Circuit modeled filter, how to?

[hifiboom]

sounds nice.

is strobe some sort of demo, or will the final synth be called strobe also?

[andy_FX]

is strobe some sort of demo, or will the final synth be called strobe also?

Strobe is the first synth in a pack of three that is due to be released this year from FXpansion. The second synth is a paraphonic string synth, and the third is an audio rate modulation synth. There will also be tool to load three synths at a time and layer them and apply effects to them and do polyphonic modulation between the sythns ie: mod env on synth1 can modulation synth2.

[aciddose]

"so what you are saying, in effect, is that your methods aren't good enough."

yes of course

the main limiting factor for modeling remains cpu power. i was mentioning before "thousands of dollars for a system to run a little titty envelope follower ..."

when i said that, i was absolutely serious. to get an accurate model of an envelope follower we need all four cores in our most powerful quad-core processors today. so, our only option is to do macro-modeling and get the "general behaviour" of circuits rather than the exact behaviours.

i believe we can produce much more advanced macro-models than we have so far in music/dsp, as the work done in the communications and scientific sectors has been much greater than here. i'm certain that you can achieve perfect (to the ear) results using macro-modeling if you continue to refine the models and equations you're applying. one important thing to remember though is something i think you haven't really agreed with so far; the macro-models if written correctly do have the ability to produce 100% accurate output. there is no reason that an electronic simulation should be more accurate, and in fact most simulations lack very important properties/behaviours which for example means certain unstable amplifier designs will not oscillate, there are many more types of circuits which are not modeled accurately.

in spice for example, however, we can build more accurate models by applying reactance to different components. we can accurately model transistors and diodes via a combination of reactive elements and current sources. many of the current mirrors in opamp spice models for example are modeled with non-linear current sources for this reason. (using the real circuit, made from transistors in the spice simulation is less accurate!)

[andy_FX]

the main limiting factor for modeling remains cpu power. i was mentioning before "thousands of dollars for a system to run a little titty envelope follower ..."

Well people pay over USD$4000 for a single compressor: GSerisComp

And I'm not sure many of them could tell the difference between the model I have done, and the real deal. I didn't model the transformer distortion and some of the other slight non-linearities in present in the log amps since I figure the real thing is meant to compress and entire mix cleanly. For something that requires more character modeling these other parts would be good.

So you can buy a computer for around USD$1000 and run 20 of my models on one at a time. Sounds like a good deal to me.

when i said that, i was absolutely serious. to get an accurate model of an envelope follower we need all four cores in our most powerful quad-core processors today. so, our only option is to do macro-modeling and get the "general behaviour" of circuits rather than the exact behaviours.

I'm not sure there would be much point in modeling like that, I've always suggested to make simplifications and optimisations to your model to allow them to run in real time without too much loss of detail.

I'm not sure what you mean by macro modeling, but if it is done by defining approximations to what is going on in a circuit then that is already what spice and qucs do. In fact you can try out your own custom equations in qucs at run time. Here is a link to a poster explaining how to do it:

http://www.mos-ak.org/munich/posters/P0 ... rinson.pdf

In this way you can easily graph things before implementing to save a little time with your simplification of more complex circuit models down to something easier to compute.

i believe we can produce much more advanced macro-models than we have so far in music/dsp, as the work done in the communications and scientific sectors has been much greater than here. i'm certain that you can achieve perfect (to the ear) results using macro-modeling if you continue to refine the models and equations you're applying. one important thing to remember though is something i think you haven't really agreed with so far; the macro-models if written correctly do have the ability to produce 100% accurate output.

Well if I knew exactly what you mean by a macro-model perhaps I can say something about it. All I know about are all the model types used in qucs and spice which are: physical, empirical, and tabular based models and everywhere an anything in between as combinations to get the job done as described here: http://en.wikipedia.org/wiki/Transistor_models

I use all of them in whichever combinations work best, and I make simplifications to great models that other people work out for me and I just have to read about them in papers and pick out the bits I want. I listen to the models and make the decision about tradeoff between cpu and quality. I'm all for developing new models to more efficiently and accurately describe circuits. I always have been.

there is no reason that an electronic simulation should be more accurate, and in fact most simulations lack very important properties/behaviours which for example means certain unstable amplifier designs will not oscillate, there are many more types of circuits which are not modeled accurately.

Still not clear what the you mean by a macro model vs an electronic simulation. Can you please let me know the difference? People that write spice and qucs models use every trick in the book to get the result they need, linear approximations, exponential things, low order polynomials, if statements, a bit of resistance, some capacitance across different paths - anything.

in spice for example, however, we can build more accurate models by applying reactance to different components. we can accurately model transistors and diodes via a combination of reactive elements and current sources. many of the current mirrors in opamp spice models for example are modeled with non-linear current sources for this reason. (using the real circuit, made from transistors in the spice simulation is less accurate!)

Yes I agree. So whos side are you arguing on now? Have you been reading too much of what I said and suddenly come about face? Are you now saying that we can improve existing spice and qucs models to get more bang for buck when it comes to cpu usage? I am certainly all for making better models. I have been all the time, but first I would have thought it easier for people starting out to learn the existing models since since there are loads of papers describing them. Do you think it is easier for people to jump in at the deep end and start creating there own spice models without going through and learning the existing models for things first, or at the very least trying them out and graphing them?

[aciddose]

"I'm not sure what you mean by macro modeling"

a macro-model is a model where you take only the behaviour of a circuit unit or module and model that behaviour, not the circuit itself.

for example, we could model a whole vca, however i believe most dsp code uses a macro model for that: multiplication.

now, how does this help if you wanted to model a minimoog vca? well, the minimoog vca still preforms multiplication, it just does additional things. rather than modeling in a bottom-up method, we can go top-down, meta-modeling might be a more accurate description?

we can apply the multiplication in combination with the required integrators, noise generation and so on. the end behaviours that we get out of this model will match perfectly with the real-world circuit, only we will have avoided using any electronics simulation in our process. we need only to take into account the influence the electronic components have on the resulting over-all behaviour of the model, not the components themselves.

i'm definitely not saying people should be using spice or any other simulator to do this work, apart from using it to get a general idea as to what behaviours a circuit will exhibit. you'll still need to be very skilled to get accurate simulation results though, so bench-top testing is really going to be required for the most accurate efforts.

i've found over time that carefully modeling elements is a complete waste of my time. when you get down to the very root of the behaviours you're trying to get, you find they are always emerging from a set of very basic functions. you should find that all electronic circuits can be easily modeled at a much higher level and that systems can be abstracted to accomplish more with less.

we could for example implement a vastly complex electronic simulation to produce a model of an oscillator. ultimately though, all emergent behaviours in this system will factor into the same set of basic functions. we will learn over time that all behaviours can be most accurately modeled via direct application of these basic functions. for example, although i have spent very little time attempting oscillator models (no time?) as i find it much easier to use real electronics; rather than modeling a full power supply and all the effects it will have on the electronics, we can simply apply brown noise and a mains signal to our reset point and thus frequency. i can not speculate more about oscillators specifically because i have not experimented much with them. i prefer my digital oscillators clean, and what i value most is their speed.

you mention that people pay $4000 for a compressor.. well, this is the market you're trying to reach, i suppose. my 'market' will pay $0 for such a thing. this should explain many of our differences of opinion.

[andy_FX]

"I'm not sure what you mean by macro modeling"

a macro-model is a model where you take only the behaviour of a circuit unit or module and model that behaviour, not the circuit itself.

for example, we could model a whole vca, however i believe most dsp code uses a macro model for that: multiplication.

Well I do this all the time. In the 1176 model I did the pre amp vca remained in it's linear area as near as made any difference to me so I replaced it with a voltage controlled voltage source in qucs (gain). I'm all for making simplifications to circuits to improve performance.

i've found over time that carefully modeling elements is a complete waste of my time. when you get down to the very root of the behaviours you're trying to get, you find they are always emerging from a set of very basic functions. you should find that all electronic circuits can be easily modeled at a much higher level and that systems can be abstracted to accomplish more with less.

Well here is a nice little schematic: 1176 schematic

I have come up with a simplification that still includes the auto release and the < 1 sample at 44100 feedback time of the entire circuit. In the end the hardest function to calculate is an exp which I already have an approximation for, the jfet is only a few x^2 type terms and the rest is linear. So it sounds like I've already done what you are talking about, which is a natural outcome of doing a model of the circuit and then simplifying and optimising your model. I never use the full models present in qucs or spice for real time use, but I do use tables generated from the full models every now and then.

you mention that people pay $4000 for a compressor.. well, this is the market you're trying to reach, i suppose. my 'market' will pay $0 for such a thing. this should explain many of our differences of opinion.

Well your market has probably paid $$ to listen to such a thing as it's used on a stupid number of recordings they they have probably bought, it's a fantastic sounding compressor. I wouldn't say that we have a difference of opinion, I have been saying the same thing all along and you have actually agreed to most of it eventually once I could get you past your persistent miss-interpretations.

[andy_FX]

we could for example implement a vastly complex electronic simulation to produce a model of an oscillator. ultimately though, all emergent behaviours in this system will factor into the same set of basic functions.

It sounds like you expect everything to converge to the aryan race of circuit modeling purity and all else will be expelled!

Well in my opinion diversity is good and I'll leave it up to others to decide which models and which simplifications they so choose to use to get the job they want done done. If there are any questions I'll happily direct people to reading material so they can work it out themselves and learn in the process.

for example, although i have spent very little time attempting oscillator models (no time?) as i find it much easier to use real electronics...

You are quite right to prefer "real electronics". So how do you get your oscillator circuit soldered onto your motherboard?

rather than modeling a full power supply and all the effects it will have on the electronics, we can simply apply brown noise and a mains signal to our reset point and thus frequency.

I agree completely. As I've said all along I think it's essential to make simplifications to models to allow them to run with however much cpu you decide to use.

You just said:

i prefer my digital oscillators clean, and what i value most is their speed.

But previously you said:

the asymmetric slewing combined with other factors give an analog pwm bass such life compared to what software can produce

So which is it? You have self admittedly "spent very little time attempting oscillator models", so are you saying you prefer clean oscillators just until you can work out something better? If you prefer clean oscillators then you must like the Strobe examples in which I have not added noise to the pitch or pwm, but I thought you said that, oh let me get the quote... "spore sounds completely lifeless and 'static'", which I was taking as a negative until you pointed out you like clean oscillators just now. Curiouser and curiouser!

[duncanparsons]

I wouldn't say that we have a difference of opinion, I have been saying the same thing all along and you have actually agreed to most of it eventually once I could get you past your persistent miss-interpretations.

This is what I've thought. It's like two guys standing either side of a door, one yelling 'it opens inwards', and the other yelling 'it opens outwards'... same door, same result, different perspectives.

Still good thread - I've learnt stuff from you both, and will be printing it all out for extended reading when the storm dies down.

thx all for the contributions

DSP

[suthnear]

ditto...

[suthnear]

p.s. (and sorry for the OT) andy, some time ago you posted an excerpt from deadmau5's 'hey baby'. At the time you said it was from a project you were working on - was that strobe?

[Ivan_C]

It sounds like you expect everything to converge to the aryan race of circuit modeling purity and all else will be expelled!

Seriously, there is a point (Godwin) where you really have to stop. I have read the full topic and I can't understand why you are both so aggressive, and where a fundamental problem is. I have said nothing until now because the talking is well documented and interesting, but I really don't like this quote. You do almost the same thing as a developper, but you have a different point of view about the VERY interesting debate analog vs digital. f**k.

[andy_FX]

I wouldn't say that we have a difference of opinion, I have been saying the same thing all along and you have actually agreed to most of it eventually once I could get you past your persistent miss-interpretations.

This is what I've thought. It's like two guys standing either side of a door, one yelling 'it opens inwards', and the other yelling 'it opens outwards'... same door, same result, different perspectives.

Still good thread - I've learnt stuff from you both, and will be printing it all out for extended reading when the storm dies down.

thx all for the contributions

DSP

Good point. There is one difference, and that is I have never said that anything aciddose has suggested is wrong or that using macro models doesn't work. I have never said that the points aciddose are making in terms of optimisations aren't good. I have even agreed with some of them.

I have always said that it's up to the person involved in the model to make such judgment calls as to particular optimisations they want, but for those people that don't know circuits that well having something as a target to make simplifications to check against is useful, which is where circuit simulators can come in useful.

Aciddose has in called me all sorts of names, and insists that I'm wrong all the time by mis-interpreting what I've been saying.

[andy_FX]

... I have read the full topic and I can't understand why you are both so aggressive, and where a fundamental problem is. I have said nothing until now because the talking is well documented and interesting, but I really don't like this quote. ...

I apologize if you didn't like the reference to purity, my intent was not to offend but to highlight intolerance.

[andy_FX]

p.s. (and sorry for the OT) andy, some time ago you posted an excerpt from deadmau5's 'hey baby'. At the time you said it was from a project you were working on - was that strobe?

Yes, deadmau5 and his other outfits use Strobe, as well as Orca+ and a compressor I did to help get things pumping in their tunes. They also use loads of other kit.

[aciddose]

"It sounds like you expect everything to converge to the aryan race of circuit modeling purity and all else will be expelled!"

the difference is in the abstract thinking - we converge to 'human' form, not aryan. the aryan form is in fact the corrupt and over-complicated mess with many additional factors involved. 'human' applies to all of us, equally, and perfectly.