Scribd Logo
Upload

Welcome to Scribd!

  • 90 views

    UConn ECE6439 2020 Midterm

    Uploaded by

    Nemo Outis
    1) The likelihood function for the observation z=x+w, where w is Laplace distributed, is given. The derivative of the log-likelihood is not defined when z=x, but the log-likelihood is differentiable almost surely. 2) The Fisher information is calculated to be 1 by evaluating the integral of the square of the log-likelihood derivative. The Cramér-Rao lower bound is 1. 3) The maximum likelihood estimate is x̂(z)=z, which is an unbiased estimate. However, its variance is 2, so it is not efficient.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd

    UConn ECE6439 2020 Midterm

    Uploaded by

    Nemo Outis
    90 views3 pages
    1) The likelihood function for the observation z=x+w, where w is Laplace distributed, is given. The derivative of the log-likelihood is not defined when z=x, but the log-likelihood is differentiable almost surely. 2) The Fisher information is calculated to be 1 by evaluating the integral of the square of the log-likelihood derivative. The Cramér-Rao lower bound is 1. 3) The maximum likelihood estimate is x̂(z)=z, which is an unbiased estimate. However, its variance is 2, so it is not efficient.

    Original Description:

    ECE6439 Estimation. Author Yaakov BarShalom.

    Original Title

    UConn ECE6439 2020 midterm

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    1) The likelihood function for the observation z=x+w, where w is Laplace distributed, is given. The derivative of the log-likelihood is not defined when z=x, but the log-likelihood is differentiable almost surely. 2) The Fisher information is calculated to be 1 by evaluating the integral of the square of the log-likelihood derivative. The Cramér-Rao lower bound is 1. 3) The maximum likelihood estimate is x̂(z)=z, which is an unbiased estimate. However, its variance is 2, so it is not efficient.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    90 views3 pages

    UConn ECE6439 2020 Midterm

    Uploaded by

    Nemo Outis
    1) The likelihood function for the observation z=x+w, where w is Laplace distributed, is given. The derivative of the log-likelihood is not defined when z=x, but the log-likelihood is differentiable almost surely. 2) The Fisher information is calculated to be 1 by evaluating the integral of the square of the log-likelihood derivative. The Cramér-Rao lower bound is 1. 3) The maximum likelihood estimate is x̂(z)=z, which is an unbiased estimate. However, its variance is 2, so it is not efficient.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    You are on page 1of 3

    probs/362mt20 March 9, 2020 Name (legible): .......................................

    ECE6439(362) MT — 2003
    Open book; Please show all your work and make sure to put page numbers. Use only one side of each sheet.

    1. Given the observation


    z =x+w
    where x is an unknown parameter and w is a Laplace distributed noise with pdf

    (1/2)ew
    
    w≤0
    pw (w) =
    (1/2)e−w w ≥ 0

    (i) Write the likelihood function (LF) Λ(x; z) of x based on z. Hint: Use the given pw with a suitable
    argument.
    (ii) Write the derivative of log-LF (LLF) w.r.t. x. Is this derivative defined for every z?
    (iii) Can one say that the LLF is differentiable almost surely? Justify.
    (iv) Evaluate the Fisher information (FI) for this problem. Hint: Carry out the expectation as the sum of
    two integrals, one for each region where you have an explicit expression of the derivative of the LLF. Ignore
    the point at which the derivative is not defined. What is the CRLB for this problem?
    (v) Find the MLE x̂(z). Hint: Sketch the LF to find its maximum. Justify your answer.
    (vi) Find E[x̂(z)]. Is this estimate unbiased?
    (vii) Evaluate the variance of this estimate. Is this estimate efficient?

    2. Given the scalar MMMSE estimation problem for x based on he observation

    z =x+w

    with noise
    w ∼ N (0, Pww )
    and the prior
    x ∼ N (x̄, Pxx )
    with x and w independent.
    (i) Find Pxz and Pzz in terms of Pxx and Pww .
    (ii) Assume Pxx is not known. Given Pxx|z and Pww , find Pxx
    (iii) Assume x̄ is not known, Find x̄ if x̂ = E[x|z] is known together with z, Pxx and Pww .
    Solution 362MT2003
    1. Given the observation
    z =x+w
    where x is an unknown parameter and w is a Laplace distributed noise with pdf

    (1/2)ew
    
    w≤0
    pw (w) = −w
    (1/2)e w≥0

    (i) Write the likelihood function (LF) Λ(x; z) of x based on z. Hint: Use the given pw with a suitable
    argument.
    (1/2)ez−x z ≤ x
    
    Λ(x; z) = p(z|x) = pw (z − x) =
    (1/2)ex−z z ≥ x
    (ii) Write the derivative of log-LF (LLF) w.r.t. x. Is this derivative defined for every z?
    
    ln(1/2) + z − x z≤x
    ln p(z|x) =
    ln(1/2) + x − z z≥x
    
    ∂ ln p(z|x) −1 z<x
    =
    ∂x 1 z>x
    It is not defined for z = x.
    (iii) Can one say that the LLF is differentiable almost surely? Justify.
    It is not differentiable at z = x, but this point in the z-space has probability zero so it is differentiable
    w.p.1, i.e., almost surely.
    (iv) Evaluate the Fisher information (FI) for this problem. Hint: Carry out the expectation as the sum
    of two integrals, one for each region where you have an explicit expression of the derivative of the LLF. What
    is the CRLB for this problem?
    Z x Z ∞ Z ∞
    2 2
    J(x) = (−1) p(z|x) dz + (1) p(z|x) dz = p(z|x) dz = 1
    −∞ x −∞

    Note that J is a “Lebesgue integral” w.r.t. the probability measure p(z|x) dz and a zero probability point
    (at which the integrand is not defined and there is no point mass) does not contribute to the integral.
    The conventional (Riemann) integral (taught in calculus) does not have an answer to such a problem.
    The CRLB is 1.
    (v) Find the MLE x̂(z). Hint: Sketch the LF to find its maximum. Justify your answer.

    x̂(z) = z

    is the peak of the LF.


    (vi) Find E[x̂(z)]. Is this estimate unbiased?

    E[z] = x

    It is unbiased because of the symmetry of the pdf pw (w) (its mean is zero).
    (vii) Evaluate the variance of this estimate. Is this estimate efficient?
    Z ∞
    var[x̂] = (z − x)2 p(z|x) dz = 2
    −∞

    i.e., not efficient.

    2
    2. Given the scalar MMMSE estimation problem for x based on he observation

    z =x+w

    with noise
    w ∼ N (0, Pww )
    and the prior
    x ∼ N (x̄, Pxx )
    with x and w independent.
    (i) Find Pxz and Pzz in terms of Pxx and Pww .

    Pxz = Pxx Pzz = Pxx + Pww

    (ii) Assume Pxx is not known. Given Pxx|z and Pww , find Pxx
    Since this is a scalar problem, one can use the “parallel resistors” formula to solve for Pxx
    Pxx|z Pww
    Pxx =
    Pww − Pxx|z

    (iii) Assume x̄ is not known, Find x̄ if x̂ = E[x|z] is known together with z, Pxx and Pww .
    Similarly, since z̄ = x̄, one can solve for x̄ the “update” equation that yields x̂
    −1
    x̂ − Pxz Pzz z Pxx
    x̄ = −1 = P (x̂ − z) + x̂
    1 − Pxz Pzz ww

    You might also like

    pFad - Phonifier reborn

    Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

    Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


    Alternative Proxies:

    Alternative Proxy

    pFad Proxy

    pFad v3 Proxy

    pFad v4 Proxy