DC blocker filter

oddson · 19 Apr 2008

streamin x; streamout y; float a0,a1,a2,b1,b2,yold,yold1,yold2,xold,xold1,xold2; stage(0){

// calculate coefficients a0= -1.08793e+008/(1+(10430.4/0.6735)+1.08793e+008); a1= (2*1.08793e+008)/(1+(10430.4/0.6735)+1.08793e+008); a2= -1.08793e+008/(1+(10430.4/0.6735)+1.08793e+008); b1= (-2*(1.08793e+008-1))/(1+(10430.4/0.6735)+1.08793e+008); b2= (1-(10430.4/0.6735)+1.08793e+008)/(1+(10430.4/0.6735)+1.08793e+008); }

// process input yold2=yold1; yold1=y; y= a2*xold2 + a1*xold1 + a0*x - b2*yold2 - b1*yold1; xold2=xold1; xold1=x;

This works (and seems to do a very good job of it too).

But if I try to sub these values (calculated using Excel from the same formulas as above):

-0.999857659 1.999715319 -0.999857659 -1.999715301 0.999715337

...I get an unstable filter.

Anyone have any suggestions?

infuzion · 20 Apr 2008

oddson wrote: But if I try to sub these values (calculated using Excel from the same formulas as above): -0.999857659 1.999715319 -0.999857659 -1.999715301 0.999715337 ...I get an unstable filter.

double floatsaren't available yet...well you can try in regular ASM.

oddson · 20 Apr 2008

infuzion wrote: double floatsaren't available yet...well you can try in regular ASM.

I know... I'm trying to over-determine the single float values so they will set to bit-level accuracy but it's not working. The max. resolution for floats does not correspond exactly to a certain number of decimal places but I thought SynthMaker's code module should be able to set them if there are more decimals than are needed but it does not seem to be working. I think it's a deficiency in the code module but I'm not 100% sure.

I wonder if I can assign them in ASM at the bit level?

infuzion · 22 Apr 2008

There are only so many bits in a 32bit float you can fill, no matter how you fill it. You could split up the math, but that could get messy & expensive, & the output will still be 32bit.

oddson · 22 Apr 2008

I'm only talking about getting the right 32 bits saved... somehow when I use the above numbers -- which should be more than accurate enough -- I get an unstable filter. But when I use calculation that should lead to the single float equivalents -- I get a filter that works quite well... but I waste a few cycles in stage zero calculating them every time -- even though they are the same every time...

1) this would appear to be a minor bug in the way SM converts decimal values into 32 bit floats 2) is there a workaround to avoid the calculation described?

MyCo · 22 Apr 2008

This could do it:

streamin in; streamout out;

float lastin; float lastout;

out = in-lastin + 0.9999*lastout;

lastin = in; lastout = out;

coeffs are: a0 = 1 a1 = -1 a2 = 0 b1=-0.9999 b2=0

oddson · 22 Apr 2008

Ok... 😊 well that's much simpler and seems to work nicely.

I've also discovered mine is inverting the signal... maybe I should leave filters for the DSP experts. 😉

It still makes me wonder about how SynthMaker's code module assigns decimal values to floats...

MyCo · 22 Apr 2008

if you want to know how a float is stored, you can just assign it to an output, and then connect my stream inspector to it. It can show you the exact hex value.

bootsy · 5 May 2008

And don't forget to scale the mult factor (frequency) depending on your samplerate, should be 1-(2*PI*frequency/samplerate).

hansdump · 18 Jan 2010

of course you can change that coeffs if you want a quick or slow dc adaption...

just have a read at this https://ccrma.stanford.edu/~jos/fp/DC_Blocker.html for more theoretical background.

dc tracking values from 1...0.9 //dctracking: a low value allows // faster tracking

here my asm implementation...

streamin x;streamout y;streamin dctracking;float xm1=0; float ym1=0;

movaps xmm0,dctracking; mulps xmm0,ym1; addps xmm0,x; subps xmm0,xm1;

//Assignment> sLeft=xmm0 movaps y,xmm0; //Assignment> sLeft=x movaps xmm0,x; movaps xm1,xmm0; //Assignment> sLeft=y movaps xmm0,y; movaps ym1,xmm0;

Sry, if I was too lazy to look at yours...

N8!

H.