SACD fundamentally flawed?
Tim Anderson
Warning: PDF:
http://www.discwelder.com/pdfs/1-Bit%20SD%20is%20Unsuitable%20paper.pdf
Quote: "Because of the insoluble theoretical problems discussed in this
paper, we are unaware of any way to generate a Super Audio
CD test disc which is both distortion-free and has a constant,
signal-independent noise floor! In the multi-bit domain, this is
easily done using standard dithering methods."
Tim
Steven Sullivan
> Warning: PDF:
> http://www.discwelder.com/pdfs/1-Bit%20SD%20is%20Unsuitable%20paper.pdf
It has been discussed here explicitly in at least one thread which started
with this post:
it has also been referenced obliquely in other threads.
--
-S.
"They've got God on their side. All we've got is science and reason."
-- Dawn Hulsey, Talent Director
Greg Berchin
wrote:
>>Has this paper been discussed anywhere?
The subject was heatedly discussed on rec.audio.high-end back in
May of 2001, under the thread: "LPCM vs DSD: The war continues at
AES #110". Try Google.
GB
Karl Uppiano
professional digital audio. But the article doesn't mention SACD
specifically. Many mass market digital audio converters these days (used in
CD, DVD, and probably SACD players) do use 1-bit Sigma-Delta conversions,
though.
"Tim Anderson" <tim...@hotmail.com> wrote in message
news:brcou...@enews2.newsguy.com...
Karl Uppiano
SACD in their article. That's very bad news for Sony. I only hope for
everyone's sake, that the single-bit architecture isn't embodied in the
standard, but is merely an implementation detail. Those two guys wrote the
book on dithering and linearizing digital audio. Their expertise is what
makes the difference between the "digital sound" that Neil Young hates so
much and "digital done right".
"Tim Anderson" <tim...@hotmail.com> wrote in message
news:brcou...@enews2.newsguy.com...
Bruno Putzeys
be a valid choice when the analog circuit does not have performance higher
than the 1-bit signal. To use this as a data format, thus binding everyone
to the noise and distortion limits, is quite another thing...
SACD is a typically Japannish invention in that it is a solution to a
nonexistent problem (decimation-interpolation), which in turn creates some
very real problems left for real engineers to solve. Some examples:
1. Splicing (editing) two DSD signals together creates a "click", even if
both represent silence.
2. Any processing (except delay :-) results in a longer word length. Getting
back to 1-bit requires another stage of deltasigma modulation. Sony dreamt
of a new signal processing paradigm operating entirely in DSD. It was not to
be - they even officially admit it now. Any quantisation mixes the signal
with quantisation noise. They can no longer be separated. This is not much
of a problem at 24 bits. At 1 bit however... well...
3. The accumulated noise from previous conversions reduces the deltasigma
modulator's headroom. After 5 conversions (e.g. level control, eq, mixing,
fader etc), the modulator already overloads at silence.
4. DSD is not distortion-free.
5. The signal bandwidth and the noise zone overlap. In a correctly designed
converter, the signal occupies the "clean zone" only, thus allowing the
noise to be filtered away. With DSD, the noise zone starts at 20kHz but the
signal bandwidth extends -by Sony's definition- to 100kHz. The SNR over
100kHz is only 30dB. Many amplifiers produce audible distortions when
presented with this noise (hence the switchable filter on many SACDs).
It was "invented" when someone took a CS5390 chip, wired the 1-bit test
outputs straight to a D/A converter and liked what he heard. Thus, the
standard was fixed at 1-bit/64fs which happened to be the internal operating
parameters of this particular chip.
This chip is now long obsolete. Current ADCs operate at rates of 128fs and
over, at 4 bits or more.
Mike Prager
> SACD is indeed fundamentally flawed. Using 1-bit as a conversion method can
> be a valid choice when the analog circuit does not have performance higher
> than the 1-bit signal. To use this as a data format, thus binding everyone
> to the noise and distortion limits, is quite another thing...
Would you care to elaborate on the implications of a
high-ranking Philips engineer taking this stance?
Mike Prager
North Carolina, USA
Karl Uppiano
> high-ranking Philips engineer taking this stance?
You really should read the VanDerKooy and Lipshitz paper. After reading the
paper, I'm surprised Sony/Philips would make such a decision myself. It's a
little bit esoteric, but the problem is obvious once it's pointed out.
It's simply this: A 1-bit converter, properly dithered to eliminate
distortion, is already running at full scale! Any audio signal you add to
that constitutes overload! If you lower the dither to provide headroom for
the audio signal, you are dithering sub-optimally, and so you can't make the
converter completely linear. Sure, you can make it work, but it can never be
optimized. VDK&L show that you can make a completely optimal converter using
only 4-bits.
The problem with high-bit parallel converters (16-bit or greater) is that
it's extremely hard to manufacture them to the tolerance necessary to get
the correct bit-weighting of better than 1/2 LSB. For a 16-bit converter, it
means resistors or capacitors need to be trimmed to 1 part in 131072, or
about 0.0007%. For 24-bit audio, it's even more ridiculous. That precision
is never realized, which means 16-bit parallel converters never achieve
16-bit resolution. In fact, the bits are referred to as "marketing bits" by
the manufacturers! That's the reason the first Philips CD players used
14-bit oversampling converters. They stood a snowball's chance in Hades of
building a parallel converter that was 14-bit accurate (maybe 12 in
practice), and 4x oversampling to get an additional two bits, for maybe
16-bit accuracy if they were lucky. Some chip manufacturers tried
self-calibration schemes. The early (pre CD-era) digital audio systems had
to be hand-tuned before each use.
The 1-bit converters carried oversampling to the extreme. If there's only
1-bit, a precision of 1/2 LSB means 1/2 full scale audio, which is obviously
very easy to accomplish. The additional bits are provided by the
oversampling process. That all works fine until you try to linearize the
system. For optimal effect, triangular probability density (TPD) dither
needs to be injected at levels up to 1 LSB. If there *is* only 1 bit, guess
what? You're at full scale before you ever even add the audio. With 4-bits,
you get 1-bit of dither and 15-bits of headroom for the audio, which when
oversampled, filtered and noise shaped, translates to any arbitrary
resolution you want (16-bits, 24-bits or whatever, depending on the amount
of oversampling used, and the digital filter design.). The precision
required to fabricate a 4-bit parallel converter is 1/2 LSB, or 1 part in
32. It is very easy to fabricate resistors or capacitors with this
precision. Add the proper amount of dither, oversampling and filtering, and
you have a theoretically perfect converter.
"Mike Prager" <hi...@ec.rr.com> wrote in message
news:8GIDb.370241$Dw6.1208561@attbi_s02...
Karl Uppiano
"With 4-bits, you get 1-bit of dither and 15-bits of headroom for the
audio..."
should read:
"With 4-bits, you get 1-bit of dither and 3-bits of headroom for the
audio..."
In article <zaKDb.130076$_M.670179@attbi_s54>,
Stewart Pinkerton
Surely the implication is that these are *facts*, and not subject to
the whims of the marketing department. What, you thought that Philips
*engineers* approved the 'perfect sound forever' slogan 20 years ago?
BTW, Sony have already admitted all his statements, hence the
existence of the 'professional' format of DSD-Wide, which is just a
fancy name for hybrid oversampled 8-bit PCM.
--
Stewart Pinkerton | Music is Art - Audio is Engineering
Bruno Putzeys
> high-ranking Philips engineer taking this stance?
It's no longer a secret. I can do this.
(
Before it was out in the open, I used to be quite vocal on this issue as
well though. The result would be my colleagues going "Shhhhh!", followed by
a burst of laughter from everyone.
)
From quite early on, Philips' position on SACD has been to emphasize on the
multichannel capability (6 full-bandwidth channels available on a
full-length album), not so much on the DSD scheme.
Karl Uppiano
previous post. I'm sure you didn't need my long-winded, simplified
explanation. Sorry about that. -- KU
"Mike Prager" <hi...@ec.rr.com> wrote in message
news:8GIDb.370241$Dw6.1208561@attbi_s02...
Mike Prager
> > Would you care to elaborate on the implications of a
> > high-ranking Philips engineer taking this stance?
>
> It's no longer a secret. I can do this.
Thanks for your reply. It's refreshing to hear the
engineering viewpoint expressed so openly.
Thomas A
The problems has been discussed here:
http://sound.westhost.com/cd-sacd-dvda.htm
and the follow-up:
http://www.fivechannels.com/artiklar/CDvsSACDvsDVD-A-followup.htm
T
Bromo
Is SACD better than Redbook CD?
Harry Lavo
news:bs4h4...@enews3.newsguy.com...
> Question:
>
> Is SACD better than Redbook CD?
In sheer sound quality, yes. Highs are airier and more extended and things
like cymbal shimmer are much more like the real thing. In general, the
soundfield is also more transparent and as a result some microdetails,
particularly in dynamics, can be heard that seem to get lost on CD. I'd say
for classical and jazz, the answer is yes. Add multichannel and modern dsd
recording and the answer is definitely yes for these categories.
The answer is more mixed for much pop music, which depends a lot on
compression and equalization to get the "sound" the producers want. Here
the tools are just becoming available in dsd and a lot of what has been done
has sounds "unfinished" relative to the cd counterparts. However, many
classic rock recordings do seem to have much cleaner sound and transparency
when the producers can go back to the original master tapes (the original
multitrack or mix tape) where the processing has already been done (e.g. you
can compress drums on input, so they are on the multi-track master, or you
can compress them in the mixdown to the stereo master). Moreover, part of
the "good" dsd remastering is in the mastering stage, since many cd's are
highly compressed in mastering for commercial viability. If you truly like
good sound, then having these remastered with less/no compression can yield
good results.
All of the above says some early seventies rock, recorded to tape with
skill, is definitely better and even excellent in surround. JT and Dark
Side of the Moon are excellent examples of this. In other cases, such as
the Rolling Stones reissues and the Dylan reissues the remaster's are
generally more transparent but the overall quality of the sound is less than
optimal because of poor original production.
So for rock, its a case by case basis. Some continue to prefer the original
cd's and I can see their point. I personally however find the remasters by
and large preferable.
Steven Sullivan
> Question:
> Is SACD better than Redbook CD?
As an archiving medium, possibly;; that's what it was intended for,
though some labels, such as Deutsche Grammaphon, have decided it's not
good enough for that. As a two-channel home playback medium
not necessarily, as the existence of excellent-sounding CDs
demonstrates. It's all in the mastering.
Thomas A
The issues around SACD has been discussed here:
http://sound.westhost.com/cd-sacd-dvda.htm
http://www.fivechannels.com/artiklar/CDvsSACDvsDVD-A-followup.htm
T