Measuring DAC Accuracy

Discussion in 'Source Measurements' started by purr1n, Mar 30, 2017.

  1. purr1n

    purr1n Finding his inner redneck

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    64,224
    Dislikes Received:
    108
    Trophy Points:
    113
    Location:
    Antarctica
    This measurement was designed to work with decent hardware like the Focusrite. This is why we only look at the performance of the 6 LSBs instead of all the bits. I crank up the input gain so the measurement device's accuracy hardly matters. BTW, the Focusrite 2i2 is actually pretty good for measurements. As always, I recommend balanced inputs or probes, even if output is single ended.

    This is what else you will need:
    1. My special WAV file. Precision triangle waves 16 bit with code words for -32 to 32, each repeating three times.
    2. ARTA or something that will record the input.
    3. WAV player that will let you set bit depth.
    4. Output of values to a .TXT file. I used ARTAs sample export format.
    The piece you don't have is my custom code to create the visualizations. This is still in development. The code is utter shit... not documented, badly formatted, not following any code standards.

    There are still avenues I would like to explore. This is new ground to my knowledge, unless JA has already written a book about it which I am not aware of.
     
    Last edited: Apr 5, 2017
    Azteca, ultrabike and Garns like this.
  2. Grahad2

    Grahad2 Red eyes from too much anime

    Friend
    Joined:
    Jan 26, 2017
    Likes Received:
    1,224
    Dislikes Received:
    2
    Trophy Points:
    113
    Location:
    Singapore
    The SE summation circuit in the Gungnir Multibit might be the culprit here.
     
  3. purr1n

    purr1n Finding his inner redneck

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    64,224
    Dislikes Received:
    108
    Trophy Points:
    113
    Location:
    Antarctica
    Well, folks here have reported their preference for the ISOMAX XLR to SE tranny box to the SE outputs, even though trannies do add a bit of distortion in the bass at higher output levels.
     
  4. purr1n

    purr1n Finding his inner redneck

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    64,224
    Dislikes Received:
    108
    Trophy Points:
    113
    Location:
    Antarctica
    Here is Modi 2 for more completeness:
    Modi 2 DNL.png

    The Modi 2 is a very resolving (subjectively speaking) DAC BTW.
     
    Psalmanazar likes this.
  5. cskippy

    cskippy Creamy warmpoo

    Friend MZR
    Joined:
    Oct 14, 2015
    Likes Received:
    4,008
    Dislikes Received:
    1
    Trophy Points:
    113
    Location:
    Tempe, Arizona
    I was just about to ask about the ISOMAX as I just picked one up. Do you happen to have one to test how it fairs with Bal>single ended conversion?
     
  6. bazelio

    bazelio Friend

    Friend
    Joined:
    Nov 16, 2015
    Likes Received:
    3,451
    Dislikes Received:
    160
    Trophy Points:
    113
    Yggdrasil to ZDS Fuzz Buster.

    [​IMG]
     
    Last edited: Apr 5, 2017
    LauSing2 likes this.
  7. winders

    winders Know-it-all boomer, prob racist

    Friend CBC
    Joined:
    Feb 13, 2017
    Likes Received:
    2,228
    Dislikes Received:
    399
    Trophy Points:
    113
    Location:
    San Martin, CA
    When I upgraded from my Yggdrasil/Lyr 2 combo using SE to Yggdrasil/MJ2 using balanced, SQ improved noticeably. I credited the MJ2 for all of that improvement. Maybe some of that SQ improvement has something to do with Yggdrasil's balanced output.
     
  8. purr1n

    purr1n Finding his inner redneck

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    64,224
    Dislikes Received:
    108
    Trophy Points:
    113
    Location:
    Antarctica
    MJ2 is a much much more capable amp than the Lyr 2 though.
     
  9. winders

    winders Know-it-all boomer, prob racist

    Friend CBC
    Joined:
    Feb 13, 2017
    Likes Received:
    2,228
    Dislikes Received:
    399
    Trophy Points:
    113
    Location:
    San Martin, CA
    That is what I understood. Which is why I was willing to give all the credit to the MJ2. I guess I should have given some credit to going balanced in the first place.

    Some of the improvements in detail, depth, separation, and clarity could be due to a cleaner balanced signal as you saw with Gungnir Multibit.
     
  10. drfindley

    drfindley Secretly lives in the Analog Room - Friend

    Friend
    Joined:
    Sep 25, 2015
    Likes Received:
    1,534
    Dislikes Received:
    1
    Trophy Points:
    93
    Location:
    Palo Alto
    Is it possible to measure further up than the output stage? I guess might give something like the Gungnir Multibit an unfair advantage, but I'm curious how much the output stage matters in these graphs. That said, I don't think it's very relevant.
     
  11. Scott Kramer

    Scott Kramer Friend

    Friend
    Joined:
    May 3, 2016
    Likes Received:
    1,259
    Dislikes Received:
    3
    Trophy Points:
    93
    Wait... This isn't the Gumbreya output is it— did you use that by mistake?
     
    bazelio likes this.
  12. Johan

    Johan Rando

    Banned
    Joined:
    Feb 23, 2017
    Likes Received:
    12
    Dislikes Received:
    29
    Trophy Points:
    8
    Location:
    Resident troll and general fucker
    seems like the schiit dacs have some flaws, both with SE outputs and some digital anomalies/glitching

    *flame sheild on*

    I am a proud Gungnir Multibit owner
     
  13. Serious

    Serious Inquisitive Frequency Response Plot

    Friend BWC MZR
    Joined:
    Sep 28, 2015
    Likes Received:
    1,867
    Dislikes Received:
    0
    Trophy Points:
    113
    Location:
    near Munich, Germany
    I think it's nearly impossible to tell what is from linearity errors and what is noise, unless the errors really are exactly the same every time (like with the Modi MB). That doesn't seem to be the case with the AGD S19.
    Gungnir Multibit Balanced looks much more like I'd expect. I doubt the zero crossing glitch was fixed, it's just that it's at a much lower level with the Gungnir MB than with the Modi MB. I could try to measure mine using Rag as an amp between DAC and ADC, but I doubt that I can even see it with my equipment.

    Gumbreya would be interesting to measure, but I don't think that it would measure any better than a regular Gungnir Multibit.
     
    Last edited: Apr 6, 2017
    Psalmanazar likes this.
  14. cizx.6

    cizx.6 Just couldn't stay away...

    Joined:
    Apr 3, 2017
    Likes Received:
    393
    Dislikes Received:
    3
    Trophy Points:
    63
    Location:
    US
    This seems contrary to what I've come to believe reading this site. Is the current consensus that for Gungnir and Yggdrasil, using XLR to RCA cables into a single ended amp is better than straight RCA to RCA?

    I read @Marvey 's comment as referring to measurement equipment only, and in the other direction (RCA out on DAC to XLR in on measurement device).

    Sorry if this is off topic, just trying to parse the info I'm seeing.
     
    mrflibble likes this.
  15. purr1n

    purr1n Finding his inner redneck

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    64,224
    Dislikes Received:
    108
    Trophy Points:
    113
    Location:
    Antarctica
    Nope. Straight from XLR jacks, not direct from chips.
     
    Scott Kramer likes this.
  16. purr1n

    purr1n Finding his inner redneck

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    64,224
    Dislikes Received:
    108
    Trophy Points:
    113
    Location:
    Antarctica
    Yeah. Pretty much difficult to separate noise from distortion, from DA linearity errors. However noise from resistors or power circuits for a ladder or sigma-delta circuit, discrete or otherwise, are still very much part of this. Performance at the outputs is what ultimately matters. For DAC, it's very much impossible to separate all the pieces, and it wouldn't make sense just to measure one just one aspect at one point in the circuit.

    To a small part, we can separate random noise. Noise averages out. The light colored line in the plots is the average over 100 cycles. The S19 seems to have a bell curve distribution to the values for each point. There are some visualization techniques I am working on that will make this more apparent.

    Yes, Gungnir Multibit still has glitch. Just not at 16 bits, or 17, or 18. If look closely, there is a tiny bump at the zero crossing. Hahaha.
     
    Last edited: Apr 6, 2017
  17. Azteca

    Azteca Friend

    Friend
    Joined:
    Sep 26, 2015
    Likes Received:
    1,503
    Dislikes Received:
    1
    Trophy Points:
    93
    Location:
    Cincinnati, Ohio
    I've got a Pulse Xfi (balanced outs) and an RME ADI-2. Also have a Geek Out 1000 (1st gen blue dude) with SE output. Would be interested to run these tests.
    I think it's a great idea to build up a corpus of data while we figure out the best way to visualize it.
    Thanks for the good work.
     
  18. purr1n

    purr1n Finding his inner redneck

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    64,224
    Dislikes Received:
    108
    Trophy Points:
    113
    Location:
    Antarctica
    Some XLR to SE cables short ground to negative. I don't know the effect of this. I thought Atomicbob or someone else had a measurement of this. I can certainly check it out.

    Now if the XLR to SE cable just takes one leg of the differential output, we would be using only one of the two chips, and who knows what else.
     
    cizx.6 likes this.
  19. purr1n

    purr1n Finding his inner redneck

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    64,224
    Dislikes Received:
    108
    Trophy Points:
    113
    Location:
    Antarctica
    I have a few more sets of wave files I am working on to check flipping of bits. Will include the file and instructions how to obtain the data.
     
    ultrabike likes this.
  20. purr1n

    purr1n Finding his inner redneck

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    64,224
    Dislikes Received:
    108
    Trophy Points:
    113
    Location:
    Antarctica
    Top = Gungnir Multibit from XLR on back panel
    Bottom = Gungnir Multibit direct from chips (test points on PCB)
    Gumby Direct vs XLR.png
     
    drfindley and Serious like this.

Share This Page