Convolution

The process of convolution is used in several Guitarix Modules: [Convolver] (both stereo and mono), [Amp_impulse], and cabinet. The convolution engine is jconvolver. Guitarix's two convolvers allow the user to select and manipulate the wav files used to convolve the live signal. The others allow adjusting pre-existing impulse responses (IR). See below for explanation.

Convolution is the folding together of a sound and a space so that it sounds like it is occurring "in" the space. It is analogous to the ground in the visual concept of figure and ground. Convolution is performed by making a sort of digital fingerprint of the target space by recording an impulse, like a balloon pop or a swept sine wave, then removing the impulse, leaving only the response to it (IR) without the original signal, like the sound after the pop everything in the signal that is not pure sine wave. In a real space, this leaves the sound of the space without the isource sound included any more. Hardware as well as spaces have these digital fingerprints, and they can be constructed synthetically too. The fingerprint is called an impulse response (IR). Convolution is when any other sound is then be played "through" the IR (the actual process is a kind of complex moving multiplication not unlike a moving average in statistics). The result is that the input sound resonates in the audible space of the impulse response, so a singer in a dead room can be made to sound as if she is singing in the third pew of St. John the Divine Cathedral as heard by a listener twenty feet away, if the balloon was to be popped in the third pew, and the microphone used to record it was twenty feet away. Or through a particular 4x10 open backed cabinet recorded on a single shure sm57 from 1 inch, slightly off axis. Or through an expensive compressor at a particular setting.

Any sound can be used as the IR to be convolved with any other sound. The formerly popular vocoder effect is a good example of this form of convolution, where the fingerprint of say a synthesizer is folded together with a voice. The impulse response does not even have to be the sound of a real space. Convolution signals can be constructed mathematically as easily as by recording.

Convolution is an excellent way to model real speaker/microphone configurations for two reasons. First, the fingerprint of the speaker/microphone combination is a very short complex echo. The shortness of the echo makes it much less processor-intensive than longer reverberation tails like a cathedral, which increase the processing power required exponentially in relation to the length of the IR. Second, speaker output is ideally somewhat linear compared to other parts of the signal chain. Unlike an amplifier, which should respond differently at different volumes of both the input signal and the amp, a speaker responds to being driven "close enough" at different volumes that an impulse recorded at the speaker's optimal output level will be good enough for any setting.

One issue in using "equipment" impulse responses is that the signal chain is then followed by whatever hardware amplification and speaker combination follows, which further colors the results, taking away from the accuracy of the modeled sound. This is not as much of a problem for guitar amps, because the frequency response of a good monitoring setup is relatively flat with a much greater frequency response than guitar speakers. The obvious way around this for recording is to record the digital signal directly, so it is only really an issue for live use.