> From: Anil Patel <anil.patel@AUTODESK.COM>
> [cut]
> If anyone has a way of measuring pulse widths accurately, please let me know.

The '84 does have a 16-bit counter, i.e. the prescaler (which is not
directly accessible) and TMR0.  The trick is to somehow find out
the contents of the prescaler at a particular point.  Although I have
not tested the following idea, it should work.  Polling is used
rather than interrupts because:
        . interrupt latency is relatively tricky to deal with
        . since this is a servo application, there's plenty of
          time between pulses to accomplish something else.

Set the prescaler to 1:256 so that TMR0 increments every 128us.  Now
poll the input until it goes high.  Clear TMR0 (which also clears the
prescaler).  Poll input until it goes low.  The combination of
prescaler/TMR0 holds the low and high byte of the count (units of
0.5us).  Save the high byte (TMR0), then poll TMR0 until it changes --
you can just check its LSB.  While polling, an auxiliary counter is
incremented to get an estimate of the remaining count left in the
prescaler.  If the complete polling loop is N instruction cycles, then
the aux count is multiplied by N to get the prescaler remainder, then
subtracted from 256 to estimate the prescaler count at the time the
input went low -- this becomes the low byte of the overall timer
count.

In reality, you will need a few other corrections to account for
differing code path delays etc., however you should be able to get a
resolution of about 3us.

Regards,
SJH
Canberra, Australia