I've read my share of those kind of threads, and enough opinions on it as well as research. If all of those findings aligned it would be something. But they don't.
So I rely on what I tested, several times (yes I've heard pré ringing in several of those tests) and try and stick to minimum phase following the frequency bandwidth I get at the listening spot. Which isn't exactly the same as linear phase throughout it all, but more like what 'a perfect source' might be able to do.
Heck, I'm not even trying to convince anyone, so I should just stop typing.
So I rely on what I tested, several times (yes I've heard pré ringing in several of those tests) and try and stick to minimum phase following the frequency bandwidth I get at the listening spot. Which isn't exactly the same as linear phase throughout it all, but more like what 'a perfect source' might be able to do.
Heck, I'm not even trying to convince anyone, so I should just stop typing.
I think the pre cursor paper by the same group was actually more relevant and useful.
https://acris.aalto.fi/ws/portalfil...udspeaker_Group_Delay_Characteristics_AAM.pdf
I feel much the same but it is hard when is it so simple to rig up an ABX that anyone can try for themselves and hear the difference, cutting out the usual uber objective viewpoint of "nothing matters" and anything reported as sounding different is automatically delusional ravings of the subjective mind.
I don't have specific samples that I saved, maybe some old Fir corrections but it would also need the matching recorded impulse and I'd need to convolve it with a piece of music or something like that to be able to listen for it.
To be quite honest, what I described as being able to hear the strong on-set of signals, that won't work if there's lots of pré-ringing before the peak. In fact, the on-set part of the signal can actually get soft if there's too much correcting going on (especially when there's high frequency parts in the ringing). Beside that bass build-up effect Mark mentioned, it's something to watch out for. My FIR correction is applying pure minimum phase correction above 200 Hz. Well, except for the mid/side EQ that I do in linear phase.
Another controversieel subject I'm sure...To test it, I've exaggerated the corrections I applied with automated tools like DRC-FIR and a high number of cycles used for the correction in linear phase.
I've been fanatical about it, trying to learn as much as I can about FIR corrections over a period of at least half a year. There isn't much I haven't tried or compared.
I don't believe the length of the FIR file, nor the sample rate is the cause of excessive pré-ringing, except that the high tap count makes a precision correction at bass frequencies possible, especially at lower sample rates. It probably was build up of bass due to a mis-match of signal correction. The corrective FIR filter starting at a lower frequency than the measurement and trying to make up for what wasn't there.
Thank you so much for posting the convention paper from this group. Great stuff. IMHO figure 11 is a very succinct summary of group delay audibility versus frequency.
This thread has split in two directions: (1) audibility of pre-ring, and (2) audibility of group delay. Understand that pre-ring and group delay are not the same thing; you can have pre-ring without group delay (a zero-phase filter does this) and you can have group delay without pre-ring (practically any minimum-phase filter does this).
Also, group delay per se is not bad -- you can have five minutes' group delay but if it's the same at all frequencies then the only effect is that the signal comes out five minutes after it went in. It is changes in group delay with frequency that may be a problem. Personally, I believe that group delay that decreases monotonically with increasing frequency is probably an objective worth pursuing.
Also, group delay per se is not bad -- you can have five minutes' group delay but if it's the same at all frequencies then the only effect is that the signal comes out five minutes after it went in. It is changes in group delay with frequency that may be a problem. Personally, I believe that group delay that decreases monotonically with increasing frequency is probably an objective worth pursuing.
Section 2 of that paper is especially relevant to the issue of audibility of pre-ring.
Based on some recent posts that might be misconstrued I want to point out what group delay is, and is NOT.
Group delay has nothing to do with the "elapsed time". To understand this, consider the mathematical definition of group delay (-d/dw of phase). If a signal is delayed by some amount of time but then arrives intact with no phase changes it has experienced zero group delay. An example of this is "digital" or "bucket brigade" type delay. All frequencies experience the same amount of delay, so group delay is zero. OTOH, minimum phase systems (like filters or even drivers themselves) that alter frequency also produce frequency dependent phase changes, which by definition is the same as frequency dependent delay, which IS group delay.
Phenomenologically speaking, group delay is NOT the total delay, it is the delay of some portion of a packet of information (e.g. of the audio signal) with respect to the average delay for that packet as a whole (e.g. average over all frequencies, or perhaps WRT the first arrival of the packet).
Group delay has nothing to do with the "elapsed time". To understand this, consider the mathematical definition of group delay (-d/dw of phase). If a signal is delayed by some amount of time but then arrives intact with no phase changes it has experienced zero group delay. An example of this is "digital" or "bucket brigade" type delay. All frequencies experience the same amount of delay, so group delay is zero. OTOH, minimum phase systems (like filters or even drivers themselves) that alter frequency also produce frequency dependent phase changes, which by definition is the same as frequency dependent delay, which IS group delay.
Phenomenologically speaking, group delay is NOT the total delay, it is the delay of some portion of a packet of information (e.g. of the audio signal) with respect to the average delay for that packet as a whole (e.g. average over all frequencies, or perhaps WRT the first arrival of the packet).
I disagree. Mathematically, group delay is the (negative of) slope of the phase vs. frequency line. Pure time delay will produce a constant slope in the phase vs. frequency line, thus pure time delay results in nonzero group delay. In this case, group delay [-d(phi)/d(omega)] equals phase delay [-phi/omega].
Where group delay really becomes significant is when the slope of the phase vs. frequency line is not constant. Then group delay becomes a local phenomenon, because the local slope can be very much different than the overall slope. Group delay is therefore important for signals that are modulated, because it describes the delay of the envelope of the modulated signal.
Sorry that is not correct...
For a better understanding, please read the whole first paragraph, especially the last two sentences:
https://en.wikipedia.org/wiki/Group_delay_and_phase_delay
...which are...
For a modulation signal (passband signal), the information carried by the signal is carried exclusively in the wave envelope. Group delay therefore operates only with the frequency components derived from the envelope.
Which means that the average delay of the "envelope" does not matter. In essence GD is only for the minimum phase part of a system, and pure all-pass type delay is different and not part of that.
By your definition, you can reduce group delay just by standing closer to the loudspeaker! Think about it.
Last edited:
+1
Maybe we should use the term group-delay distortion when we are talking about the unwanted effects of group-delay. Unfortunatley only recently studies appeared that deal with the audibility of the crossover-induced type of group delay distortion. And there are still no exact results AFAIK. Many older studies were dealing with the audibility of narrow band group delay peaks (like the famous Blauert and Laws paper). But crossovers usually have high group delay below their crossover frequency and then decreasing with rising frequency. Some of them have a slightly peaking group delay curve (almost every one that is using filters steeper than Bessel).
I agree about 50% with your statement. The reason why I don't fully agree is because eaxcly for the subject we are discussing here, namely crossovers with added phase-linearisation, the suming of the group delay intrinsic to the crossover and the added delay from the phase equaliser result in a constant delay over all frequencies. I.e. "elapsed time".
Regards
Charles
Well, you can disagree with the mathematical definition [-d(phi)/d(omega)] if you want to. I choose not to.
I've seen this disagreement over what 'group delay' means a number of times.
It boils down to whether of not one thinks constant (pure) delay is also a form of group delay, in addition to group delay that is frequency dependent which everyone agrees about.
Personally, I don't think it makes sense to include constant delay as a form or group delay. I think all measurements that attempt to show phase / group delay should have all constant delay removed from them...simply for the measurements to be readable, show anything of use..
(It's probably also worth reminding that constant delay, evidenced by a sloping straight phase trace, must be plotted against a linear frequency scale to be straight.
It boils down to whether of not one thinks constant (pure) delay is also a form of group delay, in addition to group delay that is frequency dependent which everyone agrees about.
Personally, I don't think it makes sense to include constant delay as a form or group delay. I think all measurements that attempt to show phase / group delay should have all constant delay removed from them...simply for the measurements to be readable, show anything of use..
(It's probably also worth reminding that constant delay, evidenced by a sloping straight phase trace, must be plotted against a linear frequency scale to be straight.
I can get on board with that, as long as everyone makes their assumptions clear.