Btw on dam1021 we might want to bypass the 5V LDO. TI 78xx lists 17mR as output resistance, I assume the ripples add up over the chain?
See full mod here: Reference DAC Module - Discrete R-2R Sign Magnitude 24 bit 384 Khz
Result looks quite clean. But maybe extra work is needed to save muting circuit? Reference DAC Module - Discrete R-2R Sign Magnitude 24 bit 384 Khz
See full mod here: Reference DAC Module - Discrete R-2R Sign Magnitude 24 bit 384 Khz
Result looks quite clean. But maybe extra work is needed to save muting circuit? Reference DAC Module - Discrete R-2R Sign Magnitude 24 bit 384 Khz
Last edited:
Ideally we could also do reclocking after shift register like totaldac, and only afterwards buffer with MOSFETs. The clock signal can probably be taken from the board directly with careful wiring. Vref caps can also be mounted directly on such an add on board as SMT parts.
I think it’s a worthwhile experiment to take on as a community. There shouldn’t be too much initial costs to cover. But it definitely needs to be a group effort pooling ideas and resources...
I believe the only theoretical question that remains to be answered is whether mosfets are best for “buffering”, and whether they have low enough variances between units in propagation delay time.
I think it’s a worthwhile experiment to take on as a community. There shouldn’t be too much initial costs to cover. But it definitely needs to be a group effort pooling ideas and resources...
I believe the only theoretical question that remains to be answered is whether mosfets are best for “buffering”, and whether they have low enough variances between units in propagation delay time.
Last edited:
How to Read a Power MOSFET Datasheet - Jason Sachs
"Sometimes you learn with experience which device parameters have a lot of part-to-part variation, and which ones will tend to tightly cluster around the mean. It’s not something that will be stated in the datasheet; you’ll have to read the manufacturer’s appnotes, contact their applications department, or do your own tests to make those kind of conclusions."
----------
Also found this: www.epboard.com/eproducts/protoadapter.htm
Maybe we can custom order some TSSOP-16 to DIP adaptors?
"Sometimes you learn with experience which device parameters have a lot of part-to-part variation, and which ones will tend to tightly cluster around the mean. It’s not something that will be stated in the datasheet; you’ll have to read the manufacturer’s appnotes, contact their applications department, or do your own tests to make those kind of conclusions."
----------
Also found this: www.epboard.com/eproducts/protoadapter.htm
Maybe we can custom order some TSSOP-16 to DIP adaptors?
Last edited:
Tolerance Analysis in MOSFET Analog Integrated Circuits | Sylvain Feruglio and Patrick Garda - Academia.edu
If it's within a few percent of 10ns we're probably fine considering there might also be some averaging going on, which could use more mathematical analysis later on. But first someone more familiar with the industry might want to get a more accurate estimate for our particular case...
If it's within a few percent of 10ns we're probably fine considering there might also be some averaging going on, which could use more mathematical analysis later on. But first someone more familiar with the industry might want to get a more accurate estimate for our particular case...
Btw, the current design also suffers from the possible tolerance mismatch in the shift registers' switching characteristics, even though the shift registers might be balanced within their 8 bits. I'm not even sure if it's better to have "segmented" glitches or some randomness in every bit. Reclocking after shift register might help a bit, provided that the flip-flops are matched more closely, but it's probably similar. In any case, for dam1021 reclocking after shift register might help reduce the jitter from FPGA.
If this experiment bears fruit, which is reasonable to believe based on our current evidence and theories, the ultimate solution that I see is a custom chip that has perhaps 8 or 16-bits for either the purpose of impedance buffering alone, or combined with flip-flop reclocking, whichever has better performance or cheaper if both are equivalent in audible performance. It might offer a small improvement in jitter performance but huge reduction in total PCB area needed, which would make for a great non-DIY product. This would be a much more impressive effort and is probably best taken on by a single individual with the resources, both technical and financial. Perhaps Soren would be interested if the proposed experiment proves to be worthwhile.
If this experiment bears fruit, which is reasonable to believe based on our current evidence and theories, the ultimate solution that I see is a custom chip that has perhaps 8 or 16-bits for either the purpose of impedance buffering alone, or combined with flip-flop reclocking, whichever has better performance or cheaper if both are equivalent in audible performance. It might offer a small improvement in jitter performance but huge reduction in total PCB area needed, which would make for a great non-DIY product. This would be a much more impressive effort and is probably best taken on by a single individual with the resources, both technical and financial. Perhaps Soren would be interested if the proposed experiment proves to be worthwhile.
Last edited:
Soekris,
I can tell you one thing. You designed a great and finished product! Your dad sounds fantastic. I have to confess that I do not like the buffered output as that one sounds a bit harsh to me, but I look at the buffer as a kinda bonus. Need balanced and more output out of the box -- here you go.
Regarding all this modding stuff. We are all DIYers here and some people just have an itch for performing DIY and that's why we have this forum. I agree that it is getting a bit too much in your product discussion and probably has to move to other section like normal Digital Line Level section.
I performed that Vref cap mod (I have v4 of the dac) to have my own opinion on that one. What I can say here. Does the sound changed after? Maybe the background got a bit "darker"... would I noticed that in a blind testing? Probably not. It is an easy thing to do and takes like 5 minutes, so if you feel like performing your own DIY -- go for it. If not -- not a problem as Soekris designed one good piece of digital gear for us.
Another thing I can not really understand people comparing about not enough bass, this DAC has more bass than any dacs I've had in my system, so I guess it is a matter of perception and the rest of your equipment
. Cheers!
You're probably right that the LVC595 and the one on your later products are the best shift registers that one can find for the job, thank you for the explanation - I'm not familiar with all types of shift registers available and their characteristics. But my relative lack of experience in electronic design does not automatically disqualify everything that I have to say.
The comment you made this time is finally reasonable and not simply passive aggressive and suspiciously defensive. But I don't think I can accept your point that discrete MOSFET is "too many complications without improvements", not a priori. The most valuable piece of knowledge we have obtained in the last few years of DIY seems to be that the Vref matters a great deal. Correct me if I'm wrong, but I believe that you as an engineer believe in measurable results and empirical data, preferably with the explanations provided by theories, but that they are not always necessary. This enabled you to create this wonderful discrete DAC that too few have attempted before you and provided it at an extremely reasonable price compared to your competitors. But it also led you to believe that resistors with lower tolerances (among other things such as significantly better Vref) do not matter in normal conditions because they only cause THD distortion and we cannot hear THD <0.1% based on our sciences.
I agree with the latter point, but it is actually far from clear that monotonicity glitches only affect frequency-domain, which modern technology already excels at, but not time-domain, for which we have far too few tools of analysis but which is a central component of music listening experience, if nothing else. The fact that everyone seems to agree that Soekris DAC as well as other R2R DAC have very natural timbre compared to Delta-Sigma DACs suggests that there is indeed something about timbre that our previous (mostly delta-sigma) DACs have not done well enough at, and that there might be lots of room for improvement. Our lack of scientific or theoretical understanding is further evidenced by the incredible confusions that online audiophile communities have over the debate of Delta-sigma vs R2R. This is not to say that no one has had a strong preference or even belief in either, but that no one, and I mean no one, has presented sufficient empirical evidence or theory on why it is. Those that believe that an ever decreasing THD is a sign of progress in our audio technology is just as foolish, in my opinion, as those who believe a $50000 cable with sophisticated theoretical underpinnings on why it propagates signal better, sounds better or is even essential. All evidence points to the fact that there is no audible difference in these technical distinctions. However, what interested me last Friday was the fact that our Vref improvements have yielded such significant improvements on sound quality. I believe there is no way that it can be explained by our current theories or accepted measurements such as THD, IMD or linearity. This naturally invites curiosity and after some investigation it occurred to me that perhaps the differences are hidden in the not well understood dimension of timbre in sound. As the textbook I shared earlier suggests, there may be as many as 34 different dimensions of timbre, few if any of which has received sufficient treatment in audio product design or subsequent measurement tests. If we find ourselves inclined to attribute the Vref improvements to the somewhat obscure dimension of timbre, then it is only natural to ask to what extent we can further benefit from such improvements in this direction. I should also add here that it is intuitively plausible to me that Vref ripples audibly impact timbre, for our best knowledge on instrument timbre suggests that it is but special types of ripples, some with very small signal levels.
This leads me to a reexamination of the Vref situation. I say "reexamine" because I was previously convinced that the Soekris DAC has achieved technical perfection to the best of our knowledge and if people claim that something is better, it would either be coloration by distortion or total blind luck in development, the former being much more likely. Coming back on point, according to the in-depth technical explorations carried out by Paul three years ago (who was driven out of the community by the way because he was quite sensitive to the criticism he received, I'd say unjustly, and also lack of any appreciation from Soren himself, who have been repeatedly improving the Vref capacitance configuration on his recent products and no doubt benefiting from the experimentations done in the DIY community. But I digress here). According to Paul's simulations (which I do trust to quite an extent), the stock Vref buffers on rev.2 has about 50mR impedance, not including the internal resistance of the shift register. It is likely that 300uF capacitance on rev.4 and perhaps even a bit more on rev.5 (exact amount undisclosed by Soren despite multiple requests for information) will reduce it by a significant amount, but this doesn't matter much in my argument. Now, our empirical data tells us that adding lots of Vref (I have ~5000uF low-ESR capacitance each rail, others have tried more), yields very audible, almost shocking changes, and if I may, improvements, in the sound signature - stronger and deeper bass without losing the tightness, better dynamics etc. At the same time, simulations suggest that successful Vref modding will reduce the output impedance to about 6mR flat. We shouldn't compare impedance directly, for ripple also depends on other factors. Paul's simulation puts the improvement on rev.2 with and without caps to be about a 350uV reduction in ripple, less in rev.4/5 due to a few more ceramics added. The shift register, which has ~300mR internal resistance, would contribute about 810uV of ripple (all worst case calculations, i.e. all bits on to off). Furthermore, I believe the simulations are fully confirmed by zfe's measurements which showed a -74.2db RMS noise in Vref, corresponding almost perfectly with Paul's calculation of -72.95db RMS noise based on 810uV (shift register) + 68uV (improved Vref) correlated sum, -73.62db uncorrelated sum.
I believe I'm right to claim that it is worth investigating whether reducing the internal resistance of the shift register would further improve sound quality in very audible terms since we hear dramatic changes (or as I believe improvements) with a reduction of <350uV ripple, when there is still about 800uV of ripple that we can get rid of by working on the shift register. You can go into effective n-th bit calculations here and arrive at horrifying results but I wouldn't be too worried as all research put the audible number of bits threshold to be fairly low, making it at least an irrelevant metric for now. (I say for now because most research is done on delta-sigma DACs which may have covered up the problem and made the audibility limit appear much more lenient than it is, but this is not something we can test right now nor concerned with in Soren's R2R thread.)
I double checked Infineon's MOSFET catalog, which seems to have much better switching characteristics and Rds than ON, Vishay and possibly TI, and this model is extremely impressive. The idea is obviously to use only Q2 channel to prevent imbalances, and the only reason for dual channel is that it's actually much, much better in switching time than single N-channel models. Basically <5ns on all switching characteristics, and 1.7mR Rds at 4.5V Vgs (we would be supplying 4V hypothetically in dam1021). Our shift register on dam1021 has a very, very similar performance, except its Rds (internal resistance) is about 200 times the MOSFET. Nexperia marks both typical and maximum switching times, not sure if it just means Nexperia measures carefully or that shift registers have higher tolerances, but regardless their switching performance is very similar. The "problems" that Soren mention are probably switching noise, speed and maybe delay differences. I don't know how much switching noise matters at 3Mhz, and the speed of the MOSFET certainly looks sufficient for a DAC that Soren measured to have a speed about 70nS 10/90% rise/fall time. The delay difference is a big problem though, even if we assume that the tolerances are good to within 1ns. But this is problem that shift registers suffer from too, and even if each shift register is perfectly consistent in its 8-bits, which I'm not fully convinced of, there is still variations in the 8 shift registers used per channel. Is the delay variations better or worse when we have a MOSFET for each bit rather than a shift register for every 8 bits? It might need more numerical investigation. I would not see this as equivalent to jitter, nor is ns-level jitter sufficiently proven to be audible. In any case, whether or not we're comparing against shift registers as a benchmark, the timing errors should best be viewed as a compromise with the amplitude errors, if we can regard ripples as such. I think that it's fully possible that we find that ns level timing errors in each MOSFET, which may be much more benign than global jitter at the same scale, turn out to be much less significant than the 800uV ripple. I hope I've made my argument clear that the "complications" may be minimal and the "improvements" significantly audible. At the very least, it seems to me convincing evidence that Soren's statement on this issue is not to be accepted without further justification.
Soren has a history of occasional untrustworthiness in this thread, especially regarding modding. He claimed repeatedly that rev.4/5 sounds great and do not need modding. Yet the consensus is that Vref caps on rev.4/5 are best characterized as mandatory. I say this not to discredit Soren, for he is right most of the time, but only to show that my skepticism may be justified. At this point I won't say much about why Soren sometimes offer opinions he knows to be not entirely truthful.
I'm not sure if encouragement works on Soren. He has made it very clear in his past comments that he only implements a feature (which can be taken as making any non-negligible investment of his time) if it generates significant sales. You can find verbal proof in this thread. Also, a few people here might also testify that friendliness and encouragement in the broad sense have not seemed to help solicit (much-deserved) assistance.
I'll be completely frank here and claim that he's only saying what he did to ask for positive feedback which is good for business. I do agree with your comment that his DAC is very good and I enjoy it very, very much. However, this does not mean that we're forbidden to explore room for improvement, or any talk of it blasphemy.
Majority also doesn't have a monopoly on truth, since you brought it up...
I wish you've said more in this thread.... Did you apply the low-res mod in the end to get rid of "wooly bass" caused by 4000uF capacitance or?....
PS: Please stop polluting
Nice build! Definitely a good idea to get rid of the blue EMI-emitter that comes with AZ Lite
I may be obnoxious to you, but you would do well to explain how "my passions have gotten the best of me". Maybe, maybe you're right and I know that I can benefit from a lesson.
Last edited:
Soekris,
Another thing I can not really understand people comparing about not enough bass, this DAC has more bass than any dacs I've had in my system, so I guess it is a matter of perception and the rest of your equipment
Cheers!
Hey all of us found that the vref mod was what improved the bass. Are you sure about your statement? Would you care to take off maybe the caps in one channel to verify? If you have 0.6mm lead diameter caps you're right it should be very fast...
Am I the only one that finds paragraph 7 of post 8725 above not only disrespectful but a thinly-veiled personal attack? Uncalled for in a social forum.
With respect, ynmichael, you state ‘you know you can benefit from a lesson’ - might I suggest two things:
An application of the age-old guiding principle of good writing, namely, Clarity, Accuracy and Brevity.
An immersion in a little more electronic engineering theory before launching volumes of what often comes across as ‘learn as you go’ theorizing. It diminishes your credibility. Or at least start your own thread. You earlier claim you are reluctant to do that as ‘you don’t have the time’. Forgive me for thinking that a joke based on your efforts above.
Here Endeth The Lesson
PS As this post is off-topic I will not be responding to any reply. I suggest you do not reply either as it would be further off-topic and, who knows, the lesson that might end up being learnt is that it’s entirely possible an entire thread could ignore any further posts.
With respect, ynmichael, you state ‘you know you can benefit from a lesson’ - might I suggest two things:
An application of the age-old guiding principle of good writing, namely, Clarity, Accuracy and Brevity.
An immersion in a little more electronic engineering theory before launching volumes of what often comes across as ‘learn as you go’ theorizing. It diminishes your credibility. Or at least start your own thread. You earlier claim you are reluctant to do that as ‘you don’t have the time’. Forgive me for thinking that a joke based on your efforts above.
Here Endeth The Lesson
PS As this post is off-topic I will not be responding to any reply. I suggest you do not reply either as it would be further off-topic and, who knows, the lesson that might end up being learnt is that it’s entirely possible an entire thread could ignore any further posts.
Thank you for the candid advice. I really appreciate it and it's also very good to know how people perceive of my comments. But you know I have to reply.
Brevity is hard when there are lots of information to convey (also not all the time in the world to edit). There's only so much you can do to compress. If you only see the short version, how do you decide what to accept as more plausible? Is the vendor, by default, the arbiter of truth? I didn't say it earlier but that is part of Soren's strategy, mixing technical discussion with marketing, and not executed well in my opinion.
This is not EE stackexchange so I've always assumed that some exploration is allowed. If you have to discredit what I said because I learn as I go, I think you're missing out on valuable information but there's nothing more I can do.
I did not expect to post 8725 but there seems a need after Soren's claims and everyone's immediate support. No need to say more about it.
My "thinly-veiled attacks" are statements of fact. I can't say I'm sorry if it hurt your feelings. It's easy to want to put your trust in someone and not have to go through the reasoning. Sometimes it's necessary and it works, sometimes there're better choices. I don't know if statements pointing out someone's mistakes that have an impact on others are or ought to be banned on forums, but I hope they be allowed.
I do appreciate your comment very much, but what did you think of the technical issues? That was the whole point. The others were there to help make sure that the important parts don't get thrown directly in the trash as some would like.
Edit: Perhaps you might find useful to identify what exactly my conclusion would mean for your past opinions and decisions, if it proves to be right. I only recently realized that I found it hard to trust my judgements after facts indicate that I might've been wrong about something - there's this worry or doubt floating around that I couldn't quite place, and it did not help obsessions, including audio. For me, understanding why I made the mistake helped a lot, contrary to the popular advice that time heals, which I find to be escapist and probably damaging in the long run. I had the experience many times in this diy project, and some parts of it are well-documented right here in this thread.
Last edited:
Sometimes, it's simply that things are beyond what we could reasonably predict. I'm sure Soren understands this well as an entrepreneur and designer.
——
Also, Soren it just occurred to me that you might feel like you’re getting undeserved criticism, and I now understand since you believe your project is very, very good, and it is. I have a strange personality I guess, so don’t think of me as a normal customer. But do note and I have said this many times in many discussions in this thread, that though I may be putting forth lots of ideas for improvement which could be interpreted as trying to point out flaws in your design, it is not. If you read my previous posts you would know that I like your design a lot in terms of its performance and engineering. In fact, your DAC is probably the only brand that I would care to buy from in today’s market. I can’t think of a higher compliment.
——
Also, Soren it just occurred to me that you might feel like you’re getting undeserved criticism, and I now understand since you believe your project is very, very good, and it is. I have a strange personality I guess, so don’t think of me as a normal customer. But do note and I have said this many times in many discussions in this thread, that though I may be putting forth lots of ideas for improvement which could be interpreted as trying to point out flaws in your design, it is not. If you read my previous posts you would know that I like your design a lot in terms of its performance and engineering. In fact, your DAC is probably the only brand that I would care to buy from in today’s market. I can’t think of a higher compliment.
Last edited:
It's been a while. Some speculations were clearly wrong in hindsight, but most were interesting though took up quite a bit of time. It's really both musing for dam1021 and reflection into myself.
It has proven to be even harder to generalize some things for others than I thought, and I did not ever take the challenge lightly. So take from it what you will. Perhaps it would be of interest one way or another, even if just simply for amusement
This is no smearing campaign. Occasionally accusations are leveled but I don't believe they are false. But judge for yourself on these issues - I guess that's what you're expected to do on public forums.
Btw, got a pair of HD800 and they sound wonderful with some gentle EQ correction for the factory FR curve and 6khz peak. Just sharing personal experience, not advice.
Actually, speculation about Rds of shift registers impacting R2R network is probably wrong (though the propagation delay between shift registers or even bits can still be impactful, theoretically). So if I said Soren erred by not adding mosfet switches to control the bits, I was wrong.
This is probably why I never wanted a thread - seems that I'd have to fight back on every unfair criticism to keep an "online reputation"...
Well, have fun if you must
Load my 44,1 Linear (F4) filter posted in Filter brewing for the Soekris R2R (#1633)
I and some other like it - maybe also you, ynmichael
//
I and some other like it - maybe also you, ynmichael
//
Load my 44,1 Linear (F4) filter posted in Filter brewing for the Soekris R2R (#1633)
I and some other like it - maybe also you, ynmichael
//
Interesting. I'm currently upsampling in Fb2k right after PEQ, but I'll follow its development. Is there any reason that it should sound any different/better than stock 4k linear?...
I'd still recommend that you try ABX to reject the null hypothesis. But it's starting to come back to me that Paul's filter sounded much sharper/harsher and had less bass than Soren's 4k filters, despite both being linear phase. So perhaps there is an audible difference (perhaps), even when the numbers all look good enough. Even if it's hard to completely rely on theory, for we don't know really well what matters to our ears and what doesn't, and even if it's hard to create testable conditions, for filter design often requires compromise and changing multiple variables at once, we should still be very wary of preferring changes (assuming that it's proven to be audible) that cover up or somehow compensate for defects in the rest of the signal chain rather than bring us closer to the elusive ideal of perfection in filter design. This is why I prefer theory when its available, especially in the subjective realm of audio... It doesn't mean that using listening tests to guide filter design is pointless - it's just going after a different thing and we should be careful about it. But you taught me all this
If we don't follow reliable metrics in our development, or theorize and discover new metrics, the usefulness of our efforts can be put under great doubt (this isn't to say that objectively better filters would definitely have great utility for everyone). For our judgement in audio based on listening tests is much coarser than we think, due to signal chain distortions/colorations and perhaps more significantly, colorations and distortions in the recordings that we listen to, even when we do hear a difference.
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.