Importance of impulse response

[fluid]

This is quite extreme example having sealed "sub" with 4th order XO at 80 Hz, 8th order XO at 380 Hz and 8th order XO at 3 kHz. Pressure loss in leading edge of ideal "slam" is about 8 dB.

Can you explain which points the 8dB loss comes from in the graph? I can't seem to work it out.

 

[kimmosto]

Can you explain which points the 8dB loss comes from in the graph? I can't seem to work it out.

In the beginning max. magnitude of step and impulse are ca. 40% from max. magnitude of minimum-phase version. 20log10(0.4)=-8dB.
That kind of perfect edge is rare signal in music so actual "dynamic distortion" due to bad timing varies from from 0 dB to few dBs. Perceptibility is highly related to music material. For example Harman's playlist (titled "Critical (Best) Music Tracks for Speaker and Room EQ Testing" on ASR) may not contain tracks which are able to reveal the problem.

 

[fluid]

Perceptibility is highly related to music material. For example Harman's playlist (titled "Critical (Best) Music Tracks for Speaker and Room EQ Testing" on ASR) may not contain tracks which are able to reveal the problem.

Do you have any suggestions of tracks you know to be revealing?

 

[fineMen]

It’s like you’re writing a ransom note with someone else’s words. You do this often. You should stop.

Because You don't see how it comes together: there is no such thing like a "single cycle of every frequency". That concept simply doesn't exist.

Sorry man but I can't get anything out of that incoherent rambling.
... It's like trying to communicate with text generator AI... then again those are getting quite advanced.

May be that You still have to think about it, now even shorter: cone moves (accelerated), pushes air, air pressurizes, pressure modulation moves outward as a wave. Example: cones moves according to electrical input of two frequency packages, one of low freq/ the other w/ high freq/. Problem as stated: when, in a timely meaning, is that high freq/ package sent to the cone? It depends on the group delay low versus high, and that may change. In one case the high freq/ package hits the cone when it is, due to the low freq/ package at its peak of excursion, in the other case when the cone is at its rest position. Difference? Depends on the linearity of the cone's motor and other mechanical parts. If the cone is moving non-linear, the group delay would make or take harmonic and intermodulation distortion. Hence changing group delay may cause different sound, but only because of the non-linearity of the loudspeaker. Or the other way round, it is not the time arrival at the ears, but the changing harmonic content what makes the difference. (People never were able to consciously qualify the slight difference that was attributed to group delay changes.)

I gave you numbers: group delay of 5ms is sufficient to move the relative position of the high freq/ package relative to a 50Hz centered low freq/ package by a quater of a wavelength, shifting position on top of the peak to at rest, namely the extremes.

As long as you don't account for inevitable non-linearities especially with speakers, the group delay thing is second. At least one should acknowledge that this happy conversation started with a question regarding the merits of a 1-way speaker, that shows a visually 'nice' impulse (mostly, if not only due to lacking bass), but fails miserably in nearly all other fields. I think the impedance was quite o/k ...

 

[thewas]

Problem as stated: when, in a timely meaning, is that high freq/ package sent to the cone? It depends on the group delay low versus high, and that may change. In one case the high freq/ package hits the cone when it is, due to the low freq/ package at its peak of excursion, in the other case when the cone is at its rest position.

A simplified example: define the group delay, the arrival time of the h/f package as 0, then in one case bass is delayed by say 10ms, but in the other case by 15ms. That means that given an additional low frequency of 50Hz the h/f package could sit on the peak of the excursion of the membrane, while in the other case it would sit on just it's rest position.

I gave you numbers: group delay of 5ms is sufficient to move the relative position of the high freq/ package relative to a 50Hz centered low freq/ package by a quater of a wavelength, shifting position on top of the peak to at rest, namely the extremes.

I am afraid that a simple calculation would have spared you and us from your group delay modulation theory:
Lets assume a 2 cm midwoofer cone travel peak to peak (in reality it is usually even less), the corresponding delay difference would be only 0.06 ms, so practically irrelevant in terms of group delay audibility.

 

[kimmosto]

Do you have any suggestions of tracks you know to be revealing?

Piano recordings usually have adequate hammer hit for "force measurements" by skin, though hammer hit is totally lost in some recordings. I have not listened the last album with minimum phase speakers, but it should be okay:
Trio Töykeät, High Standards (whole album):

Stella Starlight Trio, e.g. Tainted Love:

I've tested also some snare drum samples with headphones, but not with speakers. All-pass compatible with multi-way having steep slopes sounds like pressure/volume loss of ca. 2 dB.

 

[RayDunzl]

I believe you can listen from -450 to 450us. It is not a really a sweep as our ears can perceived. If we take -450us to 450us as a full cycle. it is around 1kHz. It should be much higher physically in sound. You can generate it as a sound clip and measure it through sound meter to the actual frequency that our ear perceived.

The playback in-room of a Single Sample Full Scale input gives an audible "click".

An in-room recording of that "click" and the calculated Impulse Response from a log swept sine are remarkably similar.

I did not attempt to analyze the frequency content, as it would obviously be primarily a high frequency. Low frequency ambient and microphone noise would surely spoil any measurement attempt.

Importance of impulse response

Pardon the interruption, but... Long ago it bothered me that software could take a sine sweep (no edges) and create a measurement (impulse or step) that does have "edges". So I set out to find the difference between "real" and "calculated" values. Real: Single Full Scale Bit sent through...

www.audiosciencereview.com

 

[Schollaudio]

How do we know what's good, bad or indifferent with our systems in our rooms?

These measurements are apples and oranges but show driver differences and combined. I'm sure if these were taken outside they would look different.

Fist one is tweeter only no crossover woofer not connected at 1 foot in room. Second is woofer only 3rd order crossover tweeter not connected 1 foot in room. Third is the system at 9 feet in the room.

 

[MAB]

Because You don't see how it comes together: there is no such thing like a "single cycle of every frequency". That concept simply doesn't exist.

I will agree that I don’t know everything. You are making it difficult to have any reasoned discussion, so please be constructive and you and I will learn. And delta functions are great for… learning. Perhaps you aren’t translating well. If so you should be more careful when you lift other people’s language. And more careful in general, you seem to attack and belittle rather than engage. Again, maybe the language barrier is preventing you from clarity. And if you don’t understand, you should ask.

 

[fineMen]

I am afraid that a simple calculation would have spared you and us from your group delay modulation theory:
Lets assume a 2 cm midwoofer cone travel peak to peak (in reality it is usually even less), the corresponding delay difference would be only 0.06 ms, so practically irrelevant in terms of group delay audibility.

Thank You very much for asking back. I was actually not referring to the excursion as the reason for a time-delay. Is the scenario so easily misunderstood, I wonder?

The group delay originates somewhere else, maybe in an all-pass filter, in the inductivity of the motor's coil, an x-over pp.

The lines depict the cone's movement as the excursion. The pretty hills shall illustrate a sine burst with limited bandwith (s/ e/g Linkwitz' site). It makes a difference in regard to distortion if the h/f burst occurs at one or the other time referred to the timing of the l/f burst.

 

[thewas]

The group delay originates somewhere else, maybe in an all-pass filter, in the inductivity of the motor's coil, an x-over pp.

But this would be rather the same for those two positions, also we should keep in mind that the driver itself behaves mainly like minimum phase system in its passband. The group delay is mainly a result of the transfer functions of the drivers and the crossover and can be thus even modelled and predicted accurately.

The lines depict the cone's movement as the excursion. The pretty hills shall illustrate a sine burst with limited bandwith (s/ e/g Linkwitz' site). It makes a difference in regard to distortion if the h/f burst occurs at one or the other time referred to the timing of the l/f burst.

And this difference is from (group) delay point of view as I wrote above 0.06 ms which was your point.

 

[fineMen]

But this would be rather the same for those two positions, also we should keep in mind that the driver itself behaves mainly like minimum phase system in its passband. The group delay is mainly a result of the transfer functions of the drivers and the crossover and can be thus even modelled and predicted accurately.


And this difference is from (group) delay point of view as I wrote above 0.06 ms which was your point.

Alas, your understanding isn't right. Why shouldn't it be possible to introduce an extra time delay for the high frequency part, I just change the input signal for the driver. For the group delay high f/ versus low f( I can select nearly every value and so shift the timely relation between the low frequency burst relative to the high frequency burst to my liking.

The timely position of the high f/ burst can be freely chosen in relation to the maximum excursion of the cone, and the latter is determined by the low f/ burst.

 

[danbei]

@thewas I think the point is about distortion from over excursion, whether than the position of the cone introducing group delay:
- In the first case, not much IMD because the high frequency content makes the cone move in its linear range
- In the second case, the high frequency content makes the cone move out of its linear range because it is already at its excursion limit due to the low frequency content. Hence, more IMD.
I have no idea if this is a thing to actually worry about, just trying to reformulate.
If it's the case I guess it then raises the following question: when an experience concludes group delay correction is audible, is it because of the group delay being actually audible or because of the IMD "profile" being different?

 

[MAB]

@thewas I think the point is about distortion from over excursion, whether than the position of the cone introducing group delay:
- In the first case, not much IMD because the high frequency content makes the cone move in its linear range
- In the second case, the high frequency content makes the cone move out of its linear range because it is already at its excursion limit due to the low frequency content. Hence, more IMD.
I have no idea if this is a thing to actually worry about, just trying to reformulate.
If it's the case I guess it then raises the following question: when an experience concludes group delay correction is audible, is it because of the group delay being actually audible or because of the IMD "profile" being different?

Seems this went from “importance of impulse response” to “importance of not driving at or near max excursion because of nonlinearities”. It does need to be worried about in this corner case, otherwise not. Not exactly relevant to the thread though.

 

[NTK]

The talk about IMD is an elementary mistake on fineMen's part.

The phase relationship between tones in a general signal is arbitrary. The crests of two tones in the source signal may align or not align. A phase shift by the loudspeaker can make the resulting crest factor lower or higher, depending on the source signal.

And if the two (or multi) tone signal is long enough, the crests will eventual align, regardless of the "initial" phase difference. Here is a little bit of math:

From the last line, you can understand the summed wave as an amplitude modulated signal. The right hand term gives a sine wave at the average frequency of the two original waves. The left term gives the "beat frequency" of the summed wave which is half the difference of the frequencies of the original waves. Since both sinusoidal terms can oscillate between -1 and 1, the max amplitude of the summed wave is 2, regardless of the phase difference of the 2 waves.

[Edit]: Took me a few tries to get my math right.

 

[thewas]

Alas, your understanding isn't right. Why shouldn't it be possible to introduce an extra time delay for the high frequency part, I just change the input signal for the driver. For the group delay high f/ versus low f( I can select nearly every value and so shift the timely relation between the low frequency burst relative to the high frequency burst to my liking.

The timely position of the high f/ burst can be freely chosen in relation to the maximum excursion of the cone, and the latter is determined by the low f/ burst.

And once again I showed you that for the maximum possible different position of the high frequency signal in relation to the woofer position the delay difference is only 0.06 ms. If we had drivers with 40cm excustion that diifference would become 1.2 ms and then could be relevant, but such a wide band driver would have other much more problematic issues.

Seems this went from “importance of impulse response” to “importance of not driving at or near max excursion because of nonlinearities”. It does need to be worried about in this corner case, otherwise not. Not exactly relevant to the thread though.

Exactly, in this topic we are discussing about time related distortions.

 

[fineMen]

And once again I showed you that for the maximum possible different position of the high frequency signal in relation to the woofer position the delay difference is only 0.06 ms. If we had drivers with 40cm excustion that diifference would become 1.2 ms and then could be relevant, but such a wide band driver would have other much more problematic issues.

I already told You that the group delay does not originate in the cone excursion. I have simply two bursts of which one carries a lower frequency package, the other carries a higher frequency package. I'm pretty sure that you personally get the idea of such frequency packages (https://linkwitzlab.com/mid_dist.htm, but shorter).

Maybe I have a mixing console, a digital sound workstation, whatever. I'm pretty much able to mix the two bursts together. And I can mix them together into one "signal" so that the high frequency burst occurs first in time before the low frequency burst (a). I mix another signal, in which the two occur at the same time (b). Yes, we can do that today.

In result I have two different signals, bear with me.

Would the two sound different to a human's hearing apparatus, when converted into sound by a loudspeaker? The proponents of a so called "time correct" speaker say "sure, because of the time difference".

My question is: could it be that it is not the time difference, but the non-linearity of the speaker that makes the difference? Obviously, sorry for colloquial language, the speaker has a lot more work at a time, if the two packages fall into the same time slot. If they spread out in time the speaker has less to do, because the workload is distributed.

Now comes the hard part. Such a shift in time can occur within the speaker itself. Signal (b) is fed in, but the cone, sorry for colloquial language, moves like (a), and because of that the sound waves would look as if signal (a) was fed into an ideal speaker. Since the non-ideal speaker has that extra group delay, it would more easily produce the sound waves, which might sound different. Hard to grasp, I know. But such a situation might occur with even a tiny group delay of just 5ms between bass and mid-frequencies.

('Think it doesn't make sense to ask for pre- and post-masking within the human ear, still in discussion? 'Think most of you shiver when I show you how viciously cone excursion modulates the acoustic output. )

 

[thewas]

I already told You that the group delay does not originate in the cone excursion. I have simply two bursts of which one carries a lower frequency package, the other carries a higher frequency package. I'm pretty sure that you personally get the idea of such frequency packages.

Maybe I have a mixing console, a digital sound workstation, whatever. I'm pretty much able to mix the two bursts together. And I can mix them together into one "signal" so that the high frequency burst occurs first in time before the low frequency burst (a). I mix another signal, in which the two occur at the same time (b). Yes, we can do that today.

In result I have two different signals, bear with me.

Such would be signal related delays and those are irrelevant for what we are discussing here, namely audio system related delays.

Would the two sound different to a human's hearing apparatus, when converted into sound by a loudspeaker? The proponents of a so called "time correct" speaker say "sure, because of the time difference".

Depending on the delay those can be audible or not, quite some research has been done in that field which had been linked in this thread.

My question is: could it be that it is not the time difference, but the non-linearity of the speaker that makes the difference? Obviously, sorry for colloquial language, the speaker has a lot more work at a time, if the two packages fall into the same time slot. If they spread out in time the speaker has less to do, because the workload is distributed.

As said the loudspeaker cannot really change the delay outside of its group delay due to filter functions and cone excursion.

Now comes the hard part. Such a shift in time can occur within the speaker itself. Signal (b) is fed in, but the cone, sorry for colloquial language, moves like (a), and because of that the sound would be as if signal (a) was fed into an ideal speaker. Hard to grasp, I know. But such a situation might occur with even a tiny group delay of just 5ms between bass and mid-frequencies.

As stated above such can have impact to intermodulation distortion but is not temporal (phase/group delay) related and thus not topic of this thread.

 

[fineMen]

Such would be signal related delays and those are irrelevant for what we are discussing here, namely audio system related delays.
As stated above such can have impact to intermodulation distortion but is not temporal (phase/group delay) related and thus not topic of this thread.

Funny, what else to say?

 

[danbei]

Spoiler: Off Topic

The talk about IMD is an elementary mistake on fineMen's part.

The phase relationship between tones in a general signal is arbitrary. The crests of two tones in the source signal may align or not align. A phase shift by the loudspeaker can make the resulting crest factor lower or higher, depending on the source signal.

And if the two (or multi) tone signal is long enough, the crests will eventual align, regardless of the "initial" phase difference. Here is a little bit of math:

View attachment 245941

From the last line, you can understand the summed wave as an amplitude modulated signal. The right hand term gives a sine wave at the average frequency of the two original waves. The left term gives the "beat frequency" of the summed wave which is half the difference of the frequencies of the original waves. Since both sinusoidal terms can oscillate between -1 and 1, the max amplitude of the summed wave is 2, regardless of the phase difference of the 2 waves.

[Edit]: Took me a few tries to get my math right.

I believe your math is correct, but do the tone signal are actually long enough in musical content?
I just pulled out Izotope RX on an audio file which max true peak level is -0.04 dB FS and applied a phase shift of -45deg, the max true peak level is then +3.08 dB FS. This is actually more of a difference than I anticipated.