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.
398 399 400 401 402 403 404 405 406 407
409 410 411 412 413 414 415 416 417 418
  1. KeithEmo
    You said it - but I won't disagree with you there [​IMG]
    However, you seem to be under a slight misconception about digital filters. It's not that "the D-S process produces digital noise - which must then be filtered out". All digital files consist of the actual information you want, mixed in with extra "stuff" (the energy content of the "steps" that occur at the sample rate). That extra stuff MUST be removed somehow, which is only right, since you only get back the original signal after it is removed. (Think of it like making an exact copy of a statue out of stone. You rough it out, then you sand it, and finally you polish it. And, in the end, it hopefully ends up being just like the original. The process isn't "bad" because sanding is one of the steps, and the statue looks bad until you sand it.) So, both R2R DACs and D-S DACs, and both NOS and oversampling DACs, end up with a signal which must be processed through a filter to remove the "excess junk" and get back the signal you want. Oversampling basically alters the characteristics of the noise that is present in such a way that it is easier to filter out. (And the "accusation" is that, even though the result is easier to filter out, and so there is less of it left after it's filtered, it's so much nastier than the original that even the tiny amount that survives is more audible than the original noise.)
    The real debate, or claim, or what you will, is that the specific types of noise used/produced by the D-S process is either somehow "inherently worse" than other types, or that it's more of a problem because it isn't being properly removed, and so there's more of it left to be heard, or both. I see people deeply concerned that the noise floor on a specific D-S DAC might be modulated, which could be unpleasantly audible... yet they forget that the noise floor they're worrying about is at -125 dB, which is about 30 dB BELOW the noise floor on a CD. 
    The problem I have with many of these discussions is that they start out with a technical argument, but end up with "I know the problem is present, therefore it MUST sound bad" - which is simply faulty logic. There are many reasons why two DACs might sound different, and whether each happens to be R2R or D-S is just one of them. I would very much like to see someone conduct an actual study to determine if most people can actually hear the difference between R2R DACs and D-S DACs. Unfortunately, in order to do such a test, they would have to eliminate all the OTHER variables, of which there are a lot - many of which can be difficult to eliminate. (You must first find an R2R DAC which can match the noise, frequency response, and distortion performance of the D-S DAC you want to compare it to... which may be a difficult and expensive proposition. Otherwise, we're left wondering if you're simply interpreting the flaws as virtues. An R2R NOS DAC that rolls off the high-frequency response by 2 dB at 20 kHz is going to sound so obviously different because of that roll off that you aren't going to hear any more subtle differences that might be present - nor will you be able to confirm that it IS otherwise identical.)
  2. KeithEmo
    My reference to cost cutting was based on my interpretation of your wording - and that it seemed to me to suggest that you were using "cost cutting" as a disparaging term. Personally, while I find that many products do indeed suffer a loss in performance because "costs have been cut", the reverse is often true - that new technology sometimes allows us to achieve better performance at lower cost. Also, in many cases, there is a tradeoff - but not always in performance. (I have a little digital pocket recorder that cost me $200. Aesthetically, it is a "cheap little plastic piece of junk"; yet, in terms of actual audio performance, it easily outperforms the Ampex "studio" open reel tape recorders a friend of mine used to collect - even though they cost ten times as much, weigh about fifty times as much, and are clearly "better made".)
    My point is that the fact that they're cheaper IN NO WAY suggests that they don't work as well - or that they do. 
    To take the discussion back to technical "ground".... I'm a bit curious about precisely what you think a D-S DAC would be unable to do satisfactorily. (You seem quite certain that there is something inherently wrong with them. I'm cusious how you think they "mess up the signal".) Please note that I make no claim that D-S DACs are perfect... I'm pretty sure I know which things they do well, and which they do less than perfectly, and I'm curious which of their flaws you believe "must" still be present in current versions to the degree that they overshadow the known flaws of other types.
    (Incidentally, I do agree with you that many early DACs, of both types, sounded pretty bad. I just don't see a specific correlation between the badness of the sound and the topology - beyond that fact that "most cheap DACs do use D-S chips these days". I could also say that most $2 bottles of wine taste much worse than most $2 bottles of beer; but that in no way suggests a generalization that "beer tastes better than wine".) 
    Argo Duck and richard51 like this.
  3. BassDigger
    You seem to be mis-quoting me, again.
    I'll leave the technical stuff for you; I just know what I like the sound of.

    I guess wine is more expensive to produce (and it's sold in bigger bottles). :wink:
  4. evillamer
    The ess sabre white paper has highlighted transient issues with some sigma delta dacs. Although this may seem like one of the only few problems with sigma delta dacs but we humans(survival) are more sensitive to fast(or loud) transient sounds than other form of sound waves. We are probably less sensitive to the frequency drop from NOS than we are to detecting issues with the transient sound and jitter.

    Even though the Ess claims that their sabre dac can process transients without a hitch, the transient response of the sabre dac is still behind well implemented R2R dacs in certain music instruments like cymbals/shakers. As in the ess sabre dac is not free from anomaly as they claim. This is something which some might notice and some might not, depending on the ear of the beholder and the equipment in the chain. Some might also find that the ess sabre tends to recreate transients way too fast than what is considered as normal/natural(as compared to other sigma delta or r2r dacs)


    Page 3:

  5. ciphercomplete
    I hate NOS dacs personally so I wouldn't be on my 4th (5th? I can't remember) R2R dac if any of them sounded like the few NOS dacs I have heard.  Maybe try a standard or oversampling R2R dac instead.
  6. KeithEmo
    I do apologize if you think I've mis-quoted you..... here is the actual quote from your post:
    "Although I have no idea how a good modern Delta-Sigma implementation compares (I lost interest in DS years ago), historically, there seems to have been a definite jump backwards (in terms of SQ), and it has all the hallmarks of cost-cutting!"
    To me it seems pretty clear that you are saying...
    1) That you haven't heard a good recent Delta-Sigma DAC
    2) That you are certain that the sound quality in recent years has gone downhill - and that this seems to be the result of "cost cutting"
    Do let me know if I'm missing something......
    No need to fuss about it - but I do my best to get details right (although I'm only human) [​IMG]
  7. KeithEmo
    However, it's difficult to determine which part of the process to "blame" - and how much credit or blame each deserves. All digital oversampling filters introduce some ringing (which is an error in transient response). And this same type of error will also be exhibited by an R2R DAC that uses an oversampling filter. Likewise, ALL DACs require a reconstruction filter, which will also introduce errors. However, in terms of sound quality, these are all qualitative claims. (We can see that errors exist, and we can measure them, but we can't correlate them directly with sound quality. For example, there's no way to quantify whether an extra 3 mS of ringing at -10 dB will sound worse than an extra 0.7% THD, or a 5 dB higher noise floor.)
    To me, here are a few of those questions I'd like to see "worked on":
    1) Assuming we are using oversampling, and so a digital filter, which introduces errors in transient response, do the oversampling filters used in D-S DACs introduce WORSE errors than those introduced by the filter itself. (In other words, if I use the same amount of oversampling for an R2R DAC and a D-S DAC, will the D-S DAC have WORSE errors, or THE SAME errors as the R2R DAC, or equally significant but audibly DIFFERENT errors?)
    2) I agree that Sabre DACs tend to emphasize transients (according to what I hear, when comparing them to other DACs). However, it would be nice to know if the Sabre DACs are OVER-emphasizing transients, or the others are UNDER-emphasizing them. Either they're right and everybody else is wrong, or everybody else is right and they're wrong, or they and everybody else are all wrong - just in different ways. (In some of their early marketing literature, Sabre actually said that they "tuned" their sound "based on what users liked in focus groups". This doesn't actually suggest whether the choice they went with was the most accurate - or not.)
    3) How do the errors DACs introduce into transients compare to the errors present in alternatives? Neither microphones, nor phono cartridges, nor speakers, have especially clean transient responses, and even A/D converters introduce transient errors. Therefore, no digital recording is "perfect" as a source, but the vinyl or tape "originals" that many people compare DACs to are also flawed. In fact, since most concerts these days use at least some "sound reinforcement", it would be somewhat difficult to find a "purely natural" original to compare to a reproduction. (Saying that "DAC X sounds more like vinyl" could mean that both share some particular good trait, or it could simply mean that the DAC has the same distortions and colorations as a typical vinyl reproduction. Without knowing which is true, unless your actual goal is "to mimic vinyl exactly", we can't know for sure if that claim is a good thing or a bad thing.) 
    There's also another facet of "digital processing" that most people seem not to understand - and that is about how "arbitrary processing" interacts with "perception".
    To take my favorite example from optics. All digital photographs have a certain unavoidable loss of detail. (If you take a perfectly sharp picture of a black and white checkerboard with a digital camera, the resulting image will contain some grey pixels, because some sensor sites will be "straddling" a boundary, and so "see" half white and half black, which will cause them to return the value for grey.) There is a process called "unsharp mask" which has been used for years to create the ILLUSION of improved sharpness in images. The process actually exaggerates the contrast at contrast boundaries - so, if you have an area of light pixels adjacent to an area of dark pixels, the process creates a very light halo at the edge of the light area and a very dark halo at the edge of the dark area. This alteration then tricks our eyes (or our brain) to "see" a sharper picture by exaggerating the edges that were DE-exaggerated by the photographic process. The net result, if done properly, is that the error introduced by the "sharpening process" "cancels out" the flaw in the digital photography process, with the net result that the picture looks more natural.
    The implications of this are interesting:
    1) ALL digital photographs are inaccurate
    2) Applying unsharp masking will make most of them even less accurate (by measurements)
    3) However, applying unsharp masking to many of them will make them APPEAR MORE ACCURATE TO A HUMAN OBSERVER
    3) is actually an understatement - because the "unsharp setting" that will make a given picture appear best to a certain observer will be different depending on the size of the picture, the lighting conditions, the content of the picture itself, and the sharpness of vision of the observer. Therefore, the setting must be chosen based on the intended use of the picture, and the size at which it is intended to be displayed. (There are high-end image processing programs which you can tell how big you plan to print the picture, and in what format you plan to print them, and which will then choose the best settings for you - based on studies of human vision and perception.)
    I quite suspect that what Sabre DACs do is the psycho-acoustic equivalent of this process.... they deliberately exaggerate transients slightly, which, in most cases, tends to cancel out an unintentional softening of transients that occurs elsewhere (perhaps due to unavoidable consequences of digital filtering - or perhaps not). The result, much like the result you get with unsharp mask, is that many sources are improved, while a few come out appearing exaggerated or unnatural. (I am using the term "arbitrary processing" to describe a situation where an error is assumed, and a change made that would correct that error, without any way of knowing if the error actually existed in a specific case to begin with or not.)
    Argo Duck likes this.
  8. KeithEmo
    I agree with you there.
    Unfortunately, there aren't a lot of them out there - and some of the lower cost ones are suspect for various reasons (a poorly designed R2R DAC is still not going to sound good - and a lot of the ones available at reasonable prices seem to be DIY projects offered by sources with dubious design credentials).
    The Schiit Yggdrasil is an oversampling R2R DAC - with an excellent provenance. [​IMG]
    (Although, assuming you really like the way it sounds, you'll still have to figure out how much of the credit goes to R2R and how much to simple "good design".)
  9. hans030390
    I'd say you'd need to compare them in a quiet, familiar environment, such as your own home. Do so with very resolving amps and headphones, the best you can find. The Hex is smooth to the point it will literally gloss over low-level details. It should not be even remotely subtle if you know what to look for. If the Mirus does the same, well, that would be interesting.
  10. skeptic
    Lots of studios make badly recorded classical and jazz music.  The more interesting question is what do/did the very best studios, e.g. MFSL, use while making great recordings like the gold discs of Muddy Waters and Louis and Ella, that sell for a mint?  Answer - custom gear built by Nelson Pass and Mike Moffat that definitely was not DS based.  http://www.vxm.com/21R.46.html
  11. evillamer
    Not sure if ESS has done right with their transient response tweaks but it seems that their transient response can sound exciting or fatiguing for long listening. It really depends on the equipment matching as well. Not sure if it's due to noise shaping/digital filters or something, I myself find that the soundstaging on the various sabres(no special dsp/filters) I listen to, seems to be more closed in(narrow) as compared to the PCM1704uk.
    I think we need some form of recording compensation caused by the distortion/transfer lose from recording equipment. I think that's where the Mastering Engineers comes in. These people would have already perform some of EQ/phase/attack/level adjustments at their end.
    As for the loss/distortion caused by dac/filters/amps/speakers/headphones in the chain, This is something only a pair of good ears(highly perceptive listener) can tell.
    This is a high technical video(not for everyone), watch from 27:30 where the lecturer explains what happens to the quantizer(within the sigma delta) when the input exceed a certain threshold causing the noise shaping loop to fall apart(sigma delta instability). He also explain the input cannot exceed the range of the quantizer(Maximum Stable Amplitude).
    I am guessing this is an area where instrumentation measurements with FFT charts, Simple Sine Wave Amplitude Chart or Frequency Response measurement charts is not going to show the issue with sigma delta stability as it's a continuous input/non-linear function issue.
    Also on another note.
    IME: Jitter plays a very important role when it comes to reproducing smooth silky sounding female vocals. Doesn't matter if your dac is R2R or Sigma Delta, as long as you have bad jitter, your female vocals will sound coarse. e.g. try a high jitter source apple tv/pc soundcard digital out to your dac.
  12. BassDigger
    Why are you apologising for what I think? You either mis-quoted me, or you didn't. (Actually, I think that there's as much mis-understanding and 'putting words in my mouth', as there is mis-quoting.) [​IMG]
    I've explained my position, and why I think what I do. (You haven't asked about the differences that I heard.)
    Although, we've previously discussed the importance of timing and jitter, unfortunately this is all beyond my comprehension. But, (Time for me to mis-quote you) you seem to be asking for technical explanations and implying that the measurements prove everything [​IMG]. (Although to be fair to you, you did express a desire to understand what might be going on.) I don't believe that we can measure and understand everything, and that there're are misunderstood phenomenon, creating misunderstood effects. I think that D-S vs R-2R is a prime example, of this.
    Regarding whether I've listened to the latest and greatest D-S: as I've said, I've explained my position. Also, R-2R has advanced; many of the early designs had some real 'howlers' of mistakes, in their implementation. But they still sound better than the later D-S designs. Are you implying that D-S has improved so vastly (and so much more than R-2R), over the last decade?
    I find it concerning that you don't seem to accept (or admit) the concept that companies will use any (and all) methods they can to increase their profit margins.
    D-S is much cheaper. That's why it's so ubiquitous. R-2R is considerably more costly. This cuts into precious profit margins; D-S can be made to sound 'acceptable'; marketing and the 'assistance' of the hifi media maintains sales, particularly when they concentrate on what D-S can do, with some degree of success (I believe that this is detail retrieval).
    The rest, as they say, is history.
    My 'underwear has remained untangled', but there's only so much 'going around in circles' that can remain interesting. [​IMG] 
  13. drez
    Yes DS has improved greatly over the last decade. OTOH the only R2R DAC I am interested in is the Yggdrasil. To be honest cost cutting and easy implementation is perfectly valid reason to use DS. Not every company is as crazy as Schiit.
  14. frenchbat
  15. BassDigger
    Cost cutting? Sure, if you want a cheap dac.
    But if you want a good one, a reasonable R-2R implementation will do the job (reproduce music) better every time, no matter how much D-S is passed off as real hifi (and yet can still cost megabucks).
    It hasn't been the consumers' costs that anyone's interested in cutting.
398 399 400 401 402 403 404 405 406 407
409 410 411 412 413 414 415 416 417 418

Share This Page