Scribd
Upload
150 views7 pages

1st Assignment PDF

This document contains the solutions to several statistical signal processing problems. It finds: 1) The CRLB and efficient estimator for estimating an amplitude A from noisy data x[n] = Arn + w[n]; 2) That there is no efficient estimator for estimating r from x[n] = rn + w[n]; 3) The MVU estimator and its covariance for estimating amplitudes from a linear model with exponential terms.

Uploaded by

Anonymous KwSOTJLS
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
150 views7 pages

1st Assignment PDF

This document contains the solutions to several statistical signal processing problems. It finds: 1) The CRLB and efficient estimator for estimating an amplitude A from noisy data x[n] = Arn + w[n]; 2) That there is no efficient estimator for estimating r from x[n] = rn + w[n]; 3) The MVU estimator and its covariance for estimating amplitudes from a linear model with exponential terms.

Uploaded by

Anonymous KwSOTJLS
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 7

Statical Signal Processing

Assignment 1

submitted by:
ASHISH RANJAN KARN
12EE35010

Q.1 The data x[n] = Arn + w[n] for n = 0, 1, ...N 1 are observed, where w[n] is WGN with
variance 2 and r > 0 is known.Find the CRLB for A.Show that an efficient estimator exists
and find variance.What happens to the variance as N for various value of r .
ANS:
1
2

P (x; A) =

e 2

NP
1

(x(n)Ar n )2

n=0

(1)

(2 2 ) 2

N 1
ln P (x; A)
2 X
=
(x(n) Arn )(rn )
A
2 2 n=0

N 1
1 X
(x(n) Arn )(rn )
2 n=0

N 1
1 X 2n
2 ln P (x; A)
=
(r )
A2
2 n=0

E[

(5)

2
NP
1

(3)

(4)

N 1
2 ln P (x; A)
1 X 2n
]
=
(r )
A2
2 n=0

>
V ar(A)

(2)

(6)

(r2n )

n=0

To show that efficient estiator exists


N 1
1 X
ln P (x; A)
= 2
(x(n) Arn )(rn )
A
n=0

1
= 2

N
1
X

NP
1
2n

(r )[

n=0

(7)

x(n)(rn

n=0
NP
1

A]

(8)

(r2n )

n=0

= I(A)(A A)

(9)
(10)

Where, A is efficient estimator &

1
I(A)

is a variance.

Q.2 If x[n] = rn + w(n) n = 0, 1, ...N 1 are observed, where w[n] is WGN with variance 2 and r
is to be estimated.Find the CRLB.Does an efficient estimator exists and if so find its variance.

ANS:
1

P (x; r) =

1
2 2

NP
1

(x(n)r n )2

n=0

(11)

(2 2 ) 2

N 1
ln P (x; r)
1 X
= 2
(x(n) rn )(nrn1 )
r
n=0

(12)

For this WGN,We cant put into a form I(r)(


r r).
So ,No efficient estimator can found.
N 1
1 X
2 ln P (x; r)
= 2
[x(n)n(n 1)r2n n(2n 1)r2n2 ]
r2
n=0

E[

N 1
1 X 2 2n2
2 ln P (x; r)
]
=
(n r
)
r2
2 n=0

V ar(
r) >

(13)

(14)
(15)

NP
1

(n2 r2n2 )

n=0

Q.3 If x[n] = A + w(n) n = 0, 1, ...N 1 are observed, where w[n] = [w[0]w[1]..w[N 1]]T .Find
the CRLB.Does an efficient estimator exists and if so find its variance.
ANS:
P (x; A) =

1
N
2

(2) det(C)

1
2

1
1
(xA1)
2 (xA1)C

W here1 = [1111]T

(16)
(17)

1 T 1
ln P (x; A)
=
[x C x 21T C 1 xA + 1T C 1 1A2 ]
A
2 A
= 1T C 1 x 1T C 1 1A
= 1T C 1 1[

1
I(A)

1 C x
A]
1T C 1 1

is a variance.

Q.4 We observe two samples of a DC level in correlated Gaussian noise


x[0] = A + w[0]
x[1] = A + w[1]

(19)

= I(A)(A A)
Where, A is efficient estimator &

(18)

(20)
(21)

Where w = [w[0]w[1]]T is zero mean with covariance matrix

C = 2

The parameter is the correlation coefficient between w[0] & w[0] .Compute the CRLB for
A and compute it to the case when w[n] is WGN or = 0 .
ANS:
(A) T 1 (A)
) C
A
A
T
(A) = [AA]
I(A) = (

(A)
=1
A
1
>
V ar(A)
1T C 1 1


1
1
C 1 = 2
1

2
2
> (1 )
V ar(A)
2 2
2
(1 + )
=
2

(22)
(23)
(24)
(25)
(26)

(27)
(28)
(29)

>
If = 0 ,V ar(A)
2
> 2
If = 0 ,V ar(A)
Q.5 We wish to estimate the amplitudes of exponentials in noise.The observed data are

x[n] =

p
X

Ai ri n + w[n], W here, n = 0, 1, 2, 3...N 1

(30)

i=1

Where w[n] is WGN with variance 2 .Find the MVU estimator of the amplitudes and also
their covariance.Evalute your results for the case when p = 2, r1 = 1, r2 = 1 and N is even.
ANS: This fits a linear model form.

X = H + W

1
1
1
r1
r
r
2
3

H= .
..
..
..
.
.
1
2
3
rN
r
r
1
N 1
N 1

A

A2
=
.

..

Ap
= (H T H)1 H T X

(31)
...
...
..
.

1
rp
..
.

...

p
rN
1

(32)

(33)

(34)

C = 2 (H T H)1

(35)

For p=2, r1 = 1, r2 = 1 &N is even

1
1

H = .
..

1
1

..
.
1 1


N 0
H=
0 N

(36)

= NI

Sincecolumnsareorthogonal.

(37)
(38)

NP
1

N n=0 X[n]
1

= H T X =
1

1 NP
N
n
1 X[n]
N

(39)

n=0

2
C =
I
N

(40)

Q.6 Consider the observation matrix

1
H = 1
1

1
1
1+

Where is small.Compute (H T H)1 and examine what happens as tends to 0 .If x =


[222]T ,find the MVU estimator and describe what happens as tends to 0 .

ANS:


 1
1
1
1
1
1
1
HT H =
1 1 1+
1 1+


3
3+
=
2
3 + 2 + (1 + )


2
2 + (1 + ) (3 + )
(3 + )
3
(H T H)1 =
2
3[2 + (1 + ) ] (3 + )2


2
3 + 2 + () (3 + )
(3 + )
3
=
2
2()
T
1 T

= (H H) H X


2
3 + 2 + () (3 + )


(3 + )
3
6
=
6 + 2
2()2
 
2
=
0

(41)

(42)

(43)

(44)
(45)

(46)
(47)

Q.7 The noise samples are uncorreleted but of unequal variance or


C = diag(0 2 , 1 2 , ....N 1 2 , )
a single n 2 were equal to zero?
Find dn and interpret the results.What would happen to A.if
ANS:
C 1 = DT D

(48)

Since

(49)
2

C = diag(0 , 1 , ....N 1 , )
1
1
1
C 1 = diag( 2 , 2 , ....
,)
0 1
N 1 2
1 1
1
D = diag( , , ....
,)
0 1
N 1
A =

N
1
X

dn X[n]

(50)
(51)
(52)
(53)

n=0

(54)
Where

dn =
=

[D1]n
1 DT D1

(55)

1
n
NP
1

(56)

m=0

m 2

You might also like