An example of that technique:
The old PLParEQ VST plugin did this, processing in blocks and once in both directions. The brochure dates back to 2005. The 10 band version included all-pass filters and linear phase crossovers. These days many VST EQ plugins feature something similar.
Zero delay violates causality. You can have small delay instead, something like 1-2ms.
Using block processing. In order to make a reversed version of your filter output, you would need to wait for the duration of the block to not violate causality. For example, if the sample rate is 192kHz and a block is 64 samples, the total delay in the filter would be 4* 64/192k = 1.33 ms only.
Yes jcga, that is what i was saying, hoped was coming across
My point about ignoring group delay is that I simply think group delay is a very misleading, hard to work with measurement, compared to working directly with phase.
But in fairness, that's probably due to the fact I align linear phase driver sections, and flat phase traces of multiple drivers are very easy to see how to correct. All that is needed is the correct delays between drivers. (when using complementary linear phase xovers)
I can see however, if working with IIR and trying to tie phase curvatures/rotations together, that group delay could be more valuable....simply because it's harder to make sense of curved/rotated phase traces.
Your opinion Charlie, not one that I, or what appears to be a growing number of folks share...
Maybe all-pass filters are key to accommodating the odd-order functions as well. That is, if an odd order filter is compensated with the closest even-order phase, the remaining first-order term would correct itself due to its inherent properties.
I'm not sure what the make of the start frequency of the sweep needing to be so low, I have trouble in my own room running super low sweeps because the speakers have enough output at 5Hz to actually cause the house frame to vibrate, so I have to start at 10Hz or 15Hz and have never seen such negative step in my own measurements, I would try 10Hz first and see if it works unless you did already.
Outdoors you would not have the same vibration issue but I wouldn't want the sweep to be too long at such low frequencies as the excursion would be significant.
There are quite a few studies on the audibility thresholds of group delay and they are quite frequency dependent. Of course audible does not mean "bad" but there are certainly differences that can be heard and where in frequency and the length of time determines if they could be an issue. 2ms at 300 Hz is on the threshold of regularly being audible, 1ms is almost always inaudible.
The biggest culprits for audible group delay would be 4th order or higher crossovers at 200Hz or lower. In my own experiments I find that there is not much to worry about above 1K and as you say the waveform can be completely mangled and sound no different but below 1K and particularly below 400Hz to 100Hz time performance matters much more and changing it is much more audible.
This matches up with the theory of time based hearing at low frequencies proving to be increasingly valid in the more recent studies than place based theories like the old Helmholtz view that phase does not matter.
There is also the question of the phase changes that the room imparts and I have found myself that equalizing gently for a gradual phase shift to bring the sum at the listening position to be more representative of minimum phase sounds better to me than no correction or a full linear phase correction. This isn't easy to do well and it is actually much easier to screw it up completely.
Certainly you can make a great sounding speaker without any form of time correction, but if you are searching for optimizations or the last few percent out of a design it is worth exploring I think.
I just checked and I'm sorry, that's not true at all. It's for only even order as the transfer functions need to be real. Sorry again.
It's definitely audible, if there is enough of it. Search for "temporal masking", specifically "backward masking". It is generally accepted that backward temporal masking is only effective for about 25 milliseconds. Whether the conditions used to measure backward temporal masking apply equally to audibility of pre-ring may be an open question.
Here is a recent paper on the audibility of group delay peaking. An interesting study and they also plot up the data from several previous studies on the subject. Peaking of the GD results in ringing in the time domain, but to get the threshold down to the 2msec level or so the researchers often rely on synthetic "click" type signals that are similar to a dirac pulse in the time domain and then listening is done via headphones. It has been shown before that listening to music signals in a room provides a high level of masking of these effects such that under those conditions the threshold is pushed up by several milliseconds. Only a pathalogical crossover filter would generate e.g. 4-5 msec of group delay peaking at say 1kHz. So, for the most part, doing EQ of phase or GD is not going to result in audible benefits unless your primary music consists of clicks...
I have personally heard pre-ring in two situations.
The first is when I've used a large number of FIR taps at a moderate sample rate, and I phase linearized the spectrum all the way through the lowest frequency response my speakers produce.
On kick drums, or something with a strong low frequency transient playing alone, there's something I would call a pre-echo,
or maybe like the sound or air sucking inward before being expulsed majorly outward. Reminds me of the sound of very powerful firecrackers going off...(we called them M-80's as kids)
Acoustic measurements of that processing setup show a step response with a large dip below t=0, like in my opening post. So, I learned to avoid such a dip by not phase linearizing low end roll-off.
Second situation is in a Fabfilter demo.
Anyway, on all my 4&5 way DIYs which use steep complimentary linear phase xovers, I have not heard any pre-ring at all. I think such xovers present no problem. And this thread shows that phase linearization of IIR xovers is OK too,
I think a system high pass should be IIR for reason stated, and I think one needs to be thoughtful/careful about how much phase-only linearization is applied to drivers' responses.
Then, the only potential pre-ring I'm aware of from multi-way linear phase xovers, is that there needs to be acoustic complementary response to fully null pre-ring, So to the extent a speaker has bad lobing problems, maybe the pre-ring could be heard. But frankly I doubt it. heck, how good are we at hearing lobing anyway, other than moving around to spots where sound gets screwed up.
Your case though, one of full range drivers is interesting. On the surface, a full-ranger without any xovers, should only need minimum phase correction, to both flatten magnitude and flatten phase at the same time.
However, the ease of doing that with a good FIR generator compared to using IIR, be it via DSP or analog, is compelling imo.
And if you're also using a sub with a xover, the case gets stronger i think.
So maybe FIRs advantage over IIR is more about ease and precision, than about phase. (for a full-ranger)
@gerchin, curious when you've heard pre-ring, and what the processing situations were....
Last edited:
I cannot say without any doubt that I have. I suspect that I heard it with some of the very early DAT machines that used extremely sharp linear-phase anti-aliasing filters. I perceived it as a smearing of transients, particularly high-frequency transients like top-hat, cymbals, etc. But I cannot say for certain that this was caused by pre-ring.
Oops, got a phone call when I was finishing this earlier post and I forgot to attach the paper... so here it is.
Thanks indeed for the paper, much appreciated.
The second order highpass transfer function may or may not be significant depending on the drivers involved. For example, a ribbon tweeter might be better served with a standard windowed-sinc highpass filter (HPF) since those roll off at an rate equal to the complementary lowpass filter (LPF), in this case not a huge problem since the crossover would occur at a high enough frequency to require a fairly short FIR. On the other hand, a low mid crossing over at say 150 Hz theoretically needs to be well behaved to at least 50 Hz (about 1.5 octaves down), but here frequency response wiggles are less audible at low frequencies because Fletcher-Munson equal loudness contours show FR variations of X dB power are perceived as smaller than X dB by the ear, and also the room itself tends to impose far greater variations. Here even a moderate Bessel LPF and delay-derived HPF would perform well with decent processing efficiency. Moreover a Linkwitz transform (LT) could be used to equalize the mid speaker to flat well below the passband though the HPF may be 12-18 dB down at the driver's ultimate -3 dB point, and the LT also usefully cleans up the mid's group delay for better matching with the woofer.
Phase response is audible depending on signal. While few people listen to clicks for enjoyment (Bat Boy may disagree), some instruments, drums in particular, do have steep attacks. Anyone who's hung out with musicians knows snare drums are rude: if a strong hit doesn't make you blink then the drummer isn't trying. Phase rotation of a 4th order Linkwitz-Riley (LR4) at 250 Hz crossover can be heard reliably once one knows what to listen for: the filter makes snares sound more polite than they really are. Bass drums are another matter; so far my ears cannot discern the effects of a 250 Hz crossover, but a LR4 100 Hz crossover blurs the distinction between kick drum and bass guitar, at least on Steely Dan tracks. I haven't tested whether phase is audible higher up, although Robert Orban in conversation indicated phase becomes less important above about 1 kHz. YMMV.
Possibly the disagreement about whether phase is audible or not may come down to the listener's preference: orchestral music tends to less percussion than pop or jazz music so chances of noticing a difference on that material are fewer and far between.
That's good paper thanks.
Personally, I think it continues to support traditional findings about group delay, that have primarily studied upper-mid through high-frequency ranges.
With this paper using a synthetic high-hat, a pink sweep (dominantly hf-vhf content), and a real castanet.
These traditional type studies imo, have been about group delay rather narrowly defined to that upper frequency range... and I have little reason to doubt them.
But I think phase audibility or group delay audibility, has to do with the low frequencies, how they integrate with the upper frequencies and set a foundation that is truly in time with the rest of the spectrum.
As measured by an excellent looking impulse response, or step response. And as heard by a greater sense of rhythm and dynamics.
Also from what i read from JJ and academics, the ear's critical bands increase their ability to detect phase as frequency lowers.
I also see the study, like virtually all previous group delay studies, relies on headphones. Ostensibly because because they isolate the listening environment, but they also have little phase rotation/group delay of their own compared to typical speakers.
I do not find headphones capable of imparting low frequency content the way speakers do. They can impart steady-state tones equally, but transient bass impact, not close for me.
So I don't think they are capable of evaluating phase/group delay in the frequency range where it appears to matter.
There's also a visceral bodily feel to good bass, imo. I believe hearing is a bodily expertise, that integrates the ears, skin, body cavities, and brain.
I think it takes speakers to get this visceral integration, and establishes the test bed necessary to test the evaluate phase/group delay.
So bottom like, i think headphone-only studies are likely to continue to reinforce long held dogma, that hasn't been fully proven thru the spectrum.
And in fairness to researchers, I think phase audibility test down low with speakers, hasn't been feasible due to the historical difficulty in achieving flat phase though xover regions......particularly the xover to sub.
It's not the attack that is relevant but rather the decay as that would mask any ringing due to group delay peaking etc. Figure 1 of the paper I attached to post #74 shows some research data down to 250 Hz on this, and the trend is a rising audibility threshold reaching already 5msec at 250Hz. This was for a test signal consisting of a 50usec half-cycle pulse, which is not at all like a real signal from an instrument, etc. It ends too abruptly, but this is the type of signal that is most often used in these studies because it does not mask time domain ringing. Music, even drums and such, will have a much higher threshold due to the signal's own die-down behavior.
Keep in mind that you can often convince yourself of just about anything in an uncontrolled, sighted test. These are rigorous scientific studies and I tend to believe them.
I seem to recall, but do not have at my fingertips, data that showed the audibility threshold rapidly increasing below 100-150 Hz. If you look at the group delay of any low frequency driver, even a closed box one, the group delay climbs to tens of milliseconds when Fc is low. Remember, the driver+box are in essence a type of filter in the system, and it is the low corner frequency that gives rise to the group delay for that driver's output. If you consider the audibility threshold not as X milliseconds but rather as a percentage of one cycle at a given frequency the value is not diverging all that much up until 2kHz or so, where it more or less levels out.
Just a hint of foreshadowing: a variant with 24 dB / octave highpass slope is under development.
I think that it's a matter of: do I care enough to try it and preserve phase or not. Can we hear phase differences at higher frequencies?
Dr. Griesinger says we can:
Apparently we can also hear it in the click sounds. The onset of notes is what Dr. Griesinger tells us makes us define if a sound is close or not. If we want to preserve as much as is in the recording, it does make sense to me to preserve the phase of whatever is left of it in the recording. Granted, we don't have control over the recording itself, but we can stay as close as possible to it.
Even to a level that I've made sure that my speakers in my room preserve it all the way up to the listener's seat (among other things, by making sure I get a very low level of early reflections).
What does it do for me? It makes me take notice, it draws attention, the proximity effect Dr. Griesinger speaks off seems very real to me. I feel I get closer to that "live vibe" of real instruments, of someone playing or singing in front of me. I wouldn't go trough all that trouble for background music, but I like what it does for the music I enjoy (which very often isn't the orchestral kind). I've tried it both ways, I prefer one of them. That's enough for me, no matter how much denying there is about the audibility. I'd recommend anyone curious enough to do the same: just try it! And not over headphones, as that is partially flawed. My headphones never made me "skip a beat", my speakers sure did. They have startled more than one listener. That's the proximity effect in action for me. But to get there we need to preserve what is there .i.m.h.o.
It just works for me, especially since we 'feel' the music as much as we hear it. So to me, the timing does matter. I perfectly understand if others might feel different. But do grant us the same courtesy. The tests show it can be heard, under some circumstances. An expert in the field (Dr. Griesinger) is convinced we can hear phase, even at higher frequencies and early reflections do detract from our ability to do so. Allow us the room to prefer it. You may feel different about it
.
And for a change, this is something we actually can measure.
Dr. Griesinger says we can:
Apparently we can also hear it in the click sounds. The onset of notes is what Dr. Griesinger tells us makes us define if a sound is close or not. If we want to preserve as much as is in the recording, it does make sense to me to preserve the phase of whatever is left of it in the recording. Granted, we don't have control over the recording itself, but we can stay as close as possible to it.
Even to a level that I've made sure that my speakers in my room preserve it all the way up to the listener's seat (among other things, by making sure I get a very low level of early reflections).
What does it do for me? It makes me take notice, it draws attention, the proximity effect Dr. Griesinger speaks off seems very real to me. I feel I get closer to that "live vibe" of real instruments, of someone playing or singing in front of me. I wouldn't go trough all that trouble for background music, but I like what it does for the music I enjoy (which very often isn't the orchestral kind). I've tried it both ways, I prefer one of them. That's enough for me, no matter how much denying there is about the audibility. I'd recommend anyone curious enough to do the same: just try it! And not over headphones, as that is partially flawed. My headphones never made me "skip a beat", my speakers sure did. They have startled more than one listener. That's the proximity effect in action for me. But to get there we need to preserve what is there .i.m.h.o.
It just works for me, especially since we 'feel' the music as much as we hear it. So to me, the timing does matter. I perfectly understand if others might feel different. But do grant us the same courtesy. The tests show it can be heard, under some circumstances. An expert in the field (Dr. Griesinger) is convinced we can hear phase, even at higher frequencies and early reflections do detract from our ability to do so. Allow us the room to prefer it. You may feel different about it
.And for a change, this is something we actually can measure
.FYI, for some historical info, I found a related thread on this subject from back in 2003:
https://www.diyaudio.com/community/threads/audibility-of-group-delay.14274/
https://www.diyaudio.com/community/threads/audibility-of-group-delay.14274/