From: henry@spsystems.net (Henry Spencer)
Newsgroups: sci.space.tech
Subject: Re: Rad hardened Pentium processors
Date: Sun, 3 Jan 1999 22:28:31 GMT

In article <368EFBB3.AB0803D0@NOSPAMerols.com>,
rk  <stellare@NOSPAMerols.com> wrote:
>> -- it's not difficult at
>> all to build 4Kx1 detectors and use spacecraft motion or a scanning mirror
>> for the second dimension.  SPOT panchromatic images are 6000x6000x8...
>
>from an instrumentation point of view, i would refer to these as 4kx1
>detectors, as that would be the natural size for data handling...

Unfortunately, the *images* are still 4Kx4K (or whatever), and that sets
the horsepower needed for compression.  You can economize on memory by
doing compression in strips -- image compression, especially lossy
methods, often does want to work with 2D data, not just individual scan
lines, but typically a narrow strip is enough -- but the number of CPU
cycles is determined by the pixel count, not the strip size.

>...that would be 36 megabytes or close
>to 300 megabits, which, if my old and feeble memory is correct,
>represents about the capacity of a tape recorder from that period...
>i would think that nothing other than record and playback to tape is done.
>perhaps you could tell us more about the spot system and how it works,
>along with how the division between on-board and ground processing.

I don't have a lot of info on it, but I do have some.

The detectors are actually 1728x1, and on Spots 1-3 they were commercial
fax-machine chips carefully selected for quality and matched response,
although Spots 4-5 are using custom chips.  There is a moderately complex
optical system which combines four chips (with some outlying pixels not
used) into a 6000x1 sensor.  There are four sensors per camera, but the
three color sensors have adjacent pixels combined electronically to make
them effectively 3000x1.  Each Spot has two identical cameras.

The data rate was, um, challenging for mid-1970s electronics, especially
with space qualification required.  8-bit digitization was the best that
could be done, and even that required a bank of four A-Ds (one per
detector chip) in parallel to handle the panchromatic sensor.

The system was constrained to have a total data rate of 50Mbps, just equal
to that of the updated-Landsat Thematic Mapper instrument, since a lot of
ground stations were being built to that spec.  Unfortunately, the pair of
cameras can generate over double that data rate.  The flow was cut in half
by simply accepting that you get to pick two streams out of four (two
cameras, each with a color stream and a panchromatic stream) and lose the
other two.

There's still a problem, though:  three color channels just fit in half of
the data rate (by design).  But a panchromatic image is four times as much
data as one of the color subimages, not three.  Compression is needed for
the panchromatic images, and there are two methods.  The first simply
drops the two least-significant bits per pixel -- not great, but usable
for some things (Landsat multispectral-scanner images are only 6 bits
deep).  The second sends every third pixel as a full 8 bits, and encodes
the rest at 5 bits each, using a custom-tailored incremental encoding,
optimized to represent small changes in intensity exactly at the cost of
introducing errors in large changes (because a large change probably
represents a boundary anyway, and thus quantitative measurement of it is
unlikely to be interesting, but precise measurements within smooth regions
are important).

Anyway, all this goes either to the transmitter or to one of two tape
recorders, each of which had 22min capacity on Spots 1-3, and twice that
on 4-5.  The tape recorders have various constraints and have historically
been quite unreliable.

>...it's always prudent to keep some low throughput, raw modes in these sorts
>of instruments, as the compression algorithms are usually tested with
>some simulated data or data acquired from an instrument thought to be
>similar...

Indeed, Galileo has sent back some incomplete images, because some of the
moon images had more fine detail than expected, and so the compression was
not as effective as had been hoped.
--
The good old days                   |  Henry Spencer   henry@spsystems.net
weren't.                            |      (aka henry@zoo.toronto.edu)