UNCLASSIFIED

AD NUMBER

LIMITATION CHANGESTO:

FROM:

AUTHORITY

THIS PAGE IS UNCLASSIFIED

AD818950

Approved for public release; distribution isunlimited.

Distribution authorized to U.S. Gov't. agenciesand their contractors;Administrative/Operational Use; 30 JUN 1967.Other requests shall be referred to Air ForceTechnical Applications, Washington, DC.

AFTAC ltr 25 Jan 1972

Illlilllll -III'

I

I I I I I I I I I I I I I I

o

30 ■

>TJ D C

, h * N S I

BEST AVAILABLE COPY

/~.

^P- AMP A Order 624 Project C

I I

ABSTRACT

0 D C 0 G C c

Undtr Contract AF 3J{6$7). 14648. both routtM of>«r«tlon ol th« Cumberlind PUt««u S«tfmoiogU*l Obt«rv»tory and conttdcrabl« appliad ra- aaarch wrm conducted. Durinf tha laat project yaar. on-tlna raal-tima datac- tloa and identification procaaaing using th« CPO Au« hary Procaiaor waa tm- plamantad and «valuatad at CPO for tha porpoaa of •tudy.ng automatic d«t«. procaailng. Tha CPO Aivxihary Proca«for computer two claaiat of detection outputs, the r»ahar analyeie-of-v fiance etatistic and tha Wiener power etatn- tlc. and one elate of identification output, tha United Kingdom tachniqua. Thai« detection outputs "»are compared (»-line against a fUad signal threshold level, providing a ecntinuous real-time "yaa-ooM output for signal. Howave-. tha lined.threshold detection levt-ls were initially difficult to determme accur- ately and. »nee determined, it was found that they were highly non-tima sta- tionary. Adaptive threshold detectors incorporated Into tha Auxiliary Pro. casaor could overcome the non-time stationarity of tha thresh i detectors. Off-line applied raaaarch waa performed to support tha on-line ^search.

^ fIJUMBB #3 WCLUtm» ^ "

«til i0B»eot is iubjset to apeelal s«p«rt eoolrsls ana aaeb IraaMrtttal to foraiga gowrsMnta or forsig oatieM^ ^f^JV ■ad« aaly «Itb prior eeproTal of äJX BU^K.TXSdt^^^J'

g o e i iii/iv «ilvlelon

III ,IL W-M

^r TABUE OF CONTENTS

Section Till« Pag«

1 INTRODUCTION l-l

Q SUMMARY AND CONCLUSIONS A. HARDWARE EVALUATION B. ON-UNE PROCESSING EVALUATION r. orr-J fE APPLIED RESEARCH D. CONCLUSIONS

B-l U-i U.2 U.5/4 U-3/4

tu SYSTEM DESCRIPTION m-i A. SYSTEM DESCRIPTION B. HARDWARE EVALUATION C. RECOMMENDATIONS

DM m-8 m-io

IV EVALUATION OF ON-LINE PROCESSING A AUXILIARY PROCESSOR PROGRAM B. DETECTION PROCESSING C. UK PBOCESSWG

lY-l IV. I IV-l IV-il

V SUPPORTING DETECTION PROCESSIHG RESEARCH A. PROCESSOR SIMULATION PROGRAM B. THEORETICAL NOISE SAMPLE C. PROCESSING PARAMETERS D. PROCESSED NOISE SAMPLES E. SIGNAL PROCESSING

V-l V-l V-I V-! v-s V.14

VI REFERENCES VI-I/2

APPENDIX

ADAPTIVE THRESHOLD DETECTION I

LIST OF TABLES

T«bl« TUI« Pag«

rv-i PROCESSOR PROGRAM DATA IV.2

IV.2 PRESENT PROCESSOR OPERATING MODE IV.3

v molmrto* ««r m^i^m MJ^trntgirt

i '

I LIST OF ILLUSTRATIONS

Figor« D»icrlpticm Pag«

I-1 AuxllUry Procttior 1-2

III. I Ftiher Tr«n«form«tion at • Function of N2 III-4

IV-1 CPO Dally Thr«ihold L«v«lf IV-5

IV-2 Data Pro* •■••d by th« MCF and Auxiliary Syfttma IV.8 During Known Signal Condition!

IV.3 Data Proccticd by tha MCF and Auxiliary Syatama IV.9 During Known Signal Condition!

IV.4 Thr«. Quarry Blatt« Proc»i««d by the MCF and IV.10 Auxiliary Syatama

tV.S CPC Dav.locordar Racordlnga. 26 February 1967 IV-lS

IV.6 CPO Devdoeordar Racordlnga. 3 February 1967 IV.14

IV.7 CPO Develocorder Recording!. 27 February 1967 IV.IS

IV.b CPO Develocorder Recordlnga, 2 March 1967 IV.16

IV-9 CPO Davelocordar Racofdli^a. 22 March 1967 IV.17/18

V. I CPO Thaoratlcal Nolta Sample V.2

V.2 Power Spectra Comparing "H «oretical and Original V-3 Notie Sample!

V-J CPO MCFO Low-Cufref liter! V-5

V.4 Ftaher Sutntii for .he Theoretical Noiae Sample Uftag V.6 No Filter, 0,75-, 1.0-, and 1.25-cp8 Low-Cut Filtere

V.S Cumulative Diatrlbtrtien for Flaher Sutüt.c! for Theoretical V.7 Noiae Sample a Uatsg Ni Filter, 0.7$., 1.0-, and 1,25-cp- Low-Cut Filtere

V-6 CPO Average Signal Spectrum V.9

V.7 Noiae Sample! I, 11. 111. and IV from 1965 CPO Noiae Library V-10

V-8 Power D«»alty Spectra of Z-10 for 1965 CPO Nelae V-U Samplea I, II, III, and IV

V.9 Flaher Statlatic for Noiae Samplea I. II. III. and IV V-U

v-10 Cumulative DütribuMon for Flaher Stati!tici for Noiie V-13 Sample! I. II. Ill, and IV

, I

c c [

L I [

vl I

0 I i I I 0 0

v-u V-12

V-13

V-14

V.15

GPO 1963 19-Ch»nnel Noiie Sampl« V-15

Fliher Statistic for CPO 1963 I9-Ch»nnel Nolte bampl« V-16

CumuUtive Dlatrlbutlon for Flshar Statlatlc for CPO V- 17 1963 19-Channel Noit« Sample

CPO Theoretical Noiee Sample Plue Wavelet for the V- 18 Infinite Velocity Caee

Cumulative Dletrlbutlon for Fleher SUtletlc for Theoretical V-19 Signal Wavelets

0 §

fl

e i i i i i i i vll/vlll Borvlcm tfhrialon

I ^-

SECTION I

INTRODUCTION

p la addition to routine op«ration of Ihm CurntwrUod Plateau Saitmological Obtarvatory, contidarabic appUad rataareb baa baan con- ducted under Contract AF33(6S7)-14b4ft to advance tb» underitardinf of array processing tecbnology applicable to email seiamtc array* used in the nuclear detection and classification problem. During tb« last project year, under VT/6704 on-line real-time detection and identification pro* cesslng was implemented and evaluated at CPO for the primary purpose of studying automatic detection processing.

r o

r

i i

On-line processing wss implemented using UM CPO Auxiliary Processor (Figure 1-1} which computes two classes of detection outputs, the Fisher analysis-of-variance statistic and the Wiener power statistic, and one class of identification output, the United Kingdom technique. The de- tection outputs were compared on-line against a fixed signal threshold Isvel for automatic detection. From this comparison a continuous real-time "yes-no11 output was provided for signal. This output, along with the de- tection »nd identification data, wae recorded on Develocorder film.

Contained in this report is a brief description of the digital processing hardware and results of the hardware evaluation. Couclusions, as well as supporting data regarding the evaluation of on-line detection and identification processing, are presented. Also, results from off-tine sup- porting applied research are covered.

1-1

^

40t2l

Figur«* !-l. AufcUwirv Prorr««oi

1-2 • •rvtc«« division

c E I r f

ü ^

ncTtoNa

•UMMAJIT AND CONCLUSIONS

r

8 o

D C B I i I

OK-UA« itmiikm» l^MttJlcAlMw *n4 «ttiomAtlc 4«t«ctloe «t CPO m 0»c»mb«r IfM w&k ih» intimiUtitm **4 op«r-

•fdMAM» Pro

i

B. ON-liNE PROCESSING EVALUATION

1. Detection Procesting

!

Evaluation of the Auxiliary Procesaor as a detection device Indicated considerable difficulty exist! with the fixed-threshold Section level« for the Fieber and Wiener outputs. Initially it -va» difficult to deter- mine Accurately the desired threshold level. Once the levels were deter- mined. It was found that they were highly non- ime stationary which meant that the false-alarm rate could not be fixed. Atcempts to adopt standard procedures to update the threshold levels on a daily (and sometimes more ofton) basis proved to be inadequate. The variations in threshold level were eignlficant enough that the automatic detection outputs were for prac- tical purposes us. ess.

Adaptive threshold detectors should be incorporated in the Auxiliary Processor. An adaption algorithm is presented in the appendix which is suitable for incorporation into the existing system. It I« felt that the Auxiliary Processor can be a significant tool in the automatic P-wave detection problem with this addition

2. Identification Proceasing

The two UK outputs were properly implemented in the Auxiliary Processor but are of limited use at CPO for classification. First, this type ol computation is designed for use on a large diameter crossarray (20 km has been used). The CPO array, which is 3. (• KTZX in diameter, lacks sufficient directional resolution and violates the assumption that noise is uncorrelated across the array. Second« the two outputs ma; be programed for only two directions, severely limiting the class of signals which may be studied. Third, classification work requires the preservation of signal waveform for alt events. This causes a ban*' dynamic range conflict with the MCF which is a detection device requiring adequate noise for coherent noise uuppression. The MCF Auxiliary Processor system is limited to a 12-bit (66-db) dynamic range on input. Since on-line emphasis was placed on MCF signal extraction and detection processing, the class of signals available for study was highly restricted. Large signals of interest were clipped on input or during inter- mediate computations and small signals of interest (e. g.. NTS shots) were not detected on the Oevelocorder display. An adequate library of events for study was not collected because of these limitations.

Work in this area was subsequently shifted to tLn MCF and Auxiliary Processor evaluation. The identification processing technique warrants study If the processor Is Installed at a more suitable array loca- tion and sufficient events can be accumulated.

[

c

I I [

[

c [ r i

i

II-2

T

8 B

0 f

0 r

0

A

0 s I I

C. OFF-LINE APPLIED RESEARCH

Off-line Dallas-based supporting research was directed primarily toward determining parameter specifications foi the Auxiliary Processor program and toward investigating properties of the Fisher out- put. Two critical parameters, the integration gate length for the detection outputs and the low-cut frequency filter specification necessary for pre- filtering the Fisher input data, were determined from studies of CPO signal properties. Effect of correlated noise on the Fisher output was studied empirically in relation to the low-cut filter specification. Compared to the Wiener outputs, signal attenuation for the Fisher computation at a function of wavenumber was determined to be significantly greater.

Thr variable threshold problem for a fixed false-alarm rate was investigated and found to be related to the RMS input noise level for the Fisher output. This i -rease could nut be related to the effect of a particular noise contributor, but it is rea^ouaole to assume that there may be a strong correlation to the mantle P-wave noise level.

D. CONCLUSIONS

Much useful information for advancing the state-of-the-art real-time automatic detection processing was gained by the implementation and evaluation of the CPO Auxiliary Processor. Both the Wiener power and Fisher statistic computations appear to be useful for detection purposes, but completely automatic detection should be accomplished with an adaptive threshold device. Such a device can be easily incorporated into the existing hardware with a relatively minor modification.

Little information was gained from computation of the UK technique on-line at CPO. Tim adverse array properties and dynamic range problem coupled with the sparseness of desired events severely limited the necessary library of events required for this study.

_ L« supported research complimented the on-line evaluation and provided a more comprehensive understanding of on-line Fisher and Wiener automatic detection processing.

11-3/4 •olenoe ••rvto«« division

I

n

0 e §

Q

I I 0

SECTION III

PROCESSOR IMPLEMENTATION AND EVALUATION

The purpose of thlf «ection it to provide a concise deecription of the CPO Auxiliary Processor, present results of the hardware evaluation conducted during the on-line operating period and recommend system mod- ifications based upon experience gained through system operation.

A. SYSTEM DESCRIPTION

A detailed system description may be found in the Instruction Manual, Operation and Maintenance Auxiliary Processor and the Specification for Auxiliary Processor Modification, Advanced MuUi-Qiannel Filter.

1. Description of the Key System Operations

a. Fisher Process

The Fisher process is a statistical signal detection process which operates on the filtered outputs of each MCF input channel. Filtered outputs are formed by the MCF in the beam-steer process using the following equation.

I

B n

uo n-i

a. (1)

n is the value of the most recent sample of the data for input channel k (where k ranges from 0 to k)

k k a through a. are constants

I+l is the number of filter points

8 8 I I

By using the CPO MCF, the auxiliary processor computes a Fisher output according to the following equation.

Nl I K3 Fisher Output - (

Kl_ P i (2)

(---) + N2(K2-£r)

m-i •oi«no» ••nriocs division

^

where

Ni and N2 are normalisation comtant«

K * 1 !■ the number of Input channel«

P is the Fisher history length

Kl, K2, and K3 are the Intermediate Fisher terms which are formed using the following equations.

K2

2

•(E E Bn-{p-l)) \p»l k«0 /

P K / \2

pal k»0 %

(3)

(4)

K3 P

] (E Bn-(p.l)) (5)

The Fisher output computed by the procctsor Is s transform of the true Fisher process which is given by

K3 - F ■

M p

K2 JO K+f fi)

Under highly coherent signal conditions, the denominator ©1 F is C, an un- defined state for the hardware aa implems nied- To avoid this condmon, P is transformed as follows:

m-i

I 0 0 E e o 0 c e c 0 c I I I I i

1 ß

i

i 0 0 0 ß 8 I 0

*

Fisher Output F ■>• B F + D

(A/C) F F + D/C

(7.)

where

B = 0

A/C « Nl

D/C = M2

The effect of thie transform as a function of N2 is illustrated in Figure lU-l. Nl sets the maximum value (routinely 777g or 512^, the largest possible output number as a result of the 9-bit output register). Selection of N2 it based on the F-value, computed for the site ambient-noise field (at CPO, F» 2. 2). This may be determined operationally by finding one or two processor-average, ambient-noise output values over a pre- determined gate by using the Fisher threshold detectors and by reading the F value for the known N2.

b. United Kingdom Process

The United Kingdom process performs the 0-lag crosscor- relation of two beam-steer outputs. The auxiliary processor forms two such outputs using the following equation.

e E 8

UK Output = P

r1 ck Si-tp-l) n-(p-l)

(8)

II1-3 ■ol*noa nmrvtomm dhrlalon

I

600

500 N2 • 1

."*ll . '! "I . -.lip—■*—^—

JO 40

N2 - 50

Nl > 100

60

C 1 c

c

{

FISHI» {D

rigui'« m-l. ri«h»r TrMtfo.-nwtioii at » Tmctktm »I Nl

m-4

I I I

n I

0 i 3 G 0 0 0 i

I ß 0 e I i I i I

^.

where

C* .. ■ j* beam-iteer output of the MCF proceüor n-{p-n

Ck .» ■ V:th beam-eteer output of the MCF proceseor n-(p-l|

J ^ k

P « number of UK history pointt

n ■ rroe* recent beam-eteer output

n-1 ■ next moit recent beam-steer output, etc.

A second UK output is generated using outpits 1 and m, where 1 6 m and j, k. ;, and m are not equal.

New t'K outputs are computed every 50 msec.

c. Wiener Power Process

The auxiliary processor forms four Wiener power outputs using the following equatior.

R s 2 Output - £ £ (Aj

r«l 8*1

(9)

where

m » n{i-1) - (r-l)S

A ■ the MCf output at time m, truncated to 11 bits plus sign m

A • the most recent MCF output (A^ , is the next most recent MCf output, etc.)

• iarnple interval sample interval /ram« tyne 50 msec

r ■ iHndow width in sec

sample interval

III-5

^-

Tb« muUlchAan«! pow«r p»»c««»or output is An R x S x iO-m««c window of »n S x SO-mttc •»mpU tntorval of an UCT output.

d. 1 rothold Dotoctor»

DigiUl throihold dotocton aro provided In tb« Auxil»«ry Proc.tior for th« Fiihor output xnd Xhm four powor output«. Tl» fhf.tbotd dotoctort provido a ••pxr.t. tignal output wh«n 1 monitor.d output .qwU or rxcoads * switch-■•loctod lovol.

2. Oonorxl ChxracUrlatics

The design philosophy and gso.rsl •ppsamncs of ths AoxillAry Processor were modeled after the basic UCT processor. The unit consists of an 80-ln. . tingle-bsy rack containing three dr.wsrs: the erlthmetic drawer, the output drawer, and the controller drawer (Figure l-lj. A« wun the CPO MCF. all input and output cabllnf passes through the tap el the cabinet

The spsre rack space in th« Auxiliary Processor may be «aed for support equipment such as the paper tape reader (PTR). data control modules, standard station timing unit, and the rack-mounted oscilloscope*

a. Input Signals

All input data signals are derived from UM basic UCT pro- cessor as described in the following listing.

• Wiener Power Process

Ibis process derive« In input from UM liCFt-4 eiftyMft. Of the 24 avsllsbl« bit« of data. 9 are selected nstag e* the Input-data truncation nwitch.

« Fisher Pioces-

Single-channel data are accepted from the MCFO otitp«t. This allows • ingle-channel prefUtertag of data pri«r to computation of lbs Fisher «utie X th« «vAiiable 24 bits of data. 9 are selected üirough the FisMr data truncation switch.

1 j

!

[

C

[

m*b

[ C t I E

I I I I

I I 0 6 i I I 0 B e i i i i i i

i

^3.

• UK Process

Th« Input It derived from two selectable beam-steer outputs. Nine of the total 24 available bits are se- lected by the UK input-data truncation switch.

b. Output Signals

The digital-to-analog converters used In the Auxiliary Pro- cessor are of the same type as those used in the MCF processor, r or the Fisher. Wiener power, and UK outputs, 9 bits plus sign are selected from the 2S-bit output register using individual output truncation switches, thus giving an analog-type gain control in 6-db increments.

The processor outputs are summarised as

• One Fisher output trace

e Two UK output traces corresponding to two programed area locations

e Four Wiener power traces corresponding to the MCF 1-4 output traces

• On« Fisher threshold trace corresponding to the Fisher output trace

e Four Wiener threshold traces corresponding to the four Wiener » ower traces

e» Program Selection

The following key variable programing modes are offered by means of panel-mounted switches,

• Fisher Process

Normalisation constants Nl and N2 are variable from 0 to 777.. History length (gate length of computation) may be selected from 0 to 999 points.

m-7

• UK Process

History length is varUbl« from 0 to 999 point*, and selection of the 4 beam-steers (2 each for UKO and UK)) is provided.

• Wiener Power Process

History length specifiable in R InterveU, 0 to 99. and S samples. 0 to 99. where the gate is determined by R intervals of S samples each.

• Threshold Detectors

Independently variable threshold levels ar« programable from 0 to 777g.

B. . HARDWARE EVALUATION

1. Reliability

The CPO Auxiliary Processor proved to be a highly reliable digital processing device. During the time the unit was operated at CPO from 30 December 1966 to 10 April 1V67, not one failure occurred. This represents a total operating time of approximately 2900 K/f. Normally the "infant mortality" period, during which most faUuree at* expected to occur because of marginal components, etc. , is set at around 2000 hr*.

In future operations this unit should continue to be quite reliable since it consists entirely of solid-state hardware with the exception of the blower motor.

2. Maintainability

Little maintenance experience was gained on the unit AM W the high reliability experience. Once installation and op«rational chechout was complete, processor maintenance was limited to routine operationt which include periodic cleaning of the blower motor air filter and lbs daily step test (this test is also a part of the MCF maimenan:* procedure and completely verifies the Auxiliary Processor logic and arlthlmeu

r c r

r

m-s

C i c I

I I I e

^

I

I I I I I I I I I I

During ihm wit* UiiUlUtloo Mid op«raticm«l cb«rkout phat«. tome m*lm.«RC« hnowl.cli« w«! Mln.d. Tb« lollowlnf empire»» cot»- clu.lon» c*o b« r««li«d »»••«I «P«» ^" Hmü»l •Jip«rUti€«t

I« Th« p. «•••or e*n o« m*intain«d by »n Individu»! tr«in«d in UCF mAlnUnAnc« If b« r««d« and tbo^ou§hly andcrtund« UM AuxiJUry Proct»»or Handbook.

• Tb« ccmfMiiar wir« Hat provided with tba •yfm *• •

• Tb« nMtfetod required to id UM thr«»bold t«v«i • «tub«* it MHteMMM »»d highly iaAccur«t« •Uc» il t***if *mf»Uiud« mM«ur»nMnt of UM rUhmr «ad WlnMr o«ttp«H« M tb« D«v«tocord«r dltpUy.

R»csBim»ndatiofii fe» •itr^lUyiai ••l«p »ad op«r«ti98 ef tb» • ytwm follow. Tb« thr.tbold »ovol «4^t»«Ä probUm li dlita.Md Ui mor« d^ftU l« SMOO« IV aod iho

C. KSCOlUAENDATlONi

In ord»r to •(mpllfy UtU*l MM« •»< ^orAttoo of UM AMslUary Frocotcor tytum. UM foUowiog cb*ag«« »r« rocommottdod for poiftibl« frttir« lllHwi bio«« of UM»« coold b« ««tlly lAcorpontod telo UM «KUüRI b*f#w«r« wtlb ^y mioof «*edtf4e«ttiw«« T^ tbrtfbold dt- l«€ior clMOf # I« eoe»id«r«d «ta«tttl«l for prop«? op«r«ltoe of UM tyttem •• «a MtooMttc d«t«€tlott dovtc«. R«comnMod«d cteog«» «r«

• SvttlAf UM risb«r tr«n«form coo«tA«t Hl to m CiMd «UM of 777 la order that UM WMrtmnm rtib«r o Mtput «OIIM will «(UM! UM iiMitlmwii nMpil ■■«b«f (9 but). OAIA cootrol woald «tUl b« •»•iUbU ta Ik« FitlMr DA ««ritcb.

I (

[

c

L

•«tttaf lb« rub«* traaafora c «ml«« of 20. ataca. fro« vbai la I «i^acMd all« ^ T for a r ra y data, ada^aat« ^tmb«f »tpiliu*^« raluaa «ad traaaforoMd Plabar m-n.

Itt tea t%w»4 w teMn a b©«« UM tbta MUIM iwnuraa

valMB tFlfar«

a t'ti BMte for dtaplaytRf lb« Tithmr itamr- tart t A. K2 «ad IU la ord«r M tMlUUH

•«ttl»t t^« F »aMr lfm Md tat«rwM#bH date trewttt« Tba«a tarn»» eoaM IM diaptay«! oa ih»

««tpttta «b«a la lb« laat

• Frp mittag a« tapMl ebaaMl MIMUMI eapablUty f«g A« riabaf IflpM. Al praa««! all IdCf in^i cbanMla «aa< la «ay baam^Maar aolpvt ajra »aad la UM F eo«r- pntiHoii. Fat a—ytaT U U «Mr« daiit«d M lap«* I S-cotnpottoat ««mat to UM liCF fat Ad ptfyw« «f I ^Mcy filtartag. d»iaytet »* dl*^«ftet a^nat w^o IdCF Jata. lb««« iap^i «o«M at praa^at to taclvted la ib«F

i i

f

m-i«

I I 0 s I

• Modifying th« thr««h»ld levtl •witch«i for the Wi«ner and Fisher outputs to be adaptive rather than fixed.

■

I I c I s s

I I I I m-u/iE alVMMSfl

I I e c

i

8 0 P

SECTION IV

EVALUATION OF ON-LINE PROCESSING

The primary objective of the Auxili&ry Procenor on-line evaluation phase wag to determine the feasibility and effectivenees of on-line automatic detection processing. As an integral part of this evaluation, it was intended thkt data be developed demonstrating the interdependence of de- tection threshold and false-alarm rate. From this data the two detection processes, Fisher analysis of variance and Wiener power, could be compared for detection capability. Data from the two UK outputs were to be studied using the British Classification scheme.

The following paragraphs describe in detail the on-line evalua- tion phase. For continuity, a description of the processor operating configur- ation is presented first. Data for this phase was developed primarily by station per onnel as a part of the routine observatory operation.

A. AUXILIARY PROCESSOR PROGRAM

The Auxiliary Processor was cperational or line at CPO from 30 December 1966 to 17 April 1967. Table IV-1 prevents a summary of the fixed program employed in the digital MCF and Auxiliary Processor during this time period. Table IV-2 presents a summary of the MCf's and beam steers employed in the ptocessor.

As indicated in Table IV-,. the threshold levels were varied daily. Integration gate lengths for all outputs were initially programed (or 2-sec, but were changed to 3-sec (60-pts) on 14 February 1966. This change in gate length computation was based upon a study of mean-signal-length con- ducted in Dallas. The Fisher, Wiener and UK computations are optimized when the gate length of integration equals the P-wave duration. A mean P-wave duration of 2.9 sec was determined (SectionV).

B. DETECTION PROCESSING

r

e I i

ResuUs from on-line analysis of the detection outputs indicate that ^he Fisher and Wiener processes were properly Implemnnted and (hat this type '»f oti-line computation may be highly effective in automatic ?-wa •» de- tection. Quamitive data demonstrating the effectiveness of the present system could not be obtained due to the effect of non-time stationary Fisher and Wiener Power noise distribution on the rrreshold detiction outputs.

IV-1

UMH

TfcblelV-1

PROCESSOR PROGRAM DATA

i

r i

MCF PROCESSOR

Channels 19

FUUr points 57

Multichannnl filtert 5

Beam-steer ■ 9

Signal conditioner All 0%

D-A converter Channels Channels

0 to 4 « -3 11 to 14 « -3

Beam-steer history 15

Time delay 28

c c [

AUXILIARY PROCESSOR

Arithmetic drawer

Output drawer

UKO 56 UK1 78 UK hietory 60 Fisher Nl tTI Fleher N2 20g Fliher history 60 MCF power R ■ 60, S ■ 1

8

UKC -3 0X1 -3 MCFO -4 MCF1 -4 MCF2 -3 MCF3 -3 Fisher -15

Fisher -3 UK 2 AC power -2

D»ta truacatloa switches

Fisher summation *7 truncation switch

Note: UKO and UK1 are absolute magnitude. Threshold switches are va.-ied Ml a daily basis.

IV-2 eoienoe servtoM dlvhtton

c

c

e s

I I D

5

I D 0

e

G

G

TitU

BSC*

BS1

BS2

BS3

BS4

BS5**

BS6**

BS7 **

BS8**

MCFO*

MCF1

MCF2

MCF3

MCF4

T*bl«IV-l

PRESrNT PROCESSOR OPERNTmO h DDE

Dticrtptlon

Straight ■um (Zl • Z 19)

North, velocity - 12.6 km/i«c (21 - 219)

Cast, velocity ■ 12.6 km/iec (21 - 219)

South, velocity ■ 12.6 km/i«c (21 - 219)

West, veloctly > 12.6 km/tec (21 ■ 219)

In-line eummetion toward USSR using 24. 6. 7. 9. 14. 15. 17, and 19

Traneverte •ummation toward USSR perpendicular to 3S5

In-line gummetlon approidmately toward NTS using 21, 2, 8. 10. 11. 12. 13. and 18

Tr»n«ver«e aummatio.' approximately iowerd NTS perpendicular to BS7

0.75 cpe low-cut filter

MCFJ* convolved with 1.0- to 2. 0-cpt bandpati filter

IPIOWGS^ convolved with 1.0- to 2. 0-cpa bandpaic filter

MCF31

MCF24 t

* MCFO «ubsection prefiltert the i9-channel data and preienU it for use in the BS subsection.

aa Used as inputs to UKO and UK1 of Auxiliary Processor.

^ A complete description of the filters is discussed in Appendix A of CPO Annual Report No. 1.

I I I IV.3 soisnoe aarvioaa JlrHHw

1. Automatic Detection

Two b»»ic problemt whicli aroie at CPO in attempting to imple- ment end evaluate automatic on-line P-wave detection with the Auxiliary Pro- cessor syttem are

*• • The defined detection threshold for a fixed false-alarm rate

was highly variable. Variations as great as 3 to 6 dh were observed during a single day.

e The method used for rapid determination of the threshold levels for automatic detection was, at best, inaccurate and cumbersome.

The two problems actually interrelate to the extent that the time variability problem could have been partially overcome by adjusting the detec- tion threshold settings often (each hour), but a rapid and a :curate technique for determining the desired threshold level for a fixed fslse-alarm rate was not available with the existing hardware and setup. The inaccuracy of the threshold determination techniques could have been overcome by a slow process of re- cording outputs on FM tape and conducting off-line analysis to determine the desired levels, but this was useleus due to the non-time stationarity of the Fisher and Wiener power output noise distribution.

a. Threshold-Levei Problem

A procedure was adopted at CPO to update the threshold levels on a daily basis (or more often, if required). This procedure, although inaccu- rate, did establish an approximate representation of the output noise distribu- tion in terms of the daily threshold-Wei setting. Threshold levels for the Fisher output and the four Wiener power outputs are shown in Figure IV-1 for most of the period the processor operated at CPO. The level variability is evident in this figure.

The exact cause of the variation in output levels io not known at this time; however, it is reasonable to suspect several sources. In the case of the Fisher process, changes in the mantle P-wave noise level and in the level of the trapped-mode noise are probably the most significant contributors. Mantle P-wave noise vtriabllity as well as changes in trapped-mode noise direction could effect the Wiener detection outputs.

Results from the CPO noise analysis indicated that the predom- ȧat

0

D r

I o D

D 0

D B D

G

e IV-5

I 1 i H

I 2

Recent work under coniract AF 33(657)-12747 revealed one •ourco (pratiure areai and itorm centers (hurricanei, tropical etormt. etc.). It is reasonable to assume the level of this energy would fluctuate noticeably as a function of distance, intensity and possibly direction and would significantly affect both detection output levels due to the coincidence with signal energy in velocity space.

b. Threshold-Level Determination Problem

Determination of the desired threshold level for the Fisher and Wiener power outputs would normally be baseJ upon a fixed false-alarm rate which could be determined from knowledge of the detection output dir.rlbution values, assuming that the distribution remains stationary over a reasonable period, such as a season. Under this assumfllon, off-line processing of de- tection output data would be reasonable to determine the amplitude distribution and subsequently the desired threshold. Detection output data could be collec- ted «i FM tape, digitUed and the distribution computed.

Since the apparent amplitude distributions at CPO were non-time stationary over periods as short as hours, ruling out off-line processing for level determination, an on-line procedure was established which, In effect, was a trade-off between accuracy and speed. This procedure, although per- formed datly and more often if required, proved to be inadequate to handle the non-time statlonanty problem. Probably this due, in part, to Inaccuracy.

The procedure determined the mean amplitude value from a 5- min period of noise using five noise measurements at approximately 1-min in- tervals. Each measurement was made over 10 sec of data and Involved de- termining the amplitude value of the peaks and troughs and arranging them to obtain a mean output level. The five measurements were arranged to deter- mise the dally mean value for each detection output. The mean was then mul- tiplied by a constant to determine the threshold value. Several constants were tried, based upon an assumed distribution, with the constant 2. 5 drtermlaed best from observed data.

In practice« this technique was limited by the absence of an accurate vertical scale on the Develocorder. A make-shift scale was applied to the film by placing the output D-A switches for the Fisher and Wiener power ontpots in the "+" and "O1' tost modes. This produced a sero and maximum value (777.) on the film which could be used as a scale. When in this test mode, risfter ami Wiener power data is not output.

c. I» "»commeodatloiis *. I

Automatic detection could be properly Implemented if the

IV-6

I i

I i I

I I I I e o i 0 0 0 0 0 e i i i

i i

threshold lev Is couia be varied accuretely esd rapidly to compentete lor i tr.bution change• in the Fieher and Wiener power otttpute. A euitabte apprMch would be the incorporation oi an adaptive threehold device into the emieiiag hard- ware which could be made to adapt to data hittory. thu# meurinj e reUttvely conaiitent faUe-alarm rate. By ueing an on-line «Jeptjve aliört»hm to update the threehold levele constantly, problem! in meaeuring dietribution. etc. could be avoidsd. Only the specification o£ a constant multiplier to be applied to tome output property such as the mean value would be required.

The appendix presents an adaptive threshold algorithm recomm-nded for incorporation into the existing Auxiliary Processor.

rhich la

2. Visual Analysis

Since problems were encountered using the automatic detection thresholds, an attempt was made to analyse the Wiener power and * laher out- puts for detection purposes. This proved to be unsatisfactory to the station analysts because the effect of smoothing of the integration gate destroy«d signal energy content and waveform, and made the transition from noiae to signal areas for srr«!! events quite gra iual contrasted to the normal sharp break de- tectable for a ?-wave arrival. Also, the "eveot signatures'. whsch were com- mon 1/ used to detect low-level signals, were not easily recognised after detec- tion processing.

In addition, the absence of a aero 'level for each t the traces made it difficult to determine /ssually relative amplitude on the trace data.

3. Compariaon of Fisher and Wiener Power

Limited comparisons of the Fisher and Wiener power processing at CPO were made by the observatory analysts while attemotiri to analyse this data for event detection, including:

e For high velocity signals, the ratio of the peak-aignal output to the SMS statistic is larger for the MCF precedes than for the Fisher output (Figures 1V-^ and IV-J,)

• The Fisher signal response to quarry blaets nh&m the Fisher outputs to be unaffected by P- and S-».«r« energy falhng in the velocity ranges 6.1 to I. km/ret and S. 25 t< 3.51 km/sec, respectively (Figure Vf*). IL*;.

The Fisher signal response property waa alao investiiated in off-line research (Section /} and found to decrease rapidly with increasing wave. number. This property proved quite useful at CPO fc dlstlpguishmg local and near regional events from teieseismic l^-wave energy and on several oc

IV-T

mi

• • • •■ • tt §••

00:11:46.7

1

04:26:41.9

• c» -

MVI

• •• • • i

■ •I • II • l«

■ »

■

«51

■-'•-••'''v*"'', v.w.v.vv»".'.'. A^•*,,|'/*

'v^^^VSA,

> -^.

06:17:37.5

•«»i

'M It I • «• • •• • 14

L '^■W »^^l^^^^^^^

-V-sA

7^

JANUARY 6, 1967

Fifyr« IV-2. Data Preceised by the MCF and Auxiliary Syitems During Known Signal Conditions

r

L n u

r

r p L

r

r

r i

rv-e I

0 0 0 0 0 0 D Q

0 I I I I I I

MCF» i «CM « PI (Ml*

mm

met i Her t men Her«

• to ■• i Mt • II • 14 IT

0«:16:51. 1

__1 JANUARY », mi

^ww^w/^V^A^

«er i ■en Men MCf « ■ ro ■It ■ it • is • I« IT

:;;;* MCP» I MCM • fIIHI«

MCM >TM

Figure IV-3. Data Proceased by the MCF «nd Auxiliary Syttmmm Durinx Known Signal Conditiont

IV-9

*

Ml n

tit in • t«

MiMi4f

*wi^l^W00MW^

I I I B E I E B B 8 i i 6 B I B i I I

^-

«••ietia »14*4 %m 4*fr%i»$ UltnHwll !»•••*• •»•»!? 4urm§ mtttrf kt*H

0. UK f*flC^M^NO

AMIYJI« «f IIM two UK •titpvtt «I CFO iii4lc«i«d ihn th>

thru« UmttMtWM «tt«l. l«r« tMM* l»y Ik« pr K^lMnMlMiM *** ll» «Ä«f .a*«4 kf CPQ •«•« pr»p«rt»««. ttMW« Un • «r«

• UM two UK oatpaU if compitt«^ from 4 ?i»#4 progiu.i *Htch M •••••£• f«ftrict» m««ilic«i «rorli to «wo •ign«i •o«rc«0. At CFO UM «Nrtfralt ««f«» r ompwliit f«r NTS »M for »o~»i».

• C1««^! i»«*«d upoi» «h« UK tclMmt rw^rci f>r*i«rv«ticN) ol tipMl ••«wf©rm. Tfc» MCF A«. ii«rf Pr«c«*«or •fMAin It Umii»4 I« 7t>^ rffwunte r«l»t« IM tn^al (U bu»). StOC« IHU lytlvr uwd prtmarilf for il«t«clio« ba««d vpan tuppimflon •f colMrool ROiM« tlM inyoi MlM l«««t mutt be tul- fuicollf higb to is*«*« •4«^%»«i« noit« MMttiir« (or MCr proccsatii«. TIHM. Urg« «IfMl« or« cUpfwd

• • UK l«i:J»B4^u« I« 4*tii^ for A^tlcoll^ to taN t«rmo4ioi« »too crooMrr«? 4*««

I« rii«f« IV.S tl^ F-co4» rmpidly

I 0

0 0 0 D G 0 0

fl

0

G e i

ii i

00

f—i

s

<

Ö u u en

» «5 -» •- ^ «■•>>»

• *•••■-•*» >♦

■ OX

00

9*

I JO

0 u t) el h «

O u o «

Q

i

« u

«.*. «. a. • • mmm w^.,

a a ■ PUU u-SttK aaa a^S»

IV-13 «oi*no« a«nrlc«a division

(M • • o

>- « <

0 Ü u to

J I- ». • « o H H ^ ■ P ,

o

».-.».•. «a« •*N(fc*».», uu u w • * « •• «b^C^ 9Ma a wuuu —

aaaa^

2

i-i

3 o u

« u V

•o u o u 0

.-< V

«

&

I > e u 60

i i B C 0

c

0

n u

IV-14 divlaion

0 i e e I

I I 0 6

I:

I 0 0 e

e i i i

m o

o

00

I

-m m * o - «.k. h. lb N * uo u w • •

•■*«■.£» www

a«

u h l«

M

«

j E o 6 v «

SI S o i o ■ ■ P o K u CD

> V Vi

M

IV-16

I I B 0 0 s e D I 0 e i o 0 I I I i E

r- l

n

0 0 0 D ! 0 Ö

0 0 0 B I I I

T. vn o

OH < Q Z o u w to

Öl rf fe * ■ m > »

• • in o

— mm ♦ o - •• * ♦ ^■ ■.».IL *. m mm ««M uuu u ■ ■ ■ • • an a • k. It.

saa

U

—<

u h rt

rsi N

* ■

s o u «

u

0 u 0 r-l

> « D 0 P- U

i > v »4

P

IV-17/18 •oianc« aarviowa dlviaion

SECTION V

SUPPORTING DETECTION PROCESSING RESEARCH

In conjunction with on-hn« opvration of the Auxiliary FVocessor, off-liAe reiearch was dtn« in Dallas. Primary research goals were to verify the logic of the Auxiliary Processor, to determine and study and parameters used with the different processes, and to study the effects cf varying these para- meters.

A. PROCESSOR SIMULATION PROGRAM

A simulation program to stuJ- the processor logic was written which used the IBM 7044 computer. This program simulated the logic of f th the MCF and Auxiliary Processor and checked for round-off errors in the processor computations. A complete description of the program may be found in CPO Special Report 3°

B. THEORETICAL NOISE SAMPLE

To provide ' apresentative CPO ambien. noise statistics for use in studying the Fisher statistic while minimizing computer requirements (i.e., number of noise jamples required for processing) a 5000-pt theoretical noise sample was synthesized. This sample was designed to have the same correlation matrix as the CPO noise-ensemble samples A, B, E, F, and I» developed under Contract AF 33{657)-12331. Due to the noise field time stationarity (CPO Special Report No. I4), it was felt that this 1963 average noise ensemble would provide a statistical representation of the CPO ambient noise field.

The technique used to develop this noise sample (Figure V-l) is discussed in Section III of the Array Research Semiannual Report 3 . After the data were developed, the noise sample was tested for the same properties as the desired correlation matrix. Power spectra were computed from random correlations using both tht theoretical and original noise samples. These power spectra (Figure V-2; show that the theoretical noise sample has the si me properties as the original noise sample and does represent true seismic data.

C. PROCESSING PARAMETERS

Choice of correct operating parameters is of primary importance in on- line operation of the Auxiliary Processor. Research in Dallas has resulted in optimizing two critical parameters; the signal gate length and corne- frequency of MCFO which is the prefilter for the Fisher process.

V-l eolenoe ••rwioee division

p I I > I I—I—> I f I I I I I ' '

^^v^yrtf^*

iii^ iMiini^ib^iitV^nMiiiiii^fcyiriiiiirMri'yii 1» ^PfTi^^nrVWTTT^^'T''rw^^^^r'TTw^TrfTT^^w -j i t

W a • '

TNMfMO

Figur« V-l. CPO Theoretic*! Noiie Sample

0 0 I D i 0 0 0

c

e c

V-2 diviston

v N*

I I I I i e 8 0 0 0

I I

I I I I

' ■ o o to to *5 $S tS f.o

FfKQUCNCY (CI^S)

Figure V-. . Power Spectra Comparing Theoretical and Original Hoite Samplet

V-3 dhrJ«*«n

1. Gate Length

To «»Ubliah the optimum gate length for th« Fieber mad Wiener compulatious, a itudy w»e m»de to determine the time duration of the primery P.weve iignal from teleeeleme recorded over e 6-dey period. The proceetee ere optimised when the computation gete length «quele tho aignel duration. Data for thie etudy came from a vieutl interpretation of the CFO 0ev*iiocorder film. P-wave duration from 57 teleeeitmic eignel« wee meecured, yielding a mean P-pulae iignal duration of 2.90 eec and a variance in P-pttlee iignal du- ration of 1. 90 aec.

Upon completion of Ale et«dy. the Auxiliery Proceeeor program lor the Fisher, Wiener power and UK proceeeee waa changed from 1. 0 to 1.0 eec. After thii adjustment, analysts reported data eaeier to analyse, since the proceesed trsces (noUbly the Fieber) were etabilUed and did not correlate ae well in time.

2. Optimum Low-Cut Filter

Several low-cut filters with verying corner frequencies were developed to determine the optimum low-cut filter to be used im the MCFO sub- section of the processor. The choice of this filter is extremely imporfairt eince it is the prefilter for the Fieher subroutine. F»Uer»ng of the input data to the Fisher process is necessary in order to remove highly-correlated low- frequency microseismic energy so that the Fieber aesumptloo of spatially «- correlated energy is approximated.

Analysis of filters applied on-line at CPO (Figure V-S) showed the optimum corner frequency to be approximately 1.0 epe. Tfrt filters wtÄ 0.75-, 10-. and l.ZS-cps cornsr frequencies were subsequent!) applied off- Jine to the theoretical noise sample (Figure V-1) and the Fisher output eona- puted for each of the three cases. Mesulls of this processing are thowm m Figure V-4. To facilitate the interpretation of thii data, a cumulativp-distrl- butioo counter wac added to the Fisher subsection cf the simulation program. Th- cumulative dietributioni for this study are ebowa in Figure V-S,

eWhen interpreting Figure V-5 and the other ctnrmlative dietrtbutions presented tn this refwrt. the horuontal scales are net linear. Cvm tbMgb thm curves may appear very similar, their true meaning can be interpreted by noting Ae last several Inches on the horltontal scale.

I I i

V.4

E I E

t E I f I I r

I I I I e 8 e i i i I §

c MCWftUW u en mm Lncnmm

rigurm

10 M

EFO MCt9 hm fhä Pr«f)tt«ri

M

v-i

I I I

i E i E e

v.t

I I I i I HI

I

I e

e G i

e i a i

i i i i

i

o

iZu.u.

i • §£££ \ =it>oo o

1 ££8» 2 1 Mo'-J-!

iii • t i w-

i i 1 o -

*■

\

\ o \ i«^

i \ \ * o t « •

\ a \ o Ml

\ \

1 \ 3 \

8 V o ill

m"

F \ N^% o

\N^ "" ^^

oj

^^^ ^^ ^^ *' '**'•••.. o ^******^T* ***"***, •

^^m*m^**~*^—m ' '^'^ "'«^^ -—WM , ^^ *••»*•**««*»**», ••«,„

1 -.1-.. 1

ounmjo-QN

h 43

H o. o i

g *

»«0 11 n . ÜJ •

•2 J »4

o a o i X.

O Z m B

3 ^.

U d5

«

V-7 dMsi»n

wmmmmm

Figure! V-4 and V-5 show that the optimum low-cut prefilter ihould have a con.er frequency of 0. 75-1. 00.

Greater suppreiiion ol coherent noise could be obtained by choosing a filter with a corner frequency higher than 0. 75 cps, but this would cause extensive lignal degradation since the predominate portion of the P-wave usually occurs between 0.6 and 1.0 cps.

The average signal spectrum was investigated to determine the optimum trade off between minimising the Fisher noiae distribution (low-cut frequency filtering) and maintaining adequate si energy. The average CPO signal spectrum was computed for an ensemble ol Kurile Islands events de- veloped under Contract AF 33(657)-12747 8 and is shown in Figure V.6. Peak signal energy occurs at 0. 68 cps. Since this result could be tuned to the Kurile Island ■ region, and not represent an average of the CPO predominant P-wave energy, additional P-wav data was derived from the CPO standard analysis forms. An average predominate P-wave frequency of 0.82 cps was determined from one month of standard station report data.

Based on the Fisher noise output distribution shown in Figure V-5, and with the knowledge that the peak-signal epectrum lies between 0. 68 and 0.82 cps, the 0.75 cps low-cut fUter (Figure V-3) was chosen for on-line application.

D. PROCESSED NOISE SAMPLES

1. Threshold Non-Time S*.ationarity

The Fisher output noite distribution non-time stationarity was studied. Four noise samples. I. U, m, and IV, wer« chosen for processing from the 1965 CPO library (Figure V 7) and represent the different types ol noise backgrounde encountered at CPO (low-I medium-U and III and high—IV). Power deisity spectra for these samples tre shown in Figure V -8. The Fisher statistic *or the four samples is shown in Figure V-9, acd the cumulative distributions for the Ffsh'r statistics are shown in Figure V-10. These illustrations show the Fisher statistic to be larger for high noise (Sample IV) and smaller for the other iamples.

Although Sample I is of low-level noise, its Fisher itatistic is *arger than those from Samples 11 and m. While the cause for this is unknown, it is possible that the mantle P-wave noise lavel was much high*.'- for Sample I

«The thooretical noise sample (Figure V-l) hts been prewhitened with a 1. 0 cps 12 db/octave low-cut filter which closely approrimates the frequency fil- tering accomplished on the MCF analog output data m the UCF signal condi- donini Irawer.

r

[

[

[

f {

i r

8 e 8 o Q

0 0 f

f

I I I I I I I

I

£

2.0 3.0 f(cps)

F.'gure V-6. CPO Average Signal Spectmm

V.9

»■ I 'I—i—r—r- -T—t—r—«—'■r-'T-'"»—r—«

ftha^iiHiLU^yi'iÜiMWhlfMihflUfllii

^-,1.1 -.1 Oiii'u.fliMUiia^iithJdlfiifcfyfiiii

11 i i i i i i i » ■»■ i » *

> ... ■ ■ j I i 1 I I I ■ •- • ........—i i < * J UHrwian

Figur* V-7. Noiie Swnplei I. U. ill. wxd IV from 1965 CPO Noil« Library

I I c r c c E [ l C f

1 c

[ [

V-iO

8 0 0 0 n

B

o i i i i i

_ —. >

is cr>

t/» u> V» V»

ü i i !

| :

CD

o

5 o

2

SdDO'UVSdO/NOIiOWQNnoaO (»-'UJIO'I Oi3AllVT3H

&

1 ^ i u *

5 IT

■fi o

u 9

4

> 9 M

I I

c

E

C L

I

I I 0 s 0 D

e o i i

i i i i

>

1 a a

o 2 u o

i M

o

I

I»

I I ■

P

V-13 »atotio» »TiflQ»» dlvialeft

than for Sairt'

^—, F t ■ » ' I I ' ■ * » ' ' ' u

I

i

I I Figure V-ll. CPO 1963 19-Channel Noise Sample

■ ■■■ » ■■'■''■** j » *

y THHISTI

V -15 »o»e«oe ••nrlo«s «fivleton

re.BHMMMU

* p

r ;

f ir»

J B

5 ie 3 3 indJiK)a3HSIJ

1 •

I

8

•

J3

M

I

V-16 •Mvle4Mi tfMalcw

I t i I

t

n f

I

I f

S e I

I I I 0 I

R e B 8 0 §

i

i i

^-

V.17

n *ftMM* t » r >

-J I—LI« r i r i T

Figure V-14. CPO Theoratical Noise Sample Plus Wavelet for the Infinite _« Ytte^ltY G»ia

V-18

f ß [

E E

L

c

c f I I f

D G I D i

I e i

s sf

I

|» i i 11

in» a a mm

liiiiii i i ! I \ I

-L I

rTUIVVS JO'QN

1 E I

o

1 I u o

o > k

'S li a o

I

i

« i H

I V-19 -■i^ - -., ■""'-

tfiwl^io«

• The size of the Fisher values rapidly increase as tie velocity of the incident signal increases.

Tne decrease in signal output level for the Fisher computation is greater than would be expected on the basis an array response comparison, and explains the Fisher property of significantly suppressing quarry blast in- formation (Section III). When compared with Wieua r pwoer processing, the signal suppression property is more severe for the Fisher output. The Wiener power output response is easily determined from the filter response, which for the CPO MC^'s in Table UI-l at 1. 0 cps is

Filter Infinite Velocity 25 km/sec 12 km/sec 8 km/sec

MCF1 -0 db -1 db ■4 db -5 db MCF2 +3 db +2 db -0 db -4 db MCF3 -Odb -1 db -4 db -5 db MCF4 -3 db -3 db -3 db -4 db

Maximum response variation in the MCF data is 7 db for MCF2, while the Fisher varies 12 db over the same velocity range (infinite velocity to 8 km/sec).

Fisher output signal degradation is partially implied from th» observation made in Section m regarding the peak-signal-to-RMS-noise output comparison with the MCF processes.

Fisher output responre can be maximised in practice if the input data is beam-steered to enhance signal energy prior to computation of the Fisher statistic. This approach, although well suited to off-line processing, would be impractical with the existing hardware because of the fixed program restriction.

!

[

0 (

[

c [

V.20

f r

I ^r

i

2.

3.

4.

5.

6.

7.

8.

9.

10.

I I I

SECTION 71

REFERENCES

Texas Inotruments Incorporated, 1967- Instruction Manual, Operation and Maintenance Auxiliary Procesinor, Contract AF 33(657)-14648, 1 Jan.

Texas Instruments Incorporated, 1967: Specification for Auxiliary Pro- craaov Modification, Advanced Multi-channel Filter, Contract AF 33(657) -14648, 7 Feb.

Seismic Data Laboratory, 1965: Analysis of Variance as a Method of Seismic Signal Detection, Rpt. 116, Contract AF 33(657)-12447, 25 Feb.

Texas Instruments Incorporated, 1967: CPO Ambient Noise Study, CPO Special Rpt. 1, Contract AF 33(657)-14648, 30 June.

Texas Instruments Incorporated, 1967: Array Research Final Rpt. . Contract AF 33(657)-12747, 20 Jan.

Texas Instruments Incorporated, 1967: Auxiliary Processor Simulator, Cumberland Plateau Observatory Spec. Rpt. 3, Contract AF 33(657) -14648, 9 May.

Texas Instruments Incorporated. 1965: Array Research Semiannual Rpt. 3, Contract AF 33(657)-12747. 3 June.

Texas Instruments Incorporated, 1964: Array Research Semiannual Tech. Rpt. 2. Contract AF 33(657)-12747, 15 Nov.

Tex-s Instruments Incorporated, 1965: Array Research Semiannual Tech. Rpt. 4. Contract AF 33(657)-12747. 15 Dec.

Texas Instruments Incorporated, 1966: Cumberland Plateau Seismo- logical Observatory Annual Rpt. 1, Contract AF 33(657). 14648. IS Sept.

VI-1/2 •eteno* ««rvloa ilvtalwi 1

I t t I t I I I Qj

I I m ■

APPENDIX

ADAPTIVE THRESHOLD DETECTION

I I I !

I I i f f E

I I I I

APPENDIX

ADAPTIVE THRESHOLD DETECTION

At present, use of the existing MCF Auxiliary Processor system for automatic detection is limited due to non time stationary Fisher and Wiener power output noise distributions. This 1: nitation could be over- come by incorporating adaptive threshold detection into the Auxiliary Pro- cessor.

An adaptive threshold detector is one which automatically ad- justs the detection threshold levels based upon the output noise history as de- fined over some specified period. Associated with the automatic detection capability must be a specified false-alarm rate and a learning rate.

Since a modification to the existing hardware is recommended, the storage requirements for the adaptive algorithm should be minimized, and the modifications should be incorporated into the present Auxiliary Processor drawer space.

A. RECOMMENDED SOLUTION

In order to meet the assumed hardware restrictions while still providing an adequate technical solution, it is recommended that the thre. hold level be determined from output history data which has been smoothed by an exponential time window. Mathematically, the output Fisher or Wiener power level would be described by a function X j + j given by:

J

whc.e:

n=0

i « time

-na

j a output channel {one for Fisher and four for the Wiener power)

XJ ■ Output value of the jth channel at time i-n+« '. i-n-fl

I This system provides for control of the learning speed with time constant a, and includes all past history data to determine the «utput characteristic JC' ,. Ths threshold would then be given hy some value

I I

l-.

T = K • YJ

where K is a constant which sets the desired false-alarm rate.

The exponential jmoothing technique provides for weighting of the output statistic x[+ j as a function of time, i.e.,the most recent data is weighted heaviest.

B. IMPLEMENTATION

The above equation can be simplified for on-time application as follows:

x| =I0 X| , e"na

i + 1 n=0 i-n + l

X?. . + i . X ' "^ i+l n = 1 i-n + l

letting

n = n 41

i + l ' Xi+ 1 n'^O ^-n' *

However, X, I xj •■n,a FsO Ai.n'

«ft

Therefore, x{ . ■ x| r Xi e

molmrm» ■«rvte«-. divtalon

(

{

f.

c c I 6 r

r

c £ I i i

I I r i i I i i

and the updated threshold is given by

J i + 1 = K X^ i +1 = K i + 1 +

XJ

The threshold computation for the one output is this reduced to

two multiples and one add operation and storage requirements are reduced to a

single word, X. •

A preliminary investigation of the hardware requirements for

implementing this adaptive threshold algorithm indicates that existing hard-

ware can be modified to include this capability with no increased core storage

or drawer space requirements.

m

3/4 •Gl»no» MH-vt«M tfhrtalan

8 Unclassified Stcurity CUmiflcatton

DOCUMENT CONTROL DATA • R&D n»€uHlr tslfltlcallan ot Ulla, txny el mbslnci «ni< Indstlng tnnnlmlKm iau»l fc« mtfnd uhmf ** »fmll n$»H to clammlllma)

I ONIOINATINO ACTIV/ITV (Corponim Milhot) """~~~--~"^~———^——-

Texas Instrumentb Incorporated Science Services Division P. Q. Box 5621. Pallas. Texas 75222

I«. MtPOMT liCUNITV C LAMIPICATION

V^la?sifK4 2li «noup

a niPOHJ TITLE

EVALUATION OF THE CPO AUXILIARY PROCESSOP.CPO SPECIAL REPORT NO. 5

« DItCNIMTIve MOT« (Typ* «I npc,l tnä Induilr, dmlm»)

Special Report > AUTNOHCJJ (Lm ntm; Mnf namm. inlilml)

Edward, James P. , III Benno, Stephen A.

Creasey, Geraldine

• HtPORT DATS

• a. C JNTMACT OM «NAMT NO.

Contract AF 33(657)-14648 k pnojict NO.

Project Code 5810

A Vl/6704

7«. TOTAL MO. 9* »*«■•

63 • • ONIOmATOK1* KBPOMT h'MOINrtJ

7*. NO. or mtvm

ill—

»A gT-n««i njPomr m

9

Unclaesified Secarity ClMtiftotlon

14 XIY WONUI

Cumberland Plateavi Observatory

CPO Auxiliary Processor

On-liine Real-Time Detection and Identification

.Automatic Detection Processing

Fisher Analysis-ol-Varinnce Statistic

Wiener Power Statistic

United Kingdom Technique

Fixed Signal Threshold Level

Adaptive Threshold Detectors

LINK A HOL«

LINK •

ROL« WT LINK C

INSTRUCTIONS

1. ORIGINATING ACTIVITY: Enl* ihm nmmm and miAimtt at Ih« contractor, ■ubcootractor, crania*. Dapartmanf of D- (anaa activity or othn organliatlon (coiporUm author) laautnc Ih« raport.

2a. RCFOR i SECURITY O-ASSIFICATION Enter Iha «««* »II •»( uriiy claaaiftcalton of tba raport. Indtcata whathar "Raalrlciad Data" la Inclwtad Martin« la to ba In accord Mc« wtlh aptiwprtata aarurtlr rafulatlona.

2b. GROUP: AMMtallc tomatfmAi^g it apactflad la DoD IH- ractlva JJOO. 10 and ktmtA Fercaa Induairlal Manual. CaUr lha grMp aaaiMr Alao. «han apptteabla. ahe« that a»