168 Iranian Journal of Electrical & Electronic Engineering, Vol. 6, No. 3, Sep.

2010
Adaptive Thresholding in Marine RADARs


M. Alaee*, M. Firoozmand*, R. Amiri* and M. Sepahvand*



Abstract: In order to detect targets upon sea surface or near it, marine radars should be
capable of distinguishing signals of target reflections from the sea clutter. Our proposed
method in this paper relates to detection of dissimilar marine targets in an inhomogeneous
environment with clutter and non-stationary noises, and is based on adaptive thresholding
determination methods. The variance and the mean values of the noise level have been
estimated in this paper, based on non-stationary, statistical methods and thresholding has
been carried out using the suggested two-pole recursive filter. Making the rate of false
alarm constant, the concerned threshold resolves the hypothesis of existence or absence of
the target signal. Performance of the mentioned algorithm has been compared with the
well-known conventional method as CA-CFAR in terms of decreasing the losses and
increasing calculation speed. The algorithm provided for detection of signal has been
implemented as a part of signal-processing algorithms of some practical marine radar. The
results obtained from the algorithm performance in a real environment indicate appropriate
workability of this method in heterogeneous environment and non-stationary interference.

Keywords: adaptive thresholding; CFAR; detection; RADAR; two-pole filter.



1 Introduction1
Detection is one of the most important subjects in
communication receivers, including radar systems. An
optimum detector determines either of the two
assumptions of existence or absence of the target
signals, with regard to the received observations and
based on a given theorem. It is obvious that there is no
receiver free from noises, thus such a decision to be
made would not be free from errors. Usually, depending
on the application, different optimization criteria such
as Bays and Newman Pierson are used in designing
detectors [1].
When radar reflected signal is contaminated with
noise and clutter effect, fixed threshold cannot keep the
false alarm ratio constant. Therefore, it would be
necessary to use CFAR circuits in decision making for
target detection in radar systems. Adaptive thresholding,
non-parametric methods and clutter maps are the three
major approaches ever have been introduced for
thresholding, with constant false alarm [2]. Adaptive
thresholding method assumes that the noise density
function is known and there are only some unknown
parameters in which should be estimated. Then, the
unknown parameters would be estimated using the

Iranian Journal of Electrical & Electronic Engineering, 2010.
Paper first received 3 April 2010 and in revised form 20 July 2010.
* The Authors are with Department of Telecommunication, Imam
Hossein University, Tehran, Iran.
E-mails: [email protected], [email protected],
[email protected], [email protected].
information of the neighbor points to the reference
point. So, the threshold will be estimated. Parametric or
non-parametric methods are based on the two
hypothesis of existence or absence of a signal, assuming
that the signal pdf and interference are known. For
example, the GLRT detector has been known as both
parametric and non-parametric detectors [3]. As the
signal density function and interference are known in
this detector, it is considered as parametric and as its
statistical specifications vary with time and the
environment variations, it is also considered as a non-
parametric detector. When signal and interference pdfs
are known, GLRT can be considered as the optimum
detector.
Similarly, the clutter map method requires storage of
the environment data in a number of different scans [4].
After these scans, the clutter map is prepared and
thresholding is performed through comparison of signal
and noise with the stored clutter map [5].
As the environments for marine radars are mainly
inhomogeneous and non-static and assuming
homogeneity may increase false alarm ratio in the
system, performance of implementing maximum
likelihood based method such as CA-CFAR, GO-CFAR
etc. is not in a good condition. The radar which must
detect targets upon the sea surface or near that should be
capable of discriminating targets from the sea waves.
Such reflections are also called sea clutters or echoes,
which, compared to noises, can be even make more
false alarms and may limit the radar detection
Alaee et al: Adaptive Thresholding in Marine RADARs 169
capability. The most important issue in correct
performance of a radar system is the detection algorithm
of it which means announcement of the existence or
absence of a target. Completely correct with real-time
decision is something impossible, unless the radar is
equipped with a rapid detection system with acceptable
accuracy. In case we are going to use adaptive or non-
parametric methods in thresholding and making
decision in radars, it is necessary to be knowledgeable
on the noise and clutter density functions and generally
on interference in designing the detector. Designing
optimum linear detectors follows with inverse
calculation of the interference covariance matrix. As
this calculation is practically difficult to implement as
radar processor units with current technology, adaptive
thresholding methods have been used in this paper for
designing detectors.
The most important problem for CA-CFAR
thresholder is its relatively high losses in heterogeneous
conditions. This is why lots of papers have been
prepared since long aiming to suggest a new method for
decreasing the mentioned losses [6], [7], [8]. Our
proposed algorithm aims to decrease the rate of CA-
CFAR losses while increase its calculation speed. For
more rapid implementation compared to the
conventional CFAR methods, the obtained threshold has
been implemented to the signal using a recursive filter
with known coefficients. The comparison decision of
this threshold, calculated on adaptive basis for radar
different ranges, has been considered as detection
criteria.
This paper has been developed in five sections. The
next section provides a model for the receiver noise, for
the received signal and the sea clutter. The third section
provides the suggested algorithm for the thresholder
along with design of the detector. Forth section shows
the results of simulations and implementation of the
suggested algorithm compared to CA-CFAR.
Furthermore the conclusion comes as the last section.

2 Modeling of Signal and Received Interference
In fully coherent radars, the signals reflected from a
target may have both modulations for the range and
phase. As sea marine radars are confronted with slow-
moving targets such as vessels or low-speed boats, there
is generally no need in detection criteria to keep the
phase and using coherency technology [9]. Thus, the
reflected signal from a non-coherent radar target can be
written as follows in marine radars:
S(t) = o

(t)c
] 2n (]
c
)t+] q
0
(t)

(1)
where o

(t) is the range of reflected signal and

0
(t) is
the initial phase of the transmitted frequency or
o
. f
d
,
which shows the Doppler frequency due to movement of
the target, is actually absent in this equation, as it is
actually lost in a non-coherent radar receiver after
passing by the envelop detector.

2.1 Signal Modeling
In radar processors, the reflection from the target
would be as follows [9]:
X = S +I
(2)
where S shows the signal with the equation provided
in Eq. (1) and I shows the interference (sum of noise
and clutter).
I = C +N
(3)
In the above equation, C shows the clutter and N
stands for the receiver noise. Followings are the vector
of signal and interference samples:

X = |x
0
x
1
x
N-1
]
t

(4)
S = |s
0
s
1
s
N-1
]
t

(5)
I = |i
0
i
1
i
N-1
]
t

(6)

2.2 Noise Modeling
The noise received in reflected signal from the target
is a combination of receiver noise or the inside noise of
the radar system and the outside noise (from the
environment). The sun noise, atmosphere noise and the
combustion noise in different sources are samples of
environment noise. Internal noise consists of noises
produced by the noise temperature of the antenna; the
phase noise resulted from oscillators and receiver
thermal noise. In frequencies higher than UHF, the
effects of outside noises are so decreased that the
received noise can just be attributed to the internal noise
of the receiver. Samples of noise are always present in
receiver output and they can never be diminished down
to zero. The receiver noise which mostly belongs to the
thermal noise due to the movement of electrons in semi-
conductors, are normally considered as complex white
Gaussian noise. In compliance with the nyquist sampling
rate, the consecutive samples of noise get uncorrelated
from one another. Following is the vector of noise
samples along with their spectrum specifications:
N =
|n(u) n(I
s
) n(2I
s
) n((N -1)I
s
)]
t

(7)
in which I
s
is the sampling time. The receiver noise
has a white spectrum and its distribution function is
assumed as Gaussian. In the following relationship, o
n
2
is
equal to the noise power.
E|n

n
]
| = _
o
n
2
i = ]
u i = ]

(8)
Thus, the noise covariance matrix can be expresses
as follows:
R
n
= o
n
2
I
n

(9)
where I
n
is the unit matrix of N N.

170 Iranian Journal of Electrical & Electronic Engineering, Vol. 6, No. 3, Sep. 2010
2.3 Clutter Modeling
Nowadays, lots of measurements have been carried
out by different researchers on the reflections from the
sea surface. The measurements have been performed
from HF frequency range up to millimeter to light waves
and various data have been collected under different
environment conditions. The data obtained from these
measurements show high dispersal of the results, even in
similar measurement conditions [6]. A reason for such
high variations is difficulty of measurement or
description of the sea state. Speed, period and direction
of wind at the sea surface, oceanic flows, sea surface
pollutions, the effects of storms in ocean surface in
distant points and causing water waves and local climatic
changes can all affect on the radar wave reflection from
the sea surface. Thus, calibration of measurements at the
sea surface is a difficult task, as this cannot be done
under controlled conditions. Therefore, radar designers
should take the above-mentioned changes into
consideration in their designs and determine the radar
performance under the effect of different conditions of
sea reflections.
The Sea is one of the distributed targets which its
echo range is depended on the dimensions of the area
lighted by the radar beam width [7]. Parameters involved
in the sea clutter can be divided into two sections;
section one consists of the parameters relating to the
radar such as carrier frequency, beam width of the
antenna, polarization of the sent wave and pulse width
and section two includes the environment-related
parameters such as water undulations, height of the sea
wave or sea force, speed and direction of the wind, etc.
So far, lots of models have been provided for
describing the surface sea clutter density function, none
of them in complete conformity with the actual
conditions, with regard to the problems already
mentioned with regard to the measurements. Rayleigh,
Weibull, Log Normal, Gaussian mixture and K family
models are of the most famous density functions
provided in some papers for clutter distribution.
As in almost all sea marine radars, non-coherent
integration is used to improve signal to noise ratio,
decision is actually made based on pulse N, where the
N value is obtained with regard to rotation speed of the
antenna, beam width and the pulse repletion frequency.
Since clutter samples are not independent, and are highly
correlated, joint distribution of N can be easily
obtained by multiplying. Gaussian models resulted from
central limit theorem (CLT) [1], can be taken into
consideration for clutters when N different distributions
are merged. Experience shows that Gaussian models are
appropriate for radars rather than those with high range
resolution and radars working in small beam width.
When radar works in these conditions Rayleigh, Log
Normal and K family distributions are introduced [10].
The particular radar in this paper is a commercial
marine radar system in X band of frequency with almost
150-m range resolution cells and about 2-degree beam
width at long ranges. As a wide area is seen as one cell,
for this type of radar, water wave changes will have no
considerable effect on the total area, against the radiated
ray. Therefore, central limit theorem is applicable here
and as the result, the Gaussian assumption can be applied
to the sea clutter. However, when weak signal detection
near a powerful clutter is considered, as the radar
functions from a smaller pulse width with a range
resolution of near 12 m, the Gaussian assumption of the
clutter is not valid and the sea wave will show its non-
stationary nature more considerably. We consider
Weibull distribution for the sea clutter in the mentioned
condition since we can obtain other introduced pdfs
generally. For example we can reach to the Rayleigh
distribution by setting 2 for the shape parameter in the
Weibull distribution. This distribution is expressed by
the following equation [4], [5]:

x
(x) =
C
B
]
x
B

C-1
exp_-[
x
B

C
_
(10)
where C and B are shape and scale parameters
respectively. In order to obtain a constant false alarm
ratio, our threshold based on C and B parameters is
according to the following equation:
I = {-ln(P
PA
)]
1
c
B
(11)
where P
FA
is the probability of false alarms. With
regard to the correlation function of clutter samples, the
clutter covariance matrix can be written as follows,
which can be produced for each clutter model.
P
c
= _
R
c
(u) R
c
((N -1)I
s
)
. .
R
c
((N -1)I
s
) R
c
(u)
_
(12)
In the above equation, I
s
is equal to time interval
of the clutter consecutive samples. In Equation 12 R
c
(u)
is equal to the clutter echo power received from the cell
under test.
With regard to the above-mentioned equations, despite
the noise covariance matrix, the clutter covariance
matrix varies with changes in the environment
conditions and the concerned cell spacing from the
radar.

3 Design of Detector and Adaptive Thresholding
Here, based on the model provided for the signal
reflected from the target, receiver noise and the clutter, a
linear processor is provided. If a linear filter with W
complex coefficient vector is used for detection, the
most appropriate form for testing the linear detection is
as follows [11]:
|W
opt
H
X|
B
1

B
0
T
(13)
where T is the threshold which depends upon the
appropriate P
FA
and obtains from the Equation 11 and
B
1
and B
0
are the existence and absence hypothesizes of
the target, respectively. The E superscript in the
above-mentioned equation stands for the Hermitian.
Selecting W coefficients for optimization of detector
depends on the clutter and noise distributions provided in
the previous section. Optimum weighting coefficients
are obtained from the following relationship.
Alaee et al: Adaptive Thresholding in Marine RADARs 171
W
opt
= yQ
-1
S (14)
Such a filter is quite well known today in the field of
detection area as the match filter. The linear detector is
one of the most important and common detectors with
known structure, which is used in showing existence of
signals. Simplicity and acceptable performance of this
detector has made it quite applicable.
Given the density functions provided in the previous
section, the problem here follows with determination of
T or threshold. For this purpose, it is necessary to
estimate the unknown pdf parameters (such as the shape
and the scale parameters). Up until now, different
methods have been presented for estimation of the
unknown parameters regarded to constant false alarm
ratio; e.g. CA-CFAR, OC-CFAR, GO-CFAR, CMLD
and TM-CFAR [3]. The suggested method which is
based on the two-pole recursive filter has a faster
implementation rate compared to the other provided
methods. Meanwhile, its losses are considerably reduced
with in longer time.
The most common type of spatial CFAR is CA-
CFAR where use of adaptive CFAR is in the common
range vector. Theoretically, the averaging CFARs (CA)
perform a statistical estimation of un-biased minimum
variance from the interference within the reference
window.
A common disadvantage of the CA-CFAR method is
its lack of resistance in coming to inhomogeneous
environments [12]. When CA-CFAR is used along the
range, it works well in a noisy environment but if there is
a sea clutter with a fast variations, its performance will
decrease. Also, if there are returns from a number of
targets within the sampling cells (reference), its
workability will considerably decrease, as these targets
will automatically help the threshold to raise [13].
Different strategies have been put forward to
improve the performance of CA-CFAR. For example,
instead of using the average of windows at both sides of
test cells, it is possible to compare average values of two
cell groups and then multiply the larger or smaller value
by a constant value and use it as the threshold level. In
case of choosing the smaller average as the threshold,
this technique is called SO-CFAR and in case of
selecting the larger average, it is called GO-CFAR. The
GO-CFAR method uses the window with larger average
range as the criteria for determination of the threshold.
Thus, the threshold level provided in this method is
higher than the normal level and weaker targets may not
be detected. On the other hand, if the data of a window
with a smaller average range is used as the criteria for
determining the threshold, a relatively lower threshold
level is provided and it may result in higher probability
of false alarm and saturation of the processor [12].
If a logarithmic detector is used before the CA-
CFAR, this technique is called LOG-CA-CFAR (LCA-
CFAR). This threshold selection method has been
suggested for improvement of the CFAR performance
against heterogeneous clutters. Output of the logarithmic
detector is processed by a CA or an optimized type such
as GOCA. Unfortunately, LCA has more losses than
CA-CFAR [1].
For reducing the losses and improving decision
making time, our proposed method uses the adaptive
detection resulted from Two-Pole Filter, with the
recursive coefficients calculated as the diagram block in
Fig. 1, so that the variance and mean values of the noise
have been calculated separately for each range cell and
get closer to the actual values in the course of time. In
Fig. 1 diagram block, C sits for the comparator whose
task is to compare the input signal frequency with the
estimated threshold.
The threshold is estimated as follows:
T = Y
b
B
(15)
The above equation is obtained from Equation 11
with this assumption:
Y
b
= {-ln(P
PA
)]
1
c
(16)
To achieve the unknown parameter from the Weibull
pdf we use maximum likelihood estimation [14]. So we
would have:
B
`
= _
1
N
x
]
C
N
]=1
_
1
C

(17)
By setting C parameter equal to one in the above
equation we will have:
B
`
=
1
N
x
]
N
]=1

(18)
That in fact this parameter is the mean value of input
information. Also by choosing C = 2 in Equation 17.
B
`
= _
1
N
x
]
2
N
]=1
_
1
2


(19)
That in fact is the ML estimation of standard
deviation for the Rayleigh distribution. Higher orders of
C will result higher orders of Weibull distribution
moments.
In order to increase the calculation velocity of
estimating the unknown parameter B

we used the
following equations. In this equation, 1
st
moment and 2
nd

moment of the interference are calculated as follows:

n
= K
1

n-1
+ K
2
input(i)
(20)

n
2
= K
1

n-1
2
+ K
2
(input(i) -
n
)
2


(21)
And also Y
b
is calculated using the following
equation:
Y
b
= N + N +
-In(P
FA
)- 1-x
(1.1N-0.1)
0.S1
+ xN
(22)
where, N is the number of lines within the matrix of
the radar data, whose threshold is determined
collectively. Meanwhile, we have the following
relationship:
172 Iranian Journal of Electrical & Electronic Engineering, Vol. 6, No. 3, Sep. 2010
Fig. 1 Proposed detector diagram.

x = g -
2.S1SS17 + u.8u28SS - g + u.u1uS28 - g
2
1 + 1.4S2788 - g +u.189269 - g
2
+ u.uu1Su8 - g
3

(23)
where g is equal to:
g = - 2lnP
FA

(24)
Therefore, the threshold will have different values in
different P
FA
s and is sensitive to it. Meanwhile, one
threshold is determined for each range cell. With regard
to the recursive relationship in Equation 20 and Equation
21, estimation of the mean will result in a closer value to
the actual value of the noise average with elapse of
longer time and the new inputs being added, as despite
CA-CFAR, the length of the averaging window here is
not limited to the M cell already determined and it is
extended to the length of the signal. Parameters K
1
and
K
2
in Equation 20 allow for calibration of the system
under different environment conditions for the designer.
It is noteworthy that the sum of parameters K
1
and K
2
in
this equation is equal to 1 and in the simplest case, when
K
1
= K
2
= 0.5, it can be observed that the recursive
average tends after some time to the actual average. Our
suggested values are 0.91 for K
1
and 0.09 for K
2
. Also
B
`
= +
2
represent a better performance in
comparison of a single moment for B.
Meanwhile, for a better performance of the suggested
algorithm, it is possible to use a limiting factor which is
related to our natural information on the system. This
limiting factor may be designed and used on hardware or
software basis. For instance, if determined values are
P
d
= u.9 and P
FA
= 1u
-6
, the required signal to noise
ratio is 13.5 dB according to tables [5], for detection
announcement. Therefore, if the input signal in the i
th

moment is higher than the obtained threshold by more
than 13.5 dB, it may not be entered into calculation of
the average and the previously-calculated value can be
entered into the algorithm as the new input.
Another point which should be considered in the use
of the suggested algorithm is determination of the initial
values for the noise variance and mean values. As the
provided algorithm is highly depended on its previous
values, determining wrong values for the initial values
may its intensive divergence. However, in designing
radars, as some parameters such as the receiver
bandwidth, environment temperature and the number of
ADC bits are previously determined, the noise initial
variance and mean values can be easily estimated.
In our suggested method, it is possible to use
statistical parameters within the cell for which the
threshold is being calculated (average, standard
deviation, etc.) or use a non-parametric method (such as
classification based on the order statistics) in
determining the threshold. Meanwhile, it is possible to
separate the concerned from other cells by means of
time lag (range), angle or a combination of such
components. Finally, as already mentioned, it is
possible to implement a series of non-linear processes
(such as limiting factor or large samples correction) on
the cell under test before estimating the statistical
parameters.

4 Simulation Results on Real Data
Standard deviation of the adaptive thresholding
suggested here is larger than the optimum detector by
almost 15 percent and this forces some losses in the
system. However, the calculating method is very simple
and workable for real time applications. This is while
implementation of the optimum detector and calculation
of inversed covariance matrix in real time is very time
consuming with the existing processors. In terms of the
effect and the amount of losses, the suggested
thresholding can be compared with Cell-Averaging
CFAR method.
Fig. 2 shows the diagram of Cell-Averaging CFAR
losses for different false alarm probabilities. As it can be
seen, length of M should be equal about 150 cells in
P
FA
= 1u
-6
to reach 0.2 dB losses compared to the
optimum detector. We assumed that we had already 10
pulses integrated for this chart. If the pulses were lower,
the losses would have been more [5]. This type of
averaging can be very time-consuming in radar
applications.
In the proposed thresholding algorithm, in order to
estimate actual loss from actual values, we simulated the
Weibull interference and noise for 1000 times, with the
mean of 0.73 and the variance of 0.09. It aimed to
investigate on the values estimated for the average and
the variance and compare it with the values already
determined. Then, we estimate the threshold using
equation 15 for both cases and calculate the pre-assumed
values from the estimation results. As the result the
standard deviation from actual values of the mean and
the variance will decrease when the samples are being
added. Meanwhile, in the worst case out of the
mentioned simulations, the thresholding losses have also
been calculated, which is almost 0.27 dB in the tenth
sample to 0.09 dB in the 70 th sample. These simulations
have been carried out in the false alarm ratio of 1u
-6
so
that the detector losses can be compared with those
provided for CA-CFAR.
Fig. 3 illustrates the losses diagram based on the
input data in the suggested detector for different false
alarm rates. It refers to this important note in using our
suggested thresholding system that in case the number of
the investigated samples is sufficiently high, the false
ratio is considerably decreased, so that in longer time,
the suggested detector acts similar to the optimum
Alaee et al: Adaptive Thresholding in Marine RADARs 173
detector. For example, in the 150 th sample, the rate of
losses has been decreased down to 0.07 dB. This is while
in CA-CFAR, selection of a larger window is inevitable
(see Fig. 2).
Diagrams of Fig. 4 and Fig. 5 shows the threshold
obtained from CA-CFAR and the proposed threshold on
a real radar signature. Meanwhile, we show in Fig. 5
that the adaptive threshold is shifted appropriate to the
P
FA
changes. In these figures, the horizontal axis shows
the time or just discrete samples entered into the
thresholding system. The vertical axis shows the
received signal amplitude after ADC and other
conventional algorithms of signal processing in marine
radars (such as STC and FTC). As delineated samples
are taken from real radar signatures, their range have
been shown without normalization for investigation on
what occurs in reality.


Fig. 2 Loss diagram based on the widow length in CA-CFAR.

Fig. 3 Loss diagram based on input data in the proposed
detector.

Fig. 4 Thresholding in CA-CFAR method on real radar
signature.

Fig. 5 Threshold sensitivity in P
FA
in the suggested detector on
real interference.

5 Conclusion
At the time being, different algorithms have been
suggested for detection and thresholding according to a
constant false alarm rate. In this paper, we aim to
suggest a fast implementing algorithm for thresholding
and making decision in sea marine radar systems. In this
paper, we have introduced statistical models for signal,
noise and clutter in sea marine radars and then we have
obtained optimum detector using a suggested recursive
two-pole filter with known parameters. Based on the
assumed density functions, we suggested a semi-
optimum detector, that beside very good calculation
speed, it shows very small amounts of losses with the
past of time. The proposed algorithm has been tested on
real radar signatures and has been implemented on a sea
marine radar system. The results indicate good
performance of the algorithm in non-stationary sea
interference.

20 40 60 80 100 120 140 160 180 200
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
M Length
L
o
s
s

(
d
B
)


P
FA
= 1e-4
P
FA
= 1e-5
P
FA
= 1e-6
P
FA
= 1e-7
20 40 60 80 100 120 140 160 180 200
0.04
0.06
0.08
0.1
0.12
0.14
0.16
Time(Sample)
L
o
s
s

(
d
B
)


P
FA
= 1e-4
P
FA
= 1e-5
P
FA
= 1e-6
P
FA
= 1e-7
0 500 1000 1500 2000 2500 3000 3500 4000 4500
0
0.5
1
1.5
2
2.5
3
3.5
x 10
4
Time (Sample)
A
m
p
l
i
t
u
d
e
P
FA
= 0.001


Signal + Noise + Clutter
Threshold
0 500 1000 1500 2000 2500 3000 3500 4000 4500
0
0.5
1
1.5
2
2.5
3
3.5
x 10
4
Time (Sample)
A
m
p
l
i
t
u
d
e


Signal + Noise + Clutter
P
FA
= 1e-3
P
FA
= 1e-6
P
FA
= 1e-9
174
References
[1] Gini
Detect
Gaussi
Navig,
[2] Odaba
New S
An Un
Christ
Applic
1-4244
[3] Ayoub
Signal
AES, V
[4] Hamad
"Analy
clutter
Issue 1
[5] Skolni
Graw H
[6] Ragha
Proces
Transa
System
[7] Smith
CFAR
IEEE
System
[8] Van C
Detect
2008,
[9] Levan
John W
[10] Billing
Measu
Andre
[11] Gini F
radar
Wiley
[12] Hamur
detecti
clutter
Applic
11, 20
[13] Simon
John W
[14] Ravid
CFAR
Signal
264, 1







F., "Sub-O
tion in a m
ian Clutter",
, Vol. 144, Iss
aee M., Ghor
Signal Detecti
nknown Back
tchurch: Telec
cations Confer
4-1557-1.
b T. F., Haimo
l Detection A
Vol. 36, No. 3
douche M.,
ysis of the cl
r", Elsevier,
1, pp. 117-123
ik M. L., Rad
Hill, 2008.
avan R. S.,
ssors for Li
actions on
ms, Vol. 28, Is
M. E. and
R Processor B
Transactions
ms, Vol. 36, Is
Cao T.-T. "
tors", Interna
978-1-4244-2
on N., Moad
Wiley & Sons,
gsley J. B., Lo
urements and
w Publishing,
F. and Rangas
detection, tra
& Sons, Inc.,
rcu A.C. a
ion in spatiall
r", Signal Pro
cations Confe
09, 978-1-424
n H., Adaptiv
Wiley & Sons,
R. Levano
R for Weibu
l Processing,
992.
I
Optimum C
mixture of K
IEE. Proc.
sue.1, pp. 39-4
rbani A., Am
on With Unkn
kground Usin
communicatio
rence, pp. 225
ovich A. M., "
Algorithm", I
, pp. 810-818
Barakat M
lutter map CF
Signal Proce
3, 2000.
dar Handbook
, Analysis
inear-Law D
Aerospace
sue 3, pp. 661
Varshney P.
Based on D
on Aerospac
sue 3, pp. 837
"Design of
ational Confer
2321-7.
davzeson E.,
, Inc., 2004. T
ow Angle Rad
d Empirical M
, 2002.
swamy M., K
cking, and cla
, 2008.
and Hizal A
ly correlated K
ocessing and
rence, pp. 84
44-4435-9.
e RADAR Sig
, 2007, TK658
on N.,"Maxim
ull backgroun
Vol. 139, Is
Iranian Journ
Coherent Ra
K-distributed
Radar, Son
48, 1997.
mindavar H.,
nown Dopple
ng Spectrogra
on Networks a
5-229, 2007, 9
"Modified GL
IEEE Trans.
, 2000.
M., Khodja
FAR in Weib
essing, Vol.
k, New York:
of CA-CF
etection, IE
and Electro
1665, 1992.
. K. "Intellig
ata Variabilit
e and Electro
7847, 1992.
low-loss CF
rence on Rad
Radar Sign
TK6575.L478.
dar Land Clu
Models, Willi
Knowledge ba
assification, Jo
A., "CA-CF
K-distributed
Communicati
40843, April
gnal Processi
80.A35.
mum likelih
nd", Radar a
ssue 3, pp. 2
nal of Electric
adar
and
nar,
"A
r In
am",
and
978-
LRT
on
M.,
bull
80,
Mc
FAR
EEE
onic
gent
ty,
onic
FAR
dar,
nals,
.
utter
iam
ased
ohn
FAR
sea
ions
l 9-
ing,
ood
and
56
det
targ
sign


Com
Ho
RF










cal & Electron
tection theory
get modeling,
nal processing
mmunication L
ssein university
circuits and Pr
nic Engineerin
Mohamma
Iran, on Ju
M.Sc. deg
Engineering
(Polytechni
Iran, in 20
working
engineering
His researc
signal pr
particularly au
statistical sign
in radar applica
Mohsen
Dezful, Ira
B.Sc de
Engineerin
Hossein, Ir
in Commu
Islamic Az
Researches
His M.Sc.
Effects i
Links". Currentl
y. His main re
ropagation wav
Reza Am
in 1982.
Science
Hossein
2005. H
the Dep
Imam
research
signal
processin
processo
Morteza
Khoram
received
honors
Of Tec
2001. H
field of
systems
measure
ng, Vol. 6, No
ad Alaee was b
une 27, 1981. H
gree in Telec
g from
ic) University O
006. Since 2003
in faculty
g of Imam Hoss
ch interests inc
rocessing, est
utomatic target
nal estimation
ations.
Firoozmand
an, in 1972. H
egree in C
ng from univer
ran, in 1997 and
unication Eng
zad university
s Unit) Tehran,
thesis was "F
in Free Sp
ly, he is a resea
esearch interest
es.
miri was born i
He received th
degree from
University, T
He is now a M
partment of Pa
Hossein Un
h interests in
processing, d
ng and its appl
ors and passive
a Sepahvand
mabad, Iran,
d the M.Sc. En
From Amirka
chnology, Teh
His current resea
f electronics,
and
ements.
. 3, Sep. 2010
born in Tehran,
He received the
communication
Amirkabir
Of Technology,
3, he has been
of electrical
sein university.
clude statistical
timation and
t classification,
and filtering,
was born in
He received the
Communication
rsity of Imam
d M.Sc. degree
gineering from
(Sciences and
, Iran, in 2006.
Fading and its
pace Optical
archer in Imam
ts are Antenna,
in Tehran, Iran,
he Bachelor of
m the Imam
ehran, Iran, in
M.Sc. student at
assive Defense,
niversity. His
nclude Radar
digital signal
lication to DSP
defense.
was born in
in 1974. He
ng. degree with
abir University
hran, Iran, in
arch lies in the
digital control
sensors &
0
,
e
n
r
,
n
l
.
l
d
,
,
n
e
n
m
e
m
d
.
s
l
m
,
,
f
m
n
t
,
s
r
l
P
n
e
h
y
n
e
l
&