[Rob Watts]

Yes - that's one of the reasons (there are other ones) why I digitally filter to 2048 FS, to remove all of these images.

 

[rayl]

So while traveling for the holidays, I was re-pondering some of the points from Rob Watts’ RMAF talks... the comment about the errors in DSD led to a question that is confusing me.

I agree with the stated flaw, BUT:

Isn’t it the case that ADCs today are delta sigma based so in fact, the DSD existed before being converted into PCM? So any flaws in the DSD will also exist in the PCM?

I.e. if the DSD were from DSD masters, and not just converted from PCM, then any shortcomings are irreparable...

But this is ultimately moot in practice as content I am interested in is in PCM and not avail in DSD... though interested to hear thoughts on an intellectual/science basis.

 

[514077]

So while traveling for the holidays, I was re-pondering some of the points from Rob Watts’ RMAF talks... the comment about the errors in DSD led to a question that is confusing me.

I agree with the stated flaw, BUT:

Isn’t it the case that ADCs today are delta sigma based so in fact, the DSD existed before being converted into PCM? So any flaws in the DSD will also exist in the PCM?

I.e. if the DSD were from DSD masters, and not just converted from PCM, then any shortcomings are irreparable...

But this is ultimately moot in practice as content I am interested in is in PCM and not avail in DSD... though interested to hear thoughts on an intellectual/science basis.

I'm also confused, but I still like my DSD music through Hugo2. In reading so many articles and opinions, it seems more political than anything. Mark Waldrep (Dr. Aix) uses the debate to flog his surround sound content at 24/96. Hans Beekhuisen explains why he likes 24/192; (his thoughts make sense to me). Paul McGowan is pro-DSD as that's the domain in where his DACs operate. Frankly, I'm not technically or mathematicly smart enough to know who's right. I'm most likely to trust Rob, since his DACs prove his ideas to any competent ear. But, as I say, I like what his Hugo2 does for my DSD music. Merry Xmas to all Chord users and builders.
Kevin

 

[Rob Watts]

So while traveling for the holidays, I was re-pondering some of the points from Rob Watts’ RMAF talks... the comment about the errors in DSD led to a question that is confusing me.

I agree with the stated flaw, BUT:

Isn’t it the case that ADCs today are delta sigma based so in fact, the DSD existed before being converted into PCM? So any flaws in the DSD will also exist in the PCM?

I.e. if the DSD were from DSD masters, and not just converted from PCM, then any shortcomings are irreparable...

But this is ultimately moot in practice as content I am interested in is in PCM and not avail in DSD... though interested to hear thoughts on an intellectual/science basis.

DSD is a special class of delta sigma; it is of course delta sigma but set to 1 bit output. There are particular problems with 1 bit:

1. They are obviously lower resolution than say a typical 5 bit delta sigma (sixteen times less open loop resolution)

2. They have a particular problem unique to 1 bit - stability. A delta sigma modulator works by having integrators amplifying the error, then quantizing the amplified error and the wanted signal. Consider a 5 level delta sigma with OP's of +2, +1, 0, -1, -2 say. If the input is +1 say, then the noise shaper will output a range of values - +2,+1 and 0. Everything is happy, all the integrators work normally. But with DSD, the outputs are constrained to +1 or -1. In this case, if the input is +1, then a delta sigma modulator would request values of +2, +1, 0 as before; but +2 is not allowed. So when the modulator asks for +2 and does not get it, then the internal error builds up within the modulator, and the OP goes unstable, or integrators saturate. In short, the behaviour of the modulator is upset. This creates massive and unique problems you only get with DSD, and not other delta sigma systems; in short the OP behaviour is signal dependent, and this causes noise floor modulation. There are AES papers showing the DSD noise floor modulation issue, and it is a problem that can't be eliminated. This is the reason why -6dB is the max for DSD; but the innate noise floor modulation problem occurs well below -6dB.

The lack of innate resolution can be fixed if you run the modulator at high enough rates (2048FS) and if you use high resolution outputs, such as 5 bits. With this approach, you can get extremely high in band small signal resolution - Dave's noise shaper will perfectly resolve a -301 dB signal - and this is essential for the perception of sound stage depth.

Another issue is timing of transients. Small signal transients have a delay that is due to the OP not being initially quantised, and relies on the error being integrated, which of course takes time. This means that small transients have a delay, and large transients do not - and this effect is easy to see on simulation with DSD. This is why DSD sounds unnaturally soft, as transients timing are not accurate enough. My pulse array (5 bits, 2048FS) does not show this behavior on simulation - a -60dB step change has no consistent delay compared to 0dB step change. This is due to the very high speed of operation, the 5 bit resolution, and the fact that one can properly dither a 5 bit delta sigma, but you can't dither a DSD system.

 

[jlbrach]

quick question...what does Rob think of usb conditioners for those of us who use our Dave with a USB connection?

 

[Rob Watts]

From a theory POV they can make it sound slightly better (warmer and smoother) with lower RF; but would not be better than a battery powered lap-top; but could make things sound brighter by actually increasing RF noise. Then people will think that the extra brightness is more transparency. My view is that they are much more likely to make it worse.

I have not heard any improvements with USB cables for example; all quality RF USB cables sound the same. I have heard a USB cable that sounded different - it had been "tuned" and actually was an RF noise generator, as it had poor RF qualities, and it made the sound quality brighter, and so worse.

But I plan to conduct more listening tests later on in 2018, just to check the possibility that something else is at play here...

 

[Triode User]

But I plan to conduct more listening tests later on in 2018, just to check the possibility that something else is at play here...

Outstanding above and beyond the call of duty.

Oh, and Merry Christmas by the way!

 

[rayl]

DSD is a special class of delta sigma; it is of course delta sigma but set to 1 bit output. There are particular problems with 1 bit:

1. They are obviously lower resolution than say a typical 5 bit delta sigma (sixteen times less open loop resolution)

2. They have a particular problem unique to 1 bit - stability. A delta sigma modulator works by having integrators amplifying the error, then quantizing the amplified error and the wanted signal. Consider a 5 level delta sigma with OP's of +2, +1, 0, -1, -2 say. If the input is +1 say, then the noise shaper will output a range of values - +2,+1 and 0. Everything is happy, all the integrators work normally. But with DSD, the outputs are constrained to +1 or -1. In this case, if the input is +1, then a delta sigma modulator would request values of +2, +1, 0 as before; but +2 is not allowed. So when the modulator asks for +2 and does not get it, then the internal error builds up within the modulator, and the OP goes unstable, or integrators saturate. In short, the behaviour of the modulator is upset. This creates massive and unique problems you only get with DSD, and not other delta sigma systems; in short the OP behaviour is signal dependent, and this causes noise floor modulation. There are AES papers showing the DSD noise floor modulation issue, and it is a problem that can't be eliminated. This is the reason why -6dB is the max for DSD; but the innate noise floor modulation problem occurs well below -6dB.

The lack of innate resolution can be fixed if you run the modulator at high enough rates (2048FS) and if you use high resolution outputs, such as 5 bits. With this approach, you can get extremely high in band small signal resolution - Dave's noise shaper will perfectly resolve a -301 dB signal - and this is essential for the perception of sound stage depth.

Another issue is timing of transients. Small signal transients have a delay that is due to the OP not being initially quantised, and relies on the error being integrated, which of course takes time. This means that small transients have a delay, and large transients do not - and this effect is easy to see on simulation with DSD. This is why DSD sounds unnaturally soft, as transients timing are not accurate enough. My pulse array (5 bits, 2048FS) does not show this behavior on simulation - a -60dB step change has no consistent delay compared to 0dB step change. This is due to the very high speed of operation, the 5 bit resolution, and the fact that one can properly dither a 5 bit delta sigma, but you can't dither a DSD system.

Very educational... !

Wwhat word size do most ADCs use? I thought it is 1 bit output from comparator.... I.e. a 1 bit stream at some Fs multiple is decimated/filtered into PCM.

Consequently if this stream were captured as our DSD file....then clearly, we can’t do better with capturing the later PCM (as we can always do the same, or better decimation/filtering if we prefer to). That’s where I am intellectually stuck.

If this stream is greater than 1 bit, then why doesn’t the industry define an n bit “DSD” to accommodate... thus as long as sigma delta ADC (as opposed to SAR) remains state of art for ADC, there is a path for a PDM format that is strictly more “pure” than PCM.

It is also possible that this just isn’t how the industry produces DSD files for distribution in practice and that is the flaw- that the DSD on an SACD is converted from PCM instead of being tapped in the ADC process pre decimation. I am in the dark... but curiosity definitely piqued since the talks as finally, I am learning the real picture, past the marketing noise and misinformation. They really were great talks.

Merry Christmas.

 

[rayl]

I am going to answer my own question as thinking out loud reminded me of a key factor. Post production is always going to be in PCM, so we already lost the original “DSD”...

Hence focusing on PCM is the unavoidable path, both practically, based on content availability, and technically, as commercial works are processed post production in PCM always.

 

[jlbrach]

https://audiobacon.net/2017/07/11/uptone-audio-iso-regen-review/

I listen to my dave with my laptop plugged in and i came across this reviewers view of the USB conditioner while using the Dave which for the record he loves....so far my impressions are positive

 

[Jawed]

I listen to my dave with my laptop plugged in and i came across this reviewers view of the USB conditioner while using the Dave which for the record he loves....so far my impressions are positive

Does your laptop have an optical output? If so, use the optical cable that came with DAVE to connect your laptop to DAVE. That'll give you a sound quality which USB conditioners aspire to, but apparently fail to achieve.

Now playing: Kaitlyn Aurelia Smith - In the World, but Not of the World

 

[jlbrach]

actually I am using dave/blu 2....i have a new macbook pro...the usb conditioner seems the best option

 

[alphaman]

Isn’t it the case that ADCs today are delta sigma based so in fact, the DSD existed before being converted into PCM? So any flaws in the DSD will also exist in the PCM?

I.e. if the DSD were from DSD masters, and not just converted from PCM, then any shortcomings are irreparable...

But this is ultimately moot in practice as content I am interested in is in PCM and not avail in DSD... though interested to hear thoughts on an intellectual/science basis.

If a delta-sigma ADC was used in the recording process -- and that recording was not intended for SACD/DSD** release (99% are not) -- that A/D would decimate the output of the delta-sigma ADC to std. PCM (16/44.1) as shown in the upper half of this figure:

(note: this textbook figure shows an older 2x oversampling A/D and D/A system . Nevertheless, the decimation process remains regardless of whether the A/D was delta-sigma or R2R)

Here is another textbook fig. which may clarify:

(An oversampling A/D converter using one-bit coding at a high sampling frequency, and a decimation filter. Also see ***)

**My definition of "DSD" is the strict definition of DSD:
https://en.wikipedia.org/wiki/Direct_Stream_Digital
Specifically: "A DSD recorder uses sigma-delta modulation. DSD is 1-bit with a 2.8224 MHz sampling rate. The output from a DSD recorder is a bitstream. "
Because of the sheer amount of data generated by a DSD recorder (or DSD ADC), only in the late 1990s was it practical (economical) to store all those bits (among other things).

BTW: For pure DSD (such as Blue Coast Records, or analog tape masters transfer to DSD recorder), I have never heard std. 16/44.1 PCM come even close to the rez. and fidelity of the true-DSD recording. Higher-rate PCM (24/192k) becomes competitive with pure DSD.

BTW 2: If you have a DSD source (SACD, etc.), you don't need a DAC. I repeat, you don't need DAC. The nature of the PDM bitstream alows it to be simply low-pass-filtered (in the analog domain).
For more info, I suggest Ted Smith's YouTube discussions here:






***Consider an example in which one-bit coding takes place at an oversampling rate R of 72; that is, 72 × 44.1 kHz = 3.1752 MHz, as shown in figure. The decimation filter provides a stop-band from 20 kHz to the half-sampling frequency of 1.5876 MHz. One-bit A/D conversion greatly simplifies the digital filter design. An output sample is not required for every input bit; because the decimation factor is 72, an output sample is required for every 72 bits input to the decimation filter. A transversal filter can be used, with filter coefficients suited for the decimation factor. Following decimation, the result can be rounded to 16 bits, and output at a 44.1-kHz sampling frequency.

 

[514077]

I am going to answer my own question as thinking out loud reminded me of a key factor. Post production is always going to be in PCM, so we already lost the original “DSD”...

Hence focusing on PCM is the unavoidable path, both practically, based on content availability, and technically, as commercial works are processed post production in PCM always.

Try nativeDSD. Not all albums are crossed over into PCM. I'm not saying DSD is the next best thing to sex. But, even on the web, there are contrasting points of view, mainly for the political reason: Buy my downloads: Don't use 24/192 which makes you stupid and worth ridicule, 96 is good enough: DSD is 32 times better than PCM, so you're wrong not to use DSD." I'buy DSD at the rate it was mastered; not lesser or greater. DSD when it's mastered in DSD. I do wish the guy who did the Genisis remasters, would have just released them in 192, as he mixed them. But, sometimes people need to be irritating, as I-no doubt-am. It's all a political bandwagon upon which each side likes to jump. I appreciate everything Rob takes the time to explain to us. But, with the freedom the internet gives us, I owe it to myself to always look for confirming or contrasting sources of info.

 

[514077]

If a delta-sigma ADC was used in the recording process -- and that recording was not intended for SACD/DSD** release (99% are not) -- that A/D would decimate the output of the delta-sigma ADC to std. PCM (16/44.1) as shown in the upper half of this figure:

(note: this textbook figure shows an older 2x oversampling A/D and D/A system . Nevertheless, the decimation process remains regardless of whether the A/D was delta-sigma or R2R)

Here is another textbook fig. which may clarify:

(An oversampling A/D converter using one-bit coding at a high sampling frequency, and a decimation filter. Also see ***)

**My definition of "DSD" is the strict definition of DSD:
https://en.wikipedia.org/wiki/Direct_Stream_Digital
Specifically: "A DSD recorder uses sigma-delta modulation. DSD is 1-bit with a 2.8224 MHz sampling rate. The output from a DSD recorder is a bitstream. "
Because of the sheer amount of data generated by a DSD recorder (or DSD ADC), only in the late 1990s was it practical (economical) to store all those bits (among other things).

BTW: For pure DSD (such as Blue Coast Records, or analog tape masters transfer to DSD recorder), I have never heard std. 16/44.1 PCM come even close to the rez. and fidelity of the true-DSD recording. Higher-rate PCM (24/192k) becomes competitive with pure DSD.

BTW 2: If you have a DSD source (SACD, etc.), you don't need a DAC. I repeat, you don't need DAC. The nature of the PDM bitstream alows it to be simply low-pass-filtered (in the analog domain).
For more info, I suggest Ted Smith's YouTube discussions here:






***Consider an example in which one-bit coding takes place at an oversampling rate R of 72; that is, 72 × 44.1 kHz = 3.1752 MHz, as shown in figure. The decimation filter provides a stop-band from 20 kHz to the half-sampling frequency of 1.5876 MHz. One-bit A/D conversion greatly simplifies the digital filter design. An output sample is not required for every input bit; because the decimation factor is 72, an output sample is required for every 72 bits input to the decimation filter. A transversal filter can be used, with filter coefficients suited for the decimation factor. Following decimation, the result can be rounded to 16 bits, and output at a 44.1-kHz sampling frequency.

Thanks. Tired from Xmas. Will watch tomorrow. From what little I know about everything, I had a thought that Rob and Ted might be 'oposite sides of the same high-end coin.