Well, most of the time you would do convolution in "gather" mode, where you multiply each input sample with one kernel tap, add them all together and that's your output sample.Thanks! I definitely will try this, when I'll give longer Bleps a try. The workflow you write down sounds like convolution, but with an IR that is dependent on the size of the phase resets.
But we can transpose the whole thing into "scatter" where we take one input sample, multiply the whole kernel with that sample and then add all these taps into the output buffer one by one.
For regular filtering, gather is generally preferable as more efficient (with some special cases where scatter might make sense), but for BLEPs the scatter convolution is perfect, because even though the input sampling rate is theoretically infinite (ie. continuous time), only the occasional "input sample" here and there actually produces non-zero residue... so BLEPs are convolution really, just in a very optimized form.
Statistics: Posted by mystran — Fri Jan 03, 2025 9:03 pm