Problem 4.3
( )
p y
1
1 , 0
0 ,
y 1
otherwise
( )
p y
0
2
3 (y +1) , 0
4
0 ,
otherwise
y 1
a. Find the Bays decision and minimum Bays risk for testing H1
versus H0 with equal a priori probabilities,
=1
and
C C
11
=0
,
C C
10
=
01
=
00
b. Find the minimax decision and minimax risk for the same costs
as in part (a).
c. Find the Neyman-Pearson decision and the for a false-alarm
Pd
probability of
UESTC-何子述等
0
f
1
1
Problem 4.3
Solution:
The likelihood ratio for this problem is
, 0
( )
L y
=
( )
p y
1
( )
p y
2
1
2
3
(y +1)
4
y
1
(a) For this case, the threshold is one. We choose H1 if
( )=
L y
1
2
3
(y +1)
4
1
or
2
y
1
3
Thus, the decision regions are :
R : 0
The Bayes’ risk is:
R : 1/ 3
1/ 3 and
0.4519
y
1)
dy
dy
1
2
(
y
r
1/ 3
1
1/ 3
y
1
0
2
1
2
0
3
4
1
2
UESTC-何子述等
Problem 4.3
(b) For this part, we want to find a threshold such that the two
conditional risks are equal, i.e.,
y
0
0
3(
4
2
y
1)
dy
=
1
y
0
dy
Solving this, we arrive at a cubic equation which must solve, namely
3
y
0
07
y
4 0
Two of the solutions are imaginary, and thus unacceptable. If we
model the cubic equation as
ax b
0
UESTC-何子述等
3
x
3
Problem 4.3
the real root can be written as =x A B
b
2
where
and
b
2
B
A
-
2
b
4
3
a
27
1/3
2
b
4
1/3
3
a
27
0=0.54793
y
resulting in
The average risk is r =0.45207
(c) In this case, we must find a threshold such that
y
0
3 y
4
UESTC-何子述等
0
2
1
dy
f
,
or
3
y
0
03
y
f
4
0
4
Problem 4.3
The solution is
where
and
A
f
2
0=y A B
2
B
f
2
4
f
2
4
f
1/3
1
1/3
1
The probability of detection is
dP
y
0
0
dy
y
0
UESTC-何子述等
5
Problem 4.6
Repeat Problem 4.3for the hypothesis
H
1 : y = s + n,
where s is a fixed real positive number and n has a pdf :
H
0 : y = s + n
( )
p n
1
n
(1
2
)
Solution :
p y H
1
1
y
2
s
1
p y H
0
1
y
2
s
1
UESTC-何子述等
( )
L y
1 (
1 (
y
y
2
2
s
s
)
)
6
Problem 4.6
(a) With equal a priori probabilities and uniform costs, we make
D1 decision if
As s is positive
The Bayes’ risk is then
L y
1
ys
0
y
D
1
D
0
0
r
1
2
0
1
y
dy
2
) ]
s
1
2
[1 (
0
1
[1 (
y
dy
2
) ]
s
1
2
s
1
tan
UESTC-何子述等
7
Problem 4.6
(b) According to the given cost, we have
P
01
P
10
y
0
[1 (
1
y
dy
2
) ]
s
y
0
1
[1 (
y
dy
2
) ]
s
0
y
0
(c) The probability of false alarm is determined by picking a
threshold such that
f
dy
0y
1
[1 (
y
s
2
) ]
y
0
UESTC-何子述等
8