IEEE Geoscience and Remote Sensing Magazine - March 2020 - 48

motion on the rail track. The deformation estimated by
GB-DInSAR is actually the projection component of the
3D deformation of the scene in the direction of the radar LOS. Through repeated monitoring for a long period
of time, the deformation trend of the observation area is
predicted by the change of the LOS component within the
observation time.
GB-DInSAR IMAGING ALGORITHM
Different monitoring scenarios have different imaging
characteristics, so the imaging algorithms must be properly
selected. There are three main types of imaging algorithms:
time-domain algorithms, range-Doppler-domain algorithms (RDAs), and frequency-domain algorithms, as presented in Figure 8. All three types of algorithms are derived
from airborne or spaceborne SAR imaging algorithms, but
there are several differences. In the linear scanning mode,
the length of the synthetic aperture is limited by the rail
track. It is therefore impossible to form a full synthetic aperture. In Doppler domain processing, the number of azimuth sampling points must be increased by zero padding
to avoid an incomplete display of the image in azimuth.
Usually, the frequency-domain algorithm needs to deal
with this problem. In the arc scanning mode, unlike the
synthetic aperture in airborne or spaceborne SAR, GB-DInSAR actually forms an angular synthetic aperture. In addition, GB-DInSAR usually does not need to consider motion
compensation issues.
TIME-DOMAIN ALGORITHMS
A time-domain algorithm is represented by the backprojection algorithm (BPA). Its basic principle is to coherently accumulate the range-compressed echo signal along the range

Time Domain
Range
Compression

Azimuth-Coherent
Accumulation

Range
Compression
RCMC

Azimuth FT

Azimuth
Compression

Azimuth
Inverse FT

Frequency Domain
2D FT

Complex Image

GB-SAR Echo Data

Range-Doppler Domain

Bulk
Compression

Stolt
Interpolation

2D Inverse
FT

FIGURE 8. A block diagram classification of imaging algorithms.

migration. Because the BPA calculates the two-way time delays accurately due to the pixel-by-pixel strategy, this kind
of algorithm can deal with various types of complex imaging geometry perfectly, even for 3D imaging [39]. Because
no geometric approximation is made during the imaging
process, the BPA is accurate for the whole scene; however,
the BPA needs much more computation because every aperture position must be examined for all image pixels [40].
BPAs can also be modified to fit the arc scanning
mode in polar coordinate [41]. Furthermore, the multicore parallel processing technique can be utilized for
polar-coordinate BPAs to solve the real-time imaging
problem [42].
RANGE-DOPPLER-DOMAIN ALGORITHMS
RDAs perform a range-cell migration correction (RCMC) to
deal with the coupling problem between range and azimuth
[43]. RDAs can be modified to fit the arc scanning mode
in polar coordinate with bulk-range migration correction
[44]. During the imaging process for RDAs, the slant range
is approximated by a Taylor series expansion. As a result,
theoretically, RDAs have a slant-range error comparable to
that of BPAs, which may affect the azimuth compression in
the azimuth matched-filtering process.
When the range curvature is less than d r /4, where d r is
the range resolution, the Keystone transform can be utilized
to realize a RCMC. Compared with the BPA, the calculation
burden of the Keystone transform is quite small [46].
FREQUENCY-DOMAIN ALGORITHMS
The Omega-k algorithm (~KA) is a classic frequencydomain algorithm. The bulk compression and residual
phase-error correction of the ~KA are performed in the
2D frequency domain. Except for the principle of stationary phase, the ~KA has no approximation. Thus, it is
highly accurate and so suggested for near-field imaging.
The drawbacks of the ~KA include the computational
burden in Stolt interpolation [47] and azimuth zero-padding operation to avoid spectral overlap, which consumes
a considerable amount of data [48].
The far-field pseudopolar format algorithm (FPFA) is a
type of frequency-domain algorithm. The focused results
are represented by pseudopolar coordinates. The FPFA
adopts slant-range approximation; therefore, the application scope of the FPFA is limited for far-field imaging [49].
For the imaging problems associated with GB arc-array
(AA) SAR systems, the GB-AA-SAR wavenumber domain
algorithm is proposed for processing ArcSAR data in a polarcoordinate format [50]. It can be applied to wide--a zimuth
viewing scenes and can successfully solve the problem of
azimuth mismatch caused by range approximation during
the imaging process of arc geometry [51].
DIFFERENTIAL INTERFEROMETRY PROCESSING
GB-DInSAR deformation measurement involves the
continuous observation of the same ground target to
IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE

MARCH 2020

IEEE Geoscience and Remote Sensing Magazine - March 2020

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - March 2020