Media Diligence

Skip to first unread message

Andrew Koenig

Jul 20, 1985, 2:59:01 PM7/20/85
Quote from a news article about Compact Discs

Another item for buyers to ask about is oversampling,
which is the ability of the player to ask the disc for
information more often than necessary.

"What makes the difference in sound at this level is how
hard the computer is working to translate," Harvey said.
Oversampling allows the microcomputer to cover up damaged
information by overplaying the material just before it
to cover up.

So THAT's how they do it!

Douglas Hamilton

Jul 23, 1985, 4:13:28 PM7/23/85

No, it's not. Oversampling refers to interpolating between the
samples actually recorded on the disk.

CD's are recorded with a 44KHz sampling rate which (remembering
your sampling theory) gives a 22KHz bandwidth. The problem is
that if one feeds the samples directly to the DAC's, the output
"staircases" from one sample to the next.

It's desirable to remove this high frequency "switching" noise but
of course any kind of low-pass filtering causes phase shift near the
cutoff frequency (also undesirable.) To minimize this effect, some
CD manufacturers interpolate much the way you would with a trig table
(remember before calculators?) to find intermediate values. By
creating "samples" at an 88 or 176KHz rate to be fed to the DACs, they've
moved the noise up to a much higher frequency, more removed from the
music, thus making it easier to filter the switching noise out without
introducing phase shift.

Oversampling has nothing whatsoever to do with error correction, which
happens at an earlier stage in the processing.
Douglas Hamilton Prime Computer,Inc.
617-626-1700 x3956 Video Products Group
492 Old Connecticut Path
Framingham, MA 10701



Jul 27, 1985, 11:06:35 AM7/27/85
Hamilton is half right, in that interpolation (in the sense that
CD players use it for oversampling) had nothing at all to do
with error correction, BUT, I disagree is a few places.
The first is that he says that "all filters have phase shift".

This statement is, strictly speaking, true, but there is a whole
class of filters (symetric and anti-symetric finite impulse response
filters) that have a TRANSIT DELAY. While this is equivelant to
a phase shift that changes linearly with frequency, it does NOT
mean that there is any effective phase shift in the signal.

In fact, these (symmetric FIR filters) filters are what are
used inthe "interpolation process", not any process that's like
what one used to do with a trig table.

The comments about switching noise are properly called aliasing,
and what interpolation does is remove the aliasing at the low
frequency, and then use a simple, and well behaved filter, to
remove the high-frequency aliasing that is generated instead.

Rabiner and Shaeffer have a good textbook on such things.

"It was great when it all began, ..."


Reply all
Reply to author
0 new messages