T
tions, which in this paper are referred to
H E R E is an important class of feed though sluggish galvanometer is used to
as s-transforms and which are, in fact, a
back control systems known as sam- detect error and its position is sampled
disguised form of the Laplace transforms;
pled-data systems or sampling servo- periodically by means of a chopper bar.
while Linvill's relation involves directly
mechanisms in which the data at one or The latter permits an auxiliary source of
the Laplace transforms of the input and
more points consist of trains of pulses or power to rotate a sizable potentiometer to
output. The principal difference between
sequences of numbers. Such systems m a y a position determined by the clamped
Shannon's and Hurewicz's relations on the
have a variety of forms, a common ex galvanometer needle. The process is
one hand, and that of Linvill on the
ample of which is shown in Figure 1. In carried out at uniform intervals and a
other, is the fact that the former yield
the case illustrated, the sampling is per sampled and clamped output is obtained
only the values of the output at the
formed on the control error by a so-called for use in the continuous part of the con
sampling instants, while the latter pro
sampler which is indicated as a mechani trol system. Bandwidth is lost through
vides the expression for the output at all
cal switch which closes momentarily the sluggishness of the unloaded galva
times, though at the cost of greater
every Τ seconds. The data at the output nometer, but the power gain is enormous.
labor.
of such a switch consist of a train of Similar devices for measurement of pres
equally spaced pulses of short duration The analysis presented in this section
sure errors, flow, or other phenomena can
whose envelope is the control error func has a dual objective: to achieve a uni
be devised along the same general lines.
tion. In some practical systems, the fication of the approaches used by Shan
In addition, there are some systems in
separation between successive pulses is non, Hurewicz, and Linvill; and to for
which the data-collecting or transmission
controlled by some characteristic of the mulate the input-output relations for the
means are intermittent. Radars and
input signal and consequently is not con basic types of sampled-data systems. In
multichannel time-division communica
stant. Such systems will not be con the next section, the problem of estab
tion links are examples of this type of de
sidered in this paper. lishing a relation between the input and
vice. Such devices may be treated, in
output will be approached from a signifi
In a typical sampled-data system such general, as sampled-data systems pro
cantly different point of view. Specifi
as that shown in Figure 1, the sampler is vided the duration of sampling is small
cally, a sampled-data system will be
followed by a smoothing circuit, com by comparison with the settling time of
treated as a time-variant system and its
monly referred to as hold or clamp cir the system.
behavior will be characterized by a system
cuit, whose function is to reproduce ap Despite the increasing use of sampled-
function which involves both frequency
proximately the form of the original error data devices in the fields of communica
and time.
function b y an interpolation or extra tion and control, the volume of published
polation of the pulse train. Following the material on such devices is still rather The basic component of sampled-data
hold circuit, there are the usual compo limited. The several different meth systems is the sampler, whose output has
nents of the feedback loop, shown in Fig ods which have been developed for the the form of a train of narrow pulses occur
ure 1 as i i and (7, comprising amplifiers, analysis of sampled-data systems are ring at the sampling instants 0, =»= Γ,
shaping networks, and the controlled closely related to the well-known mathe =ί=2Γ, . . . , where Τ is the sampHng in
member. matical techniques of solution of dif terval; see Figure 2. The frequency
It is apparent that the insertion of a ference equations. It is the purpose of coo=27r/r is called the sampling fre
sampler into an otherwise continuous con this paper to unify and extend the meth quency.
trol system in general should result in an ods described in the literature and to in For purposes of mathematical con
inferior performance due to a loss of in vestigate certain basic aspects of sampled- venience, it is expedient to treat the out
formation in the control data. Yet, sam data systems. put pulses as impulses whose areas are
pled-data systems have certain engineer equal to the values of the sampled time
ing advantages which make them pref Input-Output Relations function at the respective sampling in
erable in some applications to continu stants. (This is permissible provided the
ous-data systems. The most important of A central problem in the analysis of pulse duration is small compared with the
these advantages is the fact that error sampled-data systems is that of estab settling time of the system and the gain
sampling devices can be made extremely lishing a mathematical relation between of the amplifier following the sampler is
sensitive at the expense of bandwidth. the input and output of a specified sys multiplied by a factor equal to the time
An example of such a device is the electro tem. This problem has received con duration of the sampling pulse.) Thus, if
mechanical galvanometer and chopper siderable attention in the literature of the input and output of the sampler are
bar. In this device a very sensitive sampled-data systems, with the result denoted by r{t) and r*(/) respectively, the
^(^^)=(TT^{^-(Srj
that the expression for the Laplace trans Substituting these expressions in equa
form of the error is tion 34, there results after minor simpli
fications (40)
£:(5) = 7 ? ( 5 ) - H ( i ) C * ( 5 ) (28)
C{s)^G{s)E(,s) (29)
r*
(1) .R*(z)
Combining this relation with equation 28 R*{s).
gives
C{s)^G{s)R{s)-G{s)H{s)C\s) (30) c*
(2) . .CR*(s) CRHi)
Applying the z-transformation (see equa
tion 19 and following) to both sides of r* . .G(s)K*(s) GHiiR*(z)
this equation results in
(5)
C\s)^GR\s)-GH\s)C\s) (31)
The frequency response of this hold sys transfer function is analogous to the con 0.264+0.3682
tem is obtained b y replacing the complex ventional transfer function, the same GH*(z)=^ (64)
(0.368-2)(l-2)
frequency 5 b y jco, resulting in the follow basic technique m a y be applied in the
ing expression for the magnitude of the case of sampled-data control systems. This function is plotted in Figure 8 where
transfer function The function which is plotted is GH*{s) or it is seen that for the constants chosen, the
G*(5)ii*(s) according as the denominator system is stable; but if the loop gain is in
(60) of the z-transform of the output is of the creased b y a factor of 1.5, the system be
sm
form l+GH*{s) or l+G*{s)H*{s), see comes unstable.
This relation is plotted in Figure 6. I t is Table II. As usual, the complex fre If i t were desired to improve the margin
observed that the hold system is essen quency s is varied along a contour in the s- of stability or, for that matter, stabilize
tially a low-pass filter which passes the plane consisting of the imaginary axis and this system with a higher loop gain, the
low frequency spectrum of the impulse a semicircle enclosing the right-half of the procedure would be to add lead networks
train and rejects the displaced high-fre plane. Since the starred transfer func just as in the case of a continuous-data
quency spectra resulting from the sam tions are periodic with period ^ωο, the loci system. The major difficulty encoun
pling process. A n important property of of the functions GH*{s) or G*{s)H'^(s) re tered in this procedure is that the result
the zero-order hold system is that the rip trace themselves a t each cycle, so that s ing starred loop transfer function is not
ple at the output is zero if the input is a need be varied only from —^ωο/2 to jcoo/2 related in a simple manner to the original
constant. I n a similar way, the first order in order to obtain the shape of the locus. function, for as was shown previously the
hold system has zero ripple output for an Critical regions such as the vicinity of the starred transfer function of two networks
input function whose slope is a constant. origin in the 5-plane are handled in the in tandem is not equal to the product of
same mannter as in the case of continuous- respective starred transfer functions.
As is well known, if the signal does not
data systems. Consequently, the insertion of a correc
contain frequencies higher than one-half
Since the auxiliary variable ζ is de tive network in the feedback loop re
the sampling frequency, perfect repro
fined as €^*, it is evident that as 5 is as quires the recalculation of the starred
duction of the signal is obtained with an
signed imaginary values over one com loop transfer function. The need for re
ideal low-pass filter (that is, one with
plete cycle, ζ traces a unit circle in the z- calculation is inherent in the stabilization
unity gain and linear phase shift up to its
plane. Thus, the loop transfer function of sampled-data systems b y the insertion
cutoff frequency) whose cutoff frequency
GH'^is) can be plotted directly b y express of a corrective network in the feedback
is equal to one-half the sampling fre
ing it in the form GH*{z) and varying ζ loop. Thus, as the art exists at the pres
quency. For that matter, a n y low-pass
along the unit circle. T o demonstrate the ent time, the shaping procedure involves
network having roughly this characteris
technique, the GH*(z) locus for the sys essentially a trial and error method with
tic can be used t o extract most of the use
tem shown in Figure 7 will be plotted. the plot of the Nyquist diagram for each
ful spectrum from the impulse train. I t is
This system is seen to consist of a zero- trial set of system parameters used to as
even possible, though not advisable, to
order hold circuit and a simple linear sess the stability of the system.
dispense with the hold circuit altogether,
and rely on the low-pass characteristics of component in the forward circuit. The
the forward circuit t o perform the neces feedback transmission is unity so that the Conclusions
sary smoothing of t h e sampled data. loop transfer function is
The sampled-data feedback control sys
System Design Using s-Transforms G i s m s ) J ^ ^ ^ , T ^ l (61) tem m a y be analyzed in a systematic
manner b y applying the z-transform
The primary objectives in the design of Expanding this transform into partial method. Techniques of locus shaping
closed-cycle control systems include the fractions, there results similar to those commonly used in con
achievement of stability and an accept tinuous systems m a y be applied b y plot
able over-all dynamic performance. The Gis)H(s) ting the starred loop transfer function on
=^^-"^<7^-'-^ΙΤγ) ''''
standard technique of achieving these o b the complex plane. This is done with no
jectives consists in plotting the Nyquist On considering each term separately and approximations other than those relating
diagram of the loop transfer function and obtaining the corresponding 2-transforms to the narrowness of the pulses constitut
adjusting the system parameters until the from Table I, the starred loop transfer ing the pulse train. Once satisfactory
plot does n o t enclose the point —1.0. function is found to be loci are obtained b y the addition of ap-
There are several points and results in this study the behavior of the output also be
D iiscussion paper which bear interesting relations to tween samphng instants. Furthermore,
some of the work done by the discussor in and this is important, it is not always a
John M. Salzer (Hughes Aircraft Company, connection with digital-analogue systems.^ priori obvious whether this is or is not the
Culver City, Calif.): The objectives of this The use of starred j-transforms, such as case.
paper are: (a) to unify the z-transform R*(s), in place of infinite sums, is a welcome Where the output behavior between
(Shannon, Hurewicz) and 5-transform (Lin convenience. 3 This notation facilitates the samples is also of interest, the z-transform
vill) approaches to sampled-data systems; manipulations in dealing with sampled-data method i s still applicable but must be aug
(b) to formulate input-output relations for systems and makes the systematizations of mented by separate investigation of the out
various types of such systems; and (c) to Table II of the paper easier to comprehend. put during the sampling period. A separate
treat such systems by considering the As noted in the paper, in the case of solution based on initial conditions at the
sampler a time-variant element. starred transforms (that is, in the case of sampling instant may be used in analysis,
The paper fills a clear need in this under- sampled functions) the $- and z-transforms but a synthesis procedure would hardly be
published field, and it presents its topic con are exactly equivalent. Thus, the use of z- fruitful along these lines.
cisely and illuminatingly. It brings to transforms involves no approximation, and The alternative solution is the 5-trans-
gether two viewpoints, (a) and (c), to bear it offers certain conveniences. One ad form approach. As noted by the authors,
on the same problem and shows the way to vantage is notational, for ζ is easier to write this method leads to transforms which are
ward systematization (b) of the solution. tha-n e"'. The other is that the infinite products of rational functions of 5 and z,
Perhaps the most significant contribution number of poles and zeros of the starred 5 - and tables of corresponding transform pairs
of the paper is the variable network ap transforms are replaced by a finite set in the are not available. Nevertheless, the exact
proach. Although the applicability of this z-plane. Furthermore, the ambiguity at the analysis is straightforward, even if some
approach was recognized before, ^ the auth point of infinity in the 5-plane (due to essen what laborious, and moreover, the fre
ors' more recent investigations in the gen tial singularity) is circumvented by the use quency characteristics obtained may give a
eral field of time-varying systems are now of z-transforms. hint as to the nature of compensation
made^ to bear directly on the problem of In case the 5-transform is not a starred needed to improve the response.
sampljEd-data systems. The importance of transform, information is lost by the use of Whether the z-transform method is ap
the generalization thus afforded should not z-transform because this amounts to repre plicable or not in a particular case depends
be overlooked. senting a continuous function, c{t), by its on the question of bandwidth. If the
It is interesting to note that the variable samples, c*(t), just as it is done in numerical sampling rate is many times higher than the
network approach leads directly to the same mathematics. But whereas in numerical bandwidth of the input or of the system,
expressions as the 5-transform method. work the sampling (or tabular) interval may then the z-transform solution is expected to
Both methods are predicated on the concept be adjusted until such a representation is be a suitable representation of the continu
of using an arbitrary, characteristic input justified, in analyzing a given control system ous output. It is presumed that the ap
function [r(/) = , but by distinguishing one is faced with a sampling rate already de plications with which the authors concerned
a time-variant transfer function certain gen termined and the analysis must be made themselves were of this type. However, in
eralizations are made possible, as already correct for the existing physical situation. the design of certain systems it is often de
noted. An interesting illustration of such a For particular systems the sequence of out sirable to use the lowest permissible sam
generalization is given in the section titled put samples may not give a satisfactory pling rate consistent with the specifications.
"Response to a Random Input." picture so that it becomes necessary to In finding this limiting rate, one does not