Consider a sequence of n dimensional vector

This preview shows page 1. Sign up to view the full content.

This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: Model (HMM). A HMM can be either vector or scalar. Example 3: Continuing Examples 1 &amp; 2, let h = [h0 , h1 , · · · , hL−1 ]T and consider the observation random process rn = hT S n + nn (9) where nn is a zero-mean AWGN process. Since S n is a Markov chain, rn is a HMM. Consider a sequence of N -dimensional vector observations of the form r n = f n (S n ) + nn n = 1, 2 , · · · , K (10) where f n (S n ) represents N functions of S n which in general can be time varying, and the nn are AWGN vectors. Let r K = [r 1 , r2 , · · · , rK ] and S K = 1 1 [S 1 , S 2 , · · · , S K ]. The joint PDF of a sequence S K , conditioned on observations 1 r K , is 1 p(rK /S K ) P (S K ) 1 1 1 P (S K /rK ) = . (11) 1 1 K p (r 1 ) Given Eq (10) and AWGN nn , we have that K p(r K /S K ) = 1 1 N =1 p(rn /S n ) . (12) Assuming S n is a Markov chain, we have P (S K /rK ) 1 1 . = K p(rn /S n ) P (S n /S n−1 ) · P (S 0 ) n=1 (13) . where “=” denotes proportional to (i.e. p(r K ) in the denominator of Eq (11) is 1 not a function of S K ). 1 Kevin Buckley - 2007 5 MAP and ML Sequence Estimation for...
View Full Document

This document was uploaded on 10/12/2009.

Ask a homework question - tutors are online