So I take it that 100 us
is 100 microseconds (the suggested delay for
speakers placed 3 meters apart)?
Yes, ěs = microsecond == .001::ms
I don't feel like doing math on the speed of sound right now but I do think that I keep hearing that with this technique you are supposed to keep the speakers close together and at 60° or something pointing outwards.
Sampling rate is aprox. 40k so 1 sample every
250 microseconds?
I think you are off;
<<<(samp / ms) * 1000>>>;
gets me 22.675737. Unless I'm confused (again!) that means 22.7 microseconds per sample.
So from what your saying, this means you'd need to
interpelate 250 times per sample?
I think it's more like 23 but yes. You'd have to do that in the case that you'd have a interpolating delay Ugen with feedback built in, something I don't think we have.
I'd put a few in series without feedback to get around this as you're likely using a amount of feedback that makes a impulse drop below the noise floor of your listening environment after a few iterations anyway.
Now that I think about it (outch this
math stuff makes my brain hurt), what does it even mean to hae a delay smaller
than the time between samples?
Well, obviously there is nothing "between" digital values... but the values represent a continuous wave so we can interpolate in the hope of figuring out what the value between two samples would be if it were there.
Let's say we are a delay line and right now we are getting a value of .2 at our input and we know that a sample ago it was .1 ok? Assuming linear interpolation (for convenience sake) we could then say that half a sample ago we had a input of .15 We never actually saw that input, of cource, but we could still output .15 and pretend we got this half a sample ago as our unput. We'd still be able to calculate one such value every time the VM "ticked" us which is once per sample. I don't really see any big problem here, it's not harder then a delay of length 100000.5 samples and nobody frowns at those, I think.