Scribd
Upload
12 views

Detection Theory

Uploaded by

srinivascbit
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
12 views

Detection Theory

Uploaded by

srinivascbit
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/ 24

Introduction to Detection Theory

In our daily life, we are constantly making decisions.


Given some outputs or hypotheses, a criterion is selected, upon which a decision must be
made.
For example, in a radar signal detection problem, the returned signal is observed, and a
decision is made as to whether a target is present or absent.
Detection and estimation theory are fundamental to the design of electronic signal processing
systems for decision making and information extraction.
The electronic signal processing systems include: Communication, Radar, Sonar, Speech,
Image Processing, Biomedicine, Control, Seismology, etc.
The common goal of these systems is to decide when an event of interest occurs and then to
determine more information about that event.
Detection theory deals with the problem of decision making and task of information extraction is
dealt with by estimation theory.
Other names associated with the detection theory are hypothesis testing and decision theory.

What is meant by detection theory?


Answer: It is easy to explain if we look at some physical situations/systems that lead to
detection theory problems.
Example -1: Let’s look at a simple digital communication system shown in Fig. 1.1.

The source puts out a binary digit every T second.


Our object is to transmit this sequence of digits to some other location.

The channel available for transmitting the sequence depends on a particular situation.
Typically, it could be a telephone line, a radio link, or an acoustical channel. For
purposes of illustration, we shall consider a radio link.

In order to transmit the information, we must put it into a form suitable for propagating
over the channel. A straightforward method would be to build a device that generates a
sine wave,
Example -2:
Example-3: In a passive sonar detection system, the receiver listens for noise generated by
enemy vessels. The engines, propellers, and other elements in the vessel generate acoustical
signals that travel through the ocean to the hydrophones in the detection system.
This composite signal can best be characterized as a sample function from a random process.
In addition, the hydrophone generates self-noise and picks up sea noise.
Thus, a suitable model for the detection problem might be

This is the third level of detection problem and is referred to as a random signal in noise
problem.
In these examples we have seen that detection theory problems are characterized by the fact
that we must decide which of several alternatives is true.
There were only two alternatives in the examples cited; therefore, we refer to them as binary
detection problems.
Later in the course, we will encounter problems in which there are M-alternatives available (the
M-ary detection problem).
Next, we will look at some Estimation Theory Examples:
Example-1:

This type of system is called a PAM system.


Example-2:

Example-3:
In binary detection the receiver is either “right” or “wrong.” In the estimation of a continuous
parameter the receiver will seldom be exactly right, but it can try to be close most of the time.

Introduction to classical detection theory


In this lecture we discuss the basic ideas of classical detection theory.
Detection/decision-theory problem-The basic components of a simple decision-theory problem
are shown in Fig.

Let’s first define the various terms:


-The first component is a Source that generates an output. In the simplest form, this output is
one of two choices.
-The choices may be referred to as hypotheses and label them H0 and H1 in the two-choice
case.

-In general, the output might


Some typical source mechanisms are the following:
In the cases of interest to us we do not know which hypothesis is true.

The second and third components of the problem are a probabilistic transition
mechanism and an observation space. The transition mechanism can be viewed as a
device that knows which hypothesis is true. Based on this knowledge, it generates a
point in the observation space according to some probability law.
Example 1:
Let’s take an example to illustrate this idea:

When H1 is true, the source


generates + 1.
When Ho is true, the source
generates - 1.
Example 2:

The source generates


two numbers in
sequence. A random
variable n_1 is added to
the first number and an
independent random
variable n_2 is added to
the second. Thus

The joint probability


density of r_1 and r_2
when H_1 is true is
shown in
Fig. 2.3b (for the case
when n_1=n_2=n as
given in previous
example,
p_r=p_r1xp_r2). The
observation space is
two-dimensional and
any observation
can be represented as a
point in a plane. In
general, the observation space may be N dimensional.
-The fourth component of the detection problem is a decision rule.
-After observing the outcome in the observation space, we shall guess which hypothesis
was true, and to accomplish this we develop a decision rule that assigns each point to
one of the hypotheses.
-Suitable choices for decision rules will depend on several factors which we discuss in
this course later. We will demonstrate how these four components fit together to form
the total decision (or hypothesis-testing) problem.

SIMPLE BINARY HYPOTHESIS TESTS


Let’s consider the decision problem in which each of two source outputs corresponds to a
hypothesis. Each hypothesis maps into a point in the N dimensional observation space. Thus,

Decision Criteria

.
The first and third alternatives correspond to correct choices. The second and fourth
alternatives correspond to errors.
The purpose of a decision criterion is to attach some relative importance to the four
possible courses of action.
It might be expected that the method of processing the received data (r) would depend
on the decision criterion we select. In this lecture, we discuss two most important
decision criterion- Bayes and the Neyman-Pearson.
You can see that all the data processing is involved in computing A(R) and is not affected by a
priori probabilities or cost assignments. This invariance of the data processing is of considerable
practical importance. Frequently the costs and a priori probabilities are merely educated
guesses. The result in (15) enables us to build the entire processor and leave eta as a variable
threshold to accommodate changes in our estimates of a priori probabilities and costs.
Several Special kind of Beys Test:

First Case: Minimum probability of error criterion

Second Case: Minimax Criterion

Once the decision regions Z 0 Z1 are chosen, the values of the integrals in the equation for the
risk are determined. We denote these values in the following manner:
Now if the costs and P1are known, we can find a Bayes test.

Observe that as P1 changes, the decision regions (threshold) for the Bayes test change, and
therefore, PF and PM change, resulGng in a non-opGmum decision rule. A plot of the Bayes
risk, RB as a funcGon of P1 is shown in Fig. below:
Neyman-Pearson Tests:
In many physical situations it is difficult to assign realistic costs or a priori probabilities.

A simple procedure to bypass this difficulty is to work with the conditional probabilities PF and PD.

In general, we should like to make PF, as small as possible and PD as large as possible.

For most problems of practical importance these are conflicting objectives.


An obvious criterion is to constrain one of the probabilities and maximize (or minimize) the other.
A specific statement of this is the Neyman-Pearson criterion:

Neyman-Pearson criterion: constrain PF = α′ ≤ α and design a test to maximize PD (or


minimize PM) under this constraint.

You might also like