Kontakt EQ: which kind of filters?

[Derozer]

Hi,

do you have idea of which kind of filter Kontakt EQ use?
Manual:



I feel them nice filters, even with fast modulation, no clicks.
I dubt they are biquad.

SVF? ZDF? Any idea? I would like to use somethings similar on my plugins.
Tried Biquad by RBJ, but with fast modulation, they are pretty unstable.

Any tips is welcome. Thanks

[Derozer]

Nobody know it? Thanks

[mystran]

Nobody know it? Thanks

Likely only the authors of Kontakt really know what kind of filters they use.

That said, the trapezoidal (ie. ZDF) state-variable is a nice, efficient filter that handles modulation just fine. See The Art of VA Filter Design, chapter 4. I suppose it's not quite an "EQ cookbook" but most of what you need is there.

The RBJ-peaking EQ (and shelves as well) are slightly different though, in the sense that do half (in decibels) of the gain adjustment by moving poles and the other half by moving zeroes, where I think Vadim's version just moves zeroes (although it's possible the symmetric adjustment is covered somewhere else in the book). In either case, the RBJ approach keeps the half-gain bandwidth constant.

You can also use a trapezoidal SVF with RBJ-formulas, basically by taking the coefficients for the analog prototypes rather than the digital coefficients. The only catch is that for shelves you'll want to use the nominal cutoff (ie. shelf mid-point) for the tan(pi*f/fs) prewarp and offset the poles from there, rather than warping at the pole frequency directly.

That said, the "bandwidth prewarp" from RBJ is somewhat less than ideal. With straight-Q the bells become more narrow as they approach Nyquist. With the RBJ formula, the opposite happens. The ideal would be either somewhere in between these two extremes, or some better approximation of the analog response (eg. Orfanidis, MZTi, etc).

[EvilDragon]

I'd venture those old Kontakt EQs (1/2/3-band) and filters are just biquads. In fact, they all come from zplane fxpack.

[Z1202]

That said, the trapezoidal (ie. ZDF) state-variable is a nice, efficient filter that handles modulation just fine. See The Art of VA Filter Design, chapter 4. I suppose it's not quite an "EQ cookbook" but most of what you need is there.

The RBJ-peaking EQ (and shelves as well) are slightly different though, in the sense that do half (in decibels) of the gain adjustment by moving poles and the other half by moving zeroes, where I think Vadim's version just moves zeroes (although it's possible the symmetric adjustment is covered somewhere else in the book). In either case, the RBJ approach keeps the half-gain bandwidth constant.

In the latest edition of the book I moved the RBJ-compatible EQs to Chapter 10, covering only "naive" shelving in Chapter 4. Although looking now at the beginning of Ch.10, I admit it may look a bit daunting But then, it also covers elliptic and wide variable-slope EQs

[Derozer]

I'd venture those old Kontakt EQs (1/2/3-band) and filters are just biquads. In fact, they all come from zplane fxpack.

Really? I've tried these biquad implementations:

- RBJ Direct From I
- Nigel's Transposed direct form II
- Martin Vicanek's Transposed direct form II

They are all very clicking compared to Kontakt ones, while changing its parameters

[hugoderwolf]

If you smooth the parameter changes a bit and update the coefficients more often (like every 64 samples may already be sufficient), you should be able to get rid of the clicking, even with Biquads. You don't need fancy audio-rate modulation SVF filters for a simple EQ.

[mystran]

If you smooth the parameter changes a bit and update the coefficients more often (like every 64 samples may already be sufficient), you should be able to get rid of the clicking, even with Biquads. You don't need fancy audio-rate modulation SVF filters for a simple EQ.

I would like to argue that there really isn't any reason to ever use direct form biquads for anything, since they have essentially no advantages whatsoever.

[hugoderwolf]

I wasn't trying to argue about that. However, the filter structure starts to get relevant only once a proper parameter smoothing and updating is in place. First things first.

[Derozer]

If you smooth the parameter changes a bit and update the coefficients more often (like every 64 samples may already be sufficient), you should be able to get rid of the clicking, even with Biquads. You don't need fancy audio-rate modulation SVF filters for a simple EQ.

Of course I'm already smoothing all param. And update coefficients at audio rate. Still really clicking, compared. Not sure what I'm missing.

If I post code will help?

[mystran]

If I post code will help?

Possibly, in case there is some other problem.

[matt42]

Really? I've tried these biquad implementations:

- RBJ Direct From I
- Nigel's Transposed direct form II
- Martin Vicanek's Transposed direct form II

Of course I'm already smoothing all param. And update coefficients at audio rate. Still really clicking, compared. Not sure what I'm missing.

Direct forms wont work well if updated at sample rate. For those perhaps update every 32, or 64 samples, or something. You'll have to go by trial and error as to how blocky vs how glitchy the updates are/what you can live with.

For samplerate modulation you'll probably need to go with some kind of ZDF/TPT filter. There is an implementation in one of the technical papers on the cytomic site.

(EDIT: As Hugordewolf points out 64 samples will probably be quite smooth for simple EQ adjustments. Note he points out that you dont need fancy audio rate modulated SVF (I'm assuming some kind of ZDF/TPT structure), implying (at least I assume again) that you can't get audio rate from direct forms anyway.)

[Derozer]

I think I have got it, there was a problem on smoothing.
Now it seems as good as Kontakt one.

Defo biquad filters