NDT&E International, Vol. 31, No. 6, pp.

439–444, 1998
䉷 1998 Elsevier Science Ltd. All rights reserved
Printed in Great Britain
PII: S0963-8695(98)00043-7 0963-8695/98 $19.00+0.00

Comparison of laboratory and

simulated data for radar image
interpretation
C. Colla, C. D. Burnside, M. R. Clark, K. J. Broughton,
M. C. Forde*
University of Edinburgh, Department of Civil and Environmental Engineering,
Kings University Buildings, Edinburgh EH9 3JN, UK

It has been shown that the output from in-house ray-tracing simulation software
provides a means by which the impulse radar response of a structural system
can readily be simulated. Radar simulations were undertaken of re-bar targets in
a tank. A 900 MHz commercial antenna was simulated. The simulations
undertaken have shown the effects that target spacing and depth have on the
radar signature shapes and the results compare well with real-case survey plots.
Simulation data have been used to assess the output specification from a
commercial antenna, with interesting findings. It is expected that simulation will
allow the interpretation of actual surveys to be more readily understood, which
will in turn allow a better engineering assessment to be made of a structure
under investigation. 䉷 1998 Elsevier Science Ltd. All rights reserved.

Keywords: radar, simulation, electromagnetic

Introduction problem, the conditions of the environment where the

The research work on the simulation of radar response
presented herein is part of a wider study focused on the
investigation of composite masonry structures and masonry
arch bridges conducted at the University of Edinburgh,
where the radar technique is one of the non-destructive
methods being researched for the investigation of structures.
Ground-penetrating radar (GPR) has been used for over 30
years for resolving shallow subsurfaces in geophysical
applications and more recently in engineering. The develop-
ment of the method, especially in civil engineering for the
investigation of structures, has been restricted because of
the complexity of the systems and the need for specialist
interpretation of the data[1].
The user of a radar system is required to have knowledge of
the concepts of physics (electromagnetic properties of
materials, wave propagation and response of the structure)
and electronics (pulse emitted, operating antennae),
together with an engineering understanding of the problem
being investigated. Depending on the complexity of the
structural element under investigation, the nature of the
survey is carried out, and the antennae configuration used
during data acquisition, the radar user may or may not be
able to interpret the raw data. Depending upon his
interpretative experience, he may need to use sophisticated
signal-processing software, such as short pass, long pass,
horizontal or vertical filtering, plus signal amplification[2].
In addition, the user must be aware of the limitations of the
technique and have realistic expectations. Radar cannot
penetrate good electrical conductors, such as saltwater[3],
metal or fine reinforcing mesh[4] or wet clays. Features may
appear on the plots and, whilst some of the more common
ones are recognisable and interpretable by the practised
engineer, others are generated by complex effects of
multiple reflections and/or refractions. Not only can their
interpretation prove difficult, but they can mask valuable
information generated by true anomalies. Other sources of
data interferences include periodic radio-frequency noise
introduced by site conditions, distortion at data acquisition,
antenna ringing, poor antenna coupling and effects intro-
duced by the structure itself because of its complexity or
material grading such as clutter[5]. Each of the above
requires special digital signal processing with the risk of
degrading the data if manipulation is improperly applied.

* Correspondence to Professor M. C. Forde. E-mail: m.forde In complex cases it may be more convenient, in terms of
@ed.ac.uk time and effectiveness, to model the site situation and

439
 C. Colla et al.

predict the radar performance and structure response.
Therefore there is a need for a method of simulation which
aids the interpretation of GPR surveys by allowing rapid
simulation of varying geometries and materials. The combi-
nation of digital GPR system along with simulation software
will help a better engineering assessment to be made[6].
Radar principles and theoretical
background
where Z 1 is the impedance of the first material, Z 2 is the
impedance of the second material and, for low-loss
dielectrics, the impedance Z is given by:
ÿ
0=5
Z ¼ m0 =e0 er
where m 0 is the magnetic permeability of free space and e 0 is
the permittivity of free space.
In the case of oblique incidence they will be:

Z2 cos vi ¹ Z1 cos vt
The radar technique involves the pulsing of radar waves into Perpendicular polarisation : r¼
soil and/or construction materials. The propagation of the Z2 cos vi þ Z1 cos vt
signal is affected by the dielectric properties of the 2Z2 cos vi
t¼
materials, so that its attenuation and reflected components Z2 cos vi þ Z1 cos vt
vary accordingly. An examination of the reflected radar
waveforms enables an interpretation of the material and/or
structure under investigation. Z2 cos vt ¹ Z1 cos vi
Parallel polarisation : r¼
Z2 cos vt þ Z1 cos vi
The signal attenuation is given by:
2Z2 cos vi
 1=2
t¼
me⬘p Z2 cos vt þ Z1 cos vi
a¼q 1 þ tan2 d ¹ 1
2

where:
Radar simulation
a attenuation coefficient
q ¼ 2pf angular frequency
In order to aid the analysis of site data in terms of
m ¼ m rm 0 magnetic permeability
e⬘ ¼ e re 0 real part of dielectric engineering assessment, an initial simulation of relatively
permittivity simple geometries and material electromagnetic properties
j ¼ e⬙q conductivity was undertaken. In a second stage, more complex
tan d ¼ e⬙/e⬘ ¼ j/qe⬘ loss tangent (or dissipation engineering structures were simulated, including single-
factor) span masonry arch bridges. Results are presented herein for
re-bar detection in homogeneous media and are compared
The dielectric permittivity of a material is a complex
with experimental data from the laboratory.
function:
The modelling software uses a conventional ray-tracing
e ¼ e0 ¹ je00 process to model sinusoidal trace propagation along angles
that cover the emitting field of commercially available
where e⬙, commonly referred to as the loss factor, is antennas, and to recursively detect interfaces between
dependent on both frequency and conductivity. various subsurface features (strata and anomalies) and
The amplitude reflection, r, and transmission, t, coefficients determine their boundaries. The algorithm reproduces the
for the electromagnetic waves are: paths of individual radar wavelets from radar emitter to
receiver through the various interfaces. At each interface,
Z2 ¹ Z1 2Z2 because of the different electromagnetic characteristics of
r¼ and t¼
Z2 þ Z1 Z2 þ Z1 the two materials present, the wavelet splits into different
components and is partially transmitted through the second
material and partially reflected in the receiver direction
(Figure 1). The responses from all of the traces sent out at
different angles from the transmitter are then combined
according to the Principle of Superposition to produce a
complete, one-dimensional simulated image for each
sample point constituting the received signal waveform. In
turn, these are combined to obtain a complete two-
dimensional scan over the simulated area.
To reduce the time taken for the model to be calculated,
various termination conditions have been included. The
signal propagation arrests if:
Figure 1 A diagram showing the reflective index • the chosen two-way time limit has been reached;

440
 Radar image interpretation
Figure 2 The laboratory rig
Figure 3 (a) and (b) Comparison of laboratory and simulation data for interpretation of radar images
• the chosen number of simulated transmissions and reflec-
tions has been exceeded; or
• the signal has entered a ‘convoluted’ interface and, as a
result of the wavelet diffraction resulting from the differ-
ence in material properties, it has become trapped below
the interface.
Input to the modelling program takes the form of a text file
where geometrical dimensions, location and electro-
magnetic properties of embedded target objects and material
strata are expressed in terms of their coordinates, permit-
tivity and conductivity. The characteristics of the radar
antenna are expressed in the form of centre frequency. Other
information specific to the simulation process is also
entered, including the maximum two-way travel time in
ns, the aperture of the receiver, and the distance to be
covered by the antenna during the survey.
The main advantage of ray-tracing is mathematical
accuracy—each ray from the source is modelled with
considerable precision, allowing for more accurate place-
ment than comparable finite element processes. The
weakness normally attributed to the process—the inability
to model with accuracy the diffraction effect of the wavelet
being transmitted through interfaces—has been dealt with
in this model by the use of the equations listed in the theory
section.
Antenna and signal characteristics
The emitter frequency is set as the centre frequency for the
range being considered. Frequencies considered ranged
441
from 200 to 900 MHz. The source wavelet is assumed to be
three-phase sinusoidal in format. Absolute signal strength is
calculated in relation to the angle from the normal to the
radar emitter. The directional response function is used as
described in [7].
Target specifications
Embedded object geometrical shapes included circles and
polygons (triangles and rectangles) and were described in
terms of their equations and main electromagnetic proper-
ties (permittivity and conductivity).
Simulation specifications
Other information specific to the simulation process and
entered in the modelling input file included the following.
• The step size in angle degrees between each ray trace:
although a full beam with more than 20 ray traces per
degree would be required for complete accuracy, scans
with one to five traces per degree would provide the user
with a reasonable target image for training purposes.
• The number of scans per metre: scan number at the
density used by digital radar systems (1024 scans m ¹1
or less) could be employed, but fewer scans m ¹1 may
be convenient to obtain faster results. Too low a
number of scans m ¹1 may cause image deformation.
• The number of reflection patterns to be recorded: this
allows signal and image decomposition, so that it
becomes possible to recognise the effects of primary/
secondary reflection.
 C. Colla et al.

Figure 4 (a) and (b) Comparison of laboratory and simulation data for interpretation of radar images

Figure 5 (a) and (b) Comparison of laboratory and simulation data for interpretation of radar images

The output of the analysis is given in the same format as the
digital radar system data files, so that it can be viewed with
the same image processing software.
Laboratory and simulation set-ups
The experimental tests were conducted in the laboratory on
a rig which allowed the position and spacing of two
reflectors (3.5 cm in diameter) to be varied, as per the plan
layout in Figure 2. The distance of the reflectors from the
antenna varied from 5 to 40 cm, and the relative spacing
between the two targets varied from 10 to 40 cm. The
background medium during the tests was water and the
antenna used was a 900 MHz bow tie with a survey wheel.
Simulated models reproduced these same conditions.
Comparison of simulated and experimen-
tal results
Consider Figures 3a and 3b: the first is the laboratory plot
obtained with 20 cm reflector spacing and 20 cm depth; the
second is the output of the corresponding simulation.
In Figures 4a and 4b comparison is shown for 10 cm
separation of reflectors and 20 cm depth: laboratory data
and simulation.
In Figures 5a and 5b comparison is shown for 40 cm
separation of reflectors and 40 cm depth: laboratory data
and simulation.
Practical applications of simulation
The applications of a simulated radar survey can be listed
as: a training tool, a pre-survey forecast, post-data collection
analysis aid and system tester for evaluating the perfor-
mance of digital equipment.
Training
Ray-tracing is a more intuitive approach than alternative
processes such as FTFD (finite time/finite difference)[8].
The variables listed above are derived easily and apply in
practice just as readily as to the simulation. Thus the
simulation approach could be used for the training of new
users and for further in-depth training—provided that the
limitations of ray-tracing are clearly understood. Since the

442
 Radar image interpretation
Figure 6 (a), (b) and (c) Comparison of laboratory and
simulation data for interpretation of radar images
simulation produces less scattered images than digital radar
systems produce on site, simulated radar plots offer the
advantage of being more easily understood and interpreted.
Moreover, one of the special features of the simulation is
signal decomposition: this allows signal propagation to be
interrupted after one or more reflections, avoiding effects
such as ringing or multiples and keeping the output plot
simpler.
Consider Figures 4a and 6a, which compare laboratory
experimental data with the simulated image of two re-bars at
20 cm depth and 10 cm spacing. In Figures 6a, 6b and 6c a
443
longer time window has been used than in Figures 4a and
4b, for the purpose of showing multiple reflections.
Consider first Figure 6a. Lines 1a and 1b are the primary
reflections. Lines 2a and 2b and 3a and 3b are waves
bouncing backwards and forwards between the re-bars.
Whilst these data are achieveable mathematically using the
program, they are not attainable practically in a real-life
experiment, as can be seen from the experimental data in
Figure 4a.
If, in the case of Figure 6a, signal propagation were stopped
at the first reflection, then Figure 6c would have been
obtained. If signal propagation were continued beyond
Figure 6c to include the computer-generated signals
bouncing back and forth between the re-bars, then Figure
6b would have resulted.
The secondary reflection seen in Figure 6a, and marked up
as lines 4a and 4b, is not visible in the experimental data of
Figure 4a—as the time range setting in Figure 4a is too
short. Instead the back of the tank is visible.
Expert systems
An accurate simulation can be used to provide an
interpretation of the radar image obtained from site surveys.
Although this work is still some way from an iterative
process where radar images can be back-analysed to
produce the real site image, it is possible to compare site
data with simulated models to try a ‘best comparison’
interpretation. Similarly, the simulation may be used as part
of the preparation before or during a site visit to give an
understanding of the results obtained. The simulation
program developed at the University of Edinburgh is PC-
based; it can be run and results plotted in a matter of
minutes—making it suitable for site use. With the further
usage of image-analysis tools, a simulated survey plot could
be used to provide a template for a scan from the field to
allow the engineer to observe key differences and spot any
anomaly not yet accounted for.
Equipment evaluation
If an idealised radar image can be derived from a simulation,
then simulations could produce ‘ideal’ images of real case
surveys to be used as benchmarks for the testing of radar
equipment.
For example, specifications of radar antennae may state
that the signal will be emitted at certain angles (depending
on the type of antenna, the signal-beam width may be
approximately 60⬚ or up to 90⬚ angle). It is possible to
compare experimental data with simulated data for an
evaluation of the beam angle of commercially available
antennas.
Figures 7a and 7b simulate two re-bars at 5 cm depth and
10 cm spacing. Figure 7a simulates an emitted beam angle
of 60⬚ whilst in Figure 7b the angle is 89⬚. Figure 7c is the
corresponding laboratory experimental image obtained with
a commercial antenna. It is therefore deduced that the beam
 C. Colla et al.

• The simulations undertaken have shown the effects that

target spacing and depth have on the radar signature
shapes.
• The limitations of ray-tracing techniques have been
identified and highlighted: additional reflections due to
rays bouncing backwards and forwards between the re-
bars. This additional ‘ringing’ would not occur with a
broader fronted radar wave in practice.
• Simulation results compare well with the experimental
survey plots.
• Simulation data have been used to assess the output spe-
cification from a commercial antenna.
• It is expected that simulation will allow the interpretation
of actual surveys to be more readily understood, which
will in turn allow a better engineering assessment to be
made of a structure under investigation.

Acknowledgements
The authors wish to acknowledge the facilities made
available by the University of Edinburgh and the work of
Dr I. J. Padaratz, who was involved in the initial phase
of the computer simulation. The authors gratefully acknowl-
edge valuable discussions with Dr Parag Das and the
financial support provided by the Highways Agency,
London, UK.

References
Figure 7 (a), (b) and (c) Comparison of laboratory and simula-
tion data for interpretation of radar images
angle of the commercial antenna is actually of the order of
90⬚ rather than the 60⬚ claimed by the manufacturers.
Conclusions
• It has been shown that ray-tracing simulation provides a
technique by which the impulse radar response of a
simple structural system can readily be simulated.
1 Forde, M. C. and McCavitt, N. Impulse radar testing of structures.
Proc. Inst. Civ. Engrs Struct. Bldg. 1993, 96–99.
2 Olhoeft, G. R. Processing, modelling and presentation of ground
penetrating radar data. In Proc. 2nd Government Workshop GPR—
Advanced Ground Penetrating Radar: Technologies and Applications,
p. 23. Ohio, 1993.
3 Colla, C., Forde, M. C., McCann, D. M. and Das, P. C. Laboratory
modelling of radar propagation through masonry. In Proc. IV Int.
Conf. Non-Destructive Testing in Civil Engineering (NDT-CE ’97),
vol. 1, pp. 303–316. Liverpool, UK, 1997.
4 Padaratz, I. J. and Forde, M. C., A theoretical evaluation of impulse
radar wave propagation through concrete. J. Non-destr. Test. Eval.,
1995, 12, 9–32.
5 Padaratz, I. J. and Forde, M. C. Influence of antenna frequency on
impulse radar surveys of concrete structures. In Proc. Int. Conf. Struc-
tural Faults and Repair—95, vol. 2, Engineering Technics Press,
London, 1995, pp. 331–336.
6 McCavitt, N. and Forde, M. C., The application of the method of
convolution to the simulation of the response of masonry arch bridges
to ground probing radar. J. Non-Destr. Test. Eval., 1991, 6(3), 179–
194.
7 Goodman, D. and Nishimura, Y. 2-D synthetic radargrams for
archaeological investigation. In Proc. 4th Int. Conf. Ground Penetrat-
ing Radar. Finland Geological Survey, 1992.
8 Nelson, S. D. Electromagnetic modelling for ground-penetrating ima-
ging radar (GPIR) using 3-D finite difference time-domain (FDTD)
modelling codes. In Advanced Microwave and Millimeter-Wave
Detectors, vol. 2275, SPIE, San Diego, CA, 1994, pp. 186–195.

444