This is not for the faint of heart (it is poorly organized and not documented at all), and there is no build environment set up for OSX or Windows, but I wrote a C language API that I rolled into a Pd extern which takes compiled C files that describe feedback networks with approximately 0 time delay. This is done by modeling the network with dynamical systems and solving numerically. I gave a talk about this at ICMC this year.

Using these tools I made a program that runs in Pd that has two oscillators that can frequency modulate and sync one another and themselves. It maintains the 64 sample block size and the CPU load is minimal. It can easily be extended to N oscillators.

I can send a copy of the paper to you directly if you are interested (I don't want to hit the whole list with a pdf). Here is the project on github -- please forgive my many sins, this is not yet ready for release and it shows. https://github.com/dmedine/timelab
-David

On 01/08/2016 03:00 AM, i go bananas wrote:
> If you want shorter feedbacks than 64 samples within a patch, there is
no way around reducing block size

i still have this itching doubt in that.  i think what we're looking for is some real leap of imagination here.  
we are fine with having a block or 2 latency,,,so i'm imagining maybe there's some far out hack involving synchronised tabread~'s and delayed osc~'s, or something like that,  which would somehow do it. 

>You may be able to optimize by
putting only the very necessary (for the feedback loop) objects into the
re-blocked subpatch.

yeah that seems the most likely approach now. 

>if cpu is a limit, i guess the only solution is to create an external.

this is probably the 2nd most likely solution.  

thanks guys for your ideas and help.  

On Fri, Jan 8, 2016 at 7:34 PM, Roman Haefeli <reduzent@gmail.com> wrote:
If you want shorter feedbacks than 64 samples within a patch, there is
no way around reducing block size. You may be able to optimize by
putting only the very necessary (for the feedback loop) objects into the
re-blocked subpatch.

Then there are some classes that do internal sample-size feedbacks,
like [rpole~] or [fexpr~]. While [fexpr~] is quite flexible, it may be
even more expensive than a subpatch with blocksize=1 (at least that's
what I remember when I compared two implementations, but  this
observation might have been specific to that kind of problem).

Roman

On Fri, 2016-01-08 at 18:42 +0900, i go bananas wrote:
> Hi all, hope everyone's well.
>
> We're trying to implement a 4-op FM matrix with feedback, copying a
> patch my friend made in reaktor using a block size of 1 (sorry, don't
> know the full details of that, but he says he can get 1 sample delay
> for the feedback)
>
> Has anyone ever succeeded doing something like this in pd?  I know
> about the order forcing using subpatches like in G.05.execution.order
> help patch, but that doesn't seem like it will work here, as we still
> get DSP loop errors when trying to connect the output of one osc~ back
> into the frequency input of the others.
>
> I'm really looking for a solution that doesn't involve using blocksize
> of 1, and anyway, even doing that, still seems the only way to do
> feedback without getting DSP loop errors is with s~ / r~ pairs, which
> seem to only work at blocksize of 64 anyway?
>
>
> I don't mind adding a bit of latency to the whole system if there's
> maybe a hack to do this with tables or something,,,but am really stuck
> here wondering what to do.
>
> any ideas?
>
>
> cheers, Matt
> _______________________________________________
> Pd-list@lists.iem.at mailing list
> UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list


_______________________________________________
Pd-list@lists.iem.at mailing list
UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list




_______________________________________________
Pd-list@lists.iem.at mailing list
UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list