You are on page 1of 4

Performance Evaluations of Frequency Diversity

Radar System
Valentina Ravenni
ISL - Ingegneria dei Sistemi Logistici, Altran Group, via Antolisei 25 - 00173 ROMA - ITALY,
Direct Roma: +39-06-72643429, FAX Roma: +39- 06-72643433
E-mail: valentina.ravenni@isl-altran.it

Abstract Frequency Diversity Radar combines two or more
conventional transceivers through a common antenna, in a
multiplex arrangement on the same RF transmission channel. It
significantly improves radar detection performance and reduces
false alarms under adverse weather conditions, here in case of
normal distributed background clutter. Evaluations about
advantages between different algorithms are discussed.
I. INTRODUCTION
Employing more than one transmitter, each on a different
frequency but using a common antenna [1], the transmitted
power and the coverage of a radar system are increased. This
type of system is well-known as Frequency Diversity Radar
(Fig.1), whose pulses from different transmitters are radiated
sequentially by using an antenna squint, as shown in Fig. 2.
Thereby no increase is required in the powerhandling
capability of the wave guide system. The radar rephases in
time the received target returns and then recombines them at
video level, before the extraction process. Frequency
Diversity Radar also improves detection performance if the
echoes on different channels are uncorrelated. This is possible
if the transmission frequencies are sufficiently separated. It
means that no clutter correlation is present between different
channels, transmitting on different carrier frequencies. By the
way, target fluctuations are reduced because any target has
different reflection properties within different frequencies.
Moreover, if one of the transmitters fails, coverage will be
maintained and, as a result, the radar preserves detection
continuity.
Nevertheless, there are three main drawbacks: 1) the
necessity of a diplexer which increases the losses, limiting
practical radar efficiency, 2) the system is fairly complex
compared to a single maritime radar, and finally 3) an angle
and range re-alignment is required [2], [7].
This paper presents a performances study about detection
probability of Frequency Diversity Radar, using four
recombination algorithms, compared to a conventional radar.
II. PROBLEM STATEMENT
The detection problem can be described as follows [3].
Local receiver of each single chain properly demodulates,
filters, extracts the envelope and samples the incoming
narrowband waveform. The observed signal after the matched
filter, under two hypothesis,
1
H , for target present and
0
H
for disturbance only, is given by:


1
0
:
:
H y s d
H y d
= +

(1)

where s and d are complex numbers representing the signal
and the clutter, respectively.


Gear Box
Antenna
Diplexer
Transceiver 1 Transceiver 2
Synchroniser and
Correlator
Unit
Trigger 1
Video 1
Trigger 2
Video 2
Display Unit

Fig. 1 Simplified block diagram of Frequency Diversity Radar System.

Fig. 2 Antenna Squint.
The output x y = of the envelope detector for the cell
under test is compared with an adaptive threshold T , so a
binary decision is taken. In the absence of the useful signal,
x d = is assumed Lognormal distributed. The PDF
(Probability Density Function) with parameters and is
given by [4]:

( )
( )
2
2
ln
2
1
0.
2
x
x
f x e x
x

=
(2)

We can write ( )
0
: , H x L . The parameter ,
ranging from 0 to +, is a scale parameter, and , ranging
in ( ) 0, + is a shape parameter. ( ) ln z x = is a Gaussian
978-2-87487-001-9 2007 EuMA October 2007, Munich Germany
Proceedings of the 37th European Microwave Conference
1715
variable with mean and variance
2
, or, for short,
( )
0
: , H z N where ln is the natural logarithm.
Under the hypothesis
1
H , the signal is coherently added to
the clutter and the modules of the resultant complex number is
taken:
2 2
2 cos x s d s d s d = + = + + where is the
angle between s and d . is a random variable uniformly
distributed in [ ) 0, 2 and, if the Swerling 3 target model is
adopted, the module A s = is Gamma (Erlang) distributed
with PDF given by [5]:


( )
2
0
3
4
0
2
0,
A
A
a
A
f A e A
A
| |
|
\ .
=
(3)

but a closed form expression for the PDF of x cannot be
found. The threshold T is estimated as shown in Fig. 3 [6],
that is:

ML
T z = +
(4)

where and are maximum likelihood estimators of the
mean value and variance, respectively:


[ ]
1
2
1
1
1
N
i
i
N
i
i
w
N
w
N


=
=

(5)

N is the total number of samples, ( ) ln
i i
w x = , ln is the
natural logarithm and
ML
z is a suitable coefficient chosen in
according to the desired
FA
P value, that is:


{ }
0
Pr | ,
FA
P w T H = >
(6)
so:

1;1
1
1
FA
ML N P
N
z t
N

+
=

(7)

where
1;1
FA
N P
t

denotes the ( ) 1
FA
P quantile of the student
t distribution with 1 N degrees of freedom [6].

ln
x w
1 H
0 H

Fig. 3 Lognormal Detector.
III. ALGORITHMS OF RECOMBINATION
In according to [7], the amplitude of the composite video
depends on the type of recombination algorithm, as shown in
Fig. 4.

a
y
b
y
a a a a
y s d x = + =
b b b b
y s d x = + =
tot x

Fig. 4 Implementation of the composite video in a two channel system.

In a two channel system where
a
x = sample of the 1
st

signal,
b
x = sample of the 2
nd
signal and
, a b
x = sample of the
recombined signal, the following combinations exist:


,
,
,
2 2
, a b
Sum:
Product:
Geometric Average:
Euclidean Distance: +
a b a b
a b a b
a b a b
a b
x x x
x x x
x x x
x x x
= +
=
=
=
(8)

In a three channel system there is the third component
c
x = sample of the 3
rd
signal so, as shown in Fig. 5, the
implementation of the composite video is:

c y
a a a a
y s d x = + =
c c c c
y s d x = + =
tot x
b y b b b b y s d x = + =
a y

Fig. 5 Implementation of the composite video in a three channel system.

With the Sum algorithm, if the information of a single
channel is degraded it is compensated by that of the other
channel. With the Product, if one of the two video signals is
degraded, the performances of the product will be worse than
the sum. Better performances are obtained in case of great
distances where target echoes are weak. In case of Geometric
Average, the performances are very good for the background
noise reduction. The intensity of video signal is the average
between the amplitudes of the two single video signals.
Finally, the Euclidean Distance gives very good
performances for target detection and noise suppression.
IV. PERFORMANCES EVALUATION
Now, we asses the detection performances of frequency
diversity, in a two - channel system and in a three - channel
system, for each algorithms above mentioned, compared to
ones of a conventional radar.
The detection probability is:

{ } ( )
1 1
Pr | | ,
D tot tot
T
P x T H p x H dx
+
= =

(9)
where ( )
1
|
tot
p x H is the probability density function of the
variable
tot
x conditioned to the hypothesis
1
H , that is in
presence of the useful target.
1716
The false alarm probability is:

{ } ( )
0 0
Pr | |
FA tot tot
T
P x T H p x H dx
+
= < =

(10)
where ( )
0
|
tot
p x H is the probability density function of the
variable
tot
x conditioned to the hypothesis
0
H , that is in
absence of the useful target. Exact analytical evaluations of
(9) and (10), in presence of the useful signal and lognormal
noise are very complex, mainly because a close form for the
PDF ( )
1
|
tot
P x H and ( )
0
|
tot
P x H cannot be found.
Therefore, we have to resort to computer simulations, working
on unreal data or through standard numerical integration
routines, or alternatively to approximations. The following
figures show the result of Montecarlo simulations of False
Alarm Probability versus threshold , Detection Probability
versus Range, in case of fixed frequencies and fixed signal to
noise ratio, with a single, dual and triple system, respectively.
Curves show the combined advantages of frequency diversity
radar, using the algorithms Sum, Geometric Average or
Product and Euclidean Distance or Square Modulus. As a
matter of fact, the Geometric Average and the Product have
the same detection performances because
1
0
H
H
w T
>
<
is equal to
1
0
H
H
w T
>
<
. The same considerations can be done for the
Euclidean Distance and the Square modulus.
Euclidean Distance and Sum algorithm confirm the best
performance.
1 2 3 4 5 6 7 8
10
-5
10
-4
10
-3
10
-2
10
-1
Threshold
P
fa
SumAlgorithm


Single
Double
Triple

Fig. 6
FA
P versus for Sum algorithm.
1 2 3 4 5 6 7 8
10
-5
10
-4
10
-3
10
-2
10
-1
Threshold
P
fa
Product or Geometric Average Algorithm


Single
Double
Triple

Fig. 7
FA
P versus for Product or Geometric Average algorithm.
1 2 3 4 5 6 7 8
10
-5
10
-4
10
-3
10
-2
10
-1
Threshold
P
fa
Euclidean Distance or Square Modulus Algorithm


Single
Double
Triple

Fig. 8
FA
P versus for Euclidean Distance or Square Modulus algorithm.

Fig. 9
D
P versus R for Sum algorithm.

Fig. 10
D
P versus R for Product or Geometric Average algorithm.

Fig. 11
D
P versus R for Euclidean Distance or Square Modulus algorithm.

The following figures show the Detection Probability
versus False Alarm Probability for each algorithms.
1717
Furthermore, curves show Frequency Diversity Radar gives
superior performance compared to a conventional radar.

10
-6
10
-5
10
-4
10
-3
10
-2
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Probability of False Alarm (Pfa)
P
ro
b
a
b
ility
o
f D
e
te
c
tio
n
(P
d
)
Sum


SNR=15dB
Single
Double
Triple

Fig. 12
D
P versus
FA
P for Sum algorithm.

10
-6
10
-5
10
-4
10
-3
10
-2
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Probability of False Alarm (Pfa)
P
ro
b
a
b
ility
o
f D
e
te
c
tio
n
(P
d
)
Euclidean Distance or Square Modulus


SNR=15dB
Single
Double
Triple

Fig. 13
D
P versus
FA
P for Euclidean Distance or Square Modulus
algorithm.

10
-6
10
-5
10
-4
10
-3
10
-2
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Probability of False Alarm (Pfa)
P
ro
b
a
b
ility
o
f D
e
te
c
tio
n
(P
d
)
Product or Geometric Average


SNR=15dB
Single
Double
Triple

Fig. 14
D
P versus
FA
P for Product or Geometric Average algorithm.

V. CONCLUSIONS
Frequency Diversity Radar system is a simple technique
that significantly improves detection range performance for
the navigational radar, under adverse weather conditions. The
theoretical advantages of frequency diversity in detection of
targets against a clutter background are dependent partly upon
decorrelation of the target signals and partly upon
decorrelation of clutter samples at different frequencies. The
scenario model compare dual and triple Frequency Diversity
Radar System with conventional reference radar, when the
Swerling 3 target model is adopted and, the background
clutter is lognormal distributed. . In this case, exact analytical
evaluations of Detection Probability and False Alarm
Probability are very complex, so we have to resort to
MATLAB simulations. A set of Monte Carlo simulations have
been developed in order to estimate Frequency Diversity
Radar performance. The results of this analysis shows that
Frequency Diversity Radar gives superior performances,
corresponding to 40-60 % of detection probability and, a
reduction of false alarms corresponding to 40 50 %,
compared to a conventional radar.
ACKNOWLEDGMENT
V.R was supported by the Division of Base Radar Analysis
and Simulation Area (ABS) of ISL, Altran Group. The author
would like to thank Marco Scalzi, Leonardo Cantini, Massimo
Bertacca and, Salvatore Carbone for their help with theory,
simulations and English language.
REFERENCES
[1] P.A. Isaksen and E. Lovli, Multi Pulse Sweep System, A New
Generation Radar System, Radar 97, 14-16 October 1997, Publication
No. 449, IEE 1997.
[2] David K. Barton, Radars Volume 6, Frequency Agility and Diversity
Artech House, Inc, 1977.
[3] Gaspare Galati, Radar e Navigazione, Texmat, November 2004.
[4] Athanasios Papoulis, Probability Random Variables and Stochastic
Process, Polytechnic Institute of New York.
[5] Merrill I. Skolnik, Introduction to Radar Systems, McGraw-Hill, Inc,
2001.
[6] Maurizio Guida, Maurizio Longo, Marco Lops, Biparametric CFAR
Procedures for Lognormal Clutter, IEEE Transaction on Aerospace
and Electronic Systems, Vol. 29, NO. 3, July 1993.
[7] V. Ravenni, G. Pizziol, Frequency Diversity Radar System: Analysis,
Design and Performances, 3
rd
European Radar Conference, Sept.
2006, Manchester, UK.
1718