KVR Audio

itoa · Post by **itoa** » Tue Apr 22, 2014 8:19 am

If you look at the spectrum of self-oscillating analog filters you can see that in most cases (all?) resonance peak contains all harmonics.

http://vellocet.com/dsp/analog/SelfOscHarmonics.html

Is this caused by asymmetric saturation or DC added somewhere? What are the most common curves then?

neotec · Post by **neotec** » Tue Apr 22, 2014 8:42 am

Nearly all analogue filters contain non-linearities which get pretty strong when the filter is overdriven or goes into self-oscillation (which also saturates the filter and therefore get's distorted).

The rest depends on the circuit, components used, topology, ... a circuit modeller might help here.

itoa · Post by **itoa** » Tue Apr 22, 2014 9:52 am

Yes I know, but almost in every publication this nonlinearity is a symmetrical sigmoid

, hence my question. Where do even harmonics come from?

If you look sh-101 or ms-20 filter.. this response is almost a filtered saw...

antto · Post by **antto** » Tue Apr 22, 2014 1:19 pm

i have the feeling that the plot is in log scale while the frequencies printed below are linear

if the fundamental freq is 2kHz, then going up an octave each time you get 4, 8, 16..
or am i completely wrong?

itoa · Post by **itoa** » Tue Apr 22, 2014 1:45 pm

No it's right... Harmonics goes 1,2,3,4,5 ...

I tested my MS-20, MFB Dominion and Waldorf pulse.. all analog filters I met so far produce all harmonics.. So seems symmetrical saturation like tanh is the biggest myth here

2DaT · Post by **2DaT** » Tue Apr 22, 2014 2:36 pm

DIstortion within filter is different to external distortion. Even simple "distort the states" non-linearity produces all harmonics. ( as i remember

)

Edit: symmetric "distort the states" do not produce even harmonics.

itoa · Post by **itoa** » Tue Apr 22, 2014 2:54 pm

I see the following possible reasons:
1. components characteristic curve is asymmetrical - the question how and where?
2. a dc offset is added somewhere..
3. (unintended) control modulation?
4. These measurements (and mine) are wrong, because signal is distorted after the filter

I did some tests with my ladder filter model, to get close enough to e.g. sh101 filter I had to use quite sharp asymmetric saturations at several places, but this still generated less harmonics than the original.

Still no answer

anybody there?

camsr · Post by **camsr** » Wed Apr 23, 2014 9:17 am

Filters are extremely complicated. I like to think of them as the work of genius, probably only because I partially understand them.

1. components characteristic curve is asymmetrical

This is not the only case that can create even harmonics via waveshaping. If the signal itself is asymmetric, then symmetric waveshaping will create even harmonics.

aciddose · Post by **aciddose** » Wed Apr 23, 2014 10:35 am

Slight variations in transistors in differential stages account for a bit of the effect, asymmetric buffers account for the rest.

The asymmetry is typically 10% or greater. To get the effect of the ms-20 (korg35) you need at least 50% to 80%, plus many different asymmetrical stages. The opamp is offset with the diode clamp, clipping off the feedback from one side of the waveform. The transistors interact much like in the tb-303 or minimoog ladder filters such that the frequency is dependent upon both input signal and resonance.

The idea of using a plain tanh function is bullshit. The tanh function approximates the effect as a result of the exponential (diode function) base to emitter current transfer curve in a perfectly symmetrical differential amplifier. However this is never the case in long-tailed-pairs less than $50 - in other words pretty much all of them.

Extremely precise laser trimming is used in addition to measurement and matching of components to get accuracies anywhere near 1%.

itoa · Post by **itoa** » Wed Apr 23, 2014 12:24 pm

Thats it! Thanks a lot!

I realised with asymmetric saturation and HPF in feedback loop I get some DC at the filter output..

itoa · Post by **itoa** » Wed Apr 23, 2014 12:30 pm

I noticed that cutoff/resonance modulation depending on internal state does the job better than using wave shapers. I need to figure out the proper function yet.

aciddose · Post by **aciddose** » Wed Apr 23, 2014 8:28 pm

https://soundcloud.com/aciddose/korg35-noaudiopathclamp

Actually the cutoff / resonance modulation is a result of the structure of the filter, there is absolutely no modulation going on directly.

This is in software, entirely done using a configuration of accurate transistor/diode functions. Main difference is the signal path diode clamp isn't here, this is as if you moved the diode clamp to the feedback.

Another huge difference is I'm using a 4-pole filter for this while the ms-20 is a 2-pole.

Note that when I say transistor/diode, i mean functions that ARE NOT TANH.

I don't mean there is any ridiculous complicated spice simulation going on or anything like that. This is absolutely naive, simple c code. Nothing magic happening what-so-ever.

All you need to do is put the correct structure in with the correct functions applied and boom, the exact identical result will come out.

For comparison, this is the actual hardware.

https://soundcloud.com/aciddose/ressy-pulse

Main difference here is of course the res is slightly lower and the filter is 12db rather than 24db. So you should immediately hear the timbre difference.

itoa · Post by **itoa** » Thu Apr 24, 2014 5:38 am

Thanks again,

As for shaping function... as I see diode clamp is almost hard-clipper.

itoa · Post by **itoa** » Thu Apr 24, 2014 7:12 am

Here is my latest MS-20 Sallen-key emulation test:

http://www.mixcloud.com/giku/ms20-filte ... =cloudcast

I used a single asymmetric waveshaper in the feedback loop, tuned 'by ear'.. Resonance peak harmonic structure is identical to that produced by my ms-20 mini.

camsr · Post by **camsr** » Thu Apr 24, 2014 9:25 am

Sounds like it's ready for techno (i like techno)

Analog self oscillating filters - even harmonics