Schiit Bifrost-2 Technical Measurements

Discussion in 'Source Measurements' started by atomicbob, Sep 26, 2019.

  1. atomicbob

    atomicbob dScope Yoda

    Friend BWC MZR
    Joined:
    Sep 27, 2015
    Likes Received:
    10,304
    Dislikes Received:
    3
    Trophy Points:
    113
    Location:
    On planet
    Schiit Bifrost-2 technical measurements

    Standard Prologue
    If you are unfamiliar with audio measurements please use a search engine with the query:
    "audio measurements" or "audio measurement handbook"
    Look for publications by Richard C. Cabot and also by Bob Metzler, both from Audio Precision. There are other useful publications as well. These will provide basic knowledge.
    Interpretation of the following measurements is beyond the scope of technical measurements posts.


    The data presented were collected as follows:

    1. PrismSound dScope III, picoscope 5243B, Micsig DP10013 balanced probe
    2. Tecnec 75R spdif cable
    3. Balanced XLR cables Belden 1800F with Neutrik 110R AES connectors
    4. Single Ended cables Audioblast HQ-1 1.5 ft with Rean RCA connectors
    5. 100 Kohm load used for measurements
    6. dScope analyzer sample rate 48 KHz unless otherwise noted
    7. DAC 44.1 KHz sample rate, 24 bit depth unless otherwise noted
    9. Audioquest Forest and Schiit Pyst USB cables
    9. Vaunix Lab Brick USB hub
    10. Shielded 14AWG and 16AWG power cables

    Measurements are made in accordance with AES17:2015

    Schiit Bifrost-2
    Measurements commenced after 168 hours of warm-up.
    Measurements were performed over a period of several days.

    Index
    Post 1 - measurement setup description, highlights
    Post 2 - spdif input - Bal XLR outputs part A
    Post 3 - spdif input - SE RCA outputs part B
    Post 4 - USB input - Bal XLR outputs part C
    Post 5 - USB input - SE RCA outputs part D


    Notable highlights:
    The cross-talk behaves as if dual mono, though yggdrasil performs 10 dB better.
    Noise floor and dynamic range are nearly identical Bal or SE.
    Noise floor and dynamic range are exceptional for an 18 bit multibit instrumentation DAC.
    Jitter spurs are nearly as low as yggdrasil.
    Quite an accomplishment for the small footprint and price-point achieved.

    Well done Jason & Mike!


    Setup picture:

    Bifrost-2_small.jpg
     
    Last edited: Sep 26, 2019
    Vtory, Richgard, Hammy and 22 others like this.
  2. atomicbob

    atomicbob dScope Yoda

    Friend BWC MZR
    Joined:
    Sep 27, 2015
    Likes Received:
    10,304
    Dislikes Received:
    3
    Trophy Points:
    113
    Location:
    On planet
    spdif input Bal output measurements part A

    Bifrost-2 A04 THD+N THD nth-HD FFT spdif input Bal output
    20190924 Bifrost-2 A04 THD+N THD nth-HD 4+HD+N 60Hz FFT Bal 14dBu.png

    Bifrost-2 50 + 7000 Hz spdif input Bal output - Left Channel
    FFT_2_L1T14_1_A.png

    Bifrost-2 Gain Linearity spdif input Bal output - Left Channel
    G2_L1T3_1_A.png

    Bifrost-2 THD+N vs Frequency spdif input Bal output - Left Channel
    G2_L1T6_1_A.png

    Bifrost-2 Residual Noise spdif input Bal output - Left Channel
    FFT_2_L1T17_1_A.png

    Bifrost-2 Square Wave -1.5dBFS 10mS/div spdif input Bal output - Left Channel
    20190925-23 Bifrost-2 20 Hz sqr -1_5 dBFS 10 Vpp 10mS div Bal - spdif.png

    Bifrost-2 Square Wave -1.5dBFS 500uS/div spdif input Bal output - Left Channel
    20190925-24 Bifrost-2 20 Hz sqr -1_5 dBFS 10 Vpp 500uS div Bal - spdif.png

    Bifrost-2 Square Wave -1.5dBFS 100uS/div spdif input Bal output - Left Channel
    20190925-25 Bifrost-2 20 Hz sqr -1_5 dBFS 10 Vpp 100uS div Bal - spdif.png

    Complete spdif input Bal output analysis report pdf attached:
     

    Attached Files:

    Last edited: Sep 26, 2019
    Vtory, SVChucko, zonto and 4 others like this.
  3. atomicbob

    atomicbob dScope Yoda

    Friend BWC MZR
    Joined:
    Sep 27, 2015
    Likes Received:
    10,304
    Dislikes Received:
    3
    Trophy Points:
    113
    Location:
    On planet
    spdif input SE output measurements part B

    Bifrost-2 A04 THD+N THD nth-HD FFT spdif input SE output
    20190924 Bifrost-2 A04 THD+N THD nth-HD 4+HD+N 60Hz FFT SE 8dBu.png

    Bifrost-2 50 + 7000 Hz spdif input SE output - Left Channel
    FFT_2_L1T14_1_A.png

    Bifrost-2 Gain Linearity spdif input SE output - Left Channel
    G2_L1T3_1_A.png

    Bifrost-2 THD+N vs Frequency spdif input SE output - Left Channel
    G2_L1T6_1_A.png

    Bifrost-2 Residual Noise spdif input SE output - Left Channel
    FFT_2_L1T17_1_A.png

    Bifrost-2 Square Wave -1.5dBFS 10mS/div spdif input SE output - Left Channel
    20190925-23 Bifrost-2 20 Hz sqr -1_5 dBFS 5 Vpp 10mS div SE - spdif.png

    Bifrost-2 Square Wave -1.5dBFS 500uS/div spdif input SE output - Left Channel
    20190925-24 Bifrost-2 20 Hz sqr -1_5 dBFS 5 Vpp 500uS div SE - spdif.png

    Bifrost-2 Square Wave -1.5dBFS 100uS/div spdif input SE output - Left Channel
    20190925-25 Bifrost-2 20 Hz sqr -1_5 dBFS 5 Vpp 100uS div SE - spdif.png

    Complete spdif input SE output analysis report pdf attached
     

    Attached Files:

    Last edited: Sep 26, 2019
    SVChucko, Baten and Sqveak like this.
  4. atomicbob

    atomicbob dScope Yoda

    Friend BWC MZR
    Joined:
    Sep 27, 2015
    Likes Received:
    10,304
    Dislikes Received:
    3
    Trophy Points:
    113
    Location:
    On planet
    reserved 4/6 for USB Bal.

    USB measurements require me to configure a Win10 PC with dScope and PicoScope measurement SW installed. This may take a few more days or possibly several weeks depending.
     
  5. atomicbob

    atomicbob dScope Yoda

    Friend BWC MZR
    Joined:
    Sep 27, 2015
    Likes Received:
    10,304
    Dislikes Received:
    3
    Trophy Points:
    113
    Location:
    On planet
    reserved 5/6 for USB SE.

    Same delay as for post 4 above.
     
  6. atomicbob

    atomicbob dScope Yoda

    Friend BWC MZR
    Joined:
    Sep 27, 2015
    Likes Received:
    10,304
    Dislikes Received:
    3
    Trophy Points:
    113
    Location:
    On planet
    reserved for corrections or additions.
     
  7. AudioNut

    AudioNut Acquaintance

    Joined:
    Oct 2, 2018
    Likes Received:
    58
    Dislikes Received:
    15
    Trophy Points:
    18
    Location:
    Ottawa, ON
    Thanks for posting the full measurements in PDF form. Those are really eye-opening IMHO, especially some of the ones you haven't posted here. The 19+20kHz measurement shows a -64dB distortion product at 1kHz... WTF?
     
  8. atomicbob

    atomicbob dScope Yoda

    Friend BWC MZR
    Joined:
    Sep 27, 2015
    Likes Received:
    10,304
    Dislikes Received:
    3
    Trophy Points:
    113
    Location:
    On planet
    Be careful to watch the y-axis units and note generator settings. I am presenting most measurements with dBu on the y-axis not dBFS. dBu provides absolute values with respect to 0.775 Vrms. For example Bifrost-2 balanced output is +14.35 dBu at 0 dBFS. In the case of IMD 19+20KHz the reported stimulus in dBu must be subtracted from the dBu on the graph to find dBFS.

    Here is one example where I have changed the y-axis to dBFS. Note the 11.290 dBu stimulus for channel A (left.)

    Bifrost-2 19+20 KHz spdif input Bal output:
    Bifrost-2_spdif_44K_Bal_IMD_dBFS.png
     
    Sqveak likes this.
  9. AudioNut

    AudioNut Acquaintance

    Joined:
    Oct 2, 2018
    Likes Received:
    58
    Dislikes Received:
    15
    Trophy Points:
    18
    Location:
    Ottawa, ON
    Thanks for the clarification, but changing the axis to dBFS is misleading unless you're also using a full-scale stimulus. That 1kHz distortion product is almost certainly going to be higher than shown if you repeat the test with a full-scale stimulus; we just can't tell by how much.

    This has got to be easily audible, you almost don't need measurement equipment to hear this level of IMD. I'm curious to try it when I get home.

    (I just looked at the same test in Stereophile's measurements of the Yggdrasil, and interestingly it does also have a 1kHz peak, but it's far more benign on the Yggdrasil and much less likely to be audible there.)
     
  10. atomicbob

    atomicbob dScope Yoda

    Friend BWC MZR
    Joined:
    Sep 27, 2015
    Likes Received:
    10,304
    Dislikes Received:
    3
    Trophy Points:
    113
    Location:
    On planet
    Please refer to the signal generator stimulus reported. Note that both the 19 KHz and 20 KHz are -6.03 dBFS each, and the real time signal analyzer reports 11.290 dBu for the combined total in post #8 above. Now remember that the very first bit of data in the pdf file show Bifrost-2 as 0 dBFS = 14.35 dBu for balanced output. So the twin tone stimulus is very near maximum output. If the two tone levels are too high, then the summation will result in clipping of the source signal generated.

    Below you will find yet another representation of the same measurement, with all values reported in dBFS and dB below 0 dBFS. Please note that the signal generator is running two tones each at -6.03 dBFS just as in post #8 above and all of the attached pdfs.

    The 1000 Hz IMD tone is -64 dBu which is approximately 489 uV (microvolts.) Listening with a headphone such as HD650 having a sensitivity of 98 dBSPL / mW will result in 37 dBSPL. I would posit this inaudible when listening to music at 65 to 75 dBSPL which is considered quiet listening.

    I have annotated this one. Please note the generator settings, which are the same as all the others I've posted for IMD.

    Bifrost-2 19+20 KHz spdif input Bal output:
    Bifrost-2_spdif_44K_Bal_IMD_dBFS_dBFS_v2_annotated.png
     
    Sqveak likes this.
  11. mkozlows

    mkozlows Friend

    Friend
    Joined:
    Oct 6, 2015
    Likes Received:
    351
    Dislikes Received:
    1
    Trophy Points:
    63
    Not sure if I'm understanding that right, but is that measuring the same thing as in this Schiit-provided measurement? Because if so, then I think this comes in pretty close to what you're measuring -- 75-77ish dB at a -6 dBFS input -- so that seems to check out pretty well...

    [​IMG]
     
    atomicbob likes this.
  12. AudioNut

    AudioNut Acquaintance

    Joined:
    Oct 2, 2018
    Likes Received:
    58
    Dislikes Received:
    15
    Trophy Points:
    18
    Location:
    Ottawa, ON
    There's no point in switching to nominal voltage values here; it adds nothing to the conversation. Perceived distortion is relative to playback levels.

    I only had five minutes last night to run a subjective test with my Bifrost 2, but with the volume of my preamp (OG Saga passive) adjusted to typical playback levels for music on speakers, there's a clearly audible 1kHz tone with a -6dB 19+20kHz 44.1kHz dual sine stimulus. It's not subtle at all.

    It may be a problem with the Bifrost 2's digital filter, because that 1kHz tone seemed to go away with test tones at higher sample rates, but I only had a couple minutes to try that; I'll experiment with this more this weekend.
     
    allegro likes this.
  13. atomicbob

    atomicbob dScope Yoda

    Friend BWC MZR
    Joined:
    Sep 27, 2015
    Likes Received:
    10,304
    Dislikes Received:
    3
    Trophy Points:
    113
    Location:
    On planet
    A few more data points.

    Bifrost-2 19+20 KHz vs amplitude sweep spdif input Bal output:
    Bifrost-2_spdif_44K_Bal_IMD_vs_amplitude_dBFS_dBFS.png
    @mkozlows this agrees with the measurement provided by @schiit


    Bifrost-2 IMD vs frequency sweep (5+6 to 19+20 KHz) spdif input Bal output:
    Bifrost-2_spdif_44K_Bal_IMD_vs_frequency_dBFS_dBFS.png
    This demonstrates IMD decreases as the frequency of the stimulus is lowered.

    Attached is a light jazz excerpt with a -70 dBFS modulated 1 KHz tone embedded. How audible is it to anyone? Can anyone decode the message by listening only, no DSP processing and analysis tricks?
     

    Attached Files:

    Last edited: Sep 27, 2019
    Psalmanazar likes this.
  14. AudioNut

    AudioNut Acquaintance

    Joined:
    Oct 2, 2018
    Likes Received:
    58
    Dislikes Received:
    15
    Trophy Points:
    18
    Location:
    Ottawa, ON
    That's not a realistic representation of how this level of IMD would manifest with a musical signal. Music is not a pair of sines. This type of IMD would play out as an array of additional dynamic tones superimposed under the music.

    Some of these would be masked, others wouldn't, but it would play out as a lack of "blackground" combined with odd overtones, the latter most easily audible with solo acoustic instruments. Before you posted these measurements yesterday, I had posted subjective impressions of the Bifrost 2 that turned out to be pretty closely in line with what your measurements would predict. Not surprised at all by what you've measured to be honest.
     
  15. atomicbob

    atomicbob dScope Yoda

    Friend BWC MZR
    Joined:
    Sep 27, 2015
    Likes Received:
    10,304
    Dislikes Received:
    3
    Trophy Points:
    113
    Location:
    On planet
    This is the first item to which I responded. It was clear you missed the y-axis units in dBu rather than dBFS, understandable given most data present with dBFS as y-axis, and the correct dBFS value would be 11 dB lower at around -75 dBFS.

    This is the second item to which I responded. You implied I was being misleading, yet the signal generator stimulus signal levels were clearly on the graphics and also in the pdfs allowing determination that stimulus was at -6 dBFS each tone with an aggregate of -3 dBFS, also clearly shown on analyzer level metering.

    Absolute values aid those with the technical chops to estimate absolute sound levels present in the measurements. Then you tested with a two tone sinusoidal stimulus and noted a non-subtle response. So I inferred you were amenable to using sines to detect IMD and make comments upon audibility.

    The test excerpt flac file I posted should have been an egregious example. 1000 Hz is not on the musical scale. B5 is 987.77 Hz and C6 is 1046.50 Hz. Falling in between the low level 1000 Hz signal should sound sour. Additionally the 1000 Hz tone was modulated, not rhythmically synchronized with the music. This should further enhance audibility. However it is at -70 dBFS embedded in a realistic musical excerpt, chosen specifically to have minimal pyscho-acoustic maskers around 1000 Hz. If this egregious example isn't audible, then I posit it is even less likely to hear musically related IMD, except possibly in the twin-tone listening experience which you conducted.

    For those owning Bifrost-2 wishing to experience and judge personally the twin-tone IMD test, I have attached three more short files.

    Disclaimer: set levels wisely. I am not responsible for blown tweeters, headphones, ears, etc.

    Protocol:

    1. download files

    2. using familiar music set listening levels

    3. play training flac file: IMD 1KHz -10dBFS 10S 44x16x2 result example.flac
    to calibrate listening for 1000 Hz. This is the tone you are attempting to hear during The TEST.
    Listen only briefly at high levels or a basilar membrane threshold shift will occur and interfere with the test
    making it harder to hear the low level 1000 Hz tones

    4. play flac file: IMD 1KHz -70dBFS 10S 44x16x2 result example.flac
    this is the sound you are attempting to perceive at approximately realistic relative level

    5. The TEST - play flac file: IMD 19+20KHz -3dBFS 10S 44x16x2.flac
    You will experience a presence from the two stimulus tones. Ignore those and listen for the 1000 Hz tone heard during training.

    Can you hear the 1000 Hz IMD on your Bifrost-2 DAC?
    Is it subtle, obvious, somewhere in-between?
     

    Attached Files:

  16. purr1n

    purr1n Finding his inner redneck

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    50,968
    Dislikes Received:
    84
    Trophy Points:
    113
    Location:
    Antarctica
    I would be very cautious before associating this IMD figure with "lack of 'blackground'". The Bifrost 2 has a subjectively blacker canvas than Yggdrasil A1, known for sounding "grey", but which scores much better for the 0dbFS 19kHz+20kHz (-6.03db for each signal so the sum is 0db) dual-tone test with resultant 1kHz (IMD).

    Read any extraneous signal lower than around -70dbFS as having any certain audio quantities at your own peril. Especially, if datasets are lacking that can correlate measurement behavior to a perceived sound phenomenon. Science requires the establishment of hypothesis, and accumulated evidence to establish a theory.
     
    Last edited: Sep 27, 2019
    Lyander and Psalmanazar like this.
  17. Psalmanazar

    Psalmanazar Non Voting Friend

    Slaytanic Cliff Clavin
    Joined:
    Sep 27, 2015
    Likes Received:
    3,072
    Dislikes Received:
    103
    Trophy Points:
    113
    None of these have any correlation to perceived fidelity or if you yourself or most other people will like how the DAC sounds or not. You have to deafen yourself or stack tons of channels to raise the noise floor of the DAC dramatically and the noise of the mics and pres and air conditioning units you are summing together will rise first. Now in some ridiculously overdone orchestral recording or a movie that stuff matters but they're going to be way more concerned with pre amp noise floors.

    There's a presentation from one of the ESS guys where he points out that DACs with -80db THD+N floors sometimes measurably outperform -120 db ones in real world use because those stats are fixed, steady state noise measurements. Some of the SPL DACs measure with only something like -80 db for THD+N and they sound killer. Hardly anyone would take low end Focusrites over SPL and Schiit. Those measurements don't correlate at all with the rich but WIDE open SPL sound. There are DACs that measure similarly that are narrow, sharp, and grainy lo-fi. There are DACs that measure with amazing numbers that sound even worse.
     
    Last edited: Sep 27, 2019
  18. internethandle

    internethandle Rando

    Joined:
    Sep 23, 2017
    Likes Received:
    5
    Dislikes Received:
    0
    Trophy Points:
    3
    Location:
    long beach, ca
    Tried this out last night with Bifrost 2 -> Valhalla 2 -> HD 800's -- not 100% sure I did this correctly but to hear a tone at all while playing "IMD 1KHz -70dBFS 10S 44x16x2 result example.flac" or the TEST file ("IMD 19+20KHz -3dBFS 10S 44x16x2.flac") I have to turn the Valhalla 2's volume knob to about the 1 or 2 o'clock position, and even then it's faint. My normal listening levels are usually between 7 and 9 o'clock. If I go past 9 o'clock with the first training file, it's extremely unpleasant/loud (as it is with most music, for that matter).

    So, if I understand correctly, no I cannot hear the IMD given that I have to turn it up to well above normal listening levels.

    This is with a Bifrost 2 with newest firmware.
     
    Last edited: Nov 30, 2019

Share This Page