Maybe the smart people here can clarify my thinking about how resonances show up with various methods of measuring frequency response. Consider a system with a high Q resonance, maybe a crystal is a good example. Let's say the resonance takes a long time to build up, say many seconds. A quick frequency response sweep shows very little. An FFT of a noise signal, without a huge amount of averaging, also shows very little. You have to slow down the sweep, I assume so energy can be stored, to see the resonance clearly. My question is, is there some relationship between Q or time constant or something, that offers any guidance as to how fast/slow one has to sweep. Is Q the only factor in how fast a resonance can build, or is there some other parameter involved? Is an FFT with a noise signal (rather than a tracking generator) even appropriate for measuring this sort of system?
Hi,
Only a very low frequency high Q resonance (e.g. a bridge)
will take many seconds to build up. Probably there is relation
between the Q and the number of cycles of Fresonant but
in most cases the time is very short, even for very high Q.
Time is related related to Q/Fresonant.
rgds, sreten.
Only a very low frequency high Q resonance (e.g. a bridge)
will take many seconds to build up. Probably there is relation
between the Q and the number of cycles of Fresonant but
in most cases the time is very short, even for very high Q.
Time is related related to Q/Fresonant.
rgds, sreten.
Yes, there is a relationship, it is the basic limitation of of swept analyzers.
I have had some difficulty finding a document still mentioning it, because "modern architectures" are now so trendy that this has become a kind of hidden dirty secret, but it it still very real and present:
http://www.google.com/url?sa=t&rct=...cfPgOOdcubx4axgdUuukpAA&bvm=bv.94911696,d.bGg
Note that sampled (real time) systems are not completely immune to artifacts: even a spice simualtion of a highly selective system like a crystal can produce funny results when the resolution isn't high enough.
If I find some time, I'll post Ltspice sims showing really weird artifacts
That makes sense- I was thinking about some 100 kHz crystals I've got, where the time to come to full amplitude is several seconds, but using the formula it makes sense because the Q is crazy high on a physically large low frequency crystal. Makes me wonder about some of the old GR time standards that used a huge quartz bar.
No doubt I should go back and read the manual and app notes for some traditional spectrum analyzers. Recently I was making some response measurements on mechanical assemblies (non-audio) and the results with the built-in FFT of the scope were useless. I went back to the "old fashioned" swept sine measurements and had no trouble at all. Thanks!
Here is an example of such artifacts: it is a 100KHz crystal filter.
The first pic shows the simulated output voltage at the center frequency, with a timestep of 100ns (1% of the stimulus period, thus a suitably fine resolution -normally-)
A low frequency modulation seems to be superimposed: we see ~3.5 lobes on the 100ms duration.
Second pic is identical, except the timestep is now 80ns. Only ~2 lobes are visible, for a change that is normally insignificant.
Now, which is the correct one? In fact, none of them: in reality, the actual circuit doesn't behave like that at all, there is no VLF modulation, just a regular exponential build up towards the final amplitude, and what we see is a pure artifact, caused by the very high Q of the crystal.
Attachments
- Status
- Not open for further replies.