Audibility of group delay at low frequencies

[dasdoing]

Well yes, if you high pass the speaker where its excess group delay is high and have a subwoofer with a lower excess group delay that would help. This mainly helps lower minimum group delay due to the frequency response extension though. It’s worth mentioning the crossover between the subwoofer and speakers itself also adds excess group delay.

I forgot to consider, setting the delay incorrectly for the speakers/subwoofer can add excess group delay. But assuming everything has been set up correctly, the remaining excess group delay can’t be solved without preringing.

I am talking abotu putting delay in the signal. I have my subwoofer behind my couch atm. even though it is much nearer then my mains I get about 20ms excess group delay at the subwoofer frequencies. if I delay the MAINS 20ms excess group delay will be much flatter. this seams counter-intuitive, so I am in doubt if it is the right thing to do.

btw: I don't get why frequency response extension effects minimum phase group delay.

 

[Pio2001]

so, a perfect group delay for our speakers in our room would be 0 excess group delay? since a flat group delay seams to be only possible with totaly flat frequencie response

It is not a problem to have excess group delay at the bottom of nulls in the frequency response.
If the affected signal is silent, it doesn't matter.

Moreover, at medium to high frequencies, there are a lot of measuring errors at the bottom of nulls, and the software often mistake a negative peak in the group delay for a positive one (180° sudden jump in phase response... hard to tell if it is actually 180° up or 180° down when there is nearly no signal).

assuming everything has been set up correctly, the remaining excess group delay can’t be solved without preringing.

Why not ? Phase distortion is mathematically reversible.
Is it a problem of diffusion in the room ?

 

[hyperplanar]

I am talking abotu putting delay in the signal. I have my subwoofer behind my couch atm. even though it is much nearer then my mains I get about 20ms excess group delay at the subwoofer frequencies. if I delay the MAINS 20ms excess group delay will be much flatter. this seams counter-intuitive, so I am in doubt if it is the right thing to do.

It's not necessarily the right thing to do. The main thing to focus on is setting the delay so that the subwoofer and mains add up as constructively as possible in the crossover region (as time delays rotate the phase response in a linear manner).

It's possible some of that 20ms group delay is due to actual delay caused by the sub's DSP or other factors like that. However, if your sub naturally has a bunch of group delay due to things such as the sub being ported, then I would say it's not the best idea to just blindly set a 20ms delay for your mains to compensate. If your sub has a port it will still start playing notes in time, it just takes a while for the energy to build to its max, which is what the group delay line represents. So you may effectively be causing acoustic preringing by setting a delay of 20ms because the sub actually does start playing something before 20ms, but takes time to build up.

btw: I don't get why frequency response extension effects minimum phase group delay.

Well, this is kind of by the definition of minimum phase. Every minimum phase phenomenon has a direct linkage between the frequency response and the phase. If the frequency response of a minimum phase system is perfectly flat, then so will the phase. In reality, none of our systems go down to 0 Hz, so we always have an acoustic highpass filter represented in the frequency response, which causes phase shift and thus group delay. Extending the frequency response down, and so lowering the frequency of the highpass filter, will also push this phase shift down to lower frequencies.

 

[hyperplanar]

Why not ? Phase distortion is mathematically reversible.
Is it a problem of diffusion in the room ?

Yes, but not without preringing for excess phase. You end up having to balance between the audibility of preringing and the benefit of the excess phase compensation.

Here are step responses of my system which has a subwoofer crossed over.

 

[Pio2001]

Here are two complementary frequency responses (red and green), generated with strictly linear phase filters in Rephase, and their flat phase responses (purple and blue) :

Their phase response contains obviously a lot of excess phase.

If we take a clean truncated sine burst and apply the red transformation to it, we get the typical symmetric pre and post ringing that characterize linear phase filters.

Now, if we apply the green transformation to the result, we add yet more pre and post ringing... but with opposite polarities. The result of the two transformations is shown at the bottom :

Now, we can perform the same experiment with an all-pass filter. Its frequency response is flat, but not its phase response, here in blue.

The pink curve is another all-pass filter that has the opposite phase response.

Let's apply both of them successively to our signal :

Correcting excess phase doesn't add pre-ringing. Here, it even removes the existing pre-ringing !

Of course all these examples are performed in the digital domain. In the acoustic domain, it is impossible to reach such accuracy.

 

[dasdoing]

It's not necessarily the right thing to do. The main thing to focus on is setting the delay so that the subwoofer and mains add up as constructively as possible in the crossover region (as time delays rotate the phase response in a linear manner).

It's possible some of that 20ms group delay is due to actual delay caused by the sub's DSP or other factors like that. However, if your sub naturally has a bunch of group delay due to things such as the sub being ported, then I would say it's not the best idea to just blindly set a 20ms delay for your mains to compensate. If your sub has a port it will still start playing notes in time, it just takes a while for the energy to build to its max, which is what the group delay line represents. So you may effectively be causing acoustic preringing by setting a delay of 20ms because the sub actually does start playing something before 20ms, but takes time to build up.


Well, this is kind of by the definition of minimum phase. Every minimum phase phenomenon has a direct linkage between the frequency response and the phase. If the frequency response of a minimum phase system is perfectly flat, then so will the phase. In reality, none of our systems go down to 0 Hz, so we always have an acoustic highpass filter represented in the frequency response, which causes phase shift and thus group delay. Extending the frequency response down, and so lowering the frequency of the highpass filter, will also push this phase shift down to lower frequencies.

thats why I asked about "group delay in nature".
what is a proper aligned sub? one that starts to speak at the the same time as the mains, or one that peaks at the same time of the mains peak?

I guess you responded that in answer number 84

 

[Pio2001]

what is a proper aligned sub? one that starts to speak at the the same time as the mains, or one that peaks at the same time of the mains peak?

I've never studied subs, but in an ideal world, I'd say that it should peak at the same time as the mains.

 

[hyperplanar]

Of course all these examples are performed in the digital domain. In the acoustic domain, it is impossible to reach such accuracy.

Yeah in theory you can remove the excess phase without preringing but that's pretty much only going to be at the exact location your mic was at when the measurement was made. If the phase linearization filter is even 1 Hz off then preringing will appear. So I don't think we can correct for excess phase without preringing, in reality. And the effective acoustic crossover frequency shifts a bit as you move your head around...

A large chunk of LF group delay is from the rolloff of the speakers anyways as well. That can't be removed by a flat FR phase linearization filter without preringing. In theory you can do a minimum phase Linkwitz transform and make your speakers flat to infrasonic frequencies but not very practical is it

 

[dasdoing]

Moreover, at medium to high frequencies, there are a lot of measuring errors at the bottom of nulls, and the software often mistake a negative peak in the group delay for a positive one (180° sudden jump in phase response... hard to tell if it is actually 180° up or 180° down when there is nearly no signal).

found out what you mean,
whenever there is a narrow spike in GD there is a vertical jump of 360 degrees in unwrapped phase response, and there is always a null at this frequency in the FR. does this mean that all those jumps in phase don't realy exist? and if so, why can't this be visualy corrected in the phase response representation?

 

[jlo]

I've never studied subs, but in an ideal world, I'd say that it should peak at the same time as the mains.

No it should start at the same time.
Take an simple 6dB/oct crossover : when you compare the lowpassed and the highpassed time response (ie the step response), you see that both start at same time, energy peaks are timely separated but they add perfectly.

 

[dasdoing]

No it should start at the same time.
Take an simple 6dB/oct crossover : when you compare the lowpassed and the highpassed time response (ie the step response), you see that both start at same time, energy peaks are timely separated but they add perfectly.

why would that be an explanation for "should start"?
in nature the soundwaves would hit you at the same time

 

[jlo]

Not exactly : say you have two sine waves at 100Hz and 1000Hz starting at same time (ie two bursts) : they start at same time from zero but the first pressure peak of the 100Hz is at 2.5ms and the first peak of the 1000Hz is at 0.25ms (a quarter of period)

 

[dasdoing]

Not exactly : say you have two sine waves at 100Hz and 1000Hz starting at same time (ie two bursts) : they start at same time from zero but the first pressure peak of the 100Hz is at 2.5ms and the first peak of the 1000Hz is at 0.25ms (a quarter of period)

makes sense, but the first wave peak of the woofer is much quieter then that of the tweeter.

I actualy played around with rephase yesterday (simulating with convolving).
-made a filter that gave me near flat exess group delay at listening position. step response didn't change too much
-made another where I tried to make step response look as good as possible at listening position (this turned out to be impossible to make it realy good looking, but I have a triangelish first peak). exess group delay now has negative values below 80Hz
those filters alter the frequencie response so I couldn't avaluate the sound yet. will have to rew flatten them with measurments to compare

 

[jlo]

Concerning group-delay, the maximal value is about 0.15*order/frequency for a standard alignement of loudspeaker or subwoofer. Up to the sixth order, then GD=1/f, this value corresponds to one period and is generally accepted to be under the limit of audibility. So only the group-delay due to room modes that can be a real concern, not the group-delay due to the loudspeaker.

 

[dasdoing]

Concerning group-delay, the maximal value is about 0.15*order/frequency for a standard alignement of loudspeaker or subwoofer. Up to the sixth order, then GD=1/f

can you say that in laymen english, please?

 

[QMuse]

makes sense, but the first wave peak of the woofer is much quieter then that of the tweeter.

I actualy played around with rephase yesterday (simulating with convolving).
-made a filter that gave me near flat exess group delay at listening position. step response didn't change too much
-made another where I tried to make step response look as good as possible at listening position (this turned out to be impossible to make it realy good looking, but I have a triangelish first peak). exess group delay now has negative values below 80Hz
those filters alter the frequencie response so I couldn't avaluate the sound yet. will have to rew flatten them with measurments to compare

That is not how this works. Flat exces phase which is equal to zero means phase response is identical to minimum phase response. Such phase response would still not have GD equal to zero, but will have excess GD equal to zero.

Once you try to correct phase to be linear and equall to zero (even apart from natural woofer roll off) you will end up with pre-ringing in step response as even a single phase filter with Q>1 will introduce audible pre-ringing. Such phase fluctuations, as @jlo explained, are room related and related to the position of your speakers in the room. As pre-ringing is audible, and GD mostly isn't, it should be avoided at any cost.

 

[QMuse]

Here are two complementary frequency responses (red and green), generated with strictly linear phase filters in Rephase, and their flat phase responses (purple and blue) :



View attachment 67211

Their phase response contains obviously a lot of excess phase.

If we take a clean truncated sine burst and apply the red transformation to it, we get the typical symmetric pre and post ringing that characterize linear phase filters.

Now, if we apply the green transformation to the result, we add yet more pre and post ringing... but with opposite polarities. The result of the two transformations is shown at the bottom :


View attachment 67209

Now, we can perform the same experiment with an all-pass filter. Its frequency response is flat, but not its phase response, here in blue.

View attachment 67212

The pink curve is another all-pass filter that has the opposite phase response.

Let's apply both of them successively to our signal :

View attachment 67210

Correcting excess phase doesn't add pre-ringing. Here, it even removes the existing pre-ringing !

Of course all these examples are performed in the digital domain. In the acoustic domain, it is impossible to reach such accuracy.

This is not true - filter made of inverted phase response would most certainly cause horrible pre-ringing. It takes a single phase filter to cause pre-ringing. Here is an example of step response of a single filter with +45deg gain at 1kHz with Q=3. As you can see that single filter causes visible and audible pre-ringing in step response.

 

[levimax]

This is not true - filter made of inverted phase response would most certainly cause horrible pre-ringing. It takes a single phase filter to cause pre-ringing. Here is an example of step response of a single filter with +45deg gain at 1kHz with Q=3. As you can see that single filter causes visible and audible pre-ringing in step response.

View attachment 69574

Thanks for taking the time to explain things.... kind of hard to wrap my head around all of this. One quick practical questions, I have a 3 way active system with analog crossovers. In rephase I use the "Filter Linearization Tab" to generate minimum phase linearization filter to compensate for the phase shift and GD of my active crossovers (Which are LR 4). Do these "linearization fileters" cause pre-ringing or because they are minimum phase they don't?

 

[jlo]

can you say that in laymen english, please?

I'll try.
In low frequencies, a loudspeaker in a box is equivalent to a filter with an order (slope) that depends on the box type (second order for closed box, fourth order for bass-reflex,...). The maximal groupdelay is seen near cutoff frequency and is approximatively 0.15 * order / Fc, ie for a bass-reflex tuned at 30Hz is 0.15*4/30=20ms, or expressed in periods it is 0.15*4=0.6period. For a sixth order system (ie some bandpass design seen in subwoofers), the max GD would be 0.15*6=0.9period, so less than one period, one period being accepted by some as a limit of audibility.
This brings us to my point : as the groupdelay of the loudspeaker itself seems not to be audible, it can only be groupdelay due to room modes that may be a problem.
Hope it is clearer now....maybe...

 

[QMuse]

Thanks for taking the time to explain things.... kind of hard to wrap my head around all of this. One quick practical questions, I have a 3 way active system with analog crossovers. In rephase I use the "Filter Linearization Tab" to generate minimum phase linearization filter to compensate for the phase shift and GD of my active crossovers (Which are LR 4). Do these "linearization fileters" cause pre-ringing or because they are minimum phase they don't?

Let's first clarify that frequency filters can be minimum phase or linear phase while phase filters are just phase filters. The label which says "Minimum Phase Filters Phase LInearization" means that you are linearizing the phase of minimum phase frequency filters.

Below is the step response of a filter which linearizes LR4 XO at 1800Hz. While it does show some minor pre-ringing luckilly it is totally inaudible. Adding another one won't cause pre-ringing either, so you're safe with this.

As I said, I have also found out that adding a few phase filters with Q=1 (apart from XO phase correction fitlers) to further linearize the phase of the speaker also won't result in audible pre-ringing. Automatic EQ tools (Dirac, Acurate, Audiolense etc) are doing the same thing.

Plz note that this is step response of a filter, not measured step response of a speaker when this filter is applied.