Received: 28 September 2022

DOI: 10.1049/mia2.12358

ORIGINAL RESEARCH
- -
Revised: 23 January 2023 Accepted: 9 March 2023

- IET Microwaves, Antennas & Propagation

System‐based specifications for better design of

ground‐penetrating radar antennas

Samuel Wagner | Ababil Hossain | Stephen Pancrazio | Anh‐Vu Pham

Microwave Microsystems Laboratory, University of Abstract

California Davis, Davis, California, USA
Currently, no publicly available comprehensive set of specifications exists for the design
of ground‐penetrating radar antennas. Specifications are presented and derived for the
Correspondence
Anh‐Vu Pham.
design of a ground‐penetrating radar antenna based on the antenna's interaction with the
Email: ahpham@ucdavis.edu rest of the radar system. Typically, ground‐penetrating radar antenna engineers are given
basic parameters such as gain and bandwidth to design an antenna. An antenna can be
Funding information designed to meet those specifications but may not perform well in a ground‐penetrating
Office of Naval Research, Grant/Award Number: radar imaging system. By considering the whole system, including the operation pa-
N00014‐17‐1‐2488
rameters, expected targets, and signal processing algorithms, we can constrain the less‐
commonly considered performance aspects of an antenna to perform more reliably in
a deployment. We propose quantitative specifications for the antenna's coupling,
beamwidth, boresight radar cross‐section, pulse integrity, residual ringing, and front‐to‐
back ratio. Meeting these antenna specifications will decrease the probability that a
deficiency in the antenna design impacts the radar's image output. Finally, a summarised
design equation table that will be helpful in the ground‐penetrating radar antenna
community at large is offered.

KEYWORDS
antenna theory, antennas, radar cross‐sections, radar transmitters, ultra wideband antennas, ultra wideband radar

1 | INTRODUCTION consider all at once. Lamensdorf [14] presented popular ways

to quantify the performance of generic ultra‐wideband an-
Typical antenna design for ground penetrating radar (GPR) tennas and extended the concept of pulse integrity. Lestari [15]
primarily focuses on optimising of antenna gain and imped- performs an analysis on adaptive wire dipoles and bowtie an-
ance bandwidth [1, 2]. For example, [3] uses a figure of merit tenna input impedance to reconfigure the antenna for different
based on gain, bandwidth, and size to compare various an- ground environments and targets. Daniels [16] focuses an in‐
tennas for GPR. Gain and bandwidth requirements are depth study on a monostatic pair's impulse response in
focussed on because they relate to the most fundamental terms of late‐time ringing and clutter reduction for landmine
specifications given to a system designer: power, frequency, targets in lossy soils and describes the impact of a slow ring-
and link budget [4, 5]. However, an antenna that satisfies these down on the system's performance. Schantz [17] provides
generic requirements is not guaranteed to perform well in a additional generalised time domain analysis on ultrawideband
GPR imaging scenario. For example, despite having excellent antennas used as radiators given different input pulses
gain and bandwidth, the antenna may significantly distort the including a Gaussian signal. In GPR literature, however, there
transmitted signal [6], create unnecessary clutter reflections, seem to be no consistent and comprehensive set of quantitative
have extremely loud cross‐talk which smothers targets, or antenna specifications for the demanding application of GPR.
image with poor horizontal resolution [7, 8]. Several works, Issues caused by these phenomena can be alleviated to
such as [9–13], consider some of these phenomena, but never some degree using signal processing algorithms. Logically,

-
This is an open access article under the terms of the Creative Commons Attribution‐NonCommercial‐NoDerivs License, which permits use and distribution in any medium, provided the
original work is properly cited, the use is non‐commercial and no modifications or adaptations are made.
© 2023 The Authors. IET Microwaves, Antennas & Propagation published by John Wiley & Sons Ltd on behalf of The Institution of Engineering and Technology.

478 IET Microw. Antennas Propag. 2023;17:478–493. wileyonlinelibrary.com/journal/mia2

WAGNER ET AL.
- 479

however, a higher‐quality radar signal will improve detection with labelled phenomena caused by potential antenna
performance. To radiate a clean input signal, the antenna deficiencies.
should consider the impact of each of the ancillary phenomena In order of Figure 1 selection.
mentioned above. Though all aforementioned antennas may
function well in a system (and many do), it is possible that the 1. Antenna coupling is the direct signal from transmitting to
unconstrained phenomena could adversely impact system receiving antenna. The coupling signal is typically of high
performance. Following quantitative specifications based on magnitude with respect to the rest of the B‐scan and can
the total system integration, a newly created GPR antenna is dominate targets.
more likely to function well for a subsurface imaging applica- 2. Radar cross section (RCS) returns are multipath reflections
tion. Therefore, initial consideration of these antenna specifi- that resonate in the antenna‐ground surface‐antenna loop.
cations can save time and money by avoiding expensive The multipath reflections typically appear during the target
redesigns. return, so they may interfere with background removal and
In this paper, we explore several ancillary specifications for cause false targets.
GPR antenna design. In separate sections, we estimate speci- 3. Coupling residual is the noise left over after the removing
fications on coupling signal shape, pulse distortion and ringing, the coupling signal. The residual noise is proportional to the
gain and beamwidth for horizontal imaging resolution, radar slope of the coupling signal and can create a time‐
cross‐section, and front‐to‐back ratio. Design equations and, dependent noise floor.
where applicable, design plots are provided. A final design 4. Pulse distortion is the warping of the transmitted pulse
table describes each specification, its severity, how to calculate from its ideal input shape. Typically, pulse distortion will
each, and typical values. By verifying an antenna's performance cause ringing and elongate A‐scan peaks, which reduces the
against these specifications, an antenna engineer can avoid imaging quality.
costly re‐designs during the prototyping phase. An example 5. Hyperbola tail extent is the width of a point target hyper-
MATLAB programme for calculating all specifications is bola along the scan axis. This quantity is intimately related
available [18]. We have derived and compiled many new and to the final image's horizontal resolution and is controlled
useful secondary specifications for use in GPR antenna design by the antenna's beamwidth. A larger tail is better for
for the first time. imaging.

In addition, we will explore the impact of an antenna's

2 | ADVERSE ANTENNA EFFECTS IN front‐to‐back ratio. To create antenna specifications with
REAL B‐SCAN respect to each topic, a designer must know several basic
system parameters listed in Table 1.
Two‐dimensional (2D) GPR data is presented in a B‐scan In the course of specification derivation, we acknowledge
format. The B‐scan comprises a series of A‐scans, where that any given GPR system is uniquely complex and involves
each A‐scan is a transmitted‐reflected pulse taken at a different many variables, so accurately predicting how an antenna in-
scan position. Figure 1 shows an experimental GPR B‐Scan fluences final performance is challenging. Whenever possible,
approximations are made to deduce “lower bound”‐type
specifications, where a violation of the specification may not
ruin the radar system's integrity, but meeting the given speci-
fication most likely eliminates any problems associated with
that phenomenon.

2.1 | Application to impulse, SFCW, and

FMCW GPR systems

Three common types of GPR systems are Impulse, SFCW

(Stepped‐Frequency Continuous Wave), and FMCW (Fre-
quency‐Modulated Continuous Wave) radars. Impulse radars
operate in the time domain by sending and measuring an
electromagnetic pulse. SFCW systems operate in the frequency
domain by sending and receiving continuous‐wave signals and
measuring the complex reflection coefficient. FMCW systems
operate by chirping a band of frequencies, mixing the received
signal, and measuring the resultant beat frequencies. As all
F I G U R E 1 Diagram showing an example GPR B‐scan using real data.
Bolded and underlined labels denote returns of issue to be explored. Data is three systems fundamentally follow the radar range equation,
taken with an indoor sandbox using a 0.5–4 GHz system, so floor and wall each system type will obey the presented specifications derived
returns are present. from the radar range equation, including coupling slope,
480
- WAGNER ET AL.

TABLE 1 Fundamental system parameters with common typical values for GHz GPR.

Symbol Name Typical values

Vpp;t Transmitted peak‐to‐peak voltage 1–10 V

[f low ; f high � Upper and lower operation frequencies 0.3–3 GHz

λc Wavelength at centre frequency 0.6–0.1 m

� �
xðtÞ System input pulse (or equivalent) that spans f low ; f high Ricker pulse with σ ¼ 100 ps

σJ System RMS jitter 10–40 ps

n Number of dielectric layers 1–3

di Thickness of i‐th dielectric layer. i ¼ 0 refers to the antenna layer (height), i ¼ n refers to the 0 cm–1 m
target layer

G Antenna gain at centre frequency 0 dB–10 dBi

Gamp Path gain from receive antenna to most compressible receiver component 10–15 dB

P1dB 1‐dB compression point of most compressible receiver component 10–20 dBm

Z0 Characteristic impedance 50 Ω

TðθÞ One‐way power transmissivity through n layers for the ray path started by angle θ Determined by (1) or simulation 0.1–1

AðθÞ One‐way power attenuation due to n layers for the ray path started by angle θ Determined by (2) or simulation 0.1–1

σ Expected target radar cross section (RCS) −10: Weak, 0: Normal, 10: Very strong (dBsm)

ϵi Relative permittivity (real) of the i‐th layer. 1–10

αi Attenuation constant the i‐th layer 0.01–10 Np/m (dry–wet)

Pint Maximum expected interferer power 26 dBm

^
Γ Radar enclosure electric field reflectivity −1–0 (metal‐air)

gain‐beamwidth, and antenna radar cross‐section specifica- this paper. Two popular applications are sensing in both multi‐
tions. Additionally, depending on the receiver compressing layer and inhomogeneous media. In the following sections, the
element, all three system types may use the presented coupling proposed equations include generic terms for electromagnetic
magnitude and front‐to‐back ratio specifications. Ideally, im- boundary transmissivity and loss which can be modified
pulse and SFCW radars are effectively interchangeable through depending on the application.
a Fourier transform, so both would follow the proposed In a multi‐layer application, we are concerned with finding
distortion specifications. However, SFCW systems may have the correct value for T ðθÞ, the θ− incident value for the total
fine control over the radiated magnitude versus frequency and power transmissivity from the antenna layer to the target layer,
may employ techniques that intentionally overdrive amplifiers and A, the round‐trip power attenuation from the antenna‐
for specific frequencies. In this case, distortion analysis should target‐antenna. The choice of expression for T depends on
involve examining the response of the antenna when driven the imaging algorithm the system implements. For migration‐
with the equivalent impulse of the SFCW transmitter. FMCW based algorithms such as F‐K migration and Kirchhoff migra-
radars will not follow the discussed distortion specifications tion that specifically only consider the first target reflection and
due to their transmitted waveform. include no multipath effects, the correct value for T is the
All specifications in this paper are presented with respect multiplicative combination of each layer's transmissivity:
to impulse GPR. To transform an applicable specification to
n−1
SFCW or FMCW GPR, convert transmitted and noise RMS
T ðθ0 Þ ¼ ∏ Ti ðθi Þ ð1Þ
voltage in the presented equations to an equivalent power as i¼0
SFCW and FMCW typically operate in terms of power. To
apply time‐domain specifications to SFCW, take the inverse where Ti ðθÞ is the classical power transmissivity between the i
Fourier transform of the SFCW transfer function. and ði þ 1Þ dielectric layers with an incident angle of θi in the i‐
th medium as in ref. [19] or Appendix A. Appendix A contains
expressions for the TE‐polarized case, but (1) also works for a
2.2 | Notes on multi‐layer and TM polarisation. Irrespective of the sensing modality (impulse,
inhomogeneous media SFCW, or FMCW), systems that use a migration based on the
first target reflection (i.e. unknown layer properties) will use (1).
As the overall application space of GPR systems is significant, Additionally, we approximate the total one‐way dielectric power
we attempt to include as many application cases as possible in attenuation with simple addition as (2):
WAGNER ET AL.
- 481

!
n
X 1. If the coupling signal is too large, receiver amplifiers will
Aðθ0 Þ ¼ exp −2 αi d i sec θi ð2Þ saturate. The additional return from a buried target may be
i¼0
lost if amplifiers are compressed.
Conversely, systems that use imaging algorithms that 2. If the slope of the coupling signal is too large, coupling
consider the subsurface layer properties and depths, such as removal methods will create additional noise depending on
full inversion imaging based on the layered Green's function or the system's jitter [8].
migration including expected multipath, will not use (1), but
instead, use a full multilayer‐reflection form of the power cðtÞ is not easily predicted by antenna far‐field parameters
transmissivity. Though not reproduced here for brevity, the due to the near‐field proximity of the antennas. For this reason,
matrix method for transmission through n arbitrary dielectric we create specifications based only on cðtÞ itself.
layers is available in ref. [19]. Systems using inversion algo- To address the coupling signal's magnitude, we approximate
rithms or expecting multipath can use the full n‐layer trans- the large‐magnitude sections of cðtÞ as a sinusoid to estimate an
mission coefficient because all multiply‐reflected power will be equivalent power level. Consider a receiver containing a
helpful. compressible element with antenna‐to‐component linear power
In an inhomogeneous media sensing application, the gain Gamp and P 1dB in Watts. To maintain linearity, the coupling
concept of transmission and attenuation becomes difficult to signal's peak‐to‐peak voltage Vpp should obey (3):
quantify with simple equations. The reader is directed to refs.
[20, 21] to estimate inhomogeneous transmissivity T and sffiffiffiffiffiffiffiffiffiffiffiffiffi
pffiffi P 1dB Z0
attenuation A, or the result for T and L can be obtained max cðtÞ − min cðtÞ ¼ Vpp ≤ 2 2 ð3Þ
through full‐wave simulations of power transmissivity through Gamp
the expected subsurface.
where Z0 is the system's characteristic impedance. For
example, the HMC8412 low noise amplifier (LNA) [22] has a
3 | GPR ANTENNA COUPLING P 1dB of 79.4 mW (19 dBm) with a gain of 14.5 dB. In a 50‐Ω
SPECIFICATIONS system, the maximum coupling Vpp (3) is 1.06 V. This limit of
1.06 V implies that if a coupling signal has an instantaneous
This section examines the coupling signal of a pair of antennas. Vpp higher than 1.06 V, the LNA will be compressed. Suppose
The coupling affects GPR systems with transmitting and the LNA is compressed in a time period containing a target
receiving antennas in close proximity (Δx small), shown in reflection. The target hyperbola will be visibly distorted even
Figure 2. Some energy radiated from the transmit antenna will after coupling removal because the LNA is no longer linear.
be measured on the receive antenna. This energy is denoted as The engineer should design the antenna such that the coupling
the coupling signal cðtÞ. The coupling signal will obfuscate the Vpp is less than 1.06 V. If the system does not have an LNA,
ground return and shallow target reflections if large in then P 1dB may refer to the compression of the analog‐to‐
magnitude. Then, the coupling signal must be removed. Typical digital converter (ADC) or a mixer—whichever is the most
removal methods include the subtraction of a reference signal compressible element in the receiver chain. In this case, Gamp
[8]. The coupling signal affects the radar scan in two ways that would be the total gain to and including the most compressible
require constraint. element.
The issue of coupling slope has been addressed directly in
ref. [8], where two different specifications are presented.
Because the receiver samples the data randomly with some jitter
or phase noise, the coupling signal cðtÞ is not deterministic and
cannot be completely removed. The incomplete removal is
called coupling residual. To avoid any system‐level impacts, we
want to specify the magnitude of the coupling residual after the
removal algorithm to be smaller than our target reflection. Here,
we expand upon Equation (9) in ref. [8] to obtain a single
equation based upon basic far‐field parameters as (4):
� �
�d �
� cðtÞ� ≤
�dt �
2 3 0 1
nP
−1
pffiffiffi
6Gλ pffiσffiffi T ð0ÞAð0Þ7 B 2 di ϵi C
Vpp;t 6 c 7 B i¼0 C
κc pffiffi 6 � n �2 7uBt − C
2 2Z1−α2c σ J 4 P 5 @ c A
ð4πÞ 3=2
di
F I G U R E 2 Diagram showing coupling setup with operating height i¼0
d 0 , antenna separation Δx , sensing a target with RCS σ buried under n layers
of ϵi ; i ∈ 1…n permittivity. ð4Þ
482
- WAGNER ET AL.

In (4), Tn ð0Þ refers to the expected perpendicular power distortion is perhaps the most popular optimisation subject
transmissivity for an n‐layer dielectric subsurface determined other than gain and bandwidth in GPR literature. For example,
by (1) or simulation, and A refers to the round‐trip attenuation [9, 11, 12, 23–25] all consider the radiated waveform's integrity.
determined by (2) or simulation. The quantity in brackets in (4) Hofinghoff et al. [13] use a metric called group delay flatness to
is the square root of the well‐known radar range equation to describe the radiated pulse integrity. We desire to create a
obtain the radar equation in terms of RMS voltage. All three measurement that, above some threshold, the antenna's
terms together estimate the target reflection from a σ‐strength distortion is acceptable. If antenna distortion is too high, the
target. For a detailed derivation, see Appendix A. The single‐ distorted pulse can create misshapen targets and increase the
layer version of (3) is expanded in (4): overall clutter in the image.
Figure 4 shows an experimental example of pulse distortion
� � in a GPR scan. Using the same pair of antennas, we input both a
�d �
� cðtÞ� ≤ clean pulse and a pulse with an exponential ringing profile in two
�dt �
" pffiffiffi # " pffiffiffiffiffiffi # � �
Vpp;t Gλc σ e−2α1 d1 2ϵ1 2d0
κc pffiffiffiffi �2 u t −
Z1−α2c σ J ð4πÞ32 ðd 0 þ d1 Þ2 ϵ1 þ 1 c
ð5Þ

In (4) and (5), Z1−αc =2 is the parameter (Z‐score) of a

standard normal distribution that contains area αc =2 to the
right of itself, αc is the probability that the specification is
violated, and κc is a threshold constant (coupling), typically 0.1
to 1. (4) and (5) are time‐gated by the step function uðtÞ which
represents that coupling can be ignored before the final layer
reflection. Including the time gate means that operating at a
large height decreases sensitivity to coupling residual.
An antenna that satisfies (4) or (5) ensures that once
removed, the RMS voltage of the coupling residue has at most
probability αc to have greater than a κc ‐fraction of the reflected F I G U R E 3 Potential sources of distortion in an antenna: (1) Input
target signal's RMS voltage after the ground return. The an- reflection coefficient, (2) phase centre movement with respect to frequency,
tenna's coupling signal will not interfere after removal by (3) Internal structural ringing, (4) Unequal magnitude/phase delay in
baluns/feeding structures, and (5) Radiation pattern changes with respect to
ensuring a slight coupling residue. For example, a 500 MHz– frequency.
3 GHz system with σ J ¼ 20 ps detecting 0‐dBsm targets a
depth d 1 of 30 cm in a single layer (n ¼ 1) of ϵ1 ¼ 5 dry soil
with G ¼ 4 dB antennas at a height d 0 of 40 cm with a
Vpp;t ¼ 1 V has a coupling slope specification (4) of 160 mV/
ns with κc ¼ 1 and Z1−αc =2 = 1.96 (for αc = 0.05). Regions in
the antenna's coupling signal after a time of 2.67 ns with a
slope greater than 160 mV/ns have a 5% chance of having a
residue on order of the magnitude of the approximate target
reflection. Therefore, the engineer should take precautions to
decrease the coupling slope in those regions.
To estimate cðtÞ, a designer should first simulate a pair of
antennas in a coupling configuration to calculate the coupling
voltage transfer function versus frequency. Then, the designer
should calculate the output cðtÞ after excitation with an
equivalent GPR input pulse to verify (3) and (4).

4 | GPR ANTENNA DISTORTION

REQUIREMENTS

GPR systems typically operate over an ultra‐wide bandwidth.

Both frequency‐domain and time‐domain systems must be
F I G U R E 4 Effects of poor pulse transmission on the quality of a
concerned about antenna distortion, a quality that is difficult to background‐removed GPR B‐Scan. (a) Reference scan with clean input
control over wide frequency ranges. Antenna distortion origi- pulse (inset a). (b) Distorted scan with intentionally ringing pulse (inset b).
nates from many sources [6], some labelled in Figure 3. Pulse Colorbar is ADC units, arbitrary units for voltage.
WAGNER ET AL.
- 483

different scans. By intentionally distorting the input pulse, we transfer function Hð f Þ behind Figure 5's output pulses is
simulate an antenna with poor distortion properties in the form determined in 3‐D electromagnetic simulation by determining
of ringing. In Figure 4, the scan with increased pulse ringing has the S21 of two antennas pointed at each other 1 m apart.
significantly worse ground removal, high target ringing, and In Figure 5, χ ¼ 1 represents the ideal input pulse,
worse clutter. The poor ground removal would entirely obscure χ ¼ 0:989 represents an output pulse with very little distortion,
a shallow target, and the target ringing creates ghost images and χ ¼ 0:85 represents an output pulse with noticeable
below the target upon imaging. ringing. For pulses used in GPR, χ ∈ ð0:95; 1� represents su-
As the varied sources of distortion are complex, intertwined, perb pulse reconstruction, χ ∈ ½0:9; 0:95Þ has excellent pulse
and dependent on antenna structures, it is infeasible to attempt integrity, and χ < 0:9 starts to show ringing. A pulse with
to create a specification based only on those fundamental χ ∈ ½0:75; 0:9Þ can still be usable, but with degraded
sources. Instead, we examine the distortion on the output pulse performance.
when simulated going through a transmission system. If an antenna has a small χ, a designer should focus on
smoothing group delay, flattening the voltage transfer function,
and decreasing reflection coefficient in the operation band-
4.1 | Pulse integrity width. The limitation of using pulse integrity as a specification
is that a low pulse integrity could come from the widening of a
We first employ the pulse integrity χ in (6), defined rigorously pulse, reshaping of a pulse, or ringing. Because a pulse integrity
in ref. [14]: of χ ¼ 0:9 does not physically correspond to a certain quality
of the radar scan, we will next directly examine the pulse's
χ ¼ max½jyðtÞ ⋆ xðtÞj� ≥ κd ð6Þ ringing.
t

where xðtÞ is the input pulse that spans the system's operating
bandwidth, yðtÞ is the output pulse given xðtÞ; a ⋆ b refers to the 4.2 | Residual ringing
normalised cross‐correlation between two vectors a and b
(ja ⋆ bj ≤ 1Þ, and κd is a threshold parameter (distortion), We define another useful specification as the residual ringing in
typically 0.9. Pulses with χ > 0:9 tend to show negligible dB‐scale after the initial pulse. An example of this plot is used
distortion. In this analysis, the response from a target is idealised in ref. [26]. Residual ringing is more physically appealing as a
as a convolution of the primary target‐incident electric field with specification because ringing is directly observable after target
the impulse reflectivity of the target at a specific angle. In reality, returns in both radar scans (Figure 4) and focussed images. We
the measured electric field is a sum over the response from all examine the quantity R in (7):
possible transmission paths with the target's corresponding 0 1
bistatic reflectivities. For generality, we consider ideal targets jyðtÞj
R ¼ 20 log10 @ Au½t − t0 � ≤ κr ð7Þ
with perfectly flat frequency responses and no multipath effects, maxjyðtÞj
t
much like a boresight test in which an antenna radiates directly at
another identical antenna. To find yðtÞ, first simulate or measure
the voltage transfer function Hð f Þ between two antennas The residual ringing R should be less than some threshold
pointed towards each other at least twice the target distance parameter κr for t > t0 . Typically, κr is equal to or less than
apart. Then, apply an inverse Fourier transform to the product −20 dB20. Choosing this value for κr ensures that additional
of Hð f Þ and the Fourier transform of xðtÞ; Xð f Þ, to find yðtÞ. ringing will not be visible in radar imaging when displayed with
An example of this simulation is shown below in Figure 5. The a dynamic range of 10 dB. The parameter t0 stands for the time
at which a perfectly transmitted pulse with no ringing would
end. Therefore, all energy after t0 is considered distortion. A
simple method to estimate t0 , assuming a time‐symmetric input
pulse, is given as (8):

FWTMfenv½xðtÞ�g
t0 ¼ argmaxjyðtÞj þ ð8Þ
t 2

where FWTMðxÞ is the full width in time at a 10th of the pulse's

maximum value and is used as a representation of pulse width.
The FWTM is taken on the envelope of xðtÞ, calculated using
the magnitude of xðtÞ’s analytical counterpart. Figure 6 dis-
plays an example plot of (7). Violations of (7) due to the initial
falling edge near t − t0 ¼ 0 are ignored due to the low‐pass
FIGURE 5 Example showing three different values of pulse nature of a GPR system, which differentiates and widens the
integrity χ . input.
484
- WAGNER ET AL.

is the target‐reflected RMS voltage which is the first term in (13)

at an angle of 0. κr is a tolerance similar to (7), typically 20 dB20,
in this case equal to the minimum signal‐to‐clutter ratio. In this
expression, R0 is a positive value. If the expressions of (17) and
(13) are used, (9) simplifies to (10) below:
8
>
>
>
< � pffiffiffiffi pffiffiffiffi �
c κr �� ϵ0 − ϵ1 ��
R ≥ n−1
0
20 log10 pffiffiffiffi pffiffiffiffi
P pffiffiffi > 8πd0 � ϵ0 þ ϵ1 �
2 d i ϵi > >
:
i¼1
9
pffiffiffi >
>
Að0ÞT ð0Þ σ λc > =
− 20 log10 � n �2 ð10Þ
P
di >
3
ð4πÞ2 >
>
i¼0
;

F I G U R E 6 Example residual ringing of a 0.5–3 GHz Ricker pulse

Note here that the substitution of (12) and (9) add error
going through a simulated antenna system with κr ¼ −20 dB. This residual
waveform passes specifications. The inset plot shows yðtÞ in a linear y‐axis for from far‐field estimates, so if accuracy is needed, V0 and Vt
reference with the FWTM of xðtÞ indicated by the dashed lines. should be determined from a full‐wave simulation.
For example, a calculation of (9) for a 0.3–3 GHz single‐
layer radar application with ϵ1 = 5, d 0 ¼ 20 cm, d 1 ¼ 15 cm,
In the case of Figure 6's waveform, the specification is met G0 ¼ 5 dBi, with a target of σ¼ −10 dBsm, a soil loss of
because the first ringing peak is −22 dB less than the maximum 20 dB/m (2.3 Np/m), and a tolerance κr of 20 dB20 results in
peak value. In the case of a violation, the designer should modify a required ringdown of 44.7 dB in 2.23 ns or a ringdown of
the antenna to decrease the ringing or else risk impacted imaging 20 dB/ns. An antenna which has a ringdown R0 of more than
performance. 20 dB/ns will have a ratio of its ringing to the target‐reflected
Another method to specify residual ringing is by quanti- RMS voltage of at most −20 dB20 in that specific environ-
fying the ringdown R0 of antennas. The ringdown is the ment. Critically, lossy soil and weaker targets increase the
negative slope of a straight line which has a value greater than ringdown requirement, and stringent ringdown requirements
R in (7) for t > t0 . The waveform in Figure 6 has a ringdown of of 50 dB/ns are possible with wet soils and weak targets. For a
approximately 20 dB/ns. Specifying the antenna's ringdown in detailed discussion, see [16].
dB/ns indirectly limits the maximum reflectivity of target that
can be detected without interference from the antenna's im-
pulse response at a given depth. For closely‐spaced bistatic 5 | GPR ANTENNA GAIN, BEAMWIDTH
antennas in general, ringdown can be considered as a CONSIDERATIONS
constraint on the combined coupling and transmit waveform.
Physically, one can interpret the ringdown like the situation An antenna which achieves the minimum required gain may
shown in Figure 2 where two antennas will crosstalk and have a still image poorly. As the vast majority of GPR systems operate
corresponding ground‐bounce. If the coupling can be removed using synthetic apertures, a better specification would keep the
without amplifier compression, ringing on the ground reflec- synthetic aperture in mind when specifying both the beam-
tion or another strong reflector will smother weak targets and width and the gain.
dominate radar images as clutter. Smothering is especially With constant efficiency, an increase in gain will typically
prevalent if the ground is uneven, and the soil is lossy. decrease an antenna's beamwidth. A decrease in beamwidth
Ringdown can be constrained by the soil attenuation, target can in turn decrease the effective imaging aperture. The im-
reflectivity, and initial ringing power by setting the required aging aperture is inversely proportional to the resulting hori-
change in ringing magnitude over the target reflection time as zontal resolution, which should be small for accurate imaging.
in (9): To estimate the horizontal resolution from imaging parameters,
we use (11) [27]:
c κr Vg
R0 ≥ 20 log10 λc
nP
−1 pffiffiffi Vt ð9Þ Horizontal Resolution HR ≈ pffiffiffiffi ð11Þ
2 di ϵi 2 ϵn sin θm
i¼0
where θm is the maximum view angle, indicated in Figure 7.
where Vg is the initial voltage that ringing is referenced to which When selecting the antenna's beamwidth, denoted ΘHP for
in this case is the ground reflection. For radars without coupling half‐power beamwidth, the engineer encounters a dilemma
removal, Vg would be the voltage of the coupling signal. To given a constant antenna efficiency, a higher antenna gain
estimate the ground reflection, Vg ¼ Vrms G0 jRj where R is implies both a higher target reflection and a lower beamwidth.
given by (17) and Vrms is the RMS transmitted voltage. In (9), Vt Yet, a lower beamwidth implies a lower resolution after image
WAGNER ET AL.
- 485

received target‐reflected voltage V rms rx , and the difference

(headroom) between V rms rx and V noise is denoted V head . A
rms rms

positive value of V head indicates that the target is detected.

rms

However, a negative value of V rms head indicates that the target is

lost in noise. Only the main lobe of Gm ðθÞ is used in sensing.
(13) is calculated at the centre frequency, and we intentionally
neglect the frequency‐dependence and discussion of a point‐
spread function for sake of simplicity.
To determine the best beamwidth, (13) is scanned by
varying the antenna's‐position along its scan axis. At each
position, θg and θm are calculated using the path‐of‐least‐time
ray refraction angles (see Appendix A) and (13) is calculated for
V rms
head . To determine the maximum view angle θm , we observe
when V rms head transitions from a positive to negative value, the
target return disappears, and we declare that θm ¼ θn . A
smaller θm implies a worse resolution in (11). The simulation of
(13) will optimise for lowest horizontal resolution.
F I G U R E 7 Diagram showing the geometrical maximum view angle Figure 8a–c demonstrate the nominal horizontal resolu-
θm and half‐power beamwidth ΘHP . tions for various system archetypes versus beamwidth and
gain. Ideally, the designer will select the antenna's gain and
beamwidth for the minimum horizontal resolution.
focussing using the synthetic aperture. Then, what is the Figure 8a shows the horizontal resolutions for various
optimal beamwidth‐gain configuration for a given system? target depths for a GPR system where the antennas are air‐
To properly consider this question, we must constrain coupled at a height of d 0 = 25 cm off the ground. Under
ourselves to a class of beamwidths. We select normalised an- these system parameters, Figure 8a shows that it is generally
tenna gain patterns Gm ðθÞ with power output centred at θ ¼ 0 favourable for the system to trade gain for beamwidth. In this
which belong to a sincðθÞ class (12): case, the antenna designer should choose to aim for a beam-
width between 60 and 90°, or a gain between 3 and 4 dBi.
G0 jsincðmθÞj There is no significant loss in resolution until approximately
Gm ðθÞ ¼ G0 Fm ðθÞ ¼ Rπ ð12Þ
2π 0 jsincðmθÞjsinðθÞdθ 30° beamwidth.
Figure 8b plots the horizontal resolutions on a drone‐
By varying m in (12), we can alter the antenna's beamwidth mounted GPR system operating at 1.5 m off the ground. The
ΘHP . The sincðθÞ shape of Fm is selected as the main lobe of a additional height in a drone system causes increased path loss,
generic antenna pattern. We choose G0 = 4 to simulate realistic which causes a smaller θm for a given set of a system parameters.
antenna gains. We arrive at this number by comparing the The increased path loss of a UAV‐based GPR system favours
maximum (peak gain) of Gm ðθÞ to the beamwidth of Gm ðθÞ trading beamwidth for more gain. In the case of targets buried at
for different G0 . G0 implicitly contains the antenna's efficiency. 50 cm deep, the antenna designer should aim for a beamwidth
For intrinsically inefficient antennas such as resistively loaded between 30 and 60°, or a gain between 4 and 7 dBi.
vee dipoles (RLVDs), then G0 should be tuned to align with Figure 8c shows the horizontal resolutions for three
the proper gain‐beamwidth relation. different three‐layer detection scenarios, where each scenario is
Given a minimum RMS received voltage (noise level), we a different combination of three permittivities, depths, and
use Gm ðθÞ to simulate a simplified version of the GPR scan, electromagnetic attenuations. Scenario 1 is the base scenario
and thus approximate the horizontal resolution for a given from which scenarios 2 and 3 are derived. In scenario 1, the
antenna gain pattern. To simulate the scan, we again employ. expected horizontal resolution curve implies that a low gain/
the well‐known radar range equation, similar to the process high beamwidth antenna is best for sub‐ground imaging with
shown in (4) and Appendix A: the particular parameters given in Figure 8c. Scenario 2 has a
thicker top‐layer (30 cm) with a permittivity of 6. This com-
2 3 bination of layers leads to interesting behaviour in the
pffiffiffi 7 resolution‐beamwidth curve – for the given system settings,
Vpp;t 6 6G0 Fm ðθ0 ÞT ðθ0 Þ σ λc 7
V rms
head ≈ p ffi ffi 6 �n �2 7Aðθ0 Þ − V noise ð13Þ
rms
the resulting horizontal resolution is best when the antenna has
2 2 4 3 P di
5
ð4πÞ2 a gain of 2–4 dBi, and significantly worse when the antenna has
cos θi
i¼0 lower or higher gain. Scenario 3 changes the permittivity values
of bottom two layers. In this case, changing the last two
where now θi are the ray‐based angles in the layered medium permittivity values simply shifts the resolution‐beamwidth
indicated in Figure 7. In (13) all terms are defined in Table 1 or curve down. From Figure 8c, we can conclude that multi‐
previously. The first term on the right side of (13) is the RMS layer applications have interesting gain‐beamwidth behaviour
486
- WAGNER ET AL.

F I G U R E 8 Example beamwidth‐resolution curves (5) for various GPR systems. Nominal parameters if not specified: α = 0.01 dB/m, d0 = 25 cm, σ = 0
dBsm, G0 = 4, ϵr = 5, λc = 10 cm, Vpp = 1 V. Lower resolution is better. (a) Medium‐height GPR system with d0 = 25 cm. (b) Drone‐mounted GPR system with
d0 = 1.5 m, Vpp = 2 V. (c) Horizontal resolution for three multilayer scenarios (λc = 30 cm, Vpp = 3 V). Scenario 1: d = (25,10,30) cm, ϵ = (1,6,4), α = (0,3,1)
Np/m; Scenario 2: d = (25,30,30) cm, ϵ = (1,6,4), α = (0,3,1) Np/m; Scenario 3: d = (25,10,30) cm, ϵ = (1,10,2), α = (0,3,1) Np/m.

depending on the relative thicknesses and permittivities of T A B L E 2 Summary of qualitative specifications on antenna patterns
for GPR antennas in various applications. System parameters in caption of
their constituent layers.
Figure 8. Beamwidth‐gain conversions are based on (11).
It is important to consider the effect of soil attenuation.
Neither of the three situations in Figure 8c directly varied the Application/System Gain behaviour Beamwidth behaviour
soil attenuation. Both logically and based on (13), we can see Low‐height GPR 3–4 dBi 60°–90°
that a higher attenuation (lower value of Aðθ0 Þ will decrease
Drone GPR 4–7 dBi 30°–60°
the received voltage and make detection more difficult. There
arepthree
ffiffiffiffiffiffiffiffiffi ways to analyse soil attenuation. First, consider Weak targets/lossy soil 6–9 dBi 20°–40°
G0 σð0ÞAð0Þ as a conserved quantity in (13). Detection in an High clutter environment Higher gain Lower beamwidth
environment with high attenuation ‐ low Að0Þ – is equivalent
to having a reduced antenna gain or attempting pffiffiffiffiffiffiffiffiffiffiffito detect a
weaker target. Second, consider Fm ðθ0 Þ σ ðθn ÞAðθ0 Þ. As 6 | GPR ANTENNA RADAR CROSS‐
attenuation directly scales with expðsec θÞ detection in an SECTION SPECIFICATION
environment with high attenuation is equivalent to either
having a narrower antenna beampattern than designed or to The boresight radar cross section (RCS) of GPR antennas has
the detection of a target with further limited view angle been a long‐considered parameter as in ref. [23]. Antennas with
reflectivity. Third, a higher soil attenuation will directly increase a high RCS may introduce multiple clutter signals which can
the stringent requirements on the impulse ringdown due to obscure or drown out target reflections. The studied clutter
weaker target reflections, as explored in [16]. As a portion of signal is the electromagnetic wave which bounces between the
this ringdown is attributable to transmit‐receive antenna side- ground and antenna structure one or multiple times. Examples
lobes or main lobe overlap, a higher soil attenuation places an of RCS reflections are commonly found as horizontal striping
emphasis on high directionality of antennas. in the B‐scans of typically high RCS antennas, such as those in
In all previously considered situations, we have not yet refs. [11, 25, 28, 29].
discussed clutter. A larger beamwidth will include more The removal of these reflections can distort parts of the
clutter in the GPR image. Clutter, by nature, is random. Extra target hyperbola. If the ground profile is not flat, then removal
clutter in the imaging process will degrade the SNCR (signal of these multipath reflections will be difficult because the re-
to noise‐and‐clutter ratio) and increase false alarms. As these flections will look extremely similar to targets. If the RCS re-
specifications give a qualitative range in which to design, it is flections are poorly timed and arrive at the time‐depth
best for the designer to err on the side of narrower beam- equivalent exactly equal to the target return, the top of the
widths and higher gain, especially when operating at higher target hyperbola can be entirely smothered. These artefacts
heights. Finally, we have performed additional simulations may collect errors through the processing pipeline, leading to a
similar to Figure 8 with lowering the target reflectivity and decreased SNCR and increased false alarm rate. Figure 9 shows
increasing soil loss. As in Figure 8b, an increased difficulty of an experiment to highlight the issues of a high antenna RCS.
target sensing tends to favour higher gain antennas, from 6 to In Figure 9, we augment a pair of antennas with a metallic
9 dBi. This and other results from this study are summarised reflector to simulate increasing the antenna's RCS σ ant . The
in Table 2. antenna pair scans over a patch of uneven soil (Figure 9b). In
WAGNER ET AL.
- 487

where the left‐hand side of the equation is related to the time‐

of‐arrival (multiplied by speed of light) of the m‐th RCS
bounce and the right‐hand side of the equation is related to the
time‐of‐arrival of the target reflection at boresight. By itself,
(14) is not a complete specification and does not make a claim
about the magnitude of any RCS reflection.
(14) implies that there are three system applications in which
a GPR system is insensitive to antenna RCS: drone‐based sys-
tems operating at relatively large heights, low‐to‐medium height
systems looking exclusively for deep targets, and ground‐
coupled GPR systems operating at d 0 ≪ λc where the concept
of multipath bounces between antenna and ground due to RCS
no longer holds. When (14) is true, we must constrain the ex-
pected magnitude of the multipath RCS reflections. We elect to
constrain the expected RCS reflection magnitude against the
expected target return.
In order to find an upper bound for the antenna RCS, we
must estimate the reflectivity of the ground, jRj. After the
ground reflectivity is estimated, we construct the specification
by constraining the RMS voltage of the multipath RCS re-
flections. These reflections should be less than a scalar multiple
κrcs of the expected RMS voltage of the target return. For
F I G U R E 9 Example of increased antenna RCS creating false targets.
generality, we assume an isotropic radiation pattern. In mathe-
(a) Photograph of metallic backing reflector to simulate increased antenna matical terms, we constrain the antenna's boresight RCS σ ant as a
RCS. (b) Uneven ground surface. Pictured grid reference is parallel to floor. function of received target RMS voltage V rmsrx and other terms in
(c) Raw scan with coupling and RCS bounces labelled. (d) Kirchhoff‐ Table 1 as (15):
migrated image with true and extra false targets shown.
Vpp:t m=2 mþ1
pffiffi σ ant jRj ≤ κrcs V rms
rx ð15Þ
Figure 9d, the first (m ¼ 1) and second (m ¼ 2) RCS re- 2 2
flections appear inside the target zone as a strong target (boxed
in red). The combination of increased antenna RCS and an Equation (15) has the given form because the m‐th RCS
pffiffiffiffiffiffiffi
uneven ground surface entirely hallucinate these two targets. reflection is reflected by σ ant m times and takes m þ 1
Additionally, true targets (boxed in green) have ghost followers round‐trip reflections between antenna and ground. In (15),
(boxed in red) due to the antenna's RCS. The increased RCS V rms
rx is here defined as the first term in (12) with θi ¼ 0 ∀ i and
has dramatically increased the radar's false alarm rate. G0 Fm ðθÞ ¼ 1, explicitly defined in (16):
Though the idea of RCS as a far‐field parameter becomes 2 3
inaccurate when operating at medium‐to‐low heights, we use
pffiffiffi
RCS as a first‐order approximation of reflectivity due to its Vpp;t 6 6 σ λc T ð0ÞAð0Þ 7
7
ready availability in full‐wave simulation. Additionally, the RCS V rms
rx ¼ p ffiffi 6 � � 27 ð16Þ
2 24 3 n
P 5
of an antenna is not the only source of reflection of up‐going ð4πÞ 2 di
waves. The circuitry and mechanical holdings, generally placed i¼0

above the antenna [7], have been shown to create additional

reflections that can enhance or even dominate antenna‐based In (15), R contains the reflectivity of the ground surface at
reflections. The quantity σ ant should ideally contain the an antenna height d 0 , calculated as the reflection of a spherical
contribution of both antenna and backing structure. This wave emanated by and measured at a source point impinging
structural reflection is the principle behind Figure 9. on an infinite dielectric half‐space [30, 31]. In estimating R, a
A high antenna boresight RCS is only an issue when the proper treatment would calculate the Sommerfeld integral
time of arrival of the m‐th coupling bounce ðm ¼ 1; 2; …Þ lies formulation [32], and this approach has been performed in
in the range of the time of arrival of expected targets. Other- literature [31]. However, we explicitly value simplicity over
wise, the multiply‐reflected signals cannot affect target returns. accuracy. Following the development of [31], the magnitude of
By equating times of arrival in a boresight system condition, we reflection of a spherical wave from a planar boundary ap-
derive that the m‐th antenna RCS multipath reflection could proaches the common Fresnel plane wave refraction co-
potentially overlap a target detection when (14): efficients. Then, we can make the intuitive approximation (17):

n
X � pffiffiffiffi pffiffiffiffi �
pffiffiffiffi pffiffiffi 1 �� ϵ0 − ϵ1 ��
d 0 ϵ0 ð1 þ mÞ ≥ d i ϵi ð14Þ jRj ≈ pffiffiffiffi pffiffiffiffi ð17Þ
i¼0
8πd0 � ϵ0 þ ϵ1 �
488
- WAGNER ET AL.

(17) follows from Brekhovskikh's development [33] with a antenna beamwidth. Similarly, at small d 0 , both the implicit far‐
normal incidence angle with an approximation for kd0 large field assumption and plane‐wave approximation of (17) become
(wavenumber k). The rightmost term in (17) is the normal‐ intolerable. Therefore, (18) is useful for d 0 in a medium range,
incident reflection coefficient of the ground while the left from approximately 20 cm to 1 m. When at the edges of
term accounts for geometric spreading with an equivalent applicability, (18) will tend to over‐constrain the antenna RCS.
image source height of 2d0 . However, (18) is still useful as a simple, portable tool to generate
Combining (15), (16), and (17) together and collecting scalar values of RCS specification.
constants, we obtain the final expression for an upper bound To verify (18), the antenna engineer should simulate the
specification on the antenna RCS σ ant (18): RCS in the boresight direction. Commonly used full‐wave
simulators have the option to calculate RCS. If an antenna
20ðm þ 1Þ violates (18), the antenna engineer should be careful to reduce
σ ant ðdBsmÞ ≤ − log10 jRj þ
m the boresight RCS of the antenna. RCS reduction can be
2 3
achieved in several ways, the simplest being the introduction of
p ffi
6κ T ð0ÞAð0Þλ σ 7 32:9 ffi
ffi
20 6 rcs c 7 ð18Þ absorbing material [34, 35], or physically modifying the an-
log10 6 � n �2 7−
m 4 P 5 m tenna to present less physical cross‐section when viewed from
di boresight.
i¼0
#
Again κrcs is a threshold parameter, typically in the range 7 | FRONT‐TO‐BACK RATIO
0.1–2, chosen small for non‐flat applications and large for flat SPECIFICATIONS
applications. The value for m can be found by the first value
which satisfies (14). An example calculation of (15) is shown Constraining an antenna's front‐to‐back ratio (FBR) is a com-
below for a two‐layer application (n ¼ 2Þ in Figure 10. In mon mode of specifying the antenna's radiation pattern. Bowtie
Figure 10, an antenna which has an RCS higher than σ ant for a and RLVD antennas such as refs. [23, 36, 37] focus on or show
given target depth curve and operating height will contend improvements to front‐to‐back ratio. Designers typically view a
with multipath reflections due to antenna RCS on the order of higher FBR as better. Due to its interaction with the system's
the expected target return. enclosure, FBR is innately related to the coupling signal cðtÞ in
In Figure 10, there are several interesting trends worth (3) and (4), so some designers may use FBR as a proxy specifi-
elaborating upon. First, the general rule of thumb is – the more cation for decreased clutter from coupling. It would be useful to
difficult an object is to detect, the lower the antenna RCS have a quantitative specification for FBR such that antennas
should be. Second – the valid RCS return time (non‐shaded with an FBR greater than a minimum are unlikely to suffer
part of Figure 10) generally sees the maximum allowable an- adverse effects related to low antenna FBR. Excluding coupling
tenna RCS increase with height d 0 . clutter, which has been constrained in (3) and (4), antennas with
At large d 0 , however, the assumption that the ground a low FBR can introduce two additional errors. We will examine
reflection will scale as d 0 −1 will break down due to a finite specifications for both errors. First, a nearby interferer may
transmit a signal with power Pint incident on the back of the
radar as indicated in Figure 11, where Pint is high enough to
compress the receiver amplifiers. Second, the back‐lobe's
transmitted energy may reflect off the radar's enclosure [7, 23]
and re‐orient towards the ground (Figure 11 right side), creating
ghost images like those in Figure 9d. To maintain generality,
both specifications make use of the parameter Γ ^ estimates the
reflectivity of the enclosure and assumes the enclosure's shape is
a z‐symmetric thin sheet such that the enclosure's transmissivity
is simply related to its reflectivity. These specifications also as-
sume that the backward‐reflected beam does not interfere with
the forward beam.

7.1 | FBR for interferers

Low‐magnitude interferers are easily removed in a radar's
signal processing pipeline. SFCW and FMCW radars are
insensitive to low‐magnitude interferers because of their innate
F I G U R E 1 0 Two‐layer simulated maximum tolerable antenna RCS
σant versus operating height d0 for various target depths d2 . In the shaded frequency‐hopping operation. Equivalent‐time GPR systems
region, (13) is not satisfied and antenna RCS does not affect detection. Other are insensitive to interferers because interferers are often not
system parameters not shown: f c ¼ 1 GHz, α ¼ 0:01 Np/m. synchronized with the system's PRF. Therefore, the only
WAGNER ET AL.
- 489

ghost targets beneath real targets. An intuitive specification

arising from this system is constraining the reflected beam's
relative power as in (20):

� 2�
�Γ
^ �
FBR ≥ ð20Þ
κf

where κf is a thresholding parameter >0 and describes the

ratio between the forward and backward‐reflected beam.
Typically, κf ¼ 0:1 such that the time‐shifted superimposed
scan is at least −10dB less than �the� transmitted scan. For
example, a system with reflectivity �Γ
^ � of 0.9 requires a mini-
mum FBR of 9.1 dB to guarantee that the backward‐reflected
wave is less than −10 dB compared to the forward wave. Note
that systems without an enclosure (Γ ^ ¼ 0) can have any FBR
F I G U R E 1 1 Diagram showing (left) high‐FBR and (right) low‐FBR with respect to imaging.
^ Γ
antennas in an enclosure with estimated reflectivity Γ. ^ is assumed to be
symmetric looking in �z^ and refers to reflected electric field.

8 | APPLICATIONS OF
significant worry is a high‐magnitude interferer that can SPECIFICATIONS
compress or damage the receiver's LNA. Using the same logic
as (1), receiver compression can smother targets. The radar's 8.1 | Design table
received power from an incident interferer with power P int ,
^ will be
transmitted through an enclosure with reflectivity Γ To use the proposed specifications, we have compiled a table
equal to: and created a design flowchart. Table 3 is split into two sections:
the goal specifications and the verification specifications. The
� � �2 � antenna engineer should calculate the goal specifications before
^ � P int
Gb 1 − �Γ
embarking on a design. Upon the first design revision, the an-
tenna engineer should calculate the verification specifications. If
where Gb is the back‐lobe's gain. If the enclosure is lossy, the verification specifications are significantly violated, then the
replace the quantity in parenthesis with the enclosure's inser- antenna engineer should enter a new design revision. Then,
tion loss. To create a specification on FBR ¼ G=Gb , we once all goal and verification design specifications are met, the
reference (1) to create (19): antenna engineer can then enter the prototype phase. In the
� prototype phase, it is still recommended to measure the post‐
� �2 � P int Gamp design verification specifications because all three values may
FBR ≥ G 1 − �Γ
^� ð19Þ
P 1dB change in a real environment. However, by meeting all specifi-
cations, the designed antenna is much more likely to take
(19) describes the minimum FBR for an antenna with excellent subsurface images compared to an antenna which vi-
forward gain G, enclosed in material with reflectivity Γ ^ , con- olates one or many of these specifications.
nected to a receiver with gain Gamp and most‐compressible‐ It is worth identifying the most important specifications for
component compression point P 1dB , and experiencing an different classes of antennas. Vivaldi antennas [38] have excel-
interferer with power P int . (19) can be calculated over all fre- lent gain profiles for GPR but should focus on minimising re-
quencies. For example, a system with an antenna gain of 6 dB, a sidual ringing/ringdown and could suffer from high coupling.
P 1dB of 16 dBm, a Gamp of 14.5 dB [22], and reflectivity of 0.9 Loaded dipoles [23] have excellent ringing profiles but care
requires a minimum FBR of 17.3 dB to accept a 23 dBm should be taken to ensure that coupling residuals are allowable,
incident interferer without receiver compression. and the peak gain is appropriate. TEM and dual‐ridged horns
[39] also have excellent gains but could suffer from high radar
cross section due to a large size and metallic construction
7.2 | FBR for imaging leading to high clutter. Spiral antennas have desirable circular
polarization but unattractive distortion properties that must be
An antenna with a low FBR can plausibly create a time‐shifted properly characterised.
version of the radar scan superimposed on the desired truth Finally, we note that design of a GPR antenna is a multi‐
scan. The wave reflected from the enclosure can be thought
� �2 of faceted problem, and it may be difficult or impossible to
as a secondary source that transmits with a power of �Γ
^ � =FBR meet every specification in Table 3 for a given system. For that
relative to the forward beam. If the FBR is low enough, the reason, we have included a “severity” column which denotes
superimposed scan will impede background removal and create the qualitative importance of meeting each specification.
490
- WAGNER ET AL.

TABLE 3 Table of secondary design specifications for GPR antennas.

Parameter Effects if
(equations) Equation/Specification Severity violated Validity Example value
Goal specifications
Beamwidth (11) Simulate (11) High ‐ Poor horizontal Air‐coupled GPR 30°–60°
resolution
‐ Poor imaging
performance

Antenna boresight σant ðdBsmÞ≤ Low ‐ Distorted target Medium d0 . −6 dBsm;

RCS σant (15), 2 3 hyperbolas krcs ¼ 1
(16) 6κ Tð0ÞAð0Þλ pffiσffiffi 7 ‐ Poor ground
20ðm þ 1Þ 20 6 rcs c 7 removal
− log10 jRj þ log10 6 � n �2 7
m m 4 P 5
di
i¼0

32:9
−
m
� � �2 � Pint Gamp
Front‐to‐back FBR ≥ G 1 − �Γ
^� Low ‐ Receiver Systems concerned with 17.3 dB
P1dB
ratio compression or interference.
(Interference) damage
(17) ‐ Distorted scans
2
jΓ^j
Front‐to‐back FBR ≥ κf Medium ‐ Ghost images Systems with reflective 9.1 dB; κf ¼ 0:1
ratio (Imaging) below targets enclosures.
(18) ‐ Poor ground
removal
c κr Vg
Ringdown R0 (10) R0 ≥ Pn−1 20 log10 Vt High ‐ Weak target ‐ Determine Vg κr ¼ 20 dB;
pffiffiffi
d ϵ
2
i¼0
i i smothering (maximum ringing) R0 > 20 dB/ns
‐ Increased ‐ Determine Vt (expected
clutter target reflection)
‐ Calculate (10)

Effects if
Parameter Equation/Specification Severity violated How to simulate Example value
Verification specifications
pffiffi qffiffiffiffiffiffiffiffiffiffi
Coupling Vpp ≤ 2 2 PG1dBampZ0 Low ‐ Receiver – Two antennas in most 1.06 V
magnitude Vpp compression severe coupling config
(1) ‐ Distorted target ‐ Two‐port S‐parameters
hyperbolas ‐ Input the system's
pulse, measure cðtÞ
‐ Take maximum Vpp of
consecutive peaks
� �
Coupling slope �d � Medium ‐ Increased noise – Same setup as above 160 mV/ns after
�d � � cðtÞ�≤
� cðtÞ� (2) �dt � ‐ Poor coupling ‐ take derivative of 2.67 ns;
dt
2 0 3 1 removal output κc ¼ 1
nP−1 pffiffiffi
pffiffiffi 2 d i ϵi C
6
κc V pp;t 6Gλc σ Tð0ÞAð0Þ7 B 7
B i¼0 C
pffiffi 6 � n �2 7 ⋅ uBt − C
2 2Z1−α2c σJ 4 3=2 P
5 @ c A
ð4πÞ di
i¼0

Distortion pulse χ ¼ max½yðtÞ ⋆ xðtÞ� ≥ κd Medium ‐ Poor focussing – Two antennas Pbore- � κd ¼ 0.9
t
integrity χ (3) ‐ Poor vertical sight, at least 2 i di
resolution apart.
‐ Increased ‐ Two‐port S‐parameters
clutter ‐ Input the system's
equivalent pulse
‐ Calculate (3)
� �
jyðtÞj
Residual ringing R R ¼ 20 log10 maxjyðtÞj u½t − t0 � ≤ κr High ‐ Ghost impages ‐ Same setup as pulse κr ¼ −20 dB
t
(4), (5) ‐ Poor ground integrity
removal ‐ Calculate (4)
WAGNER ET AL.
- 491

8.2 | Discussion of specification interactions antenna engineer can improve the antenna's chances of success
in deployment.
This section contains a brief discussion on the interactions and
trade‐offs between gain, coupling, and other specifications. A C K N OW L E D G E M E N T S
First, we examine the interaction between gain/beamwidth, This project is sponsored by the Department of the Navy,
coupling slope and magnitude, and front‐to‐back ratio. All three Office of Naval Research under award N00014‐17‐1‐2488.
of these are fundamentally specifications on the spatial radiation Any opinions, findings, and conclusions or recommendations
profile of the antennas. An antenna with a wide beamwidth will expressed in this material are those of the author(s) and do not
likely have a low front‐to‐back ratio and high coupling slope and necessarily reflect the views of the Office of Naval Research.
magnitude due to sideways‐directed radiation. Conversely, high‐
gain antennas may have significant coupling if the gain is ach- A UT HO R C O N TR I B U TI O N S
ieved with a large antenna size. Therefore, adjusting the an- Samuel Wagner: Conceptualisation, Formal analysis, Investi-
tenna's gain to perfectly fit the imaging profile determined by gation, Methodology, Writing – original draft, Writing – re-
simulation of (13) may cause the antenna's coupling and front‐ view & editing. Ababil Hossain: Data curation, Investigation,
to‐back specifications to fail. As all three are inter‐related, this Writing – review & editing. Stephen Pancrazio: Data curation,
design difficulty extends to changing any single specification. Investigation, Writing – review & editing. Anh‐Vu Pham:
Another important relationship is that between boresight Project administration, Supervision, Writing – review & editing.
radar cross‐section and antenna gain. An increase in antenna
gain may lead to an increase in antenna size, which could C O N F LI C T O F I N TE R ES T S TAT E M E N T
impact the radar cross‐section of the antenna and cause There is no conflict of interest among the authors.
problems if it violates the specification (18).
DA TA AVAI L A B I L I TY S TA TE M E N T
Data available on request from the authors.
8.3 | Polarimetric GPR
O R CI D
Polarimetric GPR systems use diverse transmit and receive Stephen Pancrazio https://orcid.org/0000-0003-4459-2776
polarisations to detect hard‐to‐measure targets that have Anh‐Vu Pham https://orcid.org/0000-0001-5458-0109
depolarisation phenomena such as bent metallic curves [40] or
angled linear wires [41]. The previously developed specifica- R EF E R E N C E S
tions can be modified for polarimetric GPR by replacing target 1. Hertl, I., Strycek, M.: UWB antennas for ground penetrating radar
reflectivity σ in the equations with the specific co‐ or cross‐ application. In: 2007 19th International Conference on Applied Elec-
polarization reflectivity term. A common example is in (21), tromagnetics and Communications, pp. 1–4 (2007)
2. Travassos, X., et al.: A review of ground penetrating radar antenna design
where horizontal and vertical polarisations are both trans-
and optimization. J. Microwaves Optoelectron. Electromagn. Appl. 17,
mitted and received. 385–29 (2018)
� � 3. Coburn, W., McCormick, S.: Ultra‐wideband antenna performance
σ HH σ HV comparison. In: 2018 International Applied Computational Electro-
σ¼ ð21Þ
σ VH σ VV magnetics Society Symposium (ACES), pp. 1–2 (2018)
4. Keceli, S.I., et al.: Link budget analysis & verifications for system design of
GPR. In: 2015 16th International Radar Symposium (IRS), pp. 463–468
Similarly, specifications based on the radar‐range equation (2015)
should include a polarization mismatch term, which should 5. Wibowo, T.P., Zulkifli, F.Y.: Analysis of concrete inspection radar equa-
include the angle dependence of the polarization purity. As tions to obtain radar system specifications. In: 2019 16th International
polarimetric GPRs typically comprise more complicated and Conference on Quality in Research (QIR): International Symposium on
Electrical and Computer Engineering, pp. 1–4 (2019)
diverse deployments of antennas, care must be taken for
6. Sörgel, W., Wiesbeck, W.: Influence of the antennas on the ultra‐wideband
proper application of these concepts. transmission. EURASIP J. Adv. Signal Process. 2005(3), 843268 (2005/
03/17 2005)
7. Wagner, S., Pham, A.: Structural effect on image quality degradation in
9 | CONCLUSION ground‐penetrating radar array. In: 2019 IEEE International Symposium
on Antennas and Propagation and USNC‐URSI Radio Science Meeting,
pp. 1631–1632 (2019)
In conclusion, we have considered the context of GPR antenna 8. Wagner, S., Worthmann, B.M., Pham, A.V.: Minimizing timing Jitter’s
deployment in a whole system to create a set of useful speci- impact on ground‐penetrating radar array coupling signals. IEEE Trans.
fications. We have explored specifications on the antenna Geosci. Rem. Sens. 59(6), 4717–4724 (2021)
coupling signal, the transmitted pulse integrity, the transmitted 9. Wagner, S., Pham, A.: The ultrawideband elliptical resistively loaded vee
dipole. IEEE Trans. Antenn. Propag. 68(4), 2523–2530 (2020)
pulse ringing, the beamwidth, the antenna's boresight RCS, and
10. Wagner, S., et al.: Design of a wideband resistively loaded vee dipole fed
the antenna's front‐to‐back ratio. An example code to calculate by an even‐mode matched Marchand Balun. In: 2018 IEEE International
specifications is available. By calculating and considering these Symposium on Antennas and Propagation & USNC/URSI National
system‐based specifications early in the design process, the Radio Science Meeting, pp. 1353–1354 (2018)
492
- WAGNER ET AL.

11. Chen, G., Liu, R.C.: Design of a shielded antenna system for ground of the 15th International Conference on Ground Penetrating Radar,
penetrating radar applications. In: 2009 IEEE Antennas and Propagation pp. 739–744 (2014)
Society International Symposium, pp. 1–4 (2009) 37. Li, Y., Chen, J.: Design of miniaturized high gain bow‐tie antenna. IEEE
12. Lestari, A.A., et al.: A modified bow‐tie antenna for improved pulse Trans. Antenn. Propag., 70(1), 738–743 (2021)
radiation. IEEE Trans. Antenn. Propag. 58(7), 2184–2192 (2010) 38. Lualdi, M., Lombardi, F.: Significance of GPR polarisation for improving
13. Höfinghoff, J., Overmeyer, L.: Resistive loaded antenna for ground target detection and characterization. Nondestruct. Test. Eval. 29(4),
penetrating radar inside a bottom hole assembly. IEEE Trans. Antenn. 345–356 (2014)
Propag. 61(12), 6201–6205 (2013) 39. Liu, H., Zhao, J., Sato, M.: A hybrid dual‐polarization GPR system for
14. Lamensdorf, D., Susman, L.: Baseband‐pulse‐antenna techniques. IEEE detection of linear objects. IEEE Antenn. Wirel. Propag. Lett. 14, 317–320
Antenn. Propag. Mag. 36(1), 20–30 (1994) (2015)
15. Lestari, A., Yarovoy, A., Ligthart, L.: Analysis and design of improved 40. Guo, J., et al.: An ultrawide band antipodal Vivaldi antenna for airborne
antennas for GPR. Subsurf. Sens. Technol. Appl. 3, 295–326 (2002) GPR application. Geosci. Rem. Sens. Lett. IEEE. 16(10), 1560–1564
16. Daniels, D.J., et al.: Antenna design considerations for ground pene- (2019)
trating radar landmine detection. IEEE Trans. Antenn. Propag. 70(6), 41. Ježová, J., Lambot, S.: A dielectric horn antenna and lightweight radar
4273–4286 (2022) system for material inspection. J. Appl. Geophys. 170 (2019)
17. Schantz, H.: The Art and Science of Ultrawideband Antennas, 2nd edn. 42. Balanis, C.A.: Antenna Theory: Analysis and Design, 3 edn. John Wiley &
Artech House (2015) Sons (2005)
18. Wagner, S.: Ground Penetrating Radar Antenna Design Specs Generator.
[GitHub Repository]. https://github.com/Samuel-Wagner/Ground-
Penetrating-Radar-Antenna-Design-Specs-Generator (2021) How to cite this article: Wagner, S., et al.: System‐
19. Orfanidis, S.J.: Electromagnetic Waves and Antennas [Online]. https://
based specifications for better design of ground‐
www.ece.rutgers.edu/~orfanidi/ewa/ (2016)
20. Habib Mazharimousavi, S., Roozbeh, A., Halilsoy, M.: Electromagnetic penetrating radar antennas. IET Microw. Antennas
wave propagation through inhomogeneous material layers. J. Electro- Propag. 17(6), 478–493 (2023). https://doi.org/10.
magn. Waves Appl. 27(16), 2065–2074 (2013/11/01 2013) 1049/mia2.12358
21. Bereuter, W.A., Chang, D.C., Standards, U.S.N.B.O.: Electromagnetic
Remote Sensing of Inhomogeneous Media. U.S. Department of Com-
merce, National Institute of Standards and Technology (1977)
22. Devices, A.: HMC8412 Low Noise Amplifier, 0.4 GHz to 11 GHz.
https://www.analog.com/en/products/hmc8412.html (2021) AP P E N D I X
23. Kangwook, K., Scott, W.R.: Design of a resistively loaded vee dipole for This appendix contains the concept behind the derivation of
ultrawide‐band ground‐penetrating radar applications. IEEE Trans.
Equations (4), (12), and (17). Consider the situation of Figure 7
Antenn. Propag. 53(8), 2525–2532 (2005)
24. Li, M., et al.: Compact slot antenna with low dispersion for ground with n ¼ 1, where an antenna radiates at an oblique angle with
penetrating radar application. IEEE Antenn. Wirel. Propag. Lett. 15, respect to the ground. We aim to calculate an expression for
638–641 (2016) the received power P r from a transmitted power P t after
25. Shao, J., et al.: TEM horn antenna loaded with absorbing material for refraction through ground and reflection from target. This
GPR applications. IEEE Antenn. Wirel. Propag. Lett. 13, 523–527 (2014)
derivation ignores near‐field effects and has reduced accuracy
26. Lestari, A.A., Yarovoy, A.G., Ligthart, L.P.: RC‐loaded bow‐tie antenna
for improved pulse radiation. IEEE Trans. Antenn. Propag. 52(10), at λ‐small operating heights and depths.
2555–2563 (2004) Without the influence of ground, a transmit power P t will
27. Persico, R.: Introduction to Ground Penetrating Radar: Inverse Scat- induce (ignoring polarization losses), a receive power P r of
tering and Data Processing, pp. 144–145. Wiley‐IEEE Press (2014) approximately (A.1):
28. Ahmed, A., et al.: Design of UWB antenna for air‐coupled impulse
ground‐penetrating radar. Geosci. Rem. Sens. Lett. IEEE. 13(1), 92–96
λ2
(2016) P r ¼ P t σðθÞG2 ðθÞ ðA:1Þ
29. Jamali, A.A., Marklein, R.: Design and optimization of ultra‐wideband ð4πÞ3 ðd 0' þ d 1 'Þ4
TEM horn antennas for GPR applications. In: 2011 XXXth URSI
General Assembly and Scientific Symposium, pp. 1–4 (2011)
30. Li, J., et al.: Frequency‐dependent spherical‐wave reflection in acoustic (A.1) is the radar range equation [42] for a σ‐strength target
media: analysis and inversion. Pure Appl. Geophys. 174(4), 1759–1778 at a radial distance of d 00 þ d1 0 at an angle of θ. We consider
(2017/04/01 2017) that the transmit and receive antennas are identical and
31. Li, J., et al.: A novel expression of the spherical‐wave reflection coeffi-
collocated with a gain at elevation angle GðθÞ. The inclusion of
cient at a plane interface. Geophys. J. Int. 211(2), 700–717 (2017)
32. Cui, T.J., Chew, W.C.: Fast evaluation of Sommerfeld integrals for EM ground separates the in‐ground and in‐air angles, adds
scattering and radiation by three‐dimensional buried objects. IEEE dielectric loss, and creates a transmission coefficient at the
Trans. Geosci. Rem. Sens. 37(2), 887–900 (1999) dielectric interface. With the influence of a single dielectric
33. Brekhovskikh, L.M.: Chapter IV – reflection and refraction of spherical ground, (A.1) becomes (A.2):
waves. In: Brekhovskikh, L.M. (ed.) Applied Mathematics and Mechanics,
vol. 16, pp. 225–298. Elsevier (1980)
σ ðθ1 ÞG2 ðθ0 ÞT 2 ðθ0 Þ λ2c −α1 cos4d1θ
34. Liu, Y., et al.: Low RCS microstrip patch antenna using frequency‐selective Pr ¼ Pt � �4 e 1
ðA:2Þ
ð4πÞ3 cosd0θ0 þ cosd1θ1
surface and microstrip resonator. IEEE Antenn. Wirel. Propag. Lett. 14,
1290–1293 (2015)
35. Zhang, W., et al.: Wideband RCS reduction of a slot array antenna using
phase gradient metasurface. IEEE Antenn. Wirel. Propag. Lett. 17(12),
2193–2197 (2018)
In (A.2), the exponential term contains the soil loss of the
36. Ajith, K.K., Bhattacharya, A.: Improved ultra‐wide bandwidth bow‐tie sole dielectric layer and T ðθÞ contains the power transmissivity
antenna with metamaterial lens for GPR applications. In: Proceedings through the dielectric interface. In the transition from (A.1) to
WAGNER ET AL.
- 493

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
(A.2), d 0i ¼ d i sec θi . θ0 and θ1 refer to the transmission angles pffiffiffiffiffiffiffiffiffiffiffi ϵi
4 ϵi ϵiþ1 cosðθi Þ 1− sin2 θi
in air and ground respectively for a single‐layer case. With the ϵiþ1
Ti ðθa Þ ¼ h ðA:5Þ
addition of n ≥ 1 horizontal dielectric layers beneath the an- pffiffiffi pffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi i2
ϵi cosðθi Þ þ ϵiþ1 1 − ϵϵiþ1i sin2 θi
tennas, (A.2) becomes (A.3):

σ ðθn ÞG2 ðθ0 Þλ2c T 2 ðθ0 Þ 2

Pr ¼ Pt �n �4 A ðθ0 Þ In (A.5), we assume the incident field is transverse electric
3 P di ðA:3Þ
ð4πÞ cos θi
(TE). Now, we derive (4) and (5). In ð4Þ, we are concerned with
i¼0 the received target‐reflected RMS voltage V rx rms in a boresight
where now we have generalised the transmission T ðθi Þ and sensing test, so we take the square root of (A.3) with θi ¼ 0 ∀ i:
attenuation Aðθi Þ coefficients for n‐layers. Substituting the
values for T ðθi Þ and Aðθi Þ from (1) and (2) respectively, we pffiffiffiffiffiffiffiffiffi
Vpp;t σð0ÞGð0Þλc
arrive at a version of the radar range equation in (A.4) useful V rx
rms ¼ pffiffi � n �2 T ð0ÞAð0Þ
2 2 P ðA:6Þ
for GPR systems based on migration. ð4πÞ3=2 di
i¼0
P
n
4αi di
σ ðθn ÞG ðθ0 Þλ2c ∏n−1
2
T 2i ðθi Þ − cos θi
Pr ¼ Pt � n i¼0�4 e i¼0
ðA:4Þ
3 P di
In the single‐layer case, (A.6) is simplified in (A.7) using
ð4πÞ cos θi (A.4) and (A.5) with ϵ0 = 1:
i¼0

In (A.4), the exponential term sums up the relative losses in " pffiffiffi #" pffiffiffiffiffiffi #
each dielectric layer, the spherical spreading term adds the Gλ σ 2ϵ1 −2α d
V rx
rms ¼ Vpp;t 3 pffiffiffiffi �2 e 1 1
distance from each layer, and the power transmissivity term ð4πÞ2 ðd 0 þ d 1 Þ2 ϵ1 þ 1
considers each layer's transmissivity. (A.4) reduces to (A.2) in
ðA:7Þ
the case of a single dielectric layer. The construct θi refers to
the angle travelled in the i‐th dielectric layer with respect to the
normal. d i refers to the either the depth of the i th dielectric (A.7) is the main form of (5) and (A.6) is the main form of
layer or the target's depth in the last layer if i ¼ n. (4), excluding constraining parameters explained in Section II.
The angles θi can be solved for using an iterative approach Equations (12) and (17) are derived in a similar method. In
as in ref. [18]. Examining the Fresnel coefficients results in the particular, (12) uses the simulation of (A.3) for varying θ0 to
simplified form of power transmissivity from layer i to layer optimise the antenna's beamwidth by calculating the square
i þ 1 in (A.5): root of (A.3), and (17) uses (A.6) directly.