Scribd
Upload
142 views25 pages

Chapter - 2 Wiener Filters

Uploaded by

saadmuheyfalh
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
142 views25 pages

Chapter - 2 Wiener Filters

Uploaded by

saadmuheyfalh
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 25

Chapter 2

Wiener Filters
Dr. Saad Muhi Falih

Textbook of this chapter


adaptive filters theory and applications, Second Edition, by Behrouz Farhang-
Boroujeny
Wiener Filters
• In this chapter, we study a class of optimum
linear filters known as Wiener filters.
• the concept of Wiener filters is essential as
well as helpful to understand and appreciate
adaptive filters.
• Furthermore, Wiener filtering is general and
applicable to any application that involves
linear estimation of a desired signal sequence
from another related sequence.
2
In the theory of Wiener filters we assumed the following:
I. the underlying signals are assumed to be random
processes.
II. the filter design is done using the statistics obtained
by ensemble averaging. We follow this approach
while doing the theoretical development and analysis
of Wiener filters. However, from the implementation
point of view and, in particular, while developing
adaptive algorithms in later chapters, we have to
consider the use of time averages instead of ensemble
averages. Adoption of this approach in the
development of Wiener filters is also possible, once
we assume all the underlying processes are ergodic.

3
Mean-Squared Error Criterion

4
Mean-Squared Error Criterion

5
Mean-Squared Error Criterion
• Figure 3.1 shows the block schematic of a linear discrete-time
filter W(z) in the context of estimating a desired signal d(n)
based on an excitation x(n).
• Here, we assume that both x(n) and d(n) are samples of infinite
length random processes.
• The filter output is y(n), and e(n) is the estimation error.
• Clearly, the smaller the estimation error, the better the filter
performance. As the error approaches zero, the output of the
filter approaches the desired signal, d(n).
• the question that arises is the following: what is the most
appropriate choice for the parameters of the filter, which
would result in the smallest possible estimation error?
6
• As we want the estimation error to be as small as
possible, a straightforward approach to the design of
the filter parameters appears to be “to choose an
appropriate function of this estimation error as a cost
function and select that set of the filter parameters
which optimizes this cost function in some sense.”
• In choosing a performance function, the following
points have to be considered:
1. The performance function must be mathematically
tractable.
2. The performance function should preferably have a
single minimum (or maximum) point.

7
• The recursive (infinite-impulse response – IIR)
filters, in general, result in performance functions that may
have many minima (or maxima) points,
• whereas the non-recursive (finite-impulse response – FIR)
filters are guaranteed to have a single global minimum (or
maximum) point if a proper performance function is used.
• Because of this, application of the IIR filters in adaptive
filtering has been very limited.
• In Wiener filters, the performance function is chosen to be:

• where E[·] denotes the statistical expectation.


• In fact, the performance function ξ, which is also called mean-
squared error criterion.
8
3.2 Wiener Filter – Transversal, Real-
Valued Case

9
• The filter input, x(n), and its desired output,
d(n), are assumed to be real-valued stationary
processes.

10
11
• This is a quadratic function of the tap-weight vector w with a
single global minimum.
• To obtain the set of tap weights, which minimizes the
performance function ξ, we need to solve the system of
equations that results from setting the partial derivatives of ξ
with respect to every tap weight to zero. That is,

12
13
14
15
16
Example 3.1: Consider the modeling problem shown in
Figure 3.3. The plant is a two-tap filter with an additive
noise, ν(n), added to its output. A two-tap Wiener filter
with tap weights w0 and w1 is used to model the plant
parameters. The same input is applied to both the plant
and Wiener filter. The input, x(n), is a stationary white
process with variance of unity. The additive noise, ν(n),
is zero-mean and uncorrelated with x(n), and its
variance is v = 0.1. We want to compute the optimum
values of w0 and w1, which minimize E[e2(n)].

17
18
19
20
21
3.3 Principle of Orthogonality

• This shows that at the optimal setting of the Wiener


filter tap weights, the estimation error is uncorrelated
with the filter tap inputs, that is, the input samples
used for estimation.
• This is known as the principle of orthogonality.
• As a useful corollary to the principle of orthogonality,
we note that the filter output is also uncorrelated
with the estimation error when its tap weights are set
to their optimal values. 22
where yo(n) is the Wiener filter output when its tap
weights are set to their optimal values. Then, using Eq.
(3.40) in Eq. (3.41), we obtain:

23
3.4 Normalized Performance Function
• It is appropriate if we define the ratio

• as the normalized performance function. We may note that ζ =


1 when y(n) is forced to zero;
• that is, when no estimation of d(n) has been made. It reaches
its minimum value, ζmin, when the filter tap weights are
chosen to achieve the minimum mean-squared error. This is
given by

24
• Noting that ζmin cannot be negative,
• we find that its value remains between 0 and
1.
• The value of ζmin is an indication of the ability
of the filter in estimating the desired output.
• A value of ζmin close to zero is an indication
of good performance of the filter, and a value
of ζmin close to 1 indicates poor performance
of the filter.

25

You might also like