Topic: Help! Sacd Good Or Bad? (Read 112657 times)
0 Members and 1 Guest are viewing this topic.

## Help! Sacd Good Or Bad?

##### 2002-09-09 12:33:45
I've listened to SACD once and it sure sounded good...

I've heard people say that SACD ought to fail in transient reproduction, (because it can only encode a rise and fall relative to the last time step) so I did some calculations:

'Sampling rate' of SACD = 2.8MHz = 64x 44.1kHz

For encoding tones up to 44.1kHz at full scale, it has to reach full-scale from 0 in 64 steps, that is, at any given time there can only be 64 possible positions for the waveform.

This gives it 6 bits resolution to CD's 16 bits??

OK, suppose you only need to encode up to 22.05kHz at full scale, the number of possible positions increases from 64 to 128--7 bits resolution, big improvement

I doubt this is how DSD actually works, but this article http://www.iar-80.com/page40.html (I linked to page 40, but it seems page 1-39 may be going on and on about the sonic flaws of DSD as well) seems to take this view seriously and goes on to talk about how you try to recover musical information from the 6 bit stream.

## Help! Sacd Good Or Bad?

##### Reply #1 – 2002-09-09 13:33:26
Ok, um, does DSD assume delta modulation or delta-sigma? It seems to me that the delta argument only applies for delta modulation only.

However, I can't see what's the conceptual difference between delta-sigma modulation and PCM if delta-sigma 'quantizes the signal directly' (http://www.cs.tut.fi/~rosti/1-bit/)

What does 'quantize the delta (difference) between the current signal and the sigma (sum) of the previous difference' mean anyway?  (definition of delta-sigma?) It seems to me that on one interpretation of 'sigma', 'sigma' would yield the previous waveform position anyway, just as in delta modulation?

## Help! Sacd Good Or Bad?

##### Reply #2 – 2002-09-09 13:50:08
well, some people are apparently having problems with basic maths.
'Sampling rate' of SACD = 1bit * 2.8MHz = 64bits * 44.1kHz
meaning that you get 2^64 different values, compared to 2^16 different values for old cdaudio.

## Help! Sacd Good Or Bad?

##### Reply #3 – 2002-09-09 14:15:40
Problem is, the bits are not employed in the same way as CD?

My current understanding is that instead of using 64 bits to encode one time step, DSD uses 1 bit to encode each time step, which is 1/64th the length of a CD time step...

Each time step is simply encoded as higher or lower than the previous.

So as I was saying, if you want to traverse the full scale of the waveform in 1/22050s (corresponding to a maxiumum frequency of 22.05kHz), since you only have 128 time-steps in this time, you are limited to staying in one of 128 STEPS for the whole waveform if you don't want to get slope overload as illustrated here:

)

## Help! Sacd Good Or Bad?

##### Reply #4 – 2002-09-09 15:16:35
This is an example of Sigma-Delta, or Pulse Density Modulation:

Is DSD actually just PDM code?

## Help! Sacd Good Or Bad?

##### Reply #5 – 2002-09-09 16:29:13
The general idea is correct, but SACD quality does NOT resemble the quality of 6 or 7 bits PCM (although it would also be incorrect to say it resembles 64bit PCM).
DSD works by determining for each sampling moment wether the sum of the previous errors is above or below a certain level and encodes that in a 1 or a 0 (at least that's more or less what I understand of it). If one would have a signal in which the signal rises from 0 to full-scale in 1/44100th of a second, then DSD would indeed have a BIG problem with that. Luckily for DSD most music doesn't do that, in fact for most music it would be enough (only just, mind you) to encode the difference between this sample and the previous one in 8 bits (assuming 44100Hz and 16bps).
All in all DSD probably sounds better than PCM (at 44100Hz with 16bps), but it does need a lot of noise-shaping and other tricks to work well (to mask all the little errors incurred by the relatively coarse resolution of the steps), which is one of the reasons why I personally don't like the system. But then again I have never really listened to a SACD.

Also keep in mind that the stepsize for DSD could be different than for 16bps PCM, so also because of that it isn't really possible to compare 1bit in DSD with 1bit in 16bps PCM.

Oh, and for those thinking about the very high sampling-rate of DSD and thus assuming it encodes signals upto 1.4MHz, sorry, but it doesn't, in practice the signal is limited to 50 or 100 Khz (because of some necessary filtering, among other things).

BTW, DSD uses delta-sigma modulation and not delta modulation, delta modulation simply looks at the difference between the current value and the previous one, while delta-sigma modulation looks at the sum of the differences and tries to make that as small as possible.

## Help! Sacd Good Or Bad?

##### Reply #6 – 2002-09-09 16:39:31
So the encoding system is really just plain delta modulation? Not delta sigma? Why is that--delta sigma (the article seems to say) has better information density, and can be directly utilized by the final stages of a delta-sigma 1-bit DAC...

On the other hand, if the idea is correct...

You can only get 16 bits equivalent at 43Hz and below!
If you want to encode up to 11.025kHz fullscale sine waves you are still limited to 8 bits resolution!

Surely this is not right??

Quote
Luckily for DSD most music doesn't do that, in fact for most music it would be enough (only just, mind you) to encode the difference between this sample and the previous one in 8 bits (assuming 44100Hz and 16bps).

So does DSD use 1 bit or 8 bits in each time step?

Oh, I get it! You mean in practise it is sufficient to use 8 bits effective resolution and expect there to be no full scale sine waves above 11.025kHz, is that what you mean?

I suppose this could vary from recording to recording, e.g. for hard rock with crazy cymbals you have to go back to 6 or 7 bits

## Help! Sacd Good Or Bad?

##### Reply #7 – 2002-09-09 16:54:26
Sorry to have confused you, I didn't have a good look first (I have edited the post now). DSD does use delta-sigma modulation.

As for the thing about using 8 bits, it was just to illustrate that one usually doesn't need all the 16 bits to encode the signal, even when using simply delta modulation (or rather some kind of differential PCM, as the term delta modulation is apparantly only used for differential PCM using a 1 bit quantizer) and a sampling rate of 44100 Hz.

## Help! Sacd Good Or Bad?

##### Reply #8 – 2002-09-09 17:49:39
IIRC, the difference between delta and sigma-delta is basically this:

delta: you integrate (sum) the output
sigma-delta: you integrate (sum) the input

The (not too obvious) result is that, while delta modulation can suffer from slope overload (which is what you're describing - the 22kHz sine wave is moving far too quickly for the digital staircase to track it), sigma-delta modulation does not. However, the higher frequencies are full of noise.

Think of it this way: in a basic delta modulator, if you draw a graph of frequency against allowed amplitude, then you are allowed high amplitudes at low frequencies, but only lower amplitudes and higher frequencies (otherwise you push it into slope overload).

However, in a sigma-delta system, it's as if the signal is "equialised" on the input to reduce the higher frequencies (to prevent slope overload), and then re-equalised on the output, to bring the higher frequencies back to their correct level. The result is that the higher frequency ranges contain much more noise - the modulator has an intrinsic noise, and it's amplified at higher frequencies on the output.

This is a very badly remembered way of thinking about it, and it's not quite how it works in reality. But I hope it helps!

So, basically, SACD has about 20 bits of equivalent resolution in the audio band, but terrible amounts of high frequency noise. It also has excellent time resolution. It sounds very good - whether all that high frequency noise is a good idea is a different matter entirely!

Cheers,
David.

## Help! Sacd Good Or Bad?

##### Reply #9 – 2002-09-09 18:24:08
yup...

call it what you wan`t...

(i assume your talking hybrid sacd..)

1. this is not real 5.1 audio...
2. not by far (qualety-wise) as good as dvd-a...
3. say you wanna make backups/samplers from sacd..
you can`t ripp...a sacd unit can`t read a cdr/rw/dvdr..

4. conclusion.. just a waste of \$\$....

## Help! Sacd Good Or Bad?

##### Reply #10 – 2002-09-09 23:42:12
I don't think you can fairly say that SACD is "not by far as good as dvd-a".

Both SACD and DVD-A are overspec'ed, albeit less so for SACD. You don't need 24 bits at 192kHz to get a clean representation of an audio signal; or 24 bits at 96kHz for that matter. So there is no reason whatsoever why perfectly mastered SACD and DVD-A versions of the same music should be distinguishable from each other.

Now of course, the problem with ripping is understandable, but hey in that sense DVD-A PCM can't be ripped either.

As for understanding how SACD works, for me what has sufficed is to note that the signal-to-noise ratio of a DSD signal is lower than 0dB---in other words, there is more noise than signal---but that the vast majority of that noise has been shaped into the inaudible frequency range. So within the audible frequency range, you get all the SNR and dynamic range you want.
Michael

## Help! Sacd Good Or Bad?

##### Reply #11 – 2002-09-10 01:25:19
Quote
IIRC, the difference between delta and sigma-delta is basically this:

delta: you integrate (sum) the output
sigma-delta: you integrate (sum) the input

The (not too obvious) result is that, while delta modulation can suffer from slope overload (which is what you're describing - the 22kHz sine wave is moving far too quickly for the digital staircase to track it), sigma-delta modulation does not. However, the higher frequencies are full of noise.

Think of it this way: in a basic delta modulator, if you draw a graph of frequency against allowed amplitude, then you are allowed high amplitudes at low frequencies, but only lower amplitudes and higher frequencies (otherwise you push it into slope overload).

However, in a sigma-delta system, it's as if the signal is "equialised" on the input to reduce the higher frequencies (to prevent slope overload), and then re-equalised on the output, to bring the higher frequencies back to their correct level. The result is that the higher frequency ranges contain much more noise - the modulator has an intrinsic noise, and it's amplified at higher frequencies on the output.

This is a very badly remembered way of thinking about it, and it's not quite how it works in reality. But I hope it helps!

So, basically, SACD has about 20 bits of equivalent resolution in the audio band, but terrible amounts of high frequency noise. It also has excellent time resolution. It sounds very good - whether all that high frequency noise is a good idea is a different matter entirely!

Cheers,
David.

Some of the technical flaws of SA-CD:

* If you want to do some digital post processing, you must convert
it to PCM. If PDM has any advantage, this advantage is removed.

Digital post processing can be
- digital filters for loudspeaker/room acoustic equalization
- digital filters for splitting the signal for 2/3-way loudspeakers
- digital filters for sound control (more complex equalizers)

* PDM is also not suitable to directly drive digital power amplifiers.
It switches too often so you have to much switching losses.
So PDM must converted to PCM and then to PWM.

* PDM is may be suitable to built low cost head phone DA/C+Amplifiers.

* PDM is very sensitive to asymmetries between switch on and switch off.

* Best possible converters (noise + linearity at low levels) do NOT use PDM, but
- PWM
- 4...16 PDM convertes in parallel
Both can not be generated by PDM, but by a PCM.

*The frequency response of the output filter of a SA-CD is not defined

- So it is not possible to compensate the effect of the output filter in the recording
- It is very likely that manufacturs built gadgets with extremely wide frquency response
and huge amounts of HF noise to boast with a extremely wide frequency response.

A proposal of such a filter should look like:

* 4th order LR-filter with f_{-6dB} = 48 kHz
12 kHz: -0.03 dB
24 kHz: -0.53 dB
48 kHz: -6.02 dB
96 kHz: -24.61 dB
192 kHz: -48.20 kHz
384 kHz: -72.25 kHz

Depending on the high frequency amount the frequency overall-frequency response
can be linearized up to 60 kHz (Pop) or 80 kHz (Classic).

-------------------------------------------------------------
Another point:
I have serious doubts about the need of a higher time resolution.
- Modern perceptial encoder has lime resoution between 1.5 and 5 ms.
- Very critical signal needs time resolution of 2...0.5 ms for f=10 kHz.
- CD has something in the range of 0.2...0.3 ms for f=20 kHz and 0.04 ms for f=10 kHz.
- DVD-A (96 kHz) has something in the range around 0.015...0.02 ms for f=20 kHz and
0.01...0.015 ms for f=10 kHz.
--  Frank Klemm

## Help! Sacd Good Or Bad?

##### Reply #12 – 2002-09-10 04:22:09
> - Very critical signal needs time resolution of 2...0.5 ms for f=10 kHz.

Is .5ms the finest needed for stereo depth imaging cues ?  I heard of a hearing capability called "interaural time difference",  which trained humans can resolve to around 30 microseconds.  I'm fairly clueless about this (hearing time resolution), so could somebody set me straight about how that parameter is relevant to stereo/binaural music reproduction?  The references I found:
Binaural_Hearing and Discrimination of the Ear
seem to only mention impulse noises such as clicks.  I can't say that I know of many people who listen to recordings of clicks for pleasure...panning cymbal crashes, maybe?

## Help! Sacd Good Or Bad?

##### Reply #13 – 2002-09-10 10:44:33
Quote
I don't think you can fairly say that SACD is "not by far as good as dvd-a".

Both SACD and DVD-A are overspec'ed, albeit less so for SACD. You don't need 24 bits at 192kHz to get a clean representation of an audio signal; or 24 bits at 96kHz for that matter. So there is no reason whatsoever why perfectly mastered SACD and DVD-A versions of the same music should be distinguishable from each other.

Now of course, the problem with ripping is understandable, but hey in that sense DVD-A PCM can't be ripped either.

As for understanding how SACD works, for me what has sufficed is to note that the signal-to-noise ratio of a DSD signal is lower than 0dB---in other words, there is more noise than signal---but that the vast majority of that noise has been shaped into the inaudible frequency range. So within the audible frequency range, you get all the SNR and dynamic range you want.

yup...

a. do a comparison between a "normal/low-fi" cd-player.. that can read a hybrid
sacd.. and a good dvd. then you hear it.. (depends on your ears though..)
on a hybrid sacd.. the content is ordinary red book.. and the
channel info.. lies in a separate dsd layer.. in dvd-a.. the info lies in
the track itself.. ac3/pcm. not vob.. (in a vob file.. there is additional info..)

b. yes it is duable to ripp. 5.1 pcm.. but not a feature soft-developers
put in a history/change.log
(dvd-a pcm tracks is obsolite)

## Help! Sacd Good Or Bad?

##### Reply #14 – 2002-09-10 11:47:52
Quote
Is .5ms the finest needed for stereo depth imaging cues ?  I heard of a hearing capability called "interaural time difference",  which trained humans can resolve to around 30 microseconds.

But in this case, it must be compared to the interchannel time difference of the CD, that might be well inferior to the global time resolution of the CD !

## Help! Sacd Good Or Bad?

##### Reply #15 – 2002-09-10 11:57:47
Quote
> - Very critical signal needs time resolution of 2...0.5 ms for f=10 kHz.

Is .5ms the finest needed for stereo depth imaging cues ?  I heard of a hearing capability called "interaural time difference",  which trained humans can resolve to around 30 microseconds.  I'm fairly clueless about this (hearing time resolution), so could somebody set me straight about how that parameter is relevant to stereo/binaural music reproduction?  The references I found:
Binaural_Hearing and Discrimination of the Ear
seem to only mention impulse noises such as clicks.  I can't say that I know of many people who listen to recordings of clicks for pleasure...panning cymbal crashes, maybe?

It is the time needed to jump from silent to loud.

Inter channel time resolution of the CD is in the pico second range, for the DVD-A
in the femto second range.

This is much much much below the technical limits of the rest of the transmission chain.
Note that an uneven temperature distribution in your listening room generates interchannel
error which are 10^6 times larger. Also if you are moving some millimeters.
--  Frank Klemm

## Help! Sacd Good Or Bad?

##### Reply #16 – 2002-09-10 12:36:27
It's possible to detect an interaural time delay of ten microseconds at a frequency of 1 kHz. That's about the limit.

CD can store this accurately. It cannot store a waveform with a period of 10 microseconds (that would be a 100kHz sine wave!) - but it can accurately represent a time delay of 10 microseconds between two 1kHz waveforms.

Whether any particular DAC reconstructs the waveforms with enough accuracy to detect this time delay is another matter entirely. The ones I've tried do.

Cheers,
David.

P.S. Frank - I'm well aware of the downsides of SACD. At an AES conference two or three years ago, representatives from Sony would not even enter the same room as my tutor: He was sitting on the "High Resolution Audio committee", discussing many of the issues that you raised. They didn't like it. But format wars are not won or lost on technical matters, or even quality issues.

## Help! Sacd Good Or Bad?

##### Reply #17 – 2002-09-10 14:53:04
Quote
However, in a sigma-delta system, it's as if the signal is "equialised" on the input to reduce the higher frequencies (to prevent slope overload), and then re-equalised on the output, to bring the higher frequencies back to their correct level. The result is that the higher frequency ranges contain much more noise - the modulator has an intrinsic noise, and it's amplified at higher frequencies on the output.

The idea is to shift most of the quantization noise power (noise shaping) to higher frequencies, combined with high oversampling, and filter them out afterwards. So the final output does not contain a lot of high frequencies.

The equalization of the noise that you describe is IIRC more reminiscent of Dolby noise reduction

## Help! Sacd Good Or Bad?

##### Reply #18 – 2002-09-10 15:01:13
Quote
Another point:
I have serious doubts about the need of a higher time resolution.
- Modern perceptial encoder has lime resoution between 1.5 and 5 ms.
- Very critical signal needs time resolution of 2...0.5 ms for f=10 kHz.
- CD has something in the range of 0.2...0.3 ms for f=20 kHz and 0.04 ms for f=10 kHz.
- DVD-A (96 kHz) has something in the range around 0.015...0.02 ms for f=20 kHz and
0.01...0.015 ms for f=10 kHz.

I thought the reason to go hi-res is pretty much universally agreed on: to move the Nyquist frequency further from the upper ceiling of human hearing so that lower order filters can be used instead of brick wall filters, which are problematic any way you design them.

## Help! Sacd Good Or Bad?

##### Reply #19 – 2002-09-10 15:09:34
OT:
I'd finally figured out what the correct grammar should be for my title. It should be 'Ought SACD (to) sound like crud?'

Should the 'to' be in there?

## Help! Sacd Good Or Bad?

##### Reply #20 – 2002-09-10 21:27:55
Quote
I thought the reason to go hi-res is pretty much universally agreed on: to move the Nyquist frequency further from the upper ceiling of human hearing so that lower order filters can be used instead of brick wall filters, which are problematic any way you design them.

I don't think brickwall filters are problematic today. On the other side, SACD is problematic in other ways, as explained here.

## Help! Sacd Good Or Bad?

##### Reply #21 – 2002-09-10 22:33:39
Quote
OT:
I'd finally figured out what the correct grammar should be for my title. It should be 'Ought SACD (to) sound like crud?'

Should the 'to' be in there?

No, the 'to' should not be in there.

Also (at least in the USA) 'ought' would normally be replaced with 'should' in a question, so this would be better:

Should SACD sound like crud?

Finally, the negative version sounds even better because you are questioning the normal assumption:

Shouldn't SACD sound like crud?

If you really want to use 'ought', you can just put a question mark on the declaritive:

SACD ought to sound like crud?

## Help! Sacd Good Or Bad?

##### Reply #22 – 2002-09-11 02:51:03
Quote
Quote
I thought the reason to go hi-res is pretty much universally agreed on: to move the Nyquist frequency further from the upper ceiling of human hearing so that lower order filters can be used instead of brick wall filters, which are problematic any way you design them.

I don't think brickwall filters are problematic today. On the other side, SACD is problematic in other ways, as explained here.

Well.... they still might be problematic...

Ryohei Kusunoki published article on 1996-97 at Japanese MJ magazine about Non-Oversampling and Digital-Filter-Less DAC Concept.
He states that: "the issue is not either it is Non-Oversampling or Higher-rate-sampling, but the use of the digital filter can cause smearing in the time domain".

Arcticle is available here in english: http://www.sakurasystems.com/articles/Kusunoki.html

Here is interview from 1999 where he tells of his further research on subject: http://www.tnt-audio.com/intervis/kusunoki_e.html

In interview he says:
"I found the answer after listening to a DAC using eight DAC ICs to bring about 8-times oversampling without digital filter. The DAC's sound clearly indicated that oversampling was not the culprit of sound degrading, but the real offender was the digital filter."

"Digital filters cut off signals beyond 20kHz with a very steep curve, but needs around 2msec of time to calculate the enormous data. I think this is the reason of "diffusion of sound coherence", the characteristic tonal quality of the oversampling DAC"

47 Laboratory DAC's Model 4705 Progression and Model 4715 Shigaraki implement Kusunoki's ideas by using zero oversampling and they don't have any filter at all, no digital or analog. Here is info about them:
http://www.sakurasystems.com/products/47dac.html

There are some other High-End DAC's which follow this "school of thought" by using 0-oversampling and analog filter instead of digital one.
Audio Note Dacs are most famous of them: http://www.audionote.co.uk/dacs/dac_index.htm

On Audio Asylum thread "Non-oversampling DAC concept", Audio Note's Peter Qvortrup says that they started developing the 1xoversampling DAC concept in 1995, first functional prototype was tested in early 1996 and Audio Note UK built and shipped the first 1xoversampling DAC5 in August 1997, a full three months before Mr. Kusunoki's article was published. Thread is here:
http://www.audioasylum.com/scripts/t.pl?f=...digital&m=18753

## Help! Sacd Good Or Bad?

##### Reply #23 – 2002-09-11 04:19:30
So in other words, the answer to the thread's question is: sacd is not good

Must be a marketing trick to pass a DRM compliant format to the market.

You know, unfortunately, no matter how much "advantages" the "DVD-A" has, it also comes with some unwanted DRM "features".

So for me, the plain original DVD (not the planned DVD Audio) is enough.

PCM 24/96 is supported in the original DVD spec, and DVD burners are becoming popular. Ppl with 24/96 boards are capable of producing their own HQ discs, with no DRM crap.

Of course the DVD spec also supports dolby digital and mpeg audio layer 2.

All current DVD players can play these. No need for new "DVD-A" aware units either.
She is waiting in the air

## Help! Sacd Good Or Bad?

##### Reply #24 – 2002-09-11 05:42:14
Huh? Plain DVD does 24/96? What's DVD-A for then?

SimplePortal 1.0.0 RC1 © 2008-2018