🚀 go-pugleaf

HELP WANTED: $1000 FOR ANALOG HELP

Color Dreams Inc is offering a reward of $1000 for an answer to an
analog electronics problem we are having.  If you are good with analog 
design this ought to be a breeze for you.  Unfortunately we're all 
microcontroller people here.

Here's the problem in one sentence to save you from having to read on
if
you aren't suitable.  We want to turn a linear output amplified through
an op-amp into what is probably a log function and we can only afford
a transitor or two and some resistors, or at the most another op-amp.  

In more detail: We have a 0-450mv DC signal we are isolating from
a 9V signal.  We've run it through a cap which is clamped to ground
through
a jfet transistor every 1/2 time interval of the signal.  During the
other 
half of the signal cycle we let it run through an opamp to amplify it
into
an A/D converter.  The A/D converter has a rane of 2.5V, so we amplify
by
a factor of 15.  So we have a 1K resistor going from the signal into
the negative end of the opamp, with the + end grounded.  There's a 15K
feed back resistor. 

The output range of 2.5 volts is thus covered by only 0-160mv of the
original
signal and the rest is lost.  Unfortunately, the low range is the
portion we 
are most interested in, so we can't simply reduce amplification to
include
the high part of the signal in our 8 bit value.

We want to skruntch the rest of the signal down into this range, so
that
the 0-160mv ends up in the lower 2/3 of the A/D range and
the remainder is smoothly skruntched into the remaining 1/3.  We want
the
entire 0-450mv range to fit into the 8 bit A/D value, but we want the
most resolution in the 0-200mv range.

It's not acceptable to have a series of Zener diodes kick in causing a
linear skew to the top range.  We want a true smooth curve, sort of
hardware pre-conversion gamma correction if you prefer.  I don't know
if this
curve is the same as a log function but that's where we have been
looking.

Here's roughly what we want:


Input:			At 15X opamp	With desired correction circuit

0mv			0mv  		0mv
50mv			750mv		700mv
100mv			1.50V		1.2mv
150mv			2.25V		1.75mv
200mv			3.00V		2.00V
250mv			3.75V		2.18V
300mv			4.50V		2.30V
350mv			5.25V		2.40V
400mv			6.00V		2.46V
450mv			6.75V		2.5V

This is a really crude representation, someone who understood math
better
could probably find the function, enter it on their calculator, and
come up with better values for the last column.  The idea is that we
want best resolution in the 0-200 range but we don't want to lose the
information in the upper range with a simple x15 amplification.  We
want
a smooth curve to the top.  It's a non-linear function and we won't
accept
crude curve tilts using Zener diodes.  My C programmer says it's a log
function, y=log x, not necessarily log base 10, but I'm not entirely 
convinced.

We have a book suggesting that there is a range in the turn-on curve of
a transistor that can be put into the feedback path to shape the output
like we want, but I did calculations on it and came up with a linear
function all the way.  Perhaps the operating condition is before the 
tranisitor turns on.  Anyway, we built it and it didn't work.  

The first person to suggest a solution will receive the reward. You
don't
have to draw a skematic, just document it and we'll breadboard it.  Or
if
you have the ability to get a skematic into UUE or MIME format we can
use that if you prefer.  Or email for our fax number if you scribble
it down on a napkin between espresso breaks.

Component wise, we have a classic inverting opamp situation with a 1K
input 
and a 15K feedback and we are looking to throw a couple of
components into the feedback path if possible.  Optionally we can
allow another opamp of the same type.  We have a 50V per microsecond 
skew opamp, the dual version we could switch to is 15V per microsecond,
the desired sample rise time is 1/2 microsecond but we can accomodate
if necessary.


First answer received and used gets the reward, we aren't interested in

cheating anyone, we've already spent $750,000 on this project,
so we really will award it to the person who provides the answer,
but we offer no guarantees for legal reasons since we might receive
multiple
answers and someone might throw some legal tizzy.  So if you
participate
you will just have to trust us to be fair and forget about any time
losses
you might think you had if you don't win.   Just scribble down your
answer,
obviously for $1000 it isn't worth too much of your time, but if you
are the right person it's a quick $1000 bucks.  We can take taxes out
if you
want or treat you like a consultant.  You will receive no further
compensation
for this circuit, but we are often in need of analog advice and who
knows.
We often receive offers to consult for Taiwan companies and have to
turn them
down for lack of time, so again, there might be more than $1000 in this
for the right person.

This offer was previously posted through Jim M. with a $200 reward, but
we got only Zener diode answers and none were suitable.  It seemed like
something really easy, but I guess the $200 wouldn't buy enough coffee
to
attract anyone.

It's 12/16.  If I don't have an answer in a day or two I'll have to do
it
myself, so hurry up if you're interested.  I was leaning towards an
emitter
follower transistor in the feedback path with a diode across it to
protect
it and biased to contribute in it's partially on range.  You can
suggest the
same if you've developed the idea to it's fruition.


Email me through dan@wincam.com