Range migration algorithm for airborne squint

mode spotlight SAR imaging

H.-S. Shin and J.-T. Lim
Abstract: Since the reference signal based on the xed reference range is used in the range
migration algorithm (RMA), the RMA is not available to process an airborne squint-mode spotlight
synthetic aperture radar (SAR) data. Thus, the modied reference signal to transform a squint-mode
data to a broadside-mode data is introduced on the basis of the coordinate transformation and the
extended Taylor approximation. Then, using the principle of the stationary phase, the presented
formulation is analysed. Moreover, to compensate curvature errors, the proposed method is
extended on the basis of the subarea technique. Finally, the effectiveness of the proposed
method is demonstrated by some numerical simulations via a pulsed spotlight SAR simulator.
1 Introduction
Spotlight synthetic aperture radar (SAR) has been used to
obtain a high azimuth resolution. Several algorithms have
been developed for the reconstruction of spotlight SAR
data [16]. Conventional reconstruction methods such as
the polar format algorithm (PFA) are typically used under
the plane wave assumption [14]. However, the range
migration algorithm (RMA) technique increases the image
quality by utilising the actual spherical wave. Moreover,
since the RMA technique has the computation efciency
and produces acceptable results in the broadside situation,
it is available for SAR data processing [1, 3, 5]. However,
if spotlight images are recorded in the squinted geometry,
crucial problems arise in the RMA technique. That is,
since the RMA technique requires the reference signal
based on the xed reference range unlike the PFA tech-
nique, the phase of scene centre scatter is changed after
mixing the received signal and the complex conjugated
reference signal [1]. Thus, the RMA technique cannot
yield an acceptable performance in the squint-mode. If
this problem is not properly compensated during the SAR
data processing, the quality of the nal image in the squint-
mode SAR is degraded. Also, since squinting could increase
the exibility of the ight path, it is necessary to process the
squint-mode SAR data. Though several methods have been
studied in the stripmap SAR with a squint angle, in the case
of spotlight SAR with a squint angle, it is a still difcult
task to obtain the high quality images [7, 8]. In particular,
in the case of the squint-mode spotlight SAR, the PFA
technique has been accommodated [1, 7]. However, this
algorithm limits the quality of the nal image because
of polar-to-rectangular interpolation and has a higher com-
putational burden due to two interpolations compared to the
RMA technique with one interpolation [1]. Thus, we extend
the original RMA technique to introduce an efcient way
for the squint-mode spotlight SAR imaging. We modify a
reference signal based on the extended Taylor approxi-
mation and the coordinate transformation. Thus, using
the modied reference signal, we can transform an original
squint-mode raw data to a broadside-mode raw data. Then,
with the new coordinate, we analyse the formulation using
the principle of the stationary phase. Moreover, to compen-
sate curvature errors in the proposed method, we extend the
proposed method based on the subarea technique. Finally,
we test the proposed algorithm with some numerical
simulations via a pulsed spotlight SAR simulator.
2 Signal model of airborne squint mode
spotlight SAR
The geometric model in Fig. 1 provides the basis for a
simple airborne squint-mode spotlight SAR model. The
SAR sensor travels a straight-line ight path parallel to
the x-axis during a synthetic aperture length L. Moreover,
the SAR sensor travels at constant velocity and altitude.
To maintain uniform spatial sampling, the radar uses a
xed pulse repetition interval.
Let the SAR sensor position along the ight path be
(x
a
, y
a
, z
a
), where x
a
is an azimuth position, y
a
is the
constant ground distance between the ight path and the
x-axis and z
a
is a constant airborne altitude. Let
R
t

p
((x
a
2x
t
)
2
( y
a
2y
t
)
2
(z
a
2z
t
)
2
) be the distance
between the SAR sensor and the target. Also, let (x
c
, y
c
)
be a squinted scene centre position and u
s
tan
21
(x
c
/
(y
a
2y
c
)) be a squint angle [2]. The size of scene area is
X
0
Y
0
, where X
0
is the x-direction distance and Y
0
is the
y-direction distance of the scene area, respectively.
The linear FM (chirp) pulse with duration time T
p
trans-
mitted by the radar [1] is
s
t
t e
j2pf
c
t pgt
2

1
where t is time, the quantity f
c
represents the centre transmit
frequency and g is the FM chirp rate. Thus, the received
signal is
s
r
n; t rx
t
; y
t
; z
t
e
j2pf
c
t2R
t
=c
e
jpgt2R
t
=c
2
2
# The Institution of Engineering and Technology 2007
doi:10.1049/iet-rsn:20060080
Paper rst received 13th June and in revised form 1st September 2006
The authors are with the Department of EECS and Radiowave Detection
Research Center, Korea Advanced Institute of Science and Technology, 373-1,
Guseong-dong, Yuseong-gu, Daejeon 305-701, Republic of Korea
E-mail: [email protected]
IET Radar Sonar Navig., 2007, 1, (1), pp. 7782 77
Authorized licensed use limited to: Korea Advanced Institute of Science and Technology. Downloaded on February 8, 2010 at 00:47 from IEEE Xplore. Restrictions apply.
where c is the speed of microwave, n is the pulse number
and r(x
t
, y
t
, z
t
) is the reectivity function of the target.
3 Modied RMA
3.1 Proposed method for small scene
To use the RMA technique in the squint-mode, we need to
have the raw data in a broadside-mode. Thus, to transform
an original squint-mode raw data to a broadside-mode raw
data, we introduce the following technique.
First, as shown in Fig. 2, we rotate only the ight path.
Thus, we obtain a rotated SAR sensor position as follows
x
ra
y
ra
z
ra

cos u
s
sin u
s
0
sin u
s
cos u
s
0
0 0 1

x
a
0
0

0
y
a
z
a

3
From (3), the distance R
rt
between the rotated SAR sensor
and the target is
R
rt

x
ra
x
t

2
y
ra
y
t

2
z
ra
z
t

x
a
Dx
a
x
t

2
y
a
Dy
a
y
t

2
z
a
z
t

4
where Dx
a
x
ra
2x
a
(cos u
s
21)x
a
and D y
a
y
ra
2y
a

sin u
s
x
a
.
Then, as shown in Fig. 3, to express the coordinate of the
broadside-mode, we use a coordinate transformation as
follows
x
y
z

cosu
s
sinu
s
0
sinu
s
cosu
s
0
0 0 1

x x
c
y y
c
z

5
Let (x
a
, y
a
, z
a
) and (x
t
, y
t
, z
t
) be the rotated SAR sensor
position and the target position for the new coordinate
(x, y, z), respectively. Note y
a

p
(x
c
2
( y
a
2y
c
)
2
is the
ground distance between the x-axis and the new ight
path. Also, let R

p
( y
a
2
z
a
2
) be the xed slant distance
between the x-axis and the new ight path. Thus, the dis-
tance R

t
between the rotated SAR sensor and the target in
the new coordinate can be expressed as follows

R
t

x
a
x
t

2
y
a
y
t

2
z
a
z
t

6
Since, as it follows from the coordinate transformation,
R

t
R
rt
, if we can induce R
rt
from R
t
, we can process the
squinted data. Therefore for inducing R
rt
from R
t
and
better approximating R
rt
, we extend the Taylor approxi-
mation as follows. Using the rst-order Taylor approxi-
mation in the two variables (Dx
a
, Dy
a
), R
rt
can be
rewritten as follows
R
rt

x
a

m 1
m

Dx
a
x
t

2
y
a

m 1
m

Dy
a
y
t

2
z
a
z
t

1=mDy
a
y
a
m 1=mDy
a
y
t

x
a
m 1=mDx
a
x
t

2
y
a
m 1=mDy
a
y
t

2
z
a
z
t

1=mDy
a
y
a
m 1=mDy
a
y
t

x
a
m 1=mDx
a
x
t

2
y
a
m 1=mDy
a
y
t

2
z
a
z
t

7
where the parameter m is used to adjust bounded errors
induced by the approximation and is an arbitrary positive
constant.
y
y
Flight path
: x
a
Target area
Original centre
= (0,0)
y
a
Squinted scene
centre = (x
c
,y
c
)
New flight
path
Rotation
x
Squint angle
x
Fig. 2 Top view: rotation of ight path
y
x
New target
area
New scene
centre : (0,0)
New flight
path
a
y
Fig. 3 Top view: coordinate transformation
x
z
y
Flight path
Target
at (x
t
,y
t
,z
t
)
SAR sensor
at (x
a
,y
a
,z
a
)
R
t
Squinted scene
centre = (x
c
,y
c
,0)
Original centre
= (0,0,0)
Squint angle = 0
s
(-L/2,y
a
,z
a
)
(L/2,y
a
,z
a
)
(
Fig. 1 3D geometric model for airborne squint-mode spotlight
SAR
IET Radar Sonar Navig., Vol. 1, No. 1, February 2007 78
Authorized licensed use limited to: Korea Advanced Institute of Science and Technology. Downloaded on February 8, 2010 at 00:47 from IEEE Xplore. Restrictions apply.
Similarly, we perform the Taylor approximation for the
rst term of (7). After repetitive manipulations, we obtain
R
rt

x
a
x
t

2
y
a
y
t

2
z
a
z
t

m
k1
m k 1
m

m k
m

Dx
a

x
a
m k=mDx
a
x
t

x
a
m k=mDx
a
x
t

2
y
a
m k=mDy
a
y
t

2
z
a
z
t

m
k1
m k 1
m

m k
m

Dy
a

y
a
m k=mDy
a
y
t

x
a
m k=mDx
a
x
t

2
y
a
m k=mDy
a
y
t

2
z
a
z
t

2
8
Also, since we can express x
t
and y
t
as x
c
x

t
and y
c
y

t
,
where (x

t
, y

t
) is the target position from the squinted
scene centre, we obtain
R
t

x
a
x
c
x

2
y
a
y
c
y

2
z
a
z
t

9
In the case of airborne spotlight SAR with a small scene, the
values of (x
t

, y
t

, z
t
) is much smaller than those of (x
a
, y
a
, z
a
).
Thus, from (8) and (9), we can approximate R
rt
as follows
R
rt

x
a
x
c
x

2
y
a
y
c
y

2
z
a
z
t

m
k1
1=mDx
a
x
a
m k=mDx
a
x
c

x
a
m k=mDx
a
x
c

2
y
a
m k=mDy
a
y
c

2
z
2
a

m
k1
1=mDy
a
y
a
m k=mDy
a
y
c

x
a
m k=mDx
a
x
c

2
y
a
m k=mDy
a
y
c

2
z
2
a

R
t
R 10
where
R

m
k1
1=mDx
a
x
a
mk=mDx
a
x
c
1=mDy
a
y
a
mk=mDy
a
y
c

x
a
mk=mDx
a
x
c

2
y
a
mk=mDy
a
y
c

2
z
2
a

Note that the accuracy of the approximation increases as the

tuning parameter m becomes larger and the computational
complexity of R is O(N m), where N is the total pulse
number. That is, there is a tradeoff between the accuracy
of the approximation and the computation time of R.
Hence, to obtain an equivalent data in the broadside-
mode, we use the modied reference signal as follows
s
ref
n; t e
j2pf
c
t2

R=c
e
jpgt2

R=c
2
11
where R

s
2R. Then, if we mix the received signal (2)
with the complex conjugated signal of (11), the resulting
signal is
sn; t; x
a
; y
a
; z
a
rx
t
; y
t
; z
t
e
jFn;t;x
a
;y
a
;z
a

12
with F(n, t; x
a
, y
a
, z
a
) 2(4pg/c)(( f
c
/g) t 2
(2R

/c)) (R
t
2R

) (4pg/c
2
)(R
t
2R

)
2
.
From the phase term of (12), we obtain the following
phase term
Fn; t;x
a
; y
a
; z
a

4pg
c
f
c
g
t
2

R
s
c

R
t

8pg
c
2

R
s
R
t

R
4pg
c
2
R
t

R
2
13
where (8pg/c
2
) (R

2R

s
)(R
t
2R

) (4pg/c
2
)(R
t
2R

)
2
is
the residual video phase term. To remove it, we modify the
range deskew process [1]. From the range deskew process,
the instantaneous frequency is
f
1
2p
dF
dt
14
Evaluating this derivative yields
f
2g
c
R
D
1
2
c
dR
D
dt

2g
c
t
2

R
c

f
c
g

dR
D
dt
15
using R
D
R
t
2R

for convenience. Thus, from f 2(2g/

c)R
D
, we introduce the following phase compensation term
s
com
f
t
e
jpf
2
t
=g4p

R
s
=cf
t

16
where f
t
is the frequency variable while t is the time vari-
able. After application of s
com
( f
t
), the phase term becomes
Fn; t;x
a
; y
a
; z
a

4pg
c
f
c
g
t
2

R
s
c

R
t

R 17
From R
t
R
rt
2R and R

s
2R, we obtain the following
phase term
Fn; t;x
a
; y
a
; z
a

4pg
c
f
c
g
t
2

R
s
c

R
rt

R
s
18
Thus, from R

t
R
rt
, we obtain the following phase term
Fn;

k
r
; x
a
; y
a
; z
a

k
r

R
t

R
s
19
where k

r
(4pg/c)(( f
c
/g) t 2(2R

s
/c)). Then, the azimuth
Fourier transform of s(n, k

r
; x
a
, y
a
, z
a
) with respect to x
a
yields
Sn;

k
x
;

k
r

r x
t
; y
t
; z
t
e
j

k
r

x
a
x
t

R
2
b

k
x
x
a
d x
a
20
where R

p
(( y
a
2y
t
)
2
(z
a
2z
t
)
2
) and k

x
is an azimuth
spatial frequency. Using the principle of the stationary
phase to evaluate (20), we obtain the following matched
lter
F
mf
n;

k
x
;

k
r

k
r

R
s

k
2
r


k
2
x

R
s
21
A change of variables, known as the Stolt interpolation is
dened as k

p
(k

r
2
2k

x
2
) [1, 9]. Thus, after application
of the matched lter and the Stolt interpolation, the phase
term becomes
F
rma
n;

k
x
;

k
y

k
y

R
b

R
s


k
x
x
t
22
Hence, the signal becomes
S
rma
n;

k
x
;

k
r
r x
t
; y
t
; z
t
e
j

k
y

R
b

R
s
j

k
x
x
t
23
From(23), if we perform a 2D inverse FFT for the distributed
IET Radar Sonar Navig., Vol. 1, No. 1, February 2007 79
Authorized licensed use limited to: Korea Advanced Institute of Science and Technology. Downloaded on February 8, 2010 at 00:47 from IEEE Xplore. Restrictions apply.
targets, we can obtain the SAR images. Moreover, as shown
in Figs. 2 and 3, the reconstructed targets are not in their orig-
inal position because of the transformation. Thus, to rotate
and translate the targets to the original position, we need
the following transformation matrix
x
t
y
t

cos u
s
sin u
s
sin u
s
cos u
s

x
t
y
t

x
c
y
c

24
3.2 Proposed method for large scene
If the proposed method shown in Section 3.1 produces a
desired image for the full scene, we dene the case as the
smallness of a scene. However, as the scene size becomes
larger, curvature errors induced by the effect of the scene
size in the approximation of R
rt
increase. Since the errors
cause distortion and defocusing that vary with target
location, the curvature errors can limit the scene size. If
the proposed method shown in Section 3.1 produces an
acceptable image for the centre region of the scene area
but does not produce an acceptable image for the distant
region from the centre of the scene area, we use the
subarea technique [1]. Thus, using the following steps, we
can extend the proposed method shown in Section 3.1 for
airborne squint-mode spotlight SAR with a large scene.
Step 1: As shown in Fig. 4, we divide the scene size into a
number of smaller scenes (subareas). Note M P Q is
the total number of subareas, where P and Q are the
subarea numbers of the x-direction and y-direction, respect-
ively. Moreover, U
( p,q)
for p 1, . . . , P and q 1, . . . , Q
is the ( p, q)th subarea.
Step 2: For each subarea, we dene a new squinted scene
centre, a new squint angle and a new ground distance
between the new axis and the rotated ight path. That
is, let the scene centre position of the ( p, q)th sub-
area be u
( p,q)
(a
p
, b
q
), where a
p
x
c
(X
0
/2) 2(X
0
/
2P) 2(( p 21)/P)X
0
and b
q
y
c
2(Y
0
/2) (Y
0
/2Q)
((q21)/Q)Y
0
. Moreover, in the ( p, q)th subarea, let
u
s
( p,q)
tan
21
(a
p
/(y
a
2b
q
)) be the squint angle and
y
a
( p,q)

p
(a
p
2
( y
a
2b
q
)
2
) be the range distance.
Step 3: As shown in Fig. 4, to rotate the ight path and
transform the coordinate for each subarea, we use the
squint angles given in step 2. Then, for each subarea, we
perform the proposed method given in Section 3.1.
Step 4: To reduce the processing time, we perform the pro-
posed method via multi-processor. Then, we add independently
the subareas together to form an image of the full scene.
4 Simulation results
In order to show the validity and effectiveness of the pro-
posed method, we simulate a radar imaging of 2D targets
using MATLAB. The basic simulation parameters are as
follows:
Centre frequency ( f
c
): 10 GHz
Flight length (L): 300 m
Pulse width (T
p
): 1 ms
Bandwidth (B): 250 MHz
Pulse repetition frequency: 1000 Hz
Platform velocity (v): 100 m/s
Sampling frequency (F
s
): 350 MHz
Range ground distance ( y
a
): 5000 m
Flight altitude (z
a
): 2000 m
Tuning parameter (m): 1000
4.1 Case 1: Small scene
As shown Figs. 57, the proposed method produces the
desired results when the targets are near from the squinted
scene centre (small scene). Moreover, as the squint angle
becomes larger, the proposed method generates better
results compared with the PFA technique. In particular,
for (x
c
, y
c
) (2000 m, 200 m), the PFA technique is dif-
cult to reconstruct the desired images. Also, as shown
in Table 1, since the bounded error induced by the approxi-
mation becomes smaller by increasing the parameter m,
Fig. 4 Top view: subarea technique
IET Radar Sonar Navig., Vol. 1, No. 1, February 2007 80
Authorized licensed use limited to: Korea Advanced Institute of Science and Technology. Downloaded on February 8, 2010 at 00:47 from IEEE Xplore. Restrictions apply.
we can obtain the high-quality image. However, the com-
putation time of R increases. Thus, we choose the tuning
parameter m 1000 using a tradeoff between the azimuth
resolution and the computation time of R.
4.2 Case 2: Large scene
We set (x
c
, y
c
) (1500 m, 200 m). As shown Fig. 8, the
proposed method with one subarea generates the ghost
points and the degraded results as the targets become
more distant from the squinted scene centre (large scene).
However, the proposed method with eight subareas yields
better results and produces lower sidelobes. Moreover, as
shown in Figs. 8c and d, we obtain the following image
quality parameters. The azimuth resolution (mainlobe
23 dB width) and PSLR of the proposed method with one
subarea are about 0.8 m and 26.7 dB except the ghost
Fig. 8 Point target responses
a Our method with one sub-area
b Our method with eight sub-areas
c Azimuth response
d Azimuth response
Fig. 7 Point target responses via proposed method
a (x
c
, y
c
) (0 m, 0 m)
b (x
c
, y
c
) (1000 m, 200 m)
c (x
c
, y
c
) (1500 m, 200 m)
d (x
c
, y
c
) (2000 m, 200 m)
Fig. 6 Point target responses via PFA technique
a (x
c
, y
c
) (0 m, 0 m)
b (x
c
, y
c
) (1000 m, 200 m)
c (x
c
, y
c
) (1500 m, 200 m)
d (x
c
, y
c
) (2000 m, 200 m)
Fig. 5 Point target responses via original RMA technique
a (x
c
, y
c
) (0 m, 0 m)
b (x
c
, y
c
) (1000 m, 200 m)
c (x
c
, y
c
) (1500 m, 200 m)
d (x
c
, y
c
) (2000 m, 200 m)
IET Radar Sonar Navig., Vol. 1, No. 1, February 2007 81
Authorized licensed use limited to: Korea Advanced Institute of Science and Technology. Downloaded on February 8, 2010 at 00:47 from IEEE Xplore. Restrictions apply.
points, respectively. However, the azimuth resolution and
PSLR of the proposed method with eight subareas are
about 0.25 m and 213.75 dB, respectively.
5 Conclusions
To obtain SAR images with high quality for airborne
squint-mode spotlight SAR, we have proposed an efcient
compensation technique based on the extended Taylor
approximation and the coordinate transformation. Using
the modied reference signal, we have formulated and
analysed the phase information for the new broadside-
mode coordinate using the principle of the stationary
phase. Moreover, to compensate the curvature errors, we
have extended the proposed method based on the
subarea technique. Finally, we have shown the effective-
ness of the proposed algorithm through some numerical
simulations.
6 Acknowledgments
This research was supported by the Agency for Defense
Development, Republic of Korea, through the Radiowave
Detection Research Center at Korea Advanced Institute of
Science and Technology.
7 References
1 Carrara, W.G., Goodman, R.S., and Majewski, R.M.: Spotlight
synthetic aperture radar signal processing algorithms (Artech
House, 1995)
2 Soumekh, M.: Synthetic aperture radar signal processing with
MATLAB algorithms (John Wiley & Sons, Inc., 1999)
3 Cumming, I.G., and Wong, F.H.: Digital processing of synthetic
aperture radar data (Artech House, 2005)
4 Yuan, Y., Sun, J., and Mao, S.: PFA algorithm for airborne spotlight
SAR imaging with nonideal motions, IEE Proc., Radar Sonar Navig.,
2002, 149, (4), pp. 174182
5 Cafforio, C., Prati, C., and Rocca, E.: SAR data focusing using
seismic migration techniques, IEEE Trans. Aerosp. Electro. Syst.,
1991, 27, (2), pp. 194207
6 Cantalloube, H., and Dubois-Fernandez, P.: Airborne X-band SAR
imaging with 10 cm resolution: technical challenge and preliminary
results, IEE Proc., Radar Sonar Navig., 2006, 153, (2), pp. 163176
7 Davidson, G.W., Cumming, I.G., and Ito, M.R.: A chirp scaling
approach for processing squint mode SAR data, IEEE Trans.
Aerosp. Electro. Syst., 1996, 32, (1), pp. 121132
8 Yeo, T.S., Tan, N.L., Zhang, C.B., and Lu, Y.H.: A new subaperture
approach to high squint SAR processing, IEEE Trans. Geosci.
Remote Sens., 2001, 39, (5), pp. 954968
9 Stolt, R.H.: Migration by Fourier transform techniques, Geophysics,
1978, 43, pp. 2348
Table 1: Azimuth resolution (mainlobe 23 dB width) and computation time of R for (x
c
, y
c
) 5 (2000 m, 200 m)
Analysed
parameters
Tuning parameters
m 1 m 10 m 100 m 1000 m 10 000 m 100 000 m 1 000 000
Azimuth resolution, m 4.9 0.27 0.26 0.25 0.25 0.25 0.25
Computation time of R, s 0.047 0.047 0.266 2.453 24.563 274.313 2588.645
IET Radar Sonar Navig., Vol. 1, No. 1, February 2007 82
Authorized licensed use limited to: Korea Advanced Institute of Science and Technology. Downloaded on February 8, 2010 at 00:47 from IEEE Xplore. Restrictions apply.