Dual Allpass filters (Reducing IIR computation article)

[Tepeix]

Hi, just recently found this article.
https://www.dsprelated.com/showarticle/1269.php
(Reducing IIR Filter Computations Using Dual-Path Allpass Filters)
Seams very interesting but honestly, i could not make much with it
What do you think of this technique ?

I'm just happy i could make after a lot of try this little simple thing.
But it's limited to only one parameter for now...
Giving a fast slope for an approximate one octave.

[martinvicanek]

Nice! Allpass filters are in fact extremely useful, for example for complementary filters, fractional delays, reverbs, to name just afew. A good reference, alltough a bit old, is this:
https://authors.library.caltech.edu/120 ... ieee88.pdf

[Tepeix]

Thanks, i will read this !)
Now i just seen something about my schematic.. That we already talk but i forget a little.
It's cool to experiment with an external allpass code and delay like i do, (we get easily higher order)
but the problem with the Analyser is that it could add a one sample delay or some error in the signal order,
like you said me before...
So the plot will not be precise..
Maybe it's even better to have all in one code.. But i think that most parallel design works fine..

[Tepeix]

A very little and simple dual allpass filter i find useful. (but could probably be improved)

We could easily get low pass or shelf and control the curve but there's no Res for now..

[Tepeix]

Just find this other way to make an allpass.

It's a wave digital filter, but in this case it change nothing..
This is because the signal is separated in 2 in a way that recall how a current would travel in a circuit.
(but most allpass are like this.. not sure i understand all with this classification)

The main improvement is that we only need one multiply by allpass order.
The coefficient here might be from 0 to 2 and not -1 to 1.

[Tepeix]

Just find this about those type of allpass based filter.

When you combine the original signal or an allpass with another allpass, you could easily get complementary signals, for example a bandpass with subtraction and a notch with addition or an lp, hp....

But if you do this twice or more in series, the complementary part seams lost.

(here in green 4 allpass filter in series, adding the original signal get a lot of ripple due to phase,
and it seamed difficult for me to get the complementary hp)

But we could very easily make the original signal with the same exact phase.
Is it so much easy with other type of filter ??

Once we have an even number of filter in series, we just have to use the same number of allpass divided by two.
When they are in parallel in the filter, we could just make them in series.
(in the example i just have to add delay to compensate the delay of my system that make the SSE in series)

And now we could have the perfect complementary hp, or complementary of anything made, or the original signal in phase.

Yep, it's simple, but a great discover for me !)