>Now look at the table. It has a huge DC-offset and its zero crossing
>is not in the middle.

>>>>
 As I understood it I thought pending on the coefficients you use
under the cheb polynomal you could easily end up with a net dc
offset therefore requiring a dc block or high pass function.
These coefficients are adding and subtracting in differing amounts
of power terms, (a3)*4x^3-3x +(a4)*8x^4-8x^2+1 + etc. so the running averages
can end up non-zero. This was in a previous thread.

 Andy Archias