Bistatic Radar Imaging System For Sea Surface Target Detection
Bistatic Radar Imaging System For Sea Surface Target Detection
Bistatic Radar Imaging System For Sea Surface Target Detection
1: E3I2 Laboratory, ENSIETA, 2 rue François Verny, 29806 Brest Cedex 09 FRANCE
{comblefa,yassinmo,khenchal}@ensieta.fr
2: LSOL Laboratory, 6 Av. Le Gorgeu C.S. 93837, 29238 Brest Cedex 03 FRANCE
[email protected]
RDA.
where E0 is the electromagnetic field magnitude, Se is the
I. INTRODUCTION transmitted waveform. δtER, δtCR and δtEC represent the
transmitter-receiver, the target-receiver and the transmitter-target
New missions assigned to radar systems require propagation time, respectively. Each propagation time take into
sophisticated signal processing techniques to obtain better account the coordinates and velocities of the entities
resolution and detection capabilities. In this paper, a bistatic (transmitter, target, receiver). qr and qr are the transmitting and
r r
radar configuration [1] is proposed for sea surface target
receiving Jones vector which depend on the transmitting and
detection and localization. Bistatic radar operate with separated
receiving antenna polarization. [Gr] and [Ge] are receiving and
transmitting and receiving antennas (Fig. 1). This configuration
transmitting antenna radiation pattern matrices. [Pr] and [Pe] are
leads to a more accurate information on the observed scene and
the linking matrices between, respectively, the receiving, and the
can improve the detection of stealth targets. Of course, the
transmitting direction antenna base and the electromagnetic field
transmitting antenna can be monostatic, that is transmitting and
propagation bases. [SFSA] denotes the scattering matrix in the
receiving, and consequently it is possible to join monostatic
forward scattering alignment convention which depends on the
and bistatic data reflected by common covered areas.
geometrical and physical features of the target.
The used radar radiolink is designed in an adaptative structure
for a general configuration where the transmitter, the target and
the receiver are moving. Modules such as antennas radiation
characteristic or target scattering properties can be inserted in
accordance with the desired radar scenario. Then the presented
model permits to simulate a wide range of radar scenarii where
reflectivities from targets and clutter can be individually
processed and their characteristics can be investigated in time or
Fig. 1. Bistatic radar geometrical configuration with one transmitter frequency. The interest of this model is great because the
(E) and two receivers (R1 et R2). electromagnetic field propagation is considered in a three
dimension way which allows to take into account propagation,
This configuration has been known for a long time, but it shadow or masked effects. Whereas other simulator take only
was not used due to its complexity. Bistatic data acquisition into account the amplitude and phase of each reflectors on the
should provide additional qualitative and quantitative observed scene. Moreover, in contrast with actual SAR
measurements of scattering surface properties. So using bistatic simulators based on scalar field, this model use a “full-
configurations allow to obtain images with more information polarized” configuration. So, this model permits, for a given
than in monostatic configurations. The bistatic configuration scenario, to generate radar data which can be used in signal
should be particularly interesting in detecting target on rough processing algorithms for target detection, clutter suppression or
surface as sea surface. target classification.
Former this paper presents the electromagnetic field Finally, this signal representation is used by the bistatic SAR
propagation for a general bistatic configuration. Latter, an processing in order to generate the observed scene image.
original sea surface characterization is developed. And finally,
these representations are used in a new bistatic SAR processing III. BISTATIC SAR PROCESSING
which allows to simulate a sea surface target detection and
localization. a) Algorithm
The proposed bistatic SAR simulator based on the well-
known monostatic SAR algorithms: RDA (Range Doppler λ , (4)
∆CR ( M ) =
Algorithm) [3]. Monostatic and bistatic SAR images are (γ Tr + γ R ) cos(α )
obtained from three stages: motion compensation, distance
processing and a transverse processing. where c is the celerity, ∆f is the frequency bandwidth, α the
With regard to the bistatic imagery, motion compensation angle between L and BXY, λ is the wavelength, γTr and γR are
depends on transmitter, target and receiver distance variations respectively the angle subtended by the transmitter and the
during the acquisition. The acquired data can be expressed by receiver synthetic aperture. Finally the general resolution
s(t,u) where u is the synthetic aperture and t is the time domain. expressions are:
So, the motion compensation is done by a multiplication, in the c
frequency domain, of each acquisition by e-2πj(Dt+Dr)t. Where Dt ∆R( L) = ,
and Dr correspond to the transmitter-target and the target-
2∆f cos(β 2) cos(α ) cos(Φ B )
receiver distances, respectively. λ , (5)
∆CR ( M ) =
Since the emitted signal is wide-band, then, a matched [γ Tr cos(ΦTr ) + γ R cos( Φ R )] cos(α )
filtering using the emitted signal se is carried out on each
acquisition in order to improve the range resolution. Let x=c.t, where φB, φTr and φR correspond to the projection angle on the
the compressed signal is defined as XY-plane.
+∞
s c ( x, u ) ≈ ∫ s e* (t ’−t ) s(t ’, u )dt ’. (2) From these general equations, the monostatic resolution (for β=0
−∞ and φTr=φR) [3] and the resolution in particular bistatic
Finally, the Doppler shift caused by the transmitter and configurations [5][6] can be recovered.
receiver motions is used in order to improve the cross-range
resolution by a transverse matched filtering. This matched filter
depends on the relative speed vtr between the transmitter and
the target and the relative speed vr between the receiver and
the target. Then the reconstructed signal is given by:
+∞
I ( x, y ) = ∫ s v* ( y ’− y ) s c ( x, y )dy ’, (3)
−∞
2πjf 0 ( vtr ( u ) + vr ( u )) u
where s v (u ) = e , and y=u.
So, to carry out the bistatic radar imaging processing,
it is necessary to known precisely the geometrical
bistatic configuration. Moreover, bistatic observation
requires accurate time synchronization and antennas
pointing between transmitter and receiver. Fig. 3. Fig. 2 top view.
b) Resolution c) Simulations
Bistatic SAR images are characterized by there resolution The considered scene is made of four isotropic scatterers is
which depends on the geometrical configuration. The analytical observed in monostatic and bistatic configurations. The
expressions of the bistatic SAR are given in [4]. Fig. 2 presents scatterers locations are (-20;10), (-10;30), (0;0) and (20;-20) in
the ground plane as shown in Fig. 4.
the general setup of the bistatic configuration. Tr, Ta and R
stand for respectively the transmitter, the target and the
receiver. β is the bistatic angle, B is the bistatic bisector and
BXY its projection on the XY-plane.
Fig. 2. 3D general bistatic configuration defining resolution direction. The transmitter is the same for the monostatic and the
bistatic acquisition. The transmitted signal is a chirp
Fig. 3 is Fig. 2 top view showing the range (gray) and the characterized by the center frequency: 5GHz and a bandwidth
cross-range (black) resolution directions. of: 80Mhz. The transmitter location is (-25;-2000;10000) and
Only perpendicular motions to target-transmitter-axis (Ta- transmitted aperture antenna is 50m along the x-axis. In the
Tr) and target-receiver-axis (Ta-R) are taken into account monostatic configuration, the receiver location and received
because they contain all necessary information. L is the line aperture antenna are the same than the transmitter. The image
perpendicular to the sum of the transmitter and the receiver obtained from monostatic data is presented in Fig. 5.
motion (line N in Fig. 3). In the XY-plane, the range and the
cross-range resolutions are respectively expressed in the L-
direction and in the M-direction (perpendicular to BXY) as
follows:
c ,
∆R( L) =
2∆f cos(β 2) cos(α )
Fig. 5. Monostatic SAR image. Fig. 8. Third bistatic SAR image.
Now, different bistatic configurations are considered. The The targets locations have been found on each image.
acquisition configurations are presented in Tab. 1. Geometrical characteristics and resolutions obtained from
simulated SAR images agree with theoretical values.
First Second Third In the bistatic configurations, resolutions along x-axis are
acquisition acquisition acquisition better than in the monostatic configuration due to the fact that φB
Receiver is smaller in the three bistatic configurations than in the
(-500;-25;1000) (0;-1000;300) (-25;-500;1000) monostatic configuration. In the third bistatic simulation, the
location
Aperture receiver aperture angle is the biggest one than all simulations
(0;50;0) (0;0;0) (50;0;0) configurations, so the y-axis resolution is the best one.
antenna
Tab. 1. Bistatic configurations.
The resolution for these configurations are obtained from IV. SEA SURFACE MODELS
eq. (5) and are given in Tab. 2.
At each point of the sea surface, waves result from a sum of
waves locally generated by wind and waves coming from any
First Second Third areas. Due to these interactions, the phenomenon can hardly be
acquisition acquisition acquisition quantified. Since the ocean presents an uncoordinated aspect, the
∆R=8,17m ∆R=2,78m ∆R=5,77m sea fluctuations are classically represented by three dimensional
Resolutions
∆CR=1,2m ∆CR=12,22m ∆CR=1,2m
r
random process z( r ,t) depending on temporal and two spatial
Tab. 1. Bistatic resolutions. r
components. The vector r (x,y) stands for the spatial location of
The images obtained from the proposed bistatic the point represented in a Cartesian coordinates (x, y, z) at the
configurations are presented in Fig. 6, Fig. 7 and Fig. 8. time t. During this work we omit the time component. The
process is supposed to be stationary and ergodic. In this case, the
surface behaviour can completely be determined by the
knowledge of its spectrum and its slope density probability.
a) Sea slope distribution
As far as slopes are randomly distributed, the first form to
define sea slopes density probability was the Gaussian
distribution. It takes into account the wind speed effect.
However, this function is symmetric with the up wind direction.
This property is not consistent with the physical sense.
Cox and Munk studied the slopes distributions of smaller
waves on the sea surface; they get measures from an aircraft by
Fig. 6. First bistatic SAR image. means of photographs of the glitter pattern. Lines with equal
glitter intensity represent points on the sea surface where the
slope reflects the sun to the camera. Both, the proposed theory
and results show that the number of glitter points was the
greatest at the specular point and that the probability density
function for this glitter was nearly bivariate normal with the
variances in the upwind direction being larger than the variances
in the crosswind direction. They also underlined the asymmetry
of the upwind density function, a property that wasn’t respected
in the previous model. These observations were interpreted in
their slopes probability distribution.
b) Sea spectrum
During the 60’s, sea spectra have been studied at various
Fig. 7. Second bistatic SAR image. frequencies band (L to Ku). Classically, a sea spectrum is set to
S (k , φ ) = M (k ) f (k , φ ) , (6)
where M(K) represents the isotropic part of the spectrum
modulated by the angular function f(K, φ ). K and φ are
respectively the spatial wave number and the wind direction. In
this section an overview of models from literature: Pierson-
Moskowitz [7] and Elfouhaily[8] is proposed.
Pierson-Moskowitz model:
The semi-empirical sea spectrum presented in this section
is based upon studies by Pierson and Pierson-Moskowitz. They
defined a band spectrum depending on the wave’s nature
(capillary or gravity wave). Fung and Lee introduced little
changes to answer the continuity properties. The spectrum form
is then given by Fig. 10 Spreading function for different wind velocities.
S 1 (K ), K < K j
S (K ) = (7)
S 2 ( K ), K ≥ K j
The previous sea models will be widely used in RCS
valuation and bistatic SAR imaging.
where Kj=0.04 rad/cm so that S1(Kj)=S2(Kj).
To account for the anisotropic effect due to wind direction, V. BISTATIC SCATTERING COEFFICIENT
Pierson come up with the angular function f( φ ), ( φ is the wind ESTIMATION
direction), it was adopted by Fung to f(K, φ ) [9].
( )
f (K , φ ) = a0 + a1 1 − e − bK cos( 2φ ) ,
2
(8)
Many approaches were developed to evaluate the
electromagnetic sea surface scattering; each is available for
1− R
a 0 = (2π ) ,
−1 given hypothesis and conditions. In this paper, three models:
where b ≈ 1.5 cm 2 , a1 = and
1+ R Kirchhoff, small perturbations and two-scale models are
proposed.
0.003 + 1.92 × 10 −3 U 12.5 .
R=
3.16 × 10 −3 U 12.5 a) Kirchhoff approximation
The vector formulation of the Kirchhoff method is based
Elfouhaily model: upon the second Green theorem, witch states that the scattered
It was introduced in 1997 and was also called unified field at any point within the source-free region bounded by a
spectrum. Its objectives were to reproduce some observations closed surface can be expressed in terms of the tangential fields
such as those inferred by Cox and Munk and to introduce the to the surface. A mathematical statement of this is as follows
wave age dependency (fetch). The spectrum definition is much e − jkR 0
( ) , (11)
b) Small-perturbation model
This model was introduced by Rice. Starting from the
Maxwell equations and using the electromagnetic field
expression combined with the reflectivity coefficient, a six
differential equation system is generated. Using the first order
Fig. 9. Spectrum function for different wind velocities. Fourier development, the system is then simplified to a linear
one. This is possible when the surface irregularities are small
To cover the bidimentional domain, the unified spreading within the electromagnetic wavelength. The bistatic scattering
function was defined as follows: coefficient for either horizontally or vertically polarized incident
f (K , φ ) =
1
[1 + ∆(K ) cos(2φ )] . (10) wave is
2π
= 8 k 2α pq cos(θ ) cos(θ s ) S (k x + k sin(θ ), k y ) ,
2
This function is centrosymmetric as required by Guissard σ pq
s
Second bistatic
where kd is the two-scale cutoff, it is estimated to k/3 [11]. −4
When located in the gravity domain (K < kd), the
configuration
4,05.10 1,34.10 −3
electromagnetic scattering wave is estimated by Kirchhoff
approach. In the capillary zone, the CRS is evaluated using an Third bistatic 1,44 1,76.10 −2
configuration −2
adapted small perturbation model. The idea is to focalise on the 1,78.10 1,36
local reference of the incident wave and use the classical small Tab. 3. Scattering matrix of each acquisitions.
perturbation approach. The result is then adapted to the real
surface by a tilting process. This is necessary to average the pq, The values for a vv polarization are introduced in the
see eq.(12), over the slope distribution (as viewed by the received signal characteristic, and the obtained images are
receiver) for the large-scale waves. presented from Fig 11 to Fig. 14.
v’s ⋅v s h’s ⋅v s S v ’s ⋅v ’ S v ’s ⋅h ’ v’⋅v v’⋅h σ vvs σ vhs ,
[S ] = =
S h ’s ⋅h ’ h’⋅v h’⋅h σ hvs
’s ⋅hs
v h’s ⋅hs S h ’s ⋅v ’ σ hhs
(15)
where
σ pq ≤ ( p.v’s ) (q.v’) σ v ’ v ’ + ( p.v’s ) (q.h’) σ v ’ + ( p.h’s ) (q.v’) σ h ’s v ’
s 2 2 2 2 2 2
h’
s s
(16)
+ ( p.h’s ) (q.h’) σ h ’s h ’ + ( p.h’s ) (q.v’)(q.h’)σ h1 ’s h ’h ’s v ’
2 2 2
VII. CONCLUSION
REFERENCES