Page 2 of 4

Re: Matched Lowpass Filter

PostPosted: Mon Mar 21, 2016 9:48 pm
by Perfect Human Interface
Hi all. Just curious, is there a practical application for this? Besides saving on CPU I mean (which is never a bad thing). It might be a little over my head. :P

The EQ I use for music has an oversampling feature, but although I never have CPU issues I usually switch it off since I can't hear above 16KHz anyways. From what I can recall the oversampling can potentially introduce aliasing, or something along those lines, usually completely inaudible but could potentially create some small difference. I believe it handles white noise poorly. So could this type of filter design be used in lieu of oversampling to avoid that aliasing-or-whatever-it-was?

Re: Matched Lowpass Filter

PostPosted: Mon Mar 21, 2016 10:49 pm
by martinvicanek
Uhm, practical application: for many audiophiles analog filters sound much better than digital filters, and yes, the reason should be obvious from the spectra. Digital filters such as lowpass or peaking EQ are designed to approximate their analog counterparts, and the standard BLT frequency mapping does a poor job near the Nyquist frequency. So you can either oversample thus push the Nyquist frequency way beyond the audible band, or use a more accurate mapping, which is the subject of this thread. If oversampling is done correctly there should not be any aliasing (you need steep lowpass filters for that).

I'd say: A/B test a matched filter versus an RBJ biquad. Do you hear any difference?

Re: Matched Lowpass Filter

PostPosted: Tue Mar 22, 2016 12:27 am
by Perfect Human Interface
martinvicanek wrote:I'd say: A/B test a matched filter versus an RBJ biquad. Do you hear any difference?


Excellent idea. :D

I went ahead and put together a test schematic. With the lowpass I could not hear a difference. As I mentioned before, I can't hear above 16 KHz (I believe most adults can't as well), which is where I believe the difference is significant.

With the bandpass and highpass, I can hear a tiny difference at the highest extremes. I, personally, wouldn't describe it as "better," and might even say that the narrowing effect is preferable, but I'll welcome any counterpoints to that.

Let me know if I did anything wrong.

Edit: Fixed my goof. ;)

Edit 2: Would it make more sense if the filter crossovers could be set higher than nyquist (if that's even possible)?

Edit 3 (sorry): I forgot to try higher Q values. The bandpass peaks 10dB higher at a Q of 3 compared to 1. The matched version retains the same peak level. That's interesting. The matched BP also does seem slightly "smoother" at the extreme high-end, as you might expect, although the difference is tiny.

Re: Matched Lowpass Filter

PostPosted: Sat Aug 24, 2019 2:16 am
by wlangfor@uoguelph.ca
I learned a very interesting way of something that is also very similar to oversampling. that is a gentle slope hp and lp with linear co-efficients. If you match them correctly it sounds smooth like oversampling and still the data is somewhat easy to comprehend, also in a similar fashion. Maybe less hi tech but all the same impressive. I was thinking that due to the linear grades that there must be a bit extra.

Pretty cool too.

Matched Shelving Filter

PostPosted: Sun Sep 08, 2019 7:14 pm
by martinvicanek
Hi gang,
I revisited this topic when a guy asked me about shelving filters. Here is the result
matched1PoleShelves.fsm
(67.28 KiB) Downloaded 1440 times

And here is some reading to go along (Warning: explicit math content! :twisted: )
https://vicanek.de/articles/ShelvingFits.pdf

Re: Matched Lowpass Filter

PostPosted: Mon Sep 16, 2019 6:46 am
by guyman
Martin I don't know how I missed these posts/replys. my apologies !

Yes this is exactly what I was looking for !!! You are a genius man. There is a world of difference to my ears.

If I may, some pertinent things though to think about when you the have time. I think they would make these a superior collection:

- Complex Peaking filters (db atten)
- Complex All pass filters
- Complex versions of the shelves
- Are these modulate-able complex filters like ZDF? can we do ZDF?

The phase was inverse on the top shelf part of the matched peaking when I added the bandpass to the original signal, creating strange warp/cut at the top of the spectrum. I wish I understood the math behind all this, time to improve my learnins.

What's the phase like on all these, once we have a proper peaking, can it be blended with the original signal without phase cancelation..?

Re: Matched Lowpass Filter

PostPosted: Mon Sep 16, 2019 6:55 am
by guyman
Can we do a lower Q as well? These are genius.

Re: Matched Lowpass Filter

PostPosted: Thu Sep 19, 2019 8:22 am
by martinvicanek
guyman wrote:Martin I don't know how I missed these posts/replys. my apologies !

Yes this is exactly what I was looking for !!! You are a genius man. There is a world of difference to my ears.

If I may, some pertinent things though to think about when you the have time. I think they would make these a superior collection:

- Complex Peaking filters (db atten)
- Complex All pass filters
- Complex versions of the shelves
- Are these modulate-able complex filters like ZDF? can we do ZDF?

The phase was inverse on the top shelf part of the matched peaking when I added the bandpass to the original signal, creating strange warp/cut at the top of the spectrum. I wish I understood the math behind all this, time to improve my learnins.

What's the phase like on all these, once we have a proper peaking, can it be blended with the original signal without phase cancelation..?


On the first page of this thread I posted a more complete compilation, not just lowpass:
viewtopic.php?f=4&t=3710&hilit=shelf#p22688
Some of the things you are asking for are there.

Re: Matched Lowpass Filter

PostPosted: Fri Oct 11, 2019 3:19 am
by guyman
Been smacking my head against a wall all week :( please grant insight.

Only the complex bandpass allows me to modulate in the proper manner without noise, but the shelving portion is out of phase when combined to a bandpass, it can't be merged with the original signal. why is this? is it an error?

Re: Matched Lowpass Filter

PostPosted: Fri Oct 11, 2019 3:54 am
by guyman
perhaps I am decribing it incorrectly. The matched peaking filter works good, but the complex bandpass has a high frequency dip inverse to it's target "match" curve when combined with the original signal. Like a weird shelf.