[bug] Filter makes noise all by itself

[Matt_SF]

@d-healey Strange. I got nothing here : no crazy sound, everything works fine.
Tested on Mac & PC, both Apr. 17 builds...
Anyone else ?

[David Healey]

@Matt_SF Ah might be a Linux thing then. I tested on 2 distros and both have the issue. I just tested on Windows and no problem there.

[Frankbeat]

Tested on MacOs 10.12.6, HISE develop ’23, the noise (saw tooth to be exact) is triggered when the knob goes to zero.

[David Healey]

@Frankbeat Thanks for confirming.

[Frankbeat]

@d-healey You're welcome! Now I had the chance to do a little for you finally :-D

[David Healey]

@Christoph-Hart Any chance of getting a fix for this? I just had a user report the issue on a Windows system.

[ustk]

@d-healey Does it happen with all filter types or just some of them?

[David Healey]

@ustk I just tested with the snippet from my first post and it only happens with the Peak EQ

[ustk]

@d-healey The filter coefficients are exploding which puts the filter in an unstable state (auto-oscillation)

The value seems to be linear instead of log (which should be the case for the Gain), this might be why the filter explodes at 0, but it's hard to confirm.

Have you tried to recompile using this pre-proc HISE_USE_SVF_FOR_CURVE_EQ ?
Though I don't know if it impacts the Filter Peak EQ, it might only be for the Parametric EQ module

EDIT: Apparently it is only used in the Parametric EQ to replace the biquad with a state variable filter...

[ustk]

@d-healey Confirmed, 0 creates a gain to -100dB which makes the filter unstable. I found where it happens, now it's a matter of finding the right conversion...

[ustk]

@d-healey Alright I got it to work.
Now the question is, what is the range we want?
With the fix, it applies a -18dB to 0dB gain (cc val 0 to 127)
meaning that with these Gain values we get the associated modulation result below:

• Gain set to 0 -> cc val 0 to 127 -> -18dB to 0dB
• Gain set to +18 -> cc val 0 to 127 -> 0dB to +18dB
• Gain set to -18 -> cc val 0 to 127 -> -36dB to -18dB

So I wonder if CC val 64 shouldn't be the middle value instead, so the range becomes

• Gain set to 0 -> cc val 0 to 127 -> -18dB to18dB
• Gain set to +18 -> cc val 0 to 127 -> 0dB to +36dB
• Gain set to -18 -> cc val 0 to 127 -> -36dB to 0dB

So the modulation always adds -18 to 18 to the value the Gain is already set to, a bipolar-like modulation but around 64...
What do you think?

[David Healey]

@ustk

What do you think?

I think we should get @Christoph-Hart's opinion

[Christoph Hart]

The best option here (which I thought it was already) is that it applies the gain modulation as simple multiplication on the dB value - so if you have set it to +8 dB, the modulation goes from 0 to +8 and if you set it to -12dB, it goes from 0 to -12dB with an inverted curve (so that a modValue == 0 still means 0dB).

I think I've never used the gain modulation on a filter...

[ustk]

@Christoph-Hart Alright, trying this method! :)

[ustk]

@Christoph-Hart @d-healey Fixed and pushed