1. Introduction
Synthetic Aperture Radar (SAR) systems are microwave remote sensors that are mounted on board moving platforms in order to obtain high spatial resolution in the along-track direction by emulating the acquisition mechanism of large-aperture antennas [
1,
2]. Very common platforms used to mount SAR systems are satellites [
1,
2,
3,
4,
5,
6] (spaceborne systems), airplanes [
7,
8,
9,
10], helicopters [
11], and, more recently, drones [
12,
13] (aerial systems). Due to their peculiarities, spaceborne and aerial SAR systems are, to some extent, complementary.
Spaceborne systems guarantee very wide spatial coverage. However, they are forced to follow polar orbits, thus flying practically only along the South–North (or North–South) direction. This poses some limitations to the full exploitation of those techniques, such as Differential SAR Interferometry (DInSAR) or Along Track Interferometry (ATI), which allow us to measure only the Line of Sight (LoS) component of the remotely sensed phenomenon. Moreover, with the currently operative spaceborne SAR constellations [
3,
4], the revisiting time, that is, the time interval elapsing between subsequent observations of the same area, is on the order of several days. This makes it impossible on the one hand to illuminate the area of interest in a timely way in case of emergencies, and on the other hand to monitor the evolution of the remotely sensed phenomenon through daily or hourly observations.
On the contrary, aerial systems guarantee narrow spatial coverage. However, they can fly in any direction and, at least in principle, whenever required. Accordingly, they allow us to reach the area to observe in a timely way, and to reduce the revisiting time to a few minutes. Furthermore, aerial systems can use antennas much smaller than those installed on spaceborne platforms, due to the significantly reduced distance from the observed targets. This ensures a geometrical resolution in the along-track direction higher than that achievable with the spaceborne systems [
1,
2]. In addition, when high-frequency bands (like Ka, Ku, X, and C) are employed, aerial platforms allow us to easily obtain effective single-pass InSAR configurations [
8,
9,
10,
12,
14,
15,
16,
17,
18,
19]. Indeed, considering the usual flight altitudes of aerial platforms, for wavelengths on the order of a few centimeters (or less), the interferometric height of ambiguity [
1,
2] can be kept sufficiently low with baselines [
1,
2] as small as required by the geometrical constraints imposed by small aerial platforms. In this regard, it is recalled that aerial single-pass InSAR configurations are particularly attractive for two reasons. First, because (like any single-pass configuration) they allow for circumventing temporal decorrelation effects [
1,
2]. Second, because they allow us to strongly mitigate in the interferometric products the influence of the so-called residual errors which are typical of aerial SAR focused images due to the unavoidable inaccuracies of the navigation data used during the image formation procedure [
20].
The complementary peculiarities of spaceborne and aerial SAR systems have led in the last years to two contrasting trends, which have somehow driven the technological development of these systems.
The first trend answers to the requirement of illuminating larger and larger areas. To do this, spaceborne SAR systems are appropriate. In this case, the technological challenge to face consists of the implementation of advanced acquisition modes, such as ScanSAR [
1,
2,
21,
22], TOPS [
2,
5,
22,
23], and/or advanced optimization strategies, such as the digital beam-forming on receive technique [
24], which are aimed at widening the across-track (XT) coverage achievable with the more conventional Stripmap mode [
1,
2]. This trend, of course, enables the growth of novel methods of data reduction and analysis by artificial intelligence (AI), due to the sharply increasing spaceborne SAR data volume [
25,
26,
27,
28].
The second trend instead follows the need for guaranteeing fast and flexible monitoring, possibly at high resolution, of confined areas. For this purpose, aerial systems are appropriate. In this case, the technological challenge to face consists of the reduction of the size, weight, and realization costs of the developed SAR system. In this frame, beside the conventional pulse radar systems, Frequency-Modulated Continuous-Wave (FMCW) [
29,
30] is emerging as a very attractive solution. Indeed, unlike the pulse radar systems, which require high peak transmission power, the FMCW systems operate with constant low transmission power. In addition, the sampling frequency of the Analog-to-Digital Converter (ADC) of the FMCW SAR systems can be significantly smaller than the bandwidth of the transmitted signal. On the other hand, the operating principle of the FMCW SAR limits the maximum detectable sensor-to-target distance to a few kilometers, which is safely acceptable for the acquisition geometry of several aerial platforms, especially for the small-sized ones, which typically fly at very low altitudes. Summing up, the FMCW SAR systems are particularly tailored to small aerial platforms, since their architecture complexity, which is lower than that of the pulse SAR systems, involves a reduction of size, weight, and realization costs.
In this frame, the Institute for the Electromagnetic Sensing of the Environment (IREA) of the National Italian Research Council (CNR) has recently signed an agreement with “Elettra Microwave”, which is a small Italian company, for the scientific use of a novel single-pass interferometric airborne FMCW SAR prototype realized by the company. The system is named AXIS, which stands for Airborne X-band Interferometric System. Like any conventional FMCW radar, it operates in a bistatic configuration. Therefore, to obtain a single-pass interferometric layout, it mounts three radar antennas: one transmitting (Tx) and two receiving (Rx).
In this work, we present a first assessment of the imaging and topographic mapping capabilities of the AXIS system. To do this, we show the results relevant to the acquisition campaign carried out over the Salerno area, South of Italy, in 2018 just after system completion. In particular, during the campaign, a number of Corner Reflectors (CRs) were deployed within the area illuminated by the radar, and very accurate measurement of their positions through the Differential Global Positioning System (D-GPS) technique was carried out to provide a set of sound reference ground points. This allowed a first assessment of the quality of the focused SAR images and the Interferometric SAR (InSAR) products achieved with the AXIS system. More specifically, in correspondence with this set of reference points, the geometric resolution and the planar positioning accuracy of the focused AXIS images were measured. Then, a comparison between the DGPS measurements of the CRs’ positions and the Digital Elevation Model (DEM) generated with the single-pass InSAR AXIS data was carried out.
The presented analysis aims at providing first reference values for future research and operational activities that will be conducted with this sensor.
The work is organized as follows. In
Section 2 we provide a brief description of the system. The acquisition campaign, the processing chain applied to the SAR data, and the achieved results are described in
Section 3. The concluding remarks are reported in
Section 4.
3. Experimental Results
In this section, we present the results relevant to the SAR data acquisition campaign carried out using the AXIS system over the Salerno area, Italy, in April 2018. More specifically, a flight campaign consisting of six overlapping flight circuits was scheduled, each of them containing two antiparallel linear tracks of about 20 km. In other words, we planned to collect SAR data from 12 flight tracks: 6 overlapping tracks from southeast (SE) to northwest (NW) and 6 overlapping tracks from NW to SE. In
Figure 7, which shows an optical image of the test area, the actually flown circuits (obtained through the navigation data recorded during the overall campaign) are depicted with yellow lines.
Besides the flight campaign, we also performed a ground campaign aimed first of all at measuring the antennas’ lever arms, in order to provide the information necessary to accurately process the radar data [
35,
36]. As specified above, such measurements were carried out very precisely through a Total Station Theodolite. The ground campaign was also aimed at providing a number of sound ground control points to assess the quality of the obtained SAR images and interferometric products. To do this, 10 CRs (5 for the NW–SE and 5 for the SE–NW flight tracks) were deployed over the area illuminated by the radar, and their positions were accurately measured by means of D-GPS surveys.
The main SAR acquisition parameters are summarized in
Table 4. We recall that the minimum recordable range for an FMCW system is 0 m. In our case, we focused only a range portion of the overall acquired data by setting a near (slant) range equal to about 2478 m, corresponding to a (mean) look angle of about 20 degrees.
In
Figure 8, a block diagram of the adopted InSAR processing chain is depicted. In particular, the range compression of FMCW SAR data simply requires a Fourier transform of each range line [
30]. The azimuth compression step was carried out through a time-domain Back Projection (BP) strategy [
35,
36] by exploiting the information provided by the measured antennas’ phase centers and lever arms, the navigation data and an external DEM, namely the SRTM one [
37], of the observed area. The adopted processing strategy allowed us to avoid application of the approximations [
38] necessary to implement frequency-domain focusing approaches with integrated motion compensation [
39,
40]. This, of course, involves an increase of the computational burden, but this can be managed by means of parallel computing strategies that, for the time-domain approaches, are very easy to implement. Moreover, like all the time-domain focusing algorithms based on BP approaches, our processing strategy allowed us to focus all the SAR images in a common output grid, thus avoiding the need to apply the co-registration step [
1] to generate the SAR interferograms. Except for this latter processing step, standard InSAR processing [
1] was applied for the generation of the interferometric products. Hereafter, we focus our attention on a single-pass interferometric dataset related to a flight track flown from SE to NW. More specifically, we first analyze the obtained amplitude SAR images and then the interferometric products.
Figure 9 shows the amplitude of the multi-look complex (MLC) SAR image relevant to the whole track. The image is focused in an output grid coincident with the radar one (that is, slant range and azimuth). Note that in the right vertical axis of the figure is specified the (mean) look angle corresponding to the range coordinate reported in the left vertical axis. We remark that in the figure, a 10 range × 10 azimuth pixel averaging window was applied for visualization purposes, obtaining 15 m × 16 m pixel spacing. The details of the main processing parameters are reported in
Table 5. From the figure we note amplitude decay for small values (approximately less than 25 degrees) and high values (approximately greater than 60 degrees) of the (mean) look angle. This is in agreement with the NESZ curves of
Figure 5 (we recall that for the considered flight altitude, the region of our interest in
Figure 5 lies between the red and yellow curves).
Figure 9b instead reports a multi-look amplitude SAR image of a small area around Salerno’s airport (see the light blue box in
Figure 9a), which was processed at a higher resolution (see again
Table 5). Note, in particular, that a 2 range × 10 azimuth pixel averaging window was applied in the figure for visualization purposes, obtaining 1.5 m × 1.6 m pixel spacing. All the five deployed CRs relevant to the SE–NW track are present in this area; they are highlighted in the figure with red circles.
Processing parameters relevant to both images are listed in
Table 5. For both images, the output grid coincides with the radar one (range–azimuth). Starting from the single-pass data pair focused with the processing parameters of
Figure 9a, we obtained the wrapped interferogram and the corresponding coherence map shown in
Figure 10a and
Figure 11a, respectively. Note that, as in
Figure 9a, a 10 range × 10 azimuth pixel averaging window was applied to generate these two maps.
Figure 10b and
Figure 11b instead report the interferogram and the corresponding coherence map obtained starting from the single-pass data pair focused with the processing parameters of
Figure 9b. As in
Figure 9b, a 2 range × 10 azimuth pixel averaging window was applied to generate these two maps. It is stressed that to obtain the interferometric products shown in
Figure 10 and
Figure 11, besides the averaging window described above, we did not apply any additional filter aimed at limiting the noise effects. It is finally noted that in all the interferograms shown, we removed the topographic component provided by the external SRTM DEM used during the focusing step.
Starting from the wrapped interferograms of
Figure 10, we applied the phase unwrapping procedure from [
41,
42]. Then, we estimated the resulting unknown phase offset present in the unwrapped interferograms by applying the Phase-Based Estimate (PBE) procedure detailed in [
33,
43,
44] and exploiting the D-GPS measurements relevant to the CRs. Thereafter, we carried out the phase-to-height conversion [
1] to generate the InSAR DEM. The result relevant to the low-resolution interferogram of
Figure 10a is displayed in
Figure 12 on a geographic grid and superimposed upon an optical image of the overall test area.
A quantitative assessment of the presented SAR products was performed by exploiting the five CRs shown in
Figure 9b, along with the in situ D-GPS measurements relevant to their positions. More specifically, we carried out three different experiments by exploiting the high-resolution SAR image and interferogram generated with the processing parameters considered in
Figure 9b and
Figure 10b.
In the first experiment, we measured the geometric resolution of the Single Look Complex (SLC) image in correspondence with the five CRs. The achieved results are listed in
Table 6. We recall that the expected values are 0.33 m in azimuth and 0.75 m in range (see
Table 5). In particular, we measured a mean azimuth resolution of 0.36 m with a standard deviation of 0.008 m, and a mean range resolution of 0.74 m with a standard deviation of 0.04 m. Note that no filtering (such as Hamming [
1]) aimed at reducing the side lobe level of the point spread function (PSF) was applied. It is evident from
Table 6 that the measured resolutions are very close to the expected theoretical ones. Moreover, it can also be noted that in most cases (with the exception of CR3) the measured range resolution is finer than the expected theoretical one. This is due to the fact that we chose the output geometry of the focused image according to the mean antenna pointing direction along the azimuth direction. Accordingly, the output geometry (also named processing geometry in the literature [
45]) may be different from the acquisition geometry dictated by the actual antenna pointing direction within the azimuth aperture from which a generic target is illuminated. When this happens, the two-dimensional PSF relevant to the target is rotated with respect to the output grid [
45]. As a consequence, measuring the resolution of the two-dimensional PSF along the azimuth and range directions of the output grid leads to an apparent improvement of the lower resolution (in our case, the range one; see
Table 5) and an apparent impairment of the higher resolution (in our case, the azimuth one; see
Table 5).
As a second experiment, we measured the planar (that is, in the azimuth and range directions) positioning accuracy of the Single Look Complex (SLC) image in correspondence with the five CRs. To do this, we applied the backward geocoding procedure [
1] to the D-GPS positions of the CRs, thus calculating their expected azimuth and range coordinates in the considered SAR image output grid. Then, we compared these coordinates with those of the CRs imaged in the SLC image. The range and azimuth misalignments measured for all the CRs are listed in
Table 6. In particular, we measured a mean azimuth misalignment of 0.39 m with a standard deviation of 0.04 m, and a mean range misalignment of 0.26 m with a standard deviation of 0.10 m. It is noted that in the range direction the measured mean misalignment is lower than the resolution, whereas in the azimuth direction it is comparable to the resolution.
As a third experiment, we provided a first estimate of the vertical accuracy of the AXIS InSAR DEM. To do this, we compared the height values achieved on the AXIS InSAR DEM, in correspondence with the CR positions, with the D-GPS ones. The results are again collected in
Table 6. In particular, we measured a mean vertical error of 0.20 m with a standard deviation of 0.41 m.
4. Discussion
A discussion concerning the presented results is now addressed.
In the previous section, to carry out the analysis of the AXIS performances, we have exploited five CRs properly deployed within the illuminated area and the D-GPS surveys of their positions. In this way, we have obtained a number of ground measurements that may be considered with good approximation as the ground truth.
In correspondence with this set of ground reference points we first measured the geometric resolution of the focused AXIS images. In particular, for the SLC image we have obtained mean geometric resolutions of 0.36 m (azimuth) × 0.74 m (range), with standard deviations of 0.008 m (azimuth) and 0.04 m (range). These values turned out to be practically the same as the expected theoretical ones.
Then, we measured the planar (that is, in the azimuth and range directions) positioning accuracy of the focused AXIS images, obtaining for the SLC image a mean positioning misalignment of 0.39 m (azimuth) × 0.26 m (range) with standard deviations of 0.04 m (azimuth) and 0.10 m (range). Finally, we compared the D-GPS measurements of the positions of the CRs with the height values of the single-pass InSAR AXIS DEM in correspondence with the imaged CRs, obtaining a mean vertical error of 0.20 m with a standard deviation of 0.41 m. From these results it turns out that some systematic errors seem to affect the obtained measurements. These errors are, however, tolerable, since the mean range misalignment (0.26 m) is less than the range resolution and spacing (0.75 m) of the system; the mean azimuth misalignment (0.39 m) is comparable to the azimuth resolution (0.33 m) of the system; and the mean vertical error is just 0.20 m. It is likely that these three systematic effects are somehow related. Given the amount of the mean range bias, one possible explanation could be the presence of an uncompensated internal radar delay. In any case, further investigations on this issue are a matter of current work.
Summing up, the measured imaging and topographic mapping capabilities of the overall AXIS infrastructure (which consists of the radar system along with the complete data processing chain that leads from the acquired raw data to the generated SAR and InSAR products) well match the theoretical, expected ones.
Moreover, the imaging and topographic mapping capabilities of this low cost, compact and flexible FMCW system are comparable to those achieved using well-assessed, although more expensive, pulsed airborne X-band SAR systems, like, for instance, InSAeS4 [
9]. Indeed, in [
9], starting from an analysis similar to that addressed in this work for the AXIS system, it was measured for InSAeS4 a mean geometric resolution of 0.14 m (azimuth) × 0.49 m (range), a mean positioning misalignment of 0.08 m (azimuth) × 0.04 m (range) with a standard deviation of 0.07 m (azimuth) and 0.08 m (range), and a mean height error (of the obtained single-pass InSAR DEM) of −0.08 m with a standard deviation of 0.51 m.
5. Conclusions
In this work, we presented a first assessment of the imaging and topographic mapping capabilities of the AXIS system, which is a newborn single-pass interferometric airborne FMCW SAR system developed in the frame of cooperation between a public research institute (IREA-CNR) and a private company (Elettra Microwave S.r.l.).
In particular, we showed results relevant to an acquisition campaign carried out over the Salerno area, South of Italy, in 2018, just after the system completion. More specifically, we provided a first quantitative assessment of the quality of the focused SAR images and InSAR products achieved using the AXIS system. To do this, we exploited a number of CRs properly deployed within the area illuminated by the radar and D-GPS surveys of their positions.
More specifically, in correspondence with this set of ground reference points we measured the geometric resolution and the positioning misalignment of the focused AXIS images, obtaining for the SLC image, mean geometric resolutions practically equal to the expected theoretical ones, a mean range misalignment smaller than the resolution and a mean azimuth misalignment comparable to the resolution. Moreover, we compared the D-GPS measurements of the positions of the CRs with the height values of the single-pass InSAR AXIS DEM in correspondence with the imaged CRs, obtaining a vertical error on the order of dozens of centimeters.
The presented results, aimed at providing first reference values for future research and operational activities that will be conducted with this system, already show that the imaging and topographic mapping capabilities of the AXIS system well match the theoretical, expected ones. Moreover, they are comparable to those achievable using well-assessed, although typically more expensive, pulsed SAR systems. More generally, the presented results show that the AXIS infrastructure (which consists of the radar system along with the complete data processing chain from the acquired raw data to the generated InSAR products) may represent an appealing monitoring solution for those applications that require the use of high-resolution InSAR products.