Wasn't it nautibuoy who removed the last opamp stage from Marcel's filter board? And didn't one or two other people try the same thing, and said it sounded better that way? Where was the outcry then about changing "parameters" for a few people's own preference?
Instead of blaming them for doing that, I tried it too. You know what? The dac sounds "better" without that stage.
I would bet it would sound 'better' to most average listeners too.
Instead of blaming them for doing that, I tried it too. You know what? The dac sounds "better" without that stage.
I would bet it would sound 'better' to most average listeners too.
Markw4 is the fourth person making a variant of the DAC, after Thorp, nautibuoy and bohrok2610. I posted the KiCad database and some of the design considerations to make it easier for people to make their own variant, so the fact that some do is obviously fine with me. The fact that Mark uses different methods is, of course, completely up to him.
Nevertheless, when anyone makes a change I don't agree with, I will write that I don't agree and why not. What they then do with that remark is up to them.
Nevertheless, when anyone makes a change I don't agree with, I will write that I don't agree and why not. What they then do with that remark is up to them.
Hi Marcel,
I am hoping you will try listening at some point as we have discussed. I have offered my remarks about why to try doing it within some guidelines which have been found over time to on average be helpful. Depending on the outcome of such an experiment maybe we will both learn something from it? Only one way to find out
Best,
Mark
I am hoping you will try listening at some point as we have discussed. I have offered my remarks about why to try doing it within some guidelines which have been found over time to on average be helpful. Depending on the outcome of such an experiment maybe we will both learn something from it? Only one way to find out
Best,
Mark
Alright then, but what is the technical basis for this change? Yours have this ultra low noise bandgap ref buffered by opamp like in the ‘super-regulator’ configuration. I cannot imagine the LDO will do better
Let's maybe think about some more if that's okay? Marcel has a low noise reference circuit followed by bipolar buffered opamp regulators. All that stuff is built into something like LT304x. Its not just a reference. For something like a Vref regulator, the sound of the regulator error amplifier and the sound of the output cap can both have some effect on the dac sound. That's because Vref is a very special voltage in a dac. The dac output array may have zero PSRR. The regulator output voltage is multiplied times the shift register and resistor impedances as in I=E/R. Any variation in regulator output or cap distortion is convolved with the audio output, as viewed in the frequency domain. It would be probably too oversimplified to model the voltage reference difference as being the only significant difference between the two approaches.
You would have to ask @Thorp It could be something as simple as running out of board area, as Thorp's version is a Pi hat and has extra digital stuff on board to give it a PCM input. See https://www.diyaudio.com/community/threads/return-to-zero-shift-register-firdac.379406/post-7272503
Nope, I can claim no credit for that, I simply implemented a suggestion from the RTZ designer and I still don't know if it sounds different to the original filter stage because of selling my RTZ board to Acko for 'review' purposes - with hindsight a foolish move.
To be frank and IMO, the random application of various modifications based on opinions and personal preference, coupled with the lack of technical rigour, has completely devalued this thread, and it is a shame to see given its starting point of the amazing openness and generosity of Marcel in sharing his wonderful design.
The question that keeps coming to mind is 'why doesn't Mark design and share his own DAC'?
Anyway, it's time to use the unsubscribe function. When I get the time to build another RTZ board I'll start a separate thread to share progress and observations.
As MarcelvdG noted I use the original bandgap reference. When I assembled the board I measured the noise of the bandgap and found it to be similar to LT3045 although LF noise was slightly better than with LT3045 which may be important. Few days ago when testing some other changes I remeasured the bandgap noise of both channels while board was in operation and actually this time noise was even lower than in my original measurement. So based on these measurements bandgap reference has lower noise. The benefits from LT3042 would come from simplicity, savings in PCB area and cost.
Regarding 1/f-noise or excess noise of the FIRDAC resistors (I prefer to avoid the term "current noise" in this context, as it could easily be confused with the Norton current of the thermal noise):
The excess noise of resistors manifests itself as random resistance variations with a 1/f-like spectrum. Besides adding some low-frequency noise to the noise floor, it will also contribute to the AM sidebands around the desired signal, just like reference noise. I haven't calculated how large or small the effect is compared to the AM caused by reference noise.
The excess noise of resistors manifests itself as random resistance variations with a 1/f-like spectrum. Besides adding some low-frequency noise to the noise floor, it will also contribute to the AM sidebands around the desired signal, just like reference noise. I haven't calculated how large or small the effect is compared to the AM caused by reference noise.
So, theoretically it could be that using metal foil resistors leads to a perceivable sound difference, because of less AM modulation ?
Hans
P.s. Mark, since you perceive differences in sound between Vref decoupled with X5R or Rubicon, each supposedly having it’s pros and cons, wouldn’t it be an idea to audition with both placed in parallel ?
Hans
P.s. Mark, since you perceive differences in sound between Vref decoupled with X5R or Rubicon, each supposedly having it’s pros and cons, wouldn’t it be an idea to audition with both placed in parallel ?
Last edited:
Hans,
It turns out paralleling different caps of similar values has been tried before by various others. Usually, the subjective results are not good. It tends to bring out the worst of each cap, usually more than the best of each.
However, using caps of sufficiently different values, and with extra damping as needed, can sometimes work well. Unfortunately the best combination of that type I know about is NDA. So I can't use that, or if I do I can't tell you about it.
What I may do is replace the shift register output resistors first and see what that does. After taking note of the sound with X5R caps, maybe I can think of something else to try.
The above having been said, there is one old trick I have used before, which is easy enough to do and probably worth trying. It would be to put a resistor in parallel with each X5R cap. Might want to try about as low of a value resistor as the Vref power supply can safely handle. Or maybe the sweet spot for such a trial would be to try to set the pass transistor current to around where current gain is highest. Something approximately like that. Sometimes the trick can work surprisingly well, and other times no benefit is observed.
Mark
To quote from Dr. Earl Geddes:
"There are, to be sure, subjective distortions that are level dependent and as such are thought to be nonlinear distortion, but they are in fact linear effects that have a nonlinear perception. The testing of this hypothesis is currently underway and the results will probably be available in the future."
"The bottom line here is that we know so little about how humans perceive the sound quality of an audio system, and in particular the loudspeaker, that one should question almost everything that we think we know about measuring it. From what we have found most of what is being done in this regard is naive."
Sorry you act disappointed I am not so naive as some.
Also, every time I ask a measurement guy how they measure the sound stage width and depth performance of their equipment, the response is virtually always silence. The fact is nobody has a standard measurement for it, although there is some scientific research about how humans aurally localize in space. So why doesn't someone like AP just make a test for it? Probably because its hard, we still don't know everything how human hear so we don't know exactly what needs to be measured, and also because customers don't want to pay extra for a new test, especially one that might risk making their equipment look bad.
When there is no instrument to measure something relevant to human perception, then experimental methods may be required. For some which that can be heard, it is an question of discrimination, NOT preference. Some people don't seem to be capable of making that distinction, despite the fact that discrimination a standard part of perceptual testing. In fact, ABX is a discrimination test, not a preference test.
EDIT: When I am done with this dac, that should not be the end of story. It should contribute to an opportunity for others to listen to the variations on Marcel's design and form their own opinions as to what they think sounds 'best.'
New experimental results to report. A 120R, 1/2W metal film resistor was put in parallel with each 22uf, X5R bypass cap. Audible distortion is lower to my ears. However the possibly-exaggerated dynamic stereo effect is probably less than it was.
Vref series pass transistors are heating up some due to increased current load. They seem to be settling at about 50 degrees C, in open air with ambient air temp around 78 degrees F. Therefore probably best not to draw more resistor current than is being done now.
Also at the moment, aside from the 4 added resistors, the dac board is in stock form. Digital inputs are through the u.fl connectors. Reclocking is done just before the dac inputs. Output filter board is not in use at this time. Analog output is SE, positive phase only.
Since the above is an easy experiment/mod to try, maybe someone else would like to try it and let thread readers know what you think.
Vref series pass transistors are heating up some due to increased current load. They seem to be settling at about 50 degrees C, in open air with ambient air temp around 78 degrees F. Therefore probably best not to draw more resistor current than is being done now.
Also at the moment, aside from the 4 added resistors, the dac board is in stock form. Digital inputs are through the u.fl connectors. Reclocking is done just before the dac inputs. Output filter board is not in use at this time. Analog output is SE, positive phase only.
Since the above is an easy experiment/mod to try, maybe someone else would like to try it and let thread readers know what you think.
A little more commentary on the Marcel's dac configured as described in the post above.
Had 3 people listen to it, using the FPGA-based Simple DSD Converter, the Japanese PCM-DSD converter, and two variations of HQ Player.
Opinion of the dac in the configuration above, combined with Acko AKX-22 clocks, and Adrea Mori FIFO_Buffer board and reclocker board, is that its the best Marcel's dac has sounded so far. Comments include, "it's a little raw sounding, but very natural sounding space," "...you can hear the back of the room," "musical," "Less distortion than before."
Regarding the various DSD converters, in terms of space, dynamics, and other factors, more or less in that order, most natural and complete reproduction was judged in the following ranking:
1. Simple DSD Converter
2. PCM-DSD
3. HQ Player ADSM7EC2 + poly-sinc-ext2
4. HQ Player ADSM7EC2 + poly-sinc-lp
A couple of the converters were commented upon as, maybe being a little sibilant on voices but not a major problem. Space was judged as probably more important at this stage of dac development.
Had 3 people listen to it, using the FPGA-based Simple DSD Converter, the Japanese PCM-DSD converter, and two variations of HQ Player.
Opinion of the dac in the configuration above, combined with Acko AKX-22 clocks, and Adrea Mori FIFO_Buffer board and reclocker board, is that its the best Marcel's dac has sounded so far. Comments include, "it's a little raw sounding, but very natural sounding space," "...you can hear the back of the room," "musical," "Less distortion than before."
Regarding the various DSD converters, in terms of space, dynamics, and other factors, more or less in that order, most natural and complete reproduction was judged in the following ranking:
1. Simple DSD Converter
2. PCM-DSD
3. HQ Player ADSM7EC2 + poly-sinc-ext2
4. HQ Player ADSM7EC2 + poly-sinc-lp
A couple of the converters were commented upon as, maybe being a little sibilant on voices but not a major problem. Space was judged as probably more important at this stage of dac development.