Convolution Processing with Impulse Responses

[monoman]

I have my G2 engine now for a long time but never have posted any topic here.

I just woke up today and had following idea:

What about using the G2 for convolution processing with impulse responses.
I know there a plenty of vst reverbs using this technologie. But I want to go a step further. In theory you can use this technologie for any source. Like Focusrite's Liquid Mix emulating comps and eqs.

I'm not in to math that well but can I use the G2 for convolution processing?
Any thoughts out there?

Cherrio
-mono-

[monoman]

More thoughts (after a couple of cigarettes and coffee):

Can I use side chaining to get the desired results?

What about if I use an impulse response of a compressor to sidechain e. g. a bassdrum?

I would need a sampling tool for the G2 to get the impulse response into the G2, hook up an audio source to the 2 inputs, a sidechained envelope for the impulse response controlling the input signal. Does this make sense???

[iPassenger]

[jksuperstar]

There's two basic ways to do it:

1.) Multiply in the frequency domain. This means converting both your impulse response and input signal to the frequency domain (FFT, or other), multiply the two together (number of multiplies depends on the sampling frequency, could be 96k!! per sample). Afterward, convert back to the time domain. Not really a good method, because of the latency involved.

2.) Do it in the time domain. This involves, for EACH and EVERY sample coming in to the G2 (that is - 96000 per second, per channel), multiplying the entire impulse response by the input signal, dumping that into a FIFO like buffer, and also summing the result of 96000 multiplies (per sec, per channel). That's ALOT of math.

So, it's possible, that you could do something with a lowered sample rate, and/or a very short impulse response.

Just to put it into perspective, for a little while, there where some machines (from Sony, etc) that attempted to do impulse responses in real time. They were a few rack spaces high, and large cooling fans to help deal with the heat from all of the DSPs they put in there. They weren't cheap by any measure, also!

[Chet]

Convolution is not possible on the G2, unfortunately. There's no module for it, and there's not enough real-time memory to store a decent-sized impulse.

[fac]

Just a few ideas:

Convolution can be seen as the sum of delayed versions of the input signal, each of which is multiplied by some gain. So I'm guessing one could implement a convolutor using 8-tap delays whose outputs are all connected to a 8-input mixers. The levels on the mixer would be the coefficients of the impulse response. It should be possible to daisy-chain 8-tap delays by connecting the 8th output of a delay to the input of the next delay (these delays use relatively little CPU and RAM, so one should be able to chain quite a few). This should make it possible to implement convolutions with short static impulse responses (e.g. 8 to 32 samples/taps). May not be enough for a reverb impulse, but it may produce some interesting comb-filtering effects by modulating the inter-tap delay.

Now, let's say you build this convolutor, and add a feedback loop by mixing the convolutor output with the input signal and feeding it back to the convolutor input. Or better yet, having independant feedback control for each of the 8-tap delay modules. This would turn the FIR convolutor into an IIR system which will emulate longer impulse responses and could be worked into an usable reverb.

[3phase]

not possible? maybe a job for Tim ?

but i guess when mr singer claims it impossible it is probabbly impossible... is it impossible with reaktor aswell?

[Chet]