___,........

-'

';i:~

348

SIGNAL DETECTION

BINARY DETECTION WITH A SINGLE OBSERVATION

fYJn,(y!Hol

alarm and an optimum decision rule has to take into account the relative costs
and minimize the average cost. We now derive a decision rule that minimizes
the average cost.
If we denote the decisions made in the binary hypothesis problem by D;,
i = 0, 1, where D 0 and D 1 denote the decisions in favor of H 0 and H 1 , respectively, then we have the following four possibilities: (D;, Hi), i = 0, 1, and
j = 0, 1. The pair (D;, Hi) denotes H1 being the true hypothesis and a decision
of D;. Pairs (D 0 , H 0 ) and (D 1 , H 1 ) denote correct decisions, and (D~> H 0 ) and
(D 0 , H 1) denote incorrect decisions. If we associate a cost Cii with each pair
(D;, H), then the average cost can be written as

fYJn,Cy!Htl

0.622

1.0

accept H 1 -----~

--------accept H 0

Figure 6.2 Decision rule and error probabilities for Example 6.1.

2:2: C; 1P(D;, Hi)

i=O,l
i=O,I

(b)

From Figure 6.2, we obtain the error probabilities as

JR,

.622

-oo

3
2'1T

, ;;:;- exp
V

9
-- (y - 1) 2 dy
2

"" Q(1.134) "" 0.128

and

P(Dt!Ho)

= P(y

E RtiHo)

= ( fY!H,(YIHo) dy

JR,

(6.5)

2:2: C;iP(H1)P(D;IH1)
i=O,I
i=O.I

P(Do!Ht) = P(y E Ro!Ht) = { fY!H,(y!H 1) dy

where P( D; J HJ .is the probability of deciding in favor of H; when Hi is the true

hypothesis.
A decision rule that minimizes the average cost C, called the Bayes' decision
rule, can be derived as follows. If R 0 and R 1 are the partitions of the observation
space, and decisions are made in favor of H 0 if y E R 0 and H 1 if y E R~> then
the average cost can be expressed as

"" Q(l.866) "" .031

The average probability of an incorrect decision is

P,

349

""

P(Ho)P(Dtl Ho) + P(Ht)P(Dol H1)

3
1
4 Q(1.866) + 4 Q(1.134)

C = CooP(DoiHo)P(Ho) + CloP(DtiHo)P(Ho)
+ CotP(Dol Ht)P(H 1) + C1 1P(D 1IH 1)P(H 1)

"".055
=

CooP(Ho) { fY!H 0 (y IHo) dy

)Ro

+ CwP(Ho)
6.2.3 Bayes' Decision Rule-Costs of Errors
In many engineering applications, costs have to be taken into account in the
design process. In such applications, it is possible to derive decision rules that
minimize certain average costs. For example, in the context of radar detection,
the costs and consequences of a miss are quite different from that of a false

iY!H,(yiHo)dy

+ Co1P(H1) ( fY!H, (y IH1) dy

JRo
+ CuP(HJ) ( fY!H, (y IH1) dy

JR,

---

350

BINARY DETECTION WITH A SINGLE OBSERVATION

SIGNAL DETECTION

If we make the reasonable assumption that the cost of making an incorrect

decision is higher than the cost of a correct decision

C 10 > C 00

and

and make use of the fact that R 1 U R 0

have

= C 10 P(H0 )

C01 > Cu

= ( -oo, oo),

+ CllP(H1 ) + ( {[P(H,)(C0 ,

JRo

and R 1

n R 0 = ~'then we

C 11 )fy!H, (yjH,)]

- (P(Ho)(CID- Coo)fYIHo (yiHo)]} dy

(6.6)

Since the decision rule is specified in terms of R 0 (and R 1), the decision rule
that minimizes the average cost is derived by choosing R 0 that minimizes the
integral on the right-hand side of Equation 6.6. Note that the first two terms in
Equation 6.6 do not depend on R 0
The smallest value of Cis achieved when the integrand in Equation 6.6 is
negative for all values of y E R 0 Since C 01 > C 11 and C 10 > C00 , the integrand
will be negative and C will be minimized if R 0 is chosen such that for every

value of y E R 0 ,

which leads to the decision rule:

H0

Other Decision Rules

In order to use the MAP and Bayes' decision rules, we need to know the a
priori probabilities P(H0 ) and P(H 1) as well as relative costs. In many engineering applications, these quantities may not be available. In such cases, other
decision rules are used that do not require P(H0 ), P(H 1), and costs. Two rules
that are quite commonly used in these situations are the minmax rule and the
Neyman-Pearson rule.
The minmax rule is used when the costs are given, but the a priori probabilities P(H0 ) and P(H 1 ) are not known. This decision rule is derived by obtaining the decision rule that minimizes the expected cost corresponding to the
value of P(H 1) for which the average cost is maximum.
The Neyman-Pearson (N-P) rule is used when neither a priori probabilities
nor cost assignments are given. The N-P rule is derived by keeping the probability of false alarm, P(DdH 0 ), below some specified value and minimizing
the probability of a miss P(D 0 iH 1).
Details of the derivation of the minmax and N-P decision rules are omitted.
Both of these decision rules also lead to the form

L(y)

fY[H,(YIH 1 )

H1

f Y!H/Y IHo)

Ho

. (6.8)

'Y

where 'Y is the decision threshold. Thus, only the value of the threshold with
which the likelihood ratio is compared varies with the criterion that is optimized.
In many applications including radar systems, the performance of
decision rules are displayed in terms of a graph of the detection probability

P(H,)(COI - Cu)fr[H, (yjH,) < P(Ho)(Cw- Coo)fr[H0 (yjHo)

Choose

6.2.4

if

P(H,)(Co, - Cll)frrH,(YI H,) < P(Ho)(Cw - Coo)frru/Y I Ho)

fYIH,

In terms of the likelihood ratio, we can write the decision rule as

l
(

L(y)

'
l
l

frru,(YIH,)

u,

P(H 0 )(C 1o - C 0 o)

/Y[H 0 (Y IHo)

Ho

P(HI)(Cot - C11)

351

(6.7)

Note that, with the exception of the decision threshold, the form of the Bayes'
decision rule given before is the same as the form of the MAP decision rule
given in Equation 6.2! It is left as an exercise for the reader to show that the
two decision rules are identical when C 10 - C00 = C01 - C 11 and that the Bayes'
decision rule minimizes the probability of making incorrect decisions when
Coo = C 11 = 0 and C 10 = C01 = 1.

-A
(a)

Figure 6.3a
and Pv.

Conditional pdfs (assumed to be Gaussian with variance = a 2 ); PF;

:;--~~

352

BINARY DETECTION WITH MULTIPLE OBSERVATIONS

SIGNAL DETECTION

353

where

X(t)

PD

{s

under hypothesis H 0
under hypothesis H 1

(t)

St(l)

0.6
0.5
0.4
0.3
0.2
0.1
0.1

0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

-and s 0 (t) and s 1(t) are the waveforms used by the transmitter to represent 0 and
_1, respectively. In a simple case s 0 (t) can be a rectangular pulse of duration T
and amplitude -1, and s 1(t) can be a rectangular pulse of duration T and
amplitude 1. In any case, s 0 (t) and s 1 (t) are known deterministic waveforms;
the only thing the receiver does not know in advance is which one of the two
waveforms is transmitted during a given interval.
Suppose the receiver takes m samples of Y(t) denoted by Y 1 , Y 2 , , Ym.
Note that Yk = Y(tk), tk E (0, T), is a random variable and h is a particular
value of Yk. These samples under the two hypothesis are given by

PF~

(b)

Figure 6.3b Receiver operating characteristic (ROC).

P(D 1 /HJ) = P 0 versustheprobabilityoffalsealarmP(D 1 /H0 ) = PFforvarious values of the threshold. These curves are called the receiver operating
.characteristic (ROC) curves. (See problems 6.7-6.13.) An example is shown
in Figure 6.3. Figure 6.3a shows the conditional probabilities with P 0 and P F
indicated. Note that 2A is the difference in the means. Figure 6.3b shows example ROC curves as a function of the mean divided by the standard deviation.

6.3 BINARY DETECTION WITH MULTIPLE OBSERVATIONS

In the binary detection problem shown in Figure 6.1a, the receiver makes its
decision by observing the received waveform for the duration of T seconds. In
the T-second interval, the receiver can extract many samples of Y(t), process
the samples, and make a decision based on the information contained in all of
the observations taken during the symbol interval. We now develop an algorithm
for processing multiple observations and making decisions. We will first address
the problem of deciding between one oftwo known waveforms that are corrupted
by additive noise. That is, we will assume that the receiver observes

Y(t) = X(t) + N(t),

0:5t:5T

Yk = so.k

nk

under H 0

Yk = s,,k

nk

under H 1

where so.k = s0(tk), and sl.k

waveforms s 0 (t) and s 1(t).
Let

Y = [Yt. Y 2 ,

s 1(td are known values of the deterministic

YmF and y = [y,, Yz, . , YmF

and assume that the distributions ofYunder H 0 and H 1 are given (the distribution
of Y 1 , Y 2 , , Ym will be the same as the joint distribution of N 1 , N 1 , ,
Nm except for a translation of the means by so.k or sl.k, k = 1, 2, ... , m).
A direct extension of the MAP decision rule discussed in the preceding section
leads to a decision algorithm of the form

L(y)

6.3.1

/YiH,(y/Ht)

H,

!Yu/Y/Ho)

H0

P(H

0
--

(6.9)

P(Ht)

Independent Noise Samples

In many of the applications involving radio communication channels, the noise

that accompanies the signal can be modeled as a zero-mean Gaussian random

.....