Once again demonstrating that that best PSS measurements don't guarantee best sound, simply because how we measure is not how we hear, and because such measurements do not excite non-PSS effects.
EDIT: To be clear, most of us are using Marcel's version of the dac board.
With that I will stop here. Understood, bickering is not appreciated by the readers.
EDIT: To be clear, most of us are using Marcel's version of the dac board.
With that I will stop here. Understood, bickering is not appreciated by the readers.
Last edited:
I'm nobody in this thread, but I've learned so far that measured performance doesn't necessarily equate to perceived SQ. I would rather have a DAC that gets me more emotionally involved in the music, rather than having the absolute lowest noise floor/lowest THD and so forth. Man - it would be so great to be able to actually measure the emotional performance of a system using some kind of automated and standardized method, where everyone could generally agree upon the measurements and results.
Of course, that's highly likely to not happen. We're all at least slightly different in what we look for in our music, emotionally speaking - and some people want absolute perfection, and find that emotionally stimulating. I'm not one of those people; one of my hats is as a musician, who actually enjoys imperfection and uses that for emotional emphasis when playing my own stuff.
All I am saying is that from the consumer (meaning consuming the end result) perspective, I value emotion over technical performance, and don't care as much about something being superior, numbers-wise. So - without hearing and comparing - I'm inclined to go with what Mark is describing wrt SE mode.
Just one person's .02, apologize for rambling, still working on coffee, etc.
Of course, that's highly likely to not happen. We're all at least slightly different in what we look for in our music, emotionally speaking - and some people want absolute perfection, and find that emotionally stimulating. I'm not one of those people; one of my hats is as a musician, who actually enjoys imperfection and uses that for emotional emphasis when playing my own stuff.
All I am saying is that from the consumer (meaning consuming the end result) perspective, I value emotion over technical performance, and don't care as much about something being superior, numbers-wise. So - without hearing and comparing - I'm inclined to go with what Mark is describing wrt SE mode.
Just one person's .02, apologize for rambling, still working on coffee, etc.
I think you will find you can have both, accuracy of musical reproduction, and emotionally moving. Good recordings usually contain both, but one may get overridden by efforts to optimize the other. The best way I have found so far to get the two factors close into their artistically intended balance will have to come later (the goal being to reproduce a recording as it sounded at mastering and with its same emotional impact). Right now we are still focused on trying to figure out why people are hearing more low level musical details in SE mode. I'm sure there is still interest in trying to correlate that somehow with spectral measurements. That said, maybe spectral measurements will turn out to be a red herring; we'll have to see.
Anyway, after the holidays I am thinking about sending out an Iancanada clock on tour. Maybe two weeks for each user of a Marcel dac, as it may take about one week of continuous running for the clock to fully settle in. Of course, the price for trying it would be paying the postage to the next person in line.
Not intending to imply I want imperfection in this context, just that my priority is almost always the emotional aspect 🙂
I agree with what you've said about a balanced setup losing detail, because of non-linearities between the two sides. This corresponds with my experience; I found the SE setup more engaging, emotionally, but didn't have a clue why at the time.
I agree with what you've said about a balanced setup losing detail, because of non-linearities between the two sides. This corresponds with my experience; I found the SE setup more engaging, emotionally, but didn't have a clue why at the time.
Marcel's FIRdac is awesome , I think we all agree on this. How we implement it is every one's choice, single ended or balanced, one or the other layout? There are a lot of good ideas here, but I don't think there is one right and a lot of wrongs. Keep them coming. I , personally, do t think there is only one right and a lot of wrongs....
I have tried to figure out what the effect of the many proposed filter modifications for single-ended or quasi-single-ended operation would be by drawing the modified filter configurations and trying to transform them into something I could understand. I hope I got it right, some of these configurations look very unusual no matter how I transform them.
Obviously, in all cases, common-mode disturbances from the DAC core are no longer suppressed. In most cases, the output signal gets smaller.
Mark's slowing down the common-mode loop (not applicable to the OPA1632 filters) is the only configuration that neither disturbs the filter response shape nor the DC biasing.
Terminating one filter input with an RC network (378.75 ohm (or 390 ohm) in parallel with 8.2 nF) to ground leaves the filter response intact, but disturbs the DC bias point. Connecting the resistor to a clean 1.24 V source instead of ground could solve that.
You could consider using a 3:1 voltage divider (1515 ohm and 505 ohm) between the 4.96 V reference and ground instead of a single resistor to a clean 1.24 V, but then some of the disturbances on the 4.96 V will cancel out. Normally that would be regarded as an advantage, but as the purpose of the experiment is to get rid of the cancellation of common-mode disturbances, I'm not sure it is in this case.
Leaving the negative filter input connected to the DAC and adding a big capacitor from the negative filter input to ground leaves the DC bias point as is, but changes the response of the first filter stage. If I'm not mistaken, the first filter stage bandwidth will approximately drop by a factor of √2, the Q factor by a bit less than √2 and there will be an extra negative-real pole and zero relatively far away. Chances are that only cats and bats will hear the difference, though.
Connecting the negative filter input to the negative DAC output via an RCR filter with big resistors will affect both the DC bias point and the response of the first filter stage, similarly to leaving the negative filter input open.
Leaving the negative filter input open affects both the DC biasing and the first filter stage response. The Q factor of the first filter stage changes, like in an MFB filter stage where the open-loop gain of the op-amp is reduced to 1. I guess that will lead to an overdamped response, but I haven't calculated where the poles end up.
Obviously, in all cases, common-mode disturbances from the DAC core are no longer suppressed. In most cases, the output signal gets smaller.
Mark's slowing down the common-mode loop (not applicable to the OPA1632 filters) is the only configuration that neither disturbs the filter response shape nor the DC biasing.
Terminating one filter input with an RC network (378.75 ohm (or 390 ohm) in parallel with 8.2 nF) to ground leaves the filter response intact, but disturbs the DC bias point. Connecting the resistor to a clean 1.24 V source instead of ground could solve that.
You could consider using a 3:1 voltage divider (1515 ohm and 505 ohm) between the 4.96 V reference and ground instead of a single resistor to a clean 1.24 V, but then some of the disturbances on the 4.96 V will cancel out. Normally that would be regarded as an advantage, but as the purpose of the experiment is to get rid of the cancellation of common-mode disturbances, I'm not sure it is in this case.
Leaving the negative filter input connected to the DAC and adding a big capacitor from the negative filter input to ground leaves the DC bias point as is, but changes the response of the first filter stage. If I'm not mistaken, the first filter stage bandwidth will approximately drop by a factor of √2, the Q factor by a bit less than √2 and there will be an extra negative-real pole and zero relatively far away. Chances are that only cats and bats will hear the difference, though.
Connecting the negative filter input to the negative DAC output via an RCR filter with big resistors will affect both the DC bias point and the response of the first filter stage, similarly to leaving the negative filter input open.
Leaving the negative filter input open affects both the DC biasing and the first filter stage response. The Q factor of the first filter stage changes, like in an MFB filter stage where the open-loop gain of the op-amp is reduced to 1. I guess that will lead to an overdamped response, but I haven't calculated where the poles end up.
For the sake of argument let's assume Markw4 is onto something. So this DAC has audible issues (missing detail) when operated as intended by the designer. What this means is that there is an inherent flaw in the design or the implementation. As the designer has not been able to identify such fault the former seems unlikely so implementation may be at fault. However a far more likely explanation is that SE mode adds "detail" not part of the original recording. As already stated this "detail" can be caused by common mode noise and distortion which are likely higher in SE mode. But the "detail" can come from other sources as well e.g. from clocking. In Markw4's case it could come e.g. from clock buffer oscillator crosstalk, from insufficient power supply filtering, or from inappropriate decoupling capacitors.
Note that anybody is free to form their opinion. If you prefer SE mode so be it. However subjective opinions are not comparable so those that prefer SE mode are not more correct than those that prefer the mode designer intended. With measurements such comparisons are possible so if some device measures clearly worse than another it is a factual statement to say that the former is inferior.
Note that anybody is free to form their opinion. If you prefer SE mode so be it. However subjective opinions are not comparable so those that prefer SE mode are not more correct than those that prefer the mode designer intended. With measurements such comparisons are possible so if some device measures clearly worse than another it is a factual statement to say that the former is inferior.
Last edited:
We don't know the true superior/inferior metric of any dac here, since the unit of superiority is well known to be SINAD/Dollar (as determined by people at another website who know for a fact the defined unit of measure is scientifically based).
However, we can say that the value of a human being's worth has arbitrarily been decided to be a function of Height. Therefore taller people have decided they are superior to inferior shorter people. And measurements prove they are factually taller!
However, we can say that the value of a human being's worth has arbitrarily been decided to be a function of Height. Therefore taller people have decided they are superior to inferior shorter people. And measurements prove they are factually taller!
Last edited:
Haha! As a tall person I fully agree with this statement. (/s of course)
That said, one of my favorite people is Danny Devito, and a girl I'm very interested in is significantly shorter (and probably more emotionally mature, and definitely better looking) than myself.
Looking forward to building and experimenting with a number of iterations of this dac. Kudos especially to Mark and Marcel to furthering this effort.
That said, one of my favorite people is Danny Devito, and a girl I'm very interested in is significantly shorter (and probably more emotionally mature, and definitely better looking) than myself.
Looking forward to building and experimenting with a number of iterations of this dac. Kudos especially to Mark and Marcel to furthering this effort.
As usual, subjective SQ reports should be taken with a substantial amount of grain of salt (its mass could be quantified!) as there is possible no truth in them and the aura of certainty that they are delivered with is not seldom reversely correlated to the reporters level of Hybris.
//
//
What I know for a fact is that I have other DACs that measure clearly better in every possible measurement I have taken. That includes close-in phase noise and crosstalk. They also sound at least as good and are cheaper to build. And work as the designer intended without having to resort to alternative solutions that have inherent flaws.
As the only task of a DAC is to produce a voltage level at each time, analysing the properties of the signal at each time and finding that quality aspects (level, noise, phase, distortion) are better for A than B, all natural logic ends up that A is in fact a more true recreator of the original signal than B. It's just that simple. Musicality, stage, inner detail, PRAT... whatever, is just imagination of a listener which make some listeners prefer B before A. To say that B is "better" is nonsensical.
To say "I prefer B" is human and honest and depends on so much more on solely A and B, "I prefer B, in my system" is even more better ;-)
//
To say "I prefer B" is human and honest and depends on so much more on solely A and B, "I prefer B, in my system" is even more better ;-)
//
We know about measurements of other dacs, and even I chronicled in this thread how I started out with the RTZ dac belonging to Acko. It sounded bad, real bad. It was driven by an Amanero, with shared power supplies, etc. So, we already know this dac can sound bad and I'm sure measure bad too. We also know it can sound extraordinarily good, but overall its still should be considered a work in progress. Give it some more time, please.
Regarding what dac designer's intended versus what they ended up with in reality, its clear enough ESS has been a prime example of that. The hump distortion in ES9018, a chip they sold on the basis of good measurements was an unintended flaw. That's why they asked Scott Wurcer for help trying to figure it out. If you remember the notched distortion scope shot Scott posted here in the forum, obviously it wasn't a PSS waveform so it didn't show up well in FFT measurements. It sounded bad enough to enough people though for them to want to fix it. Didn't stop them from selling silicon in the meantime to get rich, however. I don't see them being criticized for a history of flaws and internally known errata (which all complex chips tend to have). They know some engineers only believe what they see on an FFT, thus so long as the FFTs look good, lots silicon can be sold to those engineers.
There are many other examples of products being designed and then flaws found and maybe fixed in later versions. Bruno Putzeys evolution as a product designer comes to mind. He has talked about finding problems in production units and then fixing the problems in later versions.
So, no need to insult Marcel for being no worse than ESS, in fact I would say that he is much more honest than ESS is maybe the main difference.
To summarize, kindly let us keep working on this dac. It may take a few months. If you can't stand to read about then don't, but kindly don't insist we should give up now just because there are still some things to work on that you don't like reading about.
Regarding what dac designer's intended versus what they ended up with in reality, its clear enough ESS has been a prime example of that. The hump distortion in ES9018, a chip they sold on the basis of good measurements was an unintended flaw. That's why they asked Scott Wurcer for help trying to figure it out. If you remember the notched distortion scope shot Scott posted here in the forum, obviously it wasn't a PSS waveform so it didn't show up well in FFT measurements. It sounded bad enough to enough people though for them to want to fix it. Didn't stop them from selling silicon in the meantime to get rich, however. I don't see them being criticized for a history of flaws and internally known errata (which all complex chips tend to have). They know some engineers only believe what they see on an FFT, thus so long as the FFTs look good, lots silicon can be sold to those engineers.
There are many other examples of products being designed and then flaws found and maybe fixed in later versions. Bruno Putzeys evolution as a product designer comes to mind. He has talked about finding problems in production units and then fixing the problems in later versions.
So, no need to insult Marcel for being no worse than ESS, in fact I would say that he is much more honest than ESS is maybe the main difference.
To summarize, kindly let us keep working on this dac. It may take a few months. If you can't stand to read about then don't, but kindly don't insist we should give up now just because there are still some things to work on that you don't like reading about.
Last edited:
You are free to do whatever you like. And others are free to comment on what you are doing. If you can't stand reading those comments then don't. My reason for occasionally reading this thread is simply because some members are using an output stage that I presented in this thread.
I for one do not appreciate your attitude, @bohrock2610 so you are now ignored. You were once before, but I did give you another chance, since it appeared like you could be a positive contributor. Never more, so that means I won't read any of your other posts - I'm not at all sure whether that's a bad thing, to be quite honest. I like to be influenced by positive people.
I encourage anyone else of like mind to do the same. If I recall correctly, you were banned for a period of time. Anyways good luck with your attitude, wow!
@Markw4 and everyone else, please continue, I'm literally all ears 🙂 I'm learning and very appreciative of what you are doing here.
Mods please delete as you see fit.
I encourage anyone else of like mind to do the same. If I recall correctly, you were banned for a period of time. Anyways good luck with your attitude, wow!
@Markw4 and everyone else, please continue, I'm literally all ears 🙂 I'm learning and very appreciative of what you are doing here.
Mods please delete as you see fit.
Maybe it is due to my autism, but if @bohrok2610 or anyone else has tried to insult me in recent posts, I have completely missed it. Besides, bohrok2610 made many constructive contributions to this thread, particularly alternative reconstruction filter designs and measurements. For example, without his experiments and measurements, I would not have learned that frequency-modulated idle tones can look like harmonic distortion when the carrier frequency is zero.