For those who have the interest and ability, the kinds of psychoacoustics discoveries in question here are probably only possible on a handful of platforms. Delta-sigma chips are almost immediately ruled out since there are too many confounding variables in the chip that are hard to untangle. PCM1704 may not work that great either due to the high level of integration as well as the I/V stage needed. I would not be surprised if someone finds out that there are important variables in audio reproduction long overlooked by semiconductor companies, because many of them might always have delta-sigma as a given, in both research and development.
P.S. There is potential for real audio science. In comparison ASR is no more than lab technician work. Soren, if you tried all of this please don't let us waste our time fooling around.
P.S. There is potential for real audio science. In comparison ASR is no more than lab technician work. Soren, if you tried all of this please don't let us waste our time fooling around.
Last edited:
For those who are just starting to read this thread, I want to clarify that the recent discussions should be considered a project separate from implementations. As far as I know, I agree with Soren's statement some time ago that the original goal of the project has been met, that a Soekris system can potentially be better than any Delta-Sigma based DAC if we are right that there is something bad in D-S for audio that we do not yet understand, which seems a very reasonable hypothesis to me. Soekris devices can also be considered in-class with MSB and TotalDAC according to our best theories and speculations. MSB would have to show proof that the jitter reduction from their $20k femto clock is audible to claim that they out-class the maybe 200ps jitter dam1021 and <1ps later products.
There is one caveat if you go dual-mono, which I consider technically superior in some sense without falling into the category of extravagance (think $30 resistors and $500 transformers) or unreasonable solutions (I have in mind audio-gd/denafrips circuitry). The issue is that dual-mono in dam1021 has audible (2.7-22uS) delays due to the async operation. There is potentially a firmware solution that will reduce this to completely inaudible levels or even zero, but it depends on what Soren prioritizes. Apart from this, I'd say buy with confidence if you're looking for a near state-of-the-art but non-extravagant and reasonable solution for audio DAC.
Even though we achieve near sota with the dam dacs, it's hard to ignore observations, if they do exist, that we cannot satisfactorily explain with our best theories. I suggested earlier that we look into whether Vref mods do make a difference, because if so, it might open up a direction for exploration. Not just for personal DIY, but for potentially arriving at new theories that can offer a more satisfactory explanation for the differences we do hear but cannot penetrate. If the Vref mod does indeed have an impact on sound, we'll need well-controlled tests along the way to ensure that our results are reliable and scientific. It is unlikely that the usual DIY method of claiming to hear a difference and being content with a theoretically harmless improvement will be able to take us any further, even if DIY'ers have occasionally and unknowingly hit on real treasures.
We'll have to first determine if we have any reliable and interesting data before this project can be said to begin, which in our case most likely requires ABX comparisons on the Vref mod. I believe that the project and its efforts are fully serious, even if the pursuit of better audio technology and science is perhaps not the most impactful research direction we have. I hope that many of us here who has the means to contribute will find the proposal interesting, and I'm very much looking forward to any progress that we might be able to make in this little project!
There is one caveat if you go dual-mono, which I consider technically superior in some sense without falling into the category of extravagance (think $30 resistors and $500 transformers) or unreasonable solutions (I have in mind audio-gd/denafrips circuitry). The issue is that dual-mono in dam1021 has audible (2.7-22uS) delays due to the async operation. There is potentially a firmware solution that will reduce this to completely inaudible levels or even zero, but it depends on what Soren prioritizes. Apart from this, I'd say buy with confidence if you're looking for a near state-of-the-art but non-extravagant and reasonable solution for audio DAC.
Even though we achieve near sota with the dam dacs, it's hard to ignore observations, if they do exist, that we cannot satisfactorily explain with our best theories. I suggested earlier that we look into whether Vref mods do make a difference, because if so, it might open up a direction for exploration. Not just for personal DIY, but for potentially arriving at new theories that can offer a more satisfactory explanation for the differences we do hear but cannot penetrate. If the Vref mod does indeed have an impact on sound, we'll need well-controlled tests along the way to ensure that our results are reliable and scientific. It is unlikely that the usual DIY method of claiming to hear a difference and being content with a theoretically harmless improvement will be able to take us any further, even if DIY'ers have occasionally and unknowingly hit on real treasures.
We'll have to first determine if we have any reliable and interesting data before this project can be said to begin, which in our case most likely requires ABX comparisons on the Vref mod. I believe that the project and its efforts are fully serious, even if the pursuit of better audio technology and science is perhaps not the most impactful research direction we have. I hope that many of us here who has the means to contribute will find the proposal interesting, and I'm very much looking forward to any progress that we might be able to make in this little project!
Last edited:
ynmichael could the fact that upsampling to the max sampling rate, effectively reducing the delays to the minimum more consistently each lock, be the source of the improved quality overall? I always thought at some point the delay was there and noticeable but it seemed hit or miss, might be worth getting a 1941 afterall just to have something to test against. I too have asked Soren for a sync solution in the past, even if it meant tapping an unconnected pin to do so, avoiding compatibility breakage.
The dam1021 async delay is a problem that Soren thought about but judged to be inaudible by mistake. In fact it’s the only time he’s turned out to be wrong I think. It’s understandable because some previous research put the audibility limit at 50us, but recent tests clearly showed that 5/10uS is perfectly audible too in tests. Soren already anticipated this issue when he designed the board but didn’t think it worthwhile to fix a couple of years ago, with the new information available now he might find reason to reconsider. An ideal engineering company, in my opinion, would try to address newly discovered issues to the extent that it’s possible with the existing product design. I’m not an audiophile and I probably won’t be constantly comparing everyday the very minute differences in instrument localization due to the async delays. But again it’s up to Soren to decide whether to fix this issue with a firmware update or not; it’s up to him to decide what kind of engineering company Soekris Engineering would be, and certainly how it impacts his largest and oldest group of customers to his DAC products...
Yep! It’s harder to ABX in a speaker setup since you would have to keep your head perfectly still the whole time 😉 but the difference is there of course
As I already said a couple of times, there is no problem with one channel delayed a little, no matter what you claim. It's like moving one speaker a couple of mm, I don't know about you, but I don't place my speakers and listening position that precise....
Multiple dam1021 was never intended to be connected synchronous, and it will NEVER happen. For those that require that, I offer the dam1121 and the newer dam1941 is born fully balanced.
So ynmichael, please stop repeating yourself, you're wasting mine and everybody elses time.
Last edited:
Maybe it comforts you to hear that a recording where the stereoinformation is derived fom a spaced pair of microphones (That's where the timing differences count) does not work as intended if listened under headphones. There is Stereoinformation, of course, but the ears decode them wrongly anyway, because you need the signal of each speaker on both ears to have the stereoinformation the recording engineer heard on his speakers. Those recordings are made to be heard on speakers.
I think there is really no problem with stereoreproduction on the Dam1021, not even theoretically....
It does matter when you’re using headphones. I’ve offered ABX results earlier. Is that not sufficient evidence? The question of whether dam1021 would be designed to be synchronous depends on whether asynchronous operation functions properly in dual-mono, doesn’t it? I believe you intend to offer dual-mono capability on dam1021.
If it's my best chance at landing the shot, what's the difference? I wasn't aware that we have access to a perfect targeting technology either. It's nice to take careful aim but I see no reason to refrain from taking a potshot if I have to.
It would be nice to know yes... I'm not sure why it would track to one word clock cycle either, but I also don't know if Soren is willing to share that information as relevant as it seems.
I'm not personally concerned about the music playback capability of dam1021 in dual-mono. As a matter of fact there is no way I would care enough to scrap my build and go for a different system - there wouldn't be any practical difference at all for the purpose of music enjoyment.
The main reason I came back to this topic is that I find very valuable an understanding of why our seemingly sync-capable DACs end up having audible delays. Even if near-perfect sync is only needed in esoteric situations, e.g. audiophilic comparison of the exact instrument imaging of two DACs, or non-music uses, the fact that our DAC has audible delays does raise very real questions. Obviously none of us started our dual-mono build thinking that we're getting audible delays, so why did it turn out this way?
Like I said I don't personally care much practically for the async problem, but it may still be very significant for the product on a larger scale. I found it really important to understand how and why our situation came to be, and perhaps it may be of use to others too. If I were Soren, this would also be a very important engineering problem to solve. For he fully intended to have the system fit for dual-mono use and advertised it as such, which was perfectly reasonable at the time, as many of us also believed that the uS delay is irrelevant. But new evidence clearly suggests that it is not actually the case, as much as it seemed the other way a couple of years ago. Soren doesn't necessarily have to care about my opinion on his company, but I believe it's important for me to form that opinion and an explanation of the situation based on the facts of the matter.
Last edited:
I think the topic of multi dam board delays has been sufficiently touched now. Let's move to topics where we can actually improve sound quality of the Dam dac ourselves.
I found the Vref related topic more interesting. In my opinion 1 mV Ripple on vref supply should have audible effects. I would assume that a drop in input voltage in the shift register directly correspondents to a drop of the output (plus probably all kind of non-linear or dynamic effects). This would already have effects in the magnitude of 12 bits lower. Or do I overlook something here? I assume no additional power supply rejection filtering is done in the shift register as it is designed for digital.
Fedde
I found the Vref related topic more interesting. In my opinion 1 mV Ripple on vref supply should have audible effects. I would assume that a drop in input voltage in the shift register directly correspondents to a drop of the output (plus probably all kind of non-linear or dynamic effects). This would already have effects in the magnitude of 12 bits lower. Or do I overlook something here? I assume no additional power supply rejection filtering is done in the shift register as it is designed for digital.
Fedde
Last edited:
Nope you missed nothing as far as I can tell. But why would it sound any different with 1mv ripple, or even 6mv? Of course if we can get ABX test data then we can try to come up with some new theories to explain the observation. But you see how our current theories predict that the sound wouldn't be any different since THD would still be way below 0.01% (which is still far from the audibility limit), and that we don't know what else the ripples or nonmonotonicity would affect?
Last edited:
I think that the target for ripple related distortions should be to get under distortions caused by resistor inprecisions. Assuming 0.01% or 0.02% resistors, ripple of 0.4 mV or 0.8 mV would be in line. Though ripple naturally works in a different way as resistor inprecision. I think we should target for below 0.1 mV ripple to leave some margin.
Fedde
Fedde
That's what I thought earlier too. But why? Our current understanding suggests that 0.05% should sound absolutely no different from 0.01% since THD is already low enough, barring that some of the 0.05% go really off over time. If resistor mismatch was such an important issue, we would all be working on resistor software compensation now...
- Status
- Not open for further replies.