Why no High-Res IR support?

[DLC86]

I remember that on the ax8, when you loaded an IR without the ultrares processing, it was saved as what was called a high-res IR (aka 2048 samples) and the cpu usage of the cab block didn't change when changing between normal res (1024 samples) and high/ultra-res (2048 samples) since it was running on the accelerator, so we could have double the length without taxing the cpu.

The axe fx III works pretty much the same way, loading an IR in standard res but selecting maximum IR lenght in the cab block gives you a 2048 samples IR (even though, in this case the CPU usage is higher since the axe doesn't have a dedicated accelerator and the convolution runs on the main cpu core).

On the fm9 otoh (and I suppose on the fm3 too), by loading a standard res IR you can only get a 1024 samples IR, no matter if the cab block is set to standard res or ultrares.

Could we have the same implementation as the axe 3 and ax8 please? Unless it's a hardware limitation of course (which seems unlikely to me since fm9/3 chips are basically new generation of the chips found in the ax8.. but who knows)

 

[moerker]

Two questions:

• in which scenario is this useful (I cannot wrap my head around this), and
• what are you gaining from 1024 samples with zeros (since they need to be padded)?

I have trouble understanding the issue here, tbh.
You have higher resolution IRs, which are stored correctly; and shorter IRs with less resolution, stored correctly.

 

[DLC86]

in which scenario is this useful (I cannot wrap my head around this), and

Let's say you're running out of CPU, you can select the high-res version of an IR (assuming the accelerator works the same as in the ax8) in place of the ultrares one and save CPU without compromising the low end response of the IR too much, which happens with standard res (1024 samples means you'll lose infromation below ~46Hz, with 2048 that limit is pushed to ~23Hz).
VERY useful with bass cabs or room correction IRs, but even with guitar cabs.

what are you gaining from 1024 samples with zeros (since they need to be padded)?

Why are you assuming that IRs saved as standard res are truncated at 1024 samples?
Pretty sure each slot has enough space to store a 2048 IR, since an ultrares IR size is roughly the same (and that's how it works in the axe fx and ax8 btw)

 

[yek]

AFAIK, HiRes has been discontinued. It uses even more CPU than Ultra-Res (UR is very efficient).

 

[DLC86]

AFAIK, HiRes has been discontinued. It uses even more CPU than Ultra-Res (UR is very efficient).

Just the word "HiRes" has been discontinued, in practice it's still there in the axe fx, just save an IR as standard res and set the cab block to max IR lenght .

And it uses more cpu than ultrares only on the axe fx which doesn't have a dedicated accelerator, on the ax8 it uses the same cpu as standard res

 

[yek]

Just the word "HiRes" has been discontinued, in practice it's still there in the axe fx, just save an IR as standard res and set the cab block to max IR lenght .

That's legacy support.

 

[moerker]

Let's say you're running out of CPU, you can select the high-res version of an IR (assuming the accelerator works the same as in the ax8) in place of the ultrares one and save CPU without compromising the low end response of the IR too much, which happens with standard res (1024 samples means you'll lose infromation below ~46Hz, with 2048 that limit is pushed to ~23Hz).
VERY useful with bass cabs or room correction IRs, but even with guitar cabs.

Just for clarification: standard-res IRs have a lower resolution, but, thus also "less information" on all frequencies, since all bands are merged together. It will however not cut away bass response. But I think you mean it will get more flubby, since the resolution is less.

Why are you assuming that IRs saved as standard res are truncated at 1024 samples?
Pretty sure each slot has enough space to store a 2048 IR, since an ultrares IR size is roughly the same (and that's how it works in the axe fx and ax8 btw)

There is no assumption here: storing a signal with 1024 in 2048 samples needs zero-padding. Storing a standard-res with 2048 samples won't make any difference in sound, that's what I meant with the comment (and what I read out of the first sentence). If there are just 1024 taps of information, blowing it up to 2048 won't change anything.

It all just makes sense if you are using ultra-res IRs and then make standard IRs out of it, not the way around (as I understand you).

 

[DLC86]

Just for clarification: standard-res IRs have a lower resolution, but, thus also "less information" on all frequencies, since all bands are merged together. It will however not cut away bass response. But I think you mean it will get more flubby, since the resolution is less.

Doesn't work that way. 1024 samples at 48KHz means the IR is 21.3 ms long.
A full 20Hz wave cycle is 50 ms long, so that IR can't even contain half a cycle of that 20 Hz wave. That makes you lose information, you can clearly see that by analyzing the frequency response of the IR, the shorter you trim it the more the low end response is altered (and if it's not a minimum phase IR the alterations reach even higher frequencies).
I'll post some graphs later to make it clearer.

There is no assumption here: storing a signal with 1024 in 2048 samples needs zero-padding. Storing a standard-res with 2048 samples won't make any difference in sound, that's what I meant with the comment (and what I read out of the first sentence). If there are just 1024 taps of information, blowing it up to 2048 won't change anything.

It all just makes sense if you are using ultra-res IRs and then make standard IRs out of it, not the way around (as I understand you).

Not sure what you mean by "padding", maybe "filling the additional 1024 samples with silence"? Yes, that's what would happen if you store a 1024 samples IR.

Anyway, who said I want to store a 1024 samples IR and use it as 2048 samples? That makes no sense obviously. So that was your wrong assumption, the point is storing 2048 or longer IRs and using them as 2048 samples IR.

 

[DLC86]

That's legacy support.

Call it whatever you want

 

[yeky83]

Doesn't work that way. 1024 samples at 48KHz means the IR is 21.3 ms long.
A full 20Hz wave cycle is 50 ms long, so that IR can't even contain half a cycle of that 20 Hz wave. That makes you lose information, you can clearly see that by analyzing the frequency response of the IR, the shorter you trim it the more the low end response is altered (and if it's not a minimum phase IR the alterations reach even higher frequencies).
I'll post some graphs later to make it clearer.

I dunno why you're disagreeing with @moerker here? Seems like you're saying the same thing... not sure where the misunderstanding is to cause the disagreement.

 

[DLC86]

I dunno why you're disagreeing with @moerker here? Seems like you're saying the same thing... not sure where the misunderstanding is to cause the disagreement.

Basically in his statement that a short IR doesn't alter bass response, it does indeed

 

[yeky83]

Basically in his statement that a short IR doesn't alter bass response, it does indeed

I think he knows and was differentiating between altered bass response with shorter IRs vs. the common misconception of a bass-cut with shorter IRs.

 

[DLC86]

I think he knows and was differentiating between altered bass response with shorter IRs vs. the common misconception of a bass-cut with shorter IRs.

If that's the case pardon, it wasn't clear to me from his posts.