RADIOMETRIC CALIBRATION OF THE X-SAR SYSTEM
Final Report
Manfred Zink Co-Investigators:
German Aerospace Research Establishment, DLR F. Heel DLR
Institute for Radio Frequency Technology Ph. Hartl University of Stuttgart
82230 Oberpfaffenhofen, P.O. Box 1116, Germany
Phone: +49-8153-28-2386, FAX: +49-8153-28-1449
email: manfred.zink@dlr.de
OBJECTIVES
The main objectives of this experiment are to improve the radiometric calibration and to assess the overall accuracy of the X-SAR products. This includes:
the monitoring of the sensor electronics,
the validation of the inflight antenna pattern using ground calibration receivers,
the estimation of the actual accuracy of the shuttle attitude information,
the verification of the antenna beam alignment of all frequencies and polarizations,
the determination of the absolute calibration factor from the image response of reference targets
Calibration activities can be separated into three major parts:
Before the first mission we performed a detailed analysis of the
overall system to localize the main error sources and developed
algorithms and procedures to correct these errors. During the
missions we concentrated our efforts on calibration campaigns
at the Oberpfaffenhofen super test site. Post mission activities
included the determination of the antenna pattern and the absolute
calibration factor as well as detailed performance analyses.
PROBLEMS IN X-SAR CALIBRATION
In a detailed preflight study [1] we localized and estimated the
influence of possible error contributions throughout the whole
system. From measurements in a thermal vacuum chamber [2] the
receiver gain was found to be highly sensitive to temperature
variations (~0.2 dB/oC). This temperature sensitivity
was identified as the major radiometric error source in the sensor
itself. It can be corrected by using internal calibration loops
to monitor the system gain variations with temperature and time.
The original specification on the shuttle attitude accuracy was
2o (3s-value) [3]. This
rather high uncertainty is the other main problem because it results
in an insufficient knowledge of the antenna boresight direction
and consequently in high errors in radiometric corrections. With
the antenna boresight we have to rely on the shuttle attitude
information. The actual attitude uncertainty can be estimated
from ground receiver measurements and from SAR images of rain
forest areas.
INTERNAL CALIBRATION
Figure 1 shows a block diagram of the X-SAR system including calibration relevant sub-systems. The chirp signal is generated in the frequency generation unit (FGU) and upconverted to 9.6 GHz in the frequency conversion unit (FCU). After amplification in the high power amplifier (HPA), the signal passes the circulator (C) and is transmitted by the antenna. The same antenna receives the signal reflected from the earth surface. This signal goes through the circulator, the shutter (S), the band-pass filter and limiter (FL) into the low noise amplifier (LNA). The signal is amplified in the LNA and the subsequent receiver (RX), sampled in the analog-digital converter (ADC) and stored on high density digital tapes. The receiver gain can be set in steps of 2 dB from 40 dB to 80 dB. The electronics are mounted on cold-plates in the shuttle payload bay.
To monitor essential system parameters (temperatures, powers, voltages, currents) corresponding sensors are installed throughout the whole system. The temperature sensors are located at system elements, where the temperature is representative (e.g. SAW filter for RX temperature). Fig. 2 shows the temperature measured at the SAW filter and the peak output power of the HPA as a function of the mission elapsed time (MET) during SRL-1. The cold-plates kept the system temperature at a mean value of 20oC with a very low variation of 0.95oC which corresponds to the resolution of the temperature sensor. This high stability solves the problem of the temperature sensitivity of the receiver. For the HPA output we calculated an average peak transmit power of 3.35 kW 0.1 dB. Identical measurements during SRL-2 confirmed the SRL-1 results.
The sensor was not expected to be that stable. Therefore special
circuits have been implemented for direct monitoring of gain variations.
The block diagram in figure 1 includes these calibration loops.
The TX-calibration coupler (C3) couples out part of the high power
transmit signal. This calibration signal (chirp) passes a fixed
attenuator (A2) and the receiver and provides information on the
variation of the product of transmit power and receiver gain (excluding
the LNA). The RX-gain for the TX-loop is set to the mid value
of 60 dB. The RX-calibration uses part of the FCU-output, which
is stabilized for gain variations, and checks the whole receiver
dynamic range by stepping the RX-loop attenuator (A1) in steps
of 5 dB from 0 to 60 dB (calibration level) and the RX-gain over
the whole 40 dB range.
RX- and TX-calibration are performed at the beginning and the
end of each data take. TX-calibration data are evaluated during
precision processing in order to correct the residual variations.
Those were found to be less than 0.2 dB which agrees well with
the above mentioned temperature and power stabilities.
ANTENNA PATTERN MEASUREMENTS
Adequate radiometric corrections of SAR data require knowledge
of the sensor-target geometry and the antenna elevation pattern.
Preflight measurements have been performed on the three separate
leaves of the X-SAR slotted wave guide antenna on an antenna test
range [4]. The reconstructed pattern has been calculated from
the data of the single leaves.
One of the main activities at the Oberpfaffenhofen test site was
the verification of the actual inflight SIR-C/X-SAR antenna patterns
using ground calibration receivers [5]. For that purpose we deployed
and surveyed (using GPS) 20 receivers in each frequency band over
an area of approximately 100 km extension. More than 30 colleagues
were involved in the operation of our ground calibration equipment.
The receivers were designed and manufactured by the Institute
of Navigation at the University of Stuttgart.
These instruments are capable of handling various pulse lengths
and PRFs, have a dynamic range of 65 dB and an absolute accuracy
of 0.3 dB (after software correction). The precise internal clocks
are synchronized to UTC (Universal Time Coordinated) using GPS
receivers. Each radar pulse is sampled sixteen times. The built-in
microcomputer stores the measured samples to a maximum of 8000
pulses, sets time marks, and writes the operating temperature
and the individual unit-ID into the memory. This feature allowed
us to analyze the transmitted radar pulses in detail. Figure 3
shows pulses from the X-SAR recorded by a ground calibration receiver
on data take 30.0 of SRL-1. The almost perfect rectangular-shaped
pulses gave us a first indication of the good performance of the
X-SAR transmit path including the antenna. The slight increase
in power from pulse to pulse is due to the slope of the azimuth
pattern in the main lobe.
Integration of the measured pulses provides the received power versus time of receipt, i.e. due to the platform motion an azimuth cut through the three-dimensional pattern. The time marks further allow to relate each measured pulse to an orbit position and consequently to calculate the distance between sensor and ground receiver and to transform our data into a co-ordinate system centered in the shuttle. The result is an azimuth cut as a function of azimuth angle at the off-nadir look angle of the ground receivers position. Such an X-SAR azimuth pattern is presented in figure 4. The received power after correction of the range dependence is plotted versus the azimuth angle. Within the accuracy of the orbit data the azimuth angle of the peak corresponds to the Doppler centroid derived from the SAR data. This pattern was recorded during the SRL-2 flight (data take 30.0) at the swath center. It is one example out of more than 200 azimuth patterns, which are all in good agreement with the theoretical sinc2 function. The maximum received power was -35.3 dBm, which corresponds to a ground flux density of -9.7 dBm/m2, as expected. Both pulse shapes and azimuth patterns measured during the second mission perfectly matched those recorded during SRL-1. We could not find any degradation from the first to the second flight.
For all individual receiver measurements we determined the power
maximum and the corresponding off-nadir look angle. In a first
iteration we then used the antenna look angles provided by the
shuttle attitude information system to relate these data to 0o
off-boresight angle. A polynomial fit to this cloud of measurements
was our first elevation pattern. By correlation (second iteration)
of this pattern with the data of each individual overflight we
obtained more accurate look angle estimates to align the measurements
and finally the elevation patterns shown in figure 5. The crosses
mark the receiver measurements from SRL-1 and the diamonds from
SRL-2. The solid and the dotted lines are polynomial fits of order
four to the SRL-1 and SRL-2 measurements, respectively. Data from
both missions agree very well. The solid and dotted lines are
almost coincident. The slotted waveguide antenna worked perfectly
during both flights. Except for off-boresight angles between 0o
and 4o these results are in good agreement with the
reconstructed (preflight measured) elevation pattern shown as
a dashed line. Maximum deviations of 0.8 dB are possibly due to
the influence of the SIR-C antenna structure [6].
The other information necessary for antenna pattern correction
is the antenna boresight. Uncertainties in boresight angle determination
are one of the major error sources in overall calibration. As
the specified accuracy on the shuttle attitude is only 2o
(3s-value), we compared boresight angles
estimated from our ground receiver measurements and from rain
forest data takes with shuttle attitude data to get some feeling
of the actual accuracy. Table 2 summarizes these results. The
difference between our measurements and the attitude information
is about 0.2o, the maximum deviation is 0.3o
which is far better than the specification. From Doppler centroid
estimates we know that the pitch and yaw information from the
shuttle is also accurate within 0.2o with outliers
up to 0.4o. Due to this excellent
performance of Endeavour's attitude information system it is possible to perform precise corrections and to process relatively calibrated SAR images.
a) b)
For a multifrequency/multipolarization system like SIR-C/X-SAR
it is very important that the different antenna beams are aligned
and illuminate the same swath at the same azimuth angle. Using
our calibrations receivers we measured the three-dimensional antenna
patterns in all three frequencies and in L- and C-band in both
horizontal and vertical polarizations. The time synchronization
of the receivers allows to correlate measurements from different
frequencies/polarizations and to estimate the alignement of the
five antennas. Figure 6 and 7 show the results of the beam alignment
measurements on data take 30.0 of SLR-2. The different beams are
well aligned in both azimuth and elevation direction. Please note
the almost perfect alignment of LH/LV and CH/CV in azimuth, which
is required to ensure high correlation of the successively measured
elements of the scattering matrix.
ABSOLUTE CALIBRATION
For the purpose of absolute radiometric calibration we deployed
15 trihedral corner reflectors of different size (six with 3 m,
six with 1.5 m and 3 with 1 m leg length) in the core area of
the Oberpfaffenhofen test site (cross-track extension: swath center
10 km). The trihedrals have been reoriented to the shuttle path
before each data take. Using our antenna elevation pattern from
figure 5 and the corresponding boresight angle our colleagues
from D-PAF processed radiometrically corrected X-SAR images. We
analyzed these images on a calibration workstation running our
analysis software package CALIX [7]. The determination of the
image power in CALIX is based on the integral approach [8,9].
In a first iteration D-PAF reduced the processor gain to avoid
clipping of the point target responses. We then determined the
absolute calibration factor from all corner reflector responses
of SRL-1. The factors derived from the 3 m trihedrals were systematically
lower than those from the 1.5 reflectors by ~0.3 dB. Mechanical
tolerances are the reason for this decrease in radar cross-section.
Taking this into account, the processing gain has been adjusted
in order to optimize the dynamic range of the image products.
We finally fixed the absolute calibration factor at 200000 (~53
dB) for SSC and at 1000000 (60 dB) for MGD products. The reason
for the difference between SSC and MGD calibration factors is
the difference in dynamic range. The calibrations of SSC and MGD
products were compared and found to be within 0.05 dB.
In a second iteration all Oberpfaffenhofen scenes of SRL-1 have
been reprocessed and reanalyzed to make sure that the fine-tuning
of the processor has been applied correctly. The results for the
MGD products are presented in figure 7. The absolute calibration
factors from different trihedrals are marked as crosses for all
Oberpfaffenhofen data takes of SRL-1. This plot also includes
the SRL-2 data (marked as diamonds), which have been processed
using the same processor version as for SRL-1. The individual
calibration factors are within the 1 dB band around the mean value
of 60 dB. The only exceptions are the data takes 30.00 for SRL-1
and 94.00 for SRL-2. The first one is saturated (gain setting
too high), the second one gives lower calibration factors because
our reflectors have been imaged under a squint angle of ~7o
(problems with the attitude control system). DT 126.02 (SRL-2)
was a `snow' data take; approximately 2 cm of wet snow on the
trihedrals surfaces resulted in a RCS loss of more than 10 dB.
DT 158.12 was only acquired during SRL-2 because of the extended
mission duration.
The accuracy of the end-to-end calibration of the X-SAR processing
chain has been checked using rain forest data. Five scenes with
a wide range of incidence angles and PRFs as well as the two resolution
modes have been processed and analyzed: range power profiles of
visually homogeneous areas were computed and converted from range
to incidence angles and from so
to g = so/sin(qi).
g can be assumed to be independent
from incidence angle for X-band. The accuracy of absolute calibration
is demonstrated in figure 8: all five g-profiles
are shown in one plot ranging from 24o through 62o.
This figure suggests again that the absolute calibration
accuracy of X-SAR is well within 1 dB. An exception is data take
151.05: here the raw data are severely saturated in mid-swath
resulting in a loss of image power; only the g-estimates
at the swath edges are representative.
The residual radiometric errors within the image swath are caused
by orbit and attitude uncertainties. Note that the maximum slope
of the two-way antenna pattern is 10 dB/o. From both
Doppler parameter estimation and ground receiver measurements
we know that the uncertainty of the attitude information is ~0.2o.
Additionally, a (realistic) radial orbit position error of 500
m at 25o look angle causes the conversion from look
angle to slant range to be off by 0.25o. From these
considerations it becomes evident that the calibration accuracy
deteriorates toward extreme near and far ranges. An error of 2
- 5 dB at the swath edges is not unrealistic, especially at steep
look angles. Indeed with about 5% of the X-SAR images processed
so far a miscorrection at near and far range is visible.
SYSTEM NOISE
In our preflight studies [1] we estimated the variation in the
thermal noise power due to different antenna brightness temperatures
(different contributions from the earth surface) to be less than
0.5 dB. The reason for this low variation is a signal attenuation
of more than 4 dB from the antenna to the LNA input. Ohmic losses
in the antenna feed network and in the wave guide from the circulator
to the antenna are responsible for this attenuation. Consequently,
it is not necessary to perform receive-only mode measurements
before and after each data take.
During the missions receive-only sequences were recorded using
the highest RX-gain setting. From those the system noise for fine
and coarse resolution mode was estimated and the noise equivalent
so was calculated.
Fig. 8 shows noise equivalent so
as a function of incidence angle, resolution mode, and product
type for the most often operated PRF of 1395 Hz. The elevation
antenna pattern has been neglected, i.e. this plot is only valid
for mid-swath; toward the swath edges the noise increases according
to the antenna pattern. For the user this means that in most of
the cases system noise can be neglected and noise subtraction
need not be performed. For comparison it may be interesting that
ERS-1 has a noise equivalent so
of -26 dB at 23o incidence angle.
CONCLUSIONS
The low noise floor (see Fig. 8) and the radiometric accuracy
of 1 dB are the main reasons for the high quality of the X-SAR
data. The performance of the space shuttle's attitude control
and information system was found to be far better than specified
and allowed accurate correction of the cross-track antenna pattern.
Due to the unexpected low temperature variation on the cold plates
the whole sensor electronics were very stable, especially the
receiver gain.
Our approach to measure the one-way antenna pattern with ground
receivers was shown to be feasible. Rain forest data have been
used to validate the receiver measurements and data from both
sources are in good agreement. The trihedral corner reflectors
made of solid aluminium plates provided highly accurate reference
targets for absolute calibration. Even from the 3 m reflectors
we obtained consistent results in X-band. The normalization of
the processor was accurately performed. The calibrations of SSC
and MGD products are within 0.05 dB. Data acquired under different
look angles, PRFs, resolution and quantization modes agree well
with each other (see Fig. 7). g-profiles
of the rain forest measured at X-band are flat over an incidence
angle range from 20o to 65o.
The remaining calibration error is still due to the uncertainty
of the attitude information. More accurate roll angles would further
improve the calibration accuracy especially at the swath edges.
ACKNOWLEDGMENT
We acknowledge the contribution and support of 30 colleagues (the
X-SAR calibration team) to the Oberpfaffenhofen calibration campaign.
Special thanks go to Richard Bamler, Michael Eineder, Ulrich Steinbrecher
and Helko Breit for many helpful discussions, for providing results
out of the raw data screening process and for quick turn around
of processed images.
REFERENCES
[1] Zink M., "The X-SAR Calibration Plan: Part I and II,"
DLR, Technical Report, Oct. 1993.
[2] Dornier GmbH, "Final Performance Test Report of the X-SAR (PM) Electronics,"
TR-2042-7000-DS/11-02, Friedrichshafen, Germany, Aug.
1992.
[3] Rockwell International, "PATH Product, Interface Control Document,"
ICD-I-TOP-002, JSC 18645, Jan. 1988.
[4] Alenia Spazio, "PM-Antenna Electrical Test Report,"
TR/X-SAR/2700/SS-04, Rome, Italy, Feb. 1991.
[5] Seifert P., H. Lentz, M. Zink and F. Heel, "Ground-Based Measurements of
Inflight Antenna Patterns for Imaging Radar Systems,"
IEEE TGARS, Vol. 30, No. 6, Nov. 1992, pp. 1131-1136.
[6] Measurements performed by Ball Aerospace, Broomfield, CO, 1990.
[7] Reichenbach W., "Aufbereitung und Analyse von SAR-Daten im
Rahmen der radiometrischen Kalibrierung,"
Diplomarbeit, Technical University of Munich, Aug. 1994.
[8] A. L. Gray, P. W. Vachon, C. E. Livingstone, and T. I. Lukowski,
"Synthetic Aperture Radar Calibration Using Reference Reflectors,"
IEEE TGARS, Vol. 28, No. 3, May 1990, pp. 374-383.
[9] Ulander L. M. H., "Accuracy of Using Point Targets for SAR Calibration,"
IEEE TAES, Vol. 27, No. 1, Jan. 1991, pp. 139-148.
[10] M. Zink, "Calibration and Performance Analysis of the X-SAR System,"
PIERS '94, (Noordwijk, The Netherlands), 11.-15. July 1994.
[11] M. Zink, "Calibration and Performance Analysis of the X-SAR System,"
IGARSS '94, (Pasadena, CA), 8.-12. Aug. 1994.
[12] M. Zink et.al., "Calibration of the X-SAR System,"
CEOS SAR Calibration Workshop, (Ann Arbor, Michigan),
28.-30. Sept. 1994.
[13] M. Zink, "Radiometric Calibration of the X-SAR Data,"
URSI-Konferenz, (Kleinheubach), 10.-14. Okt. 1994.
[14] M. Zink, "Results of X-SAR Calibration,"
IGARSS '95, (Florence, Italy), 10.-14. July 1995.
[15] M. Zink and R. Bamler, "X-SAR Calibration and Data Quality,"
IEEE TGARS, Vol. 33, No. 4, July 1995, pp. 840-847.