FIR filter with "visual designer"

philter5 · 8 Oct 2008

thx rl 😮

TheAudiophileDutchman · 8 Oct 2008

Great stuff! 😮

As you know, just adding some oversampling can yield a more expensive sounding EQ. 😇

From Shanghai with 😍 😉

kirkagur · 8 Oct 2008

rl wrote: Thanks to the new iFFT modules here is a little sketch for a visual FIR filter. Paint the desired frequency response using the mouse. [img]xxx[/img] Works nicely for spectral noise shaping, as a cheap sounding graphic EQ or formant filter. The frequency resolution in the lower frequency area is a bit lousy, though, but that's the nature of FIR filters. Download Link

is this filter linear phase?

if i have an IR of an expensive filter can i use this FIR filter to emulate that filter???

ronaldpassion · 7 Apr 2010

Hey, it is so interesting to me. Btw how to use it?

Why i change one bar, another thing change too?

TQ,RP!

rl · 7 Apr 2010

kirkagur wrote: is this filter linear phase?

if i have an IR of an expensive filter can i use this FIR filter to emulate that filter???

Sorry Kirkagur, I missed your questions.

- yes, it should be linear phase - yes, in principle you could do that, but it would be of limited use, because the frequency response is just a snapshot. Take a lowpass filter for example: each setting for cutoff frequency and resonance would result in another frequency response). So you could use it only for static settings.

ronaldpassion wrote: Hey, it is so interesting to me. Btw how to use it?

Why i change one bar, another thing change too?

TQ,RP!

- just put it into a signal path - I don't know exactly your question, Ronald, but I think your asking about the inaccuracy. With fewer taps, the setting is pretty coarse. The more taps are used, the stepper the rolloff becomes.

oddson · 7 Apr 2010

Rudolph... are you confident that you've got everything correct here?

I was playing with symmetric FIR coefs and it seemed I was getting much better fits with fewer taps -- even though (n-1)/2 of them have constrained coefficients. But I wasn't sure that the curves I was plotting (using the left-hand-side of the iFFT only) were valid because... well, frankly I don't really understand this stuff. But just trying to implement low-passes the ones from your device didn't seem to perform very well.

It just seems to me the inaccuracy is very high and doesn't really seem to improve as much as I would have thought when the number of taps is increased.

kirkagur · 7 Apr 2010

oddson wrote: Rudolph... are you confident that you've got everything correct here?

I was playing with symmetric FIR coefs and it seemed I was getting much better fits with fewer taps -- even though (n-1)/2 of them have constrained coefficients. But I wasn't sure that the curves I was plotting (using the left-hand-side of the iFFT only) were valid because... well, frankly I don't really understand this stuff. But just trying to implement low-passes the ones from your device didn't seem to perform very well.

It just seems to me the inaccuracy is very high and doesn't really seem to improve as much as I would have thought when the number of taps is increased.

oddson if i had your english, that`s what I would say... 🙂

rl · 8 Apr 2010

Well, when I did this experiment I found it pretty cool that SM was now able to do iFFT processing. I checked my example against known frequency responses like a brickwall lowpass filter with sinc coefficients and checked against sambeans FIR example. So far for me the result looked reasonable. But I do not claim to have the final wisdom on this. If you want to help to improve the schematic be my guest. Heck, I even have to remember what I did at that time - one and a half year ago.