> You've pointed out the main problems with the simple RC design ... the
> exponential ramp, and the inconsistencies in the threshold voltage of the
> PIC's input mean that the AN513 A to D can never be very linear, stable, or
> precise.  A possible solution may be to use a dual opamp, with one section
> used to create a constant current source to linearize the charging of the
> integrating capacitor, while the other section is used as a comparator.

One design I prefer, which I've mentioned a few times but never received
feedback about, uses just a comparator and an RC (on the 16C62x it uses
three port pins if you use the internal comparator; if you use an external
comparator it uses two).  The signal to be measured should be fed into one
input of the comparator (I'll assume non-inverting for now) and the output
of that comparator should feed a port pin.  The other port pin should go
into an RC filter; the output of that filter should feed the inverting in-
put of the comparator.

To measure the input, just run the following procedure about 10,000 times/sec
or so:

static int16 hicount, samples, measurement;

{
  if (input_pin)
  {
    output_pin = 1;
    hicount++;
  }
  else
    output_pin = 0;
  samples++;
  if (samples >= 5000)
  {
    measurement = hicount;
    hicount = 0;
    samples = 0;
  }
}

Measurement will hold the current measured value in milivolts (0-5000) and
will be updated about twice per second.  Note that the measurement procedure
must be called at a consistent rate (which should be much faster than the RC
rate) and that the measurements will only be as precise as the VDD level.
On the other hand, this is a very cheap way of getting fairly precise (but
slow) measurements that seems much more stable and noise-resistant than many
RC methods; because the software feeds the sampled comparator readings back
into the comparator circuit, a false-high reading on one sample will increase
the likelihood that the next reading will be low.