Image processing and distribution
The single-bit processor
By contrast with optical ones, radar
images require before becoming accessible a period of time for processing. This may be
longer or shorter according to the dimensions of the scene observed, sensor
characteristics and processor performance. For a basis of comparison, in the mid-eighties
a 100 x 100 km scene took up to nine hours' processing on a powerful computer. In the
early nineties, for the same scene, a couple of hours were enough, while today it takes a
quarter-hour. These improvements are due partly to more refined, optimized algorithms, but
mainly to increased computer performance. Despite this astounding rise in computer
capacity, the time to process an image is 60 times what the radar needs to record the
scene. This means that by the time the image is available the satellite is at least 6000
km away from the area viewed. This limitation rules out use of SAR images for a whole
range of applications that need real time or near real time.
The need to bridge the gap between the moment a scene is recorded and
the moment its image is available has stimulated the search for innovative solutions, both
algorithmically and in circuits. This is the context for the development of the single-bit
processor. Creating it has meant both adopting non-standard processing and designing ad
hoc hardware.
Single-bit processing
The signal picked up by the radar is coded
as 4-bit numbers, which without further processing would allow 16-gray-scale images to be
obtained. Processing improves both the geometrical resolution (i.e. the dimensions of the
smallest object that can be discriminated, going from over 100 meters to less than 10) and
the radiometric resolution (making objects with imperceptible intensity differences
distinguishable: each pixel can take on one of 64,000 possible gray levels). In single-bit
processing only one of the four initial bits is retained: after processing the image still
has 16-bit data as in the full-dynamic case. However, by contrast with the latter, though
retaining the capacity to pick out details (geometrical resolution), it has poorer
radiometric fidelity. As an example of the differences, two neighboring objects slightly
differing in intensity would in the single-bit-processed image have either the same
intensity (and therefore be indistinguishable) or a bigger intensity difference than the
real one. In any case, an observer could not notice the differences between
single-bit-processed and full-dynamic homogeneous scenes, given the limited gray levels a
person can distinguish. The differences would however be plain in the case of highly
brilliant isolated points.
Hardware
implementation
The single-bit algorithm is well suited to
being implemented using relatively simple digital logic. The main operation consists in
taking the product of two matrices, one for the data and the other for the filter. These
operations have been implemented using innovative FPGA (Field Programmable Gate Array)
components. These devices allow dedicated hardware circuits to be produced, loading at
switch-on time the programs defining the topology of connections among the individual
cells. This method offers the advantages of both hardware circuits (processing speed) and
software (since changing the start-up program allows the connections among cells and
therefore circuit behavior to be organized differently). This means that if one or more
components stop working they can be excluded while still having the processor able to
function properly.
With the advantages described, the
drawbacks should also be listed.
At the present time the components are cumbersome and need relatively
high power; however, technological evolution promises to eliminate these drawbacks before
too long. Additionally, producing them entails development costs that would not be
incurred if commercial processors were used.
Output of the
single-bit processor
The single-bit processor supplies in real
time the images taken by the SAR sensor, as well as, again in real time, the images in
amplitude and phase format that are the starting point for generating the topographical
products like the ground height maps. This latter processing is, however, done on
conventional computers off line, i.e. not in real time. Data on the alignment of the
antennas will also be supplied. For each piece of processing, data on image quality and
processing parameters used will be supplied. As regards the interferometric products,
image-coherence maps, interference fringes and unwrapped fringes will be supplied (DEM not
georeferenced).
The interferometric products today take some 2 hours' processing for 50
x 50 km regions with spacing between one height point and the next at 25m. As an example
of satellite survey capacity, two workers on the ground would take at least 5 years to do
the same work (the basis for this estimate is 4 million measuring points, 1 minute's time
for each individual measurement and a 24-hour working day, i.e. three shifts).
The precision data processing system and image
archiving and distribution system (I-PAF)
Precision data processing
Radar images, by contrast with optical ones, are taken by recording a
large number (some 2000 per second) of echoes reflected from the Earth's surface. The
signals picked up are very similar to holographic data and accordingly require highly
sophisticated, laborious processing. Once two radar images of the same area have been
obtained, interferometric processing can be done. This consists of determining the time
differences in echo return time for each pixel taken (radar signal phase differences).
This operation is particularly delicate, since small errors introduced by the system can
lead to unacceptable errors in the final product. With an eye to limiting possible errors,
the mission incorporates various error correction strategies:
The navigation system allows high-precision recording of position and
velocity data for the platform and at the same time keeps a check on the dimensions of the
interferometer, consisting of a 60-meter long arm with two antennas at its end.
The data are recorded in "strips" some 50 km wide, always
including two edges of sea. This expedient allows the data processed to be
"anchored" to a very exact reference, the sea.
- Since the recordings are made on ascending and descending orbits, the data strips
overlap in so-called crossover zones. In these zones further checks and corrections can be
done.
The processing system
Today's data processing technology allows
the system to be run on conventional systems. Processing is divided into:
Transcription on to "computer-compatible" magnetic support
of data coming from the mission, recorded on high-density tapes;
Processing of the calibration signals for the onboard instruments and
of the platform position and velocity information;
Synthesis of the interferograms and subsequent geometrical processing
to derive the height datum;
- In the final stage in the processing chain, the height data obtained from the
interferograms are mosaicized to form the products for distribution.
Products
The result of processing the data acquired
can be differentiated into two categories: image products and interferometric products.
The image products cover all the products obtained from acquiring and
processing the signals coming from one individual antenna. These images will have a format
and quality comparable with the products derived from the two previous missions. It will
accordingly be possible to analyze the data from the three missions in a broader context,
and in some cases to secure further results from the scientific experiments conducted
during the first two X-SAR missions.
The category of interferometric products covers the products deriving
from the joint processing of the data coming from the two antennas located at the end of
the arm. The interferometric products represent the operational result of the mission,
enabling mapping of the height of the Earth's surface in relation to a single reference
system and with uniform precision for the first time in history. The products generated by
the three international partners will have identical quality and format, to guarantee
interchange of results. In particular, the products associated with band X (sensor
designed by Italy and Germany) will because of their greater precision be used to
calibrate the data from band C too (US sensor). |