Fun with SM (Moog ladder filter)

malc · 20 Mar 2005

Wow! 🙂

rl wrote: Pls. note that the code is a bit messy, since it workarounds a couple of restrictions and minor bugs of the code component

Do we know about these? Feel free to remind us.

EDIT: just saw your other post on these, thanks.

Shifrin · 20 Mar 2005

Thanks for this, going to have a look (listen 😉 ) now 😀

Shif.

Tom7777 · 21 Mar 2005

As im a beginner in coding could you if possible please give some examples on how you converted certain parts as the SM version is very very different to the Csound code. Or did you use a convertion tool to help you? Seems like you know alot more about the coding modules than i currently do hehe 🙂 Thanks

jorge · 21 Mar 2005

wow rl, thanks a lot. I replaced the old moog code in the example schematic with your code, it sound really great: zip:[old link] rar:[old link]

rl wrote: Pls. note that the code is a bit messy, since it workarounds a couple of restrictions and minor bugs of the code component

Didnt sound messy at all and that's what matters the most.

lalo · 21 Mar 2005

well...simply...AMAZING!!!

great rl...

if i stay away from SM for a week is like a light-year 😉

congrats

lalo

rl · 21 Mar 2005

Tom7777 wrote: As im a beginner in coding could you if possible please give some examples on how you converted certain parts as the SM version is very very different to the Csound code. Or did you use a convertion tool to help you? Seems like you know alot more about the coding modules than i currently do hehe 🙂 Thanks

basically it's the same code, manually converted into C-style syntax. The little magic is the approximation code for exp and tanh. exp(x) uses the usual taylor series - for the range used more coefficients are needed for the desired accuracy. the usual taylor approximation for tanh(x) turned out to be totally inaccurate, so I used the fact that tanh(x) can be expressed as (exp(x) - exp(-x))/(exp(x) + exp(-x)), which can be written as 1 - 2 / (exp (2*x) + 1) again using taylor for exp(2x), giving reasonable good results for x >=0 and using the fact that tanh is an odd function, thus tanh(-x) = -tanh(x) for x <0

The code can be further optimized, e.g. instead of dividing numbers you can multiply the inverse of the number, or by forming some terms into multiply-add operations, e.g. expOut = 1 + x + x*x/2 + x*x*x/6 + x*x*x*x/24 + x*x*x*x*x/120 + x*x*x*x*x*x/720; -> = 1 + x ( 1 + x ( 1/2 + x ( 1/6 + x (1/24 + x ( 1/120 + x/720 ) ) ) ) )

the latter expression uses only 5 multiplications and 6 adds (not counting the divides) , the former uses 15 MUL and 6 ADD;

Tom7777 · 22 Mar 2005

Thanks 🙂

dharma one · 11 Apr 2005

cool, i really like how the moog filter sounds like in antti's asynth. i tried to use this for audio processing, but it sounds really grainy...

i just wire up an audio input to this filter module and connect that to audio output and exported that as vst.

for controlling resonance and cutoff i connected two knobs (float output) to the cutoff and reso inputs of the filter module. it says in the comments this takes values from 0-1 as input and the knob seems to give out 0 to 1 floats

any ideas what i'm doing wrong?

rl · 11 Apr 2005

dharma one wrote: any ideas what i'm doing wrong?

The filter was tuned to be an replacement for the filter of the famous example.osm. I suggest you experiment with the cfScale and resScale variables. The grainyness has also to do with the quality of the approximations. It's a tradeoff between accuracy and speed. You could try to add more terms to the approximations for exp and tanh

Josh · 20 Aug 2006

Ha! Just when you think you've seen it all

What would be nice is if the filter actually worked lol. Example moog doesn't make any sound and the filter code is mostly black...

AngularMomentum · 20 Aug 2006

Electrogravtiy wrote: Ha! Just when you think you've seen it all

What would be nice is if the filter actually worked lol. Example moog doesn't make any sound and the filter code is mostly black...

Same here, next to that the example.osm posted here does not contain any graphics when I load it up.

Tom7777 · 20 Aug 2006

Well what do you expect guys this is quite an old topic now started - Sun Mar 20, 2005 😉 The program has changed alot over time. Im guessing this isn't the only old example which doesn't work now 😀

Josh · 20 Aug 2006

Perhaps one of you code geniuses can fix it. Ladder filters seem to be pretty much the buisiness. Even the virus uses them. A high pass version would be cool too.

oddson · 20 Aug 2006

A very small change to the constants section is all that's needed to get rid of the compile errors:

// constants float pi = 3.1415926535; float v2 = 40000; // twice the 'thermal voltage of a transistor' float sr = 22100; float cfScale = 10; float resScale = 1.3;

Josh · 21 Aug 2006

Woo Hoo. Cool Cheers oddson 😀

AngularMomentum · 22 Aug 2006

Thanks Oddson, filter sounds nice indeed!

oddson · 22 Aug 2006

Don't thank me... I only spent 2 minutes on this... (all Kudos to Rudolph!)

The Code component will not allow any variable declarations after an assignment has occurred unless the assignment is part of the declaration.

If you see any more old code that won't compile look for this as a potential culprit.

Tom7777 · 22 Aug 2006

Yeah basically you just need to put all the floats to the top after the ins and outs and the old examples should work ok then 😀

rl · 26 Aug 2006

Here's more Moog. This time a port of the famous Stilson/Smith algorithm from music-dsp:

streamout lowpass; streamin signal; streamin cf; // Hz streamin rez; // 0..1 streamin sr; // samplerate in Hz

streamout dbg;

// Stilson/Smith Moog algorithm // adapted from http://www.musicdsp.org/archive.php?classid=3#145 // Rudolf Lindner, 2006

float output; float fc; float fc2; float fc3; float p; float i; float s[4]; float temp; float arg1, arg2; float Q; float gaintable; float x; float o1, o2;

// saturation (clamping) float satLimit=0.95;

fc =2*cf/sr; fc2=fc*fc; fc3=fc*fc2;

// cubic fit by DFL, not 100% accurate but better than nothing... p = -0.69346 * fc3 - 0.59515 * fc2 + 3.2937 * fc - 1.0072;

x=p*99+100; // square approximation of original gaintable (RL) gaintable = 0.00002*x*x - 0.0069*x + 0.9597; Q = rez * gaintable;

output=0.25 * ( signal - output ); // negative feedback

i=0; loop (4) {

temp = s[i];

arg1=output + p * (output - temp) + satLimit; arg2=output + p * (output - temp) - satLimit; o1 = (arg1 < 0)&(-1*(arg1)) + (arg1>=0)&arg1; o2 = (arg2 < 0)&(-1*(arg2)) + (arg2>=0)&arg2; output = 0.5*(o1-o2);

s[i] = output;

arg1=output + temp+satLimit; arg2=output + temp-satLimit; o1 = (arg1 < 0)&(-1*(arg1)) + (arg1>=0)&arg1; o2 = (arg2 < 0)&(-1*(arg2)) + (arg2>=0)&arg2; output = 0.5*(o1-o2);

i = i + 1; } lowpass=output;

output = output * Q;

dbg=lowpass;

I replaced the lengthy gaintable by a simple polynomial approximation. The code is written for clarity and works. There is always room for some optimization though.

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

there should also be noted that there there is a much better implementation of the algorithm of my original post developed by sambean in this forum.

Tom7777 · 26 Aug 2006

Nice, thanks rl, im guessing this is a lowpass? as the output says lowpass hehe 😀 I thought i should ask though, can you have moog HP BP BR ect also? i don't think i have seen any yet 🙂

Josh · 27 Aug 2006

Has anyone got the new filter to work? I just get an audible click. Btw what should the sample rate look like? I used 44100....

oddson · 27 Aug 2006

Works for me rlfilter.osm

rl · 27 Aug 2006

Thanks for the example. I was just too lazy to upload one 😊

Some tweaking tips: - between Cutoff knob and "0-1 to Hz" multiply the cutoff with itself, so there is more control in the lower frequency range

- replace

float satLimit=0.95;

with streamin satLimit; to have control over the saturation.

- crank up resonance and sat Limit to higher values [0..2] for nasty bass sounds

- try tanh or sqrt limiting instead of clamping

- use loop (3) { for 3 pole 18 dB "computer controlled transistor bass" sounds

cheers, Rudolf

________________________ "all your base is belong to us"

Josh · 27 Aug 2006

Cheers oddson...again lol Whoa the filter sounds good 😍 Thanks RL ....303 here we come he he he

stefan · 28 Aug 2006

rl wrote: Thanks for the example. I was just too lazy to upload one 😊

Some tweaking tips: - between Cutoff knob and "0-1 to Hz" multiply the cutoff with itself, so there is more control in the lower frequency range

Or multiply cutoff with 140 or so and feed it into a pitch2freq converter!

Josh · 28 Aug 2006

How would I go about oversampling it?

Kingston · 7 Sept 2006

I had a look at the code module and thought of a simple optimisation,

replace all of these:

denum = 1 + 2*x + 2*x*x + 4/3*x*x*x + 1; exprPos = 1 - 2/denum; denum = 1 - 2*x + 2*x*x - 4/3*x*x*x + 1; exprNeg = -1 + 2/denum; x = (x >= 0) & exprPos + (x < 0) & exprNeg;

with these:

n0=x*x*2; n1=2*x + n0; n2=2.666666667*n0*x;

denum = 2 + n1 + n2; exprPos = 1 - 2/denum; denum = 2 - n1 - n2; exprNeg = -1 + 2/denum; x = (x >= 0) & exprPos + (x < 0) & exprNeg;

I didn't count how many MULTs, DIVs and ADDs that saves but quite a lot! 😀 (like hundreds!!)

Both functions are equal by the way, except for maybe that 2.666666667 bit (cuz it's 4/1.5). I don't know what the resolution allowed is, before denormals start to happen.

Josh · 10 Sept 2006

RE the ladder filter. I'm getting a wierd wisteling noise when the cutoff is in a low postion and I play a low pitched note even when the resonance is set to 0. I whsh I could post a sample of it but I dont have webspace yet. Has anyone else experienced the same?

Tom7777 · 10 Sept 2006

Hmm, haven't had the chance to check the latest version out yet but that sounds like something to do with the resonance, maybe theres a small bug in the code or something, are you giving it Hz values or normal 0-1's?

Josh · 14 Sept 2006

Kingston wrote: I had a look at the code module and thought of a simple optimisation,

replace all of these:

denum = 1 + 2*x + 2*x*x + 4/3*x*x*x + 1; exprPos = 1 - 2/denum; denum = 1 - 2*x + 2*x*x - 4/3*x*x*x + 1; exprNeg = -1 + 2/denum; x = (x >= 0) & exprPos + (x < 0) & exprNeg;

with these:

n0=x*x*2; n1=2*x + n0; n2=2.666666667*n0*x;

denum = 2 + n1 + n2; exprPos = 1 - 2/denum; denum = 2 - n1 - n2; exprNeg = -1 + 2/denum; x = (x >= 0) & exprPos + (x < 0) & exprNeg;

I didn't count how many MULTs, DIVs and ADDs that saves but quite a lot! 😀 (like hundreds!!)

Both functions are equal by the way, except for maybe that 2.666666667 bit (cuz it's 4/1.5). I don't know what the resolution allowed is, before denormals start to happen.

You mean like so?

streamin input; streamin cutoff; streamin resonance; streamin sr; streamout sigout;

// filter based on the text "Non linear digital implementation of the moog ladder filter" by Antti Houvilainen // resonance [0..1] // cutoff from 0 (0Hz) to 1 (nyquist)

// constants float pi = 3.1415926535; float v2 = 40000; // twice the 'thermal voltage of a transistor'

float cfScale = 10 ; float resScale = 1;

// state vars float az1; float az2; float az3; float az4; float az5; float ay1; float ay2; float ay3; float ay4; float amf;

// helper vars float x; // temp var: input for taylor approximations float denum; // temp vars to split complex expressions into simpler ones float exprPos; // bug in code parser? float exprNeg;

float expOut; // output of approximations float tanh1Out; float tanh2Out;

// others float kfc; float kf; float kfcr; float kacr; float k2vg;

kfc = cutoff/cfScale; // sr is half the actual filter sampling rate kf = cutoff/(2*cfScale);

// frequency & amplitude correction kfcr = 1.8730*kfc*kfc*kfc + 0.4955*kfc*kfc - 0.6490*kfc + 0.9988; kacr = -3.9364*kfc*kfc + 1.8409*kfc + 0.9968;

x = -2 * pi * kfcr * kf; // expOut = 1 + x + (x*x)/2 + (x*x*x)/6 + (x*x*x*x)/24 + (x*x*x*x*x)/120 + (x*x*x*x*x*x)/720; expOut = 1 + x + x*x/2 + x*x*x/6 + x*x*x*x/24; expOut = expOut + x*x*x*x*x/120; expOut = expOut + x*x*x*x*x*x/720;

k2vg = v2*(1-expOut); // filter tuning

// cascade of 4 1st order sections x = (input - resScale*resonance*amf*kacr) / v2;

// (x>=0) -> tanh(x); (x< 0) -> -tanh(-x) // tanh1Out = (x >= 0) & ( 1 - 2 / ( 1 + 2*x + 2*x*x + 4/3*x*x*x + 1 ) ) // + (x < 0) & -1 * ( 1 - 2 / ( 1 - 2*x + 2*x*x - 4/3*x*x*x + 1 ) ); n0=x*x*2; n1=2*x + n0; n2=2.666666667*n0*x; denum = 2 + n1 + n2; exprPos = 1 - 2/denum; denum = 2 - n1 - n2; exprNeg = -1 + 2/denum; x = (x >= 0) & exprPos + (x < 0) & exprNeg;

x = az1/v2; n0=x*x*2; n1=2*x + n0; n2=2.666666667*n0*x; denum = 2 + n1 + n2; exprPos = 1 - 2/denum; denum = 2 - n1 - n2; exprNeg = -1 + 2/denum; x = (x >= 0) & exprPos + (x < 0) & exprNeg;

ay1 = az1 + k2vg*(tanh1Out - tanh2Out); az1 = ay1;

x = ay1/v2; n0=x*x*2; n1=2*x + n0; n2=2.666666667*n0*x; denum = 2 + n1 + n2; exprPos = 1 - 2/denum; denum = 2 - n1 - n2; exprNeg = -1 + 2/denum; x = (x >= 0) & exprPos + (x < 0) & exprNeg;

x = az2/v2; n0=x*x*2; n1=2*x + n0; n2=2.666666667*n0*x; denum = 2 + n1 + n2; exprPos = 1 - 2/denum; denum = 2 - n1 - n2; exprNeg = -1 + 2/denum; x = (x >= 0) & exprPos + (x < 0) & exprNeg;

ay2 = az2 + k2vg*(tanh1Out - tanh2Out); az2 = ay2;

x = ay2/v2; n0=x*x*2; n1=2*x + n0; n2=2.666666667*n0*x; denum = 2 + n1 + n2; exprPos = 1 - 2/denum; denum = 2 - n1 - n2; exprNeg = -1 + 2/denum; x = (x >= 0) & exprPos + (x < 0) & exprNeg;

x = az3/v2; n0=x*x*2; n1=2*x + n0; n2=2.666666667*n0*x; denum = 2 + n1 + n2; exprPos = 1 - 2/denum; denum = 2 - n1 - n2; exprNeg = -1 + 2/denum; x = (x >= 0) & exprPos + (x < 0) & exprNeg;

ay3 = az3 + k2vg*(tanh1Out - tanh2Out); az3 = ay3;

x = ay3/v2; n0=x*x*2; n1=2*x + n0; n2=2.666666667*n0*x; denum = 2 + n1 + n2; exprPos = 1 - 2/denum; denum = 2 - n1 - n2; exprNeg = -1 + 2/denum; x = (x >= 0) & exprPos + (x < 0) & exprNeg;

x = az4/v2; n0=x*x*2; n1=2*x + n0; n2=2.666666667*n0*x; denum = 2 + n1 + n2; exprPos = 1 - 2/denum; denum = 2 - n1 - n2; exprNeg = -1 + 2/denum; x = (x >= 0) & exprPos + (x < 0) & exprNeg;

ay4 = az4 + k2vg*(tanh1Out - tanh2Out); az4 = ay4;

// 1/2-sample delay for phase compensation amf = (ay4+az5)*0.5; az5 = ay4;

// oversampling (repeat same block) x = (input - resScale*resonance*amf*kacr) / v2; n0=x*x*2; n1=2*x + n0; n2=2.666666667*n0*x; denum = 2 + n1 + n2; exprPos = 1 - 2/denum; denum = 2 - n1 - n2; exprNeg = -1 + 2/denum; x = (x >= 0) & exprPos + (x < 0) & exprNeg;

x = az1/v2; n0=x*x*2; n1=2*x + n0; n2=2.666666667*n0*x; denum = 2 + n1 + n2; exprPos = 1 - 2/denum; denum = 2 - n1 - n2; exprNeg = -1 + 2/denum; x = (x >= 0) & exprPos + (x < 0) & exprNeg; ay1 = az1 + k2vg*(tanh1Out - tanh2Out); az1 = ay1;

x = ay1/v2; n0=x*x*2; n1=2*x + n0; n2=2.666666667*n0*x; denum = 2 + n1 + n2; exprPos = 1 - 2/denum; denum = 2 - n1 - n2; exprNeg = -1 + 2/denum; x = (x >= 0) & exprPos + (x < 0) & exprNeg; tanh1Out = (x >= 0) & exprPos + (x < 0) & exprNeg;

x = az2/v2; n0=x*x*2; n1=2*x + n0; n2=2.666666667*n0*x; denum = 2 + n1 + n2; exprPos = 1 - 2/denum; denum = 2 - n1 - n2; exprNeg = -1 + 2/denum; x = (x >= 0) & exprPos + (x < 0) & exprNeg; ;

ay2 = az2 + k2vg*(tanh1Out - tanh2Out); az2 = ay2;

x = ay2/v2; n0=x*x*2; n1=2*x + n0; n2=2.666666667*n0*x; denum = 2 + n1 + n2; exprPos = 1 - 2/denum; denum = 2 - n1 - n2; exprNeg = -1 + 2/denum; x = (x >= 0) & exprPos + (x < 0) & exprNeg;

x = az3/v2; n0=x*x*2; n1=2*x + n0; n2=2.666666667*n0*x; denum = 2 + n1 + n2; exprPos = 1 - 2/denum; denum = 2 - n1 - n2; exprNeg = -1 + 2/denum; x = (x >= 0) & exprPos + (x < 0) & exprNeg;

ay3 = az3 + k2vg*(tanh1Out - tanh2Out); az3 = ay3;

x = ay3/v2; n0=x*x*2; n1=2*x + n0; n2=2.666666667*n0*x; denum = 2 + n1 + n2; exprPos = 1 - 2/denum; denum = 2 - n1 - n2; exprNeg = -1 + 2/denum; x = (x >= 0) & exprPos + (x < 0) & exprNeg;

x = az4/v2; n0=x*x*2; n1=2*x + n0; n2=2.666666667*n0*x; denum = 2 + n1 + n2; exprPos = 1 - 2/denum; denum = 2 - n1 - n2; exprNeg = -1 + 2/denum; x = (x >= 0) & exprPos + (x < 0) & exprNeg;

ay4 = az4 + k2vg*(tanh1Out - tanh2Out); az4 = ay4;

// 1/2-sample delay for phase compensation amf = (ay4+az5)*0.5; az5 = ay4;

sigout = amf;

That didn't work 😕

Josh · 24 Sept 2006

is x*x*x the same as x^3 and are there any differances in cpu consumption between the two?

Shifrin · 24 Sept 2006

Electrogravtiy wrote: is x*x*x the same as x^3 and are there any differances in cpu consumption between the two?

x*x*x is indeed the same as x^3 - however you can't compare CPU consumption between those operations as the '^' operator is not supported in the code component! Indeed, in C++ the '^' operator is a bitwise exclusive OR operator 😉

Shif.

Josh · 24 Sept 2006

Cheers Shifrin 🙂

Josh · 19 Oct 2006

Kingston wrote: I had a look at the code module and thought of a simple optimisation,

replace all of these:

denum = 1 + 2*x + 2*x*x + 4/3*x*x*x + 1; exprPos = 1 - 2/denum; denum = 1 - 2*x + 2*x*x - 4/3*x*x*x + 1; exprNeg = -1 + 2/denum; x = (x >= 0) & exprPos + (x < 0) & exprNeg;

with these:

n0=x*x*2; n1=2*x + n0; n2=2.666666667*n0*x;

denum = 2 + n1 + n2; exprPos = 1 - 2/denum; denum = 2 - n1 - n2; exprNeg = -1 + 2/denum; x = (x >= 0) & exprPos + (x < 0) & exprNeg;

Or just delete a whole shedload of them. Seems to work but I have no idea exactly why or if its still oversampled or not?

polyin input; polyin cutoff; polyin resonance; polyin sr; polyout sigout;

// filter based on the text "Non linear digital implementation of the moog ladder filter" by Antti Houvilainen // resonance [0..1] // cutoff from 0 (0Hz) to 1 (nyquist)

// constants float pi = 3.1415926535; float v2 = 40000; // twice the 'thermal voltage of a transistor' float sr ; float cfScale = 3; float resScale = 1.5;

// state vars float az1; float az2; float az3; float az4; float az5; float ay1; float ay2; float ay3; float ay4; float amf;

// helper vars float x; // temp var: input for taylor approximations float denum; // temp vars to split complex expressions into simpler ones float exprPos; // bug in code parser? float exprNeg;

float expOut; // output of approximations float tanh1Out; float tanh2Out;

// others float kfc; float kf; float kfcr; float kacr; float k2vg;

kfc = cutoff/cfScale; // sr is half the actual filter sampling rate kf = cutoff/(2*cfScale);

// frequency & amplitude correction kfcr = 1.8730*kfc*kfc*kfc + 0.4955*kfc*kfc - 0.6490*kfc + 0.9988; kacr = -3.9364*kfc*kfc + 1.8409*kfc + 0.9968;

x = -2 * pi * kfcr * kf; // expOut = 1 + x + (x*x)/2 + (x*x*x)/6 + (x*x*x*x)/24 + (x*x*x*x*x)/120 + (x*x*x*x*x*x)/720; expOut = 1 + x + x*x/2 + x*x*x/6 + x*x*x*x/24; expOut = expOut + x*x*x*x*x/120; expOut = expOut + x*x*x*x*x*x/720;

k2vg = v2*(1-expOut); // filter tuning

// cascade of 4 1st order sections x = (input - resScale*resonance*amf*kacr) / v2;

// (x>=0) -> tanh(x); (x< 0) -> -tanh(-x) // tanh1Out = (x >= 0) & ( 1 - 2 / ( 1 + 2*x + 2*x*x + 4/3*x*x*x + 1 ) ) // + (x < 0) & -1 * ( 1 - 2 / ( 1 - 2*x + 2*x*x - 4/3*x*x*x + 1 ) ); denum = 1 + 2*x + 2*x*x + 4/3*x*x*x + 1; exprPos = 1 - 2/denum; denum = 1 - 2*x + 2*x*x - 4/3*x*x*x + 1; exprNeg = -1 + 2/denum; tanh1Out = (x >= 0) & exprPos + (x < 0) & exprNeg;

x = az1/v2;

tanh2Out = (x >= 0) & exprPos + (x < 0) & exprNeg;

ay1 = az1 + k2vg*(tanh1Out - tanh2Out); az1 = ay1;

x = ay1/v2;

tanh1Out = (x >= 0) & exprPos + (x < 0) & exprNeg;

x = az2/v2;

tanh2Out = (x >= 0) & exprPos + (x < 0) & exprNeg;

ay2 = az2 + k2vg*(tanh1Out - tanh2Out); az2 = ay2;

x = ay2/v2;

tanh1Out = (x >= 0) & exprPos + (x < 0) & exprNeg;

x = az3/v2;

tanh2Out = (x >= 0) & exprPos + (x < 0) & exprNeg;

ay3 = az3 + k2vg*(tanh1Out - tanh2Out); az3 = ay3;

x = ay3/v2;

tanh1Out = (x >= 0) & exprPos + (x < 0) & exprNeg;

x = az4/v2;

tanh2Out = (x >= 0) & exprPos + (x < 0) & exprNeg;

ay4 = az4 + k2vg*(tanh1Out - tanh2Out); az4 = ay4;

// 1/2-sample delay for phase compensation amf = (ay4+az5)*0.5; az5 = ay4;

// oversampling (repeat same block) x = (input - resScale*resonance*amf*kacr) / v2;

exprNeg = -1 + 2/denum; // (x>=0) -> tanh(x); (x< 0) -> -tanh(-x) // tanh1Out = (x >= 0) & ( 1 - 2 / ( 1 + 2*x + 2*x*x + 4/3*x*x*x + 1 ) ) // + (x < 0) & -1 * ( 1 - 2 / ( 1 - 2*x + 2*x*x - 4/3*x*x*x + 1 ) ); tanh1Out = (x >= 0) & exprPos + (x < 0) & exprNeg;

x = az1/v2;

tanh2Out = (x >= 0) & exprPos + (x < 0) & exprNeg; ay1 = az1 + k2vg*(tanh1Out - tanh2Out); az1 = ay1;

x = ay1/v2;

tanh1Out = (x >= 0) & exprPos + (x < 0) & exprNeg;

x = az2/v2;

tanh2Out = (x >= 0) & exprPos + (x < 0) & exprNeg;

ay2 = az2 + k2vg*(tanh1Out - tanh2Out); az2 = ay2;

x = ay2/v2;

tanh1Out = (x >= 0) & exprPos + (x < 0) & exprNeg;

x = az3/v2;

tanh2Out = (x >= 0) & exprPos + (x < 0) & exprNeg;

ay3 = az3 + k2vg*(tanh1Out - tanh2Out); az3 = ay3;

x = ay3/v2;

tanh1Out = (x >= 0) & exprPos + (x < 0) & exprNeg;

x = az4/v2; denum = 1 + 2*x + 2*x*x + 4/3*x*x*x + 1; exprPos = 1 - 2/denum; denum = 1 - 2*x + 2*x*x - 4/3*x*x*x + 1; exprNeg = -1 + 2/denum; tanh2Out = (x >= 0) & exprPos + (x < 0) & exprNeg;

ay4 = az4 + k2vg*(tanh1Out - tanh2Out); az4 = ay4;

// 1/2-sample delay for phase compensation amf = (ay4+az5)*0.5; az5 = ay4;

sigout = amf;

:)

Shifrin · 20 Oct 2006

Electrogravtiy wrote: Or just delete a whole shedload of them. Seems to work but I have no idea exactly why or if its still oversampled or not?

The sections you're removing implement what I'd guess is saturation/nonlinearities in the feedback of the filter. Shouldn't affect any oversampling - just the sound of the filter.

[edit]On closer inspection of your code it seems there are still some tanh calculations going on! Nonetheless it's cool to fiddle with these things![/edit]

Shif.

sambean · 20 Oct 2006

Hola peeps..

Here's a primitive version of the moog filter... Easier to see what's going on and stuff.. Based on a paper from tinternet....

http://magalov.myweb.hinet.net/filters/Moog.osm

Sam

Shifrin · 20 Oct 2006

Cheers Sam 😉

Shif.

TheAudiophileDutchman · 21 Oct 2006

Doesn't look very 'primitive' to me 😀 😉

Josh · 21 Oct 2006

Nice one Sam 😀

Tom7777 · 21 Oct 2006

Nice one, many thanks 🙂

Josh · 22 Oct 2006

sambean wrote: Hola peeps..

Here's a primitive version of the moog filter... Easier to see what's going on and stuff.. Based on a paper from tinternet....

http://magalov.myweb.hinet.net/filters/Moog.osm

Sam

Does it come in a poly version? A dumb question perhaps: Does filtering a mono stream use less cpu than filtering a poly stream?

sambean · 23 Oct 2006

Electrogravtiy wrote: sambean wrote: Hola peeps..

Here's a primitive version of the moog filter... Easier to see what's going on and stuff.. Based on a paper from tinternet....

http://magalov.myweb.hinet.net/filters/Moog.osm

Sam

Does it come in a poly version? A dumb question perhaps: Does filtering a mono stream use less cpu than filtering a poly stream?

It should be compatible with poly streams too, I may have 'Right-clicked' some of the connector types to mono, but those can be clicked back.. Just plug it into a poly stream, if the connectors don't change white, right click on them..

If there are less than 4 voices playing the the cpu should be the same, more than then the mono one will take less cpu...

Sam

TheAudiophileDutchman · 23 Oct 2006

@Sam: what would be the most efficient way to get a Moog HP/BP out of your example?

electrogear · 12 Nov 2006

Does it come in a poly version? A dumb question perhaps: Does filtering a mono stream use less cpu than filtering a poly stream?

I don't know the reason why, as I'm still trying to get to grips with what connectors I should be using at what times, but I've found that using the EQ with Poly gives me very high CPU (around 12% per EQ module), and with Mono quite a bit less. I'm still trying to work out where to use poly signals for effects, and what benefits there are over processing poly signals rather than Mono - It'a all a bit confusing... think I need to read up a bit more....

Cheers,

Mike

matti · 12 Nov 2006

electrogear wrote: Does it come in a poly version? A dumb question perhaps: Does filtering a mono stream use less cpu than filtering a poly stream?

I don't know the reason why, as I'm still trying to get to grips with what connectors I should be using at what times, but I've found that using the EQ with Poly gives me very high CPU (around 12% per EQ module), and with Mono quite a bit less. I'm still trying to work out where to use poly signals for effects, and what benefits there are over processing poly signals rather than Mono - It'a all a bit confusing... think I need to read up a bit more....

Cheers,

Mike

The difference between poly and mono signals is really quite simple. If you think of a 16 voice synth that has an eq connected as poly, it processes every 16 voices separately with that eq. That means it takes 16x more power to process than the mono one(SSE might affect this, but basicly you can think this way). If the effect is in mono it only uses one instance of the eq. The "poly to mono" element mixes all the polyphonic voices together and makes it into one mono channel. So you can think of the mono example just like a mixer channel with an eq. Eq isn't really needed as a poly effect.