Re: Mark IV Electronics Design

[Tom Droege <droege@wwa.com>]

Dan, Chris,

Neither of you seem to be addressing the problem.  How do you interface a data
source that *must* put out a large bolck of data at a fixed rate?  It seems
to me
there needs to be a buffer somewhere.  

It is true, that you don't usually don't need a full sized buffer.  I am
sure that some 
of the channels that have been discussed have enough buffering so that they
*seldom* overrun.

The problem is that "seldom" means that from time to time a retransmission
must
be requested.  But there is no storage in tass to allow retransmission
unless I 
put a memory card somewhere. 

You may argue that I don't need to read out the CCD continuously.   I have
been 
called in to consult where systems were designed that did not read out
continuously.
The astronomers did not like the result. 

Note the Mark III is not guaranteed to meet this requirement.  It depends
on how 
Norman's software is written.  One can also see the result of this if you
look for it.
Early on in the Mark III testing I would get a whole line shifted from read
out when
the disk look up got too long.  I don't want these problems in the Mark IV.
 So it is
designed to take the data in one continuous block, then the computer can
work on
it as irregularly as it wants with no effect.  

There are things going on in an analog data system that cannot be fixed by
any 
amount of heroic electronicing.  One example is dielectric absorption.  All
dielectrics
(read the capacitor in the CCD where a pixel is sampled and held) have
something that is 
time related to their memory of charge.  You can't fix this completely, and
it shows up as
a trail behind a star at some level.  But it is worse if you measure
irregularly since not
only the signals have a dielectric absorption effect but also the base line
levels.
So you can fix part of the problem by regular measurement.   For a system
with long
cables there are also signals that bounce back and fourth down the cables.
With 
regular clocking, part of this signal is the same each time.  These effects
vary from
1% to 0.1%.  A significant problem for a 16 bit system.  Double correlated
sampling
also helps. 

Still, dielectric absorption is not the problem it once was.  In the good
old days (WW II
Radar) people were killed by dielectric absorption.  They would turn off
the power and
short out the high voltage capacitors.   Then go in and work on the system.
 After a few
got killed, they learned to clip a shorting cable across the capacitors
while they were 
working.  Then they had to learn not to blow up the power supply when they
turned 
the power back on forgetting to remove the jumper.  Sigh!   

Tom Droege