Hello,
I think I managed to solve this a long time ago but forgot the solution. In Pd apparently it is impossible to have a feedback with "zero" delay, i.e. if a delwrite~ and delread~ objects are in feedback the minimum delay is one block, thus the only option is to use [block] and reduce the blocksize.

How can "zero" logical delay be achieved in feedback without resorting to reducing the block size?

L

Edit: just to be quick and clear: this is a trouble you stumble into when designing Karplus-Strong patches

You have to use [block~ 1] for feedback loops. There is a workaround for feedforward delay lines, but for feedback you don't really have a choice. Unless you just want to write your Karplus-Strong as an external instead.

Or use fexpr~ ?

[ fexpr~ $x + ( $y[-50] + $y[-51] ) / 2 ]

Here the feedback delay (51 samples in my example) is bounded by the blocksize, so you might need to increase that for low pitches.

I will give a try to the fexpr~ to write a biquad (that's what I'm using at the moment) in fexpr~ language and the feedback...

Despite being fairly easy to use, [fexpr~] is a bit notorious for being a cpu hog. So, I would recommend trying both approaches and see how they affect your cpu usage.

This is actually a nice reminder for me. I've been meaning to see how a biquad in [fexpr~] compares to my vanilla abstraction (which uses [czero~] and [cpole~]). I would think it would be better, but everything that I've used with [fexpr~] has resulted in a noticeable cpu spike.