
in
the
US made
a
su
ccessful t
r
ial f
light in
April
a
nd M
a
y of
20
10
sep
a
rately [1,
2], marking t
hat the US h
a
s w
on a co
mpletely new brea
kthroug
h in the are
a
o
f
hypersoni
c cruise vehicl
e
.
Meanwhil
e
, the se
rie
s
of X-51B~X-51H etc a
r
e
following th
e
pre
s
crib
ed o
r
der. Russia,
Japan
and
so
me othe
r cou
n
tries i
n
Euro
pe are al
so
striving to deve
l
op
their own ne
ar space hy
p
e
rsoni
c vehicles [3]. These vehicle
s
are divided into
two categ
o
ri
es:
low dyna
mic
one
s and
hig
h
dynami
c
on
es. Fo
r t
hose
high dynami
c
vehicl
es
wit
h
the sp
eed
of
more tha
n
5 Mach, ho
w t
o
track them
grad
ually turns to be a p
r
oblem. In the paper, the n
ear
spa
c
e
hype
rsonic vehi
cle
s
’
traje
c
tory
is
analyzed,
the
n
,
acco
rdi
ng to
its ch
ara
c
t
e
rs we studi
e
d
how to
tra
c
k hyperso
nic t
a
rget
applyin
g
improv
ed
current stati
s
tical
(CS
)
mo
del. Simulati
ons
sho
w
that the
improved
CS model exhibi
ts a very goo
d perfo
rman
ce.
2. Analy
s
is o
f
Nea
r
Space
H
y
personic Vehicle Trac
king
Acco
rdi
ng to
literatu
r
e
[4
], near spa
c
e hype
rsoni
c vehicl
es u
s
ually ad
opt
scramj
et
combi
ned e
n
g
ine an
d turb
o ro
cket engi
ne due to the
i
r flight enviro
n
ment. And in orde
r to sa
ve
fuel con
s
um
p
t
ion and redu
ce the de
ma
nds fo
r the
cooling
syste
m
, a jumping
trajecto
ry is
often
use
d
. A typical flight trajectory is de
scri
bed a
s
Figu
re
1 [5].
Figure 1. Jum
p
ing Traje
c
tory
As Figure 1 sho
w
s, the vehicl
e start
s
to jump
from the height of 30km, with th
e initial
flight
path an
gle
of 0 deg
ree.
At
that ti
me it i
s
at
th
e lo
we
st poi
nt of the
orbi
t, just the
n
t
h
e
scramj
et engi
ne ignite
s to
boo
st. Wh
en
it rea
c
he
s a
certain
heig
h
t belo
w
40
km,
the engi
ne tu
rn
s
off. And the
vehic
l
e
glides freely to the next lowes
t
point. In the t
e
rminal
part
of the glide, the
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: 630
9 – 6314
6310
vehicle raises its head de
p
endin
g
on ae
rodyna
mic fo
rce. The
n
the engin
e
ignite
s to boo
st ag
ain
and the proce
ss i
s
re
peate
d
, leadi
ng a n
on-stop flight eventually
to the sched
ule
d
destin
a
tion.
In ord
e
r to
tra
c
k the afo
r
em
entione
d jum
p
ing traje
c
tory, a pro
per m
odel i
s
n
eed
e
d
to be
built. When
we build a mod
e
l of a maneu
vering targe
t, we mainly ex
pect it not onl
y to conform to
reality but also to be co
n
v
enient in mathemat
ical treatme
nt. By analyzing th
e whole ju
m
p
ing
trajecto
ry, we divide the
idealized course of f
lig
ht into con
s
tant velocity (CV) m
o
tion
and
con
s
tant a
c
celeratio
n
(CA
)
motion, an
d
we u
s
e
CV
model a
nd CA model to track the traje
c
tory.
For the CV
motion, we choo
se white
noise pro
c
e
s
s that confo
r
ms to the
ch
ara
c
t
e
ri
st
ic
s of
t
he
trajecto
ry to
make
up th
e
unpredi
ctable
error
attac
h
e
d
in the m
ode
l. While fo
r th
e CA motio
n
,
we
cho
o
se the
CS mod
e
l. CS model
see
s
the e
s
timat
i
on of la
st m
o
ment a
s
th
e cu
rrent me
an
accele
ration,
and th
e a
c
cel
e
ration
in
the
mod
e
l
con
s
i
s
ts
of the
me
an a
c
cele
rati
on a
nd
a Sin
ger
accele
ration
pro
c
e
s
s [6]. For th
e m
odel hyp
o
the
s
is th
at the
accele
ratio
n
and th
e m
ean
accele
ration
are i
dentified
online, it i
s
e
v
en clo
s
e
r
to
the reality th
an Sing
er
mo
del an
d it h
a
s a
wide
r ra
nge
of application
s
. This m
odel
is acquir
ed o
n
the ba
sis of
doing research into g
ene
ral
maneuve
r
ing
target
s, an
d it ha
s g
o
od p
e
rfo
r
ma
nce
in tracking pla
ne ta
rgets. B
u
t the
maneuve
r
ing
freque
ncy in
the model i
s
an empi
rica
l
value took
b
y
analyzing
g
eneral targ
ets,
while the target’s m
o
vem
ent is mainly
uniform
, the es
timation of the ac
celeration may
have
rand
om ch
an
ge, resulting in erro
r co
mp
aring
with
re
ality. By combining all the analysi
s
re
sul
t
s,
we ca
n use CV model an
d CS model to descri
be
the tracking of hypersoni
c vehicl
es, but som
e
para
m
eters is needed to b
e
identified a
nd set pro
p
e
r
ly
acco
rdin
g to the charact
e
rs of the act
ual
trajec
tory.
3. Tracking
Model for Ne
ar Space Hy
personic Ve
hicle
In general, he
re we
con
s
id
er a typical lin
ear
time-va
r
yi
ng discrete system, it can be
r
e
pr
es
e
n
t
ed
a
s
:
11
Xk
k
X
k
k
Zk
H
k
X
k
v
k
(1)
Whe
r
e
X
k
de
no
ted by
n
X
kR
is th
e sy
stem
sta
t
e, k i
s
d
eno
ted by
N
k
,
1
and
Z
k
denote
d
by
q
Z
kR
is the output measurement
.
n
n
R
k
is the state
transition
matrix.
q
n
R
k
H
is the
m
easure
m
ent
matrix.
k
Q
k
,
0
~
and
k
R
k
v
,
0
~
are the mutu
ally indepen
d
ent noises th
at are u
s
ed t
o
describe th
e system
disturban
ce a
nd the mea
s
u
r
eme
n
t noise, resp
ectively.
For CS mo
de
l, the followin
g
formula
can
be acq
u
ire
d
.
t
a
a
t
a
t
x
~
)
(
t
t
a
t
a
t
a
)
(
OR
t
t
a
t
a
1
~
(2)
(3)
In Formula
(2),
t
a
~
is the ze
ro-m
ean
colo
red a
c
cele
rat
i
on noi
se, an
d it is a Singer
accele
ration pro
c
e
ss.
a
is the mean val
ue of maneu
vering a
c
cele
ration, and it
is con
s
tant in
each sam
p
lin
g perio
d.
is the maneuve
r
ing fre
que
ncy ( recip
r
o
c
al
of the maneuvering tim
e
con
s
tant).
t
is
zero-m
ean
white
noi
se
with th
e v
a
rian
ce
2
2
2
a
, in whic
h
2
a
rep
r
e
s
ent
s accele
ration va
riance of the target.
t
1
is Gau
ssi
an noi
se
with mean valu
e of
a
.
In Formula
(3
), there is a
key hypothesi
s
in CS model
k
def
k
k
k
k
def
k
a
Z
a
E
Z
a
E
a
ˆ
]
/
[
]
/
[
1
1
(4)
Whe
r
e
k
Z
is all the output me
asu
r
em
ent till the moment k, and
k
a
ˆ
is the estimation at
moment
k.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
Hyperso
nic V
ehicl
e Tra
c
ki
ng ba
sed o
n
Im
prov
e
d
Cu
rrent Statistical
Model (He G
uangj
un)
6311
The var
i
ance
2
is obtaine
d from the followi
ng formul
a.
0
ˆ
,
ˆ
4
2
0
ˆ
,
ˆ
4
2
2
max
2
2
max
2
t
a
t
a
a
t
a
t
a
a
(5)
Obviou
sly, in CS model, it is need
ed to
acqui
re
)
(
t
a
,
,
t
1
and
2
. The mean
value of
a
c
cel
e
ration
)
(
t
a
and
t
he va
rian
ce
o
f
noise
2
ca
n be
ac
q
u
i
r
e
d
from F
o
r
m
u
l
a (4)
an
d
Formul
a (5),
throug
h worki
ng out the
estimation of la
st mome
nt.
is a a
n
empi
ri
cal valu
e set
according
to
the flight cha
r
acte
rs of g
e
neral
airplan
e
, wh
en the
target m
a
ke
s a
slo
w
tu
rn
the
value is 1/60
and when it
make
s a e
s
cap
e
man
e
u
v
er the value
is 1/20. But for nea
r sp
a
c
e
hypersoni
c vehicl
es, we
can
not simpl
y
choo
se th
ese em
piri
ca
l values. Adaptive online
identificatio
n algorith
m
is u
s
ed to de
cid
e
the value of
according to the flight con
d
i
tions.
4. Online Identifica
tion of Maneuv
ering Frequen
c
y
A lot of literature on
m
aneuve
r
ing
target t
r
a
ckin
g
mod
e
l bel
ieve that th
e onlin
e
identificatio
n
of mane
uveri
ng fre
que
ncy
is ve
ry imp
o
rtant fo
r hyp
e
rsoni
c vehi
cl
e tra
cki
ng [7,
8]. But there
is no fine
en
ough te
rm
s o
f
settlement till now.
We st
udied t
w
o me
thods
of onlin
e
identificatio
n of
as
following.
4.1. Online Identific
a
tion
based on Kalman Filtering
Suppo
se the
maneuve
r
ing
frequen
cy
at moment k is denoted by
k
=
. Then the
target’s
mot
i
on mod
e
l can b
e
pre
s
ente
d
by Formul
a (1
).
The state
equation i
s
k
k
X
k
k
X
1
, and the observation eq
ua
tion is
k
v
k
HX
k
Z
.
Whe
n
1
k
=
, the filtering p
r
o
c
e
ss is
1
1
ˆ
ˆ
1
1
1
1
1
1
1
1
1
ˆ
1
1
ˆ
1
k
k
P
H
k
K
I
k
k
P
k
k
K
k
k
X
k
k
X
k
S
H
k
k
P
k
K
k
Q
k
k
k
P
k
k
k
P
k
k
X
k
k
k
X
T
T
(6)
Whe
r
e
1
ˆ
k
k
X
H
k
Z
k
an
d
k
R
H
k
k
HP
k
S
T
1
are p
r
edication
resi
dual a
nd i
t
s cova
rian
ce
matrix.
1
1
ˆ
k
k
X
,
1
1
k
k
P
and
k
Z
are inputs.
After we
g
e
t
predi
catio
n
re
sidu
al a
nd it
s cova
ria
n
ce
matrix at m
o
ment
k-1
fro
m
Fo
rmula
(6), we can
work out mat
c
hin
g
pro
babi
lity of t
he target’s motion
state betwe
e
n
the momen
t
k-1
and k. Th
e ca
lculatio
n pro
c
ess is a
s
follo
wing.
k
k
S
k
k
S
k
T
1
2
1
exp
2
1
(7)
Con
d
u
c
t no
rmalizatio
n p
r
oce
s
sing
to t
he
re
sult of li
kelih
ood
fun
c
tion
)
(
k
, we
will
get
the matching probability
)
(
k
betwe
en th
e mod
e
l bei
ng u
s
ed
an
d
the a
c
tual
model.
We
introdu
ce
d h
a
rd threshold
0
, if
0
)
(
k
we thin
k
can reflect
the target’s
motion state
truthfully, if n
o
t we co
ntinu
e
to find the value of
throu
gh iteration.
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: 630
9 – 6314
6312
4.2. Online Identific
a
tion
based on
H
Fi
ltering
Whe
n
filteri
n
g mo
del
and
targ
et motio
n
mo
de
mat
c
h
accu
ratel
y
, and th
e
statistical
cha
r
a
c
ter of system di
sturb
ance is kno
w
n already
, Ka
lman filterin
g
based o
n
line
identificatio
n i
s
the mo
st opti
m
al solution.
But for the hi
gh spee
d
an
d
high m
ane
uvering
targ
ets,
the actu
al noi
se
covari
an
ce i
s
usually g
r
eat
er th
an th
e a
s
sumed
noi
se covaria
n
ce, thus lea
d
ing
to low p
r
e
c
isi
o
n
and even the
diverge of the
filter. Then
H
filtering ba
se
d online ide
n
tification i
s
put forward.
Add
Yk
L
k
x
k
to Form
ula (1
), therei
n
k
L
is
state co
mbination
ma
trix and it
is related
to
covarian
ce
up
dating
of
H
filte
r
ing. He
re we
do not
ma
ke
any
a
s
sumpti
ons, but
for
the energy is limited.
Whe
n
1
k
=
, the filtering in
put are
1
1
ˆ
k
k
X
1
1
k
k
P
and
Yk
. Set
noise su
ppre
ssi
on pa
ram
e
ter
0
, then judge if
1
1
k
k
P
meet the con
d
ition
0
1
2
L
L
H
H
k
P
T
T
(8)
Whe
r
e
I
L
. If Formula (8
)
can
not be met, increa
se the
value of
and
judge o
n
ce
again. Else if Formul
a (8
) is true, predi
ct filt
ering cova
rian
ce matrix
of the next moment by:
T
k
e
T
T
T
k
P
L
H
R
L
H
k
P
Q
k
P
k
P
1
1
1
1
,
(9)
Whe
r
e:
T
T
k
e
L
H
k
P
L
H
I
R
R
1
0
0
2
,
(10
)
Then the
state estimation
of the target based on
H
filtering i
s
:
ˆˆ
ˆ
()
(
1
)
(
)
(
)
(
)
(
1
)
Xk
Xk
K
k
Y
k
H
k
Xk
(11
)
Whe
r
e
k
K
is filtering gai
n den
oted by:
1
)
(
)
(
)
(
T
T
H
k
HP
R
H
k
P
k
K
(12
)
Whe
r
e p
r
edi
cation re
sidu
al
and its cova
riance matrix are de
noted
by:
ˆ
()
()
(
/
1
)
(
)
(
/
1
)
T
kY
k
H
X
k
k
S
k
H
P
k
k
H
R
,
(13
)
It should
be
noted that th
e paramete
r
s Q and
R u
s
ed in
H
filtering is different
from
those in Kal
m
an filtering.
In fact,
they are weig
ht coeffici
ent se
t by us acco
rding to targ
et’s
motion mod
e
and actual
noise backg
round. They
ar
e written a
s
above is to
make it easy to
comp
are with
Kalman filtering.
Then
solvin
g likeli
hood
function th
roug
h Fo
rm
ula (5
) a
n
d
solving
ma
neuveri
n
g
freque
ncy a
r
e
similar to me
thods in
cha
p
t
er 4.1.
5. Simulations and An
aly
s
is
Apply the aforeme
n
tione
d algorith
m
to identify mane
uvering frequ
ency
online.
The
simulatio
n
scenari
o
is a hi
gh sp
eed a
n
d
high man
euv
ering ta
rget
moves a
s
Fig
u
re 1
sho
w
s: first
CA,
se
con
d
CV
and
t
h
ird con
s
tant reta
rded accel
e
ra
tion then
rep
eat from
CA. In the sim
u
lat
i
on,
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
Hyperso
nic V
ehicl
e Tra
c
ki
ng ba
sed o
n
Im
prov
e
d
Cu
rrent Statistical
Model (He G
uangj
un)
6313
the initial val
u
e of
X
k
is de
not
ed by
0
3
00000
2640
30
Xx
x
x
, (units
are m,
/
ms
and
2
/
s
m
). The M
onte Ca
rlo si
mulation time
is 100.
The mea
s
u
r
e
m
ent noise is modeled a
s
:
k
e
x
x
k
x
k
x
k
v
0
0
(14
)
Whe
r
e
deno
tes relative e
rro
r co
efficie
n
t,
X
and
X
denotes fixed measure
m
ent
error,
and
k
e
is a
ze
ro-me
an
norm
a
l p
s
eu
do-ra
ndom
num
be
r
with
vari
an
ce of 1.
The
o
b
se
rvation
noise cova
ria
n
ce i
s
as follo
wing:
k
e
E
x
x
k
x
k
x
k
R
2
2
0
(15
)
Whe
r
e
denotes arbitra
r
ily small real. Simulation resu
lts are a
s
Fig
u
re 2 an
d Fig
u
re 3 sho
w
.
Figure 2. Pre
d
iction of Ta
rget’s Man
euv
ering
Freq
uen
cy
Figure 3. Pre
d
iction of Ta
rget’s Man
euv
ering
A
ccel
e
r
a
t
i
on
It can be co
nclu
ded fro
m
the figure
s
that whe
n
the
target move
s unifo
rmly and in a
straig
ht line,
the predi
ction
of
its ma
neu
vering frequ
e
n
cy is gr
eate
r
, and th
e filtering
mod
e
l i
s
simila
r to
CV model.
Whe
n
the target
moves
with
consta
nt accel
e
ration, th
e
predi
ction
of
its
maneuve
r
ing
frequ
en
cy is very sm
all (i
n fact t
he
order
of magnit
ude i
s
10
-3
), and th
e filtering
model i
s
simil
a
r to
CA m
o
d
e
l. Co
ncl
u
sio
n
can
be d
r
a
w
n th
at the i
m
prove
d
alg
o
r
ithm p
u
t forward
in the pape
r can ide
n
tify the maneu
vering fre
q
u
ency of ne
a
r
sp
ace hyp
e
rsoni
c vehi
cle
s
adaptively on
line, thu
s
ma
king
the filteri
ng mo
del
m
a
tch the
a
c
tual
co
ndition
s m
o
re
accu
ratel
y
.
Both Kalman
and
H
filtering
based
onlin
e identificatio
n
ca
n id
entify mane
uverin
g freq
uen
cy
relatively authentically, but the latter method gi
ves a maneuve
r
ing
freque
ncy wit
h
a wider ran
ge,
that is becau
se in
H
filtering the cha
r
acte
r of interferen
ce noise is un
known.
0
20
40
60
80
10
0
12
0
140
16
0
-2
0
2
4
6
8
10
12
14
16
18
观测
时间
()
s
目
标
机率
动频
预测
值
K
a
lm
a
n
F
ilt
e
r
H-
i
n
f
F
ilt
e
r
0
20
40
60
80
10
0
12
0
14
0
16
0
-1
5
-1
0
-5
0
5
10
15
观测
时间
()
s
目
标
方
向加速
度估
轴计
X
H
-
in
f
F
i
lt
e
r
Ka
l
m
a
n
F
i
l
t
e
r
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: 630
9 – 6314
6314
Referen
ces
[1]
Xi
ao
F
uge
n. B
a
ckgro
und
of
X-37B S
pace
Ve
hicle
D
e
vel
o
p
m
ent.
Spac
ecr
a
ft Enviro
n
m
en
t Engi
neer
in
g
.
201
0; 27(5): 55
8-56
5.
[2]
W
illiam H
e
n
n
i
gan. H
y
p
e
rso
n
i
c X-
51 W
a
ve
Rider S
hatters
Record
at 3,500 Mp
h.
Stars and Strips
.
201
0; 69(4
2
): 3.
[3]
Hua
ng W
e
i. R
e
searc
h
Statu
s
of
Near S
p
a
c
e H
y
p
e
rson
ic
Vehic
l
es.
W
i
n
ged M
i
ssil
e
s J
ourn
a
l
. 2
007
;
(10): 28-3
1
.
[4]
Xu
e Yo
ngj
ia
ng,
Li T
i
fang.
T
he Devel
opm
ent o
f
Near
Spac
e
Vehic
l
es a
nd A
nal
ysis of
Ke
y T
e
chnolog
y.
Aerody
na
mic Missile
Jo
urn
a
l
. 2011; (2): 32-
36.
[5]
Z
hao J
un, M
eng
Li
ngs
ai.
Jumpi
ng T
r
aje
c
tor
y
Optim
i
za
tion
of the
N
ear S
pace
V
ehicl
es W
i
th
H
y
pers
onic Sp
eed.
Tactical M
i
ssile Tec
hno
lo
gy
. 2010; (5): 3
2
-35.
[6]
Z
hou H
o
n
g
ren,
Jing Z
h
on
gde,
W
ang Pe
id
e. Mane
uver
i
ng T
a
rget T
r
acking.
Beij
ing: N
a
tio
nal
Defens
e
Industr
y
Pr
ess. 1991.
[7]
Han C
hon
gzh
ao, Z
hu Ho
ng
ya
n, Dua
n
Z
h
ansh
e
n
g
, etc. Multi-Sourc
e
Information F
u
sion. Bej
i
n
g
:
T
s
inghua Un
iv
ersit
y
Press. 2
006: 12
4-2
94.
[8]
Xu
Ji
ngsh
uo,
Qin Yo
ng
yu
an,
Pen
g
R
ong.
Ne
w
M
e
th
o
d
f
o
r Sel
e
ctin
g A
daptiv
e Ka
lma
n
F
ilter F
a
di
ng
Fa
cto
r
.
Systems Engin
eeri
ng
and El
ectron
ics
. 2004; (26): 1
552-
155
4.
[9]
Che
n
Ji
aju
n
,
Liu
aofe
ng,
Xi
n
Jinsh
e
n
g
, Gu
X
uefe
ng. R
e
s
earch
on
Man
euver
ing
F
r
eq
uenc
y F
u
zz
y
Adaptiv
e T
a
rge
t
T
r
acking Al
go
rithm.
Journ
a
l
of Projecti
les
R
o
ckets Missi
les
and
Guid
anc
e
. 201
0; (30)
:
259-
262.
[10]
Dai
Hon
g
-d
e, Dai S
h
a
o
-
w
u, Con
g
Yu
a
n
-cai
, W
u
G
uan
g-bi
n. Per
f
ormance
Co
mparisi
on
of
EKF/UKF/CKF for the
T
r
acking
of Ba
llistic
T
a
rget.
T
E
LKOMNIKA Indo
nesi
an J
our
nal
of El
ectric
a
l
Engi
neer
in
g
. 2012; 10(
7):15
3
7
-15
42.
[11]
Kalma
n
RE. A Ne
w
Ap
proac
h to Lin
ear F
ilt
erin
g an
d Pred
iction Pro
b
l
e
m.
IEEE Transactions ASM
E
.
196
0; (20): 34-
45.
[12]
Xi
uli
ng
He, Y
an Sh
i, Jian
g
Yunfan
g. Improve
d
Base
d
on “Se
l
f-Ada
p
tive T
u
rning
Rate” Mo
de
l
Algorit
hm.
T
E
LKOMNIKA Indones
ian J
ourn
a
l of Electrica
l
Engi
neer
in
g
. 2013; 11(
6).
Evaluation Warning : The document was created with Spire.PDF for Python.