[net.audio] How many bits, really?

emrath@uiuccsb.UUCP (02/05/84)

#N:uiuccsb:5700026:000:635
uiuccsb!emrath    Feb  4 02:36:00 1984

I have read here and in the mags that some players use 14-bit converters
and some use 16-bit converters and some use oversampling.
I have not read, however, what the actual format on the disc is.
How many (data) bits per sample and how many samples per second
are actually on the disc?  (I do know the fundamental sample rate
is 44.1kHz per channel)  Does anybody have a reference?
(I don't know much about digital signal processing, but I had this one idea
that digital filtering and oversampling could be similar - interpolate
extra numbers between the known numbers and feed them to the D2A).

BTW: How do you prefer to spell disk?

gregr@tekig1.UUCP (Greg Rogers) (02/09/84)

*************************************************************************

The following is a reply to a specific question about the techniques
used with CD players and is a simple statement of facts.  I will express
no opinions about the sounds of CD's in general nor about differences
in sound (if any) between techniques.  I believe far too much subjective
opinion on these issues has already nearly destroyed the usefulness and
pleasure of participating in this newsgroup.  I hope any followup to 
this article will refrain from subjective comments as well unless supported
by some scientifically acceptable test data.    Thank You

***************************************************************************

In answer to your question about types of CD players, 14 bit vs 16 bit etc.

There are at the moment two distinctly different approaches being used in
CD players to transform the digital data to an analog signal.  As you gathered
one method (initially used by Phillips) utilizes a 14 bit D/A converter,
digital filtering, and oversampling for conversion, while the other method
(initially used by Sony) utilizes a 16 bit D/A converter and analog filtering
only.
The digital data on the CD disc has a resolution of 16 bits, and a
sample rate of 44.1 kHz on all CD's and is independent of the players
digital-to-analog conversion technique.  The actual data on the disc
is encoded in a highly complex manner to ensure the integrety of the data 
even if flaws in the disc surface destroy some of the information.  In most
cases "lost data" can be completely reconstructed "on the fly" in real time, 
in severe cases interpolation will be used if necessary (rarely needed), or
in the cases of last resort controlled muting is used (very rare, analogous
to the tick and pop eliminaters used in analog systems).  Just for information
the technique used is to first encode the data using a system known as CIRC
(Cross Interleaved Reed Solomon Code), add C&D (Control and Display) bits,
and then store the data on the disc using a technique known as EFM (Eight to
Fourteen Modulation) which is very analogous to techniques used with computer
floppy disks to simplify the hardware requirements to read and write the disks.
The reverse process is used when playing the CD which results in a complete 
recovery of the 16 bit/44.1 kHz data ready to be transformed by either of the
D/A conversion systems mentioned above.  Before describing those a little 
further it should be noted that although the techniques described above would
appear to be very complicated to people unfamiliar with them, in reality the
of CD's over many years.  What I'm trying to say is just because you ain't 
heard of them don't assume they are unproven/experimental or even new.  They
ain't.  At this point let me give you a few figures that actually describe
the capability of the error detection and correction system in terms of the
physical disc.

	Maximum COMPLETELY correctable burst length         4000 data bits
							    = 2.5 mm on disc

	Maximum Interpolatable burst length		    12,300 data bits
							    = 7.7 mm on disc

This means that scratches or imperfections in the disc less than 2.5 mm long
in the direction of the track (around the disc, on the track) will be TOTALLY
corrected!!  Imperfections longer than 2.5 mm but less than 7.7 mm will be
interpolated.  Imperfections longer than this will be muted in a very well
controlled manner.  

Returning now to the final step - Digital to Analog Conversion.  The 16 bit
D/A conversion method (Sony) is conceptually the easiest.  The 16 bit data at a
44.1 kHz sample rate is passed to a 16 bit digital-to-analog converter.  Because
the signal has been digitized the resulting output of the D/A contains
frequency components above the original audio signal.  These need to be filtered
out since they were not part of the original audio signal, and could overload
amplifiers, create intermodulation products and so forth.  In the Sony system
a high-order analog Bessel filter follows the D/A converter to filter
out these unwanted frequencies (I believe it is a 9th order filter, anyone
positively know for sure?).  The signal is then ready for your amplifier.
This simple straightforward approach seems rather expensive.  The cost of the 
precision 16 bit D/A should be higher than a 14 bit D/A as used by Phillips
and the multiple section analog filter must be trimmed to very high tolerances
which is also rather expensive.

In the Phillips system the digital data is filtered prior to the D/A converter
using a digital transversal filter.  Although the 16 bit data is only available
at a 44.1 kHz rate, three additional values of 0 are inserted between each 
actual sample within the digital filter so that data leaves the digital filter
at a 176.4 kHz rate.  (Don't let the fact that the inserted values are 0 bother
you too much,  without an understanding of the mathematics of digital filtering
this simply can't be explained.  It does however work and again nothing is 
really new about this.)  At this point the data is rounded of to 14 bits in a
rather fancy way and presented to a 14 bit D/A converter that operates at
176.4 kHz.  The output of the D/A is then filtered using an analog Bessel filter
of the 3rd order.  The net result of the digital filtering, the fancy 14 bit 
round off, and the 4 times higher sample rate D/A, is to produce a signal/noise
ratio of about 97 dB, essentially identical to the 16 bit D/A method.  Since
the digital filter and the round off circuit are both contained in a single
custom integrated circuit and the D/A is 14 bits (although has to operate
at 4 times the speed) instead of 16 bits I would expect the Phillips approach
to be considerably cheaper.  

If the Phillips system is really cheaper why didn't Sony use it also?  Well
I'm only speculating about cost and perhaps Sony can build and trim their
high order analog filter very inexpensively using automated means?  Perhaps
Sony had their 16 bit D/A already available to them from other projects
(PCM tape decks) and Phillips would have needed to develop their own 16 bit 
D/A?  Perhaps it was easier for Phillips to design their digital filter IC 
than for Sony?  If anyone else on the net has any insight as to why Sony 
went one way and Phillips went the other please speak up.  In any event 
I guess thats what makes engineering interesting, more than one way to skin 
a cat.  More interesting from the listeners point of view is if either
company really (I don't care what they say in their advertisements) believes
one approach SOUNDS better than another.  The Phillips approach does give
a better high end phase response although we can debate the audibility of
this parameter forever.  (No correct that - You can debate it, I give up
that uselessness.)  What other measureable parameters might one approach
favor over another?  I haven't thought this out yet, maybe someone else
more knowledgeable than I in this field has some ready answers.  

Well I hope this has answered the original question.  There is plenty more
that can be said but it all gets real technical real fast and probably
doesn't belong here.  If anyone finds out more technical details on Sony's
filter please let me know.  Again please DON'T take this opportunity to
open a debate over which system sounds better unless you got some real
scientific evidence to offer.

Keep hoping something useful is going on here but no longer convinced,
					Greg Rogers