Mike Keitz wrote:

> This depends on what kind of variability is found in LEDs.  If it is
> variable internal resistance, then constant current drive is better.
> If it is a variable leakage path that doesn't produce light, then
> approximately constant-voltage drive would be better.  Experience
> seems to point to the former.

  In a manner of speaking.  The equivalent circuit of a LED is a diode
with a drop of between 1.1 (I-R) through 1.5 (Red) to 2.2 or more Volts
(Green and High-brightness) in series with a relatively low resistance.
As mentioned later in Mike's discussion in regard to the Vbe threshold
of a transistor, it is (AFAIK,) the threshold voltage which is
temperature-dependent and for this reason, constant-voltage drive is
impractical.

  Nevertheless, many cheap clocks use a single limiting resistor for the
whole display (all LEDs of course coming from the same batch) and
implement the "dimmer" function by changing this.  Since (they do not
flash the colon so) they only change once per minute, the alteration in
brightness is simply not noticed!

> Another classical circuit uses another transistor to shunt drive away
> from the main one when the emitter current reaches the desired value.
...
> The disadvantage is additional parts

  Hearking back to the MOSFETs, two distinct advantages are the absence
of a "drive" resistor and the fact that (within 12V or so) the gate may
be overdriven in source follower mode when the Vdd collapses, without
any extra load on the PIC.  The MOSFET may be used as a constant-current
sink by actually using a gate drive resistor shunted by, of all things,
a pair of series LEDs; and a source resistor.  Count: two resistors,
two diodes and one MOSFET, though one of the diodes may be common to all
drivers.  Note that although the gate runs at 3V or so, the drain will
drive down to almost the source voltage of about 1V.

  What has not been mentioned in regard to the various schemes of drive
using voltages greater than the PIC supply is the requirement for two
transistors per anode driver, an NPN driven by the PIC to drive the PNP
whose emitter is at the higher voltage rail, with a base resistor in
each and for effective turnoff, a pullup between base and emitter of the
PNP.  Even using ULN arrays, this is a fair component count.

> This won't be too bad, as the very noticeable beat between the
> multiplexing and ripple will be stationary thus not very noticeable.
> If the frequencies are slightly different the effect is a moving
> pattern of bright and dimmer.

  It occurs to me that if the PIC strobe algorithm is synchronised to
the mains, four digit drives could occupy a quarter cycle each and thus
each have the same integrated brightness.

  Cheers,
        Paul B.