Pre and Post Effin Ringing and shit like that

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

  1. rtaylor76

    rtaylor76 Can't wipe his tag

    Friend
    Joined:
    Nov 9, 2015
    Likes Received:
    1,730
    Dislikes Received:
    83
    Trophy Points:
    93
    Location:
    Nashville, TN
    Home Page:
    To me it would be more interesting to study why some prefer filter a) over filter b).

    I have always preferred minimum linear phase because it has a more immediacy to the initial sound. The drum hits, plucks, etc. It only takes a few milliseconds for our brains to decide, based on the ADSR, of what an instrument is and it's localization. This is why synthesizers, like the Yamaha MOTIF, uses real samples for the first initial hit, then blend in the synthesized sound behind it, even if it has a bit of pre ringing.

    Granted, minimum phase just pushes all the wackyness to behind it and makes things sound mushy, but it is a trade-off to the initial realism and spacial cues. There is no free lunch.

    Everytime I go to linear minimum phase it sounds more glaring, sharp, and the soundstage flattens out. It is also more fatiguing. Minimum linear phase and NOS DACs w/o filters sound better to me, but this is just personal preference.
     
    Last edited: Aug 3, 2018
  2. purr1n

    purr1n Finding his inner redneck

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    50,622
    Dislikes Received:
    82
    Trophy Points:
    113
    Location:
    Antarctica
    You sure you dont have things reversed? I've always found minimum phase, or greater ringing, to sound more fatiguing relative to filters with less ringing, regardless of the existence of pre-ringing. For example, the GOV2 and PWD2 were too sharp and harsh with the minimum phase type filters.

    NOS and minimum phase are diametically opposite sound to me. The most similar to NOS in the OS world has been linear phase with slow rolloff.
     
    Last edited: Aug 3, 2018
    ultrabike, Josh83 and Erikdayo like this.
  3. rtaylor76

    rtaylor76 Can't wipe his tag

    Friend
    Joined:
    Nov 9, 2015
    Likes Received:
    1,730
    Dislikes Received:
    83
    Trophy Points:
    93
    Location:
    Nashville, TN
    Home Page:
    You are right.

    Minimum phase = sharp EQ roll off, more post ringing
    Linear phase = slow EQ roll off, pre and post ringing

    Then forget what I said. Whatever the slow rolloff is. Fixed my post.

    Also, the NOS filterless I have is the 1541x4 that naturally has slow rolloff beginning around 10k anyway on the output. Some stuff in that band can sound harsh at times, but again, it is a trade-off.
     
    Last edited: Aug 3, 2018
    Erikdayo likes this.
  4. purr1n

    purr1n Finding his inner redneck

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    50,622
    Dislikes Received:
    82
    Trophy Points:
    113
    Location:
    Antarctica
    Thank goodness. I thought I might have been in the southern hemisphere where toilets drain the wrong way.

    There is also linear phase sharp knee rolloff too. That's my preference for most DACs.
     
  5. NekoAudio

    NekoAudio Acquaintance

    Joined:
    Apr 18, 2016
    Likes Received:
    53
    Dislikes Received:
    0
    Trophy Points:
    18
    Location:
    San Jose, California
    Home Page:
    I read that to mean hardware EQ devices, like 15-band or 31-band analog graphic equalizers.
    Prism is using the CS4398 but does some proprietary pre-processing first in the DAC chain.
    The best test track I've come across for hearing (presumably) phase-related differences imparted by filters is the Cups song from Pitch Perfect, because of the isolated clapping. Based on too few samples to count as scientific evidence, I also have an impression that feeding 88.2kHz or 96kHz sample rates can reduce any such differences to inaudible.
     
    Lyander likes this.
  6. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Friend MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,555
    Dislikes Received:
    16
    Trophy Points:
    113
    Location:
    Irvine CA
    "As a side-effect of changing the level balance of different bands of frequencies, conventional EQ also changes their phase relationships. So low frequencies can be delayed with respect to high, or vice versa, depending on how the EQ parameters are set. We call this 'phase shift'."

    In general, an equalizer will indeed change phase relationships for different frequencies in the passband. However, a well applied EQ is providing a correction on the effects of the speaker or headphone down the change. The overall response should have as linear phase as possible.

    As far as group delay, usually large FIRs have large group delays, but the phase relationships remain intact and coherently combine. IIRs indeed have little latency and little group delays. But the frequencies will be out of phase with eachother. In some applications that use phase information this is a problem. In audio is probably less of a problem.

    When it comes to a large FIR with linear phase, it is straight forward to derive the group delay of all frequencies. Namely is the number of taps of the filter divided by two times the sampling rate. So if the filter is 12000 taps and the sampling rate is 192 kHz, the group delay is 6001/192k = 31ms. A 100 Hz tone has a period of 10ms. That said, the FIR should result on a 31ms delay and no phase distortion because all frequencies will be effectively delayed by the same amount. In an IIR is also not such a big deal because the phase delays will likely be in the order of a few micro seconds.

    In regards to random EQ software, I can imagine how some implementations will result in aberrations. Specially graphic ones. But I have not taken the time to analyze particular brands or implementations.
     
    Erikdayo likes this.
  7. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Friend MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,555
    Dislikes Received:
    16
    Trophy Points:
    113
    Location:
    Irvine CA
    The roll off of filter has little to do with whether it is minimum or linear phase.

    You can make a brickwall filter either way.

    As far as ringing. There is no ringing when the filter is applied to a bandlimited signal whose passband is below the passband of the filter. There will be ultrasonic "ringing" for signals whose passband is above the passband of the filter, assuming stopband is in the ultrasonic range.
     
    MrTeaRex likes this.
  8. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Friend MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,555
    Dislikes Received:
    16
    Trophy Points:
    113
    Location:
    Irvine CA
    I most definitively do not read that in the Cirrus Logic marketing webpage of shame. They are discussing DACs and CODECs. Not equalizers. Furthermore, here is their picture from their website:

    cirrus.png

    Those are not equalizers by the look of them. Those are low pass interpolation filters.

    The interpolation filter is not part of the pre-processing. It's part of the conversion. It happens right before the delta sigma.

    The part offers "selectable interpolation filters" through their "control port interface". That said, I don't see linear/minimum phase digital filter selection for the CS4398. I just see different roll-offs options.

    The three options I see for that part have a group delay ranging from 4.6/Fs to 9.4/Fs. LOL! Maybe they are all minimum phase or some random "advanced" compromise! FUCK!

    AKM does clearly offer minimum and linear phase options. Their AK4497EQ minimum phase options have group delays ranging from 5/Fs to 6/Fs. Their linear phase options include "Slow Roll Off" and "Sharp Roll Off"(Traditional) with GD of 6.5/Fs and 29.2/Fs respectively. The "Slow Roll Off" is a highly windowed, short linear phase filter. I recommend the "Sharp Roll Off" one. Stay away from the "Short Delay Sharp Roll Off" and "Short Delay Slow Roll Off" options unless you need low latency. Also stay away from the "Low Dispersion" one, though it's not as bad.

    Note again that there is a "Sharp Roll Off" (linear phase) and a "Short Delay Sharp Roll Off" (minimum phase). Just as there is a "Slow Roll Off" (linear phase) and a "Short Delay Slow Roll Off" (minimum phase) options. Roll off, sharp or slow, does not imply linear or minimum phase.

    Anyhow. I see at least a good option with AKM. I don't see it with Cirrus. Damit! Fuck you Cirrus!
     
    Last edited: Aug 3, 2018
    maverickronin likes this.
  9. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Friend MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,555
    Dislikes Received:
    16
    Trophy Points:
    113
    Location:
    Irvine CA
    @NekoAudio, actually Prism could probably indeed pre-process by upsampling and applying their own filter. Then maybe bypass the Cirrus craptastic filters. But that's a lot of headaches. I would just buy the AK4497EQ part, unless the Cirrus CS4398 has something the AK4497EQ doesn't.
     
  10. rtaylor76

    rtaylor76 Can't wipe his tag

    Friend
    Joined:
    Nov 9, 2015
    Likes Received:
    1,730
    Dislikes Received:
    83
    Trophy Points:
    93
    Location:
    Nashville, TN
    Home Page:
    You can play with this and see how different mixing software platforms use their filters in sample rate conversion (downsampling).

    http://src.infinitewave.ca/

    You can see there are examples of linear phase brickwall filters as well as mininum phase brickwall filters. Just compare the difference between say Ableton 9.11 vs Ableton 8.2. Both have similar looking passbands, but the impulse response looks vastly different.

    Then you look at Amadeus Pro 2.4.2 that has a very gentle sloping filter, a great looking impulse response, but the sweep looks bad.
     
  11. Armaegis

    Armaegis Friend

    Friend BWC
    Joined:
    Sep 27, 2015
    Likes Received:
    3,684
    Dislikes Received:
    2
    Trophy Points:
    113
    Location:
    Winnipeg
    I don't have any sources, but I've read that somewhere in all the mathemagic that 88.2 or 96kHz are ideal sampling rates to work in because it allows the headroom to shift the number buggery into the inaudible zone, but is not so high that ultrahigh frequency garble can find it's way back down.
     
    maverickronin and ultrabike like this.
  12. maverickronin

    maverickronin Friend

    Friend
    Joined:
    Mar 17, 2018
    Likes Received:
    316
    Dislikes Received:
    2
    Trophy Points:
    63
    Location:
    Midwest, USA
    I just noticed that "Short Delay Sharp Roll Off" is the default on ADI-2 DAC.

    One the one hand why. On the other hand, it's not like I heard a difference...

    Edit: Added a page from the manual

    [​IMG]
     
    Last edited: Aug 3, 2018
  13. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Friend MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,555
    Dislikes Received:
    16
    Trophy Points:
    113
    Location:
    Irvine CA
    Looking at a filter response in time domain can be misleading. One way to evaluate performance in time domain would be to pass a real band limited audio signal through the filters. Align latencies as much as possible (minimum phase filters can have fractional latency) and take the difference from the original. Take the mean square difference. The smaller the difference, the better.

    It may not be perfect, specially with fractional GD relative to sampling rate. And indeed phase issues (which are linear) are probably not much to worry about.

    If doing frequency responce measurements or room correction, intuitively I would suggest the linear phase filter though. Latency should not be much of a concern with that AKM solution given their low GD.

    BTW the analog reconstruction filter is an IIR and it will be more detrimental. But hopefully the most problematic areas are in the digital filter stop band and in the ultrasonic band.

    Minimum phase option is not necessarily terrible. But linear phase is definitively not a problem. The pre ringing claim is what gets me a bit irritated, thats all.
     
    Last edited: Aug 3, 2018
    Erikdayo likes this.
  14. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Friend MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,555
    Dislikes Received:
    16
    Trophy Points:
    113
    Location:
    Irvine CA
    BTW, that time domain test I suggested will be even more difficult if not done fully in the digital domain because using an ADC will result in some random phase delay on top and it also has a bunch of analog and digital filters. Recording the results from a speaker driver will be even more difficult.

    Not impossible though. But some aproach to account for fractional delay alignment will be required. Can’t do much about the anti aliasing filters in the ADC. One would have to use the same ADC filters in all cases to keep things fair.

    In the end other issues likely dominate. Just don’t buy into the pre ring kool aid.

    EDIT: I acutally have an idea about how to test this even w my POS 2i2. Might doit when I get back.
     
    Last edited: Aug 3, 2018
  15. NekoAudio

    NekoAudio Acquaintance

    Joined:
    Apr 18, 2016
    Likes Received:
    53
    Dislikes Received:
    0
    Trophy Points:
    18
    Location:
    San Jose, California
    Home Page:
    I thought you were referring to the Audio Masterclass web page when you were quoting conventional EQ changing phase, which didn't seem to be about any specific DAC chipsets or specifically restricted to the digital domain. My bad.
    Yes, they're doing something like that. I don't know all of the details but there's a blurb on page 2 of this review from HiFi News [PDF].
    Yes, that's also my understanding and I vaguely recall seeing phase measurements as a function of sample rate of a DAC using a "minimum phase filter" that supports this. I might have taken those measurements myself, but I don't think it was me.
     
  16. MRC01

    MRC01 Rando

    Joined:
    Dec 10, 2018
    Likes Received:
    21
    Dislikes Received:
    2
    Trophy Points:
    3
    Location:
    Earth
    Pre-ringing is real, in the sense that under worst-case conditions you can hear it -- absolute silence, then an impulse with lots of HF energy like the sudden loud THWACK of a stick, or castanets. But minimum phase is a cure that may be worse than the disease. It eliminates the pre-ring but actually rings louder and longer after the transient, and adds phase distortion. So with minimum phase the impulse hits cleaner but its decay, the actual sound, is more distorted. Worse, that high frequency phase distortion is there all the time, not just when there are impulses.

    However I think this "linear vs. minimum" is an artificial dichotomy because the amplitude of that impulse ripple is related to the slope of the anti-aliasing filter. Most examples use a very steep narrow transition band filter to exaggerate the effect for educational purposes. With high sample rates, or oversampling lower rates like 44.1k, (e.g. the real world) the filter has a much wider transition band with a more gradual slope, shrinking the amplitude of that ripple so small it is negligible.

    All this makes me wonder why people debate this so much. Probably because it's fun to argue.
     
    slankoe and famish99 like this.
  17. AudioNut

    AudioNut Acquaintance

    Joined:
    Oct 2, 2018
    Likes Received:
    58
    Dislikes Received:
    15
    Trophy Points:
    18
    Location:
    Ottawa, ON
    In a traditional standalone DAC, yes that's true, but what that Cirrus web page is talking about is AV receivers/soundbars/wireless speakers, etc. (I'll refer to these as "AV tech products").

    These days, virtually all consumer AV tech products use sample rate conversion to some nominal internal clock rate prior to DSP and prior to the DAC stage. So in practice the most important interpolation/reconstruction filtering happens well before the DAC, and in that sense what Cirrus is saying is accurate. (Obviously the DAC stage applies additional interpolation/reconstruction, but the effects of doing 2x interpolation on a 192kHz internal signal are far less significant than an 8x oversampling filter on a 44.1kHz signal.)

    Even in the pro audio realm, it's becoming more common for newer pro audio interfaces to use ASRC on digital inputs by default rather than clocking to the incoming digital signal. You usually still get the option to disable this on pro audio gear (though not always).
     
    famish99 likes this.
  18. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Friend MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,555
    Dislikes Received:
    16
    Trophy Points:
    113
    Location:
    Irvine CA
    Nope. Bullshit.

    @NekoAudio was talking about the CS4398 DAC chip. Not Cirrus AV receivers/soundbars/wireless speakers. Or some SW prior to sending data to the DAC chip.

    And yes, a lot of shit can happen before things hit the IC. Obviously.
     
  19. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Friend MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,555
    Dislikes Received:
    16
    Trophy Points:
    113
    Location:
    Irvine CA
    Minimum phase is shit relative to linear phase. It has no reason to exist other than reduced latency.

    This, "under worst-case conditions you can hear it" is COMPLETE AND TOTAL BULLSHIT.

    Fucking get it. The "ringing" on the filter corresponds to the filter transcients. Again, TRANSCIENTS. Once the filter delay taps are filled to the main tap, the filter can be considered to have reached STEADY STATE. On steady state a linear phase filter will be way more accurate than a minimum phase filter. AND THERE WILL BE NO FUCKING RINGING AT THE OUTPUT OF THE FILTER, except at the very begining and at the end. EVEN IF YOU HAVE FUCKING DRUMS PLAYING THROUGH OUT THE MIDDLE OF THE AUDIO FILE.
     
    SoupRKnowva and Lyander like this.
  20. AudioNut

    AudioNut Acquaintance

    Joined:
    Oct 2, 2018
    Likes Received:
    58
    Dislikes Received:
    15
    Trophy Points:
    18
    Location:
    Ottawa, ON
    You're using the term "ringing" in a sense that isn't really what signal processing folks mean by ringing. Of course any FIR filter settles to zero in finite time; that doesn't mean it doesn't produce what signal processing folks mean by ringing artifacts.
     

Share This Page