BodePlot strange result (Non Linear and Sample Rate Adapt)

[Tepeix]

Hi,

Doing some measurement on allpass, i get strange result using this bode plot.
It's one from synthmaker era, done by Dofius, that i find very useful.

Normally i'm supposed to get the same result for each graph.
They each calculate exactly the same thing (or i'm wrong???),
But for the first one the same calculation are done twice (one for the out and one for the feedback)

Why different result ? And which one is the truth ?
I suppose the first one but like to be absolutely sure

[martinvicanek]

That's a bit tricky. If you build the filter in FS with discrete elements according to a textbook block diagram, the result will depend on the order in which you connect the parts. It is better to do it in code, where you have control over processing order.

[Tepeix]

Thanks !
I will do with code to be sure.
Could reproduce the 2 different solutions with a code :

Code: Select all

feed = mem[index];        /// First graph
p1=in+(feed*all1);
///feed = mem[index];     /// second graph if placed here
mem[index]=p1;
out=feed+(p1*all2);
   
index = index + 1;
index = (index<delay)&index;

So finally it's just one sample later for the feedback that make this.

[Tepeix]

This time a very strange Bode Plot result

There's only one code. The code reduce the high frequency when testing with a synth.
(in some maybe nonlinear process)

But the Bode Plot show the same reduced volume for all frequency.
What could cause that ??

[Tepeix]

That was maybe a stupid question..
Never really understand how the bode plot works exactly,
But it send an impulse that is 1 for the first sample then analyze the result.

My code works when the signal go up and down, maybe it need a +1 then -1 or even more complex impulse to be analyzed ?
I tried to send an inverted impulse with one sample delay. And get a little curve that show less treble.
But this curve might be underestimated. the trick is not precise at all.
(Edit: this trick don't works at all.. acting like some sort of prefilter.. )

Or maybe also it's because of a non-linear process?
I read somewhere that it's more hard to analyze the result when using non linear think.

[martinvicanek]

You are right in saying that the Bode plot really only makes sense for a linear system. Think of the single impulse as a signal which contains waves of all possible frequencies. You pass that to your device and at the output you do a Fourier analysis to see which frequencies got amplified or attenuated. But this only makes sense if the result is the same when you send the waves for eeach frequency individually, i.e. when your system response is linear.

A nonlinear device like yours will not only alter the amplitude and phase of a sine wave, but, in general, will also alter its shape. So it is not well characterized by a spectral response function, regardless of how you measure it.

[RJHollins]

A nonlinear device like yours will not only alter the amplitude and phase of a sine wave, but, in general, will also alter its shape. So it is not well characterized by a spectral response function, regardless of how you measure it.

I wish more people, using PluginDoc, analyzing 'analogue' flavor plugins, would better understand this.

Thanks Martin.

[Tepeix]

Yes thanks !) I was confuse.
Was thinking that even with non-linear we would see a little what happens.
But the result is clearly unusable.
Maybe i could make some code with fixed input that replace the non linear part to test different volume input.
But would i take the time ? Finally testing with a synth, a scope and FFT give an idea of what happens.
Maybe we have to join the author of PluginDoc to said them to upgrade.

[Tepeix]

Again a little stupid question

Is it possible to adapt the bodePlot to different sample rate ?

I find one way using the oversampling module and adapting the filter.
But it could only works for double sample rate.

I also try to resample the array but it doesn't seams to works for now..

[Tepeix]

I was thinking that removing the analiser and replacing it by blue code and mem i could get a plot that adapt to sample rate..
But it don't works, doing the same as a normal plot...
Strange.
Next try will be to resample the array..