On  using RC timing as an A/D...  It seems that if you're using
an RC circuit instead of a current source driving a capacitor (as is done
in dual slope A/D converters), you have to deal with linearizing since
the time to charge a capacitor to a ceratin threshold is not linear with
the applied voltage.
        Looking at AN513, I see that it's a single slope A/D.  I like
dual slope A/Ds since they "self calibrate" by using the reference input
for the "de-integrate" phase.  Here, it appears that they are suggesting
an occasional single slope A/D conversion of a reference input, then
doing some math.  The dual slope A/D takes care of the math for you.   In
addition, the circuit in AN513 is relying upon an input port changing
state at some stable voltage.  How stable is the threshold, especially
with temperature and power supply variations?  Ideally, we'd add a
comparator to drive the input port.   Most dual slope A/Ds also have an
autozero phase.  For example, the MC14433 puts a sample of the total
offset voltage (due to the integrator op amp and the comparators) in a
capacitor.  That capacitor is then placed in series with the input during
the unknown integrate phase, canceling out the offset.  I did a project
with the MC14433 about 20 or 25 years ago.  It worked extremely well.
with zero input, the thing ALWAYS read 0.  Reversing polarity of the
input ALWAYS resulted in an exact reversal of the reading (positive to
negative, or back).  The MC14433 is a 3.5 digit A/D with a multiplexed
BCD output.
        One tricky thing on using an A/D to measure an absolute voltage
(instead of a voltage ratio) is the stability of the reference.  When I
first did the 14433 project, I just used a zener reference and a carbon
film resistor voltage divider.  I could put my finger on any of those
parts and watch the digits roll by.  Not very good!  I ended up using an
LM399H reference (6.95 volts with 0.5 ppm/C tempco) and a couple metal
film 5 ppm/C resistors.  I always wished National or someone would make
an LM399H (which gets its stability by including a temperature stable
oven on the chip) with a voltage divider on it.  Having the voltage
divider at a constant temperature would sure help!
        Another interesting dual slope A/D is the Maxim MAX134CPL.  This
is a digital multimeter on a chip.  The bus interface is a bit slow, so
some tricks have to be used to hang it on a processor bus.  It's
something like 4.5 digits.  It uses "residual multiplication" dual slope
A/D conversion.  It's the only time I've seen that technique, and it's
pretty clever.  In a normal dual slope A/D, to get more resolution, you
typically use a longer conversion time or a faster clock.  The unknown
voltage is integrated for a fixed time.  The reference is "de-integrated"
for a variable time until the integrator voltage gets back down to zero.
The time it takes to get back to zero is the A/D conversion.  There is
always some overshoot on the de-integrate phase, since the integrator
output does not pass through zero on exactly a clock change.  In residual
multiplication, they then integrate this "residual" integrator output and
multiply it by a scaling factor.  It was pretty clever.  It gives a
higher conversion rate than normally possible for that resolution using
dual slope.  The 134 does not include a zero capacitor.  It does have a
zero phase, but outputs a result that must then be subtracted from other
conversions.  This is the trick of moving hardware to software... saved a
capacitor!
        Another interesting dual slope A/D is the ICL7135.        It's
similar to the MC14433 but 4.5 digitsl.  It also has a multiplexed BCD
output, but in my project I didn't use that.  Instead I used the busy
line, the polarity line, and the clock line.  ANDing the busy and the
clock results in a burst of clock pulses.  The number of pulses in the
burst varies with the analog input voltage.  The A/D section in this
project operated on a floating power supply and was then opto coupled
down to the processor section.  The busyANDclock was then dumped into the
counter on a 6522A.  20,001 was subtracted from the resulting count, and
there was the A/D conversion in hundreds of microvolts.
        Summarizing...  If all you need is 8 bits of A/D, why not use a
chip with it on it?  It then takes NO external parts and works well.  If
more resolution is used, there are several single chip solutions that
don't require you to put a bunch of op amps and switches together.  Why
not take advantage of the expertise that went into the design of the A/D
chips that are out there?


Harold