The monitoring of natural media and man-made structures at small scale, through ground based SAR (GB-SAR) interferometers, or more generally terrestrial radar interferometers (TRI), as it is referred to when not exploiting SAR techniques, has rapidly grown in the last decades. The most consolidated applications of this technique are the monitoring of open mines, natural processes (i.e., landslides, glaciers), and artificial facilities: For the reader’s convenience, an overview of this technique and its main applications is available in [1
]. These applications are based on interferometric techniques, where the main goal is to provide deformation maps, estimating the evolution of the surface kinematics under observation. Usually, in these measurements, the understanding of the scattering behaviour is mainly related to the amplitude information, which is analyzed for image interpretation purposes, and to select good points from the interferometric point of view, estimating the behaviour of some parameters as coherence (COH) and dispersion of amplitude (DA) [3
]. Almost the totality of terrestrial radar measurements discussed in the literature are acquired using linearly polarized (LP) sensors, and mainly using a single linear co-polar configuration, usually co-polar vertical (VV). It is well known that polarimetry can provide highly profitable information in different applications of radar imaging of natural media, such as rough, bare, and vegetated soil, snow, and glaciers: The availability of polarimetric data provided by recent SAR satellite missions has strongly pushed studies based on spaceborn, large scale, data [4
]. On the other hand, so far in GB-SAR imaging, this approach has been rarely used, and only a few case studies have been discussed in the literature [5
]. The main reason of this lack of research is probably due to the fact that commercial terrestrial radars do not provide multi-polarization capability, and they usually operate in single co-polar configuration (mostly vertical). In addition, the main use of these terrestrial apparatus is focused on interferometric processing for deformation measurements, where polarization features are of minor concern, and have not been deeply investigated so far.
The few studies where multi-polarization terrestrial radar data are analyzed are generally based on data acquisitions performed through laboratory prototypes and wide band sensors, not necessarily working at the Ku band, the spectrum allocation reserved to terrestrial interferometric systems [9
]. In [10
], for example, the authors showed some results of polarimetric acquisitions obtained through a terrestrial radar prototype, VNA based (details about this technical issue in [11
]), and operating at the C band to investigate the characteristics of terrain targets, using as a classification tool the H/A/Alpha polarimetric decomposition [12
], with encouraging results for improving the classifications of natural targets, thus providing a better understanding of their scattering mechanism. In a further paper [13
], an improved version of the same apparatus used in [10
] was calibrated through reference artificial targets, and 3D polarization-sensitive images acquired in different seasons were reconstructed from the acquired data, observing differences among the polarization signatures. In [14
], the authors investigated the polarimetric response at the X and C-band using a ground-based synthetic aperture radar, aiming at providing two and three-dimensional images of indoor and outdoor wheat canopy samples, respectively, to understand the scattering processes and the role of soil under the vegetation on the cross- and co-polarimetric response. In [15
], multi-temporal analysis of the polarimetric features of an urban environment using a ground-based dataset in the X-band was carried out. The temporal behaviour of the entropy, H, was estimated to provide a description of the polarimetric stability of the urban scenario. More recently, the upgrade of a commercial apparatus with polarimetric capabilities, the Gamma Remote Sensing’s GPRI-II, namely KAPRI, was tested [17
]. Other papers have been published where polarimetry is used for the specific goals of characterizing advanced reflectors for calibration [18
]. Circular polarization (CP) backscattering using a GB-SAR has been investigated in [19
], and calibration of the used apparatus was carried out with a rigorous approach in a controlled environment, and with reference targets. In particular, in [19
], the authors provide for the first time the results of a set of SAR laboratory experiments, where circular polarization acquisitions are discussed.
Anyway, circular polarization has been analyzed in a very few cases, and the interest to investigate the CP response of natural media, the topic of this paper, comes from the general features of this polarization, among which its ability to be less affected by the multipath effect is outstanding. It is well known that when the transmitted circular polarized wave is reflected from a surface, along a direction whose incidence angle is higher than the Brewster angle [21
], it turns the handedness of the rotating electromagnetic field. In Figure 1
, we resume the features that characterize the propagation of CP in the presence of symbolic targets.
After a bounce, the backscattered field propagates in the opposite direction with respect to the transmitted signal. The same rotation direction that would be described as “right-handed” circular polarization, RHCP, for the incident beam is “left-handed” circular polarization, LHCP, for propagation in the receiver direction, and vice versa. For this reason, in the case of a single bounce, a radar provided with the transmitting and receiving antennas with the opposite direction of rotation receives a strong signal. On the contrary, if the two antennas are co-polar, i.e., with the same direction of rotation, the detected signal is very low. When two bounces occur before reaching the receiving antenna, the situation is the opposite. In general, for an odd or even number of bounces, the situation is analogous; although due to the rapid decrease of the signal strength occurring after multiple bounces, the phenomenon is of minor concern. Therefore, by combining co-pol (RHCP and RHCP or LHCP and LHCP) and cross-pol (RHCP and LHCP or LHCP and RHCP) pairs of transmitting and receiving antennas, the contribution from even or odd bounds can be enhanced or minimized. Furthermore, using linear polarized antennas, the reflecting surface does not reflect the signal precisely in the same plane, therefore, the signal strength is weakened. Since circular polarized antennas send and receive in all planes, the signal strength suffers a minor decrease. In circularly-polarized systems, the reflected signal is returned in the opposite orientation, largely avoiding conflict with the propagating signal; the result is that circularly-polarized signals can better penetrate and bend around obstructions. Multi-path is caused when the primary signal and the reflected signal reach a receiver at nearly the same time; linear polarized antennas are more susceptible to multi-path due to the increased possibility of reflection.
Infrastructures and natural media often exhibit non-regular surfaces, and targets’ distribution, and multiple reflections frequently occur: CP measurements are theoretically expected to provide better performances in terms of phase stability and targets’ identification, due to the reduced multipath interferences which characterize this polarization configuration. Furthermore, in literature examples, the potential of CP measurements to detect targets with a specific shape can also be found, especially in studies based on planetary missions based on radar imaging sensors [22
]. On the other hand, the main drawbacks of the CP system are: (i) A lower efficiency in antenna gain, (ii) a complex design especially when CP is requested in a large bandwidth, and, finally, (iii) a lower diffusion and availability.
This paper reports some experimental results obtained acquiring data through a terrestrial radar, using different combinations of linear and circular polarized antennas. The polarization diversity is evaluated with an empirical approach comparing and analyzing the different responses, looking for the main differences between circular and linear combinations. The study introduces the topic, showing some tests carried out with simple targets using a pair of patch antenna arrays designed to operate at the Ku band frequencies reserved for TRI/GB-SAR use. The two circular polarized patch arrays specifically designed and developed for this study [24
] were mounted as transmitting and receiving antennas in a largely diffused GB-SAR commercial system, the Ibis-L manufactured by Ids. The antennas can be used, in principle, with any similar apparatus working in the same band, provided that the used radar sensor is capable to interface through standard connectors or waveguide flanges. The study tries to evaluate, on the basis of some experimental data, if the circular polarization response of natural media and artificial targets can improve the interpretation of the radar images, with respect to the standard co-polar VV configuration, commonly used in TRI. The goal is to investigate how different polarization combinations in terrestrial radar interferometry affect the coherence and amplitude dispersion of natural media, and in particular whether the circular polarization can improve the identification of stable scatters.
The paper is organized in five sections. After this general introduction, the instrumentation characteristics and the methods used to acquire the radar response with different polarizations are described. The following section describes the results of some experimental tests designed to estimate the polarimetric features of single targets and a heterogeneous scenario, including an urban area surrounded by vegetation. First, simple tests using the radar sensor in a real aperture radar (RAR) mode are carried out to assess the capability of the system to provide a sufficient polarization purity. Then, the results of a two-days experimental campaign, carried out in an area including a small village affected by a large landslide, and monitored using the radar in the GB SAR mode, are presented and commented on. Different combinations of receiving and transmitting antennas are used to evaluate the behaviour of some parameters of main concern to individuate coherent areas in radar interferometry. A discussion section is then dedicated to the main outcomes of the study. A conclusions section is dedicated to a summary of the paper content, and suggestions for planning further experiments.
4. Discussion of the Results
Based on the experimental data analyzed in the previous section, we can assess and discuss the following outcomes. First, the radar system arranged with the novel antennas, despite the lack of a rigorous polarimetric calibration, was demonstrated to be appropriate to provide a polarization capability to distinguish single and multiple bounce responses from reference targets. Although fully polarimetric high performing systems have been used by other researchers to investigate the response of natural surfaces ([15
]), so far, these systems have not measured circular polarizations. In this preliminary study, we demonstrated that the arranged system was adequate for attainment of the study’s goals. This was first demonstrated by the tests carried out in the RAR mode, measuring the response of the wall and of the light pole. Both range profiles shown in Figure 8
and Figure 10
, corresponding to the backscattering responses of these two targets clearly distinguish between different responses for co- and cross- polarized antennas. Nevertheless, it must be underlined that the CP antennas’ gain is low, and their efficiency is small, due to reasons already commented on in Section 2
. The three polarimetric configurations, i.e., the radar acquisitions with VV, co- and cross- circular polarization, were carried out with sufficient accuracy: The value of the 23 dB difference between the co- and cross response from the wall test gives an estimate of this sensitivity. Also, the lightpole data confirm that the co- and cross-pol CP responses significantly diverge, as expected: A difference higher than 20 db of SNR (see Figure 10
A) was achieved.
In the GB SAR data analysis, we focused on the radar response considering two parameters: DA and COH, which are the main parameters related to the statistical reliability of interferometric processing. High values of coherence and/or low values of DA are quality indexes related to the accuracy of the retrieved interferometric parameters. In the short time interval data set, all the tested polarization configurations showed a sufficient dynamic range to detect some outstanding targets whose response confirmed the expected difference between the single and double bounce effect. Although the goal to disclose a straightforward classification capability of the circular polarization responses was not sufficiently proven, some remarks can be made.
Among the two analyzed parameters, coherence appears to be better indicator for image interpretations. Indeed, by focusing on single targets (see Figure 17
and Figure 18
), specific behaviours where single and double bounces are separated (see Figure 19
) were found.
The relative phase between RR and RL was calculated and analyzed, but probably due to a significant atmospheric phase screen affecting the monitored area, no affordable remarks can be made, at least for the short time interval. On the other hand, the RR/RL amplitude ratio seems to provide useful information: The ratio was not homogeneous on all the structures, and the odd bounce modes prevailed. About the importance of this parameter, this confirms the results available from the literature coming from other application fields [23
]. The presence of an evident double bounce effect in the urban area was confirmed by a detailed analysis of the backscattering geometry: See Figure 19
and Figure 19
Regarding the long time series, where the co-pol LP (V) and the cross-pol CP were compared, the main outcome is that the DA of RL is quite uniform and assumes values similar to the areas subjected by relevant decorrelation in void areas, such as wood or grass, while COH RL is high on trees despite the large temporal interval for long series. Analyzing the two PDF distributions, it can be seen that the use of a circular polarization, despite a lower SNR, does not jeopardize the quality of the images; on the contrary, it provides a higher coherence on stable scatters.
On the basis of these preliminary experiments, we can assess that radar monitoring based on cross polarized circular antennas is able to provide detailed information about the scattering mechanism, which is not available with the conventional VV linear polarized observation. An important aspect to be considered is the low gain of the used CP antennas with respect to that of the LP antennas: Although in this study it did not jeopardize the measuring conditions, it affects the SNR of the measurements, and can be improved by upgrading the antenna design, or using different antennas with a higher gain. This technical issue must be taken into account in order to improve the measuring conditions in further experiments.