Login / Register 0 items | $0.00 New @ KVR
itoa
KVRist
 
478 posts since 3 Sep, 2009, from Poland

Postby itoa; Tue Jan 09, 2018 11:52 am What makes SEM filter so special :)

I modelled some filters, basing on their circuits (more or less), using TPT and Newton-Raphson metod with oversampling. The models sound not exactly the same but share a lot of character with the original.

Well.. until I sat on SEM filter :) The only significant nonlinearity I see on the scheme are diodes in damping path. I've tried to make everything with them, but my model doesn't even touch the SEM character. (especially this beautifully buzzing high-end and filter modulation).

I know, I'm asking about black art, but any tip? Where should I go?
giq
itoa
KVRist
 
478 posts since 3 Sep, 2009, from Poland

Postby itoa; Tue Jan 09, 2018 11:54 am Re: What makes SEM filter so special :)

(deleted)
giq
JCJR
KVRAF
 
2308 posts since 17 Apr, 2005, from S.E. TN

Postby JCJR; Tue Jan 09, 2018 1:19 pm Re: What makes SEM filter so special :)

A rather obvious perhaps uninformative reply-- Maybe you will get better-- Maybe some of the following is wrong somehow, dunno--

It is a fairly "standard" second-order state variable filter using CA3080 Operational Transconductance Amps as the tuning elements. The OTA's occupying the same place in a state variable circuit one would find the two frequency-setting resistors in a fixed-frequency state variable filter.

The FETs appear wired as "nominally unity gain" voltage followers, probably used to present a very high input impedance and lower output impedance, of the variable-current-drive of the OTA's (acting as "variable resistors") driving the frequency setting capacitors C20 and C21.

Each pair of OTA+Cap makes a first-order RC lowpass, and such circuits work best when "very lightly loaded". The high input impedance of the FETs accomplishes this. When "heavier loaded" with lower impedance buffers, the performance of the RC becomes "less ideal" and therefore the performance of the state variable filter would also be "less ideal".

OTA pins 2 and 3 are differential voltage input similar to opamp input stages. OTA pin 5 is the iABC current-- Essentially a path thru some transistors into the negative power supply pin of the chip. If you don't draw any current out of that iABC pin, then the OTA acts like a very high-value resistor, and as you draw more current out of the iABC pin the OTA turns on more and more, acting like progressively lower-value resistor. That pin cares about current not voltage. You could decide you want to draw maybe 0.25 ma or whatever, and pick out the right value of resistor and hook iABC thru the resistor to ground (pulling current from negative power supply pin thru the resistor into ground). Or use another value of resistor and hook up between iABC and the positive power supply rail. Or whatever, so long as you draw the proper amount of current to turn on the OTA the exact amount you want. Typically in VCA's and VCF's, transistors were used so that you can feed control voltage into a resistor network, gradually turning on the intermediate transistor, and the transistor is responsible for pulling the desired current out of iABC to tune the filter to the desired frequency.

Most OTAs including CA3080 are voltage in and current out, with the gain also controlled by iABC current as described above. Some of the later OTAs had a wider linear input voltage range, but CA3080 would distort if driven very hard. You could double-check, but my flawed memory remembers CA3080 "pretty clean" from a max of 10 millivolt up to maybe 100 millivolt input. Depending on what is deemed "clean enough" you could hit them harder. So far as I recall OTA input distortion was gentler-sounding than most solid-state clipping. CA3080 was also (in my opinion) a fairly noisy part, so you could "guarantee low distortion" with an input so low as to be swamped in noise, or you could "guarantee low noise" by hitting the input hard enough to make maybe 5 or 10 percent distortion, but then the signal is loud enough to swamp the noise.

Unless you can find a document listing typical AC signal levels at the OTA inputs, you would probably need to get an SEM and oscilloscope together in the same room, probe circuit points, and find out how hard Mr Oberheim was hitting those CA3080s. And knowing typical OTA input drive on the SEM filter, you might guestimate how much distortion is coming out of the CA3080s. I'm pretty sure there are CA3080 spec sheets with charts of input voltage versus distortion.

I doubt that the FET followers would be a significant source of nonlinearity but it is wild guessing only.

SEM was pretty early tech. The opamps listed 301 and 741 were good early-tech parts with more noise and distortion, lower slew-rate, etc than you would probably want to put in a newer design. Just sayin, rather than the discrete FETs, a designer a few years later might have used a simpler circuit with higher-quality FET-input opamps, so the discrete FETs are not even necessary in the circuit.

The only other thing I notice in my ignorance-- The "most typical" state variable filter would put the tuning capacitors in the feedback loop of opamps, making them "true integrators" rather than "well-buffered first-order lowpass filters". For instance, you would maybe have the OTA output directly connected to one of the opamp inputs and the capacitor in the opamp feedback loop.

OTOH possibly there is something about the topology Mr Oberheim chose, which would have made a "better tracking" filter than something more "standard looking". Or as mentioned above, maybe Mr Oberheim would have done it different if he could have got inexpensive high-quality FET input opamps when he designed the circuit? Just wild guessing.
mystran
KVRAF
 
4979 posts since 11 Feb, 2006, from Helsinki, Finland

Postby mystran; Fri Jan 12, 2018 1:59 pm Re: What makes SEM filter so special :)

itoa wrote:Well.. until I sat on SEM filter :) The only significant nonlinearity I see on the scheme are diodes in damping path. I've tried to make everything with them, but my model doesn't even touch the SEM character. (especially this beautifully buzzing high-end and filter modulation).


Buffers are non-linear and OTA output current is rail-limited; the latter is potentially relevant because non-linearity of the buffers reduces the negative feedback (and in general fairly small feedback non-linearity can cause some funny stuff), which can then drive the actual capacitor voltage against rails (and a model that doesn't limit the OTA output voltage swing will blow up at this point).

Maybe there's more, but that's the most obvious stuff that comes to mind.
Image <- plugins | forum
hugoderwolf
KVRist
 
176 posts since 1 Apr, 2009, from Bochum, Germany

Postby hugoderwolf; Fri Jan 12, 2018 2:37 pm Re: What makes SEM filter so special :)

Yeah the SEM is a really funny one. The circuit is much unlike most other filter topologies in that less feedback gives you more resonance. That turns some basic intuitions on the head.

In the end, my conclusion was that this filter isn't meant at all to be driven hard. Mostly it operates at very healthy levels and very linearly. As far as I know it's also hardly possible at all to get it into self-oscillation.

However, you most definitely need the OTAs nonlinear. The diodes are much less important as far as I remember.
mystran
KVRAF
 
4979 posts since 11 Feb, 2006, from Helsinki, Finland

Postby mystran; Fri Jan 12, 2018 5:53 pm Re: What makes SEM filter so special :)

hugoderwolf wrote:In the end, my conclusion was that this filter isn't meant at all to be driven hard. Mostly it operates at very healthy levels and very linearly. As far as I know it's also hardly possible at all to get it into self-oscillation.


On the other hand, the Wasp filter has essentially the same topology, except this time the "buffers" are not even approximately linear (in any region really) and this results in all kinds of "weird shit" (please apologise my language) whether or not you actually try to drive the thing.
Image <- plugins | forum
Kraku
KVRian
 
1399 posts since 13 Oct, 2003, from Prague, Czech Republic

Postby Kraku; Sat Jan 13, 2018 5:24 am Re: What makes SEM filter so special :)

mystran wrote:On the other hand, the Wasp filter has essentially the same topology, except this time the "buffers" are not even approximately linear (in any region really) and this results in all kinds of "weird shit" (please apologise my language) whether or not you actually try to drive the thing.


Not trying to hijack the thread, but what would be a nice realtime approach to modeling the Wasp filter's buffers? This is something I would be interested in trying out to learn more about our mystic art 8)
mystran
KVRAF
 
4979 posts since 11 Feb, 2006, from Helsinki, Finland

Postby mystran; Sat Jan 13, 2018 9:20 am Re: What makes SEM filter so special :)

Kraku wrote:
mystran wrote:On the other hand, the Wasp filter has essentially the same topology, except this time the "buffers" are not even approximately linear (in any region really) and this results in all kinds of "weird shit" (please apologise my language) whether or not you actually try to drive the thing.


Not trying to hijack the thread, but what would be a nice realtime approach to modeling the Wasp filter's buffers? This is something I would be interested in trying out to learn more about our mystic art 8)


You could probably just approximate the CMOS inverters with some suitable pseudo-sigmoid and solve the same way you'd solve for any other static non-linearity. As far as the general behaviour of the circuit goes, the exact shape is not really even very critical.

That said, you absolutely MUST limit the OTA output to supply for this to work, or it'll blow up faster than you can say "fish."
Image <- plugins | forum
Z1202
KVRian
 
954 posts since 11 Apr, 2002

Postby Z1202; Mon Jan 15, 2018 2:47 am Re: What makes SEM filter so special :)

mystran wrote:(and a model that doesn't limit the OTA output voltage swing will blow up at this point).
Would you care to elaborate a bit? Quite a few years ago I have made a ZDF model of the SVF modeling exclusively the diodes in the damping path (not exactly diodes, the curve was closer to an inverted tanh than to sinh, but essentially the same idea). It didn't seem to blow up. The thing is, as soon as the first integrator's output gets too high, the signal in the damping feedback path gets very large. Of course, it's not a rigorous proof of stability, but as I said, it seemed fully stable.
User avatar
Richard_Synapse
KVRian
 
845 posts since 19 Dec, 2010

Postby Richard_Synapse; Mon Jan 15, 2018 6:50 am Re: What makes SEM filter so special :)

itoa wrote:I modelled some filters, basing on their circuits (more or less), using TPT and Newton-Raphson metod with oversampling. The models sound not exactly the same but share a lot of character with the original.

Well.. until I sat on SEM filter :) The only significant nonlinearity I see on the scheme are diodes in damping path. I've tried to make everything with them, but my model doesn't even touch the SEM character. (especially this beautifully buzzing high-end and filter modulation).

I know, I'm asking about black art, but any tip? Where should I go?


A basic SVF implementation will not give you that character, you will need a better model. At least this is what I noticed when doing SPICE simulations for some SVF variations, it did not result in the same output as a basic SVF.

Perhaps you could try mystran's circuit emulator in the other thread :)

Richard
Synapse Audio Software - www.synapse-audio.com
itoa
KVRist
 
478 posts since 3 Sep, 2009, from Poland

Postby itoa; Mon Jan 15, 2018 7:00 am Re: What makes SEM filter so special :)

Richard_Synapse wrote:
itoa wrote:I modelled some filters, basing on their circuits (more or less), using TPT and Newton-Raphson metod with oversampling. The models sound not exactly the same but share a lot of character with the original.

Well.. until I sat on SEM filter :) The only significant nonlinearity I see on the scheme are diodes in damping path. I've tried to make everything with them, but my model doesn't even touch the SEM character. (especially this beautifully buzzing high-end and filter modulation).

I know, I'm asking about black art, but any tip? Where should I go?


A basic SVF implementation will not give you that character, you will need a better model. At least this is what I noticed when doing SPICE simulations for some SVF variations, it did not result in the same output as a basic SVF.

Perhaps you could try mystran's circuit emulator in the other thread :)

Richard


Looking at this scheme. http://www.synthfool.com/docs/Oberheim/Oberheim_SEM1A/oberheim_sem_1a.gif Its the same well known SVF structure + a few possible nonlinearities (OTAs). Some people say 301A is crucial here. I've tried a lot of them :) nothing moved my model closer to this character.
giq
mystran
KVRAF
 
4979 posts since 11 Feb, 2006, from Helsinki, Finland

Postby mystran; Mon Jan 15, 2018 8:01 am Re: What makes SEM filter so special :)

Z1202 wrote:
mystran wrote:(and a model that doesn't limit the OTA output voltage swing will blow up at this point).
Would you care to elaborate a bit? Quite a few years ago I have made a ZDF model of the SVF modeling exclusively the diodes in the damping path (not exactly diodes, the curve was closer to an inverted tanh than to sinh, but essentially the same idea). It didn't seem to blow up.


It's not the integrator output that blows up. It's the integrator itself. The diode feedback path will only see the buffer output. This won't ever be a problem in any real circuit because the OTA won't push the capacitor voltage beyond the supply rails, but in a simulation you can end up accumulating arbitrary amounts of DC (which won't be cancelled out because the feedback paths don't see it).

If the buffers are completely linear, then this problem doesn't exist and the diode limiting is sufficient.
Image <- plugins | forum
JCJR
KVRAF
 
2308 posts since 17 Apr, 2005, from S.E. TN

Postby JCJR; Mon Jan 15, 2018 9:50 am Re: What makes SEM filter so special :)

Maybe somewhere, someone has poked around with a scope and published typical signal levels at various points of the circuit.

Cursory examination of the schematic set here-- http://www.synfo.nl/servicemanuals/Oberheim/SEM-1A_SCHEMATICS.pdf it appears about everything is powered on typical +/- 15 volt rails. I'm too lazy to look it up but tis doubtful that the 301 could output un-clipped all the way to the rails, and so far as I recall the 741 will miss the rails by at least a couple of volts, somewhat dependent on frequency.

So the max "theoretical" peak-to-peak voltage is 30 volts and perhaps the max clean opamp outputs, just from old memories, probably not even 26 volts peak to peak and maybe as low as 20 or 22 volts peak to peak. The OTA input is attenuated by the 220 ohm / 100220 ohm divider, so that a 30 volt peak to peak input would only present about 66 millivolts, and more realistic maybe 20 volts peak to peak input would only present about 43 millivolts to the OTA input. As best I recall "real clean" CA3080 performance might be as low as 10 mv input, but 43 mv isn't hitting the OTA real hard, though there would be some input differential stage distortion, and as best I recall OTA input distortion was "mellower" than typical transistor hard clipping. Almost "tube-like" sometimes.

Just sayin, if the circuit is seeing only 5 volts P-P or less typical operating levels, distortion MAY be rather negligible in the OTA and FET buffers. The unity gain FET voltage follower will only get real distorted at higher levels, and I wouldn't be surprised to see assymetrical distortion in that case, different nonlinear curves on positive and negative extremes, much as an overdriven class A tube follower.

Regardless what levels Mr Oberheim used in the actual SEM, I suppose there are enough hints there to guestimate the kinds of distortion at various small and large operating levels.

The linear-to-exponential voltage to current converter driving the OTA pulls "about a 15 volt drop" thru a 15 kohm resistor, so the max "flat out" iABC drive (filter wide open) would be in the ballpark of 1 mA, which so far as I recall is about the max recommended highest-gain iABC gain. But in normal operation it may be that the OTA doesn't really see anything nearing a full 1 mA iABC drive, dunno. Maybe it tunes to the top of the audio band using much less than 1 mA iABC. The frequency-setting caps are not very big and the frequency-setting caps are the only load on the OTA outputs (FET input impedance negligibly high) so at high frequencies maybe the OTA could output enough current to easily clip driving into the load of the small cap. At lower frequencies (lower OTA current-gain, lower iABC gain-control current) it would be less likely that the OTA could reach the rails, even driving such tiny capacitors.

For mystran and others who better-understand filter theory, I ask whether any "oddities of sound" in this filter might come from a state variable filter based on two "well buffered, nearly ideal first order lowpass filters". Almost all the "textbook" and commercial state variable filters are based on a pair of true integrators rather than a pair of 1-poles. The DC difference being, that with the pair of 1-poles, a DC or very low frequency input, the lowpass outputs would never exceed the inputlevel * gain. But the "true integrators", the integrator output from DC input, would ramp up (or down) in a perfect straight line until it pegs on the power supply rail.

Of course in a state variable based on a pair of true integrators, the multiple feedbacks assure DC stability of the entire circuit, but each true integrator left to its own devices would act completely different than a "nearly ideal 1 pole lowpass filter". So maybe when you make the state variable filter out of apples rather than oranges, that could affect the transfer function? I don't know enough filter theory to reason that out.

EDIT: Its been years since messing with analog. Possibly with only a capacitor on the OTA output, no load resistor to limit OTA output voltage gain, the OTA driving the cap acts identical to a true integrator. Would be easiest solved with a screwdriver and oscilloscope. :)
mystran
KVRAF
 
4979 posts since 11 Feb, 2006, from Helsinki, Finland

Postby mystran; Mon Jan 15, 2018 10:28 am Re: What makes SEM filter so special :)

JCJR wrote:Of course in a state variable based to a pair of true integrators, the multiple feedbacks assure DC stability of the entire circuit, but each true integrator left to its own devices would act completely different than a "nearly ideal 1 pole lowpass filter". So maybe when you make the state variable filter out of apples rather than oranges, that could affect the transfer function? I don't know enough filter theory to reason that out.


Every SVF is built out of "true integrators" at least as far as the frequency range of interest is concerned. There's no local feedback here, the OTA just tries to dump the desired current (ie. input voltage times CV current) into the cap as long as the cap voltage doesn't get too close to rails (at which point the OTA obviously has to give up).

If the global feedback is not enough to keep things stable, then things won't be stable, which is why saturating buffers (eg. abused CMOS inverters) lead to "weird stuff" and in fact in a real-world circuit, even a seemingly linear one the feedback is never precise enough to completely cancel DC, but it's usually enough to keep the caps stable anyway; they just end up with some fixed DC offsets until the feedback catches up, typically depending on CV and hopefully more or less negligible in the cutoff frequency range of interest.
Image <- plugins | forum
JCJR
KVRAF
 
2308 posts since 17 Apr, 2005, from S.E. TN

Postby JCJR; Mon Jan 15, 2018 1:41 pm Re: What makes SEM filter so special :)

Thanks mystran. Yeah I was being dumb the last two paragraphs before the edit, thinking the OTA driving a cap without a load resistor WOULD NOT act "close enough" to a true integrator. So most likely topology wouldn't explain any "unique filter sound".

I never had a SEM but repaired a few oberheim 4 and 8 voice SEM poly synths, and modified at least one but can't recall what was done. Playing them, was something I thought would be a fun synth to have at home but kinda big and heavy and not very practical for general gigging.

Because the "programmer module" did not remember enough of the SEM knob settings to be quick enough changing patches live, except perhaps in a very controlled concert gig playing a "carved in stone" set list, or maybe just using the thang for synth horn blats all night long.

It was so long ago, but I don't recall much "unusual" about the filter sound, just second order state variable. The distortions would depend on typical operating level. Someone interested enough could analyze the VCO circuit and mixer component values to guestimate signal levels into the filter, unless somewhere those values are already published. For instance it wouldn't be worth figuring out the FET voltage follower positive and negative nonlinearities if the signal never gets close enough to the rails for it to matter.
Next

Moderator: KVR Moderators (Main)

Return to DSP and Plug-in Development