You will probably find that there is no collinearity in this baseline model. Even the r=1 case,
f1(t)=cos(PI*(t-t0)/(tN-t0)) is not a linear function, which the
-polort 1 default baseline creates.
However, although
t and
f1(t) are linearly independent, they are fairly similar. So it would make sense to use
-polort 0 to have the builtin baseline model reduced to just a constant. This would reduce the chance of numerical problems from near-collinearity.
Have you tried this method with 3dDeconvolve yet? I'm curious to hear your results. If there
is a problem, you could reduce the set to a smaller basis using some sort of PCA.
Also, you can use the little-known AFNI program
1deval to calculate these additional baseline time series.
Zorkon the Magniloquent