Progress In Electromagnetics Research, Vol.

140, 6389, 2013

IMAGING ENHANCEMENT OF STEPPED FREQUENCY

RADAR USING THE SPARSE RECONSTRUCTION
TECHNIQUE
Bo Pang1, 2, * , Da-Hai Dai1, 2 , Shi-Qi Xing1, 2 ,
Yong-Zhen Li1, 2 , and Xue-Song Wang1, 2
1 School

of Electronic Science and Engineering, National University of

Defense Technology, Changsha 410073, China
2 State

Key Laboratory of Complex Electromagnetic Environment

Effects on Electronics and Information System, Changsha 410073,
China

AbstractBased on the observation that sparsity assumption is well

satisfied in the synthetic aperture radar (SAR) imaging applications,
there is increasing interest in utilizing compressive sensing (CS) in
SAR imaging. However, there are still several problems which should
be concerned in CS-based imaging approaches. Firstly, inevitable noise
and clutter challenge the performance of CS algorithms. Secondly, the
super-resolving ability of CS algorithms is not sufficiently exploited
in most cases. Thirdly, nonideal characteristics of mutual coherence
affect the performance of CS algorithms in complex scenes. In this
paper, a novel CS imaging framework is proposed for the purpose
of improving the imaging performance of stepped frequency SAR.
Meanwhile, a super-resolving imaging algorithm is proposed based on
the nonquadratic optimization technique. Simulated and rail SAR
measured data are applied to demonstrate the effectiveness of the novel
framework with the proposed super-resolving algorithm. Experimental
results validate the superiority of this method over previous approaches
in terms of robustness in low SNR, better super-resolving ability and
improved imaging performance in complex scenes.

Received 4 March 2013, Accepted 29 April 2013, Scheduled 22 May 2013

* Corresponding author: Bo Pang ([email protected]).

64

Pang et al.

1. INTRODUCTION
Synthetic aperture radar (SAR) is a microwave sensor which is
capable of producing high-resolution images of the earths surface [1
5]. Having the advantage of weather independence and all-day
operation capability, it is widely used in many military and civilian
applications [68].
Conventional SAR imaging schemes usually process the received
signal using matched filter (MF) [911]. Although they are easy
to implement, one common shortcoming of these MF based schemes
is the resolution limitation to system bandwidth, which complicates
localization of point scatterers for automated recognition tasks. In
order to overcome this disadvantage, linear frequency modulation
(LFM) and stepped frequency (SF) waveforms are often utilized for
generating wide bandwidth. Compared to LFM waveform, the SF
waveform reduces the requirement on hardware and thus has been
widely employed to increase system bandwidth. Unfortunately, as
there exists some tradeoff between resolution and imaging range
width [12], the bandwidth of SF system can not be increased infinitely.
Therefore, how to acquire super-resolution imaging ability under
limited bandwidth has attracted widespread concerns in recent years.
One idea to overcome this limit is to use compressed sensing
(CS) theory. As an emerging technique, CS has brought about
a breakthrough to sparse signal reconstruction. According to this
theory, the exact recovery of an unknown sparse signal can be
achieved from limited measurements by solving a sparsity constrained
optimization problem. Furthermore, this method possesses superresolving ability, overcoming the limitation imposed by bandwidth and
synthetic aperture [13]. Recent publications have shown the great
potential of CS theory in various applications. In [14], it is argued
that a radar system can eliminate the need for the matched filter in
the radar receiver and reduce the required receiver analog-to-digital
conversion bandwidth by utilizing CS theory. In [15, 16], CS theory is
applied to ground penetrating radar imaging, although only a small
subset of the measurements are used, the CS theory still obtains
sparser and sharper target images compared to the standard backprojection method. For wide-angle imaging, where the isotropic point
scattering assumption is violated, the CS is employed to improve the
resolution [17, 18]. In tomographic SAR (Tomo-SAR), CS theory is
exploited to overcome the poor resolution and aliasing effect brought
by limited overall baseline and non-uniform inter-track distance, and
better tomographic reconstruction results of man-made objects such as
buildings and stadiums have been acquired [13, 19]. For SAR imaging,

Progress In Electromagnetics Research, Vol. 140, 2013

65

especially the imaging of man-made targets, the scattered signal can

be deemed as a few point-like scatterers contributions. In this sense,
sparsity assumption is well satisfied for SAR imaging [20], which paves
the way for utilizing CS theory in SAR imaging. However, there are
still several problems that should be concerned.
Firstly, some researchers have realized the problem that the
performance of CS-based imaging algorithms degrades as the SNR
decreases [21, 22]. Therefore, how to obtain robust reconstruction in
the presence of strong noise is challenging. Secondly, there are various
methods to implement CS imaging, such as basis pursuit (BP) [23],
matching pursuit (MP) [24], orthogonal matching pursuit (OMP) and
so on. Among them, the OMP and its variations are most widely
used for their convenience and effectiveness [1, 2527]. Nevertheless, as
special cases of greedy methods, the super-resolving abilities of OMP
and its variations are poor [21, 28], i.e., the super-resolving ability of
CS is not exploited sufficiently in many occasions. Thirdly, rather
than realistic SAR scenes, simple and clean test scenes are usually
investigated [2931]. In these scenes, simple targets such as several
isolated point scatterers take up only a small part of the scene, while
the rest part of the scene is free of strong targets. In this case, only the
small part of the scene including strong targets should be investigated
and the influence from other parts of the scene can be negligible, which
is too ideal.
Aiming at aforementioned problems, this paper focuses on the
improvement of SF SAR imaging quality based on the CS theory.
The contribution of this paper can be summarized as follows. Firstly,
we propose a new CS imaging framework for SF SAR imaging, in
which an operator is introduced to counter the influence of noise
or clutter. Secondly, in order to exploit the super-resolving ability
of the CS algorithms sufficiently, we provide an extension of the
nonquadratic optimization technique [32] which was proposed by Cetin
to the application of SAR imaging. Consequently, the super-resolving
ability of the proposed algorithm is much better than that of OMP
and MF based algorithms. Thirdly, in the proposed framework, the
mutual coherency characteristic is more satisfactory. Therefore, even
if a complex scene considered, the influence from other targets can be
excluded. In this paper, a complex scene including a vehicle and some
trihedral corner reflectors is investigated, and better imaging results
are obtained using the proposed framework.

66

Pang et al.

2. SIGNAL MODEL
In SF SAR, wide bandwidth is generated with a series of pulses with
carrier frequencies increasing from pulse to pulse. Therefore, we do
not require the wide instantaneous bandwidth, which will mitigate
the hardware burden for the radar system. Assuming the SF signal
transmitted by radar is

sT ( ) = wr
exp (j2f (m) )
(1)
Tp
where f (m) = f0 + mf denotes the carrier frequency of mth pulse;
f0 is the start carrier frequency; f is the frequency step; the range
of m is 1 B/f and B stands for the synthetic bandwidth; Tp
is the pulse width; is the range time; wr ( /Tp ) is the envelope of
transmitted pulse which is usually chosen as rectangular pulse or LFM
pulse signal as described in Figure 1.
f

f
N Pulses

..

N Pu l ses

..

B
f

Tp

Tp

f0

f0
TR

(a)

TR : Pulse repetition
interval

TR

(b)

Figure 1. Two kinds of time-frequency profiles for stepped frequency

radar. (a) Rectangular pulse envelope, (b) LFM pulse envelope.
For a point scatterer situated at (x, y), its echo can be expressed
as

t y/v
2R(t; x, y)/c
sR (, t) = g(x, y)wr
wa
Tp
Ts

2R(t; x, y)
exp j2f (m)
c

(2)

where c is the velocity of light; t is the slow time; wa ((t y/v)/Ts )

is the azimuth envelope (azimuth beam pattern); Ts is synthetic
aperture time; v is velocity of radar platform; g(x, y) denotes the

Progress In Electromagnetics Research, Vol. 140, 2013

67

reflectivity amplitude of the target situated at (x, y) and x stands

for
p range coordinate, y stands for azimuth coordinate; R(t, x, y) =
x2 + (vt y)2 is the traveling path of electromagnetic wave from
radar to target at slow time t.
Taking sref (, t) = exp(j2f (m) ) as the reference signal, the
demodulated signal can be expressed as
sRD (, t) = sR (, t)sref (, t)

ty/v
2R(t;x,y)/c
4f (m)R(t;x,y)
= g(x,y)wr
wa
exp j
(3)
Tp
Ts
c
where the superscript * stands for conjugate. In order to take
advantages of digital signal processing, the demodulated signal should
be sampled at range time to get discrete signal. Assuming rectangular
pulse envelope is used, the range time sampled signal can be expressed
as

t y/v
4f (m) R (t; x, y)
sRD (f , t) = g (x, y) wa
exp j
(4)
Ts
c
Then, we perform the Fourier transform on sRD (f , t) and get the
expression of two-dimensional frequency spectrum as

sRD (f , ft ) = g (x, y) wa

cxft

c2 f 2
Ts 2v 2 f (m) 1 4v2 f 2t(m)

c2 f 2
4xf (m)
y
1 2 2 t
exp j
exp j2 ft (5)
c
4v f (m)
v

Up to now, the signal model for a point target situated at (x, y) has
been derived. Nevertheless, for a scene consisting of many targets, the
measured signal should be represented as a superposition of the echoes
reflected from all targets which are illuminated by the radars beam.
In this sense, the received signal should be written as

ZZ

cxft

sRD (f , ft )=
g(x, y)wa

2f 2
c
Ts 2v 2 f (m) 1 4v2 f 2t(m)
G

4xf (m)
c2 f 2
y
exp j
1 2 2 t
exp j2 ft dxdy (6)
c
4v f (m)
v
where G stands for the scene illuminated by the radars beam.

68

Pang et al.

In order to apply CS algorithm to two-dimensional SAR imaging,

the two-dimensional distribution of targets reflectivity should be
sampled to get the discrete form as

g (x1 , y1 ) . . . g (x1 , yQ )
..
..
..

G=
(7)
.
.
.
g (xP , y1 ) . . . g (xP , yQ )
where P is the number of samples in range direction and Q the number
of samples in azimuth direction. For the purpose of super-resolving, the
sample interval should be smaller than the Fourier resolution cell. In
the following experiments, the sample interval is chosen as one-third
of the Fourier resolution cell. Therefore, P and Q are usually large
numbers.
Based on (7), the received signal in (6) can be expressed as

sRD (f (m), ft )=

Q
P X
X
p=1 q=1

exp j

g(xp , yq )wa

cxp ft

c2 f 2
Ts 2v 2 f (m) 1 4v2 f 2t(m)

4xp f (m)
yq
c2 f 2
1 2 2 t
exp j2 ft (8)
c
4v f (m)
v

Following, the two-dimensional signal shown in (8) are sampled in the

ft direction (the signal is already discrete in the f direction) to get

sRD (m, n) =

Q
P X
X
p=1 q=1

g(xp , yq )wa

cxp ft (n)

q
Ts 2v 2 f (m) 1

c2 ft2 (n)
4v 2 f2 (m)

s
!

yq
4xp f (m)
c2 ft2 (n)
1 2 2
exp j2 ft (n) (9)
exp j
c
4v f (m)
v

After that, the two-dimensional signals sRD (m, n) and g(xp , yq )

are expressed in lexicographically ordered vector as

sRD (1, 1)
g(x1 , y1 )
..
..

.
.

sRD (N, 1)
g(xP , y1 )

..
..
g=

sRD =
(10)
.
.

s (1, M )
g(x , y )
1 Q
RD

.
..
.

.
.
sRD (N, M )
g(xP , yQ )

Progress In Electromagnetics Research, Vol. 140, 2013

69

where sRD is a M N 1 vector, g is a P Q 1 vector.

Then Equation (9) can be expressed as
sRD = Ag

(11)

where A is a M N P Q dictionary which can be expressed as

A = [a (1, 1) , . . . a (N, 1) , . . . , a (1, M ) , . . . a (N, M )]T

(12)

where the superscript T stands for transpose.

a (m, n) = [a (m, n, 1, 1) , . . . , a (m, n, P, 1) , . . . ,
a (m, n, 1, Q) , . . . , a (m, n, P, Q)]T

(13)

cxp ft (n)

a(m,n,p,q)=wa
q

c2 f 2 (n)
Ts 2v 2 f (m) 1 4v2 ft2 (m)

4xp f (m)
yq
c2 ft2 (n)
exp j
1 2 2
exp j2 ft (n) (14)
c
4v f (m)
v
In the presence of noise, we should solve equations
sRD = Ag + n

(15)

to reconstruct the two-dimensional distribution of the targets

reflectivity g. Since the set of equations in (15) are conventionally
underdetermined, the CS theory is usually resorted to.
3. CS THEORY
According to CS theory, it is possible to recover sparse signal from
a number of measurements which are much less than the number of
Nyquist rate samples. Nevertheless, the key condition for CS theory
to hold up lies in the sparsity or compressibility of the signal. A vector
x CL is said to be K sparse when there exists a basis satisfying
x = s while s CL has only K L nonzero elements. Comparing
to sparsity, compressibility requires the elements of s follow a power
decay law with K strongest coefficients [33], which is less rigorous and
can be satisfied by more real-world signals.
Now, let us consider a measurement matrix (sensing matrix)
CM L with M L, then the measurement equations can be
written as
y = x = s = s
(16)
where = is called dictionary. However, since this set of
equations is underdetermined, exact solution of s from Equation (16) is

70

Pang et al.

challenging. In mathematical sense, underdetermined equations have

infinitely many solutions. Within CS theory, the K sparse vector can
be estimated from M O(K log L) measurements by utilizing the
sparsity of the signal, which means solving the following optimization
problem:
(P0 )

sCS = arg min ksk0

s

subject to

y = s

(17)

where, ksk0 returns the number of nonzero elements in the vector

s. Unfortunately, (P0 ) is a N-P hard problem and is computational
difficult. Instead, the problem is usually solved by a relaxed version
(P1 )

sCS = arg min ksk1

s

subject to

y = s

(18)

where ksk1 returns the sum of the absolute values of all the elements
in the vector s.
In practice, the noise and clutter from measuring and background
are inevitable, therefore the actual problem we should solve is
(P1 )

sCS = arg min ksk1

s

subject to

ky sk2

(19)

where stands for the noise level.

From above discussion, it is possible to note that (P1 ) is the
approximate solution of (P0 ) and (P1 ) takes the influence of noise
into account. Two natural questions are whether the approximation
holds up and what is the performance of the algorithm in the presence
of noise. One sufficient condition for both (P0 ) and (P1 ) to have the
same solution and for (P1 ) to stably recover the sparse signal in the
presence of noise is known as the restricted isometry property (RIP).
A matrix is said to satisfy the RIP provided there exists a constant
s (0, 1) making
(1 s ) kvk22 kvk22 (1 + s ) kvk22

(20)

holds up for any K sparse vector v. The RIP essentially states that any
subsets of K column chosen from are nearly orthogonal. Generally,
the smaller s , the better noise resistance performance of the algorithm.
However, in practice, there is no computational feasible way to
check RIP properly, as it is combinatorial in nature [20]. Fortunately,
there exist some alternatives. One of them is mutual coherence, which
is defined as
|hi , j i|
() = max
(21)
i6=j ki k2 kj k2
where i stands for the ith column of matrix . From another aspect,
the mutual coherence can be viewed as the largest off-diagonal element

Progress In Electromagnetics Research, Vol. 140, 2013

71

H ,
where
is obtained by normalizing each column of
of matrix
[34]; superscript H denotes the complex conjugate.
However, the mutual coherence () only provides the largest
coherence between different columns of , which occurs between two
close columns. In other words, mutual coherence () only describes
H
but not its full characteristics. It is
the local characteristic of
suitable when targets only take up a small part of the scene, while
the rest part of the scene is free of strong targets. Nevertheless, when
a large scene which comprises targets deployed dispersively in it is
investigated, the influence of other targets (i.e., the coherence between
other columns) can not be negligible. In this situation, the mutual
coherence () is not sufficient to ensure stable reconstruction.
are
H
In this paper, the full characteristics of matrix
investigated but not only its largest off-diagonal element. Comparing
H
is more robust in predicting the performance of the CS
to (),
algorithm, especially in situation where a large scene comprises targets
deployed dispersively in it. In the following sections, it can be noted
that the mutual coherence characteristic is improved by the proposed
algorithm.
4. NUMERICAL ITERATIVE SUPER-RESOLVING
IMAGING ALGORITHM
As mentioned earlier, the MF based algorithms face the problem
of Fourier resolution limited by the radar bandwidth and synthetic
aperture. Nevertheless, the radar resolution is very important to the
understanding of the image, especially for the application to man-made
structures, which motivates the approaches for enhanced resolution.
The super-resolving ability of CS algorithm has been represented
in many literatures [32, 35, 36].
However, there are various
implementations for CS, and the resolving abilities of different
implementations differ. For example, the OMP and its variations
are widely exploited for their convenience and effectiveness. However,
the super-resolving ability of these algorithms are poor [21]. In this
paper, a regularization method is proposed based on nonquadartic
optimization technique proposed by cetin [32]. The derivation of this
algorithm can be summarized as follows.
First, we formulate the imaging problem as the following
regularization problem

g
= arg min ksRD Agk22 + kgkkk
(22)
g

where term ksRD

Agk22

is used for preserving the data fidelity of the

72

Pang et al.

solution; is the scalar parameter to balance the emphasis on data

fidelity or signal energy; kkk denotes the lk -norm. In this algorithm,
we constraint k 1 since the smaller value of k implies less penalty
on large pixel values as compared to larger k and results in better
preservation of the scatter magnitudes [32].
In a following, we denote
J (g) = ksRD Agk22 + kgkkk

(23)

as the objective function. In order to minimize J(g), we should first

calculate the differential of J(g) with respect to g. However, in order
to eliminate the nondifferentiablity of the lk -norm around the origin
when k 1, J(g) should be modified as
J (g) = ksRD

Agk22

PQ
X

|gi |2 +

k/2

(24)

i=1

where is a constant small enough not to affect kgkkk and is chosen

as = 105 in our experiment. Then the differential of J (g) can be
expressed as
J (g) = H (g) g 2AH sRD
(25)
where H(g) = (2AH A + k(g)), superscript H denotes the complex
k
conjugate, (g) = diag{1/(|gi |2 + )1 2 }. The objective is to find
a g satisfying J(g) = 0. Noting that H(g) is the function of
unknown targets scattering reflectivity g, we can not simply obtain the
estimation of g by letting g = 2H1 (g)AH sRD . Instead, an iterative
algorithm should be utilized for solving this problem. Examining the
gradient expression of (25), H(g) resembles as a coefficient matrix
multiplying g. Consequently, H(g) is taken as the Hessian matrix.
Then the iterative algorithm can be expressed as
gn+1 = gn H1 (gn ) J (gn )
gn k22 /kgn k22

(26)

We terminate iteration (26) when kgn+1

< . Where
is a small positive termination constant and is set as = 106 in our
experiment. The geometry demonstration of this iterative algorithm
is shown in Figure 2.
Another idea to overcome the resolution limit is to use modern
spectral estimation methods such as MUSIC (Multiple Emitter
Location and Signal Parameter Estimation), ESPRIT (Estimation of
Signal of Parameters Via Rotational Invariance Techniques) rather
than MF based methods. The modern spectral estimation achieves
super resolution in direction of arrival (DoA). Images generated by
these approaches are inherently free of sidelobes, and the resolution

Progress In Electromagnetics Research, Vol. 140, 2013

J (g)

73

J (gn )

H -1 (g n ) =( ( J(gn )))-1
H
= (2A A+k(gn )) -1
= (gn -g n+1) ( J(gn ))-1

( ( J(g)) = 0
gn+1

gn

Figure 2. Geometry demonstration of the iterative algorithm.

relies on the precision of estimation. Wherein, ESPRIT [37, 38] exploits
the rotational invariance of sub-array to realize the estimation of DoA.
Comparing with MUSIC, ESPRIT has the advantage of computational
efficiency, and eliminates the demand of spectral peak searching.
Furthermore, the ESPRIT algorithm can achieve the Cramer-Rao
lower bound (CRLB) on location error variance for the sum-ofreflectors model with sufficiently high signal-to-noise ratio [39, 40].
Therefore, in this paper, the super-resolving results of ESPRIT are
provided to demonstrate the super-resolving ability of the proposed
algorithm.
From above discussion, it can be noted that the proposed
algorithm is an iterative frequency domain CS algorithm. Therefore,
for simplicity, it is acronymized as IFCS in the following discussion.
5. EXPERIMENT RESULTS
In this section, the superiority of IFCS is demonstrated using simulated
and real data. The imaging results of Omega-K algorithm and
time domain CS (i.e., the CS theory is applied directly to the time
domain echo) imaging algorithm proposed in [12] are also represented
for comparison. Among them, Omega-K [4143] algorithm is a
representative MF based imaging algorithm. Since the data are
processed in the 2D frequency domain with little approximations in
Omega-K algorithm, it is often deemed as an accurate algorithm and
used as a reference to evaluate other imaging algorithms.
5.1. Noise Resistant Capability
Firstly, we demonstrate the noise resistant capability of IFCS on
a simple simulated scene which consists of four point scatterers

74

Pang et al.

Table 1. Scatterer properties.

Scatter
Scatter
Scatter
Scatter

1
2
3
4

Range
54.5
52.4
56
55.4

Azimuth
0
0
1.5
1.5

Amplitude
0.5
1
0.3
1

Table 2. Simulation parameters.

Parameter
Carrier Frequency
Bandwidth
Radar Velocity
Frequency Step
Sweep Period
Radar Height
Azimuth Beam Width

Value
10 GHz
500 MHz
0.1 m/s
1 MHz
120 ms
2.5 m
2.5

with different reflectivity amplitude. The coordinate and reflectivity

amplitude of these scatterers are listed in Table 1. The simulation
parameters are listed in Table 2.
Figure 3 shows the color-coded (color are coded by the reflectivity
amplitude of the scatterers) imaging results of the simulated scene with
different imaging algorithms. For IFCS, the parameters are chosen as
k = 0.1 and = 1000 empirically. Considering the inevitable noise
from measuring and background, the simulated data are added with
white Gaussian noise, which is commonly used for radar measurement
noise. From up to down, the SNR levels are 0 dB, 10 dB and 10 dB
respectively.
From the color-coded images shown in Figure 3(a) to Figure 3(c),
it can be noted that when SNR is high, both the time domain CS
algorithm and IFCS can reconstruct the scatterers well. Namely not
only the positions of scatterers are correctly reconstructed but also
their reflectivity amplitude information is well preserved. Furthermore,
comparing to the imaging result of the Omega-K algorithm, the
absence of sidelobes makes the images generated by the time domain
CS and IFCS much preferable, considering that the sidelobes will
prevent a better discrimination of closely located targets. However,
when SNR degrades to 0 dB, artifacts will appear in image provided

57

57

56

56

56

55
54

55
54

-1

-3

-2

Azimuth (m)

-1

-3

56

56

54

Range (m)

56
55

55
54

-3

-2

Azimuth (m)

-1

(f)

57

57

56

56

55
54

55
54
53
52

52
-1

-2

Azimuth (m)

53

53

-2

-3

Range (m)

Range (m)

54

54

(e)

55

55

Azimuth (m)

56

52
-3

-1

(d)

57

52

52
-1

53

53

53

(c)
57

-2

-1

(b)
57

52
-3

-2

Azimuth (m)

57

Range (m)

Range (m)

54

Azimuth (m)

(a)

Range (m)

55

52

52
-2

75

53

53

53
52
-3

Range (m)

57

Range (m)

Range (m)

Progress In Electromagnetics Research, Vol. 140, 2013

-3

-2

-1

-3

-2

-1

Azimuth (m)

Azimuth (m)

Azimuth (m)

(g)

(h)

(i)

Figure 3. Color-coded imaging results of a simulated scene with 4

scatterers having different reflectivity amplitudes under different SNR
levels, color is coded by the reflectivity amplitude of the scatterers.
(Upper plots) SNR = 10 dB. (Middle plots) SNR = 0 dB. (Lower
plots) SNR = 10 dB. (a) Omega-K, (b) time domain CS, (c) IFCS,
(d) Omega-K, (e) time domain CS, (f) IFCS (g) Omega-K, (h) time
domain CS, (i) IFCS.
by the time domain CS algorithm (shown in Figure 3(e)) while the
image provided by IFCS is still free of artifacts (shown in Figure 3(f)).
When SNR further degrades to 10 dB, the amplitude of artifacts will
rise, making the weak scatterer (i.e., scatterer 3 situated at the top
right part of the scene) submerged in artifacts. Nevertheless, IFCS
exhibits its robustness and stability in strong noise cases (shown in

76

Pang et al.

Figure 3(i)), in which targets are detected with accurate amplitudes at

the right positions, and no artifacts appear.
The better noise resistant capability of IFCS can be owed to the
azimuth Fourier transform included in it. After Fourier transforming,
the signal energy will concentrate to several Doppler bins, while the
noise energy is still spreading over the whole frequency domain, leading
to the improvement of SNR. However, for the time domain CS imaging
algorithm, it has been proven that the estimation error of the unknown
signal is approximately proportional to the noise level [44]. Therefore,
it is not difficult to understand that the performance of the time
domain CS imaging algorithm degrades and artifacts increases as SNR
decreases.
5.2. Super-resolving Performance

57

57

56

56

56

55
54

55
54

54

53

53

52
-3

52
-3

52
-3

-1

-2

Azimuth (m)

-1

55
54
53

-2

Azimuth (m)

-1

52
-3

1
0.8

0.6
0.4

0.6
0.4

0.2

0.2

0.2

52

53

54

55

Range (m)

56

57

52

53

54

55

Range (m)

56

57

Amplitude

1
0.8

Amplitude

1
0.8

0.4

-1

Azimuth (m)

0.6

-2

Azimuth (m)

0.8

Amplitude

Amplitude

55

53

-2

Range (m)

57

56

Range (m)

57

Range (m)

Range (m)

In order to evaluate the super-resolving performance of IFCS, scenes

which comprises two closely separated scatterers in range/azimuth
direction are simulated with the parameters listed in Table 2.
Afterwards, different algorithms are applied and the imaging results
are presented in Figure 4 to Figure 11. The SNR is set to 10 dB.
The Fourier resolutions for range and azimuth direction are 0.3 m
and 0.35 m respectively. For IFCS, the parameters are still chosen
as k = 0.1 and = 1000 empirically.
Firstly, the situation of two closely separated scatterers in range
direction is investigated, and the results are presented in Figure 4 to
Figure 7. From Figure 4 it can be noted that the Omega-K imaging

0.6
0.4
0.2

52

53

54

55

Range (m)

56

57

52

53

54

55

56

57

Range (m)

Figure 4. Imaging results of two closely separated scatterers in range

direction using the Omega-K algorithm. One scatterer situated at
(54.5, 0), the other one situated at (54.9, 0), (54.8, 0), (54.7, 0) or
(54.6, 0) from left to right. (Upper plots) Two-dimensional imaging
results. (Lower plots) Range profiles corresponding to Upper plots.

57

56

56

55
54

55
54

55
54

53

53

53

52

52

52

40

60

80

100

20

Sequence of Monto Carlo Run

40

60

80

Range (m)

57

56

Range (m)

57

56

20

55
54
53
52

100

20

Sequence of Monto Carlo Run

40

60

80

20

100

0.8

0.8

0.4
0.2
0
51

0.6
0.4
0.2

52

53

54

55

56

0
51

57

Amplitude

1
0.8

Amplitude

0.6

0.6
0.4
0.2

52

Range (m)

53

54

55

56

0
51

57

40

60

80

100

Sequence of Monto Carlo Run

Sequence of Monto Carlo Run

0.8

Amplitude

Amplitude

77

57

Range (m)

Range (m)

Progress In Electromagnetics Research, Vol. 140, 2013

0.6
0.4
0.2

52

53

54

55

56

0
51

57

52

53

54

55

56

57

Range (m)

Range (m)

Range (m)

57

57

56

56

56

55
54

55
54

54

53

53

52

52

52

40

60

80

100

20

Sequence of Monto Carlo Run

40

60

80

Sequence of Monto Carlo Run

54
53
52

20

100

55

40

60

80

20

100

0.8

0.8

0.8

0.8

0.4
0.2
0
51

0.6
0.4

52

53

54

55

Range (m)

56

57

0.6
0.4
0.2

0.2
0
51

Amplitude

Amplitude

0.6

53

54

55

Range (m)

56

57

51

60

80

100

0.6
0.4
0.2
0

0
52

40

Sequence of Monto Carlo Run

Sequence of Monto Carlo Run

Amplitude

Amplitude

55

53

20

Range (m)

57

56

Range (m)

57

Range (m)

Range (m)

Figure 5. Imaging results of two closely separated scatterers in range

direction using the OMP algorithm. One scatterer situated at (54.5,
0), the other one situated at (54.9, 0), (54.8, 0), (54.7, 0) or (54.6, 0)
from left to right. (Upper plots) Range profiles of 100 Monte Carlo
runs. (Lower plots) Range profile corresponding to 100th Monte Carlo
Run.

52

53

54

55

Range (m)

56

57

51

52

53

54

55

56

57

Range (m)

Figure 6. The imaging results of two closely separated scatterers in

range direction using the IFCS algorithm. One scatterer situated at
(54.5, 0), the other one situated at (54.9, 0), (54.8, 0), (54.7, 0) or
(54.6, 0) from left to right. (Upper plots) Range profiles of 100 Monte
Carlo runs. (Lower plots) Range profile corresponding to 100th Monte
Carlo Run.
algorithm cannot resolve the scatterers falling into one resolution cell
and suffers from sidelobes. Comparing to Omega-K imaging algorithm,
the images obtained by the OMP algorithm is free of sidelobes.
However, the super-resolving ability of the OMP algorithm is still

78

Pang et al.

55
54

53

52

52

52

-2

-1

-3

-2

Azimuth (m)

-1

54

54

53

53
52
-3

Estimated Position
True Position

56
55

55

53

-3

Estimated Position
True Position

56

Range (m)

54

Estimated Position
True Position

56

Range (m)

Range (m)

55

57

57

57

Estimated Position
True Position

56

Range (m)

57

-2

-1

-3

-2

-1

Azimuth (m)

Azimuth (m)

Azimuth (m)

57

57

56

56

56

55
54

55
54

54

53

53

52
-3

52
-3

52
-3

-1

-2

Azimuth (m)

-1

55
54
53
52

-2

Azimuth (m)

-1

-3

0.8

0.8

0.8

0.8

0.4
0.2
0
-3

0.4

-1

Azimuth (m)

0
-3

0.6
0.4
0.2

0.2
-2

Amplitude

Amplitude

0.6

-2

-1

Azimuth (m)

0
-3

-1

Azimuth (m)

0.6

-2

Azimuth (m)

Amplitude

Amplitude

55

53

-2

Range (m)

57

56

Range (m)

57

Range (m)

Range (m)

Figure 7. Imaging results of two closely separated scatterers in range

direction using ESPRIT algorithm. One scatterer situated at (54.5,
0), the other one situated at (54.9, 0), (54.8, 0), (54.7, 0) or (54.6, 0)
from left to right.

0.6
0.4
0.2

-2

-1

Azimuth (m)

0
-3

-2

-1

Azimuth (m)

Figure 8. Imaging results of two closely separated scatterers in

azimuth direction using the Omega-K algorithm. One scatterer
situated at (54.5, 0), the other one situated at (54.5, 0.4), (54.5,
0.3), (54.5, 0.2) or (54.5, 0.1) from left to right. (Upper plots)
Two-dimensional imaging results.
(Lower plots) Range profiles
corresponding to upper plots.
limited. When two scatterers move into one resolution cell, the results
become unsatisfactory. Firstly, the scatterers are not right located in
reconstructed images. For example, the scatterers situated at (54.5, 0)
and (54.7, 0) are located at (54.4, 0) and (54.6, 0) instead. Secondly,
the amplitude information of the scatterers is not well preserved by
the OMP algorithm. As a result, the reconstructed amplitude of
two scatterers with equal amplitude appears to have considerable
discrepancy. Figure 6 shows the imaging results of IFCS. It is pleased
to see that the super-resolving performance is much improved by
using IFCS. Two scatterers are stably reconstructed with accurate
position and amplitude until the range direction distance between them
is closer than one-third Fourier resolution cell. Figure 7 shows the
imaging result using the ESPRIT algorithm with a priori information

1
0
-1

1
0
-1

1
0
-1

-2

-3
40

60

80

100

0
-1
-2
-3

-3
20

40

60

80

100

20

Sequence of Monto Carlo Run

Sequence of Monto Carlo Run

-2

-3
20

Azimuth (m)

Azimuth (m)

-2

40

60

80

20

100

Sequence of Monto Carlo Run

1

0.8

0.8

0.4

0.4
0.2

0.2
0
-3

0.6

-2

-1

0
-3

Amplitude

1
0.8

Amplitude

0.6

0.6
0.4
0.2

-2

-1

0
-3

60

80

100

0.6
0.4
0.2

-2

Azimuth (m)

Azimuth (m)

40

Sequence of Monto Carlo Run

0.8

Amplitude

Amplitude

79

Azimuth (m)

Azimuth (m)

Progress In Electromagnetics Research, Vol. 140, 2013

-1

0
-3

-2

Azimuth (m)

-1

Azimuth (m)

57

56

55

1
0
-1
-2

1
0
-1

Amplitude

Amplitude

Amplitude

Amplitude

Figure 9. Imaging results of two closely separated scatterers in

azimuth direction using the OMP algorithm. One scatterer situated
at (54.5, 0), the other one situated at (54.5, 0.4), (54.5, 0.3), (54.5,
0.2) or (54.5, 0.1) from left to right. (Upper plots) Azimuth profiles
of 100 Monte Carlo runs. (Lower plots) Azimuth profile corresponding
to 100th Monte Carlo run.

54
53

0
-1
-2

-2
52

-3

20

40

60

80

-3

100

40

60

80

100

20

Sequence of Monto Carlo Run

40

60

80

-3

100

20

Sequence of Monto Carlo Run

1

0.8

0.8

0.4

0.4

-2

-1

Azimuth (m)

0
-3

0.6
0.4
0.2

0.2

0.2
0
-3

0.6

Amplitude

1
0.8

Amplitude

0.6

-2

-1

Azimuth (m)

0
-3

40

60

80

100

Sequence of Monto Carlo Run

0.8

Amplitude

Amplitude

Sequence of Monto Carlo Run

20

0.6
0.4
0.2

-2

-1

Azimuth (m)

0
-3

-2

-1

Azimuth (m)

Figure 10. Imaging results of two closely separated scatterers in

azimuth direction using the IFCS algorithm. One scatterer situated
at (54.5, 0), the other one situated at (54.5, 0.4), (54.5, 0.3), (54.5,
0.2) or (54.5, 0.1) from left to right. (Upper plots) Azimuth profiles
of 100 Monte Carlo runs. (Lower plots) Azimuth profile corresponding
to 100th Monte Carlo Run.
about the number of scatterers. Blue cross and red circle stand for
the true and estimated positions of the scatterers respectively while
their size is directly proportional to the amplitude of the scatterers.
Similarly to the results obtained using IFCS, two scatterers are well
reconstructed until they are closer than one-third Fourier resolution

80

Pang et al.

55
54

55
54

53

53

53

52
-3

52

52

-2

-1

-3

-2

Azimuth (m)

-1

Azimuth (m)

Estimated Position
True Position

56

Range (m)

54

Estimated Position
True Position

56

Range (m)

Range (m)

55

57

57

57

Estimated Position
True Position

56

Estimated Position
True Position

56

Range (m)

57

55
54
53
52

-3

-2

-1

-3

-2

Azimuth (m)

-1

Azimuth (m)

Figure 11. Imaging results of two closely separated scatterers in

azimuth direction using ESPRIT algorithm. One scatterer situated at
(54.5, 0), the other one situated at (54.5, 0.4), (54.5, 0.3), (54.5, 0.2)
or (54.5, 0.1) from left to right.
cell, which demonstrates the super-resolving capability of IFCS from
another aspect. However, comparing to the ESPRIT algorithm, IFCS
eliminates the step of model order selection, i.e., IFCS adaptively
chooses the number of scatterers.
Afterwards, the situation of two closely separated scatterers in
azimuth direction is investigated. Since similar conclusion can be
established, the results shown in Figure 8 to Figure 11 are not analyzed
in detail here.
5.3. Improvement of Mutual Coherence Characteristics
In this section, not only the largest off-diagonal element of matrix
H ,
i.e., (), but also its full characteristics corresponding to the

5.5 m 6 m scene in Section 5.1 are investigated. Table 3 shows the

parameter () for the time domain CS algorithm and IFCS. It seems
that if only () is used for evaluation, the performance of the time
domain CS algorithm and the IFCS algorithm is comparable.
Table 3. The () parameter of different CS framework.
()

Time domain CS
0.31

IFCS
0.34

H .
It can be noted
Figure 12 shows the representation of
H

that the elements of distribute following a band crossing

the matrix from the upper left corner to the bottom right corner.
Theoretically, the mutual coherence decreases as the distance between
H
exhibit
the columns augments. Nevertheless, the elements of
some fluctuations in time domain CS framework when distance between
the columns augments (shown in Figure 12(a)). On the contrary,
the mutual coherence characteristic of IFCS is preferable. On one

Progress In Electromagnetics Research, Vol. 140, 2013

500

500

1000

1000

1500

1500

2000

2000

2500

2500

3000

3000
500 1000 1500 2000 2500 3000

(a)

81

500 1000 1500 2000 2500 3000

(b)

H .
(a) Time domain CS,
Figure 12. Representation of matrix
(b) IFCS
hand, the large mutual coherence concentrate around the principal
H .
On the other hand, when distance between
diagonal of matrix
the columns augments, the mutual coherence decreases dramatically,
indicating the negligible influence from further targets. The advantage
of the mutual coherence characteristic in IFCS is demonstrated in next
section using measured data.
5.4. Real Data Imaging Results
In this section, the measured X-band stepped frequency SAR data are
used to demonstrate the effectiveness of IFCS. The test scene located at
the suburb of Changsha, China, is a slightly undulated field covered by
weeds and a few bushes. In this test site, a rail SAR experiment system
is established with a vector network analyzer (VNA) mounted on the
platform which is moving along the rail with preset velocity. While
the platform is moving, the VNA transmits SF signal and collects
echoes under the control of a computer. The experiment parameters
are shown in Table 4. Figure 13 presents the photograph of the rail
SAR experiment system and its components.
Figure 14 shows photographs of the targets deployed in the
test scene. As having the ideal point-like scattering properties and
exactly scattering mechanism, the trihedral corner reflectors are widely
utilized to demonstrate the performance of imaging algorithm. In our
experiment, four trihedral corner reflectors are deployed in the test
scene, as shown in Figure 14(c) to Figure 14(e). Trihedral corner
reflector A is deployed on a metal stair with some corner structures.
It is expected to be stronger in SAR image as the contribution of
the metal stair included. Trihedral corner reflectors B1 and B2 are
two targets which almost locate at the same azimuth. However, since
B2 is deployed above ground with a height of about 1 m, these two
targets separate in range direction with about 0.7 m. Trihedral corner
reflector C deployed on the ground is the target closest to radar. In

82

Pang et al.

(b)

(a)

(c)

(d)

Figure 13.
The photograph of rail SAR experiment system.
(a) Overview of the experiment system. (b) Rail. (c) Agilent VNA
(PNX-5242A). (d) Transmitting and receiving antennas.
Table 4. Experiment parameters.
Parameter
Carrier
Frequency

values
10 GHz

Radar
Bandwidth

500 MHz

Frequency
Step

1 MHz

Sweep
period
Platform
Velocity
Azimuth
Beam Width

120 ms
0.1 m/s

Parameter
Depression
Angle
Reference
Antenna
Height
Intermediate
Frequency
Bandwidth
Source
Port Power
Sweep
Type

values
3.8
4.34 m

1 MHz
15 dBmw
Step
Frequency

18

addition to simple corner reflectors, a more complicated target, i.e., a

vehicle, is deployed in the test site to validate the performance of IFCS
in reconstructing complex targets. The photograph of the vehicle is
shown in Figure 14(b).
Figure 15 shows a 2D image of the test site obtained by the

Progress In Electromagnetics Research, Vol. 140, 2013

(a)

83

(b)

(c)

(d)

(e)

Figure 14. The photograph of targets deployed in test scene. (a) The
overview of the targets. (b) The vehicle. (c) Trihedral corner reflector
A. (d) Trihedral corner reflectors B1 and B2. (e) Trihedral corner
reflector C.
65 70 75

75
70
65

Range (m)
35 40 45 50 55 60

Range (m)

60
55
50
45
40

30

30

35
-10

0
Azimuth (m)

(a)

10

B2
A

B1
C

-15 -10 -5 0 5
Azimuth (m)

10 15

(b)

Figure 15. (a) Imaging result of the test site obtained by Omega-K
algorithm. (b) Ground truth.
Omega-K algorithm and the ground truth. A remarkable consistency
can be found between them. Therefore, in the following discussion,
the image shown in Figure 15(a) is taken as a reference to evaluate
the performance of other algorithms. However, one aspect worth to
describe in Figure 15(a) is the noisy look, especially in areas without
strong point scatterers. The source of noise can be summarized as three
folds. Firstly, in order to accelerate frequency scan, the intermediate

Pang et al.
42

42

40

40

40

38
36

38
36

34

34

32

32
-8

-6

-4

-2

Range (m)

42

Range (m)

Range (m)

84

38
36
34
32

-8

-6

-4

Azimuth (m)

(a)

-2

-8

-6

-4

-2

Azimuth (m)

Azimuth (m)

(b)

(c)

Figure 16. Color-coded imaging results of four corner reflectors.

Color is coded by the reflectivity amplitude of the scatterers.
(a) Omega-K, (b) time domain CS, (c) IFCS.
frequency bandwidth of VNA is set as 1 MHz in our experiment, which
will shorten the frequency scanning period. However, the SNR is
sacrificed instead. Secondly, the returns are inevitably interfered by
clutter. Thirdly, as the coherent nature of SAR system, the returns
are contaminated by the speckle noise.
Figure 16 shows the imaging results of four trihedral corner
reflectors (A, B1, B2, C) using different algorithms. From imaging
result of the Omega-K algorithm (shown in Figure 16(a)), it can be
noted that trihedral corner reflector A appears to be stronger than
the others, as the contribution of the metal stair included. For the
time domain CS algorithm, although four trihedral corner reflectors are
right reconstructed, the blurring of the image caused by some artifacts
is also obvious. On the contrary, IFCS generates image with much less
artifacts. Moreover, trihedral A appears to be stronger than the others,
i.e., the reflectivity amplitude information is preserved by IFCS.
Figure 17 shows the imaging results of a complicated target, i.e., a
67

67
66

65
64
63
62

65
64
63
62
61

61
60
-2

66

Range (m)

Range (m)

Range (m)

66

10

60
-2

65
64
63
62
61

10

60
-2

Azimuth (m)

Azimuth (m)

Azimuth (m)

(a)

(b)

(c)

10

Figure 17. Color-coded imaging results of a vehicle. Color is coded

by the reflectivity amplitude of the scatterers. (a) Omega-K, (b) time
domain CS, (c) IFCS.

Progress In Electromagnetics Research, Vol. 140, 2013

85

vehicle, using different algorithms. For the time domain CS algorithm,

it can be seen that the outline and detail of the vehicle are not well
preserved and the image is contaminated by some artifacts. However,
the image obtained by using IFCS is much preferable. Firstly, the
shape and geometry features of the vehicle are well preserved, which
permits the effective use of region-based target recognition algorithm.
Second, the vehicle is represented as a set of scatterers in image
generated by using IFCS. Although precise super-resolving arguments
are not as easy for this complicated target, IFCS seems to capture
more details about the vehicle. Even some weak scattering centers
such as the one corresponding to the head of the vehicle (marked as
red rectangular in Figure 17(a)) which is weak in image obtained by
the Omega-K algorithm is enhanced in image obtained by using IFCS.
Thirdly, the sparse reconstruction obtained by CS-based algorithm has
been proven to be more effective than the images obtained by the
Omega-K algorithm when the automatic system is used for detection
and recognition [45]. Fourth, IFCS shows its robustness in the presence
of noise and clutter, as the image obtained by IFCS is dense and clean.
6. CONCLUSION
In this paper, a novel CS framework for stepped frequency SAR
imaging is proposed. Meanwhile, a super-resolving algorithm (i.e.,
IFCS) is proposed based on the nonquadratic optimization technique in
order to enhance super-resolving ability. The preferable characteristics
of the proposed framework with super-resolving algorithm are firstly
verified using a scene consists of four point scatterers with different
reflectivity amplitudes. In addition to the absence of sidelobes,
IFCS exhibits better performance than time domain CS under low
SNR. Afterwards, we simulate the scenes which comprise two closely
separated scatterers in range/azimuth direction to validate the superresolving ability of IFCS. The imaging results of OMP and ESPRIT are
also represented for comparison purpose. The results indicate that the
super-resolving ability of IFCS outperforms OMP and is comparable
to that of ESPRIT. Apart from simulated data, an X-band stepped
frequency rail SAR experiment system is established to demonstrate
the effectiveness of IFCS with real data. Another aspect worth to
mention is that not only simple corner reflectors but also a vehicle is
deployed in the test site, which is more persuadable. Consequently, in
images obtained by IFCS, the corner reflectors and the vehicle are right
located with accurate reflectivity and reduced artifacts. Furthermore,
more details about the vehicle are captured by IFCS.

86

Pang et al.

REFERENCES
1. Chan, Y. K. and V. C. Koo, An introduction to synthetic
aperture radar (SAR), Progress In Electromagnetics Research B,
Vol. 2, 2760, 2008.
2. Zhang, M., Y. W. Zhao, H. Chen, and W.-Q. Jiang, SAR imaging
simulation for composite model of ship on dynamic ocean scene,
Progress In Electromagnetics Research, Vol. 113, 395412, 2011.
3. Sun, J., S. Mao, G. Wang, and W. Hong, Polar format algorithm
for spotlight bistatic SAR with arbitrary geometry configuration,
Progress In Electromagnetics Research, Vol. 103, 323338, 2010.
4. Mohammadpoor, M., et al., A circular synthetic aperture eadar
for on-the-ground object detection, Progress In Electromagnetics
Research, Vol. 122, 269292, 2012.
5. Koo, V. C., et al., A new unmanned aerial vehicle synthetic
aperture radar for environmental monitoring, Progress In
Electromagnetics Research, Vol. 122, 245268, 2012.
6. Lim, T. S., C.-S. Lim, and V. C. Koo, Autofocus algorithm
performance evaluations using an integrated SAR product
simulator and processor, Progress In Electromagnetics Research
B, Vol. 3, 315329, 2008.
7. Storvold, R., E. Malnes, Y. Larsen, K. A. Hogda, and
S. E. Hamran, SAR remote sensing of snow parameters in
Norwegian areas-current status and future perspective, Journal
of Electromagnetic Waves and Applications, Vol. 20, No. 13, 1751
1759, 2006.
8. Li, Y., G. H. Lv, and X. Z. Liu, Moving-target velocity estimation
in a complex-valued SAR imagery, Progress In Electromagnetics
Research, Vol. 136, 301325, 2013.
9. Park, S.-H., J.-I. Park, and K.-T. Kim, Motion compensation
for squint mode spotlight SAR imaging using effcient 2D
interpolation, Progress In Electromagnetics Research, Vol. 128,
503518, 2012.
10. Liu, Y., Y. K. Deng, and R. Wang, An extended inverse chirp-z
transform algorithm to process high squint SAR data, Progress
In Electromagnetics Research, Vol. 138, 555569, 2013.
11. Wu, J., Z. Li, Y. Huang, Q. Liu, and J. Yang, Processing
one-stationary bistatic SAR data using inverse scaled fourier
transform, Progress In Electromagnetics Research, Vol. 129, 143
159, 2012.
12. Yang, J. G., J. Thompson, X. T. Huang, T. Jin, and Z. M. Zhou,
Random-frequency SAR imaging based on compressed sensing,

Progress In Electromagnetics Research, Vol. 140, 2013

13.

14.
15.
16.
17.
18.

19.

20.
21.

22.

23.
24.

87

IEEE Trans. on Geoscience and Remote Sensing, Vol. 51, No. 2,

983994, Feb. 2013.
Budillon, A., A. Evangelista, and G. Schirinzi, Threedimensional SAR focusing from multipass signals using compressive sampling, IEEE Trans. on Geoscience and Remote Sensing,
Vol. 49, No. 1, 488499, Jan. 2011.
Baraniuk, R. and P. Steeghs, Compressive radar imaging, Proc.
IEEE 2007 Radar Conference, 128133, Waltham, MA, Apr. 2007.
Gurbuz, A. C., J. H. McClellan, and W. R. Scott, Compressive
sensing for GPR imaging, Proc. 41st Asilomar Conference on
Signals, Systems Computers (ACSSC), 22232227, Nov. 2007.
Gurbuz, A. C., J. H. McClellan, and W. R. Scott, Compressive
sensing for subsurface imaging using ground penetrating radar,
Signal Processing, Vol. 89, No. 10, 19591972, 2009.
Austin, C. D., E. Ertin, and R. L. Moses, Sparse signal methods
for 3-D radar imaging, IEEE Journal of Selected Topics in Signal
Processing, Vol. 5, No. 3, 408423, Jun. 2011.
Varshney, K. R., M. Cetin, J. W. Fisher, and A. S. Willsky,
Sparse signal representation in structured dictionaries with
application to synthetic aperture radar, IEEE Trans. on Signal
Processing, Vol. 56, No. 8, 35483561, Aug. 2008.
Zhu, X. X. and R. Bamler, Tomographic SAR inversion by l1norm regularization The compressive sensing approach, IEEE
Trans. on Geoscience and Remote Sensing, Vol. 48, No. 10, 3839
3846, Oct. 2010.
Patel, V. M., G. R. Easley, D. M. Healy, and R. Chellappa,
Compressed synthetic aperture radar, IEEE Journal of Selected
Topics in Signal Processing, Vol. 4, No. 2, 244254, Apr. 2010.
Xing, S. Q., Q. F. Liu, D. H. Dai, Y. Z. Li, and X. S. Wang,
Polarimetric SAR tomography using l21 mixed norm sparse
reconstruction method, Progress In Electromagnetics Research,
Vol. 130, 105130, 2012.
Zhang, L., M. D. Xing, C. W. Qiu, J. Li, J. L. Sheng, Y. C. Li, and
Z. Bao, Resolution enhancement for inversed synthetic aperture
radar imaging under low SNR via improved compressive sensing,
IEEE Trans. on Geoscience and Remote Sensing, Vol. 48, No. 10,
38243838, Oct. 2010.
Chen, S., D. Donoho, and M. A. Saunders, Atomic decomposition
by basis pursuit, SIAM Journal on Scientific Computing, Vol. 20,
No. 1, 3361, 1999.
Tropp, J. and A. Gilbert, Signal recovery from random

88

25.
26.

27.
28.

29.

30.
31.
32.

33.

34.

35.

Pang et al.

measurements via orthogonal matching pursuit, IEEE Trans. on

Information Theory, Vol. 53, No. 12, 46554666, Dec. 2007.
Needell, D. and R. Vershynin, Greedy signal recovery and
uncertainty principles, Proc. SPIE, Vol. 6814, J-1J-12,
Bellingham, WA, 2008.
Alonso, M. T., P. L. Dekker, and J. J. Mallorqui, A novel strategy
for radar imaging based on compressive sensing, IEEE Trans.
on Geoscience and Remote Sensing, Vol. 48, No. 12, 38243838,
Dec. 2010.
Needell, D. and J. A. Tropp, CoSaMP: Iterative signal recovery
from incomplete and inaccurate samples, Appl. Comp. Harmonic
Anal., Vol. 26, 301321, 2008.
Xing, S. Q., Y. Z. Li, D. H. Dai, and X. S. Wang,
3D reconstruction of manmade objects using polarimetric
tomographic SAR, IEEE Trans. on Geoscience and Remote
Sensing, 455458, 2012.
Jouny, I., Compressed sensing for UWB radar target signature
reconstruction, IEEE 13th Digital Signal Processing Workshop
and 5th IEEE Signal Processing Education (DSP/SPE) Workshop, 714719, 2009.
Herman, M. and T. Strohmer, Compressed sensing radar, Proc.
IEEE International Conference on Acoustics, Speech and Signal
Processing (ICASSP), 15091512, 2008.
Herman, M. and T. Strohmer, High-resolution radar via
compressed sensing, IEEE Trans. on Signal Processing, Vol. 57,
No. 6, 22752284, Jun. 2009.
Cetin, M. and W. C. Karl, Feature-enhanced synthetic aperture
radar image formation based on nonquadratic regularization,
IEEE Trans. on Image Processing, Vol. 10, No. 4, 623631,
Apr. 2001.
Candes, E., J. Romberg, and T. Tao, Robust uncertainty
principles: Exact signal reconstruction from highly incomplete
frequency information, IEEE Trans.on Information Theory,
Vol. 52, No. 2, 489509, Feb. 2006.
Donoho, D., M. Elad, and V. N. Temlyakov, Stable recovery
of sparse overcomplete representations in the presence of noise,
IEEE Trans. on Information Theory, Vol. 52, No. 1, 618,
Jan. 2006
Potter, L. C., E. Ertin, J. T. Parker, and M. Cetin, Sparsity and
compressed sensing in radar imaging, Proceedings of the IEEE,
Vol. 98, No. 6, 10061020, Jun. 2010.

Progress In Electromagnetics Research, Vol. 140, 2013

89

36. Samadi, S., M. Cetin, and M. A. Masnadi-Shirazi, Sparse

representation-based synthetic aperture radar imaging, IET
Radar Sonar & Navigation, Vol. 5, No. 2, 182193, 2011.
37. Roy, R., A. Paulraj, and T. Kailath, ESPRIT A subspace
rotation approach to estimation of parameters of cisoids in noise,
IEEE Trans. on Acoustics, Speech and Signal Processing, Vol. 34,
13401342, Oct. 1986.
38. Roy, R. and T. Kailath, ESPRIT Estimation of signal
parameters via rotational invariance techniques, IEEE Trans.
on Acoustics, Speech and Signal Processing, Vol. 37, 984995,
Jul. 1989.
39. Blu, T., P.-L. Dragotti, M. Vetterli, P. Marziliano, and L. Coulot,
Sparse sampling of signal innovations, IEEE Signal Processing
Magazine, 3140, Mar. 2008.
40. Xing, S. Q., D. H. Dai, X. S. Wang, and T. Wang, Twodimensional ESPRIT super-resolution feature extraction using
fully polarized measurements and its performance analysis, Acta
Electronica Sinica, Vo1. 37, No. 12, 26812687, 2009.
41. Guo, D., H. Xu, and J. Li, Extended wavenumber domain
algorithm for highly squinted sliding spotlight SAR data
processing, Progress In Electromagnetics Research, Vol. 114, 17
32, 2011.
42. Cafforio, C., C. Prati, and F. Rocca, Full resolution focusing
of SEASAT SAR images in the frequency-wave number domain,
Proc. 8th EARSel Workshop, 336355, May 1988.
43. Bamler, R., A comparison of range-doppler and wavenumber
domain SAR focusing algorithms, IEEE Trans. on Geoscience
and Remote Sensing, Vol. 30, No. 4, 706713, Jul. 1992.
44. Jafarpour, S., W. Xu, B. Hassibi, and R. Calderbank, Efficient
and robust compressed sensing using optimized expander graphs,
IEEE Trans. on Information Theory, Vol. 55, No. 9, 42994308,
Sep. 2009.
45. Cetin, M., W. C. Karl, and D. A. Castanon, Feature enhancement and ATR performance using nonquadratic optimizationbased SAR imaging, IEEE Trans. on Aerospace and Electronic
Systems, Vol. 39, No. 4, 13751395, Oct. 2003.