home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!cis.ohio-state.edu!zaphod.mps.ohio-state.edu!uwm.edu!csd4.csd.uwm.edu!info-high-audio-request
- From: matthews@eecs.ucdavis.edu (Thomas W. Matthews)
- Newsgroups: rec.audio.high-end
- Subject: Re: Digital processing of sound
- Date: 14 Nov 92 19:24:51 GMT
- Organization: Division of Electrical Engineering and Computer Science, UC Davis
- Lines: 74
- Approved: tjk@csd4.csd.uwm.edu
- Message-ID: <1e89huINNjss@uwm.edu>
- References: <1dof19INN85b@uwm.edu> <1drc7hINNl9q@uwm.edu>
- NNTP-Posting-Host: 129.89.7.4
- Originator: tjk@csd4.csd.uwm.edu
-
-
- I just read this section:
-
- >However, we may be able to show that the process of filtering at 22kHz can
- >inject substantial noise and distortion back down into the audio band.
- >Consider a tone of 20kHz, being amplitude modulated at 7kHz. This might be a
- >_very_ simple representation of a cymbal. Ignoring the time domain, the
- >spectrum of this signal (a 20kHz tone regularly varying in amplitude) will
- >consist of three spikes - one at 20kHz, where we would expect, and one at
- >each of 20kHz +/- 7kHz, ie 13kHz and 27kHz.
- >
- >In the ordinary course of events, playing back these three signals will
- >result in the original signal. This happens in the human sensing mechanism,
- >because the body is still sensitive to the 27kHz tone, it combines all three
- >and we 'hear' the original cymbal.
- >
- >However, pass this signal through a brick wall filter at 22kHz, and we now
- >have only two tones, one at 20kHz and one at 13kHz. No matter how well these
- >are then reproduced by the playing equipment, there is no 27kHz sideband to
- >combine with the 13kHz sideband to modulate the 20kHz carrier. The result is
- >a distorted version of the original 20kHz tone, and a new and completely
- >unrelated 13kHz tone that wasn't in the original signal!
-
- To me, this implies that the process of filtering ALONE will introduce
- aliassing. It will not. (if the filter has non-linearities, intermodultion
- distortion can be produced, but that is not the issue here).
-
- The writer does make an interesting point: A 20 kHz. tone cannot be
- amplitude modulated at 7 kHz. unless a 27 kHz. component is present.
- Traditional thinking says that since one cannot hear 27 kHz. by itself,
- its presence or absence will not affect the perception of the amplitude
- modulated sound described. Have any listening tests ever been done
- that isolated this one specific issue? I would like to thank the author
- for bringing up a very "sound" argument against deliberate filtering.
-
- The posting goes on:
-
- >Proponents of Nyquist (but who haven't studied his work) will say that his
- >theorem states that all signals up to half the sampling frequency can be
- >reproduced. This is precisely what's happening in this case, but the
- >information content of the signals below half the sampling frequency is being
- >eroded, and noise is being added.
-
- Acording to standard definitions used in communications theory,
- the "information content of the signals below half the sampling frequency"
- never did include the 27 kHz. component. (As a matter of fact, the 13 kHz.
- component had the same information as the 27 kHz. component because the
- modulating 7 kHz. tone was real-valued. This is why single-sideband AM
- transmission works.) I recognize that I'm being a little picky about
- terminology here, and I hope I haven't crossed over the line into being
- a jerk over this minor point :-) .
-
- The last paragraph:
-
- >It is this injection of noise back into the audio band that is subjectively
- >the most objectionable component of digital sound. The ear is apparently
- >tolerant of noise and/or distortion, as long as it is a harmonic of the
- >original signal. It is extremely sensitive to signals that are not harmonic.
- >I was told that this is easy enough to demonstrate in a laboratory using
- >signal generators - when a test tone is playing, harmonic signals (several %)
- >may be introduced with very little subjective effect, but fractional
- >percentages of unrelated signals are extremely easy to detect. The ear is
- >very good (because it practices every day, in the world around us) at picking
- >out small sounds from background signals.
-
- makes a strong argument that aliasing is far less tolerable than the same amount
- of harmonic distortion. Once again, though, aliasing arises from the sampling
- operation, not the filtering. And to avoid aliasing when the Nyquist rate
- is so close to the desired bandwidth, sharp cutoff filters must be used. An
- earlier posting makes a srrong case in favor of realizing these filters in the
- digital domain (but only because they've got to be there somewhere!).
-
- Tom Matthews
-
-