TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol.12, No.6, Jun
e
201
4, pp. 4237 ~ 4
2
4
2
DOI: 10.115
9
1
/telkomni
ka.
v
12i6.467
9
4237
Re
cei
v
ed O
c
t
ober 9, 20
13;
Revi
se
d De
cem
ber 19, 20
13; Accepted
Jan
uary 23, 2
014
DOA Estimation by Fourth-Order Cumulants without
Source Enumeration and Eigen Decomposition
Elhafiz Ab
da
lla Bakhit Ya
goup, Zhi
w
e
n Liu, and Yougen Xu
Scho
ol of Infor
m
ation a
nd El
e
c
tronics, Beij
in
g
Institute of
T
e
chn
o
lo
g
y
, Beij
ing 1
0
0
081, Ch
ina
Corresp
on
din
g
auther, e-mai
l
: hafiz_kuttum
@
yah
oo.com, z
w
l
i
u
@
bit.e
du.
cn,
y
o
uge
n
x
u
@
bit.ed
u
.cn
A
b
st
r
a
ct
A new
al
gor
ithm for d
i
recti
on of
arriva
l
(DOA
) esti
mat
i
on
is pr
op
os
ed.Usi
ng fo
urth-ord
er
cumul
ants a
n
d
mo
difi
ed M
U
SIC (MUl
i
pl
le
SIg
nal
Cl
assifica
ti
on) alg
o
rith
m. How
e
ver,
it do
es
n
o
t
re
quir
e
an
y
eig
end
eco
m
po
sition of th
e cu
mu
la
nt matrix
of the rece
iv
ed
data a
nd so
ur
ce en
umerati
o
n. It also eli
m
in
ates
the n
e
e
d
for
kn
ow
ledg
e
of the
spati
a
l c
har
acteristics
of
the
nois
e
a
n
d
int
e
r
f
erence. T
h
is
meth
od
o
n
ly
u
s
es
the conj
ug
ate
spatia
l sig
nal
of di
fferent se
nsor pos
itio
ns. Computer
si
mu
lati
on resu
lts are provi
d
e
d
to
de
mo
nstrate t
he p
e
rfor
ma
nc
e of
the
prop
o
s
ed a
ppr
oach
and c
o
mp
are t
h
e
m
to D
C
I (Diag
o
n
a
lly-
l
oa
de
d
Conj
ug
ate corr
elati
on
matrix Inverse p
o
w
e
r) meth
od.
Ke
y
w
ords
: fourth-order cumulants,
DCI, MUSIC, DOA estim
a
tion
Copy
right
©
2014 In
stitu
t
e o
f
Ad
van
ced
En
g
i
n
eerin
g and
Scien
ce. All
rig
h
t
s reser
ve
d
.
1. Introduc
tion
Dire
ction
of
arrival
(DOA) estim
a
tion i
s
o
ne
of the
impo
rtant
co
ntents
of a
r
ray sig
nal
pro
c
e
ssi
ng, b
e
ca
use of its importa
nt appl
icatio
n
s
in ra
dar, so
na
r, co
mmuni
cation
s, etc.
There are m
any array mo
dels, an
d alg
o
rithm
s
are a
v
ailable for e
s
timating the
DOA of
sou
r
ces. Am
ong the
s
e,
Multiple Sign
al Cla
ssi
fi
cati
on (M
USIC)
[1] algorithm
that belong
s to
sub
s
p
a
ce m
e
thod
of eig
enstructu
re
i
s
the
supp
e
r
resolution
method, a
n
d
it ha
s g
o
o
d
perfo
rman
ce
and
widely u
s
ed.
Ho
weve
r, MUSIC a
n
d
som
e
mo
di
fied versi
o
n
s
of MUSIC [2
-8]
requi
re th
e n
o
ise
ch
ara
c
t
e
risti
cs
of the se
nsor
s b
e
kn
own an
d the total n
u
mbe
r
of si
g
nals
impingin
g
on
the array to
be kno
w
n o
r
to be
exact
l
y estimated
in advan
ce.a
nd al
so the
r
e
is
algorith
m
whi
c
h is
able to
achi
eve high
accuracy
[11,
12]. We pro
pose a ne
w
dire
ction findi
ng
algorith
m
to overcome th
e aforem
enti
oned
sho
r
tco
m
ings by u
s
i
ng high
-o
rde
r
cumula
nts m
a
trix
to re
pla
c
e th
e conju
gate
sou
r
ce
co
rrel
ation m
a
trix [
9
]. The
algo
rithm ha
s l
o
w co
mputatio
n
a
l
compl
e
xity in com
pari
s
o
n
with the M
U
SIC al
go
rith
m and
still h
a
s g
ood
pe
rforma
nce. Both the
comp
uter
si
mulation
s an
d the expe
ri
ment of t
he
dire
ction findi
ng sy
stem h
a
ve bee
n giv
en to
illustrate the
performan
ce
of the algorithm.
The rest of t
h
is p
ape
r is
orga
nized a
s
fo
llows. Sect
ion 2 introdu
ce
s the
syste
m
model
and
DCI me
thod. Se
ction
3, the
ne
w algo
rithm i
s
de
scribe
d i
n
detail. Se
ction 4
pre
s
e
n
ts
simulat
i
o
n
re
sult
s t
h
at
sho
w
t
he ef
f
e
ct
iv
ene
ss of
the
prop
osed al
g
o
rithm. Finall
y
, we con
c
lu
de
this pap
er in
se
ction 5.
Thro
ugh
out t
he p
ape
r, lo
wer-ca
s
e
bol
dface
it
alic l
e
tters
den
ote
vectors; u
p
per-case
boldfa
c
e itali
c
letters
re
prese
n
t matri
c
es, a
nd l
o
wer a
nd
upp
e
r
-ca
s
e itali
c
l
e
tters stan
d
for
scalers. Th
e symbol
is u
s
ed for
conj
ug
ation ope
ratio
n
s a
nd notati
on
T
x
and
H
x
re
pre
s
ent
transpo
se a
n
d
co
njugate t
r
an
spo
s
e,
re
spe
c
tively. We use
E
x
,
cu
m
x
and
to indicate the
expectatio
n
o
perate
r
, the cumulant
s
and
kron
ecke
r produ
ct, sepa
ra
tely.
2. Sy
stem Model and DCI
2.1. Sy
stem
model
Assum
e
there are
M
far-field
narro
wb
and
sign
als
,
1
,
2
,
...
.....
...,
m
tm
M
s
, impinging on
auniform line
a
r array (ULA
) with
N
senso
r
s from differe
nt directio
ns
,
1
,
2
,
.
...
..
...
,
.
m
mM
The array out
put vector
co
uld be re
presented by:
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 6, June 20
14: 4237 – 4
242
4238
1
M
mm
m
ts
t
t
xa
n
(1)
Whe
r
e
12
si
n
s
i
n
si
n
,
,
.
.
.
...
...
..
...
...
..
.
mm
N
m
T
jj
j
m
ee
e
a
is the steerin
g vect
or of the
m
th sign
als,
21
/
n
nd
, and
is the wavelength of
sign
al,
d
is the element spa
c
e.
Then rewritin
g Equation (1
) in matrix form, we obtain:
ts
t
t
xA
n
For
1
,
2
,
..
..
..
..
..
,
tK
(2)
Whe
r
e:
12
,
,
..
...
..
...
..
T
N
tx
t
x
t
x
t
x
is
1
N
obse
r
vation vector
(sn
apshot vector),
12
,
,
.......
........
,
T
M
Aa
a
a
is
NM
array manifold matrix,
12
,
,
...
..
..
...
..
..
..
...
..
,
T
M
ts
t
s
t
s
t
s
is
1
M
signal ve
ctor,
and
12
,
,
...
..
..
..
..
...
..
.
T
N
tn
t
n
t
n
t
n
is
1
N
noise
vector.
The cova
ria
n
c
e matrix of the data re
ce
i
v
ed by the se
nso
r
array, denoted by:
H
xx
E
tt
Rx
x
2
ss
RI
(3)
Whe
r
e
H
ss
s
RA
Γ
A
and
H
s
Et
t
Γ
ss
denote
s
the covarian
ce m
a
trix of radiatin
g sign
als
and
I
is the
NN
identity matrix.
The co
njug
ate covari
anve
matrix of the array, denote
d
by:
..
TH
xx
s
s
Et
t
R
Rx
x
A
A
(4)
Whe
r
e
.
T
ss
E
tt
Rs
s
is call
ed the co
njug
ate sou
r
ce co
rrel
a
tion matri
x
.
2.2. Diagona
lly
Loaded Conjugate Co
rrelation Ma
trix In
v
e
rse Po
w
e
r Me
tho
d
(DCI)
This m
e
tho
d
ca
n e
s
timate DOA of narro
wband
non
circula
r
si
gnal
s with
out
eigen
de
comp
osition
an
d
source
enu
me
ration.Thi
s
m
e
thod
exploit
s
the
conju
g
a
te correlatio
n
betwe
en the receive
d
sig
n
a
l
s at different
sen
s
o
r
po
sitions.
Here we u
s
e
only conj
ugat
e array correl
ation matrix
..
TH
xx
s
s
Et
t
R
Rx
x
A
A
First we sq
ua
re the above
eguatio
n to obtain He
rmiti
an matrix.
..
.
.
H
TH
H
H
xs
x
x
x
x
ss
ss
CR
R
A
R
A
A
R
A
A
C
A
(5)
The DOAs ca
n be obtain
e
d
by searchin
g
the
peaks of the followin
g
spatial
spe
c
trum.
H
L
DC
I
L
H
x
D
aa
S
aa
(6)
Whe
r
e:
1
xx
N
DC
I
(
7
)
Whe
n
L
, the DCI method ap
proa
ch
es M
U
SIC [9].
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
DOA Estim
a
tion by F
ourth
-Orde
r
Cum
u
lants with
out Source… (El
hafiz Abdall
a
Bakhit Yago
u
p
)
4239
3. The Propo
sed Algori
t
h
m
.
The p
r
o
b
lem
is th
at give
n the
array
output
,
1
,
2
,
....
...
tt
K
x
where
K
denote
s
the
numbe
r of sn
apshots, e
s
timate the DO
A paramete
r
,
1
,
2
,
....
....
,
m
mM
of the impingi
ng sig
nal
s.
We shall work with a fou
r
th orde
r cumul
ant of
received array outpu
t which
can b
e
expre
s
sed
as:
12
3
4
1
2
3
4
1
2
3
4
,,
,
c
u
m
x
x
x
x
E
x
x
xx
E
x
x
E
xx
13
2
4
1
4
13
E
xx
E
x
x
E
xx
E
x
x
(8)
We can u
s
e
matrices to e
x
press
cumu
l
ants by usi
n
g
Krone
cker p
r
odu
cts.
.
HH
x
EE
E
Cx
x
x
x
x
x
x
x
..
HH
EE
xx
xx
(9)
Whe
r
e
x
C
is the fourth ord
e
r
cumul
ants of
sign
al vector
x
.
Con
s
id
erin
g the pro
p
e
r
ties
of Krone
cker
prod
uct
s
we
have.
H
xs
Cb
C
b
(10)
Whe
r
e
s
C
is fourth orde
r cum
u
lant matrix o
f
signal,
..
.
HH
HH
s
CE
E
E
E
E
ss
ss
ss
s
s
s
s
s
s
(11)
and
12
....
....
....
...
N
bb
b
b
11
...
....
...
....
...
...
NN
aa
a
a
(12)
We cons
truc
t further the following f
ourth
orde
r cumula
nts matri
c
e
s
[10].
..
HT
T
T
xx
x
x
x
x
x
x
x
Q
Q
xx
xx
E
t
t
vec
vec
CZ
Z
R
R
R
R
I
R
R
I
(13)
Whe
r
e
tt
t
ZZ
Z
,
tt
t
ZX
X
,
""
ve
c
M
denotes
1
PG
vector f
o
rme
d
from
the elem
ents of a
P
G
matrix
M
by
st
ac
kin
g
it
s c
o
lum
n
s,
a
nd
1
,
1
,
...
....
....
.,
1
T
Q
I
which is
a
1
Q
vector
with a
ll element
s b
e
ing on
e.In a
ddition
""
denot
es the
khat
ri-Rao p
r
o
d
u
c
t (element
Krone
cker p
r
odu
ct) i.e:
11
,
.
.
....
....
....
GG
M
Hm
h
m
h
(14)
In which
12
,
,
....
....
....
,
G
M
mm
m
and
12
,
,
...
.....
....
...,
G
Hh
h
h
.
We then fo
rm
the following
matrix:
1
xx
N
DC
I
(15)
Whe
r
e
is a scalar
whi
c
h
is sm
aller th
an the lea
s
t eigenvalu
e
of
x
C
,
N
I
denotes a
22
NN
identity matri
x
. Let
1
M
m
m
v
be the do
minant
eigenve
c
tors of
x
C
according to
M
largest
eigenvalu
e
s
1
M
m
m
. W
h
er
es
1
NM
m
m
u
be the rem
a
ining
NM
eigenve
c
tors
asso
ciated wi
th the
zero eige
nval
ues, it then follow that:
1
11
MN
M
H
xm
m
m
m
m
mm
Dv
v
u
u
(16)
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 6, June 20
14: 4237 – 4
242
4240
From the
sub
s
pa
ce o
r
tho
g
onality we
kn
ow that
()
0
H
mk
uB
for
1
,
....
.....,
mN
M
and
1
,
..
..
..
..
..
..,
kM
. Thu
s
the
di
rectio
n
of arrival ca
n b
e
o
b
tained
by
searchin
g the
pea
ks of the
followin
g
spat
ial spe
c
tru
m
expre
ssi
on.
1
H
F
O
CD
CI
L
H
x
bb
S
bD
b
22
11
()
LH
MN
M
LH
LH
mm
m
mm
bb
bv
b
u
(17)
Whe
r
e
L
x
D
denotes the
L
-fold powe
r
of
x
D
and
L
is an integ
e
r, and
mm
.
Let
mm
then:
22
11
()
H
FOC
D
C
I
MN
M
LH
H
mm
m
mm
S
bb
bv
b
u
(18)
Whe
n
L
, then
0
L
m
, and (18) to
be MUSIC al
gorithm u
s
in
g
fourth ord
e
r
cumul
ants
x
C
.
4. Simulation Resul
t
s an
d Analy
s
is
In this sectio
n, com
puter
simulatin
s
a
r
e
prese
n
ted t
o
demo
n
st
rat
e
the pe
rformance of
fourth order cumula
nts
DCI meth
od.
The fo
llowi
ng perfo
rma
n
ce m
easures a
r
e u
s
ed
to
investigate th
e perfo
rman
ce of DOA esti
mation tech
ni
que.
F
i
r
r
s
t
ly th
e
po
w
e
r
sp
ec
tr
um p
l
o
t
s
,
wh
en
t
here
we
re
five source
s i
m
pingin
g
on
the ULA
with di
re
ction
s
[-4
0
-20 0
2
0
40], the S
N
R is
10
dB an
d the
sna
p
sh
ot numb
e
r i
s
1000. T
he
sp
atial
spe
c
tru
m
of FOC-DCI can
be estimate
d
corre
c
tly as shown in the Figure 1.
Figure 1. Estimate Five Sources u
s
in
g F
O
C-DCI Algo
rithm, SNR=
10 dB, Number of
Snapshots=5
0
0
Secondly the estimation accura
cy, assume the array is illumi
nat
ed by two si
gnals from
[-30 50]. All result
s are av
erag
ed via 1000 Monte
-
carlo sim
u
latio
n
run
s
. The root-me
an-sq
uare
err
o
r (R
MSE)
v
e
rs
us sig
n
a
l–to-n
ois
e
r
a
tio
(
S
NR)
of FOC-
DC
I, SOC
-
DC
I and FOC
-
D
C
I,
N
C
-
Root MUSIC, NC-s
tandard ESPRIT
as
s
h
own in Figure 3 and
Table 1 res
p
ec
t
i
vely. And th
e
RMSE versu
s
the
numb
e
r of sn
ap
shot
s are
depi
cted
in Figu
re
2 a
nd Ta
ble 2.
We
can
see f
r
om
Figure 2-3 a
nd Tabl
e 1-2, that
generall
y
the FOC-DCI method h
a
s bette
r esti
mation preci
s
ion
comp
ared wit
h
SOC-DCI, NC-Root
MUSIC and NC-stand
ard ES
PRIT.
-8
0
-6
0
-40
-20
0
20
40
60
80
-60
-50
-40
-30
-20
-10
0
direc
t
ion
of
arriv
a
l(
o
)
F
O
C
-
D
C
I
s
pec
t
r
um
(
d
B
)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
DOA Estim
a
tion by F
ourth
-Orde
r
Cum
u
lants with
out Source… (El
hafiz Abdall
a
Bakhit Yago
u
p
)
4241
Figure 2. RM
SE of the FOC-DCI an
d DCI
Estimator versu
s
Num
b
e
r
of Snapsh
o
ts
Figure 3. RM
SE of the FOC-DCI an
d DCI
Es
timator vers
us
SNR
Table 1. The
RMSE of FOC-DCI, NC-st
anda
rt ESPRIT and NC-Root
MUSIC versus SNR
SNR in (dB)
RMSE in degree
s
NC-standa
rt ESP
RIT
NC-R
O
OT M
U
SIC
Proposed metho
d
(F
OC-
DCI)
5 0.1368
0.1301
0.1252
7 0.1037
0.08497
0.09739
9 0.06714
0.0537
0.04995
11 0.05659
0.05036
0.03828
13 0.03853
0.03666
0.03121
15 0.02694
0.02876
0.02121
17 0.03182
0.02538
0.007071
19 0.02476
0.02138
0.007071
Table 2. The
RMSE of FOC-DCI, NC-st
andart
ESPRIT and NC-Root MUSIC
versus Num
b
er of
Snapshots
Number of sna
p
shots
RMSE in degree
s
NC-standa
rt ESP
RIT
NC-R
O
OT M
U
SIC
Proposed metho
d
(F
OC-
DCI)
50 0.09845
0.06974
0.07364
150 0.05843
0.05525
0.05525
250 0.047
0.03367
0.02828
350 0.03567
0.03466
0.03466
450 0.03685
0.02931
0.02121
550 0.02256
0.02079
0.01932
650 0.0328
0.02526
0.007071
750 0.02059
0.01678
0.01932
850 0.02623
0.02022
0.007071
950 0.02388
0.02254
0.01236
4. Conclusio
n
We
pre
s
e
n
te
d ne
w di
re
cti
on findin
g
al
gorithm
for
n
on
circula
r
si
gnal
s, which
based
on
the
fou
r
th
o
r
d
e
r cum
u
lant
s of
the data re
ceived
by the array. It’s
a f
a
s
t
DOA
es
timation method,
whi
c
h d
o
e
s
not re
qui
re dete
c
ting
the num
b
e
r of in
cid
ent sig
nal
s and
perfo
rming
eigen
de
comp
osition to a
c
hieve sign
al
noise
s
epa
ration. An important topi
c which rem
a
ins
unsolved i
s
on the o
p
timum sel
e
ctio
n
of
. Both the theoretical
formulatio
n
and
comp
ute
r
simulatio
n
sh
ow that the F
O
C-DCI has
agoo
d perfo
rmance
Ackn
o
w
l
e
dg
ements
This work was supp
orte
d by Nation
al Natural Scien
c
e Fo
un
dation of Ch
ina (no.
6107
2098, 6
1
0720
99, 613
3
1019
).
100
150
20
0
25
0
30
0
35
0
400
450
500
0
0.05
0.1
0.15
0.2
0.25
num
ber
of s
napshot
s
R
M
S
E
(
degr
ee)
F
O
C-
DC
I
SO
C-DC
I
0
2
4
6
8
10
12
14
16
18
20
0
0.
05
0.
1
0.
15
0.
2
0.
25
SN
R
(
d
B
)
R
M
SE(
o
)
FO
C-
DCI
S
O
C-
DCI
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 6, June 20
14: 4237 – 4
242
4242
Referen
ces
[1]
Schmidt RO. Multiple
em
itter l
o
catio
n
a
nd s
i
g
nal
par
ameter
estimatio
n
.
IEEE Transactions
on A
n
tenas
and Pro
p
a
gatio
n
. 1986; 3
4
: 27
6-28
0.
[2]
Gouno
n
P, Ad
net
C, Gal
y
J.
Loc
alis
atio
n
a
ngu
lair
e
de s
i
g
nau
x n
on c
i
rcul
arizes.
T
r
a
i
tme
n
t du
si
gna
l
.
199
8; 15: 17-2
3
.
[3]
Abei
da H, D
e
l
m
as JP. MUSIC-like
estimati
on of
d
i
rectio
n
of arriva
l for
non-c
i
rcul
ar so
urces.
IEEE
T
r
ansactio
n
s o
n
Sign
al Proc
e
ssing
. 20
06; 54
: 2678-2
6
9
0
.
[4]
Chev
ali
e
r P, F
e
rréol
A, Alb
e
r
a
L. H
i
gh
reso
luti
o
n
d
i
rectio
n
findi
ng from
h
i
gh
er or
der sts
t
ics: the 2q-
MUSIC algorithm.
IEEE
transaction Signal Pr
ocessing
. 20
06
; 54: 2986-
299
7.
[5]
T
ang JH, SI
XC, Ch
u P. Imp
r
oved
MUSIC
alg
o
rithm
bas
e
d
o
n
fo
urth-ord
er cum
u
la
nts.
Systems
an
d
Electron
ics En
gin
eeri
n
g
. 20
1
0
; 32; 256-
259.
[6]
W
e
i P, Xiao
X, Li L.
T
he four
th-ord
er
cumul
a
nts
based
sp
e
c
tral estim
a
tion
metho
d
an
d it
s ap
plic
atio
n
to directio
n find
ing.
Jour
nal of
electro
n
ics
. 19
95; 12(4): 3
30-
335.
[7]
W
ang Q, Che
n
H, Z
hao G. A
n
impr
oved
dir
e
ction
fi
ndi
ng
alg
o
rithm b
a
se
d on T
oep
litz
Appro
x
imatio
n.
Sensors
. 20
13;
13: 746-7
57.
[8]
He Z
,
Li Y, H
uan
g Z
.
A mo
difie
d
hig
h
ord
e
r cumula
nt MUSIC alg
o
rith
m for signal D
O
A estimation.
Journ
a
l of Elec
tronics
. 200
0; 17(4): 31
9-3
2
4
.
[9]
Qi X, Xu Y, Fu S, W
ang Y, Liu Z
.
High r
e
sol
u
tion DOA
estimati
on of
non-c
i
rcul
ar si
gna
ls w
i
thout
source e
n
u
m
er
ation a
nd e
i
ge
ndec
o
m
pos
itio
n
. Proc. W
C
SP. 2009; 9: 1-4.
[10]
Xu Y, Li
u Z
.
Noncircu
larit
y
rest
oral for multi-ant
enn
a blin
d be
amfor
m
ing.
Multid
i
m
Syst Signa
l
Processi
ng
. 20
10; 21: 13
3-16
0.
[11]
Lin
g
w
e
n Z
.
S
parsit
y
base
d
AOA Estimation f
o
r Emitte
r Loc
aliz
ation.
T
E
LKOMNIKA Indo
nesi
a
n
Journ
a
l of Elec
trical Eng
i
ne
eri
n
g
. 201
2; 13: 7
69-7
74.
[12]
Lan
g
x
ia
ng Z
,
Yao
w
u S, Yira
n S, Lifei D, Hong
w
e
i S. Elec
tromagn
etic Ve
ctor
Sensor arr
a
y
param
ete
r
estimatio
n
method.
T
E
LKOMNIKA Indon
esi
an Jour
nal
of Electrical E
ngi
ne
erin
g.
201
3; 11
: 2860-2
8
6
8
.
Evaluation Warning : The document was created with Spire.PDF for Python.