Schiit Yggdrasil Measurements

Discussion in 'Source Measurements' started by atomicbob, Oct 31, 2015.

  1. atomicbob

    atomicbob Friend

    Friend
    Joined:
    Sep 27, 2015
    Likes Received:
    3,087
    Trophy Points:
    93
    Location:
    On planet
    2015.12.12 update - corrected THD and THD+N measurement. Please go to post 6 below to see the correct data. The incorrect measurement with the OS driver interaction appears in the attachments to this post (#1.)

    I will never know how the first three hours of the yggdrasil sounded as I let the DAC warm up for an hour and proceeded to make technical measurements. If the measurements presented are familiar territory, then you should be as impressed as I am. If such graphs are not familiar, then please note that these graphs are extended, some down to -160 dB for example. Extreme. Much farther than most other graphs you will see elsewhere on other DACs on the internet or in brochures. So be careful when making comparisons. Watch the X and Y axes in other data. The graphs presented here are not marketing graphs. They have not been prettied up as is often the case in mainstream consumer goods.

    Schiit claims THD <0.006% THD+N (more stringent) measured Left 0.00233% Right 0.00239%
    Schiit claims IMD <0.007% IMD measured Left 0.00067% Right 0.00162%
    Schiit claims SNR >117dB Dynamic Range measured Left 121.7 dB Right 122.0 dB

    Schiit has specifications for these measurements published on their website and in their operating instructions. In my measurements, my yggdrasil performed better than Schiit's conservative specifications in every single measurement. Every single one. Well done Schiit!


    Commentary added to the end of this post.

    The data presented were collected as follows:

    1. PrismSound dScope III, picoscope 2205, FLIR E63900
    2. DAC balanced output used for measurements except oscilloscope square wave response which used unbalanced output
    3. 100 Kohm load used for measurements
    4. 44 KHz sample rate, 24 bit depth
    5. USB input - cables to go USB cable
    6. Balanced cables Tecnec Canare starquad Neutrik XLR cables
    7. Unbalance cable BlueJeans Belden RCA
    8. Vaunix Lab Brick USB hub

    Commentary:

    The previous THD+N graph measurements were artificially high due to a driver / OS interaction on my computer. The corrected measurement now appears in post 6 below. Update made 2015.12.12.

    Designers are often faced with trade-offs for various reasons. I believe the Schiit design team has made trade-offs that result in best auditory experience over best measurements possible on all fronts. I have a colleague who gives a "13 dB miracle" demonstration showing that SNR isn't the best predictor of auditory experience. So THD+N may not be that influencial on the final sound while jitter may have a bigger effect. Those of you who will reference the 1992 and 1998 AES papers on the topic of jitter may note that no trained subjects were used to determine the masking thresholds. With many more listeners now well experienced in digital audio listening we might see much lower thresholds if the experiments were repeated today and trained listeners employed.
     

    Attached Files:

    Last edited: Dec 12, 2015
    ultrabike, auri, zonto and 3 others like this.

  2. atomicbob

    atomicbob Friend

    Friend
    Joined:
    Sep 27, 2015
    Likes Received:
    3,087
    Trophy Points:
    93
    Location:
    On planet
    Sorry to spread this over multiple posts. More measurements from 3 hr on time.
     

    Attached Files:

    auri and Feuergeist76 like this.
  3. atomicbob

    atomicbob Friend

    Friend
    Joined:
    Sep 27, 2015
    Likes Received:
    3,087
    Trophy Points:
    93
    Location:
    On planet
    Yggdrasil measured at 408 hrs on time.

    Test setup is same as described for the measurements taken at 3 hrs on time with the following exception: A PicoScope 5243B was used in place of the 2205. The 5243B has 15 bit 2 ch ADC capability and higher sample rates than the 8 bit ADC used on the 2205.

    I will try to include a few extra screenshots though this posting system only allows 8 per post and I only reserved two posts for these measurements. Most measurements were so close to the original to be considered of no consequential change. Except one spectacular change. If the 3 hr on time jitter spectrum was impressive, take a look at the 408 hr on time jitter spectrum. OMG! My congratulations to Mike Moffat. That is a truly impressive achievement!

    The THD+N graph measures artificially high due to a driver / OS interaction on my computer. An update will be forthcoming.
     

    Attached Files:

    Last edited by a moderator: Oct 31, 2015
    auri and Feuergeist76 like this.
  4. atomicbob

    atomicbob Friend

    Friend
    Joined:
    Sep 27, 2015
    Likes Received:
    3,087
    Trophy Points:
    93
    Location:
    On planet
    Yggdrasil measurements at 408 hrs on time, part B.
    I have included a 1KHz -70 dBFS time domain signal to demonstrate that the blib seen in the -90 dBFS version is pretty insignificant.
     

    Attached Files:

    Last edited by a moderator: Oct 31, 2015
    auri and Feuergeist76 like this.
  5. Marvey

    Marvey Loves sex and records

    Staff Member Friend
    Joined:
    Sep 24, 2015
    Likes Received:
    25,325
    Trophy Points:
    113
    Location:
    Calabasas, CA
    Take with a grain of salt. Using Focusrite POS. Input is 16/96 (with gain shoved up), so might be some aliasing. Low noise floor, but quite a bit of crap in the treble. Ignore the uV - I didn't calibrate. However, it does give us some idea what is going on.

    16 and 24 bit data -90db undithered 1kHz sine wave. Looks like Yggy dithers the 16bit signal.
     

    Attached Files:

    ultrabike, auri and Feuergeist76 like this.
  6. Marvey

    Marvey Loves sex and records

    Staff Member Friend
    Joined:
    Sep 24, 2015
    Likes Received:
    25,325
    Trophy Points:
    113
    Location:
    Calabasas, CA
    There measurements were done on the QA400 with the QA190 differential probes. The balanced outputs of the Yggy were used. Since the Yggy's outputs are so hot, the differential probes were put on /10. The use of the probes increase the noise floor of the measuring equipment about 10db. 0dBr in the graphs was set represent the max output of the Yggdrasil. ARTA was used to generate the 16/44.1kHz signals from a PC to the Yggy via USB and unfortunately DirectSound. At a later date, I'll make WAV files to playback directly via WASAPI in JRiver. 16384 points were used for most of the FFTs. A noise lower floor could have been obtained with a larger window, but processing would have taken longer and I didn't want to wait. Finally, my home is a noisy environment with lots of LED, CFB, dimmer, and Ethernet over A/C electronics. Right channel is displayed to get a sense of the limitations of the measurement equipment.

    A lot of these measurements are comparable to those seen in Stereophile with the exception that the AP gear used by JA is better.

    #1) 1kHz @ -90dBr. Low level signal @ 1kHz vs. noise floor.
    #2) 1kHz @ -1dBr. The standard 1kHz measurement with the DAC running at almost max output.
    #3) 1kHz @ -1dBr. Same as above, but this time we are taking a close peak at the spectrum below 1kHz.
    #4) 19kHz + 20kHz. IMD. Slighter wider band view.
    #5) PN white noise. Wide band view to note extension and nature of filter slope.
    #6) 1kHz @ -60dBr. Mid level signal.
    #7) 19.1kHz. Wide band view.
     

    Attached Files:

    ultrabike, auri and Feuergeist76 like this.
  7. atomicbob

    atomicbob Friend

    Friend
    Joined:
    Sep 27, 2015
    Likes Received:
    3,087
    Trophy Points:
    93
    Location:
    On planet
    2015.12.12 Here is the updated and correct THD and THD+N measurement for my yggdrasil:

    THD THD+N
    upload_2015-12-12_14-5-48.png
     
    auri, Feuergeist76 and zonto like this.
  8. winders

    winders Friend

    Friend
    Joined:
    Feb 13, 2017
    Likes Received:
    487
    Trophy Points:
    63
    Location:
    San Martin, CA
  9. Marvey

    Marvey Loves sex and records

    Staff Member Friend
    Joined:
    Sep 24, 2015
    Likes Received:
    25,325
    Trophy Points:
    113
    Location:
    Calabasas, CA
    I wouldn't read too much into what Mike is saying. There seems to be all sort of rumors going around the Internet. One person on HF was saying stuff like Mike and Jason were handling out brand new Analog Boards (that fixed the "blackness" problem) in cardboard boxes with attached NDAs forms to important people at RMAF. LOL! Or that we had a SBAF meet in somewhere California where these DACs with the Analog Boards were exhibited. (Maybe there was some confusion, as at the recent Pasadena meet, there was a tweaked Gumby with direct out from the chips using Freya buffer as the output stage). I've also been getting questions like, I heard a new Bifrost is coming out? Should I wait? And I'm like, not to my knowledge. I don't know. I'm kind of wondering if there is just some confusion with the Gen 5 USB boards or hacked Gumbys, or maybe some of the tweaks on the output stage (jumper a few parts, remove a few parts) I've tried.

    P.S. Yes. There was a software update that fixed the glitch that one dude was saying he could hear at Computer Audiophile. This has actually been a pretty poorly kept secret, as a Computer Audiophile member already let this on. The fix happened like a year and half ago. (The funny thing is that the guy who said he could hear the glitch probably heard a fixed unit). I've heard both pre and post fix DACs. Can't tell the difference. We can't hear small glitches the size of a few LSBs that only occur under specific conditions in a 24-bit -90db signal. I'm sure Schiit "fixed" this to keep measurbators like John Atkinson @ Stereophile happy. Atkinson said that the glitch was caused by the design choice to use 20-bit filters. He's an idiot. An idiot for making the wrong conclusion*, and an idiot from simply not asking Schiit the cause of the glitch.

    *The use of 20-bit filters would have resulted in more easily seen discrete steps in the -90db sinewave, not a glitch at the point where all the bits flip at the same time.
     
    Last edited: Oct 12, 2017 at 9:04 AM
    Grahad2, 9suns, dmckean44 and 2 others like this.
  10. winders

    winders Friend

    Friend
    Joined:
    Feb 13, 2017
    Likes Received:
    487
    Trophy Points:
    63
    Location:
    San Martin, CA
    I was suggesting that new measurements were in order to show that the “glitch” at -90dB was gone. That’s all. I was not suggesting that there was a new analog board or what effects it might have if it did exist.
     
  11. schiit

    schiit SchiitHead

    Friend
    Joined:
    Sep 28, 2015
    Likes Received:
    2,843
    Trophy Points:
    93
    Location:
    Valencia, CA
    Home Page:
    With all the speculation about changes on both the analog and digital side, I should probably do a chapter about the realities of modern manufacturing (that no, sorry, not everything is exactly the same, that no, sorry, happy elves do not lovingly craft every single part, and that yes, parts go obsolete and necessitate changes.) In fact, the output devices we used in Magni 2 went obsolete twice, one time which necessitated fairly significant changes. That's life, that's how it goes.
     
    Last edited: Oct 12, 2017 at 9:19 AM
    Thad E Ginathom and jowls like this.
  12. atomicbob

    atomicbob Friend

    Friend
    Joined:
    Sep 27, 2015
    Likes Received:
    3,087
    Trophy Points:
    93
    Location:
    On planet
    The 'glitch' only occured at zero crossing in the DAC output. A 90 dB down signal may occur ANYWHERE in the range of +/- 8388608 quantization levels of the DAC output. So if a -90 dB 1 KHz signal was superimposed on a -30 dB 50 Hz signal, the glitch would only occur every time the composite signal crossed zero. This implies an occurrence once out of 441 samples at 22.6 uS per sample. For real world music, a bell like sound riding on a bass signal is highly unlikely to stimulate the glitch as often as this contrived example.

    Hence, I was not concerned about this glitch and find it implausible that anyone can hear it.
     
  13. Marvey

    Marvey Loves sex and records

    Staff Member Friend
    Joined:
    Sep 24, 2015
    Likes Received:
    25,325
    Trophy Points:
    113
    Location:
    Calabasas, CA
    I'm responding to all the people asking me about dumb rumors, all of which started after RMAF.
     
  14. Marvey

    Marvey Loves sex and records

    Staff Member Friend
    Joined:
    Sep 24, 2015
    Likes Received:
    25,325
    Trophy Points:
    113
    Location:
    Calabasas, CA
    WHAT! You betrayed me! You are telling me that this Magni 2U I've been using as a instrumentation amp for headphone measurements has been improved and you didn't tell me about it or offer me an upgrade! I could have gotten better distortion measurements by 0.0033% on the HD650!
     
    Grahad2 and schiit like this.
  15. dmckean44

    dmckean44 Friend

    Friend
    Joined:
    Nov 17, 2016
    Likes Received:
    212
    Trophy Points:
    43
    Location:
    San Diego
    And disturb all the crazy cat ladies that have taken over your thread?
     
    monacelli and schiit like this.
  16. Scott Kramer

    Scott Kramer Friend

    Friend
    Joined:
    May 3, 2016
    Likes Received:
    496
    Trophy Points:
    63
    Screen Shot 2017-10-12 at 12.35.21 PM.png
    I'm surprised @schiit didn't do the DAC's and Analog output stages on separate boards. Those x4* AD5791BRUZ's (the 20bit R2R chips) are $100 bucks a pop on mouser.

    *1 per phase, per channel
     
  17. Marvey

    Marvey Loves sex and records

    Staff Member Friend
    Joined:
    Sep 24, 2015
    Likes Received:
    25,325
    Trophy Points:
    113
    Location:
    Calabasas, CA
    It's like what Mike said: 1) Not worth it to sell upgrades unless there is a signifcant improvement, because any upgrades are going to be expensive; 2) If it's easy to roll in, like the glitch fix, just do it as a running change.
     
  18. Grahad2

    Grahad2 Friend

    Friend
    Joined:
    Jan 26, 2017
    Likes Received:
    860
    Trophy Points:
    93
    Location:
    Singapore
    I think he's talking about manufacturing since if 1 chip spoils or is DOA or goes magic smoke up, you have to scrap the expensive board until someone can manually rework it. Or at least that's why semiconductors have so many redundancies for larger ASICs.
     

Share This Page