CIC Decimation by factor 100000?

[Marcel D. (diablokiller999)]

Hi guys!
I need to filter some samples that come in at 10MHz down to 100Hz, 1KHz 
and 10KHz. I thought using a CIC filter with a FIR afterwards to cut out 
aliased spectral components should be a good idea to do this.
After all, everything should be implemented into a FPGA. But after some 
simulation in Scilab, this seems like a complex job. My -3dB frequency 
is around 3000Hz if I'm not doing sth. wrong in my Scilab program. 
Decimation Rate is 32, Waiting Cycles is also 32 while the number of 
stages is 3.

Should I try some more calculations and implement it the way I planned 
or is there a better solution to my needs I don't see right now?

[Gerald (Guest)]

Hi,

CIC filter is a good way.
I would try to decimate with ~3 concatenated CIC Filter, each with a FIR 
CIC-compensation filter.
If you have access to Matlab, it is a quite easy task, you can even save 
the filter as vhdl.

[signal processing engineer (Guest)]

this is the most crappy idea thinkable to stack several CICs. This all 
can be done witin one step if done correctly. Also stacking FIRs is not 
efficient and can be done in one step with appropriate numbers of TAPs.

[Marcel D. (diablokiller999)]

So was my attempt with decimation down to 1KHz and a FIR to 100Hz a good 
approach? I have a Scilab-Program to determine the attenuation and want 
-3dB frequency to be at 1000Hz, but it crashes everytime I get below 
5KHz ^^
Has anyone a better solution? Maybe for Octave?

clear all;
z=poly(0,'z');
R=256; // CIC divider Ratio
M=3;  // CIC stages
N=1;// CIC delay stages
Scale=0;
num=(z**(R*N)-1)**M;// Based of 1-z**(-R*D)
den=(z**(R*N)-z**((R*N)-1))**M;// Based of 1-z**-1
Pd=syslin('d',num,den);

//Frequency Response
fsample=10e6/2**Scale;// Input sample frequency
fmin = 0.00001;// min frequency
fmax = 1; //max. Frequency (normalized), Nyquist frequency at 0.5
//Pd=Pd**M;
Pdnom= (Pd+0.001)/(R*N)**M;//normalized transfer function
//scf(2); bode(Pdnom,fmin,fmax);

[frq1,rep]=repfreq(Pdnom,fmin,fmax,0.0001);//Frequency response in 0.001 steps

OutdB=10*log(abs(rep));//Amplitude in dB
frq_plot=frq1.*fsample;
scf(3); //clf; 
plot("nl",frq_plot, OutdB,'black');
xlabel('Frequency in Hz');
ylabel('Magnitude in dB');
xgrid(2);
title('Frequency Response of the first and the second CIC decimation Filter ')

[Marcel D. (diablokiller999)]

Tried a bit Octave, found some small programs to deal with. By building 
a CIC with 2048 decimation, 3 stages and 1 delay time I get a graph that 
tells me my -3dB frequency is at 1272Hz. Afterwards I designed a FIR 
with hamming window/1272Hz sampling rate and get around 104Hz @ -3dB and 
very steep attenuation. Does this look like a good design choice, do the 
values even make sense or did I screw up somewhere? ^^
Yeah, I'm fairly new to filters indeed...

[Michael W. (Guest)]

where is the problem using a cic with more then ratio 2048 ?

What is the target for a) 10MHz filtering, ... b) ... c) 100 kHz?

Which shall be the edge freuency?

[Martin O. (Guest)]

If you decimate by a high ratio the number of bits needed in the CIC 
grow,
especiially with higher order CIC stages. Therefore it might be better
to cascade CIC stages.

[Derrick Corea (Guest)]

Thanks for the above code.

[Michael W. (Guest)]

Martin O. wrote:
> If you decimate by a high ratio the number of bits needed in the CIC
> grow,
> especiially with higher order CIC stages. Therefore it might be better
> to cascade CIC stages.

you are not telling us, that some tiny little bits will increas an FPGA 
design much more, than tripeling it by using it three times?

An FPGA decimation with CIC is notinh else than a counter and a sampler 
for the counter and some adds and sub. More bits will increase the 
counter size.