Page 1
www.scichina.cominfo.scichina.com
www.springerlink.com
Synthetic aperture radar tomography sampling
criteria and three-dimensional range migration
algorithm with elevation digital spotlighting
TAN WeiXian1,2,3†, HONG Wen1,2, WANG YanPing1,2, LIN Yun1,2,3 & WU YiRong1,2
1 State Key Laboratory of Microwave Imaging Technology, Beijing 100190, China;2 Institute of Electronics, Chinese Academy of Sciences, Beijing 100190, China;3 Graduate School, Chinese Academy of Sciences, Beijing 100190, China
Based on the general geometric model of multi-baseline Synthetic Aperture Radar Tomography (To-moSAR), the three-dimensional (3-D) sampling criteria, the analytic expression of the 3-D Point SpreadFunction (PSF) and the 3-D resolution are derived in the 3-D wavenumber domain in this paper. Con-sidering the relationship between the observation geometry and the size of illuminated scenario, a 3-DRange Migration Algorithm with Elevation Digital Spotlighting (RMA-EDS) is proposed. With this algo-rithm 3-D images of the area of interest can be directly and accurately reconstructed in the 3-D spaceavoiding the complex operations of 3-D geometric correction. Finally, theoretical analyses and simu-lation results are presented to demonstrate the shift-varying property of the 3-D PSF and the spatial-varying property of the 3-D resolution and to demonstrate the validity of the 3-D RMA-EDS.
synthetic aperture radar tomography (TomoSAR), three-dimensional (3-D) SAR imaging, 3-D sampling criteria, 3-D resolution,
3-D range migration algorithm with elevation digital spotlighting (RMA-EDS)
1 Introduction
The conventional two-dimensional (2-D) synthetic
aperture radar (SAR) only provides 2-D images
along the slant range and the azimuth direc-
tion, which represents the projection of the three-
dimensionally (3-D) distributed objects over the 2-
D plane. Therefore, it has the disadvantages of
shading, foreshortening and layover due to the in-
fluence of relative geometry between radar and the
investigated scene and other factors. With Inter-
ferometric SAR[1] (InSAR) technique, it is feasible
to generate the digital elevation model (DEM) of
the illuminated area. However, the elevation sam-
pling number and spacing are both limited, severe
elevation ambiguity occurs, and only the surface
height information is obtainable. That is to say,
the distribution of the objects in 3-D space remains
unknown.
Received January 25, 2008; accepted June 20, 2008
doi: 10.1007/s11432-009-0003-2†Corresponding author (email: [email protected] )
Supported by the National Science Fund for Distinguished Young Scholars (Grant No. 60725103), the National Natural Science Foundation of
China (Grant No. 60602015), the National Key Laboratory Foundation (Grant No. 9140C1903030603) and the Knowledge Innovation Program
of Chinese Academy of Sciences (Grant No. 07QNCX-1154)
Sci China Ser F-Inf Sci | Jan. 2009 | vol. 52 | no. 1 | 100-114
Page 2
Tomographic imaging technique was originally
developed in the field of medical imaging. Since the
middle of the 1980s, Farhat[2,3], Jakowatz[4], Knaell
et al.[5] have analyzed the relationship between to-
mographic imaging and SAR imaging. Since the
late 1990s, Reigber et al.[6] extended tomographic
imaging concept to 3-D SAR imaging, proposed the
basic concept multi-baseline SAR tomography (To-
moSAR) imaging and reconstructed the first air-
borne TomoSAR 3-D imagery. However, the pro-
posed 3-D range migration algorithm (RMA) was
deduced without considering the influence of the
3-D sampling spacing. In addition, the second syn-
thetic aperture was considered to be formed in the
normal direction perpendicular to the line-of-sight
and the azimuth direction, the center of the eleva-
tion synthetic aperture was symmetrical about the
scene in the elevation direction in the TomoSAR
geometry.
Contemporaneously, She et al.[7] reconstructed
the 3-D satellite SAR image with ERS-1 (European
Remote Sensing satellite) data using beamforming
method in elevation focusing. Later on, Fornaro et
al.[8,9] developed the spectral estimation techniques
and inversion methods with long-term spaceborne
ERS-1/2 data. Pi et al.[10] also developed a spec-
tral estimation technique for tomographic imaging.
Recently, Fortuny et al.[11] have developed the 3-
D SAR near field imaging algorithms and provided
a lot of significant results for SAR 3-D imaging
technique. A version of 3-D RMA was proposed to
verify the planar scanning geometry. But the algo-
rithm implies that the middle of the elevation syn-
thetic aperture is symmetrical about the scene in
the elevation direction. The relationship between
the size of the ground-range and elevation direction
is not considered. Since 2005, Frey et al.[12] have
been engaged in the time-domain back-projection
algorithm for TomoSAR experimental data focus-
ing.
In this paper, we introduce a more general geo-
metric model for TomoSAR imaging, propose the
3-D range migration algorithm with elevation digi-
tal spotlighting (RMA-EDS) for the general geome-
try and extend the existing 3-D RMA developed by
Reigber[6], Fortuny[11] and Soumekh[13]. The paper
is organized as follows. In section 2, we introduce
the general imaging geometry. In section 3, we de-
rive the 3-D sampling criteria, and present the 3-D
Point Spread Function (PSF) and the 3-D resolu-
tion of TomoSAR system. In section 4, we make
a review of the work in refs. [6, 11, 13], and pro-
pose a 3-D RMA-EDS algorithm which is suitable
for processing the chirped and stepped frequency
TomoSAR data with less elevation sampling num-
ber. In section 5, we conduct the simulation exper-
iment to validate the 3-D RMA-EDS and illustrate
the capabilities of removing the layover and fore-
shortening in the traditional 2-D imaging geome-
try. Then we test the quality parameters of the
3-D TomoSAR image and reveal the shift-varying
property of the PSF and the spatial-varying prop-
erty of the 3-D resolution of the TomoSAR system.
Finally, in section 6, we draw some conclusions and
outline the future work.
2 TomoSAR imaging geometry
We consider the imaging scenario in a 3-D spatial
domain OXY Z, where X-axis, Y -axis and Z-axis
refer to the ground-range, azimuth (flight track)
and elevation direction, respectively. The illumi-
nated area is located at one side of the flight direc-
tion, as shown in Figure 1. A synthetic aperture,
which relates to the variation of the instantaneous
aspect angle of the radar antenna azimuth beam,
is formed with the movements of the radar in the
azimuth direction. An additional synthetic aper-
ture, which relates to the variation of the off-nadir
angle, is formed in the elevation direction (Z) with
the multi-pass flight, multi-baseline or antenna ar-
ray. Note that the elevation synthetic aperture is
perpendicular to ground-range and azimuth OXY
plane in this geometric model, but not to the line-
of-sight of radar and azimuth plane[6]. When the
center of the elevation synthetic is equivalent to
zero, the geometry is simplified into the model
mentioned in refs. [6, 11, 13], and the model can
be considered as Down-Looking SAR[14] when the
X-axis and Z-axis are used to denote elevation and
ground-range direction, respectively.
For convenience, we define the coordinates sys-
TAN WeiXian et al. Sci China Ser F-Inf Sci | Jan. 2009 | vol. 52 | no. 1 | 100-114 101
Page 3
tem PUVW of target Pn in Figure 1, which re-
flects the relationship between target Pn and the
observation angle. The origin of PUVW is set at
the position of Pn, and PUVW varies with the
target Pn, U -axis refers to the average line-of-sight
direction pointing to Pn, V -axis is parallel to Y -
axis, and W -axis is perpendicular to PUV plane.
That the center of elevation synthetic aperture is
equivalent to any values is the first point in this pa-
per which is distinguished from that of the existing
geometries[6,11,13].
Figure 1 SAR tomography imaging geometry.
3 TomoSAR signal modeling and sam-pling criteria
3.1 TomoSAR signal model
Provided that the targets are aspect-angle inde-
pendent in the 3-D spatial domain with the size
of X0 × Y0 × Z0 (ground-range swath × azimuth
length × elevation height). The point target Pn is
located at (xn, yn, zn) with the reflectivity function
δn(xn, yn, zn), which is assumed to be frequency in-
variant. The SAR sensor is positioned at (x, y, z)
and is flying on the x = 0 plane with a constant
effective velocity vA. The center of the target zone
is (Xc, 0, 0). Therefore, xn ∈ [(Xc −X0/2), (Xc +
X0/2)], yn ∈ [−Y0/2, Y0/2], zn ∈ [−Z0/2, Z0/2].
The center points of the elevation synthetic aper-
ture LH are located at the line x=0, z=Zc. If the
linear frequency-modulated (FM) chirped pulses
p(t) with bandwidth B are transmitted, the ideal
echoed signal from Pn is
s(t, x = 0, y, z) = δn(xn, yn, zn)p(t− 2Rn/C), (1)
where
Rn =√
(Xc + x′n)2 + (y − yn)2 + (Zc + z′ − zn)2
is the range between the sensor and the target,
y is the sampling in the azimuth direction and
y ∈ [(yn − LA/2), (yn + LA/2)], C is the speed
of light, LA is the azimuth synthetic aperture size
determined by the −3 dB azimuth beam width
ΨA. z′ is the sampling along the elevation direc-
tion and z′ = z − Zc, z′ ∈ [−LH/2, LH/2]. x
′n =
xn − Xc, x′n ∈ [−X0/2,X0/2]. After the carrier
frequency has been removed, the slant range fre-
quency matched filtering is performed via
H1(ft) =
∫
[p(t− 2Rc/C) exp(−j2πfct)]
× exp(−j2πftt)dt
∗
, (2)
where Rc =√
X2c + Z2
c is the reference range, ft
is the slant range frequency corresponding to the
slang range sampling time t. p(t) = exp(j2πfct +
jπKt2
r )rect(t/T ), if |t| 6 T/2, rect[t/T ] = 1, else
rect[t/T ] = 0. Then the filtered function presented
in the slant range wavenumber domain is
SRC(Kω, y, z′) = δn exp−j2[Rn −Rc]Kω, (3)
whereKω = 2π(fc+ft)/C is the wavenumber of the
transmit signal andKω ∈ [Kω min,Kω max],Kω min =
2π(fc − B/2)/C,Kω max = 2π(fc + B/2)/C,Kc =
2πfc/C.
The echoed signal in slant range wavenumber do-
main from the whole target area can be denoted by
the linear sum of the backscattering information of
different targets in slant range wavenumber domain
according to the Born approximation,
SSRC(Kω, y, z′) =
∑
n
SRC(Kω, y, z′). (4)
3.2 TomoSAR sampling criteria
Both data acquisition and signal processing require
that the data should be discretized for SAR system.
To discretize the radar signal without aliasing, the
Nyquist sampling criteria should be satisfied. In
this paper, the Nyquist sampling criteria for To-
moSAR is analyzed.
3.2.1 Slant range sampling. The wavenumber
support of the echoed signal SSRC(Kω, y, z′) in the
102 TAN WeiXian et al. Sci China Ser F-Inf Sci | Jan. 2009 | vol. 52 | no. 1 | 100-114
Page 4
range direction is Ωs = [2Kω min, 2Kω max]. In or-
der to perform Fourier analysis without aliasing,
the phase difference between the two points should
satisfy 2(Kω max−Kω min)∆R 6 2π, where ∆R is
the distance between the two targets in the slant
range direction. We have
fs =1
∆t=
C
2∆R>
[
C(Kω max −Kω min)
2π
]
= B,
(5)
where fs and ∆t are the slant range sampling fre-
quency and sampling time interval. In practice,
slant range oversampling ratio is usually set be-
tween 1.1 and 1.2[15] to provide a guard band. As
for the stepped frequency system[11], the frequency
sampling spacing ∆f is governed by
∆f 6C
2
√
X20 + Z2
0
cos[ψA/2]
−1
. (6)
3.2.2 Azimuth sampling. In TomoSAR, the in-
stantaneous wavenumber observed along the Y -
axis is
Ky =∂
∂y[arg(SRC(Kω, y, z
′))]
= − 2Kω sin θn(z′) sinφn(y), (7)
where θn = arcsin√
(Xc + x′n)2 + (y − yn)2/Rn
is the instantaneous incidence angle at differ-
ent heights z′ and φn = arcsin[y − yn]/√
(Xc + x′n)2 + (y − yn)2 is the azimuth instanta-
neous aspect angle.
Therefore, the azimuth wavenumber support
band of (4) is Ωy = [−2Kω sin θn(z′) sin φn(ΨA/2),
2Kω sin θn(z′) sin φn(ΨA/2)], where φn(ΨA/2) is
the largest azimuth aspect angle of target Pn with
respect to the sensor. The whole support band
BY in the azimuth direction is approximately equal
to 4Kω sin(ΨA/2); therefore, the azimuth Nyquist
sampling spacing is
∆y 6π
2Kω max sin(ΨA/2). (8)
Hence, the pulse repetition frequency (PRF) is
PRF =βvA
π2Kω max sin(ΨA/2), (9)
where β ∈ (1.1, 1.4)[15] is the azimuth oversampling
ratio. The influences of the range and azimuth am-
biguity ratio[13,15] are not considered here.
3.2.3 Elevation sampling. In TomoSAR, the
imaging scenario is illuminated by the sensor with
different incidence angles, which is similar to the
Spotlight SAR[4] mode along the elevation direc-
tion. The instantaneous elevation wavenumber is
Kz′(Kω,Ky; z′)
=∂
∂zargFTy[SSRC(Kω, y, z
′)]
= −√
4K2ω −K2
y cos θn(z′), (10)
where FTy indicates the Fourier transform (FT)
of the signal SSRC(Kω, y, z′). In accordance with
(10), the support band of FTy[SSRC(Kω, y, z′)] in
the elevation wavenumber domain is
ΩZ =⋃
Ky
⋃
x′n,z′
[−2Kω cos θn(z′)
2Kω cos θn(z′)]. (11)
In (10) and (11), the average elevation Doppler
centroid of Pn changes with its spatial position so
that the whole elevation Doppler support is spread
by the variant elevation Doppler centroid. The
spread support band observed in the elevation di-
rection becomes
BZ = (2Kω max cos θn min−2Kω min cos θn max), (12)
where θn max and θn min are the maximum and mini-
mum instantaneous incidence angle from the sensor
to the target area. Therefore, the elevation Nyquist
sampling spacing is determined via
∆z0 6πRmin
[Kω max(LH + Z0)], (13)
where Rmin ≈ (Xc −X0/2) if |Zc| 6 (Z0 + LH)/2,
else
Rmin ≈√
(Xc −X0/2)2 + (|Zc| − (Z0 + LH)/2)2.
However, the echoed signal sampled with the
above sampling spacing has already included re-
dundant information. With a reference signal
H3(Kω,Ky, z′),
H3(Kω,Ky, z′)
= exp
j√
Xc + [Zc + z′ − zn]2
√
4K2ω −K2
y
, (14)
the elevation sampling spacing could be re-
duced without damaging any spectrum informa-
TAN WeiXian et al. Sci China Ser F-Inf Sci | Jan. 2009 | vol. 52 | no. 1 | 100-114 103
Page 5
tion of the targets. Then multiply the signal
SSRC(Kω,Ky, z′) (azimuth FT of SSRC(Kω, y, z
′))
with H3(Kω,Ky, z′) and take the FT in the eleva-
tion direction for the every given Ky considering
z′ ≪√
X2c + Z2
c and zn ≪√
X2c + Z2
c , we can get
SSAFRC(Kω,Ky,Kz′)
∼∑
n
exp(−jRre
√
4K2ω −K2
y)
· sin c
LH
π
[
Kz′ − ε√
4K2ω −K2
y
]
, (15)
where Rre = (√
[Xc + x′n]2 + [Zc − zn]2 − Rc) and
ε = zn/√
[Xc + x′n]2 + [Zc − zn]2 − Zc/Rc.
Thus the whole elevation bandwidth BZC is re-
duced by multiplying with H3(Kω,Ky, z′), that is
BZC
≈ 2Z0Kω max/√
[Xc + x′n]2 + Z2
c + 2π/LH
≈ 2Z0Kω max/√
[Xc + x′n]2 + Z2
c , (16)
which is much smaller than the support BZ of
SSAFRC(Kω,Ky, z′). So the elevation sampling
spacing can be reduced to ∆z.
∆z 62π
BZC
=π√
(Xc −X0/2)2 + Z2c
[Kω maxZ0]. (17)
The elevation oversampling ratio is set between
1.1 and 1.2 to provide a guard band. Sampling with
(17), the elevation sampling number is reduced, but
some additional steps are needed to recovery the
spectrum of the targets accurately. The problem
is to be discussed in section 4. If the elevation
sampling distance is greater than 2π/BZC , it will
cause severe elevation ambiguity. It is possible to
gain the information of the strong backscattering
targets with the inversion methods, spectral esti-
mation technique and so on[8−10].
3.3 TomoSAR point spread function
The TomoSAR point spread function (PSF) and
the 3-D resolution are analyzed in this subsection.
Assume that the elevation sampling distance sat-
isfies the restriction in (13). Take elevation and
azimuth FT, in accordance with the method of sta-
tionary phase (MSP)[16], the 3-D wavenumber sig-
nal is
S(Kω,Ky,Kz′)
= δn(xn, yn, zn)A2 expj2KωRc
× exp−j[Kz′(−Zc + zn) +Kyyn]
× exp−jxn
√
4K2ω −K2
y −K2z′, (18)
where A2 ≈ −j4πKωxn(4K2ω − K2
y − K2z′)−1 is a
slowly fluctuating amplitude function which does
not play any important role in the following anal-
ysis. Hence, the influence of this term is not con-
sidered.
To analyze the point spread function (PSF) of
target Pn, a Cartesian coordinate system PUVW
of the target Pn has been defined in Figure 1.
The elevation Doppler centroid of target Pn is ap-
proximately −2Kω cos(ξnc), and it changes with
the position of the targets. Assuming that the
azimuth Doppler centroid is equivalent to zero,
the translation relation in wavenumber domain be-
tween OXY Z coordinates and PUVW coordinates
is denoted by
KX
KY
KZ
=
sin ξnc 0 cos ξnc
0 1 0
− cos ξnc 0 sin ξnc
KU
KV
KW
, (19)
where ξnc = arctanxn/(Zc − zn) if (Zc − zn) 6= 0,
else ξnc = π/2.
Eqs. (18) and (19) imply that the 3-D spatial
wavenumber support of target Pn is a part of a
“hollow sphere” with radii 2Kω min and 2Kω max,
and the received signal of target Pn is the 3-D
spherical band-pass signal, as shown in Figure 2(a).
The size of spherical surface is proportional to the
size of azimuth and elevation synthetic aperture.
When the “hollow sphere” is projected onto the
KU -axis and KV -axis plane, the support band BUn
in the KU -axis direction is an annulus spanning a
wavenumber of BUn = 2[Kω max −Kω min].
When the 3-D “hollow sphere” is projected onto
the KU -axis and KW -axis plane in Figure 2(b), the
support band BV n in the KV -axis direction is an-
other annulus with BV n = BY ≈ 4Kω sin(ΨA/2).
When the 3-D “hollow sphere” is projected
onto the V -axis and W -axis plane, it is an-
other annulus which spans a spatial wavenum-
ber of BWn in the W -axis direction with BWn ≈
2KωncLH sin(ξnc)r−1hn , where rhn = [(Xc + x′
n)2 +
(Zc − zn)2]0.5,Kωnc is the wavenumber center with
respect to every azimuth wavenumber Ky, given by
Kωnc =1
2
√
4K2ω −K2
y |max +√
4K2ω −K2
y |min
.
(20)
104 TAN WeiXian et al. Sci China Ser F-Inf Sci | Jan. 2009 | vol. 52 | no. 1 | 100-114
Page 6
Figure 2 SAR tomography imaging geometry. (a) 3-D wavenumber support; (b) support projected onto OXZ plane.
As for the narrow band signal, the spherical re-
gion of the support is not very apparent. The sup-
port of the target in the 3-D wavenumber domain
could be viewed as an approximately regular cube.
Combining (18) with (19), the 3-D PSF can be ex-
pressed as
PSF (un, vn, wn;xn, yn, zn)
∼ sinc
BUnun
2π
sinc
BV nvn
2π
sinc
BWnwn
2π
, (21)
where the sinc function is defined as sinc(x) =
sin(πx)/(πx). The translation relation between
OXY Z and PUVW is stated as
un
vn
wn
=
sin ξnc 0 − cos ξnc
0 1 0
cos ξnc 0 sin ξnc
x− xn
y − yn
z − zn
.
(22)
As shown in Figure 3(a), the PSF of target Pn is
a 3-D sinc-like function. Figures 3(b)–3(d) are the
slices of the PSF on the different slices in 3-D space
with dynamic range 30 dB. The impulse response
width (IRW) is defined as −3 dB width (resolu-
tion) of the main lobe of the impulse response, and
the 3-D resolution is approximated by
ρUn ≈ [0.886C]/[2B],
ρV n ≈ [0.886π]/[2Kω min sin(ΨA/2)],
ρWn ≈ [0.886πrhn]/[Kω minLH sin(ξnc)],
(23)
where ρUn, ρV n and ρWn represent the resolution of
target Pn along the U -axis, V -axis and W -axis in
the 3-D spatial domain, respectively.
The transform (22) shows that the orientation
of the 3-D PSF relates to the target coordinates
and the PSF at elevation and azimuth plane is
generated by rotating the 3-D sinc-like function
by ξnc. Therefore, the PSF of target Pn in To-
moSAR system is shift-varying. From (23), we
can see that the resolutions are dependent on the
position of the target and affected by the imag-
ing geometry. Therefore, the resolutions of target
Pn in TomoSAR system are spatial-varying. The
detailed description of shift-varying and spatial-
varying property will be demonstrated with the
simulation in section 5. The shift-varying and
spatial-varying property of the 3-D PSF of To-
moSAR is the second point in this paper which is
distinguished from the work of Reigber et al.[6,11,13].
4 3-D range migration algorithm with ele-vation digital spotlighting
In this section, 3-D range migration algorithm with
elevation digital spotlighting (RMA-EDS) which is
distinct from refs. [6,11,13] is presented, as shown
in Figure 4. The 3-D RMA-EDS can operate for
any case of Zc and the elevation sampling with
(17). In addition, the algorithm is suitable for
processing the chirped and stepped-frequency SAR
data.
TAN WeiXian et al. Sci China Ser F-Inf Sci | Jan. 2009 | vol. 52 | no. 1 | 100-114 105
Page 7
Figure 3 Pn’s PSF processed by the method[11] under the condition (13). (a) PSF in the 3-D space; (b) PSF at OXY slice; (c) PSF
at OY Z slice; (d) PSF at OXZ slice.
Figure 4 Block diagram representation of RMA-EDS for the
general TomoSAR geometry with elevation sampling restriction
(17).
As for image formation, the imaging geometry is
extended to a more general case in which the cen-
ter of elevation aperture Zc is arbitrary, and the 3-
D image reconstructed accurately with RMA-EDS
needs no geometric correction. As for SAR sys-
tem, the raw data volume becomes smaller with
the sampling spacing (17) and the data acquisi-
tion time is shorter, the elements of antenna ar-
ray becomes much less and the configuration of the
system gets more compact. For accurately recon-
structing the 3-D image without geometric correc-
tion, the following problems should be considered:
• Usually, the center of the elevation synthetic
aperture is not symmetrical about the illuminated
scenario in the elevation direction.
• The elevation Doppler centroid updates with
the position of the target and spreads the elevation
spectrum in the elevation wavenumber domain.
• The elevation sampling with (17) will cause
the spectral aliasing if the elevation FT is used di-
rectly.
• The ground-range swath of the illuminated
area is usually much wider than the height of the
illuminated area.
In the following subsections, the main operations
will be formulated or described according to the
signal flow in the processing chain in Figure 4. The
RMA-EDS is the third point in this paper which is
distinct from that of the existing algorithms[6,11,13].
4.1 Slant range and azimuth processing
In this step, slant range matched filtering and az-
imuth FT is performed.
As for the linear FM pulse SAR system, the
matched filter H1(ft) in the slant range frequency
domain is performed first. As for the stepped fre-
quency SAR system, only a phase compensation
function H2(Kω) multiplication with the reference
range Rc is carried out,
106 TAN WeiXian et al. Sci China Ser F-Inf Sci | Jan. 2009 | vol. 52 | no. 1 | 100-114
Page 8
H2(Kω) = exp[j2KωRc], (24)
where Kω ∈ [Kω min,Kω max]. Then the processed
signal in the slant range wavenumber domain could
be written as
S1(Kω, y, z′) = δn exp−j2[Rn −Rc]Kω, (25)
instead of considering the form of the transmitted
signal.
The azimuth and elevation 2-D FT[6,11,13] could
not be performed directly because the elevation
support is much larger than the elevation sampling
rate. Therefore, only the azimuth FT could be
performed before the elevation spectral alias-free
recovery; that is,
S2(Kω,Ky, z′) = δnA3 expj2RcKω
× exp−jKyyn
− jRs
√
4K2ω −K2
y, (26)
where Rs =√
[Xc + x′n]2 + [Zc + z′ − zn]2 and
A3 = exp(−jπ/4)(4K2ω −K2
y)−1.
4.2 Elevation digital spotlighting
processing
This step is to perform the elevation wavenumber
alias-free recovery under the elevation sampling re-
stricted condition (17). As shown in Figure 5, the
ground-range swath is much larger than the height
of the target, and the equivalent height RH is usu-
ally higher than the real height Z0. The eleva-
tion Doppler centroid −2Kω cos(ξnc) of target Pn
spreads the whole elevation Doppler support.
Step 1. The elevation spectrum extension is
conducted by zero-interleaving with the factor γ.
γ =
⌈
RH
Z0
⌉
=
⌈
(cos θn min − cos θn max)
Z0/√
(Xc − 0.5X0)2 + Z2c
⌉
, (27)
where ⌈∗⌉ represents the smallest integer that is
larger than or equal to ∗.
Note that zero-interleaving is different from that
used in ref. [18], where it is utilized to remove the
whole azimuth aliasing. Here it is used to extend
the spectrum due to the elevation Doppler centroid
variation but not the sampling spacing with (17).
Step 2. Zero-padding is performed consid-
ering the relation between the length of elevation
synthetic aperture and the height of the area of in-
terest to avoid the circular convolution aliasing[13].
Step 3. After the multiplying the elevation
signal with H3(Kω,Ky, z′), the elevation spectrum
is reduced to ∆z−1, where z′ is the elevation sam-
pling after the above operations.
Figure 5 The ground-range and elevation plane in TomoSAR
geometry.
Step 4. The elevation and slant range signal
are transformed to the polar domain through the
slant range inverse FT and the elevation FT. The
spotlight filter which is similar to that of ref. [13]
is introduced.
H4(Kω,Ky,Kz′)
=
1 if |Rn sin(θc + θn) −Xc| 6 (X0/2),
&|Rn cos(θc + θn) − Zc| 6 (Z0/2),
0 else,
(28)
where θn represents the instantaneous incidence,
and Rn is the slant range, θc = arccos(Zc/Rc).
With this filter, the area of interest can be recon-
structed without any ambiguities through limiting
t and θn.
Step 5. Perform the slant range FT and
zero-padding the elevation. Then the alias-free
S3(Kω,Ky, z′′) is recovered via the multiplication
of the elevation signal with H5(Kω,Ky, z′′) and el-
evation inverse FT,
H5(Kw,Ky, z′′) = exp(j2RcKω)
× exp(−j√
4K2ω −K2
y
√
X2c + (Zc + z′′)2), (29)
where z′′n = (n − NH/2 − 1)∆z0, n = 1, 2, · · · , NH
and NH = 2⌈max(γZ0, LH)/∆z0⌉.
TAN WeiXian et al. Sci China Ser F-Inf Sci | Jan. 2009 | vol. 52 | no. 1 | 100-114 107
Page 9
Step 6. Finally, a phase function should be
multiplied to transform the elevation band-pass
signal to the low-pass signal S4(Kω,Ky, z′′) before
the elevation Fourier transforms, and the filter is
H6(Kw,Ky, ξnc) = exp[jKωncz′′(Zc/Rc)]. (30)
Step 7. After performing the elevation
Fourier transform of S4(Kω,Ky, z′′), we have
S5 (Kω,Ky,Kz′′) ∼ exp[−j(Kz′′(zn − Zc))]
× exp[−j(Xc + x′n)
√
4K2ω −K2
y −K2z′′ ]
⊗ σ(KZ) × exp[−jKyyn + j2KωRc], (31)
where KZ = Kz′′ + KωncZc/Rc,Kz′′ = 2π(n −
NH/2 − 1)/(∆z0NH), n = 1, 2, · · · , NH . ⊗ and σ
indicate the convolution operation and the delta
function, respectively.
4.3 3-D wavenumber filtering
Combining the above operations, the echoed sig-
nal is transformed to the 3-D wavenumber do-
main. Then the 3-D matched filtering in the 3-D
wavenumber domain is performed, that is,
H7(Kω,Ky,Kz′′)
= expjXc
√
4K2ω −K2
y −K2Z
× exp−j2RcKω−jZc(Kz′′ +KωncZc/Rc).(32)
Subsequently, with (32) the filtered signal of tar-
get Pn in the 3-D wavenumber is obtained.
S6 (Kω,Ky,Kz′′)
= S5(Kω,Ky,Kz′′)H7(Kω,Ky,Kz′′)
∼ δn(xn, yn, zn) exp−j[KZzn +Kyyn]
× exp−jx′n
√
4K2ω −K2
y −K2Z. (33)
4.4 3-D STOLT mapping
3-D STOLT mapping is performed in 3-D
wavenumber domain to transform the signal in the
sphere coordinate into the uniformly sampling sig-
nal in the 3-D Cartesian coordinate (KX ,KY ,KZ)
for inverse FT. Eq. (33) is redefined as
S6(Kω, Ky,Kz′′) ∼ δn(xn, yn, zn)
× exp−jznKZ(Kω,Ky,Kz′′)
× exp−jynKY (Kω,Ky,Kz′′)
× exp−jx′nKX(Kω,Ky,Kz′′), (34)
where
KX =√
4K2ω −K2
y − [Kz′′ +KnωcZc/Rc]2,
KY =Ky,
KZ =[Kz′′ +KnωcZc/Rc].
(35)
Eq.(35) indicates the 3-D STOLT mapping[6],
which is the extension of 2-D STOLT
mapping[17,19]. With the 3-D STOLT mapping
in the 3-D wavenumber domain, the nonorthog-
onality of the received signal axes is rectified by
mapping the signal in (Kω,Ky,K′′z ) domain into
the signal S7(KX ,KY ,KZ) in (KX ,KY ,KZ) do-
main; in other words, the 3-D STOLT mapping
essentially replaces wavenumber in the direction of
wave propagation by wavenumber observed in the
ground-range (X), azimuth (Y ) and elevation (Z)
direction.
S7(KX ,KY ,KZ)
≈∑
|KX−l∆KX |6NIX∆KX
|KY −m∆KY |6NIY ∆KY
|KZ−n∆KZ |6NIZ∆KZ
S′6(Kω,Ky,Kz′′), (36)
where NIX , NIY and NIZ denote the half-length of
the sinc interpolation kernel used for STOLT map-
ping. ∆Kω,∆Ky and ∆Kz′′ are the wavenumber
sampling spacings of S6(KX ,KY ,KZ).
S′6(Kω, Ky,Kz′′)
= S6(Kωl,Kym,Kz′′n)
× sinc3KX ,KY ,KZ, (37)
where Kωl,Kym and Kz′′n are the discretized
wavenumber of S6(Kωl,Kym,Kz′′n)
sinc3 KX ,KY ,KZ
= sinc[KX − l∆KX ]/∆KX
× sinc[KY −m∆KY ]/∆KY
× sinc[KZ − n∆KZ ]/∆KZ. (38)
During the course of STOLT interpolation, the
number of the samples in 3-D wavenumber domain
KXKYKZ grid is
NX = 2⌈(KX max −KX min)/(2∆KX)⌉,
NY = 2⌈(KY max −KY min)/(2∆KY )⌉,
NZ = 2⌈(KZ max −KZ min)/(2∆KZ)⌉,
(39)
where the spatial wavenumber spacing ∆KX 6
2π/X0,∆KY 6 2π/Y0,∆KZ 6 2π/(γZ0),KX min,
108 TAN WeiXian et al. Sci China Ser F-Inf Sci | Jan. 2009 | vol. 52 | no. 1 | 100-114
Page 10
KX max,KY min,KX max,KZ min and KZ max are the
minimum and maximum value of KX ,KY and KZ ,
respectively.
As in (38), the interpolation kernel consist of
three sinc functions. For 3-D mapping, one method
is to utilize two times 1-D sinc interpolation
(The first interpolation is performed in azimuth
and slant range wavenumber domain. Another
one is conducted in ground-range and elevation
wavenumber domain). The 3-D mapping in (36)
is based the evenly spaced data, the interpolation
along KY and KZ direction could be skipped when
∆KY = 2π/(NY ∆y) and ∆KZ = 2π/(NZγ∆z).
Therefore, the most direct 3-D STOLT mapping
could be obtained by 1-D sinc interpolation[15,17,19]
along KX direction through considering the az-
imuth wavenumber Ky and elevation wavenumber
Kz′′ at the same time.
4.5 3-D inverse FT and image space
determination
After the 3-D STOLT interpolation, the 3-D SAR
image can be obtained with 3-D inverse Fourier
transform. The reconstructed target area is
f(xn, yn, zn)
∼∑
n
δnPSF (un, vn, wn;xn, yn, zn)
∼∑
n
δn(xn, yn, zn)sinc
(
BUnun
2π
)
× sinc
(
BV nvn
2π
)
sinc
(
BWnwn
2π
)
. (40)
The ground-range, azimuth and elevation sam-
ples of the reconstructed 3-D image correspond to
the above uniformly spaced samples in the 3-D spa-
tial domain OXY Z
xn = Xc + (l −NX/2 − 1)∆X ,
yn = (m−NY /2 − 1)∆Y ,
zn = −Z0/2 + (n−NZ − 1)∆Z ,
(41)
where l = 1, 2, . . . , NX ,m = 1, 2, . . . , NY and
n = 1, 2, . . . , NZ .∆X = 2π/(NX∆KX), ∆Y =
2π/(NY ∆KY ), and ∆Z = 2π/(NZ∆KZ) are the
evenly spaced samples.
As the step of zero-interleaving has been per-
formed with the factor γ to recover the elevation
spectrum, the actual samples of the reconstructed
image in the 3-D spatial domain should be reduced
to NZ × NY × [NZγ−1], where [NZγ
−1] represents
the integer part of NZγ−1. The focused 3-D image
is generated in ground-range (X-axis), azimuth (Y -
axis) and elevation (Z-axis) direction, so the geo-
metric correction is not needed in the reconstruc-
tion.
4.6 Discussion of the approximation
The main operations used in the 3-D RMA-EDS al-
gorithm are the Fourier transform, elevation spec-
trum alias-free recovery and STOLT mapping. As
for the Fourier transform, the only approxima-
tions in the algorithm lie in the slant range Fourier
transform, azimuth Fourier transform and eleva-
tion Fourier transform. The transforms are based
on the assumption of a high time-bandwidth prod-
uct in slant range direction, azimuth and elevation
direction. Usually in the TomoSAR system, these
approximations could be ignored. As for eleva-
tion spectrum alias-free recovery, the data colleted
with sampling spacing (17) has included redun-
dant information. With the aforementioned steps,
the elevation spectrum could be recovered without
any approximation. As for the STOLT mapping,
2-D or 1-D wavenumber interpolation is used in
wavenumber domain, and the signal in the Sphere
coordinates wavenumber domain could be mapped
to Cartesian coordinates accurately.
5 TomoSAR simulation experiments
In the above sections, the sampling criteria, PSF
of TomoSAR and the 3-D RMA-EDS are analyzed.
In this section several simulation experiments are
conducted to demonstrate the 3-D resolution ca-
pabilities, the shift-varying properties of the point
spread function of TomoSAR, the spatial-varying
property of the 3-D resolution and so on. For sim-
plicity, the echoed raw data is generated by (4).
In order to test the validity of RMA-EDS, the sys-
tem parameters are chosen as listed in Table 1. In
addition, the TomoSAR experiment has been per-
formed in microwave anechoic chamber and pro-
cessed with this algorithm[20].
For a 200 m×100 m×40 m (ground-range, az-
imuth, elevation) target zone, the elevation sam-
TAN WeiXian et al. Sci China Ser F-Inf Sci | Jan. 2009 | vol. 52 | no. 1 | 100-114 109
Page 11
pling spacing is about 0.62 m according to (17) if
the 3-D RMA is used directly. The antenna aper-
ture angle is 0.575, so the azimuth antenna size is
about 2.99 m. With the 3-D RMA-EDS, the eleva-
tion sampling spacing can be set at 2.03 m. Hence,
the sampling number in the elevation direction is
51 when the volumetric height is 40 m.
Table 1 Parameters used in the simulation experiment
Parameter name Value
Center frequency (GHz) 10.00
Signal bandwidth (MHz) 100.00
Elevation aperture length (m) 101.55
Center of elevation aperture (m) 5000.41
Azimuth resolution (m) 1.32
⊥Line-of-sightresolution (m) 1.30
Line-of-sight resolution (m) 1.33
Sampling number of azimuth 226
Sampling number of elevation 51
Azimuth sampling spacing (m) 1.2030
Elevation sampling spacing (m) 2.03
Range step frequency size (MHz) 0.40
Center of elevation target zone (m) 0
Azimuth length of the targets area (m) [−100 100]
Elevation height of the targets area (m) [−20 20]
Center of ground-range target zone (m) 5000.00
Ground-range width of the targets area (m) [−100 100]+ 5000
As shown in Figure 6, in order to exhibit 3-D
resolution capability and overcome the foreshort-
ening and layover effects with TomoSAR imaging
technique, the relative positions of 168 point tar-
gets are given. 3 point targets are evenly located
in the line x = 4910, z = −18 with azimuth inter-
val 24.00 m in ‘A’ area in Figure 6(c) and (d). 18
point targets are evenly located at the plane z − 6
(x − 4913) = 0 with azimuth interval 6.00 m and
ground-range interval 3 m in ‘B’ area in Figure 6(c)
and (d). 8 point targets are symmetrically located
in the circle z = −18, (x − 4952)2 + y2 = 242 with
angle interval 45, 7 point targets are evenly lo-
cated in the line x = 4952, y = 0 with elevation
interval 6 m in ‘C’ area in Figure 6(c) and (d). 28
point targets are evenly distributed at the plane
z − (x − 4982) = 0 with azimuth interval 6.00 m
and ground-range interval 3 m in ‘D’ area in Figure
6(c) and (d). 28 point targets are distributed at the
plane z− 0.25(x− 5024) = 0 with azimuth interval
6 m and ground-range interval 4.00 m, correspond-
ing to the elevation interval 1.00 m in ‘E’ area in
Figure 6(c) and (d). 76 point targets are evenly lo-
cated in the edges of the rectangular parallelepiped
with azimuth length 48.00 m, ground-range width
36.00 m and height 36.00 m shown in ‘F’ area in
Figure 6(d).
The conventional SAR data collecting geometry
has a property of cylindrical symmetry. Therefore,
the targets distributed in the elevation direction
could not be resolved in the conventional 2-D imag-
ing geometry with the method[13] of 2-D SAR fo-
cusing. Meanwhile, the serious foreshortening and
layover effects occur in the conventional 2-D imag-
ing geometry, as shown in Figure 7.
(1) Layover. As shown in Figure 7, the posi-
tion of target ‘B’ area on the 2-D SAR image is
reversed with that of target ‘A’ area for the in-
fluence of layover. In addition, the targets are lo-
cated at the center line of ‘C’ area perpendicular to
the horizontal plane, and the phenomenon of lay-
over becomes the most serious. The target plane
z − (x − 4982) = 0 is approximately perpendicu-
lar to the line-of-sight of the SAR sensor, and the
distances between the targets and the sensor are
approximately equal, so 28 point targets are super-
posed into 9 points. The amplitude of the super-
posed target becomes stronger with the increasing
number of the targets, e.g. ‘G’ and ‘H’ are stronger
than that of other 7 targets, as shown in Figure 7.
(2) Foreshortening. One example is the target
‘C’ area as shown in Figure 6(a), two targets are
located in the circle in the ground-range direction
with distance 34.94m in between, while in the cor-
responding 2-D SAR image as shown in Figure
7(a), the distance is only about 23.81 m. An-
other example is the target ‘E’ area, whose slope
is about 14 and is much less than the incidence
angle. Therefore, the foreshortening occurs, and
the large area consisting of large number of tar-
gets is reduced to a small area, and the imaging
area is brighter than the area nearby, as shown in
Figure 7. Also the size of the area 12.70 m (slant
range)×48.00 m (azimuth) in 2-D SAR image is
much smaller than the real area 24.74 m (slant
range)×48.00 m (azimuth). Note that both lay-
over and foreshortening occur in target ‘F’. They
are more difficult to be resolved. Therefore, the
110 TAN WeiXian et al. Sci China Ser F-Inf Sci | Jan. 2009 | vol. 52 | no. 1 | 100-114
Page 12
Figure 6 The targets and the projections onto OXY , OY Z and OXZ. (a) The targets in 3-D space; (b) projections onto OY Z plan;
(c) projections onto OXY plane; (d) projections onto OXZ plane.
Figure 7 The 2-D SAR images obtained at different height. (a) 2-D SAR image at z=4948.82 m; (b) 2-D SAR image at z=1969.13
m.
conventional SAR system could not avoid the
above problems due to the observation geometry,
and cannot resolve the targets in the 3-D spatial
domain.
With the 3-D RMA-EDS, the 3-D structure of
the targets can be reconstructed and resolved. As
shown in Figure 8(a), the 3-D image is displayed
at 3.01dB contour. As expected, the whole space
structure is very consistent with the real situa-
tion in Figure 6(a). The image is reconstructed
in OXY Z space, and thereby, it needs no geomet-
ric correction after focusing. In Figure 6(b), the
projected targets in the solid black boxes are very
clear, while in Figure 8(b) the projected targets in
the solid black boxes are blurred since the line-of-
sight resolution and the ⊥Line−of−sight resolution are
about 1.30 m and 1.48 m, respectively. Because the
ground-range position of these targets is different,
they can still be resolved in the 3-D SAR space, as
shown in Figure 8(a), (c) and (d).
The expanded 3-D SAR image of target ‘C’ area
is shown in Figure 9(a), and Figure 9(b) and (c)
displays the details of target ‘C’ area at different
layers with a dynamic range of 20 dB.
As shown in Figure 3, the PSF of the tar-
get is a 3-D sinc-like function, and the recon-
struction of the PSF is performed without any
weighted window. In this paper, the IRW, the
peak sidelobe ratio (PSLR) and the integrated side-
lobe ratio (ISLR) are measured to verify the per-
formance and focusing accuracy of the 3-D RMA-
EDS. Three point targets are set in another simu-
lation to demonstrate the property of the 3-D PSF,
e.g. P1(3080.00 m, 24.00 m, 12.50 m), P2(5000.00
m, 0 m, 0 m) and P3(7920.00 m, −24.00 m, −12.50
m). The theoretical 3-D IRW is calculated with
(24) and shown in Table 2.
On the one hand, the IRWs of the target varies
with the position of target and the measured re-
sults shown in Table 2 are consistent with the the
oretical value listed in Table 3, that is to say, the
resolution of the TomoSAR is spatial-varying. On
TAN WeiXian et al. Sci China Ser F-Inf Sci | Jan. 2009 | vol. 52 | no. 1 | 100-114 111
Page 13
Figure 8 The 3-D SAR image and the projections on the different planes. (a) The 3-D TomoSAR image; (b) projection onto the
OY Z plane; (c) projection onto the OXY plane; (d) Projection onto the OXZ plane.
Figure 9 The 3-D SAR image of ‘C’ and the layers at different planes. (a) Target ‘C’ area; (b) layer at x =4951.8 m; (c) layer at
z = −17.56 m.
Table 2 Theoretical IRWs of different targets in the 3-D spatial domain
Target position (xn, yn, zn)Ideal IRW (m)
U V W
P1(3080.00 m, 24.00 m, 12.50 m) 1.33 1.33 1.45
P2(5000.0 m, 0 m, 0 m) 1.33 1.33 1.30
P3(7920.0 m, −24.0 m, −12.5 m) 1.33 1.33 1.45
Table 3 Image quality parameters measured in PUV W coordinates
Target position (xn, yn, zn)IRW (m) PSLR (dB) ISLR (dB)
U V W U V W U V W
P1(3080.00 m, 24.00 m, 12.50 m) 1.35 1.34 1.42 −13.13 −13.54 −13.21 −10.16 −10.62 −10.16
P2(5000.00 m, 0 m, 0 m) 1.37 1.33 1.30 −13.22 −13.42 −14.23 −10.10 −10.51 −10.18
P3(7920.00 m, −24.00 m, −12.50 m) 1.33 1.34 1.43 −13.81 −13.27 −15.32 −10.55 −10.35 −10.61
112 TAN WeiXian et al. Sci China Ser F-Inf Sci | Jan. 2009 | vol. 52 | no. 1 | 100-114
Page 14
Figure 10 The SPFs (top) of target ‘P1’, ‘P2’ and ‘P3’, the slices (bottom) of the corresponding PSFs of the targets at OXZ plane.
(a) Target ‘C’ area; (b) layer at x =4951.85 m; (c) layer at z = −17.56 m.
the other hand, the orientation of the 3-D SPF of
different targets relates to nc, as shown in Figure
10. The rotation angles of SPF of targets ‘P1’, ‘P2’
and ‘P3’ at ground-range and elevation plane are
31.63, 45.00 and 57.74, respectively. Thus the
PSF of the target is shift-varying.
6 Conclusion and future work
In this paper, we have provided a treatment of sam-
pling criteria, point spread function (PSF) and 3-
D reconstruction algorithm issues about the To-
moSAR imaging technique theoretically.
(1) The 3-D sampling criteria of TomoSAR is
derived in the 3-D wavenumber domain and then
verified both numerically and experimentally.
(2) The shift-varying property of the PSF of To-
moSAR and the spatial-varying property of the 3-
D resolution for the case of narrow band system
are demonstrated with numerous simulation exper-
iments and analyses in 3-D wavenumber domain.
(3) The 3-D RMA-EDS, which is suitable for
any case of Zc, the elevation sampling spacing with
(17), the case in which ground-range swath of the
illuminated area is wider than the height of the
area, the chirped and stepped-frequency SAR sys-
tem, has been proposed and verified with the sim-
ulation and the measured image quality parame-
ters. In addition, with 3-D RMA-EDS, the 3-D
target area can be reconstructed without any am-
biguities. The focused 3-D SAR image needs no
geometric correction and can be displayed directly
in the 3-D spatial domain.
(4) Compared to the conventional SAR tech-
nique, TomoSAR imaging has the capabilities of
providing the 3-D image of the area of interest.
The simulation experiment of multi-baseline To-
moSAR is performed, and the results processed by
RMA-EDS are presented to demonstrate that the
foreshortening and layover that seriously affect the
interpretation of the 2-D SAR image could be re-
moved very well.
The main limitation of the proposed algorithm
is the need for assumption of the azimuth constant
effective velocity and elevation uniformly sampling.
Therefore, the problems of sampling nonunifor-
mity, irregularity, sparsity and motion compensa-
tion technique in TomoSAR imaging that occur
frequently in practice are not addressed in this pa-
per. In addition, as the imaging scenario is illu-
minated by multi-baseline or array antenna, the
coherence of the backscattering information of the
targets at different heights or different incidences
is worth studying.
The authors would like to thank the anonymous reviewers and
the readers for their valuable comments and suggestions.
TAN WeiXian et al. Sci China Ser F-Inf Sci | Jan. 2009 | vol. 52 | no. 1 | 100-114 113
Page 15
1 Goldstein R M, Zebker H A. Interferometric radar measure-
ments of ocean surface currents. Nature, 1987, 328(20): 707–
709
2 Chan C K, Farhat N H. Frequency swept tomographic imag-
ing of three-dimensional perfectly conducting objects. IEEE
Trans Antennas Propag, 1981, AP-29(2): 312–319
3 Farhat N H, Werner C L, Chu T H. Prospects for three-
dimensional projective and tomographic imaging radar net-
works. Radio Sci, 1984, 19(5): 1347–1355
4 Jakowatz C V Jr, Wahl D E, Eichel P H, et al. Spotlight-
mode synthetic aperture radar: a signal processing approach.
Boston: Kluwer Academic Publishers, 1996
5 Knaell K K, Cardillo G P. Radar tomography for the gen-
eration of three-dimensional images. IEE Proc-Radar Sonar
Navig, 1995, 142(2): 54–60
6 Reigber A, Moreira A. First demonstration of airborne SAR
tomography using multibaseline L-band data. IEEE Trans
Geosci Remote Sens, 2000, 38(5): 2142–2152
7 She Z S, Gray D A, Bogner R E, et al. Three-dimensional
space-borne synthetic aperture radar (SAR) imaging with mul-
tiple pass processing. Int J Remote Sens, 2002, 23(20): 4357–
4382
8 Fornaro G, Lombardini F. Three-dimensional multipass SAR
focusing: experiments with long-term spaceborne data. IEEE
Trans Geosci Remote Sens, 2005, 43(4): 702–714
9 Fornaro G, Serafino F. Imaging of single and double scatter-
ers in urban areas via SAR tomography. IEEE Trans Geosci
Remote Sens, 2006, 44(12): 3497–3505
10 Li J, Min Rui, Pi Y M. SAR stereo imaging algorithm based
on amplitude and phase estimation. In: 1st Asian and Pacific
Conference on Synthetic Aperture Radar Proceedings. Huang-
shan: IEEE Press, 2007. 385–388
11 Lopez-Sanchez J M, Fortuny-Guasch J. 3-D Radar imaging us-
ing range migration techniques. IEEE Trans Antennas Propag,
2000, 48(5): 728–737
12 Frey O, Morsdorf F, Meier E. Tomographic processing of multi-
baseline P-band sar data for imaging of a forested area. In:
Proceedings of IGARSS. Barcelona: CIMNE, 2007. 156–159
13 Soumekh M. Synthetic aperture radar signal processing with
matlab algorithms. New York: John Wiley & Sons, 1999
14 Weiß M, Ender J, Peters O, et al. An airborne radar for
three dimensional imaging and observation-technical realisa-
tion and status of ARTINO. In: 6th European Conference on
Synthetic Aperture Radar. Dresden: Deutschen Nationalbib-
liografie, 2006
15 Cumming I G, Wong F H. Digital processing of synthetic aper-
ture radar data: algorithms and implementation. Boston:
Artech House, 2005
16 Stamnes J J. Waves, rays, and the method of stationary phase.
Optics Express, 2002, 10(16): 740–751
17 Cafforio C, Prati C, Rocca F. SAR data focussing using seis-
mic migration techniques. IEEE Trans Aerosp Electron Syst,
1991, 27(2): 195–207
18 Prati C, Rocca F. Focusing SAR data with time-varying
doppler centroid. IEEE Trans Geosci Remote Sens, 1992,
30(3): 550–559
19 Stolt R H. Migration by Fourier transform techniques. Geo-
phys, 1978, 43(1): 23–48
20 Tan W X, Wang Y P, Hong W, et al. SAR Three-dimensional
imaging experiments with microwave anechoic chamber SAR
data. In: Proceedings of Asia Pacific Microwave Conference.
Bangkok: Bangkok University Press, 2007. 1423–1426
114 TAN WeiXian et al. Sci China Ser F-Inf Sci | Jan. 2009 | vol. 52 | no. 1 | 100-114