home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: rec.audio
- Path: sparky!uunet!stanford.edu!nntp.Stanford.EDU!kong
- From: kong@leland.Stanford.EDU (Kong Kritayakirana)
- Subject: CD Sound Quality (again?) -- Ideas
- Message-ID: <1992Dec28.084012.15920@leland.Stanford.EDU>
- Sender: ?@leland.Stanford.EDU
- Organization: DSG, Stanford University, CA 94305, USA
- Date: Mon, 28 Dec 92 08:40:12 GMT
- Lines: 63
-
- Digital Recording: Part I: ADC
- ------------------------------
- Assuming you have good microphones and pre-mics, etc. needed to make good
- sounding recordings. You should use something like 22-bit 128x oversampling
- ADC. (Hey maybe we'll be able to get away with 20 bits but I love
- overengineering) Why?
-
- Why 22-bit?
- -----------
- To give headroom for recording engineers so they don't have to set the
- recording level too high and risk having digital "clipping," 2 extra bits
- here provide 12dB headroom and should be pretty safe if he sets the
- recording level to about -10dB when the full orchestra blasts.
-
- Extra 4 bits are used for proper dithering down to 16-bit resolution. When
- proper dithering is done, the quantization noise will become uncorrelated
- with the quantized signal and hence no digital "artifacts" of undithered
- signals (if you want to see what no dithering can cause, pick up a copy of
- Audio magazine and see the plots of undithered vs. dithered low level sine
- waves. The undithered one contains a lot of harmonic distortion. The
- dithered one has somewhat higher noise floor, but harmonic distortion is
- hardly present).
-
- So 2-MSB-bit headroom + 4-LSB-bit for dithering and we have pretty cool
- 16-bit digital audio sound with real 98.1dB dynamic range.
-
- Why 128x oversampling?
- ----------------------
-
- The culprit in digital recording is the so-called "antialias filter." The
- filter supposedly filters out everything above half the sampling freq. If we
- use no oversampling during ADC, the ANALOG antialias filter must cut off
- very sharply somewhere near 22kHz. This is BAD for sharp analog filter is
- very hard to be designed and implemented w/low distortion and phase shifts.
- So if we go 128x oversampling (again I LOVE overengineering) the antialias
- filter has to cut everything above 64x44.1kHz = 2.8MHz. The ANALOG antialias
- filter is a lot less likely to screw anything up in 0-22kHz range. Here's
- the picture:
- 2.8MHz "lowpass" 22-bit 128xOS
- Analog Source --> Antialias Filter --> A/D converter --> Decimation/Dither
-
- Aside: The antialias filter used in the diagram above can be very mild with
- very gentle rolloff (and hence has low noise and distortion). It
- must provide good attenuation above 2.8MHz and good pass
- characteristics below 22kHz. The filter characteristic in
- 22kHz-2.8MHz can be anything. Who cares?
-
- So once we get the 128x oversampled data we can do the much less harmful
- digital filtering to filter anything above 22kHz out and use the decimation
- technique to convert it down to 44.1kHz sampling rate. Combined with the
- technique suggested above to cut 22-bit data down to 16-bit, now we have
- very clean 44.1kHz 16-bit digital data ready to be printed on CD.
-
- Comments? Suggestions? Any part left out? Email me.
-
- Coming up next: Why oversampling during DAC (besides marketing reasons)
- ----------------------------------------------
- Kong Kritayakirana (kong@leland.stanford.edu)
- Remotely reading and posting to rec.audio from
- a nice beach somewhere in south pacific. Damn
- the technology that helps us to get in touch.
- ----------------------------------------------
-
-