Homework 9

1. Let {Xi , i = 1, 2, . . .} be a sequence of independent random variables with common mean

E [Xi ] = and common variance Var [Xi ] = 2 . The sample mean based on the first N random
variables in the sequence is
1 N
N = Xi .
N i=1
How large should N be to ensure that


P | N | > 0.05 0.005?
2.

(a) Compare the Chebyshev bound and the exact probability for the event {|X | c}
as a function of c for X distributed uniform on [b, b].
(b) Compare the Chernoff and Chebyshev bounds to the exact probabilities for X a Laplacian
random variable with parameter a = 2 for c = 1, 2, 3, 4, 5. You may use Python, MATLAB,
or whatever mathematical tool you like, to evaluate the Chernoff bound.

3. If the characteristic function of a random variable X is

 
3

2
5
sin 2
e j416 /2 ,
X () =

5 j
determine E[X].
SS-1. Let X be a Poisson random variable with parameter .
(a) Let = 100. Fill out the following table (show how you got each value).
True
Chernoff
Generalized
CLT
probability
bound
Chebyshev Inequality
Estimate
P(X 105)

0.3216

0.8843

1.8

0.31

P(X 120)

0.0282

0.1528

1.4

0.023

SS-2. Problem 4.30 in Stark and Woods:

Let X have a Cauchy pdf
fX (x) =

( 2 + 2 )

Compute the characteristic function (cf) X () of X.

SS-3. Problem 4.31 in Stark and Woods:

Let
Y=

1 N
Xi
N i=1

where the Xi are independent identically distributed Cauchy r.v.s with

fXi (x) =

1
[1 + (x )2 ]

i = 1, . . . , N.

Show that the pdf of Y is

fY (y) =

1
,
[1 + (y )2 ]

that is, is identical to the pdf of the Xi s and independent of N. (Hint: Use the c.f. approach.)
4.

(a) In a radar system used to detect an airplane, a continuous wave signal is transmitted on
two adjacent frequencies and the return signals are processed to produce two outputs
X and Y . If only ground clutter (i.e., no plane) is present, X and Y are independent
Gaussian random variables with mean zero and variance 1/4. If a plane is present,
the X and Y are jointlypGaussian random variables with mean zero, variance one, and
correlation coefficient 3/4.
If the probability that a plane is present is 2/3, determine the maximum a posteriori
(MAP) criterion for deciding whether a plane is present.
(b) In another radar system, the outputs X and Y are always uncorrelated, zero-mean
Gaussian random variables. However, the variance of the random variables is one if no
plane is present and four if a plane is present. If the probability that a plane is present is
2/3, determine the maximum likelihood (ML) criterion for deciding whether a plane
is present.

SS-4. Problem 5.9 in Stark and Woods

SS-5. For each of the following matrices, determine if it is a valid covariance matrix for some
random vector. In the case that it is a valid covariance matrix, determine if the random
variables with that covariance matrix are linearly independent.
(a)

4
1.6 1.6
1
0.5
K1 = 1.6
1.6 0.5 0.694

(b)

4
1.6 1.6
1 0.6
K2 = 1.6
1.6 0.6
1

(c)

4
1.6 1.6
1
0.5
K3 = 1.6
1.6 0.5
1
(d)

4
1.6 1.6
1
0.5
K4 = 1.6
1.6 0.5
1
5. Part (a) of problem 5.10 in Stark and Woods
6. Problem 5.15 in Stark and Woods

299

PROBLEMS

5.6. Let Xi, i = 1,..., n be n mutually orthogonal random vectors. Show that
E

= EE[||Xi||2].

x<
i=l

i=l

(JKnfc Use the definition ||X||2 = XTX.)

5.7. Let Xj,z = 1,... ,n be n mutually uncorrelated random vectors with means /^ =
E[Xi\. Show that
2"

= ^[||Xi-Mi||2].

(*-- M i )
i=i

i=l

5.8. Let Xi,i = 1,... ,n be n mutually uncorrelated random vectors with i[X;] = /xi}
i = 1,..., n. Show that

= EK.

E
i=l

i=l

where Ki [(Xi - &)&* - Mi)T]5.9. Explain why none of the following matrices can be covariance matrices associated
with real random vectors.
2 - 4 0'
4 3 1
0 1 2
(a)
5.10.

4
0
0

6 1+j 2'
1-j
5
-1
2 - 1 6

0
0"
6
0
0 -2

(b)

(c)

4
6
2
6
9
3
9 12 16
(d)

(a) Let a vector X have 2?[X] = 0 with covariance Kxx given by

KXX = [y/2 4
Find a linear transformation C such that Y = CX will have
Kyy =

1 0
0 1

Is C a unitary transformation?
(b) Consider the two real symmetric matrices A and A' given by
A ^

a b
b c

a' b'

A ' b' c'

Show that when a = c and a' = c', the product A A' is real symmetric. More
generally, show that if A and A7 are any real symmetric matrices then AA'
will be symmetric if AA7 = A'A.

PROBLEMS

301

(b) Show that the first-order approximation to the eigenvectors is given by

n

where &# = 0jAK0i/(Ai - A.,) i ^ j and bi{ = 0.

5.14. (a) Let Ai > A2 > ... > An be the eigenvalues of a real symmetric matrix M.
For i > 2, let <i,02>- >0i-i De mutually orthogonal unit eigenvectors
belonging to Ai,..., Xi-i. Prove that the maximum value of uTMu subject
to ||u|| = 1 and u7^ = ... = \iT<j>i_1 = 0 is A*, that is, A* = max(uTMu).
(b) Prove Theorem 5.5-3 using Theorem 5.5-2.
5.15. Let X = (Xi, X2, X3)7, be a random vector with fj, = E[X.\ given by
M=(5,-5,6)T

and covariance given by

"5
K =

2-1
2 5 0
-10
4

Calculate the mean and variance of

Y = ATX + B,
where A = (2, -1,2)T and B = 5.
5.16. Two jointly normal r.v.'s Xi and X2 have joint pdf given by
2
fx1x2(xux2) = 7=exp[-f(x? + |xix2+x^)].
7TV '

Find a nontrivial transformation A in

(SHCS)

such that Fi and y2 are independent. Compute the joint pdf of Yi, Y2.
5.17. Show that if X = (JCU..., Xn)T has mean fj>= (/jlu ..., jzn)T and covariance

then the scalar r.v. Y given by

has mean

and variance