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.

8.3. A REPARAMETERIZATION OF THE MODEL REDUCTION PROBLEM 143<br />

This allows us <strong>to</strong> employ a joint framework in which <strong>to</strong> study the H 2 model-order<br />

reduction problem for various co-orders k.<br />

In the co-order k case equation (8.23) involves N equations and k − 1 additional<br />

constraints, in terms of N + k − 1 unknowns (the N coefficients of ã(s) and the k − 1<br />

unknown parameters ρ 1 ,...,ρ k−1 in ρ(s)).<br />

We now reparameterize by introducing the N quantities x i := ã(δ i ), for i =<br />

1,...,N, <strong>to</strong> characterize the polynomial ã(s). Note that a polynomial of degree N −1<br />

has N degrees of freedom and is fully characterized by its values at N distinct points.<br />

These values are related <strong>to</strong> the coefficients of the polynomial ã(s) inalinear way, by<br />

means of a Vandermonde matrix involving the interpolation points. To be explicit,<br />

let V (β 1 ,β 2 ,...,β N ) denote a Vandermonde matrix in the variables β 1 ,...,β N :<br />

⎛<br />

V (β 1 ,β 2 ,...,β N )=<br />

⎜ .<br />

.<br />

.<br />

⎝<br />

1 β N βN 2<br />

1 β 1 β 2 1 ... β N−1<br />

1<br />

1 β 2 β 2 2 ... β N−1<br />

2<br />

.<br />

... βN−1<br />

N<br />

⎞<br />

. (8.26)<br />

⎟<br />

⎠<br />

As in [50], note that the quantities x i := ã(δ i )=ã N−1 δ N−1<br />

i + ...+ã 1 δ i +ã 0<br />

(for i =1,...,N) depend linearly on the coefficients ã 0 ,...,ã N−1 through the fixed<br />

Vandermonde matrix V (δ 1 ,...,δ N ) in the following way:<br />

⎛<br />

⎜<br />

⎝<br />

x 1<br />

x 2<br />

. .<br />

x N<br />

⎞ ⎛<br />

⎟<br />

⎠ = ⎜<br />

⎝<br />

ã(δ 1 )<br />

ã(δ 2 )<br />

.<br />

ã(δ N )<br />

⎞<br />

⎛<br />

⎟<br />

⎠ = V (δ 1,...,δ N ) ⎜<br />

⎝<br />

ã 0<br />

ã 1<br />

. .<br />

ã N−1<br />

⎞<br />

⎟<br />

⎠ . (8.27)<br />

Likewise the quantities ã(−δ i )=ã N−1 (−δ i ) N−1 + ...+ã 1 (−δ i )+ã 0 in Equation<br />

(8.23) depend linearly on the coefficients of ã(s) <strong>to</strong>o. This relation can be expressed,<br />

by using the matrix V (−δ 1 ,...,−δ N ), as follows:<br />

⎛<br />

⎜<br />

⎝<br />

ã(−δ 1 )<br />

ã(−δ 2 )<br />

.<br />

ã(−δ N )<br />

⎞<br />

⎛<br />

⎟<br />

⎠ = V (−δ 1,...,−δ N ) ⎜<br />

⎝<br />

ã 0<br />

ã 1<br />

.<br />

ã N−1<br />

⎞<br />

⎟<br />

⎠ . (8.28)<br />

In this expression existence of the inverse of the Vandermonde matrix V (δ 1 , ..., δ N )<br />

is guaranteed by the poles δ 1 ,δ 2 , ...,δ N being all distinct, as its determinant is given<br />

by:<br />

det V (δ 1 ,...,δ N )= ∏ (δ i − δ j ). (8.29)<br />

i>j<br />

Combining Equations (8.27) and (8.28) allows <strong>to</strong> express the quantities ã(−δ i )in<br />

terms of x 1 ,...,x N through fixed Vandermonde matrices involving the zeros δ 1 ,...,δ N

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

Saved successfully!

Ooh no, something went wrong!