So what exactly IS an IR? (technically speaking)

OK, I know what an IR is. (Impulse Response) It's a computer file that contains information that is used by the Axe-FX to closely emulate the captured sound of a speaker, cabinet, and microphone in a specific position.

But what is it, actually? How is that data presented to the computer?

Is it a sound? A captured swept tone? A data file containing specified frequencies, levels, and mathematical descriptions of the harmonics generated for a given input frequency?

What's actually in that file?

Since it appears that IRs from various sources are interchangeable, at least to some extent, which of the formats of these various IRs are compatible with the Axe-FX units, and which are not?
 
Thanks, that's helpful. I'm still not exactly sure of how the IR represents the sound of the speaker yet but after reading through it again and thinking about it for a while I'll wake up tomorrow knowing more than I do tonight.

One thing that interests me is the dynamic behavior of speakers, as exhibited by their cone breakup modes as well as their dynamic compression. I don't know that an IR can properly replicate this, as any given speaker is likely to generate fewer harmonics due to cone breakup at low levels and more harmonic content will be generated by the mechanism of cone breakup at high output levels. It's not exactly a surprise to any guitar playing veteran that you have to push many speakers to a certain volume level before the speakers start adding their own grind to the sound.

I can't see how that dynamic behavior can be captured in an IR. It would seem to me that instead, an IR is taken at a specific volume level and
will be a most accurate representation when the user sets his rig to about that same volume level. At substantially different volume levels, it should not be so accurate.

I think that in the future, there will eventually be a new breed of impulse response capture and playback, one that captures the speaker at a variety of power levels from minimum to max rating, and the capturing system analyzes the output to derive a model of speaker distortion vs. drive level. That would seem to me to be an evolutionary step forward in impulse response capture.
 
Thanks, that's helpful. I'm still not exactly sure of how the IR represents the sound of the speaker yet but after reading through it again and thinking about it for a while I'll wake up tomorrow knowing more than I do tonight.
It's in the name. The file is the speaker's response to an impulse. The "impulse" is a pop/click sound, and the response of the speaker to the impulse is also a sound file (e.g. a .wav file). But Fractal uses their own proprietary file format for copy protection.
One thing that interests me is the dynamic behavior of speakers, as exhibited by their cone breakup modes as well as their dynamic compression. I don't know that an IR can properly replicate this, as any given speaker is likely to generate fewer harmonics due to cone breakup at low levels and more harmonic content will be generated by the mechanism of cone breakup at high output levels. It's not exactly a surprise to any guitar playing veteran that you have to push many speakers to a certain volume level before the speakers start adding their own grind to the sound.
"Dynamic behavior of speakers" is a myth. Cone "breakup" results in a linear response (not distortion) fully captured in IRs. And there's no audible "dynamic compression" going on.

So "push the speakers" is a myth, and you're actually pushing the amp rather than the speaker.
I can't see how that dynamic behavior can be captured in an IR. It would seem to me that instead, an IR is taken at a specific volume level and
will be a most accurate representation when the user sets his rig to about that same volume level. At substantially different volume levels, it should not be so accurate.
A guitar speaker's response can be captured with just a few microvolts input, and its response will be the same as if you gave it a buncha volts.
I think that in the future, there will eventually be a new breed of impulse response capture and playback, one that captures the speaker at a variety of power levels from minimum to max rating, and the capturing system analyzes the output to derive a model of speaker distortion vs. drive level. That would seem to me to be an evolutionary step forward in impulse response capture.
Fractal Audio has shared they've already tried this extensively in their research, and that it made not a bit of difference.
 
  • Like
Reactions: Rex
So you're saying that a speaker is a LINEAR device? That is at variance with the opinions put forth by a variety of speaker design engineers. Particularly in the high end range of the speaker market, driver designers spend the majority of their engineering efforts on trying to ensure that the driver's motion remains purely pistonic without harmonic breakup modes coming into the picture at high drive levels.

I do not have access to a laser interferometer but it may be possible for me to come up with a test setup to sweep a given speaker at various output levels, using a very clean source and amplifier, at levels from the threshold of audibility to the 1 dB compression point that is defined by many speaker designers as the full power handling capacity, and capture the spectral output at each. Say at 10 dB increments.

I am sure beyond any doubt that the harmonic content will change as a function of output. It won't be the same spectral plot when normalized to the same level.

Speakers do not exhibit linear behavior for a double handful of reasons.

If you only take an IR at a specific level, then that IR will only be accurate at that level. It cannot be otherwise. Because speakers are non-linear devices.

https://www.klippel.de/fileadmin/_m...linearities–Causes_Parameters_Symptoms_01.pdf
 
Clif’s comments...
Oct 21, 2015
...IR technology is great. It's used by EVERY modern modeling product (and "profiling" product despite what some would want you to believe) because it works very well. The only thing it does not capture is nonlinear effects. At this time DSPs do not have the requisite power to model the nonlinear aspects (well they do but then they wouldn't be able to do much else). However the nonlinear aspects are minimal and almost always dwarfed by the nonlinear aspects of the amplifier.
 
I’m not sure but were there another DSP chip or core added to a guitar processor, IRs might give in to: speakers in cabinets under test, thiele-small parameters extracted, whatever else might be needed...and then it’s response gets “modelled’ like everything else in this beautiful box. :)
 
To state that the nonlinear aspects of speaker performance are "minimal" is, in my opinion, a subjective determination. (And my opinion is also subjective.)

That's fine, but I do not consider those nonlinearities to be "minimal" when it can be easily demonstrated that simply changing a single speaker for another one that fits in the same cabinet can have a dramatic effect on the sound. Such an obvious difference can not be "minimal".

My Axe-FX setup is simple: AFII to Alesis stereo power amp (RA-100), to Hartke 4x12 angled cab loaded with Celestion G12M25 reissue Greenbacks, well broken in. I make two assumptions which I believe are reasonable: 1: That the output drive stage of the AFII is linear in its behavior, so that the signal delivered to the power amplifier varies only in total level and harmonic distortion content does not appreciably change over its full range of output. 2: That the power amplifier is also very linear until it reaches its rated power output of 75 WPC into a nominal 8 ohm load, or 100 WPC into a nominal 4 ohm load.

So, working on the belief that these assumptions are accurate, then whether the system is driving the 4x12 at subaudible levels, or at "speakers will die shortly, right after eardrums" levels, there will be no variance across that range, with regard to the harmonic content going to the speakers.

Yet, there really is no doubt that the speakers sound different when being pushed pretty hard. There are audible cone harmonics that only come out when it gets loud. I can hear them clearly even when using earplugs and earmuffs for double hearing protection.

Those cone harmonics are probably still there at lower total volume levels, but they are at a lower level relative to the total output. Their behavior is less linear at higher output levels.

With the amount of DSP power now available, I predict that it's only a matter of someone taking the time to create a system that more accurately models speaker behavior than the current IR model, which reflects speaker behavior very well but only at the one volume level it was measured at.

Since Fractal Audio is unquestionably the leader in DSP amplifier modelling, it stands to reason that the brains behind these wonderful products should hopefully be interested in pushing the modelling of the speaker system to new heights of performance as well. And that will require a more detailed approach to modelling the full-range dynamic performance of the speaker system.

Beat the competition to the punch. Create a multi-dimensional IR capture and playback methodology that captures the speaker performance at a range of output levels which will capture how it responds differently at those various levels.

Eventually, somebody's going to do it. Might as well be Fractal to get there first.
 
Would it really sound that much more "realer" though ?

The idea is great in theory, my realtime dynamic nature and all, but it already sounds like a real speaker to me in 99.9% of cases, be it a mic'd cab in the studio, a mic'd cab going to the FOH system et al., so would there be any real world benefits people could actually hear ?

I had the Nebula (I think it was called) stuff a number of years ago which were supposed to be more dynamic, but they never sounded great. It was kind of a dead end format honestly, don't know if its still around even.

Sure IR's are static, and we always want to push modeling technology further, but at what end cost to the user ? How much would I end up having to pay for all that R&D, and how long would I have to wait to get all the speaker combinations that I personally use modeled ?

If someone comes out with something better, awesome, but do I want to pay $299 extra for it ? Can I hear a difference ? Does it only include a 4x12 and a 2x12 and a 1x10 and I'm going to be waiting 8 months + for a new model to come out ?

Then what ? Someone "cracked it", but if no one buys it, or licenses it, or we find we just can't hear a difference, what becomes of the tech ?

Would be great if Cliff says one day " I had a breakthrough, firmware 7.xx now has dynamic IR's or whatever, free update as always" and we all go yay!, sounds awesome Cliff...

But is that going to be likely ? Is it going to pay off financially for him ? Or... Is he better putting his time into modeling a new amp or something ?

Guess its ultimately not a question of "can", but "will"....
 
It is true that..now no jokes...
an overhung driver: the voice coli can travel outside of the gap...this would introduce non-linear response based on travel
an underhung driver: the voice coil does not travel outside the gap...ergo...more linear response
 
Last edited:
good stuff pulled from the wiki by clif:
"The myopic only see the IR as a capture of the speaker's "unadulterated" response. As I stated before I believe the future is treating IRs as capturing the entire recording chain including mics, preamps, etc. and have pushing in that direction.


So I see this element of IR capture more like profiling and less like modeling...
 
I note that the document I linked earlier defines not less than NINE factors in speaker design that are non-linear. Just pointing that out.

Everything gets cheaper with time as the technology to develop it matures. Today's smartphone, as common a gadget as anything, simply could not exist 20 years ago for a million dollars a copy.

I have yet to get a really exact match between the sound of my "real" amps and cabinets and the best I've been able to get out of my AFII using the models based on them. It's close but there is a difference. It stands to reason that the whole point of a product like an Axe-FX is to get as close as is practical to having the same performance and sound. And it WILL get closer still with every passing generation and update. The IRs can be looked at as a point of incremental improvement.

I don't really expect that the IRs of today will be the industry standard 10 years from now. Do you?
 
we do have several parameters related to speaker behaviour, including motor drive which models speaker compression as far as i'm aware
 
So you're saying that a speaker is a LINEAR device? That is at variance with the opinions put forth by a variety of speaker design engineers. Particularly in the high end range of the speaker market, driver designers spend the majority of their engineering efforts on trying to ensure that the driver's motion remains purely pistonic without harmonic breakup modes coming into the picture at high drive levels.

I do not have access to a laser interferometer but it may be possible for me to come up with a test setup to sweep a given speaker at various output levels, using a very clean source and amplifier, at levels from the threshold of audibility to the 1 dB compression point that is defined by many speaker designers as the full power handling capacity, and capture the spectral output at each. Say at 10 dB increments.

I am sure beyond any doubt that the harmonic content will change as a function of output. It won't be the same spectral plot when normalized to the same level.

Speakers do not exhibit linear behavior for a double handful of reasons.

If you only take an IR at a specific level, then that IR will only be accurate at that level. It cannot be otherwise. Because speakers are non-linear devices.

https://www.klippel.de/fileadmin/_migrated/content_uploads/Loudspeaker_Nonlinearities–Causes_Parameters_Symptoms_01.pdf
I'm aware of that paper. You seem strangely familiar with some relevant technical info for a person who didn't know what an impulse response is. I'm confused on how to address you now :p

I'm going to respond to you by linking you to forum posts made by someone in the top of the field of speaker design. It should give you a lot of insight in layman's terms. After reading them, you can go investigate with academic papers if you'd like.
To state that the nonlinear aspects of speaker performance are "minimal" is, in my opinion, a subjective determination. (And my opinion is also subjective.)

That's fine, but I do not consider those nonlinearities to be "minimal" when it can be easily demonstrated that simply changing a single speaker for another one that fits in the same cabinet can have a dramatic effect on the sound. Such an obvious difference can not be "minimal".

My Axe-FX setup is simple: AFII to Alesis stereo power amp (RA-100), to Hartke 4x12 angled cab loaded with Celestion G12M25 reissue Greenbacks, well broken in. I make two assumptions which I believe are reasonable: 1: That the output drive stage of the AFII is linear in its behavior, so that the signal delivered to the power amplifier varies only in total level and harmonic distortion content does not appreciably change over its full range of output. 2: That the power amplifier is also very linear until it reaches its rated power output of 75 WPC into a nominal 8 ohm load, or 100 WPC into a nominal 4 ohm load.

So, working on the belief that these assumptions are accurate, then whether the system is driving the 4x12 at subaudible levels, or at "speakers will die shortly, right after eardrums" levels, there will be no variance across that range, with regard to the harmonic content going to the speakers.

Yet, there really is no doubt that the speakers sound different when being pushed pretty hard. There are audible cone harmonics that only come out when it gets loud. I can hear them clearly even when using earplugs and earmuffs for double hearing protection.

Those cone harmonics are probably still there at lower total volume levels, but they are at a lower level relative to the total output. Their behavior is less linear at higher output levels.

With the amount of DSP power now available, I predict that it's only a matter of someone taking the time to create a system that more accurately models speaker behavior than the current IR model, which reflects speaker behavior very well but only at the one volume level it was measured at.

Since Fractal Audio is unquestionably the leader in DSP amplifier modelling, it stands to reason that the brains behind these wonderful products should hopefully be interested in pushing the modelling of the speaker system to new heights of performance as well. And that will require a more detailed approach to modelling the full-range dynamic performance of the speaker system.

Beat the competition to the punch. Create a multi-dimensional IR capture and playback methodology that captures the speaker performance at a range of output levels which will capture how it responds differently at those various levels.

Eventually, somebody's going to do it. Might as well be Fractal to get there first.
https://www.thegearpage.net/board/index.php?posts/25799697/
https://www.thegearpage.net/board/index.php?posts/25856337/
https://www.thegearpage.net/board/index.php?posts/9060321/
https://www.thegearpage.net/board/index.php?posts/25501462/
https://www.thegearpage.net/board/index.php?posts/25501357/
https://www.thegearpage.net/board/index.php?posts/21117615/

And one by FAS creator Cliff:
https://forum.fractalaudio.com/thre...g-speaker-cabinets.135191/page-2#post-1599277
I note that the document I linked earlier defines not less than NINE factors in speaker design that are non-linear. Just pointing that out.

Everything gets cheaper with time as the technology to develop it matures. Today's smartphone, as common a gadget as anything, simply could not exist 20 years ago for a million dollars a copy.

I have yet to get a really exact match between the sound of my "real" amps and cabinets and the best I've been able to get out of my AFII using the models based on them. It's close but there is a difference. It stands to reason that the whole point of a product like an Axe-FX is to get as close as is practical to having the same performance and sound. And it WILL get closer still with every passing generation and update. The IRs can be looked at as a point of incremental improvement.

I don't really expect that the IRs of today will be the industry standard 10 years from now. Do you?
Here's Celestion saying they can't hear the difference, and demonstrating it:
https://www.celestionplus.com/news/speaker-impulse-response-hear-difference/

If you're not getting an exact match between real cabs and AFII cabs, it's likely not due to some inherent limit in IRs.
I recommend that you take your own impulse responses of your own cabs with your own setup. Then compare that IR to your own setup. As of now, you're comparing someone else's mic'ed cab to your own mic'ed cab, and of course they won't be the same.
 
Last edited:
As for how to address me, my first name is Chris. I go by that. :) My technical level is somewhere between technician and engineer. I've been an electronic technician all my life. I'm 54 as of this time. I'm also insatiably curious about all things technical so anything that interests me is quickly researched to get an idea of what it is that I'm looking at. Some things merit further study, others don't pique my interest quite so deeply, but at least then I've gained basic familiarity with another technology or concept.

I know what an IR IS, I just wasn't sure of how that data is presented within the IR file. I'd actually listened to some IRs (not Fractal's format) that presented as .wav files and sure enough, you hear a click. I was not ready to say "Oh, that's all that is, just a click." I didn't know if that audible click was actually even supposed to be there or if it was a data artifact. Just as you can play a data CD on many older CD players and hear the data which is definitely not music.

If people have investigated what I'm asking about and concluded that there's no real audible benefit to, let me describe it as a multi-dimensional impulse response (each dimension being an IR captured at a different output level) , then I'm cool with that. I'm just doing thought experiments,
which are the kind that don't require me to buy more equipment. :cool:
 
For practical purposes, think of IRs as a very high resolution equilizer that's used to model the frequency response of a speaker in a cabinet being captured by a specific microphone placed in a specific way.

Yes, IRs are a linear filter - interesting non-linear effects like speaker power compression are actually handled inside the amp block on FAS units.

However, unlike other EQs you use, IRs capture the frequency response of these setups VERY precisely. For the Ultra Res IRs, you can think of it as something very much like a 8000 band graphic EQ. The extremely high resolution allows you to capture very small details of the speaker/cab/mic system.
 
Back
Top Bottom