Audio, hi-fi and car audio - Why do SACDs sound better? (Soft troll)

In article , Jim H
wrote:
Jim Lesurf wrote in uk.rec.audio
[snip]

The problem for me, here, is that - as in some other places in what
you write - you seem to use terms in ways that differ from the
standard definitions employed by metrologists and information
theorists/engineers in textbooks, etc.

I got this from one of my old (e-) textbooks, A Practical and
Authoritative Guide to Contemporary English, Science Terms:
Distinctions, Restrictions, and Confusions

"Imagine that you are a scientist with a measurement or calculation to
your credit that has taken years of meticulous work. When your results
are published, you find that your techniques are praised for their
precision and your results are criticized for their lack of accuracy.

Need to interpert the above with care as the words could be assumed to have
various meanings. For example, people might interpret the word "precision"
in the above to mean "rigor" or "clarity" or a lack of ambiguity. These are
all slightly different, and may not be the same as the definition in terms
of metrology/IT. See below... :-)

How is this possible? We usually think of accuracy and precision as
pretty much the same thing. But in science, these words are used in
significantly different ways. A result is considered accurate if it is
consistent with the true or accepted value for that result.

Yes, the standard definition is IIRC along the lines of the last sentence
above.

The precision of a result, on the other hand, is an indication of how
sharply it is defined.

Yes, as above. :-)

[snip]

The problem is that any sampled digital signal pattern which was
sourced from an 'analog' sound-pattern original will then normally be
reconverted back into an 'analog' form when you come to listen to it.
This process, if carried out according to the requirements of basic
information theory, results again in a 'continuous' output waveform
that can convey as much information as was in the original.

Ok. I suppose there's no such thing as a completely digital system,

Bear in mind that 'analog' and 'digital' are *models*, they are not
physical reality. They are ways in which we try to use physical properties
to convey or hold information. The EM fields, etc, don't know anything
about 'analog' or 'digital', and the currents on wires behave according to
Maxwell (if we are correct) in either case.

and so past the digital - analogue conversion the signal can no longer
be considered to have the benefits of digital.

Yes. In audio, the purpose of the digital signal pattern is to provide the
information from which an analog pattern is created which should bear a
strong similarity to the 'original' analog pattern(s) which were sampled.

Unless we can measure things with infinite precision and accuracy (we
probably can't since the random processes in this universe make this
meaninless if our current understanding of physics and information
theory are correct) it becomes meaningless in practice to think of the
orginal as being 'irrational' (by which I suspect you actually mean a
'real' which requires an infinite number of digits, as opposed to a
finite 'integer').

By irrational I mean what I was taught at school and used at college, a
number which is not expressible in terms of a over b, which would mean
an infinite number of digits, if expressed decimally. An irrational
number is part of the set of Real numbers.

Am I right in suspecting you program in c?

Yes. :-) Also C++, BBC BASIC, FORTRAN, and Java as well as occasional
burst of assembler when I feel the need to make my head sore... ;-

However my point is that when we are talking about physical quantities like
EM fields we must be careful with 'irrational' since we don't know the
precise value, only approximate ones with finite accuracy and precision.
Irrationals are so 'by definition' if you see what I mean. Information
theory and physics generally stops us from saying if a given electric field
level, say, is 'irrational' in value or not. Also, it might depend upon our
units. Consider converting angles in radians to angles in degrees. :-)

For the above reasons, I think it makes sense to stick with reals or
integers when discussing most physically real quantities relevant here.

*Any* measurement or recordings (including analog ones) will contain
'errors' (imperfections limiting the information content) in the forms
of noise, distortions, etc. One of the basic axioms of information
theory is that 'analog' and 'digital' systems have the possibility of
being indistinguishable in this respect when conveying signal pattern
*provided they are done adequately*. In both cases, therefore, the
problems are in the domain of engineering, not an inherent theoretical
difference.

Yes, at the recording. But, once digital masters have been made the
descrete data should not contain any further error by the time it
reaches its destination.

Unfortunately, noise and other forms of error still arise in the digital
domain. The advantage of digital signals is that they have some immunity
and they make processes like error detection and correction easier than
with analog. However errors will still occur due to noise, etc.

The distribution of data analogue containing continuous values will
always introduce futher errors.

The problem with analog is that (for the normal systems relevant here) the
system has no way to reliably seperate noise and distortion introduced by
the medium of transfer from the intended signal.

By distribution I mean either the copying to a physically distributable
medium, or some other kind of transfer, such as radio or the internet.

I suppose to be completely correct, when a say a digital signal 'should
not contain any further error' I really mean 'may not' in that it is
conceivable and likely that by the time the data reaches your DAC it has
been unchanged by the copying, and is identical to when it left the
studio.

Yes. Ideally, it will be unchanged by transmission/storage. However both
random and systematic errors do occur. Fortunately, we can employ
redundancy, etc, to deal with this successfully *most* (but not all) of the
time. Hence, for example, a CD with a channel bit error rate of 1 in a
thousand can then usually play with no output errors at all.

[snip]

And then a lot of audiophiles aren't interested in resonable ;)

Long may it remain so! :-) So far as I am concerned, one of the reasons
we have mad progress over the years is that people keep finding that they
want improved sound systems, recordings, etc.

Btw, anyone know what kind of interpolation is normal for cd audio? If,
as I suspect, polynomial what is the highest power?

Vendor and system specific I think. Nominally, they should all be using a
corrected sinc if they just follow information theory. However when you
look at the various inband response patterns it becomes clear they differ
at times. I imagine that many makers just impliment with Philips or the
other chosen chip-makers supply as standard. However I'm pretty sure that
makers like Meridian 'do their own thing'.

I suppose it could just be linear, in which case only two points are
considered at a time. I remeber reading that some soundcard or other
uses '8-point interpolation'. Typically vague, but if this means what I
think it does, that's a seventh- order polynomial aproximation and a
startling boast,

Need to distingish between 'interpolation' and 'reconstruction' here. In CD
audio 'interpolation' is usually used to mean the recovery of a sample that
has gone missing from the series that should be on the CD, hence requiring
some form of educated guesswork. 'Reconstruction' is the process which may
include generating oversamples which fit in between the known (from the CD)
sample values. In principle. these are not guesswork at all, but the values
we *would* have obtained if we'd bothered to measure the original at these
instants. This follows from ensuring the original was sampled according to
the sampling theorem, etc.

Assuming you mean reconstruction...

Even the first generation Philips chipset used far more than the closest
two sample values to interpolate the oversamples and perform digital
filtering. Don't know values for current chipsets, but I'd guess than
between tens and hundreds of samples are used for each interpolation. May
even be based upon 'infinite' response methods where the total becomes
vague but 'large'.

For interpolation something similar will apply, but again may differ from
one maker to another.

For vinyl it is less clear what the smallest possible unit is,
because for most day to day purposes we consider solid objects to be
continuums, so I suggested vinyl atoms as a side point.

In practice, if you work out the random noise levels that would arise
due to the molecular size of the material used, the stylus area, and
thermal effects, you should find you end up with a noise level not
very far below the best LPs. The problem here is that many LPs are
relatively poorly made compared with this physical limit. Hence many
LPs produce more noise than this.

In practice, The smallest possible unit of vinyl recording is probably
the width of the cutter's needle.

This will change during recording of the LP, though. :-)

One thing that always surprises me is that records don't seem to sound
any worse at the middle than at the edges, despite the music being
stored on less physical medium. If anything the sound is better near
the middle. Any idea why this is?

Well, my experience in the past is that a number of LPs do sound quite
different near the center than near the edge.

Bear in mind that those cutting the LP are aware of this problem. One trick
they used to employ was a pot connected to the cutter head via a bit of
string. As the cutter moved in, it pulled the string and rotated the pot.
The pot was part of a treble control, turning down the level of treble near
the end of the LP. Experience showed LP makers that if you change this
slowly over the 20mins or so of an LP most people do not notice. :-)

Also, skilled LP makers would manipulate the sound level, treble level, and
difference signal level, according to the circumstances in order to
disguise any problems. I assume they still do this, but don't know if
string is still involved. :-)

[snip]

However, the above is nothing to do with 'digital' as such. That just
makes the explanation easier. We could just as easily have used two
analog dial meters - one with a tiny, cramped scale, the other with a
much larger scale with finer divisions and longer pointer. Similarly,
as an example, we could also have compared a wooden stick metre rule
with a clock gauge when trying to measure a physical length.

Yes, but I was really talking about the properties in both as a
distribution medium. To be honest, with this being a topic about medium,
I didn't think about the recording stage. In my mind the masters had
been made and I was thinking about what happens from there.

The output from the recording is the input to the distribution. From
your example, it wasn't the taking, but the publishing of the
metrologist's results I was trying to discus. It gets difficult to give
a parallel now, I can't think how s/he could realistically publish in
analogue.

Above is a fair point. However when using LP or CD (or other systems) to
convey music, my own interest is with how well the (assume stereo) pair of
signal patterns emerging from the amplifier matches than being sent to the
system used to cut/make an LP or CD. I regard the cutter or CD burner as
part of the channel in this respect.

Slainte,

Jim

--
Electronics http://www.st-and.ac.uk/~www_pa/Scot...o/electron.htm
Audio Misc http://www.st-and.demon.co.uk/AudioMisc/index.html
Armstrong Audio http://www.st-and.demon.co.uk/Audio/armstrong.html
Barbirolli Soc. http://www.st-and.demon.co.uk/JBSoc/JBSoc.html