jmejia@gmail.com

Does anyone know where to look for real-world data on the charectoristics of specific vintage hardware? (ie. moog or arp filters, pultech eq's, ..anything nice sounding and famous really)

I'd like to experiment a little with designing clones in MSP (figuring eq's or filters are a simpler place to start than vintage compressors), but I'm hoping in this day and age I can skip the hardware analysis step and use someone elses data!

emulating-vintage-hardware-filters-and-eqs

Quote: jmejia@gmail.com wrote on Sun, 26 March 2006 10:02

----------------------------------------------------

> Does anyone know where to look for real-world data on the charectoristics of specific vintage hardware? (ie. moog or arp filters, pultech eq's, ..anything nice sounding and famous really)

easiest way is convolution, you'd need the spat library (i think that's what it's called) for high quality IR.

if you want to do real modelling with circuits and all, i think you're a long way from home.

that said, the synthi emulation is pretty cool, 

http://couprie.pierre.free.fr/texte.php3?id_article=21

i feel this is more of an external devloping kind of thing, but i may be completely off on that...

Quote: jmejia@gmail.com wrote on Sun, 26 March 2006 10:02
----------------------------------------------------
> Does anyone know where to look for real-world data on the charectoristics of specific vintage hardware? (ie. moog or arp filters, pultech eq's, ..anything nice sounding and famous really)
----------------------------------------------------
easiest way is convolution, you'd need the spat library (i think that's what it's called) for high quality IR.

if you want to do real modelling with circuits and all, i think you're a long way from home.
that said, the synthi emulation is pretty cool, 
http://couprie.pierre.free.fr/texte.php3?id_article=21

Csound has implementations of Moog VCOs and filters. I haven't used 

them, so I can't attest to their quality, but what with Csound being 

GPL, there should be source code that you might learn from. (or you 

One big issue with emulating analog is producing the jitter in the 

signals associated with it, for which rand~ is a good starting point. 

(or my preferred combo: a rand~ controlling a rand~)

Also, doing some slightly varying wave-shaping might be interesting.

Csound has implementations of Moog VCOs and filters.  I haven't used 
them, so I can't attest to their quality, but what with Csound being 
GPL, there should be source code that you might learn from.  (or you 
could just use them in csound~)

One big issue with emulating analog is producing the jitter in the 
signals associated with it, for which rand~ is a good starting point.  
(or my preferred combo: a rand~ controlling a rand~)
Also, doing some slightly varying wave-shaping might be interesting.

I am interested in modelling, not using convolution. I don't know anything about circuitry modelling (although I do have a lot of nice schematics, and a few diy prototypes lying around)

I'm not sure if it would be easier to look at spectrum analysis data from a particular object (as I was thinking in my previous post), or to analyze the circuit itself... again, it's something I haven't looked into before but am interested in learning about. It seems like a schematic could be turned into to a useful algorithm for a filter in msp, and even more useful if inherent physical flaws of the hardware components are considered. (tolerance of resistors, heat stress on components, idiosyncracies of specific old parts, etc)

I'm just looking for a push in the right direction for learning how to do this sort of thing. Maybe theres a book that would help? I know how to read circuits, and have some knowledge of what they do, but I don't know exactly what aspect to study furthur.

I have looked at that synthi emulation before, it's impressive.

Quote: peter.mcculloch@gmail.com wrote on Sun, 26 March 2006 11:10

> Csound has implementations of Moog VCOs and filters.

I'll check out that code, thanks for the tip!

Your advice Re: electronic jitter sounds useful too!

Quote: peter.mcculloch@gmail.com wrote on Sun, 26 March 2006 11:10
----------------------------------------------------
> Csound has implementations of Moog VCOs and filters.  

On 26-Mar-2006, at 20:10, Peter McCulloch wrote:

> One big issue with emulating analog is producing the jitter in the 

Altogether, Litter Power provides a rich source of materials for this 

sort of endeavor. In case you're interested.

Peter Castine +--> Litter Power & Litter Bundle for Jitter

iCE: Sequencing, Recording & |home | chez nous|

Interface Building for |bei uns | i nostri|

On 26-Mar-2006, at 20:10, Peter McCulloch wrote:
> One big issue with emulating analog is producing the jitter in the  
> signals associated with it

Altogether, Litter Power provides a rich source of materials for this  
sort of endeavor. In case you're interested.

--------------    http://www.bek.no/~pcastine/Litter/    -------------
Peter Castine             +--> Litter Power & Litter Bundle for Jitter

> easiest way is convolution, you'd need the spat library (i think

> that's what it's called) for high quality IR.

Convolution would only work for linear filtering, for that purpose

buffir~ would be fine (as filters do not need very long impuls

resposes), though its not really handy, as it can only remember on

The specialty of "real world filters" like Moog and Arp filters is their

distortion part. No way to do that with convolution.

Linear filters are pretty well covered by a combination of filtergraph~

and biquad~ or by svf~ or by reson~ or...

Might be fun to put some distortion somewhere in the (feedback)path to 

Knowing about the general functionality of filters, nonlinear distortion 

and sound would be an necessary requirement though. Can't really avoid 

that, even if someone would hand you a ready patched solution.

junior wrote:
> easiest way is convolution, you'd need the spat library (i think
> that's what it's called) for high quality IR.

Convolution would only work for linear filtering, for that purpose
buffir~ would be fine (as filters do not need very long impuls
resposes), though its not really handy, as it can only remember on
single setting (no filtersweeps...).

The specialty of "real world filters" like Moog and Arp filters is their
distortion part. No way to do that with convolution.

Linear filters are pretty well covered by a combination of filtergraph~
and biquad~ or by svf~ or by reson~ or...

Might be fun to put some distortion somewhere in the (feedback)path to 
emulate old analog nasties.
Knowing about the general functionality of filters, nonlinear distortion 
and sound would be an necessary requirement though. Can't really avoid 
that, even if someone would hand you a ready patched solution.

  [][]  [][][]  [][]  [][][]
[][][][][][][][][][][][][][][]

         Stefan Tiedje
         Klanggestalter
     Electronic Composition
               &
         Improvisation

            /~~~~~
     \   /|() ()|
     ))))) )|  |  |( \
     ///     _/)/ )))))
             ___/   ///

-------------------------x----
--_____-----------|-----------
--(_|_ ----|-----|-----()----
-- _|_)----|-----()-----------
----------()------------x-----

14, Av. Pr. Franklin Roosevelt,
94320 Thiais, France
Phone at CCMIX +33-1-57 42 91 09


yeah, that's fine... I was just hoping I could get my hands on data on specific filters. for example..adding distortion to the feedback path of a linear filter is all well and good...but I'd like to data regarding how much, when and where the specific filter I'm modelling distorts. (among other charectoristics) I am not interested in generic analog emulation, but emulating a specific device. (although I'm not too picky about what device it is!) I also don't want a ready patched solution...just some more knowledge, and hopefully some real-world specs and/or an understanding of how to model from a schematic. I'm not afraid of heavy reading...

You may also want to check the US Patent Office. A number of those 

old filters were patented and the patent applications will have had 

A long and winding road, and it's not as if success is guaranteed. If 

you do find something, the good news is that the patent protection 

will almost certainly have long since expired, freeing you to do what 

On 27-Mar-2006, at 11:16, Jesse wrote:
> I'm not afraid of heavy reading...

You may also want to check the US Patent Office. A number of those  
old filters were patented and the patent applications will have had  
circuit diagrams attached.

A long and winding road, and it's not as if success is guaranteed. If  
you do find something, the good news is that the patent protection  
will almost certainly have long since expired, freeing you to do what  
you will.

i completely disagree with those who said that

you can look into electronics diagrams of hardware

the normal way of programming a vintage EQ would be

making a series of recordings with the gear and

then for example write a FIR filter from this data.

i completely disagree with those who said that
you can look into electronics diagrams of hardware
to be able to biuld a software version.
the normal way of programming a vintage EQ would be
making a series of recordings with the gear and
then for example write a FIR filter from this data.


Quote: Roman Thilenius wrote on Mon, 27 March 2006 04:48

> i completely disagree with those who said that

> you can look into electronics diagrams of hardware

> to be able to biuld a software version.

> the normal way of programming a vintage EQ would be

> making a series of recordings with the gear and

> then for example write a FIR filter from this data.

That was exactly the direction I was thinking in when I first posted this thread, I was hoping to steal someone elses recordings or even better the analyzed data.

In lieu of finding such luck, in terms of 'a series of recordings' could I do it with impulse-type recordings (to analyze...NOT for convolution), or would I need an exhaustive set of recordings of real-world musical sounds?

I'm thinking if I send an impulse through a hardware filter with incremental recordings that might be enough to develop a data set for writing a filter.. any thoughts?

Quote: Roman Thilenius wrote on Mon, 27 March 2006 04:48
----------------------------------------------------
>
> i completely disagree with those who said that
> you can look into electronics diagrams of hardware
> to be able to biuld a software version.
> the normal way of programming a vintage EQ would be
> making a series of recordings with the gear and
> then for example write a FIR filter from this data.
----------------------------------------------------

That was exactly the direction I was thinking in when I first posted this thread, I was hoping to steal someone elses recordings or even better the analyzed data.
In lieu of finding such luck, in terms of 'a series of recordings' could I do it with impulse-type recordings (to analyze...NOT for convolution), or would I need an exhaustive set of recordings of real-world musical sounds?
I'm thinking if I send an impulse through a hardware filter with incremental recordings that might be enough to develop a data set for writing a filter.. any thoughts?


>> i completely disagree with those who said that

>> you can look into electronics diagrams of hardware

>> to be able to biuld a software version.

>> the normal way of programming a vintage EQ would be

>> making a series of recordings with the gear and

>> then for example write a FIR filter from this data.

Both approaches are possible, either has its traps and pitfalls.

What I've seen of NI's work indicates they disagree with your 

Anyone know which approach Propellerheads take?

>> i completely disagree with those who said that
>> you can look into electronics diagrams of hardware
>> to be able to biuld a software version.
>> the normal way of programming a vintage EQ would be
>> making a series of recordings with the gear and
>> then for example write a FIR filter from this data.

What I've seen of NI's work indicates they disagree with your  
complete disagreement. So go disagree.-

I'd also love to know how Universal Audio does it... the UAD-1 must have the most authentic sounding emulations I've heard/used.

Quote: Stefan Tiedje wrote on Tue, 28 March 2006 02:39

> The only way to really emulate is the schematics. You need to know the 

> nonlinear behaviour of each component. especially transistors, FET, 

This method really interests me, and your arguments agains FIR make some sense. What specs should I pay attention to in terms of a given transistor in the circuit? Can the non-linear behavior be derived from the data sheet on the part? (I'm hoping this isn't something I'll need to measure myself with meters and scopes on the component itself...

To give a concrete example, one idea I have is to model a moog-style multi-ladder filter from some DIY schematics I've built from in the past. The top and bottom pairs of the ladder utilize the Analog Devices SSM2210 NPN transistors. (matched pairs)

Where would I begin in terms of emulating the behavior of these transistors? I know basically what they do, but what will give me insight about their non-linear behavior, and where would I begin in terms of writing an algorithm that mimics them? The datasheet is here if anyone can help:

Again, I realize there is a lot of work cut out for me if I go this route...and I'm not afraid to hit the books if I need to learn more about electronics, but I'd like some pushes in the right direction.

> The only way to really emulate is the schematics. You need to know the 
> nonlinear behaviour of each component. especially transistors, FET, 
> opamps etc. 

 This method really interests me, and your arguments agains FIR make some sense. What specs should I pay attention to in terms of a given transistor in the circuit? Can the non-linear behavior be derived from the data sheet on the part? (I'm hoping this isn't something I'll need to measure myself with meters and scopes on the component itself...

 To give a concrete example, one idea I have is to model a moog-style multi-ladder filter from some DIY schematics I've built from in the past. The top and bottom pairs of the ladder utilize the Analog Devices SSM2210 NPN transistors. (matched pairs)
 Where would I begin in terms of emulating the behavior of these transistors? I know basically what they do, but what will give me insight about their non-linear behavior, and where would I begin in terms of writing an algorithm that mimics them? The datasheet is here if anyone can help:
http://tinyurl.com/l8eub

On Mar 27, 2006, at 6:00 AM, Jesse wrote:

> I'd also love to know how Universal Audio does it... the UAD-1 must 

> have the most authentic sounding emulations I've heard/used.

According to an interview in Tape Op some months back, they do 

attempt to model the system at the component level. But, as I 

remember thinking when reading the article, this could mean a lot of 

different things. One thing it certainly doesn't mean is that they 

have a model of the system as a network of all of its discrete analog 

components with their salient nonlinearities all running in real 

time. That's just too much computing, and much of it irrelevant 

Emulating individual components in MSP is unlikely to be a way 

forward. Circuit simulation cannot really be done in the linear flow- 

through-unit-generators world of MSP, for a couple of reasons. One 

is the feedback and the signal vector size. Even with signal vector 

of 1 you will be adding a single-sample delay to filter structures 

which should have (practically) zero delay, which will throw off your 

simulation. For a lot of filters/circuits simple oversampling might 

be a way around this problem, but then you're probably out of real 

time already. Another problem, more basic, is that you need to 

model both current and voltage through the system. It's not possible 

to understand this flow as unidirectional, as with audio in MSP. The 

sum of current at each junction, at each time step, must be zero, 

which means that you have to solve in one way or another for the 

effect of each component on all of its neighbors. Look at how 

something like SPICE works if you want to affirm how difficult this is.

What is doable is to break a big circuit down into functional blocks 

which are understood as filters. Then the color-providing 

nonlinearities which people focus on can be added to this basic 

structure. Some of these may correspond mainly to single components 

in the analog circuit such as vactrols, transformers or output gain 

stages. So it's not entirely wrong when people say they are 

"modeling components," but it's not the whole story.

To do this kind of stuff, there is no way around knowing some filter 

design / basic DSP. My favorite starter book is still "Digital Audio 

Signal Processing, an Anthology", John Strawn, Ed. You also need to 

understand the analog circuits, for which "The Art of Electronics" by 

Horowitz and Hill is great. Also, one can learn a lot of context by 

lurking on the music-dsp list, where this kind of thing is discussed 

all the time. All in all, this is a rewarding multi-year voyage.

>
> I'd also love to know how Universal Audio does it... the UAD-1 must  
> have the most authentic sounding emulations I've heard/used.

According to an interview in Tape Op some months back, they do  
attempt to model the system at the component level.  But, as I  
remember thinking when reading the article, this could mean a lot of  
different things.  One thing it certainly doesn't mean is that they  
have a model of the system as a network of all of its discrete analog  
components with their salient nonlinearities all running in real  
time.  That's just too much computing, and much of it irrelevant  
besides.

Emulating individual components in MSP is unlikely to be a way  
forward.  Circuit simulation cannot really be done in the linear flow- 
through-unit-generators world of MSP, for a couple of reasons.  One  
is the feedback and the signal vector size.  Even with signal vector  
of 1 you will be adding a single-sample delay to filter structures  
which should have (practically) zero delay, which will throw off your  
simulation.  For a lot of filters/circuits simple oversampling might  
be a way around this problem, but then you're probably out of real  
time already.   Another problem, more basic, is that you need to  
model both current and voltage through the system.  It's not possible  
to understand this flow as unidirectional, as with audio in MSP.  The  
sum of current at each junction, at each time step, must be zero,  
which means that you have to solve in one way or another for the  
effect of each component on all of its neighbors.  Look at how  
something like SPICE works if you want to affirm how difficult this is.

What is doable is to break a big circuit down into functional blocks  
which are understood as filters.  Then the color-providing  
nonlinearities which people focus on can be added to this basic  
structure.  Some of these may correspond mainly to single components  
in the analog circuit such as vactrols, transformers or output gain  
stages.  So it's not entirely wrong when people say they are  
"modeling components," but it's not the whole story.

To do this kind of stuff, there is no way around knowing some filter  
design / basic DSP.  My favorite starter book is still "Digital Audio  
Signal Processing, an Anthology", John Strawn, Ed.  You also need to  
understand the analog circuits, for which "The Art of Electronics" by  
Horowitz and Hill is great.  Also, one can learn a lot of context by  
lurking on the music-dsp list, where this kind of thing is discussed  
all the time.  All in all, this is a rewarding multi-year voyage.

Quote: randall jones wrote on Tue, 28 March 2006 12:49

> According to an interview in Tape Op some months back, they do 

> attempt to model the system at the component level. But, as I 

> remember thinking when reading the article, this could mean a lot of 

I read that article as well, and I remember being pretty unclear as to what that really meant too. It'd be nice if they out-right told us their secrets - but that would be communism wouldn't it.

> What is doable is to break a big circuit down into functional blocks 

That sounds good, so in my above example, it might make sense to consider the transistor ladder as one (the primary) of these functional blocks? (as opposed to modelling each transistor) Now I just would need to figure out how to mimic a transistor ladder in MSP.............

>Then the color-providing nonlinearities which people focus on can be added to this basic 

> structure. Some of these may correspond mainly to single components

Then I think I would be correct in assuming that the 'key' component that makes my ladder in question sound like the ladder in question would be the SSM2210 NPN transistor I mentioned earlier. I know these things are distinctive...but how do I find out why exactly they are distinctive in some sort of emulatable terms? Maybe they introduce some very specific type of jitter that could be emulated in approximately the same position in an MSP patch? Looking at the data sheets I posted above, I have no idea where to look for such a thing.

> My favorite starter book is still "Digital Audio 

> Signal Processing, an Anthology", John Strawn, Ed.

Thanks! I'll pick that one up right away!

> You also need to understand the analog circuits, for which "The Art of Electronics" by 

I do have that book, but I haven't gotten too far into it yet.. It's on my list of things to do.

>Also, one can learn a lot of context by 

> lurking on the music-dsp list, where this kind of thing is discussed 

> all the time. All in all, this is a rewarding multi-year voyage.

Thanks for all your good advice! I'm excited to learn more about all of this!

Quote: randall jones wrote on Tue, 28 March 2006 12:49
----------------------------------------------------

> According to an interview in Tape Op some months back, they do  
> attempt to model the system at the component level.  But, as I  
> remember thinking when reading the article, this could mean a lot of  
> different things.  

> What is doable is to break a big circuit down into functional blocks  
> which are understood as filters.  

>Then the color-providing nonlinearities which people focus on can be added to this basic  
> structure.  Some of these may correspond mainly to single components  

> My favorite starter book is still "Digital Audio  
> Signal Processing, an Anthology", John Strawn, Ed.  

> You also need to understand the analog circuits, for which "The Art of Electronics" by  
> Horowitz and Hill is great.  

>Also, one can learn a lot of context by  
> lurking on the music-dsp list, where this kind of thing is discussed  
> all the time.  All in all, this is a rewarding multi-year voyage.
>

> opamps etc. All the information should be available, and you probably 

> need either a vector size of one, or need to code it in Java or C.

there is 2 sorts of mastering compressors and they 

all function more or less the same way and have 

but they all soound different, because they use 

from looking at the circuit plan you do not know how the

tube or transistor or spool influences the attack of a

bassdrum sound at -13db, only from recroding it you will 

there are even products out there where you can choose 

with what type of tubes you want to have your product 

equipped. the schematics will remain the same.

> The only way to really emulate is the schematics. You need to know the 
> nonlinear behaviour of each component. especially transistors, FET, 
> opamps etc. All the information should be available, and you probably 
> need either a vector size of one, or need to code it in Java or C.

not at all.
there is 2 sorts of mastering compressors and they 
all function more or less the same way and have 
similar schematics.
but they all soound different, because they use 
different components.
from looking at the circuit plan you do not know how the
tube or transistor or spool influences the attack of a
bassdrum sound at -13db, only from recroding it you will 
know.

emulating vintage hardware filters and eq's