That's not quite right.
The filter fits a line around each point using local weightings derived from a Gaussian function (of width sigma). The fitted function (only locally a line) forms the low-pass version of the signal and this is removed from the original function when doing
high-pass filtering. Note that the big difference between this and a more traditional high-pass filter is that any global linear component in the data is 100% removed, regardless of the sigma value. Apart from that it shares most of the standard high-pass
filter characteristics as sigma is varied (small sigma gives a closer fit in the low-pass function and hence removes more signal/frequency content).
I hope this helps.
Hello FSL experts,
I have a question about how the high-pass filtering algorithm works. In my pre-stats report it says:
"highpass temporal filtering (Gaussian-weighted least-squares straight line fitting, with sigma=64.0s)"
My understanding is that the procedure first convolves each voxel's time-series with a Gaussian filter, and then subtracts this low-frequency component, leaving only the higher-frequency signal. Is this correct?
Is the sigma mentioned in the report the sigma for the filter in the time domain (in units of seconds), or in the frequency domain (in units of seconds/cycle)?
Thank you in advance,
Connor