Lecture 5

Spectrum Estimation
Periodogram Based
.Periodogram
.Modified Per.
.Bartlett
.Welch
Classical Methods
(Fourier Transform
Based)
ACF Based
.Blackmen Tukey
Filterbank Approach
.Minimum Variance
Nonclassical Methods
(Non Fourier Based)
Family of Non Parametric Methods
In practice we have one set of finite duration data.
This approach has 2 practical problems
1. Cant do the expected value
2. Cant do the limit

The periodogram is a method that ignores them both
Based on the definition of PSD:
In practice we compute it using the discrete Fourier transform (DFT)
(possibly with zero padding) which computes the discrete time
Fourier transform (DTFT) at discrete frequency points
Periodogram Definition (recalled)
2
1
( ) lim ( )
2 1
M
jwn
x
M
n M
S w E x n e
M

=


=
`
+

)

2
1
0
1

( ) ( )
N
jwn
PER
n
S w x n e
N

=
=

Periodogram-Viewed as a filterbank
Although the Periodogram is Always implemented using
the DFT, a filterbank interpretation is also very helpful.

Defining the impulse response of an FIR filter as


1
, 0 <
( )
0,
i
jnw
i
e n N
h n
N
ow

The frequency response of this filter then becomes

=

=

Periodogram-Viewed as a filterbank (cont)

= +
= =
=

2
( 1)
i
y N
Note that if one chooses n=N-1 the term gives the
periodogram
2 2
1 1
2
2
( 1 ) ( 1)
0 0
2
1
0
1 1
( 1) ( ) ( )
1

( ) ( )
i i i
i
N N
j N k w j N w jkw
i
k k
N
jkw
PER
k
y N x k e e x k e
N N
x k e NS w
N


= =

=
= =
= =

If a WSS random process x(n) is filtered with h

= + = +
= = =

( ) 1
i
jw
i
w w
H e
=
= Since
The power spectrum of x[n] and y[n] are equal at frequency w
i
If the bandwidth of the filter is small enough then the power spectrum of
x[n] may be assumed to be approximately constant over the passband
of the filter, then the power of in yi[n] will be approximately
{ }
2
2
1
[ ] ( ) ( )
2
1
( ) ( )
2
i i
jw jw
i x i
jw jw
x x
E y n S e H e dw
w
S e S e
N
t
t
t
t

=
A
~ =
}
Therefore
{ }
2
( ) [ ]
i
jw
x i
S e NE y n ~
{ }
2

( ) [ ]
i
jw
x i
S e NE y n ~
{ }
2
2

[ ] [ 1]
i i
E y n y N =
Thus if we are able to estimate the power in y
i
[n], then the power
spectrum at frequency w
i
may be estimated as follows
One simple (crude) way to estimate the power is to use a one
point sample average
This is equivalent to
2
1
2
2
0
1
[ 1] [ ]
i
N
jkw
i
k
y N x k e
N

=
=

2
2 1

( ) [ 1] [ ]
i i
jw jkw
x i
S e N y N x k e
N

~ =
Thus
This is equivalent to the periodogram
Basic design steps of MVSE
Design a bank of band pass filters hi[n] with center
frequency wi so that each filter rejects the maxiimum
amount of out of band power while passing the
component at frequency with no distortion

Filter x[n] with each filter in the filterbank and estimate
the power in each output process yi[n]

Set Sx(wi) equal to the power estimated in step 2 divided
by the filter bandwidth
The idea is to steer our filter to a particular frequency
and reject the signal at all other frequencies
{ }
2
2
1
[ ] ( ) ( )
2
1
( ) ( )
2
i i x i
x i x i
E y n S w H w dw
w
S w S w
N
t
t

t
t

= =
A
~ =
}
Goal for the Minimum Variance
Spectral Estimation Method
Figure out a way to design each filter in the filterbank to
minimize the leakage. This will lead to a data dependent design.
Collect Data - then - design Filters

Design Filters that minimize the sidelobes while
keeping the height of the mainlobe at 1.
Design Goals:
1. Want Hi(wi)=1
this will let through the desired Sx(wi).
2. Minimize the total output power in the filter
This is equivalent to minimizing the sidelobe contribution
even though the integral includes the desired frequency wi.
Work on the term
1
0
( ) [ ]
p
jwn
i i
n
H w h n e

=
=

1 1
*
0 0
1 1
* ( )
0 0
1
[ ] [ ] ( )
2
1
[ ] [ ] ( )
2
p p
jkw jlw
i i i x
k l
p p
jw l k
i i x
k l
h k e h l e S w dw
h k h l S w e dw
t
t

t
t

= =

= =
( (
=
( (

=

}

}
( )
1
( ) ( )
2
jw l k
x x
S w e dw r l k
t
t
t

=
}
The frequency response of the filter h
i
[n] is
where p (<N) is the order of the filter
Noting that
1 1
*
0 0
[ ] [ ] ( )
p p
i i i x
k l
H
i x i
h k h l r l k

= =
=
=

h R h
Work on minimizing the Matrix Form
For each i, minimize
H
i i x i
= h R h
( ) 1
i i
H w =
2 ( 1)
[1 . . . ]
i i i
i i
jw j w j p w
i
e e e

=
h e =1
where e
Subject to
Solve the above constrained optimization
problem using Lagrange multipliers
( 1)
i i
H H
x i x
J = h R h h R
i.e., choose h
i
and to minimize J
= c
T
MV Estimate of an AR(1) process
Let x[n] be an AR(1) process
with autocorrelation

The power spectrum of x[n] is ;

Given r
x
[k] for k<p the minimum
variance spectral estimate is:

2
[ ] , <1
1
k
x
r k
o
o
o
=

2 2
1 1
( )
1 2 cos( )
1
x
jw
S w
w
e
o o
o

= =
+

1
1

( )
MV
H
x
p
S w
e R e

+
=
2
2
2
2
1
2
1
(1, , ,..., )
1
1 0 ... 0 0
1 ... 0 0
0 1 ... 0 0
...
0 0 0 ... 1
0 0 0 ... 1
p
x
x
R Toep
R
o o o
o
o
o o o
o o
o o
o

| |
|
+
|
|
+
=
|
|
|
+
|
|

\ .
MV Estimate of an AR(1) process
2
1

( )
2 ( 1)(1 ) 2 cos( )
MV
p
S w
p p w o o
+
=
+ +
Note that minimum variance estimate converges
to the true estimate as p tends to infinity
MW Estimate of a Complex exponential in noise
1
1
( ) ( )
jnw
x n Ae w n = +
1 1
j
A A e
|
=
2
w
o
1
2
1
( ) ( )
jkw
x w
r k A e k o o = +
2
1 1 1
I
H
x w
R A e e o = +
1 1 1
2
1
1, , ,...
T
jw j w jpw
e e e e ( =

Let x(n) be a random phase complex exponential in white noise
where
and | is a rv uniformly distributed over [-t, t].
If the variance of w(n) is
then the autocorrelation of x(n) is
Thus the autocorrelation matrix is

where
Woodbury matrix identity is







where A, U, C and V all denote matrices of the correct size.
Specifically, A is n-by-n, U is n-by-k, C is k-by-k and V is k-
by-n.
1 1 1 4
1 1
1 1
2 2 2
1
1
1 1
2
1
1 1
( 1)
1
H
H
w
x
H
w w w
w
A e e
A
R I I e e
A
p A
e e
o
o o o
o

(
= =
(
+ +

+
1
1
1 1 2 2
1
2
2
1
1 2
1
1 1

( )
1
( 1)
/( 1)
1
( 1)
MV
H
x
H H
w w
w
H
w
p p
S w
e R e
A
e I e e e
p A
A p
e e
p A
o o
o
o

+ +
= =
(

(
+ +

=
+

+ +
Using Woodburys identity

Thus,

MW Estimate of a Complex exponential in noise
MW Estimate of a Complex exponential in noise

1 1
( )
1 1
0 0
( )
p p
jkw jk w w H jkw
R
k k
e e e e e W w w

= =
= = =

2
2
1
1 2
1

( )
/( 1)
1 ( )
( 1)
w
MV
R
w
S w
A p
W w w
p A
o
o
=
+

+ +
1
2
2
1
1
2
1

( ) ( 1)
( 1)
1
( 1)
w
MV w
w w
w
S w p A
A p
p A
o
o
o
=
= = + +
+

+ +
Since

From this expression we see that the MV estimate
attains its maximum at w=w1 with

We can express the minimum variance estimate as
MV Estimate of a Complex exponential in noise

2
2
1 1 1
1

( ) ( )
1 1
w
x x
w S w A
p p
o
o = = +
+ +
2
1 1

( )
x
w A o =
1
2

( )
MV w
w w
S w o
=
~
Therefore the estimate of the power in x(n) at frequency w=w1 is
If the signal to noise ratio is large then
Furthermore if p>>1 and ww1, then W
R
(w-w
1
)~0
function Px = minvar(x,p)
%MINVAR Spectrum estimation using the minimum variance method.
% The spectrum of a process x is estimated using the minimum
% variance method (sometimes called the maximum likelihood method).
% x : Input sequence
% p : Order of the minimum variance estimate - for short
% sequences, p is typically about length(x)/3
% The spectrum estimate is returned in Px using a dB scale.
%
x = x(:);
R = covar(x,p);
[v,d]=eig(R);
U = diag(inv(abs(d)+eps));
V = abs(fft(v,1024)).^2;
Px = 10*log10(p)-10*log10(V*U);
end;

The Minimum Variance Spectral Estimation:
MATLAB code
function R=covar(x,p)
%COVAR Generates a covariance matrix/
%USAGE R=covar(x,p)
%
% Generates a p x p covariance matrix for the sequence x.

%---------------------------------------------------------------

x = x(:);
m = length(x);
x = x - ones(m,1)*(sum(x)/m);
R = convm(x,p)'*convm(x,p)/(m-1);
end;