Circuit modeled filter, how to?

[Urs]

Analogue or digital?

http://www.skirize.adbe.org/UnknownVirusArp.mp3

I'd say digital - is there an analogue synth with a digital reverb?

[djsubject]

Analogue or digital?

http://www.skirize.adbe.org/UnknownVirusArp.mp3

I'd say digital - is there an analogue synth with a digital reverb?

Opps,

i did add reverb to the audio file

My bad,

so you was right about the digital reverb

Subz

[djsubject]

well, technically the virus isnt analogue, nor are the juno series. they're hybrids. so it all really depends upon the particular definition of 'analogue' being used. 100% analogue? i can tell you no, it isnt.

how?

well it is sequenced isnt it?

not a virus! (just the arp is from the old Zebra Vs Virus thread)

Subz

[gnjp]

At the end of the day no one will give a shit
what route you took to get there - if it sounds good...

[aciddose]

wait a minute.

has this thread just somehow transformed into a digital vs. analogue thread?

i was under the impression the thread was "modeling a filter: how to?" and that we were arguing about which way is potentially the most effective to do so.

we've also completed our argument coming to this conclusion:

- if you want to attempt to create an exact replica of something, you should probably use methods to model that something as accurately as possible.

- if you're interested in only the attributes of that thing, like the 'sound' of a filter for example, it may not be most effective to attempt a perfect model; some have voiced that they believe the exact models sound best while others disagree.

- something about picasso.

[djsubject]

wait a minute.

has this thread just somehow transformed into a digital vs. analogue thread?

No,

i just like to take things off topic



Subz

[andy_FX]

hm, the "intentionally mismatching" idea sounds a little bit like conspiracy-nut kind of stuff to me. i'm sure a couple companies may have done this (moog perhaps? sounds like his kind of thing) however most of the engineers out there are worried about one thing and one thing only: cost!

I think cost saving definitely caused quite a difference in sound between certain analog sythns because components with such high variation were used.

Andrew

[aciddose]

certain synths, certainly. we come back to the normal distribution of parameters however and in the majority of circuits you should actually find that what isnt trimmed actually averages to a relatively narrow range.

all engineers with experience know that for certain circuits you need not match components perfectly. for example you might think that opamp adders need to be matched absolutely in order to get perfect unity gain - that isnt true though, not at all. if the circuit that you're controlling has a scale trim, like any filter with a expo stage for example, the gain of the input stage becomes irrelevant as the scale trim adjusts that gain. you might also think that major differences might be encountered due to input scales not matching, for example. it isnt going to matter much when you've got 5% error on an envelope input however as you can compensate for that error by allowing a 105% range. for cases where the error is negative, you'll get 100%. for cases where it is positive, you'll get 110%. the extra 10% isnt going to matter generally, say for a cutoff control. you simply have to adjust the input control in a slightly different way.

as far as i know, nobody has ever attempted to patch a synthesizer depending only upon the numbers printed around the dials - even if they did, to get more than 5% accuracy you'd have to be using awful expensive pots.

http://xhip.cjb.net/temp/public/prtiess.mp3

[andy_FX]

...
we've also completed our argument coming to this conclusion:

- if you want to attempt to create an exact replica of something, you should probably use methods to model that something as accurately as possible.

- if you're interested in only the attributes of that thing, like the 'sound' of a filter for example, it may not be most effective to attempt a perfect model; some have voiced that they believe the exact models sound best while others disagree.

- something about picasso.

I disagree with your summary. I have never made a value judgment on explicitly stating what needs modeling, that is up to each individual to optimise and include whatever they think is important in the model. So, I think what is more accurate is:

- analyse the circuit in question.
- determine which bits you think are important to model
- model them to the level of accuracy required by your particular choices of trandeoff between detail and cpu.


Andrew

[aciddose]

well, that is pretty similar to what i said, only phrased in a different way.

i intended to say exactly the same thing, only i framed it more in a "lump everything into one" A/B way.

another good point to add to those two:

- think for yourselves, people!

[andy_FX]

well, that is pretty similar to what i said, only phrased in a different way.

i intended to say exactly the same thing, only i framed it more in a "lump everything into one" A/B way.

another good point to add to those two:

- think for yourselves, people!

The subtle difference is in not telling people what is important to model and what isn't, and letting them make the choice and tradeoffs they think best. I'm glad that you, in effect, agree to the approach I actually pointed out in my first post

Andrew

[aciddose]

well, in your first post you were talking about a lot of awfully complicated electronic simulation wizardry and i was of the opinion that wasnt really a good starting point for a beginner to take off from.

you said:

"The easiest case as an example to illustrate this is the simple circuit in most distortion stomp boxes, a diode, resistor and cap. You cannot model this with a voltage to voltage lookup table for the diode and then have a one pole low pass filter, it just doesn't give the right results, but you can make simplifications like this to save cpu if you so choose."

my argument which started us off arguing about the merits of doing things your way (component simulation / hybrid) rather than my way (macro-models) was that you can indeed model that with a simple integrator and lookup table, and model it _perfectly_ getting sample-for-sample accurate results just the same as the electronic simulation would give you. obviously you'd need to oversample in the same manner the simulations do, and use exactly the same function, in this case the correctly scaled diode equation.

i've already posted the code proving that.

you mentioned that you could hear a big difference between the three approximations: n^2, n/(n+d)*(1+d), and the diode equation. well, yes anybody could.. i was saying though that the harmonics are very similar and if you bother to scale the inputs correctly to get a similar level of saturation out of the plain n^2 function, you wont be able to hear the difference in most situations. the second approximation actually _is_ the diode equation if you use the right input parameters. what i said back then in one of my posts was that i decided that doing the real equation there wasn't worth my time; ah but apparently the whole argument was.

the facts are, you can macro-model everything perfectly. there is nothing about these systems which makes for only one possible solution to get the same outputs. there are many possible solutions and it is even possible to do a completely linear transform like a modified fft (already suggested in this thread too) and get exactly the same sample-to-sample output as the simulation gives you. my method of cheap macro-modeling is going to be easiest for the newbies to grasp and also will take the least cpu power.

my other statements still stand:

the most important issues are not meddling with modeling parameters like the diode equation in my example code. the most important problem with that code is actually: aliasing! you must use considerable oversampling or use band processing to avoid that. in a four-stage series integrator filter like the MFOS ota filter linked in the start of this thread you have another very serious issue: group delay / phase!.

it is still my opinion that you must solve those problems first, then you can play with modeling. you should also first macro-model, and then go into greater depth with simulation if you find the macro-models to be inaccurate. (unlikely in my opinion, but i'll just say, ok, it might happen.)

you yourself had absolutely no idea how to model a simple integrator/clamp and you thought in fact that it was "not possible" to do so with a "simple integrator and non-linear function look-up table". in my opinion this is actually the result of you having become dependent upon your simulation code. you might be best _yourself_ to take my advice given here to williamk and others!

[andy_FX]

I stuffed up the quotes the first time so edited to get them right by adding the "" characters around the quoted persons name.

well, in your first post you were talking about a lot of awfully complicated electronic simulation wizardry and i was of the opinion that wasnt really a good starting point for a beginner to take off from.

You keep on bringing up magic etc. What is it with you? Did you play too much D&D when you were little? All the stuff I said is easy. If you don't know circuits you have to start somewhere, and a circuit simulation program (that is free and cross platform) is a very good place to start. You seem to forget that you know circuits, new people don't.

you said:
"The easiest case as an example to illustrate this is the simple circuit in most distortion stomp boxes, a diode, resistor and cap. You cannot model this with a voltage to voltage lookup table for the diode and then have a one pole low pass filter, it just doesn't give the right results, but you can make simplifications like this to save cpu if you so choose."

Yes this is correct, apart from the order that I got wrong that I already pointed out. If you do a one pole low pass filter, then put the output in dsp land into a waveshaper you will not get the same result as an unbuffered lowpass filter and waveshaper - period. Look at the graphs

my argument which started us off arguing about the merits of doing things your way (component simulation / hybrid) rather than my way (macro-models) was that you can indeed model that with a simple integrator and lookup table, and model it _perfectly_ getting sample-for-sample accurate results just the same as the electronic simulation would give you. obviously you'd need to oversample in the same manner the simulations do, and use exactly the same function, in this case the correctly scaled diode equation.

The people we are talking to don't know how to work this stuff out, they are learning circuits. If you already know what you are doing after having spent 10 years doing this stuff then perhaps it's ok, but otherwise these guys need help it how to work this stuff out and not waste time with trial and error to get there.

i've already posted the code proving that.

you mentioned that you could hear a big difference between the three approximations: n^2, n/(n+d)*(1+d), and the diode equation. well, yes anybody could.. i was saying though that the harmonics are very similar and if you bother to scale the inputs correctly to get a similar level of saturation out of the plain n^2 function, you wont be able to hear the difference in most situations. the second approximation actually _is_ the diode equation if you use the right input parameters. what i said back then in one of my posts was that i decided that doing the real equation there wasn't worth my time; ah but apparently the whole argument was.

I'll post code and audio examples to show otherwise, but I will have to do it when I have some spare time.

the facts are, you can macro-model everything perfectly. there is nothing about these systems which makes for only one possible solution to get the same outputs. there are many possible solutions and it is even possible to do a completely linear transform like a modified fft (already suggested in this thread too) and get exactly the same sample-to-sample output as the simulation gives you. my method of cheap macro-modeling is going to be easiest for the newbies to grasp and also will take the least cpu power.

My point is that to generate the macro model you need to know what is going on - how do you learn these skills? You yourself pointed out your already analysed all these circuits you are talking about. This doesn't help someone else come along and model a different circuit. You are being selfish in your brash statements.

my other statements still stand:

the most important issues are not meddling with modeling parameters like the diode equation in my example code. the most important problem with that code is actually: aliasing! you must use considerable oversampling or use band processing to avoid that. in a four-stage series integrator filter like the MFOS ota filter linked in the start of this thread you have another very serious issue: group delay / phase!.

Read step one of what I said originally. I'm not sure if you don't listen or you just have an extremely selective memory, either way it's bloody annoying. How many people here have posted backing you up? How many people have appreciated your particular point of view? I am not arguing with you so please don't complement yourself thinking you have engaged me intellectually in any way shape or form. I am pointing out basic facts you seem unable to handle.

it is still my opinion that you must solve those problems first, then you can play with modeling. you should also first macro-model, and then go into greater depth with simulation if you find the macro-models to be inaccurate. (unlikely in my opinion, but i'll just say, ok, it might happen.)

My first post, again, already pointed out that you need to have some decent corrections happening in the linear case. The problem with you suggestion with macro modeling is that unless you already know how to do it you won't know how to do it.

you yourself had absolutely no idea how to model a simple integrator/clamp and you thought in fact that it was "not possible" to do so with a "simple integrator and non-linear function look-up table". in my opinion this is actually the result of you having become dependent upon your simulation code. you might be best _yourself_ to take my advice given here to williamk and others!

Please refer to my earlier comment. I said that you cannot model the system with a one pole low pass filter and then take it's output and shape it with a diode shaper, look at the graphs and for once try to get it through you impressively thick skull what I am saying. That took such a long time to wade through your inane comments, please keep them shorter next time, I have things to get getting on with, but cannot leave such tripe to stand un-answered.

Andrew

[andy_FX]

you mentioned that you could hear a big difference between the three approximations: n^2, n/(n+d)*(1+d), and the diode equation. well, yes anybody could.. i was saying though that the harmonics are very similar and if you bother to scale the inputs correctly to get a similar level of saturation out of the plain n^2 function, you wont be able to hear the difference in most situations. the second approximation actually _is_ the diode equation if you use the right input parameters. what i said back then in one of my posts was that i decided that doing the real equation there wasn't worth my time; ah but apparently the whole argument was.

For the right inputs they will sound identical as well, eg silence. I've never really heard of a right or wrong input, I don't straight jacket my audio by saying some is right and some is wrong.

Andrew

[aciddose]

do not be a fool, i said parameters. have you ever used the diode equation? it's a simple function, the one i posted in the source, with the constants calculated a particular way.

"If you do a one pole low pass filter, then put the output in dsp land into a waveshaper you will not get the same result as an unbuffered lowpass filter and waveshaper - period. Look at the graphs"

you did look at my source, didnt you? this is the same output as the circuit. it is exactly the same.

no, in my source my output does not exactly match your graph. i thought maybe you were intelligent enough to do so if you desired. you obviously lack a fundamental understanding of these systems. for the same parameters (frequency, scale, etc) and same inputs the outputs will match exactly. depending upon exactly what methods you're using to model electronic components the electronic model may even be a hell of a lot less accurate. ignoring the reactive attributes of the semi-conductive region of the diodes in the clamp, no other effects occur beside the function:

d = a - iclamp(b)
b += d * c

in order to be perfectly accurate your function iclamp() must be exactly the diode equation, matching the parameters of which ever diodes you're attempting to model. iterative calculation based upon current sources and the equation gives voltage. in this case we only care about current, so the unmodified equation can be used. 'c' must be exactly the value based upon R/C in the circuit, taking the sample rate into account. then the output will be exactly the same.