1. This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
    By continuing to use this site, you are consenting to our use of cookies.

    Dismiss Notice

The Hopelessly Derailed ODAC/Objective DAC Anticipation/Discussion Thread

Discussion in 'Sound Science' started by maverickronin, Apr 29, 2012.
5 6 7 8 9 10 11 12 13 14
16 17 18
  1. Currawong Contributor
    Kiteki, FYI, square waves don't occur naturally in music, so that the Octave can reproduce it without ringing is irrelevant. It distorts sine waves as any NOS DAC does from, say, 10 kHz and above. I do like it as a DAC though, especially with high-res music (88.2k and above), as that negates the distortion and, I feel, sounds slightly clearer.
  2. maverickronin
    He was talking about that in the context of some pretty unnatural music though.
    IDK if this is a case where it would actually matter but if I have time I'm going to check it out in the future.
  3. mikeaj
    answers in bold below
    Impulse response is not going to capture all the nonlinearities showing up as the different distortion figures (at different volumes and frequencies, no less).  I'm not even sure if it would be really close, considering how poor the earbuds do in some regards.
    I'm somewhat clueless at nonlinear systems and instrumentation, so I could be missing something though.
    If you reference Tyll's InnerFidelity post referencing Soaa-, note that the mathematical equivalence only works 100% for linear systems.  It's why the graphs don't line up that well for some headphones.
    Check the frequency of the square waves though, and the number of ripples.
    With enough bandwidth and the right filtering, you can have beautiful-looking square waves show up on the oscilloscope.  Do they sound more beautiful to the human auditory system, which is more limited in bandwidth?
  4. maverickronin
    I knew it wouldn't be perfect but is it that far off?  I guess its because the ibuds would be too non-linear and you'd need a more complicated transfer function with a few other measurements then?
  5. mikeaj
    Transfer function is just the Laplace Transform of the impulse response.  The more nonlinear the system is, the less accurately the impulse response characterizes the input/output behavior, I should think.  There are plenty of techniques for characterizing nonlinear systems, measuring them, and so on, but I don't know any of them since they're outside the very basics and not particularly related to anything I do.  In a lot of systems of interest, behavior is close enough to linear and accuracy of the model not so important, that simplifying assumptions can be made.  Not so here.  As for really complicated systems like weather or something in between, they use different approaches and analyses.
    As you can see from just the THD plots at 90 dB and 100 dB, the impulse response would have a different shape at different volumes.  That by itself would already be a big pain to account for.  Play the same thing 10x louder and you don't get the same response back 10x louder (I'm being loose here with the word "louder").
    It should still be pretty close though, I guess?  I don't have any good intuition on this one, so maybe I'm wrong.
  6. frenchbat
    I didn't know there was a name for that. Thanks for the link :wink:

  7. kiteki
    Thanks for the answers mikeaj you're the first person that can answer anything seriously.
    - It didn't say National used the listening room as an input to the design however it did say a To-99 package sounded better to several of them and they were looking to find out why, until they were all fired.
    - "The ringing comes from the lack of ultrasonics."  In the impulse response?  How does anyone know the pre-echo / post-ringing or lack thereof in the IR isn't audible?
    I don't know where you found the measurements on the Quad, the Octave is here - http://www.nosminidac.nl/downloads/octavehificritic.pdf, (there is a duo and hex as well)
    - "acoustic systems have enough peculiarities that you need a much more complicated model"  I think it's pretty intuitive that if you take a Stax, severely tailor the FR, increase noise etc. it will be near impossible to identify on paper (blind), despite still sounding... hm... just like a Stax.  Clearly acoustic systems are very difficult especially in 2012 yet I'm flooded with "Audio measurement = perfect look at the CSD/IR *** you can't see the 'stax-ness' can you!?" so then I start to question this 'science'.
  8. mikeaj
    The ringing is a consequence of the frequencies being cut off, but the ringing is probably not completely harmless (as hopefully I didn't imply earlier).  I'm not sure of the time scales and frequencies involved here, but in (poor) lossy compression, pre-ringing artifacts are some of the worst offenders.  On 44.1 kHz sampling rate, ringing due to the lowpass filtering should be more pronounced at the highest frequencies, but there may be some issues at frequencies some people can kind of hear.  It depends on the steepness of the filter—the flatter the frequency response, the more trouble you can get in the time domain.  On second thought, I'm pretty sure the ringing is problematic.  Hm, maybe I need another class in DSP...
    Also, even if ringing is at very high frequencies above what you can hear, that can still impact the performance of the amplifier and headphones, which have to deal with that energy, potentially making changes in the audible range.
    It should be noted that very short-duration impulses are pretty uncommon in any recording, but to me that's halfway an excuse too.  That said, if you look at the spectrum of most recordings, there's not a lot of energy above 20 kHz, and you need energy near the 22.05 kHz limit on 44.1 kHz audio to get ringing like that.  I think most modern audio DACs shouldn't do too badly here (if it makes a significant difference), but that's just speculation from me.  It would seem like a company advocating a gentler filter, no oversampling, etc., would pick out a relatively egregious example with a highly synthetic test tone—just note the frequencies they're dealing with, and figure out the context.
    I didn't find measurements of the Quad.  That THD figure was just one the website, so I was guessing that the real result was not going to be much better than what they claim.  Thanks for the link to those Octave results.
    Acoustic measurements can still tell you a whole lot.  They just need to be interpreted correctly.
  9. stv014
    Here is an example of ringing I created by resampling a single sample pulse from 44100 Hz to 192000 Hz, and applying a linear phase FIR filter that rolls off from 0 dB at 20 kHz to (theoretically) infinite attenuation at 22.05 kHz (the intended magnitude response is cos(((f - 20000) / (22050 - 20000)) * (π / 2))²). The IR display is zoomed vertically to show even small amounts of ringing, and has a total length of 1.5 ms (288 192 kHz samples). For comparison, a "short block" in MP3 has a total length of 384 samples (8.7 ms at 44.1 kHz), and 256 samples in Vorbis (5.8 ms at 44.1 kHz). You can also compare it to CSD graphs of headphones to find out how significant (or not) the effect is.
    fr.png     ir.png
    I would guess the ultrasonic garbage from a NOS DAC is likely to cause more issues in amplifiers than the ringing of a linear phase lowpass filter, which is limited to the range of the roll-off (20 to 22.05 kHz in the example above).
    Of course, with a good oversampling DAC that supports high sample rates, it is also possible to emulate the various types of filtering (even a NOS-like response) in software.
    Edit: some graphs from a real DAC (PCM1792 on a Xonar STX sound card, 44.1 kHz playback and 192 kHz recording frequency); the top one is using the "sharp rolloff" DAC filter mode, while the bottom one is "slow rolloff". The last picture shows how the colors map to actual levels on the spectral graphs (every second is 10 dB quieter, so it goes from 0 dB to -120 dB):
    stx1fr.png     stx1spec.png
    stx2fr.png     stx2spec.png
    The spectral displays have a total length of 5 ms (960 samples). To me, the default "sharp rolloff" mode seems to be a reasonable compromise between rolled off treble, ultrasonic content, and ringing.
    By the way, a NOS DAC can ring if the edge/pulse in the input is anti-aliased and is not at an integer sample boundary.
  10. kiteki
    You can emulate this?  http://www.nosminidac.nl/sample__squarewave.html
  11. stv014
    It is obviously not possible to reproduce exactly the same waveforms with limited bandwidth. But the 96 kHz available at 192 kHz sample rate does allow for shortening the ringing significantly (as well as shifting it to a higher frequency) at the expense of ultrasonic imaging products.
    ir2.png     square.png
  12. stv014
    Judging from the measurements of other NOS DACs, the high frequency intermodulation distortion and low level linearity could be suspect, but those are of course not specified.
    That is, of course, assuming that it is really -125 dB. This value is also specified as "noise generated by electronics", which could have some creative interpretations (e.g. only the datasheet noise level of an IC).
  13. kiteki
    So we have some potential design improvements here for the ODAC -> O2
    - To-99 op-amps (speculation: akin to the like of reported USB cable and capacitor improvements? EMI / RF rejection?)
    - NOS emulation via software
  14. maverickronin
    You've got a lot of chiptune right?  Why don't you go through it an pick out some samples with square waves for some testing?
  15. kiteki
    It's been a while since I've seen pure square-waves in the oscilloscope, an issue is that most music is composed with layers (channels) so you might have pure square waves in channel 1, however they're obscured by the music in channel 2, 3 and 4 since it's all compiled into a single wave in the end, that doesn't lessen the effect though, apart from the drivers difficulty in presenting that information, which is why multi driver setups and harder materials like Titanium perform better in this respect, especially with complicated electronic music.
    When I find some songs I'll upload them somewhere.
    Edit:  In the meantime I recommend this To-99 for the O2 - http://www.ebay.com.au/itm/1x-OPA2111BM-OPA2111-To99-Op-amp-IC-replace-NE5532-4558-/350530775385 - http://www.ti.com/lit/ds/symlink/opa111.pdf
5 6 7 8 9 10 11 12 13 14
16 17 18

Share This Page