Well, I tried comparing it with the ARMA spectrum produced by a good scheme (from HOSA toolbox), and the result of this code is not worth anything.
I've never seen any kind of spectral estimate go so wrong--something must be incorrect in the implementation or the theory.