1. Introduction
According to [
1], ultra-wideband (UWB) technology in radars basically extends into three different areas according to their application range: short, medium, and long range. Leading components within each of these areas include: Ground-or-Surface Penetrating Radars (GPR), Terrain surveillance radars, and Synthetic Aperture Radars (SAR), respectively. At the moment the most widely used UWB radar is GPR, which has been commercialized by more than a dozen companies around the world [
2].
Due to the widespread proliferation of electromagnetic devices, the radio spectrum has become a limited resource, so that the “broad intrusion” of the GPR signals in an already crowded spectrum has recently resulted in proposals that establish rules and regulations regarding the use and characteristics of these devices [
3,
4].
The main advantage of using UWB technology for surface radars is determined by the need for high vertical resolution (also known as down-range or depth resolution), which is closely related to the radar operational bandwidth [
5]:
where Δv is the vertical resolution of the system, c is the velocity of an electromagnetic signal in vacuum, and Δf is the bandwidth used by the system.
Due to the use of UWB technology, remote sensing devices known as radar were divided into two broad categories: radars (themselves) and sensors. In terms of its design, the main difference between them lies in the receiver chain. Radars usually have amplitude peak detectors whereas sensors preserve the signal for further analysis [
6], which can provide valuable additional information. In this sense, current GPR equipment should be reclassified as sensors.
Most GPR equipment meets the required bandwidth specifications for a UWB device using very short-time pulses or impulses transmitted at baseband (without an intermediate carrying frequency). That is why such GPR systems are called time domain systems. Each trace obtained by the radar is composed of distorted and attenuated versions (as a result of the propagation medium) of the pulse emitted by the antennas. Part of the antennas construction is still a process done mainly by hand and even antennas from the same company and with the same nominal frequency present, for instance, slight differences in terms of this emitted wavelet or in the radiation pattern. It is also important to notice that since GPR antennas usually consist of two dipoles – one for transmission and another for reception – the effective wavelet recorded will be dependent on the characteristics of both antennas, not just the transmitter.
Therefore, in order to make a good interpretation of the GPR records and extract as much information as possible from the signal recorded during processing, a deep knowledge of the type of emission used is essential because the characteristics of the detected reflections (length and shape of the reflected pulse, overlapping of constructive or destructive reflections, etc.) directly depend on the characteristics of the wavelet emitted by the antennas [
7,
8]. In addition, advanced processing techniques such as deconvolution or specific algorithms for target recognition require specific knowledge of this signal for proper operation.
Within the field of numerical simulation, it is also useful to work with the real source wavelet of the system. The goal of the simulation is to obtain a synthetic record very similar to that obtained in the field, which could aid in data interpretation. To provide practical results, modulation schemes in computer simulators should be able to incorporate, in addition to real antenna configurations and appropriate descriptions of the material properties, a precise model of the signal emitted by the antennas [
9]. However, the manner in which these modelling programs make use of this wavelet should be considered since there are several methods, some of which are somewhat complex to implement if the intention is to use a user-defined wavelet.
With all this in mind and an aim to deepen the knowledge of our system, we have designed a set of experiments with a relatively simple methodology so that they can be performed without specialized instrumentation. A basic knowledge of the radar operation and some additional simple tools are enough for a user to carry out the tests to study the particular characteristics, in the time and frequency domains, of the wavelet emitted by the antennas.
The characteristics of the emitted signal in the time domain are related to pulse shape, number of cycles or semi-cycles composing the pulse, temporal length, and geometric expansion. On the other hand, frequency analysis identifies characteristic parameters of the signal in this domain, such as the frequency spectrum used, relative power per spectral line, antenna central frequency, etc. Additionally, because the antenna consists of two dipoles, the direct signal between both is always present in any radar trace, so it is also interesting to study it in detail. For the realization of the proposed methodologies, a spacious laboratory has proven to be adequate to test medium- to high-frequency antennas as evaluated in this work.
3. Methodology
To obtain the direct signal between dipoles, the antenna under test (AUT) was placed in an environment free of any reflectors that might overlap the registration of this signal. The direction of maximum radiation of the dipoles pointed into the air, leaving a plane containing both dipoles parallel to the ground [
Figure 4(a)]. The antenna was subjected to a warm up time of 10 min before starting the tests. This warm up time consisted of keeping the equipment acquiring data triggered by time at fast speed.
A methodology often used to achieve a very approximate version of the emitted signal is to obtain a reflection of this wavelet from an object or surface [
18–
20]. Given its characteristics of linearity, homogeneity, isotropy, and non-dispersion, air becomes an ideal propagation medium in this case, ensuring that the waveform received is not affected by any of these effects that, to a greater or lesser extent, are present in all environments in which a study with a GPR is carried out. The main correction factor associated with this reflection is the geometric attenuation of the signal. This factor is also analyzed from the obtained results.
With this in mind, two simple tests were performed with the AUT. The first is shown in
Figure 4(c), where each antenna was placed against a wall that is covered with aluminium foil because its high conductivity (σ = 3.54 • 10
7 S/m) and reflection coefficient in air (0.99996), which is close to a perfect electric conductor (PEC). On the other hand, this ensures that the back reflection due to the wall-air interface will not overlap the first reflection. The dimensions of the PEC were 180 × 180 cm.
The scheme followed in the second test can be observed in
Figure 4(b). In this case, a reflector consisting of a metallic bar 3 cm in diameter was used. As can be appreciated, the dipole’s orientation is parallel to the axis of the bar, maximizing the intensity of the reflection in the E-plane. The metallic bar is placed just in the plane of separation between the two dipoles.
In the two tests discussed (PEC and metallic bar), instead of a single measurement, a series of successive measurements was carried out with a gradual separation between the antenna and reflector. In so doing, these tests could be used to characterize both the geometric attenuation of the signal and determine the minimum distance of overlap between the direct signal and a first reflector.
Finally, one last test was conducted over a paved area where a large pipe was known to be present [
Figure 4(d)]. The main goal here was to carry out a profile transversal to the longitudinal axis of the pipe in order to obtain a clear reflection of the emitted wavelet on the pipe.
5. Discussion
In this work, a methodological proposal has been presented with the main objective of determining the characteristics of the source wavelet emitted by a GPR ground-coupled antenna. In order to do that, three different tests were presented: two in air with a similar methodology but using a metallic bar in one case and an aluminium surface in another, and a test pipe buried in lossy dielectric. These results have been compared and demonstrate that it is possible to draw the following conclusions.
From the test results, it has been possible to obtain an approximation of the emitted wavelet. The analysis in time of the length and shape of the wavelets has allowed an approximate characterization of these signals. The frequency domain analysis has revealed the frequency ranges used by our antenna and its spectral power distribution. The type of methodology employed in the experiments is not a direct method, in the sense that what is obtained is not a direct version of the signal, but a reflection on an object or surface. In this sense, the work of [
21] shows that the difference between both methods is very small.
Regarding the analysis of the signal between dipoles, as can be observed in the results, there is a fairly good correlation between the direct signal and the others obtained from the experiments in air (
Figure 11). The best correlation is achieved when the air pulses are inverted, which is consistent with the change of polarity suffered by a propagating pulse that impinges on a metallic object or surface.
The minimum distances between the AUT and the surface to obtain a clear reflection of the latter without interfering with the first have been obtained. In this sense, two approximate separation levels have been established. In the first level (10 dB), the separation is almost complete and there is only an overlap between the components of the pulse of lower energy. From the second level (3dB), the higher energy components begin to overlap and it becomes very difficult to distinguish between the two signals. These distances can be used to get a clear reflection from the surface without aliasing and calculate the permittivity of the surface layer, which in turn leads to the detection of changes in the composition of this layer. Such a methodology has proven to be useful in the study of roads [
17,
22]. Here one must also take into account the loss in penetration depth of the antenna in addition to other effects discussed in the section of previous considerations.
To calculate the attenuation curves obtained from the amplitude peaks of the reflected signals in the PEC and metallic bar, the term 1/R
n, commonly used to characterize the geometric expansion of a wave front, was used. In the case of the PEC, a value of n = 2 has been used, which would be the theoretical value for a spherical wave and a plane reflector of this type (
Table 1). It should be remembered that this approach assumes that the object is far enough from the antenna, which explains why the adjustment with this equation is not entirely accurate. The values obtained are completely correlated in both experiments for the 500 MHz antenna. For equation adjustments, values within the overlapping area between the direct signal and the reflection have not been considered as shown in
Table 2. However, these amplitude values are used (after application of a moving average filter in two dimensions) to adjust a general exponential regression curve.
As discussed in the methodological section, the aluminium surface used as a reflector in the experiments has a size of 1.80 × 1.80 m. This size is insufficient to contain the entire footprint of the antenna when the antenna is located away from the reflector. In any case, the size of the PEC was chosen so that, for the maximum distance of the experiment (5.80 m), at least the first Fresnel zone of the antenna was contained. At this distance, the radius of the first Fresnel zone will be close to 1.80 m.
When isolating a specific reflected wavelet within the entire recorded trace with the objective of comparing it with other wavelets, it is important to establish a criterion on which parts of the trace can be considered part of this signal.
Figure 13 illustrates the problem where a target wavelet is just subsequently cropped tighter.
In
Figure 13, the sequence on the left represents the process by which the reflection has been isolated from one of the traces of the experiment on the PEC. The first image in the sequence has little reflection amplitude compared with the direct signal but from the second image is magnified because the direct signal has been erased and the scale self-adjusted to its maximum remaining amplitude. The sequence on the right represents the spectrum associated with the current trace, so one can follow its evolution as the reflection of interest is extracted from the recorded signal.
The last graph of the sequence in
Figure 13 represents the portion of the signal that seems to be associated with the reflection. The larger peaks are clear and easily distinguishable, but it is not so easy to determine the beginning and end of the pulse since the random noise in the trace has similar amplitude values than the lower peaks of the pulse. In this sense, the fact that the experiments are based not on a single measurement, but on a series of measurements at different distances, helps to clarify which part belongs to the pulse and which is due to the presence of random noise or clutter in a particular trace. Still, when comparing reflections, small changes in the criteria to establish the limits of this signal could give rise to non-negligible variations in the size of the pulse, bandwidth, and central frequency. To illustrate this effect,
Figure 14 shows the changes in the frequency spectrum of the pulse obtained in
Figure 13 as the pulse is confined within more stringent limits. Thus, when reflected pulses are isolated and compared as in this study, it is important to consider possible contributions of this effect on the results.
Experiments carried out with the buried pipe clarify the antenna behaviour in common materials of interest. There is a decrease of the central frequency and bandwidth associated with a widening of the pulse in time. As discussed before, this is because, first, the medium acts as a low-pass filter, so that the pulse broadens in time narrowing its associated frequency band, and, second, the proximity of the medium makes the antenna frequency spectrum move to lower frequencies. It also highlights the energy loss of the reflection in the metallic bar with respect to the reflection at the PEC for the 500 MHz antenna, possibly due to the low directivity of this antenna.