• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Importance of impulse response

NTK

Major Contributor
Forum Donor
Joined
Aug 11, 2019
Messages
2,660
Likes
5,820
Location
US East
To illustrate the "sensitivity" of using impulse responses to evaluate the bandwidth of a system, here are some of animated plots. Impulse responses are very revealing for high frequency limitations but not for low frequency.

IR_1.GIF


IR_2.GIF
 

ernestcarl

Major Contributor
Joined
Sep 4, 2019
Messages
3,106
Likes
2,313
Location
Canada
OK. I bite. For numerical computation reasons, frequency scaled by 1000, i.e. 0.02 -> 20 Hz and 20 -> 20 kHz. Time scale is in milliseconds.
System: 2nd order HP, Q = 0.7, at 20 Hz and 2nd order LP, Q = 0.7, at 20 kHz.

View attachment 245429

Unreachable in the vast majority system setups, but it might be useful as a point of comparison. Attachment wave file IR below is created with rePhase which anyone can import into REW.

1669166123670.png

*centering does not make a significant difference, but it helps keep the file size smaller with the number of taps used -- magnitude reduction below 10Hz is reduced.
**eh... actually, there's barely a difference below 3Hz even with the default "middle" centering. Shortens the impulse delay from 682.67 ms down to 60 which is significant for other reasons which is not relevant here.



Impulse responses are very revealing for high frequency limitations but not for low frequency.

In that case, then simply use the other graphical views already available in the program. One can still extract valuable information from it in other ways.
 

Attachments

  • 20Hz-20kHz_impulse_wave.zip
    923.3 KB · Views: 39
Last edited:

kemmler3D

Major Contributor
Forum Donor
Joined
Aug 25, 2022
Messages
3,022
Likes
5,635
Location
San Francisco
OK. I bite. For numerical computation reasons, frequency scaled by 1000, i.e. 0.02 -> 20 Hz and 20 -> 20 kHz. Time scale is in milliseconds.
System: 2nd order HP, Q = 0.7, at 20 Hz and 2nd order LP, Q = 0.7, at 20 kHz.

View attachment 245429
Whoa, thanks @NTK!

There you have it, folks. The perfect audible band IR has a rise time of about 0.06ish ms and returns to zero by about 0.4ms. At least according to me eyeballing this mockup. To be fair I think the LP filter is a little low / slow (could set it to a steeper filter at 22khz maybe) but y'all get the idea.

From now on you can call all speakers with a longer IR than 0.5ms "slow", you're welcome.

It does make me wonder how close a real speaker can get to the theoretical ideal here. The D&D 8C looks to take around 1 millisecond with full DSP on. Which when we're considering mathematical ideals, is actually really impressive... https://www.soundonsound.com/reviews/dutch-dutch-8c
 
Last edited:

RayDunzl

Grand Contributor
Central Scrutinizer
Joined
Mar 9, 2016
Messages
13,204
Likes
16,985
Location
Riverview FL
There you have it, folks. The perfect audible band IR has a rise time of about 0.06ish ms and returns to zero by about 0.4ms.

Hmm...

Ok, for grins:

Black - "perfect" impulse from above
Red - electrical measurement here, preamp out, one channel
Blue - speaker pair output
Green - speaker pair output with DRC

1669177997651.png
 

kimmosto

Active Member
Joined
Sep 12, 2019
Messages
213
Likes
498
Since you mention “sound stage”… does this mean you listen in stereo as well to evaluate speaker performance? I ask because there are those who only do mono listening test evaluations for a variety of reasons.
Of course (also) in stereo. Mono can reveal something, but stereo can reveal something else. Or may I say, stereo can reveal everything relevant for stereo listening (with two ears) which is normal usage for stereo system with two loudspeakers. This is too obvious to be totally ignored.
It is quite natural for human being that 2D-3D image created by stereo system can steal focus from something else, but that is bad excuse to skip evaluation by listening for example diffraction and other multi-source timing problems, tolerance for totally different dimensions of listening triangle, sound stage logic and integrity, sound stage projection balance and also resolution balance because sound stage (in stereo) creates part of acoustic resolution. Mono can suit for basic studies (with unexperienced listeners) of tonal balance and coloration, but that is limited view.
 

RayDunzl

Grand Contributor
Central Scrutinizer
Joined
Mar 9, 2016
Messages
13,204
Likes
16,985
Location
Riverview FL
Another grin...

Black - "perfect" impulse from above
Red - impulse from preamp out, playing a 20-20k sweep
Blue - impulse from 20-20k sweep file generated by REW and imported and measured

1669182873408.png


Looks like the DAC/Preamp (red) did a pretty good job of mimicking the stimulus (blue).

I wonder what this says about reading impulse response graphs?
 
Last edited:

fluid

Addicted to Fun and Learning
Joined
Apr 19, 2021
Messages
691
Likes
1,196
There you have it, folks. The perfect audible band IR has a rise time of about 0.06ish ms and returns to zero by about 0.4ms. At least according to me eyeballing this mockup. To be fair I think the LP filter is a little low / slow (could set it to a steeper filter at 22khz maybe) but y'all get the idea.

From now on you can call all speakers with a longer IR than 0.5ms "slow", you're welcome.
Raw impulse responses are dominated by high frequencies and as such are not very helpful at understanding the timing behaviour of speakers by themselves. A filtered IR, Step response or ETC is much more useful.

Here is a Dirac Pulse unfiltered, it is all over and done with in no time, but think about it, that short time period cannot encapsulate a 20Hz frequency.

Dirac Pulse no filter.png


When you filter it at 125Hz 1/1 all of a sudden it blows out to over 80ms. Same pulse different viewpoint.

Dirac Pulse 125Hz filter.png
 

antcollinet

Master Contributor
Joined
Sep 4, 2021
Messages
7,410
Likes
12,294
Location
UK/Cheshire

ernestcarl

Major Contributor
Joined
Sep 4, 2019
Messages
3,106
Likes
2,313
Location
Canada
Okay, so we kind of already agree it's not enough to check the the default IR graph view alone. That's fine since there's a lot more information that can be gathered even just from a single in-room measurement.

I made some re-adjustments to the crossover DSP EQ settings in my couch 5.1c setup to determine if I can lower GD a little bit further. The same mains and sub (unequalized) measurements were posted here. Nothing's physically changed in the room other than the DSP. The improvement in the time domain was rather modest, but there's no additional hit in the processing time delay from the original.

1/6 smooth, linear peak
1669192516176.png 1669192530617.png

1669192671976.png


1669192704688.png


1669192727323.png


I can't do much more about the phase kink around 125 Hz as it's caused by the room. Attempts to over-correct it with rePhase causes unwanted pre-ringing.




1669192749487.png 1669192756879.png 1669192768544.png 1669192785005.png 1669192790760.png 1669192837505.png 1669192841254.png


I think I got the idea from @BYRTT some time ago that the less difference we see between the actual measurement ~vs~ the extracted "ideal" min phase response version, the better your in-room measurements probably are... dunno how universally true that is, but it seems like a good alternative point of comparison.



I rarely listen to plain old 2.0 stereo so it's either upmixed to a variation of "pseudo-surround" MCH which works very well most of the time to improve the sense of envelopment in my dry room -- occasionally adding more processing like phase shuffling to the left and right channels and/or adding extra reverb, surround volume level adjustments etc. -- or, simply, keep the center channel in summed mono mode without a sub when listening to audiobooks/podcasts...
 

Attachments

  • COUCH minimal GD (60ms FIR adjusted) FDW 15.zip
    3 MB · Views: 30

fineMen

Major Contributor
Joined
Oct 31, 2021
Messages
1,504
Likes
679

ernestcarl

Major Contributor
Joined
Sep 4, 2019
Messages
3,106
Likes
2,313
Location
Canada
Guys, you know what You are doing with the powerful tools the contemporary computer technology gives to you? I personally, with all due respect doubt that, to a degree of being close to be positively sure.

So what do you suggest? Play all channels (e.g. 2.0/5.1/7.1) without checking if the system works well together? Sub unequalized and mains unequalized; no checking time of flight and time alignment; no bass management etc? I haven't really learned anything interesting from any of your long-winded warnings.
 

fluid

Addicted to Fun and Learning
Joined
Apr 19, 2021
Messages
691
Likes
1,196
Guys, you know what You are doing with the powerful tools the contemporary computer technology gives to you? I personally, with all due respect doubt that, to a degree of being close to be positively sure.
I'm 100% certain that if you have a point to make it would be more useful than making snide remarks.
 

fineMen

Major Contributor
Joined
Oct 31, 2021
Messages
1,504
Likes
679
I haven't really learned anything interesting from any of your long-winded warnings.
Nice term, the 'long-winded'. To correct for 20ms of a group delay @120Hz or so is clearly feasible, no doubt about it! But where did it come from to begin with? What is 'min phase'? What does the DSP do? In short, what can be learned from it?
I'm 100% certain that if you have a point to make it would be more useful than making snide remarks.
As with the 'scientific papers' you introduce once in a while: no explanation, no specific reference. What does it mean: '125Hz 1/1'? Is it the filtered branch alone, or the sum after re-combination with the low-pass branch? In either case, how does it relate to what?

There's nothing bad with fumbling around, for sure. But publishing some kind of 'result' always needs some foundations to be explained clearly. As to connect it to common knowledge. It sounds trivial, but from my practical experience doing 'science' I know that it may become the tricky part in the process. Do take the 'short-winded' detour, no good.
 

fluid

Addicted to Fun and Learning
Joined
Apr 19, 2021
Messages
691
Likes
1,196
As with the 'scientific papers' you introduce once in a while: no explanation, no specific reference. What does it mean: '125Hz 1/1'? Is it the filtered branch alone, or the sum after re-combination with the low-pass branch? In either case, how does it relate to what?

There's nothing bad with fumbling around, for sure. But publishing some kind of 'result' always needs some foundations to be explained clearly. As to connect it to common knowledge. It sounds trivial, but from my practical experience doing 'science' I know that it may become the tricky part in the process. Do take the 'short-winded' detour, no good.
The papers I linked to earlier were to provide background that touched on the audibility of group delay. They did not need an explanation or a reference they were meant to be read and demonstrate that not everyone has come to the conclusion that phase doesn't matter. The research is far from conclusive so it is really up to the individual to read and weight it. I did in fact pull an image and provide a small synopsis of one that I thought had a generally useful recommendation of keeping the group delay under 1ms to at least 300Hz.

125Hz 1/1 is a Full octave filter with a centre frequency of 125Hz as applied in REW, the screenshot was made with settings in for clarity. Perhaps you do not use REW the explanation of the filtering is contained within the manual https://www.roomeqwizard.com/help/help_en-GB/html/graph_filteredir.html

The explanation is in the post, when you filter or reduce the bandwidth of a Dirac Pulse it looks completely different. A compact impulse indicates a high bandwidth.
A referenced explanation from Bob McCarthy on page 363 of his book Sound Systems Design and Optimization
"Time bandwidth product is the relationship between the length of the time record and the bandwidth. The relationship is reciprocal, therefore the combined value is always one. A short time record creates a wide bandwidth, while a long time record creates a narrow bandwidth."
 

Killingbeans

Major Contributor
Joined
Oct 23, 2018
Messages
4,089
Likes
7,550
Location
Bjerringbro, Denmark.
I don't understand why mixed flat FR and impulse respond.

They are two different graphical representation of the same data (when you add phase). One can be derived from the other using math.

Crappy impulse response guarantees crappy frequency + phase response and vice versa.

Look at post #121:


But some ASR will scream at you can't hear 40kHz. However, people don't scream at ribbon tweeter that can hit 40kHz too.

Personally I can't hear a thing above 16kHz. I won't scream at people who claim to hear musical information at 40kHz, but I'll struggle to keep a straight face.

I look at tweeters that go to 40kHz as free redundancy. As long as I'm not expected to pay extra for it, I don't really care.

Finally, you have super tweeter at 100kHz, which don't make sense if you do not know how to use it.

Super tweeters don't make sense at all, no matter how you use them. Period.
 

kongwee

Major Contributor
Joined
Jan 22, 2022
Messages
1,024
Likes
276
They are two different graphical representation of the same data (when you add phase). One can be derived from the other using math.
They are different depending what you look for. Fast rise and fall only can be seen on impulse respond. Mostly importantly, how much time for the transient to silent fast. Of course, the data is not enough only limited on single impulse graph from reviewer. There is frequency respond, however just a single number, 19kHz, 24kHz, 40kHz doesn't tell you the range of a tweeter. Phase is important in crossover point and time alignment. Of course all are correlated in loudspeaker design. That is the designer job.

1 meter ESL can even go 100khz if you can pump that kind of current or handle super low impedance into it. People buy them not become it is go 100kHz. I have two studio monitor where tweeter is metal tweeter. They just sound more smooth and non fatigue. Their cabinets have good damping factor not to waste the attribute of these tweeter. They can go at 24kHz. Of course with a silent cabinet, soft dome is enough for everyone. That kind of loudspeaker can easily goes over $1000. Add a few hundred you can get metal tweeter. You can compare Yamaha HS series vs MSP series. Their FR does not different much. Sonically they are very different if you are really doing studio work.

Finally, ASR fave 8361A's tweeter is metal type. It can go 24kHz but genelec do not need to reveal it. In fact all genelec's tweeters are metal including for home use.
 

thewas

Master Contributor
Forum Donor
Joined
Jan 15, 2020
Messages
6,755
Likes
16,210
Mostly importantly, how much time for the transient to silent fast.
The spectral decay is much more useful to visualise that than the impulse response, here for example of a well controlled loudspeaker
1669216865448.png

and here from a less well controlled one
1669216832805.png
 
Top Bottom