Lmms: Playing a sample or instrument with LS Filter's cutoff frequency near max mutes the project's audio.

[Sawuare]

The project's audio can be unmuted by doing any of these:
• Reloading the project.
• Reloading the effect.
• Disabling the effect.

[zonkmachine]

Confirmed, NaN...

[zonkmachine]

@Sawuare I assigned this to me but maybe you wan't to have a go at fixing this?

[PhysSong]

How can I reproduce that? Could you please provide steps to reproduce?

[Sawuare]

How can I reproduce that? Could you please provide steps to reproduce?

1. Load a sample in a sample track or load an instrument.
2. Add the LS Filter effect to what you loaded in the step above. Increase the cutoff frequency to max.
3. Play the sample or play a note.

[PhysSong]

@Sawuare I just reproduced that right before you post the comment. :smile:

[Sawuare]

@Sawuare I assigned this to me but maybe you wan't to have a go at fixing this?

@zonkmachine Sure, but how?

[zonkmachine]

Sure, but how?

We have a super special debug flag when compiling but it's only for Linux and if I remember it correctly you are on Windows, no?

[PhysSong]

https://github.com/LMMS/lmms/blob/68c85c8e0579827994759573698c5b27f9c71832/plugins/LadspaEffect/swh/util/biquad.h#L105-L118

Here alpha blows up for high frequencies, because omega approaches to π and sn to 0. Here a2 is set to NaN since it is zero times infinity.
Possible workaround is: f->a2 = 1 - a0r * 2.0;

[Sawuare]

We have a super special debug flag when compiling but it's only for Linux and if I remember it correctly you are on Windows, no?

Yes, I'm on Windows. Is there an alternative?

[zonkmachine]

I got this with the fpe flag. It happened when I tried to open the pluggin controls. In my book that isn't a bad call to powf. If I enter -0,00401606411^0,36787945 in a calculator I get −0,131367367 ...

Program received signal SIGFPE, Arithmetic exception.
0x00007ffff459893f in __ieee754_powf (x=x@entry=-0,00401606411, y=y@entry=0,36787945) at ../sysdeps/ieee754/flt-32/e_powf.c:125
125 ../sysdeps/ieee754/flt-32/e_powf.c: No such file or directory.
(gdb) bt
#0  0x00007ffff459893f in __ieee754_powf (x=x@entry=-0,00401606411, y=y@entry=0,36787945) at ../sysdeps/ieee754/flt-32/e_powf.c:125
#1  0x00007ffff459d1db in __powf (x=-0,00401606411, y=0,36787945) at w_powf.c:27
#2  0x00000000004f08b7 in linearToLogScale (min=88,2000046, max=22050, value=0) at /home/zonkmachine/builds/lmms/lmms/include/lmms_math.h:223
#3  0x00000000004f1945 in AutomatableModel::inverseScaledValue (this=0x19f5d30, value=0)
    at /home/zonkmachine/builds/lmms/lmms/src/core/AutomatableModel.cpp:272
#4  0x000000000061818d in Knob::drawKnob (this=0x18dcf00, _p=0x7fffffffb430) at /home/zonkmachine/builds/lmms/lmms/src/gui/widgets/Knob.cpp:445
#5  0x000000000061957f in Knob::paintEvent (this=0x18dcf00, _me=0x7fffffffb810) at /home/zonkmachine/builds/lmms/lmms/src/gui/widgets/Knob.cpp:685
#6  0x00007ffff7122300 in QWidget::event (this=0x18dcf00, event=0x7fffffffb810) at kernel/qwidget.cpp:8533

[zonkmachine]

Yes, I'm on Windows. Is there an alternative?

I don't know how it works on Windows unfortunately. @PhysSong Is it possible to invoke the gnu debugger, gdb on Windows?

[PhysSong]

Yes. You can run gdb on Windows. However, it will provide less information than Linux.

[tresf]

@PhysSong Is it possible to invoke the gnu debugger, gdb on Windows?

Last we tried this, msys2 was broken with the debug flags. https://github.com/LMMS/lmms/pull/2205#issuecomment-126917960

[musikBear]

Increase the cutoff frequency to max.

That is the reason. It will produce a sound with absurd low freq. If you have ambient time, like several hours, the wave-form will dead slowly normalize and sound is re-established
Look at the monitor-display, if im right, a horizontal line will be glued to the bottom of the display
Could actually be interesting to add a graphic-EQ or SPAN and see what they display in this situation
It is not the only filter-plugin that can cause this

[zonkmachine]

if im right

Probably not. See the explanation above.

[PhysSong]

@gi0e5b06 No, you're doing definitely wrong math there. That is a little bit long to explain here.
Anyway, the real problem is alpha, which blows up toward infinity. In strict sense, infinity isn't a number. It means it is larger than any number. However, we usually define 1 over infinity to be 0.
And there is a clear fact, 0 * infinity is not defined. That's what we call NaN.

[PhysSong]

If I enter -0,00401606411^0,36787945 in a calculator I get −0,131367367 ...

@zonkmachine You did a wrong calculation. -a^b is -(a^b)(here ^ denotes power, not the xor operator in C++), not (-a)^b. Such value can be defined using some mathematical techniques, but it never can be a real number. Moreover. there are some ambiguities with that, too.
So (-0,00401606411)^(0,36787945) doesn't make sense in powf() and causes an arithmetic exception. That's a separate issue, I guess.

Edit: //! @brief Scales value from logarithmic to linear. Value should be in min-max range.
It says everything.

[zonkmachine]

wrong calculation. -a^b is -(a^b)

Precisely that error. Doing it again I get an imaginary part. And yes, we're talking floating point operations.

[zonkmachine]

Is there an alternative?

Given how this issue is developing? A friendly advice. Stay away from it. You may not come out of it in one piece.

[PhysSong]

@gi0e5b06 The problem is:
abs(sn)<10E-10 looks harmless, but it is satisfied only if the frequency equals to the nyquist frequency.
abs(M_LN2 / 2.0 * bw * omega / sn)<0.05 is definitely wrong, since omega / sn is always larger than 1, bw is between 1.0(resonance 0.0) and 0.1(resonance 1.0), M_LN2 / 2.0 is 0.34657359 ... , so it may cause issues.

[PhysSong]

Unlike most of SWH filters, LS filter can have cutoff near the nyquist frequency(e.g. 0.5 * sampling rate). The usual value is 0.49 or 0.45(at least in other SWH plugins) times the sampling rate.
https://github.com/LMMS/lmms/blob/46ef3c6a11e9713c24aa3e9b01c658b2941a58d6/plugins/LadspaEffect/swh/ls_filter_1908.c#L273

It is the same for upstream. @swh Is there any special reason for that?