The high-pass filter can have a great effect on the subsequent parameter estimation. To explore, how a certain setting (default is 128) influences the time course of the regressor you can use the spm_filter function. First, we build a HRF-regressor:

```reg = [zeros(10,1); ones(10,1); zeros(30,1); ones(10,1); zeros(30,1); ones(10,1); zeros(30,1)]; ``` ```bf = spm_get_bf %TR is 2.2 in our case U.u = reg; U.name = {'reg'}; % create the regressor convolved with the HRF convreg = spm_Volterra(U, bf.bf); ``` In your SPM.mat you can find this kind of regressors under SPM.xX.X.

spm_filter is a function that either creates the low frequencies to be removed or removes those frequencies from the time course. The input variable K defines all necessary parameters:

```K.RT = 2.2; K.row = 1:130; % 130 corresponds to the length of convreg K.HParam = 128; % cut-off period in seconds ```

Here, I defined K.row as it also happens internally. But I guess, ones(1,130) should also work, as it seems that K.row just serves as a mask. Now, we compute the low frequencies that should be removed from convreg.

```nK = spm_filter(K); ```

nK.X0 contains the frequencies that can be removed. A second call of spm_filter removes these frequencies.

```Y = spm_filter(nK, convreg); ``` In SPM.mat these regressors are stored in SPM.xX.nKX and are depicted in the design matrix. As you can see, the high-pass filter cut-off period can greatly affect the regressors that are used for the parameter estimation:

```K.HParam = 256; figure; plot(spm_filter(spm_filter(K), convreg)); ``` ```K.HParam = 64; figure; plot(spm_filter(spm_filter(K), convreg)); ``` 