Delta Sigma: An Inconvenient Truth

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

  1. Jay

    Jay Facebook Friend

    Joined:
    Oct 2, 2015
    Likes Received:
    172
    Trophy Points:
    43
    Location:
    GA, USA
    I’ve found some merit in this thread because it caused me to go down a rabbit hole researching articles and opinions on the webs. While I’ve always posited that implementation matters more than the chip, I also believed that some chips were superior to others. I mean, based on engineering, that had to be true, right?

    While going down my rabbit hole, I read more about upsampling, oversampling, and digital filters than I ever thought I’d have the attention span for.

    I’m now much more cognizant of the importance of and effects of the filters. So, for that alone, this thread was worth it to me.
     
  2. RobS

    RobS RobS? More like RobDiarrhea.

    Joined:
    Feb 21, 2019
    Likes Received:
    1,419
    Trophy Points:
    93
    Location:
    .
    @Jay share anything worth reading please? Maybe I'm doing this wrong but how I look at buying DACs is I'm buying it for the filter. Seems that is the most important aspect and one where it impacts the sonics the most. The Schiit Multibit stuff folks either like it or dislike it due to the custom properitary filter.

    In all seriousness, the Convert-2 might be the least delta sigma sounding DAC I've heard, or rather one that exhibits the least delta sigma attributes. Like purr1n said earlier, I can see how it can fool someone into it being an R2R, it's lines are the fastest I've heard from a DAC. Must have something to do with the filtering, cause the dual mono AD1955 based DC-1 did not sound nearly as fast. But the Convert-2 is the exception to the rule IME. It's jarring to go back to other DS DACs when I can hear all the digital sounding artifacts. That's been something of a revelation with the Yggdrasil. It has this subtlety to it that you only notice when it's no longer in your system. Like going down to cheaper DACs, I'm missing information in the recording I know should be there and its unsettling.
     
  3. Jay

    Jay Facebook Friend

    Joined:
    Oct 2, 2015
    Likes Received:
    172
    Trophy Points:
    43
    Location:
    GA, USA
    Basically, the summaries of what I read is that the filter is extremely important with phase and timing, which is far more influential with staging and imaging than the chip (at least any modern chip) and the output stage. Since some filters only work on multibit/R2R chips and not DS, it’s tough to tell for sure if the magic people hear from multibit/R2R dacs is from the chips themselves or the filters they employ. Maybe it’s a synergy of both.

    I’m sure this is well known by many here, but it was all new to me.
     
  4. RobS

    RobS RobS? More like RobDiarrhea.

    Joined:
    Feb 21, 2019
    Likes Received:
    1,419
    Trophy Points:
    93
    Location:
    .
    Yeah that makes sense, especially after reading the quotes k4 posted from Moffat. If you believe him, most DS filters lose phase information because they are destroying the original samples, so you lose optimization in the time domain.

    I'm sure the Yggdrasil chips do matter for the filter's performance, it needs to be bit perfect to do what the filter wants with phase and everything else.

    What I don't get is k4's claim that oversampling and DSP is a "parlor trick". Hope he can elaborate.
     
  5. Vtory

    Vtory Audiophile™

    Pyrate MZR
    Joined:
    Sep 12, 2016
    Likes Received:
    10,831
    Trophy Points:
    113
    Location:
    East Coast
    With standard linear phase fir filters, I believe there's no much room to lose. By its construction, phase response should be a linear function of frequency that ensures systematic cohesiveness all over the spectrum. In my journey in DS dacs when filters seletable, this has been the most solid choice to go. Some manufacturers opt out this with or without knowing what they're doing (mostly latter but some former), or give users options to choose. Arguably anything deviating from LP FIR does screw phases to some extent. May work (rarely) or may not work (more probably). No universal answer in the end.

    All the above applies to DS where out-of-band stuff must be filtered out. OS MB may apply but I'm not entirely sure to what extent it is the case.

    That's an interesting comment where we may differ. In fact I thought the opposite. To me, delta sigma sounding is mostly defined by (1) less distinctive staging depth and layering, (2) (ironically or counter-intuitively) less flowy or continuous perception, and (3) unnatural presentation of small signals (both dynamics and details). Not necessarily bad things at all. Anyway Convert-2 meets all the three criteria. I'm thinking it rather a very representative DS sounding done right. Modius (bal out) also categorized here. That's why I thought it a poor men's CV2.

    Pro iDSD in its DSD upsampling mode actually fooled me in #2 and #3 (and #1 to less extent). It really sounds like some hybrid.
     
    Last edited: Jun 28, 2020
  6. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    89,778
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    It's no parlor trick. It attempts to solve a problem.

    Oversampling eases the requirements for the analog anti-aliasing or reconstruction or low pass filter for CD content. CD content will beat at its sampling frequency at 44.1kHz, which isn't that far, just slightly more than an octave above 20kHz. There's probably some crap at 22.05kHz too. All this crap needs to be filtered for various reasons: preventing aliasing - higher frequency stuff mucking up stuff in the audio band, killing ultrasonics that could fry your tweeters or send your amp into oscillation, avoiding IMD, etc.

    Without oversampling, an extremely aggressive or steep filter needs to be used. This means more stages, more parts, and a filter transition band that bleeds deep in the audio band that results in early roll-off in and phase shift.

    This has nothing to do with the digital filter in OS DACs. All DACs have an analog filter too. Just that no one talks about it because it ain't sexy. What oversampling offers is a more faithful reproduction of the original signal. The entire deal with "NOS is the most pure unmolested way" is bullshit.

    What isn't bullshit is what approach different people may prefer.
     
    Last edited: Jun 28, 2020
  7. RobS

    RobS RobS? More like RobDiarrhea.

    Joined:
    Feb 21, 2019
    Likes Received:
    1,419
    Trophy Points:
    93
    Location:
    .
    I'm not sure I understand what you mean by unnatural presentation of small signals and details regarding DS DACs. Heck that would make more sense applied to the Yggdrasil. Convert-2 had the most "natural" presentation of those small signal details but its in the macrodynamics that were extremely exaggerated, sometimes too much. I have not heard that level of exagerration from other DS DACs.

    Convert-2 less flowy? I profoundly disagree. Or I may misunderstand what you mean. I'm with you on the staging and layering, definitely not top shelf. Haven't heard a DS DAC do that well. Supposedly Solaris has the best soundstage out of DACs and that's DS.
     
  8. k4rstar

    k4rstar Britney fan club president

    Pyrate BWC
    Joined:
    Jun 11, 2016
    Likes Received:
    6,942
    Trophy Points:
    113
    Location:
    Toronto, Canada
    what I meant by parlor trick is referring to the subjective result, not the technical theory for why its necessary. again, all of this sounds really good on paper, or else no one would bother. The perceived subjective effects of signal manipulation in the digital domain (including digital filters, oversampling, upsampling, noise shaping, decimation, anything you can accomplish with a DSP or ASRC) are something that to me may be initially impressive but gets old. It took me a long time to realize this and it required experiments with DACs, digital transports (CD and file-based), and playback software. In the end I decided to apply the concept of the shortest path (within reason) that I had already been applying to my amplifier and speaker system to my digital chain. It got the results I wanted.

    a lot of this can be thought of in terms of error correction. oversampling and corrective filters can be thought of as just another form of error correction, not unlike feedback in a tube amp; re-clocking schemes; etc. All of these engineering ideas came about as a means of achieving better measured performance. To think that a filter should be optimized for the time domain operates under the assumption that filtering/up-sampling/oversampling is of benefit to begin with. I don't believe in this, so I reject designs that require it, such as delta-sigma.

    continuing to use the Yggdrasil as an example because I know you're familiar with it, the subjective benefits of the filter can be described as an enhanced sense of space/air and relative instrument localization. there's absolutely no way of knowing of this is 'true' or not - but I do feel that these effects are superfluous to music and I can live without them. what I can't live without is the essence of music - expressiveness, intonation, rhythm, the feeling that the performer is engaged in their playing, the feeling of absolute presence or space even during silence, and so on. The Yggdrasil definitely doesn't fail here but what I have heard from NOS suggests it falls behind. it just took some work to find a NOS DAC that wasn't brutally compromised in other areas. THIS is where things become personal, do you really like the effect the Moffat DACs give? have you heard NOS DACs that didn't have flaws you can't live with? Important questions!

    p.s. I never heard the convert2 and I don't think I said I did. based on what I have read about it here I doubt I would live with it for any extended period of time.
     
  9. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    89,778
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    That's bad analogy. You can't just throw science away and in this case we are not talking about steady state measurements a la ASR. We can argue that NOS involves even more signal manipulation because of their steeper filters where the transition band goes so deep into the audio band. We are talking both frequency (HF rolloff) and time domain (HF phase shift) here. That is analog signal manipulation.

    If you prefer NOS, just say that you prefer it.
     
    Last edited: Jun 28, 2020
  10. Vtory

    Vtory Audiophile™

    Pyrate MZR
    Joined:
    Sep 12, 2016
    Likes Received:
    10,831
    Trophy Points:
    113
    Location:
    East Coast
    @RobS

    Unless we say both have the same level of naturalness, that's very fine that we differ in perception. That's all about what we regard as reference points. I (as having very little live performance experience compared to sbaf averages) personally think yggdrasil the most natural and convincing to me. It's my ground truth for every subjective evaluation. All the relative distances drawn from it. It's not strange that my arguments automatically translate into the opposite statements from another reference points.

    During my loaner period with CV2, I honestly thought it a little non-linear in small signal volumes, let alone exaggerated macrodynamics. Some sounds quieter than I'd expect, some louder. Like I said above, it's also possible to blame it for my own reference coordinates (likely unusual).
     
  11. k4rstar

    k4rstar Britney fan club president

    Pyrate BWC
    Joined:
    Jun 11, 2016
    Likes Received:
    6,942
    Trophy Points:
    113
    Location:
    Toronto, Canada
    I prefer NOS, but will take OS R2R over D/S any day, twice on Sunday

    I'm not trying to paint NOS as the perfect technical solution, it's obviously far from it. To choose between 1) non-oversampling R2R converter with its brute force real-time conversion errors and imperfections of the original data stream and 2) delta-sigma conversion producing a filtered 1-bit datastream with all the corresponding artifacts such as phase distortion can only be answered after in-depth real-world comparisons.

    I think music is a strict relationship of signal and time. if the time portion is interrupted as it is with D/S designs, no amount of clever digital manipulation will restore it. the Schiit filter is something of a happy medium but still too compromised for my taste.
     
    Last edited: Jun 28, 2020
  12. RobS

    RobS RobS? More like RobDiarrhea.

    Joined:
    Feb 21, 2019
    Likes Received:
    1,419
    Trophy Points:
    93
    Location:
    .
    It's cool. I was trying to understand where you are coming from and put your impressions in the right context. Sorry if it looked as if I was disputing you. I do agree on the loud/quiet sounds. That's an aspect of it that started to grate over time. After the initial excitement wore off, it became more "one note" sounding. I needed more subtlety and nuance so I started looking elsewhere.

    I have not heard any NOS DACs. If there's one that can do all genres well, with minimal flaws, then I'd like to check it out. I know you sent me one over PM but I see that as too much of a gamble and don't have an out if I end up not liking it. I know from reading here that Metrum definitely won't be my cup of tea.

    I can live with the softer sounding Yggdrasil, as I can enjoy more genres of music than the Convert-2. Its flaws are not dealbreakers to me. Looking at other DACs recommended here...havent seen much I want to try. Polite, relaxed, mellow, etc. are stuff I hate which seems to discount a lot of other DAC favorites.

    I look for hyper-enganging stuff where music demands my full attention and I don't want to do anything else. I can't listen to my current speaker setup during work cause I'm too distracted and want to dance. What's the point of expensive soundsystems if you can't dance? I don't sit when I listen to music. Even with headphones.

    From what you told me in PM, yeah I agree. Maybe too many compromises in certain genres you enjoy. I can't listen to Classical with Convert-2, for example.
     
  13. Melvillian

    Melvillian Friend

    Pyrate
    Joined:
    Jul 3, 2017
    Likes Received:
    1,345
    Trophy Points:
    93
    Location:
    Los Angeles
    If you had to do some work to find the right NOS R2R DAC, doesn’t that reinforce the point that implementation is the most important thing and not the perceived general drawbacks or benefits of D/S, R2R OS, and R2R NOS?
     
  14. k4rstar

    k4rstar Britney fan club president

    Pyrate BWC
    Joined:
    Jun 11, 2016
    Likes Received:
    6,942
    Trophy Points:
    113
    Location:
    Toronto, Canada
    i'm obviously biased but IMO there's a double standard - implementation matters more for R2R and less for D/S. i feel there's just a performance ceiling for D/S DACs when it comes to transmission of musical nuances due to the nature of the conversion that even the best ones can't do much about - that was kind of the basis of my original post.
     
  15. dark_energy

    dark_energy Friend

    Pyrate
    Joined:
    Sep 27, 2015
    Likes Received:
    1,040
    Trophy Points:
    93
    Location:
    Forest
    I wonder where the bottleneck is. Maybe the reconstruction with a filter where you rebuild the signal with the help of caps.
     
  16. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    89,778
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    NOS designs will introduce more phase shift than OS designs. There's no getting away from a steep analog filter. Maybe it's the other way around? That you prefer phase distortion?

    Maybe this expressiveness that you hear with R2R is the result of a platform which allows, even encourages more inventive designs, i.e. IV conversion, output stage, digital filter (or NOS), etc. DS chips tend to do everything in a single package + opamp + a cap here and there for the analog filter. Folks have tried NOS DS designs (yes, it's possible) with interesting results. The other thing we are left at is that progess with DS chips has been minimal, with only AKM and ESS coming up with their latest and greatest after 15 years, while making things simpler (embedding more functions into the chips). Audio design is much like cooking. Perhaps R2R is more like cooking from scratch whereas DS is the Hello Fresh box (or even worse the Martha Stewart box)?

    The only arguments that I can buy is that R2R is the native language of PCM and that with any kind of conversion something is lost, and that ultra-high frequency switching of DS or hybrid DS approaches may have a resultant sound that is different. I prefer to leave it here in simple terms, simple thought exercises rather than going down the rabbit hole of getting too sciency. The math behind DS simply wins. Same with DS measurements, frequency domain or time domain, DS just wins.

    See here: https://www.superbestaudiofriends.o...easuring-dac-accuracy.4184/page-3#post-131301 and https://www.superbestaudiofriends.o...easuring-dac-accuracy.4184/page-3#post-131354

    I'm a little bit concerned about the dissemination of misinformation where people think DS DAC output is some kind of fuzzy noise embedded recreation of the original signal when it's not. A lot of what I'm hearing sounds like R2R-NOS marketing bullshit from some audiophile company with corresponding lame examples, pictorals, and graphic analogies. The math behind DS is sound, just as sound as the math behind how the sinc function can reconstruct 44.1kHz - which is heavily aliased by nature because we are so close to Nyquist.

    It's hard for non-math people to understand, but the reconstruction of PCM isn't just connecting the dots. The beauty of the math is how the original signal can be almost magically recreated with the squiggles to the left and right of the dots they pass through.

    FWIW, I do prefer R2R. But I won't pad my preference with claims that DS somehow is a poor approximation of the real thing fudged with noise-shaping, digital manipulation and skulduggery. Heck, PCM is already an approximation: quantization noise anyone? Real world sound quanta can't exactly be mapped to a limited set of integer numbers.
     
    Last edited: Jun 29, 2020
  17. RedFuneral

    RedFuneral Facebook Friend

    Joined:
    Jun 8, 2016
    Likes Received:
    213
    Trophy Points:
    43
    Location:
    CT, USA
    I've owned 2 NOS DACs, an Ultra-Fi DAC41 & Metrum Flint. The Ultra-Fi is hyper engaging with the fastest transients I've heard in a DAC. It's also kind of rough & rowdy, like if Grado made a DAC. It's a must listen for anyone who believes NOS will always be polite & sleepy. The big catch is that I don't think it really holds up on any technical front, the strong transient perception is probably from distortion, aliasing, or some other NOS flaw. That being said its the only component I've owned 5yrs+ and I used it exclusively during that time.
    The Metrum fits the stereotype perfectly, it kind of confused me that I still liked it with aggressive metal music. I regret selling it. Similarities between the DACs were warmth, 'pure' sounding transients despite falling on different ends of the speed spectrum, lack of high treble, and WTF is this performance with atmospheric electronic music(ambient and the like.) They're an adrenaline rush with techno though, the music feels like it's playing faster and the warmth adds weight to all the notes.

    The general marketing point with NOS is the lack of ringing, the 'perfect' impulse response. I was a true believer when I owned my UF but later getting a Metrum with the same NOS impulse and a rounded sounding transients makes me question things. It still triggered that adrenaline response to fast music despite the laid-back sound. What are the chances aliasing or IMD make things seem more exciting? What if they take studio recorded music and make it sound more like it was recorded in an imperfect room(as most people likely experience live music of the non symphonic variety?)
     
  18. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    89,778
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    A steep analog filter will cause plenty of ringing. I guess no one mentioned that. Or if no filtering, improper reconstruction, especially at 44.1kHz, could sound good or different or have a cool effect.

    It's all over the place. I'd say the Holo Spring had rather sharp transients and was less round sounding than Gungnir A1. The Metrum Octave (1st gen Metrum) sounded so polite and boring that I wanted to kill myself. The Sonnet Morpheus seemed like a return to this form as opposed to the most recent Metrums (before the company's dissolution) which I rather liked. The lower-end Soekris stuff, which isn't particularly rounded also made me want to kill myself out of boredom. AGD Master 7 in NOS mode was rather interesting, dense, lush, inviting, but ultimately too much of a good thing. It's too hard to generalize NOS as having a certain kind of transient response, tone density, level of expressiveness, etc. I say this because some folks almost swore off NOS from the Morpheus DAC thread, i.e. "If NOS sounds like this, then I'm out". There's more to NOS than how one NOS implementation may happen to sound like.

    For studio recordings, NOS has its charms because it might make us feel as if we were in the middle of or right up to the band. For orchestral from speakers, NOS is a big showstopper if you want to feel like you are in Disney Concert Hall or at the SF Symphony. I think it's important in these kinds of discussions to include recordings. From a different point of view, it's NOS, it's staging that could be the parlor trick.
     
    Last edited: Jun 29, 2020
  19. skem

    skem Friend

    Pyrate
    Joined:
    Nov 1, 2017
    Likes Received:
    1,911
    Trophy Points:
    93
    Location:
    Charles River
    From what I understand, many people find first order speaker crossovers to be unengaging and less dynamic, even though they are reproducing the sound without phase shifts. By contrast, I find first-order (or crossover-free) transducers to preserve timbre and stage better and I prefer them. Thus, I suspect that de-phased harmonics makes music sound more alive even though it’s artificial (an actual parlor trick).

    Since so many components in a chain are likely to contribute to phase shifts that chasing phase coherence is generally ill advised. Moreover, very few people have probably heard a phase-coherent system, so whatever the sound of phase coherence might be, it is not a widely appreciated thing that people become accustomed to. Finally, in electronic or highly processed music, it offers no objective benefit. But let’s say it is a thing for minimally processed instrumental music and binaurally mic’ed room recordings.

    I suspect this thing is why NOS has its adherents. Some people who pay particular attention to stage and timbre find NOS offers far more realism (e.g., me) and I suspect the reason has to do with phase shifts across multiple instrumental harmonics.

    Note: the phase coherence of harmonics is preserved in a listening room regardless of distance from the sound source.
     
    Last edited: Jun 29, 2020
  20. Serious

    Serious Inquisitive Frequency Response Plot

    Pyrate BWC MZR
    Joined:
    Sep 28, 2015
    Likes Received:
    2,594
    Trophy Points:
    113
    Location:
    near Munich, Germany
    @skem phase shifts across instrument harmonics should be virtually 0 for any linear phase OS DAC. Even with minimum phase filters most of the phase shift happens in the octave from 10-20kHz, so most of the signal still won't be affected. But as ultrabike demonstrated elsewhere, the waveform gets noticeably affected by the phase shift from minimum phase filters, be it in oversampling DACs or (to a lesser degree since the filters are likely not as steep) in NOS DACs due to analog low pass filters or simply analog bandwidth limits.

    As you mention speakers will have far more phase shift anyway. I'm not sure if you've seen it, but I've made files you can import into a convolver to hear the effect from popular speaker crossovers yourself:
    https://www.superbestaudiofriends.org/index.php?threads/crossover-phase-shift-audibility.9313/

    I don't think first order crossover speakers are inherently less dynamic. If anything I think what you hear is more due to the fact that the drivers have to have wide bandwidth and smooth response, which often means lower efficiency designs and more correction filters in the crossover. I find my OB speakers to sound a helluva lot more dynamic than most "normal" floorstanders and they are as close to minimum phase as they can be considering the drivers used.

    I'm not sure which speakers you've heard, maybe we just disagree in our description of sound. I do think the steeper filters can add an unrealistic edge to the sound FWIW.

    I also think phase shifts in the audio band should generally be negligible from most components other than speakers. Microphones will also have some phase shift at the extremes considering they are less linear and more bandwidth-limited than electronics. I'm currently experimenting with using FIR filters to get a close to linear phase response (as opposed to minimum phase) from 20Hz - 20kHz, but I'm still undecided as to if it's worth it or not.

    BTW: I think headphones off a DAC with a linear phase filter could easily be considered a phase coherent system. So I think in reality many people have experienced a phase coherent system.
    I also don't think binaural recordings are as great phase-accuracy-wise as you might think. They will require EQ to not sound emphasized in the upper midrange. IMO the EQ filters used here should simply be minimum phase filters to "restore" the original phase information. And once we get past 4-5kHz-ish phase accuracy goes to shit with binaural recordings due to the peaks and dips in the FR.
     
    Last edited: Jun 29, 2020

Share This Page