This preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
Unformatted text preview: Chap.11 Spectral Representation 11.1 Factorization and Innovations This section discusses a signal modeling method to represent a real WSS RP ( ) t X as the output of a minimumphase system ( ) L s with an input of white noise process ( ) t i . If ( ) t X can be treated in this way, it is called a regular process. Definition: A LTI system ( ) L s is minimumphase if both ( ) L s and its inverse system 1 ( ) ( ) s L s Γ = are causal and stable. Here, stable means that their impulse responses ( ) t l and ( ) t γ are of finite energy. ( ) L s is minimumphase iff both ( ) L s and ( ) s Γ are analytic in the right half plane of s , i.e., all their poles are in the left half plane of s . A RP is regular if it is linearly equivalent to a whitenoise process ( ) t i in the sense that ( ) ( ) ( ) t t d γ α α α = i X and ( ) ( ) R τ δ τ = ii ( ) ( ) ( ) t t d α α α = X i l { } 2 2 ( ) ( ) E t t dt L = < X l ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) R R τ τ ρ τ δ τ τ τ τ τ = = = XX ii l l l l ( ) ( ) d τ α α α = + l l τ and ( ) ( ) R R τ τ = XX XX ( ) ( ) ( ) S s L s L s = and 2 ( )  ( )  S L ϖ ϖ = ( ) L s is called the innovations filter of ( ) t X , 1 ( ) ( ) s L s Γ = is called the whitening filter of ( ) t X , and ( ) t i is called the innovations of ( ) t X . The problem of determining ( ) L s can be stated as follows: Given a positive even function ( ) S ϖ , find a minimumphase system ( ) L s such that 2  ( ) ( ) L j S ϖ ϖ = . The PaleyWiener condition 2 log ( )  1 S d ϖ ϖ ϖ < + says that an ( ) S ϖ which satisfies the above condition has a solution. Note that ( ) S ϖ cannot contain line spectrum and cannot be bandlimited (which implies predictable instead of regular). This problem is, in general, difficult to be solved. However, for the special case that ( ) L s is a rational function of s , ( ) S ϖ is a rational function of 2 ϖ . It is hence easily to find ( ) S s and decompose it to obtain ( ) L s . This is proved in the following: Since 2 * ( )  ( )  ( ) ( ) ( ) ( ) S L j L j L j L j L j ϖ ϖ ϖ ϖ ϖ ϖ = = = ∴ ( ) ( ) S S ϖ ϖ = This implies that 2 2 ( ) ( ) ( ) A S B ϖ ϖ ϖ = . Changing variable by / s j ϖ = , we obtain 2 2 ( ) ( ) ( ) A s S s B s = . So the poles and zeros of ( ) S s appear in the format of symmetric pairs j s s = and j s (and * j s L * j s for complex j s ). Fig. 112(a) shows an example. In this case, we can extract all poles and zeros in the left half plane and use them to form ( ) L s . To be more specific, we first factorize ( ) S s ( ) ( ) ( ) ( ) ( ) N s N s S s D s D s = ( ) ( ) L s L s = , where ( ) ( ) N s D s contains all poles and zeros in the left half plane (see Fig. 112(b))....
View
Full Document
 Fall '08
 SinHorngChen
 Fourier Series, Autocorrelation, dt, Autoregressive model

Click to edit the document details