[unfortunateson]

Quote:

Originally Posted by JaZZ Just in case it's not obvious to everybody: DSD as distribution format makes no sense. Logic would call for PCM with 24 bit and 192 or 96 kHz.

I agree. In fact, I remember reading a scientific essay showing mathematically that at "one-half the data rate [PCM] outperforms DSD on every count! DSD is a profligate wastrel of capacity."

 

[theaudiohobby]

Quote:

Originally Posted by unfortunateson I agree. In fact, I remember reading a scientific essay showing mathematically that at "one-half the data rate [PCM] outperforms DSD on every count! DSD is a profligate wastrel of capacity."

That is rather funny, the last time I looked DXD at 24/352 did not outperform DSD except with respect to out-of-band noise, however DSD handily outperforms it in other areas.

 

[JaZZ]

Quote:

Originally Posted by theaudiohobby Big laugh...

So, of the scenarios listed below, which would you consider the most logical in the sense of least signal degradation?

A) DSD recording --> conversion to PCM for sound editing --> conversion to DSD for SACD production --> conversion to PCM --> D/A conversion

B) PCM recording --> sound editing --> conversion to DSD for SACD production --> conversion to PCM --> D/A conversion

C) PCM recording --> sound editing --> D/A conversion

 

[Glassman]

I'd render the choices slightly different:

1) editing in analog -> 1bit/64Fs SDM A/D -> SACD -> 1bit/64Fs SDM D/A

2) editing in analog -> 1bit/64Fs SDM A/D -> SACD -> conversion to PCM for filtering -> 5bit/128Fs SDM D/A

3) 1bit/64Fs SDM A/D -> conversion to PCM for editing -> 1bit/64Fs SDM -> SACD -> conversion to PCM for filtering -> 5bit/128Fs SDM D/A

4) 5bit/128Fs SDM A/D -> conversion to PCM for editing -> 1bit/64Fs SDM -> SACD -> conversion to PCM for filtering -> 5bit/128Fs SDM D/A

5) 5bit/128Fs SDM A/D -> conversion to PCM for editing -> DVD-A -> 5bit/128Fs SDM D/A


those represent:

1) ideal situation from Sony/Philips marketing brochures never utilised in practice

2) 'pure DSD' recordings

3) recordings made with DSD mashines, edited in digital

4) DXD recordings

5) DVD-A recording

 

[Ferbose]

Glassman, I am amazed that you know all these facts about A/D and D/A.
Your posts are highly informative.
Ideally, we want pure DSD to be 1-bit all the way from recording to playback.
We want pure PCM to be 16 or 24 bit all the way, too.
In reality, neither is happening.
Argueing which format is inherently better may be fultile.
But SACD seems to be getting more nods from audiophile labels and audiophiles.
Perhaps it is just easier or cheaper to get good sound using the SACD architecture.

 

[theaudiohobby]

Quote:

Originally Posted by Glassman I'd render the choices slightly different: 1) editing in analog -> 1bit/64Fs SDM A/D -> SACD -> 1bit/64Fs SDM D/A 2) editing in analog -> 1bit/64Fs SDM A/D -> SACD -> conversion to PCM for filtering -> 5bit/128Fs SDM D/A 3) 1bit/64Fs SDM A/D -> conversion to PCM for editing -> 1bit/64Fs SDM -> SACD -> conversion to PCM for filtering -> 5bit/128Fs SDM D/A 4) 5bit/128Fs SDM A/D -> conversion to PCM for editing -> 1bit/64Fs SDM -> SACD -> conversion to PCM for filtering -> 5bit/128Fs SDM D/A 5) 5bit/128Fs SDM A/D -> conversion to PCM for editing -> DVD-A -> 5bit/128Fs SDM D/A those represent:

My comments in italics

1) ideal situation from Sony/Philips marketing brochures never utilised in practice

Not correct

2) 'pure DSD' recordings

Not correct, upsampling from 1bit 64fs to 5bit 128fs sigma delta does not require an intermediate PCM stage

3) recordings made with DSD mashines, edited in digital

Nope, I know that the Meitner and Grimm Audio D/A converters do not do this..also if DSDwide is used for editing then this will not be the case, accepted that is far easier to edit PCM, but even then,in a number of implementations, it is only the edited portions that are converted to PCM, maybe part of the reason why some heavily edited genres do not work well in SACD

4) DXD recordings

5) DVD-A recording

 

[Glassman]

1) the only converter I know of is DSD1700 and I didn't see it used lately, also if you check Burr Brown's web, there is no stock of them.. maybe some early high end players used them, but today? no

2) how come? what would be the point? of course it's low pass filtered before entering another sigma delta modulator, you absolutely don't want to encode all that out-of-band noise from single bit modulated stream.. and filtering in other words means working in PCM, so..

3) I'd like to see how you would attenuate by say 6dB in DSD without going PCM, please show me.. since SDM encoded data has so little bit depth, you cannot apply effects on the original bit depth and since you have to increase the bit depth, you then have to bring it back to the original and you do this by running it through the sigma delta modulator again and before you do so, you have to get rid of the unwanted out-of-band noise, otherwise the second encode would be pretty rubbish, that means you have to low pass filter and that's pure PCM operation..

it's also important to note that PCM has nothing to do with samplerate, those PCM operations I'm talking about can happen at 32bit/64Fs.. it also may be of interest that sigma delta modulators are heavily used also on redbook CDs, in this case it's called noise shaping and it usually operates at the couple of least significant bits of the 16bit PCM.. these techniques really overlap in practice, the only case when they don't is clean path from SDM A/D to SDM D/A, which is not common practice in vast majority of cases..

 

[theaudiohobby]

1) the only converter I know of is DSD1700 and I didn't see it used lately, also if you check Burr Brown's web, there is no stock of them.. maybe some early high end players used them, but today? no

And if the manufacturer is not using an off the shelf chip, or using discrete circuit ala Meitner or Grimm Audio, it is possible and still is a commercially valid approach.

2) how come? what would be the point? of course it's low pass filtered before entering another sigma delta modulator, you absolutely don't want to encode all that out-of-band noise from single bit modulated stream.. and filtering in other words means working in PCM, so..

Out-of-band noise is mute for a CD release and the oversampling process is how 1bit/64fs is converted to 5bit/128fs.

3) I'd like to see how you would attenuate by say 6dB in DSD without going PCM, please show me.. since SDM encoded data has so little bit depth, you cannot apply effects on the original bit depth and since you have to increase the bit depth, you then have to bring it back to the original and you do this by running it through the sigma delta modulator again and before you do so, you have to get rid of the unwanted out-of-band noise, otherwise the second encode would be pretty rubbish, that means you have to low pass filter and that's pure PCM operation.

Editing in DSDWide without going to PCM is already a commercial reality and I have exchanged emails in the past with some of most vocal critics of DSD and he conceded that it is doable, the specifics are outside the scope of this discussion, but it does not revert to PCM, neither is low pass filtering a PCM operation, downsampling and oversamplling are not PCM operations.

it's also important to note that PCM has nothing to do with samplerate, those PCM operations I'm talking about can happen at 32bit/64Fs.. it also may be of interest that sigma delta modulators are heavily used also on redbook CDs, in this case it's called noise shaping and it usually operates at the couple of least significant bits of the 16bit PCM.. these techniques really overlap in practice, the only case when they don't is clean path from SDM A/D to SDM D/A, which is not common practice in vast majority of cases.

Yes, but it has everything to do with word length, In PCM, wordlength size determines performance, for example 24bit/96kHz will have superior performance to a 16bit/196KHz converter owing to the larger wordlength albeit over a more limited bandwidth as per Nyquist fs/2. However in SDM, performance is proportional to the sample rate increase, 1bit 64fs is always inferior to 1bit 128fs, the wordlength cannot get any smaller but the sampling points can be closer hence quantisation noise is smaller, that is a fundamental difference. Actual converters may overlap in practice, i.e. PCM to SDM conversions and vice versa, but they are fundamentally different encoding processes.

 

[Glassman]

I'm happy to see you're continuing with this debate, I appretiate that!

1) in proprietary, everything is possible, I was talking more about the actuall implementation of the technology in wide practice..

2) I'm affraid I don't understand what you're trying to say here, could you please rephrase and be more specific?

3) DSDwide is just a 4bit SDM, nothing more as far as I know and as such it works the same as ordinary DSD.. I still don't see the possibility to edit SDM without increasing the bit depth and at least light filtering before encoding back to SDM.. it's the same as with editing PCM and applying noise shaped dither, once you do anything to it, you have to perform the noise shaped dithering again, in fact it is usually performed only at the end of all operations.. both oversampling and decimating require filtering to work and it absolutely have to work in PCM, or in other words with high resolution amplitude information.. prove me wrong - show me the principle of operation of a system working with noise shaping..

addendum) we're moving on point here.. if you take bare PCM, that means without any applied noise shaping, then it's performance is determined solely by the bit depth and is the same over the whole usable bandwidth, which is fs/2, this is in direct contrast to noise shaping system, whose performance deteriorates with increasing frequency and is given by sampling rate together with bit depth.. holding to your examples, 16/192 can be better than 24/96, it depends on actuall implementation and what you are after.. clearly 16/192 will have better properities for content between 48 and 96kHz

if that's what you're interested in.. when you apply sufficient noise shaped dither on the 16/192, you can probably achieve >20bit performance at the beginning and still >12bit at the end of the usable bandwidth.. which is better? this is the balancing between performance distribution among various frequencies, bit depth and sampling rate.. DSD has incredible bandwidth, however it's performance is usable only at the very beginning of the available spectrum.. higher samplerate is better as well as higher bit depth.. now let me ask you, what do you think is better, 1bit/128Fs or 2bit/64Fs? we can of course continue with 4bit/32Fs through 8/16, 16/8, 32/4, 64/2 to 128bit/44.1kHz

it is clear that anything past 24bit is useless as we're absolutely not able to record or playback with such precision, it's only useful when doing math with the signal, but not on the distribution medium.. but that 16bit/352.8kHz seems pretty viable option, with a light use of noise shaping on the few least significant bits, we can achieve greater than 24bit performance in the beginning of the spectrum with at least 20bit at 20kHz and decent performance after that with residual out of band noise at levels bellow 8bit.. the thing is, with commonly used PCM hi-res formats now at 24bit, there is no reason for using noise shaping on them, there's no reason to aim for better than 24bit performance.. DSD as it is now has really good precision at lower frequencies, but only so-so performance on higher frequencies, which is the nature of SDM, but the problem really is the amount of out-of-band noise one has to deal with..

the wordlength cannot get any smaller but the sampling points can be closer hence quantisation noise is smaller
the quantisation noise is determined solely by the bit depth and it is the same regardless of samplerate used, higher samplerate only provides more space to shape this quantisation noise and allow greater performance in the beginning of the usable bandwidth..

the funny thing is, noise shaped dither can make 16bit perform as >20bit, but it needs a space to operate (ultrasonic frequency range), which unfortunately on CDs doesn't have

and on the other hand, with hi-res formats, where the space is (at least between 20-48kHz), there is no need for noise shaped dither anymore due to the 24bit bit depth

before making a decision, we have to determine what we demand.. it is probably >20bit precision over the whole audible frequencies, we also want certain headroom to allow for much gentler band limiting, resulting data should fit onto DVD in six plus two channels, anything else? and of course we want as little side effects as possible.. pure 24bit/96kHz PCM matches all of the above and it is the simplest approach too, easy to work with, no problems bound with it..

although in my eyes, having 24bit precision at ultrasonic frequencies is a waste of space, to me it would be more clever to improve on impulse response, which means increasing the bandwidth and of course decreasing the bit depth, that way we can still have the same or nearly the same precision in audible band, but a lot more extended ultrasonic region and hence better impulse response, still with sufficient precision.. 16bit/176.4kHz is what I'm speaking about, this is the same data amount as 1bit/64Fs DSD and a better format in my eyes..

 

[theaudiohobby]

Quote:

Originally Posted by Glassman I'm happy to see you're continuing with this debate, I appretiate that! 1) in proprietary, everything is possible, I was talking more about the actual implementation of the technology in wide practice..

Here I believe you mean LSI implementations, because both products I mentioned are commercial products available on the open market.


Quote:

Originally Posted by Glassman 2) I'm affraid I don't understand what you're trying to say here, could you please rephrase and be more specific?

All I am saying that filtering is not a PCM process, neither decimation or oversampling for that matter, 5bit/128fs is decimated to 1bit/64fs, and vice 1bit/64fs oversampled to 5bit/128fs with no intermediate PCM stages.

The points you raised in your third point and addedum are best addressed over the weekend, it is a long discussion

when I have time, however in answer to an important rhetorical question you raised, The performance 1 bit/128fs is superior to 2bit/64fs except wrt to the ultrasonic noise spectrum. There are other practical issues involved that may make 2bit/64fs preferable such as downstream processing as a result of the more benign ultrasonic noise spectrum, but it terms of absolute performance 1bit/128fs is superior to 2bit/64fs.

With respect to my previous post, there is one issue that need revisiting, 16bit/192KHz with same reasonable implementation will have superior performance 24bit/96kHz, except with respect to the absolute avaliable dynamic range even within limits of the Nyquist frequency of the later i.e. below 48kHz, reasons there are simply more samples and the higher sampling frequency gives better options wrt to the chosen filtering techniques. If you implied this in your response, then ignore this paragraph.

 

[Glassman]

can't wait till the weekend

I'm particulary intrested in example on how to decimate 5bit/128Fs to 1bit/64Fs without band limiting to 32Fs, in other words low pass filtering.. or how such a low pass filtering works in limited amplitude resolution..

power to ya