oversampling?

February 12, 2010 9:46 AM

Are there any crackpot signal processors in the house?

I am working on a data set which is sampled right at Nyquist in X, in Y, and in time for reasons of budget. I hate this. I have a vague memory of seeing an erstwhile bona fide document many years ago regarding sound engineers recording at 10 X Nyquist and noticing a difference on extreme subtleties (I swear the example they used was rapid pitch changes by opera soprano Cecilia Bartoli). I am sure the consensus expert opinion is a little room past Nyquist to roll off your filter in the Digital to Analog conversion process is scientifically factually the only thing real humans ever should pay for. My Oppenheim and Schafer is well-worn. Still I am not convinced. Does anybody remember the document that I remember? Is there another forum I could post this question without compromising my professional reputation?

posted by bukvich to technology (7 answers total)

What's the specific problem you're trying to address?

If your data is already recorded and sampled, you can't really do anything more to it to address aliasing issues. The aliasing or lack thereof has already happened.
posted by Jpfed at 10:26 AM on February 12, 2010

I don't know what paper you're looking for, and perhaps you already know this, but Rupert Neve is the king of hi-rez audio crackpots. He talks about this, and has definitely mentioned studies where they've been able to find people who can distinguish shifts in cutoff in the 200kHz range, or at least behavior that motivates him to engineer for that frequency.
posted by rhizome at 10:33 AM on February 12, 2010

Jpfed: the specific problem I am trying to address is I want to resurrect the document from my past that I poorly recall which said 10 X nyquist worked somewhere once.

rhizome: I am an audio dilettante; I work in seismic. I don't recall hearing of Rupert Neve but from what I see so far on the first couple search result pages returned from google your answer is looking very good. If you are willing to volunteer names of queen, prince, or jester of audio crackpots I am interested in reading it.

Thank you so far!
posted by bukvich at 10:47 AM on February 12, 2010

Looks like a very strange definition of Nyquist to me. Are you sure they weren't just arguing that the top frequencies for the specific domain was much higher than previously assumed?
posted by effbot at 11:19 AM on February 12, 2010

Neve is kind of in a class by himself, that's what makes people crackpots! He is by far the most outspoken proponent of high-bandwidth audio, of which his Tape-Op interview gives a decent sketch. I don't really have much but Google beyond that, I guess I'm a dilettante as well, only from another direction. :)
posted by rhizome at 11:42 AM on February 12, 2010

Trying to "confirm that something worked somewhere once" is a noble goal I guess. But the purpose of the Nyquist frequency isn't really to set an upper bound on sampling frequency. The point is to set a lower bound, below which signal quality will degrade substantially.

10x Nyquist would "work" every time it just requires faster equipment and more samples per unit time. There is nothing inherently wrong with it, it is just wasteful (not in a "pollute the earth" kind of way, but in a "not really any redeeming incremental improvement for the effort" kind of way).
posted by milqman at 12:57 PM on February 12, 2010

The critical factor in digital signal processing is the anti-aliasing low-pass filter before sampling. This filter is supposed to remove all frequency content of the input signal above the Nyquist frequency, which is one-half the sample rate. Perfect anti-aliasing filters do not exist. They have some finite roll off beyond the nominal cutoff. Therefore you want to over-sample beyond the nominal cutoff frequency to reduce aliasing.
posted by JackFlash at 2:59 PM on February 12, 2010