Acceptable group delay

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Your listed occupation is engineer- do you know Matlab well? The reason I ask is because I investigated this question some time back by processing songs/test tracks in Matlab and then A/Bing them with the original... it's pretty easy to do if you know Matlab and what better way to convince yourself.

I tried various amounts/types of group delay at various frequencies... for example an 8th order LR crossover at 100 Hz. That can be done with the built-in functions "butter" and "filter." I tried funkier group delay shapes by using the function "iirgrpdelay" which creates an IIR filter with prescribed group delay. My conclusion was that it is definately audible at some point, but:

- the threshold is beyond anything I'd see under "normal" operation (an LR8 crossover at 100 Hz for example)
- it takes specialized source material to pick it out unless it's really rediculous
- with music, even when it's audible there isn't necessarily a strong preference for the original (it's not offensive)

I'm sorry, but I don't have the numbers on me. At the time I moved on because it didn't seem worth further consideration.
 
People want to believe that Group Delay is some magic number that will tell them if their bass is "fast" or "tight", etc... I have heard sealed subwoofers with a Qtc of 1.3+ in a car that sounded "tight" and "fast". This goes against all "logic", right?, that an underdamped woofer in a boosted environment would sound good?

I don't have a magic bullet that will give you the sound you are after, but I know where you need not look, IME. Perhaps group delay and Qtc are not all they are cracked up to be in evaluating system perceived bass performance. Using the car example, perhaps high output, low distortion and environmental acoustics are more important.

I have done the Matlab experiments (also long ago) and I will concur that for realistic levels of delay in a well designed system, group delay is essentially inaudible, and when audible (for very high order systems) it is subtle with musical signals - changes in spectrum (frequency response) are much more noticeable.
 
Nice thread so far.

Was anyone able to come to any conclusions in regards to how noticeable group delay can be at any higher frequencies? I've always assumed that it was, but I'd be interested to hear about anyone's experiences on the subject of group delay.
 
Carefull here as generalizations of this topic will get you into trouble.

MOST studies have shown GD to be inaudible, BUT there are studies which have shown it to be audible in certain circumstances. Suffice it to say that at LF its never audible. It begins to be audible (in those special cases to be discussed below) at about 1 kHz, rises in importance until about 3-4 kHz and then falls dramatically.

The conditions under which it can be audible are headphones (of course) and directional loudspeakers in not too reflective a room. Early room reflections make GD inaudible, but if these early reflections are controlled then it can be audible. BUT the bigger effect is that the audibility of GD is SPL level dependent. In other words it may be audible at higher SPLs but NOT at lower SPLs. This is a critical factor if you want to talk about "audibility" because now you have to talk about "at what sound level".
 
  • Like
Reactions: 1 user
BHTX said:
Was anyone able to come to any conclusions in regards to how noticeable group delay can be at any higher frequencies? I've always assumed that it was, but I'd be interested to hear about anyone's experiences on the subject of group delay.

The reference everyone seems to cite on this is an article by Blauert and Laws published in JACS about 30 years ago. They give a table of thresholds:

500 Hz 3.2 ms
1 kHz 2 ms
2 kHz 1 ms
4 kHz 1.5 ms
8 kHz 2 ms


What I found experimenting was more or less consistent with this, but to hear the difference at these levels you really need specialized sounds. I could only hear a difference by using a recording of castanets. With music you can multiply these thresholds by at least 2.

Something else to consider is that these numbers are a very simplified picture of what they are trying to describe- the phase response of the speaker. I think (no evidence, just speculation) that the shape of the group delay curve is probably more important than the absolute value of the group delay at a given point - i.e. a rapid change in group delay carries as much weight as a large amount of group delay. Maybe a useful thing to look at is the derivative of the group delay curve? I don't know.

Another question I've never been able to answer is why does everyone look at group delay in time units rather than cycles? It seems to me that cycles (or radians) is a more natural way to describe what is happening to the signal. In time units, you need to know the frequency and the group delay in order to describe the change in the waveform, but in cycles you need only know the group delay to describe the change in waveform.

Displaying in cycles can change the way the data is interpretted. For example, in the table above you might be tempted to think 2 kHz is the most sensitive region and that 1 kHz and 8 kHz are similarly sensitive. If you display the same data in cycles you get

500 Hz 1.6 cycles
1 kHz 2 cycles
2 kHz 2 cycles
4 kHz 2.7 cycles
8 kHz 4 cycles

Thus we see 500 Hz is the most sensitive region in terms of the change in the waveform. I'm not sure what the correct way to view this is. Maybe someone else can add some input.
 
Thats very interesting and I tend to agree the more reasonable way to look at it. I also tedn to think that sharp changes in the GD would be the more audible rather than a gradual and broad GD effect. But those kinds of details have not been studied. The most pertinent study is the latest one by Brian Moore (not first author - don;t rememebr the others) in JAES. They claim no audibility under normal circumstances. But I find their normal circumstances to not be what I do so I tend to think that there will be situations where GD could be audible in practice. Two cyles of GD at LFs is a lot of delay.
 
  • Like
Reactions: 1 user
Suffice it to say that at LF its never audible
It is audible on some signals : try a low frequency sawtooth with this soft phase audibility. You should hear the GD effect on headphones. But on loudspeakers and real music, I'm not sure.
Same for mid/hi frequencies : GD of a typical crossover (LR or butterworth or so) is audible with the Zwicker triple tone but on music, I don't think so...
 
2 kHz 2 cycles?

At 2 kHz the wave length is 17cm. For 2 cycles it is 34cm. Many larger speakers have the cabinet width approaching 34cm. The cabinet diffraction would create the effect of "GD" of 2 kHz 2 cycles.

Is it the reason why GD is not audible with loudspeakers but headphones?
 
Roundovers indeed reduce or elliminate diffraction. Mathematically diffraction is proportional to the second derivative of the surface relative to the wavenumber. So for a given curvature the diffraction will go away above some frequency. As the curvature gets larger this frequency gets lower.
 
HiFiNutNut said:
Is it the reason why GD is not audible with loudspeakers but headphones?

With headphones you remove a lot of variables like room acoustics, cabinet diffraction, crossovers, etc... Good headphones can be very hifi, moreso than all but the best loudspeakers in many ways, but they don't image well....unfortunately. Crossfeed can sometimes remove the "in the head" sensation, but it isn't a panacaea.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.