Delta Sigma: An Inconvenient Truth

Discussion in 'Digital: DACs, USB converters, decrapifiers' started by k4rstar, Jun 21, 2020.

  1. crenca

    crenca Friend

    Friend
    Joined:
    May 26, 2017
    Likes Received:
    550
    Dislikes Received:
    5
    Trophy Points:
    93
    Location:
    Southern New Mexico
    You guys might very well be right, in that an high(er) level understanding of digital audio might not be an antidote to the spaghetti of analogy that I quoted from @k4rstar (no offense intended @k4rstar), the common audiophile beliefs and the marketing machine that leverages them, and/or even if it is, it does not matter to the more important scheme of cloud sourcing impressions and correlating them with known touchpoints...but darn it, I can dream! :rolleyes:
     
    monacelli likes this.
  2. GoodEnoughGear

    GoodEnoughGear Evil Dr. Shultz‎

    Friend
    Joined:
    Oct 25, 2015
    Likes Received:
    2,272
    Dislikes Received:
    11
    Trophy Points:
    113
    Location:
    Cape Town, South Africa
    Nice - please share that sheet if you can, already workday here but I want to futz with this later.
     
  3. k4rstar

    k4rstar Done his time

    Friend BWC
    Joined:
    Jun 11, 2016
    Likes Received:
    3,972
    Dislikes Received:
    88
    Trophy Points:
    93
    Location:
    Toronto, Canada
    Home Page:
    @crenca no offense taken, i'm glad that @purr1n and some of you other guys showed up to save the thread from its dumpster fire and also constructively bodychecking me

    I think it's not a secret at this point I'm a hard-line subjectivist and don't put much stock in measurements or audio science. I realize that is not in line with the moderate ethos of this forum and I respect that - the 'too objectivist for subjectivists' thing really applies to me :)

    to @monacelli's point I will provide some sort of retrospective impressions of the DACs I have owned (too many to remember at this point) when I have a little time, for some impressionist Yang to the theoretical Yin
     
    crenca, monacelli, je2a3 and 3 others like this.
  4. purr1n

    purr1n Finding his inner redneck

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    58,795
    Dislikes Received:
    100
    Trophy Points:
    113
    Location:
    Antarctica
    I've got nothing against being purely subjective and as far as DACs go, it's all purely subjective. The measurements are so good that there's practically zero correlation of measurements with sound quality, and zero correlation of architecture or processing to personal preference. I haven't found a single measurement that will tell me whether I will like or dislike a DAC unless the measurements were so shitty that they were on transducer level shitty. I may see certain signs of NOS in frequency response, but this could be attributed to a gentler filter on OS, and even then I may not even hear it as being rolled (in the case of the Holo Spring for example).

    My acceptance of DAC evaluations being purely subjective does not mean that I will not address common misconceptions from NOS proponents that NOS is the most pure and that OS is somehow based on bad math because NOS happens to be what they prefer. I've seen these arguments for way way too long and I thought this was a opportunity to clarify this stuff.

    @Hands loves the NOS stuff, but he's not going to bring up how Sony got a hard-on for OS in the early 80s, 1-bit DSD in the late 80s, or Wadia and Theta did their own filters in the early 90s, and how all this was the wrong path because dithering, approximations, noise shaping, all sucked the life out of music or introduced effects which were parlor tricks that one got bored with in three months time.

    All this shit, the math, has been around with PCM / R2R since the early days, especially on the ADC and maybe mastering side. It's been around even before digital because the technology for digital hadn't been invented yet.
     
    Last edited: Jun 30, 2020 at 7:08 AM
    Biodegraded, crenca, Luckbad and 9 others like this.
  5. Psalmanazar

    Psalmanazar Most improved member; A+

    Slaytanic Cliff Clavin
    Joined:
    Sep 27, 2015
    Likes Received:
    3,951
    Dislikes Received:
    121
    Trophy Points:
    113
    DSD is 1-bit at 2822.4 khz. Make it look like a vaguely normal sine wave. The analog filter looks simple but now try to dither it and keep the distortion down. Sony could never do it.
     
    dark_energy likes this.
  6. Thad E Ginathom

    Thad E Ginathom Friend

    Friend
    Joined:
    Sep 27, 2015
    Likes Received:
    7,456
    Dislikes Received:
    11
    Trophy Points:
    113
    Location:
    India
    According to my math-limited understanding, the term High-Resolution was taken from imaging and stuck onto audio, where it has about as much meaning as "digital" on the packing of a set of cheap headphones. But the marketing people won, and it stuck. Which doesn't give it any more meaning
    I never took a calculus class at all, and still don't really know what it is. I used to wish I could understand this stuff so that I could argue with audiophools authoritively, but that's a fool's errand, because I have seen actual, genuine digital authorities fail to do so. The digital audiophools have twisted their reality to the extent that many believe that Nyquist-Shannon is an attempt to explain their mysical-unicorn music, rather than being the foundation it is built on. No point in even mentioning dates to them, let alone anything more advanced.

    The lesson is: fuck audiophools. If one wants to understand the magic, and has the mental equipment to do so, then, no doubt it is a worthwhile pursuit. I may not even remember correctly now, but I think I got as far as getting a faint shadow of a clue as to why this sync thing means that the output of digital audio is not the stair-step picture beloved of pono purveyors. (And fuck Neil Young, too, notwithstanding some beautiful music way back then)

    Hats off to all of you that can get your heads around this math, science, technology and its implementation: I'll watch admiringly from afar

    :bow:
     
  7. purr1n

    purr1n Finding his inner redneck

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    58,795
    Dislikes Received:
    100
    Trophy Points:
    113
    Location:
    Antarctica
    Try n=-10 to 10. It's iterative, the more we do, the better the reconstruction. You are now ready to teach. Showing the original dots, the samples, better illustrates the magic behind the math.

    Yeah, we definitely don't want to connect the dots, but that's what happens without a proper reconstruction filter.
     
  8. spwath

    spwath Collegiate hijinks master

    Friend BWC
    Joined:
    Dec 13, 2015
    Likes Received:
    5,517
    Dislikes Received:
    27
    Trophy Points:
    113
    Location:
    Hartford, CT
    Hey, I love Excel. And have been away from college all summer. So if someone gives me an Excel homework assignment, I'll do it.

    One I was a bit confused about in the interpolative method, what exactly is the set x[n]? I was trying to figure that out but couldn't get it. Once I figure that out I'll graph using that method.
     
  9. wormcycle

    wormcycle Friend

    Friend
    Joined:
    Aug 13, 2016
    Likes Received:
    811
    Dislikes Received:
    25
    Trophy Points:
    93
    Location:
    Toronto, ON, Canada
    That's great example how easy is to cross from subjective to incomprehensible, if the term subjective becomes so loose that simply stops making sense.
    I thought in from the subjective point of view what you hear would decide whether the things like filtering/up-sampling/oversampling are of benefit or not.
    Now you are saying that you a priori do not believe that they are beneficial, therefore you reject any design that uses them and therefore what...
    How do you reconcile it with:
     
    Raimei Templar and yotacowboy like this.
  10. k4rstar

    k4rstar Done his time

    Friend BWC
    Joined:
    Jun 11, 2016
    Likes Received:
    3,972
    Dislikes Received:
    88
    Trophy Points:
    93
    Location:
    Toronto, Canada
    Home Page:
    no not exactly a priori. I meant I reject them because of my listening experiences. any additional theorizing, even if it sounds like audiophool bullshit to Marv, is just my attempt to attach some sort of explanation to an otherwise esoteric phenomenon (why does NOS sound better to me even though it has as many if not more issues with the science behind it?) without using established audio science as reasoning because it obviously does not support the phenomenon, other than to say I am deaf and like high-frequency roll-off and phase shift.

    I understand if it comes across as technobabble. I concede Marv's approach is easier for most to accept and get behind.
     
  11. monacelli

    monacelli Friend

    Friend
    Joined:
    Jul 23, 2017
    Likes Received:
    1,044
    Dislikes Received:
    0
    Trophy Points:
    93
    Location:
    The Land of Enchantment
    x[n] is the traditional notation used for the discrete-time signal, i.e., the "dots". By contrast x(t) is the continuous time signal, which you were approximating by using ever-increasing sampling rates (more and more dots). So when you use Excel to generate samples of a sine wave, say in Column B, each entry of the column corresponds to x[n] for a particular n. So x[0] is in the first row, x[1] in the second row, and so on. The Whittaker-Shannon interpolation formula Marv linked to is a particular case of a mathematical operation called convolution. It turns out that the inverse Fourier transform of an ideal reconstruction filter is a sinc function (of infinite length), which helps to explain where the formula comes from.
     
    spwath likes this.
  12. Hands

    Hands Overzealous Auto Flusher - Measurbator

    Staff Member Friend MZR
    Joined:
    Sep 27, 2015
    Likes Received:
    10,450
    Dislikes Received:
    39
    Trophy Points:
    113
    Location:
    Colorado
    Home Page:
    I do more listening on OS DACs these days than I do NOS, though I am selective either way. Depends on my mood, what music I'm listening to, what other gear I'm using, etc. I think the combination of aging a few years and refining my gear made the difference in me no longer being tied at the hip to NOS.

    If given the choice between NOS and varying levels of OS, I'm most likely to go for something like 2X or 4X OS. Either that or some form of slow roll-off filter. (It's not always clear if the slow roll-off filters in various products oversample at a lower rate or not.)

    But even on something like the previous gen Bifrost Multibit, assuming one of its last iterations before the BF2 launch, I have no issues with what seems to be an 8X OS filter. Granted, I think the stock I/V opamp sounds like poop, but the DAC being OS in and of itself is totally OK with me. Usually I prefer it over NOS.

    Still, there's no real consistency. I found the Matrix XSP to be somewhat veiled and dull no matter what I did. Meanwhile, an old school, Audio-GD, WM8741 DAC sounded clearer, more lively, and overall just better (more like R2R, TBH) even on a lower OS rate setting.

    And I'm testing out an AK4137 SRC board with my DIY NOS DAC. Even at a lowly 2X OS rate, which usually doesn't sound a whole lot different than NOS, I'm leaning towards thinking it makes the DAC sound like insufferable ass. Might be some other technical issues at hand here. Seems to measure OK, at least. (Though I think the AK4137 was used in the Holo Spring and also sounded bad...hmm.)

    All this just to say:

    1. I'm not a good candidate these days for a NOS poster boy.
    2. Nothing seems to line up consistently between what should sound good, in theory to me, and what actually sounds good in practice.


    Regarding NOS frequency response and filters, do you mean analog filters on the output stage or something else? I thought the NOS droop was inherent to NOS itself even if there's nothing else, filter-wise, to affect the audible response range.
     
  13. yotacowboy

    yotacowboy McRibs Kind of Guy

    Friend
    Joined:
    Feb 23, 2016
    Likes Received:
    3,285
    Dislikes Received:
    11
    Trophy Points:
    93
    Location:
    NOVA
    Home Page:
    OK, I'm going to preface this with a note that my response doesn't intend to get down into the fussy philosophical side of things, but I just want to expose a point that's made often in a post-Wittgenstein world.

    Bridging the gap between nonsense and a new sense of something is actually important. Or, as you said, paraphrasing, from subjective to incomprehensible, albeit, the other way around.

    Perhaps think of it like this [oversimplified]: k4rstar has a set of experiences that could be mostly ineffable for him. NOS DACs might have a quality that's simply ineffable, but tangible. And when he tries to describe that mostly ineffable quality, he's reduced to language and logic that don't "fit" the quality (technically, qualia). So how do you explain to others something that's not at all quantifiable, and barely qualifiable to someone, anyone?

    And when he tries to use the logic that suits other, more effable or logically tautological experiences in order to gain meaning, it might come across as proposing something logically paradoxical, or linguistically incomprehensible.

    I know, I know, this is splitting the goofiest of hairs, but I do think it's something that's actually happening ITT.

    (which is kind of neat to witness an actual "beetle in a box" discussion actually happening to a Witt "fan". JFK i'm a nerd.)
     
  14. cameng318

    cameng318 Rando

    Joined:
    Jun 14, 2020
    Likes Received:
    12
    Dislikes Received:
    0
    Trophy Points:
    3
    Location:
    U.S.
    Got the homework done. It's kind of unfair since I just got out of college a month ago. However, I doubt any of my classmates can figure this out in 2 hours.

    I used sinc filter from -5 to 5 to OS it to 176.4 kHz. For the truncation I used 4 bits, that the effect is dramatic enough. Actually I messed up the levels a bit, 4 bits gave me 17 levels, and dithering/noise shaping made a few more levels, but I'm not worrying about overflowing over 1 in Excel. Also I didn't check the coefficients for the IIR filters, but I guess they are near the ballpark. Getting 80% score on a homework is good enough for me.

    Sheet and result here:
    https://drive.google.com/file/d/1gsKXqAVTT3PD4gTkOMWpWSVWMcBQMq9s/view?usp=sharing

    This is a really great exercise that excites my electrical intuition. I hope my answer can help anyone that wants to understand DAC math.
     
    Psalmanazar, Dzerh, Robert777 and 2 others like this.
  15. Mr.Sneis

    Mr.Sneis Friend

    Friend
    Joined:
    Oct 2, 2015
    Likes Received:
    942
    Dislikes Received:
    1
    Trophy Points:
    93
    Location:
    Phoenix, AZ
    I have owned more r2r dacs than I care to remember. I also haven't read any post outside of the OP, but I don't understand why this has to be an us vs them thing. If you truly love music none of this shit matters.

    You can have great DS dacs, you can have great R2R dacs.

    On the flipside you can have shitty R2R and shitty DS just as well.

    You will find generous use of mediocre opamps, cheated xlr outs, cheap power supplies, and snake oil across all dacs.
     
    Last edited: Jun 30, 2020 at 6:34 PM
    bixby, sphinxvc, Psalmanazar and 12 others like this.
  16. purr1n

    purr1n Finding his inner redneck

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    58,795
    Dislikes Received:
    100
    Trophy Points:
    113
    Location:
    Antarctica
    Come to think of it some of the expensive DACs I disliked the most were R2R: Wadia (the later PCM1704), Resolution Audio Cantata, Linn Numerik, Metrum Octave.
     
  17. RobS

    RobS treboR (Psalmanazar Groupie)

    Contributor
    Joined:
    Feb 21, 2019
    Likes Received:
    1,730
    Dislikes Received:
    82
    Trophy Points:
    93
    Location:
    Appalachia
    Do other R2R DACs have an overly smoothing way like the Yggdrasil does? Like harsh recordings are a bit too smoothed over in the transiensts and I was wondering if this is an R2R thing. How's Rockna fair here?
     
  18. Melvillian

    Melvillian Friend

    Friend
    Joined:
    Jul 3, 2017
    Likes Received:
    911
    Dislikes Received:
    2
    Trophy Points:
    93
    Location:
    Los Angeles
    I don’t think the A2 Gungnir Multibit does this from what I heard at the Schiitr. Seems to be a characteristic of Bifrost 2 and Yggdrasil A2
     
  19. RobS

    RobS treboR (Psalmanazar Groupie)

    Contributor
    Joined:
    Feb 21, 2019
    Likes Received:
    1,730
    Dislikes Received:
    82
    Trophy Points:
    93
    Location:
    Appalachia
    This might sound idiotic but I want my shitty recordings to sound like shit. Yggdrasil makes everything sound too good. I know that sounds absurd but I would like to have a lo-fi setup with a cheap DS DAC for those albums I have that are ragged and mastered horribly. But good recordings sound fucking sublime on Yggdrasil.

    I sometimes have a hard time discerning poor recordings on the Yggdrasil. Like stuff that has been brickwalled to shit doesn't sound as bad as it should. Or maybe I have stopped being so neurotic about this stuff.

    I've thought about doing that. Gungnir A2 would probably make a better match with XA25. Scales back the warmth of the Yggdrasil A2. Starting to discover too much warmth puts me to sleep. Yggdrasil has a bit of warmth (more so from its SE outputs) but not overly so like Gungnir A1. But hard getting that balance of "cool" without going sterile.
     
    Last edited: Jul 2, 2020 at 8:23 PM
  20. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Friend MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    9,068
    Dislikes Received:
    17
    Trophy Points:
    113
    Location:
    Irvine CA
    Fuck you guys, I love DS DACs that don't suck.
     
    bixby, nishan99, YMO and 16 others like this.

Share This Page