Importance of impulse response

[thanossapiens]

Well, actual frequency does not matter much, it's more about the additional crossover point increasing the number of ways by one.

Assume you had a main speaker which is good down to 40Hz, say a 3-way speaker.

When you properly integrate a subwoofer you now have a 4-way system, splitting the range below the midrange driver in two sections, one above the subwoofer XO and one below it.

This split generates an increased group delay at and below the subwoofer XO frequency and that the offending property, assuming a classical split function like Linkwitz-Riley 4th order acoustic or the like is being used.

And 80Hz (THX standard frequency) is pretty much at or even above the low-frequency core event of kick drums or the attack of plucked upright bass etc, now coming in later than the rest.

Hope it is clearer now what I meant to say.

Isn't it possible to somewhat eliminate the delay by bringing the subwoofer closer?

 

[fineMen]

Isn't it possible to somewhat eliminate the delay by bringing the subwoofer closer?

I cannot answer this question directly. I doubt such a trick would be feasible, though. I think I posted something about the necessity to acknowledge the other part. It is a cross-over, so the sub has a partner. If the cross-over is a good one, and most are, the other speaker will somehow mirror the visually 'bad' impulse from the sub. In doing so it will compensate the bigger part of the sub's misbehaviour. If, along Your suggestion, the sub was not time-aligned that wouldn't be possible anymore. Hence most likely it would make thing worse.

Just as an anecdote: someone proposed a time-corrected substraction filter that he claimed had perfect time alignment, no frequency dependent group delay, but a constant latency. I can't remember the details, but the math was sound. The implementation needed a DSP like miniDSP. Hint: the wavelength @100Hz is still about 3,3 meters.

I never followed it further, because (a) in the bass the group delay is scientifically proven of less importance, especially in smaller rooms for obvious reasons, and (b) even with a notoriously 'bad' 3rd order bandpass sub it was easy to come below 8ms of a group delay, once the amplitude response was equalized. Sound was and still is very good, but I argue it is for the latter--good amplitude response.

 

[ernestcarl]

I'd like to see data on the detectability.

If data is lacking, one can test this... if you own a KH120 or KH310, for example, ABX tests can be done with foobar2000 (will need the ABX plugin). Set one file as your untouched default control and another copy that is convolved with the desired phase EQ. I don't care about convincing anyone else, but I've proven to myself that I can hear phase correction/manipulation e.g. tweeter to mid-woofer xo phase linearization and pano phase shuffler. So it ain't always imaginary by default... Yes, it may be subtle enough to not always be clear whether the filtered response is even better/preferred or not, but at the very least, such things can be detected with controlled testing.

 

[ernestcarl]

In-room frequency response examples from full IR swept sine measurement data can be hard to come by... however, here's one mdat file compilation I found over gearslutz/gearspace today:

Bunch of Gearslutz REW results in one .mdat file - Gearspace.com

A question that gets asked a lot on this forum is: “How good are my REW measurements?”. And the way I’ve always approached this is to compare my own room results with those of other forum members, especially if they share a similar room size. I think it’s

gearspace.com

I had a look at the in-room the measurements in this set today and... some of these just look a bit worse off in one way or another e.g. phase, decay, RT, waterfall... some peculiarities can also already be seen in the IR graph as well:

The first one (red trace) may be a summed L+R sweep that's not centered precisely enough -- i.e. operator error.

 

[fluid]

Well that's just the thing. I'd like to see data on the detectability. My experience is that most acoustic phenomena that are 'subtle but detectable' are often actually imaginary.

There has been a lot of research done about aspects of audibility but a lot of it is with headphones and not speakers and sometimes the choices they made seem slightly odd.

On the audibility of all-pass phase in electroacoustical transfer functions
https://vbn.aau.dk/ws/portalfiles/portal/227876517/2007_M_ller_et_al_AES_Journal.pdf

Another masters thesis with both headphones and speakers but oddly high order FIR filters were tested.
http://lib.tkk.fi/Dipl/2008/urn011933.pdf

Audibility of Group-Delay Equalization
https://acris.aalto.fi/ws/portalfiles/portal/66449704/Audibility_of_Group_Delay_Equalization.pdf

Group Delay Distortions in Electroacoustical Systems
https://www.diyaudio.com/community/attachments/blauert-group-delay-pdf.930472/

A LISTENING TEST SYSTEM FOR MEASURING THE THRESHOLD OF AUDIBILITY OF TEMPORAL DECAYS
https://assets.ctfassets.net/4zjnzn...hreshold_of_Audibility_of_Temporal_Decays.pdf

 

[fluid]

Isn't it possible to somewhat eliminate the delay by bringing the subwoofer closer?

Group delay of filters is frequency dependent, so a pure delay or removing one will not undo it.

Edit: Although there are methods like the Harsch crossover where delay can be used with specific High Pass and Low pass slopes to improve the time response, but that comes with it's own problems.

 

[kimmosto]

audibility and preferences related to frequency variable group delay?

fluid already answered to audibility, but I'd like to comment about "preference" studies. The following by KSTR is excellent starting point:

b) develop a preference for the linear-phase version. It takes some training and obviously some passion for the topic in general...

Preference result of some study does not necessarily mean that any...majority of the listeners has experience and knowledge:
a) What kind of music material is needed / what instruments to play to perceive quality of timing
b) How minimum phase speaker sounds compared to speaker with very long excess GD at LF
c) Which one is more authentic i.e. correlates better with original signal or music recording.

I think (possibly wrong) that preference studies are primarily for commercial purposes: to find out the most important and easily detectable sound features for common consumer (to maximize sales and profit ). Why to put resources to features which are not important and easily perceivable for majority. Some part of that majority could have strong faith to brand (fanboy) or "science" focusing to preference of majority calculated with statistical functions (sciencephool). Both groups could grow some denial and deafness; "brand X engineers know what they do" or "Toole says there is no consensus". Okay, also group preferring minimum phase applications could grow false beliefs due to some coincidences in own studies, but false conclusions could fade out in few decades.

So personal preference for minimum phase speaker is result of wide and long experience with different speakers and own studies. Opinions and studies by others are not important - at least for me. This reflects also to my commercial designs. Minimum phase when it's technically possible. Otherwise as low excess group delay as possible at LF without damaging any other important feature such as directivity. No subwoofer.

 

[morillon]

rePhase

rePhase, a loudspeaker phase linearization, EQ and filtering tool. rePhase is a free FIR generation tool for building fully linear-phase active crossovers with arbitrary slopes. It also let you manually compensate for the phase shifts of your loudspeakers and existing crossovers, be it active or...

rephase.org

 

[KSTR]

Isn't it possible to somewhat eliminate the delay by bringing the subwoofer closer?

Indeed there are approaches with setback of the higher way to get some trading of constant time-of-flight delay for variable group delay across a narrow frequency range, see for example the Samuel Harsch crossover design.

I think it is not well suited to subwoofer application because

• the highpass is only second order,
• the setback is substantial (1/2 wavelength at XO frequency, like ~1.7 meters at 100Hz),
• the phase tracking is compromised, we have ~90deg offset at the intersect which is ~3dB down,but it increases at lower frequencies, becoming so large that addition of signal is already destructive (sum of both ways is less than the louder one).

The latter two points lead to very complex and almost unpredictible room modes excitation and to high sensitivity to placement, plus the variation across the room is large.

In contrast, the classical textbook subwoofer XO splitting function is a 4th order Linkwitz-Riley with following properties:

• 4th order on both slopes, and more importantly, exact same phase and same group delay on both ways (this requires a lot of work and knowledge and tools, no way you have this by default with any arbitrary mains + arbitrary sub configuration, only few products are fully developed as sets with a perfectly matching sub for specific main speakers),
• this lead to perfect summation, for a given frequency (shaped sine burst) the time domain output of both ways look exactly the same, just than one is louder than the other,
• by this also tolerant to misplacement of at least 1/3th of the XO wavelength with low magnitude and phase error. Room mode response is more predictible.

And when we now apply DSP phase-unwrapping with the mentioned inverse allpass IR we can keep all these properties and have transient perfectness as well, as the icing on the cake.

 

[KSTR]

I never followed it further, because (a) in the bass the group delay is scientifically proven of less importance, especially in smaller rooms for obvious reasons, and (b) even with a notoriously 'bad' 3rd order bandpass sub it was easy to come below 8ms of a group delay, once the amplitude response was equalized. Sound was and still is very good, but I argue it is for the latter--good amplitude response.

wrt b), I fully agree, flat magnitude response vs. frequency is the all-important factor where 90% of the efforts should aim at. Then again, once you have that, linearising phase gives a further subtle improvement... and if one uses a DRC package like Acourate or Dirac etc, phase linearisation at the listening position/area is done anyway -- unless you're optimizing for lowest latency.

FIR phase linearisation further opens a way to even compensate for the phase from a steep slope of the system highpass when it is at a rather high frequency, like 50Hz in small monitors. Say you have 6th order alignment, 4th order acoustic ported design plus 2nd electrical subsonic filter. Around 50Hz this produces a lot of phase and if you partially unroll the phase to be similar to a 3rd or 2nd order system you can considerably speed up the core bass region... at the cost of sometimes a bit "fluffy" character of the very lowest frequencies at and below 50Hz. The "center of gravity" of shaped sine bursts at various frequency now wanders much less in time than with 6th order but for the lowest notes the bursts are rendered a bit overstretched, taking more time in total... and starting a bit too early, by this.

 

[fineMen]

wrt b), I fully agree, ... . Then again, ...

I personally think, if sort of "time delay" was relevant for the human hearing, we should address the topic in relation to hearing aids. Some people need such in order to participate in social life without distress. As soon as I think of the entitlement of lone hifi-enthusiasts (in contrast to music enthusiasts) I start to feel a bit dizzy after my just taken breakfast. I had my arguments with Dr. Earl Geddes, but finally he went back into the field of real psychoacoustics / physiology and I was fine with him, all forgiven, more power to him!

You cannot name any single recording that was lacking due to the made-up problem sort of discussed here and now. You ignore the production process of a recording. Even with electronica as a musical genre, the artist would check the tone using speakers, most probably phase 'distorted' ones. If they were in the camp of "classy" hifi at all.

Reiterated I see 'scientific papers' waving in the wind like buddhistic prayer flags with no clue given what the argument is about specifically and so on. It's a void, I'm out.

 

[Holmz]

Isn't it possible to somewhat eliminate the delay by bringing the subwoofer closer?

It would need to be like a trombone slide where it comes forwards where the groupd delay is high and move back where the group delay is low.
And then if there, “are two or more tones gathered”…

 

[Holmz]

…

Reiterated I see 'scientific papers' waving in the wind like buddhistic prayer flags with no clue given what the argument is about specifically and so on. It's a void, I'm out.

Hum… Like the Dali speakers?

 

[thanossapiens]

Indeed there are approaches with setback of the higher way to get some trading of constant time-of-flight delay for variable group delay across a narrow frequency range, see for example the Samuel Harsch crossover design.

I think it is not well suited to subwoofer application because

• the highpass is only second order,
• the setback is substantial (1/2 wavelength at XO frequency, like ~1.7 meters at 100Hz),
• the phase tracking is compromised, we have ~90deg offset at the intersect which is ~3dB down,but it increases at lower frequencies, becoming so large that addition of signal is already destructive (sum of both ways is less than the louder one).

The latter two points lead to very complex and almost unpredictible room modes excitation and to high sensitivity to placement, plus the variation across the room is large.

In contrast, the classical textbook subwoofer XO splitting function is a 4th order Linkwitz-Riley with following properties:

• 4th order on both slopes, and more importantly, exact same phase and same group delay on both ways (this requires a lot of work and knowledge and tools, no way you have this by default with any arbitrary mains + arbitrary sub configuration, only few products are fully developed as sets with a perfectly matching sub for specific main speakers),
• this lead to perfect summation, for a given frequency (shaped sine burst) the time domain output of both ways look exactly the same, just than one is louder than the other,
• by this also tolerant to misplacement of at least 1/3th of the XO wavelength with low magnitude and phase error. Room mode response is more predictible.

And when we now apply DSP phase-unwrapping with the mentioned inverse allpass IR we can keep all these properties and have transient perfectness as well, as the icing on the cake.

I'm still pretty confused by this stuff, but from what I understand, a single driver speaker doesn't have these problems?

 

[gnarly]

I'm still pretty confused by this stuff, but from what I understand, a single driver speaker doesn't have these problems?

Practically speaking, correct.
The two major factors that mess up an impulse and step response are the phase rotations from crossovers; and different times-of-flight from drivers to the ears simply based on geometric distances.

A single full-range driver has neither. Too bad they can't ever play very loud, or really have good frequency extensions.

Circling back to the importance of an impulse response.....
There's obviously a lot of opinions about it...

But i know for certain if you were to strike up a conversation with a speaker designer from almost any highly regarded prosound company (live, theatre, install); and tell them you don't think a speaker's anechoic impulse response is very important, you'll find that you will no longer be taken seriously....

 

[pma]

You would need an ideal point source - not easy to build

 

[thanossapiens]

Has any company tried to make an active single driver speaker, that utilizes EQ to somewhat fix its issues etc?

 

[KSTR]

Has any company tried to make an active single driver speaker, that utilizes EQ to somewhat fix its issues etc?

Quite a few companies offered active monitors with wideband drivers in the past, e.g. Klein+Hummel M51/M52, Fostex 6301, ... but these are usually meant as additional signal control monitors when space is at a premium with less emphasis on absolutely neutral sound quality (flat response, first and foremost). Reference class models/makes from small/unknown companies may exist that I'm not aware of, though.

But some full-range enthusiasts have used DRC (digital "room correction", that is, flat magnitude and phase at the listening position) which basically realizes active speakers with dedicated shaping of the response at (or just upstream of) their amplifiers. Usually with very good, sometimes breathtaking results...

 

[Yannick Willox]

Studies have shown that listeners don't have a preference between speakers that have time aligned driver impulses, and speakers without time aligned driver impulses. Does this answer your question? As long as the frequency response is even and follows the listeners preferred curve, how the impulse response looks is mostly unimportant.

These studies seem either old or flawed.
I have seen more recent acoustics papers, describing why the overuse of diffusion in concert halls destroys the musical information and ability to time/play properly together as musicians. One of the conclusions was, hard "time-coherent" reflections are actually necessary onstage, to enable interaction between musicians.
A highly diffused stage will sound literally incoherent, and while musicians can hear each other, their brain actually refuses to understand.

How would these findings be compatible with the conclusion that time-coherence is inaudible in loudspeakers ?

 

[Yannick Willox]

To tack on to this, IRs are often recorded in spaces because you can use them as a convolution filter to simulate the reverberation as an effect on other audio. Take an IR in a church and now you have "church reverb" for your DAW. I wrote an article on the technique for Electronic Musician back in 2005 I think it was.

For that, you want a truly full range, flat, omnidirectional speaker, or you just shoot a starter pistol or pop a balloon. Turns out the cops don't love the starter pistol approach, though.

OT : I have been shooting my own IRs, with my studio monitors. I do not really agree that one needs an omnidirectional speaker. There are almost no musical instruments with an omni response, most are directional, some highly directional. Eg. a trumpet is very directional. If one puts a (with omni speaker) sampled hall on a trumpet spot, the acoustic will sound weird. A trumpet never triggers the real space in this way !

Ideally one would need to sample an acoustic with many different&directional speakers.