Predicting Aural Detectability of Aircraft in Noise Backgrounds - DTIC

UNCLASSIFIED

AD NUMBER AD902828

NEW LIMITATION CHANGE

TO Approved for public release, unlimited

distribution

FROM Distribution authorized to U.S. Gov't. agencies only; Test and Evaluation; Jun 1972. Other requests shall be referred to the Air Force Flight Dynamics Laboratory, Attn: FY, Wright-Patterson AFB, 45433.

AUTHORITY AFWAL,

per ltr dtd 21 Jul 1976

THIS PAGE IS UNCLASSIFIED

AAFFDL-TR-72-16

PREDICTING AURAL DETECTABILITY OF AIRCRAFT IN NOISE BACKGROUNDS

00•

00 SANFORD FIDELL KARL S. PEARSONS RRICARDA L. BENNETT BOLT BERANEK AND NEWMAN, INC.

-

TECHNICAL REPORT AFFDL-TR-72-17

-DDC JULY 1972

SEP7 9T

Distribution limited to U. S. Government agencies only; test and evaluation; statement applied 28 June 1972. Other requests for this document rnit be referred to AF Flight Dynamics Laboratory/FY, Wright-Patterson AFB, Ohio 45433.

AIR FORCE FLIGHT DYNAMICS LABORATORY AIR FORCE SYSTEMS COMMAND WRIGHT-PATEIERSON AIR FORCE BASE, OHIO

NOTICE When Government drawings,

specifications,

or other data are used

for any purpose other than in connection with a definitely related Government procurement operation, the United States Government thereby incurs no responsibility nor any obligation whatsoever; and the fact that the government may have formulated, furnished, or in any way supplied the said drawings, specifications, or other data, is not to be regarded by implication or otherwise as in any or

manner licensing the holder or any other person or corporation, conveying any rights or permission to manufacture, use, or sell any patented invention that may in

any way be related thereto.

• ..... I.o.... -

-.

• l rli I

..... .......

•r.......... ...... .. .....

41

Copies of this report should not be returned unless return is required by security considerations,

notice on a specific document. AIR FORCEM

LI

24-&72/160

contractual obligations,

or

a

6

a

-

.4

PREDICTING AURAL DETECTABILITY OF AIRCRAFT IN NOISE BACKGROUNDS

SANFORD FIDELL KARL S. IPEARSONS RICARDA L. BENNETT

II Distribution limited to U. S. Government agencies only; test and cvaluation; statcment applied 28 June 1972. Other requests for this document must be referred to AF Flight Dynamics Laboratory/FY, Wright-Patterson AFB, Ohio 45433.

m

--

-

mJ

FOREWORD This report was prepared by Bolt Beranek and Newman, Aero-Acoustics Branch, Dynamics Laboratory,

Vehicle

Dynamics Division,

Wright-Patterson

Contract F33615-71-C-1220.

Inc.,

for the

Air Force Flight

Air Force Base,

Ohio,

The work described herein is

under

a part of

the Air Force Systems Command exploratory development program to predict

the noise environment of flight

directed under Project Flight Vehicles,"

1471,

vehicles.

"Aero-Acoustics

Task 1471 02,

The work was

Problems in

Air Force

"Prediction and Control of Noise

Associated with USAF Flight Vehicles".

Capt.

R.

P.

Paxson of the

2,

1972 for

Aero-Acoustics Branch was the task engineer. Manuscript was

released by the

authors on May

publication as an AFFDL Technical Report. This Technical Report has been reviewed

and. is

approved.

SWALT . OW Asst. for Research & Technology Vehicle Dynamics Division

li

ABSTRACT Laboratory experiments

were undertaken to develop improved aural

detection criteria for light aircraft. Specifically two series of psychoacoustic judgment tests were conducted to determine the applicability of the psychophysical Theory of Signal Detectability (TSD) to prediction of the aural detectability of light aircraft noise signatures in

jungle noise backgrounds.

The first

series of

tests produced data supporting development of a simplified graphical prediction method based on TSD. The second testing program validated the precision and accuracy of the prediction method under quasi-realistic

listening conditions.

Predicted

levels of performance were typically within one or two dB of the data averaged for all

I

observers.

1

!ii

L

TABLE OF CONTENTS

SECTION

I. II. III. i.

.................

. 2

PSYCHOPHYSICAL THEORY OF SIGNAL DETECTABILITY

6 6

........

PSYCHOACOUSTIC JUDGMENi'T TESTS ..... Description of Tests IA and IB ......

........

a) b)

............ Method for Test IA .......... ............. Method for Test IB ....

6

c)

Experimental Design for Tests IA and IR

6

2. 3. 4.

Results of Tests IA and IB ........

5.

Discussion and Summary ..........

..........

............. Description of Test II .............. Results of Test II ......

8 10 12 .. 14

............

18

PROPOSED AURAL DETECTION CRITERIA .. ......

19

1.

Development of Criteria

1...........19

2.

Procedures for Evaluating Aural Detectability of Aircraft . . . . . . . . . . . . . . . . . Comnarison of Predicted and Subjective Determination of the Aural Detectability of .............. the Schweizer 2-33 ........ Other Factors Affectinp the !%ural -etectlon ................. of Aircraft ...........

IV.

3. 4.

APPENDIX

INTRODUCTION ..........

V.

CONCLUSIONS ......................................

I.

SPECTRA OF TEST SIGNALS .........

II. III. IV. V.

INSTRUCTIONS TO OBSERVERS

STATISTICAL SUMMATION METHOD

.........................

iv

24 26

41

...........

..

44 50

..........

CONDENSED SUMrMARY OF PROCEDURES FOP EVALUATING THE AURAL DETECTABILITY OF AIRCRAFq ......

REFERENCES ..................

24

37

...........

..........

OBSERVERS AND EQUIPMIET .....

2"

..

52 58

LIST OF ILLUST7ATIONS Figure

Pg

1

A FAMILY OF RECEIVER OPERATING CHARACTERISTIC CURVES

27

2

BACKGROUND NOISES FOR TEST IA AND IB .. .......

28

3

RESULTS FOR TEST IA FOR BACKGROUNDS 1, 2,

4

RESULTS FOR TEST IA - MEAN AND STANDARD DEVIATION OF SIGNALS FOR ALL BACKGROUN'DS ..... ............. ..

...

3 ........

29

30

5

TEST IB -- WITH BACKGROUND I COMPARISON OF STATISTICAL SUMMATION PROCEDURE AND d' max RULE.•.•.• 31

6

PREDICTED LEVEL CALCULATED WITH d' max RULE FOR BACKGROUND I - TEST IA AND TEST IB

.....

..........

32

7

BACKGROUND NOISE FOR TEST I1 ...............

33

8

RESULTS FOR TEST II WITH JUNGLE NOISE BACKGROUND

34

9

PREDICTED AND OBSERVED SIGNAL LEVELS FOR VARIOUS CORRECT DETECTION RATES IN TEST II (ABSCISSA IN OVERALL SPL) ............. . . ................. ...

35

10

CHART FOR PREDICTING DETECTION LEVELS (FOR 50% CORPECT AND 1% FALSE ALARM RATES) USING BACKGROUND NOISE SPECTRUP IN ONE-THIRD OCTAVE BANDS ............. . 36

11

BLOCK DIAGRAM OF PLAYBACK SYSTEM FOR JUDGMENT TESTS .

12

CHART FOR PREDICTING DETECTION LEVELS (FOR 50% CORRECT AND 1% FALSE ALARM RATES) USING BACKGROUND NOISE SPECTRUM IN ONE-THIRD OCTAVE BANDS

.

.

.

.

.

.

.

13

BACKGROUND NOISE AND AIRCRAFT FLYOVER NOISE SAMPLE

14

PREDICTED DETECTION IZVEL EXAMPLE FOR AIRCRAFT FLYOVER NOISE IN NOISE BACKGROUND . . . . . .

v

•

r'

"rl".......

.

.

..

.

49

55

.

56

.

57

LIST OF TABLES Table I. II. III. IV. V.

Page SIGNALS FOR DETECTION TEST IA ....... SIGNALS FOR DETECTION TEST IB

...........

SIGNALS FOR DETECTION TEST II

.......

ADJUSTMENT OF 50% DETECTABILITY VARIOUS FALSE ALARM4 RATES . .

..............

7

..........

9

............

15

LEVEL FOR 22

..............

ADJUSTMENT OF 50% DETECTABILITY LEVEL TO OTHER PERCENT CORRECT WITH 1% FALSE ALARM RATE .. ......

23

APPENDIX I.

I. II. III.

SIGNALS FOR TEST IA ................................. SIGNALS FOR TEST IB ......... SIGNALS FOR TEST II

.................

38 ..

...................

39 40

APPENDIX III. I.

EQUIPMENT EMPLOYED FOR SIGNAL PRESENTATION AND ANALYSIS

vi

00

47

SECTION I. INTRODUCTI ON Recently there has been some interest in designing quiet light For such designs it is necessary to know the noise aircraft. Although some investi-

levels of aircraft which are detectable. gations have been carried out to develop detection levels,

some improvement is

methods for determining

required.

Therefore a

development program was undertaken to improve aural detection criteria for light aircraft Aural detection of light aircraft noise by human observers under a complex task.

field conditions is

depends not only upon the

It

physical characteristics of the aircraft noise signature and the ambient noise of the listening environment, but also upon the sensitivity of the otserver,

his state of attention,

information available to the observer,

a priori

and the costs and payoffs

associated with the outcomes of the observer's detection decisions.

"The research program described in the current report was undertaken to develop a relatively straightforward method of predicting human performance in the aural detection task so that reliable information on the probability of detection of current and future aircraft would be available to both design and tactical personnel. The theoretical approach adopted to simplify and quantify the various factors influencing detection performance is physical Theory of Signal Detectability. theory to the aural detection task in Section II.

length in

Section IV is

detection criteria.

Application of this

the field is

described at

Section III presents the methods and

results of three experiments method.

the psycho-

conducted to validate the prediction

a detailed discussion of the proposed aural Section V summarizes

the study.

A number of appendices

procedures,

data,

the main conclusions of

containing finer detail on

and measurements accompany the main body of the

report. -1--

SECTION II PSYCHOPHYSICAL THEORY OF SIGNAL DETECTABILITY As mentioned briefly In the introduction, performance in

prediction of human

the aural detection task demands understanding of The psychophysical (Reference 1) provides a

both physical and psychological factors. Theory of Signal Detectability (TSD)

theoretical framework for joint consideration of these factors. The underlying assumption of the current application of TSD is that human detection performance may be likened to an ideal energy Thus, the basic physical quantity upon which detection process. human detection performance rests is the signal to noise ratio. For purposes of simplicity, this quantity is measured in independent one-third octave bands according to procedures discussed in greater detail In the following section. an energy detector could make detection decisions in First, any sample of acoustic energy would the following fashion. In theory,

be assumed to derive from one of two distributions. butions are those of noise alone,

The distri-

and signal plus noise.

For

computational convenience the two distributions are assumed to be In practical terms, the normal in nature and of equal variance. distance between the means of the two di-tributions is

given by

As the signal to noise ratio the measured signal to noise ratio. A standardized increases, the decision task becomes simpler. statistic known as d' (d prime) may be calculated to determine the sensitivity of an energy detector working on any given set of defined as the distance between The d' statistic, distributions. the means of the two distributions divided by the standard deviation of the noise distribution, is monotonic with the signal to noise ratio and directly proportional to the quantity 2E/No (twice the signal energy divided by the noise power per unit bandwidth).

--2-

Next,

the probability

that the sample of acoustic energy could

have been generated by the noise process the signal plus noise processes is

alone and by the sun of

evaluated.

An optimal manner

of combining the information represented by these two probabilities is

to form their ratio,

It

has been shown (Reference

ratios are optimal in

referred

to as

a likelihood or odds ratio.

1) that decisions based on likelihood

the sense that all available physical

informat-on about the presence of the signal is such a statistic. represents

inccrporated in

The numerator of the statistic

cominjon2y

the probability that the observation belongs to the

distribution of signal plus noise, the probability

while the denominator represents

that the same observation

belongs

to the

distribution of noise alone. It

must be emphasized that even an ideal enerFgy detector operating

in

this fashion is

ratio is

1 (i.e.

fallible.

, it

is

to the noise alone ou,

If, equally

for examPle,

the

1X.kký2Ihood

likely that the obs:-va-tion belongs

to the signal plus noisc

distrbution),

the

decision must be based on grounds other than the physical vation.

In

fact,

it

should be apparent

necessary to make a decision. expressed In

that a criterion

presence of a signal whenever the

(e.g.,

'report

likelihood ratio

the criterion must be based upon non-physical

the

exceeds

factors.

10"),

Foremost

outcomes.

of any detection decision

are readily enumerated.

A "hit" may be defined as a decision to report signal when the signal is

aiWay

factors are the costs arid payoffs

associated with the decision

The four outcomes

is

Although the criterion may be

terms of a likelihood ratio

among these non-acoustic

obser-

actually present.

the presence cf thc

A "miss" m•ty

be

defined as a decision not to report the presence of a signal when it

is

in

fact present.

A "false alarm" results

--3--

from a decision

to

report the presence of the while a "correct

signal in

rejection"

the absence of a signal,

represents

the presence of the signal when it

is

a decision not in

fact not present.

The values of the four decision outcomes context of the situation in

which

are determined by the

the decision must be made.

Certain situations may favor a high false

alarm rate

likelihood ratio criterion with a very low value), situations

(i.e.,

may favor a very high correct rejection rate

fundamental sensitivity

of the observer,

is

of the criterion employed in

entirely independent Thus,

the same observer,

a lax

while other

stringent likelihood ratio with a very high value).

decision.

to report

(i.e.,

a

Note that the

as expressed in

d'

units,

rroducing a

of fixed sensitivity,

is

capable

of producing widely divergent behaviors. In some situations, the observer may achieve an extremely high hit (albeit In

at the expense of a concomitantly

other situations,

with a concomitant

reduction in

the false

is

Operating CharacterXl'c".

rate,

alarm rate.

false alarm rates

observer of fixed sensitivity

for an individual

referred to as a "Receiver

Figure 1 is

of a family of ROC curves which is is

fAlse alarm rate).

the observer may achieve a very low hit

The plot of hit rates versus

accelerated,

high

rate

a schematic representation

parametric in

d'.

The negatively

positively sloping function depicted by the ROC curve

a direct consequence of the assumptions

of the common TSD model.

The remaining non-acoustic Information which an ideal ooserver must consider in

making a rational decision is

probability of occurrence of a signal. observer has some indication available)

of the probability

will include a signal.

In

In

the a priori

other words,

the ideal

(from whatever sources may be that any given observation interval

terms

-4

of the problem at hand,

an

observer in

a realistic

expectations

detection task certainly has some

about the frequency of occurrence of overflights,

even before he begins the listening task. This information is composed in is

most concisely expressed as an odds ratio,

a manner similar to a likelihood ratto.

usually the probability that a signal is

observation interval,

while

that the noise alone is It

The numerator

prestnt during an

the denominator is

the probability

present during the observation interval.

has been s),own that Bayes'

Theorem (Reference

optimal way to combine the a priori

2)

provides an

information with the likelihood

ratio resulting from a particular observation.

Bayes'

Theorem is

applied to the current situation by observing that the a posteriori odds in

favor of the occurrence of a signal during a particular

observation interval are equal to the simple product of the a priori interval,

odds (the odds in

favor of occurrence of a signal in

known before the observation is

ratio (the odds in information The reader is

fi.vor of occurrence

contained in

made)

the

and the likelihood

of a signal based on the

the observation).

referred to Reference

1 for more detailed discussions

and mathematical proofs of the theoretical concepts presented above.

L4

SECTION

III

PSYCHOACOUSTIC JUDGMENT TESTS

1.

Description of Tests IA and IB

Psychoacoustic experiments of the audibility

The intent of Test IA was to establish

jungle noise backgrounds. aircraft

of a number of different

noise signatures.

the manner in

components of light

The intent of Test IB was to determine of the individual

which the detectabilities

to influence

combine

noise signatures embedded in

of light aircraft

the detectabilities

IA and IB below)

of TSD to prediction

conducted to determine the applicability

were

a)

(referred to as Tests

components

overall detectability.

Method for Test IA

Twelve

synthetic signals were selected as representative of the noise signatures

portions signals

fell

into three

The twelve

categories of spectral characteristics,

including narrow band noises,

broadband noises,

and pure tones in

Two rates of frequency modulation

low and medium frequency ranges. and two amounts

of light aircraft.

of

of amplitude modulation were applied to several of

the signals as well.

Observers heard the signals in

background noise environments.

each of three

The background noises were similar

to those of the day and night ambient noise in

the Jungle and to

the more familiar noise background of modern inhabited areas. Table I

catalogs the twelve signals

Appendix I spectra.

and three background noises.

contains one-third octave band analyses of the signal Figure 2 plots the spectra of the background noises.

Ten observers between the ages of 16 and 24 years were employed in Test IA.

Observers were audiometrically

Recommendation R-389.

screened according to ISO

Observers were required to depress a

response switch corresponding to the observation interval in

-6s

which

F

F >

C

CC

0~

>

~

i

>,~ 0.-.

_

~

OU~~~~~

C,

0V

E-

C.. C.

Lt

CMt

c'C. C

C..

C..

1''.1

.-

3='

~

ico

CC CC. -

UC m 01

S

ZI

II" IC'~~

C

' CO

'

-tC.'

C

C.

qC

>~~~c Cg-

0r C.

~

CC

L. Cdo .-

0

4Ol,

c-'Cl

12-0

a''

C')

y

02 C V.

C.7C- F

U

2C

0'0 0It E-

x.

co

U............

7

*-

0'

C'lC

C-

0~.

0.

S '

C

F

F

'

C )

they thought they heard a signal in task.

a two alternative forced choice

Administration of experimental

control of an automated psychophysical b)

conditions was under the laboratory.

Method for Test IB

Fourteen signals were presented to observers for subjective judgment under conditions similar to those of Test IA.

The

fourteen signals included seven previously employed in Test IA, three signals constructed of quadruple combinations of the above seven signals, and four recorded aircraft flyovers. Table II summarizes

the nature of the signals employed in

observers,

of whom eight had participated

in

Test IB.

Test IA,

Ten

served in

Test IB. c)

Experimental

The trial

procedure employed in

based adaptive Testing (PEST), given trial switches

Design for Tests IA and IB Tests IA and IB was

technique known as Parameter Estimation by Sequential described in

detail elsewhere

(Reference

3).

On a

an observer was required to depress one of two response to indicate to the computer his decision about which

observation interval contained the signal. were four seconds period.

a computer-

Observation intervals

long and separated by a one second intratrial

Feedback was immediately presented in

lighted response switch to indicate

the form of a

the interval in

which the

signal actually had occurred. Fourteen hours per observer were required to complete the thirtysix (twelve of Test IA.

signals by three background noises)

listening conditions

Seven experimental sessions per observer were scheduled

at two hours each. two hour sittings,

Approximately eight hours,

divided into four

were required for each observer

fourteen listening conditions of Test IB.

--8-

to complete

the

acI

0-1.

E - 3-

C

C x

CC

LL a-4

'1

:'t0 0

0L.

C

C

rC

I- z-

I4

>

0)0)

.4)

uýl -

o.

4)

ca

(3

I

I

I

Each observer

completed five repeated determinations

correct detection point in

succession in

Test IA.

was provided between sets of determinations; was provided after

each half hour in

observers encountered tests

In

of the 75%

A brief

break

a five minute break

the anechoic

chamber.

the various listening conditions in

different random orders.

Appendix III

contains

All both

further

detail on observers and equipment employed in these tests. 2.

Results of Tests IA and IB

The principal findings of Test IA are presented graphically in Figures 3 and 4. deviations

Data in

detectabilities

all

Test IA was a statistical

of signals

third octave bands.

detail

in

Appendix

observers'

of the 75% detection level for

each noise background.

The zero line of Figure 3 The vertical

distances of data points from the zero line are as deviations

i4idication of the variability predicted performance

described in

The data of Figure 3 are the averages of

the predicted level of performance.

interpretable

The prediction

summation of the

The prediction method is

IV.

the form of

calculated from signal to noise ratios

five determinations

each signal in represents

figures are plotted in

from a predicted level of performance.

scheme employed in in

these

from prediction.

therefore directly

Figure 4 provides

an

associated with the deviations from

for each of the twelve signals,

collapsed

over the three noise backgrounds. It

must be realized that Figures 3 and 4 express only deviations

from predicted levels,

not absolute

levels.

Since there is

a

37 dB(A) range of predicted levels among the twelve signals and three noise backgrounds, prediction

is

and since the average deviation

only 2.8 dB,

a plot of absolute levels would show an

excellent fit of the data to the prediction.

-10a, |U

from

The goodness of fit of the predicted d's to the data of Test IA confirmed the applicability of the TSD to aural detection of light aircraft

noise in

various noise backgrounds.

No special problems

were encountered in predicting the detectabilities of "peculiar" signals; In

i.e.,

those containing amplitude or frequency modulation.

order to produce the best fit

detectabilities

it

of the predicted to the observed

was necessary

to assume that the observers'

efficiency was 0.4 with respect to an ideal energy detector. value

is

consistent with previously reported

particularly

in

and immediate

feedback available

4),

to observers.

the principal findings of Test IB

in

graphical

Since the intent of Test 1B was to determine the manner in

which the detectabilities are

(Reference

light of the well defined observation intervals

Figure 5 presents form.

findings

This

of discrete

components of a.rcraft

noise

combined by observers to yield overall detectability,

different prediction Figure 5 demonstrates by the statistical

schemes were entertained. the fit

two

The upper plot of

of the data to the predictions made

summation rule discussed in

lower plot of Figure 5 demonstrates

the fit

Appendix IV.

The

of the data to

The predictions made by a d' max rule discussed in Section IV. average deviation of the data from the predicted values for the statistical

summation predictions was -1.66

dB,

while the average

deviation from the predicted values One statistical

procedure

for the d' max rule was -0.33 dB. for evaluating the significance of the

difference between means of different value of the t

statistic

samples is

(with 13 degrees

difference between the average

the t

of freedom)

test.

The

for the

deviations from predicted values

for the d' max and statistical summation rules was 4.2. The this large or likelihood of obtaining a value of the t statistic larger by chance alone is

less than 0.01.

Thus,

that the d'max rule provided significantly better the statistical

summation rule.

-11-

it

was

concluded

predictions

than

The inclusion of seven signals of Test IA in Test IB presented an opportunity to assess the test-retest reliability of experimental The average differences in conditions over a period of months. the observers' detection performance for the seven signals This difference of less included in the two tests was 0.8 dB. than 1 dB was interpreted as confirmation of the stability of test conditions and the generality of predictions based on signal detection theory. The greater accuracy of prediction afforded by the d' max rule was It appears that the observers in Test IB "locked on" not expected. to the most detectable component of the complex signals (the three quadruple combinations and the four flyovers) and ignored informaGreater familiarity with tion present in other spectral regions. the signals gained through extensive experience might encourage observers to modify their detection strategy to listen for more than one component of complex signals. Figure 6 summarizes on one plot the effectiveness of the d' max rule. All of the signals of Tests IA and IB are plotted in this figure on the same basis as in previous figures in this section. above the predicted line, twelve points fall Twelve points fall The accuracy of below the line, and two fall exactly on the line. The precision of t.'e prediction the prediction is thus confirmed. is

indicated by very small deviations of the predicted from the

observed values.

3.

Description of Test II

The design of Test II was intended to permit assessment of how well the prediction procedures developed in Tests IA and IB worked Accordingly,-Test II was in quasi-realistic listening conditions. administered as a vigilance task,

-12-

in

which observers were ignorant

of the time of presentation of a signal. (Figure

7)

was heard at all

times,

Continuous jungle noise*

while no other cues were

provided as to the time of occurrence of a signal. The observer's

task was to depress a response button whenever he decided that a signal vas in fact being presented. A response made while the signal was presented, or within a half second grace period following the conclusion of the signal, counted as a correct detection (hit).

A response made in

counted as a false alarm.

the absence of a signal was

Failure to respond during the presen-

tation of a signal or the grace period was registered as a miss. The instructions

to observers are included in

Signal levels corresponding to 25%,

50%,

Appendix II.

and 75% correct detection

performance were estimated using the PEST procedures. misses were employed to determine

the required level

Hits and changes.

False alarms were tabulated to permit estimation of false alarm rates but had no direct bearing on the progress of a PEST run.

A

payoff schedule which penalized observers five cents for each false alarm, to be subtracted from a fixed bonus of $1.00 for each experimental session, was introduced to control the false alarm rate.

Three PEST runs were administered to determine

three performance

each of the

levels of the psychometric function.

The duration of intertrial

intervals

(the waits between successive

presentations of signals) were determined eccording to an exponential

distribution of waiting times.

The exponential distribution

was selected so that observers would have no information about the time of occurrence likely longest *

of the next signal.

Thus,

it

was equally

that a next signal would be heard immediately or after the delay.

The recording of Panamanian jungle noise was made available through the courtesy of Mr. Richard Wells of General Electric Company, Schenectedy, New York.

-13-

Two different exponential collected for all

distributions were employed.

signals with a distribution of waiting

that ranged from four seconds duration of the signal tapes hundred seconds.

Data were times

(the minimal wait occasioned by the themselves)

to approximately

one

Data for signal six were also collected with a

distribution of waiting times four times as long.

The

longest

wait between signals in

the longer distribution was approximately

five and a half minutes

(385

seconds).

The purpose of collecting

data under the longer waiting time distribution was to permit assessment of the extrapolation of detection performance

to even

greater waiting times. The order of administration of experimental

conditions

different levels of correct detection performance) as was the order of testing of signals. cautioned in

(i.e.,

was randomized,

Observers were

carefully

their instructions to avoid the "gambler's fallacy"

of expecting a signal to occur imminently simply because

it

has

not occurred for a subjectively long period of time. Except

for the changes in

trial

procedures noted above,

equipment and physical conditions of Tests IA and IB. employed in

Test II

Table

and aircraft

III.

may be found in to observers

Appendix in

Test II

III. included tones,

flyovers from Tests IA and IB,

as shown in

Detailed one-third octave band analyses of the

signals and background appear in 4.

similar to those

Further detail of the observers and equipment

The signals presented noises,

of testing were

the

Appendix I.

Results of Test II

The major findings of Test II Figure

8 displays

are summarized in

averaged data for all

mean and standard deviations for all

-14-

Figures

observers,

8 and 9.

showing the

signals under investigation

r

N CC CC

CC CC"

E 1 N CC

4) E F

Cr C, CCC

3

.0

-.N

0 -

C-

I 0

'-4

4,. .0

0 3 .0

N

0

4)0 C

C'

N

C-

-.

C,.

I N 0! 44)

.3

I *

I

I I

C)

C

I

I

I

I

I

I

I I

-

H

C-'

C-)

I

N

I U'-'

z o r4 4,4

E-' 1C)

-

*'-C U ,-4 -C

EII

i0

.0

CC

C)

-

C.-.

0

I

I

I

I

I

I

0 0-C

CC

44

4.)

.3 CC C-C

,; C,

a

.0 'C)

I

"I

0 0 0C-Cl

o

a-, 0

0 I

-

-I )

n

I I

I

I I

I

I

I I

I

I I

4'

0 C"' -C-

E

0 C-) 4) 42

o

Z

4)

o, t

i.-.

42

N

4-C

4-C

).C

4-. 42

4-. 44

C-

C)

CC

114

0 .-4 .3

3-. .-4 .3

Z

;

o -

L

5..

0

0

40 .3

E-

V C-.

w 4

C

rV

4

CC4

CC

.'

If"

'.0

1'-

04

-15-

_-

_

-

The shaded area about the mean at the 50% correct response rate. As may indicates one standard deviation above and below the mean. be seen, these standard deviations are quite small, on the order of 1 or 2 dB In most cases. Predictions based on the method using d'

max

specified in

Section IV

of this report (solid arrow) and upon the method developed by the Air Force Flight Dynamics Laboratory (AFFDL) and specified in References 5 and 6 (AFFDL method, dotted arrow) are also displayed The method developed by BBN can be used to predict in Figure 8. detection levels for various correct response rates and false alarm The rates chosen as described in Section IV for the prerates. diction method are 50% for correct response and 1% for false alarm. Predictions for all signals based on TSD by BBN are typically Predictions based on the AFFDL method within 2 dB of the data. differ generally in level from the predictions made by BBN, and in One difficulty some cases are considerably divergent from the data. encountered in

using the AFFDL method was deciding whether or not a

For example, signals 3 and 4 pure tone existed in the signal. The higher were borderline cases so two predictions were made. predicted levels in Figure 8 were the result of assuming no pure tone content while the lower levels resulted from a "pure-toneThe mean subjective response did lie between present" assumption. the two predictions but the differences in predictions encompassed a range of 13 dB, an amount greater than desirable. As stated above, it appears that predictions based on the BBN method came closer to the data than those based on the AFFPL This is further illustrated by noting that BBN predictions method. for five of the seven signals were well within a standard deviation of the mean data while the AFFDL method produced predictions for only two to three of the seven signals which fell similarly close to the mean (depending on the tonal content assumptions).

-16-

The greatest differences between the two prediction methods were observed for the sixth and seventh signals. for signal 6,

an aircraft

1 dB from the mean, AFFDL prediction.

flyover with no tonal

was about

content,

as opposed to 9 dB from the mean for the For signal

0.5 dB from the mean, 15 dB different

The BBN prediction

7,

the BBN prediction was about

while the AFFDL method predicted a level

from the mean.

not contain pure tones,

Thus,

for those signals which do

the AFFDL prediction method produces poor

estimates of the 50% correct where the tonal content is

detection level and for those signals

in

question,

the range of predictions

may be as high as

13 dB.

Figure 9 displays

the BBN predictions and the observed results

(averaged

over observers again)

detection levels.

for the 25%,

50%,

Since the AFFDL method produces no predictions

for levels other than 50% correct detection, display such predictions

in

Figure 9.

for 75% of the cases)

at all

no effort

Once again,

between observed and predicted levels is between

and 75% correct

agreement

quite good (within 2 dB

three detection rates.

the observed and predicted slopes is

suggesting the feasibility

was made to

Agreement

especially noteworthy,

of extrapolation to other detection

rates than those measured. Predicted valaes

for the differences

correct detection levels,

and the 50% and 75% correct

levels were 1.6 and 1.1 dB, differences

between the 25% and 50%

respectively.

actually observed were

detection

The corresponding

1.2 dB and 0.9 dB,

respectively.

Although both the expected and observed values suggest the narrowness of the "detection

aperture",

it

values increase with increases in Table IV of Section IV.

should be noted that the

the false alarm rate,

The differences

as seen in

between 30% and 50%

correct detection levels and 50% and 70% correct detection levels becomes

6.5 dB and 2.5 dB,

respectively, for a 25% false alarm

rate.

-17F.

.

,-1

. .

.

•uI

•

T

I

I

1

|~~-

I•'

.

..

-

.

1

i.

.

. . .

.. I

..

I...

.i

I

I

I

n

5.

Discussion and Summary

Two series of subjective judgment tests were conducted to determine the applicahtlity of the psychophysical Theory of Signal Detectability (TSD)

to prediction of the levels at which light aircraft

are aurally detectable in jungle noise backgrounds. Tests IA and IB demonstrated the applicability of TSD to the current problem, while providing data for the evolutiun of a simplified graphic prediction method. Test II pr'ovided verification of the prediction methcd under quasi-realistic listening conditions. A number of grounr's were found on the basis of the current research to prefer the TSD approach to prediction of aural detectability to other approaches,

such as the AFFDL method specified in

Reference

5.

First, the TSD method provides a better fit, on the average, to the data. Second, it is more general, in that it permits prediction of performance at levels other than 50% correct detecJion and at specified false alarm rates. Third, it is simpler, since it requires no distinctions to be made on the basis of the tonal content of aircraft noise signatures. Fourth, the TSD approach is fixmly based in theory, and thus capable of extension to novel situations,

unforeseen detection strategies,

-18-

'S.

and so forth.

SECTION IV PROPOSED AURAL DETECTION CRITERIA 1.

Development of Criteria of this report has outlined the nature of the sensi-

Section II tivity in

measure d'

and its

interactions with non-acoustic factors

determining human detection performance.

straightforward prediction method, made to permit simplification

In

developing a

a number of assumptions were

of the method,

and to make maximal

use of acoustic information about the levels of aircraft background noises.

First,

it

was assumed that observers have

diffuse a priori information about the presence is

of aircraft

during

This assumption was made of necessity,

the detection task. it

and

not pcssible to quantify this sort of information in

graphic prediction men,,hod.

Second,

it

was necessary

since a

to assume a

false alarm rate consi2stent both with hearing threshold data tThich appear in

the graphical method and with the nature of the

detection task in will be found in

the field. later

Justification

for these assumntions

discussions.

The essence of the proposed method for predicting the detectability of light aircraft noise in Jungle noise backgrounds is that human performance

is

based on the signal to noise ratio in

one-third octave band to which human sensitivity is basic data employed In

the single highest.

the proposed method are therefore

The

the one-

third octave band spectra of the aircraft noise and of the jungle noise. These data are used to determine the va'ue of the sensitivity measure d',

as discussed in

which estimates d'

is

th.

previou,

section.

The eauation

as follows: d-

'

S(W)(

N(

)

an expression of the efficiency of a human observer with respect to an ideal energy detector, S is the signal level in where n is

a one-third octave band, i1 is same one-third octave band,

the background noise level in

and W is

the

the one-third oct-,e bandwidth.

Equation 1 may be solved for the signal to noise ratio for specific cases of interest. For example, experimental data collected in For Tests IA and IB (see Section III) yield a value of 0.4 for n. reasons discussed below, a 50% correct detection rate and a 1% false alarm rate were selected as values of especial interest for present purposes. These values yield a d' of 2.32 (Peference 7), so that 10 log (S/N)

Thus,

for example,

a signal in

= 10 lojr

2.32

(2)

the one-third octave band centered

at 1000 Hz would be detected 50% of the time, with a false alarm rate of 1%, when the signal is 4.2 dB below the level of the noise in the same one-third octave band. Procedures for Evaluating Aural Detectability of Aircraft

2.

A special graph bas been prepared (Figure 10) to facilitate determination of the level at which aircraft will be detectable This figure embodies the assumptions with specified probabilities. noted above and the conditions of Equation 2 in order to reduce the mechanical aspects of the prediction scheme to a minimum.

The

plot of the threshold of pure tones (Reference 8) in Figure 10 may be interpreted as the level at which an average human observer is capable of detecting a signal 50% of the time in the absence of No plotting should be done below this external background noise. curve.

Figure 10 also includes a special set of ordinates

(indicated by the sloping grid lines)

,

-20-

for use in plotting the

background noise spectrum in background noise is may be used produce

(in

When the

one-third octave bands.

plotted on the special. grid of Figure 10, it

conjunction with the pure tone threshold)

to

the predicted one-third octave band detection levels.

These Intermediate

detection levels are then to be used directly

with a plot of the signal spectrum in generating the

one-third octave bands for

final prediction.

Knowing the detection

level for a given background noise and the

one-third octave band sound levels of the aircraft altitude, difference

at a given

the one-third octave band which exhibits the greatest can be determined.

the aircraft

This difference

is

the amount that

noise must be reduced to meet the 50% detectability

requirements.

The change in

altitude necessary to meet the

requirements may be determined by the method outlined in Reference

6.

procedure is

A condensed summary of the entire evaluation given in

Appendix V.

Figure 10 may also be employed to produce predictions

of detection

performance at other false alarm rates and other percentages of correct detecion. Such predictions are derived by adjusting the 50% correct detection rate via the correction terms contained in Tables IV and V. Some understanding of the assumptions implicit in Figure 10 may be helpful in making such adjustments. The false alarm rate upon which Figure 10 is based is 1%. This rate was selected to provide agreement with test results discussed in Section III and field judgment test results gathered by AFFDL (Reference 5). The 1% false alarm. rate implies that the observer is extremely cautious, only reporting signals when he has considerable confidence. However, someone selected to listen for hostile aircraft might be encouraged to adopt a fairly lax criterion for reporting their presence, since a strict criterion

-21-

4

TABLE IV ADJUSTMENT OF 50% DETECTABILITY LEVEL FOR VARIOUS FALSE ALARM RATES

False Alarm Rate

d'

Adjustment Added to 50% Detectability Level

1% 5%

2.32 1.64

0 dB -1.5 dB

10%

1.?8

-2.5 dB

15% 20%

1.04 .84

-3.5 dB -4.5 dB

25%

.68

-5.5 dB

30%

.52

35%

4o%

.38 .26

-6.5 dB --8.0 dB

45%

.13

-9.5 dB -12.5 dB

=4

rt

-22-

TABLE

V

ADJUSTMENT OF 50% DETECTABILITY OTHER PERCENT CORRECT WITH

Correct Detection Rate

d'

_

FALSE ALARM RATE

Correction to be Added to 50% Detectability Level

10%

1.04

-3.5

20%

1.148

-2.0

30%

1.80

-1.0

40%

2.06

-0.5

50% 60%

2.32 2.58

0 +0.5

70%

2.84

+1.0

80%

3.16

+1.5

90%

3.60

+2.0

-23-

_

1%

LEVEL TO

could lead to an intolerably high miss rate. alarm rate of 25% might be appropriate.

In

In

that

case a false

the long run,

an

observer with a flase alarm rate of 25% would be correct three times out of four when he reported the presence of a signal. 3.

Comparison of Predicted and Subjective Determination of the Aural Detectability of the Schweizer 2-33.

Some judgment tests

with four observers were conducted in

situation by AFFDL (Reference

5)

to determine

the altitude

Schweizer 2-33 should fly to be "Just detectable". the test

claimed to hear the plane at 2200 feet altitude

All observers

(Reference

5).

communication with AFFDL revealed that one observer also

claimed to hear the aircraft

at 2400 feet.

estimated by using data supplied in method outlined in utilizes

the

The results of

indicated an altitude of 2200 to 2400 feet.

Personal

a field

this

report.

This figure is

in

Reference

Actually,

these data resulting in

The altitude

close agreement

5 and the prediction

the example

a predicted

may be in

Appendix V

altitude of 2200 feet.

with the subjeotive

test

results.

Results were also obtained with the AFFDL method which predicted an altitude of 1300 feet. predicticn methods for this

The lack of agreement between two particular case is

absence of tonal components in 4.

the aircraft

attributed to the

noise.

Other Factors Affecting the Aural Detectability of Aircraft

Assumptions about the probability of occurrence of aircraft flights

in

the field depend upon tactical

unrelated to acoustic

considerations.

are beyond the scope of this report, incorporate assumptions It

is

ro'-od in

of this

in

higher levels

Since such considerations no attempt was made to

that strong a priori

of aircraft

determining detection rates.

not imply recognition

entirely

sort into the prediction method.

passing, however,

about the presence or absence

information,

over-

Further,

or identification.

information

could play a large

role

a correct detection does Recognition may require

than those associated with detection.

-24-

00

m

I

M

Knowledge gained from the conduct of the two test number of observations

about the possibility

that human detection performance

extent

noise ratios in

by simple amplification in

is

the low frequency (below

noise spectra,

aircraft

aural

of facilitating

listening aids.

detection by semi-sophisticated

series permits a to the

First,

determined by signal to 250 Hz)

portion of

it

may be possible to enhance detectability

in

this frequency

region.

Such increments

signal to noise ratios are practical because of the peculiar

nature of jungle noise,

which contains relatively

energy at

little

low frequencies. Thus,

a crude horn of dimensions

favoring low frequency

might function as a primitive listening aid.

Electronic

transmission amDlifi-

cation of the low frequency portion of the spectrum could be even more effective.

Substantial gain would be needed to render audible

the portion of the aircraft

noise signature which enjoys the least

masking from jungle noise. Second, in

enhancement of detection performance may also be achieved

another way.

It

has been demonstrated

(Refdrence

4)

that

multiple observers making independent decisions are capable of Thus, two or three well better performance than single observers. trained observers working independently (perhaps with the helo of some listening device) might well be capable of higher hit rates The improvement in sensitivity (and than a single observer. hence, detection performance) so achieved could be on the order of 3 to 10 dB.

-25-

SECTION V CONCLUSIONS Six major conculsions may be drawn from the present research.

1.

The psychophysical Theory of Signal Detectability provides reasonable predictions levels

detectability embedded in 2.

for light aircraft

Human aural detection of aircraft in

which human sensitivity

Agreement

noise signatures

various background noises.

signal to noise ratio

3.

(within 1 or 2 dB of the data) of

to be based on the

seems

the single one-third is

)ctave band to

highest.

between predictions based on TSD and those based of

the AFFDL method (Reference

3)

is

best (within

for signals with definite tonal content. pure tones the AFFDL method is

difficult

from the TSD method by 9 to 15 dB. method under predicts

3 dB generally)

For signals lacking to apply and differs

In

general the AFFDL

the aural detectability

of the signals

without tonal content.

4.

The TSD prediction method specified in

Section IV of this

report works equally well for aircraft

noise signatures

containing broadband noises and pure tones, distinctions 5.

to be made between them.

The efficiency

of a human observer with respect to an ideal

energy detector in 6.

without requiring

detection situation is

a quasi-realistic

For a fixed false alarm rate of about

1%,

0.4.

1.2 dB separate the

25% and 50% correct response points on the psychometric function describing aircraft

detectability,

while 0.9 dB separate the

50% and 75% correct response points.

-26-

C)t

o< c

z

10

ILU

0

0

U--

U

UL

CI

C

U-

el

o0

040

-27-

0*

w

D,

00

LU

L

C.

E

4

-o

4I'.

wi

.F

U

I

>

C-4 -0o

00

Cl) U

*00

00

C)

(n

CN

zu/smuAp Zo

F

.0

*'

(op)

*OAOI

~-28-.

*jnm-d punoS

00 C'L)

~00 k^ u CI-

w L00 ceLL.

LU

0)

CC

tn03 OJJ IOAOIP040P*Jd

9p

Ul -29-~

U0101A,

0

z 0

z

z >

>

%0

,-o 4-

4-(A

(A~ UJQ I.-

oeU cc-

-30-

S2

01

0

]-Jl

1

C)C)

J|

Signals Preicte•d Level Calculated with Statistical Sumfmotion Rule

mox

0

S0

c

0 0

0 1.2...

..

I

6

.

-5 1 2

3

7

8

9

10

J II 7 8 9 10 Signals P~redilcted Level Calculatedl wltý d' n* Rule 4

5

6

TEST I B - WITH BACKGROUND

FIGUlRE 5. F

0

0

0

0

STATISTICAL RU LE

-

SUMMATION

12.

13

11

I 12

I 1,1

14

.

.

..

.

.

.

.

i 14

I - COMPARISON OF PROCEDURE AND d' m ax . I-31-

tq

i

i

C)

-~-

K:'

-C

Cr)

~~0

0

C-)

ff

apC

0

32

UT

-IVi~

C

0

I--

-~

00

2

9Z

0>

o

2E

w

0 0 0

0 -

U-

r WO/WUA

wooo*OJ

(op)

JDAVI *JrI"Old

PuflOS

'-33-

pop

~

-

~

-

=-=-

cc-

z

•l

0 U

*

0

~I(3

U

.e

V.

0.,o••••p.• ,

U-

V)1I

I

l1

SP

ul IC*"rmOd

PInoS

110

S-34-

0

to tCL

00

_________

00 -J

A