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.
419 420 421 422 423 424 425 426 427 428
  1. castleofargh Contributor
    we're all similar people presented with the same options. we feel something and get some idea as to why. then from time to time, measurements seem to conflict with those impressions and ideas. and that's where we're not so similar anymore:
    -when I'm presented with objective data contradicting my ideas and memories, if I can test my idea with some control, I'll do it and see what happens. when I can't, I consider the conflict to be big enough of a problem not to decide that my feelings and ideas are conclusive. and if I absolutely have to pick a side, I'll side with the objective data(but that's really something I'd rather avoid).
    -based on the last few posts, it's pretty obvious that not only you take your feelings and ideas very seriously, but when they conflict with measurements, you decide to suspect the measurements of being flawed or incomplete instead of second guessing yourself and the quality of your experiences.
    you bring up some fairly reasonable questions about measurements. we obviously don't get enough specs by default. and with the natural tendency from manufacturers to "forget" the variables showing bad specs for their gears, having apparently good specs are even less of a proof that we'll get transparency. but focusing on that alone is pure fallacy. we obviously also have to deal with our testing conditions and all the possible issues we can just call "human error". from the sighted anecdotal experiences, the accuracy of our memories, the quality of interpretation, jumping to conclusion, etc. all very real very relevant concerns. trying to find which set of measurements is effectively presenting what we feel, that should obviously come after making sure that what we feel was induced by sound in the first place. something the average audiophile is not going to do, because of ignorance, laziness. or the usual idea that preconceptions, placebo, logical fallacy, etc, all somehow belong in the box labelled "it only happens to others".

    I've followed discussions about the sound of delta sigma vs R2R, the specific sound of some chipsets and related stuff. I'm personally very curious about this, I've tried to test my fair share of gears and conditions, but right now I have no confidence claiming that anything is real or more than circumstantial. I've very clearly failed to pass blind tests using DACs with different chipsets. I can blame my low listening skills and my ears not growing younger, but whatever the reason, those DACs were still good enough to fool me.
    I also have been able on rare occasions, to identify some DACs despite them having the same chipset.
    the usual idea that R2R will have less linearity and more aliasing or treble roll off, while delta sigma has more noise, even such generalities based on the designs can be contested if we go pick the right DACs. some R2R stuff have impressive linearity, some delta sigma have impressive SNR.
    so all in all, I'm confident that at a statistical level we can find patterns and correlations, but who has gone through enough gears in a rigorous enough way to call his results statistically significant? I sure didn't.
    GrussGott and bfreedma like this.
  2. KeithEmo
    I can think of a few reasons....

    1) Many DACs, especially more esoteric designs, employ additional upsampling in addition to the oversampling inside the DAC. In fact, many tout it as a feature.... Note how many DACs "upsample to 384k" or "upsample to DSD" using some proprietary firmware or FPGA before sending the data to the DAC, usually claiming some sort of audible improvement as a result...and some customers believe those claims (I'm not saying either way). However, you could take it as a simple statement of fact - that they do NOT do any of that extra sample manipulation.

    2) Being more cynical, you might suggest that enough people seem to believe that "not oversampling" is a virtue that they feel they'll sell a few more DACs by encouraging the misunderstanding. (Non-oversampling DACs share several drawbacks, which tend to make their technical performance look relatively poor... competing against them makes the Antelope DAC look better by comparison.)

    intlsubband likes this.
  3. KeithEmo
    I'm not an expert on the math of designing filters, but I'm pretty sure that, while the overall response is very important, the design of the individual stages doesn't count for much. Therefore, if they're using several stages, it would only be the aggregate response of all of them that matters. Therefore, the design of a single stage could seem poor, but a cascade of several such "poor stages" could actually produce an excellent aggregate response. of course, it also may not. Most DAC chips offer a variety of oversampling options to the designer, many of which may be better or worse in various different ways... and, presumably, some of the choices and combinations simply don't perform very well in important ways.

    Sabre DACs are also interesting in other ways. For example, you may be familiar with ASRCs (asynchronous sample rate converter). Many DACs incorporate an ASRC to eliminate jitter (I believe Benchmark was the first). To horribly oversimplify how they work, an ASRC converts the incoming digital audio data to a new sample rate. In the process, the data is re-clocked to a new local clock, so any jitter associated with the incoming clock is eliminated, or drastically reduced. Some people have an aesthetic objection to the fact that "every single sample of the original data is discarded and replaced with new interpolated data". However, from a technical perspective, all that counts is that the analog data that comes out the end is improved because the effects of the incoming jitter are removed. In most DACs that have this feature, it is implemented using a dedicated "ASRC chip" - often an Analog Devices AD1896, a TI SRC 1492, or the equivalent from Cirrus Logic. Sabre DACs have a similar jitter reduction method built in.... it actually introduces corrections into the data as part of the upsampling process. Again, horribly oversimplified, jitter causes problems because you end up with the correct value being submitted to the converter stage at a slightly wrong time, resulting in a wrong output value. A standard ASRC addresses this by correcting the timing. Sabre's method, according to their conceptual description, essentially calculates the error, then adjusts the value of the data sample itself to "compensate" for the error introduced by the faulty timing. I have no particular opinion about which method is "better", but both incorporate complex DSP calculations to make corrections to compensate for errors in the data - so both offer the opportunity for the "corrections" to introduce changes that might be audible...

    Some DACs that use a separate ASRC allow you to bypass it so to evaluate any audible changes it might introduce for yourself. As usual, when it comes to complex digital filters, the actual changes such devices cause tend to be complex and somewhat difficult to describe. For example, several of our Emotiva DACs have included an ASRC, with the option to disable it, and it most definitely produces a subtle alteration of the sound. If you start out with a source that is extremely high in jitter, then the difference is usually what I would consider to be a small but distinct improvement. However, if you start out with a signal that has relatively low jitter, there is still a slight audible difference in sound - but it becomes more difficult to describe, or to identify as "better or worse". As a broad generality, many digital filters produce subtle differences which can be measured using various metrics. Often, however, it becomes a matter of personal taste which one is "better" - and whether a given person notices one or the other at all.

    I should also note that there's a middle ground. Most commercial DAC chips offer the designer the option of using the internal filters, often a choice of several different ones, or of adding their own external filter altogether... and many DAC vendors do just that. However, actually designing digital filters is very complex. Many DAC vendors have their own filter designs which they insist are better than the included ones. Interestingly, if you read a lot of reviews in the sorts of magazines that do serious math, you sometimes find that a certain DAC has "a unique sound" because the filters it incorporates are simply badly designed, or designed to perform very well in certain specific ways at the expense of other important parameters, and the designs of one or two of them have simply been found to be based on faulty math. Companies like Analog Devices and AKM spend a lot of money optimizing the performance of their products; some of us think they know what they're doing, and see little reason to "reinvent the wheel" when the options they provide work really well. (In terms of deliberately "boutique sound", many non-oversampling DACs incorporate gentle roll off filters, which have a very clean ringing characteristic, but also a very slow high frequency roll off. You end up with a DAC that has very little ringing, but a frequency response that is -3 dB at 20 kHz with a 44k input signal. As a result, plucked strings sound clean and sharp, but it becomes difficult to tell if the "smooth sound" you hear is really due to the lack of ringing - or simply to the rolled off high end. And, yes, if your speakers had a rising high end, the two flaws might cancel out perfectly... producing some of that legendary "synergy" people talk about. )

    JayNYC and Curtis R. Emma like this.
419 420 421 422 423 424 425 426 427 428

Share This Page