Distinguishing ceramic from film capacitors electrically?

[T3sl4co1l]

I im aware of the standard piezo microphonics what im wondered about was "studies" made on fans and hardisk vibrations. Surely this depends om cap material, mounting material PCB, distance of vib source, microphonic bandwidth, resonant frequencies of/on the board, etc, etc, etc.

Well, yeah, it depends on all of those things.  What good would a study be?

Alternately -- are you simply looking for a claim that "measurements in [typical server rack] showed XmV of noise attributable to microphonics"?  What significance would you draw from it?  Is the measurement bigger or smaller than you were expecting, and if so, why?  If they are able to attribute the signal to some source or another -- if indeed the signal is separable at all -- since, that's the point of such an environment, there's just a lot of stuff going on?

From a systems point of view, the approach is simple -- there is only one mechanism by which voltage gets induced: strain (flexural or otherwise) of the component of interest.  Knowing the dimensions and elastic modulus of the component, you can attach it to a mechanical model of the PCB, and other components and enclosure and vibration sources, and it's all just a mechanical filter network from source to target.  Whatever the gain is, that is your answer.

Tim

[MT]

Well, yeah, it depends on all of those things.  What good would a study be?

I dunno, just interested.

Alternately -- are you simply looking for a claim that "measurements in [typical server rack] showed XmV of noise attributable to microphonics"? 

As said not looking for claims just interested to see outcome if any, clearly there are in Daves tap oscilloscope tests,
Servers i dont think it matter much so much due to clock/date noise anyway.

What significance would you draw from it?

Earth is flat?

Is the measurement bigger or smaller than you were expecting, and if so, why?  If they are able to attribute the signal to some source or another -- if indeed the signal is separable at all -- since, that's the point of such an environment, there's just a lot of stuff going on?

Tap the front of the oscilloscope.

From a systems point of view, the approach is simple -- there is only one mechanism by which voltage gets induced: strain (flexural or otherwise) of the component of interest.  Knowing the dimensions and elastic modulus of the component, you can attach it to a mechanical model of the PCB, and other components and enclosure and vibration sources, and it's all just a mechanical filter network from source to target.  Whatever the gain is, that is your answer.

You excluded unknown resonances that might appear here and there.

[T3sl4co1l]

What unknown resonances?  They are only unknown if you didn't include them in your model!

You could even take a finished product to a vibe lab, give it a shake (at lowish amplitudes so as to maintain linearity) and measure acceleration at a component (or its voltage directly).  The empirical method solves any issues with modeling.  (Or even, do both and adjust the model until you've matched reality, and bring those insights into the next model and so on. )

Outcomes are simple for the most part: most digital logic isn't going to care, and most analog circuitry is going to have either some coupling factor (via PSRR, or such caps directly in the signal path), or some parametric effect (e.g., microphonics acting upon a compensation capacitor in a PLL, or the digital equivalent being supply variation causing jitter).

The oscilloscope test is simply the former: caps in the signal path, the amplitude of which may or may not depend on range, hence you might get 10mV of noise on one range or 1V on another.

Hard drive racks are sensitive to noise simply because they contain mechanical elements, extremely precisely positioned; I would guess the actuators are entirely digital control, so that there's no capacitor in their signal path.  Supply noise may be a small contributor, but again it may manifest more as jitter, and the data streams are self-clocking and heavily error-corrected, so it should make little difference to that end.

Tim