Prev: GUI for driving voltages in a scan
Next: Help with correlation
From: Irina on 11 Aug 2010 11:37
Hello,

Could please anyone tell me if there might be a huge in outcome results of correlation if two variables are processed with two different filters?
Specifically, we compare NIR data and BOLD signal. NIR data is filtered using MATLAB filtfilt function, and then correlated with BOLD, which is processed using AFNI 3dFourier filter.

Do you think that results of correlation will be SIGNIFICANTLY improved if both signal are passed through the same filter?

Thank you very much,
Irina
From: Wayne King on 11 Aug 2010 12:05
"Irina " <irina.schelkanova(a)ryerson.ca> wrote in message <i3ug3g$7te$1(a)fred.mathworks.com>...
> Hello,
>
> Could please anyone tell me if there might be a huge in outcome results of correlation if two variables are processed with two different filters?
> Specifically, we compare NIR data and BOLD signal. NIR data is filtered using MATLAB filtfilt function, and then correlated with BOLD, which is processed using AFNI 3dFourier filter.
>
> Do you think that results of correlation will be SIGNIFICANTLY improved if both signal are passed through the same filter?
>
> Thank you very much,
> Irina

Hi, Irina the frequency response of the zero-phase filter obtained in filtfilt() by filtering the data, reversing it, and filtering it again has a magnitude response that is the square of the original filter's magnitude response and a phase response of zero. This results in a passband ripple and stopband attenuation that is twice that of the original filter. As a simple example:

h = fdesign.lowpass('Fp,Fst,Ap,Ast',0.2,0.25,0.5,40);
d = design(h);
b = d.Numerator;
[H,W] = freqz(b,1,512);
plot(W,20*log10(abs(H)),'k','linewidth',2);
hold on;
plot(W,40*log10(abs(H)),'b','linewidth',2);
legend('Original Filter','zero-phase filter');

If you zoom in on the passband you will see the passband ripple of the zero-phase filter is twice that of the original filter.

Without knowing the filter specifics and signal characteristics it's hard to say what's big, but hopefully that describes the difference. Of course the output of the conventional filter is delayed based on the phase response while the phase response of the zero-phase filter is zero.

Hope that helps,
Wayne
 | 
Pages: 1
Prev: GUI for driving voltages in a scan
Next: Help with correlation