DSP Robotics Support

I am working on a digital waveguide modeling of a string. the algorithm compares given string plucked recording by comparing the signal with a one period delayed version to extract feedback filter. It turned out to be very sensitive to precision of delay - rounding delay to integer samples works badly so I used linear interpolation. However, that turned out to be even worse, because linear interpolation messes with high frequency contend too much.

Does anyone have better interpolators implemented or guide me where to look for some algorithms?

The stock delay in Synthmaker has a nice clean spline to it, else google optimal 4. Probably has it.

Edit: liniar is not that bad compared to no interpolation..
You will only lose more. Liniar has about the best high frequency response around.
The spline will have a bit better post niquist, Ie no sagging in. optimal is worse, but the best around full spectrum.

There was a pack of interpolators with visual graphs for synthmaker which should work in flowstone too here:

http://www.synthmaker.co.uk/forum/viewt ... f=4&t=9671

they aren't actually set up as delays but it's fairly easy to connect them up to multiple outs on a basic integer delay.

the stock 'interpolated delay' uses allpass interpolation. if you only want a static delay that should work (does have a small rounding error which can build up in feedback loops it seems) so not so good for the actual synthesis part but ok.

Oversampled linear is another way to get good results i think, could look at the 'oversampling toolkit' from synthmaker:
http://synthmaker.co.uk/dokuwiki/doku.p ... ersampling.

Good luck with your waveguides, I've been working on them a lot too. will put up a demo of some stuff soon including some drums and brass.

thanks for advice, I had hard time finding something useful.
In the end allpass interpolation was good enough (I've came up with a way to decrease the sensitivity to the artifacts). The project is already working very fine, you may check here http://www.dsprobotics.com/support/viewtopic.php?f=3&t=2364

yeah the allpass is often fine, I've been making nonlinear cymbal waveguides though which are pretty unstable with all the modulation so I've started looking beyond that. modelling a whole string with all delays replaced with series allpass interpolators worked pretty well but with predictably ridiculous CPU, might as well be doing FDM by that point which uses half the modules (as both directions use the same masses and springs). I'm still reliant on it's brother the dispersive allpass for detuning effects.

It seems that the best advice for people using flowstone for audio is to search the old synthmaker forum also, not necessarily that obvious from this site though.