Professional Documents
Culture Documents
=
=
(1)
And the Doppler shift of moving targets relative to the
AEW radar is expressed as
( ) 2 cos cos
a t
d
V V
f
|
+
=
(2)
Where
a
V and
t
V are the flying speed of AEW and
target respectively. is the angle between the course of
AEW and the line of sight of radar. | is the angle of the
course of target and the line of sight of radar. u and are the
azimuth angle and the elevation angle of the AEW radar
beam respectively. H denotes the altitude of the platform. R
is the slant range from the platform to a clutter patch.
For the convenience of the statement about the problem,
it might as well be supposed that one AEW radar detects one
target which is flying at constant speed. The scenario is given
in Figure 1.
0 0.5 1 1.5 2
x 10
5
0
0.5
1
1.5
2
2.5
3
3.5
x 10
5
X direction/m
Y
d
i
r
e
c
t
i
o
n
/
m
AEW
T
a
r
g
e
t
___________________________________
978-1-4244-8443-0/11/$26.00 2011 IEEE
Table 1.The system parameters of AEW radar
Parameters of AEW radar value
Speed of AEW radar 140m/s
Carrier frequency 1.3GHz
Pulse duration V
Bandwidth 1MHz
Pulse repetition frequency 2435Hz
Doppler threshold 100Hz
Table 2. The parameters of target
parameters of target value
Speed of target 500m/s
Radar cross section 0.1m
2
0 20 40 60 80
-4000
-3000
-2000
-1000
0
1000
2000
3000
Bl i nd
doppl er
zone
Sample number
D
o
p
p
l
e
r
f
r
e
q
u
e
n
c
y
/
H
z
Figure 2. The Doppler history of target
relative to single radar
-1 0 1 2 3
x 10
5
-3
-2
-1
0
1
2
3
4
x 10
5
T
a
r
g
e
t
X direction /m
Y
d
i
r
e
c
t
i
o
n
/
m
Detected points
undetected points
Figure 3. The result of single radar detection
3. Principles of target detection based on multi-AEW data
fusion
3.1 Network Structure of the detection system
The target detection structure based on multi-AEW data
fusion can be classified as two kinds of centralized and fully
distributed. In the former structure, observations of the local
sensors are directly transmitted to the data fusion centre.
Though it can acquire well detection performance, it is highly
constrained by the capacity and error of the channels between
the local sensors and the data fusion centre and aggravates the
computation load of the data fusion centre.
Therefore, the distributed system is applicable when all
sensors are located at the same site, or the capacity of the
channels between the sensors and the data fusion centre
allows practically error-free transmittal of the sensor
observations to the central processing site.
In the latter system, each local sensor makes local
decisions and then transmits them to the data fusion centre.
Compared with the centralized system, it decreases the traffic
of the channels. But the system performance worsens because
the transmittal of the local observations is not allowed and
partial decisions are made at each local site. Despite the
declining performance of the fully distributed system to a
degree, it possesses some advantages, such as lesser traffic,
better reliability and lower cost. Accordingly, in this paper the
fully distributed system is introduced in Figure 4.
Local Sensor1
Local Sensor2
Local SensorN
Data
Fusion
Center
(DFC)
1
z
2
z
N
z
1
u
2
u
N
u
0
u
Communication
Channels
Target
Figure 4 .Fully distributed target detection system
3.2. Principles of target detection
In Figure 4, it is supposed that the fully distributed
system is consisted by the data fusion centre and N AEW
radars. The binary hypotheses problem is described as
0
: H Target is inexistent
1
: H Target is existent
Each radar of the system makes its decision
{ }( ) 1, 2, ,
i
u i N = = u "
dependently according its observation
i
z
( ) 1, 2, , i N = " and threshold
k
q , where
1 0
1 0
1 ( ) ( )
0 ( ) ( )
i k i
i
i k i
if P z H P z H
u
if P z H P z H
q
q
>
<
(3)
The vector u is transmitted to data fusion centre and
composes the observation vector of the centre. A global
decision
0
u is then obtained at fusion centre based on the
decision vector uand the threshold
0
q . where
1 0 0
0
1 0 0
1 ( ) ( )
0 ( ) ( )
i i
i i
if P u H P u H
u
if P u H P u H
q
q
>
<
(4)
It is proved in [2] that the optimal performance of
CFAR detection will be achieved in local sensors and fusion
centre under Neyman-Pearson criterion. If the global false
alarm rate of the system is given, the fully distributed data
fusion system can be depicted as the following compound
function
( ) { } ( )
0 0 0
: 0,1 ,
N
T T R u T = u
(5)
where
0
T is the fusion function of the fusion centre,
{ }
1
N
i
i
T T
=
= denotes the local decision function variety.
{ } ( ) : 0,1 ,
N
i i i i
T R u T z =
(6)
only when the system probability of false alarm
0 f
P o s , the system probability of detection
0 d
P can be
maximized.
It is of great importance to choose a appropriate
threshold
0
T . In [3], an adaptive distributed fusion algorithm
under Neyman-Pearson criterion is given, which needs
training to obtain convergence. However, both the time that a
target moves across a resolving cell and the quantity that a
radar samples are limit. And the emergence of a target is
occasional, so it is difficult to satisfy the convergence demand
of the mentioned algorithm.
=
| |
=
|
\ .
_
(7)
( )
0
1
1
N
N i
i
d d d
i
N
P P P
i
=
| |
=
|
\ .
_
(8)
where
( )
!
! !
N
N
i i N i
| |
=
|
\ .
(9)
If let
0 f
P o =
, the local detection threshold q can
figure out accordingly.
3.3. Generalized likelihood ratio detection of local radar
sensor
It is supposed that each AEW radar antenna array
consists of M cells, the pulse number between a Coherence
Processing Interval(CPI) is K . The receiving data of the
th
n
cell and the
th
k snapshot is defined as
nk
z . The array data
vector of the
th
k pulse is given by
( )
1 2
T
s k k Mk
Z k z z z ( =
"
(10)
( )
s
Z k ( ) 1, 2, k K = " can be organized as the
following 1 MK dimension column vector
( ) ( ) ( ) 1 2
T
T T T
s s s
Z Z Z Z K ( =
"
(11)
Under the binary hypothesis, Z is given as
1
0
S C N,
C N,
b H
Z
H
+ +
=
(12)
where b is the amplitude of target echo, Cand N are
the clutter and noise vector respectively, S is the target
space-time steering vector.
( ) ( )
t s
S S S
t s
=
(13)
( ) ( ) ( )
( ) ( ) ( )
s
S 1 exp ,
exp 1 ,
s s
T
s
j
j M
u
u
=
"
(14)
( ) ( ) ( )
( ) ( ) ( )
t
S 1 exp ,
exp 1 ,
t t
T
t
j
j K
u
u
=
"
(15)
where denotes the Kronecker product.
( )
2
, cos cos
s
d t
u u
=
( )
4
, cos cos
a
t
d
V
f
t
u u
=
( )
s
< and ( )
t
< is the spatial steering vector and the
temporal (Doppler) steering vector for a signal.
Considering an adaptive fully space-time filter structure
and the input vector Z , the signal-only filter output is given
as
H
y W Z =
(16)
where W is the weight vector.
The generalized likelihood ratio in adaptive space-time
processing advanced in [6] is
1
2
1
1 1
0
S
1
S S 1
H
Z
GLRT
H H
Z Z
H
R Z
R Z R Z
L
H
q
>
A =
< (
+
(
(17)
where
1
1
1
L Q
f
L P q
+
| |
= |
|
\ .
; L is the number of range
samples; Q is the dimension of processor, in the case of fully
space-time processing
Q MK =
,
( )
0 0
H
Z H H
R E Z Z =
denotes the
covariance matrix of clutter plus noise, it can be estimated by
1
1
L
H
Z l l
l
R Z Z
L
=
=
_
(18)
where
l
Z is the
th
l data vector of samples. The false
alarm threshold of the processor has no relation to the clutter
covariance matrix and the signal amplitude. It is decided only
by the dimension of processor and the given local probability
of false alarm. Consequently, it possesses the characteristic of
CFAR detection.
4. Simulation of target detection based on multi-AEW
radar data fusion
The fusion system is supposed to be consist of one
fusion centre and three local radar sensors, namely, 3 N = .
The system parameters of each radar sensor are uniform and
given in Table 1. The parameters of target are listed in Table
2. The simulation scenario is illustrated in Figure 5.
0 1 2 3
x 10
5
-1
0
1
2
3
x 10
5
X direction/m
Y
d
i
r
e
c
t
i
o
n
/
m
AEW1
A
E
W
2
AEW3
T
a
r
g
e
t
X direction /m
Y
d
i
r
e
c
t
i
o
n
/
m
Detected points
undetected points
Figure 8. The target detection result of AEW2
-1 0 1 2 3
x 10
5
-3
-2
-1
0
1
2
3
4
x 10
5
T
a
r
g
e
t
X direction /m
Y
d
i
r
e
c
t
i
o
n
/
m
Detected points
undetected points
Figure 9. The target detection result of AEW3
Let
dj
f and
dcj
f ( ) 1, 2, , j N = " denotes the Doppler
shifts of a moving target and the clutter relative to the
th
j
AEW radar. Respectively, it is shown in Figure 7 that the
necessary condition of effective target detection which can be
depicted that at least one { } 1, 2, , k N e " exists and satisfy
dk dck k
f f eO (19)
That is to say, as long as one of the target's Doppler
shifts does not fall into its corresponding radar's blind
Doppler zone, the continuous and complete target plots
presented in Figure 10 will be obtained after fusing in fusion
centre under the "1 out of N " rule.
-1 0 1 2 3
x 10
5
-3
-2
-1
0
1
2
3
4
x 10
5
T
a
r
g
e
t
X direction /m
Y
d
i
r
e
c
t
i
o
n
/
m
Figure 10. The result of fully
distributed system target detection
5. Conclusion
Focusing on the detection problem of target buried in the
blind Doppler zone of AEW radar, the fully distributed CFAR
target detection configuration based on multi-AEW radar data
fusion has been presented after analyzing the evolving
process of target Doppler shift relative to radar sensors. The
fusion rule, generalized likelihood ratio of target detection
and the probability of false alarm under the condition of fully
space-time processing are derived. It is justified that the
advanced technique can improve the probability of target
detection and the quality of target track dramatically in case
the necessary condition is satisfied. Moreover, the technique
is significant on the aspect of detecting low radial speed
and/or weak targets.
References
[1] J.M.C. Clark, P.A. Kountouriotis,and R.B. Vinter, A
Methodology for Incorporating the Doppler Blind Zone in
Target Tracking Algorithms, The 13th International
Conference on Information Fusion Proceedings,pp.1481-
1488, July 2010.
[2] S.Thomopoulos,R. Viswanathan, and D.Bougoulias,
Optimal distributed decision fusion, IEEE Trans. on
Aero. and Elec. Sys., vol. 25, pp. 761-765, September
1989.
[3] Ansari N,Chen J G Zhang Y Z,Adaptive decision fusion
for unequiprobable sources[J],IEE Proceedings of Radar
sonar and Navigation.1997,144(3):105-111.
[4] Q.Yan and R.S.Blum, Distributed signal detection under
the Neyman-Pearson criterion," IEEE Transactions on
Information Theory, vol.47, no.4, pp. 1368-1377, May
2001.
[5] Ruixin Niu and P. K. Varshney, Performance Analysis of
Distributed Detection in a Random Sensor Field. IEEE
Transactions on Signal Processing, vol.56, no.1, pp.339-
349, Jan 2008.
[6] E.J. Kelly: An adaptive detection algorithm. IEEE Trans.
AES 22 (1), 1986, pp. 115-127.