Thoughts on a bunch of DACs (and why delta-sigma kinda sucks, just to get you to think about stuff)

[amalgamist]

Alright so who bought the non-working Theta Gen V for $1030 ? 

 

 

[jbarrentine]

 I can determine the type/ ages of violins used by each performer. 

 
I almost spit my drink out. 

 

[prot]

July 4th we are having a meet here in Ottawa. At which will be a Yggy and my NAD M51. We are going to do some well designed, multiple subject, multiple trial blind listening tests. I'm getting a Yggy by years end, and while I suspect it will be a significant step up in sound quality I can't help but think some of this talk about DS being garbage is a little exaggerated. Seriously, to listen to some of the posts you would think that using a DS DAC is essentially the equivalent of listening to the cheapest Emerson receiver with nasty Radio Shack homemade speakers from the 70s. seriously, it sounds like people are suggesting this post-apocalyptic wasteland of sound degradation. While I have a DS DAC, frankly I don't build my identity around it, and I'm quite willing to accept that there are inherent flaws worth overcoming, but some posts here really make DS seem like a technology invented by degenerate freaks who failed community college electronics out to push garbage on people for crazy money. Enough already!

And yes, I did read the title of the thread.

+1 for your D-S comments
+10 for the test idea. I'd recommend you open a thread in the science section and ask for advice.. many exp people there and there are many specifics for audio tests. If you have time I'd go for two tests: one "which dac you prefer" and one "can you hear a diff". And I think evillamer is right, a third Dac would help.
Good luck with the test.

 

[hans030390]

Transparency
Both the Audiomeca Ambrosia and SF SFD MK II are more transparent . Further and deeper sound stages with better spatial location of instruments. They combine a whole coherent scale of the macro dynamic swings and flow of music while maintaining the tiny individual pieces that make up the whole tapestry. The micro dynamic inflections are easily heard that let one determine the timbre of each instrument and the actual playing. I can determine the type/ ages of violins used by each performer. The audial never allowed that.

 
You said Model S MK I, right? I'm sure the the gains between I and II are very small, but worth considering. Either way, not sure what the Audiomeca is running in terms of DAC chips/modules, but based on what I've read about the SFD MKII and what I know about or have personally heard with UltraAnalog chips, some of these differences you mentioned don't surprised me. The UA modules are very nice. Even then, basically all of the differences you mentioned seem pretty on par when comparing a non-oversampling design (Audial) with an oversampling design (basically everything else). Oversampling does tend to help with staging, spacial location, and separation on a large and small scale. More transparent in that regard. I'm in the minority in that I think oversampling mucks up some other elements (making some aspects besides those sound less real in comparison to my ears), but, hey, most disagree with me, and that's why I recommend oversampling to most others.

 

[wahsmoh]

  Alright so who bought the non-working Theta Gen V for $1030 ? 

 

Ouch. I saw that auction and didn't follow it much longer but that really surprises me

 

[Sonic Defender]

+1 for your D-S comments
+10 for the test idea. I'd recommend you open a thread in the science section and ask for advice.. many exp people there and there are many specifics for audio tests. If you have time I'd go for two tests: one "which dac you prefer" and one "can you hear a diff". And I think @evillamer is right, a third Dac would help.
Good luck with the test.

Thanks, appreciate the thoughts and comments. If time permits a third DAC is possible, but I worry that too much time would be involved. Anyway great suggestions on all fronts. Thanks mate.

 

[hans030390]

Both the Audiomeca Ambrosia and SF SFD MK II are more transparent . Further and deeper sound stages with better spatial location of instruments. They combine a whole coherent scale of the macro dynamic swings and flow of music while maintaining the tiny individual pieces that make up the whole tapestry. The micro dynamic inflections are easily heard that let one determine the timbre of each instrument and the actual playing. I can determine the type/ ages of violins used by each performer. The audial never allowed that.

 
Oh, in relation to my last message, also curious what source you fed these DACs with....I'm assuming some sort of USB->SPDIF converter. (Not trying to imply something wasn't up to par, just curious for my own reasons.)

 

[KeithEmo]

  I think the problem with all the theorist and those that strictly follow the Nyquist theory(44.1Khz is enough to capture/reproduce everything in audio) or there abouts(think of those who subscribe to hydrogen audio and etc) forget that there's a little thing in life/engineering known as "headroom".
 
In theory, you can design a 1.8m height door for a 1.7999m high human being to use, but do you think the human being going through the door can do it fast and smooth? This is not taking into account thermal expansion of the door/wall/floor.
 
Likewise 44.1KHz makes it diffcult on the digital/analog filters, and no filters(*not sure if it includes yggdrasil) in the world can perfectly filter without causing all sorts of unwanted distortions like gibbs effect, post-ringing, pre-ringing, phase distortion and etc.
 
There's a 1990s AES article which they tried to push for 48KHz as the replacement(DVD Audio) standard for CD audio, because it relaxes the requirements of the D/A filtering component.
 
As for the bits part... think of it as making bread. By right if you using 16kilograms of yeast, you should make 16kilograms of bread. However during the process of making bread, you lose some of the yeast somewhere/somehow and the end result is the bread is 14kilograms(or thereabouts) thus not bit perfect. You would want more yeast(bits) to begin with so as to recreate the bread at 16kilogram.
 
There's this famous baker know as Sabre, he takes 24kilograms of yeast, but only produces about 15kilograms worth of bread.

 
I wanted to comment on your comment - and on two other things as well.
 
1) Nyquist - What many people forget is that Nyquist is more of a "limit" than a "promise". Essentially, it is theoretically possible to provide all the information necessary to define a continuous sine wave as long as you have more than two samples per period (in other words - in order to "store" a sine wave, you need a sampling frequency at least slightly over twice its frequency). However, what this really means is that, if your sample rate is too low, then it won't be able to do the job. In other words, a CD, sampled at 44.1k CANNOT store information above 22 kHz. This does not constitute a promise that a CD WILL accurately store information up to 22 kHz - just that it's POSSIBLE - considering the sample rate. Nyquist also assumes a continuous sine wave (the math says you can totally define A CONTINUOUS SINE WAVE if you know a little bit more than two data points per cycle - it assumes an endles stream of identical cycles - which is what you have in a continuous sine wave). It doesn't necessarily hold if you have a non-sine-wave, or a combination of multiple sine waves, or a transient wave form.
 
2) While it's true that oversampling has flaws,so do all the other alternatives. Some people seem to have the idea that the only reason oversampling is used is that many manufacturers are too cheap to do things "the right way". The actual reality is that there are no methods that work perfectly - regardless of what you spend. In order to convert a 44k sample to analog without oversampling, you MUST have a reconstruction filter that is flat to 20 kHz, yet is down AT LEAST 70 dB to 80 dB at 23 kHz. (Otherwise there will be image information and distortion in your output that will distort it). Designing an analog filter that meets this requirement isn't just a matter of cost - it's difficult to impossible to do. (Instead of oversampling you would need to use a filter made up of several op amps and a lot of very carefully matched parts to even get close.) You must also remember that the physical world, and all sources, and even our ears, are also flawed. For example, while all digital filters ring at least a little - so do microphones and speakers. In other words, many of the signals that DACs can't seem to reproduce perfectly are also the sort of signals that don't exist at all ion real life and, if they did, microphones wouldn't be able to record them accurately, speakers couldn't play them, and vinyl and analog tape couldn't reproduce them accurately either.
 
3) When you talk about a compromise, you need to talk about both the pluses and minuses. For one thing, I've head some reasonable claims that, due to how the process works, Delta-Sigma DACs are much more sensitive to jitter than NOS DACs. If true, this means that, ALL ELSE BEING EQUAL, a certain amount of jitter added to the signal will cause a Delta-Sigma DAC to produce more distortion products than a typical NOS DAC. However, this is NOT a flaw per-se... it merely sets a requirement. (It means that, if you're using a Delta-Sigma DAC, it's important that you deliver it a low-jitter signal. But, if it turns out that reducing the jitter by a factor of ten solves that problem, then perhaps trading "the filter problem" for "the jitter problem" is a useful tradeoff after all - as long as you recognize the problem and design your device accordingly. However, it does mean other parts of the design may need to be adjusted.)
 
Unfortunately, a lot of smaller vendors simply lack the design knowledge to get everything right. (To use my example. You could design a NOS DAC whose input section had relatively a lot of jitter and the jitter won't affect the sound quality that much - but the filter issue will be a problem - which is why many NOS DACs have very poor high frequency response. However, if you were to simply drop a Delta-Sigma chip into that design, while it would solve the filter problem easily, you might find that it didn't sound very good - because your input stage produced enough jitter to cause the D-S chip to distort. However, it also might turn out that reducing that jitter is relatively simple... and, once you do that, the overall performance is better.) My point is that a lot depends on how the different parts of the design interact and work together, and on how well the designers understand this.
 
The obvious answer to your "headroom" comment is that you are correct. The sample rate and bit depth for CDs was chosen based on the idea that it was "just good enough" - after meeting the "requirement" of being able to fit a full album on a CD using the current maximum data rates practical at the time. The simplest way to ensure sufficient "headroom" is simply to use 24/96 - which gives you a lot more dynamic range and a much wider frequency response. This allows you to use a simpler and easier to design filter. (If you use a high enough sample rate to begin with, then there's no reason to OVERSAMPLE at all - because you've eliminated the problem it solves.)

 

[zach915m]

Just got a Gen Va in.  Will try to write up some comparison's to my trusty Progeny at some point.  But WOW, first impression is very good...ofcourse!

 

[goobicii]

44.1khz can reproduce constant sinewave only up to about 1600hz,higher and the loudness jumps up and down due to insuficient sample rate
 
yesterday I made experiments,and I make this claim "anything less that 1536khz is lossy format"

 

[evillamer]

Interesting read for those who want know how cd audio got its 74mins max play length.

http://www.exp-math.uni-essen.de/~immink/pdf/beethoven.htm

 

[wahsmoh]

  Just got a Gen Va in.  Will try to write up some comparison's to my trusty Progeny at some point.  But WOW, first impression is very good...ofcourse!

Thanks Zach. I'm a Progeny version A owner. Would be interested to know just how close the Progeny gets to the Gen Va at a fraction of the price. 

 

[Sapientiam]

  44.1khz can reproduce constant sinewave only up to about 1600hz,higher and the loudness jumps up and down due to insuficient sample rate
 

 
This is only true on an unfiltered NOS DAC - the image frequencies 'beat' with the signal and cause amplitude fluctuations. Install the appropriate reconstruction filter (most NOS DACs don't do this) to sufficiently attenuate those images and the problem is solved. Its not 'due to insufficient sample rate' as far as I'm aware.

 

[diamondears]

I wanted to comment on your comment - and on two other things as well.

1) Nyquist - What many people forget is that Nyquist is more of a "limit" than a "promise". Essentially, it is theoretically possible to provide all the information necessary to define a continuous sine wave as long as you have more than two samples per period (in other words - in order to "store" a sine wave, you need a sampling frequency at least slightly over twice its frequency). However, what this really means is that, if your sample rate is too low, then it won't be able to do the job. In other words, a CD, sampled at 44.1k CANNOT store information above 22 kHz. This does not constitute a promise that a CD WILL accurately store information up to 22 kHz - just that it's POSSIBLE - considering the sample rate. Nyquist also assumes a continuous sine wave (the math says you can totally define A CONTINUOUS SINE WAVE if you know a little bit more than two data points per cycle - it assumes an endles stream of identical cycles - which is what you have in a continuous sine wave). It doesn't necessarily hold if you have a non-sine-wave, or a combination of multiple sine waves, or a transient wave form.

2) While it's true that oversampling has flaws,so do all the other alternatives. Some people seem to have the idea that the only reason oversampling is used is that many manufacturers are too cheap to do things "the right way". The actual reality is that there are no methods that work perfectly - regardless of what you spend. In order to convert a 44k sample to analog without oversampling, you MUST have a reconstruction filter that is flat to 20 kHz, yet is down AT LEAST 70 dB to 80 dB at 23 kHz. (Otherwise there will be image information and distortion in your output that will distort it). Designing an analog filter that meets this requirement isn't just a matter of cost - it's difficult to impossible to do. (Instead of oversampling you would need to use a filter made up of several op amps and a lot of very carefully matched parts to even get close.) You must also remember that the physical world, and all sources, and even our ears, are also flawed. For example, while all digital filters ring at least a little - so do microphones and speakers. In other words, many of the signals that DACs can't seem to reproduce perfectly are also the sort of signals that don't exist at all ion real life and, if they did, microphones wouldn't be able to record them accurately, speakers couldn't play them, and vinyl and analog tape couldn't reproduce them accurately either.

3) When you talk about a compromise, you need to talk about both the pluses and minuses. For one thing, I've head some reasonable claims that, due to how the process works, Delta-Sigma DACs are much more sensitive to jitter than NOS DACs. If true, this means that, ALL ELSE BEING EQUAL, a certain amount of jitter added to the signal will cause a Delta-Sigma DAC to produce more distortion products than a typical NOS DAC. However, this is NOT a flaw per-se... it merely sets a requirement. (It means that, if you're using a Delta-Sigma DAC, it's important that you deliver it a low-jitter signal. But, if it turns out that reducing the jitter by a factor of ten solves that problem, then perhaps trading "the filter problem" for "the jitter problem" is a useful tradeoff after all - as long as you recognize the problem and design your device accordingly. However, it does mean other parts of the design may need to be adjusted.)

Unfortunately, a lot of smaller vendors simply lack the design knowledge to get everything right. (To use my example. You could design a NOS DAC whose input section had relatively a lot of jitter and the jitter won't affect the sound quality that much - but the filter issue will be a problem - which is why many NOS DACs have very poor high frequency response. However, if you were to simply drop a Delta-Sigma chip into that design, while it would solve the filter problem easily, you might find that it didn't sound very good - because your input stage produced enough jitter to cause the D-S chip to distort. However, it also might turn out that reducing that jitter is relatively simple... and, once you do that, the overall performance is better.) My point is that a lot depends on how the different parts of the design interact and work together, and on how well the designers understand this.

The obvious answer to your "headroom" comment is that you are correct. The sample rate and bit depth for CDs was chosen based on the idea that it was "just good enough" - after meeting the "requirement" of being able to fit a full album on a CD using the current maximum data rates practical at the time. The simplest way to ensure sufficient "headroom" is simply to use 24/96 - which gives you a lot more dynamic range and a much wider frequency response. This allows you to use a simpler and easier to design filter. (If you use a high enough sample rate to begin with, then there's no reason to OVERSAMPLE at all - because you've eliminated the problem it solves.)

You think the Yggy is all about that recycled megaburritto filter to avoid huge costs in designing a better filter?

 

[frenchbat]

You think the Yggy is all about that recycled megaburritto filter to avoid huge costs in designing a better filter?

 
Oh so now the megaburrito filter is cheaped out. Man, what happened to make you so bitter ? The filter of your dream has responded to your sollicitations ?
 
Or maybe you've switched to another one.
 
[Opens the centerfold page of Stereophile]
 
Let me guess it's the Chord one, right. Oh yeah .... So many taps, it's just tapalicious