Step three is as follows:
1. use function 'aic' to identify the order of AR process
2. according to the order of AR process, use function 'ar' to estimate the parameters of AR process, for example variable r in your paper.
3. Whitening the AR process: use function 'filter',
for example, for AR(1) in your paper, Yt' = filter([1 r1],1,Xt);
The above mechanism is simple, take the AR(1) as an example, Yt' = Xt'-r1Xt';
Using z transform, the equation can be transformed as Yt' = Xt'(1-r1*(z^-1)), because Xt-1' = Xt'*(z^-1);
4. After calculating the Yt', and the Tt (which you know how to deal with),
Yt is simply the sum of whitened noise Yt' and the linear trend Tt .
The difference between Xt and Yt is just Xt with color noise and Yt with white noise.
Actually, I would recommend you estimate the b in the linear trend associated with AR process, if you are interested in exploiting that, the published paper I wrote as follows explains that in details.
"Long run surface temperature dynamics of an A-OGCM: the HadCM3 4×CO2 forcing experiment revisited" (link: http://www.springerlink.com/content/r07p33551lq854mn/).
Hope this would help