How to make an IR? "Deconvolving?"

TimmyO

If I understand the basic concept, an IR is made by suirting a known signal (either a pulse or a sweep) into the item to be emulated, and recording the output, then kind of removing that original (known) signal again, right? 

And if that's correct, then it is as possible to make an IR of say an overdrive pedal (albeit at a single setting)

Does anyone know what tools might be used to do it, what the steps are and what type of signal to use? 

Roland

Easy enough to do, but not particularly useful. Put a known signal into the pedal, and compare the output with the direct signal. You'd get the EQ of the pedal, but not the other aspects like compression and waveform distortion.

TimmyO

Surely the output signal has been compressed and distorted, and captured as such? 

stratman3142

My DSP knowledge is a bit rusty but I think basic convolution works because a linear system can be characterised by its impulse response. Compression and distortion are non-linear system characteristics.

I expect more complex processing could include an amplitude dependency. Is that how a Kemper works I wonder (I'm getting out of my depth now without reading up on it)?

PolarityMan

It's not a huge jump to what kemper do with profiling though.

Roland

TimmyO said:

Surely the output signal has been compressed and distorted, and captured as such? 

True, but it's not uniform across the frequency range. The compression and distortion that you often hear in an amp is frequency dependent. Many amps and pedals are designed to distort more in certain frequency ranges, so they boost those frequencies before distorting, and reduce them afterwards. Hence frequency dependent distortion characteristics. Depending on how the pedal circuit is designed you can get those factors, plus clipping which adds a harsher edge to the sound.

stratman3142

Roland said:

TimmyO said:

Surely the output signal has been compressed and distorted, and captured as such? 

True, but it's not uniform across the frequency range. The compression and distortion that you often hear in an amp is frequency dependent. Many amps and pedals are designed to distort more in certain frequency ranges, so they boost those frequencies before distorting, and reduce them afterwards. Hence frequency dependent distortion characteristics. Depending on how the pedal circuit is designed you can get those factors, plus clipping which adds a harsher edge to the sound.

Even if compression and distortion were uniform across frequency (hypothetically) the compression/distortion would still be a function of the amplitude of the input signal (i.e. non-linear). So, I think you'd need to (in effect) take a number of impulse responses of varying amplitude so that you could do some interpolation. That's thinking in the time domain. There may be other approaches in the frequency domain.

stratman3142

The more I think about it, Kemper profiling must be taking account of non-linear characteristics.

I did a bit of a search and there might be a clue at the link below:

https://www.reddit.com/r/DSP/comments/3d5uhc/how_does_the_kemper_profiling_amplifier_work/?st=j6z7jdj1&sh=6aadcca9

The link above has another link to what's referred to a 'dynamic convolution'. It sounds plausible but I don't know if it's correct regarding the Kemper.

It would be good to know if anyone has any info on the theory behind Kemper profiling and whether that relates to the question of the OP.

TimmyO

Roland said:

TimmyO said:

Surely the output signal has been compressed and distorted, and captured as such? 

True, but it's not uniform across the frequency range. The compression and distortion that you often hear in an amp is frequency dependent. Many amps and pedals are designed to distort more in certain frequency ranges, so they boost those frequencies before distorting, and reduce them afterwards. Hence frequency dependent distortion characteristics. Depending on how the pedal circuit is designed you can get those factors, plus clipping which adds a harsher edge to the sound.

Is that not why they use a swept signal rather than a pulse for some IRs?

stratman3142

TimmyO said:

Roland said:

TimmyO said:

Surely the output signal has been compressed and distorted, and captured as such? 

True, but it's not uniform across the frequency range. The compression and distortion that you often hear in an amp is frequency dependent. Many amps and pedals are designed to distort more in certain frequency ranges, so they boost those frequencies before distorting, and reduce them afterwards. Hence frequency dependent distortion characteristics. Depending on how the pedal circuit is designed you can get those factors, plus clipping which adds a harsher edge to the sound.

Is that not why they use a swept signal rather than a pulse for some IRs?

A linear time invariant system can be characterised by either its impulse or its frequency response. There's a Fourier transform relationship between the two (see link below).

https://dsp.stackexchange.com/questions/536/what-is-meant-by-a-systems-impulse-response-and-frequency-response

But, for a non-linear system (e.g. distortion) I still think you'd need to do that characterisation for a number of signal amplitudes.

oaf

You can use something like this:
http://www.voxengo.com/product/deconvolver/

or for something cleverer look at stuff from Acustica Audio
https://www.acustica-audio.com/
They've done something different with Nebula 4 (some kind of additional licencing/training thing I think, I forget the details) but v3 allows you to make your own. There's an article on it in SOS if you're interested:
https://www.soundonsound.com/techniques/sample-your-gear-acustica-audio-nebula

Distortion units don't always sample so well, particularly at the more extreme settings.

oaf

Here's an article on using Reaper and its included Reaverb plugin
https://www.soundonsound.com/techniques/reaverb-part-2