Impulse response loader Rack Extension

This forum is for discussing Rack Extensions. Devs are all welcome to show off their goods.
Post Reply
User avatar
Kategra
Posts: 327
Joined: 18 Jan 2015

28 Feb 2018

I don't like RV7000 for IR, done some tests and it seems that it has low resolution in applying a frequency response.
Is there any good RE (with no, or very low latency) that can load IR?

User avatar
Loque
Moderator
Posts: 11170
Joined: 28 Dec 2015

28 Feb 2018

No. I am wondering, that there is not one yet (high quality ofc).
Reason12, Win10

househoppin09
Posts: 536
Joined: 03 Aug 2016

28 Feb 2018

Kategra wrote:
28 Feb 2018
I don't like RV7000 for IR, done some tests and it seems that it has low resolution in applying a frequency response.
Is there any good RE (with no, or very low latency) that can load IR?
Can you elaborate on the tests you did and the problems you found? I've never noticed any obvious problems using IRs with the RV7000 and would like to know what to look out for.

User avatar
Kategra
Posts: 327
Joined: 18 Jan 2015

01 Mar 2018

househoppin09 wrote:
28 Feb 2018
Kategra wrote:
28 Feb 2018
I don't like RV7000 for IR, done some tests and it seems that it has low resolution in applying a frequency response.
Is there any good RE (with no, or very low latency) that can load IR?
Can you elaborate on the tests you did and the problems you found? I've never noticed any obvious problems using IRs with the RV7000 and would like to know what to look out for.
You can read it here: viewtopic.php?f=4&t=7498356

Basically, frequency response above 3000Hz is off by a few dBs.


RV7000 bad response above 3000Hz, even 20Hz to 50Hz looks differently than the original.
Image

Original response
Image

househoppin09
Posts: 536
Joined: 03 Aug 2016

01 Mar 2018

Ahh, I missed that thread the first time around! Very informative, thanks.

User avatar
QVprod
Moderator
Posts: 3488
Joined: 15 Jan 2015
Contact:

01 Mar 2018

That thread doesn’t seem to signify whether the problem is RV7000 or an issue with convolution in general. Ostermilk’s comnents lean toward the latter. There are no IR loading REs probably because of RV7000 having the capability already. Have you tried a convolution VST to see if the results are ‘better than RV7000? I personally wouldn’t expect an IR to be a perfect null with the EQ curve it’s based off.

avasopht
Competition Winner
Posts: 3931
Joined: 16 Jan 2015

02 Mar 2018

QVprod wrote:
01 Mar 2018
That thread doesn’t seem to signify whether the problem is RV7000 or an issue with convolution in general. Ostermilk’s comnents lean toward the latter. There are no IR loading REs probably because of RV7000 having the capability already. Have you tried a convolution VST to see if the results are ‘better than RV7000? I personally wouldn’t expect an IR to be a perfect null with the EQ curve it’s based off.
No, that's down to what the RV7000 is doing.

The process of convolution applied to an impulse will give you back the convolution exactly.

But if you've got a 10 second impulse you don't want to be doing raw convolution, so you'll do two things:
1. Convert the convolution to a fourier transform (DFT)
2. Process small windows with incoming audio with the DFT (using FFT), and then combining them.

Applying DFT will require '*windowing,' which will affect the spectrum.

*: windowing here is a function applied to a window (because, well, .. it's technically a verb so it's not the same word :silent: ).

And even using FFT can be pricey, because 10 seconds of audio is the equivalent of trying to run a zero-phase filter at 441 khz (that's 441khz, not 44.1khz).

So some convolution devices will perform multirate convolution. Multirate convolution allows you to run large impulses at significantly lower costs. The low frequency components can be processed at lower sample rates, and then combined with higher rate convolutions (which can use smaller windows). If your convolution isn't eating your CPU, chances are that it's doing multirate convolution.

The most important question is really whether there is a discernable difference, because we're talking about -69dB at +20khz, so not only are you not going to hear it, you're also not going to hear it ;)

scratchnsnifff
Posts: 1423
Joined: 21 Sep 2016

02 Mar 2018

I have noticed using the IRS from trash2 and throwing them into the Rv7000, they sound way lower in volume. But maybe it’s because of the distortion
Mayor of plucktown :evil:

User avatar
QVprod
Moderator
Posts: 3488
Joined: 15 Jan 2015
Contact:

02 Mar 2018

avasopht wrote:
02 Mar 2018
QVprod wrote:
01 Mar 2018
That thread doesn’t seem to signify whether the problem is RV7000 or an issue with convolution in general. Ostermilk’s comnents lean toward the latter. There are no IR loading REs probably because of RV7000 having the capability already. Have you tried a convolution VST to see if the results are ‘better than RV7000? I personally wouldn’t expect an IR to be a perfect null with the EQ curve it’s based off.
No, that's down to what the RV7000 is doing.

The process of convolution applied to an impulse will give you back the convolution exactly.

But if you've got a 10 second impulse you don't want to be doing raw convolution, so you'll do two things:
1. Convert the convolution to a fourier transform (DFT)
2. Process small windows with incoming audio with the DFT (using FFT), and then combining them.

Applying DFT will require '*windowing,' which will affect the spectrum.

*: windowing here is a function applied to a window (because, well, .. it's technically a verb so it's not the same word :silent: ).

And even using FFT can be pricey, because 10 seconds of audio is the equivalent of trying to run a zero-phase filter at 441 khz (that's 441khz, not 44.1khz).

So some convolution devices will perform multirate convolution. Multirate convolution allows you to run large impulses at significantly lower costs. The low frequency components can be processed at lower sample rates, and then combined with higher rate convolutions (which can use smaller windows). If your convolution isn't eating your CPU, chances are that it's doing multirate convolution.

The most important question is really whether there is a discernable difference, because we're talking about -69dB at +20khz, so not only are you not going to hear it, you're also not going to hear it ;)
That explanation still leaves things open ended. I understand what you're saying, but are better results achieved with a different convolution device? Otherwise I would think it's convolution in general following the processes you outlined. If we look at impulse response as a picture, afaik pictures often have some amount of distortion in the image which causes things like curved lines that would otherwise be straight. Sometimes that distortion is noticeable and sometimes not which can often be the fault of the lens used. Changing lenses can solve that problem or at least minimize distortion. Judging by the graphs above I can agree that the difference between the two may not actually be audible.

User avatar
selig
RE Developer
Posts: 11685
Joined: 15 Jan 2015
Location: The NorthWoods, CT, USA

02 Mar 2018

QVprod wrote:
avasopht wrote:
02 Mar 2018
No, that's down to what the RV7000 is doing.

The process of convolution applied to an impulse will give you back the convolution exactly.

But if you've got a 10 second impulse you don't want to be doing raw convolution, so you'll do two things:
1. Convert the convolution to a fourier transform (DFT)
2. Process small windows with incoming audio with the DFT (using FFT), and then combining them.

Applying DFT will require '*windowing,' which will affect the spectrum.

*: windowing here is a function applied to a window (because, well, .. it's technically a verb so it's not the same word :silent: ).

And even using FFT can be pricey, because 10 seconds of audio is the equivalent of trying to run a zero-phase filter at 441 khz (that's 441khz, not 44.1khz).

So some convolution devices will perform multirate convolution. Multirate convolution allows you to run large impulses at significantly lower costs. The low frequency components can be processed at lower sample rates, and then combined with higher rate convolutions (which can use smaller windows). If your convolution isn't eating your CPU, chances are that it's doing multirate convolution.

The most important question is really whether there is a discernable difference, because we're talking about -69dB at +20khz, so not only are you not going to hear it, you're also not going to hear it ;)
That explanation still leaves things open ended. I understand what you're saying, but are better results achieved with a different convolution device? Otherwise I would think it's convolution in general following the processes you outlined. If we look at impulse response as a picture, afaik pictures often have some amount of distortion in the image which causes things like curved lines that would otherwise be straight. Sometimes that distortion is noticeable and sometimes not which can often be the fault of the lens used. Changing lenses can solve that problem or at least minimize distortion. Judging by the graphs above I can agree that the difference between the two may not actually be audible.
I would tend to agree - you need to show three responses to determine if the problem is with the RV7000 or not: the original, the RV7000, and a convolution done “right”.


Sent from some crappy device using Tapatalk
Selig Audio, LLC

User avatar
Marco Raaphorst
Posts: 2504
Joined: 22 Jan 2015
Location: The Hague, The Netherlands
Contact:

02 Mar 2018

Kategra wrote:
28 Feb 2018
I don't like RV7000 for IR, done some tests and it seems that it has low resolution in applying a frequency response.
Is there any good RE (with no, or very low latency) that can load IR?
This is the best for frequency response IRs: http://www.kvraudio.com/product/nadir-by-ignite-amps

It has length settings which might solve the issue you're running into. Must say that I don't have any issues using custom made amp IRs with RV7000. Those are frequency response files of course.

User avatar
Kategra
Posts: 327
Joined: 18 Jan 2015

03 Mar 2018

I have made new tests.

Methodology:
1. Created an one sample pulse in Audacity, saved in 1 second long file stereo wav 24 bit 48000hz. -> file name = "Pulse"
2. Recorded the pulse file trough Sonarworks AK701 preset (with minimum phase and -10.1 db output) -> named the file "IR Sonarworks".



3. Reason 10 - create 3 audio tracks, each loaded with the same 24bit 48000hz audio clip 10 seconds of white noise.

4. Track one added Sonarworks with same preset as above -> mix Chanel solo -> exported loop 10 second. Normalized in Audacity and frequency plotted.
5. Track two added RV7000 with "IR Sonarworks" (EQ & Gate =off, IR length 100%, Stereo, Decay=127,Predelay=0, Gain=0, LF Damp=20hz, HF Damp=0, High EQ=0, Wet 100%-> mix Chanel solo -> exported loop 10 second. --> mix Chanel solo -> exported loop 10 second, Normalized in Audacity and frequency plotted.
6. Track three added Nadir VST, loaded "IR Sonarworks" (tried both mono and stereo routing with another IR load in second cabinet - same result as in mono ) selected Extreme Quality, mix Chanel solo -> exported loop 10 second. --> mix Chanel solo -> exported loop 10 second, Normalized in Audacity and frequency plotted.

Here is how each Frequency plot looks in Audacity (please observe that the dB scale is not exactly the same for each plot as I can not specify the dB range). Nadir looks worse

Sonarworks
Sonarworks White noise -9dB.PNG
Sonarworks White noise -9dB.PNG (111.4 KiB) Viewed 6912 times



RV7000
RV7000 White noise -9dB.PNG
RV7000 White noise -9dB.PNG (113.46 KiB) Viewed 6912 times



NADIR
Nadir White noise -9dB.PNG
Nadir White noise -9dB.PNG (118.5 KiB) Viewed 6912 times


In my opinion the graphs show that RV7000 is more acurate when it comes to the Freq response.

User avatar
Kategra
Posts: 327
Joined: 18 Jan 2015

03 Mar 2018

An this is how the Sonarworks Headphone VST looks like:
Sonarowrks AKG701 Average.PNG
Sonarowrks AKG701 Average.PNG (97.65 KiB) Viewed 6857 times

User avatar
QVprod
Moderator
Posts: 3488
Joined: 15 Jan 2015
Contact:

03 Mar 2018

Yep RV7000 is definitely more accurate in this example. Nadir also seems to have a lo-pass filter set around 20khz. Both have some hi-pass filtering. That said, the level differences of most the peaks in both graphs within the 20Hz-20kHz range compared to the original are within 1dB of each other so the difference after adding gain should be mostly inaudible.

User avatar
Marco Raaphorst
Posts: 2504
Joined: 22 Jan 2015
Location: The Hague, The Netherlands
Contact:

03 Mar 2018

Cool test. So rv7000 is even better than nadir!

User avatar
EnochLight
Moderator
Posts: 8405
Joined: 17 Jan 2015
Location: Imladris

03 Mar 2018

Kategra wrote:
28 Feb 2018
I don't like RV7000 for IR, done some tests and it seems that it has low resolution in applying a frequency response.
Kategra wrote:
03 Mar 2018
I have made new tests. ..In my opinion the graphs show that RV7000 is more accurate when it comes to the Freq response.
OK so, we're back to using RV7000 then? :D
Win 10 | Ableton Live 11 Suite |  Reason 12 | i7 3770k @ 3.5 Ghz | 16 GB RAM | RME Babyface Pro | Akai MPC Live 2 & Akai Force | Roland System 8, MX1, TB3 | Dreadbox Typhon | Korg Minilogue XD

User avatar
Kategra
Posts: 327
Joined: 18 Jan 2015

04 Mar 2018

EnochLight wrote:
03 Mar 2018
Kategra wrote:
28 Feb 2018
I don't like RV7000 for IR, done some tests and it seems that it has low resolution in applying a frequency response.
Kategra wrote:
03 Mar 2018
I have made new tests. ..In my opinion the graphs show that RV7000 is more accurate when it comes to the Freq response.
OK so, we're back to using RV7000 then? :D
For headphone correction will clearly use the original (Sonarworks).

For guitar.. I don't know...I'm really thorn between buying IR pack from Celestion to use after Vermilion amp or actually buy a cabinet with Vintage 30 and run it in my Blackstar combo ht-5r. Problem is, I never played through any real amp other than the Blackstar, so I don't know if vintage 30 is going to satisfy me.
Finding the tone i want got so high to my head that it distracts me from my 🎸 practice :)))

User avatar
EnochLight
Moderator
Posts: 8405
Joined: 17 Jan 2015
Location: Imladris

04 Mar 2018

Kategra wrote:
04 Mar 2018
For headphone correction will clearly use the original (Sonarworks).

For guitar.. I don't know...I'm really thorn between buying IR pack from Celestion to use after Vermilion amp or actually buy a cabinet with Vintage 30 and run it in my Blackstar combo ht-5r. Problem is, I never played through any real amp other than the Blackstar, so I don't know if vintage 30 is going to satisfy me.
Finding the tone i want got so high to my head that it distracts me from my 🎸 practice :)))
I guess my point is, you determined that the RV7000 seems to perform just fine. Correct? At least your last test indicates that it's more accurate as far as frequency response is concerned.
Win 10 | Ableton Live 11 Suite |  Reason 12 | i7 3770k @ 3.5 Ghz | 16 GB RAM | RME Babyface Pro | Akai MPC Live 2 & Akai Force | Roland System 8, MX1, TB3 | Dreadbox Typhon | Korg Minilogue XD

User avatar
Kategra
Posts: 327
Joined: 18 Jan 2015

04 Mar 2018

EnochLight wrote:
04 Mar 2018
Kategra wrote:
04 Mar 2018
For headphone correction will clearly use the original (Sonarworks).

For guitar.. I don't know...I'm really thorn between buying IR pack from Celestion to use after Vermilion amp or actually buy a cabinet with Vintage 30 and run it in my Blackstar combo ht-5r. Problem is, I never played through any real amp other than the Blackstar, so I don't know if vintage 30 is going to satisfy me.
Finding the tone i want got so high to my head that it distracts me from my 🎸 practice :)))
I guess my point is, you determined that the RV7000 seems to perform just fine. Correct? At least your last test indicates that it's more accurate as far as frequency response is concerned.
I've determined that it's more accurate than Nadir in freq response; Maybe there is room for a better RE IR loader than RV7000 :question: Really don't know..
Though, I have to admit, the free IR from Celestion, (the Vintage 30 mix, by Cenzo Townshend) sounds awesome trough it!

User avatar
Marco Raaphorst
Posts: 2504
Joined: 22 Jan 2015
Location: The Hague, The Netherlands
Contact:

04 Mar 2018

I am currently using an IR cab I made myself. Can't find differences between RV7000 and NadiR I should say. Only thing I often consider now is the CPU usage.

avasopht
Competition Winner
Posts: 3931
Joined: 16 Jan 2015

05 Mar 2018

selig wrote:
02 Mar 2018
I would tend to agree - you need to show three responses to determine if the problem is with the RV7000 or not: the original, the RV7000, and a convolution done “right”.


Sent from some crappy device using Tapatalk
Convolution on an impulse gives you back the IR exactly (that's why it's called called the impulse response ;) ). And when I say exactly, I mean exactly to the bit level representation of the IR.

It's not a matter of done "right," that's just the algorithm, and there is only one way to perform convolution (well two ways, but they're effectively the same and give the exact same result against an impulse). Applied to a signal you may see slightly different results between the two approaches due to different orders of floating point operations, but the error does not accumulate beyond the length of the IR.

User avatar
selig
RE Developer
Posts: 11685
Joined: 15 Jan 2015
Location: The NorthWoods, CT, USA

05 Mar 2018

avasopht wrote:
selig wrote:
02 Mar 2018
I would tend to agree - you need to show three responses to determine if the problem is with the RV7000 or not: the original, the RV7000, and a convolution done “right”.


Sent from some crappy device using Tapatalk
Convolution on an impulse gives you back the IR exactly (that's why it's called called the impulse response [emoji6] ). And when I say exactly, I mean exactly to the bit level representation of the IR.

It's not a matter of done "right," that's just the algorithm, and there is only one way to perform convolution (well two ways, but they're effectively the same and give the exact same result against an impulse). Applied to a signal you may see slightly different results between the two approaches due to different orders of floating point operations, but the error does not accumulate beyond the length of the IR.
That was the point of my example - if you’re not seeing the “right” results there’s something else going on, and you need to compare different devices before you can identify where that difference may actually be coming from (or to identify the problem as pilot error).


Sent from some crappy device using Tapatalk
Selig Audio, LLC

User avatar
Marco Raaphorst
Posts: 2504
Joined: 22 Jan 2015
Location: The Hague, The Netherlands
Contact:

05 Mar 2018

What might be tricky btw with the IRs is the attack of the files. What kind of attack do these EQ need? As fast as possible imo. This can be tricky.

For cabinet sims changing the attack is very interesting. I can create IR which are even more dynamic that the "real" cabs by changing the attack of the file.

I must say I work from ear and check how it feels while playing guitar. I guess for EQ some of the same issue might be part of the process.

Post Reply
  • Information
  • Who is online

    Users browsing this forum: No registered users and 25 guests