Time Domain SAR Raw Data Simulation using CST and

Image Focusing of 3-D Objects

Adnan Saeeda and Olaf Hellwicha
a
Computer Vision and Remote Sensing, Technical University Berlin, Berlin, Germany

ABSTRACT
This paper presents the use of a general purpose electromagnetic simulator, CST, to simulate realistic synthetic
aperture radar (SAR) raw data of three-dimensional objects. Raw data is later focused in MATLAB using
range-doppler algorithm. Within CST Microwave Studio a replica of TerraSAR-X chirp signal is incident upon
a modeled Corner Reflector (CR) whose design and material properties are identical to that of the real one.
Defining mesh and other appropriate settings reflected wave is measured at several distant points within a line
parallel to the viewing direction. This is analogous to an array antenna and is synthesized to create a long
aperture for SAR processing. The time domain solver in CST is based on the solution of differential form of
Maxwells equations. Exported data from CST is arranged into a 2-d matrix of axis range and azimuth. Hilbert
transform is applied to convert the real signal to complex data with phase information. Range compression,
range cell migration correction (RCMC), and azimuth compression are applied in time domain to obtain the
final SAR image. This simulation can provide valuable information to clarify which real world objects cause
images suitable for high accuracy identification in the SAR images.
Keywords: Raw Data, Range-Doppler Algorithm, SAR Image, CST, RCS, TerraSAR-X, Corner Reflector,
MATLAB

1. INTRODUCTION
Based on acquired/given object geometries SAR image data can be simulated for certain antenna positions.
Then electromagnetic surface characteristics as well as radiometric sensor characteristics must be considered.
Methods with large computational efforts allow to simulate the SLC signal pixel by pixel considering all
object details. For this purpose, we will make use of the CST software of Computer Simulation Technology AG,
Darmstadt. The software will allow to model scatterers of different materials arranged in variable geometries.
Relevant categorization in SAR simulation is made by Giorgio at el.1 introducing ’SAR Image simulator’ and
’SAR raw data simulator’. In a SAR image simulator, the object of simulation is the image which is generated by
appropriate input data with no availability of any intermediate product. SAR raw data simulators are developed
to provide test data for processing algorithms or radiometric analysis. Another classification about evaluation of
backscattering is ’point’ and ’extended scene’ simulator. Point simulator is focused on system parameters and
might rely upon an artificial scene. Extended scene simulator needs a more sophisticated backscattering model.
Image simulators are applied for the purpose of providing a priori knowledge about man-made objects. For
instance, ray tracing methods are applied in Wohler et al. (1980)2 simulating SAR image signature of airplanes.
The main motive was to provide the test data for automated object identification. This simulation was limited to
direct backscattering only. Recently, Balz et al.3 approximated the double reflection in his simulation approach.
Brunner et al.4 presented a ray tracing approach for simulating appearance of basic building shapes.
RaySAR (Auer et al.5 ) is a 3-d SAR simulator based on ray tracing methods, implying an enhanced version
of the open-source software POV-Ray for simulating radar signals in 3-d, i.e. azimuth, range, and elevation. As a
result, the simulated signal contributions can be compared to object geometries reconstructed by interferometric
SAR methods. The focus of the simulation concept is on local urban scenes imaged by very high-resolution
Further author information: (Send correspondence to Adnan Saeed)
Adnan Saeed: E-mail: [email protected]
Olaf Hellwich: E-mail: [email protected]

Image and Signal Processing for Remote Sensing XXIII, edited by Lorenzo Bruzzone, Francesca Bovolo
Proc. of SPIE, Vol. 10427, 1042716 · © 2017 SPIE · CCC code: 0277-786X/17/$18 · doi: 10.1117/12.2281186

Proc. of SPIE Vol. 10427 1042716-1

(VHR) SAR sensors and on the geometric correctness of simulated signals. Random scattering is neglected as
the main goal is to identify and understand deterministic reflection effects occurring at individual man-made
objects. A combination of two basic reflection models is used to simulate the spatial distribution of diffuse and
specular reflection signals. RaySAR provides images in two steps. First, the modeled scene is sampled by an
enhanced version of the POV-Ray ray tracer. Thereafter, the detected signal contributions are imposed with a
pixel grid and summed in order to derive the final image.
CohRaS (Coherent Raytracing SAR Simulator),6 is a SAR simulator based on ray tracing methods. The ray
tracer itself is based on the concept developed by Amananatides and Woo.7 This simulator is intended mainly
for the simulation of small scenes with high resolution for creating training data for classifiers and sample data
for the training of image analysts. The only restriction imposed on the geometry of the 3D model, is that all
polygons need to be convex. The focus is set on the fast calculation of many small images, as this is needed
for classification. SARViz [15] is a real-time SAR image simulation system. The preface real-time is used in the
computer graphics sense, which, in our case, means more than 20 simulations or frames per second. The size
of each frame is typically 1024x768 pixels. SARViz uses the rasterization approach implemented on Graphics
Processing Units (GPU), which allows very fast simulations, but has certain limitations regarding the geometric
and the radiometric accuracy. For example, rasterization does not allow simulating multiple bounces, because
the path of the waves throughout the scene is not known.
Simulators for providing raw data are based on physical models describing reflection phenomena. Speck et
al.8 presented a raw data simulator based on separate models for 3-d objects such as buildings or vehicles and
background.
The use of a General-Purpose EM simulator for simulating SAR is rare. Some of the basic experiments on
the generic EM simulator FEKO, to test its suitability for the SAR simulation was done by Amit and Bernard
et al.9 Their focus was towards generating a synthetic aperture radar (SAR) image database of ground targets,
using a generic EM simulator.
The paper is divided into three more section as follows: Section 2 details the CST simulation workflow and
setting used to setup the simulation. Section 3 gives an insight about the post processing of data via Matlab.
Last section conclusions are drawn.
Begin the Introduction below the Keywords. The manuscript should not have headers, footers, or page
numbers. It should be in a one-column format. References are often noted in the text and cited at the end of
the paper.

2. SIMULATION
CST Microwave Studio
CST microwave studio is a part of CST studio suite used in this project, which is a package for electromagnetic
design and analysis in high frequency range. A powerful graphical solid modelling frontend simplifies the process
of creating the structure to be simulated. After the model is constructed a fully automated meshing is applied
before the start of simulation engine. Software also contains several different simulation techniques to best suit
various applications.

Time Domain Solver

Time domain solver offers the largest flexibility which can obtain the entire broadband frequency behavior from
one single calculation run. The time domain solver is less efficient for structures that are electrically much smaller
than the shortest wavelength, which is not the case in simulating a corner reflector here used.

Workflow
The simulation workflow can be overviewed via following sequence: after defining working units, background
material, the structure is modelled, frequency range is defined, ports, boundary and symmetry conditions are
set, mesh is visualized, plane wave exited by a chirp signal is incident on the scene and reflected field data (time
domain) is measured by several equidistant far-field probes placed parallel to the object.

Proc. of SPIE Vol. 10427 1042716-2

Units
Millimeters, Kelvin and Gigahertz are units set for dimensions, temperature and frequency respectively. When a
unit is changed all expressions including this unit are multiplied with the change factor e.g. changing frequency
from Hz to MHz all frequency expressions will be multiplied by 1x106.

Material Properties
Basic available materials are PEC (perfect electrically conducting material) and vacuum. Lossy material Alu-
minum selected from materials library is used. For linear behavior dielectric and magnetic material properties
determine the ratio of electric field and electric flux density. Three right angular Aluminum sheets of side lengths
1000mm and thickness 2mm are aligned at right angle resulting in a corner reflector.

Boundary Conditions
Several boundary conditions are available each applicable to any face of the bounding box, electric boundary
operates like a perfect electric conductor where the tangential components of the E-field s and normal components
of the magnetic flux are zero. This means that the fields are normal to the boundary and magnetic fluxes are
parallel to the boundary. The opposite case applies for a magnetic boundary where tangential components of
magnetic fluxes and normal components of electric fields are zero. Open boundary operates like a free space
open ( add space) boundary is used which adds some extra space (vacuum) between the bounding box and open
boundary, waves can pass this boundary with minimal reflections. Designed model is oriented in a way that
symmetry planes could be defined. One electric field symmetry plane is defined in this case which reduces the
calculation time by a factor of two.

Excitation Signal
Plane wave is defined as an excitation signal, where the simulation amplitude in v/m (volts per meter) is recorded.
Instead of using default signal to excite the plane wave, a chirp of duration 10 ns is imported to CST. This chirp
is the replica of the signal used in TSX. Moreover, the chirp is filtered by a hamming window filter, which makes
it gradually increasing and then decreasing signal also suppressing sidelobes. Such a gradual change in the signal
almost completely removes noises and artifacts that are generated by sudden changes.

Eu0e50rtC0iip Hemming
104005
-CliqNmrg
A0000

90000

40000

30000

30000

40000

au*
.90000

-11+0050
10

Figure 1. Simulated chirp signal of 10 nsec as seen in CST after Hamming window filter.

Figure 2. Plane wave excited by chirp from one boundary side to be incident on the modelled reflector placed at origin.

Proc. of SPIE Vol. 10427 1042716-3

E-Field Probe
Probes are used to record one particular component of electric field in a specified location during the transient
analysis. For the time domain solver, the result of probe is the time signal. Having several equidistant probes
to observe the scattered signal is analogues to synthetic array which imitates the motion of SAR sensor. Each
probe contains the phase shifted signal compared to the adjacent probe. The phase shit being dependent on the
distance between the probes can be adjusted as per requirement.

Figure 3. Several E field probes to measure transmitted and reflected signal as modelled in CST.

3. SAR PROCESSING
Basic Range Doppler Algorithm (RDA) suited for the processing of relatively low squint angles is used, discussed
in detail by Cumming and Wong.10 Range compression can be performed with fast convolution when the data
are in azimuth time domain. A range Fast Fourier Transform (FFT) is performed followed by a range matched
filter multiply and then the inverse transform is applied. The fact that current RAW data is not very large
having convolution in time domain is also not very time consuming and is preferred to avoid FFT artifacts.
Azimuth FFT transforms the data into range doppler domain, range cell migration which is range time and
azimuth frequency dependent is performed in range-doppler domain. The targets having same slant range but
separated in azimuth time all have same doppler history. By virtue of Fourier transform shift theorem the energy
of all these targets will be positioned exactly in the same frequency samples, with each target having a different
linear phase component. In short targets having same range of closet approach transforms into one trajectory
in RD domain. This property is exploited in RDA to gain computing efficiency.
RCMC is performed in range doppler domain which straightens out the curved trajectories that now run
parallel to the azimuth frequency axis. There are two ways to implement RCMC, it can be performed by range
interpolation option in RD domain. The amount of RCM to be corrected is given by

λ2 R0 fn2
∇R(f n) = (1)
8V r2
Where Vr, Ro and fn are radar velocity, closed range of approach and azimuth velocity respectively. Another
implementation involves the assumption that RCM in range invariant over a finite range region. In such case
RCMC is implemented using FFT a linear phase multiply and then inverse transform. The phase multiplier for
a given frequency is given by

4π∇R(f n)
Grcmc = exp(j c ) (2)

The correction amount is held constant for small range blocks

Proc. of SPIE Vol. 10427 1042716-4

 Raw Data

20

40

L
E 60
'Ñ
a
BO

100

120
500 1000 1500 2000 2500 3000 3500 4000
Range

Figure 4. RAW-Data in 2-d matrix after export from CST plotted in Matlab

Range Compressed (Time Domain)

40

L
7
E 60
A
a
SO

100

120
1000 2000 3000 4000 5000 6000 7000
Range

Figure 5. Range compressed data in Matlab, increase in number of range bins resulted by convolution

Proc. of SPIE Vol. 10427 1042716-5

 RCMC Curve
120

100

BO

L
E 60
N

40

20

o
0 1000 2000 3000 4000 5000 6000 7000
Range

Figure 6. The amount of range shift as seen in Matlab.

After RCM -Correction

2000 3000 4000 5000 6000 7000

Range

Figure 7. After RCMC applied the scattered data is now parallel to Azimuth axis

After RCMC correction azimuth compression is conveniently applied as frequency domain matched filter at
each range gate. inverse transform results in compressed final image.

Proc. of SPIE Vol. 10427 1042716-6

 Final SAR Image

11:.1_111

,
n .c,jc

Figure 8. Final compressed point seen with azimuth scaled to maintain range azimuth image aspect ratio.

Final SAR Image (7X Zoom)

3400

3500

3600

3700
L
g 3800

3900

4000

4100

4200

4300
2900 3000 3100 3200 3300 3400 3500 3600 3700 3800
Range

Figure 9. Imaged magnified at the scatterers location resembling a cross signature typical to a corner reflector.

4. CONCLUSIONS AND FUTURE WORK

Use of a general purpose electromagnetic simulator, CST, to simulate realistic synthetic aperture radar (SAR)
raw data of three-dimensional objects was presented. A basic point scatterer was modelled simulated and
processed. All possible steps were taken into account and the parameters were kept closest to real TerraSAR-X
data. Results were precise and can be verified in the existing literature. Currently, simulating complex objects is
under progress. No computational optimization algorithms are considered yet but can be a part of future work.
This simulation can provide valuable information to clarify which real world objects cause images suitable for
high accuracy identification in the SAR images.

Proc. of SPIE Vol. 10427 1042716-7

 REFERENCES
[1] Franceschetti, G., Migliaccio, M., and Riccio, D., “The sar simulation: an overview,” in [Geoscience and
Remote Sensing Symposium, 1995. IGARSS’95.’Quantitative Remote Sensing for Science and Applications’,
International], 3, 2283–2285, IEEE (1995).
[2] Wohlers, M., Hsiao, S., Mendelsohn, J., and Gardner, G., “Computer simulation of synthetic aperture radar
images of three-dimensional objects,” IEEE Transactions on Aerospace and Electronic Systems (3), 258–271
(1980).
[3] Balz, T. and Stilla, U., “Hybrid gpu-based single-and double-bounce sar simulation,” IEEE Transactions
on Geoscience and Remote Sensing 47(10), 3519–3529 (2009).
[4] Brunner, D., Lemoine, G., Greidanus, H., and Bruzzone, L., “Radar imaging simulation for urban struc-
tures,” IEEE Geoscience and Remote Sensing Letters 8(1), 68–72 (2011).
[5] Auer, S., Hinz, S., and Bamler, R., “Ray-tracing simulation techniques for understanding high-resolution
sar images,” IEEE Transactions on Geoscience and Remote Sensing 48(3), 1445–1456 (2010).
[6] Hammer, H. and Schulz, K., “Coherent simulation of sar images,” (2009).
[7] Amanatides, J., Woo, A., et al., “A fast voxel traversal algorithm for ray tracing,” in [Eurographics], 87(3),
3–10 (1987).
[8] Speck, R., Hager, M., Garcia, M., and Süß, H., “An end-to-end-simulator for spaceborne sar-systems,” in
[4th European Conference on Synthetic Aperture Radar EUSAR 2002], 237–239, VDE Verlag GmbH (2002).
[9] Mishra, A. K. and Mulgrew, B., “Generation of sar image for real-life objects using general purpose em
simulators,” IETE Technical Review 26(1), 18–27 (2009).
[10] Cumming, I. G. and Wong, F. H., [Digital processing of synthetic aperture radar data: algorithms and
implementation], Artech House (2005).

Proc. of SPIE Vol. 10427 1042716-8