Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Scribd
Upload
51 views

Exercise 2 Question + MATLAB

This document contains 4 exercises related to statistical signal processing: 1. Shows that the regularity condition does not hold for data with a uniform distribution, so the CRLB cannot be applied. 2. Derives (a) the log-likelihood function, (b) the CRLB, and (c) the MVU estimator for data modeled as a signal in white Gaussian noise. 3. Asks to (a) check if the regularity condition holds, (b) find the CRLB, and (c) give the MVU estimator for data with a Poisson distribution. 4. Asks to find the CRLB for the parameter ρ using the given Fisher information, for a

Uploaded by

Houssam Menhour
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
51 views

Exercise 2 Question + MATLAB

This document contains 4 exercises related to statistical signal processing: 1. Shows that the regularity condition does not hold for data with a uniform distribution, so the CRLB cannot be applied. 2. Derives (a) the log-likelihood function, (b) the CRLB, and (c) the MVU estimator for data modeled as a signal in white Gaussian noise. 3. Asks to (a) check if the regularity condition holds, (b) find the CRLB, and (c) give the MVU estimator for data with a Poisson distribution. 4. Asks to find the CRLB for the parameter ρ using the given Fisher information, for a

Uploaded by

Houssam Menhour
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

521348S Statistical Signal Processing I

Exercise 2

1. If the data x[n] for n = 0, 1, 2, ...N − 1 are I.I.D with uniform distribution U [0, θ], show
that the regularity condition does not hold such as:
∂ ln p(X; θ)
" #
E ̸= 0 for θ > 0
∂θ

Hence the CRLB can not be applied to this problem.


2. The data
x[n] = Arn+4 + w[n]
for n = 0, 1, 2, ...N − 1 are observed, where w[n] is WGN with variance σ 2 and r > 0
is known.

(a) Write the log-likelihood function for the observed vector X. Write all terms exactly
and simplify.
(b) Derive the CRLB for the parameter A by using the theory of signals in WGN.
(c) Derive the MVU estimator for A.
Note: you can also solve (b) or (c) using the standard approach. Final answer
must be simplified anyways.

3. Data x[n], n = 1, · · · , N is observed as IID RVs where x[n] has Poisson distribution
with probability density function
λx[n] −λ
, x[n] > 0
(
e
p (x[n]; λ) = x[n]!
0 , x[n] < 0
with expected value E[x[n]] = λ.

(a) Is regularity condition hold for ∂ln(p(X;λ))


∂λ
.
(b) Find CRLB.
(c) Give MVU estimator for λ.
4. If we observe independent bivariate Gaussian samples x = [x[0], x[1]]T , the noise w =
1 ρ
" #
[w[0], w[1]]T has covariance matrix C = . Let ρ is the scalar parameter in
ρ 1
which x N (0, C(ρ)), then the Fisher Information reduced to
 !2 
N ∂C(ρ)
I(ρ) = tr  C−1 (ρ) . (i)
2 ∂ρ

By using above I(ρ), find CRLB for ρ.

1
Matlab Project Assignment
Instructions*Please read carefully*
Return your solution to Moodle before 20.09.2022 10:30. Late returning
will lead to deduction of points. It is strongly recommended not to leave
returning to last minute but to do this well before the deadline.
Upload your solutions to Moodle as ZIPPED file containing all files needed
to run your codes. Your codes must be ready to run without any additional
user configuration.
Use filename MATLAB_Group.zip . In addition to the well-commented program, make
sure the report includes the following:
• The full name and the student number for each group member.
• Clear answers to the problems posted (including for example figures obtained by
running the simulation program).
Plagiarism is strictly forbidden and will lead to immediate rejection!
You will get separate grade for each MATLAB exercise and you must need passing
grade in order to pass the whole course.

Task
The data
x[n] = θn + w[n]
for n = 0, 1, 2, ...N − 1 are observed, where w[n] is zero-mean WGN with variance σ 2 .
The objective is to estimate θ.
2
It is known that the Cramer-Rao lower bound is CRLB = PNσ−1
n=0
n2

PN −1
x[n]n
(a) Show that the estimator θ̂ = Pn=0
N −1 2 is efficient by Monte Carlo (MC) simu-
n=0
n
lation.
(b) Generate figure that shows that simulated and theoretical probability density
functions agree for the estimator in (a). Make sure that your figures includes
labels and legends.
Hint: See MATLAB Lecture Materials in Moodle for help on this topic.
(c) What happens when N = 1 and N → ∞ ? Explain why.

Use the following parameter for the simulations:


M C = 10000, % Number of Monte Carlo loops
N = 10,
σ2 = 2
θ = 10.

You might also like