20.05.2014 Views

link to my thesis

link to my thesis

link to my thesis

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

142 CHAPTER 8. H 2 MODEL-ORDER REDUCTION<br />

One may now take the n coefficients of the monic polynomial a(s) <strong>to</strong>gether with<br />

the n coefficients of b(s) and the N − n coefficients of q(s) as the N + n optimization<br />

parameters in conjunction with the N + n equations associated with the terms in<br />

(8.18). This leads <strong>to</strong> a polynomial system of equations, which is structured in the<br />

sense that the coefficients of b(s) show up in a linear fashion, but which may still be<br />

quite hard <strong>to</strong> solve algebraically because the system is not in Gröbner basis form. By<br />

means of a suitable reparameterization this structure can be reworked.<br />

When the model-order is reduced by k from order N <strong>to</strong> order n = N − k, then<br />

the degree of q(s) is at most k − 1. We shall employ the notation:<br />

q(s) =q 0 + q 1 s + ...+ q k−1 s k−1 0. (8.19)<br />

A reparameterization is used which involves the N zeros δ 1 ,...,δ N of the given polynomial<br />

d(s), which upon substitution in<strong>to</strong> Equation (8.18) enables the elimination of<br />

the polynomial b(s) from the problem. As stated before, we here require the technical<br />

assumption that the N zeros δ 1 ,...,δ N of d(s) are all distinct.<br />

We first focus on the generic situation where q 0 0. For i =0, 1,...,k − 1we<br />

define ρ i = qi<br />

q 0<br />

and let the polynomial ρ(s) be:<br />

ρ(s) = q(s) =1+ρ 1 s + ρ 2 s 2 + ...+ ρ k−1 s k−1 . (8.20)<br />

q 0<br />

Note that ρ(s) is of degree ≤ k − 1 and that ρ 0 = q0<br />

q 0<br />

=1.<br />

Then Equation (8.18) can be manipulated as follows. Multiplying both sides by<br />

q 0 and introducing:<br />

ã(s) :=q 0 a(−s) (8.21)<br />

yields:<br />

e(s)ã(−s) − q 0 b(s)d(s) =ã(s) 2 ρ(s). (8.22)<br />

Note that ã(s) is not monic, in contrast <strong>to</strong> a(s).<br />

Evaluation of Equation (8.22) at the N distinct zeros δ 1 ,...,δ N of the polynomial<br />

d(s) gives:<br />

e(δ i )ã(−δ i )=ã(δ i ) 2 ρ(δ i ), (i =1,...,N). (8.23)<br />

Note that in Equation (8.23) it holds that ã(δ i ) 0, and that the equations no longer<br />

involve the polynomial b(s). This equation constitutes a system of N equations in N<br />

unknowns.<br />

When studying the H 2 model-order reduction problem for co-order k, the polynomial<br />

ã(s) is, for future computational convenience, treated as a polynomial of degree<br />

≤ N − 1, while its degree is actually N − k. Fork = 1 we write:<br />

ã(s) =ã N−1 s N−1 +ã N−2 s N−2 + ...+ã 1 s +ã 0 . (8.24)<br />

For k > 1 we use a similar notation, where we then have <strong>to</strong> add the additional<br />

constraints:<br />

ã N−1 = ···=ã N−k+1 =0. (8.25)

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!