APPEMDI Z
Re1rcdt.-, th h,
,,ti fthe Atroneuttral Fngirteersny Qeys.., fr,e. M. 19S ,uu.
Analog Equipment for Processing Randomly
Fluctuating Data
FRANCIS B SM1TH
L oriçey
Aerar,t,cdi Lborto,, NA.CA
Errineota!
tuthec of hLeing. u!!er, and trntsphcç,rturbulenceproduce d.ta hch ut not bIe
to nai
bthe iu procei o rneuringor '3.ating disirete valuesof rc force, pre re. or morrer.t. !ntead, 'bei rancorrit
tuctudting qudniit3e, ivcì'ed, it be,r,met nces%arv to u,r
istica1 metho-1 tt aria! u an4 decri!,r the pheoomrno',
rtng instigated The rvpe o!înformttor, uizaL
reur&
are pry,bahihlvdi ribetkris. power spectra, and cro,., pectrAIn ardu t't aod the tiror sod .zpenseinv1'ved us.ng ccii. entIoyi.iI numerica' methcid for securing he vriou, statlst&cai
a',e's.
the NACA Ix%es a Iniquie u? recordu,i d&t* on rnagl,eric tapeand pl*yio it bsktntoeectron.cs-alLganavzev which perforai the desiret! *n*iys1 andautorriaticafly pint thiui!,
Tht 'nagnetc tape -ecording ssternavidthe itaog analyur. are described in this paper and their appbczrioo to aunnauttoal
probern. t i}tu'sred A diaiuwortof the compa.rst.,ve accuracy and reiabi1ity o? the numerira! and analog me(hos is also
n-clud.
i INTROt,ucTrorc
Ip
aeronautical research and development it is possibleMeCE OP lift gXPERIMaNTAL Wi)&X undertaken iii to set up eper1rnents u that the amplitude of any quan-tity being measured remains very nearlY cotstarit dur-ing a given test run In thi, case a precise nieasure clthe static level is sought and any uctuatiofls in the level occurring during the run are considered to be
"noise which must be eliminated by fiiter or (aired throtigh on the time-history record
Howevei. many phenomena being studied produce data in which randon amplitude uctuatiors are
in-herent and measurement o average or (aired 1eve only
is of little or no &1ue Typical examples are gurt
aim-in errors encountered in fire control systems, loads irrt-posed ori aitplanes by turbulence or bufeting, streSses
produced by ergire notse or vibration, and aircraft
landing gear hads caused by rough runways Tn thtse cises, the information sought must be secured by atial-Presented at ihr Ekuunic Aids ro the Aircrafi igd,..q!r
eoo,it Tweiy TP ted Aiii.JMeeting, AS, \e 'i't:k, Januat 2 .
.erufliutI(iI 'i,trt-h Tnstrirnent Rse.rçh
i , itOu
ysis and stdu of thf' ra,nlj:iI itmpi:tude fluctu4t'Jris ahúut the neti '- alue.
Iht' ¡i',-t 'ractuc! wiv ti
esri1-e qu.antittte1vthese atiplitude u ariatt'r.s is to use- the st.tistcal
tr-ch-mi.lueS Which ha Ot,it dt'-ibei fir rilvzingstationar.
raridurn !im seriec fh' s'mp'lest and tr.ost faintli,&.
of these t.hn:qur
i thc probabihtv clistrihutt'i.which, fl)r exa:nple, has beei use-d for yeaxs to de'scrtLi
atmospheric turbulence Recently,
tht
generalize'!harrrictmc arialvss t«hniques have been developed ari are being sunecstuilv applte-d t many rand'm- tvv-data analviis priblers
The theuiretiral aspects of these techniques have beei
discussed by Wener» T.iev. and Ricr.
andspecif-applications t' aerodvnaimc prùblerns have beer
de-sctibed by Clenieritson,' Lie-pmanri,' Summers,' Press,
Chilton.' and othcrs
These authors have demnnstrated that statistical
techr.ique are extrerielv useful itt aeronautical
re-search. However, il ccnver.tional time-history records
arid ntrnencal data w&rk-up procedures are used to secure the varius reiuir,d analyses, the application o
these teduniqoes to actual samples cf experimental data s a time-consuming and expensive process. This is
es-pecially true tri th case of power and cross spectra de-terminations, ¡n addition to the digital computer
eu-pense involved tn making the thousands of required numerical calculations, several days may be required to read the necessary number of data points (rom the
tune-history r&cord. These factors often seriously
re-strict the number of statistical analyses that might be
made, or force one to limit seu-erelv the lenr..h of the
data samples and to accept the consequent poor st
tistical reliability.
To overcome these irnitations the NACA baa com-bined commercially available electronic equipment with
NACÀ-develot.4 components to prrvide as'. analog data pra-esSun Iacditv which prriduces the required
sta-tisticai anali ses i'ipidJy and ine.erisively
-cility oonshts uf toux bai.- cieITle-tltS' l) n-i
tape data storage and playback system, 1 a probahi!
analyzer, and (4) a cross pectraI drnsty analyzer. This p&per will descnbe these pieces of eqiprnent. in the order listed. aM wifl bnefly ijiuctrate the kpph cation of the varioub types of analyses to aer'riautical
TeSITh.
(Ifl
MAGNETIC Ta REcoRtG At PtA5CK
SYSTPM
An essential jart ol the data proces1ng facJ*tv is (he magnetic tape ¡ecncding eqUipTntut illustrated in Fig I
This equiprntnt provides a iiwui' for storing the data in such way that it can be accurately r produced iii a form most suitable for analog procecring.
Instead of recording data diretv on magnetic (ae a frequencvmodulated carnet svsteit ic Used. Thi permits the storage of very kw fre1unrv and d.'. and eliminates much oí thc 5U71OUS amplitude
vSìi
tier, which would otherwmr be caused by tape ux1rer-fections and ttpe wea.r It also makes possible thr di red recnrding of FM adit tClcflLrter dat.i.Where the data to be processed are p ductd b a stationary facility such as a wind tunnel, the tape rt corder and FM rnrdu!ators are located at the fa i!it Where the data are prûdticed by an airplene. hdieoptir.
ne rirkt-propdJed rnissik, the dit axe tranmstter! to a tape reeor'kt on the ground by a r1 te1eiuet.r li,ik Or. if the data to be proctssed have beenp vuuslv
re-corded as a trace de
ctit,i on íihtoí'he paper
orwt4DTUr4NE.. OP
r---
----OUMO TES! FMDATA N
:.
ROQ(LT MOOtL r FMAOIO
OP APt. ANf DATA -
fLEM( rEJi
TAPf-GUST VrLocrry
lic
I 1.01 PiOAjjJ7y .01L o i3j$Î vLLCtT'r, f pa Fm 2. PrcbbÍity distbtö,i. 1PuT f(I FEFE.iT1Aa. FfR$CE AMPLiFIER -SC APffi NG LiMfTE OUTPUTAJ1LT'
4-A/ERAGIP4G L.. CIRGUIT R EGO R DER u-rtilni, a tr:4ncrihilg :1i.i :ised jti convert t;n ,tc
If ecl ..r- to .t fre&ii'r'i. i t.Ii_4tkd rarrtcr h1.'h is Ti
u t
(rse n4 iiiinet u t i
i.
ri r r i. . ¿real rleI cf t1,hil,tv n h.niiltir: rli1. .ìnlIrks -'..q:ii'.lr-:iiii,
,Lnd frequticv r,lrlgI-r. .zoi ht ,ay.ri1 ritliv Lai tot .s bh as I t. Iv chrigiug i
t j* 'r e'
t)t
.aiiipfe- : I*
ti
5k vr'lilillutes in duiat tin r.l t 'ii taviijn fr d &. tc trø(Xi c' '.,e
(i.
I)()()\('t
scihe rtt'irrde4 and repr io the FM s' stem t riu urr t . bi.
.icd f ,
cli ulr.JTiri th (ri lt't.i
n. hfite
-trr!er aiid ici ord,n Jzre tiThe anaJ' .u;al,ui m.t}e»ds zrquirt thiit data
ph' he sa:itd j Idrgr
r: r q tiriii-.. b) the ng t ut i;utnt L,rris.-tiii:ti t-at:h sauipk rn the trt.eI 'plii t-' ! t. .1
k ifltiilUl1L'A l.sor bruire. heitig pI.cved
1)1(1.. ii: to I in FM k -ra .ìuL i 'N j n. hc ao.gfv rr lic thtte t pt-c A 4riaivjs-r thu uii whnle lata c'rded t!s tai rua'1 l)e' r.ie-s-i its- a prTsfithl1It%T b;
tribuu.ri analyst .
p'tv'r
eitras,lrsitv 3fla!v:rr.i 51lt't'r.l deii-tt' ariai ,- 'j !eSe
tltrt'i-udy-ier 4nd tOe vpe l ct.iti'ti.:aJ ul(nrull.tr,cu pr
duçt'l by tich will tic d i.i.'.el
tri th'- f0tli v:l:g .icrt utr.s 1 r his paper
III
PR iBAB!I ITY i )1STUi.'TlO1'i he terri fuictscìn shc.ri u th top of Fig 2 its-a r-its-arvimv vits-ar) trig :1uits-antzty of the (vpe to he etnideied throi..ghc.i'. this paper 'í.)re way ti 4r !vr such a function IS ti deterrrjne ts 7rÛbabihV ;. trihutrot. This :s a rueasure - the prtportion cf tita1 thr during which th mpitude of the varying quai
tity exceeds gwen Ieves.
For example, suppose the tute functionghown at t'ri
tcip it.f the f.gure were a plot cf the g'ast velocities en
couritet-d by
n airrkae
virig through rough air The intensity of the nzrhujence could be showr. bypht-ting the probabiity that thegusts wcìud exceed spei.c vekcities aganst gust veiocitv. The gl.ists iu the at mosphere represent«! by the dstnbuticr. shown would
k
ARCNA'JTICi EO EfNC, EvIfA
DATA í.cOoro
V
OI$liBi.jT Ø ,-LOOP-/
PL*YBACJArO9Ç
FE) A4ALYZR'cso AP4ALiZ(
ANALOG QJlCMNî
VO kADOM1
t1'
Q-I&.VflI P(.QT!4iAY
pTs
¿00 400 So $PtA$ t.OS?o
4.Dribtkn of .cg
,.r kii4s
beexpecte'itnexceor4ft.persec about 5pex
't if the tiens and to xcted 2C ft per aec. only about per cent of the time.
The electronic yzer used to detercic probability distrbutiois is iJl'.istrated in Fig. 3.
Th data b..np]e
revorded cn ari endie lp of rnag'netk tape is contLo uoutly played into the analyzer, and U instazitanems amplitude is conpared to a rtferttce voltage Ea, The level of the reference voltage is deternzined by the p-Etien of the poteiitmmeter slider. If the input voltage
s larger than the erence voltage. the açlifier ll
be saturated and the outpi.lt of the bmiter will be at level B, if the tnput is lesa than the reertnce vçiitaje,
theamplifier will be cnt off and the output of th
lin-iter circnit will be at level A. The eTcentagt of tune
that the leve! is at H then represeriu the p hdity f the iriptit being above the ñzed level ¿. By ave'Mirzg the output of the limiter this probabthty nay be read
directly and recorded By slowly diai ing the level
-f the reference voltage and rivirg the paper on the re-rorder in ayncbronizaton. coriplete plot cf the prob-ability dietribution rray be obtained.
To illustrate the use of the analyzer. Fig 4 shows a
distribution of witig b.iffeting cads obtained dur-ui a gi-a4uai pull-up at high speeds. The solid line
it
the analyzer record and the circled points are values deter-mined nurizericslly.Set-up and r.nnir
time on theeiectronic analyzer 'wu about 2C to 30 nin. and about 7
man-hours were -ecuired to read the tie-bistc'ry
rec-ord, calculate the mean. and dete,iine nimiericaiiy the circled points.
(IV)POER SPECTRAL t)tK5iT
The probahiity dstcibitin fezr:bes the intensity
charactenstics o the data but does not dtscribe their frequency or spectral chsrcteristics.
Nor de it
fur-nub iufortnatson ufiC:erit for calculating system input-. output relationships or transer furiottors. If this sortof intormalion is required a second tYpe nl statistical
analysis called the polcer sI.ectxal denaity anaiysi may
be ued
Power rjectxal itensity anatyss islomet,rnrs referred to as 'gtrneralired harmonic analysis" and is s3.mIar to the f aizuliar Fourier sier.es harmonic frequency analysis except in the following tespects' (l Fourier series anal
Vs
is applicable to repeutave functions while general.izd han'nv'
s.tvsis s appfü aì'Ae to stationary ran.i''i
'it
iuLcnon3. i!hn1 rated n Fig 5. the ipectn1m s Ti »re.en te-d ar a coiv in u cus -urve r ther than as d.is.rete rmsriaonkal!v-re!ated freoue-ncies, and 3;
the sçrum is pPtte-i ri terr.s of men-squared am
ttud
lier mit bandwtdm.! ouk-ale-nt toaver,1.t-pwec per unit bindwtdth n th
electrical siem; .stead of unule n-wae anplitude.
The .wr' spevtrux- then repeesents the distnbutic'n f energy ovei the frequency sp«tru.rni Fos example, the por ti-Lni of the tct.aJ energy induded io the frequency
band f to f in the ilure iS represented by the area
under the curve be ween f and
One extremely ucful feature çet the pcwer spectra
ccncent is the simple nput-outp-ut spectra rehtionshtp iUustra-ed in Fig. For linear systems, the power spsctra.l density of the system's output G,f ;se.qal to
the power spectral density of the mr.put G' times the
squared absolute value of the trausfer function Y(fi.
Thus, we ha'.e a re. ationship among the three nariablea su':h that if two of th three ame ru:wri. the thu-ct may
bedrterr,irtd.
1'his sort of Lnput-output.tnansier íurçti'n
rel4ti'r
ship is one whi'h is familiar to the electrical engineer
who is otten fortunate enough to be able to work with
strictly sinusvidal signa. lt should be note-J that the power- spectra c-';ncept now enables th aen-onauticiii
-meer. wh. often his rio choice ut to work with
ran-pso
/4
IlVWs
T\Y'3D, 6 (f) -..-- r'noÑ I' O4J'rPJT
so, '
f)I Y(f) o
.- a1(f)
i f Yff)ft
6. Lnprt oisput c-'.,a-e, spectral dresii reiaiorikuy W*N M.LJ!v*
'Cs UNIT
L
5 Powe' SZ,«tral ier.sity.
ip4tjT
INPUT
7
CONSTANT BANOWIOIN
VJNA$LE ÇLT
RECORDER
Pro. 7 Power ipoetral den,i(v &flatvzeT
domly fluctuating type nf signa}s,' to perform the same, $i*t of 'circuit analysis' as the electrical enneer
For example again consider the problem of ali air-plant flying through rough air: the airplane can be con-siderrd as a mechanism having a transfer function 1(f). the turbulence can be considered to be the input G,(f, and the att-pl.anes response can be consid. red as the
out-put G.(f).
Thus, by measuring an airplane's ransfer function and its response to atmospheric turbulence, it k pos-sible to use the plane as an instrujrtent (or measuring turbulence. as Clementson' and Surrirners' did. Or, if a specitic turbulence spectrum is assumed awl th
air-plane's tranti' function is known, it is possib!e.tn irt-dict the planes response to the turbulence.
Also, by using the techniques developed by Ricv. it is possible in many instances to use power spectra to calculate such things as the number of zero crossings and the number of times the inplítudeof a randomly vary in quantity esceeds certain le.-e!s
Power spectral '-lensity may be nu!nerleally calculated by a procedure outlined by Tukey.' Very briefly sinn-marized, this procedure is as fotlows First, rrid the time-history record point by point; second, calculate the autoc-orrelation function of the data saniple by t'vai-uattrig the mtegral
ff(t ((i +
for several d.ret.e values of time lag T and' third, de-termine the power spectral density of the data by taking the Fourier cosine transform of the autocon'elatoîi function
This process is e4uiValent to passing a tunable, c)n-stant-bandwtdth filter of known cbsracterhtics over the data and measwing the time a.'erage of the square of the alter's output. Except for squaring of the filters output and me difference in the lilt-er's characteristics. the resulta are alnost identical to those obtained from the familiar spectrum analyzers that have been used foe years by eorn,muntcation, and sound engineers
This suggests then that there are two types of dec tronic analog equipment which might be used to
de-tt'riii,nr 1sa'cr Mplctra: one type parallels the nume:-kai process 1w first determining the autocorretation functi'iii tid tht'n taking its transform to get power spectral density, the other omits the autocorrelation function entirely and measures the spectrum directly by means of a scanning electrical filter.
Since scanning filter types cf analyzers are compara-tively simple electronic devices and are commercially available, the NACA uses the direct spectrum measure-ment approach
This type of analyzer is illustrated in Fig ' and oper-ates in the following manner The data sample stored On a çontinunU Iriop of magnetic tape is applied to the bandpass filter Any frequency components in the data 'ahich tall within the filter's pass band will be passed by the ñlter. squared. averaged, and then applied to a c.liret- s-riting re--order. The filter is initially' Set at the tow end of the. freuuericy range and slowly scans ujwu-d throuh the spectrum until the entire [requencv range of interest has been covered; at the saine time the recorder purer is moving under lite atylus so that a con-tinuous iii t of power spet-tral dt':iity against Frequency is obtdirled
The scarinni sjs'etl ot the analyzer is normatIv con sers-ativetv aljust't? so that about thiee passes of the it-ata sanqile n the 1oop are made during the time re quired (or the tUter to a.uì one filter bandwidth Uiukr tht»c coniiitr',ns, the tin&e required for complete analysi, of .ì tvpn.'al record is IO to l.' min. Faster scanning speeds nav be used, hut me "smearing' 01 thi' spectrum might result
The bandwidth cf th scanning flittr 'ri the NACAs equipment cari be adjusted to values ranging froml/2 cycles per sec- to 0O cycles per sec and the true-time range of frequencies which can be analyzed is from 3
cycles per sec to l'OO() CYCICS per sec.
}y taking advantage of the pmihk changes irr tape speed preciously nit-nrtirned, it is pnrsible tu obtain equivalent filter hamivodths q4 001 cycles per sec 'ir -less and t, handle frequencies ranging f rom a fw hua-dredths of a cv.k to 50 or 60 kc. By changing tape speeds when' necessary, it is also poti.ible to handle
05 O-4 EtiCTO$iC £Id*1fZt O COP1J'TtIi POIS 0-t Q IO PtQUt5IC'Y, CRI
Pro 8 Power ipsetral disity of iriag ûirar Ios4s
AERONAJTICAL ENG'N(EiNG Q(VtEW
MAY955
SPtCTRA.. b Cpi o-z F1LT OUTPUT
W1MMM
/ i SQUARING AVERAGING C4RCUtTJTh
* SCANNING MOTORrecd lengths ranging from a few tenths of a second to several minutes in duration.
To illustrate the use of the power spectrum analyzer,
Fig. shows the speetnim of the shear kadson a
fighter-type airplane wing under buffeting conditions. The
Continuous curve was obtained from a niagnetk tape record played through the analyzes-and thecircled points
were obtained by reading the time-history record and
numerically calculating the spectrum.
The results (rom the two mcthxls differ by a
maxi-mum of about 3.0 per cent at the low-frequency end of
the spectxnsn. This difference is due, in part at Least.
to a large, undesired, very-low-frequency component
in the original data sample which hadto be attenuated
before the analysis. The electrical high-pass filter used to attenuate this component was not identical to
the equivalent numerical bighpasa filter used (os- the
same purpose; therefore. some difkietiers in the
spec-trum at the low-frequency end weretobe expected
Some difference between the analyzer and the
nu
merical values also result from the fact that the shape and bandwidth of the analyzers scanning filter were not identical to the equivalent filter resulting from the numercal process..etually, differences of 10 per cent are riot partic-ularly alarming. We are dealing with statistical proc-esses such that, even with errorless data processIng
schemes, the results obtained from repeating the sante
ir'iht easily vary
as much as plus or inirwa 2. to30 per cent. More will be said about this sub sect later. Here, to read the record and determine the spectrum using an automatic digital computer required about 10
man-hours; the time required for setting up and run-¿
ning the electronic analyzer was less than 30 min.
(V) Cioss SPECTRAL DEwsir'r
It has been shown that the power spectral denstv
furnishes information regarding the frequency content
of fluctuating quantities which is not pro'ided by the probability distribution. However, where the phe-noinenon being investigated involves study of twoor
ANALOG EOLJLPMfNT FOR RANDOMLY
lUCT'JATlNG DATA
117o
- 'ç
g4(f)_u.( Y(f }s. g0(f) i(f) u
?c 9.
Croe tra1 d.nsty.- CO-POWER - - - QUAD POWER %% 00-SPECTRUM MULTIPLIER LAVERAGIP4cJ
L*v]-4 Ir Q4JAO-SPECTRUS4I SCAHHIN6 [MuLr1PuER J MOTOR DUAL RECORDER¡11f., Io Crou sprtr1l drri%itvanaIvZeT
s
more related fluctuating quantities. which is usually thr case, some knowledge of the titile or phase correlation
between the two quantities might a1si be required
This infornmtìomi is not provided by the power spectra of the individual functions but niay be obtained from another statistical process called crtss spectra analysis
The cross spectrum betwccn two time functions ii a vector quantity and two spectra are required to fur
nish the ('t,IH1Iett' cross spectral density. These arc
illustrated in Fig. 9. The real part is called the
"co-power spectrum and indicates the product of the
in-phase frequency coiliporients iii the two functions The
unagunirv part is called the quadrature spectrum" and indicates the product nl ihr 90 out-of -phasefrequency components in the two functions. The absolute value and phase angle of the cross sprct.rurn are determined by vectorially combining the in-phase and quadrature
.spec tra.
To gain an understanding of the physical significance of cross spectra, consider two fluctuating time functions whose power spectra are represented(in Fig. 9 as PSD1 and PSI),. The cross spectrum between the two (uric tions will indicate only those frequencies inPSI)1 which
as-e also contained in PSI)1 and which bear a specific.
iionrandoni phase relationship to the frequencies in PSI)1. For example, PSD1 and PSD, may include
power in th same frequency band, but, if the
corre-sponding frequency components are entirely
independ-ent of one another, the phase between them would be random and the cross spectral density would be zero throughout the band. However, if some of the
(re-q uency components in PSI)1 bear a definite phase or time relationship to the corresponding components in
PSI),, this relationship will be indicated in the es-riss
spectrum.
¡I determination of the phaseresponse of a linear system One useful application of the cross spectrum is in the
subjected to a random-type input.
Recall that the
power spectra of the input and output made poible calculation of the absolute amplitude of a system'straris*er f unction but did not I urnith any knowledge of the phase response By u*ug cro spectra, both the
118
ECN4uTCa ENGjNfER:r'G
.t.----MA1,
95f
PSD(iI i cso: Et: L NijT SPC1J - SPtC1'N-sPr.rRu
- buaßcfIfrd
I -TAdFf , fjNC1ïQl r. 4, MPtJTvC( mc 15O o c tOV()4C'f, psTio Il. Crow .çrctra of wing he'din mouieet
s shown by Lee,' the transfer functhrn F(,f) equìls the input-output (4$ p ctruan G(ß divided by the power
spectrum of the tnput G11(f)
An interesting feature of this relationship rs the fact
that the equtior holds true even thougb other
nde-pendent random noises a.re present in the output. This is true since the input-output crois spectrutti wiE ignore
the presence of unc'orrelated random thctuations in
the output.
The -oss spectra has other useful, practical apl1ica-tions to systems with multiple inputs and outputs. but
these are too involved to diicuss here. Â good illus-tration of this usage, howe'rer, is included in Summer,
paper on tmopberic turbulence measurenients
The numerical proceis foi- determining c-ois spectrais
even more lengthy and expensive than that required
for determirnng power spectra. Howe-vet, the
numer-ical i-nuIts e be duplicated by the analog process
Wtxstr-ated in Fig. IC). The two fluctuating data sam-ples recorded on a continuous ioop of dual -channel tape are simultaneously applied tu two synchronized filters of the same type used with the power spectral density analyzer. However, instead of squanng and averaging the outputs of the filters ai would be done to determine
power spectra, the two outputs are fed into a
multi-plies whose output is averaged anti automatically plot-ted against frequency to furnish the co-power spectral density. At the same time, one of the filtri- outputs ii
run through a 90° phase shifter and again Multiplied
by the otit put of the other filter. The product is then
averaged and plotted tri give the quadrature power
spectral density.
To illustrate the use of the aois spectrum analyzer,
Fig Il shows
me data secured from a buffetingair-foil in a supernic wind tunnel.
The double-peaked spectrum is the power spectral density of the fluctuat-ing aerodynamic forces existfluctuat-ing at a par1,cu3ar soan-vue st*tion and the single-peaked spectrum is the powerspectral density of the fluctuating bending mcment existing at the root of the wing. The cross spectral
densities between these two quantities as determined
by the &naiyzer are shown in the center of the figure. The calculated a.mplitude and pLuie respolise showr. at the bottoni of the figure indicate that the wing is
sim-llar to a spring rruss system wth low damping. This
simple example does not previde arrriynarnc data il any particular value hut w
cht'n
tr demorstrare the use igl the analvzir and the type of priblern to whtchis apthcable.
T-a s.i up and run thnte fçrir i,,ectra n the.detm,nr air aFçzers required ab.L1t I
h,ur
F rtad thr tnrehistory records and run the sauìe finar a'alyseedigitaIlv
would mecire ui-i rstimndted -40 tü 4) mnn hc'ur - wiiih explains why nc mthcally calculated check points
are inwu in the figure.
('1) Accu*&cv aND RLIarLrrv CcinvarTç's
Electronic analog analyzers might introduce errors
in the ai-dei of piusor minus O per cent where a digital computer can pr(xesS data to lurost ai-tv drgr.n ¿f
pi-e-c-iiiun that might pritically be de-iired Cnnsitqunt1y. a given dst.a sariple can be statistically .analyìed by a
digital computer more precisely than
y an analog
machine.However, it mgt be remembered that we are dtairr.g
with statistical qmntiùes Consequently. if an exper
iment or test is repeated several times. sizeab]e vaa tiorus in experimental results c-ecu! een if the initna merits and ..a1yzing proceases introduce n. error a
ail. For example consider again the wing ioadspowe
spectrum, previously shown irr Fig & which is fairly
typic-al of the type of problerris encountered Tltç sta
tisuca! rnliabtlity mr this ease was such thai unlyone
out of three tests would be expected to 'ie)cl power
es-timates within plus or minus IO per cent ut the u-un
value. li the length of the dst& sair.plecriuld be
in-creased by a factor f 20. however, i out of 20, instead
of nut of 3. of the tests would be espected to vild re-suIts within plus nr minus I. per ..tnt of the true value.
In analysi of actual data samples, then, considera-tion e-f the statistical aspects of the problem usually
shows that improved reliability i-night be ubtairied, not by improving computational accuracy, but by analvz-Ing longer data samples.
Longer samples are not alwayi possible, especially
in airplane or niissume ighti, where steady flight con-ditions often cannot be main taand for longer than a few seconds. Where longer samples cari be secured,
however, the electronic analyzers can economically handle sai-ripies of stich length that analysis by any other process would b-e quite impractical.
(VII) CocLeu!No Rgaxs
In condulon, the various statistical analyses
re-quired in aeronautical research a,r4 development cs.ri be secured r*pidly a.nd economically by the use of elec-tronic analog analyzers. The accuracy of the analog
artalyurs in processing i specific data sample is sorne-what inferior to that available from digital computers.
This (actor is normally outweighed. however, by the analog equipments c4pabílity for economicaIly
han-dung longer data sample.s and providing improved sta-tistical reliability.
ANALOG EQL'IPMN T FOR
RANDOMLY FLUCTUAI
NG DA I
fr 1)9
lt appears that more
terve u
of thr
naiognaty-lu
t'hniqur might íj,dbtat
£urtha.r ad' cc.znent inthe apç,&ation of st2tstiçaI mrthod t aerodynamic
problems.
Wrie,, Noi1iett. Axrpu*. ¡*k so*, a4 SuiioMa
çf Sww T%?TE Sf4!J th &i(ta'wIL .
iioai.
TbTthaoky Pre, Me.e.cbuiew
inailatof Tuiojy.
Cam,ridi. Mus.. a.d Jon À.ky & tozia, ln* ,\pw ï'rk.
1.949.I Tukey, Jitn W, Th Smptwg Theo'y
..'f Pm'r 5p«waa
&i'.etes, Smposiwi u Apptwatici
of Atrrt1tàon
t&is te PyJ Prbkr»
S.od o
ONP.. Depart.men of (lie Navy). Waod5 Hole, M.ss pp ,'ur.e 314,
194g
'Rkc. S C.
't!
4älyiu ol'asàs !Qoi. Pt.
E
a.od U, 13e9 Syst T&d Jour , Voi XXIII, o 3, p'
-132.
July. 1944; P Jil .nd IV. Vol XXIV. i. p 4f
Jaauarv. 194.
C'eihgn1t C.. Aa ¿t,,vwa
'f thr Ps
sec-l.s,t Draii',.aèr
tSe. Z) Ths
, a4tKI 1aor, Mschuas
Iisiture of TcçhnJoy, Cssnt,rida.Mass May. l9iO.
'Lipm,,, H W., Oa ii
r4Sloa,,f i&i.tftca! C.v.r'4,f*Ji*g P'ìem, Joort.J ot ¿ht Arn4uU4* &etce,.
Vc. 19, No. 12, p I92.
'5wuntrs, R A
A Ñ&cis:eeaj Dfsc'soq çf !Afc- csk AimespErs Tiir.,,Àea,re, Rep. T (& i) Tb,eaM. ¡oit jmezoeno's Latorstúrv. Maia.huvtrs I bruto
M T-olog', C,n
tviugt Maie. May, IG4.
Prii, deny, sod koi, Johi
C, .Seiac .4frpicshoas'
Ge*ra4a,d !i'a'e',s,c tno.{yu i. C, Laadr oa
cl lbs Aa'onawtI &'envtl, Vol. , No 1, pp 173e,
6t, January, lW IClijiton, &ob«-t G.
eo.a'C1uss of
4& Tar.
hkacs QéiinF's
'-Øiraioa Vas,, Afaarda s,
w,e, 'ACA TN 3318, 1954
'Lst, Y 'e.'
.4 .aa's
of .%itziai .VIAg, la LO aa¿i,ìø Pv'ai, rh. Rti,.
'k t$i. Re,arch Lb*toey o? 5loet'onacs. M aCc.e(ts L 3gtL M T,v-IooiGE.,Setrmr
1,