Noise Problem

[Tom Droege <tdroege@veriomail.com>]

Possibly I have found my noise problem.  The CCD requires a 20-24 volt 
supply to power the output.  This is an inconvenient supply to find in a 
combination.  So I use a +/- 5 volt and a +/- 15 volt linear set to power 
the system.  This leaves the problem of how to get +20.  To make +20 I made 
a little printed circuit board that mounts on the +/- 15 volt supply.  It 
picks off the transformer winding and uses a voltage doubler to make +33 
that it regulated to +20 with an LM317 regulator.  Since the load is small, 
I just used RC filtering.  It appears that I sized the R a little too 
large, and the regulator would sometimes drop out.  It is not clear why it 
chose to drop out early in the morning, but that is what it did.  This then 
makes lots of noise where you do not want to make noise.   This is the very 
front end of the CCD gain chain.

I finally noticed this when I started plotting the power supply voltages 
over time just to look to see if there was anything funny going on.  I was 
really looking for an oscillation.  Then I noticed that the +20 was jumping 
around more that it should, and also was decreasing over time.

Some of you may recall that I am concerned about the final software using 
the built in measuring system.  The general plan is to monitor each voltage 
and make a history.  One should compute the mean value over time and 
compare to a set point range.  If the supply goes out of this set point 
range sound alarm #1.  As a second check, the sigma should be computed for 
the measurements over time.  If the sigma increases over time, then sound 
alarm #2.

The first alarm detects things like low line voltage and such.  The second 
detects a deteriorating supply.  Usually filter capacitors that are drying 
up.  These tests are really worth making.  If I had implemented them, I 
would have saved hundreds of hours looking for this problem.

I am not quite out of the woods yet.  Something else happened last night 
that I cannot explain.  I ran for several hours taking data.  I then 
stopped the program and looked at the data.  It looked fine, except that 
there were clouds and the data was pretty fuzzy.  I then restarted the run 
for the rest of the night.  It immediately failed in a new failure mode, 
and took no good data for the rest of the night.  Sigh!

But slowly I beat out the bugs.

Arne and Michael.  You will want to make a change to this little 
board.  The supply is shown on drawing SC 3.  For a start, look at the +20 
with an oscilloscope.  If you cannot see any ripple at the 10 mv level, 
then you are probably OK and for the moment do nothing.  You can find a +20 
volt test point on the ADC board.  Or you can look at the output of the 
little board that is tacked onto the +/- 15 volt linear supply.  If you see 
ripple, then it will probably cause an increase in noise level.  One 
solution is to just crank down the pot on the little board for a lower 
voltage.  This will probably not affect the operation much, and will reduce 
the ripple problem.  The next step is to reduce the input filter 
resistor.  I am now using 200 ohms.  You could just parallel the existing 
resistors with a 200 or 240 ohm resistor.  A 1/4 watt or 1/2 watt should be 
OK.  I will later on make an official change.  I want to test a little more 
first.

The symptom for this problem is unexplained very wild data.  It might be OK 
over part of the frame and then swing wildly between + and - full 
scale.  You may remember that we have had a "bright pixel" problem where a 
pixel here and there is near full scale.  I thing that this was the 
cause.  Note that the double correlated sample scheme covers a lot of 
problems.  It does it's best to dig signal out of noise.

Tom Droege