Beruflich Dokumente
Kultur Dokumente
Severe Interference
Tri-Tan Van Cao
Defence Science & Technology Organisation
PO Box 1500, Edinburgh, SA 5111, Australia. E-mail: tan.cao@dsto.defence.gov.au
sample in the range/Doppler dimensions is called a cell. The discarded in addition to the largest ranked samples [11],
test cell is the cell at which a detection decision has to be etc. A variety of combinations of different CFAR algorithms
made. The interference power in each test cell is estimated are also proposed, e.g., [12], [13], [14], etc.
using its surrounding cells which are termed reference cells. However, if the actual number of interfering samples ex-
In the range-Doppler map, the reference cells form a reference ceeds the assumed value, robustness of these modified CFAR
window. The interference estimation is simply the sample algorithms is no longer guaranteed. In addition, none of the
mean of the power available in the cells within a reference above mentioned CFAR algorithms exploits the statistical
window. The adaptive threshold is then formed by multiplying properties of the sample in the test cell during reference sample
the interference estimate with a constant, the value of which is selection.
determined by the required false alarm rate. A few immediate Recently, a CFAR detection algorithm known as Switching
neighbours (known as guard cells) on each side of the test cell CFAR (S-CFAR), that takes into account the amplitude of the
are excluded from the estimation to prevent possible power test cell for reference sample selection, has been proposed
spill-over from the test cell. in [15]. The unique structure of the S-CFAR is its switching
Under the condition that the sample in each reference action, which allows the processor to select: (i) either only
cell is independent and identically distributed (iid) and is the small thermal noise samples for interference estimation
governed by the exponential distribution, performance of the when interferers are present, or (ii) the whole CFAR window
CA-CFAR processor is optimal (in the sense that the detection to minimise the false alarm rate when there are no interferers.
probability is maximised for a given false alarm rate) when Continuing this research theme, closed-form analysis of the
7
-
: + M c
[
D b
7
- V
+ M
on Swerling I target/background model is first given in this are the probabilities of detection when there are exactly 7
-
paper. Application of the S-CFAR for target detection in the samples in . is the statistical average performed over ,
-
5
presence of an unknown number of interfering samples is then the probability density function (pdf ) of the sample in the a
on .
2. T HE S-CFAR A LGORITHM .
X
A generic CFAR processor receives input from the square law A. Computation of
[
D b
.
detected video range samples (i.e., range cells). Assume that Let be the number of interfering samples contained in a
g
a
the amplitude in each cell is an iid random variable with an CFAR window. One has:
exponential pdf described by: [ n p q [
D r
X X
(1)
[
D b
k m
[
D
q
b
(6)
U
D
>
$ &
7
-
+
$
g
(7)
( (
q
b
can k
,
-
[
D
q
b
[
D
q
b
[
D
q
b y
X
[
D
q
b
(8)
are partitioned into two sets and as follows: , - , /
X
k k k
[
D
q
b
s
s
-
7
1
0
(2) , , and
-
is the probability that the test cell survives {
[
D
q
b
,
-
following section,
X
and
X
are computed.
- 2
,
x
{
D b D b
[ q [ q
5 5 5
+
k
0
1
1
/
,
(9)
2
0 0 0
@
5
target is declared in the test cell if: By setting in (9), the probability that a thermal noise
2
2
(i)
8
(3)
-
(10)
0
: < : + M
b
F H O P
-
7 7
5
-
@ B C D
- >
7
R
(ii)
sample is sorted to is: ,
-
(4)
/
0
: + M
<
(11)
F H
- V
7 7
b
+ R S
0 U
5
/
>
<
/
+
M
,
-
Inequalities (3) and (4) mean that the S-CFAR switches is then:
between the sample set and the whole reference window, ,
-
p
(12)
g
C s
b b b
s
/
k
/
k
5
+
From (3) and (4), the detection probability of the S-CFAR is: s
b
- 7 -
+
r
D D
[ p n [
X X
b b
+ +
c
M D b D b
[ [
5
- -
- /
s
$
< < ]
S
R S
k k
-
7
U
D
U
D
> >
R S
-
(13)
/ a a
(5) 2 2
S ] _
is then: Combining (19) with (6), and then averaging over the pdf of
, one has:
W
(14) 8
4
Let:
C !
"
" + "
(15)
C $ R
#
+
" " @
! !
%
C
- - - -
H
0 0 0 0
! !
>
C $ R
(21)
#
samples sorted to ! %
&
!
+ < <
8 8
9 ; 9 ;
4 4
= =
- -
0 0
"
+ "
5 7 >
(
9 ;
I
D
N N
G
? ? H
H
H
@ B F F
v
2 2
C C
R C $ R
3
N
H H
?
U ? ?
@ B F
>
M K + F v + + F
N
5 7
"
C R C $ R
J
$ C $ R $ $ C $ R
8
9 ;
1
U ?
4
(22)
K M
K W F
>
"
>
2
5 7 >
(16)
in which and are defined in (18) and (20), respec-
v + +
where satisfies X
$
U
W
, and U
, ,
C
tively.
N
$ C
$
C
for `
C
. Using (16) , one
has:
B. Computation of
.
a
One has:
D e
8
4
c ( f
!
(23)
"
+ < <
8 8
9 ; 9 ;
4 4
= =
- -
0 0
"
+ "
5 7 >
@ B
D
F
?
G H
I
=
9 ;
F
detection when there are samples sorted to ( ),
&
`
2 2
C C
? H H U
F K
5 7
C R C $ R $ C
J
+
8 8
9 ;
4 4
(24)
K M @ B
J
>
"
N
2
u
5 7 >
W v + F
8
r s t
9 ; q
(17)
J o
u {
2
C >
K @ B
5 7
u
"
are samples in , of which are interfering samples;
&
(18)
+
v X }
|
$
&
G
H H
C $ R
, one has:
$ 5 >
$
$
(25)
" " " "
! ! % ! % !
a
! ! ! !
@
+
where and are the thermal noise samples sorted to #
#
C
! !
- - - -
& &
0 0 0 0
>
"
+ "
! !
% %
C $ R
& &
N
H H
?
F
J
U
R C $ R
M K
- -
"
0 0
C
"
+ "
H
u
>
(
W v + F
8 8 8
r s t r s t r s t
9 ; 9 ; q
=
I 9 ;
8
9 ;
4 4 4
4
J o
=
u {
2 2
2 2
> K @ B
C C
"
+ +
I 9 ; 8 8
u 9 ;
4 4
5 7 5 7 >
5 7
(19)
2
G
?
@ K @ B F
5 > 5 7
where:
N
?
H
5
@
1
W F
N
2
4
8
9 ;
$ $
J
N
G
U ?
H
M @
K @ B F
N
5 > 5 7
(20)
"
+ $ @
H
$ $ $
C $ R
(26)
where satisfies . Using (26) one in which and are defined in (20) and (28), respec-
`
/ /
has: tively. Finally, substituting (21) and (30) into (5), one has:
&
!
d d d d
%
# +
$ $
P
N F
3 3
6
y
1 1
E
. .
4 4 . / .
- - - -
F
6
q
/ .
- -
~
< >
;
3 3
;
< >
? I I
6
1 1 1 1
8 8
,
, ,
6 6
3 4
3 4 4 4
9
6
q
/
L ; < > P
< > Q
P
N
I
6 N F
,
8
B B
!
J
G
9
+
-H
6
y
,
7 C O
C E
. . / .
[ - - - -
C E
%
!
F
6
q
>
% `
] / +
P
U W X
< > T
,
,
/
b
3 3
6
a J
C E 7
I I
E 7
6
. 1 1 1 1
[
- %
> d
C O
3 4
3 4 4 4
H d
B
!
6 q
(32)
]
+
%
, ,
(27)
d d
4. S-CFAR D ESIGN U NDER S EVERE I NTERFERENCE .
(28)
%
/
i
,
j
,
-
.
tire search region of an electronic support system which is
designed to detect signals from all possible directions [17]).
/
3 3
I I
4 4 4 4
3
3
6
q
P
< > U W X
s
8 ,
-H
C O
P
< > T
U W X
H
]
%
`
/ +
[
>
occupying more than half of the clear region), not much can
be done from a CFAR perspective.
,
6
C E 7 a J
[ -
>
.
u
!
]
,
%
,
%
,
H
]
,
H j
conventional CFAR window with cells is employed.
6
]
still have the same statistics as that of the noise in the test
x
N F
P
cell. Since the number of interfering samples is unknown and
6 y
,
the S-CFAR detector can be tuned to tolerate a large number
6
q
-
.
F
-
.
-
.
of interfering samples (as demonstrated later in this paper),
the largest CFAR window with cells is considered.
~
3 3
1
I
3 4
1
3
I
4
1
4
1
4
6
F
d
d
With the nominal value (half window size), Fig.
1 shows
!
i +
d
%
6 q
6 6
(30)
$ $ $
probability
i
6
i
,
i i
j j
H
H H
u u
I
d
! !
G J / + / / C E 7 +
% %
, ,
6 q 6 q
C E 7
Detection Probability
−3 S (2N=64, N =16)
i
0.6
log(Pfa)
0.3
−4 CA (2N=16, N =0)
i
0.4 0.4
−5 α =0.5
0.6
0.2
−6
−7 0
10 12 14 16 0 5 10 15 20 25 30
β SNR (dB)
Fig. 1: Selection of for window size . Fig. 3: S-CFAR and CA-CFAR detection performances.
0 1
NT=32 S
−1 27 OS
22 0.8
−2 CA
17
Detection Probability
12
−3 0.6
log(Pfa)
Ni=0
−4
−5 0.4 Ni=32
Ni=8
−6
0.2
−7
−8 0
5 10 15 20 0 5 10 15 20 25
β SNR (dB)
Fig. 2: Selection of for window size and . Fig. 4: Detection performances in worst case.
)
>
too far from the CA-CFAR curve. is evident that the detection performance of the large window
After selecting the CFAR window size and the sorting S-CFAR processor at SNR above 10dB is still better than that
threshold multiplier , the S-CFAR false alarm probability of the CA-CFAR processor in terms of CFAR loss.
Fig. 4 shows the detection probabilities at
of
0 2
' > ?
D
E G
(
( '
,
hence more CFAR loss in an homogeneous environment) is same SNR as that of the target. Parameters of the OS-CFAR
are selected such that it has the same CFAR loss as that of
) *
environment. The CA-CFAR fails with only contaminated H
)
0 2
.
) *
(
) 4
, '
, , and
.
) *
(
8
(
9
(
) 4
5. P ERFORMANCE A NALYSIS . case even with the target SNR of 20dB). These results show
From an interference estimation point-of-view, as a reference that the problem of target masking due to severe interference
sample is located further from the test cell, it is less repre- can not be solved by simply extending the CFAR window of
sentative of the noise level in the test cell. For this reason, the CA-CFAR or OS-CFAR detectors.
to be realistic in the following performance comparisons, a Fig. 5 shows the detection thresholds at of
0 2
quarter of the designed S-CFAR window is expected to contain the large window S-CFAR ( , , ' > ?
(
8
(
9
D
E G
) and OS-CFAR (
) 4
,
) in worst
' > ? '
of the designed S-CFAR processor ( ) case. There are three groups of targets: ranges [17-23], [25-
' > ?
0 2
)
>
36], and [40-51]. Nearly half of the range profile is filled with
@
20
R EFERENCES
15 [1] M.I. Skolnik. Introduction to Radar Systems. McGraw-Hill Book
Company, USA, 3rd edition, 2001.
10 [2] H.M. Finn and R.S. Johnson. Adaptive detection mode with threshold
control as a function of spatially sampled clutter level estimate. RCA
5 Review, 29(3):414–464, September, 1968.
[3] P. Antonik, B. Bowies, G. Capraro, and L. Hennington. Intelligent Use
of CFAR. Kaman Sciences Corporation, USA, 1991.
0
0 15 30 45 60 70 [4] G.V. Trunk. Range resolution of targets using automatic detectors.
Range IEEE Transactions on Aerospace & Electronic Systems, 14(5):750–755,
September, 1978.
Fig. 6: CFAR detection thresholds.
[5] G.V. Hansen and J.H. Sawyers. Detectability loss due to greatest-of
selection in a cell averaging CFAR. IEEE Transactions on Aerospace
& Electronic Systems, 16:115–118, 1980.
[6] H. Goldman and I. Bar-David. Analysis and application of the excision
interfering targets/jamming samples of the same SNR=15dB. CFAR detector. IEE Proceedings, Radar, Sonar & Navigation, volume
135, Part F, pages 563–575, December, 1988.
The large window OS-CFAR detects nothing due to the large [7] H. Goldman. Performance of the excision CFAR detector in the presence
CFAR loss as analysed above. The large window S-CFAR of interferers. IEE Proceedings, Radar, Sonar & Navigation, volume
detects one (largest) sample in the first group, two samples 137, Part F, No. 3, pages 163-171, June, 1990.
[8] M.A. Khalighi and M.M. Nayebi. CFAR processor for ESM systems
in the second group, and one sample in the third group. applications. IEE Proceedings, Radar, Sonar & Navigation, volume 147,
No. 2, pages 86-92, April, 2000.
Fig. 6 shows the detection thresholds at of
), and SO-CFAR (
). There
are three groups of targets: ranges [16-20], [27-40], and [46- multiple target situations. IEEE Transactions on Aerospace & Electronic
50], all of the same SNR=18dB. The OS-CFAR detects only Systems, 13(4):338–343, July, 1977.
two samples in the first group, whereas it misses all samples [11] P.P. Gandhi and S.A. Kassam. Analysis of CFAR processors in
nonhomogeneous background. IEEE Transactions on Aerospace &
in other groups. The missing of the second group is due to Electronic Systems, 24(4):427–445, July, 1988.
the fact that the actual number of interfering samples (i.e., 14) [12] M.E. Smith and P.K. Varshney. Intelligent CFAR processor based on
is greater than the nominal number (i.e., 8) the OS-CFAR is data variability. IEEE Transactions on Aerospace & Electronic Systems,
36(3):837–847, July, 2000.
designed to tolerate. The missing of the third group is due [13] S. Hinomas and M. Barkat. Automatic censored CFAR detection for
to the low SNR of the target/interfering samples. In general, nonhomogeneous environments. IEEE Transactions on Aerospace &
as analysed in [11], an SO-CFAR can detect the first and the Electronic Systems, 28(1):286–304, January, 1992.
[14] M.A. Khalighi and M. Bastani. Adaptive CFAR processor for nonhomo-
last samples in a group of targets. However, in this case SO- geneous environments. IEEE Transactions on Aerospace & Electronic
CFAR can only detect three largest samples in the first group. Systems, 36(3):889–897, July, 2000.
The missing of the second group is due to the presence of [15] T.V. Cao. A CFAR thresholding approach based on test cell statistics.
IEEE 2004 Radar Conference Proceedings, pages 349-354, April 2004,
the nearby first and third groups, while the missing of all the Philadelphia, USA.
third group is due to the low SNR of the samples at the right [16] G. Morris and L. Harkness. Airborne Pulsed Doppler Radar Systems.
edge of the group. The designed S-CFAR detects four samples Artech House, USA, 1996.
[17] D.C. Schleher Electronic Warfare in the information Age. Artech House,
in the first group, nearly most of the samples in the second USA, 1999.
group, and up to three samples in the third group.