Analog filters, that sound "physical"

[itoa]

We can often hear that certain filter sounds "physical", like real resonating object/instrument.

In most cases the more we "push" analog filter the lower resonance frequency is. This is caused by internal saturation. What is interesting, this phenomenon is opposite in real world. Materials are more or less stiff and this stiffness causes higher pitched resonances for stronger excitations. (like membrane, string etc.)

This is just an observation.. that surprised me

Of course I can simulate this behaviour by using a rising curve instead of saturation, but..
do such a filters exist in synths?

[camsr]

A lowpass filter is by definition a damper, as such it limits slew rate.

[itoa]

Of course, like membranes and strings . I'm talking on large signal nonlinear behaviours or real circuits.

[antto]

In most cases the more we "push" analog filter the lower resonance frequency is. This is caused by internal saturation.

if you mean to push the resonance beyond self-oscillation - well, i have one analog filter here (x0xb0x) and the resonant frequency remains the same no matter how much i "push" the feedback
feedback level for self-oscillation on that filter is 17.0, i can push it to 79.9

what analog filters have this property? (i'm curious)

[itoa]

I observed this in my nonlinear LPF models and the following real synths:
- MFB dominion X sed: all 3 LPF ladder filters (separate circuits).
- MS 20 - sallen key filter

Mind model
If you use saturation in integrator like buf += c*tanh(in-buf) or buf += c*(tanh(in) - tanh(buf)) -- the difference is smaller than in linear case. Isn't it? So it's the same like lowering c (cutoff).

Diode ladder used in 303 is far more complicated... many things are coupled.. it's like coefs are modulated by the signal, so it's a completely different disco

[mystran]

Please, don't take this the wrong way.. but you really should stop thinking about "analog filters" and start thinking/learning about what is going on in particular circuits. For what it's worth, there's no such thing as "analog filter behavior" as you can build a circuit with pretty much whatever behavior you want (well, within some practical limits). Various existing filter circuits behave quite differently with respect to each other and while there can be some common themes, trying to think in "general" is practically useless.

More specifically, tanh() is just an approximation that happens to describe an ideal matched long-tailed pair and some similar configurations of BJTs. How and where these configurations happen in a particular circuit (and what their relatively level is with respect to other stuff) can have drastic effects (often more so than the exact shape of the curve) on how a given circuit behaves when driven.

[itoa]

Hi Mystran and kudos again

but I am waiting for someone who will refer to the original subject.

I know that many things can be modelled in analog and digital domain, I know that "LPF by definition is a damper" and various existing filters behave differently. AFAIK tanh is quite precise when it comes to approximation of transistor saturation (but this is not important here). Maybe I used wrong terms...

So lets start again: for (all/many) LPF resonant filters (used in available synths) it's cutoff freq gets lower for large excitation and resonance. It's the opposite to the real world phenomenons. Having in mind a lot of mechanical / electrical analogies I found it quite interesting. Maybe some synths (drum synths?) are equipped with filters that work this way. Maybe it's desirable in some cases. I just wanted to know.

I like looking at a problem from different perspectives, like most filters can be expressed as vibrating mass-springs systems. Use nonlinear springs with e.g. x^2 restoration force and you will get a squelchy resonance like 303. There are viscoelastic joints, nonlinear dampers etc.. This is the unity! Maybe we should learn physics instead of electronics?

[Zaphod (giancarlo)]

we measured (and reproduced) a lot of harmonic distortion in many analog filters, if it answers the question. Yes the effect is much more balanced in that way.
Harmonic distortions often is generated even by simple components (transistors), very often this kind of non-linearities is a bit lost when you simplify and model them optimizing for speed. This is my guess analyzing results (sampling software and comparing it with sampled hardware)

[camsr]

I think you are confused a little. What real world phenomenon mimics a low pass filter AND raises it's resonance frequency under large load?

[Urs]

I think you are confused a little. What real world phenomenon mimics a low pass filter AND raises it's resonance frequency under large load?

If the input/feedback contains a DC component…?

Most of the time I see the filter frequency go down when resonance is turned past self oscillation, but for instance in our Doepfer equivalent of the EMS Synthi filter frequency goes up.

[itoa]

The answer: every
A resonant filter is obvious equivalent to a physical vibrating object being struck. e.g. RLC filter is equivalent of mass-spring system. (force = voltage, velocity = current, damper = resistor, mass = inductor etc..). This works as well for pneumatic and other systems. (isn't this fascinating ?)
some random examples:
- every resonating object is a nonlinear, resonating filter, that has it's frequency, phase and impulse response
- vocal cords is an oscillator (exciter), vocal tract is a resonating (multipole) filter.
- TR-808 kick circuit is a bp resonating filter that is excited by impulse and act like a membrane.

If you hit a membrane, string, stiff spring hard it's initial resonance frequency will be higher and you notice a fast pitch drop. The restoration force is nonlinear, but this nonlinearity is the opposite to these I found in most analog filters (I'm not sure on bridged-t and some diode ladders).

[Urs]

But still, only one ot of 50+ analogue filters we have around goes "up" in pitch with higher load/resonance. Every single other one of them drops in filter frequency.

[itoa]

URS: thank you! This is what I wanted to know

I know many of us like to be trapped into known analog circuits (dogmatic prison , but it's the opposite: they were modelled after the real phenomenons to mimic real instruments...

URS: I modelled a naive filter with rising waveshaping curve in integrators.
http://www.mixcloud.com/giku/filter-test/

You can hear a small pitch drop when the filter gets impulse. And this mimics the physical behaviour.

camsr: http://multimechatronics.com/images/upl ... nalogy.pdf

[camsr]

itoa, thanks for the link, very good reading.

Resonances move up in frequency in a physical body when stiffness is increased. For example, here is a sound transmission, example:


And in relation to your question:

http://www.neurophys.wisc.edu/h&b/textbook/mid_ear.html

The electrical analogy of increased stiffness would be a limiting of charge.

[itoa]

camsr: great so you answered your own question.. these are functions of different masses and stiffness factors.

the simplest explanation is that material restoration force gets greater for larger vibrations. If you push a membrane you increase it's tension, at the beginning this force is almost linear but (theoretically) grows to infinity soon - you can't push it more. The bigger tension, the higher vibrational frequency (x^2 relationship).
In metal objects this makes higher frequencies travel faster.

Well...
URS found only one filter that behaves this way. The other filters do the opposite (analogy: the restoration force gets smaller), that is quite a nonsense (or not ?

Do designers don't care? Isnt't this aspect important? I think this may be crucial for drum like sounds.