Having designed many PLL's over the years (mainly RF & uwave), resonant peaks between cap's can often cause problems. This is mainly due to the need to decouple effectively between DC and "blue light"! LF due to the loop filter frequencies and "blue light" to supress harmonics of the RF. It is a common solution to actually add some series R with selected caps to kill the amplitude of any peaks. This, rather counter-intuitively, can also be effective in killing EMC related problems.

I did see an article (in print) some years ago describing this in some detail but have never come across it on the web. For myself I try to maintain a list of common caps measured in a 50 Ohm system with details of their SRF, peak attenuation etc. This enables relatively rapid assessment of where peaks are likely to cause problems.

I certainly need to update this list at some stage and often think I should create a spreadsheet so I can quickly analyse a particular set of caps. One of my regular clients (a major processor vendor) still gets berated by me when they bring out new CPU's with clocks up to a few GHz and still recommend decoupling with 0.1uF on every power pin. They seem to refuse to catch on that their clocks are way above the SRF of the decoupling!

Having said all that, I have never really encountered such a problem in audio gear.

Thanks for that!
Nobody wants to change the 0.1uF value! Someone must have stocks in 0.1uF caps......

I worked with a Austrian designer who started in audio in the "glory" days. He avidly recommended a series resistor for decoupling caps too. He said it gave him more control over what frequencies he let on the power bus. This seems to gel with what your saying.

Well 0.1uF is still very useful as CPU's also generate lots of sub-harmonics, they just don't do much at the clock frequency.

One other thing I do is to usually spread values in 2 decade steps i.e. 100:1 (depending on packages and dielectrics) which of course aims at about 10:1 steps in SRF (for similar cap packages). This usually controls the magnitude of resonant peaks which seem to get much worse with closer SRF's. Of course physical layout is also paramount to this working effectively. Some engineers think I am being daft when I specify things like 47pF decoupling caps but then they seem to have no idea what the SRF of a 47pF 0603 is! Nor do they appreciate the benefits of COG/NPO dielectrics over X7R etc. etc.

Add to that the fact that ceramic caps are also basically piezo electric devices and they can get very bemused!

Temperature Stability is good thing.

Another find:
http://www.designers-guide.org/Design/bypassing.pdf

Link555, post: 368746 wrote: Temperature Stability is good thing.

I'm not sure if you're referring to my comment about COG/NPO v X7R here but that is not what I was meaning. Due to normal tolerance spreads on a production run I do not find temperature stability at all important for decoupling as it will still have less effect than the tolerances. If the decoupling is properly sorted out, temperature should not be a major factor. If it is then it needs a rethink. I have had to test many designs over -55C to +125C and have yet to see a serious temperature problem.

What I meant by the comment on ceramic dielectrics was the difference in the losses (tan delta) of the different dielectrics which can have a significant impact on decoupling schemes (just like ESR).

Of course you were probably not referring this at all! :<)

Cool thanks for that, I now get what your saying. My head was somewhere else.....I saw a linear opto-isolator design where the decoupling caps were spec'd X7R, and that coupled with opto-isolator caused the output to swing %35 at 55C. When they put in NPO caps the swing became more manageable. Thats what I was thinking.