Airist R2-R DAC Measurements

Discussion in 'Source Measurements' started by purr1n, Jun 4, 2018.

  1. purr1n

    purr1n Finding his inner redneck

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    68,331
    Dislikes Received:
    112
    Trophy Points:
    113
    Location:
    Antarctica
    See here for comparison to S19 discrete resistor DAC.

    NOTE: PRE-PRODUCTION MEASUREMENTS.

    As expected, not great measurements.

    0db 1kHz.
    Second, third, and forth order are about 10db less than the S19 measurements, but higher order are higher than S19. Note that spikes will be more evident and stand out more from the AverLAB measurements because the noise floor on the AverLAB is so much lower than from the QA400 with the diff probes.
    RDAC 1kHz 0db-1.jpg

    -60db 1kHz
    Lower 2nd order, but higher 3 order and a few spike higher than S19.
    RDAC 1kHz -60db-1.jpg

    -90db 1kHz
    Practically zero distortion except 10th and 20th order.
    Did Airist pay closer attention to precision matching / calibrating the LSBs???!!! Is this why the DAC sounds better that what the above measurements suggest? Or does the RDAC not like higher output levels?
    RDAC 1kHz -90db-1.jpg
     
    Last edited: Jun 5, 2018
  2. purr1n

    purr1n Finding his inner redneck

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    68,331
    Dislikes Received:
    112
    Trophy Points:
    113
    Location:
    Antarctica
    Let's work back up 1 bit at a time. -84db, -78db, -66db, etc.

    RDAC 1kHz -84db-1.jpg
    RDAC 1kHz -78db-1.jpg
    RDAC 1kHz -66db-1.jpg

    Interesting... low level output from the 4 LSD toggles is quite good. @ultrabike: any ideas?
     
    Last edited: Jun 4, 2018
    ultrabike and Vtory like this.
  3. purr1n

    purr1n Finding his inner redneck

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    68,331
    Dislikes Received:
    112
    Trophy Points:
    113
    Location:
    Antarctica
  4. Hands

    Hands Overzealous Auto Flusher - Measurbator

    Staff Member Friend MZR
    Joined:
    Sep 27, 2015
    Likes Received:
    11,357
    Dislikes Received:
    40
    Trophy Points:
    113
    Location:
    Colorado
    Home Page:
    Most curious to see what the FR looks like at the top end as well as impulse or square wave ringing characteristics to see if we can better understand the type of filter they've inplemented.
     
    cskippy likes this.
  5. Vtory

    Vtory Illogical Spock

    Friend MZR
    Joined:
    Sep 12, 2016
    Likes Received:
    6,317
    Dislikes Received:
    39
    Trophy Points:
    113
    Location:
    Greater Philadelphia Area, PA
    This is what I found in the homepage

    Seems that there is a big gap between measurements and subjective perceptions. Measurement results are neither very surprising to me nor discouraging me. That's what I initially assumed for cost-limited discrete r2r implementations.. (Not as extremely as S19 though).

    Based on the distortion results posted, can guess what 16b/24b -90dbFS sine waves would look like. But will be happy if Marv posts additional data points.

    I kinda have a memory that MSB's (fucking price) dac sounded much better than JA's measurements suggested. Maybe this is a scale-downed version.
     
    Last edited: Jun 5, 2018
  6. CEE TEE

    CEE TEE Free Agent

    Staff Member Friend IEMW
    Joined:
    Sep 25, 2015
    Likes Received:
    1,429
    Dislikes Received:
    0
    Trophy Points:
    93
    Location:
    California Bay Area (San Francisco)
    Mass production units will measure cleaner due to some additional shielding which will be applied.
    After feedback on the first power supply and regulation implementations, this was re-worked.
    Shouldn't change audible sound though.
     
  7. monacelli

    monacelli Friend

    Friend
    Joined:
    Jul 23, 2017
    Likes Received:
    1,588
    Dislikes Received:
    0
    Trophy Points:
    93
    Location:
    The Land of Enchantment
    Thanks, @purr1n. I’m scratching my head over this one a little. @ultrabike is better at interpreting these than I am, but this plot seems to imply about ~10.4 bits (ENOB) at around 0 dBFS. For sure I don’t expect a discrete R-2R DAC to measure like a D-S, but the subjective impressions and the measurements are a little farther apart than I would expect. I wonder if this is a USB vs. SPDIF thing, or if there is something else going on.
     
  8. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Friend MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    9,620
    Dislikes Received:
    20
    Trophy Points:
    113
    Location:
    Irvine CA
    10.5 bits may actually sound not as bad as it seems. SINAD here may be limited by wideband noise inherent in the R-2R architecture used (not narrowband harmonic distortion/noise). If that is the case, it's not easy to tell how things will sound just by looking at the ENOB.

    This is going to sound a bit philosophical and stream-of-consciousness like, but to distinguish a set of tones, even an FFT needs a set of N-point blocks. It needs to integrate in time the samples in order to "identify" the tones in frequency. I believe our brains also need to integrate samples, in some way, to distinguish sound. Wideband random noise tends to integrate to nothing if zero mean. Not so with time correlated narrow band noise like harmonic distortion. This "integration" may dig more "bits" so to speak (sort of what some folks refer to as "coding gain" in "spread-spectrum"). But who knows.

    Again, in a nutshell, hard to say how 10.5 bits (ENOB) is going to sound. It likely depends on the type of noise limiting the ENOB. This is one reason I kind of criticized Amir and his performance characterization using Amir-bits. The other is the fact that Amir-bits are not even ENOB.
     
    Last edited: Jun 5, 2018
    Dzerh, monacelli, Lyander and 2 others like this.
  9. purr1n

    purr1n Finding his inner redneck

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    68,331
    Dislikes Received:
    112
    Trophy Points:
    113
    Location:
    Antarctica
    Let me run more tests just to be sure...
     
  10. purr1n

    purr1n Finding his inner redneck

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    68,331
    Dislikes Received:
    112
    Trophy Points:
    113
    Location:
    Antarctica
    RDAC FR-1.jpg
     
    Torq and dmckean44 like this.
  11. purr1n

    purr1n Finding his inner redneck

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    68,331
    Dislikes Received:
    112
    Trophy Points:
    113
    Location:
    Antarctica
    Just thinking out loud, but the 10.5 bits is for a LOUD 0dbFS signal. Wondering if looking at "bits" in this fashion might not be reflective of real musical content. Maybe for brickwalled recordings at high compressed / high volumes. But the fact is, we don't listen to 0dbFS signals. What's an average dbFS or VU of a musical recording? How does volume change from passage to passage? These considerations should in effect change at how we look at number of bits. As we can see in the THD+N plot. the RDAC performs extremely well up from -110dbFS to around -45dbFS. Maybe this is why the derived ENOB in the traditional sense doesn't totally work here. For many soft and medium volume passages of highly dynamic recordings, the RDAC wouldn't perform less well than its SD brethren.

    None of these measurements are surprising for a discrete ladder using 0.05% tolerance parts. However, the lower levels signals are performing better than expected. The higher levels worse than expected (still better than S19 discrete ladder DAC, at least for the lower harmonics). Scratching my head on this one. Maybe the designer of the RDAC can give us some insights.
     
    Last edited: Jun 7, 2018
    Vtory likes this.
  12. Vtory

    Vtory Illogical Spock

    Friend MZR
    Joined:
    Sep 12, 2016
    Likes Received:
    6,317
    Dislikes Received:
    39
    Trophy Points:
    113
    Location:
    Greater Philadelphia Area, PA
    upload_2018-6-7_3-29-9.png

    This is just one example of highly compressed modern track (Lady Gaga's Bad Romance). Without additional digital gains of the transport, it seems that dac will utilize from -10dBFS to 0dBFS (0.8Vrms to 2.5Vrms). As your distortion spectrum suggests, there will be a lot of (at least more than Modi 2) harmonics in those output voltage range. Not sure if that's associated with perceptive sweetness commonly reported by the three reviews.


    upload_2018-6-7_3-37-35.png

    This is a less compressed example - Hotel California (live) from Hell Freeze Over. Roughly -16 to 0 dbFS.

    PS. Of course short term crest factors might be bigger, but no more than 30dB in the above examples.
     
    Garns, monacelli and Cspirou like this.
  13. Magnetostatic_Tubephile

    Magnetostatic_Tubephile Friend

    Friend
    Joined:
    Dec 27, 2017
    Likes Received:
    783
    Dislikes Received:
    44
    Trophy Points:
    93
    Location:
    GoogleChrome, EU
    I thought that this was Linear phase OS DAC, am I missing something with that top end rolloff aka NOS? Or does it only do 2x oversampling?

    (Apologies if my question is dumb.)
     
  14. Vtory

    Vtory Illogical Spock

    Friend MZR
    Joined:
    Sep 12, 2016
    Likes Received:
    6,317
    Dislikes Received:
    39
    Trophy Points:
    113
    Location:
    Greater Philadelphia Area, PA
    FR response of rdac looks like slow-roll-off (both linear and minimal phase filters are almost indistinguishable in fr plot), as opposed to sharp-roll-off. AFAIK roll-off of NOS generally starts much earlier (at 6k or so).
     
    purr1n likes this.
  15. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Friend MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    9,620
    Dislikes Received:
    20
    Trophy Points:
    113
    Location:
    Irvine CA
    Yes. Performance at low levels seems to be limited by thermal noise, and not by distortion.

    I think ENOB at 0 dBFS is a measure of total dynamic range. How relevant 10.5 bits is with real music content? Perhaps -45 dBFS is a bit low for the top end of the primo amplitude range for available content, but perhaps not catastrophic.
     
  16. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Friend MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    9,620
    Dislikes Received:
    20
    Trophy Points:
    113
    Location:
    Irvine CA
    Yes. It seems it maybe NOS given the 2 dB loss at 20 kHz.

    I think all audio signals should be 2x oversampled, even in NOS.
     
    Vtory likes this.
  17. Hands

    Hands Overzealous Auto Flusher - Measurbator

    Staff Member Friend MZR
    Joined:
    Sep 27, 2015
    Likes Received:
    11,357
    Dislikes Received:
    40
    Trophy Points:
    113
    Location:
    Colorado
    Home Page:
  18. purr1n

    purr1n Finding his inner redneck

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    68,331
    Dislikes Received:
    112
    Trophy Points:
    113
    Location:
    Antarctica
    Supposedly there's a sample rate converter. No idea to what frequency. The low pass frequency seems low. Even a soft knee filter at x2 shouldn't roll off that early. Probably a choice on the part of the designer.
     
  19. purr1n

    purr1n Finding his inner redneck

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    68,331
    Dislikes Received:
    112
    Trophy Points:
    113
    Location:
    Antarctica
    Exactly, we'd like to see the THD+N get progressively lower for signals with peaks higher than -45dbFS, but in light of the analysis of the music files from @Vtory, the situation doesn't seem catastrophic. I'm geeked because this opens up "new" ideas, new ways of looking at measurements, taking into account realistic usage.

    The funny thing is that good quality recordings without the loudness, the compression, are going to do better on the RDAC.
     
    LetMeBeFrank, ultrabike and Vtory like this.
  20. purr1n

    purr1n Finding his inner redneck

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    68,331
    Dislikes Received:
    112
    Trophy Points:
    113
    Location:
    Antarctica
    Reality check for the AverLAB results in post #1 using the Focusrite. I think the initial results from the AverLAB are indeed legit. Note possible ground loop /hum issues here with the Focusite - hence the 60Hz spike and odd harmonics of it - not seen on the AverLAB. Again, nothing too unexpected for a discrete resistor ladder.

    RDAC 1kHz.png
     
    Last edited: Jun 7, 2018

Share This Page