Sentinel-1 Active Radar Calibrator Dean Rowsell(1), Muthu Gandi(1), Trevor Stuber(1), Andrew MacNeill(1), D’Arcy Hart (1), Desmond Power(1),

Harry Jackson(2), Paul Snoeij(3), Björn Rommen(3), Ignacio Navas-Traver(3), Dirk Geudtner(3), Ramon Torres(3)

(1)C-CORE, Canada (2)Steppes House Services Limited, United Kingdom (3)European Space Agency, ESTEC, The Netherlands

INTRODUCTION

The Copernicus program, formerly known as Global Monitoring for Environment and Security (GMES), is set to establish a European capacity for Earth Observation and is the most ambitious program of its kind to date.

With its six Sentinel missions acquiring Earth data from space and a vast system of sensors on Earth, Copernicus will provide a unified system through which the data acquired are fed into a range of thematic information services. These are used to protect and manage land, the marine environment and the atmosphere, in order to provide emergency response and security, and to monitor climate changes.

Copernicus is an initiative led by the European Commission (EC) in partnership with the European Space Agency (ESA), which manages the space component, and the European Environment Agency (EEA).

The first satellite (Sentinel-1A) of the Sentinel-1 European Radar Observatory mission, a constellation of two polar orbiting satellites, providing robust datasets for the EU Copernicus Services, was launched on a Soyuz rocket from Europe's Spaceport in French Guiana on 3 April 2014. The two C-band radar satellites will provide continuous all-weather day/night imagery for land and ocean services.

To fully exploit the information provided by the Sentinel-1 SAR, the signal received from the earth surface must be analysed quantitatively. This requires the SAR system to be calibrated in such a manner as to directly relate image pixel intensity with a mean surface backscatter coefficient in the region represented by the pixel. Such calibration is termed a radiometric calibration, and must account for all temporal and spatial variations in the transfer function of the SAR.

Radiometric calibration must be performed as part of the normal operation of the SAR. The calibration process is divided into two components—internal and external calibration. Internal calibration provides an assessment of radar performance using internally generated calibrated signal sources, especially in the context of pre-flight testing. External calibration makes use of ground targets of known RCS to render an end-to-end calibration of the SAR system, thereby assessing the impact of those elements that are difficult, if not impossible, to assess using internal methods (e.g., antenna beam pattern).

External calibration methods can involve the use of: passive and precisely constructed targets, such as corner reflectors and spheres; natural terrain with known backscattering properties, such as the Amazon forest; or active radar calibrator. The latter class of calibration target is the Sentinel-1 transponder.

SENTINEL-1A ACTIVE RADAR CALIBRATOR

In its simplest form, an active radar calibrator, or transponder, comprises a receiving antenna, an amplifier and a transmitting antenna representing a constant RCS. The ESA transponder converts the analogue received signal to digital form, performs signal processing in the digital domain, and converts the signal back to analogue prior to retransmission. Such architecture has the advantage of being able to implement transponder functions that are otherwise difficult or impossible to achieve with purely analogue hardware [1, 2, 3].

For example, a precise, programmable delay, as used in the Sentinel-1 transponder is more easily achieved with a digital architecture. Further, transponder transfer function compensation across the signal bandwidth that accounts for temporal variations are more easily, if not solely, accommodated digitally—a very important factor in this application.

The transponders were developed and fabricated for ESA by C-CORE, Newfoundland, Canada and are currently installed at:

-­‐ ESA/ESTEC, Noordwijk, The Netherlands -­‐ Royal Netherlands Meteorological Institute, De Bilt, The Netherlands -­‐ National Aerospace Laboratory (NLR), Marknesse, The Netherlands

HARDWARE ARCHITECTURE

The hardware architecture is depicted in Fig. 1, comprised of a Sheltered Subsystem and an Antenna Subsystem.

Fig. 1. Hardware Architecture

The Sheltered Subsystem is housed within a NEMA-4X enclosure, which is further encased within an outside main frame consisting of tubular aluminium. The main frame provides a support base for the antenna subsystem, which sits atop this frame. The antenna subsystem comprises a number of support members for the antenna, the Antenna Interface Module and waveguide components, while at the same time paying close attention to the associated loading on the Pan-Tilt unit. All custom-fabricated aluminium components have a finish applied, suitable to withstand the environment. The entire physical assembly is designed to mount on an adaptor plate, which is pre-installed at each transponder site prior to shipping the transponders.

Sheltered Subsystem components

1. Microwave Transceiver – This subsystem is custom designed and fabricated at C-CORE from connectorised microwave components and a custom power supply. Its role is to down-convert the SAR signal from 5.405 GHz to baseband (75 MHz centre); up-convert the processed SAR signal from an IF of 150 MHz (produced by the DAC) to 5.405 GHz; and, generate all local oscillator signals (LOs) required within the transceiver, and required externally for sample clock and test references.

2. Host Server - The host server is the central control platform for the entire transponder. The host application

and HMI resides on this platform. The server chassis houses three modules. The Digital Signal Processor (DSP) provides ADC and DAC capabilities and the entire signal processing function in-between. The signal processing function comprises system compensation, delay and target strength. The I/O Controller exercises digital control over various elements of the transponder and provides an interface to the server for digital and analog feedback for diagnostic purposes. The GPS Clock synchronizes the Host Server internal real-time clock with Coordinated Universal Time (UTC) for accurate scheduling of calibrations, and provides a means to accurately time-stamp the acquired azimuth pattern.

3. Power Supply - The power supply provides DC power to the Microwave Transceiver and the Antenna Subsystem. The power supply is controllable from the Host Server, enabling shut-down of these modules. Associated with the power supply is an uninterruptible power supply (UPS). The UPS is the backup power supply that will continue to power the Transponder for a limited time in the event of a mains power failure to enable the Server to shut down the Transponder in a controlled manner and then shut down the Server and UPS module.

4. External Communications – A link is provided via fibre optic cable to site facility Internet communications. 5. Environment Control - The environment control module provides heating and air-conditioning sufficient to

maintain an operating temperature within the shelter of 10-35°C. It also provides a dehydrator to pressurize the Waveguide Assembly and Antenna to inhibit internal corrosion.

Antenna Subsystem components

1. Antenna - The antenna is a Potter Horn, fabricated by Thomas Keating Ltd., which exhibits exceptional side-lobe response, mitigating any negative effects of backscatter for surrounding terrain or objects.

2. Antenna Interface Module (AIM) - The AIM provides a signal path between the antenna and microwave transceiver with a switchable loopback function for self-calibration. Diagnostic functions are also provided by the AIM. The Waveguide Assembly is included in this module, consisting of waveguide components and switches for directing the microwave signal from the transceiver to the antenna.

3. Pan-Tilt Unit - The Pan-Tilt Unit provides the antenna pointing capability for the transponder.

TRANSPONDER CALIBRATION

The general approach is to establish an internal loopback reference response to which the transponder is aligned, through filter H1, during external calibration and at any point in the future. During external calibration, an internal calibration aligns the internal loopback response to this reference. The external response (with the disk) is then obtained, and compared to the ideal transponder response (RCS = 70 dBm2 across the full band). The residual essentially portrays deviation in the response of the portion of antenna subsystem outside the internal calibration loop, as compared to its expected response. This residual is fixed for all future time, until such a time that the external calibration is repeated, and a new residual obtained. The residual is added to H1, following any future internal calibration, to establish the desired RCS across band. Fig. 2 shows the model used to describe the transponder calibration.

Internal Self-Calibration

Transponder self-calibration takes place as part of normal transponder operation, typically immediately before and after a SAR calibration. Self-calibration determines the change in the transponder transfer function from external calibration to the present point in time. Self-calibration exploits a waveguide loop-back in determining this change, as well as the thermal characteristic of the circulator.

Fig. 2. Transponder Calibration Model

All gain elements can be considered system transfer functions that model the gain magnitude and phase over the bandwidth of the transponder. When performing a loopback, the goal is to precisely characterize the transfer function of the loopback path in order to provide compensation via a system compensation filter. The digital loopback chirp pulse is generated by the processor and converted by the DAC. Upon capturing the signal from the loopback path, through the ADC, a compensation filter is derived that re-instates the exact loopback response that prevailed during external calibration.

When streaming and processing a signal from an external source (i.e. during SAR calibration or external calibration) the processor is configured to apply the compensation filter derived to the digitized signal. The correct gain is applied to the received signal in order to impart the desired transponder RCS. Internal self-calibration ignores the effects of elements external to the loopback path, which are addressed in the external calibration and gain error budget.

External Calibration

The external calibration is carried out with the gain set-up described above. Although it is carried out at multiple range positions to evaluate multipath, the responses obtained at these positions are integrated to render a single response. So for the purposes of this discussion, a single range position and response will be presented. Fig. 3 shows views of the external calibration test setup.

Data was collected at each of 33 rail stops comprising 32 equally spaced 0.25 m intervals, for a total length of travel of 8.0 m. The exception is for transponder #2, where only every second rail position was evaluated, for a total of 17 rail stops comprising 16 equally spaced 0.5 m intervals. At each rail stop, three data sets, or scenes, were collected—1st background, target, and 2nd background. For background data, the target was pointed as far as possible from the transponder in azimuth, in a clockwise direction.

Each data set comprises 20 binary files derived from 20 independent triggers—each data file comprising 10 FIFO cycles. There are actually 110 FIFO cycles programmed into the transponder, but only the last 10 cycles are retained. This is to provide for a stabilization of the power amplifier response in cases where the power amplifier is gated. This will have no bearing on the case where the power amplifier is continuously on. However, the same methodology is followed in both cases for consistency.

Fig. 3 External Calibration Test Setup Transponder on rails (top left) pointed towards external calibration disk on roof of four-storey building (~19 m height from ground) (top right and bottom).

A single FIFO cycle contains 150 distinct pulses, each at a single frequency, covering a 150 MHz bandwidth in 1 MHz steps. Each pulse is 300 nsec, or 45 samples, in duration. A PRF of 150 kHz is established by separating the starting edge of each pulse by 1000 samples. A FIFO cycle contains 150,000 samples in total—therefore, 150 pulses. A full calibration package (with 33 rail positions) contains 2.97 Gsamples, or 11.88 GB, of raw data.

Data analysis

It is impractical to ingest such a large calibration package into Matlab. Therefore, some means of paring down the data is needed for management purposes. Each pulse is 45 samples in duration, but along the full length of the rails, the pulse delay varies by exactly 8 samples (1 sample per meter). Ignoring the pulse edges, a window of 25 samples would be adequate to capture a sufficient portion of the pulse at any position on the rails, using a fixed start index. This reduces the memory requirement by a factor of 20, to just over 1 GB, which becomes manageable. Further disk savings is attained by storing the data in a compressed Matlab .MAT format, resulting in a storage requirement of about 225 MB. The end result is three 5-D arrays—one for each of 1st background, target, and 2nd background. Each array is a 10 X 20 X 25 X 150 X 33 (or X 17) sample array.

The average of the 10 FIFO cycles and 20 trigger events is first computed. A polar average is used for this purpose as this preserves the amplitude in the presence of any phase variation across each sample. The resulting matrix comprises three dimensions containing 25 samples, 150 frequency steps, and 33 (or 17) rail positions. All further data

manipulation is carried out in a coherent manner on the complex samples. Amplitude results are presented by averaging the magnitude of all samples comprising 25-sample pulse window.

Background data was obtained for the purposes of coherently subtracting the background from the target-plus-background data, thereby rendering only target information (i.e., clutter reduction). This is only practical if the background remains stable between the background, and the target-plus-background scenes. This will be less than perfect due to actual changes in background (e.g., moving vehicles, people, or other scatterers), and phase instability over the course of the experiment at each rail position (it typically takes a few minutes to re-align the disk and horn between data sets). To mitigate these effects, two backgrounds were obtained before and after the actual target acquisition. This provides some options for background reduction. First, one of the backgrounds may be corrupt due to moving scatterers, and can thus be eliminated from the analysis. Second, in the event of any phase drift between scenes, averaging the 1st and 2nd backgrounds may present a phase average that better coincides with the background phase during the target-plus-background acquisition (which takes place between the two backgrounds). For the following analysis, the default procedure averages the two backgrounds before subtraction from the target scene. In cases where exceptions need to be made (which may be specific to just a single rail position), these exceptions will be noted.

Fig. 4a shows the average target amplitude over the pulse window. Fig. 4b and Fig. 4c shows the average background level over the pulse window for the 1st and 2nd backgrounds respectively. For each graph, the surface was integrated (using absolute, not dB values), yielding the mean amplitude of each surface, as recorded in each Fig.. Comparing the mean target amplitudes to the mean background amplitudes, one can see that there is a difference of about 32 dB. So in an integrated sense, the target levels are not sufficiently higher than the background levels to yield sufficient certainty in the target level estimation. A background rejection of at least 40 dB is required for this purpose. However, by subtracting the two backgrounds from one another, one obtains an estimate of the achievable background rejection. Here, an additional background rejection of 14 dB is observed. Therefore, one expects that by subtracting a background scene from a target scene, 46 dB of background suppression can be achieved. This is in an integrated sense, and not applicable to each and every frequency/range bin.

Fig. 4d shows the background-suppressed target data. Here, the background suppression is very evident in comparison to Fig. 4a. The range and frequency effects on the data are also clearer in this result. The next step is therefore removing range and frequency artefacts that are not associated with the transponder response. These are a result of the transponder position along the rails, and near-field calibration disk effects. Once these effects are removed, the remaining surface is representative of the transponder response, from which a correction for the transponder response can be obtained.

The transponder position along the rail affects the signal level, so a reference range corresponding to the furthest rail position is chosen, to which the signal levels at all other rail positions are transformed. To carry out the signal transformation, a range transformation is first required to convert each rail position to range-to-disk, as the rail and the line to target are not parallel; subsequently the signal level is adjusted accordingly.

The calibration disk exhibits a varying RCS within the near field, and thus, a correction is needed in this respect. Correction numbers for the disk RCS as a function of range and frequency were based on accurate EM calculations.

The correction numbers are interpolated and mapped to the actual frequency domain of the calibration data, and ranges mapped on a line between the antenna and the disk (as derived in the previous section).

The next step is to consider all of the frequency and range bins, each comprising a 25 sample pulse window. To obtain a magnitude/phase representation for each bin, an FFT is performed. The FFT size is carefully chosen as the least common multiple of the number of samples per bin (25) and the number of frequency bins (150). This will avoid any leakage into adjacent FFT bins by ensuring each FFT bin is exactly centred on a frequency bin. Otherwise, ripple will result over the response surface.

Fig. 4a Average Target Amplitude for Transponder 1 (84.69 dB)

Fig. 4b Average 1st Background Amplitude for Transponder 1 (52.44 dB)

Fig. 4c Average 2nd Background Amplitude for Transponder 1 (52.42 dB)

Fig. 4d Average Target - Background Amplitude for Transponder 1 (84.66 dB)

The spectral characteristics for each range and frequency bin, previously obtained, are analysed to determine the peak magnitude response. A transfer function for the external loop is established by dividing the response by the data that was applied to the loop through the delay FIFO, and applying the corrections from the previous two sections. This transfer function also incorporates the RCS margin that reflects the excess loop gain owing to the combination of disk size and range. The external loop responses of transponder 1 with the RCS margin, range correction and the disk near-field correction applied is shown in Fig. 5. The response clearly shows the residual frequency dependency that is primarily due to the antenna.

Fig. 5 External Loop Transfer Function of Transponder 1

Investigation into the response surfaces for each of the three transponders reveals a close-to-planar response surface. Therefore, to evaluate the slope of the response in the range direction (which should ideally be zero), a planar fit was carried out. First however, it is observed that the band edges seem to have a more dramatic range effect than the remainder of the surface. Therefore, the planar fit excludes all portions of the surface outside a 90 MHz bandwidth. The planar fit is carried out in a least-squares sense. Data-tips are next used to extract the frequency and range slopes from the surface.

Across the three transponders, an average frequency slope of 1.02 dB was found. The antenna calibration data showed a slope of about 0.5 dB over the same frequency range. Multiplying this by a factor of two in consideration of two-way travel through the horn, the expected slope is 1.00 dB. The variance in slope across the three transponders is attributed to variance in waveguide assembly (other than the horn). This is plausible considering the number of waveguide components and flange interfaces in this assembly, in addition to the circulator.

The magnitude of the response surface is averaged along the range dimension to generate the range-independent magnitude response of the external loop. The inverse of this magnitude response is then calculated to render the frequency response of the desired correction.

Correction Filter Calculation

The corrections derived in the previous section must be converted to an impulse response of an FIR filter, which will be convolved with that of the internal loopback correction, to render the final correction filter. This section describes the process and results of converting the external calibration corrections to impulse responses.

Since the correction derived above is an amplitude correction, a frequency sampling-based methodology is employed for the correction filter.

The first step is to eliminate the undesired portion of the response. Due to the ripple at the 100 MHz band edges, a 90 MHz band is chosen for this purpose. The mean in-band response is then calculated and used as an interpolation point at 0 Hz.

Based on the 150 frequency samples an impulse response for the correction filter is calculated. The correction filter is then shifted to baseband, and since it is now in complex form, it is decimated to a rate of 150 MSPS, matching the baseband rate in the digital processor. By way of example, the TR2 response is shown below.

Fig. 6 Correction Filter Response of Transponder 2

CONCLUSIONS

The Sentinel-1 transponders are a set of three high-precision active radar calibrators for external characterization of the Sentinel-1 C-Band SAR instrument. The transponders provide the capacity to mimic a fixed target with a precisely calibrated radar cross-section. This paper presented an overview of the hardware architecture and the calibration strategy and final implementation of the transponders.

IN MEMORY

D’Arcy Joseph Hart (56) suddenly died on August 7th 2014. In 2009 D’Arcy became the project manager for the Sentinel-1 transponder project at C-CORE, where he was responsible for the design, development and installation of the three calibration devices for the Sentinel-1 radar satellite launched in April 2014. It is very sad that D’Arcy could not enjoy the very impressive results from both the transponders and the satellite.

REFERENCES

1. A. Woode, Y-L Desnos and H. Jackson, “The development and first results from the ESTEC ERS-1 ARC unit”, IEEE Transactions on Geoscience and Remote Sensing, Vol. 30, No. 6, Nov 1992.

2. R.K. Hawkins, L.D. Teany, S. Srivastava and S.Y.K. Tam, “RADARSAT Precision Transponders”, Adv. Space Res., Vol. 19, No.9, pp. 1455-1465, 1997.

3. H. Jackson, I. Sinclair, and S. Tam, “Envisat-ASAR Precision Transponders”, CEOS SAR Workshop, 26-29 October 1999, ESA-SP4508.