If the TR is 2 seconds, the sampling rate is .5 Hz. That means that the
presence of anything greater than .5 Hz will cause aliasing in the data.
However, the HRF acts as a low pass filter. So if you upsample and then
introduce jitter, then the output of the convolution can't contain anything
greater than .5 Hz. Otherwise, when you downsample, there will be aliasing
noise introduced into the data. The issue is, in your particular
experiment, the level of that aliasing whether it is acceptable. The other
question is that if the filters are going to remove everything above the .5
Hz anyway, I am unsure about why the upsampling is done at all. The
requirement to keep the voxel vs. time data below half the sampling rate is
the Nyquist Theorem. FIR is a digital filter, so its frequency
response can be viewed in Matlab. If the TR is set at .5 sec, then the
sampling rate is 2 Hz and any jittering greater than 1 second would not
introduce aliasing.
Linda Seltzer
UC Davis
|