Schiit Yggdrasil Stereophile Review + Measurements

Discussion in 'Digital: DACs, USB converters, decrapifiers' started by purr1n, Jan 20, 2017.

  1. JayC

    JayC Resident Crash Test Dummy

    Pyrate
    Joined:
    Oct 26, 2015
    Likes Received:
    980
    Trophy Points:
    93
    Location:
    Eindhoven
  2. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    89,777
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    Let's not post links to reddit. I end up wasting too much time there. I'd rather these people set up a blind test to prove or disprove whatever. Until these people do, they are just armchair quarterbacking. All of these people just post links here and there and in many instances do not understand the nuances of what they are talking about. They really don't understand measurements or the math behind this stuff. I myself barely do - this keeping in mind that I've written the code to extract certain visualizations from the headphone impulse response measurements. This stuff takes me time to study and understand. I doubt sexually frustrated redditors who exhibit outrage at Moffat's comments on HF here or there are going to fare better in understanding this stuff.
     
    Last edited: Jul 17, 2017
  3. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Pyrate MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,960
    Trophy Points:
    113
    Location:
    Irvine CA
    1. I disagree with Mike about Parks-McClellan. It does optimize phase. The optimization Mike proposed is constrained differently though. For all practical purposes a Parks-McClellan and a Burrito Salsa by @schiit filter with thousands of taps are both going to perform great, at the expense of latency. The discussion offered by ilkless @ reddit is IMO pretty aimless though.
    2. Delta-Sigma vs Multi-Bit DAC is a discussion that has been around for ages. For all practical purposes the multi-bit DAC offered by @schiit performs great. Do I believe Delta-Sigma is worse than Multi-Bit? No. But again @schiit Multi-Bit implementation is very good.
    Whatever one may think, very few main stream DACs offer a linear phase brick wall digital filter like @schiit. And very few main stream DACs offer a well implemented not-end-of-life audio Multi-Bit DAC. Whatever I don't agree with @baldr or do, there is no denying that Jason and Mike's products are very competitive and perform very well.

    After a quick read of that reddit post I think ilkless is sort of out in the weeds, and seems to have an ax to grind.
     
    Last edited: Jul 17, 2017
  4. JayC

    JayC Resident Crash Test Dummy

    Pyrate
    Joined:
    Oct 26, 2015
    Likes Received:
    980
    Trophy Points:
    93
    Location:
    Eindhoven
    I too spent some time today trying to take apart that post and some of the comments, trying to see whether there was any weight to it but I was getting really lost.
    I just wanted to learn and I figured where better than here, from people I find more reliable sources to break things down or just point me in the right direction.
    You are right though, probably best to forget about reddit when it comes to stuff like this.

    Yup, totally seems that way. Some of the commenters are pretty deluded too, with them quoting JA's Yggdrasil review seriously and other nwavguy campers.
     
  5. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    89,777
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    I get sucked into it too; but ultimately proof is in the pudding, whether the car uses real corinthian leather or a DAC uses closed form math.
     
  6. AstralStorm

    AstralStorm Friend

    Pyrate
    Joined:
    Dec 18, 2015
    Likes Received:
    106
    Trophy Points:
    33
    Speaking of Parks-McClellan, it does have one drawback - it does not run with oversampling and can generate discontiguous FR or phase inbetween bands optimized. If you instead optimize a closed form low order spline interpolator or multilevel interpolator like it, you don't have the problem.

    There are better non-closed-form optimizations available nowadays - constrained nonlinear optimizers, you just need to put in a really good set of constraints. Parks-McClellan's min-max is quite outdated. Heck, it is sometimes outperformed by ordinary least squares except on the only measure it fully forces - the stopband.

    For example, I had extremely good research results when forcing an approximation to monotonicity in FR constraint in addition to stopband level, integrated passband ripple and maximum steepness in transition band, minimizing number of taps. This with interior point nonlinear optimizer from Octave and there are actually better ones out there.
     
  7. monacelli

    monacelli Friend

    Pyrate
    Joined:
    Jul 23, 2017
    Likes Received:
    1,649
    Trophy Points:
    93
    Location:
    The Land of Enchantment
    Is this bit about linear phase filters true? The science behind linear phase and generalized linear (affine) phase FIR filters has been understood for quite some time. There is really not a lot of mystery there any more. Of course, no realizable filter is truly "brick wall," as there is always some non-negligible transition width. So FIR filters can really distinguish themselves in three ways (1) transition band width, (2) passband ripple, and (3) stopband ripple. And then there is the matter of minimizing the filter order in circumstances where it's necessary to do so. That's about it. You can get fancy with polyphase implementations if you need to maximize computational efficiency, but even that type of thing is very well understood.

    Are most audiophile DACs really failing to implement well-designed FIR filters? That would be pretty scary and make me skeptical about a lot of the engineering going into these products in general.
     
  8. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Pyrate MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,960
    Trophy Points:
    113
    Location:
    Irvine CA
    I would say number of filter taps belongs to (4) filter order.

    Schitt is sacrificing (4) filter order (efficiency, power consumption, size, and latency), for (1) transition bandwidth, (2) passband ripple, and (3) stopband ripple.

    If (4) filter order is not a problem, then for the same number of crazy large number of taps, a Parks-McClellan and a Schitt's closed-form-whatevers (optimal in some random sense), may converge to pretty much the same thing and maybe off by an academic mosquito fart.

    Note that as the order of the filter grows, convergence of some numerical optimization algorithms becomes difficult due to precision limitations. For example, think 9000 taps for Bifrost, and 18000 taps for Gungnir.

    And with all that said, I don't think every audio DAC uses that many taps on their FIRs. Nor offer the same stop band rejection, transition band, ripple, and linear phase given filter order.
     
  9. Darren G

    Darren G Friend

    Pyrate
    Joined:
    Apr 4, 2017
    Likes Received:
    589
    Trophy Points:
    93
    Location:
    Pennsylvania, PA
    Ironic, but I've wondered, does Yggdrasil sound good (in part) because the intermediate samples it creates are a little off? What I mean is it's so easy to get seduced by the race toward accuracy, but take a microscopic look at tape, the particles are randomly distributed, and vinyl is hardly entirely perfect either. Thing is that the earliest CD's, before dither was a thing, might have been accurate within the limits of quantization errors, but repetitive accuracy ends up being annoying.
     
  10. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Pyrate MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,960
    Trophy Points:
    113
    Location:
    Irvine CA
    The interpolation process by Yggdrasil is likely more accurate than most other implementations. It's not really that off from ideal. It's actually the other way around. The filter order is huge. The Yggdrasil's filter performance has been actually measured. It's awesome. As in the interpolated samples will be more accurate than your average DAC.

    Vinyl turntable wobble, pitch, noise and so forth are not awesome. CD's un-dithered 16-bit, low order non-linear phase brick-wall recording, dynamic range compressed and poor remasterization are not awesome.

    IMO vinyl wins against many releases not because vinyl's non-idealities are addictive. It's because many digital releases are so fucked up relative to the vinyl releases, that the non-idealities of the vinyl media pale in comparison to what the studios do to the digital releases.
     
    Last edited: Jul 29, 2017
  11. Muse Wanderer

    Muse Wanderer Friend

    Pyrate BWC
    Joined:
    Oct 7, 2015
    Likes Received:
    946
    Trophy Points:
    93
    Location:
    Malta
    Classical music recordings may be an exception as sound engineers producing the recording tend to have direct feedback from the musicians and conductors with the primary aim for life-like dynamics, timbre and tone. Recordings from the 60s on labels such as Decca that were transferred from analog to digital tend to sound 'organic' despite being digital. Much newer recordings such as those on BIS (e.g. Suzuki Cantata boxset) or Harmonia Mundi are more detailed and sound sharper, probably as a result of their digital source.

    Personally I love most of my digital classical music, whilst many newer 'popular' music recordings whilst still good sounding may lose that oomph on the Yggdrasil. However listening to MFSL with Moffat GAIN recordings such as U2 Joshua Tree and REM Murmur is so freakin' good!
     
    Last edited: Jul 29, 2017
  12. landroni

    landroni Friend

    Pyrate
    Joined:
    Aug 12, 2016
    Likes Received:
    1,164
    Trophy Points:
    93
    What genuinely surprises me is that studios are intent on doing this even for 'audiophile' releases, like 'high-res' or MQA. Why would they feel the need to compress things for this niche? It's not like they'll end up played back on the radio... And these people probably have the right equipment to notice the difference.
     
  13. Muse Wanderer

    Muse Wanderer Friend

    Pyrate BWC
    Joined:
    Oct 7, 2015
    Likes Received:
    946
    Trophy Points:
    93
    Location:
    Malta
    There are some interesting anecdotes showing how intimate and engaged musicians and sound engineers become during their recordings.

    Igor Stravinsky, one the greatest 20th century composers, conducted his whole repertoire with Columbia Symphony Orchestra in the late 60s. It was issued on Columbia label (now Sony) on a 22 disc box-set. The music is marvelous and the box-set is such a bargain. On CD 8, after his sublime symphony in E flat, there are few pieces with the composer directing the orchestra and talking directly to the sound engineers at the other end of the recording equipment. There are some amusing parts such as when he describes how he views the microphone on track 10: "Now the listener is a microphone. He is a very disagreeable listener".

    On track 8, the engineers ask Stravinsky to play a part of Pulcinella slower. He ask "Why?, I like it better!". The engineer says that it is faster than the previous part and wants it to match the previous tempo. Stravinsky acknowledges and says his version is how it should sound. The engineer concedes but asks for a repeat as he will not be able to slice the tape properly. "Oh I see, I see, I will make this concession!" was Stravinsky's answer. Engineer responds "Everything in the name of technology". Stravinsky responds "Good technology!".

    The best part is his conversation on the 10th track, Symphony in C, after he conducts a piece that sounded excellent to him but not to the engineers. The trumpets were only faintly heard by the engineer. Stravinsky noted that the part was 'pianissimo' and it was supposed to sound that way. The engineer was quite adamant that the recorded sound was not acceptable and that he could change the sound afterwards if needed. Stravinsky asks "Can you?!" and the engineer responds "Believe me". Stravinsky responded 'I trust you!" but surely checked the final mix to ensure that what HE envisaged was the final result.

    There are various extracts of these recordings. Here is an example, when Stravinsky is literally obsessed to get the best sound from his orchestra.

    Part 2 of this video includes the time when the composer and the chief musicians go to the recording studio to study the result (time 1:18) ....

     
    Last edited: Jul 29, 2017
  14. Darren G

    Darren G Friend

    Pyrate
    Joined:
    Apr 4, 2017
    Likes Received:
    589
    Trophy Points:
    93
    Location:
    Pennsylvania, PA
    I have become a big fan of videos of live events (I enjoy watching the musicians perform); some of these suck so bad I can barely listen, but many are quite good after a few listens.

    Another + point for Yggdrasil; the audio delay is minimal. If this was a DAC with a billion (exaggeration) taps, a long delay/latency, that wouldn't work.

    Examples -

    Chris Botti Live in Boston (many guest stars, amazing all around, and oh the sound... excellent)
    Toto Live in Montreaux (not the Toto I remember, but damn they get down)
    Creed Live 2009 (holy crap the electric guitar work and crowd energy is some kick-butt shit)
     
    Last edited: Aug 1, 2017
  15. monacelli

    monacelli Friend

    Pyrate
    Joined:
    Jul 23, 2017
    Likes Received:
    1,649
    Trophy Points:
    93
    Location:
    The Land of Enchantment
    This criteria (minimal latency) is probably an argument for OS vs. NOS. With linear phase filters, the delay is deterministically related to the filter order (as you implied). So if you designed a nice, high-order filter (say M=16001 taps) and ran it at 44.1 kHz, you'd end up with a very perceptible delay ((M-1)/2 samples, so roughly 180 ms for redbook). But if you oversample and do your filtering at the higher sampling rate, you still end up with a group delay of 8000 samples, but each of those samples accounts for less time. So at 88.2 kHz, the delay would be 90 ms, etc. The trade-off is that filtering at the higher sampling rate is more computationally expensive. (Disclaimer: I have no clue about Schiit's designs, aside from forum posts and whatnot.)
     
    Last edited: Aug 1, 2017
  16. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Pyrate MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,960
    Trophy Points:
    113
    Location:
    Irvine CA
    There are no billion (exaggeration) taps. There are thousands of taps. And the audio delay should definitively not be minimal. This per Jason:

    http://superbestaudiofriends.org/index.php?threads/schiit-multibit-bifrost.105/page-6#post-2450

    There are a lot of claims there about how awesome the filter is. But here is my take:
    1. Parks-McClellan optimizes time and frequency domain.
    2. Schiit's filter is an approximation to the ideal filter, closed-form solution or not. This because the ideal filter is not realizable.
    3. Optimization is not an "educated guess". That is a gross miss-representation.
    4. Thousands of taps will have audible delay.
    5. Parks-McClellan likely has lower error bounds than the Schiit's closed-form filter when approximating the ideal filter for a given filter order (which is kind of academic given the number of taps).
    There is a lot of discussion about this in that thread.

    I feel it is pointless to put down the Yggdrasil on the account of it's digital filter performance (other than delay - if that affects at all your application). It's indeed fairly good. It is also pointless to call Yggdrasil a flawless piece of perfection. I would encourage people to do their own research and think for themselves whenever possible.
     
    Last edited: Aug 1, 2017
  17. TonyNewman

    TonyNewman Validated by Tyll removing Utopia from WOF

    Joined:
    Jul 16, 2016
    Likes Received:
    290
    Trophy Points:
    63
     
  18. AllanMarcus

    AllanMarcus Friend

    Pyrate
    Joined:
    Oct 23, 2015
    Likes Received:
    2,969
    Trophy Points:
    113
    Location:
    Los Alamos, NM
    Home Page:
    I'm a little confused. Darren was praising the Yggdrasil. You make it sound like you are arguing with his praise. Are you saying the delay is significant? Didn't you suggest the delay could be a small as .1 ms?
     
  19. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Pyrate MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,960
    Trophy Points:
    113
    Location:
    Irvine CA
    You are not confused. The Yggdrasil has many things going for it. But minimal delay is not it. I'm definitively not praising that. And for some applications minimal delay may be a must. For some others probably not so much.

    As far as 0.1 ms, do you mean this? > http://superbestaudiofriends.org/index.php?threads/schiit-multibit-bifrost.105/page-6#post-2490

    That is for a minimum phase filter where the main tap is 10 taps in @ 96kHz sampling rate. Turns out the Yggdrasil uses a linear phase filter where the main tap is 9000 taps in (assuming 18000 taps). For a 96 kHz sample rate one gets 94 ms of delay.

    I think for lip synch things have to be around +/- 22 ms before corrections need to be applied.

    Something is gotta give. You want a linear phase, brickwall filter, with -120 or more dB of rejection, uber flat passband, and some random optimization that keeps the original samples the same and not much degradation on the interpolated ones? You need a large filter and you are going to pay with delay.
     
    Last edited: Aug 1, 2017
  20. frenchbat

    frenchbat Almost "Made"

    Joined:
    Sep 26, 2015
    Likes Received:
    1,369
    Trophy Points:
    93
    @ultrabike isn't the Yggdrasil and Gungnir Multibit's tap length 18000 as per Jason's post you have linked ?

    EDIT : Thanks for clarifying.
     

Share This Page