home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: rec.audio
- Path: sparky!uunet!zaphod.mps.ohio-state.edu!sol.ctr.columbia.edu!news.columbia.edu!cunixa.cc.columbia.edu!gmw1
- From: gmw1@cunixa.cc.columbia.edu (Gabe M Wiener)
- Subject: Re: Anti-aliasing on the recording end?
- Message-ID: <1993Jan21.205811.9048@news.columbia.edu>
- Sender: usenet@news.columbia.edu (The Network News)
- Nntp-Posting-Host: cunixa.cc.columbia.edu
- Reply-To: gmw1@cunixa.cc.columbia.edu (Gabe M Wiener)
- Organization: Columbia University
- References: <shetline-210193103508@128.89.19.74>
- Date: Thu, 21 Jan 1993 20:58:11 GMT
- Lines: 44
-
- In article <shetline-210193103508@128.89.19.74> shetline@bbn.com (Kerry Shetline) writes:
- >It has obviously become popular in digital playback systems to use
- >oversampling as a way to provide digital-domain filtering, allowing for the
- >use of much less severe analog filters. But what's going on these days on
- >the recording side?
-
- Well, the common way for a long time was to use 2x oversampling. Let me
- explain why.
-
- On playback, the problem is this....how do you turn the quantized wave
- into smooth analog information? And there were (are) two choices.
- Either you filter the hell out of the thing (brick-wall style) or you
- oversample and interpolate, and filter gently later. Thus, the more
- oversampling you do, the less severe your analog filtering has to be.
- This is called anti-imaging filtering.
-
- On the record side, the problem is different. Once again, you need to filter
- out everything above 20 kHz, but not for the same reason. The problem here
- is with ultrasonic harmonics, namely their interaction with the sampling
- clock. A 20 kHz complex wave, even though we only hear it as a sine wave
- (fourier, ya know), still has ultrasonic harmonics, and the interaction
- between 40 kHz and 44.1 kHz is, ah, quite audible :-). So in a regular
- 44.1 system, you'd have to have a solid cutoff at 22 kHz or else anything
- above that would alias. In an oversampling system, the clock rate is run
- much higher, say 88.2 or 176.4. Now, you simply sample everything four
- times, and discard three of the four samples. The advantage here is that
- now your clock signal is much, much higher, and you can have a much
- gentler slope filter. If you have a 2X system, your clock is 88.2, so
- you can attenuate gently until 66.2 (or 60.2 usually... a 2 kHz guard
- band is standard). The same applies for higher rates.
-
- Once the data is sampled, the extra samples are decimated and then the
- 44.1 kHz signal is recorded.
-
- There are more modern techniques (1bit, etc) which you can read about in
- Pohlman if you're really interested :-)
-
-
-
- --
- Gabe Wiener - Columbia Univ. "This 'telephone' has too many shortcomings
- gmw1@cunixa.cc.columbia.edu to be seriously considered as a means of
- N2GPZ in ham radio circles communication. The device is inherently of
- 72355,1226 on CI$ no value to us." -Western Union memo, 1877
-