Motu M4 Measurements Collection (Noob alert!)

Discussion in 'Audio Science' started by Vtory, Feb 9, 2021.

  1. Vtory

    Vtory Audiophile™

    Pyrate MZR
    Joined:
    Sep 12, 2016
    Likes Received:
    10,831
    Trophy Points:
    113
    Location:
    East Coast
    Disclaimers
    • The title mainly implies device (dac or amp) measurements performed by Motu M4 albeit first posts will focus on loopback configuration (M4 measured by M4) though.
    • Of course noob in the title means me, not readers. lol
    • This is something I'm newly learning, prepare enough salt and patience for any of my posts in this thread. Also be wary of zero control of known confounders.
    Links to some specific posts

    [​IMG]

    Just for a whim, started to mess with another kind of measurements which I haven't done yet.
    Totally unsure about how reliable this device is.. but Motu gave a good impression to sbaf seniors through higher ultralite products, which drove me to one of their budget options, M4.

    After finishing asio setup (super easy) and basic calibration, I rushed to take a quick loopback measurement (bal out to bal in). And below is what I got in the first try.

    [​IMG]

    WTF. Cannot help laughing. Indeed a big head scratcher.
    This shouldn't be true. Too good to believe, really. Something probably went wrong. Should look into more later.

    Any advice regarding typical noob traps will be all appreciated.

    Will update this thread as I collect more data points. Ultimately I want to measure other non acoustical output products (dac, preamp, head amp, etc) via M4 like I did with EARS.

    EDIT: The graph above doesn't make any sense (wrong option chosen). Please refer to the following post (#2)
     
    Last edited: Feb 14, 2021
  2. Vtory

    Vtory Audiophile™

    Pyrate MZR
    Joined:
    Sep 12, 2016
    Likes Received:
    10,831
    Trophy Points:
    113
    Location:
    East Coast
    Figured out what I did wrong.

    upload_2021-2-9_23-18-11.png

    I shouldn't have chosen loopback in the input. It seems to take virtual signal instead of real adc input. Thus, choosing loopbacks made results ridiculously good (noise free, no physical limitation, etc).

    Below is closer to what I initially expected. Haha..

    [​IMG]
     

    Attached Files:

    Last edited: Feb 9, 2021
  3. Vtory

    Vtory Audiophile™

    Pyrate MZR
    Joined:
    Sep 12, 2016
    Likes Received:
    10,831
    Trophy Points:
    113
    Location:
    East Coast
    Adding some standard measurements. All measured with Motu M4 bal out 3 --> bal in 1 (EDIT: M4's balanced input 1 is not line in but mic in, which involves unnecessary gain/processing -- not appropriate for assessing pure dac performance). Input gain control not used.

    [​IMG]

    [​IMG]

    [​IMG]

    [​IMG]

    I kinda remember I had worse results with older focusrite (many years ago; different computer; different electricity; different software), but still have little idea regarding decentness of this dac/adc. Will try to measure some cheap dacs in the house later.
     
    Last edited: Feb 11, 2021
  4. Armaegis

    Armaegis Friend

    Pyrate BWC
    Joined:
    Sep 27, 2015
    Likes Received:
    7,466
    Trophy Points:
    113
    Location:
    Winnipeg
  5. Vtory

    Vtory Audiophile™

    Pyrate MZR
    Joined:
    Sep 12, 2016
    Likes Received:
    10,831
    Trophy Points:
    113
    Location:
    East Coast
    @Armaegis That's super sweet! I'd love to experiment that myself. Will pm you soon.
     
  6. Vtory

    Vtory Audiophile™

    Pyrate MZR
    Joined:
    Sep 12, 2016
    Likes Received:
    10,831
    Trophy Points:
    113
    Location:
    East Coast
    EDIT: Graphs updated again on 2021-02-13 to make y-axis consistent

    Updating graphs with more appropriate setting. I realized I used mic in rather than line inputs, and that didn't do any justice -- oppositely it worsened the result by limiting input (by 8db!) and adding potential mic circuit shits.. Oh well.

    All the graphs updated now with the setting of line out 3 to line in 3.

    [​IMG]

    [​IMG]

    [​IMG]

    [​IMG]

    [​IMG]

    All the signal strengths adjusted to get 0dbFS(peak) out of dac's line outs.
     
    Last edited: Feb 12, 2021
  7. Vtory

    Vtory Audiophile™

    Pyrate MZR
    Joined:
    Sep 12, 2016
    Likes Received:
    10,831
    Trophy Points:
    113
    Location:
    East Coast
    Something interesting found.

    Let me first show ARTA 1khz FFT measurement (using ASIO in and out).

    (Aside: I'm testing both ARTA and REW atm to find out which one suits better for my purposes)

    [​IMG]

    It doesn't look that different from 1khz fft plot taken from REW. Probably due to various causes, they do show some discrepancy though (ARTA look tad cleaner harmonics and minutely lower distortion summaries). At least I don't want to nitpick that for now.

    ARTA also allow us to use WDM drivers. We all know WDM sucks to ears. But I was curious how they look like in the measurement.

    See below.

    [​IMG]
    (ARTA WDM in and out)

    Omg.. I'd emphasize no hardware changes AT ALL.

    Some bits rounding. Weirdly done jitters. Harmonic characteristics changed too. It's interesting changes well below -100db relative to the fundamental.. but I strongly believe all these associated with our perception more or less.

    To rule out possibilities caused by WDM "input" error, I took measurements with REW+ASIO (only for input). ARTA still generated sine signal via WDM.

    [​IMG]

    This time REW made a cleaner plot. But characters largely similar between the two. For the record, I want to make it clear that both REW and ARTA used the same FFT size (64k) and quite similar averaging method (exponential).

    Wait, didn't REW have the option of "Java" driver? Is that another name of WDM? Let's find out.

    [​IMG]

    LMAFO. What the f*** ...
    Completely no idea what is going on. To me, this looks even worse than WDM... (please correct me if I am wrong)

    I also cross-measured on ARTA as well (REW generating signal via "Java" while ARTA taking ASIO input).

    [​IMG]

    No idea why the noise level beyond 3khz shaped that way. But other than that, strange patterns persisted.

    Takeaways
    1. Let's bypass any kind of unnecessary software mixers.
    2. Measuring software per se may make quite a change. Let's be aware.
     
    Last edited: Feb 10, 2021
  8. atomicbob

    atomicbob dScope Yoda

    Pyrate BWC MZR
    Joined:
    Sep 27, 2015
    Likes Received:
    18,653
    Trophy Points:
    113
    Location:
    On planet
  9. Vtory

    Vtory Audiophile™

    Pyrate MZR
    Joined:
    Sep 12, 2016
    Likes Received:
    10,831
    Trophy Points:
    113
    Location:
    East Coast
    Thanks. I think I've read those before. But now reading those posts again, seems I can understand every sentence much better now than the last time lol. Will try -0.1db tricks myself to see if I can reproduce those results with M4, too.
     
  10. Vtory

    Vtory Audiophile™

    Pyrate MZR
    Joined:
    Sep 12, 2016
    Likes Received:
    10,831
    Trophy Points:
    113
    Location:
    East Coast
    Like I said yesterday, I experimented to figure out if tiny bit smaller signal helps WDM or not.

    Let's start from WDM 0dbfs out results. I kept option dialog open to show exact parameters I used. WDM signal generated by ARTA without adding any intentional jitter.

    [​IMG]

    Many unexpected harmonics at unexpected frequencies. Based on frequencies of those harmonics, I believe some kind of intermodulation occur between 333khz and main signal as weird harmonics shown at 333hz, 1660hz, 2330hz, 3670hz, etc. No idea what they are or whether they're associated with some kind of clipping or clipping protections.

    [​IMG]

    Hmm.. 0.1db difference didn't make meaningful change in my configuration. The result was more or less same with the above. But I rather wanted to keep going ..

    [​IMG]

    Ok, now 333hz and intermodulation all gone suddenly. Maybe ARTA's math isn't as smart as Bob's dedicated $$$ signal generators.

    [​IMG]
    [​IMG]

    The results look fairly persistent after a certain point.

    Finally, as reference, adding ASIO loopback result.

    [​IMG]

    Still I'm seeing higher order harmonics stronger in WDM outs. Maybe that's why we still hear some inferiority with WDM even in real music (no music is like time-invariant full scale signal lol)? But I will digress.
     
  11. Vtory

    Vtory Audiophile™

    Pyrate MZR
    Joined:
    Sep 12, 2016
    Likes Received:
    10,831
    Trophy Points:
    113
    Location:
    East Coast
    upload_2021-2-11_22-12-31.png

    By the way, quite unexpectedly I am enjoying M4's playback performance. Particularly with its hp out connected to M200 (vmoda). Meatless, anti-elegance, but there's something recreationally addictive that bonds me to the music. Honestly I feel like it stack well against good budget aio such as ifi zen dac.

    Will post about this once I collect more thoughts.
     
  12. atomicbob

    atomicbob dScope Yoda

    Pyrate BWC MZR
    Joined:
    Sep 27, 2015
    Likes Received:
    18,653
    Trophy Points:
    113
    Location:
    On planet
    -0.1 dBFS won't make any difference. The point where distortion is lowered is at -0.2 dBFS, which is below the Windows mixer limiter set for -0.131 dBFS. Re-read the DDC section in my post to see this point.
     
  13. Vtory

    Vtory Audiophile™

    Pyrate MZR
    Joined:
    Sep 12, 2016
    Likes Received:
    10,831
    Trophy Points:
    113
    Location:
    East Coast
    Damn.. either my eye or brain malfunctioned. Why did I think it -0.1db? Thanks for calling that out.
    Anyway tried -0.2db. Seems it worked with wdm. Glad I could reproduce this.

    [​IMG]
     
  14. Degru

    Degru Facebook Friend

    Joined:
    Jul 22, 2017
    Likes Received:
    140
    Trophy Points:
    33
    Location:
    Portland, OR
    I decided to try comparing measuring WDM/DirectSound on my end as well, although using the analog loopback of my Thinkpad x201s sound card. I generated a 1khz sine of 0dbfs and of about -1dbfs (amplitude 0.9) using Audacity, and exported them to 44100hz WAV files. I then played them in Foobar, using DS and WASAPI outputs with the Windows audio settings for the output device set to 44100hz. I fired up ARTA, and used its spectrum analyzer (in WDM mode) capturing the laptop's analog input with a samplerate of 96khz (Windows setting for input device at 96khz as well). I used an FFT of 65536 with around 24 rounds of linear averaging each time. All audio devices were set to 0db level in Windows and I made sure nothing was clipping.

    [​IMG]
    [​IMG]
    [​IMG]

    (These should be labelled -1db not -4db, oops)

    [​IMG]
    [​IMG]

    I am first of all sort of shocked at how clean the measurement came out on literally a 2010 laptop. It certainly does not sound good to the ear with music. But, you can clearly see the limiter kicking in with a 0db sine wave through WDM/DS. It's also evident that anything below the limiter's threshold is not impacted. You can also prevent the limiter from kicking in by simply applying a preamp reduction in EQ APO as well.

    An interesting test to try is playing a floating-point 32 bit file that's significantly louder than 0db. With WASAPI this will just be clipped off but with DS it will be pulled down by the limiter.
    [​IMG]
    [​IMG]
     
    Last edited: Feb 12, 2021
  15. Vtory

    Vtory Audiophile™

    Pyrate MZR
    Joined:
    Sep 12, 2016
    Likes Received:
    10,831
    Trophy Points:
    113
    Location:
    East Coast
    Thanks for another interesting data point.

    Btw, what do you mean by 0db? Relative to what?

    Looking "0 db" made the limiter activate, my initial guess was "relative to full scale out of your soundcard". Then you labeled "+10db" waveform later, which hardly make sense to me. By definition, FS is maximum, so couldn't be louder than that. So, it would be better to clarify unit for each.

    Decibel is just a "ratio" (such as 2 times or 0.5 times). Just some sort of monotonically transformed one. Without anything after that, it could be confusing.

    Another thing I want to ask was, if I understood correctly, laptop sound out isn't quite fixed line out. That means it's hard to define FS. Was there any way in your laptop to bypass voltage and current gains post analog signal conversion part?
     
  16. Degru

    Degru Facebook Friend

    Joined:
    Jul 22, 2017
    Likes Received:
    140
    Trophy Points:
    33
    Location:
    Portland, OR
    0db as labelled in the graph means the signal in the file is 0dbfs.

    +10db means just that; floating point WAV files can store values above 0dbfs, which can be attenuated back down but will otherwise be clipped if played into a DAC directly.

    As for the sound card settings, Windows actually lets you right click the unit under the "Levels"tab in the audio device properties and change it to db instead of percent. I changed these to 0db (which is 100% for the output and around ~40% on the input). It of course still won't be outputting standardized line level voltages, but this should presumably prevent any unwanted gain or attenuation and thus maximize SNR. The sensitivity of the input is a bit lower than the voltage of the output so that is why you don't see it reach 0dbfs on the measurement. The important thing for this test is the level that Windows mixer sees coming from the music player, since that's what actually triggers the limiter.

    Edit: -4db should actually be -1db on the last two graphs; I just realized Audacity must have calculated the loudness wrong :|
     
    Last edited: Feb 12, 2021
  17. Vtory

    Vtory Audiophile™

    Pyrate MZR
    Joined:
    Sep 12, 2016
    Likes Received:
    10,831
    Trophy Points:
    113
    Location:
    East Coast
    Measurements taken for M4's unbalanced line out (left -- out 3 -- only).

    [​IMG]

    [​IMG]

    [​IMG]

    [​IMG]

    [​IMG]

    Some results (compared to bal out) look interesting. But thoughts to come later.
     
  18. Vtory

    Vtory Audiophile™

    Pyrate MZR
    Joined:
    Sep 12, 2016
    Likes Received:
    10,831
    Trophy Points:
    113
    Location:
    East Coast
    Thanks for providing additional details. Helpful for me to better understand your results.

    I'm still struggling to figure out 32bit wav file though.

    My understanding below.
    • You originally generated 0dbfs sine signal
    • +10db gain applied and saved to 32bit float file (= 10dbfs peak at the time of saving)
    • Playback software automatically attenuates it playable (unsure if it just add -10db or with a little extra negative gain)
    So, I am puzzled why only WASAPI showed clipping. After reading the file in right ways, the peak should be in line with 16- or 24-bit wav's 0dbfs. Because it didn't touch WDM limiter, I believe the effective (attenuated) peak would be lower than -0.13dbfs. What am I missing?
     
  19. Degru

    Degru Facebook Friend

    Joined:
    Jul 22, 2017
    Likes Received:
    140
    Trophy Points:
    33
    Location:
    Portland, OR
    Foobar (when using DS output) sends the full 32 bit floating point signal to Windows, and windows does processing in floating point as well, including the limiter. It is the WDM limiter that is bringing the 10dbfs signal down to -0.13 before it is being output to DAC. WASAPI exclusive mode of course doesn't pass through the limiter so it gets converted directly to 24bit non-FP which clips off the higher samples.
     
    Last edited: Feb 13, 2021
  20. Vtory

    Vtory Audiophile™

    Pyrate MZR
    Joined:
    Sep 12, 2016
    Likes Received:
    10,831
    Trophy Points:
    113
    Location:
    East Coast
    Ok, seems it is foobar's wasapi plugin that does naive jobs. lol
    DAW apps (at least major ones) attenuate when they read such files. That's why I was confused.

    Anyway, great to know I should never ever use 32 bit option in foobar wasapi.
     

Share This Page