Hi Mark,
You may find these papers of intereest;
http://www.bodziosoftware.com.au/Pre_Post_Ringing_IR_And_Pulses.pdf
http://www.bodziosoftware.com.au/Pre_Ringing.pdf
Peter
You may find these papers of intereest;
http://www.bodziosoftware.com.au/Pre_Post_Ringing_IR_And_Pulses.pdf
http://www.bodziosoftware.com.au/Pre_Ringing.pdf
Peter
Thanks zbig001, PLB
I've tried various impulse tests...balloon pop etc, along with a ton of dynamic music, and haven't heard anything that sounds like something arriving too soon. I'll try the 1-sample, perfect impulse.
Yes, I've read those papers several times. Great resource.
My take from the pulse paper is that linear-phase pre-ringing can only be heard with very high Q, high boost peaking EQs. Shelving, cuts, even notches, don't seem to pose a problem. And that linear-phase crossovers are plain superior to min-phase crossovers, when using the same type and order on both sides of crossover. Both pre and post ringing appear to totally cancel out. Even high orders look to be no problem. It makes for a hell of a strong case for using lin-phase crossovers IMO.
I've listened and measured to lin-phase FIR usage of 64ms delay, and hear nothing but the tightest dynamics I've ever heard....and have become a near zealot regarding lin-phase. But I'd like to find any test that I can hear, if I'm just missing pre-ringing's effects....always learning, right ?
I've tried various impulse tests...balloon pop etc, along with a ton of dynamic music, and haven't heard anything that sounds like something arriving too soon. I'll try the 1-sample, perfect impulse.
Yes, I've read those papers several times. Great resource.
My take from the pulse paper is that linear-phase pre-ringing can only be heard with very high Q, high boost peaking EQs. Shelving, cuts, even notches, don't seem to pose a problem. And that linear-phase crossovers are plain superior to min-phase crossovers, when using the same type and order on both sides of crossover. Both pre and post ringing appear to totally cancel out. Even high orders look to be no problem. It makes for a hell of a strong case for using lin-phase crossovers IMO.
I've listened and measured to lin-phase FIR usage of 64ms delay, and hear nothing but the tightest dynamics I've ever heard....and have become a near zealot regarding lin-phase. But I'd like to find any test that I can hear, if I'm just missing pre-ringing's effects....always learning, right ?
That 1-sample pulse is an absurdly violent test method...
Maybe even CA converter can respond with some ringing.
If despite this ringing will be too faint:
- generate loud pulse, close to 0 dBFs
- excite ringing with cutoff filter where hearing is most sensitive (3-4 kHz)
- use "brickwall" type steepnes
- no fancy window function, rectangular windowing only
- long IR file, because the more taps the longer ringing
Maybe even CA converter can respond with some ringing.
If despite this ringing will be too faint:
- generate loud pulse, close to 0 dBFs
- excite ringing with cutoff filter where hearing is most sensitive (3-4 kHz)
- use "brickwall" type steepnes
- no fancy window function, rectangular windowing only
- long IR file, because the more taps the longer ringing
Yes, such microscopic impulse tests are kind of absurd I guess.
The only reason I even try such tests is I have yet to hear any pre-ringing using 96dB oct lin-phase crossovers, and would like to know if pre-ringing truly exists.....audibly that is, without crazy Q's or gain boosts.
So far, I have to say my impression is that FIR pre-ringing goes deep in the myth bucket...
The only reason I even try such tests is I have yet to hear any pre-ringing using 96dB oct lin-phase crossovers, and would like to know if pre-ringing truly exists.....audibly that is, without crazy Q's or gain boosts.
So far, I have to say my impression is that FIR pre-ringing goes deep in the myth bucket...
Very good demonstration of pre-ringing resulting from the use of linear phase EQ right here: https://m.youtube.com/watch?v=UFrbKqRuKxI
It’s at the 5 minute mark
It’s at the 5 minute mark
Yes, thank you. I've seen that.
He states he uses steep slope/high-Q EQ's to cut lower frequencies. I do think high-Q EQ's using linear-phase can be heard to pre-ring... (and heard with min-phase too as post ring, though not heard in his example).
I know to avoid high-Q EQ boost filters, whether lin or min phase. Both are audibly bad IME. And I know to avoid linear-phase crossovers which are not used the same on both sides of crossover. His low cut filter made out of EQ seems to be pretty equivalent to a one sided crossover. I also think he's selling his Fabfilters...(as he says his EQ example is extreme for normal music, probably tailor chosen to compare his filter) So all told, I don't really think he's made much of a case demonstrating pre-ringing.
I'm left thinking still, that linear-phase crossovers in FIR seem fine, as do reasonable min-phase EQ's embedded into FIR along with the crossovers.
He states he uses steep slope/high-Q EQ's to cut lower frequencies. I do think high-Q EQ's using linear-phase can be heard to pre-ring... (and heard with min-phase too as post ring, though not heard in his example).
I know to avoid high-Q EQ boost filters, whether lin or min phase. Both are audibly bad IME. And I know to avoid linear-phase crossovers which are not used the same on both sides of crossover. His low cut filter made out of EQ seems to be pretty equivalent to a one sided crossover. I also think he's selling his Fabfilters...(as he says his EQ example is extreme for normal music, probably tailor chosen to compare his filter) So all told, I don't really think he's made much of a case demonstrating pre-ringing.
I'm left thinking still, that linear-phase crossovers in FIR seem fine, as do reasonable min-phase EQ's embedded into FIR along with the crossovers.
Impulse and steps are directed at digital circuits to elucidate some information about them. They are useful as a test only. No music signal is remotely like an impulse. Ergo pre-ringing is a theoretical concept and is not heard.
And no - a cymbol crash, drum hit, rifle shot are not impulse signals either.
If you FFT a square wave signal you get harmonics 1,3,5,7,9 etc to how however many you want to analyse too. So a 10Khz square wave is unreproducable as even the 3rd harmonic exceeds the bandwidth of a listenable signal
And no - a cymbol crash, drum hit, rifle shot are not impulse signals either.
If you FFT a square wave signal you get harmonics 1,3,5,7,9 etc to how however many you want to analyse too. So a 10Khz square wave is unreproducable as even the 3rd harmonic exceeds the bandwidth of a listenable signal
There is no way you can feel that. Pre ringing is at max several hundreds of samples long, so at max a few milliseconds (which is already a lot). The magnitude is so small you would not feel it, and so short you would not notice.
I think the point you're making is these test signals don't really describe the audio music we hear. If that's your point, I really agree, and I'm heavily into testing !
That's the whole reason I'm asking 'who has heard pre-ringing'...for themselves and under what conditions.
For grins to support what you said, I've attached your 10kHz square wave example,... an o'scope grab i just made, sig generator though a good RME soundcard
That's the whole reason I'm asking 'who has heard pre-ringing'...for themselves and under what conditions.
For grins to support what you said, I've attached your 10kHz square wave example,... an o'scope grab i just made, sig generator though a good RME soundcard
Attachments
Earlier I mentioned filter used to cutoff speaker. There is always some energy centered around cutoff frequency, when lowpass or highpass filter is used. You can feel it, just set frequency in midbass range.
You can even test that without pre-ringing going on
Can you feel the sound before you hear it? If not, you will never feel pre-ringing. The time it takes the sound to travel from speaker to your ears is about as long as the pre-ringing would maximally be By now probably all our ears will pretty much be used to some amount of pre-ringing. Almost every DAC uses brick-wall FIR filters that manifest some kind of pre-ringing, and almost all FFT based audio compression schemes also will have more or less ringing going on. It will be mostly masked by the audio signal itself (which is what those compression schemes rely on anyway).
When using FIR filtering, it will mainly depend on the specific filter used, active frequency frequency and the amount of filter tabs. More tabs will make the pre-ringing longer, but will also decrease its magnitude. Obviously audio delay can be a problem here depending on the use case. for Live audio or movies, FIR filtering low frequencies is often not doable or yields un acceptable levels of pre-ringing.
The question of whether pre-echo is audible depends on the context in which such filters are used (and also on the length of those filters). For example:
Where filters are used to compensate a natural post-echo, such as that of the low frequency roll-off of a loudspeaker, pre-echo is advantageous with no detrimental audible effects: It actually improves bass reproduction and gives the impression of extra bass extension (but that is another subject). In this case, the pre-echo cancels the post echo to leave a linear phase response.
The situation is different where linear phase filters are used for loudspeakers crossovers. Here the two spatially separated drivers mean than the ideal summed response is only apparent along some axis (or within a fraction of a wavelength from that axis). Pre-echo is then apparent at positions away from these more ideal listening points, where the higher the order of the filter, the more likely the pre-echo is audibly detrimental. Even in an ideal response, however, the reflections may have their own pre-echo (and different to that of the perceived source) and therefore can be audibly significant.
Anti-aliasing filters have been investigated at some depth, not least because most are inherently high-order (even "brickwall") implementations. Notably, the higher order the filter, the greater the extent of its effects pushing downwards into audible frequencies: For purely (nominally) ultrasonic filters, pre-echo is no more audible than the frequencies effected. Craven's JAES paper (March 2004) is a good read for anyone feeling the need to minimise pre-echo in anti-aliasing filters - even those that have been used earlier in the recording chain. Certainly the effect of pre-echo is less detrimental than aliasing distortion (when it is apparent).
As for the recording chain (and ignoring the signals that are routed through multiple ADC and DAC configurations in generating a final "recording"), filters exploiting pre-echo are often used for artistic purposes. Many drums sounds, for example, are generated by time reversing the naturally recorded signal and therefore have a significant pre-response. (And you certainly don't feel their cone displacement before you hear them).
Where filters are used to compensate a natural post-echo, such as that of the low frequency roll-off of a loudspeaker, pre-echo is advantageous with no detrimental audible effects: It actually improves bass reproduction and gives the impression of extra bass extension (but that is another subject). In this case, the pre-echo cancels the post echo to leave a linear phase response.
The situation is different where linear phase filters are used for loudspeakers crossovers. Here the two spatially separated drivers mean than the ideal summed response is only apparent along some axis (or within a fraction of a wavelength from that axis). Pre-echo is then apparent at positions away from these more ideal listening points, where the higher the order of the filter, the more likely the pre-echo is audibly detrimental. Even in an ideal response, however, the reflections may have their own pre-echo (and different to that of the perceived source) and therefore can be audibly significant.
Anti-aliasing filters have been investigated at some depth, not least because most are inherently high-order (even "brickwall") implementations. Notably, the higher order the filter, the greater the extent of its effects pushing downwards into audible frequencies: For purely (nominally) ultrasonic filters, pre-echo is no more audible than the frequencies effected. Craven's JAES paper (March 2004) is a good read for anyone feeling the need to minimise pre-echo in anti-aliasing filters - even those that have been used earlier in the recording chain. Certainly the effect of pre-echo is less detrimental than aliasing distortion (when it is apparent).
As for the recording chain (and ignoring the signals that are routed through multiple ADC and DAC configurations in generating a final "recording"), filters exploiting pre-echo are often used for artistic purposes. Many drums sounds, for example, are generated by time reversing the naturally recorded signal and therefore have a significant pre-response. (And you certainly don't feel their cone displacement before you hear them).
Anti-aliasing filters is a very important but separate issue regarding conversion of quantized signal back to analog form.
However, I mentioned aliasing in context of negative side-effects that appear when we transform a sound file/track by a way of convolution.
Here we have a digital/quantized signal all the time!
If I want to apply an effect to the sound or carry out the filtration, and I want to do it by convolution of two tracks - the music signal and the file that is to be used as a filter, unfortunately most software will bring strong aliasing here.
The only thing that can save us is:
1) internal oversampling feature in convolution plugin, or
2) sampling frequency of the filter higher than sampling frequency of the signal.
You guys can find out for yourself.
If some of you are not sure if his FIR setup introduces aliasing, then on this page is a very effective test file
Online Digital Aliasing Sound Test
The problem of aliasing can be assessed simply by hearing.
However, I mentioned aliasing in context of negative side-effects that appear when we transform a sound file/track by a way of convolution.
Here we have a digital/quantized signal all the time!
If I want to apply an effect to the sound or carry out the filtration, and I want to do it by convolution of two tracks - the music signal and the file that is to be used as a filter, unfortunately most software will bring strong aliasing here.
The only thing that can save us is:
1) internal oversampling feature in convolution plugin, or
2) sampling frequency of the filter higher than sampling frequency of the signal.
You guys can find out for yourself.
If some of you are not sure if his FIR setup introduces aliasing, then on this page is a very effective test file
Online Digital Aliasing Sound Test
The problem of aliasing can be assessed simply by hearing.
Last edited:
I don't thinking anti-aliasing and FIR utilization have much to do with each other.
The DAC and its particular anti-aliasing implementation seems a completely different issue to me.
The UD-501 allows anti-aliasing to be switched Off, set On normal-slope, or set On steeper-slope, up to 384kHz if i remember right.
I've listened through good phones, and looked on scope to the various sample rate and anti-aliasing combinations. Even 44.1 sounds comparatively fine with a good AA filter implementation. My personal opinion is there are far more important things to tend to for better audio.
Like FIR on its own...I still ask....
has anybody actually heard pre-ringing on their system using linear-phase FIR ?
The DAC and its particular anti-aliasing implementation seems a completely different issue to me.
The UD-501 allows anti-aliasing to be switched Off, set On normal-slope, or set On steeper-slope, up to 384kHz if i remember right.
I've listened through good phones, and looked on scope to the various sample rate and anti-aliasing combinations. Even 44.1 sounds comparatively fine with a good AA filter implementation. My personal opinion is there are far more important things to tend to for better audio.
Like FIR on its own...I still ask....
has anybody actually heard pre-ringing on their system using linear-phase FIR ?
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.