Hi Antonio,
Just to be sure, your want to generate a time series that is
basically noise (meaning that you do not plan to put any signal
in it - reponses to stimuli), is that right?
1. Maybe the blur step is the most (only?) important one.
You could synthesize time shifting (if you do that), but volume
registration (or more like un-registration) seems like it would
be quite difficult. It would be easy to apply motion parameters,
but not so easy to simulate natural subject motion (over a TR).
Of course, that depends on what you want to study. If you just
want Monte Carlo simulations of regression with noise, maybe the
blur operation is all that matters.
2. It might make sense to get the stdev from the residual time
series generated by 3dDeconvolve. It makes sense to me to use
the variance without any signal in it. But perhaps that is not
what you prefer.
3. Start with your original dataset (as -a), but don't use 'a'
in the expression. Then make up the data ('t' means time).
Most data that I see processed is done as scaled shorts. So it
seems quite reasonable to me (for what that's worth).
- rick