18.08.2013 Views

Soner Bekleric Title of Thesis: Nonlinear Prediction via Volterra Ser

Soner Bekleric Title of Thesis: Nonlinear Prediction via Volterra Ser

Soner Bekleric Title of Thesis: Nonlinear Prediction via Volterra Ser

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.

3.2. NONLINEAR MODELING OF TIME SERIES 33<br />

Equation (2.9) (linear prediction model) is a time-invariant linear system where<br />

I have replaced the output by the one-step ahead prediction <strong>of</strong> the input. Similarly,<br />

I can consider equation (3.12) and construct a time-variant nonlinear prediction<br />

operator with a nonlinear <strong>Volterra</strong> model <strong>of</strong> the form (order <strong>of</strong> p, q, r):<br />

xn =<br />

+<br />

+<br />

p<br />

ai xn−i<br />

i=1<br />

q q<br />

bjk xn−jxn−k<br />

j=1 k=1<br />

r r r<br />

clms xn−lxn−mxn−s<br />

l=1 m=1 s=1<br />

+ εn , (3.13)<br />

where ai, bjk, and clms are the linear, the nonlinear quadratic, and the nonlinear<br />

cubic impulse responses <strong>of</strong> the nonlinear system, respectively, also known as <strong>Volterra</strong><br />

kernels. The first term on the right hand side <strong>of</strong> equation (3.13) represents the<br />

classical linear prediction problem with a prediction operator <strong>of</strong> order p. The second<br />

and third terms <strong>of</strong> equation (3.13) represent the expansion <strong>of</strong> the signal in terms<br />

<strong>of</strong> quadratic and cubic nonlinearities. The modeling error is given by εn. Note<br />

that equation (3.13) is the nonlinear AR formulation as an extension <strong>of</strong> linear AR<br />

modeling.<br />

The last expression can also be written in prediction error form<br />

εn = xn −<br />

−<br />

−<br />

p<br />

ai xn−i<br />

i=1<br />

q q<br />

bjk xn−jxn−k<br />

j=1 k=1<br />

r r r<br />

clms xn−lxn−mxn−s . (3.14)<br />

l=1 m=1 s=1

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

Saved successfully!

Ooh no, something went wrong!