detection1.pdf

15
Detection Theory Example 1- Radar 2- Communications 3- Speech 4- Sonar 5- Control 6- ... Denition Assume a set of data {x[0],x[1],...,x[N 1]}  is availab le. T o arriv e at a deci sion, rst we form a function of the data or  T (x[0], x[1],...,x[N 1])  and then make a de- cision based on its value. Determining the function  T  and its mapping to a decision is the central problem addressed in Detection Theory. 1

Transcript of detection1.pdf

Page 1: detection1.pdf

8/14/2019 detection1.pdf

http://slidepdf.com/reader/full/detection1pdf 1/15

Detection Theory

Example

1- Radar

2- Communications

3- Speech

4- Sonar

5- Control

6- . . .

Definition

Assume a set of data {x[0], x[1], . . . , x[N − 1]}   is available. To arrive at a decision,

first we form a function of the data or T (x[0], x[1], . . . , x[N −

1]) and then make a de-

cision based on its value. Determining the function  T  and its mapping to a decision

is the central problem addressed in Detection Theory.

1

Page 2: detection1.pdf

8/14/2019 detection1.pdf

http://slidepdf.com/reader/full/detection1pdf 2/15

Introduction

Example: BPSK phase detection

2

Page 3: detection1.pdf

8/14/2019 detection1.pdf

http://slidepdf.com/reader/full/detection1pdf 3/15

Introduction

Detection Problem

The simplest detection problem is to determine if a signal is present or not. Note

that such a signal is always embedded in noise. This type of detection problem is

called binary hypothesis testing problem. Assuming the received data at time  n to

be x[n], the signal s[n] and the noise w[n], the binary hypothesis testing problem is

defined as follows

H0   :   x[n] = w[n]

H1   :   x[n] = s[n] + w[n]

Note that if the number of hypotheses is more than two, then the problem becomes

a multiple hypothesis testing problem. One example is detection of different digits

in speech processing.

3

Page 4: detection1.pdf

8/14/2019 detection1.pdf

http://slidepdf.com/reader/full/detection1pdf 4/15

Page 5: detection1.pdf

8/14/2019 detection1.pdf

http://slidepdf.com/reader/full/detection1pdf 5/15

Detection Problem

Example continued

The probability density function of x[0] under each hypothesis is as follows

 p(x[0];H0) =  1

√ 2πσ2 exp−   1

2σ2 x2

[0]

 p(x[0];H1) =  1√ 

2πσ2exp

−   1

2σ2(x[0]−1)2

Deciding between H0   and H1, we are essentially asking weather   x[0]   has beengenerated according to the pdf  p(x[0];H0) or the pdf p(x[0];H1).

5

Page 6: detection1.pdf

8/14/2019 detection1.pdf

http://slidepdf.com/reader/full/detection1pdf 6/15

Page 7: detection1.pdf

8/14/2019 detection1.pdf

http://slidepdf.com/reader/full/detection1pdf 7/15

Chi-Squared (Central)

A chi-squared pdf arises as the pdf of x, where x =vi=1

x2i , if xi is a standard normally

distributed random variable. The chi-squared pdf with   v   degrees of freedom is

defined as

 p(x) =

1

2v2 Γ(v

2)

xv

2−1 exp

−12x

, x > 0

0, x < 0

and is denoted by  χ2v.   v  is assumed to be integer and  v

≥1. The function Γ(u)   is

the Gamma function and is defined as

Γ(u) =

   ∞

0tu−1 exp(−t)dt

7

Page 8: detection1.pdf

8/14/2019 detection1.pdf

http://slidepdf.com/reader/full/detection1pdf 8/15

Chi-Squared (Noncentral)

If   x =vi=1

x2i , where   xi’s are i.i.d. Gaussian random variables with mean   µi   and

variance σ2 = 1, then x has a noncentral chi-squared pdf with  v  degrees of freedom

and noncentrality parameter λ =v

i=1 µ

2

i . The pdf then becomes

 p(x) =

12

x

λ

v−2

4 exp− 1

2(x + λ)

I v2−1

√ λx

, x > 0

0, x < 0

8

Page 9: detection1.pdf

8/14/2019 detection1.pdf

http://slidepdf.com/reader/full/detection1pdf 9/15

Neyman-Pearson Theorem

Detection performance measures

Detection performance of a system is measured mainly by two factors:

1. Probability of false alarm:  P FA = p(H1;H0)

2. Probability of detection: P D = p(H1;H1)

Note that sometimes instead of probability of detection, probability of miss detec-

tion, P M  = 1−P D   is used.

9

Page 10: detection1.pdf

8/14/2019 detection1.pdf

http://slidepdf.com/reader/full/detection1pdf 10/15

Neyman-Pearson Theorem

Problem statement

Assume a data set  x = [x[0], x[1],...,x[N −1]]T  is available. The detection problem

is defined as follow

H0   =   T (x) < λ

H1   =   T (x) > λ

where  T   is the decision function and  λ   is the detection threshold. Our goal is to

design T  so as to maximize P D  subject to P FA < α.

10

Page 11: detection1.pdf

8/14/2019 detection1.pdf

http://slidepdf.com/reader/full/detection1pdf 11/15

Neyman-Pearson Theorem

Neyman-Pearson Theorem

To maximize P D  for a given P FA = α decide H1  if

L(x) =  p(x

;H1) p(x;H0)

 > λ

where the threshold λ is found from

P FA = {x:L(x)>λ}

 p(x;H0)dx = α

The function L(x) is called the likelihood ratio and the entire test is called the likeli-

hood ratio test (LRT).

11

Page 12: detection1.pdf

8/14/2019 detection1.pdf

http://slidepdf.com/reader/full/detection1pdf 12/15

Neyman-Pearson Theorem

Example: DC level in WGN

Consider the following signal detection problem

H0   :   x[n] = w[n] n = 0, 1, . . . , N  −1

H1   :   x[n] = s[n] + w[n] n = 0, 1, . . . , N  

−1

where the signal is s[n] = A for A > 0 and  w[n] is AWGN with variance σ2. Now the

NP detector decides H1  if

1

(2πσ2)N 2 exp

−   1

2σ2N −1

n=0 (x[n]−A)

21

(2πσ2)N 

2

exp−   1

2σ2

N −1n=0  x2[n]

  > λ

Taking the logarithm of both sides and simplification results in

Aσ2

N −1n=0

x[n] > lnλ + N A2

2σ2

Since A > 0, we have finally

1N 

N −1n=0

x[n] >   σ

2

N A ln λ + A2   = λ′

12

Page 13: detection1.pdf

8/14/2019 detection1.pdf

http://slidepdf.com/reader/full/detection1pdf 13/15

Neyman-Pearson Theorem

Example continued

The NP detector compares the sample mean x̄ =   1N 

N −1n=0  x[n] to a threshold λ

. To

determine the detection performance, we first note that the test statistic T (x) = x̄ is

Gaussian under each hypothesis and its distribution is as follows

T (x) ∼  N (0, σ

2

N  )   under   H0

 N (A, σ2

N  )   under   H1

We have then

P FA = P r(T (x) > λ′

;H0) = Q

  λ′ 

σ2/N 

and

P D = P r(T (x) > λ′

;H

1) = Q   λ′−A σ2/N 

P D  and P FA  are related to each other according to the following equation

P D = QQ−1(P FA)

− N A2

σ2

13

Page 14: detection1.pdf

8/14/2019 detection1.pdf

http://slidepdf.com/reader/full/detection1pdf 14/15

Receiver Operating Characteristics

The alternative way of summarizing the detection performance of a NP detector

is to plot   P D   versus   P FA. This plot is called Receiver Operating Characteristics(ROC). For the former DC level detection example, the ROC is shown here. Note

that here   NA2

σ2  = 1.

14

Page 15: detection1.pdf

8/14/2019 detection1.pdf

http://slidepdf.com/reader/full/detection1pdf 15/15

Minimum Probability of Error

Assume the prior probabilities of H0  and H1  are known and represented by  P (H0)

and P (H1), respectively. The probability of error,  P e, is then defined as

P e = P (

H1)P (

H0

|H1) + P (

H0)P (

H1

|H0) = P (

H1)P M  + P (

H0)P FA

Our goal is to design a detector that minimizes  P e. It is shown that the following

detector is optimal in this case

 p(x|H1) p(x|H0)

 > P (H0)P (H1)

 = λ

In case P (H0) = P (H1), the detector is called the maximum likelihood detector.

15