Delft University of Technology

Faculty of Electrical Engineering, Mathematics, and Computer Science

Circuits and Systems Group

ET 4386 Estimation and Detection

Richard C. Hendriks

Example 1 - Lecture 2

Let x1 , ..., xN be iid measurements from a Poisson (λ) distribution with marginal pmf
λ xn
p(xn ; λ) = e−λ ,
xn !
and with expected value E[xn ] = λ.

∂ ln p(xn ;λ)
(a) Calculate ∂λ and show that the regularity condition is satisfied.

(b) Determine the CRLB for Var[λ̂] under the pmf p(x; λ)
(c) Give the MVU estimator for λ.

Example 2 - Lecture 2

Let x[0] and x[1] be two measurements of a constant in correlated Gaussian noise:

x = A1 + w,

with x = [x[0] x[1]]T , w = [w[0] w[1]]T and 1 = [1 1]T . The noise w is zero mean Gaussian
with correlation matrix:  
2 1 ρ
C=σ .
ρ 1
Parameter ρ is the correlation coefficient between w[0] and w[1].
(a) Compute the CRLB for A.
(b) Explain what happens when ρ = 1, ρ = −1 and ρ = 0.
(b) Give the MVU estimator for A.
Answer Example 1 - Lecture 2

N
−λ λxn = e−N λ λ( n=1 xn )
P
QN
(a) p(x; λ) = i=1 e xn !
QN
n=1 xn !
PN
∂ ln p(x;λ) xn
∂λ = −N + n=1
λ
E[ xn ]
h i P
∂ ln p(x;λ) n=1N Nλ
The regularity condition holds as, E ∂λ = −N + λ = −N + λ = 0.
 ∂ 2 ln p(x;λ) − N
P
n=1 xn
(b) = .
h 22
∂λ i λ2
∂ ln p(x;λ) N 1 λ
E ∂λ2
=−λ. The CRLB is then given by Var[λ̂] ≥ h
∂ 2 ln p(x;λ)
i = N
−E
∂λ2

(c) From question (a) we know that

PN
∂ ln p(x; λ) n=1 xn
= −N + .
∂λ λ
This can be rewritten as
 
PN
∂ ln p(x; λ) N  n=1 xn

=  −λ
.
∂λ λ |
|{z} N
{z }
I(λ) λ̂

This is exactly the form

∂ ln p(x; λ)
= I(λ)(λ̂ − λ).
∂λ
PN
xn
The MVU estimator is thus given by λ̂ = n=1
N .

Answer Example 2 - Lecture 2

h i
(a) p(x; A) = (2π|det(C)|)−1/2 exp − 12 (x − A1)T C−1 (x − A1)
∂ ln p(x;A)
∂A = xT C−1 1 − A1T C−1 1
2
∂ ln p(x;A)
∂A2
= −1T C−1 1
2 (1−ρ2 ) 2
Var[Â] ≥ 1T C1−1 1 = σ 2−2ρ = σ2 (1 + ρ)

(b) When ρ = 1, the two noise samples w[0] and w[1] are fully correlated and x[0] and x[1]
will be equal. This means we have only one independent sample and Var[Â] = σ 2 . When
ρ = 0 the two samples are completely independent and we get the result Var[Â] = σ 2 /2.
For ρ = −1, adding x[0] and x[1] will completely cancel the noise (x[0] + x[1] = A) and
thus Var[Â] ≥ 0.
∂ ln p(x;A)
(c) From ∂A = xT C−1 1 − A1T C−1 1 we can write
 
∂ ln p(x; A) T −1  T −1 −1 T −1
= xT C−1 1 − A1T C−1 1 = 1


C 1}  1 C 1 x C 1 −A

∂A | {z | {z }
I(A)
Â

The MVU estimator is thus given by

−1
 = 1T C−1 1 xT C−1 1.