Hi! Good Morning!!!

I'm writting for second time at this forum because i need your help one more time.

As I said in the last message i'm doing my final proyect with Pure Data. In this case i'm trying to develope a Wah-Wah effect which is based in a parametric filter which is moved in its main frequency along the time.
I found a path in this forum a few weeks ago which is called autowah.It's based on a parametric filter which change its main frequency according to a LFO which we can change its frequency and shape.

Then, here is my problem. I was developping this parametric filter wich i told before basing it in a paper which i found surfing the internet. The problem is that after i developed it in Matlab (a mathematics software) and run it right, i tried to translate it to Pure Data enviroment and it didn't run. I´m not sure if it's a patch's fault or a Software fault, because i couldn't fix the problem whatever i tried.

Explanation of the problem:

i have it implemented as i implemented it in matlab, and the filter coefficients are right if you compare it whith the Matlab ones. Then, it should work however it doesn't run. I controlled the problem and it appears when the output signal is put in the delay inlet which makes the main loop. The filter is based in a direct line from the main signal imput wich is summed to a delayed one sample signal, and two sample delayed one from the same signal. Add to it, the output is delayed in the same way one and two samples, and summed to the rest of the signals before the output. The problem appears when this output signal is delayed and summed at the end. Then the mean of the output signal (calculated with env~) is over four hundred dB and at the end most of the signals are inf# or over four hundred dB, except the main signal ways which bing values near to ninety dB. Then, I have tested the delays whith impulses in their entrance and they run well.

Thus i'm looking for your advices to make it runs well. The filter is similar to vcf~ but i need to do it by myself.

I left a link to my personal DropBox account where are all the files you need in Pure Data or even in Matlab(i have a problem uploading the things to this page). Add to it there is a video recording which explains perfectly which is the problem. (it is in spanish (i'm sorry)).

There you could find a carpet with the autowah.pd files found in this forum.
Another folder which contains the matlab files where the main one is which is called pruebaeqbanda2.m. In the last folder you could find the pure data files where the main one is called pruebafiltrobueno.pd. Add to it there is a .pdf file which explains the mathematic teory behind the filter and a video which i show how it works.

Thank you before your help, and excuse me about my english (i promisse i will do it better for the next time)

link: https://www.dropbox.com/sh/8mlb5xnnf5l8qot/OJ4IBX6OgG

Pd passes signals from object to object as vectors, or "blocks", not one at a time. By default, the block size is 64. This is important when working with feedback filters because it means that the fed back signal is delayed by at least the block size. So what's happening is that your y[n-1] and y[n-2] are actually y[n-65] and y[n-66]. You can use [block~ 1] set that feedback delay down to one sample, though you'll likely see an increase in cpu usage. You'll also need to trade out [send~] and [receive~] with [tabsend~]/[tabreceive~], because the first pair doesn't like block sizes below 64.

Also, keep in mind that it will already be delayed by one sample, so you'll need to fix those delay abstractions accordingly.

All that said, if you want to make your life easier, take a look at [fexpr~]. You can pretty much just plug your filter equation right into it.

Thank you very much Maelstorm!!

You were right, i have achieved to make it work with your advices, however i had to use another delay line for the main stream, and i delayed just for one sample the other ones , because the other way runs worse to me(maybe i didn't know how to run that).

I promise to upload that patch improved in a few days. I hope to use all your advices in the new implementation.

See you (every body) soon!!!

Hi! Good morning another time! (I hope you are better than me)

I'm becoming crazy about my patch. It runs, however it doesn`t run as i want.
I've tried the throw~ and catch~ boxes, i've tried to use send~ and receive~ boxes and even i tested it with delwrite~ and delread~ how ever none of them runs as i want. I know the problem must be in the buffer length because i can't have a narrow peak in the frequency as the vcf~ and bf~ do. I know maybe i did not develope a filter whith order higher as they have, how ever, when i test it in matlab there is no problem. In addition, in matlab, my filter amplify the signal and attenuate it, how ever in pure data, signal is just attenuated in the main frequency.

I would like to know how it runs like that, how ever i can't find and fix the problem.

Have you any other idea? If you have anyone i would like to hear that as crazy as it looks.

Thank you for reading my bad english.

I hope the new files will be uploaded to the link above before four o clock.

i forgot to say that when i'm using any of the linking boxes ( send, receive, throw,catch,... ) I have an error which says:

sigcatch 1243-signal1: unexpected vector size
throw~ 1243-signal1: vector size mismatch

or

sigsend signal1: unexpected vector size
receive~ signal1: vector size mismatch

if i link with a direct wire from the patch output to the inlet of de delays into the patch which we need to delay the output signal, appears an error which say that is being a loop, and it make that the send of signal stops working.

The patch is now aviable. if you couldn't download, please, say to me. The main patch is called "yafunciona4.pd".

Thanks very much and read you soon.

The "vector size mismatch" errors come from the fact that [send~]/[receive~] and [throw~]/[catch~] don't like block sizes under 64, so they don't work properly with [block~ 1].

However, [delwrite~]/[delread~] work perfectly fine. You just have to make sure the delay time for [delread~] is exactly one sample long. It expects the time in milliseconds, though, so you'll have to calculate the length in ms based on the sample rate.