Detection and Estimation Theory - Assignment 4 | ECE 561, Assignments of Electrical and Electronics Engineering

Material Type: Assignment; Class: Detection & Estimation Theory; Subject: Electrical and Computer Engr; University: University of Illinois - Urbana-Champaign; Term: Spring 2009;

Typology: Assignments

Pre 2010

Uploaded on 02/24/2010

koofers-user-le0
koofers-user-le0 🇺🇸

5

(1)

10 documents

1 / 2

Toggle sidebar

This page cannot be seen from the preview

Don't miss anything!

bg1
ECE 561: Detection and Estimation Theory
Spring 2009
Issued: April 8, 2009
April 8, 2009
1)
Read the treatment of asymptotic robustness and robust detection on
pages 250-263 of Levy, and provide a concise, one-or-two page summary
of the most important results in the text.
2)
Under which conditions on the pdf’s and loss function does there exist
a natural ordering among the performances of MMSE, MAE, and MAP
estimators (in class, we showed what the conditions are for all three to be
identical. This questions asks you to establish when can you claim one to
be strictly larger than the other two, for all possible pairs and one triple).
3)
Consider the modified LLS problem, in which you seek an estimator of
Xof the form ˆ
X=AY+b, but with Arequired to be upper or lower
triangular. As an example, you could also consider another form of a LLS
estimator with structural constraints on A. How would you solve this
problem? When does the LLS estimator allow for Ato be triangular?
4)
Prove the following statements about the Schur complement D/M of a
matrix Dwith respect to the block matrix M, where Mconsists of blocks
A, B, C, D, respectively:
a) det(M) = det(D)×det(D/M).
b) The inertia of a Hermitian matrix Mis an ordered triple, IM=
(m>, m<, m0), where m>,m<, and m0denote the number of positive,
negative, and zero-valued eigenvalues of M. Show that
IM=ID+ID/M .
1
pf2

Partial preview of the text

Download Detection and Estimation Theory - Assignment 4 | ECE 561 and more Assignments Electrical and Electronics Engineering in PDF only on Docsity!

ECE 561: Detection and Estimation Theory

Spring 2009

Issued: April 8, 2009

April 8, 2009

Read the treatment of asymptotic robustness and robust detection on pages 250-263 of Levy, and provide a concise, one-or-two page summary of the most important results in the text.

    Under which conditions on the pdf’s and loss function does there exist a natural ordering among the performances of MMSE, MAE, and MAP estimators (in class, we showed what the conditions are for all three to be identical. This questions asks you to establish when can you claim one to be strictly larger than the other two, for all possible pairs and one triple).
    Consider the modified LLS problem, in which you seek an estimator of X of the form Xˆ = A Y + b, but with A required to be upper or lower triangular. As an example, you could also consider another form of a LLS estimator with structural constraints on A. How would you solve this problem? When does the LLS estimator allow for A to be triangular?
    Prove the following statements about the Schur complement D/M of a matrix D with respect to the block matrix M , where M consists of blocks A, B, C, D, respectively: a) det(M ) = det(D) × det(D/M ). b) The inertia of a Hermitian matrix M is an ordered triple, IM = (m>, m<, m 0 ), where m>, m<, and m 0 denote the number of positive, negative, and zero-valued eigenvalues of M. Show that

IM = ID + ID/M.

c) Can you explain in an intuitive way how the Schur complement arises in the expressions for the LLSE error-covariance matrix?

Assume that, as in class, D is a non-singular matrix.

    a) In class, we showed that there exists a fundamental relationship be- tween the variance of an estimator, and the Fisher information over the pdf of the observables (the Cramer-Rao inequality). Are there natural extensions for bounding the third order moment of the estimation error? What kind of properties would the ”information measure” have to have to make this plausible?

b) Find the Fisher information contained in n independent Bernoulli trials. c) Write a short report on the so-called Jeffreys prior distribution, related to Fisher information. Do you see where this prior could be used in our study of estimation theory?