Anda di halaman 1dari 4

SAR Processing with Motion Compensation using the Extended

Wavenumber Algorithm
Emmanouil Alivizatos, Athanasios Potsis, National Technical University Of Athens, Greece
Andreas Reigber, Technical University Berlin, Germany
Alberto Moreira, German Aerospace Center (DLR), Microwaves and Radar Institute, Germany
Nikolaos Uzunoglu, National Technical University Of Athens, Greece

Abstract
Modern Synthetic Aperture Radar (SAR) systems are continuously developing into the direction of higher spa-
tial resolution and new modes of operation. This requires the use of high bandwidths, combined with wide azi-
muthal integration intervals. For focusing such data, a high quality SAR processing method is necessary, which
is able to deal with more general sensor parameters. Wavenumber domain (Omega-K) processing is commonly
accepted to be an ideal solution of the SAR focusing problem [1]. However, it is mostly applicable on space-
borne SAR data where a straight sensor trajectory is given. In case of airborne SAR data, wavenumber domain
processing has certain limitations in performing high-precision motion compensation.
In this paper, a detailed description of the motion errors in the wavenumber domain, as well as a motion com-
pensation technique in this domain is formulated. The correction of the motion errors in the two dimensional
spectral domain can result in very accurate second order motion compensation. This procedure can also be com-
bined with a 2D sub-aperture technique, which results in a fully azimuth-frequency adaptive block processing
scheme. The reason why the wavenumber MoCo can be very critical especially in low frequency-widebeam and
high squinted SAR data, is that in these cases wavelength dependent corrections become mandatory.

1 Introduction for processing airborne data with strong motion errors


and variable Doppler centroid in range or/and in azi-
1.1 Importance of the motion compen- muth direction.
sation procedure On the other hand, in the Omega-K algorithm [4], [5],
due to its processing architecture a high precision mo-
A crucial problem in most airborne SAR sensors is tion correction cannot be introduced in its processing
the compensation of motion errors, induced by at- scheme. Compared with the ECS two-step motion
mospheric turbulence (i.e. the compensation of compensation (MoCo), only one step MoCo can be
changes of the platform velocity vector in orientation applied [6]. In its very basic implementation it starts
and/or in magnitude). If not corrected, the image normally with range compression, allowing a first or-
quality will considerably degrade [2]. der MoCo. However, the range dependent part of the
Airborne sensors, in contrary to spaceborne sensors, MoCo, which can only be applied after the correction
always show deviations from the ideal flight track. of RCM, is not possible in Omega-K [6].
SAR imaging from such unstable platforms requires
an accurate measurement of the antenna position dur-
ing the flight and a modified processing scheme, 1.3 Extended Wavenumber Algorithm
which takes into account the non-linear movement of (Extended Omega-K)
the sensor.
Several processing schemes, which try to compensate
1.2 The ECS and Wavenumber Do- this problem, using different approaches, can be
main approach for motion compensation found in the literature [7], [8]. The technique de-
scribed in [6] for example is based on the Taylor ap-
Chirp Scaling (CS) algorithm [3] was developed proximation of the phase in the two dimensional spec-
mainly to avoid interpolations, which are necessary tral domain performs a scaling of the data in range by
when we had to deal with strong range-cell migration a scaled inverse range FT. In contrary the EOK algo-
data (e.g. when wide azimuth beam data have to be rithm, first proposed in [9], extends the basic Omega-
processed). An improved version of the CS was the K processing procedure and proposes a modified Stolt
ECS algorithm [2], which was developed originally mapping in order to apply a robust and high accurate
MoCo algorithm even in case of high squint angles transforms the initial wavenumber domain (kx, ω) to
and long synthetic apertures. (kx, kr) where:
The MoCo procedure in the EOK algorithm [9] is per- 2
⎛ ω + ωο ⎞ 2 ω
formed in two different steps. A first order MoCo kr = ⎜ ⎟ − kx − o (1)
takes place right just after or before the range com- ⎝ c/2 ⎠ c/2
pression and a second order MoCo after the Modified ωο is the center frequency, c is the speed of light and
Stolt mapping, following in principle the same MoCo kx is the Doppler wavenumber (kx = 2*π*fD/v, fD is the
correction procedure as in ECS. Doppler and v is the platform velocity).
A problem of the Extended Omega-K approach ap- After the Stolt Mapping further motion compensation
pears in case of high squint and wideband SAR data cannot be performed due to the fact that the azimuth
processing, because of the range and wavenumber chirps don't have any range dependence more (i.e.
dependence of the motion errors. A solution of this constant length and Doppler rate).
problem is presented in this paper using a sub- For this reason, a modified Stolt mapping [2] is per-
aperture technique. formed:
2 2
⎛ ω + ωο ⎞ 2 ⎛ω ⎞ 2 (2)
kr = ⎜ ⎟ − kx − ⎜ ο ⎟ − kx
2 Extended Wavenumber Do- ⎝ c/2 ⎠ ⎝ c/2 ⎠
main Algorithm The data are not azimuth focused and a second mo-
tion compensation step can be integrated. From the
2.1 Extended Omega-K processing above equation it becomes evident that the Modified
steps Stolt mapping introduces also a wavelength depend-
ent effect to the azimuth signal (and also to the mo-
The main processing steps of EOK, as analytically tion errors). An interpretation of this effect as well as
described in [2], are: a possible solution is proposed in the following sec-
i) First order Motion Compensation tions.
ii) 2D FFT
iii) Range compression
iv) Modified Stolt Mapping 2.2 Wavelength dependent nature of
v) 2D IFFT motion errors
vi) Second Order Motion Compensation
vii) Azimuth FFT In this section a formulation of the problem is per-
viii) Azimuth Compression formed, based on the signal analysis of a target’s re-
ix) Azimuth IFFT sponse before and after the modified Stolt mapping.
Range compression of the data is performed if a The response of a point target in the signal domain
chirped pulse has been transmitted. After the modified can be expressed as [11]:
Stolt mapping and the 2D IFFT are performed the δ( x - x , r − r ) → (3)
o o
data are RCM corrected but still azimuth unfocused.
Important to note is that the original formulation of ⎛ 2 R( x − xo ; ro ) ⎞ ⎛ 2 R( x − xo ; ro ) ⎞
go ⎜ t − ⎟ exp⎜ − jωo ⎟
the Omega-K algorithm does not allow for the second ⎝ c ⎠ ⎝ c ⎠
order motion compensation with enough accuracy.
This is due to the fact that after the Stolt mapping all where xo is the azimuth position of the target, ro is the
2
the SAR data in the signal domain will have the azi- range of closest approach, R( x; ro ) = ro + x2 and
muth chirps with the same length and Doppler rate
go(.) is the complex envelope of the transmitted sig-
(i.e. no range dependence). This leads to a mismatch
nal. Without loss of generality we choose xo = 0 and
between the SAR data and the motion compensation
the 2D response in the ω-kx wavenumber domain, us-
phase correction to be applied. The EOK algorithm
ing the method of stationary phase [11], is:
with a modified Stolt interpolation and with a
subaperture approach for motion compensation repre- ⎡ ⎛ ω + ωo ⎞
2 ⎤
PP ( k x ,ω ) ≅ exp ⎢ − jro ⋅ ⎜ (4)
⎟ − kx ⎥
2
sents one accurate solution for this limitation of the
⎢⎣ ⎝ c/2 ⎠ ⎥⎦
standard Omega-K algorithm.
while the geometry of the signal spectrum is charac-
2.1 Modified Stolt Mapping terized by the following relation [1]:
x kx ω + ωo
The modified Stolt mapping, introduced to the EOK =− ⇒ k x = − sin θ ⋅
2
ro + x 2 ω + ωo c/2 (5)
algorithm [2], provides the possibility for improved
second order motion compensation. In the conven- c/2
tional Omega-K algorithm, the Stolt mapping proce- x
sin θ = (6)
2
dure [11] corresponds to an interpolation, which ro + x 2
t ω
x kx
2D Fourier
Transform

2D Inverse
Fourier
Transform

(a)
kr kr
x
kx
Azimuth
Inverse
Fourier
Transform

Azimuth
Fourier (c)
Transform

(b)

The different curves in the 2D spectrum domain are


Fig. 1 (a) Correspondence between the different as-
shown in Fig 1b. Furthermore, the image of the re-
pect angles (left) and the straight lines in the signal
sponse in the kr-azimuth domain is illustrated. To re-
spectrum (right), (b) 2D spectrum (left) and azimuth -
turn back to the signal domain, only an inverse FT in
‘range frequency’ (right) of the target’s response after the kr direction is needed. After the inverse FT, targets
modified Stolt mapping (simulation parameters: 120 are separated in range. But a problem appears in
MHz center frequency, 30 MHz bandwidth, PRF = every target’s response because of the way the inverse
100 Hz and –25˚ to 25˚ aspect angle) and (c) wave- FT in kr is calculated. This effect can be verified by
length dependent position shift of the motion errors in observing that every horizontal line away from the
the phase response. azimuth center of the target truncates curves corre-
From the above equation becomes evident that there sponding to different aspect angles. As a result of this,
is an one-one correspondence between the angle θ in the signal domain, an integrated error from an in-
and the parametric straight line, in the wavenumber terval of the target’s illumination time is accumulated
domain, given by Eq. 5. This correspondence is also to the target’s final signal domain response (see Fig
demonstrated in Fig 1a. 1c).
It must be noted, that because the deviation from the This effect becomes even more important in high
nominal track changes as the platform moves, the de- squint processing, because an integrated motion error
viation is only a function of the target’s range and as- from the entire target’s illumination time is accumu-
pect angle but also of azimuth. As a consequence, dif- lated. It must be noted that this wavelength dependent
ferent aspect angles (equivalently azimuth positions position of the motion errors exists also to the initial
for a specific target) of the aperture correspond to dif- raw data, but the second term of the phase response
ferent deviations. These lines can be characterized as after the Stolt mapping (see Eq. 7) is not kr depend-
iso-MoCo lines. ent, in contrary to the phase response before the Stolt
The modified Stolt mapping is performed by the mapping. Consequently, the expansion is correct for
change of variable given by Eq. 2. The response in the center frequency (kro = 2 ωo/c) but not for other kr.
the modified kr-kx wavenumber domain is:

⎡ 2 ⎤ 2.3 Solution of the problem using sub-


⎛ω ⎞
PPmod − SM ( k x ,ω ) ≅ exp ( − jro ⋅ k r ) exp ⎢ − jro ⋅ ⎜ o ⎟ − k x 2 ⎥ aperture technique
⎢⎣ ⎝ c/2⎠ ⎥⎦

(7) Block processing technique can be performed to the


and the straight line (see Eq. 5) that corresponds to a data in order to compensate the above-described ef-
certain angle becomes (after the change of variable) a fect. Based on this technique, for every azimuth-range
parametric curve defined by the following equation: position, the motion errors from different angles are
unwrapped and then compensated.
2
⎛ω ⎞ (8) After performing a 2D FFT to every block of the im-
k r = k x ⋅ cot θ − ⎜ o ⎟ − k x
2

⎝ c / 2⎠ age, the parameter θ of the curve that passes from


every (kx, kr) point is calculated from Eq. 8:
2
the computational burden and requires extra precau-
⎛ω ⎞ 2 tion to avoid unwanted phase jumps at the boundaries
kr + ⎜ o ⎟ − k x
⎝ c / 2⎠ (9) of the sub-apertures. A new version of the algorithm,
cot θ = without the use of sub-aperture, is currently under de-
kx
velopment. Instead of correcting fully the problem in
Afterwards, the kx’, which is the kx where the curve the wavenumber domain, a correction of the expan-
(see Eq. 8) crosses the horizontal axis (kr = 0) is cal- sion for a reference range is performed, using a suit-
culated: able phase function.

ωo kx
k x' =
c/2 ⎛ 2 ⎞
2 (10) 4 Literature
⎛ ω ⎞
k x + ⎜ kr + ⎜ o ⎟ − k x ⎟
2 2
⎜ ⎝c / 2⎠ ⎟ [1] C. Cafforio, C. Prati, and F. Rocca: "SAR Data Focusing Using
⎝ ⎠
Seismic Migration Techniques", IEEE Trans. on Aerosp. and
In case of high squint, the above equation becomes: Elec. Sys., Vol. 27, No. 2, pp. 194-207, 1991.
[2] A. Moreira, J. Mittermayer and R. Scheiber: ”Extended Chirp
k x' = k xd + sin( θsqu int ) ⋅ cos( θsqu int ) ⋅ kr (11) Scaling Algorithm for Air- and Spaceborne SAR Data Proc-
essing in Stripmap and ScanSAR Imaging Modes”, IEEE
Where kxd is the kx wavenumber corresponding to the Trans. on Geosc. and Rem. Sens., Vol. 34, pp. 1123-1136,
squint angle and θsquint is the squint angle. Sep. 1996.
The kx and kx’ both correspond to two different azi- [3] R.K. Raney, H. Runge, R. Bamler, I. Cumming and F. Wong:
muth positions (according to stationary phase theo- ”Precision SAR Processing without interpolation for range
rem). The deviation (function of the azimuth position) cell migration correction”, IEEE Trans. on Geosc. and Rem.
for every point of the wavenumber domain (kx, kr) Sens., Vol. 32, pp. 786-799, July 1994.
corresponds to a specific azimuth position. This spe- [4] G. Franceschetti and G. Schirinzi ”A SAR processor based on
cific azimuth position corresponds to the sum of the two dimensional FFT codes”, IEEE Trans. On Aerosp. Elec.
reference azimuth position x’ of the specific block and Sys., Vol. 26, pp. 356-366, Mar. 1990.
the time difference between the previously described [5] R. Bamler: “A comparison of range-Doppler and wavenumber
azimuth positions. Furthermore, it can be expressed as domain SAR focusing algorithms”, IEEE Trans. on Geosc.
the deviation, which corresponds to the reference and Rem. Sens., Vol. 30, No. 4, pp. 706-713, July 1992.
azimuth position shifted by the following time inter- [6] C. Cafforio, C. Prati, and F. Rocca: ”Full Resolution Focusing
val: of SEASAT SAR Images in the Frequency-Wavenumber
Domain”, Int. Journal of Rem. Sens., Vol. 12, pp. 491-510,
⎛ kx k x' ⎞ (12) 1991.
time _ difference = r' ⋅⎜ − ⎟
⎜ 2 2 2 ⎟ [7] G. Fornaro: ”Trajectory Deviations in Airborne SAR: Analysis
⎝ k ro − k x k ro − k x ' 2 ⎠ and Compensation”, IEEE Trans. on Aerosp. and Elec. Sys.,
ω (13) Vol. 35, No. 3, pp. 997-1009, July 1999.
k ro = o
c/2 [8] R. Lanari: ”A new method for the comparison of the SAR range
where r’ is the reference range position of the specific cell migration based on the chirp Z-transform”, IEEE Trans.
block. on Geosc. and Rem. Sens., Vol. 33, pp. 1296-1299, Sept.
Accurate motion compensation is achieved by per- 1995.
forming the above procedure for every block and [9] A. Reigber, A. Potsis, E. Alivizatos, N. Uzunoglu and A.
every point (kx, kr) of the 2D spectrum. Moreira, “Wavenumber Domain SAR Focusing with Inte-
grated Motion Compensation”, IGARSS 2003.

3 Conclusions and Future Work


A detailed description of the motion errors position in
the wavenumber domain as well as a motion compen-
sation technique in this domain was analytically for-
mulated in this paper. It was demonstrated, that the
correction of the motion errors in the two-dimensional
spectral domain could result in very accurate 2nd order
motion compensation. This procedure was combined
with a 2D sub-aperture technique resulting in a fully
azimuth-frequency adaptive block processing scheme.
Although the proposed MoCo algorithm successfully
compensates in several SAR system operational
modes, the use of sub-aperture techniques increases