Sigmoid transfer function troubles
Hi community.
I'm trying to implement a sigmoid transfer function for my audio input. The function I need to use is:
F(f) = h + ktanh(f)
I've tried to implement it in pd (see attach) but not sure that what i've done is correct. I also wasn't sure if there were any sigmoid objects already out there in pd land.
Any thoughts?
The math in your abstraction is just fine. However, typically in Pd you want the audio that is being processed to be the leftmost inlet. It's just the convention and will avoid confusion if you make sure to follow it. Also, I personally would make the other [inlet]s into audio [inlet~]s. That way, you have the option from outside the abstraction on whether or not you want to smooth the parameters with something like [line~].
Thanks Maelstorm. Will try that out. I've come to be believe that you're an expert on filters.
Have you had/seen any successful fractional delay interpolation filters in pd? I made one last year, but it was pretty clumsy. Wondered if you know of any good ones.
I wouldn't really say I'm an expert of filters. I just know enough to find my way around them. 
I messed with some interpolation filters a while ago, but I can't seem to find the files right now and don't really remember which ones they were. But when it comes to building filters in Pd, as long as they are simple IIR or FIR filters, they can be built with the raw filter objects (such as [czero~], [cpole~], etc.), though it will require some math on your part to break the filter up into a series of first-order filters. Obviously, the higher the order, the more of a pain in the ass this is, and potentially more processing to calculate the coefficients.
More complicated filters can be done in [fexpr~], though it is know to be a cpu hog. It is, however, easier to use, as you can basically just plug the equation right in.
If you can't do it using either methods, than it's probably safe to say you'll have to go with a clunky implementation using [block~ 1] or write it as an external instead.
What was the filter you implemented?
Yop,
in 2000 I used a first order allpass fractionnal delay implementation in a violin string/bow interaction model. It had to be followed by a lowpass in order to improve stability. Here is a document where this method is explained : http://www.atelierenbois.be/includes/tfe.pdf . The core is presented from page 43, and there's a addendum at page 91. Sorry, this is in french.
Anyway, not sure this will be of some help...
Arnaud
Well you certainly know more about them than I do Maelstorm! Anyway. The last attempt I made at an interpolation filter was 4-point lagrange, but as I said, it was pretty clumsy and i'm not entirely sure it functioned properly as it was a pretty rushed effort!
Anyway i've attached it just so you could get a look. Maybe you can see where things went wrong.
And Arnaud, thats great that you've got experience and thanks for the pdf, but my french is far from workable so as much as I partly recognise some of the maths, there's not much I can extract from it. Did you produce your waveguide in pd?
Basically this is all for a flute waveguide i'm currently working on thats having a few teething issues to say the least. In it's current form it kind of sounds like a flute, but doesn't behave as such. For example once you create the resonance, turn off the air jet or excitation, the tone should stop because nothing is keeping the resonance going, yet this doesn't happen, which says to me some form of missing link in the loss filters. I recon it's the fairly dramatic oversimplification of the filter thats causing it, but wondered if any of you guys would take a look?
Should I post the waveguide on here or make another topic, Maelstorm?
Any thoughts?
Oops! Looks like something went wrong!
