Denafrips Ares 2 R2R DAC Measurements and 'NOS' Investigation

Discussion in 'Digital: DACs, USB converters, decrapifiers' started by GoldenOne, Jul 12, 2021.

  1. GoldenOne

    GoldenOne Friend

    Friend
    Joined:
    Jul 17, 2020
    Likes Received:
    611
    Trophy Points:
    93
    Location:
    England
    Denafrips Ares 2: https://www.denafrips.com/ares

    Subjective review here:


    Test Setup:
    - Audio Precision APx555 B-Series Analyzer with 200kOhm input impedance
    - Van Damme Star-Quad shielded SPC XLR cables with Amphenol Connectors
    - USB Source: AMD PC via ifi iGalvanic 3.0
    - Audioquest Carbon USB and SPDIF cables
    - All measurements shown in this post are taken with USB input, 44.1khz sample rate and 'sharp' filter unless otherwise noted
    - Device was turned on and left to warm up for 24 hours before testing
    - Full reports containing additional data and test configurations are attached
    - Exact analyzer/filter configurations for each measurement are detailed in the full reports
    - APx Project is included in the reports folder should anyone wish to repeat this set of measurements or use it for another DAC

    This DAC is an odd one. Subjectively whilst it's too soft sounding for me to want it as an 'only' dac, I do quite like it for many genres.
    For an R2R dac especially at this pricepoint it seems to do very well. BUT, the claim that it is a NOS dac is absolutely untrue. The 'NOS' mode on this dac is simply linear interpolation at 768khz/705.6khz. It is not NOS.

    Additionally, whilst measuring this DAC there have been some....oddities which haven't occurred on other devices. It seems that Denafrips is doing a fair amount of DSP on this DAC which hinders some measurements.
    John Atkinson of stereophile found that he was completely unable to even do a linearity test as the transfer function made it present a strange stepped zigzag result.
    Whilst testing this device I found that swept sine tests behaved unusually, with frequency response results often looking either outright wrong or with a strange 'rippling' toward the end which differed in appearance dependent on the length of the sweep.

    Additionally the THD+N and IMD values change depending on the level, which whilst not necessarily intentional, might help to 'inflate' this device's Dynamic range test results by about 10dB.

    If there are any other measurements you'd like to see they're most likely in the reports. But if not, let me know and I'll add them to my sequence for future tests.

    If anyone sees any issues with any measurements or configs please let me know and they'll be addressed.

    FULL REPORTS:

    Reports available here: https://drive.google.com/drive/folders/1Lvy6m79jos2vK-kiipRGUW8aSd0nMpX7?usp=sharing

    Reports contain the following measurements:

    - 1khz Sine View
    - 15khz Sine View
    - -90.31dBfs Sine View
    - Phase Response
    - Filter Ultrasonic Attenuation
    - 20hz-90khz RMS Noise Level
    - Device Output Level (Vrms)
    - 20hz-20khz Noise RMS Level
    - Idle Noise FFT
    - 1khz FFT (0dBfs and -3dBfs)
    - 50hz FFT (0dBfs and -3dBfs)
    - Effective Number of Bits (0dBfs and -3dBfs)
    - THD+N (0dBfs and -3dBfs)
    - THD+N / Frequency
    - Dynamic Range (AES17)
    - SNR
    - IMD (SMPTE)
    - 50hz/7khz IMD SMPTE FFT
    - IMD SMPTE vs Output Level
    - Linearity
    - Linearity (No Bandpass)
    - SINAD/Output Level
    - Crosstalk / Frequency
    - DC Offset when Active
    - DC Offset when Idle
    - 44.1khz Jitter
    - 48khz Jitter
    - 90khz Bandwidth test
    - 32 Tone FFT
    - 1.2Mhz Bandwidth Noise FFT
    - 1.2Mhz Bandwidth 1khz FFT (0dBfs and -3dBfs)



    1khz 0dBfs:

    upload_2021-7-12_15-47-51.png

    1khz -3dBfs:

    upload_2021-7-12_15-48-13.png


    THD+N / Frequency:
    (Note: The dip above 10khz is likely not due to this device, this is something that occurs on many devices I've tested and I think it's something to do with the analyzer, will talk to AP about it)

    upload_2021-7-12_15-55-1.png


    Linearity:

    upload_2021-7-12_15-56-10.png

    Interesting to note that on the Ares 2, Linearity without a bandpass filter looks almost the same, which is unusual, most dacs the linearity without bandpass filter veers upwards due to noise.

    upload_2021-7-12_15-56-54.png


    SINAD(THD+N) / Output Level:

    Strange behaviour here with a couple 'jumps' leading up to -60dB.
    This is a little concerning as it could imply that the AES17 dynamic range measurement, which uses a -60dB sine, may not give a fully realistic result here.

    upload_2021-7-12_15-57-34.png
    upload_2021-7-12_15-57-43.png
    upload_2021-7-12_15-59-32.png

    IMD (SMPTE) vs Output Level:

    upload_2021-7-12_15-59-59.png
    upload_2021-7-12_16-0-6.png
     
    • Like / Agree Like / Agree x 14
    • List
    Last edited: Jul 12, 2021
  2. GoldenOne

    GoldenOne Friend

    Friend
    Joined:
    Jul 17, 2020
    Likes Received:
    611
    Trophy Points:
    93
    Location:
    England
    Dynamic Range (AES17):

    upload_2021-7-12_16-6-1.png

    SNR:

    upload_2021-7-12_16-6-15.png

    Crosstalk:

    upload_2021-7-12_16-6-57.png

    -90.31dBfs Sine (Recorded with 192khz ADC sample rate):

    upload_2021-7-12_16-7-50.png

    Filter Ultrasonic attenuation:

    upload_2021-7-12_16-8-10.png

    Impulse Response:

    Sharp:
    cSepz4e7Mj.png

    Slow:
    NSoEq3pVo7.png

    "NOS":
    d3o9X5rrIK.png

    "NOS" Close zoom:
    3ud9Wgr1ST.png
     
    • Like / Agree Like / Agree x 2
    • List
    Last edited: Jul 12, 2021
  3. GoldenOne

    GoldenOne Friend

    Friend
    Joined:
    Jul 17, 2020
    Likes Received:
    611
    Trophy Points:
    93
    Location:
    England
    Idle Noise Spectrum upto 1.2Mhz:

    upload_2021-7-12_16-11-13.png

    44.1khz Jitter (On USB):

    NOTE: Jitter measurements were the same on both SPDIF and USB, suggesting that the buffering system is in place on both input methods and that the performance of the source device is unlikely to impact things much.

    upload_2021-7-12_16-11-55.png

    48khz Jitter (On USB):

    upload_2021-7-12_16-12-10.png

    Multitone:

    upload_2021-7-12_16-12-22.png
     
    • Like / Agree Like / Agree x 4
    • List
    Last edited: Jul 12, 2021
  4. GoldenOne

    GoldenOne Friend

    Friend
    Joined:
    Jul 17, 2020
    Likes Received:
    611
    Trophy Points:
    93
    Location:
    England
    REPORTS:

    Reports available here: https://drive.google.com/drive/folders/1K8VmpPChqwJfOrZD_Zo9iC2f91AOCK8O?usp=sharing

    Reports contain the following measurements:

    - 1khz Sine View
    - 15khz Sine View
    - -90.31dBfs Sine View
    - Phase Response
    - Filter Ultrasonic Attenuation
    - 20hz-90khz RMS Noise Level
    - Device Output Level (Vrms)
    - 20hz-20khz Noise RMS Level
    - Idle Noise FFT
    - 1khz FFT (0dBfs and -3dBfs)
    - 50hz FFT (0dBfs and -3dBfs)
    - Effective Number of Bits (0dBfs and -3dBfs)
    - THD+N (0dBfs and -3dBfs)
    - THD+N / Frequency
    - Dynamic Range (AES17)
    - SNR
    - IMD (SMPTE)
    - 50hz/7khz IMD SMPTE FFT
    - IMD SMPTE vs Output Level
    - Linearity
    - Linearity (No Bandpass)
    - SINAD/Output Level
    - Crosstalk / Frequency
    - DC Offset when Active
    - DC Offset when Idle
    - 44.1khz Jitter
    - 48khz Jitter
    - 90khz Bandwidth test
    - 32 Tone FFT
    - 1.2Mhz Bandwidth Noise FFT
    - 1.2Mhz Bandwidth 1khz FFT (0dBfs and -3dBfs)
     
    • Like / Agree Like / Agree x 1
    • List
    Last edited: Jul 12, 2021
  5. GoldenOne

    GoldenOne Friend

    Friend
    Joined:
    Jul 17, 2020
    Likes Received:
    611
    Trophy Points:
    93
    Location:
    England
    NOS or NOT?

    I first became a little suspicious/confused when I noticed that the Ares 2 performs near identically (in fact a hair better) in NOS than it does in OS mode.

    This is very unusual and almost all NOS dacs will perform slightly to considerably worse in NOS than they do with OS. But not here.....

    NOS:
    upload_2021-7-13_2-47-29.png

    Sharp OS:
    upload_2021-7-13_2-47-46.png

    So, I looked at the impulse response and sure enough we aren't seeing a NOS IR at all, it's a triangle!
    [​IMG]

    We can also see a strange stepping behaviour which is at either 705.6khz or 768khz depending on the input sample rate. Suggesting 16x oversampling is occuring.

    Looking on a wideband FFT even when idle we can see a large spike which is either at again 768khz or 705.6khz
    upload_2021-7-13_2-49-52.png

    Looking at a 15khz sine also shows linear interpolation behaviour:
    upload_2021-7-13_2-50-58.png

    Additionally, the FR is about -7.9dB down by 22.05khz, which is exactly what we'd expect for linear interpolation. And not the -3.9dB we'd expect for true NOS.

    Everything seems to point to this being linear interpolation. But whatever it is, it certainly isn't NOS.
     
    • Like / Agree Like / Agree x 4
    • Epic Epic x 2
    • List
    Last edited: Jul 12, 2021
  6. GoldenOne

    GoldenOne Friend

    Friend
    Joined:
    Jul 17, 2020
    Likes Received:
    611
    Trophy Points:
    93
    Location:
    England
    Reserved 5
     
  7. GoldenOne

    GoldenOne Friend

    Friend
    Joined:
    Jul 17, 2020
    Likes Received:
    611
    Trophy Points:
    93
    Location:
    England
    Reserved 6
     
  8. Marvey

    Marvey Super Friend

    Staff Member Friend BWC
    Joined:
    Jun 6, 2021
    Likes Received:
    1,284
    Trophy Points:
    93
    [​IMG]

    Check bandwidth of the test. The sudden dip after 10kHz can happen if the bandwidth is set to 20kHz. Trying setting bandwidth to 90kHz and turn off any filters. Makes sense right, especially if the distortion goes higher orders past H2. A 20kHz cap will lower the distortion numbers in such a case.

    [​IMG]

    Looks like they glued two ladders together. It's not an uncommon approach in the "NOS/R2R DAC arena" to get better numbers at lower signal levels instead of just one ladder.

    Could be interesting to show a spectrum of a -36dbFS sine and compare to a -42dbFS one.

    As for your description of the sound, Denafrips does seem to have a house sound.
     
    • Like / Agree Like / Agree x 2
    • List
    Last edited by a moderator: Jul 12, 2021
  9. GoldenOne

    GoldenOne Friend

    Friend
    Joined:
    Jul 17, 2020
    Likes Received:
    611
    Trophy Points:
    93
    Location:
    England
    That makes sense. I guess the issue then is that setting either the signal path filter to 40khz or a 192khz SR with no filters means that the THD+N values will be higher than with the 20khz filter anyway due to it factoring in stuff above 20khz. And so the level of noise above 20khz could then impact things and perhaps skew the measurement.

    Not sure what the ideal way to address this would be but maybe the easy answer is just include both. I have the wideband noise FFT in the reports too so that should give some context for a 40khz filter measurement.

    Thank you


    Yep that'd definitely make sense. I'm not too sure what to make of that in terms of things like dynamic range as it in some ways could be considered 'cheating' as higher vs lower level signals will behave and likely sound different. But then I guess it's perhaps one just to leave up to our ears to evaluate.

    But most definitely why tests like this are important and why a 1khz tone definitely doesn't tell the whole story!


    Unfortunately it's boxed up and shipped out now so I can't do any further tests. But, I may be able to get some measurements of a terminator plus soon so I'll see if that exhibits the same behaviour and see about doing it on that.

    In regards to their house sound that does seem to be the case and whilst I don't have 'proof', I imagine it's a result of their DSP.
    They've revised the DSP board on their terminator several times without changing the DAC itself, and there are a fair few oddities in the behaviour of this dac that indicate DSP of some sort going on which would definitely contribute to that.

    Whether it's to try to get more accuracy out of the ladders or a deliberate sound 'tuning' I don't know
     
  10. Armaegis

    Armaegis Friend

    Friend BWC
    Joined:
    Sep 27, 2015
    Likes Received:
    4,969
    Trophy Points:
    113
    Location:
    Winnipeg
    It's something like a 3db or 6db improvement for every doubling right?
    I remember seeing builds for the old DDDAC1794 that commonly had 4 boards stacked (only one time did I ever hear of an 8-board stack).
     
  11. Marvey

    Marvey Super Friend

    Staff Member Friend BWC
    Joined:
    Jun 6, 2021
    Likes Received:
    1,284
    Trophy Points:
    93
    Different kind of stacking. You are referring to paralleling DAC chips to linearize through averaging. Similar to this approach would be hardware balanced with one DAC chip on each side.

    This is gluing two DAC ladders together. One ladder does 0 to -39dbFS. The other does -40 and below. The second ladder is run at a lower voltage. Some middle bit of the digital code word then becomes the most significant bit of the second ladder.

    I'm not convinced this method sounds good. Lots of signal around -40dbFS with real music. We are suddenly changing distortion profiles as we go up and down the "glue" point.

    The UltraAnalog modules of years' past also took this approach - I think. Would be curious to see how they implemented it. I liked the sound of the UltraAnalog stuff. The modern "NOS glued DAC ladder" approach I am not totally convinced of yet.
     
    • Like / Agree Like / Agree x 5
    • List
  12. Beefy

    Beefy Almost "Made"

    Joined:
    Apr 10, 2021
    Likes Received:
    256
    Trophy Points:
    63
    Location:
    Canada
    The seemingly better way to glue to ladders together is with the sign-magnitude approach. Instead of switching from one ladder to the other at a particular amplitude, you have one ladder that takes the + signal, and one ladder that takes the - signal. This is the way that the PCM1704 did it, now Soekris, MSB, et al. Betcha don't see a distortion hump with any of those.
     
    • Like / Agree Like / Agree x 2
    • List
  13. 7seven

    7seven Acquaintance

    Joined:
    Oct 28, 2018
    Likes Received:
    33
    Trophy Points:
    18
    Location:
    Ireland
    Actually all the Soekris DACs are segmented types like this. sign magnitude and then 'thermometer' DAC for MSBs.
     
    • Like / Agree Like / Agree x 1
    • List
  14. Beefy

    Beefy Almost "Made"

    Joined:
    Apr 10, 2021
    Likes Received:
    256
    Trophy Points:
    63
    Location:
    Canada
    Yes and no. I think segmentation on a single ladder with a single Vref is quite different to what Marv is questioning what might be happening here. But without knowing the exact architecture of the Denafrips, it's hard to know.

    Either way, there's definitely a transition here that I've not seen on other R-2R DAC measurements. Then again, I've not seen many DACs measured this well!
     
  15. earnmyturns

    earnmyturns Smartest friend

    Friend
    Joined:
    Sep 25, 2016
    Likes Received:
    2,923
    Trophy Points:
    113
    Location:
    Palo Alto
    Home Page:
    • Like / Agree Like / Agree x 2
    • List
  16. Beefy

    Beefy Almost "Made"

    Joined:
    Apr 10, 2021
    Likes Received:
    256
    Trophy Points:
    63
    Location:
    Canada
    • Like / Agree Like / Agree x 1
    • List
  17. GoldenOne

    GoldenOne Friend

    Friend
    Joined:
    Jul 17, 2020
    Likes Received:
    611
    Trophy Points:
    93
    Location:
    England
    Also an interesting thing I saw on the Ares 2 in 'NOS'

    If you do an impulse response test, then you get the triangular IR shown above. But if you take the derived IR from the transfer function measurement on the AP, it shows a sinc IR.

    This does not happen on an actual NOS dac, or even on ADI-2 in 'NOS' os mode
     
  18. cameng318

    cameng318 Friend

    Friend
    Joined:
    Jun 14, 2020
    Likes Received:
    208
    Trophy Points:
    33
    Location:
    NYC
    Maybe it has a filter with rectangular window function. It would turn a rectangle impulse into a triangle.

    The derived IR might be bandwidth limited to show the sharp corners, so it would look like sinc.
     
  19. NekoAudio

    NekoAudio Acquaintance

    Joined:
    Apr 18, 2016
    Likes Received:
    57
    Trophy Points:
    18
    Location:
    San Jose, California
    Home Page:
    FWIW, the data above 20kHz should still be considered as relevant for the analog output measurement as it has a potential impact on the downstream input circuit or to emit EMI. UL testing measured my DAC's analog output well into the MHz range in order to determine if it contained any undesirable high frequency signals, with it being my responsibility to address (i.e. filter/remove) any such signal.
     
  20. JeremiahS

    JeremiahS Acquaintance

    Joined:
    Sep 30, 2015
    Likes Received:
    55
    Trophy Points:
    28
    Location:
    Jakarta, Singapore
    Would you say that something like the SFD-1 is a better buy than the Ares 2 since they are priced similar?
     

Share This Page