1. This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
    By continuing to use this site, you are consenting to our use of cookies.

    Dismiss Notice

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

Discussion in 'Dedicated Source Components' started by purrin, Dec 5, 2013.
179 180 181 182 183 184 185 186 187 188
190 191 192 193 194 195 196 197 198 199
  1. prot
    I'm no EE or audio expert (IT background) but here's my interpretation of the wiki data on R2R (2kg of salt as prerequisite!).
    It's not hard to achieve 24 or any other bit depth with R2R. That only depends on the no of resistors and the circuit topology. Problem is the accuracy of those 24bits. IIUC, the resistor accuracy and matching are the main factors there and it mostly applies to the higher order bits. You get a 24bit result no matter what resistors you use but the last X bits may or may not be correctly interpreted.

    Question is, how do those errors affect the sound?. And that's even harder to say. It may generate a sound coloration that is actually pleasing ... some sort of tube-like coloration.
    The fact that R2Rs have a known problem with the highs rolloff seems to fit well with my "explanation". And generally (another 2kg of salt here) if you err on changing 1s into 0s your sound will become thinner. If you err the other way it *may* become fuller, louder, deeper.
    That's only *very* generic bits theory and logic, though. And I may be talking way above my pay grade :). Would be very helpful to hear something from ppl with hands on experience.
  2. purrin
    There's babbling around, but that doesn't change the FACT that the D-A used in the Yggdrasil will resolve the full 16-bits essentially without any error. Again, what part they are using is not exactly a secret now.
    I'm sure the TotalDuc sounds resolving to you, but another fact remains, you really haven't had any significant experience in terms of high-end DACs both new or vintage, unlike a lot of people who have heard the Yggy.
  3. purrin
    That's pretty much it. You can make a 64-bit discrete R2R DAC, but the errors will be so significant that you end up with much less effective number of bits. Even 13.5 or 14 bits is pushing it with discrete resistors because even power circuits become a factor.
    Smoothes things out. Makes things rounder sounding. Removes microdetails, ambient cues. People make all sorts of excuses: "Oh, the Metrum doesn't lack resolution, it just pushes the details into the background." Um, yeah right. Where did that string decay on that Eagles recording go?
    Based on listening tests on my gear, I can distinguish up to 18-bits. (Taking a good 24bit hires recording and converting down to 20,18,16,14 bits, etc.) The effects are subtle. I doubt most people without sufficiently resolving gear will be able to tell the difference with 12 bits. I cannot hear any sonic advantages past 48kHz or 44.1kHz sampling rate. 
    Argo Duck likes this.
  4. prot
    So, that would be another "24bit" component that can barely solve 16bit. Looks like those bits are quite relative in the audio world. And the 24bit sigmas are also max 22bit. And who knows how many other limiting components you have in the playback chain.
    Does anyone else feel like something stinks in audioland? I'm an IT guy. If my 24bit algorithm actually works on 16bit, I'll be fired the next day. Apparently not the same in audioland. You can just put a 24bit label on anything and noone cares. You even get enthusiastic reviews for it. It may not be as simple as that, but for me it still sounds like shady business and crookery with numbers.
    Oh well, as long as it all sounds "amazing" and "so much better than redbook", everyone is happy.
  5. arnaud Contributor
    To the point you can judge products without hearing them, TotalDuc being one, amazing! :)
  6. Articnoise

    If not a secret any longer that dac chip is used in the Yggdrasil?

  7. prot
    +1. would like to know that too
  8. Sorrodje
    Purrin, You've not heard the dac but you suggest it does not resolve more than 14 bits despite the SNR measurements (indeed provided by Brien himself) and you suggest the resolution is as deficient that an octave ( I confirm easily the octave lacks dynamics and resolution ) . Smells like a soviet lawsuit. guilty before judgement :wink:
    I have the Octave and the A1 Side by side. Will try to give impression about resolution. I'm as cheapstake as you so if the totaldac does not bring special to the table against the Metrum, it 'll send back the A1 and recover my money . it's as simple as that. And it would even not be a proof. Maybe my ears or my gear is too bad to hear enough resolution after all. then the Octave will largely suffice.
    I did not really understand the technical principles though. Forget me for that :wink:
  9. prot
    Thx for the confirmation. Wasnt sure that my pure math interpretation works as such

    Again from a pure math/bits background, that should be true if you simply remove the bits above 14 or so. If you keep em and ignore the errors the sound should be somewhat more detailed ... and possibly with more "weight". But since the errors are unpredictable (afaik), one cannot really say. As mentioned, you could very well end up with a sort of pleasung tubelike coloration. That would match the SQ impressions posted about the current R2R dacs.

    Tried 16 vs 14 bit once and I think I heard the diff (faint detail cues) but not sure I could pass a proper abx. And couldnt hear any diff above 44/48khz either. My system is however lowfi by your standards: modded mmax plus dac and sfaber speakers.
    Btw, I dont think my modded mmax fits into your "shrill sigma highs" category. Sounds quite smooth to me and very detailed with the sparko opamps. A lot of that smoothness may come from the sfabers but a mytek 192 was definitely more "strident" in the same setup.
  10. purrin
    Indeed, it is rather amazing! Yes, based on my own experience with so many DACs, examining their designs, I can almost get a sense of how something will sound.**
    I'm more than inclined to believe that the TOTALDuc will have great tonal, textural, liquid qualities with no delta-sigma treble hash. However, I have always, always, been suspicious of the resolving capabilities of R2R DACs. The fact is, delta-sigma designs do on average (for audio applications) have better accuracy than R2R designs, especially the discrete or magical black-box solutions. Cheap, small, and resolving were always the advantages of delta-sigma and why delta-sigma was permeated 99.999% DACs sold today.
    • On the current iteration of the Metrum DACs, which I and even some owners had indicated, are weak in the resolution department, uses an industrial DAC chip. It's been speculated what this chip is on other forums (Metrum scratches the markings out - how ghey). I looked up the spec sheets, and yikes - it's absolutely craptastic in terms of accuracy. (again, this might not be the right spec-sheet - only speculation.) Explains a lot and one of the reasons (there are obvious other reasons) why Metrum had to stack 4, 8, 16 of them to get good accuracy from them.*
    • I've enjoyed the MSB DACs tremendously. The Analog and Platinum (worse than the Analog in some respects) are craptastic in terms of resolution, but have great tonal, textural, liquid qualities with no delta-sigma treble hash. I almost, almost bought one, but I decided otherwise because I knew I wanted the resolution of the best delta-sigma DACs out there. Even the uber MSB Diamond... more resolving. It has the better black-box modules. Still not enough - not as good as the best delta-sigma DACs I've heard.
    • The old TDA1540 (14-bit with 4LSB) and TDA1541/2 (16-bit with 1-2LSB error - this spec is very very doubtful, I've heard that it's no better than 15-bits, and even then it's probably worse than that.) as used on the Valab and TeraDACs. They also suck in terms of resolution.
    • The most resolving R2R chip I've heard is probably the PCM63 followed by the PCM1702 and maybe the UltraAnalog Modules (two high accuracy 8-bit chips stacked with 4 resistor ladders). And even then, these R2R solutions need to be stacked (doubled up or more) to get close to the resolution of really good modern delta-sigma DACs. The Audio-GD M7 I had used 8x1704 chips - and even then, the imaging sucked because the digital filter implementation sucked.
    • I'm sure you've also heard of the Soekris R2R DAC board. My friend Misterrogers has been playing with it. Sounds great he says. I trust him. He's heard the Yggy. The Soekris R2R is not in the same league of the Yddrasil in terms of resolution.
    • Also, I've found that the practice of stacking chips (evens out errors) does increase resolution. I stacked 3x AKM4393s, and yes, that sounded more resolution than 1x AKM4393. DAC engineers don't stack chips for fun. There's a reason for it.
    Bottom line: While there isn't enough data to write a paper, I'm at least satisfied, at least for myself, that DAC accuracy does make a huge difference in sound quality.
    As far as crossing the TotalDuc off my short-list: I know what my sonic priorities are. When it's published in Six Loons (a webzine that says everything is great) that the TotalDuc has no better than 14-bits of resolution, I know I can cross that DAC off my list, especially when it costs 7000-10000 euros. **** that. I'm not saying the TotalDuc will not be a great DAC. I'm sure it is. I'm just not willing to pay so much for something that will be resolve the full 16-bits. Also, keep in mind that I'm running Yggy v0.99 through its paces trying to break it right now. Ho Lee *** - I would have never expected how much musical information is still contained on old CDs of questionable audio engineering quality from the 80s. I mean, why would I want to spend $10K and then down convert to 13/14-bits every CDs that I have?
    Another thing regarding the blathering from Schiit: I like to trust but verify. This is why I went out to get Theta Gen V (I suspected same / similar DSP) and audition a four dozen other DACs, a dozen of them in them in the mult-thousand dollar range. I wanted to find out if Mike Moffatt and Jason Stoddard were full of **** of the accuracy thing, missiles, close-form filters, some of their opinions of certain chips of yesteryear, etc.
    *On this thought, at least industrial, medical, military R2R chips do publish DNL/INL accuracy plots. They don't do that for audio R2R chips - which is kind of scary if you think about it.
    **"It's the implementation, not the D-A chip / circuit" - that's the biggest load of BS I've heard. Yes, implementation does matter, but D-A chips all do have a "sound" to them.
    kazsud likes this.
  11. Articnoise

    Interesting this that you say that more resolving = better accuracy.

    If that’s a fact how do the Yggdrasil compere to a good vinyl player like VPI or Linn LP 12 when looking strictly on level of accuracy?

  12. purrin

    Wrong measurement. I am talking about DAC accuracy / error. Read below.
    No Soviet lawsuit. Think of this this way: If I want a good handing car, and Car and Driver said a car weighed almost 1800kg, why would I consider buying that 1800kg car, especially if it's really expensive?
    The Man himself already said its 14 bits at best. I happen to want a 16-bit DAC. I don't want another TDA1540 from 1982. (I appreciate his honestly, but really he has no choice - no way going around laws of physics and manufacturing limitations of discrete resistors)
    It's the Man's opinion that 14 bits is all we need. I happen to disagree.
    LOL, not all Sabre sounds bad to me. LFF looked like he wanted to kill me or gave me this "are you serious" look when I made him listen after 6 seconds on the Mytek.
    The errors with R2R or true multi-bit DACs should be predictable in certain ways. There some variance because of temperature, but what happens is translation between the bitcodes and voltage output gets messed up. For example, here is a perfect DAC:
    0001 = 0.1 V
    0002 = 0.2 V
    0003 = 0.3 V
    0004 = 0.4 V
    0005 = 0.5 V
    0006 = 0.6 V
    In reality, because of resistor variances, we end up getting something like this (making up random numbers):
    0001 = 0.12 V
    0002 = 0.19 V
    0003 = 0.33 V
    0004 = 0.32 V
    0005 = 0.47 V
    0006 = 0.61 V
    The relation between bitcode and voltage output can be plotted as INL (do a Google search). It's measure of relative accuracy. You have also heard the term monotoncity - used by zdfx in his shilling Yggy post. (do Google search on that term too.) Fascinating stuff and a really unique angle to look at things. We usually hear a lot more focus with DSD, hires, et. al., on frequency / bandwidth / jitter on the X-axis but pay no attention to error on the Y-axis. Accuracy is like jitter on the Y-axis.
  13. Sorrodje
    what measurements is wrong and why ?  Just want to understand :wink:
    How to understand this answer ? (quoted from http://www.head-fi.org/t/693798/thoughts-on-a-bunch-of-dacs-and-why-i-hate-chocolate-ice-cream/2820#post_11384596)
  14. purrin
    We are talking apples and oranges. I gave hints in my prior post what I was talking about. Will explain later - meeting at work - unless someone wants to chime in first.
  15. Sorrodje
    Sorry . My english and my technical skills are maybe too weak.
179 180 181 182 183 184 185 186 187 188
190 191 192 193 194 195 196 197 198 199

Share This Page