· A COMPARISON OF PILOT DESCRIBING FUNCTION MEASUREMENT TECHNIQUES
by
C. E. Froste11
,6
'.
1911
•
•
A COMPARISON OF PILOT DESCRIBING FUNCTION MEASUREMENT TECHNIQUES
by
Co Eo Froste11
Manuscript received June,
19710
ACKNOWLEDGEMENT
The author wishes to express his appreciation to his supervisor, Dro Lo Do Reid, for his support and guidance throughout the projecto
The author is indebted to UTIAS for providigg the opportunity to do this work o A special thanks goes to Hart House, University of Toronto and the
Student Union of the Technical University in Helsinki, Finland for their student
SUMMARY
This work is aimed at assessing the techniques used to calculate human
pilot describing functions. The study considers data analysis methods based on~
(a) cross power spectral density of pilot input, output and error;
(b) cross power spectral density of pilot output and error;
(c) Fourier transform of pilot output and error.
Taped records of human pilot performance from previous investigations in
a compensatory control task with random input signals of continuous power spectra
were on hand and provided a pilot data base. The same data were used to exercise
each method, permitting direct comparison of the results. Data are presented as
amplitude and phase plots of measured describing functions using an average of
a reasonably large amount of data as well as single experimental runs.
A comparison of the linear. model fit parameter defined in two ways gave
1. 2.
4.
5.
6.
8.
9.
TABLE OF CONTENTS NOTATION INTRODUCTION MATHEMATICAL BACKGROUND2.1 Power Spectral Density Application
2.2 Direct Fourier Transforms LINEAR FIT PARAMETERS
3.1
P~
when Yl(jw) is used 3.2 P22 when Y2(jw) is used3.3
A Comparison of p2 Defined in Two Ways EXPERIMENTAL SET UPIDENTIFICATION OF AN ANALOO PILar RESULTS
6.1
Comparison of Variations6.2
Comparisons of MeansarHER METHODS FOR MEASURING PILar DESCRIBING FUNCTIONS
7.1
Parameter Model7.2
Orthogonal Filters7.
3
Impulsive ResponseCOMPARATIVE EXPERIMENTS IN THE :bITERATURE CONCLUSIONS REFERENCES APPENDIX TABLE FIGURES PAGE 1 1 1
3
5
5 66
8
9
9
9
10 10 10 11 11 11 1213
A(S) e(t) F(x(t» i(t) n(t) o(t) p(t) R (T) xy s T x Y(s) 2 p (w) w <P (w) xy NarATION
Transfer function of the aircraft dynamics Error signal
Fourier transform of x(t) Input signal
Pilot's rernnant Pi lot' s output
That Pilot's output due to n(t)
The cross correlation between x(t) and y(t), called auto correlation if x(t)
=
y(t)The Laplace transform variable Sampling period, sec.
The Laplace transform of x(t) The pilot describing function
1
-<Pnn(w)
Time delay, sec. Frequency, rad/sec.
The cross spectral density of x(t) and y(t) called auto power
1. INTRODUCTION
Many previous studies have investigated the human operator in a system
with the task of minimizing the system error signal. The display is called
com-pensatory if only the system error is displayed to the pilot. A block diagram of
such a system is shown in Fig. 1. The difference between the input to the system and the actual state of the system is defined to be the tracking error. If the
input signal has a random appearing nature and the task involves single axis track~ ing~ then we have a single degree of freedom, random input tracking task. By
changing the cut off frequency of the input power spectrum (Fig. 2), the difficulty
of the task can be altered. The RMS value of the input was 0.5 in. An input with
a higher cut off frequency is more difficult to track. Changing the controlled system dynamics also affects the difficulty of the task. Three different cut off frequencies and position and rate control dynamics were used in this study.
A successful approach to identifying pilot describing functions has been
the frequency response measurement from a continuous serve analysis model, although
sampled data models and optimal control theory models have also been used. Of ten
pilot describing functions have been measured using the cross power spectral den~ sity of pilot input, output and error (Ref. 1 to 5). The accuracy of the result is good. in most frequency ranges. However, the computer time required is
con-siderable.
The purpose of this study is to measure the describing functions by
employing the cross power spectral density of pilot output and error. This
method discussed in Refs. 2, 5,
6,
7,8
and9,
is well suited for use in investi=gating the pilot as a single element and requires less computations than the method of cross power spectral density of input, output and error.
In Ref. 10 the use of direc~ Fourier transforms has been suggested and
further developments are covered by Ref. 11, 12 and 13. This method has been exercised and compared to the above mentioned cross correlation methods. Use of the direct Fourier transform method allows considerable simplification in the
computation. The technique, however, when applied with a random input signal is
doubtful.
Ot her techniques for identifying the describing funétions are discussed in section
7.
Comparisons of some of these techniques found in the literatureare summarized in section
8.
2. MATHEMATICAL BACKGROUND2.1 Power Spectral Density Application
The basic system is shown in Fig. 19 where the human operator is a
non-linear element and the aircraft a linear element. The pilot can be represented by a linear system
Y(jw)
plus a remnant term n(t) as shown in Fig. 3a. Sincethis figure is a linear system with two inputs, iet) and n(t)~ the superposition
principle applies and the system may be represented by the sum of the systems
of Fig. jb.
One pqssibili ty. is to choose the describing function
Y(j
w)
to minimizethe I.\MS value of p~t), .the portion of the total signal being fed to the aircraft
that results from·the remnant. Let Y
(jw)
correspond to minimizing of the RMS. 1
<p. (w)
ln
o and Y
l (jw)<P. (w) lO
<P. le
(W)
Another possibility is to minimize the integral square value of' n(t). The re sult is derived in Ref'. 2,
Since Yl(jw) and Y
2(jw) are obtained by employing dif'ferent criteria it is not expected that they are generally identical. In Ref'.
6
the dif'f'erence betweenthese two methods is shown to be
Y (jW) = Y (jw)
-1 2
where <P (w) is the auto power spectral density of' the remnant n(t) related to
n
2n2
Y
2 (jw) (minimizing of' the RM3 value of' n( t) ).
In order to demonstrate the dif'f'erence between using Yl(jw) and Y 2(jw) consider the problem of identif'ying experimentally a system such as the human
operator shown in Fig. l.{lfor simplicity's sake assume that in actual f'act the system to be identified consists of a linear element Y (jw) whose output is
summed with the output of a random noise generator r(t), giving a system looking like Fig. 3a with Y replaced by Y
p and n(t) by r(t). It is f'urther assumed that
i(t) and r(t) are QQcorrelated.l
If the identification of' Y (jw) is carried out through the use of'
p
Yl(jw) then one obtains:
But
0
= ë
Y
(s) +r
p <P. (w) lO <P.(w) •
le <P. (w)=
Y ( jw) <P. ( w) s inc e <P. ( w)=
0 and lO p le lrY (jw)
=
Y (jw) and the identification is exact in theory,1 p
If the identification of' Y (jw) is carried out through the use of'
p
Y
2(jw) then one obtains:
-But 0e
Y
(s) +r
p <P (w) eo <P(w)
ee <P ( w) = Y (jw) <P ( w) + '$' _', (w) and eo p ee er <P (w) Y 2(jw)=
Yp(jw) + <per(w)·•
•
The identification is not perfect and the error involved depends on
the amount of correlation between e(t) and r(t). To see the implication of this
consider the following:
e = i -
0
A(s) o=
r +e
Y (s) p e i -r
A(s) -ë
AY (s) ore
p i 1+ AY (s) Pr
A(s) - l+AY (s) P <p (W)=
ee <p rr (W) 1 IA(jwj'12 <P •• (w) + --'--=~--:2~ 11 +AY (jw) 12 H 11 +AY (jw) 1 (since <P. (w) =0)
~r and <p (w) er p p -A*(jw) <p (w) (l+AYp(jW))* rrA*(jw)~rr(w)
11 + AYp(jW) 12 (l+AY (jw»* (<P .. (w)+ IA(jW) 1 2 <p (w)} p l~ rr if <P ••(w»>
<P (w) then Y 2(jw)=
Yp(jw) and ~~ rr if <P •. (w) «<P (w) then Y2(jw)
= -
~(.
) •
Thus the extent of the measure-'~~ rr JW
ment error depends on the size of <P (w) relative to <P •• (w) •
rr ~~
The model used in this example is of ten put forward as a reasonable
approximation to the human operator. This indicates that care must be taken if Y
2(jw) is used to find human operator describing functions. In the event
that another form of nonlinear system is under study, it is not possible to
indicate from the above analysis whether Yl(jw) or Y
2(jw) is more useful .
2.2 Direct Fourier Transforms
In Ref. 10 it is suggested that a direct Fourier transform can be
used to measure pilot describing functions. The cross power spectral density
method is employed by Fourier transforming the cross correlation between two signals x(t) and y(t) that are non-zero between + T.
r
T
Estimate of R (T)=
~T
x(t)y (t+T)dt xy ~ -T Estimate of F(x(t) ) Estimate of <P (w) xYl~r
T e -jwt x (t )dt-T
,2T{~T,rT
x(t)y(t+T)dt }e-jWT dT -2T -TNote that the ~ 2T limits come from the fact that -2T
<
T<
2T. However, this estimate of the cross power spectral density of x(t) and y(t) is a very bad estimate, not usually used when~the cross correlation technique is employed withrandom signals. In section 4 the estimate of @ (w) used in this experiment
and in Refs. 1, 2 aI\d 3 is outlined. . xy
In Refs. 10, 12 and 13 it is shown that F*[xJ F[yJ
2T
is identical to ~
(w
)
.
Following this the describing functions are identifiedxYl <P. (w) 10 2 from := ":"'<P -. --'-(
w-..
)
-1e 2!lEl
:=FreT
It will be shown in section 6.1 that this estimate of power spectrum leads to
large variability in the data when applied to random signals. If the proof in Refs. 10, 12 and 13, showing that <P (w) is identical to <P (w), is assumed
xYl xY2
to hold in the case of the exact formulation for power spectral density, then it is, as shown in Appendix A, possible to prove that any two signals are corre~ated unless one of them is identically zero, thus indicating that the
original function chosen (<p ) was a bad estimate of power spectral density. xYl
According to Refs. 10 to 13 the linear fit parameter
2 P (w)
I<p.
(w)I
2102
=
F*[iJ F[oJ F[iJ F*[oJ'";"'<p-
. .
--r(
w ... )r--;'<P--.(-wT) F*
U ]
F [ i ] F*[0 ]
F[0 ]
11
2 002
1.0
This erroneously indicates linearity under all circumstances.
However it is felt that direct Fourier transforms are a valid inter~
pretation under the following conditions (Ref. 14):
1. The input is a sum of sine waves
2. The transforms are evaluated using the same sine and eosine functions for
both transforms over the same run length
3. The transforms are evaluated only at the input frequencies.
Beyond these restrietions the use of the direct transforms is questionableo
•
30 LINEAR FIT PARAMETERS
Since the controlled vehicle is a linear system, with the transfer
func-tion A(jw), we define a linear fit parameter p(w) to measure the linearity of the
hvman operator aloneo If the pilot behaves in a nearly linear fashion, then
p(w) will have values close to unity and the remnant will be small, while low .
values of p(w) indicate more nonlinear performance and the corresponding remnant
will be large.
2
301 PI when Y
l (jw) Is Used
In Fig. 3a nr(t) is the remnant and is uncorrelated with the input i(t)o
Define ~ nlnl (w)
1
-~
(w)
00where ~ (w) is the auto power spectral density of the remnant related to the
n
ln1
minimizing of the RMS value of p(t). From Fig. 3a
u
=
ö
-
ë
Y (s)1 1
~
(w)=
~
(w)+
IYl(jw)r2~
(w)-Yl(jw)*<I:> (w)-y1(jw)
~
(w) nln l 00 ee eo oe Since ~ eo (w)=
~ oe (w)* ~ (w) = ~ (w) + nlnl 00 IYl('jw)12~
ee (w) _ Y (jw)* 1~
oe (w)* - Y 1 (jw)~
oe (w) 1 From Fig. 3aö
= Y l (s)(î
-
A(s) 0) + n-
1 -o Yl(s)I
+uI
1 + A(S)Y l (s)When i(t) and nl(t) are assumed uncorrelated then~. (w)
~nl
o
~ (w) 00 Y l (jw) Il +AYl (jw) 2 2~ii(w)
+ Il+Ail(jw)Y l (jw) 2 2 <P (w) <P •• (w) +
(I
1I
-1) <P (w) 2 nlnl 1 +AYl ~ jw} I I l+AY1~jW} nlnl P l (w) = 1 - <P (w) Y l (jw) 2 2 00 1 +AY l (jw}I
<P • • (w) +I
1I
<P (w) I I l+AY1(jW) nlnl 2 It <p (w) «<p .• (w) then Pl ~ 1.0 and the linear model is a perfect fit.
nlnl I I
If <P (w) »<p .• (w) then
P~ ~
1 - /1 + AY1(jW)/2 and does not go to zero
nlnl I I
as one might anticipate.
3.2
P~
when Y2(jw) Is UsedFrom Fig. 3a we obtain,
Ö
= ll2 + Y2(s)
ë
<P (w)
=<P
(w)+
/Y2(jw)/2<p (w) since it was found when Y2(jw)
00 n2n2 ee
<p (w)
eo
<P
(w)
ee
(minimizing the RMS value of n(t)
with e(t), i.e., <P
(w)
=
o.
) (Ref. 2) that n
2(t) has zero correlation
en2 <P (w) 2 n2n2 Again de fine P
2
(w)=
1 <P(w)
00IY
2(jw) 1 2 <p (w) ee 1.0If <P n n (w) »<P (w) then P22
~
0 and thus behaves as expected. 2 2 ee3.3 A Comparison of p2 Defined in Two Ways
<pnn(w)
From Fig. 3b we obtain p
=
n -
p A(s) Y(s),11 + AY(jw) 12 <P Çw). If in practice pp <P. (w) 10 Yl = <P.
(w)
l e is quite close to (1 + A(s)Y(s)p -n=
•
-
.
cP
(w) eocP
(w)
ee then Yl
=
Y2=
Y, nl=
n2=
n and ~l=
P2=
p. (This will be shown true forthe present data in section
6).
cP
(w)Define p2(w)
=
1 -~
• . Then in the case where Yl
=
Y2=
Y 00 2 2 2 2 cPnn(w)-cP(w)
[
2 ] cP(w)
p (w) -Pl(w)=
P(w)
-P2 (w)=
cPoo(W)pp=
Il+AY(jw) I -1cP!~(w)
=
~
[ll+AY(jW) 12~l
] (1_P2(W) ) 2 IcP. (w) 12 also P (w) = cP.~(w)cP
(w) 1.1. 00 (Ref. 1) 2In the past P ha1 usually been used when employing Yl(jw)
In this case it can be shown by using the equations of section 3.1 that
2 P
(w)
=
IY1(jw) 12 cP ..(w)
1.1. 2 If cP(w)«
cP .. (w) then P ..., 1.0 ninl l.l.If cP (w) »cP .. (w) then p'2""O and thus behaves as expected.
n l n1 1.1. cP.
(w)
1.0cP.
(w)
•
1.eNow
P~
orP~
can be found from p2(w) - {ll+AY(jW) 12_1} {1_p2(W)} =p~(w)
2 2 2 2
Figure
4
is a sample comparison of single run Pl ' P2 and P3 (based on Pl as
defined in section 3.1, and 2
2
I
cP eo (w)1 )P2 cP (w)cP
(w)
•
ee 00
The fit is good except for the first few frequency points. In Ref. 1 it was
shown that the experimental accuracy in calculating p2 for the low frequency
points is poor. In this experiment P~ is calculated with p2 as a base
( P3(w) 2 = P 2 (w) + ( 1 l+AY(Jw) . 12 -1) (l-p 2 (w)) ).
Since for the low frequency points, especially in a rate control task, the term
11 + AY(jw) 12_1 is large and is multiplied by a very small n~er (1_P2 ), a small
a measure of the linear model fit when y
=
1 <Pio(w) <P. (w) ~e is employed. As shown in Fig. 4 the equivalent pi is more accurate frequency points than P~. Therefore it is suggested in future studies based on<P.
(w)
~o <P.(w)
~e if<P (w)«<p .. (w). nlnl ~~at the first few low 2
that P
l should be used
2 2
The model fit parameters P
l and P2 are based on the remnant net). A
t · . 2 d 2 .
small remnan w~ll g~ve P
l an P2 close to unity, thus ind~cating close to linear behaviour. In a physical sense this is an excellent measure of the linearity of the pilot alone. In Figs. 15 and 19 p2~ is shown.
2
If we are interested in the system from a control system engineering point of view then we are more interested in the signal going into the aircraft and especially the portion pet) due to the remnant. p2 (based on pet) ) is a good measure of linearity in this case.
4. EXPERIMENTAL SET UP
Data from earlier experiments at UTIAS (Ref. 1) were used. The
equipment used to provide this data has been described in detail in Ref. 1. The main parameters will be repeated here.
The facility used consisted of a modified CF-100 fixed-base flight simulator cockpit coupled with an EAI TR-48 analogue computer. The signals iet), o(t) ~d eet) were recorded in digital form af ter passing through an EECO ZA37050 analogue-to-digital converter. The sampling rate was 20 samples per second. Of the 190 seconds long experimental run, 180 seconds (T) were recorded and used. The maximum length of lag in terms of samples (NLAGS) was 200 giving T
=
9.95 seconds in the correlation functions. The general procedure was to find ~e auto or cross correlation R (T), then multiplying this by axy
particular function
aCT)
before estimating <P(w)
by Fourier transforming R (T).xy xy
This leads to more acceptable spectral window shapes. Here the "Ranning window" was used
(a(T)
= ~ (1 + cos(m/T ) ) . (For further details see Ref. 1).m
Only compensatory data were analyzed here. The aircraft dynamics were a) position control (0.114 in/deg) and b) rate control (0.338/s in/deg/sec as measured from joystick input by the pilot to display motion). Both cases (K and lis) were further divided into three parts depending upon the cut off frequency of the random input signal. Low (2 rad/sec), medium (4 rad/sec) and high (6 rad/sec) cut off frequencies were used. Figure 2 shows the spectral shape of these input signals (L, Mand R).
Initially each of these six conditions consisted of six experimental runs byeach of six subjects, i.e.,
36
runs per condition. A few runs had to be skipped due to bad recording. Table 1 shows the number of runs to calculatethe means and standard deviations for each condition.
The spectral calculations were done on an IBM 7094 computer . The describing function Y (jw) and corresponding p2(w) for one run require~ 57.9 seconds. Some simplitications could be done to calculate Y
2(jw) and P2(w) •
The time used to analyze one run was 39.5 seconds allowing a time saving of
32%. In this experiment anormal Fourier transform was used to calculate Y
3
(jW
).
At the end of the experiment a trial with a fast Fourier transform was performed requiring only 5.9 seconds to analyze one run. The time saving was 90% over the Yl(jw) calculation. Note that no p2 calculation is performed in the case ofY~(jW). All calculations used the same time records and found the pilot des
cri-b1ng function at 25 frequency points.
5. IDENTIFICATION OF AN ANALOO PILOT
An analog pilot, Y(s)
=
87.5/(s + 3) deg/in, performed experimentalruns utilizing the normal experimental set up and input signal levels. The
result, Fig. 5, gives us an insight into the accuracy with which the digital
programs (Fourier trans~orms ~d cross power spectral density of output and
error) can identify pilot describing functions. The top of the triangal symbol
locates the data positions. As in Ref. 1 where the cross power spectral density
of input~ output and error was used, no problems were encountered with the po
si-tion control task. In the rate control tasks the performance was excellent except
for the first frequency point. This problem is due to low power levels for the
signals o(t) and e(t) as described in Ref. 1.
6. RESULTS
Figures 6 to 11 show the amplitude and phase plots of the describing
function Y
3(j
w)
,
i.e., data analyzed by Fourier transforms of the output signaland error signal as described in section 2.2. In the figures K stands for
posi-tion control and lis for rate control. In the amplitude plots the left corner
a:nd. in the phase plots the right corner of the triangle symbol indicates the mean and the bars show plus and minus one standard deviation. Figures 12 to 14 and
16 to 18 show the amplitude and phase plots of the describing function when the
analyzing process is base~ on cross power spectral density of output and error.
The model fit parameter P2 (calculated from I~
(w)
12/~ (w)~(w) )
plus andeo ee 00
minus one standard deviation is shown in Figs. 15 and 19, where the right corner
of the triangle symbol indicates the mean. The describing function Yl(jw) ,
based on cross power spectral density of input, output and error has been plotted
(~ one 5) in Ref. 1 for the same experimental data (see Figs. 36 and 37).
6.1 COmparison of Variation
As shown in Figs. 12 to 14 and 16 to 18 the variation in the describing function when cross power spectral density of output and error is employeè, iF small and smooth and very close to the result in Ref. 1 (Figs. 36 and. 37), where cross power spectral density of input, output and error was used. Wh en Four~er
transforms are used, Figs. 6 to 11, the variation is large and rough.
The cross power spectral density methods use data from a short experi -mental run and attempt to predict the pilot describing function that would be found for an infinitely long run. The variation from one experimental r~~ to
/
another is expected to be small and smooth. On the other hand if Y
3(jW) and e(t)
is given for a particular run, then o(t) can be calculated exactly from Y (jw) = F(o)/F(e) for that particular
3
min. run, since the Fourier transformtec~n
i
que
calculates the describing function that fits the short experimental interval exactly and thus the variation from run to run is large (when random inputs a
1
eused)? although the means of a large number of experimental runs can be expected to approach the desired describing function.
6.2 Comparison of Means
The means of Yl(jw), Y
2(jw), Yj(jw) and corresponding model fit
para-meters are plotted in Figs. 20 to 27. The cross spectral density methods
(Yl(j w) and Y
2(jw) ) are very close to each other over the whole frequency range for all six conditions. Yl(jw) and the corresponding
P~
(calculated asp~)
are plotted as a line. Y2(jw) and the corresponding
p;
are represented by a trianglesymbol, where the top is the data position. Y
3(j w) is plotted as a plus sign. The means of p2 as measured in Ref. 1 are represented by a cross.
The Fourier transforms technique (Y
3(jw) ) gives a good approximation of the pilot describing function if a large number of runs are averaged although the large variance in the data reduces its usefulness.
The above roentioned results are verified in Figs.
28
to 35? where thepilot describing functions based on one experimental run per condition by the same typical subject, have been plotted.
The linear fit parameters
p~
andp;
show a pilot behaviour quite closeto linear. The plots are quite flat and close to unity. This indicates a fairly constant remnant n(t). Since p2 in Ref. 1 drops off in the middle of the fre -quency range? p(t) is built up in the closed l09P system.
7.
OTHER METHODS FOR MEASURING PILOT DESCRIBING FUNCTIONSTechniques for identifying de~cribing functions other than those used
in this experiment will briefly be described in this section. In section
8
comparisons found in the literature are discussed.
7.
1
Parameter ModelThe parameter model method assumes a particular describing function model for the pilot dynamics and then solves for the parameters in that model. With proper programming the parameters can be made to converge to values which minimize the differeryce between system and model outputs. Although the stability
and speed of convergence of such parameter trackers is of concern? the technique
has the adv~ntages of being physically easy and inexpensive to implement (requir=
ing onlyan analog computer). The method is restricted in that only a limited
set of systems, which have the specified form, can be adequately identified.
The model used in Ref.
15
has the forma
3
s + a4
-Àswhere the time shift À accounts for any pure time delay in Y(s). Estimates of
the parameters al~ a
2, a3, and a4 were determined by a quasilinearization
tech-nique described in Ref. 15.
In Ref. 16 two different model forms were used. The three parameter
model was
Y(s)
and the two parameter model was Y(s) A(s)
=
crossover model as developed in Ref.
17.
7.2
Orthogonal FiltersK e
s -TS
This model is called the
The orthogonal filter method (Ref.
18)
is somewhat more general thanthe parameter model. It assumes that the unknown system dynamics can be modelled
by a series of transfer functions of the form (Ref. 15)
-Às { b
Y(s)
=
e 1 1 +T lS +
Estimates of the parameters b
l,b2,b
3
, •••
etc., can be determined by amulti-regression technique (Ref.
7).
7.3
Impulse ResponseThe impulse response method (Refs.
'
7, 9,
12 and 19) assumes a verygeneral input-output relationship that can be represented by the form (Ref. 15)
T
-Às
r
m -TSY(s)
=
e ~ g(T)e dTo
where g(T) is an impulse response function that is assumed to be zero for T
<
0and also zero for T
>
T • The calculation of the impulse response function atdiscrete times, g(o), g(Nt), g(26t), etc., is shown in Ref. 15.
8.
COMPARATIVE EXPERIMENTS IN THE LITERATUREA few comparisons of techniques for measuring pilot describing func
-tions can be found in the literature. The techniques of Fourier transforms,
parameter models, orthogonal filters and cross power spectral density are co
v-ered in Ref. 16. However, the experiment compared techniques on single runs
only and thus there is no variation in the data shown. Furthermore the input
was in all cases of sum of sinusoids. These methods provided good measurements
in the region of system crossover frequency. This is the frequency where the
product of the absolute values of the pilot dynamics and of the aircraft dynamics
passes from greater than unity to less than unity. The more computationally expen
deteriorated when signal levels were low. Controlled dynamics of the form lis
or 1/s2 and the pilot's ability to control either very wellor poorly reduced
signal levels over certain frequency ranges outside the region of crossover
frequency.
In Fig. 38 (from Ref.16) the Fourier transform method is compared to
the two and three parameters model for an analog pilot (known system). The
Fourier transform technique is good (since the input is a sum of sinusoids) but
the parameter models are accurate only in the region of crossover.
In Fig. 39 (from Ref. 16) the Fourier transform, two and three para-meters models, orthogonal filters and cross correlation are compared for a human
pilot. Since this comparison is based on a single run and no variation in the
describing function based on different techniques can be shown, it is difficult
to draw any conclusions.
In Ref. 15 three different identification methods (the parameter model,
orthogonal filters, and impulse-response techniques) were applied to the
identi-fication of both simulated (i.e., known) systems and piloted systems. According
to Ref. 15 the three methods were shown to estimate adequately the pilot descri
-bing functions. However, the input signals were a sum of sinusoids. No
varia-tion of the describing funcvaria-tions could be shown since the experiment consisted
of a single run per condition.
In Ref. 17 the two parameter model (the crossover model) is compared
to the cross spectral density of input, output and error. Figure 40 shows a
typical pilot comparison from Ref. 17. The fit is good as long as the linput
power in the region of crossover is high 'enough to allow accurate par~eter
tracking. In Fig. 40 the input signal had a power spectrum with a cut off
fre-quency of
4
rad/sec(M).
A cut off frequency of 2 rad/sec(L)
was also used inthis experiment. No comparison of variations of the describing functions was
done.
9. C ONC LUS I ONS
1. The overall agreement between the pilot describing function measured by
cross power spectral density of input, output and error and by cross
power spectral density of output and error is very good. This second
technique can successfully be used to measure the pilot describing
func-tion and the linear model fit parameter, when the nonlinear and noise
components are small. This technique allows a 32% saving in computer
time.
.
2. If a large amount of data is available, the Fourier transforms method
gives a good approximation to the mean, although the large variance
reduces its usefulness as an experimental technique (when random inputs
are employed) relative to other approaches. The computer time saving was
90% as compared to the cross spectral density method of input, output
and error.
3. As a measure of the'lin
2
arity of the human pilot, the linear model fitpara~eters
PI(w)
andP
2
(w)
based on the reID?ant n(t), are preferableto P , provided that ~ . .
(w)
»~(w)
and ~(w»>
~(w).
I I nln
1. Reid, L. D. 2. Reid, L. D. 3. Gordon-Smith, M. 4. McRuer, D. T. Krendel, E. S. 5. McRuer, D. T. Graham, D. Krendel, E. S. Reisener, W. Jr., 6. Elkind, J. I . 7. Elkind, J. I. 8. Wingrove, R. C. Edwards, F. G. 9. Wingrove, R. C. Edwards, F. G. 10. Taylor, L. W. Jr., 11. Smith, H. J. 12. Taylor, L. W. Jr., REFERENCES
IIThe Measurement of Human Pilot Dynamics in a
Pursuit Plus Disturbance Tracking Task". UTIAS
Rept. NO.138, University of Toronto, April 1969.
IIThe Design of a Facility for the Measurement of
Human Pilot Dynamicsll
• UTIAS Tech.Note No.95,
University of Toronto, June
1965.
IIAn Investigation into Certain Aspects of the Des
-cribing Function of a Human Operator Controlling a
System of One Degree of Freedomll
• UTIAS Rept. No.
149, University of Toronto, Feb, 1970. IIDynamic Response of Human Operatorsll
• WADC TR
56-524, October 1957.
IIHuman Pilot Dynamics in Compensatory Systems 11 •
AFFDL-TR-65-15, July 1965.
IIA Comparison Between Open and Closed-Loop
Measure-ments of Dynamies Systemsll
• Bolt, Beranek and
Newman Inc. Memorandum Rept. 8224-4, Ma.rch 1963.
IIFurther Studies of Multiple Regression Analysis
of Human Pilot Dynamic Response. A Comparison of
Analysis Techniques and Evaluatiorrof Time-Varying
Measurements 11 • ASD-TDR-63-618, March 1964. IIMeasurement of Pilot Describing Functions From
Flight Test Data wi th an Example from Gemini Xli. Fourth Annual NASA-University Conference on Manual
Control. NASA SP-192, March 1968, pp.119-134.
IIA Technique for Identifying Pilot Describing
Functions from Closed-Loop Operating Records". NASA TN D-6235, March 1971.
IIDiscussion of Spectral Human-Response Analysis!l.
Second Annual NASA - University Conference on
Manual Control. NASA SP-l~8, March 1966, pp.403-412. IIHuman Describing Functions Measured in Flight and
on Simulatorsll
• Second Annual NASA - University
Conference on Manua.l Control. NASA SP-128, March
1966, pp.279-290.
!IA Comparison of Human Response Modelling in the
Time and Frequency Domains". Third Annual NASA
-University Conference on Manual Control. NASA
13. Taylor, L. W. Jr., 14. Young, L. R. Windblade, R. 15. Wingrove, R. C. 16. Shirley, R. S.
17.
Jackson, G. A. 18. Elkind, J. I. Starr, E. A. Green, D. M. Darley, D. L. 19. Goodman, T. P."Relationships Between Fourier and Spectral
Analysis" • Third Annual NASA - Uni versi ty Conference
on Manual Control. NASA SP-144, March 1967, pp.183-186.
"Summary". Second Annual NASA - University
Con-ference on Manual Control. NASA SP-128, March 1966,
PPD 1-11.
"Comparison of Methods for Identifying Pilot Descri
-bing Functions from Closed-Loop Operating Records".
NASA TN D-6235, March 1971.
!IA Comparison of Techniques for Measuring Human
Operator Frequency Response". Sixth Annual NASA
-University Conference on Manual Control. AFIT,
AFFDL, 1970, pp.803-869.
"Measuring Human Performance with a Parameter Tracking
Version of the Crossover Model". NASA CR-910,
October 1967.
"Evaluation of a Technique for Determining Time
-Invariant and Time-Variant Dynamic Characteristics
of Human Pilots". NASA TN D-1897, May 1963.
"Determination of System Characteristics from Normal
APPENDIX A
Consider two independent stationary random signals x(t) and y(t) with
amplitude probability funetions that are symmetrie about zero.
Pl(x) is symmetrie ab out x
=
0P
2(y) is symmetrie ab out y
=
0(00
P l (x) )C. dx=
0 ~_00
andJ
oo P 2(y) y dy=
0-00
Now lim1 fT
2T x(t) y(t+T) dt -T T-7OO=
J
00JOOp
1 (x) P 2 ( y) xy dxdy -00 -00 { sinee~(x/y)
=
Pl(x)}beeause x(t) and y(t) are assumed independent.
o
And 1r
oo -jWT cp(w)
R (T) e dT xy 21T xy -00 0 R (T) == 0 xyHowever this does not require cp
(w)
or cp(w)
=
0xx yy
Now eonsider the identieal si tuation but apply the
cp
(w)
xylim
~T
F*
(x)F(y)
for cp
(w)
=
0 for allw
xy
requires
~:moo ~T F~(K)
F (y)=
0T
or that either
F*
(x) orF
(y)=
ATET T
assumption that
where A is a constant and E
<
1/2 as T-7OOwhieh means that either
cp
(w)
or cp(w)
= 0 for allw .
xx yy
Aeeording to this it is impossible to have two eompletely uneorrelated signals
TABLE 1 TASK K lis FREQ. TYPE L M N L M N S 1
6 6 6
6 6 6
S 26 6 6
5 6 6
S3
6 6 5
5 6 6
S4
6 6 6
4
6 6
S5
6 6 6
6 6 6
S6
5 6
4
5 6 6
TarAL35 36 33
31 36 36
I
I
I
- - i I
System
Outl63' I Error Observational, I Dynamics of Operator Dynamics of
I
or State
~ I Si~al Sensing, and I Limb/Manip- Output Contro11ed
-c I
..
I Visual Computational I ulator Vehic1e orI
Input E1ements I Combination System
I I
I
put
-
--~HUMAN OPERATOR
1.0~
__
~~____
~~y---~--__
-
o-
• .-1 '.-1 .et-
-
3-• .-1 • .-1
te.
L M HO.Ol~---~---*----~~~
__
~
10.0
Figure 2
w
radiansI
sec. Input Power Spectran(t)
i(t) e(t) o(t) m(t)
Y(s)
.
A(s)-Figure 3 aServo system for the compensatory task.
i(t) o'(t) Y(s) a A(s)
.
.
-+
n(t) Y(s)..
p(t)-
A(s)o·s
x
0·0
2
x
RHO**2 (1)
6.RHO**2 (2)
RHO**2(3)
4
5
8
10
12
14
RAD/SEC
Figure 4 Comparison of the linear model fit parameters, one run, typical subject, M l/s task.
30·0
~---AC-T-UA-L-Y-=8-7-.S-I-(-5+-3-)-DE-G-I-IN--~
6
PHI (EO) IPHI (EE)
z
+
F (0)IF
(E) ~20· 0
+ - - - " < : - - - 1 ~~
M 115 TASK
w
g
~10.0
+---i!!!~---I ---.J 0.... ~ -<: ~o·
0
+---+---+---+-+---+---+-~-+---+--+-~-~-+--+---+----+2
4
5
8
10
12
14
RAD/SEC
15
o·
0
+---+---+---+-+---+---+-~-+---+--+-~-~-+--+---+----+~
-50· 0
+---~---1
-100.0
+ -_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ -...JFigure 5 Analog pilot, M l/s task.
a
=
Z
p
100·0
--~
F(O)/F([)
+
ANO -
1
STANOARO DEVIATIDN
..--.
z
t--t!
~
w
0~ ~
~
::J10·0
~ ~ 'lr-~}~ ~}}r ~ q~} ~r!
~
}
t--t ~ -.J~
~
<-1·0
2
4
5
8
10
12
14
15
RAD/SEC
0·0
l-50·0
It
,... ~~
-100·0
t
t-~
[ L-150.0
-200DO
t ~t
t
~ ~C
Itt
~t
c
~ ~~ t
t
t t t
D
v-250·0
Figure 6 L K tas k.W III
100·0
10-0
1-0
0-0
-50·0
:l§
~-150-0
-200·0
-250-0
ID
~
F(OJ/F(EJ
+
AND - 1 STANDARD DEVIATION
!}
2
4
5
8
10
12
14
15
RAD/SEC
~L
~
t
D
t
t
~
f'D
Iit
D
~t
t
~t
D
~
C
t
11 L-Figure 7 M K task.100-0
10·0
1-0
0·0
-50·0
(.!J~
-100·0
W III <:::::r
[ L-150.0
-200·0
-250·0
t
I~t
....F(OJ/F(EJ
+
AND - 1 STANDARD DEVIATION
2
4
5
8
10
12
14
16
RAD/SEC
L
-t-t-e
D
D D
Ii t,. ~D D D
D D D D
C
tD
j) Figure 8 H K task.100·0
.--.z:
f--1 "-~~
~
10·0
:::J I -f--1~
<.1·0
0·0
-50·0
~ ~~
-100·0
bi
~
CL-150-0
-200-0
-250-0
F(O)/F(EJ
+
AND - 1 5TANOARD DEVIATION
~
~
~~~q~l
~
~
~ ~ ~
q q
~ ~
1
f-2
4
5
8
10
12
14
15
RAD/SEC
I~ ~
t t
t
~
t t
C ~t
~
t
'" t
t
Ct
t
t
l\ vt
tt
t Figure 9 .L 1/ s tas k.100·0
.--..:z
I-t "-~w
0w
010-0
::::J .-I-t .-.-J 0... ~ <. 11·0
2
4
5
8
10
12
14
15
RAD/SEC
0-0
'-t
t
~ ~-50-0
It
t L-.--.. ~~
-100-0
~ ~
{t
D
~D ..
"
tQ
ëE
-150-0
D D D
C
t
DC C
t
D
-200-0
-250-0
Figure 10 M l/s task.100·0
F(OJ/F(EJ
+
ANO - 1 STANDARD DEVIATION
" "
q
d
~
q
~ ~
r
q
r
!
f
ê
10 -0
+---t:~~~ -i1-~-t'l~-H-t+-+--+-~=-...t---+-f----!--+-~--=--t-+--+-+--l
... .-.J r·~
<.1-0
2
4
6
8
10
12
14
16
RAD/SEC
0·0
-50-0
t
t
t
{t
t
t
ID
" t t
...C3
o
-100-0
w
~
CL-150.0
~t
t
tt
Dt
t t
.
t
t
t
tb
-200·0
-250·0
Figure 11 H 1/ a taak.l00~O ...-.
z
...
,~E3
~
10~O] ]
F1-!HEOJ
~!(EEl
+
M(]
-
1 Si
AN:W\U
[(\I
IA TIeN
~~
t--t~
.
. . .
. .
• ,. . .
. .
.
.
.
2
4
6
8
10
12
14
16
~
I I L L I I I I.
-50·0
Ij
Ii
-§
-100·0
-
~~
+[~~~~
I ,-200·0
~ ~ ~ ~ ~ ~ ~
-~
t
~ ~
j ~ ~ ~.
Figure 12 L K task.100·0
f f
PHHEDJ IA-lI
(EE!
+
At{]
-
1 STAt{]AR[] lIVIA TIo-J
]~1l+.
I1·0
2
4
6
8
10
12
14
16
RAOIS1.1:
O· 0
+--+---f--+--+---+--+----+--+--+--+----t~__+___+__+__+-~·o~~---~
...
c..s
~
-
100·0
+----~-,---____l
:
~~~
~
··150·0
+---+--1't--t+-:T--+--=-~----__I
{{{i
-200·0
+---~_±__+_~~__I-250,.0
+--_ _ _ _ _ _ _ _ _ _ _ _
- - - - . 1 Figure 13 M K task.100·0
...
~
PHI (EO)
IR·n
(E[)
+
AND .
-
1 STAtaRD DEVIATICN
~
~
l~~
I
W.O~~~-~~-~~-~-~-~-~~-~-~-~-~-~~-l-~-~~~
1-0
2
4
5
8
10
12
14
16
apo
~RAD/SEC
-so·o
[
.-....~~~
§
...
-100·0
1{
~-l5l.0
[{{{1[1
-200·0
-250·0
+ - - - '
Figure 14 H K task.1KO
. . f - - - ,
.~ ~ ~ ~ ~ ~ ~ ~ ~ ~
[
~ ~ ~
1
~
{ {
~ ~
{ { {
1
~
RHOM*2
+
ANO
-
1 STANOARD DEVIATION
~
PHI (EO) /PHI (EU
2
4
6
8
10
12
14
16
RAD/SEC
1-0
+ - - - . . , - - ,
{t~itt~i~t~tttttt{{{{{1
I
0·5 {
~*.2+
ANJ -
1
ST~lIVIATI(}J
A·U (EO)/A-U (EE)
o
-0
+_____l~__f____4!----+--+---l :----+--+---+---+-. __f___+__~+__+2
4
6
8
10
12
14
16
0~5
RHO**2
+
AND - 1 5TANDARD DEVIATI(}J
PHI (ED)/PHI (E[)
~
o
~0
+--I---+__+__+_-+--~I---+. -+---l~I---+__+__+__+.__+_-'--+
2
4
6
8
10
12
14
15
RADI~C
100-0
....-.
z:
1-1
"-
PHI(EOJ/PHI(EEJ
+
ANO - 1 STANOARD DEVIATION
~
w
0w
010-0
:::J I--1-1!
~
-<
1-0
2
4
5
8
10
12
14
15
0-0
RAD/SEC
--50-0
{~~tt+
~~ii~[{
C
tse
ËE
-150.0
v ~t
C~
t
~t
11t
t
D
-200-0
-250·0
Figure 16 L l/s task.100·0
1·0
0·0
-50·0
w
U1 <.::r:
EL-150.0
-200·0
-250·0
PHI(EOJ/PHI(EEJ
+
AND - 1 STANDARD DEVIATION
2
4
5
8
10
12
14
15
RAD/SEC
-{(j:~~
~
ct
-L~~
~~~ct~~
+'
c
~{{C
t t
~t
t Figure 17 M 1/ s task.100·0
PHI (EOJ/PHI (EEJ
+
AND - 1 STANDARD DEVIATION
1·0
2
4
5
8
10
12
14
15
0·0
-50·0
,... ~~
-100·0
...se
cr:
-150.0
RAD/SEC
i.+
I+~q:~~'
.
c
i~J
~ ~ ~
{
{
{~ ~
C
D C
"t
D D
D C
v-200·0
-250·0
Figure 18 H 1/ s task.RHO**2
+
AND - 1 STANDARD DEVIATION
PHI (EDJ /PHI (EE)
O· 0
+--+--+--+--+--+--t--t--t---1t----it----it----i--t--t--t--+0·5
2
4
5
8
10
12
14
15
RAD/SEC
~
RHO**2
+
AND - 1 STANDARD DEVIATION
PHI (EDJ /PHI (EE)
O· 0
+--+--+--+--+--+---t--t--~t----it----it----i--t--t--t--+0·5
2
4
5
8
10
12
14
15
RAD/SEC
[ RHO··2
+
AND
-
1 5TANDARD DEVIATIDN
PHI (EO) /PHI (EE)
0·0
+--+--+--+--+--+---+---t--t--I--t----il---t--t--t----+---+2
4
5
8
10
12
14
15
RAD/SEC
100·0
1·0
PHI (10) /PHI (IE)
6PHI (EO) /PHI (EE)
+
F
(0)
/F
(E)
+
2
4
5
8
10
12
14
15
RAD/SEC
0·0
+-~~--+-~~--+--r~--+--r~--+-~~--~4-50·0
~~
-100·0
w
$
~
~t
:r:: +4-~-150·0
+---~~~~--~~---~-200·0
+---~~.---~-
250.0
+-_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ ----.JZ. f-1
"-CJ Wo
100-0
1·0
0·0
PHI (IDJ IPHI (IE)
6.
PHI (EDJ IPHI (E[)
+
F(O)/F(EJ
2
4
5
+
+
8
10
12
14
15
RAD/SEC
-50-0
+ " < -CJ~
-100
-0
+ - - - ' > 0 . . , ; , - - - 16-+
~w
+
SQ
$~+
~
-150-0
+ -______________6-~~~~---__4
-200-0
+---~~~__1-250-0
100·0
w
g
10·0
l -r-! .-J~
-<
L'J1·0
0-0
-
50-0
E::i
-100-0
·
-
200-0
-250·0
PHI (10) IPHI (IE)
6PHI (EO) IPHI (EE)
+
F
(0)IF
(E)
2
4
5
8
10
12
14
RAD/SEC
Figure 22 Comparison of mean values, H K task.
15
n;
:l1li...
ZP
o o1-0
~;r~x---~
0·5
X 6~66666 X Xx x x x x x x x x
x x x x x
- RH]**2 OF PHI (IO) /PHI (IE)
6
RHO**2 OF PHI (EO) /PHI (EE)
x
REF-
1-x
xxx
0-0
+-~-+~--~~-r~--r-+-~-+~r-+-~-+~2
4
6
8
10
12
14
16
RAD/SEC
1-0
~;-~~---~
xXxxxxxx
x x x
0-5
x x x
x x
- RHO**2 OF PHI (IO) /PHI (IE)
6
RHO**2 OF PHI (EO) /PHI (EE)
0-0
x
REF.
1-2
4
6
8
10
12
14
16
RAD/SEC
1·0
x
XX>.<
6 6 6 6X
X X
X
0-5
x x x x
x x
x
x x
x x
-
RHJ**2
Cf
PHI (10) /R-U (IE)
6
r;H]**2 OF PHI (EO) /PHI (EE)
x
REF· 1·
0·0
2
4
6
8
10
12
14
16
RAD/SEC
~
100-0
10-0
- PHI
(Io)
IPHI (IE)
ó
PHI (EO) IPHI (EE)
+
F (0)IF
(E)2
4
5
+
8
10
12
14
15
RAD/SEC
0-0
+-~~--+-~--~~-+--~~-+--+-~~--+-~-++
-50-0
; - - - . . . : : . . . : : - - - 1~
-100 -0
+ - - - = - - - 1w
V1 -<: I D--150 -0
+ - - - " ' - . . i ! " F - . . , . . - - - l+
-
200-0
-+---~-250-0
~
100·0
10·0
1·0
~+
- PHI (Io)/PHI (IE)
6
PHI (EO) /PHI (EE)
+
F (0) /F (E)
2
4
5
+ +
+
8
10
12
14
15
RAD/SEC
0·0
+-~-+~--+-~~-4--~4-~-+--~+-~-+-+-50· 0
+---=-~---__I~
-100.0
+---.A.~---______l
~
CL-150. 0
+---~~_:______.__l-200·0
+ - - - l
-250.0
+--_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
---l~
100·0
10-0
+
1·0
PHI (IO) IPHI (IE)
6
PHI (EO) IPHI ([E)
+
F
(0)IF ([)
2
4
5
8
10
+
+ +
+
+
15
-50.0
+---..:-~---J~
-
100· 0
+---~=__.,.__---..l
w
~t+
~6t±+
CL-
-150.0
+-_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
~~--=6~-
200·0
+---~-250-0
+--_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
--.J1·0
+---~
x
x x x
.~x x x x x x X X Xx xx6.6.
IX
I~.
I0·5
6.
-
RHJ**2
Cf
PHU IDJ /PHI (IE)
6.
RHO**2 OF PHI (EDJ /PHI (EE)
x
REF· 1.
0·0
2
4
5
8
10
12
14
15
RAD/SEC
1·0
x
6.
X
X
X
x x x x x x
x
O·S
-
RHO**2 OF PHI (IDJ /PHI (IE)
6.
RHD~fft2OF PHI (EDJ /PHI (EE)
x
REF· 1·
0·0
2
4
5
8
10
12
14
15
RAD/SEC
1·0
x
x
x
x x
x x
x
x x x X
6.6.
0·5
- RHO**2 OF PHI (IDJ /PHI (IE)
6.
RHO**2 OF PHI (EDJ /PHI (EE)
x
REF· 1·
0·0
2
4
5
8
10
12
14
15
RAD/SEC
Figure 27 Comparison of mean values, L l/s, M l/s, and H l/s task.
100·0
- PHI (IO) /PHI
(IE)
..--. 6.
PHI (EO) /PHI (E[)
z
+
F
(O)/F
(E)
H "-c...:::J~
:t-g
10·0
6: I -H~
<:+
+
1·0
2
4
5
8
10
12
14
15
0·0
RAD/SEC
-50·0
+ i ;
-.---. c...:::J~
-100-0
4 - - -...- - - 1
w
~
Cl..·-
150. 0
+---"""--~r___X'_.::_____==__---___:r_-....+ - - - i
+
-200·0
+---~~+
+
+
-250·0
- f - - - '
100-0
PHI (IO) /PHI (IE)
6.
PHI (EO) /PHI (EE)
:z:
+
F (0) /F (E)
t--I "-l:J W 0~
10-0
:=J f -+
t--I+ +
+
~+
D--~<-+
1·0
2
4
5
8
10
12
14
15
0-0
RAD/SEC
-sopo
+ - - \ - - : r - - - l ~~
-100·0
+---J~=---l
w
Ul~
CL-150-0
+---E~~-____=:___---l-200·0
-250·0
+
6 . 6 .+ +
100·0
+
- PHI (10) /PHI (IE)
6.
PHI (EO) /PHI (EE)
+
F (0)IF
(E)+
w
g
10·0
+---~~~~--~---~ r-I-t ---1 [l... ~ <. ~1·0
0·0
-50·0
~
-100·0
w
Ul~
[l...-150.0
-200·0
-250·0
~+
+
+
+
2
4
5
8
10
12
14
15
RAD/SEC
6:+
+
6.+
+ +
6 . 6 . 6 .+
+
1·0
O·S
0·0
2
1·0
0·5
0·0
2
1·0
0·5
6. 6.RHO**2 OF PHI (IO)/PHI (IE)
6.
RHO**2 OF PHI (EO) /PHI (EE)
X
REF. 1·
4
5
8
10
X XX X X X
X
RHO**2 OF PHI (IO)/PHI (IE)
6.
RHO**2 OF PHI (EO) /PHI (EE)
X
REF. 1.
4
5
8
X10
X
X
XRHO**2 OF PHI(IO)/PHI(IE)
6.
RHO**2 OF PHI (EO) /PHI (EE)
x
REF· 1.
X X X X X X12
14
15
RAD/SEC
6. X X X X X X X X12
14
15
RAD/SEC
6. X X X Xx
X X0·0
+-~-4--~~~-4--~~-+~--+-~-+~~+-42
4
5
8
10
12
14
15
Figure 31 Comparison, one run, typical subject,
RAD/SEC
100-0
- PHI
(Io)
/PHI (IE)
6-
PHI (EO) /PHI ([[)
z.
+
F
(0)
/F
(E)
t-t ""-CJ W+
0+
+
w
+
+
010·0
::J 6..6..6-+
6.. 6.. f -t-t.1.+
+
-1+
0....+
:2 <.1·0
2
4
5
8
10
12
14
15
0-0
RAD/SEC
-50·0
CJ~
-
100·0
l.J.JS!2
ËE
-lS0ftO
+---""'~---_l
-
-200.0
+ - - - . . . : + - - - l r - - - - !+
-2S0ftO
100·0
1-0
+
0·0
-50·0
... ~~
-100·0
w
Ul <::r:
CL-150.0
PHI (10) IPHI (IE)
6.
PHI (EDJ IPHI (EE)
+
F
(DJ IF (E)
2
4
5
~+
+
+
+
+
8
10
12
14
15
RAD/SEC
-200·0
+---~+
-250.0
+--_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ - - - 1100-0
PHI (Io) /PHI (IE)
+
..--. 6-
PHI (EO) /PHI (EE)
:z
:I-
F
(DJ
/F (E)
+
1---1'-
+
~w
0+
w
g
I -10·0
1---1 -.J CL ~<-+.
8
10
12
14
16
RAD/SEC
0·0
~~~~~+-~-+--r-+-~-+--~4-~-4--~~
+
-50·0
+---,,5...,...---~---1+
+
~~
-100·0
w
Ul<-
+
ËE -150. 0
+---=r:...~=::__::..,._____l
+
-200·0
+ - - - , - - - 1 +
-250.0
+ - - - r - - - 1 Figure 34 Corr.parison, one run, typical subject, H l/s task.1·0
6 6 XXX 6oXX 60 6oX~
X X X X X X0·5
-
RHO* *2 OF PHI
(10)/PHU IE)
6
RHO**2 OF PHI (EO) /PHI (EE)
X
REF. 1.
0·0
2
4
5
8
10
12
14
15
RAD/SEC
1·0
X~A 6 X X X X X X X X X X X X0·5
6-
RHO**2 OF PHI
(10)/PHI (IE)
6
RHO* *2 OF PHI (EO) /PHI (EE)
X
REF. 1·
0·0
2
4
5
8
10
12
14
15
RAD/SEC
1·0
X X Xxxx
Xx
x
0·5
X X-
RHO* *2 OF PHU
10)/PHI (IE)
1
6
RHO**2 OF PHI (EO) /PHI (EE)
x
REF. 1.
0·0
2
4
5
8
10
12
14
15
Figure 35 Comparison, one run, typical subject,
RAD/SEC
L l/s, M l/s, and H l/s task.lOOr---r---~,_---_r~
1 MEASURID DATA .± I~AvERAGm
lOVER. TRAINEn SUBJEx:TS
- FrrrID PILOT MODEL
_I:!:E 10.~---~r-_r---~~---_+~ 1.0 L...L __ .1....-....L __ L....J... __ L..-L __ ..L..-L--IL... ____________ -L..J _EEE ·100 ~---~,_--~---~---+~ &l Ol '" ~ ·200 1---+---f---=--+-+-f> ... -+4+--II 10 I' . I · · I
I
f
I
f-I
.1 I-.1 W RADlANs/sEC.Figure 36 and p 2 from ref. 1, M
I
l O O , . - - - . - - - , - - - r 1
X KUSURED DATA.±. 'C"'AVERAOEO
I OVER S TRAINEO SUBJECTS
- Frrrm Pll.OT MODEL . . EEE ID. ~:::+=I..J..-=.~~'::""::"'--=----=----t----n ·100 l---l---!--t'-.rft.,.---H ·200 ~--_4----+--____:Ht1j 10 1$ u
..
'-I .1! t!
I
1
r.) IWlIANS/SEC.! !
1111
I
I
1 -100 PH1\..CiE - - _ (degrees) 4AY --180 -220 -260 .1 w (rad/sec)
Figure 38 Comparison from ref. 16. A= 1 / s2.
X
Fourier tran8forms - - - Three parameter model - - - Two parameter model8-8 4.4s+1 1 Y(s)= 0.75 8+8 0.048+1 O. 128+1