At 04:19 PM 11/9/99 -0500, Dave wrote:
> > There is apparently never a reason to use higher than 100 uF or lower than
> > 0.1 uF on the regulator output. And with lower values, you don't need a
> > discharge resistor or diode to protect the regulator.
>
>Important point, the 100 or whatever is IN ADDITION to the 0.1uF.
>The larger caps don't work so well at 1-2 MHz, which is what the 0.1 is
>there for.

Again, I'm not an EE. However, I'm not sure I agree with this 'conventional
wisdom'. The point of concern is the impedance of the capacitor at the
frequency of interest. Although smaller capacitors may have less lead
inductance, the higher capacitance of the larger ones may result in a lower
impedance at that frequency, giving better performance.

Any analysis I've seen has always indicated that higher capacitance is
better, even at reasonably high frequencies (e.g., 20 MHz). E.g., a 10 uF
tantalum is superior to a 0.1 uF tantalum or ceramic.

Perhaps Paul or Michael have something to contribute on this issue.

Robert A. LaBudde, PhD, PAS, Dpl. ACAFS  e-mail: ral@lcfltd.com

Well, at work I play with laser drivers working at 2.5 GBits/s, and electrolytics certainly don't do very much at that frequency.  I've never personaly measured the losses in an electroylitic, but if you take a look at almost any RF circuitry, you won't find many high value decoupling caps.  Maybe one or two for low frequency bypass, the rest will be small ceramic devices.  As you say, larger caps will have a higher inductance and ESR than smaller ones, and this will become the dominant factor at high frequencies, swamping the very small capacitive reactance.

As further evidence that the 2200uF output cap in the original design was superfluous, this is straight from the Nat Semi datasheet:

"Normally, no capacitors are needed unless the device is situated more than 6 inches from the input filter capacitors in which case an input bypass is needed. An optional output capacitor can be added to improve transient response"