Pre and Post Effin Ringing and shit like that

Discussion in 'Blind Testing and Psychoacoustics' started by ultrabike, Aug 17, 2016.

  1. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Pyrate MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,960
    Trophy Points:
    113
    Location:
    Irvine CA
    Decided to move this to a new thread to keep the Multi-Bit Schiit thread on track...

    As @Serious mentioned, AFAIK the MegaComboBurrito is a linear phase filter which will have both "pre and post ringing".

    You want both "pre and post ringing" in order to approximate an ideal interpolator closely. The price you pay is latency.

    From my perspective "Pre-echo" and "Post-echo" stuff seems like some bullshit terms coined by some folks to solve problems that do not exist at the expense of creating real problems.

    The only reason I see to use a no "Pre-echo" filter is if you want lowest possible latency at the expense of some accuracy.
     
    Last edited: Aug 18, 2016
  2. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    89,777
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    Meridian-Pre-post-ringing.gif

    It's a tradeoff. See how much more nasty the overshoot and post-ringing is. I don't like minimum phase or apodizing filters. At least I've never liked them on the DACs which have allowed me to change the settings. The argument is that pre-ringing is unnatural and doesn't exist in the real world, and that post-ringing would be masked by the waveform. On first thought, this might make sense... until we realize that waveforms of impacts (say snare drum hits) in real music have content both before and after the impacts which are likely to mask the (lower amplitude) ringing.

    [​IMG]

    So I concur with @ultrabike's assessment that "Pre-echo" and "Post-echo" stuff seems like some bullshit terms coined by some folks to solve problems that do not exist at the expense of creating real problems. But seriously, screw theory. Minimum or apodizing filters just sound like ass.
     
  3. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Pyrate MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,960
    Trophy Points:
    113
    Location:
    Irvine CA
    Yup! A good evaluation of how these filters behave is exactly what The Merv (TM) did. Pass a signal through both filters and see what it does to the original.

    A linear phase filter will have an associated latency (half the size of the filter roughly). During this latency there will be transients, much like the ones one gets on an analog electric network. But after that, the signal comes out from start to finish largely untouched relative to the original.

    A minimum phase filter such as those apodizing and ballz cutting filters will have minimum transients and delay, but kiss phase awesomeness goodbye, and because of that, some nasties may indeed happen.
     
  4. SSL

    SSL Friend

    Pyrate
    Joined:
    Nov 12, 2015
    Likes Received:
    1,239
    Trophy Points:
    93
    Based on the measurements of Schiit MB on this forum, they are of course using a linear phase filter, but it also has a steep roll-off. Imaging and square wave measurements are indicative.

    Pre- and post-rining with linear phase will not only be low in amplitude, but also at Nyquist; so it is very unlikely to be audible with real material.
     
  5. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Pyrate MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,960
    Trophy Points:
    113
    Location:
    Irvine CA
    It is also a 18000 tap filter, so the step roll off will not necessarily affect the phase.

    Pre- and post-ringing amplitude should not matter. After transient (latency), it should pretty much be inaudible relative to the original... as in well over 120 dB inaudible (> 20-bits) . Remember, the ideal perfect interpolator is supposed to be a square window in the frequency domain which is a sync in the time domain. The problem is that such a thing goes from - to + infinity. So it does not exist and it can only be approximated. Several optimizations to do so.

    In other words, the ideal filter has infinite pre- and post-ringing. The nice thing is, the area under the pre- and post-ringing region (IR power) converges (it is not infinite). Furthermore, it does so relatively fast. Which is why we can have pretty good approximations in the first place.

    Anyhow, the point is that the more pre- and post-ringing there is, the more accurate things will be. So forget about hearing it or not. Again, the more there is, the less you'll hear filter issues. Things will come out a bit late though.
     
    Last edited: Aug 17, 2016
  6. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Pyrate MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,960
    Trophy Points:
    113
    Location:
    Irvine CA
    BTW I know what I'm saying can be a bit counter intuitive. So that's why I'm bringing it up.
     
  7. SSL

    SSL Friend

    Pyrate
    Joined:
    Nov 12, 2015
    Likes Received:
    1,239
    Trophy Points:
    93
    Just to clarify, by "more rining" you mean more:
    • Duration?
    • Amplitude?
    • Frequency?
     
  8. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Pyrate MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,960
    Trophy Points:
    113
    Location:
    Irvine CA
    Ringing seems to mean everything outside the main lobe in the impulse response... which seems to oscillate.

    However, the sinc function has a very specific behavior in terms of duration, amplitude and frequency of the ... "ringing". It is pretty much sin(x)/x...

    Approximations may use windowing techniques (either in frequency or time domain, or both), perhaps optimization techniques minimizing power, this or that. Most of them will trade this for that. All of them have some sort of ... "ringing" or whatever.
     
    Last edited: Aug 17, 2016
  9. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Pyrate MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,960
    Trophy Points:
    113
    Location:
    Irvine CA
    In other words, the ideal "ringing" duration, amplitude and frequency obeys the sin(x)/x equation... or #sin(@x)/(@x), where @ controls the period and # is the amplitude... x is the normalized time variable. That's the ideal (and un-realizable) interpolator.

    Good approximations behave as close as possible to this ideal interpolator.
     
  10. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Pyrate MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,960
    Trophy Points:
    113
    Location:
    Irvine CA
    Another thing that can be confusing about ringing is that in the frequency domain, a brick-wall corresponds to the sync function which is infinite.

    In realizable low-pass filters there are passband, transition, and stopband regions. They are not infinitely step, because they are not infinite long in time domain. Trade-offs in these regions will have different sorts of ... "ringing" in the time domain. And linear phase can be maintained, at the expense of this or that.

    This is not what should be expected out of a speaker or headphone driver IMO, since at that point one may not very concerned about the transition or stopband regions as long as the passband keeps a decent phase behavior (seldom the case). Hell, some folks may advertise their cans are capable of 100 kHz or some crazy shit like that... This is not what you necessarily want out of an interpolator in a DAC.

    +++

    Anyhow. The point is Schitt's DAC erect burrito is linear phase, has ringing, pre and post, and is good for it. Better IMO that all minimum phase apodizing filters... unless you are concerned about latency in certain applications.
     
    Last edited: Aug 17, 2016
  11. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Pyrate MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,960
    Trophy Points:
    113
    Location:
    Irvine CA
    Moved things to this new thread to keep things on track on the other one...

    Another thing to consider:

    Brick-Wall filtering by itself is not "bad". It depends on implementation.

    Brick-Wall filtering is typically associated with an Elliptic, Chebyshev or Butterworth IIR filter (analog or digital). These filters can provide a pretty step drop from passband to stopband, and be way more practical than a 18000-tap FIR filter. The price we pay for that is a horrid and shitty phase response, specially near the transition band (high frequencies).

    There is not much that can be done in the analog side of a DAC for that, and therefore the request to have signal sampled at 96 kHz instead to reduce the effects of the analog equivalent IIR filtering.

    However, in digital, a FIR filter can still be pretty much Brick-Wall through brute force (18000-taps) and have super nice phase response.

    The tradeoff is that an 18000-tap FIR filter is not only going to have more latency, it's going to require quite a bit of fire power to implement vs a 5th or 7th order IIR Brick-Wall filter. But don't automatically assume that Brick-Wall is bad just because it's Brick-Wall.
     
  12. briskly

    briskly Friend

    Pyrate
    Joined:
    Oct 3, 2015
    Likes Received:
    59
    Trophy Points:
    18
    Location:
    New York
    The stuff about linear phase, transition width, and pre-ring should just follow from the sampling theorem and Fourier identities, right? It gets awkward to think of a minimum phase filter as correcting some sort of fault in the stopband filter with that in mind.

    It is harder to directly track phase shift with high frequencies, but then again I'm not sure how important "ringing" artifacts are fs/2 either.
     
    Last edited: Aug 18, 2016
  13. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Pyrate MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,960
    Trophy Points:
    113
    Location:
    Irvine CA
    Minimum phase is used to reduce latency.

    Linear phase will have the property of having all frequencies aligned at the passband. Minimum phase will not. The issues for minimum phase go below fs/2.
     
  14. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Pyrate MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,960
    Trophy Points:
    113
    Location:
    Irvine CA
    OK.... here you go.

    This is a 200 tap linear phase low pass filter:

    linear_phase.png

    linear_phase_fr.png

    Here is the equivalent minimum phase filter:

    minimum_phase.png

    minimum_phase_fr.png

    As you can see the phase in the minimum phase filter is not linear in the passband. Frequencies will not be in phase with each other and input and output data will not match up. High frequencies will be affected the most. This affects signal below fs/2.

    Passband also exhibits some ripple relative to the minimum phase equivalent. Stopband rejection also degrades.
     
  15. briskly

    briskly Friend

    Pyrate
    Joined:
    Oct 3, 2015
    Likes Received:
    59
    Trophy Points:
    18
    Location:
    New York
    The minimum phase filter wouldn't require the delay line unlike the linear phase. The minimum phase filter has a non-linear phase characteristic in the passband.
    I didn't mean to suggest otherwise in the first place.

    What I meant to refer to is hearing the phase error of minimum phase.
     
  16. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Pyrate MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,960
    Trophy Points:
    113
    Location:
    Irvine CA
    And here are the results when passing signal through these filters:

    Linear phase:

    linear_phase_results.png


    Minimum phase:

    minimum_phase_results.png

    The red is the original signal.

    The blue is the signal through the filter.

    Things match up pretty good with the linear phase filter relative to the minimum phase one. But signal comes out after 100 samples in the linear phase filter case, and 7 samples later in the minimum phase filter case.

    Note: The original signal is also delayed by 100 (it was bandlimited to fs/2 with the linear phase filter), so for the linear phase signal appears around 200 samples later. For the minimum phase about 107 samples later... This is also why the linear phase filter phase plot in degrees goes to 10000 or such. Phase is closely related to delay.

    Here are the two cases again, but zoomed out:

    Linear Phase:

    linear_phase_res_zoom.png

    Minimum Phase:

    minimum_phase_res_zoom.png

    Again, signal is right on top of the original with linear phase. Not so with minimum phase.
     
    Last edited: Aug 18, 2016
  17. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    89,777
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    You forgot to conveniently delete the phase plot and expand the Y-axis for the minimum phase.
     
  18. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Pyrate MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,960
    Trophy Points:
    113
    Location:
    Irvine CA
    Yes. Phase issues may not be as obvious as other types of issues.

    It may be because phase issues arise even due to room and environment and we adapt to become tolerant of it. But they are there and they may be heard as some weird coloration or such. Poor filtering may also strain the analog front end. All of these issues may add up to nasties.
     
    Last edited: Aug 18, 2016
  19. Madaboutaudio

    Madaboutaudio Friend

    Pyrate
    Joined:
    Sep 28, 2015
    Likes Received:
    545
    Trophy Points:
    93
    Location:
    Singapore
    Basing on my personal subjective experiments with slow roll off(minimum phase) and sharp roll off(linear phase) adjustment with the following dacs:

    Audio-gd Master 7 DSP digital filter
    IBasso DX90 ES9018K2M digital Filter
    IBasso DX100 ES9018S digital filter
    Sony ZX2 DSD filter

    I find that slow roll off makes music sounds smoother but the trade off is a gooey and wonky warm sound, also creates a more hollow soundstage.

    Sharp roll off has a sharper sound(especially on the attack of the transient) but the trade off is more grittiness to the sound. I find that sharp roll off works best across a broad genre of music.
     
  20. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Pyrate MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,960
    Trophy Points:
    113
    Location:
    Irvine CA
    You can make a linear phase filter with slow and smooth roll off if you like...
     

Share This Page