031051S Exercise 1

Alejandro Flores ([email protected])

September 8, 2020

1. Consider the discrete Poisson distribution:

λn
Pn = e−λ
n!
n = 0, 1, 2, 3, ... and λ > 0.
Pinf
(a) Show that this distribution is normalized: n=0 Pn = 1.
(b) What is the first moment µ?

(c) What is the variance σ 2 ?

(a)
∞ ∞
X X λn
Pn = e−λ (1)
n=0 n=0
n!
∞
X λn
= e−λ (2)
n=0
n!

Then, the sum presented is a well-known series:

∞
X xn
= ex (3)
n=0
n!

Using 3 in 2, we get:
∞
X
Pn = e−λ eλ (4)
n=0
=1 (5)

(b)
∞
X
µ = E [Pn ] = nPn (6)
n=0
∞
X λn
= ne−λ (7)
n=0
n!

1
031051S Alejandro Flores ([email protected]) Exercise 1

Considering the convention 0! = 1, it’s easy to see that, in 7, the first term of the sum (n = 0) is 0, so,
without losing any information, we write:
∞
X λn
µ = E [Pn ] = ne−λ (8)
n=1
n!
∞
X λn
= e−λ n (9)
n=1
n(n − 1)!
∞
X λn−1 λ
= e−λ (10)
n=1
(n − 1)!
∞
X λn−1
= λe−λ (11)
n=1
(n − 1)!

We perform a change of index k = n − 1 in 11 and we get:

∞
X λk
µ = E [Pn ] = λe−λ (12)
k!
k=0

We use 3 in 12 and we get:

µ = E [Pn ] = λe−λ eλ (13)
=λ (14)

(c) In the following proof we use similar arguments than in the mean, in particular, if the first term of
a sum is 0, we set the sum to start one element after. Also, when helpful, we make index changes
accordingly.
2
σ 2 = E (Pn − µ)2 = E Pn2 − E [Pn ]
   
(15)
∞
X
= n2 Pn − µ2 (16)
n=0
∞
X λn
= n2 e−λ − λ2 (17)
n=1
n!
∞
X λn−1 λ
= e−λ nn − λ2 (18)
n=1
(n − 1)!n
∞
X λn−1
= e−λ λ (n − 1 + 1) − λ2 (19)
n=1
(n − 1)!
∞
X λk
= e−λ λ (k + 1) − λ2 (20)
k!
k=0
∞ ∞
!
−λ
X λk X λk
=e λ k + − λ2 (21)
k! k!
k=0 k=0
∞ ∞
!
k−1 k
Xλ λ X λ
= e−λ λ k + − λ2 (22)
(k − 1)!k k!
k=1 k=0
∞ ∞
!
l k
−λ
X λ X λ
=e λ λ + − λ2 (23)
l! k!
l=0 k=0

2
031051S Alejandro Flores ([email protected]) Exercise 1

We use the expression on 3 and get:

σ 2 = E (Pn − µ)2 = e−λ λ λeλ + eλ − λ2

 

(24)
λ −λ 2
=e e λ (λ + 1) − λ (25)
2
= λ (λ + 1) − λ (26)
2 2
=λ +λ−λ (27)
=λ (28)

2. Let Xn be an IID sequence of Gaussian random variables with zero mean and variance σ 2 ,
and let Yn be
Xn + Xn−1
Yn =
2
(a) What are the mean and covariance of Yn ?

(b) Is Yn a Gaussian random variable?

(a) For the mean, we have:

 
Xn + Xn−1
mY = E[Yn ] = E (29)
2
1
= E [Xn + Xn−1 ] (30)
2
Given the linearity of the expectancy, we have:
1
mY = E[Yn ] = (E [Xn ] + E [Xn−1 ]) (31)
2
Since Xn is IID the mean is time invariant and it is also equal to 0 (zero-mean).
1
mY = E[Yn ] = (0 + 0) = 0 (32)
2

For the covariance, we have:

σY2 = E (Yn − mY )2
 
(33)
= E (Yn − 0)2
 
(34)
" 2 #
Xn + Xn−1
=E (35)
2
1 
= E (Xn + Xn−1 )2

(36)
4
1 
= E Xn2 + Xn Xn−1 + Xn−1 2

(37)
4
1
E Xn2 + E [Xn Xn−1 ] + E Xn−1
   2 

= (38)
4
(39)

3
031051S Alejandro Flores ([email protected]) Exercise 1
 
Since Xn is zero-mean, E Xn2 is the covariance of Xn . Since Xn is IID, Xn and Xn−1 are independent,
and E [Xn Xn−1 ] = E [Xn ] E [Xn−1 ]. Also, its covariance is time-invariant, so we have:
1 2
σY2 = σ + E [Xn ] E [Xn−1 ] + σ 2


(40)
4
1
2σ 2 + (0)(0)


= (41)
4
σ2
= (42)
2

(b) Is Yn a Gaussian random variable?

Solution in figure 1.

Figure 1: Question 2 literal b

3. Consider the stochastic process defined by

Y [n] = X[n] + βX[n − 1]

where β ∈ R and X[n] is a zero-mean WSS process with auto-correlation function given by
Rx [k] = σ 2 α|k| for |α| < 1.


(a) Compute the PSD Py ejω of Yn .

(b) Form the auto-correlation matrix with filter length N = 2, compute cross-correlation
Rxy [n].

Solution in figures 2 and 3.

4
031051S Alejandro Flores ([email protected]) Exercise 1

Figure 2: Question 3, literal a

5
031051S Alejandro Flores ([email protected]) Exercise 1

Figure 3: Question 3, literal b

6
031051S Alejandro Flores ([email protected]) Exercise 1

4. Find the eigenvalues and eigenvectors of the given matrix

 
1 ρ
R= , −1 ≤ ρ ≤ 1
ρ 1

Solution in figure 4.

Figure 4: Question 4

7
031051S Alejandro Flores ([email protected]) Exercise 1

5. Find the derivative of the following functions

(a) f (x) = ||x||2 with respect to complex vector x ∈ CN ×1 .

(b) f (x) = xH Ax with respect to complex vector x ∈ CN ×1 . Note that A is a constant square
matrix.

(c) f (X) = trace(X) with respect to real-valued matrix X ∈ RN ×N .

Solution in figures 5 and 6.

8
031051S Alejandro Flores ([email protected]) Exercise 1

Figure 5: Question 5, literal a and b

9
031051S Alejandro Flores ([email protected]) Exercise 1

Figure 6: Question 5, literal c

10