The Hybrid Cram
´
er-Rao Bounds on Elevation in
Compressive Sensing SAR Tomography
Bo Yang
∗
, Huaping Xu
∗
, Yanan You
∗
and Xiaozhen Xie
†
∗
School of Electronic and Information Engineering
Beihang University, Beijing, China
Email: yangb89@buaa.edu.cn, xuhuaping@buaa.edu.cn, yyngc@126.com
†
College of Science, Northwest A&F University, Yangling, Shaanxi
Email: xiexzh@nwafu.edu.cn
Abstract—Tomographic Synthetic aperture radar (Tomo-SAR)
as an advanced 3-D imaging technique has been achieved great
interest and development recently, especially the successful ap-
plication of compressive sensing (CS) in this field. For simplicity,
additive Gaussian noise has been often assumed to be identical
independent distribution (i.i.d) to obtain the Cram
´
er-Rao lower
bound (CRLB) on elevation estimation. In this work, a realistic
hybrid Cram
´
er-Rao bound (HCRB) on elevation estimation
is derived for CS Tomo-SAR in presence of the correlated
noise. Concretely, a small random parameter representing the
correlation of noises is introduced, which complies with the
normalized Laplacian distribution in range of (−1, 1). Then
the probability density function of observations conditioned the
unknown deterministic sources parameter are correlated with
a small nuisance parameter in non-diagonal elements of the
noise covariance matrix. Computer simulations are performed
for analyzing the proposed HCRB of elevation. Moreover, the
root mean squares errors of two estimators based on CS have
validated that the proposed HCRB approximates the realistic
radar scenario better than CRLB.
Keywords—Synthetic aperture radar (SAR), SAR tomography
(Tomo-SAR), Cram
´
er-Rao lower bound (CRLB), hybrid Cram
´
er-
Rao bound (HCRB), correlated noise, elevation accuracy.
I. INTRODUCTION
Tomographic Synthetic aperture radar (Tomo-SAR), the
extension of SAR interferometry (InSAR), is an advanced
imaging technique that multiple views are utilized to map the
scattering power at different heights, thus enabling the 3-D
imaging of the scenes. Since Tomo-SAR was first demon-
strated the capability in height resolution by airborne system
([1]), this technique has gained great interest and development
recently in the field of spaceborne multibaseline InSAR, espe-
cially for the scenes with steep topography or a high spatial
density ([2], [3], [4]). With the idea that 2-D SAR image pixel
is actually the projection of the 3-D scene scattering along
the elevation direction onto the 2-D azimuth-range plane,
Fornaro et al. ([4], [2]) first validated the capability in terms
of spaceborne and formulated the layover problem as a linear
inverse problem about the amplitudes, phases and elevations
of the targets.
Beginning with the Tomo-SAR data model, a number of
estimators have been proposed to achieve the 3-D reconstruc-
tion with multipass SAR data over the same scene (see [5], [6],
[7], [8], [9], [10] for reference). Owe to the scattering number
in a generic pixel is sparse compared with the discretization
level of semidicrete operator, the Tomo-SAR estimators based
on the theory of compressive sensing (CS) has been widely
proposed and yielded great success in super-resolution of
elevation ([5], [6], [8]).
The lower bound of elevation estimation is necessary to
predict the performance and assess the accuracy in elevation
estimation. Based on the assumption that the probability den-
sity function (pdf) of observations conditioned the unknown
deterministic sources parameter are identical independent dis-
tribution (i.i.d), Zhu et al. [11] proposed the lower elevation
bound of the sources and analysed the influence of a few
parameters by the theory of Cram
´
er-Rao lower bound (CRLB)
for CS Tomo-SAR estimators. Considering that the noises in
one stack containing background clutter and thermal noises
are independent from each other, the CRLB of unknown
parameters has been easily derived by referring the calculation
of [12] and [13].
Nevertheless, in most cases, especially for the urban areas
where there are many artificial structures varying in multi-pass
acquisitions, the noises are correlated [14]. In this work, the
estimation accuracy on elevation is deduced for CS Tomo-
SAR with correlated noise. A small random parameter on
behalf of correlation coefficient of noises is introduced, which
complies with the normalized sparse distribution in range of
(−1, 1). Then the noises containing background clutter and
thermal noises are correlated with this nuisance parameter. We
have derived the calculated formulas of the hybrid Cram
´
er-
Rao bound (HCRB) in presence of the correlated noise, and
demonstrated the proposed bound is more general than CRLB
results presented in [11], which is the special case of ours. At
last, some curves are drawn to analyze the derived HCRB, and
simulation shows the estimating precision approaches HCRB
more than the CRLB.
The rest of this conference is organized as follows. Section
II reviews the Tomo-SAR model and introduce the nuisance
parameter about the correlated noise. In Section III, the process
of calculating the HCRB of elevation accuracy is presented.
With the parameter of Terra-SAR system, the derived bounds
are analysed and confirmed with two estimators in Section IV.
Finally, the conclusions are drawn in Section V.
4th Int. Workshop on Compressed Sensing on Radar, Sonar, and Remote Sensing (CoSeRa 2016)
128