Jay's 1024 cabs?

[Amphibian]

Is one suppose to use a mic sim with those? If one would use the mic sims with Jay's cabs, how does that affect the quality of a 1024 point IR? Isn't the mic's still at 512 point? Someone care to explain how this works?

 

[Jay Mitchell]

Is one suppose to use a mic sim with those?

You certainly don't have to. I've never used the mic sims myself, because I've been interested in recreating the natural sound of the cabinet with as little coloration as possible.

If one would use the mic sims with Jay's cabs, how does that affect the quality of a 1024 point IR? Isn't the mic's still at 512 point? Someone care to explain how this works?

You would use the mic sim on the first half of the IR only. Because the mic IR is only ~11ms long, it can't affect any portion of the speaker IR that occurs more than that amount of time after the initiation of a sound. The practical effect of this is that the speaker sound is modeled with higher resolution than the mic sound. Given the behavior of most microphones (smoother, with less detail than loudspeaker behavior), this won't cause a problem.

 

[Amphibian]

Let me see if I get this right:

1. The amp block is split, one side of it is sent to a shunt and one sent to a delay.

2. The shunt feeds the left cab, since it is hard panned to the left. The delay, on the other hand, delays the signal 10 ms and sends it to the right cab, since it's panned to the right.

3. You put a mic simulation on the Left cab, which affects the first 512 points of the IR. You then leave the Right cab intact, without a mic sim?

4. The cabs are panned to center to give the illusion of a single cab with higher resultion.

Question: Wouldn't the obvious thing be to put the mic sim on both parts of the IR? Isn't it like wearing fitover sunglasses. You would need both sides for it to work...!?!?! What am I missing? :geek: :geek:

 

[Jay Mitchell]

Question: Wouldn't the obvious thing be to put the mic sim on both parts of the IR?

After some additional thought, I'm pretty sure that's correct. The effect of recombining the two IRs with a 10ms delay on the second half is to create one long IR, all of which should be convolved with the 512-point IR of a mic sim. I'd never considered any of this before now, because it doesn't make any difference to me. Because there is so little total energy in the second half of the 1024 IR, I don't believe it will make much difference.

 

[javajunkie]

Question: Wouldn't the obvious thing be to put the mic sim on both parts of the IR?

After some additional thought, I'm pretty sure that's correct. The effect of recombining the two IRs with a 10ms delay on the second half is to create one long IR, all of which should be convolved with the 512-point IR of a mic sim. I'd never considered any of this before now, because it doesn't make any difference to me. Because there is so little total energy in the second half of the 1024 IR, I don't believe it will make much difference.

It doesn't. I like the coloration on the r121, so I use it on most of the custom cabs I've use. I tried it with it on/off on the 2 cab w/ 1024 IR. I could not really hear a difference.

 

[scarr]

Question: Wouldn't the obvious thing be to put the mic sim on both parts of the IR?

After some additional thought, I'm pretty sure that's correct. The effect of recombining the two IRs with a 10ms delay on the second half is to create one long IR, all of which should be convolved with the 512-point IR of a mic sim. I'd never considered any of this before now, because it doesn't make any difference to me. Because there is so little total energy in the second half of the 1024 IR, I don't believe it will make much difference.

It doesn't. I like the coloration on the r121, so I use it on most of the custom cabs I've use. I tried it with it on/off on the 2 cab w/ 1024 IR. I could not really hear a difference.

That's interesting. So having the extra 512 samples there makes a notable difference in the sound, but their signature being reshaped by a mic IR doesn't really matter? I'm trying to figure out the logic behind that, but it's too soon after waking up to work yet.

 

[Jay Mitchell]

That's interesting. So having the extra 512 samples there makes a notable difference in the sound,

As I have pointed out many times, the difference is notable but subtle.

but their signature being reshaped by a mic IR doesn't really matter?

A subtle difference in a subtle difference can easily fall below the threshold of perceptibility. I suspect that, given the ability to do a direct A/B very quickly, the difference would be audible, at least with some speaker IRs and some playback systems. Whether or not it is audible enough to make a practical difference is an open question. It's never concerned me one way or the other, because I have never used mic sims.

I'm trying to figure out the logic behind that, but it's too soon after waking up to work yet.

There's really nothing to figure out. Either you hear a worthwhile difference, or you don't. The same applies to 512- vs. 1024-point IRs. Even though I have no use for them, I have chosen to make 512-point versions of all my custom IRs available, specifically for folks who either could hear no difference or who feel that the difference is not great enough to justify the resources consumed by the setup required to use 1024-point ones. I definitely hear a difference with my presets and my FRFR system, and that difference is, for my purposes, sufficiently worthwhile to easily justify the dedication of one of the available delay blocks to that use. That doesn't mean that I think everyone - or even anyone - else will hear a similarly worthwhile difference, however.

 

[AlbertA]

Let me see if I get this right:

1. The amp block is split, one side of it is sent to a shunt and one sent to a delay.

2. The shunt feeds the left cab, since it is hard panned to the left. The delay, on the other hand, delays the signal 10 ms and sends it to the right cab, since it's panned to the right.

3. You put a mic simulation on the Left cab, which affects the first 512 points of the IR. You then leave the Right cab intact, without a mic sim?

4. The cabs are panned to center to give the illusion of a single cab with higher resultion.

Question: Wouldn't the obvious thing be to put the mic sim on both parts of the IR? Isn't it like wearing fitover sunglasses. You would need both sides for it to work...!?!?! What am I missing? :geek: :geek:

Techincally you shouldn't apply the mic to the other half, because the result is not a 1024 point IR convolved with a mic Ir, it's something else. Like Jay said in the beginnin just use it in the first 512 samples. You can't extend the mic IR to samples it doesn't have. I.e. the goal is to convolve a 512 point IR with a 1024 point IR.

Sonically tough, it's all subjective so do whatever sounds best to you.

 

[Mike Snider]

And...as Jay mentioned, the signifigant energy is found in the first IR. The 1024 IR is split into two halves, but are not equal "sounding". If you listen to each side of the IR you will notice one is perceptivly louder than the other- but when used together, ceate a more finite sample.
To my ears the difference between the 512 and 1024 are signifigant and 1024's are now all i use.

 

[javajunkie]

Question: Wouldn't the obvious thing be to put the mic sim on both parts of the IR?

After some additional thought, I'm pretty sure that's correct. The effect of recombining the two IRs with a 10ms delay on the second half is to create one long IR, all of which should be convolved with the 512-point IR of a mic sim. I'd never considered any of this before now, because it doesn't make any difference to me. Because there is so little total energy in the second half of the 1024 IR, I don't believe it will make much difference.

It doesn't. I like the coloration on the r121, so I use it on most of the custom cabs I've use. I tried it with it on/off on the 2 cab w/ 1024 IR. I could not really hear a difference.

That's interesting. So having the extra 512 samples there makes a notable difference in the sound, but their signature being reshaped by a mic IR doesn't really matter? I'm trying to figure out the logic behind that, but it's too soon after waking up to work yet.

I think you just hear most of the change in sound coming from the first part of the IR that you lose much of the difference of adding a mic on the second half.

 

[scarr]

That's interesting. So having the extra 512 samples there makes a notable difference in the sound,

As I have pointed out many times, the difference is notable but subtle.

but their signature being reshaped by a mic IR doesn't really matter?

A subtle difference in a subtle difference can easily fall below the threshold of perceptibility. I suspect that, given the ability to do a direct A/B very quickly, the difference would be audible, at least with some speaker IRs and some playback systems. Whether or not it is audible enough to make a practical difference is an open question. It's never concerned me one way or the other, because I have never used mic sims.

[quote:3ol6bi9p]I'm trying to figure out the logic behind that, but it's too soon after waking up to work yet.

There's really nothing to figure out. Either you hear a worthwhile difference, or you don't. The same applies to 512- vs. 1024-point IRs. Even though I have no use for them, I have chosen to make 512-point versions of all my custom IRs available, specifically for folks who either could hear no difference or who feel that the difference is not great enough to justify the resources consumed by the setup required to use 1024-point ones. I definitely hear a difference with my presets and my FRFR system, and that difference is, for my purposes, sufficiently worthwhile to easily justify the dedication of one of the available delay blocks to that use. That doesn't mean that I think everyone - or even anyone - else will hear a similarly worthwhile difference, however.[/quote:3ol6bi9p]
The extension to my line of thinking is that if the sonic signature of the second half (which would be varied with/without the mic) isn't as important as just plain having it, how much difference would it make if you just reused one "second half" response across the different cabs? Wouldn't that just be a "subtle difference in a subtle difference" too? Reusing one tail could give you four extra User Cab slots. Now I think about it, this concept would extend to the existing/factory cabs as well. You could "make them 1024-point" by pinning the same tail to them, couldn't you? Maybe a different one for close/distant micing, but that would give access to more "1024"-point IR's.

This is 100% theoretical, but it might be worth looking into to expand the options on a single AFx.

 

[Jay Mitchell]

if the sonic signature of the second half (which would be varied with/without the mic) isn't as important as just plain having it, how much difference would it make if you just reused one "second half" response across the different cabs?

That has close to a zero probability of working. Keep in mind that the contribution to the total sound of the latter portion of a continuous IR may be quite subtle, but that, if you concatenate the first half of one with the second half of another, the combined IR will no longer be continuous. The discontinuity at the splice can easily create spectral artifacts that are not particularly subtle.

Reusing one tail could give you four extra User Cab slots.

See above. There's no way to make the splice seamless, and that's a dealbreaker. If using 1024s is inconvenient, my suggestion is to just use 512s and fuhgeddaboutit.

 

[scarr]

if the sonic signature of the second half (which would be varied with/without the mic) isn't as important as just plain having it, how much difference would it make if you just reused one "second half" response across the different cabs?

That has close to a zero probability of working. Keep in mind that the contribution to the total sound of the latter portion of a continuous IR may be quite subtle, but that, if you concatenate the first half of one with the second half of another, the combined IR will no longer be continuous. The discontinuity at the splice can easily create spectral artifacts that are not particularly subtle.

By that point, aren't they mostly the same basic shape? If you matched the level appropriately, I don't think they would be terribly discontinuous. How many dB are the responses normally down by that point?

To draw a crude analogy, If I took the tail of a snare drum after it dropped 40+ dB (just guessing) and removed it, it would change the sound of the snare. It would be somewhat subtle (especially in a mix), but notable. Extending that tail would be audible to some. I know that I could stick the tail of a different snare on there though, and while it wouldn't sound the same, it would sound closer to the unabbreviated snare than simply leaving the tail out. As most snare hits have the same basic shape (exponential decrease), some simple steps (like looking for a zero-crossing) could eliminate any discontinuity. At that low a level, I don't know if you could hear the difference between a rosewood snare and a birch. An IR is a somewhat different beast in certain ways, but in many ways very much the same. It isn't optimal, but it could be a much closer approximation.

If using 1024s is inconvenient, my suggestion is to just use 512s and fuhgeddaboutit.

That and being happy with the close cabs is why I haven't messed with User Cabs myself. I'm trying to generate some ideas for the people who do like/enjoy/use them and might want to get more out of them. Or just get more of them in the AFx.

 

[Jay Mitchell]

By that point, aren't they mostly the same basic shape?

There is no definition of "same basic shape" that is mathematically rigorous. Piecewise continuity for, say, the first three time derivatives (including the zeroth) would be a good bet for splicing a second half without adding artifacts. That means that one size won't fit all, which is what I've been trying to get across.

If you matched the level appropriately, I don't think they would be terribly discontinuous.

You're making far too many assumptions here. Having to match levels "appropriately" (meaning precisely) means one size won't fit all.

To draw a crude analogy, If I took the tail of a snare drum after it dropped 40+ dB (just guessing) and removed it, it would change the sound of the snare.

That is an inappropriate analogy. The technique you're suggesting works - after a fashion - with DSP-based reverb algorithms, where it is common practice to use IRs for the early reflections but prohibitive in terms of the required resources to generate entirely from stored IRs. The tail is created statistically as a means to save resources. This inevitably involves sonic compromise, and that compromise is audible if you A/B the difference.

In the case of loudspeaker IRs, the additional length of the IR is not audible as a tail. Insted, it simply creates more detail in the frequency response of the speaker. You can easily see the difference with an IR in .wav format using AlbertA's utility.

I know that I could stick the tail of a different snare on there though, and while it wouldn't sound the same, it would sound closer to the unabbreviated snare than simply leaving the tail out.

We're not talking about something that is audible as a decay - a sub-20ms IR has no audible "tail."

trying to generate some ideas for the people who do like/enjoy/use them and might want to get more out of them.

You can't get there from here. You won't get the benefit of a more authentic, detailed response with a generic tail, but you will get some added spurious response anomalies.

Edit: I just took a look at two different IR in some detail at the 10 ms point. You can't even count on their values having the same sign (both positive or negative) at this point, so a seamless splice is completely out of the question.

 

[scarr]

I'm not arguing that this would be a precise alternative. I know it isn't. Doubling the number of points of your pure IR is going to be the optimum solution. I'm thinking about in between solutions.

As you point out, other technologies use a similar concept with reverbs. A reverb is just an impulse response, as is a snare drum hit. I realize that a snare's IR is something we appreciate directly, whereas reverb/cab IR's are interpreted indirectly, but they're all still IR's.

I'm not wrapping my head around all the math that's actually going on with this (I'm trying to remember all my transfer function conversions/relations), but if changing the second half of the response curve with a mic's "EQ" doesn't provide a noticeable difference in the benefits of using a 1024-point IR, unlike not having the last 512 samples at all, then having something similar to the actual response is better than nothing.

 

[Jay Mitchell]

I'm not arguing that this would be a precise alternative. I know it isn't.

What you're missing is that it isn't even a viable alternative. It won't get you closer to the actual sound of the cab you're trying to model. The correctly-done 512 will be a hands-down better alternative.

As you point out, other technologies use a similar concept with reverbs.

Not applicable here, however. Extrapolating a reverberant tail using a mathematical algorithm isn't an available option in this case. It definitely would not work to splice a generic reverberant tail onto an early IR, and it won't work for a cab IR either.

I realize that a snare's IR is something we appreciate directly, whereas reverb/cab IR's are interpreted indirectly, but they're all still IR's.

There's more to it than that. There is still ringing going on with speaker IRs at the 10ms point, and this ringing swings in both positive and negative directions. Tacking on the second half of an unrelated 1024-point IR will almost always create a significant discontinuity, and it will add audible, but entirely nonphysical, features to the response of the speaker.

if changing the second half of the response curve with a mic's "EQ" doesn't provide a noticeable difference in the benefits of using a 1024-point IR, unlike not having the last 512 samples at all, then having something similar to the actual response is better than nothing.

Not if "similar to the actual response" creates audible response features that the actual speaker doesn't have. You're much better off with a 512.

This is going to a whole lot of effort just to add four available user IRs. If you want more than five IRs (how many opportunities do you get to select among five cabs during a gig?), then just use the 512s.

 

[scarr]

I'm not arguing that this would be a precise alternative. I know it isn't.

What you're missing is that it isn't even a viable alternative. It won't get you closer to the actual sound of the cab you're trying to model. The correctly-done 512 will be a hands-down better alternative.

As you point out, other technologies use a similar concept with reverbs.

Not applicable here, however. Extrapolating a reverberant tail using a mathematical algorithm isn't an available option in this case. It definitely would not work to splice a generic reverberant tail onto an early IR, and it won't work for a cab IR either.

[quote:7p938whd]I realize that a snare's IR is something we appreciate directly, whereas reverb/cab IR's are interpreted indirectly, but they're all still IR's.

There's more to it than that. There is still ringing going on with speaker IRs at the 10ms point, and this ringing swings in both positive and negative directions. Tacking on the second half of an unrelated 1024-point IR will almost always create a significant discontinuity, and it will add audible, but entirely nonphysical, features to the response of the speaker.

if changing the second half of the response curve with a mic's "EQ" doesn't provide a noticeable difference in the benefits of using a 1024-point IR, unlike not having the last 512 samples at all, then having something similar to the actual response is better than nothing.

Not if "similar to the actual response" creates audible response features that the actual speaker doesn't have. You're much better off with a 512.[/quote:7p938whd]
A generic reverb tail would be closer to a genuine reverb response than a gated reverb, wouldn't it? Is that a fair analogy, or if not, why?

Part of me understands that the IR is a time domain representation of a frequency domain response curve. Another part of me knows it has to make sense in the time domain as well, and I think that's what I'm getting stuck on, for better or worse. Why is cutting the data off after 512 points better than approximating the next 512?

This is going to a whole lot of effort just to add four available user IRs. If you want more than five IRs (how many opportunities do you get to select among five cabs during a gig?), then just use the 512s.

My thinking was that it could be extended to the existing cabs as well, so it isn't purely about getting four extra user cabs. I can run through more cabs than that in a gig easily, if I have the opportunity. Heck, I've come close to running through that in a single patch before.


I'm not sure, but part of all this might be a difference in what you see as the goal versus what I see as the goal. If you're trying to get the Cab block to sound as close as it can to a particular configuration, and different from each alternative, then the "generic tail" wouldn't help that. Going back to the reverb analogy (which I believe to be truly analogous here, but please point out where it breaks down if I'm wrong), if you want to represent the sound of a particular room, a fake/generic tail isn't going to add to the characteristics of the room. It won't help you gauge exactly how much the wall are damping the sound or what kind of tiles are on the floor. If your IR is gated at a certain point though, you've lost a lot of that finer detail anyway. My point is that adding the tail end of a reverb sampled in Carnegie Hall to the initial one second IR of Music Hall in Cincinnati will give a better sense of dimension and decay in a large space than a reverb that cuts off after the first second. It isn't more characteristic of Music Hall, but it is more representative of the response of a hall. For whatever (possibly misguided) reason, I see a relationship between that and the cab IR's.

To (over)simplify it for other people who might still be reading this (and part of why I'm continuing this here is to hopefully edify others about some of this too), I'm thinking it is like having the picture of the front half of a car. Obviously having a picture of the back half of that car will best represent the car. What do you do if you only have pictures of the front halves of cars and you're trying to illustrate what a car is to someone? Putting a picture of the back of a Carolla onto a picture of the front of a Camry will give a better indication of what a car (not necessarily the Camry) looks like than leaving the front half dangling in space. Now sticking the back end of an F150 on a Camry might not be as accurate of a picture of what a Camry is, but it would still better illustrate how the vehicle drives. Moreover, the back of an F150 could be stuck on pictures of the front of a mess of different trucks and you might not be able to tell it apart from the "real" truck. The extend of how generic you make the "IR extension" will effect how accurate it is, but if you stick to 4x12's for a particular "IR extension", you might be able to get something that at that point is minimally different from any particular 4x12's natural IR.

 

[Jay Mitchell]

A generic reverb tail would be closer to a genuine reverb response than a gated reverb, wouldn't it?

That's not the choice that has to be made. There's always some attempt to simulate decay, whether it's a complete IR - which is extremely demanding of cpu resources to convolve - or the early portion of a reverberant IR with a statistically-derived tail.

Is that a fair analogy, or if not, why?

No. IR's aren't "gated," which is another way of saying that the window function that is used is not rectangular.

Why is cutting the data off after 512 points better than approximating the next 512?

Neither is what occurs or would occur.

First, the 512-point IR is not "cut off" (i.e., it doesn't have a rectangular data window), its data is manipulated mathematically so as to smooth out the the transition to the required zero value at its end. If you just "cut it off" at the end, you'd get audible response anomalies and features that have nothing to do with what the real speaker does. The data window has an effect on the response, but the effect is much more subtle than the one created by a rectangular window. A longer, but authentic, IR substantially reduces the effect of this window. The audible improvement in a 1024-point IR over a 512-point one is largely due to this one effect. By the end of a 20-ms time period, the response of almost any speaker has decayed naturally into the noise, and, as a result, the window function has a much smaller effect on the sound.

Second, what you propose would do a singularly poor job of "aproximating the next 512" points. The worst, and most important, effect is that it would create a discontinuity at the splice point. See above. This would create response anomalies that are at least as bad as - and potentially worse than - those that would be created by a rectangular-windowed 512-point IR.

You're missing a key element of a synthesized reverberant tail: the DSP system creates the tail and matches it to the early portion. It is therefore continuous and free of significant window-generated anomalies.

Because the existing factory IRs are already windowed to 512 points, they all decay smoothly to zero at the end of the ~11-ms data record. Because of this, adding a "second half" - which would invariably begin at a nonzero value - to one of them is guaranteed to produce spectral artifacts. With the first half of a 1024-point IR, the endpoint value is not windowed to zero (because it is to be used with its second half). Used with the correct second half, the values are continuous across the splice. Used with a cpu-generated second half, the same would be true, just as with a synthesized reverb tail. This is not an option in the Axe-Fx. Adding an arbitrary second half will not produce the same result.

Edit: I'm trying to get across the fact that discontinuous IR data that is nonrepresentative of the actual device IR will create distracting high-frequency response anomalies that don't sound like a speaker, thereby defeating the purpose of the generic added section.

 

[scarr]

While I didn't mean strictly "gated" (poor choice of words on my part), your point about how the IR's are already windowed is valid, which makes my whole idea of adding anything to them pointless. At some point I was going to ask if the extension was purely a way of getting the windowing out of the range of the IR's data, but that got lost somewhere along the way in trying to come up with analogies for my line of thinking.

I get your point that the discontinuity is the big concern here. In a theoretical world, if that were somehow eliminated, or the IR's engineered/manipulated in some way to have them line up properly, would a generic extension IR help like the DSP generated reverb tail, or would it still just be random noise thrown into the signal?

 

[Jay Mitchell]

I get your point that the discontinuity is the big concern here. In a theoretical world, if that were somehow eliminated,

The first half of a 1024-point IR ends on an unknown value. The second half begins at a nearby value such that the combined function is continuous. Eliminating any possible discontinuity would require forcing the first half to end at a known value. The only defensible choice for that value is zero, and the only technique for forcing the function to zero at the end of the IR while minimizing undesirable artifacts is data windowing. As you've acknowledged, concatenating a spurious second half with a first half that has been windowed is not a productive exercise.

Implementing a statistical tail extension would potentially require more resources than just extending the IR length to 1024 points, and we have a pretty good idea that that isn't feasible.