: s'.ui minr:MBmuti»i:tiHia8nMnMMM!

aMinmnlnMinttilK

McORAW«HIU SERIES IN

mtro

.jumniuusiiimHHJnmimiWHtmsmi'ims
621.8 G35c
Gibson & Tuteur 1058249

Control system components

RDS FROM POCKET

XTROT REMOVE

PUBLIC LIBRARY
FORT WAYNE AND ALLEN COUNTY, IND.
ACPL ITEM
DISCARDED
CONTROL SYSTEM COMPONENTS
 McGraw-Hill Series in Control Systems Engineering
John R. Ragazzini and William E. Vannah, Consulting Editors

Cosgriff • Nonlinear Control Systems

Gibson and Tuteur • Control System Components

Goode and Machol • System Engineering

Laning and Battin • Random Processes in Automatic Control

Ragazzini and Franklin • Sampled-data Control Systems

Savant • Basic Feedback Control System Design

Smith, O. J. M. • Feedback Control Systems

Truxal • Control Engineers’ Handbook

Control System Components

JOHN E. GIBSON, Ph.D.

Associate Professor of Electrical Engineering
Purdue University

FRANZ B. TUTEUR, Ph.D.

Associate Professor of Electrical Engineering
Yale University

McGRAW-HILL BOOK COMPANY, INC.

New York Toronto London

1958
 CONTROL SYSTEM COMPONENTS

Copyright © 1958 by the McGraw-Hill Book Company, Inc. Printed

in the United States of America. All rights reserved. This book, or
parts thereof, may not be reproduced in any form without permission
of the publishers. Library of Congress Catalog Card Number 57-12581

THE MAPLE PRESS COMPANY, YORK, PA.

 PREFACE
1C58249
Although a fairly large number of books dealing with the theory of
servomechanisms has become available in recent years, most of them
do not concern themselves to any great extent with the components
required to implement the theory. The texts that have been written
primarily about components have in general consisted merely of qualita¬
tive descriptions of available equipment. In the opinion of the authors
of the present work a more quantitative approach to the subject of
components is needed, and this book is an attempt to fill this need.
In making this attempt some compromises have, of course, been neces¬
sary. The number of components available for use in feedback control
systems is so large that even a short description of each could easily
fill several volumes. On the other hand it is possible to go into so much
detail in the description and analysis of only a very small number of
components that if the length of the book is to be kept within reason it
becomes impossible to discuss many commonly used components at all.
We have attempted to steer a middle course in this respect by presenting
a fairly detailed but by no means exhaustive analysis of a number of
what it is hoped are typical components. Then, in order to make the
coverage of the field somewhat more complete, shorter descriptions of a
somewhat larger group of devices have also been included. In most
cases these are similar to the typical components discussed in detail.
The decision on what was to receive detailed coverage, what was to
receive only brief discussion, and what was to be left out altogether has,
of course, been somewhat arbitrary. In general we have tried to exclude
discussion of conventional devices that are used in standard fashion
in control systems and we have considered in greatest detail those devices
that seem to us to be of primary interest to the control-system engineer.
Our purpose has been to present engineering principles and methods
of analysis rather than specific discussions of commercial devices. Cata¬
logues of equipment and tables of data that may be found in standard
handbooks have not been included. In a number of cases mathematical
analyses are not completed; however, it is hoped that sufficient back¬
ground has in all cases been developed so that the reader should have
little difficulty in completing these analyses for himself. Usually, sug¬
gestions for such extensions will be found in the problems that follow
each chapter.
VI PREFACE

The reader of this text is assumed to have a basic knowledge of servo

theory or to be acquiring it concurrently. In particular, it is assumed
that he has had an introduction into the Laplace transform method of
analysis and is familiar with such concepts as transfer functions and
complex impedance. These notions are reviewed briefly in Chapter 1
(mainly to introduce our notation which is slightly unconventional)
and are assumed to be known in all of the other chapters. Also, a certain
amount of electrical background is essential. For most of the topics
covered in this book the electrical engineering background obtainable
in one of the standard survey courses given to nonelectrical engineering
majors should be sufficient. However, unfortunately for the non¬
electrical major, many of the components of control systems are elec¬
trical. Also, although the authors tried valiantly to keep the point of
view of the nonelectrical control-system engineer in mind, they suffer
from the handicap of being electrical engineers themselves. Hence
it is clear that the greater the electrical background, the better. This
applies particularly to Chapter 2 in which a knowledge of standard
vacuum-tube theory is assumed.
On the whole the individual chapters of this book are self-contained
and can be considered in any order, although naturally the authors
prefer the order here followed. If the order is changed, it is advisable
that the first twelve sections of Chapter 1, or at least Sections 1.10 to
1.12, be read first, since these sections contain the basic review material
on the Laplace transformation and the Laplace transform nomenclature
that is used throughout the book.
The authors have tried to treat each one of the major topics for the
nonspecialist. For this reason a number of the chapters start with
material that will appear elementary to the specialist and then proceed
to more advanced levels. In general, the introductory sections contain¬
ing the background material are kept as short as possible and no attempt
at completeness has been made. Instead only the details necessary for
the work at hand have been included. In most cases the central purpose
of the analysis is the derivation of the transfer function of the com¬
ponent. For components described by nonlinear differential equations
linearization techniques are presented and the usual assumptions that
are made in order to obtain a linear approximation are discussed.
Large parts of certain chapters of the book have been presented in a
graduate course on control-system components at Yale University for a
number of years, while most of the other parts have also been presented
in undergraduate and special courses offered to mechanical and electrical
engineers in industry.
John E. Gibson
Franz B. Tuteur
 CONTENTS

Preface. v

Chapter 1. Electric Networks for Control Systems. 1

1.1 Introduction 1.2 Resistors 1.3 Variable Resistors 1.4 The Precision
Variable Resistor 1.5 Mechanical Characteristics of the Precision Variable
Resistor 1.6 Nonlinear Precision Variable Resistors 1.7 Temperature Coeffi¬
cient of Resistance 1.8 Capacitors 1.9 Inductors 1.10 The Analysis of
Simple Networks 1.11 Steady-state Frequency Response from the Transfer
Function 1.12 Asymptotic Representation of the Frequency Response 1.13
Common RC Networks 1.14 The Synthesis of RC Networks. Introduction
1.15 Properties of RC Networks 1.16 Properties of Two-terminal RC Networks
1.17 Transfer Functions Realizable with Four-terminal RC Networks 1.18 L
Sections 1.19 Cascading of L Sections 1.20 Approximate Ladder-network
Synthesis 1.21 Approximate Synthesis of L Sections 1.22 An Exact Synthesis
Method of L Sections for Simple Transfer Functions 1.23 The Exact Synthesis
of L Sections. General Method 1.24 Exact Synthesis of RC Driving-point
Impedances 1.25 Numerical Example 1.26 Bridge-T and Twin-T Networks
1.27 The Use of Amplifiers with RC Networks 1.28 Reliability Problems

Chapter 2. D-C Amplifiers.61

2.1 Introduction 2.2 Drift in D-C Amplifiers 2.3 Effect of Negative Feedback
on Drift 2.4 Analysis of Simple Triode Circuits 2.5 The Cathode Follower
2.6 Input Impedance of the Cathode Follower 2.7 Drift Due to Component
Variation 2.8 Cathode Follower with Plate-load Resistor 2.9 Equivalent
Circuit for the Cathode Follower 2.10 The Triode Amplifier with Cathode
Bias 2.11 The Difference Amplifier 2.12 The Miller Circuit 2.13 Phase
Inverters 2.14 The Cathode-follower Inverter 2.15 The Paraphase, or Bal¬
anced, Inverter 2.16 The Cathode-coupled Inverter 2.17 Interstage Coupling
Networks 2.18 The Transistor D-C Amplifier 2.19 Analysis of Simple Tran¬
sistor Circuits 2.20 Single-stage Circuits 2.21 Drift in Transistor Amplifiers
2.22 Multistage Amplifiers 2.23 The Chopper-stabilized D-C Amplifier 2.24
Design Considerations 2.25 Triode Amplifier with Resistance-coupling Net¬
work 2.26 Design of Triode Amplifiers with Neon-tube Type Interstage
Coupling Network 2.27 Design of Cathode-follower Circuits 2.28 Design
of Difference Amplifiers Problems

Chapter 3. Power Amplifiers.115

3.1 Introduction 3.2 A-C Power Amplifiers 3.3 D-C Power Amplifiers
3.4 Thyratron Amplifiers 3.5 Control of Firing Angle 3.6 The Thyratron
Vlll CONTENTS

Amplifier with an Inductive Load 3.7 The Full-wave Thyratron Amplifier

with Inductive Load 3.8 Thyratron-amplifier Transfer Function 3.9 The
Thyratron Amplifier with a D-C-motor Load 3.10 The Effect of Armature
Inductance 3.11 D-C Motor Driven by Full-wave Circuit 3.12 Thyratron
Circuits for Reversing Control 3.13 Other Thyratron Applications 3.14
Relay Amplifiers 3.15 Relay-amplifier Stability 3.16 Relay-amplifier Static
Characteristic 3.17 The Frequency Response of a Relay Amplifier 3.18
Magnetic Amplifiers. Introduction 3.19 Operation of the Series-connected
Magnetic Amplifier 3.20 Time Constant of the Series-connected Amplifier
3.21 Reactors Connected in Parallel 3.22 Magnetic Amplifiers with Feedback
3.23 The Self-saturated Magnetic Amplifier 3.24 The Ramey Magnetic Ampli¬
fier Problems

Chapter 4. D-C Machines.179

4.1 Introduction 4.2 Generated Voltage in a Simple D-C Generator 4.3 The
Effects of Hysteresis and Eddy Currents 4.4 Factors Affecting Terminal
Voltages of a Loaded Generator 4.5 Figure of Merit 4.6 The Rototrol Gen¬
erator 4.7 Analysis of the Rototrol Generator 4.8 Pilot Generators with
Several Control Windings 4.9 The Regulex Generator 4.10 The Amplidyne
Generator 4.11 Analysis of the Amplidyne 4.12 The Equivalent Circuit
of the Amplidyne 4.13 Inaccuracies in the Amplidyne Analysis 4.14 Com¬
parison of the Various Generator Types 4.15 D-C Motors with Separate
Excitation 4.16 Relation between the Transfer Function and the Speed-
torque Curves 4.17 An Equivalent Circuit for the D-C Motor 4.18 Inac¬
curacies in the Motor Analysis 4.19 Other D-C Motor Types 4.20 The Field-
controlled Motor Problems

Chapter 5. Synchros and Related Devices.220

5.1 Accuracy and Precision 5.2 Elementary Operation of Synchros 5.3 Static
Errors and Residual Voltages 5.4 Velocity Errors 5.5 Two-speed Systems
5.6 The Differential Generator 5.7 Synchro Repeaters 5.8 Variable-re¬
luctance Transducers 5.9 The Microsyn Problems

Chapter 6. Demodulators and Modulators.249

6.1 Fundamental Concepts 6.2 Analysis of the Half-wave Discriminator
6.3 Full-wave-discriminator Circuits 6.4 Other Triode Discriminator Circuits
6.5 Diode Discriminators 6.6 Electromechanical Demodulators 6.7 The
Clamping Discriminator 6.8 Modulators 6.9 Drift in Modulators Problems

Chapter 7. A-C Motors.276

7.1 Introduction 7.2 Construction Features 7.3 Theory of Operation 7.4
Theory of Operation of A-C Servomotors 7.5 Approximate Transfer Function
of the A-C Servomotor 7.6 Figures of Merit for A-C Motors 7.7 Motor
Characteristics in the Presence of Finite Control Impedance 7.8 Unbalanced
Stator Windings 7.9 Single-phasing of an Induction Motor 7.10 The Induc¬
tion Motor as a Tachometer 7.11 Other A-C Servomotors. The Shaded-pole
Induction Motor Problems

Chapter 8. Mechanical Networks and Gears.305

8.1 Introduction 8.2 Springs, Masses, and Dashpots 8.3 Mechanical Equal¬
izers 8.4 Electromechanical Networks for Carrier Servos 8.5 The Lancaster
 CONTENTS IX

Damper 8.6 Gears 8.7 Design of Gear Trains for Minimum Inertia 8.8
Gear Ratio for Load Matching 8.9 Backlash in Gears 8.10 Ball-screw
Actuator Problems

Chapter 9. Mechanical Components.334

9.1 Introduction 9.2 Differential Gears 9.3 The Universal Joint 9.4 Strain
Gauges 9.5 The Flyweight Tachometer 9.6 The Gyroscope. Introduction
9.7 The Gyroscope. Precession 9.8 Gyroscope. Equations of Motion 9.9
Practical Gyroscopes 9.10 Gyroscope Applications 9.11 Application of
Gyroscopes to Inertial Navigation Problems

Chapter 10. Pump-controlled Hydraulic Systems.363

10.1 Introduction 10.2 Basic Types of Hydraulic Control Systems 10.3 The
Pump-controlled Hydraulic System 10.4 Analysis of the Pump-controlled
System 10.5 Transfer Function of the Pump-controlled System 10.6 Other
Pump and Motor Types 10.7 Piston Pumps 10.8 Vane Pumps 10.9 Gear
Pumps 10.10 The Ball Pump 10.11 Hydraulic Transmission Lines Problems

Chapter 11. Valve-controlled Hydraulic Systems.386

11.1 Introduction 11.2 Spool-type Pilot Valves 11.3 Pulsed Operation of
Hydraulic Valves 11.4 Flow-pressure Relations in Orifices and Valves 11.5
Axial Hydraulic Reaction Forces in Spool-type Valves 11.6 Radial Hydraulic
Forces in Spool-type Valves 11.7 Graphical Analysis of Flow Control Valves
11.8 Linearized Small-signal Analysis 11.9 Flapper Valves 11.10 Nozzle
Valves 11.11 Slide Valves 11.12 Two-stage Valves 11.13 Pressure-regulat¬
ing Devices 11.14 Choice of Operating Pressure for Hydraulic Systems Prob¬
lems

Chapter 12. Pneumatic Systems.435

12.1 Introduction 12.2 Pneumatic Flow 12.3 Pneumatic Flow through an
Orifice 12.4 Compressible Flow in Pipes 12.5 Compressibility as an Equiva¬
lent Spring 12.6 Pneumatic Transmission Lag 12.7 Pneumatic-Electric
Analogs and Control Functions 12.8 A Pneumatic Control System Problems

Chapter 13. Pneumatic Components.461

13.1 Introduction 13.2 Comparison of Weight of Electric, Hydraulic, and
Pneumatic Systems 13.3 Choice of Operating Pressure 13.4 Pneumatic
Power Supplies 13.5 Pneumatic Control Valves 13.6 Pneumatic Motors
13.7 Actuators Problems

Index 481
n
 CHAPTER 1

ELECTRIC NETWORKS FOR CONTROL SYSTEMS

1.1. Introduction. The vast majority of servo control systems in use

at the present time employ electronic amplifiers and electric networks.
This is true in spite of the facts that electronic components are often less
reliable than other types and that conversion equipment is usually
required to convert nonelectric signals into electric ones, and vice versa.
The main reason for the popularity of electronic components is, of course,
the relative ease with which electric signals can be manipulated. Thus,
amplification of electric signals is easily accomplished by electronic cir¬
cuits. While mechanical, hydraulic, or pneumatic amplifiers do exist and
are widely used, they are much more complex and more expensive and, in
general, perform much less well than electronic amplifiers. Similarly,
mechanical, hydraulic, or pneumatic devices can be built to perform the
function of frequency selection, i.e., to amplify some signals and to attenu¬
ate others on the basis of frequency. However, electric networks are
ordinarily much preferred for this purpose, because they are easy to
design and to construct and, except for certain specific instances,1 their
performance is superior to that of the other types of devices.
In the present chapter we deal with the electric networks used in
control systems. These networks consist primarily of resistors and
capacitors, although inductors are occasionally used. Before the networks
are discussed, therefore, these three components are considered.
1.2. Resistors. Resistors are commonly made of a composition of
carbon or from high-resistance wire. The fixed carbon resistor is manu¬
factured in three standard power ratings: and 2-watt sizes; and
in three standard tolerances: 20 (uncommon and no less expensive
than 10 per cent tolerance), 10, and 5 per cent. Figure 1.1 gives the
standard color code for resistance value and tolerance. It is becoming
common practice to print the resistance on the body of the resistor in
addition to color coding it.
The size of a carbon resistor is not indicative of its resistance value,
since the composition is varied for various resistance values, but it does
indicate the approximate power rating. Carbon resistors are made on a
1 R. Adler, Compact Electromechanical Filter, Electronics, April, 1947, pp. 100-
105. W. Roberts and L. Burns, Jr., Mechanical Filters for Radio Frequencies,
RCA Rev., yol. 10, pp. 348-365, 1949.
1
2 CONTROL SYSTEM COMPONENTS [Chap. 1

Table 1.1. Standard Values of RMA Carbon Resistors

10 ohms 33 ohms
12 39
15 47
18 56
22 68
27 82
and powers of 10 thereof

mass production basis, and the Radio Manufacturers’ Association (RMA)

has established standard production resistance values. These values are

Figure or Figure or
■ /First significant figure number of number of
Color zeros Color zeros
~ ^—Second significant figure Black 0 Green 5
Brown 1 Blue 6
■ '^"Multiplier Red 2 Violet 7
. Orange 3 Gray 8
Tolerance Yellow 4 White 9

Color % tolerance
None 20
Silver 10
Fig. 1.1. Resistor color code. Gold 5

given in Table 1.1. The values are chosen so that the percentage change
in resistance between each of the values is approximately constant. Values
at one-half the standard percentage step are also produced but are not so
widely stocked and may be somewhat more expensive.
Wire-wound resistors may be obtained in any resistance value and
wattage rating. Standard RMA values in the 5- and 10-watt ratings
are commonly available from stock. Several companies specialize in
supplying precision resistors to any accuracy required.
1.3. Variable Resistors. Sometimes the term “variable resistor” is
applied to a device which has only two terminals. One terminal goes to one
end of the resistor, while the other terminal is connected to the wiper arm.
The symbol is given in Fig. 1.2a.
O

The name “potentiometer” is then

restricted to a device which has both ° 0 vww^p/wwv
ends of the resistance brought out ^ ^
to terminals and a third terminal for
,i .. , • Fig. 1.2. Variable-resistor and potenti-
the connection to the wiper arm, , u ,
as shown in Fig. 1.26. This nomen¬
clature is not universal, and both names may be used interchangeably.
Variable resistors either are made from a carbon composition or are
wire-wound. The carbon resistor is less accurate and less expensive.
The carbon variable resistor usually has a tolerance of 5 per cent of rated
Sec. 1.4] electric networks for control systems 3

value. The wire-wound variable resistor can be made to extremely accu¬

rate specifications. Since World War II a practically new industry has
sprung up to manufacture variable resistors to standards which would
have been unbelievable 10 years earlier.
1.4. The Precision Variable Resistor. The precision variable resistor,
widely used in analogue computers and as a pickup device in servo¬
mechanisms, is manufactured in two forms: both are wire resistors. (A
process which uses a deposited film of resistance material holds some
promise, but it is not as yet of commercial importance.) The first form
is the conventional wire-wound resistor held to close tolerances and with
quality materials used. The resistance wire is wrapped on a “card,” or

Fig. 1.3. Cutaway of precision variable resistors: (a) single-turn; (6) helical-wound
resistor; (c) multiturn slide-wire. (Courtesy G. M. Giannini and Co.)

form, which is then bent into a circle. The wiper bears on the edge of the
card and moves from turn to turn (see Fig. 1.3).
Accuracy of a variable resistor, or “linearity,” is commonly defined in
two ways. So-called “best linearity,” “independent linearity,” or nor¬
mal linearity is the maximum departure of the actual resistance curve from
the best straight line which can be drawn through the actual curve, this
departure being expressed as a percentage of the total resistance of the
4 CONTROL SYSTEM COMPONENTS [Chap. 1

device (see Fig. 1.4). Note that the straight line is not drawn through
the origin, and thus a better figure for linearity is obtained than may be
obtained if the zero-based linearity is used. Zero-based linearity is the
maximum departure of the actual resistance curve from the best straight
line drawn through the origin. This is the more natural definition, but
precision-resistor manufacturers have adopted normal linearity almost
universally. A single-turn wire-wound resistor may be obtained from
stock with 0.1 per cent normal linearity for resistance values of 5,000 ohms
and above. The body diameter of this device is of the order of 3 in. If
either the resistance or the body diameter is smaller, the number of turns
of wire must be reduced, and thus the linearity is reduced.
The resolution of such a device is also important. Resolution of a varia¬
ble resistor is defined as the smallest change in resistance which can be
detected, expressed as a percentage of the total resistance of the device.

End voltage (or resistance). End voltage (or resistance)

Voltage of potentiometer - Voltage of potentiometei-
across across
terminals terminals
L inearity curve of / Linearity curve of /
potentiometer—- / potentiometer —/s'f'
/y /
Straight line — Linearity tolerance / lr /'
(limit lines) V //
’ y / /
curve or /l / / /
Voltage Voltage
at moving at moving Linearity ' '
contact contact tolerance S//
(minus)-\ ///y/

,y dr / Straight line
approximating curve
/ A of potentiometer
/
Total /
/
/ but passing through
Total
zero point
travel / travel
T~o
End voltage or Travel of contact Endvoltage or Travel of contact
resistance of resistance of
potentiometer potentiometer
(o) Normal linearity (£) Zero-based linearity

Fig. 1.4. Linearity of variable resistors.

The smallest resistance change detectable in a wire-wound resistor is that

of one turn of wire, and thus

_1_
Resolution = X 100% (1.1)
total number of turns of wire

T}^pically, a single-turn (360° card) wire-wound resistor with a body

diameter of 3.5 in. has a resolution of about 0.04 per cent at 10,000 ohms.
A larger body diameter or higher resistance would result in better resolu¬
tion (a smaller percentage).
It is possible to construct a wire-wound resistor with the card not bent
into a circle but rather with the wire spiraled in the form of a helix. This
allows more turns of resistance wire and increases the linearity and
improves the resolution. While the usual number of turns of the helix is
Sec. 1.5] electric networks for control systems 5

10, resistors are commercially available with as many as 40 or 50 helical

turns. Figure 1.36 shows a 10-turn helical resistor.
In another type of variable resistor the resistance is not wound on a
card so that the wiper moves from turn to turn of resistance wire; instead
the wiper is always in contact with the resistance wire. The high-resist¬
ance wire is shaped in a spiral, usually of 10 turns. This is merely a
modification of the linear slide-wire potentiometer. The advantages of
the slide-wire resistor are infinite resolution (0 per cent by the standard
definition), low capacitive and inductive effects, and excellent zero-based
linearity. The disadvantages are the difficulty of maintaining linearity
for resistance values above 50,000 ohms and the somewhat higher cost.
Slide-wire variable resistors are available in resistance values up to 25,000
ohms with zero-based linearity as good as 0.025 per cent and normal
linearity as good as 0.01 per cent. The zero-based linearity of the slide-
wire device excels that of the wire-wound resistor in general, and the nor¬
mal linearity of the slide-wire resistor compares favorably with that of
the wire-wound resistor in the resistance range of 10,000 to 25,000 ohms.
At lower values of resistance the normal linearity of the slide-wire resistor
is better than that of the wire-wound device.
1.5. Mechanical Characteristics of the Precision Variable Resistor.
Precision resistors are usually mounted within a steel, aluminum, or plastic
case. The shaft bearings on the standard models are precision ball bear¬
ings which require no lubrication. Most models will withstand acceler¬
ations of up to 100 times that of gravity along any axis without damage.
The multiturn models are equipped with mechanical stops at each end of
the resistance wire. Single-turn units may be equipped with mechanical
stops, or in some models continuous mechanical rotation is possible.
Standard models with ball bearings require about 1 oz-in. starting torque,
and low-torque models with ball bearings are available which require less
than 0.25 oz-in. starting torque. Models are commercially available
with jewel bearing which require a maximum of 0.003 oz-in. starting
torque. Usually the starting torque may be divided by 2 to find the
running torque required.
Many manufacturers do not seem to realize that, in addition to the
quantities mentioned above, the control engineer requires the moment of
inertia of the rotating parts of the variable resistor, in order to compute
the mechanical time constant of the unit into which the resistor is con¬
nected. Therefore this datum is not always available. Assuming the
usual lightweight construction, a resistor with a body diameter of 3 in.
may have a moment of inertia of about 0.05 oz-in.2. A resistor with a
body diameter of less than 1 in. may have a shaft moment of inertia of
less than 0.005 oz-in.2, and one manufacturer has a regular production
model with a starting torque of 0.005 oz-in. and a moment of inertia of
5 X 10-6 oz-in.2. Many standard models have fittings which make
6 CONTROL SYSTEM COMPONENTS [Chap. 1

it possible to connect several resistors coaxially on the same shaft; this is

called “ganging” the resistors. If three or more resistors are ganged, the
shaft should be brought out through the last resistor and supported by a
bearing to prevent twisting of the first resistor’s shaft by the weight of
the ganged resistors.
1.6. Nonlinear Precision Variable Resistors. It is sometimes neces¬
sary to cause the voltage at the resistor wiper arm to be a particular non¬
linear function of shaft rotation. This might be necessary, for instance,
in synthesizing the optimum switching function in a relay servo or in the
scheduling control for a jet fuel-control system. One method of accom¬
plishing this is to wire the resistance card deliberately to give the required
function. The disadvantages of this method are that it is very difficult
to construct the winding accurately and, second, that the resolution varies
with the spacing of the resistance wire unless the width of the card is
varied in accordance with the function. A second method, which is more
practical, consists in arranging a linear resistor with taps every few
degrees. It is impossible to locate taps any closer than 10 mechanical
degrees to each other, and these taps must be installed at the factory dur¬
ing the manufacturing process. Padding resistors may be installed across
the taps to shape the resistance characteristic in a series of straight-line
segments that approximate the desired curve. This is called resistance
;padding. The linear-resistor characteristic must have at least as steep
a slope as the steepest slope of the desired characteristic, since it is impossi¬
ble to increase the slope of the linear resistor by placing a padding resistor
in parallel. Note further that the slope can never be less than zero, since
this would require negative resistance.
A second method of padding the linear resistor to approximate a curve
with straight-line segments is called voltage padding. Voltage padding1
consists in placing voltages across the taps of the linear resistor to shape
the characteristic of the resistor’s output voltage versus shaft position
to approximate a desired nonlinear function; for typical example see
Fig. 1.5. Voltage padding is more flexible than resistance padding
because there are no restrictions on the slope of the desired characteristic.
Theoretically, it is merely necessary to calculate the voltage required
between taps and to connect a suitable voltage source between the taps.
Actually, several other factors must be considered. First, it is necessary
that the power-dissipation rating of the variable resistor not be exceeded.
Not only must the over-all rating be observed, but also the power dissi¬
pated between any two taps must not be excessive. The minimum total
value of resistance of the variable resistor is established by the segment
with the maximum voltage gradient.
1 G. R. Korn, Design and Construction of Universal Function Generating Potenti¬
ometers, Rev. Sci. Instr., Vol. 21, no. 1, p. 77, January, 1950.
Sec. 1.7] ELECTRIC NETWORKS FOR CONTROL SYSTEMS
__- 7

A second factor that must be considered is the internal impedance of

voltage supplies that are placed across the taps. The simplest method of
obtaining the proper voltage gradient is to use a fixed battery and a volt¬
age-dropping resistor. A more sophisticated approach is to employ a
regulated, adjustable voltage supply that has a very low output
impedance.
Finally, the current drawn by the wiper arm must be considered. Fre¬
quently the potentiometer resistance is made low enough that the voltage
drop produced by the wiper current is negligible. Alternatively, if the
resistance connected between the wiper arm and ground is constant, it
may be more accurate to take the loading into account in the design and
to adjust the padding in such a way
$ $ £
that the loaded potentiometer has &
§
$
§ §
the desired characteristic. ft! ft!
+ + + + +
It is possible to combine the sev¬
eral methods of obtaining nonlinear
resistors. For instance, voltage
padding may be used between a few
taps and then the characteristic be¬
tween the voltage pads may be
shaped by resistance padding. Re¬
sistance padding can only decrease
the volts per degree of the charac¬
teristic (i.e., make it closer to zero),
and therefore the unpadded resistor
Fig. 1.5. Voltage padding designed to
must be assigned a slope at least as
generate a sine function.
steep as the greatest slope of the
desired characteristic, and voltage padding must be used to establish
negative slopes.
Most manufacturers of precision variable resistors are willing to under¬
take the design and construction of nonlinear variable resistors to the
customer’s specifications.
1.7. Temperature Coefficient of Resistance. Most physical materials
exhibit some variation of resistance with temperature. On the sub-
microscopic level, resistance is presumably the interference with the flow
of electrons, or current, by the random motion of atomic particles of the
material. At absolute zero temperature there is no random motion of
these particles, and thus we would expect the material to have zero resist¬
ance there. This is borne out, in general, by experimentation, but as in
most simplifications there are exceptions. At very low temperatures the
curve of resistance does not approach zero in a straight line, and, most
particularly, there- are certain materials, such as carbon, whose resistance
increases with a decrease in temperature in the normal operating range.
8 CONTROL SYSTEM COMPONENTS [CHAP. 1

The empirical relation for the temperature variation is discussed in most

elementary texts on electricity1 and is given in Eq. (1.2),

Rh = Rc[ 1 + oc(th — tc)] (1*2)

where Rh = hot resistance

Re = cold resistance
th = hot temperature
tc = cold temperature
a = temperature coefficient of resistance
Tables for a for various materials are given in engineering handbooks.
As a result of the way in which a is defined, it is dependent not only on
the material, but also on the cold temperature at which it is computed.
Usually the tables list the a computed for a cold temperature of 20°C.
This value is usually adequate for most calculations. However, should
a very exact calculation be necessary for a cold temperature other than
20°C, a new a may be computed by setting up a linear ratio.
The temperature coefficient of resistance of the material used in making
precision variable resistors is a very important consideration. In the
lower resistance values a low-nickel-high-copper alloy is usually used.
It has a resistivity of 60 ohms/unit volume and a temperature coefficient
of 0.0071/°C. In the intermediate resistance values, Advance Alloy
(50-50 copper nickel), with a temperature coefficient of 0.00002 and a
resistivity of 294, is used. In the higher resistance values, Nichrome V
(nickel chrome), with an a of 0.00013 and a resistivity of 650, is normally
used. Karma wire, with a resistivity of 800 and an a of 0.00002, is some¬
times used in high-resistance elements, although its life and corrosion
characteristics are not quite so good as those of Nichrome V.2
1.8. Capacitors. Several types of capacitors are of commercial impor¬
tance. The air-dielectric variable capacitor has one set of plates mounted
on a shaft which may be turned with respect to the other set of plates.
The shaft is eccentric with the fixed set of plates, and as it is turned, the
effective spacing between the two sets of plates is varied, thus varying the
capacitance. The air-dielectric variable capacitor is more commonly used
in the tuning circuits of radios and electronic devices than in control
systems.
The tubular capacitor, or “paper’7 capacitor, is widely used in standard
electronic circuits. It consists of two thin conducting sheets of tin foil
or silver foil separated by a sheet of dielectric. Usually the dielectric is
impregnated paper. The sheets are then rolled into a compact tube,
using one additional sheet of dielectric to prevent shorting the conducting
1 See, for example, Cook and Carr, “Elements of Electrical Engineering,” John
Wiley & Sons, Inc., New York, 1951.
2 Data furnished by the Helipot Corporation, South Pasadena, Calif.
Sec. 1.8] ELECTRIC NETWORKS FOR CONTROL SYSTEMS 9

plates. The tubular capacitor is an inexpensive element, but it should

not be used in critical circuits. The dielectric tends to deteriorate with
age, and the capacitor is not stable with respect to temperature and
humidity. The tubular capacitor may exhibit an appreciable leakage
current in the larger sizes. Leakage current is the current that flows from
one plate to the other in a capacitor if the dielectric is not perfect. The
leakage path may be represented approximately as a resistor in parallel
with the capacitor.
Recently a new silicon dielectric material has been used in tubular
capacitors. This new dielectric results in a physically smaller capacitor
and reportedly improves to some extent the undesirable characteristics
mentioned above.
The oil-filled capacitor provides more capacity in the same physical
space than does the tubular capacitor but, because of the metal case
required, it is more expensive. For capacities of greater than 0.5 pi and
for rated voltages of greater than 100 volts the oil-filled capacitor is
generally used, since it is difficult to make paper or mica capacitors in
these large sizes.
The mica capacitor is more dependable and more expensive than the
tubular capacitor and is available in the same ratings. The construction
consists of alternate layers of metal foil and mica as a dielectric. Every
other metal plate is connected together, and the assembly is incased in
bakelite.
High capacities in a small space are available in the electrolytic capacitor.
However, the usual electrolytic capacitor is polarized, i.e., only voltages
of one polarity may be applied to it without damage. The electrolyte is
a solution of boric acid which acts as the negative plate. An aluminum
anode acts as the positive plate, while a thin film of aluminum oxide which
forms on the anode acts as the dielectric. In the so-called “dry ” electro¬
lytic capacitor, the electrolyte is damp paper soaked in boric acid or
ethylene glycol. The solution will ionize and break down if a reverse-
polarity voltage is applied to it. This makes the usual electrolytic capaci¬
tor useless for a-c-voltage applications. The polarized electrolytic
capacitor is used in d-c power-supply filters and in bypass applications in
electronic amplifiers.
It is possible to construct an electrolytic capacitor with two anode
plates immersed in a common electrolyte, for a-c service. This non¬
polarized capacitor is equivalent to two polarized electrolytic capacitors
connected in series, i.e., back to back, with the negative terminals com¬
mon. The anode area must be four times the area of a polarized capacitor
of the same capacity. The nonpolarized electrolytic capacitor is useful
in applications where the exact value of capacitance is not important, as
in phase-splitting service for single-phase, a-c induction motors.
 10 CONTROL SYSTEM COMPONENTS [Chap. 1

1.9. Inductors. Unlike resistors and capacitors, inductors are not

commonly mass produced to standard specifications. This is due to the
extremely wide range of parameters and ratings that are encountered.
Air-cored inductors are usually made only in sizes up to a few milli-
henrys, since they become very bulky for larger inductance values. For
this reason air-cored inductors are little used except in high-frequency
communication applications.
Iron-cored inductors are smaller and more compact than air-cored
devices and are available in inductance values of up to several hundred
henrys. In the past decade very-high-permeability alloys such as
Deltamax, permalloy, and Supermalloy have become available for use as
core material. Deltamax typically will support 90,000 maxwells/in.2
with about 0.4 amp-turn/in. at the knee of the magnetization curve.
Permalloy saturates at 39,000 maxwells/in.2 with about 0.2 amp-turn/in.,
and Supermalloy saturates at 42,000 maxwells/in.2 with 0.12 amp-turn/in.
From the intense interest in improved core materials for magnetic ampli¬
fiers further advances are to be expected in magnetic steels. Inductors
using these newer, sharply saturating materials must be designed to oper¬
ate below the knee of the curve. Grain-oriented core materials conven¬
tionally are sold in U-shaped laminations or in the form of rolls of “tape.”
In either case, bending, squeezing, or heating will damage the magnetic
properties. Powdered-iron toroids are also used extensively and result in
inductors having extremely low iron losses.
Although practical resistors and capacitors approach mathematical
perfection, a “pure” inductance cannot be constructed. Inductors are
subject to copper loss from the resistance of the wire with which they are
wound. Furthermore, the iron core results in iron losses due to hysteresis
and eddy currents. A very important parameter that must be considered
in choosing an inductor for a given application is, therefore, the quality
factor Q, defined as the ratio of power handled to power dissipated. It
may be shown1 that for inductances this is approximately equivalent to

0-3)

where co = frequency, radians/sec

L = inductance, henrys
R = equivalent resistance including the iron core loss, ohms
This definition of Q indicates properly that the performance of induc¬
tors is poor at low frequencies, but it does not show that performance is
also poor at high frequencies. As a result of interwiring capacitance,
coils resonate at some frequency, and the performance deteriorates
1 Ramo and Whinnery, “ Fields and Waves in Modern Radio,” John Wiley & Sons,
Inc., New York, 1953.
Sec. 1.10] ELECTRIC NETWORKS for control systems 11

rapidly at frequencies above this value. Modern toroidal coils carefully

made on high-quality cores of about 1 in. diameter will exhibit a peak Q of
about 200 in a frequency range of 2 to 5 kc. The Q falls rapidly above
and below the peak frequency. Typically the Q would be 100 at 10 kc
and 1 kc and would be less than 5 at 100 cps.
Since it is impossible to obtain a high-Q coil for operation below 100 cps,
coils are seldom deliberately used as circuit elements in d-c servos. Of
course, this statement does not apply to the inherent inductance of motor
windings and solenoids, etc. In a-c servos, where a 400-cps carrier is
often used, inductances are occasionally employed. Even here, however,
the size, weight, difficulty in shielding, low quality factor, expense, non¬
reproducibility in mass production, and nonlinearities argue against the
use of inductors.
1.10. The Analysis of Simple Networks. Networks are analyzed by
the application of Kirchhoff’s voltage and current laws. Since most net¬
works used in control systems have a com¬
mon ground connection and seldom con- X
tain mutual-inductance elements, the node e{
method of analysis1 normally results in sim¬
pler equations and yields the results with
Fig. 1.6. RC rate network.
less labor than the mesh method. The
result usually of primary interest in control-system applications is the net¬
work transfer function, i.e., the relation between input and output voltage.
Although it is not our purpose in this text to present an extensive dis¬
cussion of network analysis or of Laplace transform techniques, a simple
example will be worked out, partly by way of review, partly to establish
the system of nomenclature used in later chapters.2 The network con¬
sidered is the simple RC “rate network” shown in Fig. 1.6.
To analyze this network by means of the node method, we equate the
currents leaving the node e0 to zero. It is frequently assumed that any
device connected to the output terminals of the network does not “load”
the network; in other words, no current flows to the right from the node
e0. With this assumption we have,

cCir-f) + i = ° <'-4>
1 Gardner and Barnes, “Transients in Linear Systems,” John Wiley & Sons, Inc.,
New York, 1942, pp. 38-43.
2 Readers not familiar with the Laplace transform method for analyzing networks
are referred to Gardner and Barnes, ibid., and also to Chestnut and Mayer, “Servo¬
mechanisms and Regulating System Design,” John Wiley & Sons, Inc., New York,
1951, vol. 1, pp. 66-96. For a mathematical treatment of the Laplace transformation,
see Doetsch, “Theorie und Anwendung der Laplace-transformation,” Dover Pub¬
lications, New York, 1943,
12 CONTROL SYSTEM COMPONENTS [Chap. 1

The transfer function of the network is defined as the ratio of the

Laplace transform of the output voltage to that of the input voltage with
all initial conditions set to zero. Accordingly, we transform Eq. (1.4) to
obtain

C4EM - EMI + ^ EM = o (1.5)

We use the shorthand notation e = E(s); hence Eq. (1.5) may be rewritten
in the form

Cs(e0 -id+^o = 0 (1.6)

Solving for eQ, we obtain the transfer function

e0 _ RCs
(1.7)
% ~ RCs + 1

It may be shown1 that the quantity RC has the dimensions of time, It is,
therefore, referred to as a time constant and given the symbol T. With
this notation, Eq. (1.7) can be written in the equivalent form

K _ Ts
(1.8)
Ci Ts + 1

Equations (1.7) and (1.8) are in the time-constant form. This form is
identified by the fact that the constant term is unity. The time-constant
form is one of two standard forms in which transfer functions are com¬
monly expressed. The other common form, which may be referred to as
the root form, is given for the simple network discussed here by

= S = S (1 9)
Ad s + (1/22C) « + (I/?7) V '

This form is usually more convenient when the Laplace transform is to be

inverted to obtain the time function to which it corresponds.
The transfer function for simple networks can usually be obtained more
rapidly by using the principle of the voltage divider. This principle
states that in a series circuit the ratio of the voltage across any element or
group of elements to the total applied voltage is equal to the ratio of the

1 C is defined as charge/volt, and R is defined as volts/amp; thus, dimensionally,

volts charge charge

(R)(C) =
amp volt amp

and current is the rate of flow of charge, or charge/time; thus

charge
RC = time (sec)
charge/time (sec)
SEC. 1.11] ELECTRIC NETWORKS FOR CONTROL SYSTEMS 13

impedance of the element or elements to the total impedance of the cir¬

cuit. In order to apply this principle, use is made of the concept of
generalized impedance, defined as the ratio of the Laplace transform of
the voltage across an element to the Laplace transform of the current in
the element. Thus, the impedance of a capacitor is 1 /Cs, that of an
inductance is Ls, and that of a resistance is R. Using these ideas, we have
for our network
e0 _ R _ RCs
(1.10)
R T (l/Cs) RCs T 1
The transfer function is thus obtained in a single step.
1.11. Steady-state Frequency Response from the Transfer Function.
The Laplace transform method of analysis is perfectly general. There

Fig. 1.7. (a) Magnitude and (6) phase angle of the transfer function for the circuit
shown in Fig. 1.6.

have been no limitations placed on the voltage that is applied to the cir¬
cuit except that it exists only for positive time. If we are willing, how¬
ever, to restrict ourselves to sinusoidal voltages and to steady-state oper¬
ation, we may solve for the response of any circuit to any frequency given
merely the transfer function of the circuit. It is only necessary to sub¬
stitute yco for s (j = — 1) and then to let co be any value of frequency
in which we are interested.1 The magnitude of the resulting complex
number is the amplitude ratio of output to input, and the angle is the
phase difference between output and input. For the circuit considered
in Fig. 1.6 this information is available in Eq. (1.11). The complex num¬
bers in the numerator and denominator are converted to polar form and
divided. The resultant magnitude is shown in Fig. 1.7a, while the angle
between input and output is shown in Fig. 1.76.
1 See Chestnut and Mayer, op. tit., pp. 99-103.
14 CONTROL SYSTEM COMPONENTS [Chap. 1

e0* j»T CO r/90' CO T

/90° — tan-1 co T
Si 1 +juT Vl + (cor)Vtan-1 0>T \/l + (<*T)2
(l.H)
1.12. Asymptotic Representation of the Frequency Response. The
transfer functions of networks with lumped parameters such as R’s, L’s,
or C’s always consist of the ratio of products of simple factors such as s or
Ts + 1. In the simple example of Fig. 1.6 the transfer function consists
only of the simple ratio of two factors of this type, but, as will be shown
presently (Sec. 1.16), more complicated networks will in general contain
more of these factors in both the numerator and the denominator. For
this reason it is convenient to represent the amplitude characteristic given
in Fig. 1.7a on a log-log scale. This enables one to add or subtract the
characteristic curves of the typical factors to build up the total character¬
istic. Logarithmic representation also permits showing a wide range of
variables in a small space.
The logarithmic representation of a factor of the type Ts + 1 is simpli¬
fied by first considering its asymp¬
totic behavior for very small and
very large s = jco.
For coT <3C 1, we have

log |1 -F j<aT\ « log 1=0 (1.12)

while, for coT ^ 1,

Fig. 1.8. Asymptotic representation of log |1 + juTI ~ log a)T

log |l + jcoT|. = log co + log T (1.13)

Thus the asymptote for low frequencies is the log-co axis, while for high
frequencies the asymptote is a straight line with a positive slope of 1,
intersecting the log-co axis at the point co = 1/T. These two asymptotes
are shown in Fig. 1.8 as the solid lines. The actual characteristic
log |1 + jwT\ is shown dotted. Note that the actual curve does not
depart very far from the asymptotes at any point. For real T the maxi¬
mum departure occurs at the point where the two asymptotes meet.
This is the so-called “break point,” for which co = 1/T. At this point
the value of the actual characteristic is \/2; hence the departure of the
asymptotes from the true curve is of the order of 30 per cent. Since the
asymptotes represent the actual curve so closely, it is common practice to
draw only the asymptotes to represent the amplitude characteristic of a
transfer function. The resulting curve is the asymptotic amplitude
diagram.
* The bar indicates that ja> has been substituted for s.
Sec. 1.12] electric networks for control systems 15

For the simple network of Fig. 1.6 we obtain the asymptotic representa¬
tion by taking the logarithm of the magnitude of the frequency ratio
[Eq. (1.8) in which s = ju>]. We obtain

= log T + log co - log 11 + jwT\ (1.14)

Thus there are three components. The asymptotic characteristics for the
three components are shown in Fig. 1.9a, while the total characteristic for
the rate network of Fig. 1.6 is shown in Fig. 1.96. The dotted line again
represents the actual curve.

Fig. 1.9. Asymptotic representation of rate-network transfer function.

For more complicated functions the procedure is exactly the same,

except that there are more factors whose characteristics must be com¬
bined. Under these conditions a somewhat simpler technique for drawing
the asymptotic diagram is convenient. It is based on the following rules:
1. Start with a frequency low enough that, for all the factors of the
form 1 -f- Ts, |Ts| < 1. As far as the asymptotic representation is con¬
cerned, all of these factors then are equal to unity.
2. The low-frequency asymptote may now be drawn.
3. As the frequency is increased, the Ts product in some factors will
become greater than unity. If such a factor appears in the numerator,
the slope of the asymptotic diagram increases by 1, while if it is in the
denominator, the slope decreases by 1. If there is more than one factor
having the same T, the slope is increased (or decreased) by the number of
such factors appearing.
4. The time constants T need not always be real numbers. If they are
complex, they must occur in complex-conjugate pairs. As far as the
asymptotic diagram is concerned, a pair of factors with complex-con¬
jugate time constants is treated exactly as if they were a pair of factors
with equal real time constants; i.e., they result in a slope change of ±2.
However, since complex time constants indicate a resonance condition in
the transfer function, the actual curve may depart very markedly from
16 CONTROL SYSTEM COMPONENTS [Chap. 1

the asymptotic curve near the break point. The reader is referred to
texts on servomechanism theory for a complete discussion of this problem.1
The procedure is best illustrated by an example. Thus, consider the
transfer function

Ks(T2s -f* 1)
1 > Tl > T2 > T, > Ta > Tb
(Tis + l)(Tbs + 1 )(T4s + l)(Tbs -f- 1)
(1.15)

For s = jo) < 1/Ti, the asymptotic representation is identical to that of

the transfer function Ks. It has a +1 slope, and for s = 1 its level is K.
As the frequency increases to cc = 1/Ti, the first denominator term
becomes effective, and the slope changes to zero. When to = 1 /T2, the
slope changes to+1, etc. The com¬
plete curve is shown in Fig. 1.10.2
The magnitude of the transfer
function is sometimes given in
decibels (db). When used in con¬
nection with transfer functions, the
decibel value is usually defined as

'out
db = 20 log (1.16)

Fig. 1.10. Typical asymptotic diagram. The decibel expression is no more

convenient in use than any other
logarithmic ratio. Since the product of two numbers is equivalent to the
sum of their logarithms, addition is substituted for multiplication. The
use of the actual numerical ratios combined with the log-log plot will be
found more convenient than the decibel function in the' design of correc¬
tive networks. The use of the actual ratio makes it possible to read
required information directly from the plot without conversion, and fur¬
thermore the ambiguity in the definition of the decibel is avoided.
1.13. Common RC Networks. The five simple RC networks given
in Table 1.2 are very commonly used in servo systems, and for many of
the simpler systems no other networks are required. The table was con¬
structed by first drawing the physical network configuration and then
analyzing it, as shown in Sec. 1.10 above. In all cases the assumptions
are made that the driving source has zero output impedance and that the

1 James, Nichols, and Phillips, “Theory of Servomechanisms,” Radiation Labora¬

tory Series, vol. 25, McGraw-Hill Book Company, Inc., New York, 1947, pp. 144,
176. Also Chestnut and Mayer, op. cit., pp. 302-315.
2 The logarithmic plot may be made by plotting the logarithms of the values on
linear graph paper, or by plotting directly the values scaled logarithmically. The
latter procedure is universal; thus, in the remainder of the text we shall label the axes
with the functions rather’than their logarithms.
Sec. 1.14] electric networks for control systems 17

Table 1.2. Table of Commonly Used IlC Corrective Networks

Asymptotic amplitude
Physical Transfer
Names characteristic vs.
configuration function
frequency

S_o RCs cj = V/fc

! Rate net-
C Si RCs 4- 1 <j
-o work; high-
°—)h
\ pass filter
ei
O—

Reset net¬ So 1
R work; inte¬ 6i RCs + 1
O—vWA
gral con¬

i troller; low-
pass filter

Proportional R2C2S -f- 1

o-AW-
plus reset Ji (R1 + RidCiS + 1
>R? network;
lag network

Proportional 50 R2 RiCis 1
51 R1 + R2 / R1R2 \
jii *
plus rate
network;
lead net¬
\Ri + R2J
C is + 1

work

Proportional So (R2C28 -(- l)(/?iCis -f- 1)

■j£l
plus rate 6i l(Rl + R2)C2S + l] (g^j-Cu + l)
plus inte¬
R, (approximate; see Sec. 1.21)
gral con¬
troller ; lag-
c2
x lead net¬
work

network is not loaded. The transfer function for the last network given
is approximate and has been obtained by a technique explained in Sec.
. .
1 21
1.14. The Synthesis of RC Networks. Introduction. The problem
of synthesizing a physical network to meet a prescribed transfer function
is one that has engaged the attention of a long and distinguished list of
investigators. In general, the problem is complicated, and the methods
for solving it are correspondingly involved. Fortunately, a solution to
the general network-synthesis problem is not normally required in control
applications. One reason for this is that the transfer functions that are
encountered in control systems are usually quite simple, and ordinarily
18 CONTROL SYSTEM COMPONENTS [Chap. 1

they need not be obtained with great precision. Hence, approximate

synthesis methods are in almost all cases completely adequate; in fact, one
might say that exact synthesis procedures are rarely justified.
A second important reason for the simplification of the synthesis prob¬
lem is that, as a result of the disadvantages listed in Sec. 1.9, inductances
are very rarely used in networks designed for control applications. The
restriction of the problem to that of RC network synthesis moves a large
number of transfer functions that are theoretically realizable by an unre¬
stricted network into the region of unrealizable functions. It therefore
reduces the scope of the synthesis problem considerably. This limitation
on the choice of permissible transfer functions is not, however, a very
serious disadvantage, since almost all the transfer functions normally
required in control systems are found to be of the type realizable with RC
networks. It is usually found that the few transfer functions that cannot
be synthesized by RC networks alone can be synthesized approximately
by RC networks together with amplifying circuits.
1.15. Properties of RC Networks. Before the synthesis of networks
is considered, a few remarks concerning some of the properties of RC net¬
works and the physical realizability of transfer functions are in order.

io) Two-terminal network {/?) Three-terminal network

Input Output

(c) Four-terminol network

Fig. 1.11. Network types.

The subject of network analysis and synthesis is much too large and com¬
plex to be covered adequately in a few paragraphs, and the following
statements can serve at best as only a brief summary. For more exten¬
sive treatments the reader is referred to standard works1 on the subject.
Networks may be classified according to the number of terminals by
which they are connected to other circuits. Although a network may
have as many terminals as it has nodal points, most of the networks in
common use are two-, three-, and four-terminal networks, as shown in
Fig. 1.11. Note that a three-terminal network is a special type of four-
1 Bode, “Network Analysis and Feedback Amplifier Design,” D. Van Nostrand
Company, Inc., Princeton, N.J., 1945. Guillemin, “Communications Networks,”
John Wiley & Sons, Inc., New York, 1948, vol. 2.
 Sec. 1.16] electric networks for control systems 19

terminal network in which both input and output have a common ground
connection. Networks with more than one set of input terminals are
occasionally encountered in control systems, particularly in places where
a number of signals must be added or subtracted. This type of network
will not, however, be discussed in any detail in this chapter. We shall be
concerned primarily with three-terminal networks, since this is the type
most commonly used in control systems.
A three- or four-terminal network may contain a large number of com¬
ponents, but it is usually possible and convenient to think of it as being
made up of a relatively small number of two-terminal networks. For
example, the ladder network shown
in Fig. 1.12 is made up of the two-
terminal networks Zi, Z2, Z3, Z4, etc.,
each one of which may contain a
number of resistors and capacitors.
The properties of the larger structure
are related to those of the two-termi¬ Fig. 1.12. Ladder network.
nal networks of which it is composed,
and it is, therefore, appropriate to consider first the properties of two-
terminal networks.
1.16. Properties of Two-terminal RC Networks. A two-terminal net¬
work is characterized by a single function, its driving-'point impedance,
defined as the ratio of the voltage applied to the two terminals to the
resulting current. In common with all other network functions, the
driving-point impedance is a function of frequency or, more generally, of
the complex variable s. The most important property of a driving-point
impedance of any passive network made up of lumped parameters such as
resistors, capacitors, or inductances is that it can be expressed as the ratio
of two polynomials in s as follows:
ansn + an^is"-1 + + no
z= (1.17)
+ 6m_ is™-1 + +
This may be demonstrated by writing all the Kirchhoff mesh equations
for the network and solving for the current as a function of the voltage at
any two terminals.1 Equation (1.17) may be written in the equivalent
factored form:
/
y _ K {s ~ ai)(s — a2) • ’ ' (s — an)
(1.18)
(« - 0l)(« -02)•••(«- Pm)
In accordance with standard convention the a s and p’s are referred to as
the zeros and poles, respectively, of the network function.
Guillemin2 has shown that the poles and zeros of RC driving-point
1 Guillemin, ibid., p. 208.
2 Ibid., p. 212.
20 CONTROL SYSTEM COMPONENTS [Chap. 1

impedances must be real and negative and that they must obey the
separation principle as follows:

0 > dl > Oil > 02 > > • • • (1.19)'

This implies that the zeros and poles must be simple; i.e., factors of the
form (s — a)n cannot occur in the transfer function unless n = 1. It fol¬
lows that an asymptotic representation of an RC driving-point impedance
must take the form shown in Fig. 1.13 and that the actual driving-point
impedance must satisfy the relation

0 > dj!?g > -1 (1-20)

- d(log w) -

Bode1 has shown by means of an argu¬

ment involving the energy stored or
dissipated in a network that the phase
angle of an RC driving-point imped¬
ance is restricted to negative values
Fig. 1.13. Asymptotic diagram of an
between 0 and — 90°. This result
RC driving-point impedance.
would also have been suggested by the
well-known relationship between the slope of the asymptotic diagram and
the phase angle.
Further implications of Eqs. (1.19) and (1.20) are that the number of
poles may be equal to the number of zeros or may exceed the number of
zeros by 1. In the former case the driving-point impedance approaches
the impedance of a resistance at high frequencies, while in the latter case
it approaches that of a capacitance. A driving-point impedance may be
infinite or finite, but not zero, at zero frequency, and it may be finite or
zero, but not infinite, at infinite frequency.
All of these properties of EC driving-point impedances may be sum¬
marized by saying that a network made up of resistors and capacitors will
in general have an impedance function that is at all times somewhere
between the impedance function of a pure resistance and that of a pure
capacitance. The extremes are in general approached only asymptoti¬
cally. Stated in this way, the conclusions concerning the properties of
RC driving-point impedances become almost obvious.
1.17. Transfer Functions Realizable with Four-terminal RC Networks.
Passing now to a consideration of three- and four-terminal networks, we
find that such networks are characterized by three network functions,
such as, for example, the driving-point impedances at the input and out¬
put terminals and the transfer function between input and output volt¬
ages. Since our chief interest in this section is the transfer function, we
investigate the limitations on the transfer functions realizable by RC
1 Bode, op. cit., par. 7.-11.
Sec. 1.17] electric networks for control systems 21

networks. In this discussion it is assumed that the driving source sup¬

plying the input to the network has zero output impedance and that a
load of infinite impedance is connected across the output terminals.
The restrictions on transfer functions obtainable by RC networks are
perhaps most simply obtained by noting that any transfer function that is
realizable by an RC network may be realized in the form of a symmetrical
lattice,1 or bridge, as shown in Fig.
1.14. In order to find the output
voltage, we observe that

en = e<2, ci (1.21)
ZB
However ei = (1.22)
Za + Zb
Za
and e2 =
Za + Zb
A Fig. 1.14. Symmetrical bridge, or lattice,
e_o . Za Zb
Hence A
(1.23) network.
e: Za + Zb

The impedances ZA, ZB, and ZA + ZB are driving-point impedances and

must, therefore, obey the rules governing driving-point impedances deter¬
mined above. The poles of ZA + ZB and the poles of ZA — ZB are identi¬
cal and cancel out of Eq. (1.23); hence the nature of the transfer function
is determined completely by the zeros of ZA + Zb and ZA — ZB. It fol¬
lows immediately that the transfer function must be expressible as the
ratio of two polynomials in s. Since the poles of the transfer function are
the zeros of the driving-point impedance ZA + ZB, they must be negative,
real, and simple. The zeros of the transfer function, on the other hand,
are not similarly restricted and may in general lie anywhere in the complex
plane. If the zeros are complex, they must, however, occur in complex-
conjugate pairs. This is required, since otherwise the coefficients of the
numerator polynomial would be complex numbers, a physical impossi¬
bility since these coefficients are combinations of real R’s and C’s. The
transfer function cannot have a pole at the origin, since the gain for zero
frequency in that case would become infinite. This property may also
be deduced from the fact that a driving-point impedance cannot have
a zero at the origin [Eq. (1.19)]. Finally the degree of the numerator can¬
not exceed that of the denominator, since otherwise the gain of the net¬
work would become infinite at infinite frequencies. This restriction may
again be deduced also from the number of zeros of ZA — ZB and of
Za + Zb•
Before leaving this discussion, we note that, if inductances as well as
resistances and capacitances could be used, the only change in the restric-
1 J. L. Bower and P. F. Ordung, The Synthesis of Resistor-Capacitor Networks,
Proc. IRE, vol. 38, no. 3, pp. 263-269, March, 1950.
22 CONTROL SYSTEM COMPONENTS [Chap. 1

tions on the form of the permissible transfer functions would be in the

location of the poles. For a network composed of R’s, L’s, and C’s, the
poles may be complex-conjugate and are otherwise restricted only to the
left half plane, the latter restriction arising from the fact that the network
would otherwise be unstable. Thus, as long as transfer functions having
complex poles are not required, RC networks are completely adequate to
meet the demand for any realizable transfer function.
1.18. L Sections. Although any transfer function that is realizable
by an RC network can be synthesized by a lattice structure, the network
most commonly used in control applications is the ladder network shown
in Fig. 1.12. This type of network is relatively easy to design, particu¬
larly by approximate methods, and despite the fact that it is considerably
less versatile than the lattice, it usually meets the requirements of the
designer of control systems.
In the following paragraphs we consider an approximate synthesis
method based on cascading L-type networks to form a ladder structure.
Exact methods of ladder-network synthesis exist but usually result in
networks having a considerably lower gain than the approximate design.
For details on the design of exact ladder networks and also the design of
lattice networks, the reader is referred to the literature.* 1 2 3 4 5
The form of the L section is shown in Fig. 1.15. It is a simple voltage
divider, and its transfer function is
Zt given by

Zo ? = H = (1.24)
6i
e* Zi +
We make the standard assumptions
Fig. 1.15 L-section type of RC
that the source impedance of the input
network. voltage source is zero and that the load
impedance is infinite at all frequencies.
The properties of the network are then easily deduced from the properties
of driving-point impedances derived previously. The restrictions on the
transfer functions obtainable with L sections may be stated as follows:
1. The poles and zeros must be negative and real.
2. The poles must be simple.
3. There can be no more than two successive zeros (or poles) without an
intervening pole (or zero).
4. The number of zeros cannot exceed the number of poles.
5. There cannot be a pole at the origin.
1 J. T. Fleck and P. F. Ordung, Realization of a Transfer Ratio by Means of an
R-C Ladder Network, Proc. IRE, vol. 39, pp. 1069-1074, 1951. Bower and Ordung,
op. cit.
Sec. 1.18] electric networks for control systems 23

6. The maximum rate of increase or decrease of the magnitude of the

transfer function cannot exceed the first power of frequency; i.e., the
transfer function H(s) must satisfy the inequality

^ d{log \H\) ^
(1.25)
~ d(logco) -

7. The magnitude of the transfer function cannot exceed unity at any

frequency, i.e.,
HI < 1 (1.26)

Equation (1.24) indicates that the poles and zeros of the transfer func¬
tion are produced by the poles or zeros of Z2 or Zx + Z2. Since these are
both driving-point impedances, the poles and zeros must be real and
negative, as noted in point 1 of the list above. Poles of the transfer func¬
tion result only from zeros of Zi + Z2, since a pole in Z2 is also a pole of
Z\ + Z2 and will therefore cancel out. Since zeros of driving-point
impedances must be simple, restriction 2 follows. Note, however, that
zeros may occur in pairs, because they are produced by either poles of
Zi + Z2 or zeros of Z2. Restriction 3 follows from the separation prop¬
erty of the poles and zeros of driving-point impedances. Thus suppose
that H has a pole due to a zero of Zi + Z2. The next singularity of
Zi + Z2 must be a pole; however, if this is a pole of Z2, no singularity in
H results. Zi + Z2 may now again have a zero, resulting in two succes¬
sive poles in H. The next singularity in H must now, however, be caused
by either a pole in Zi or a zero in Z2. Either of these singularities results
in a zero in H. Thus there can be no more than two successive poles. A
similar argument may be used to show that no more than two successive
zeros may occur. Restrictions 4 and 5 are restrictions on RC-network
transfer functions in general and need not be discussed again. Restric¬
tion 6 is related to inequality (1.20). The maximum rate of decrease of
the transfer function occurs when the numerator decreases at its maximum
rate and when the denominator is constant; the maximum rate of increase
occurs when the numerator is constant and when the denominator
increases at its maximum rate. Thus restriction 6 follows. Restriction
7 is almost obvious, since the network contains no amplifier, but it may
be shown more rigorously to be a consequence of the limitation on the
phase angles of Zx and Z2. Thus, suppose Eq. (1.24) to be rewritten as
fdllows:
Z2 1
H = (1.27)
Zi T Z2 (Zi/Z2) + 1

Since the phase angles of Z2 and Zi are restricted to the fourth quadrant,
24 CONTROL SYSTEM COMPONENTS [Chap. 1

we may write
-90° < < 90° (1.28)

Hence § + i>i
■^2
(1.29)

and restriction 7 is proved.

1.19. Cascading of L Sections. Restrictions 2, 3, and 6 may be over¬
come by cascading several L sections to form a ladder network of the form
shown in Fig. 1.12. If the input
impedance of the second L section is
'l sufficiently high relative to the out¬
put impedance of the first section, so
that the second section presents
Fig. 1.16. Cascaded networks.
negligible load on the first, then the
transfer function through both sections is approximately equal to the
product of the transfer functions of the individual sections. This is
easily demonstrated.
Suppose a network having a transfer function Hx and an output imped¬
ance Zox drives a network having a transfer function Hy and an input
impedance Ziy (see Fig. 1.16). When the second network is not con¬
nected, the output of the first network, ex, is given by

€\ = e%Hx (1.30)

With the second network connected, however, the voltage becomes

Z,iy
ex = e,H2 Z OX “f" Ziy
(1.31)

Therefore, the output voltage from the second network is given by

7
TT TT
(1.32)
II

i
T““l

x11 y nr ry
^ ox ”1 Aiy
A

to HxHy
or A (1.33)
ei 1 + ( Z0X/ Ziy )
It is clear that, if
Zox
«1 (1.34)
^7iy
e0
then A - HXHy (1.35)
ei
The maximum permissible value of \ZoxIZiy\ depends on the accuracy
desired, but a commonly accepted maximum is of the order of 0.1. Nor¬
mally only a very rough estimate of \Z0X/Ziy\ is required. If more than
two L sections are to be cascaded, the above argument is easily extended,
and inequality (1.34) must be met at each network junction.
Sec. 1.20] electric networks for control systems 25

In computing the input and output impedances of an L section, it is

convenient to note that the input impedance of network Hy is given by

Ziy = Zlv + Z2y = (1.36)

riy

while the output impedance of network Hx is given by

Zox = = Z1XH* (1.37)

A lx “t- zj 2x

Hence ~ = Zl^Ik (1.38)

When three or four L sections must be cascaded, the application of

inequality (1.34) may require unrealistic component values, since the
inequality requires the impedance level of each succeeding section to be
several times higher than the preceding one. Impedance levels larger
than 1 megohm are usually undesirable, with their probability of excessive
noise pickup. Hence isolating amplifiers or cathode followers (see Sec.
2.5) are quite frequently used to prevent loading between sections in the
more extensive networks.
If the number of cascaded L sections is N, then restrictions 2, 3, and 6
of the list in Sec. 1.18 are amended as follows:
2a. The poles of the transfer function may be approximately iV-fold.
Note that the requirement that the poles must be simple was made in
general for all RC networks. Hence, even with N cascaded sections it is
not possible to have N exactly coinciding poles; however by making
Zox/Ziy sufficiently small, a number of poles may be brought as close
together as is desired.
3a. There may be 2N successive zeros (or poles) without an intervening
pole (or zero).
6a. The maximum rate of increase or decrease of the magnitude of the
transfer function cannot exceed the vVth power of frequency:

-N < < N (1.39)

- d(log a) ~

1.20. Approximate Ladder-network Synthesis. In synthesizing a net¬

work to obtain a prescribed transfer function, it is first necessary to ascer¬
tain that the desired transfer function meets all the requirements listed
above. When this has been done, the number of L sections required is
determined by examination of the amended restrictions 2a, 3a, or 6a.
Thus, for instance, if a transfer function having an asymptotic gain dia¬
gram with a —3 slope is desired, at least three L sections would be needed.
The apportionment of poles and zeros among the various L sections is to
some extent arbitrary, and any order of cascading the parts is usually
26 CONTROL SYSTEM COMPONENTS [Chap. 1

permissible as long as inequality (1.34) is properly satisfied. The specific

design chosen from this range of possibilities should be that which mini¬
mizes the size of the larger capacitors and makes inequality (1.34) easy to
meet. In order to accomplish the former, it is usually best to place the
sections having zeros and poles of smallest magnitudes in the last position
of the cascade.
Examples will be worked out in a later section to provide further clarifi¬
cation of the design procedure. Before a design can be completed, how¬
ever, a method must be available for the synthesis of the individual L sec¬
tion. We present here three methods, one approximate and the others
exact.
1.21. Approximate Synthesis of L Sections. The following approxi¬
mate synthesis method is based on the idea that a capacitance connected
in a circuit with resistances acts as an open circuit at very low frequencies
and as a short circuit at very high frequencies. For each capacitor in the
circuit, therefore, there exists only a relatively small frequency range for
which its impedance is comparable to that of the resistances surrounding
it. In this frequency range we consider the capacitor as being active.
For other frequencies, where the capacitor represents either an open cir¬
cuit or a short circuit relative to the resistances connected to it, we con¬
sider it inactive, since it has negligible effect on the circuit.
The synthesis method is approximate rather than exact because it
assumes that the region of activity of a given capacitor is sharply bounded.
Under this assumption a capacitor is either active or it is inactive, and no
intermediate conditions are possible. This implies that the change from
activity to inactivity takes place at distinct boundary points on the fre¬
quency scale. In order to locate these boundary points, suppose first
that a driving-point impedance contains, at a particular frequency, no
active capacitors. The impedance is then made up only of resistances,
and the slope of the asymptotic impedance diagram (as in Fig. 1.13) must
be zero. When a capacitor becomes active, it changes from an open cir¬
cuit to a short circuit as the frequency increases. The presence of an
active capacitor, therefore, causes a reduction in impedance with fre¬
quency as shown by the segments of —1 slope in Fig. 1.13. It is most
natural to take the break points of the asymptotic impedance diagram as
the boundary points defining the activity of a capacitor. In accordance
with the usual convention, the frequencies of the break points are equal
in magnitude to the poles and zeros of the impedance. Hence, if it is
assumed that there is never more than one active capacitor in an imped¬
ance, we may think of this capacitor as being responsible for the pole and
the zero defining its zone of activity.
A further important point in the synthesis procedure is that there must
never be more, than one active capacitor in the entire L section being
Sec. 1.21] electric networks for control systems 27

synthesized. Although this is not necessarily true for L sections in

general, we are certainly free to design an L section so that this stipulation
is met. We permit the limiting case where one capacitor becomes active
at the same point at which another one becomes inactive, since otherwise
the synthesis of transfer functions having two coincident zeros would not
be possible.
Consider now the general two-terminal impedance Z, containing only
one capacitor (see Fig. 1.17). At frequencies low enough for the capacitor
to be inactive, the impedance is effectively a resistance R. At intermedi¬
ate frequencies the capacitor becomes active and the impedance decreases,
and at high frequencies the capacitor is again inactive. Hence over the
entire frequency range the impedance may
be expressed by
TlS+ 1
Z = R (1.40)
T 2s + 1 Fig. 1.17. Impedance con¬
where T2 > Tx taining one capacitor.

If a voltage source E is connected across the impedance, the resulting cur¬

rent is
j _ E _ E(T2s + 1) (1.41)
Z " R(Tis + 1)
Rearranging gives
R(TlS + 1)7 = (T2s + 1 )E (1.42)

Equation (1.42) may be thought of as the transform of the differential

equation

RTi ^ + RI = T2 d-^~ + E (1.43)

at at
Suppose now that the impedance is open-circuited, Then the current 7
is zero, and Eq. (1.43) becomes

dE
T 2 ~77 ~b E — 0 (1.44)
dt
Assuming that the voltage has some initial value Eq, the solution for E as
a function of time is
E = E0e-t/T> (1.45)

Similarly, we may solve for the current that flows when the two terminals
of the impedance are short-circuited together. Under this condition
E = 0, and Eq. (1.43) becomes

Tl d4 + I = 0 (1.46)
dt
or I = he-t/T i
(1.47)
28 CONTltOL SYSTEM COMPONENTS [Chap. 1

From Eq. (1.40) we see that l/7\ and 1 /772 are, respectively, the zero and
the pole of Z. Thus we arrive at the following conclusion:
When a driving-point impedance contains only one capacitor, it has
only one pole and one zero. The pole is the reciprocal of the discharge
time constant obtained on the open-circuited impedance, and the zero is
the reciprocal of the discharge time constant of the impedance when it is
short-circuited.
*2
-WVW-
if This rule enables us to write down
the driving-point-impedance func-
*1
o—AW/ -° tions by inspection of the circuit.
*3
Thus, consider the impedance shown
WWW
in Fig. 1.18. At low frequencies the
Fig. 1.18. RC impedance. capacitor is an open circuit; hence
the resistance is i?i + Rz. The pole is the reciprocal of (j?2 + Rz)C.
The zero is the reciprocal of [R1Rs/(Ri -f Rs) + R^C. Hence the imped¬
ance function is

Z = (7?i + /?3)
[R 2 + (RiR*)/(Ri + R-i)]Cs + 1
(1.48)
(R2 Rd)Cs + 1

In an L-section network like that shown in Fig. 1.15, a pole in the trans¬
fer function is produced only by a zero of Z\ + Z2 that is not simultane¬
ously a zero of Z2. Hence for a network containing only one active
capacitor a pole in the transfer function is always the reciprocal of a dis¬
charge time constant of the entire short-circuited L section. Zeros in the
transfer function are produced by zeros of Z2 or poles of Zi, since a pole of
Zi corresponds to a pole of Zi + Z2. Thus zeros are given either by the
discharge time constant of Z2 short-circuited, or of Zi open-circuited.
To synthesize an L-section network, we first draw up the network
qualitatively according to the following rules:
1. If the asymptotic representation of the transfer function has a zero
slope, the network has no active capacitors in it.
2. If the asymptotic representation of the transfer function has a +1
slope, a capacitor in Z\ is active, and if it has a — 1 slope, a capacitor in Z2
is active.
3. A zero in the transfer function at zero frequency requires a series
capacitor, not bridged by any resistance in Zx. A zero at infinite fre¬
quency requires a shunt capacitor across Z2.
The procedure is best illustrated by means of a few typical examples.
Consider the transfer function whose asymptotic diagram is shown in
Fig. 1.19. The network must contain two capacitors, one active in the
interval coi ^ co ^ co2 and the other active m the interval co3 < CO < C04.
We consider these two intervals separately. A network to synthesize the
transfer function to the left of co3 is shown in Fig. 1.20a. Note that at low
Sec. 1.21] electric networks for control systems 29

frequencies the capacitor acts as an open circuit; hence the network gain
is unity. At intermediate frequencies the capacitor becomes active and
reduces the gain, and at high frequencies the capacitor is short-circuited,
and the network gain remains con¬
stant at a reduced level. For fre¬
quencies above w3 the network gain
must increase. By rule 2 this re¬
quires a capacitor in Zi. If the net¬
work gain is to be less than unity
at high frequencies, however, the
Fig. 1.19. Example of transfer function.
capacitor must short out only part #
of R\. Thus the complete network shown in Fig. 1.206 results.
To complete the design quantitatively, we note that the break at cox
represents a pole in the transfer function of magnitude coi. This pole is
caused by a zero of the input driving-point impedance of the circuit of

o-AA/VA wvw-
*1 Rz

O-

{a) First step in synthesis [b) Complete network

Fig. 1.20. Synthesis of the transfer function of Fig. 1.19.

Fig. 1.20a and is therefore the reciprocal of the discharge time constant
of that circuit when the input is short-circuited. Thus

— = (Ri + Ri + Rz)C\ (1.49)

COi

The break at o>2 represents a zero of Z2 which is equal to the reciprocal of

the discharge time constant of Z2 when its two ends are short-circuited;
i.e.,

— = RzCi (1.50)
co2

The next zero, corresponding to co3, is due to a pole of Z\. It is, therefore,
the reciprocal of the discharge time constant of Zi when it is open-
circuited;

— = R2Cz (1.51)

Finally, l/o>4 must be equal to the discharge time constant of the entire
30 CONTROL SYSTEM COMPONENTS [Chap. 1

network with the input terminals connected together and with C1 replaced
by a short circuit.
1 _ C^R^iRi Rz)
604 R1 -f- R2 T R3
Although other equations may be written about the circuit of Fig. 1.206,
it will be found that they are not independent from the four given above.
Thus we have only four independent equations. There are, however, five
unknown parameters in the circuit: two capacitors and three resistors.
This situation is encountered in all methods of transfer-function synthesis,
since the impedance level of the net¬
work is not specified by the process.
The value of any one component of it
the network may therefore be set :/?i
arbitrarily, depending on the require¬
ments of the remainder of the circuit.
ia)
Thus, suppose the circuit to be driv¬
ing the grid of a vacuum tube. Then
the maximum resistance between
grid and ground might be specified as Rz .Cs
o-MW-
less than 1 megohm, so that we would
have the additional requirement that
Rz + R\ < 106 ohms. If then we
arbitrarily let R\ -f- R2 = }/\2 megohm,

R. 'T''Co

ic)
Fig. 1.21. Network transfer function. Fig. 1.22. Steps in the design (see text).

we have five equations, which may be solved to find the values of the
five parameters. Since most of the equations are, in general, quite
simple, like Eqs. (1.50) and (1.51), it is usually possible by a judicious
choice of the arbitrary parameter to simplify the solution very consider¬
ably. Thus, in the foregoing example, if we assign a value to Ci, Eq.
(1.50) may be solved immediately for R3, and Eq. (1.49) for Ri + R2 + R3.
Then division of Eqs. (1.51) and (1.52) yields an equation having only R1
as the unknown, so that all the resistances may be obtained easily. C2
is then obtained from Eq. (1.51).
As a second example, consider the transfer function shown in Fig. 1.21
which has a zero both for zero and for infinite frequencies. The +1 slope
extending down to zero frequency requires a series capacitor in Z\ by rule
Sec. 1.21] electric networks for control systems 31

3. Hence the transfer function to the left of co2 is synthesized by the net¬
work shown in Fig. 1.22a. At the frequency a>i the capacitor Ci becomes
inactive. In order to obtain the — 1 slope between co2 and o>3, a capacitor
in Z2 must become active in this interval; however, unless there is a series
resistance in Z i, such a capacitor cannot have any effect on the transfer
function. Thus the network must take the form shown in Fig. 1.225.
The final break at oh must be produced by another capacitor in Z2, so that
the network of Fig. 1.22c follows. Note that a capacitance is shunted
across Z2 as required by rule 3.
To evaluate the R’s and C’s, we proceed as in the previous example and
obtain the following equations:

— — (R2 + Rs 4 Ri)Ci- (C2 and C3 are open circuits) (1.53)

OH

1 _ C2Ra(Rz 4 R2) “ (Ci is short-circuited and

(1.54)
co2 R4 R2 T" R3 C3 is open-circuited)
(Ci is short-circuited and
— = r,o 2 (1.55)
0)3 C3 is open-circuited)
1 _ C3R2R3 (Ci and C2 are short-
(1.56)
C04 R<i R3 circuited)

We have four equations but six unknowns. One more equation may be
obtained, as in the last example, by assigning an arbitrary value to one of
the components, but this still leaves one additional equation to be deter¬
mined. This additional equation is required to specify the gain level of
the network. Thus, suppose that the gain of the transfer function
between the frequencies 001 and w2 is set at g, where g < 1. Then since Ci
is assumed short-circuited and C2 and C3 open-circuited in this frequency
interval, we have
_ Rz 4 Ra
~

(1-57)
R2 4 Rz 4 Ra
~ ~

Thus we have the necessary six equations, which may now be solved for
the six components.
It should be noted that the question of gain level did not appear in the
first example because the level was tacitly assumed to be unity at fre¬
quencies below coi. Had a level of less than unity been desired, then an
additional resistor (probably across C\ in Fig. 1.205) would have been
placed in the circuit. An additional equation would then have been
needed to specify this resistance.
Normally one is interested in as high a gain level as possible, since low
gain must be made up with amplifying circuits. The question naturally
arises, therefore, as to how large g may be made. For a transfer function
of the form shown in Fig. 1.19 the answer is obvious: the maximum gain
is unity. However, for network functions of the type shown in Fig. 1.21,
32 CONTROL SYSTEM COMPONENTS [Chap. 1

the answer is not obvious, and it cannot be obtained from the approximate
synthesis procedure described here. All that can be said at this point is
that g < 1, since, by Eq. (1.57), Ro = 0 if g = 1.
The essential difference between the transfer functions of Figs. 1.19 and
1.21 is that the latter contains two successive poles, while the former does
not. It will be shown in connection with the exact synthesis method to
be described subsequently (Sec.
1.23) that this question of gain level
arises generally only when two suc¬
cessive poles are specified for a
transfer function. An approximate
idea of the maximum value of g per¬
Fig. 1.23. Transfer function with two missible in such cases may, there¬
successive poles.
fore, be obtained by considering a
transfer function having only two poles. The asymptotic diagram of
such a transfer function must have the form shown in Fig. 1.23, and an
analytic expression of H is
__ _ Zo _gcc2s_
H = (1.58)
Zi Z2 (s + o?i)(s -f- C02)
This may be rewritten as
Z\ + Z 2 _ 1 _|_ Z\ _ (s + COi)(s + 0)2)
(1.59)
Z2 Z2 geo 2S
Z1 S2 -T (cO 1 + CO2)s “b COiO>2 — ge02S
so that (1.60)
Z2 ge^2S
The roots of the numerator must be real and negative. By a short
algebraic manipulation it may be shown that this requirement leads to

9 < 1 -
S
A plot of this function is shown in
(1.61)

Fig. 1.24. Note that for co2 = wi

the gain is zero; this means that a
network having two coincident poles
cannot be synthesized. This was,
of course, proved previously by use
of a different argument. However,
even if the poles are separated by a
factor of 10, the maximum gain is
0 * 0 - \fe)
only 0.45. This indicates that, un¬ Fig. 1.24. Maximum gain for network
having transfer function of Fig. 1.23.
less they are very far apart, succes-
sive poles should be avoided as much as possible. When a number of L
sections must be cascaded for other reasons, it is therefore often best
Sec. 1.22] electric networks for control systems 33

to locate successive poles in different L sections. Finally, it should

be noted that no difficulty is experienced in synthesizing a transfer
function of the form shown in Fig. 1.25 by a single L section. Here the
gain in the region having two poles
is assumed to be considerably below
the limiting value given by Eq.
(1.61), and the approximate method
should give good results.
Of considerable interest in con¬
nection with any approximate
method is the question of how good
Fig. 1.25. Transfer function easily realiz¬
the approximation is. While in the able by single L section.
network-synthesis problem no gen¬
eral answer applicable in all cases can be given, a number of points can be
made as follows:
1. If the network contains only one capacitor, the approximate and
exact methods give identical results.
2. Any pole or zero of the transfer function that is produced by a pole
or zero of an impedance having only one capacitor is given exactly by the
approximate method. Thus in the network of Fig. 1.206 the zeros o>2 and
a>3 are exact, since the first is due to a zero of Z2 and the second is due to a
pole of Zi, and both of these impedances have only one capacitor.
3. Any zero of the transfer function produced by a pole of Zi is exact.
This is due to the fact that the approximate method is inaccurate only in
locating the zeros of driving-point impedances.
4. In general, the approximate method gains in accuracy as the zones
of activity of the various capacitors contained in the impedances are
moved farther apart on the frequency scale. It is clear that under this
condition the approximation that a capacitor is either open- or short-
circuited outside its active zone becomes better and better.
1.22. An Exact Synthesis Method of L Sections for Simple Transfer
Functions. For the relatively simple network functions that are used in
the great majority of control systems, the approximate synthesis method
discussed in the previous section is quite easily made exact. The pro¬
cedure is best illustrated by an example. Thus consider the transfer func¬
tion diagramed in Fig. 1.19 above. By means of the qualitative reasoning
discussed in connection with this figure, we obtain as before the general
network shown in Fig. 1.206. The exact transfer function of this network
is obtained by treating it as a voltage divider:

_R 3 + (1/6*1$)_
R 3 + (1/Cis) + /2i + [Rz/iRzCzs + 1)]
_(RzCis T 1)(R2C2S -j- 1)_
(R\ + RzjRzC2Cis2 -f- (R1C1 + R3C1 -f* R2C1 T I£2(72)s 4- 1
34 CONTROL SYSTEM COMPONENTS [Chap. 1

From Fig. 1.19 the transfer function is given by

[(s/c02) + l][(s/w3) + 1]
H = (1.63)
(s2/CO1CO4) + [(1/COi) 4" (l/c04)]s + 1
By direct comparison we obtain

R2C2 = —
co3
(1.64)
flsCl = -
C02
Substituting these values in Eq. (1.62) and equating Eqs. (1.62) and
(1.63), we obtain

1 1 1 / 1
CO1C04
+ — H-Js + 1 = — ( — + R\C\ ] s2
CO 1 CO4/ CO3 \co2

+ 1 R\C\ + R2C?! + -+-) s + 1 (1.65)

C02 CO3/

By equating the coefficients of the s2 term, we obtain

R&i = (1.66)
CO16O4 C02

Substituting this value for R\C\ in Eq. (1.65) and equating the coefficients
of s, we obtain finally

= —

CO 1
+ --—

CO4 CO1CO4
- —

CO3
(1.67)

Equations (1.64), (1.66), and (1.67), along with the specification of one of
the R’s or C’s to provide the desired impedance level for the network,
constitute a set of five equations, from which the five network parameters
are easily obtained.
This method may be used for other simple networks having only a few
poles and zeros. As the networks become more complicated, considerable
ingenuity may have to be exercised in particular cases to make the sub¬
stitutions indicated above. Also the method does not optimize the gain
level for transfer functions having two successive poles, like that shown
in Fig. 1.21. Such optimization would have to be a trial-and-error proc¬
ess with larger and larger values of gain assumed until an unrealizable
situation (a negative value for an RC product, for instance) appears.
Despite its shortcomings, the method has considerable merit where appli¬
cable, since it is considerably simpler than the general exact synthesis
method to be described in the next section.
1.23. The Exact Synthesis of L Sections. General Method.1 The
L section considered in this section is that shown in Fig. 1.15. Suppose
1 J. L. Bower, P. F. Ordung, and J. T. Fleck, “The Synthesis of Resistor-Capacitor
Networks,” Yale \Jniv..Dept. Elec. Eng. Tech. Rept., August, 1948, pp. 21-24.
Sec. 1.23] electric networks for control systems 35

that the desired transfer function meeting all the restrictions for L sec¬
tions discussed above has the form

(S + o:i)(s + 0:2) (s + 0:3)

H = = h (1.68)
Z\ + Z2 (S + dl )(S + /?2 )(S + ^3)

In this expression h is a constant gain multiplier which is to be made as

large as possible. Equation (1.68) may be rewritten as follows:

Z\ + Z2 _ , Z\ _ 1 (s d~ T~ 02)(s H~ ^3)
(1.69)
Z2 Z2 h (s + ai)(s + «2)(s + 0:3)
Defining
(s + + &2)(s + ^3)
F(s) A (1.70)
(s + ai)(s + 0:2) (s + ol 3)
we find that
Zi 1
(1.71)
h [F(s) ~ h]
From this equation it is clear that poles of Z1 or zeros of Z2 result from
poles of F(s), while zeros in Z\ or poles in Z2 result whenever F(s) — h = 0.

The zeros of F(s) — h are best found graphically, particularly if F(s) is of

higher than the second degree. A convenient method is to make a plot of
F(s) for negative real values of s. A typical plot is shown in Fig. 1.26 for
the case 0 < ai < di < P2, < oc2 < (33 < as. The abscissa of the plot is
significant only in certain regions and may be made approximately loga¬
rithmic in s with the ends distorted to include zero and infinity. Further¬
more F(s) does not need to be accurately plotted, and a rough sketch
suffices.
A horizontal line representing an arbitrary value of h is drawn on this
sketch, and each intersection of this line with the curve for F(s) is there¬
fore a zero of F(s) — h. In order to determine the maximum value of h,
36 CONTROL SYSTEM COMPONENTS [Chap. 1
we note that, in the example given, F(s) — h — 0 is a cubic equation and
therefore has three roots. For these roots to be negative and real the
line representing h must intersect F(s) three times for negative real 5.
Hence the maximum value of h is seen to be equal to F{ — 71), the maxi¬
mum of F(s) occurring between and /32. A horizontal line correspond¬
ing to this value of h is shown dotted in Fig. 1.26. In general, a similar
limitation on the maximum value of h can always be found.
The example given illustrates again the effect of successive poles in the
transfer function on the gain level of the network. Note that the limit
on /imax is set by the peak in the F(s) function occurring between the two
poles of H, that is, between and /?2. It should be clear from the con¬
struction that for maximum gain these poles should be separated as far as
possible.
The exact values of hmax and the zeros 71 and y2 must be known before
the synthesis can be made. These values are best found by a simple
trial-and-error procedure, as will be shown later in an example.
The distribution of zeros and poles to the two impedances Zi and Z2 is
now made as indicated by the o’s and x’s at the bottom of Fig. 1.26.
Starting at the left, the pole at X\ must be assigned to Zi as a pole, since
the smallest singularity in any impedance must be a pole. At the point
s = 71 there are two coincident zeros of F(s) — h; these are produced by a
zero in Z1 and a pole in Z2. For the three remaining singularities there
is a variety of possible arrangements. It is usually desirable to keep
the total number of poles in the two impedances as small as possible, since
each pole calls for a capacitor (see Sec. 1.24). The arrangement shown
in Fig. 1.26 results in a design calling for a minimum number of capacitors,
but it is not the only arrangement that does so.
The construction up to this point has now given us the impedances Z\
and Z2 in the form
s + 71
Z! = Ki (1.72)
s+ on
(s -j- a2) (s -f- 0:3)
Z2 = K (1.73)
(s + 7i)(s + 72)

The a’s and 7*s are known, but K1 and K% are not yet determined. The
ratio K1/K2 is found most easily by using Eq. (1.71) evaluated at a con¬
venient value of s. Thus, for s = — 00 ? Eq. (1.71) becomes

Zi Kl _ 1 _ 7N
(1.74)
Z2 k2 h(1 h)

One unknown parameter now remains. As noted previously, this arises

from the fact that in synthesizing a transfer function we do not consider
the impedance Level-of the resulting network. Thus the single unknown
Sec. 1.24] electric networks for control systems 37

parameter that remains is arbitrary and permits adjustment of the imped¬

ance level.
1.24. Exact Synthesis of RC Driving-point Impedances. The final
problem remaining is the synthesis of driving-point-impedance functions
such as those obtained in Eqs. (1.72) and (1.73) in terms of R’s and C’s.
This may be done in a variety of ways,1 two of which are given here.
In the first method we make a partial-fraction expansion of the imped¬
ance function. Thus, for the impedance given in Eq. (1.73) we have:

(<*2 + <23 ~ Yl ~ T2)S + tt2Q:3 ~ TlT2

= K2 + (1.75)
(s + 7i)(« + 72)
K3 K4
— K- 2 + (1.76)
s + Yl $ + 72

Note that the partial-fraction expansion is valid only for proper fractions;
hence the impedance function of Eq. (1.73) must first be converted by
division into the sum of a constant and a
proper fraction, as shown in Eq. (1.75). -WAV-
The constants i£3 and K± are found by
standard techniques. Equation (1.76)
suggests that Z2 consists of three com¬ C
1L
ponents in series. The first component
may be identified as a resistance K2, while Fig. 1.27. Resistor and capacitor
in parallel.
the other two both consist of a parallel
combination of a resistor and a capacitor, as shown in Fig. 1.27. This
may be shown by evaluating the impedance of this combination:

R/Cs 1/C
Z = (1.77)
R + (1 /Cs) s + (1 /RC)

Thus the impedance Z2 is synthesized as shown in Fig. 1.28, in which the

values of the components are in ohms and farads.

K-
3/7i
Kt4//?
-vWW- -vWW-
K*

1L _1 (_

13 !/K,
• /‘4
Fig. 1.28. Synthesis of Z2 (first method).

The second method of synthesizing the impedance function is again

illustrated by using Z2 of Eq. (1.73) as an example. We write
1 Guillemin, “ Communication Networks,” John Wiley & Sons, Inc., New York,
1948, vol. 2, pp. 211-216.
38 CONTROL SYSTEM COMPONENTS [Chap. 1

S ($ + 7i)(s + 72)
K2 $($ + <22) ($ + 0:3)

A1
+ +
D,
K2 \ S

131 H-j-1-j-
$
B2S
I
S -f 0:2

0:2
B3s
S + 0C3
$4-0:3,

(1.78)

where B1 = A1/K2, etc. The form of Eq. (1.78) suggests that the admit¬
tance Y2 consists of three components in parallel. The first of these is the
conductance Bh and the other two both consist of a capacitor and resistor
in series, since the admittance of such a combination is given by

Cs/R
(1.79)
Cs + (1 /R)
Hence by the second method the impedance is synthesized in the form
shown in Fig. 1.29, with resistance and capacitance values again given in
ohms and farads.
We may now complete the design OCy
of the transfer function given in Eq.
(1.68). Using the first method of
driving-point-impedance synthesis,
we obtain the network of Fig. 1.30,
in which the values of resistances
and capacitances are given in ohms
and farads, and where K1/K2 is
given by Eq. (1.74).

Fig. 1.29. Synthesis of Z2 (second Fig. 1.30. Network having transfer func¬
method). tion of Eq. (1.68).

1.25. Numerical Example. As an example of the synthesis method,

consider the transfer function

(s + 10) (s + 20) (s + 50)2

H = h( (1.80)
(s +!)($ + 2 )(s + 200) 2(s + 500)
In this function ho is a variable gain parameter which is to be maximized
in the design. The asymptotic a diagram of this transfer function is
shown in Fig. 1.31a. The function has four successive zeros and three
successive poles.and can therefore be realized by a two-section ladder net-
Sec. 1.25] electric networks for control systems 39

work. In Fig. 1.316 and c are shown the a diagrams of two L sections,
which together will realize the transfer function of Eq. (1.80). The two
component transfer functions are
(s + 20) (s + 50)
x x (s + l)(s + 200)(s + 500)
(1.81)
(s + 10) (s + 50)
v v {s + 2)(« + 200)

In making the distribution of poles and zeros between the two component
network sections, we keep in mind that the gain level between two succes¬
sive poles has an upper bound for realizable transfer functions (see Fig.

Fig. 1.31. a diagrams of over-all transfer function and component transfer functions.

1.24). The over-all network gain is therefore maximized by keeping the

required gain level between the two poles (as between s = —200 and
$ = —500 in Fig. 1.31) as low as possible. For this reason the pole at
s = — 1 and the zero at s = —20 are assigned to Hi rather than to H2.
It should be noted that there is in general some optimum division of this
sort, which occasionally may require the introduction into the component
40 CONTROL SYSTEM COMPONENTS [Chap. 1

networks of poles and zeros that do not exist in the over-all transfer func¬
tion. The reader is referred to some of the problems at the end of this
chapter which illustrate these principles (see, for instance, Prob. 1.14a).
We proceed by first synthesizing Hx. A sketch of F(s) for the transfer
function Hx is shown in Fig. 1.32, with the horizontal line for the maximum

s F(s)
-0.1515 85.67
-304 85.62
-308 85.67
-310 85.65
-320 85.06

Fig. 1.32. Design of Hx.

value of hx indicated. A few values of the trial-and-error solution for the

peak value of F(s) are shown in the table at the right. They give a
maximum value of hx of 85.67, with the maximum occurring at s = —308.
Also the intersection of the line for hx = 85.67 with F(s) is found by trial
and error to occur at s = —0.1515. Also indicated on Fig. 1.32 are the
zeros and poles of Z\ and Z2. The two poles of F(s) at s = —20 and
s = — 50 could have been interchanged without significant difference in
the final result, but otherwise the arrangement must be as shown. We
have, therefore,
s + 308
Z lx
s + 50
(1.82)
s T 20
(s + 0.1515)(s + 308)
The ratio K\x/K2x is best evaluated for s = 0; from Eq. (1.71)
Klx (308)2 X 0.1515
s=0 K2x 50 X 20 807 (10° - 85'67)
(1.83)
K lx
0.01167
K 2x
The partial-fraction expansion of the Z’s is

258 \
Zlx = 0.01167AT,
s + 50/
( 0.06447 0.935 \
Z2x = K 2x
\s + 0.1515 + s + 308/
Sec. 1.25] electric networks for control systems 41

The complete design of Hx is shown in Fig. 1.33, in which the values are
expressed in ohms and farads. K2x is left arbitrary to permit adjustment
of the impedance level.
The design of Hy proceeds in a completely analogous manner. A sketch
of F(s) is shown in Fig. 1.34. The maximum value of hm&x in this case is
F(0), because larger values of h would require zeros or poles in one of the
impedances for positive value of s. This is ruled out by the restrictions
on realizable impedance functions. The intersection of the horizontal
line for hmax with F(s) at s = —770
0.0602K2x
can be easily determined in this ex¬
ample by a direct algebraic solution.
The distribution of zeros and poles

Fig. 1.34. Synthesis of Hy.

to the two impedances shown in the figure minimizes the number of

capacitors. We have
s + 770
Z 1V = K1,
s + 50
(1.84)
s + 10
= K,y s

The ratio K\v/K2y may be evaluated for 5 = — oo and is given by

K ij/ 1
(1 - 0.8) = 0.25 (1.85)
K 2y 08
720
Hence Z\y = 0.257^22/ (1 +
8 + 50,
10 A 2 V
Z 2y = K^y +

The final design of Hy is shown in Fig. 1.35, with resistances and capaci¬
tances in ohms and farads.
The final step in the design is to determine the ratio of K2x to K2y so
that, when Hx and Hy are connected, the second L section will not load
 42 CONTROL SYSTEM COMPONENTS [Chap. 1

the first one excessively (see Sec. 1.19). In general, this step calls for a
calculation of the ratio of the impedances of the two sections, as required
by Eq. (1.34). Since one order of connection may have advantages in
the size of capacitors required, it is good practice to compute the circuit
for both orders of connection.
The impedance ratio ZoxIZiy for the case when network Hx precedes Hy
is most easily found from Eq. (1.38).

^OX 1x^3 X11 y fJ\x ir

Ziy Z2y Z2y

0.7998K2xs(s + 20) (s + 50) (s + 308)
(1.86)
K2y(s +!)(« + 2)(s + 200) 2(s + 500)
An asymptotic plot of this function shows the maximum to occur for
1 < to < 2 with a value of about 0.006K2x/K2y. If the impedance ratio
is to be less than 0.1 for all frequencies, the ratio of K2y to K2x should be
greater than 0.06. The network resulting for a ratio of 0.06 is shown in
Fig. 1.36.
If Hy precedes Hx, we consider the ratio

Zix z2x n
^ 0 K2y (s + 0.1515)(s + 10)(s + 50)(s + 308)(s + 770) , Q .
K2x (s + 1 ){s + 2)(s + 200)2(s + 500) 1 ;

The maximum value again occurs for 1 < o> < 2 and has a value of less
than 40K2y/K2x- Thus, if the ratio of the impedances Zoy/Zix is to be
less than 0.1, K2x/K2y > 400. The
3.6 KZy resulting network is shown in Fig.
-VWW"—
0.25 K^y 1.37.
-VWW— The final decision as to which cir¬
cuit is better depends now on prac¬
-1(-
0.00555'
Kty
tical circuit considerations. Thus,
K,2y
if the network were to drive the
.OJ_
grid of a vacuum tube, a maximum
K2y d-c resistance between grid and
ground of 1 megohm might be speci¬
Fig. 1.35. Final form of Hy. fied. This specification would be
met in the circuit of Fig. 1.36 with
a A of about 3 X 106, while in Fig. 1.37 K would be about 3 X 104.
(It is assumed that the network driving source has zero d-c resistance.)
Under these conditions the largest capacitance in Fig. 1.36 would be about
5.1 ijlf, while in Fig. 1.37 it would be only 3.3 gf. Since large capacitors
are bulky, the network of Fig. 1.37 has a slight advantage. In general,
it will be found that the size of the largest capacitors is minimized by
Sec. 1.26] ELECTRIC NETWORKS FOR CONTROL SYSTEMS 43

0.0602 K 0.216 K
—A/VW— -WWA-
0.01167 K 0.015 K
■wwv— t-\A/VW-< -o
Input Output
-l(- Hf—
0.332 0.0926
K K
0.06 K
0.426K

1.67
K

0.00304 K

Fig. 1.36. Final network with Hx preceding Hy.

3.6 K 24.1 K

Fig. 1.37. Final network with Hy preceding Hx.

placing the network section with the lowest frequency poles toward the
end of the cascade.
1.26. Bridged-T and Twin-T Networks.1 In this section we discuss
two networks that are commonly used to obtain transfer functions with

1 Chestnut and Mayer, “Servomechanisms and Regulating System Design,” John

Wiley & Sons, Inc., New York, 1955, vol. II, pp. 187-209. Valley and Wallman,
“Vacuum Tube Amplifiers,” Radiation Laboratory Series, vol. 18, McGraw-Hill
Book Company, Inc., New York, 1948, pp. 384-391. James, Nichols, and Phillips,
“Theory of Servomechanisms,” Radiation Laboratory Series, vol. 25, McGraw-Hill
Book Company, Inc., New York, 1947, pp. 117-124. L. Stanton, Theory and Applica¬
tion of Parallel-T Resistance Capacitance Frequency Selective Networks, Proc. IRE,
vol. 34, pp. 447-456, 1946.
44 CONTROL SYSTEM COMPONENTS [Chap. 1

complex zeros. The bridged T is the simpler of the two networks, but
the complex zeros produced by it can have only a relatively small imagi¬
nary part. The twin T is more complicated but also more flexible, per¬
mitting the synthesis of transfer functions with purely imaginary zeros
(i.e., complete rejection of one fre¬
._\(°1_ quency). The twin T also can be
designed with a nonminimum phase
o- >-VWW--VWVv—1 --o type of transfer function; i.e., it
/?, Ro
may have zeros with positive real
Input cz Output
parts.
The form of the bridged T is
shown in Fig. 1.38. The transfer
Fig. 1.38. Bridged-T network.
function can be found by applica¬
tion of standard techniques (the node method gives the result most
rapidly) and can be put into the form

K _ tt( \ _ R1R2C1C2S2 + Ci(Ri + R2)s + 1 , 0

Ji ~ {S) ~ R1R2C1C2s2 + [Ci(#i + R2) + RiCJs + 1 { }
By differentiation of the magnitude of the frequency function \H(jcc)*\, it
is found that the function has a minimum for

1
(1.89)
" = = VRiRiCxC*
It is convenient, therefore, to normalize the frequency variable s by
defining

v = — = s
a>c
(1.90)

The transfer function is simplified further by making use of the additional

definitions:
CxiRt + R2) = —
C0c
and (1.91)

With these definitions the transfer function may then be put in the form

p2 + 2fp + 1
H(p)
a + l\ (1.92)
v2 + 2r + V + 1
2r )
Physical realizibility requires both f and a to be positive real numbers.
Thus f cannot be made equal to zero, and the network cannot completely
reject the frequency coc. In fact, it is seen from Eq. (1.92) that, as f and
therefore the minimum gain of the network are decreased, the two poles
Sec. 1.26] electric networks for control systems 45

of the transfer function move farther and farther apart, and therefore the
phase lag introduced by the network at frequencies well below coc becomes
larger. This is illustrated by the attenuation and phase-shift curves

0.1 0.2 0.4 0.6 0.8 1.0 2.0 4.0 6.0 8.0 10.0
"/ft/c
(a)

Fig. 1.39. Transfer characteristics of bridged-T network.

given in Fig. 1.39. Since the complex zeros produced by the network are
quite often used to cancel a set of complex poles produced by some other
component of the loop, the large phase lag produced at low frequencies is
sometimes objectionable.
It should be mentioned that, although the bridged-T network of the
46 CONTROL SYSTEM COMPONENTS [CHAP. 1

form shown in Fig. 1.38 seems to be most commonly used, it is also possi¬
ble to construct the network with re¬
sistors and capacitors interchanged.
The basic form of the transfer func¬
tion as given by Eq. (1.92) is not
changed by this procedure.
The twin-T network is shown in
Fig. 1.40. Its transfer function is
Pig. 1.40. Twin-T network. again most easily found by use of
the node method and can be put into the form

I = Hw = m (L93)
where
N(s) = fli7?2fi3CiC2CV + (Ri + RJRzCiCzs* + 728(Ci + C2)s + 1 (1.94)
and
A(s) = R\R2R3C\C2C3s3 + [(Ri T RijR^C\C2 -f- (R2 -f- Rz)R\C2C3
R iR %C\C s]s2 + [Ri(C2 + C 3) + R2C2 + Rz{C\ + C2)]s + 1 (1.95)

Since the twin T is most often used as a rejection network, it is desirable

to determine, first, under what conditions there is a null in the transfer
function. The numerator of the transfer function is a cubic polynomial
and therefore has three zeros. If the network is to have a null at the fre¬
quency ccc, two of the zeros must bejcac and —jcoC) and N(s) must have the
form

N(s) =
uc2y
+ OJr + - + 1 (1.96)
7
where y is the third zero, which must be negative and real. Comparison
with Eq. (1.94) indicates that

—2 ~ (^1 R^RzR2 (1.97)

1 _ RiR2R3C\C2C3
(1.98)
^7 “ -Ri(Ci + C2)

The second relation is obtained by dividing the coefficient of 53 by that of

s. Equating Eqs. (1.97) and (1.98) and canceling common factors gives
for the null condition

C3 _ R9(Ri + R2)
(1.99)
C1 -b C2 R1R2
where n is any real, positive number. It will be shown presently that n
has an optimum value depending on the ratio of Ri to R2 and that of C1 to
Sec. 1.26] electric networks for control systems 47

C2. If the network is symmetrical, i.e., if R\ = R2 and C1 = C2, it will be

found that the optimum value for n is 1.
Note that, if the three capacitors are fixed, the adjustment of any one
of the three resistors will suffice to meet the null condition and cause the
circuit to reject one frequency. However, if the null is to occur at a pre¬
scribed frequency, at least two resistors must be adjusted simultaneously,
and the variation of one of the resistors is a fairly complicated function of
the value of the other. In practice the circuit is quite often constructed
using a three-gang potentiometer for the three resistances. The three
resistances are then all equal, so that the null condition is met with n — 2.
Variation of the three resistors permits a smooth variation of the rejection
frequency.
If the null condition is met, the transfer function of the circuit can be
simplified considerably. To do this, we find it convenient to let

(7 2 = ctC 1, R2 = fiRi (1.100)

Substituting for R2 in Eq. (1.99), we obtain

«! = «1 (1.101)
1 + j8

and using this result and Eq. (1.100) in Eq. (1.97), it is found that

RiC1 (1.102)

Finally, the frequency variable s is again normalized with respect to the

rejection frequency coc by defining

s
(1.103)

Substitution of all of these definitions and results [Eqs. (1.100) to (1.103)]

into Eq. (1.93) yields then, after some algebraic manipulation, the trans¬
fer function

JT / 1 +
\ _ P“

(1.104)
~ H{v) ~ 1 + Kp +
n( 1 + a) + «(1 + ff)
where (1.105)
y/&(3n
Thus, for a given null frequency the circuit behavior is completely deter¬
mined by n, by the ratio of R2 to R1, and by the ratio of C2 to C\.
Normally it is desirable to have as sharp a null as possible at the rejec¬
tion frequency. The sharpness of the null may be defined in a number
of ways, but a convenient one is to consider the slope of the magnitude
ratio \H(ju/uc)\ at the null as a measure of sharpness. Accordingly we
48 CONTROL SYSTEM COMPONENTS [Chap. 1

define the optimum value of K as the value giving the steepest slope of
|H(jco/c0c) | for co = coc. By differentiation of the magnitude of Eq. (1.104)
with p set equal to jco/wc, it is found that the slope at co/coc = 1 is — 2/K.
Hence the optimum value of K is the smallest realizable value. In deter¬
mining this value, we must keep in mind that a, (3, and n must be positive
real numbers. It is convenient to express K as the sum of two terms:

K = (1.106)

Note that the two terms indicated by the parentheses are both positive
and that K is therefore minimized by minimizing each one separately.
The first term has the form (x + 1/x). By differentiation this function
is easily shown to be minimum for x = 1. Hence one condition for mini¬
mum K is
afi _ j
n (1.107)
or n = a(3

Substituting this value of n into the expression for K gives

K = 2 + a + i (1.108)

It is clear that the minimum value of K is 2 and that it is approached by

making a as small as possible and (3 as large as possible. In practice, if
the twin T drives a grid of a vacuum tube, R<i should normally not exceed
1 megohm. Also, since R i should be about ten times as large as the source
impedance of the input, a practical minimum on R\ is about 10,000 ohms.
Hence a practical maximum value for (3 is about 100. Similar consider¬
ations limit a to about 0.01. Hence the practical minimum for K is of the
order of 2.02.
The twin T is often built symmetrically, i.e., with a = 0 = 1. Under
these conditions the best value of n is seen to be 1, and K = 4, or twice its
optimum value. As can be seen in Fig. 1.41, this value of K produces a
decidedly larger phase lag at frequencies below the rejection frequency
than the optimum value does.
Both the bridged T and the twin T are used to some extent as series
equalizers in a-c servo systems. When used in this way, the bridged-T
network has an effect similar to that of a standard d-c lead network (see
Sec. 1.13), while the twin T produces pure differentiation; i.e., the output
signal amplitude is proportional to signal frequency, and the output phase
leads the input phase by 90°.
The reason for this may be explained as follows: Consider first a
typical all-a-c system, such as the one shown in Fig. 7.1. If the shaft of
Sec. 1.26] electric networks for control systems 49

the input synchro in that figure is given a displacement varying sinus¬

oidally with time, then, as is shown in Sec. 5.2, the error signal going into
the amplifier has the form

e = Em (cos cost cos ioct) (1.109)

where cos is the relatively low signal frequency at which the shaft of the
synchro is being oscillated, while coc is the carrier frequency, which is

100

-100
0.1 0.2 0.4 0.6 0.8 1.0 2.0 4.0 6.0 8.0 10.0

[&)

Fig. 1.41. Characteristics of twin T for various values of K.

assumed to be at least an order of magnitude greater than the signal fre¬

quency. Furthermore, it is shown in Sec. 7.4 that to a first approxima¬
tion the torque generated by a two-phase servomotor supplied with a sig-
50 CONTROL SYSTEM COMPONENTS [Chap. 1

nal of the form of Eq. (1.109) is given by

T — Tm cos cost (1.110)

i.e., it is proportional only to the signal envelope. An equalizing network

inserted between the synchro and the motor to provide phase lead at sig¬
nal frequencies should, therefore, advance the phase of the signal envelope.
Hence, if the input to the network is a signal of the form given by Eq.
(1.109), the output should be

e0 = Eom cos (a),t + 6) cos uct (1.111)

where 6 is the desired phase lead. By use of standard trigonometric
identities the input signal [Eq. (1.109)] may be written in the form

e = [cos (a>c — us)t + cos (coc + co.)t] (1.112)

showing that the signal consists of two sidebands, one at the frequency
o)e — cOs and the other at the frequency coc + cos. Similarly the output
signal may be expanded into the form
Tp
e0 = {cos [(<oe — ccs)t — 6] + cos [(coc + cos)t + 0]} (1.113)

Thus we see that the function of the network must be to provide phase lag
at frequencies below the carrier frequency and phase lead at frequencies
above the carrier frequency. Inspection of Figs. 1.39 and 1.41 indicates
that both the bridged T and the twin T will perform this function, pro¬
vided that they are tuned so that their null frequency coincides exactly
with the carrier frequency. Note, however, that they perform the oper¬
ation only approximately, since their characteristics are symmetrical with
respect to the logarithm of frequency, while the desired characteristic, as
shown by Eq. (1.113), should be symmetrical with respect to the frequency
itself. This is not, however, a serious objection if the signal frequencies
are low with respect to the carrier frequency. A much more serious diffi¬
culty with this type of series equalization is that the performance depends
very critically on the carrier frequency, and even relatively small shifts
from the nominal value result in a serious loss of performance. This is
due to the fact that the network characteristics depend on the absolute
frequency of the signal rather than on its ratio to the carrier frequency.
To examine this problem quantitatively, and also to indicate an
approach toward a design procedure, suppose that the signal frequency is
given by
ct>s = ccoc (1.114)
Then the upper- and lower-sideband frequencies are, respectively,
wc(l + c) and xoc(l — c). The discussion is illustrated by using the
Sec. 1.26] electric networks for control systems 51

bridged-T network but is easily extended to the twin T (see Prob. 1.19).
The transfer function is given in Eq. (1.92). In order to get the frequency
function, let p — ju = jco/coc so that the normalized upper- and lower-
sideband frequencies become simply j(l + c) andj(l — c), respectively.
Substituting into Eq. (1.92) and replacing 2f + (a + l)/2f simply by
2fd, we obtain, after some algebraic rearranging,

c(c + 2)
1 +j
H (ju) 1 2r(c + i) (1.115)
U c(c + 2)
i +j
2?d(c + 1)
for the upper sideband and
c(c — 2)
i - j
H(ju) 1 2r(c - i) (1.116)
td c(c - 2)
i - j
2Uc ~ 1)

for the lower sideband. When the network is to be used to provide phase
lead in a servo, is made very much larger than f. Hence, for small c the
denominator is approximately equal to unity, and the transfer function
becomes approximately

c(c + 2)
H(jco) « £ 1 ±3 (1.117)
2f(c ± 1).

An approximate design procedure now becomes apparent. Suppose, for

example, that the carrier frequency of the servo is 60 cps and that it is
desired to provide a phase lead of 45° = 7r/4 radian at a signal frequency
of 3 cps. Then c = 0.05, and since for a 45° angle the real and imaginary
parts are equal, £ = 0.05. From Eq. (1.92) we have that, for the net¬
work to be physically realizable, £d must be greater than 20. This ful¬
fills the requirement above that £* ^Note, however, that the gain of
the network at the carrier frequency (c = 0) is £/{d, which in this case is
less than 34oo- Thus, in order to provide a reasonable phase lead at low
signal frequencies, the network must introduce a large amount of attenu¬
ation. Furthermore it is clear that the equalizer characteristics available
from this type of network cannot compare in variety or complexity with
the characteristics of d-c equalizer networks discussed earlier in this
chapter. A-c servos therefore tend to have poorer performance than
servos in which equalization is applied to the signal frequencies directly.
The effect of carrier-frequency shift on network performance can now
also be illustrated by an example. Suppose that the frequency shifts
5 per cent, or 3 cps. Then as far as the network is concerned, c for the
upper sideband becomes %q = 0.1, while for the lower sideband c is zero.
52 CONTROL SYSTEM COMPONENTS [Chap. 1

With the same value of as before, the phase lead of the upper sideband
is about 63.5°, while the phase of the lower sideband is zero. By invert¬
ing the process used to derive Eq. (1.113), it can be shown that the signal-
frequency phase shift is approximately equal to one-half the difference
between the phase shifts of the sidebands, which in this case amounts to
about 31°. Thus a small shift in carrier frequency results in a large loss
of phase lead. A further effect is that the carrier phase is also shifted by
about 30°. As shown in Chap. 7, this results in reduction of torque gener¬
ated by the motor and excessive heating.
Although the commercial 60-cps supply frequency is usually maintained
very close to its nominal value, this is not true of most 400-cps-circuit
supplies. The frequency of currently available aircraft supplies cannot
be guaranteed to be regulated to better than ± 10 per cent, or ±40 cps.
Such a frequency variation will make networks of the type discussed here
quite useless. The regulation problem of aircraft supplies may be solved
in the next few years, thus making high-performance a-c servos practical
in aircraft.
1.27. The Use of Amplifiers with RC Networks.1 The variety of
transfer functions available with RC networks can be extended greatly by
using networks in conjunction with amplifiers. The subject is too large
to be dealt with completely here, and only some of the possibilities are
indicated by the following examples.
A common procedure is to use networks in the return path of a feedback
amplifier, as indicated in Fig. 1.42.
If the amplifier gain is G and if the
Amplifier
network transfer function is H,
then the transfer function through
e{
the entire circuit is
A

Network en G
1 + HG
Fig. 1.42. Feedback amplifier with RC
network in feedback path. 4 if HG » 1 (1.118)

Thus, provided the loop gain is sufficiently high, the approximate transfer
function is the reciprocal of the network transfer function.
This principle can be applied in the generation of transfer functions
having complex poles by making the network in the return path of the
feedback amplifier a twin T. Qualitatively such an amplifier will have
a gain of unity at very low and very high frequencies, but at the null fre¬
quency of the twin T, the gain is G, since the feedback is not operative
there. Clearly the transfer function is not the exact reciprocal of the
twin-T transfer function. Even though the transfer function of the twin T
1 Valley and Wallman, op. cit., pp. 391-408.
Sec. 1.27] electric networks for control systems 53

has imaginary zeros, that of the feedback amplifier cannot have imaginary
poles, since this would require infinite amplification at the null frequency.
The discrepancy can, however, be made as small as desired by making the
gain of the amplifier, G, large enough.
Another interesting application of the twin-T circuit in connection with
a feedback amplifier is shown in Fig. 1.43. This circuit is referred to as
the rejection amplifier, and it has the effect of sharpening the null of the

Fig. 1.43. Rejection amplifier.

twin T. The operation of this circuit is most easily explained by deter¬

mining the over-all transfer function e0/e{. Application of standard tech¬
niques and the transfer function for the twin T [Eq. (1.104)] gives

r (P2 + 1)
ea _ (p2 + Kp -f -1) v2 + i (1.119)
* ~ , g (p2 + i) K
V + 1
^ (P2 + Kp + i) «?+l)
where G is the amplifier gain. Comparison with Eq. (1.104) indicates that
the effect of the circuit is essentially to change the K in the denominator
of the transfer function of the twin T to K/{ 1 + G). In the discussion
of the previous section it was shown that the slope of the magnitude of the
frequency function at the null was —2/K for the twin T; hence it is clear
that the action of the amplifier is to make the slope steeper, and therefore
to sharpen the null.
In both the examples given, the transfer functions that were obtained
are of the type that could also be obtained by the use of inductances.
Thus, complex poles are commonly obtained by simple RLC resonant cir¬
cuits, while the sharp rejection characteristic of the second example may
be shown tQ be identical to that of a bridged-T network in which one of
the resistances is replaced by an inductance.1 This bears out the state¬
ment made in Sec. 1.14 that, by the use of RC networks with amplifiers,
practically any transfer function that can be realized with resistances,
capacitances, and inductances can be realized without actually having to
use inductances.
A final example illustrates the use of a difference amplifier (see Sec.
2.11) to generate a nonminimum phase transfer function. The circuit
is shown in Fig. 1.44.
1 Valley and Wallman, op. cit., p. 385.
54 CONTROL SYSTEM COMPONENTS [CHAP. 1

The difference amplifier is supposed to have infinite input impedance,

and its output voltage e0 is given by G{e 1 — e2). Then, since
A
ei 1 , 02 Ts
Ts + 1 and Ji = Ts + 1
(see Sec. 1.13), the over-all transfer function is
60 = g 1 — Ts
(1.120)
e,- 1 + Ts
This transfer function is of the all-pass type; i.e., its frequency ratio
has a constant magnitude of G, but the phase shift between output and
input changes from 0 to 180°. This transfer function can also be obtained,
R

Fig. 1.44. Use of difference amplifier to produce nonminimum phase transfer function.

perhaps more simply, by a bridge network in which each arm contains

either a resistor R or a capacitor C, and in which the R’s and C’s alter¬
nate. The disadvantage of a bridge circuit is, however, that there is no
common ground connection between output and input. Hence a circuit
of the sort shown in Fig. 1.44 might be preferable.
1.28. Reliability. The great complexity of modern electronic equip¬
ment and, to a lesser extent, other forms of equipment and the resulting
failures in operation have led in the past few years to an intensive study
of the basic problem of reliability. This interest is especially great in the
military services and was probably brought to a head when it was shown
during the Korean War that at any one time, two-thirds of all the elec¬
tronic equipment in the theatre was inoperative because of failure.
Indicative of this rapidly growing interest in reliability is the fact that
in 1954 almost no contracts for military equipment explicitly mentioned
reliability, while at the present time almost all contracts require proof of
reliability by performance. Secondly, reliability is at the present time
placed first in order of importance in design contracts by the military,
even ahead of the meeting of performance specifications.
The major reasons for failure of electronic and other devices are four-
Sec. 1.28] ELECTRIC NETWORKS FOR CONTROL SYSTEMS 55

fold. First, complexity. Second, improper design and over-loading of

components. Third, failure due to environmental conditions such as
heat, humidity, and shock. Fourth, improper maintenance.
That equipment designed in the past few years is more complex than
its predecessors is indicated by the fact that the number of vacuum tubes
carried in military vehicles from aircraft carriers to intercontinental
bombers has increased by a factor of 4 or 5 since World War II. Even if
each element were just as or more reliable, this increase in complexity
would cause a dramatic increase in the failure rate.
We define reliability as the probability that a device will operate
properly under given environmental conditions for a specified length of
time. The time specified may be 20 years for tubes installed in the trans¬
atlantic telephone cable, 100 hours for a long-range bomber, or 1 hour for
a guided missile.
All of the elements in a piece of electronic equipment do not have the
same reliability, of course. Resistors are perhaps 0.9999 reliable, while
vacuum tubes are perhaps 0.7 or 0.8 reliable. It is inevitable that the
vacuum tube will continue to be the most unreliable component in elec¬
tronic equipment because of the high temperatures involved in its oper¬
ation and the delicate nature of its construction. However, by proper
design and quality control its reliability will certainly be improved con¬
siderably. To find the over-all reliability of the equipment, we need only
take the product of the reliability of each component, provided we make
the more or less realistic assumption that the reliability of one component
does not affect the reliability of the other components. Since we are
interested in only the first failure, this seems justified. If it is further
assumed that all components are equally reliable, simply for demonstra¬
tion’s sake, the over-all reliability is the reliability of the individual com¬
ponent raised to a power equal to the number of components in the unit.
In Fig. 1.45 is shown a plot of such a calculation. The required reliability
of the individual components is plotted as a function of the desired
over-all reliability of the equipment. The number of components in the
equipment is the parameter. Notice, for example, that in a unit with
1,000 components, 99.9 per cent reliability of the individual elements will
result in about 37 per cent reliability of the over-all unit, and if the indi¬
vidual reliability is raised to 99.99 per cent, the unit is still only 91 per
cent reliable.
It is obvious from this chart that the designer must strive for simplicity
and that a modification which requires a considerable increase in com¬
plexity should probably not be accepted even if its performance is con¬
siderably better than the simpler model’s. It is apparent also that the
list of preferred tube types must be narrowed considerably and then
adhered to religiously if we are to achieve reliable electronic components.
56 CONTROL SYSTEM COMPONENTS [Chap. 1

Designers must be encouraged not to design new circuits for each new
device but to use standard circuits that are mass-produced and under rigid
quality control.
The second cause of unreliable devices is the attempt by the designer
to wring every last bit of performance from electronic components. The
statistical life history of a batch of components or complete units, what¬
ever their type, usually falls into a pattern such as that shown in Fig. 1.46.

Fig. 1.45. Over-all reliability of a device as a function of the reliability of the individual
components.

The curve may be divided into three regions. Region A, or the infant
mortality region, consists of failures that occur in the break-in or shake-
down period. The Navy’s insistence on a shakedown cruise and specifica¬
tions by buyers for 10 to 20 hours operation of a unit at the factory are
both good procedures, as is seen from this chart. Next comes region B, the
operating region. If the unit is well designed, the failures that occur in

Fig. 1.46. Typical life history of a group of devices or components. Region A is the
break-in region, B represents the operating region, and C is the wear-out region.

this region are few and random in nature and time of occurrence. Finally,
region C, or the wear-out region, is a bell-shaped curve in a well-designed
piece of equipment.
It has become a vogue in the guided-missile field to attempt to design
equipment for short operating life. The argument goes that the unit
need operate for only an hour or two at the most; thus, it is wasteful to
Sec. 1.28] electric networks for control systems 57
design for more. We can see the fallacy of this argument from Fig. 1.46.
The break-in period blends into the wear-out region and there is essen¬
tially no period of trouble-free operation. We can categorically state
that it is impossible to build a reliable, short-lived piece of equipment,
and our missile program will be successful only when this is realized.
In general, the designer only shortens region B when he pushes compo¬
nents up to and beyond their design limits. It has been found, for exam¬
ple, that operating vacuum tubes at 120 per cent of their rated plate dissi¬
pation will result in about 200 per cent the failure rate of tubes operated
at their design value. Several manufacturers have recently set up as
standard-design procedure that vacuum tubes are to be operated at a
maximum of 50 per cent of their rated plate dissipation. This has
resulted in a 4:1 improvement in reliability.
The third cause of unreliability is operation under difficult environ¬
mental conditions. Extrapolation of present trends indicates that
environmental specifications, such as ambient temperature, shock, and
operation at low atmospheric pressures, will continue to become more
severe. Derating of conventional components will soon fail to provide
an operating margin. Extensive research on new materials and methods
for component manufacture is imperative.
Finally, unreliability is due to improper maintenance. This may be
the result of improperly trained technicians, lack of thought on the proper
layout for ease of maintenance, or simply, equipment complexity.
Usually all three causes are at work. One solution gaining favor is the
use of modular packaging. Modular packaging of subunits of an assem¬
bly reduces repair to simply finding the troublesome module and replacing
it. This is not only quicker and easier but is apparently more economical
in the long run.
PROBLEMS
1.1. By means of resistance padding, construct the best approximation to the square
law y = 2x2, where x is shaft position in radians and y is resistance in ohms. Let
1/max = 10,000 and use a maximum of three segments.
1.2. It is desired to construct by voltage padding the response function shown in
Fig. 1.47. Use four intermediate voltage taps.

Radians
58 CONTROL SYSTEM COMPONENTS [Chap. 1

1.3. By resistance, voltage padding approximate the cubic equation

y = x — x2 + 0.5rc3

where x is shaft position in radians and y is the output voltage.

1.4. Draw the asymptotic a diagram for the functions:

fc(0.1s + 1) (0.05s + 1)
a‘ s(0.01s + 1)(0.00001s2 + 0.001s + 1)
, s(5s + l)(0.2s + 1)
(10s + l)(s + 1) (0.1s + 1) (0.02s + 1)

1.5. A network consists of two sections, H i and H2. The output impedance of
Hi is given by
Aq(0.5s + 1) (0.02s + 1)
(2s + l)(0.1s + 1) (0.01s + 1)

and the input impedance of H2 is

^ k2(s + 1) (0.05s + 1)
i2 s(0.25s + 1)

Find the ratio of k i to k2 such that section H2 does not load section Hi.
1.6. Two lead networks (see Table 1.2) are to be cascaded to produce an over-all
transfer function with a +2 slope in the asymptotic a diagram. The attenuation at
low frequencies is the same for the two networks. Show that the ratio of impedance
levels required to prevent loading between the two networks is a maximum when the
two lead networks have the same break frequencies coi and co2.
1.7. Write down by inspection the impedance functions of the networks shown in
Fig. 1.48.

R
WVvV-
R C
o-MW-|f
c
(o)

id)
Rz
WW—
—AW- —AW—4
Rz c
< >—v/VW—o
c ft*
y WWW

ic) id)
Fig. 1.48

1.8. For the driving-point impedances containing more than one capacitor shown
in Fig. 1.49, write the impedance function by inspection by assuming that for any
particular frequency only one capacitor is active. Also assume that the active zone
for C2 comes at a higher frequency than that of Ci, etc.
Problems] electric networks for control systems 59

Rz
R\
—AW—
C\
"](■
*3 Cz
■VWW-
■tt

[o)

R\ Rz
[—vWv^- -MW-

Hf- /?3 ^3 R,\

C\ -1^ L-WW—If—*
Id)
Fig. 1.49

1.9. Design a single L-section network by the approximate technique to synthesize

the following transfer functions:

0.2s (0.05s + 1)
a. H
(s + 1) (0.01s + 1)
(0.5s + 1) (0.05s + l)2
b. H
(2s + 1) (0.2s + l)(0.01s + 1)
,_s(0-2s + 1)_
c. H (,h is to be as large as possible.)
(s + 1) (0.05s + 1) (0.01s + 1)
(0.1s + 1) (0.025s + 1)
d. H
(5s + 1) (0.01s + 1) (0.002s + 1)

1.10. Design the networks of Prob. 1.9a and b by the exact method described in
Sec. 1.22.
1.11. Use two L sections to synthesize the following transfer functions. Use the
approximate method described in Sec. 1.21 to obtain the L sections, and adjust the
ratio of impedance levels to prevent loading.

1.25s2(0.4s + l)2
a. H
(10s + 1) (2s + l)(0.1s + l)2
(0.05s + l)2(0.01s + 1)
b. H
(s + 1) (0.005s + l)2
(s + 1) (0.01s + 1)
c. H
(0.1s + l)2
(0.0143s + l)3
d. H
(s + l)(0.1s + 1) (0.001s + l)2
s2(0.005s + 1)
e. H (Design for maximum h.)
h (0.1s + l)2(0.05s + 1) (0.02s + 1)

1.12. Use the exact method of Sec. 1.22 to synthesize the individual L sections for
the transfer functions given in Prob. 1.11. Compare the resulting designs to those
obtained in Prob. 1.11.
1.13. Design a two-terminal impedance to have the following impedance function:

(0.02s + 1) (0.0025s + 1)
Z = 10,000
(0.05s + 1) (0.01s + 1)

Use both methods given in Sec. 1.24.

60 CONTROL SYSTEM COMPONENTS [Chap. 1

1.14. Use the exact synthesis method described in Secs. 1.23 to 1.25 to obtain the
following transfer functions. Use the minimum number of L sections. In all cases
design for maximum h.

s2(s + 500) (s + 2,000)2

a. H =
h (s + 10)2(s + 50)(s + 10,000)2
, 8(8 + 10) (8 + 20) (s +40)
b. H =
(8 + 1 )(8 + 3) (8 + 5) (8 + 80)
h _(8 + 70)3_
c. H =
(s + l)(s + 10) (s + 1,000)2

1.16. Find the input impedance and output impedance of the bridged-T network
of Fig. 1.38. Assume the source impedance is zero and the load impedance infinite.
1.16. Repeat Prob. 1.15 for the twin-T network of Fig. 1.40, in which Ri = R2 =
YzRz and Ci = C2 = 2C3.
1.17. A twin-T network for which a = (3 = n — 1 is to be cascaded with a simple
low-pass network (Table 1.2, second row). The pole in the low-pass network is to
coincide with the null of the twin-T network. Find the ratio of impedance levels
required to prevent appreciable loading between the two networks. Assume that the
low-pass filter follows the twin T.
1.18. By the use of a bridged-T network together with one other L section, design
a cascaded network to synthesize the following transfer function:

(2s + 1) (0.01s2 + 0.05s + 1) (0.005s + 1)

(10s + l)(s + 1) (0.01s + 1) (0.001s + 1)

Check that the two sections do not load each other appreciably.
1.19. Show that a twin-T network tuned to the carrier frequency of an a-c servo
system has an effect on the carrier envelope that is similar to the effect of a perfect
differentiating circuit on a low-frequency signal.
1.20. Design a bridged-T network to provide a 30° phase lead at a signal frequency
of 6 cps for a servo using a 400-cps carrier. What is the effect of a 10 per cent shift
in carrier frequency on the network characteristic?
1.21. An improperly tuned bridged-T network produces a phase shift of du in the
upper sideband and a phase shift 6i in the lower sideband. Show that the effective
phase lead of the envelope frequency is approximately A (6U — 6i). Also show that
the phase shift of the carrier is approximately A(0u + 0j).
 CHAPTER 2

D-C AMPLIFIERS

2.1. Introduction. D-c amplifiers are required as control amplifiers

in a large class of electromechanical servomechanisms. In the following
treatment it is assumed that the reader is familiar with the elementary
rules of analysis and design of standard audio amplifiers, and the emphasis
will, therefore, be primarily on points in which d-c amplifiers differ from
the more conventional types.
A d-c amplifier is capable of amplifying signals at very low frequencies,
including zero frequency or direct current. This means that the distinc¬
tion between signal voltage and power-supply voltages, which in a con¬
ventional audio amplifier is obvious from the difference in frequency bands
occupied, more or less completely disappears in the d-c amplifier and
becomes largely a matter of definition. Thus, the output voltage is as
much a function of the operating voltages as of the designated signal
voltage and may have any value, even though the input voltage is zero.
Since any power-supply-voltage variation or circuit-parameter variation
will affect the output voltage, a d-c amplifier is subject to drift, i.e., a
variation of the output independent of the designated input. Drift is
undesirable, and its minimization poses one of the major problems in the
design of d-c amplifiers.
Another difficulty arises from the fact that the low signal frequencies
that must be handled make it impossible to use capacitors or transformers
for interstage coupling or for any other separation of signal and power-
supply or bias voltages. A d-c amplifier is characteristically direct-
coupled. Hence the design of interstage coupling networks is another
important problem facing the .designer of d-c amplifiers.
Further requirements usually made of d-c amplifiers are similar to those
made of audio amplifiers and require only brief mention. The gain of the
amplifier should be stable and constant over as large a frequency range
as possible. Any spurious output, usually referred to as noise, should be
held to a minimum. It is usually desirable to have a high input imped¬
ance and a low output impedance (ideally infinity and zero, respectively).
Ordinarily the output voltage should be zero when the input is zero, and
an adjustment is usually provided somewhere in the circuit to accomplish
this.
61
62 CONTROL SYSTEM COMPONENTS [CHAP. 2

2.2. Drift in D-C Amplifiers. There are three major factors that cause
drift in a d-c amplifier. These are:
1. Variations of power-supply voltages
2. Changes of components with temperature and age
3. Effects of heater-voltage variation
Only the third of these is assumed to be unfamiliar. Drift is caused by
heater-voltage variations because changes in cathode temperature result
in changes of the initial velocities of the electrons emitted, so that the
electrode voltages required to maintain a given electron flow must change.
It has been found that, if the plate current is small compared to the total
emission from the cathode, this effect is essentially independent of the
plate current. The heater-voltage effect is best expressed in terms of the
amount by which the cathode voltage must change relative to the other
electrode potentials in order to hold the plate current constant. For
oxide-coated cathodes, this amounts to about 0.1 volt for a 10 per cent
change in heater voltage about the normal value, and it is relatively
independent of the tube type.1
Various methods are used to counteract these effects. An obvious
remedy is to regulate carefully all power supplies including heater sup¬
plies. High-quality components, wire-wound resistors, etc., will tend to
minimize the effect of component variations. A number of cancellation
methods have been devised to minimize the effect of cathode-temperature
variation, and some of these will be described in detail below. These
methods usually make use of the fact that the difference between the
voltage changes required in two tubes to keep the plate current constant
is much less than the change re¬
quired in either tube alone. When
b -~7^°.
_ 60 +^ they are applicable, push-pull cir¬
(a)
cuits are very effective in reducing
drift from all three causes, particu¬
larly if matched tubes and compo¬
nents are used. Negative feedback
is usually ineffective, since it does
not ordinarily affect the signal-to-
[t>)
noise ratio of an amplifier. This is
Fig. 2.1. Block diagrams of circuits used
for the calculation of the effect of nega¬
demonstrated below.
tive feedback on drift. 2.3. Effect of Negative Feedback
on Drift. Suppose that the ampli¬
fier shown in Fig. 2.1a has a gain G and generates an internal noise having
the value v at the output. (Drift may be thought of as noise of very low

1 An excellent discussion of this phenomenon can be found in Valley and Wallman,

“Vacuum Tube Amplifiers,” Radiation Laboratory Series, vol. 18, McGraw-Hill Book
Company, Inc., New York, 1948, pp. 421-424.
Sec. 2.3] - d-c amplifiers 63

frequency.) Then the output is

e0 — Gei + v (2.1)

Suppose we attempt to improve this performance by using feedback.

Since feedback alone will reduce the gain, we must add a preamplifier Gi
ahead of the original amplifier G, and, to be realistic, we must assume that
the additional amplifier also contributes noise. Let this noise, measured
at the output of Gi, be v\. The input voltage to G\ is

C\ @e0 (2.2)
Therefore e0 = G\Ge\ + Gv i + v
— GiGci — GiG(3e0 Gvi T v
_ G\G . Gv i v
(2.3)
1 T GiG/3 1 1 T GiG(3 1 T GiG/3

For the feedback to have full effect, it is desirable that

Gffl » 1 (2.4)

In this case, the denominator 1 + GiG/3 in Eq. (2.3) may be replaced

approximately by GiGp, and Eq. (2.3) may be rewritten in the approxi¬
mate form

~+ i1,1 + oh ” (2-5)
In order to have a valid comparison between the feedback amplifier and
the original amplifier, we make the gain of both amplifiers the same; i.e.,
1//3 = G. Then

eQ = Gei + jy- v\ + ^7- v (2.6)

We see that, although the original noise v has been reduced by the gain
of the preamplifier, Gi, an additional noise term (G/Gi)vi has been added.
If we assume that the amount of noise added by an amplifier is propor¬
tional to the amplifier gain, i.e., if v\ = (Gi/G)v, then we see that the feed¬
back amplifier has about the same noise output as the original amplifier.
The above argument does indicate, however, that if the preamplifier
were noise-free, the noise output of the feedback amplifier would be con¬
siderably reduced. It also indicates that, in the design of a feedback
amplifier, all care should be taken to make the first stages as noise-free as
possible.
Getting back to the problem of drift, we shall find that it is possible to
construct d-c amplifiers with zero drift and very high d-c gain. The gain
does, however, drop sharply as the frequency is increased to values as
high as 1 or 2 cps. If such an amplifier is used as a preamplifier in a feed¬
back circuit like that of Fig. 2.16, the drift introduced by the main ampli-
64 CONTROL SYSTEM COMPONENTS [Chap. 2
fier may be reduced to as small a value as is desired. The fact that the
gain of the amplifier drops at higher frequencies is of no consequence, for
as long as the inequality (2.4) is satisfied, the gain is given by 1/(3 and is
essentially independent of the gain of the forward path of the feedback
loop. This principle is employed in the chopper-stabilized d-c amplifier1
described in detail later.
2.4. Analysis of Simple Triode Circuits. Before considering the opera¬
tion and design of complete amplifiers, we must examine the behavior of
single stages rather completely, with particular emphasis on the fact that
they are to be used in d-c amplifiers. The quantitative effect on drift of
the three factors listed in the last section, as well as questions of gain and
terminal impedances, will be discussed.

Fig. 2.2 Linear triode plate characteristics.

For purposes of analysis, the vacuum tube will be assumed completely

linear with plate characteristics as shown in Fig. 2.2. The dotted lines
represent the actual characteristics, while the solid lines represent the
linear approximations.
The linear characteristics are parallel straight lines, equidistant for
equal increments of grid voltage. The distance between the intercepts
of two characteristics differing by a unit grid voltage is /jl volts on the epk
axis. The line for egk = 0 does not, in general, pass through the origin,
and its intercept on the epk axis, usually positive, is Ex volts. It should
be noted that these characteristics are derived from the published non¬
linear curves by drawing a tangent to the actual characteristic passing
through the operating point. The distance between the lines egk) is
determined by the average distance around the operating point. Thus
the exact shape of the linearized characteristics depends very much on
the operating point chosen.
By assuming these linear characteristics, the equation for the plate

1 E. A. Goldberg, Stabilization of a Wide-band D-C Amplifier for Zero and Gain,

RCA Rev., June, 1-950, pp. 296-300.
Sec. 2.4] d-c amplifiers 65

current can be found to be

tp (@pk ~l- P-@gk Ex) (2./ )

Tp

where the voltages epk and egk are the plate and grid voltages, respectively,
measured relative to the cathode. The tube may therefore be represented

egk

T
ek

Fig. 2.3. Linear equivalent cir¬ Fig. 2.4. Equivalent circuit of

cuit. heater-voltage effect.

by the equivalent circuit shown in Fig. 2.3, where the arrows on the
voltages represent the assumed positive sense.
This circuit is accurate only for normal cathode temperature, since it
does not provide for the insertion of the cathode-temperature-variation
effect. The initial velocity of the electrons results, approximately, in a
“virtual” cathode voltage differing from the cathode voltage measured
at the terminal. The effect of heater-voltage
variation may, therefore, be most easily repre¬
sented in the equivalent circuit by a voltage source
en inserted between the cathode terminal and the
cathode, as shown in Fig. 2.4. Hence the equiv¬
alent circuit of Fig. 2.3 may be generalized to
include heater-voltage variation by considering
the terminal marked “cathode” as the “virtual”
cathode inside the tube and by connecting a eg
voltage source eh between this terminal and the
outside connection. The voltage eh may be de¬
fined as T
Fig. 2.5. Triode equiv¬
dek alent circuit with heater-
eh Ae/
de/ iP = c voltage effect.

where Aef is the change in filament voltage. This leads to the equivalent
circuit of Fig. 2.5, which applies to all triodes. For this circuit the equa¬
tion for the plate current becomes
66 CONTROL SYSTEM COMPONENTS [Chap. 2

where egk is the voltage between the grid and the virtual cathode, as
shown in Fig. 2.4.
2.5. The Cathode Follower. Cathode followers are used primarily as
impedance changers, since they offer a very high input impedance and
very low output impedance. The voltage gain is slightly less than unity.
The actual circuit is shown in Fig. 2.6a and its equivalent in Fig. 2.66.

Fig. 2.6. Cathode follower.

The output voltage is ek, measured between cathode and ground. The
current iQ is the load current and is included for generality. It will be
noted that the cathode is not returned to ground but to a negative
voltage Ekk- This is necessary if the output voltage ek of the cathode fol¬
lower is to swing through negative as well as positive values of voltage.
The complete equation for this circuit may be derived by use of Kirch-
hoff’s current law. The expression for the current leaving the node ek is

(ek — Ekk) -p- + (ek + eh — i*-cgk + Ex — Ebb)-i0 — 0 (2.9)

■tck r r>

But inspection of Fig. 2.4 reveals that

cgk (eg 6h C&) (2.10)

Hence Ck [ —I-+ — \ — Ekk yt ~ (Ebb ~ Ex)-eg —

\Jdk Tp rp t Ri V

+ Ch l -—b —) ~■ i0 — 0 (2.11)
'Tp ^PI

cg — + (Ebb ~ Ex)-b Ekk tt ~ eh ———- + i0

and ek = ' prn rp Hk r1V
1 1 4~ M

Rk rp
_ uRkeg + Rk(Ebb ~ Ex) rpEkk + + Rkrpi0 — (1 + fi)RkCh
(2.12)
Rk(u 1 + )
Sec. 2.6] D-C AMPLIFIERS 67

In deriving this equation, no assumption other than linearity has been

made. Hence this equation is correct for a-c or d-c, over the range of
variation of the variables for which the assumption of linearity may be
made to hold. In particular, the quiescent value of ek may be found by
setting eg = ek = ia = 0 and inserting the proper values for Ebb, Ex, and
Ek
Generally, however, we are interested in specializing this equation to
obtain expressions for gain, output impedance, effect of power-supply-
voltage and component variations on drift, etc. This is done by partial
differentiation. Thus for the gain we have

dek fj.Rk
G = 1 (2.13)
deg (n + 1 )Rk + rp M 4" 1

The approximations made to obtain the final result are that (/x + l)Rk ^>rp
and n^>> 1. For a typical triode (6SL7) common values are /x = 70,
Rk = 200,000 ohms, rp = 50,000 ohms; hence these approximations are
very good. Similarly we may find the output impedance

2? _ dek _ Rkvp v (9 1A\

diQ (/u + l)Rk T /x + 1 gm

Common values for the output impedance of a cathode follower are from
500 to 1,000 ohms, corresponding to a gm of from 1,000 to 2,000 nmhos.
Note that the value of Rk does not appear in the approximate expression
for the output impedance.
The drift due to changes in Ebb or Ekk is found from Eq. (2.12) to be

dek 1 _ 1
(2.15)
1

dEbb (m + 1 )Rk + rP jU + 1 H
dek rp TP ^ 1
and (2.16)
dEkk (m T l)Rk + Tp (/x + 1 )Rk gmRk

Note that drift from these two sources is rather small, especially if Rk is
large. The drift due to heater variations is, however, large; it is given by

dek _ — (1 + ii)Rk
(2.17)
deh (1 + n)Rk + rp

Thus the drift due to heater-voltage variations is essentially equal to the

change in or about 0.1 volt for every 10 per cent of heater-voltage
change.
2.6. Input Impedance of the Cathode Follower. Since the cathode
follower is used as an impedance-changing device, there is considerable
interest in the value of the input impedance. Up to now, this has simply
been assumed infinite. In practice, however, there is some leakage from
the grid to the plate and cathode in any tube, which results in a finite
68 CONTROL SYSTEM COMPONENTS [Chap. 2

input impedance. Since the two leakage paths from the grid are in
parallel, it is more convenient to consider the input admittance. We
define leakage admittances between grid and plate and between grid and
cathode as shown in Fig. 2.7. These admittances are assumed to be
measured on a dead tube, i.e., on a tube for which \x = 0.
The current ig must be equal to

ig — — Ebb) Y op + (eg — ek)Y0k (2.18)

Therefore the input admittance is

dig
— Ygp + ( 1
deg (

— Ygp + 1 - Y gk
(m + 1 )Rk + tP/
\r* + rP)Y gk Y gk
= Ygp + Ygp “F (2.19)
(/J, + 1 )Rk + Tp +1
where the assumption has been made that Rk » rp, so that Rk + rp « Rk.
Thus we see that, although the cathode fol¬
lower has no effect on the grid-to-plate
admittance (which is the same as the grid-
to-ground admittance, since the plate is at
“signal ground7’), the grid-to-cathode ad¬
mittance, measured when /x — 0, is reduced
approximately by the factor 1/(1 + /i)
when the tube is operating. This result
should be.compared with that obtained with
a conventional grounded-cathode amplifier,
where the input admittance is equal to the
Fig. 2.7. Input-admittance
paths in cathode follower.
sum of the grid-to-cathode admittance plus
the grid-to-plate admittance multiplied by
the gain of the stage.1 This comparison shows the cathode follower to
have a much lower input admittance than an amplifying stage.
2.7. Drift due to Component Variation. Equation (2.12) can also be
used for computing the drift caused by component variations. Thus, for
instance,

dek + {Ebb — Ex) + rvi0 — (a* + 1 ){Ekk + e*)]

dRk [Rk(v + 1) + rp]2
+ (Ebb — Ex) + r pi0 — (ix + 1 )Ekk]
(2.20)
[(m + 1 )Rk + Tp]‘
1 Reich, “Theory and Applications of Electron Tubes,” McGraw-Hill Book Com¬
pany, Inc., New York, '1944, pp. 93-96.
Sec. 2.9] d-c amplifiers 69

since eb <C Ekk. Also we have

d@k Rjc[(Rk H- Rk:(Ebb Ex) EkTpto Vp(Ekk "I- 6h)\ /.-) <)i
dJI [(A* + 1 )Rk + rj2 1 j

These relations and similar ones for other component variations are
useful, not only for the computation of drift, but also to determine the
maximum range of deviation of the output voltage as a result of compo¬
nent tolerances. If such a range is known, it becomes a relatively simple
matter to specify the range of balancing potentiometers required in the
amplifier. Note that, in the application of formulas for drift due to com¬
ponent variations, all voltages and currents involved in the circuit must
in general be known.
2.8. Cathode Follower with Plate-load Resistor. Occasionally, a
plate-load resistor RL is inserted between the plate and the positive supply
voltage of a cathode follower. This occurs particularly in situations
where the tube operates simultaneously as an amplifier stage and a
cathode follower. The analysis derived thus far applies to cases of this
sort almost without change; all that is required is the replacement of rv in
all the relations by rv + Rl. This does, however, have the result, par¬
ticularly if Rl is large, that a number of the approximations made to
simplify the final expressions are no longer completely valid.
2.9. Equivalent Circuit for the Cathode Follower. In the analysis of
complicated circuits involving cathode followers, it is sometimes con¬
venient to make use of an equivalent circuit. Such a circuit may, of

Fig. 2.8. Cathode-follower equivalent circuit.

course, take any number of forms. It is merely required that the expres¬
sions for gain, drift, and impedance derived from the equivalent circuit
correspond to those derived previously. The circuit shown in Fig. 2.8 is
suggested for situations where only the gain, output impedance, and
drift rate due to Ebb, Ekk, and eh are of interest. This circuit does not give
the correct answer in problems involving component variations. The
boxes marked n, fi + 1, and l/(/x + 1) are perfect amplifiers with zero
output impedance. The input and output voltages, as well as the output
current, are to be thought of as voltage and current variations about some
fixed, unspecified quiescent value. For this reason it is not necessary to
70 CONTROL SYSTEM COMPONENTS [Chap. 2

include the intercept voltage Ex, which is always assumed to be constant.

Note that the equivalent circuit remains valid even if Rk is removed, i.e.,
becomes infinite. This fact has occasional practical significance.
2.10. The Triode Amplifier with Cathode Bias. Before deriving
expressions for the performance of the triode amplifier, it is convenient to

ia) Actual circuit id) Equivalent circuit

Fig. 2.9. Triode amplifier with cathode bias.

determine the effect of the cathode resistor. Thus, for the circuit shown
in Fig. 2.9, the plate current is

_ @p Ex ~t- f^^gk &h

(2.22)
yp
rv + Rk

(The voltage e'k is a cathode-signal voltage inserted for generality.) Since

^gk @g @h Rk^p &k (2.23)

the current becomes

• _ ep — Ex -f- neg — (n T l)(eft -T e'k) — \xRkiP

P tp + Rk
eP ~ (Ex + Ch + ek) + n(eg — ek — ek)
(2.24)
7V
where r'p = rp + (/x + 1 )Rk (2.25)

If Rk = 0 in Eq. (2.25), it is obvious that, for a tube without a cathode

resistor, the plate current is given by the same expression, except that
r'p = rv. Hence we arrive at a result of general importance: Whenever a
tube is viewed only from the three terminals—plate, grid, and lower end
of cathode resistor—it may be replaced by an equivalent tube whose plate
resistance is r'p = rp + (/x + l)Rk and whose other characteristics are
unchanged.
We make use of this fact to derive the performance equations for a
Sec. 2.10] D-C AMPLIFIERS 71
triode amplifier. The equivalent circuit is drawn in Fig. 2.10 in accord¬
ance with the results obtained in Eq. (2.24), and a node equation is
written at node ep, the output node. The voltage e'k is retained for
generality.
We have

ep — Ebb | ep — Ex veg — (eh + ej.)(l + aO • _ n

— "T" # 'Z-o (2.26)
Ri V

EbbEp + ExRl — nRLeg + (m + 1)-Rl(^a + e*) + RlEpi0

or ep (2.27)
Rl + t'p

To find the gain, we differentiate ep with respect to eg and obtain the

familiar result

q _ dep ijlRj P-R'i

(2.28)
de n Rl + E Rl + Tp + ([x + l)Rj,

The output impedance is

de-n RlK _ Rl[tp -|- (m T l)7th]

Zo = (2.29)
di0 Rl + E Rl + + (n + 1 )Ri

This could have been found also by inspection of Fig. 2.10, since all
voltage sources in that figure are assumed to have zero impedance.
The drift due to Ehb is

de p r:
dEbb Rl + En
rP T (/x -f~ 1 )Rk
(2.30)
Rl + rp -j- (n + l)Ri

This again could have been found

by inspection. The drift due to
heater-voltage variation is obtained by
differentiating ep with respect to en:

dei (m + 1 )Rl
(2.31)'
deh Rl + tp -j- (ix -f- 1 )Rji
Here we find again, as with the cath¬
ode follower, that filament-voltage Fig. 2.10. Equivalent circuit of triode
variation causes a very large amount amplifier with cathode resistor.
of drift, since the effective cathode
voltage en acts on the tube in approximately the same way as the grid
voltage, which is the desired signal.
Drifts due to variations of the components may be obtained by differ¬
entiating Eq. (2.27) with respect to the desired component. Specific
computations are left to the reader.
72 CONTROL SYSTEM COMPONENTS [Chap. 2

If the tube is operated as a grounded-grid amplifier with cathode input,

the voltage ek will become the signal voltage. The gain of the tube for
this type of operation may be computed from Eq. (2.27), and the result is
identical with Eq. (2.31):

dep _ (m + 1 )Rl
(2.32)
de'k Rl + rp + (yu + l)Rk

Comparison of this equation with (2.28) shows that the gain for the two
methods of signal introduction is about the same, the gain for cathode
input being (yu + l)//x times that for grid input. More important is the
fact that the sign change associated with amplification when the signal is
applied to the grid does not occur when the signal is applied to the cath¬
ode. This fact is of importance when it is desired to apply negative feed-
back around an amplifier.
One disadvantage of applying the
signal to the cathode is that the input
impedance is very low. The input
impedance may be deduced from one
of the results obtained for the cath¬
ode follower; i.e., it is the sum of Rk
and the output impedance of a cath¬
ode follower with a plate-load resistor
Rl and an infinite cathode-load re¬
sistor. Figure 2.9a shows the cir¬
cuit; the impedance in question is
that “seen” by ek. Hence

Z, = Rk + r* +,R,l (2.33)
Fig. 2.11. The difference amplifier. M + 1

This is usually much less than the input impedance at the grid.
2.11. The Difference Amplifier. We have seen that one of the most
serious causes of drift is the variation of heater voltage, with its influence
on the effective cathode voltage. A number of methods have been
devised to counteract this effect, one of the most useful being the differ¬
ence amplifier, shown in Fig. 2.11. The two triodes may be in separate
envelopes;however, the two sections in a single envelope are usually better
matched and will, therefore, give better results than two triodes picked at
random.
To analyze the circuit, we suppose that the cathode of the left-hand
section is disconnected at the point marked x in the diagram. The right
half of the tube is then a standard cathode follower, and if for the moment
we consider only the- effect of the grid voltage on the output, we have
Sec. 2.11] D-C AMPLIFIERS 73

_CgZlJ-zR’k_
(2.34)
0*2 + 1 )Rk + T p2

The subscript 2’s refer to the right-hand triode. Also the impedance to
ground seen by the left-half triode is the output impedance of the cathode
follower. It is given by

Z = (2.35)
(1 + M2)Rk + rP2
Hence the circuit may be redrawn in the equiv¬
alent form of Fig. 2.12. The output voltage
e0 for this circuit as a function of eg\ and eg2 is
given by the combination of Eqs. (2.28) and
(2.32), with Rk in both equations replaced by Z
of Eq. (2.35). Thus,

— niRLCgi T- On T 1 )R
L (/*2 + 1 )Rk + VV2
T p2Rk
Rl + Tp\ + Oil + 1)
(m2 T" 1 )Rk + Tp2
(2.36) for difference amplifier.

The subscript 1 refers to the left-hand triode. This equation may be

simplified by the assumptions that Ox2 + 1 )Rk » rp2 and that mi = M2 and
rp 1 = rp2; hence
nR L,(@g2 6gl)
(2.37)
Rl + 2 rp
It should be noted that, since the plate circuits of the two sections are
different, one Avould expect the triodes to have different operating points
and, therefore, different values for m and rp. Thus, although Eq. (2.37)
is usually remembered because of its simplicity, it is not accurate. It is,
however, correct to conclude that the output voltage is a function of the
difference of the two grid voltages. In many circuits this property is of
considerable use. For instance, one of the grids may be connected to a
variable bias supply for zero adjustment of an amplifier, or a feedback
voltage may be introduced into an amplifier through one of the grids, etc.
The effect of heater-voltage variations may be computed by inserting
voltages ehi and e^2 between the “ actual ” cathodes and the cathode termi¬
nals of the left- and right-hand sections, respectively. An analysis similar
to that carried out for egi and eg2 results in

(mi + l)RLehi — (mi + 1 )Rl 7—\ h2

(m2 + f)Rk ~r tV2

Rl + Tp\ + (mi + 1) -1 p2/-1 ' vv

rp 2 + Rk{n 2 + 1)
(m + 1)-Rl(cai — ehi ) (2.38)
Rl + 2 rp
74 CONTROL SYSTEM COMPONENTS [Chap. 2

where the simplification is again due to the assumption of large Rk and

equal ^ and rp in the two tube halves. The importance of this result lies
in the fact that it is the difference between the initial-velocity voltages
eni and that is amplified by the stage. This difference is, on the aver¬
age, at least ten times smaller than either one of the two voltages, par¬
ticularly if two triode sections in the same envelope are used. Thus this
circuit, while giving substantially the same amplification as a single triode
stage, results in a reduction of drift by a factor of ten or more.
The drift performance of this circuit for variations of the positive and
negative supply is not quite so favorable but is still quite acceptable.
The computation for Ebb can be carried out along the same lines as indi¬
cated for the grid voltages and offers no difficulty. The computation for
drift due to Ekk variations is best carried out by substituting the equiv¬
alent circuit of Fig. 2.8 for the cathode-follower section of the difference
amplifier. Both of these computations are left to the reader.
2.12. The Miller Circuit. Although the difference amplifier makes
possible a large reduction in the effect of heater-voltage variations, it will

[a) Complete circuit id) Right-holt section

Fig. 2.13. The Miller circuit.

not, in general, reduce the drift from this cause completely to zero.
Since no two tubes have exactly the same characteristics, a circuit which
is to be insensitive to heater-voltage variation must contain an adjust¬
ment to accommodate different tubes. Such a circuit was described by
Miller1 and takes the form shown in Fig. 2.13.
The analysis proceeds as for the difference amplifier. We assume the
existence of the heater-effect voltages em and ek2 in the two sections of the
tube and, for the moment, assume the left-hand section disconnected at
point x. The right-hand section may then be considered as a cathode

1 S. E. Miller, Sensitive D-C Amplifier with A-C Operation, Electronics, November,

1941, p. 27.
Sec. 2.12] d-c amplifiers 75
I
follower with a plate-load resistance. Its input voltage is

Ri _ R\
6g2 p i p &k p' (2.39)
III "T II2 /t/c

(see Fig. 2.135). The voltage at the break x may then be deduced by the
superposition of Eqs. (2.13) and (2.17). It is necessary, however, to
replace rv by rp2 + Rl2 in accordance with the discussion in Sec. 2.8. Also
eQ in Eq. (2.13) must be replaced by Eq. (2.39). The result is

ek =
— (^2 + l)Rkeh2 + /JL2Rk(Rl/Rk)€k

R k(ll 2 + 1) + rp2 + Rl2

_~ (m2 H~ 1)Rk6h2_
(2.40)
Rk(ll2 + 1) T" rp2 + Rl2 — M2A1
where Rk = R1 + R2.
We may now redraw Fig. 2.13a as Fig. 2.14, where Ra is the output
impedance of the cathode follower of
Fig. 2.135. It may be seen from this
figure that, as long as

(m 2 + 1 )Rk
Ch1 — 6h2
Rk(ll 2 + 1) + Tp2 + R L2 — II2R1
(2.41)

the net voltage in the cathode circuit of

the left section is zero, and the heater-
voltage effect is completely canceled.
Since the factor multiplying is an ad¬
justable constant, because of the presence
of R1, the circuit can be made insensitive
Fig. 2.14. Equivalent circuit for
to heater-variation effects as long as there the Miller circuit.
is a constant ratio between ehi and e^2-
This constant ratio exists for a fairly large range of filament voltages
around the proper operating value. Accordingly, if we let

Cm = ac^2 (2.42)
we can solve for the value of R1 that results in zero drift. The result is

Ri = i)(l +-) + rp2 + — (2.43)

af \ in/ M2
l\ rr2 + Rl2
~ Rk (1 (2.44)
a) + M2

The chief function of Rl2 is to prevent overloading of the right-hand sec¬

tion of the tube, and it must be large enough for this purpose. On the
other hand, excessively large values of Rl2 result in a decrease in gain for
76 CONTROL SYSTEM COMPONENTS [CHAP. 2

the circuit [see Eq. (2.49) below]; hence it should not be larger than
necessary.
In practice it is undesirable to make the cathode resistor, Ri R2 = Rk,
constant, as implied in the previous analysis, since any adjustment of the
variable tap affects the bias voltage of both grids, making it very difficult
to prevent either one or the other of the two tube sections from cutting off.
A more practical form of the circuit is shown in Fig. 2.15. R% and R±
together form the effective plate load of the right half section and serve to
reduce the effective Ebb seen by this sec¬
tion. Thus ip2 is kept small and does not
produce a large bias through the common
cathode resistor. R2 is constant and small,
just enough to keep grid 2 slightly nega¬
tive with respect to its cathode. (A small
amount of grid current in the right-half
section is actually of little consequence.)
Rli should be large, both for maximum
gain and for wide operating range, since
the adjustment of Ri required for drift
Fig. 2.15. Practical form of Mil¬
compensation does move the operating-
ler circuit.
point of the left-half section over fairly wide
limits. (It has only a small effect on the right half.)
The condition for perfect balance now becomes

_j_ rp2 + R'l2

R\ — {R\ T R2)
M2
(2.45)
~ R2(a — 1) -|-(rp2 + R'l2)

R$R 4
where
R% R4
The reader may convince himself, by using typical values in the circuit
of Fig. 2.15, that reasonable variations of a can be compensated for with¬
out causing tube cutoff.
In order to find the gain of the circuit or its sensitivity to changes in the
positive supply voltage, it is necessary first to compute the effective
cathode resistor seen by the amplifying section. This is best done by
writing an expression for the voltage change ek as a function of a current ik
flowing into the right-half section (see Fig. 2.136). Using Eqs. (2.13),
(2.14), and (2.39) and the fact that rp must be replaced by U>2 + R'l2j We
have
_ ikRk{rp2 T R'L2) , ii2Rk{R\/Rk)ek
Rk(v 2 + 1) + ^p2 + R'l-2 tf*(M2 + 1) + Tp 2 + R'l 2
_ 1_ikRk(rp2 T R'l2) ____
(2.46)
Rk(n2 + 1) + rP2 + R'l2 ~ M2-R1
Sec. 2.14] d-c amplifiers 77
*
Hence the effective cathode resistor of the left-half section (R0 of Fig. 2.14)
is
£k _ _Rk(rp2 T R'l2)_
(2.47)
ik Rk(n2 + 1) + 'f'p‘2 + R'l2 — M2^1

If the circuit is properly adjusted to eliminate drift due to heater-voltage

variations, we may use Eq. (2.43) to eliminate R\ with the simple result
that

R„ = (2.48)
M2 + 1

where a is defined as in Eq. (2.42). Now the expression for gain found
previously for the triode amplifier with cathode bias [Eq. (2.28)] may be
used, with the result:

de7 M 1R LI
G =
der, R LI + Tpi + (mi + l)(^p2 T" RfL2)a/ (M2 + 1)
HiRli
if mi = M2 (2.49)
Rli + Tpi + a(r P2 + R'lz)

The computation of the drift due to changes in Ebb is slightly more

involved, since the drift is due both to the direct effect of Ebb on the plate
voltage [as in Eq. (2.30)] and to its “indirect” effect on the cathode volt¬
age ek of the right-half section and thereby on the left-hand plate voltage.
The details of this computation are left to the reader. For the properly
adjusted circuit the result is

dep ~ rpi -h cl{Rli H~ rV2 H~ R'l2)

(2.50)
dEbb rp i + Rli + u(rp2 + R'l2)

where we again assume that mi = M2- It is clear that, for a = 1, Eq

(2.50) becomes Aep = AEbb. Hence the Miller circuit does not compen
sate for changes in the positive supply voltage very
well, and for best results the positive supply must be
closely regulated.
2.13. Phase Inverters. Whenever it is possible,
push-pull amplifiers are used because of their relatively
drift-free operation. However, if the input signal is
with respect to the ground, “single-ended,” a phase
inverter is required to convert the signal into the
“double-ended” form required in push-pull circuits.
A number of fairly satisfactory circuits are available Fig. 2.16. The

for this purpose. cathode-follower

inverter.
2.14. The Cathode-follower Inverter. The simplest
form of inverter is the cathode-follower type shown in Fig. 2.16. For the
output voltage eai the circuit looks like a triode amplifier with cathode
78 CONTROL SYSTEM COMPONENTS [Chap. 2

bias; hence Eq. (2.28) applies and results in

e0i _ /j.R ^ ..
(2.51)
eg (jj. -f- 2 )R + rp
The output voltage eo2 is obtained by treating the circuit as a cathode
follower with plate-load resistance R; hence Eq. (2.13) may be used and
gives
@o2 flR
(2.52)
eo (m + 2 )R + rp
Hence we have here exactly that e0\/eo2 = — 1, the ideal inverter action.
Unfortunately, if eg is near ground potential, e0i is at a relatively high
positive d-c level, whereas eo2 is near ground. This is no disadvantage in
an a-c amplifier, but since most d-c amplifier coupling networks introduce
some signal loss, the voltage e0i will generally be attenuated somewhat
before it can be applied to a push-pull stage. This effect may, of course,
be compensated by making the plate-load resistor larger than the cathode
resistor, but then the circuit loses
some of its simplicity. Another dis¬
advantage of this circuit is that it
has no gain. Its major advantage
is its simplicity and the fact that only
one tube is required.
2.15. The Paraphase, or Balanced,
Inverter. This inverter takes a form
similar to that shown in Fig. 2.17,
the common circuit used in a-c am¬
Fig. 2.17. The paraphase inverter.
plifiers. Replacing the capacitor by
any of the d-c coupling networks described below in Sec. 2.17 will make
this circuit applicable to d-c operation.
The circuit analysis is simplified if it is assumed that
1. The two tubes are identical; i.e., rp and n are the same.
2. Ri = R2 » Rl.
3. Rg » R\.
4. The frequency is high enough for the impedance of the coupling
capacitor to be negligible.
5. The inverter action is sufficiently good that the current in the cath¬
ode resistor Rk is constant; hence the cathode voltage ek is constant.
The following three equations may then be written for this circuit:
_ — hRl
(2.53)
6pl ~ flz. + rp e“'
_ —/jlRl
(2.54)
ep2 ~ Rl + r„
(2.55)
Sec. 2.16] d-c amplifiers 79

Substitution of (2.55) in (2.54) yields

— /jlRl epi + ep2 — fiR L

6p2 @p i (2.56)
Rl + rp 2 hRl 4“ 2Rl d- 2r.

If is. very large, this becomes approximately ep2 = — eph a typical

value of — ep2/epi being of the order of 0.95. Thus the circuit is not a per¬
fect inverter; however, it supplies gain, and the two voltages epi and ep2
are both at the same level relative to ground. A disadvantage of the cir¬
cuit is that it is essentially a feedback circuit, and under certain condi¬
tions, particularly when neon-tube couplers are used (see Sec. 2.17), it
may oscillate at a high frequency. This can often be corrected, at the
expense of poorer frequency response, by connecting a small capacitor
between the second grid and ground. It can be shown that the inverting
action can be made perfect, i.e., — ev2/evi can be made exactly unity, by
making R2 slightly greater than R i. The precise relation is

R2 T" Ri _ ijlRl
(2.57)
R2 — Ri Rl 4~ tp

2.16. The Cathode-coupled Inverter. A relatively simple inverter,

having most of the advantages and only few of the disadvantages of the
two inverter circuits discussed thus far,
may be obtained by modification of the
difference amplifier described in Sec.
2.11. The modification consists of in¬
serting equal plate-load resistors in both
plate circuits (Fig. 2.18) and grounding
one of the grids. (In practice the grid
shown grounded may be connected to a
variable-bias voltage and may serve as
a balance adjustment.) If the two
tubes shown are the two sections of a
double triode, it is a good assumption Fig. 2.18. Cathode-coupled inverter.
that rp and n for the two sections are the
same. Note that theoretically the two sections have the same operating-
point in this circuit.
The equations derived for the difference amplifier apply here, except
that rp2 must be replaced by rp + RL. If we make this substitution in
Eq. (2.36) and let eg2 = 0, we have

_ jlR L@i
(2.58)
Rl 4~ 4“ (m 4“ l)Iik-(rP 4~ Rl)/[(h 4~ 1 )Rk 4- rp + Rl]
80 CONTROL SYSTEM COMPONENTS [CHAP. 2

In computing ep2, becomes the eg2 of Eq. (2.36) and egi is zero; hence

_(m H~ l)Rk/[(n T~ l)-^fc ~T rP -|~ Rl]_

@p2 (2.59)
Rl + Tp + (m T" 1)RkiXP "b Rl)/[(h + 1 )Rk + f'p + Rl]
Cp2 _ _ _(/X ~t~ 1 )Rjc_
Hence (2.60)
£pi (m + 1 )Rk + tp + Rl

Thus this circuit is also not a perfect inverter. However, an increase of

Rk will improve the inversion characteristics, and in practice it is found
that this circuit inverts at least as well as the paraphase circuit. The cir¬
cuit has the further advantages of a balanced output, high gain, sim¬
plicity, no tendency to oscillate, and the possibility of very simple balance
adjustment by use of the second grid.
Where extremely accurate inverting characteristics are required, it is
possible to make Rk effectively very large by using a triode as the cathode
resistor, as in Fig. 2.19. A pentode
would be even better, but the difficulty
of supplying its screen voltage and the
fact that a triode gives almost perfect
inversion usually make the triode more
desirable.1 The effect of this addition to
the circuit is simply to replace the Rk in
Eq. (2.60) by rp3 + (ju3 + 1 )Rk, the resist¬
ance seen from the plate of the third tube.
Thus this tube results in a very large
effective cathode resistor without the dis¬
advantage of the very large negative
Fig. 2.19. Cathode-coupled in¬
verter with constant-current tube
supply voltage that would be necessary
in cathode return. to obtain this same value of Rk without
the tube.
2.17. Interstage Coupling Networks. In addition to the problem of
drift, the other major problem that distinguishes d-c amplifiers from a-c
amplifiers is the design of interstage coupling circuits. These circuits
must pass the signal with minimum attenuation and at the same time
permit each stage to operate at the correct quiescent level. When the
cathodes of all stages are grounded, the requirement is essentially to
reduce the quiescent level of the plate of the first stage to that of the grid
of the second stage without undue loss of signal.

1 Another constant-current device that could be used here is the temperature-

limited diode. This is a diode whose filament is operated at reduced voltage, so that
the plate current is determined by cathode emission and is practically independent
of plate voltage. This method, although theoretically very simple, would require a
very closely regulated filament voltage for the temperature-limited diode, since the
circuit would otherwise exhibit excessive drift.
i
Sec. 2.17] D-C AMPLIFIERS 81

The simplest method for doing this is the bias battery (see Fig. 2.20).
If the quiescent voltage of the plate is eb volts and that of the grid ec volts,
the battery voltage must be eb — ec volts. Ideally the battery has zero
internal impedance and perfect voltage
stability; under these conditions the cou¬
pling is close to ideal and has the following
advantages:
1. No signal loss.
2. No power-supply drain.
3. Low-impedance coupling, hence good
frequency response and small noise pickup.
Fig. 2.20. Interstage coupling
4. Coupling introduces no time lag or with bias battery.
frequency distortion.
In practice this coupling has several serious disadvantages, and it is used
only very rarely. These disadvantages are:
1. Batteries have limited life.
2. Voltage does not remain constant, results in drift.
3. Practical batteries are physically large; hence the capacitance to
ground is large, and frequency response is adversely affected. Bulkiness
is also a very important disadvantage
when miniature size is a requirement.
Another coupling that can be used is
the gas-tube coupling (Fig. 2.21). In a
voltage amplifier only very little power
will be handled by the coupling, and for
this reason the smallest neon tubes, such
as the NE-2, are used. These tubes have
a voltage drop that is almost completely
independent of the current through the
Fig. 2.21. Gas-tube interstage tube, as long as the region of normal glow
coupling network.
is not exceeded.1 The voltage drop va¬
ries considerably from tube to tube, rang¬
ing from about 49 to 74 volts with an average value of 62 volts. The
current may vary from 0.03 to 0.3 ma, but the design value is 0.1
ma. In the design of this coupling, one must keep in mind that
glow tubes require a relatively high voltage to strike the arc. For the
NE-2 neon tube this is about 90 to 100 volts. Although the voltage is
nominally independent of current, there is a small increase as the current
is increased; the effect is as though there were an internal resistance of
about 10,000 ohms. The fact that ionization is involved in the conduc¬
tion processes through the tube gives rise to a very pronounced time lag

1 Reich, “Theory and Applications of Electron Tubes,” McGraw-Hill Book Com¬

pany, Inc., New York, 1944, p. 417.
82 CONTROL SYSTEM COMPONENTS [Chap. 2

between applied voltage and resulting current at frequencies of about

4,000 cps and higher. This effect may be represented approximately by
an inductance whose value depends on the parameters of the rest of the
circuit but may reach values of several henrys.
The coupling has the following advantages:
1. Full gain
2. Low-impedance coupling
3. Low power-supply drain
4. Possibility of using only a small negative supply voltage
There are also a number of disadvantages, however, some of them, such
as 1, 2, and 3 in the following list, sufficiently serious to outweigh the ad¬
vantages in many critical applications. These are disadvantages:
1. General unreliability of arc tubes.
2. Small instability of glow results in a certain amount of internally
generated noise.
3. Occasionally the circuit breaks into spontaneous relaxation oscilla¬
tions.
4. Poor high-frequency response due to large inductive effect referred
to above.
5. If negative supply voltage is not large enough, glow tubes may some¬
times go out during very large signal swings, thus causing very heavy
distortion.
In recent years a new and highly efficient coupling device has been
developed in the form of the so-called Zener diode. This is a silicon-
junction rectifier operated in the reverse direction. Such rectifiers show
a very sharp inverse breakdown characteristic in which the voltage is
almost exactly constant for large variations of current. Zener diodes are
available in voltage ratings ranging from 2 to over 600 volts and with
current ratings of 10 ,ua to 10 ma. Usually the higher-voltage units must
be operated with smaller currents to avoid excessive power dissipation in
the rectifier junction. Zener diodes are much more reliable than gas tubes
and have a much smaller effective resistance. The resistance is a function
of the voltage for which the diode is designed, typical values being about
5 ohms for a 6-volt diode, and 1,000 ohms for a 60-volt diode. Zener
diodes generate a certain amount of noise when operated near the knee of
their characteristic; however, if sufficient current is permitted to flow
through them (typically about 0.2 ma in small diodes) their noise can be
reduced to a negligible value. Another disadvantage is that the break¬
down voltage is somewhat temperature-dependent. The temperature
coefficient for diodes having approximately a 5-volt rating is zero, so that
it is theoretically possible to obtain a coupling with zero temperature
dependence simply by connecting a sufficient number of 5-volt diodes in
series. However, the temperature coefficient of higher-voltage units is
Sec. 2.17] D-C AMPLIFIERS 83

so small that in most applications this refinement is not necessary.

Zener diodes are obtainable from most of the manufacturers of semicon¬
ductor devices, such as Texas Industries, Transitron, etc. Many of these
manufacturers also publish application notes in which the characteristics
of the diodes are described in considerable detail.
A very commonly used coupling is the resistance coupling shown in
Fig. 2.22. With this coupling there is a signal loss which becomes very
marked if Ekk is not large. The coupling -of*/bb
requires a considerable current, which has
the result of decreasing the effective supply f R1
*-AW—
voltage to the plate of the first stage (see Sec.
2.25). If the impedance level of Ri and R2 is A
-J 1 a
Vr-
raised to minimize the current drain, a high-
impedance coupling results; this is more sus¬
-of.kk
ceptible to noise pickup. In conjunction
Fig. 2.22. Resistance inter-
with the input capacitance of the following
stage coupling.
stage, this resistance also represents a low-
pass RC filter with a relatively small passband. This results in poor high-
frequency response, and although it is possible to compensate for this by
bridging Ri with a capacitor, an optimum design must be made. This is
usually a trial-and-error procedure. Despite these disadvantages, this
coupling is very commonly used, because it is much more rugged and reli¬
able than the first two described.
If additional tubes are no disadvantage, the level changer1 is occa¬
sionally convenient. One form of this
circuit, in which a triode is used, is
shown in Fig. 2.23. The level changer
consists of the tube V3 and the resistors
Ri, R2, R3, and Rk. The tube acts es¬
sentially as a constant-current device;
hence the IR drop through R1 is con¬
stant, and no signal loss occurs. Ad¬
justment of the level may be made by
adjusting the grid tap of the level
Fig. 2.23. Triode level changer.
changer. R1 must be large to prevent
excessive current drain, and the tube,
as a constant current source, has ideally an infinite resistance [actually
rp + (/j, + 1)/2J. Therefore this coupling is a high-impedance coupling.
Hence, except for the fact that the signal is transmitted with no loss, this
coupling has the same disadvantages as the resistance coupling.
Up to now we have considered only circuits in which all the cathodes
1 See Chestnut and Mayer, “Servomechanisms and Regulating System Design,”
John Wiley & Sons, Inc., New York, 1955, vol. II, pp. 172-173.
84 CONTROL SYSTEM COMPONENTS [Chap. 2

are grounded. However, it is sometimes possible, particularly in push-

pull circuits, to permit the cathode voltages of successive stages to become
more and more positive, so that it becomes possible to connect the plate
directly to the grid of the following stage. This represents an ideal
coupling, since it does not attenuate the signal and introduces no imped¬
ance. An amplifier embodying this principle is shown in Fig. 2.24.

2.18. The Transistor D-C Amplifier. For many control applications

transistor d-c amplifiers are very much superior to vacuum-tube ampli¬
fiers. Some of the obvious advantages of transistors are their small size
and their low power requirements. Since they have no filaments, the
large amount of power wasted in heating vacuum-tube filaments may be
saved in a transistor amplifier. This reduces the problem of cooling,
which becomes quite formidable when many vacuum tubes are crowded
into a small space. Also, the absence of filament power removes one
source of noise, the 60-cps hum picked up from the filament leads in
vacuum-tube circuits. An additional power saving comes from the fact
that low-level transistor-amplifier stages operate with collector voltages
of a few volts whereas typical plate voltages used on vacuum tubes are on
the order of several hundred volts. Transistor amplifiers are potentially
much more reliable than vacuum-tube amplifiers. The life of a well-
designed transistor is well above 50,000 hours as compared with the 2,000
hours that is standard with vacuum tubes.
Transistors also have, of course, some disadvantages compared to
vacuum tubes. The major disadvantage is their sensitivity to tempera¬
ture variations and the fact that they cannot be used at all at very high
temperatures. For germanium transistors the maximum operating tem¬
perature is usually well below 100°C, and even for silicon transistors oper¬
ation at temperatures much above 150°C is usually undesirable. This
disadvantage severely limits the application of transistors to control sys¬
tems. A less important disadvantage is that transistors are electrically
Sec. 2.19] D-C AMPLIFIERS 85

less rugged than vacuum tubes; i.e., they are easily damaged or destroyed
by relatively minor overloads. A further disadvantage is the large varia¬
tion of parameter values among units having nominally the same charac¬
teristics. This problem will undoubtedly become less severe as manufac¬
turing and selection processes are improved. Transistors cannot be used
at very high radio frequencies but this is of no consequence in a control
system. Finally, transistors are extremely sensitive to nuclear radiation.
Since transistor action takes place by means of minute atomic impurities
(the positive and negative charge carriers referred to below), any dis¬
turbance of this action by radiation particles has a significant effect on
the operation of the device.
A junction transistor may be thought of as a sandwich of three sections
of semiconductor crystal. The central section is referred to as the base
while the outside sections are the emitter and collector respectively. In
most transistors the junction between collector and base has a larger area
than that between emitter and base, but except for this the junctions are
identical. There are two types of transistors. In the p-n-p type the
collector and emitter consist of material in which the electric current is
carried by holes, or positive charge carriers, while in the base the cur¬
rent is carried by electrons or negative carriers. The other transistor type
is the n-p-n, where the base consists of positive material and the emitter
and collector of negative material.
The operation of a transistor may be explained by thinking of the two
junctions as rectifying junctions. The collector junction is biased in the
reverse direction, and in the absence of emitter current only a very small
inverse current flows in the collector circuit. The emitter junction is
biased in the forward direction, so that only a small voltage between
emitter and base is needed to cause relatively large emitter currents to
flow. The flow of emitter current through the base results in the produc¬
tion of a large number of charge carriers within the base, and these are
picked up by the collector, resulting in a flow of current in the collector.
Thus, we may define transistor action as the control of current in the
negatively biased collector junction by current in the positively biased
emitter junction. (The terms “positive” and “negative” are correct for
p-n-p transistors; they would be reversed for n-p-n transistors.) An
important transistor parameter is a, the ratio of collector current to
emitter current. In junction transistors a is slightly less than one, typical
values ranging from 0.95 to 0.98. In point-contact transistors a normally
exceeds one.
2.19. Analysis of Simple Transistor Circuits. In analyzing transistor
circuits it is convenient to assume small signal operation about a fixed
operating point just as in vacuum tubes. In the following discussion we
consider only junction transistors, since point-contact transistors are not
86 CONTROL SYSTEM COMPONENTS [Chap. 2

used in control amplifiers. Also, the transistors will be assumed to be

of the p-n-p variety unless a statement to the contrary is made.
Consider a transistor connected, as shown in Fig. 2.25a, with the base
grounded. This circuit may be analyzed as a standard four-terminal
network if we assume that the emitter, base, and collector all have an
electrode resistance re, rb, and rc
ic
respectively, and that in addition,
the collector voltage v2 for fixed
collector current ic is given by the
product of the emitter current ie
and a fictitious resistance which,
for the moment, will be designated
by rx. We obtain
ie
vi = (re + rb)ie + rbic (2.61)
re+rb-
V2 = rxie + (rb + rc)ic (2.62)
rb^ rc It should be noted that none of
.0 oa, the four resistances, re, rb, rc, or rx,
Wc
have any physical existence within
[b) the transistor; they are in fact
defined by Eqs. (2.61) and (2.62).
Explicitly the definitions are
re+rb rb+rc
v\ A dvi

Wc
0 v
ocu Ie
rb =

A
die
dvi
ie = const.
(2.63)

= — rb (2.64)
di e ic = const.
ic)
Fig. 2.25. Transistor in grounded-base
and
connection: (a) actual circuit; (5) a-c dV‘
rc. — — rb (2.65)
equivalent circuit; (c) d-c equivalent dir ie — const.
circuit showing lco.
In the definition of rx we prefer to
introduce the current gain a already mentioned above. For the directions
assumed for the currents ic and ie, this parameter is defined by
dic
a = — (2.66)
dL V2 = const.

since the actual collector current in a p-n-p transistor is out of the collec¬
tor. From Eq. (2.62) we obtain that
die
(2.67)
die V2 — const. rb -F rc
Hence we have that
= a
rb + rc
or rx = a(rb + rc) (2.68)
Sec. 2.19] D-C AMPLIFIERS 87
A possible equivalent circuit from which Eqs. (2.61) and (2.62) could have
been obtained directly is shown in Fig. 2.256. This equivalent circuit is
not quite correct for d-c amplifiers since it indicates that ic = 0 when
ie = 0. In practice a small current Ico flows in the collector circuit even
when ie = 0. To account for this current the circuit of Fig. 2.256 may
be modified as shown in Fig. 2.25c.1 The magnitude of Ic0 depends on
the transistor and the temperature. In a typical transistor (General
Electric 2N43A) Ic0 ranges between 5 and 10 ^a at a temperature of

Fig. 2.26. The grounded-emitter circuit: (a) actual circuit showing load resistance and
battery; (6) equivalent circuit.

25°. The temperature dependence of Ico is one of the major factors caus¬
ing transistor d-c amplifiers to drift. The variation of Ic0 is given
approximately by
La = I cooek{T~ro) (2.69)

where Icoo = Ic0 at the temperature T0, and k varies from 0.06 to 0.09 with
an average value being 0.08.* This relation is valid for T0 of about 25°C
and T — T0 less than 50°C. It should be noted that for k = 0.07, Ico
doubles for every 10°C change in temperature.
Since the grounded-base connection discussed up to now has a current
gain of less than one, direct coupling of stages cannot produce current
gains greater than one, and the voltage or power gain for a multistage
amplifier would therefore be the same as for a single stage. For this
reason the most commonly used connection is the grounded-emitter con¬
nection shown in Fig. 2.26a. This connection is closely analogous to the
grounded-cathode connection for vacuum tubes. To obtain the equa¬
tions for this connection we may redraw the equivalent circuit of Fig.
2.256 in the form shown in Fig. 2.266. Note that the input current is
now ih = ii = —ie — ic, and the output current is ic = as before.
While it would be possible to write the equations for this circuit in the
same form as Eqs. (2.61) and (2.62), i.e., with the voltages as the depend-
1 Hunter, “ Handbook of Semiconductor Electronics,” McGraw-Hill Book Com¬
pany, Inc., New York, 1956, pp. 13.2-13.4.
* Ibid.
88 CONTROL SYSTEM COMPONENTS [Chap. 2

ent variables, it is desirable to write them in a different form in which the

input voltage and output current are the dependent variables:
higl\ | flre^ 2 (2.70)
f2 = hfei 1 T /ioe^2 (2.71)
The A 'parameters appearing in these equations can be defined in terms of
re, rb, rc, and a used above.1 A possible procedure is to first set up the
equations for the equivalent circuit of Fig. 2.266 with Vi and v2 as the
dependent variables:
Vi = (re + rb)i i + rei2
v2 = [re + rh — a(rb + re)]ii + [rc + rb + re — a(rb + rc)]i2
Rearranging gives
— (re + rb)i i = —Vi + rei2
v2 — [re + rb — a(rb + rc)]n = [rc + rb + re — a(rb + rc)]i2

Solving for and i2 results in Eqs. (2.70) and (2.71), and by direct com¬
parison we find that
re[re + rb — a(rb + rc)] re
hie = re + rb — + rb
rc + rb + re — a{rb + re) 1 a
re re
hre
rc + rb + re - a(rb + rc) rc( 1 — a)
re + rb — a(rh + rc) a
hf e (2.72)
rc + rb + re — a(rb + rc) 1 — a
1 1
hoe
re + rb + rc — a(rb + rc) rc(l — a)

In making the approximations, advantage has been taken of the fact that
rc is usually very large, in the megohm range, while rb and re are typically
less than 1,000 and 50 ohms respectively.
The equations for the transistor in terms of the h parameters and the
subscript notation used are now being accepted as standard by industry.
The subscripts i, r, f, and o refer to input, reverse, forward, and output
respectively, while the subscript e refers to the fact that the parameters
are defined for the grounded-emitter connection. For the grounded-base
and grounded-collector connections the subscripts 6 and c are used. A
double subscript notation using the numbers 1 and 2, i.e., hu, hi2, h21, h22,
instead of h{, hr, hf, h0, is also in common use, and for some applications
z and y parameters are occasionally used.2

1 The h parameter representation is one of the standard ways of representing a four-

terminal network. Other methods include z parameters, y parameters, etc. See
Guillemin, “Communications Networks,” John Wiley & Sons, Inc., New York, 1935,
pp. 134-138.
2 Hunter, op. cit., sec. 11. Also see Shea, “Transistor Circuit Engineering,” John
Wiley & Sons, Inc., New York, 1957, p. 22.
Sec. 2.19] D-C AMPLIFIERS 89

In terms of the h parameters, the equivalent circuit of the transistor

takes the form of Fig. 2.27a. Typical values for the h parameters are
also given in this figure.1 Note that hie is a resistance while hoe is a con¬
ductance. The advantage of the h parameter representation is that in
most practical circuits the load admittance 1 /RL, which effectively shunts
hoe (see Fig. 2.26a), is much greater than hoe, so that hoe can often be

<3
M

Typical values:
hie ~ 2,000 ohms
V\ hre - 6x 10'4
hfe — 50
hoe- 2.5 x 10~5 mhos
o o

io)

Uf)
Fig. 2.27. Equivalent circuit for grounded-emitter circuit in terms of h parameters:
(a) exact circuit; (6) approximate circuit.

neglected. Similarly the voltage hreVz in the input circuit is usually much
less than the voltage drop through hie, so that the input impedance is
approximately hie. Thus the very simple approximate equivalent circuit
of Fig. 2.27b results. In this circuit the input circuit is completely
decoupled from the output, making the transistor a unilateral current
amplifier with the current gain hfe.

-15
-A. = PSD u nmn
^-
200
o -10
E 150
S" 100
-5
50
-0
0
0 -10 -20 -30
TC I
volts

Fig. 2.28. Characteristic curves for grounded-emitter connection.

Most manufacturers now supply values for the h parameters of the

transistors they produce rather than values for re, rb, or rc. Also, the h
parameters can be read directly from the characteristic curves usually
furnished. A typical set of characteristics is shown in Fig. 2.28. Note
1 Hunter, op. cit., p. 11.6.
90 CONTROL SYSTEM COMPONENTS [Chap. 2

that two families of curves are required: an input family and an output
family. From the input family one obtains hie, the slope of the curve at
the chosen operating point, and hre, the vertical distance between two
curves for which the difference in collector voltage is unity. Similarly,
from the output family one obtains hfe, the vertical distance between
curves for which the difference in base current is unity, and hoe, the slope
of the curve. From the shape of the curves it is apparent that the
parameters hie and hfe can be obtained with good accuracy, while hre and
hoe are difficult to get accurately. Fortunately, the important parameters
are the ones that can be obtained accurately. It should also be noted
that the output characteristics are almost parallel and equidistant for
constant increments of base current. This means that the transistor is
a good linear amplifier of the input current; i.e., a sinusoidal input current
will result in sinusoidal output current or voltage. On the other hand,
due to the curvature of the input characteristic, a sinusoidal voltage
applied to a transistor will in general yield a nonsinusoidal output. It is
therefore usually desirable to drive the base from a high-impedance, or
constant-current, source. It should be pointed out that the h parameters
are basically small-signal parameters just like the n, rp, and gm of a vacuum
tube, and that they vary with the operating point. They also vary with
the operating temperature. Thus the typical values given in Fig. 2.27
can vary by factors of two or three or more. Most manufacturers of
transistors do, however, supply charts showing what this variation is.
The equivalent circuit of Fig. 2.27a again does not take into account
the fact that with zero base current the collector current is not zero. In
order to obtain an equivalent circuit that properly takes into account all
of the d-c conditions of a transistor, we may go back to the circuit of
Fig. 2.25c, and convert it to the common-emitter form in terms of h
parameters. The details of this conversion are left to the reader, but it
results approximately in introducing an additional voltage source
— Icohfehre/hoe into the input circuit, and an additional current source
— hfeho into the output circuit (see Prob. 2.9).
2.20. Single-stage Circuits. The current gain, voltage gain, power
gain, input impedance, and output
impedance of a transistor may be
easily obtained from the four-ter¬
'R, minal representation shown in Fig.
2.29. In this representation it is as¬
sumed that the currents i\ and i2 and
Fig. 2.29. Four-terminal network. the voltages V\ and v2 are related by
Eqs. (2.70) and (2.71).
In order to find the current gain, we note that in Fig. 2.29, v2 = —i2Rl-
Substituting the value of v2 in Eq. (2.71), we immediately find that
Sec. 2.20] D-C AMPLIFIERS 91
hf e
(2.73)
1 "f" hoeR’L

The approximation is justified since typically Rihoe 1. (See typical

values of the h parameters in Fig. 2.27a; Rl is usually less than 5,000
ohms.)
To obtain the voltage gain we replace i2 in Eq. (2.71) by —v2/Rl, and
solve Eqs. (2.70) and (2.71) simultaneously. The result is

hfe{R l/hie)
(2.74)
Vi h fe'vre
{eh
1 T" Rlhoe
hneh
oe,,jie

Here again a simple approximate result may be obtained by letting

RjJnoe <jC 1, and noting that hfehre/hoehie is typically about 0.6. The
approximate result is
Gv ~ hfe(Rl/hie) (2.75)

By a similar procedure we may find that the input impedance may be

written
(hfehre) (Ri^hce)
(2.76)
h {hiehoe) (1 —t- RL,hoe)

and the output admittance is

(hfehre) {hie)
y. = = h oe 1 - hoe (2.77)
Vi {hiehoe) {hie ~b Rg) _
The power gain is simply the product of current and voltage gain. It
should be noted that the expressions for current gain, input impedance,
and output impedance reduce approximately to one of the h parameters.
Also the voltage gain is approximately equal to the product of current
gain and ratio of load resistance to input resistance. These approximate
results could have been obtained directly by use of the approximate
equivalent circuit of Fig. 2.27h.
Although the grounded-emitter circuit is the one used most often, the
grounded-collector and grounded-base circuits are also used occasionally.
The form of these circuits for p-n-p transistors is shown in Fig. 2.30. For
n-p-n transistors, the battery voltage should be reversed. The simplest
way to obtain the various gains and impedances for these circuits is first
to compute new h parameters for them. These h parameters may then
be used instead of the common-emitter parameters in Eqs. (2.73), (2.74),
(2.76), and (2.77). This is possible since these equations describe the
characteristics of the generalized four-terminal network of Fig. 2.29. It
must be noted, however, that the approximations made in these four
equations apply in general only to the grounded-emitter connection.
92 CONTROL SYSTEM COMPONENTS [Chap. 2

As an example of the process of obtaining the modified h parameters, we

consider the common collector circuit of Fig. 2.306. By comparison with
Fig. 2.26a we note that
^ le ^ 1c ^ 2c

^2e P 2c

%le he

he he he

where the subscripts e refer to grounded emitter and c to grounded collec¬

tor. Rewriting the equations for the grounded-emitter connection, Eqs.

Fig. 2.30. (a) Grounded-base circuit; (6) grounded-collector circuit.

(2.70) and (2.71), and rearranging results then directly in

^ lc hiehc “h (1 hre)V 2c

he (1 —b hfe)hc ~b hoeV 2c

by direct comparison, we obtain:

h^c hiie
(2.78)
hfc — — (1 + hfe)
hoc

By a similar but somewhat more complicated process, the common-base

parameters may be obtained in terms of the common-emitter parameters.
For the common-base connection we have approximately

hib — hie/ (1 + hfe)

hieh
le'^oe
hrb ~ h re
1 + hfe
hfe
hfb —a = (2.79)
1 + hfe
hoe
hob
1 "b hfe
The approximations used in obtaining the common-base parameters are
that 1 » hre and that hieh0e <3C hfe. Both of these are justified as may be
seen from the typical values of the he parameters in Fig. 2.27a.
Sec. 2.20] D-C AMPLIFIERS 93

The modified h parameters may now be used in Eqs. (2.73) to (2.77) to

find the current and voltage gain, input impedance, and output admit¬
tance for the various connections. The exact expressions for these char¬
acteristics will have exactly the same appearance for all three connections,
if the modified h parameters are used, but the approximations that can
be made are quite different for the three connections. In Table 2.1 the

Table 2.1

Approximations
Zi Zo Gi Gv

•
made

Common hie 1 + hfe hfe RLhoe 1

(1 + hfe )
base 1 + hfe hoe 1 + hfe 'vie Rg/hie » 1
0 < *'•£” < 1
fvie'voe

Common 1 , Rl RLhoe « 1
hie hfe hfe h.
emitter hoe ttie Rg/hie » 1

° < h/ir’
flie'voe
<1

Common Rg RLhoe <'5C 1

Rl(1 + hfe) — (1 + hfe) 1
collector 1 + hfe Rg/hie » 1
(1 -f- hfe)RL/hie 1
(1 + hfe)/(Rghge) 1

approximate characteristics for all three connections are summarized.

The expressions are all in terms of common-emitter h parameters. The
approximations made for each connection are also given. It should be
noted that these approximations are not applicable in all circuits of prac¬
tical importance, particularly the requirement that Rg/hie 1 is not
always met. We note that the input impedance is very small (ca. 50
ohms) for the grounded base, intermediate for the grounded emitter, and
highest for the grounded collector. The lowest output impedance is
found in the common collector, and the highest in the grounded base.
Only the common-emitter connection has a voltage and current gain
greater than unity; this is the chief reason why it is used more frequently
than the other two connections.
It is often necessary, particularly in d-c amplifier circuits, to consider
circuits having resistances other than the load resistance Rl. Particu¬
larly important is the common-emitter circuit with a resistor in the
emitter lead, the common emitter with a feedback resistor between collec¬
tor and base, and the common collector with a resistor in the collector
lead. All of these modifications of the basic circuits are most conven¬
iently handled by computing modified h parameters for them. This is
94 CONTROL SYSTEM COMPONENTS [Chap. 2

done in a manner similar to that explained in connection with Eq. (2.78).

A complete list of modified h parameters is given by Hunter.1 In general
the effects of additional resistors of this sort are the same as in vacuum-
tube circuits. For instance, a resistor in the emitter lead of a common-
emitter connection results in a marked decrease in voltage gain, and an
increase in the input and output impedance.
2.21. Drift in Transistor Amplifiers. The main cause of drift in tran¬
sistor amplifiers is the temperature variation of Ico, the collector-to-base
current that flows when the emitter current is zero. The variation of this
current with temperature is given in Eq. (2.69). As was pointed out in
connection with this equation, Ico approximately doubles for every 10°C
change in temperature. A second effect causing drift is the change
of the h parameters themselves with temperature. This is quite severe in
some transistors. The drift due to Ico may result in a condition of thermal
runaway where an initial increase of Ico due to an external temperature
rise results in a further temperature rise in the transistor, which, in turn,
results in a further increase in Ico, etc., until the transistor either destroys
itself or drifts into a relatively inoperative condition. Even if thermal
runaway does not occur, the variation in the h parameters and in Ico
generally tends to shift the operating point of the transistor away from
the optimum value, and thus to magnify the drift.
There are several ways of combating drift. An obvious method of
alleviating the problem is to operate the transistors with a large enough
collector current so that Ico is only a small fraction of the total collector
current. In this way the percentage change of collector current resulting
from a given change in temperature may be reduced. A related method
is to design the networks that supply bias to the terminals in such a way
that changes in Ico result in only small changes of operating point. Shea2
defines a stability factor for current as the ratio of a change in the
quiescent operating value of the emitter to a temperature-induced varia¬
tion in Ico. Similarly, he defines a voltage stability factor for the col¬
lector-to-base voltage. These stability factors can be expressed in terms
of the resistors of the biasing network and the power-supply voltages.
In designing a biasing network, limits on the permissible shift of operating
point are first set by taking into account the permissible changes in the
transistor parameters. These limits together with the variation in Ico
expected (a function of the expected temperature change) are then used
to set an upper limit on the stability factors. The biasing network may
then be designed by the use of the relationships existing between the
stability factors and the components of the biasing network. Complete
design tables are given by Shea, and the reader is referred to this book

1 Hunter, op. cit., pp. 11.26-11.27.

2 Shea, op. cit., chap. 3. *
Sec. 2.22] D-C AMPLIFIERS 95

for details of the procedure. For the common emitter circuit, Shea’s
results can be summarized as follows:
1. The poorest situation results when the base is driven from a high-
resistance source, and the emitter is grounded.
2. Placing a resistor in the common emitter lead improves stability.
3. Reducing the output resistance of the driving source improves
stability.
4. A feedback resistor connected between the collector and base
improves stability.
It should also be noted that a large load resistor in the collector lead
tends to protect a transistor from thermal runaway, since an increase in
Ico causes a decrease of collector voltage. This is true even though the
voltage stability factor as defined by Shea increases with RL.
Another method of reducing drift is to use a difference amplifier1 as
shown in Fig. 2.31. This circuit is
completely analogous to the vac¬
uum-tube difference amplifier dis¬
cussed in Sec. 2.11, and its output
is effectively equal to the difference
in the two base signals multiplied
by a gain factor. Also, as far as
changes in Ico are concerned, the
output will be a function of the dif¬ Fig. 2.31. The transistor difference
ference of the currents in the two amplifier.
transistors. Thus the circuit will
discriminate against variations in Ico in the same way as the tube circuit
of Sec. 2.11 discriminates against changes in eh. The complete analysis
of the transistor version of the circuit is left as an exercise for the reader.
The difference amplifier is only one special case of using a transistor as
a compensating element to reduce drift in a d-c amplifier. Other methods
of using transistors, diodes, and temperature-sensitive elements in this
way are described in the literature.2 Most of these methods require that
the amplifier be adjusted for least drift after construction.
2.22. Multistage Amplifiers. When transistor amplifier stages are cas¬
caded to form a multistage amplifier it is necessary, just as in vacuum-
tube amplifiers, to design interstage coupling networks which will permit
proper voltage and current biases to be applied to the coupled transistors.
This problem is, however, often much simpler in transistor amplifiers than
in vacuum-tube amplifiers. This is due to three factors. First the col-

1 D. W. Slaughter, Feedback Stabilized Transistor Amplifier, Electronics, vol. 28,

May, 1955, pp. 174-175.
2 Shea, op. cit., chap. 5. Hunter, op. cit., p. 13.17. E. R. Kretzmer, An Amplitude
Stabilized Transistor Oscillator, Proc. IRE, vol. 42, pp. 391-401, 1954.
96 CONTROL SYSTEM COMPONENTS [Chap. 2

lector may often be operated at voltages of only 1 or 2 volts, particularly

in low-level stages. Secondly, if transistors of the same type (either all
p-n-p or all n-p-n) are used, the base and collector voltages have the same
polarity. Finally, it is possible to alternate between p-n-p and n-p-n
transistors.
A typical three-stage circuit illustrating some of these points is shown
in Fig. 2.32. In the first stage a resistor is employed in the common-
emitter lead to provide some bias stabilization. In picking a value for
this resistor it should be kept in mind that a very large value will increase
the input impedance and reduce the voltage gain, but that it does not
have much effect on the current gain. As long as the input impedance
of the transistor is less than the source resistance of the generator Rg, the

current delivered by the generator and the current amplification of the

stage will not be reduced by A3. Thus a compromise value that will not
make the input impedance too high should be used. Typically i?3 is on
the order of 1,000 ohms. The plate-load resistor R2 may be chosen in the
same way as in vacuum-tube circuits, by use of the load line. It should
not be too large, so that the quiescent collector current is very much
greater than the current flowing for zero base current. The biasing
resistor R i is now chosen to produce the correct emitter current as
demanded by the choice of operating point. It may be computed by
assuming that the base-to-emitter voltage is approximately zero. For
zero ei the quiescent value of the emitter current Ie is then equal to

r _ _Aj_ Rg _ E\Rg
R\ + (RgRs)/(Rg + Rs) Rg + R% R\{Rg -f- Rz) + RgR%

Since Eh Ie, Rg, and R% are known, Ri may be found. The operating
point of the first stage, and therefore the d-c level of the amplifier output,
can be varied by adjusting Ri.
The second stage is an n-p-n stage; hence the base must be biased in the
positive direction. If the positive voltage A3 + E2 is very much greater
in size than the negative collector voltage of the first stage, the gain
loss in the coupling network is small. This fact would dictate a choice of
small collector operating voltage for the first stage. R± and A5 are
Sec. 2.23] D-C AMPLIFIERS 97

designed so that the base of the second stage operates at the correct value
of quiescent emitter current.
In the coupling between the second and third stages no resistance is
needed at all since the collector of the second stage and the base of the
third stage can both operate at the same positive voltage (approximately
E2). As a matter of fact, in low-level stages it is sometimes possible to
dispense also with resistor R$ and to let the input impedance of the third
stage act as the collector load resistance for the second stage. A very
simple circuit results.
The emitter of the third stage is connected to the positive voltage E\
so that the output voltage e0 can be zero for zero e*. Instead of using a
separate battery E2, a bleeder may be used between the positive supply
and ground, or more elegantly, a silicon diode (Zener diode) may be used
instead.1
2.23. The Chopper-stabilized D-C Amplifier. Although the difference
amplifier and the Miller circuit discussed in previous sections are capable

Fig. 2.33. D-c amplifier using modulation and demodulation.

of reducing drift to a value low enough for d-c amplifiers of moderate gain,
these relatively simple circuits cannot meet the requirement for almost
zero drift that is made in the most critical situations. D-c amplifiers used
in analogue computers or amplifiers used to amplify the extremely minute
voltages generated by thermocouples are representative of this class of
problem.
For such applications it is necessary to “ detour ” around the drift prob¬
lem introduced by d-c amplifiers and to use a-c amplifiers instead. A
typical circuit to accomplish this is shown in block-diagram form in Fig.
2.33. Here the signal is passed first into a modulator, which converts the
d-c (or very low frequency) signal into a relatively high frequency signal
that may then be amplified in a standard a-c amplifier. Since a-c ampli¬
fiers with very high gain are relatively easy to construct and since they
do not drift, a very large amount of drift-free amplification may be
obtained, provided that the modulator itself is fj’ee of drift. After the
signal has been amplified to the desired level by the a-c amplifier, it is
passed through a demodulator to recover the original d-c, or low-fre¬
quency, information.
Although a number of electronic modulator and demodulator circuits
1 Hunter, op. cit., p. 13.10.
98 CONTROL SYSTEM COMPONENTS [Chap. 2

are available (see Chap. 6), most of them are subject to some drift.
Hence, for critical applications the electromechanical vibrator, or chop¬
per, is used (see Fig. 2.34). These devices have in recent years achieved
a very high degree of perfection; they are hermetically sealed and suffi¬
ciently rugged to withstand large amounts of acceleration, shock, and

(6)
Fig. 2.34. Electromechanical chopper: (a) disassembled; (b) schematic.

temperature variation. The terminals are usually brought out to a

standard tube socket, and the outside dimensions are comparable to those
of standard or even miniature vacuum tubes. Further, their life expect¬
ancy under normal operating conditions is comparable to that of vacuum
tubes, and they may, therefore, be designed into a circuit in exactly the
same way as a tube would be.
Sec. 2.23] D-C AMPLIFIERS 99

A convenient method of connecting the chopper is shown in Fig. 2.35,

where a single chopper performs the functions of both the modulator and
demodulator. If the chopper is used in this way, it must be of the short-
circuiting type; i.e., when the vibrator reed is at the center of its travel,

Fig. 2.35. Chopper-stabilized d-c amplifier.

all three contacts are connected together. With this type of chopper
either the output or the input is always grounded. There is therefore no
possibility of a feedback connection being established between output and
input through the capacitance existing between the open vibrator con¬
tacts. Such feedback might result in undesirable oscillations.
The operation of the circuit is best understood by assuming the input
signal to be a low-frequency sine
wave and examining the wave¬
shapes at various points in the cir¬ Time
cuit, as shown in Fig. 2.36. Figure
2.36a shows the sinusoidal input
voltage. If it is assumed that the
input impedance of the a-c amplifier
is infinite compared to the resist¬
ance Ri in Fig. 2.35, the input volt¬
age (a) appears at the input to the
amplifier without attenuation when
the input contact of the chopper is
open; otherwise the voltage is zero.
This situation is illustrated in Fig.
2.365. Note that the amplifier
Fig. 2.36. Waveshapes of chopper-
input voltage still contains a com¬ stabilized amplifier.
ponent at the low signal frequency.
However, if we assume that the amplifier does not pass this low
frequency, then the amplifier output voltage will have the form shown in
Fig. 2.36c.1 This output will be passed to the low-pass filter only when
the output chopper contact is open; hence there results the voltage shown
1 It is shown in Chap. 6 that the amplitude of the signal component of the wave¬
shape of Fig. 2.366 is one-half of the input amplitude. The high-pass filters in the
a-c amplifier effectively subtract this component. The reader may convince himself
by performing this subtraction graphically that Fig. 2.36c is obtained.
100 CONTROL SYSTEM COMPONENTS [Chap. 2

in Fig. 2.36d. The low-pass filter serves to remove the chopper frequency
from the output, so that the final result is ideally an amplified reproduc¬
tion of the input (Fig. 2.36a). Note that, in the example carried through
here, the amplifier was assumed to have an even number of stages, so that
there is no sign reversal between the voltages of Fig. 2.366 and c. The
final reversal between input and output is due to the fact that the chopper
inherently applies a signal to the input at the same time that it shorts the
output, and vice versa. It should be clear that an odd number of ampli¬
fier stages would have resulted in a reversal between signals 6 and c and
no reversal between a and d.
Although the amplifier described above and others operating on the
modulator principle make it possible to achieve very high gain with
negligible drift, they have the very serious disadvantage that they can

Fig. 2.37. Two-channel wideband d-c amplifier.

amplify only very low frequency signals. The reader may easily prove
to himself by the method used in connection with Fig. 2.36 that, if the
input frequency had been equal to the chopping frequency, the output
would have been a d-c voltage. This indicates that amplification of sig¬
nals at chopper frequency is impossible (see also Chap. 6). Even for
signal frequencies of the order of one-tenth of the chopping frequency it
would be difficult to design a low-pass filter to follow the amplifier which
would remove the carrier and pass the signal without too much attenu¬
ation. The obvious means for alleviating this difficulty, i.e., using higher
chopper frequencies, is limited by the fact that most choppers are built
to operate at either 60 cps or 400 cps and that it is very difficult to build
an electromechanical device to operate reliably at frequencies in the
upper audio range. Hence, when it is desired to have high gain with
zero drift over a wide frequency band extending down to zero frequency,
it becomes necessary to employ a two-channel arrangement, like that
shown in Fig. 2.37. Here the signal is split into high- and low-frequency
components by means of the input filters shown preceding the amplifiers
in the two channels. Actually, the high-pass filters shown with the high-
frequency channel might be the usual RC coupling networks used in a-c
amplifiers. The crossover frequency at which the signal shifts from the
Sec. 2.23] D-C AMPLIFIERS 101
high-frequency channel to the low-frequency channel must be low enough
so that negligible energy at chopper frequency enters the modulator,
since this will give rise to spurious signals, as indicated previously. A
second low-pass filter, similar to the one shown in Fig. 2.35, is required
in the low-frequency channel to remove the chopper frequency from the
output. The summing circuit might take the form of the difference
amplifier described in Sec. 2.11. In this case it is, however, necessary
that there be a sign reversal in one channel relative to the other.
It should be noted that, unless special care is used in the design of the
transfer functions of the two channels, there will be a dip in the frequency
at which the signal shifts from one channel to the other. Although the
transfer functions of the two channels required to avoid this will not be
described here, an indication of how the problem can be solved approxi¬
mately is given below in connection with Eq. 2.80. Negative feedback is

Zf

also commonly used to stabilize the gain and has the additional advantage
of providing approximately constant gain for the composite amplifier even
if the two channels do not have exactly the same gain. Thus it is possible
to obtain an amplifier with substantially constant gain over a wide fre¬
quency range extending down to direct current, while at the same time
the drift is virtually zero.
In many applications, such as the operational amplifiers used in
analogue computers, it is not necessary to have constant gain at all fre¬
quencies. Large amounts of negative feedback are always employed in
these amplifiers, and the primary requirement is zero drift. In this case
the ideas concerning drift elimination by means of negative feedback dis¬
cussed in Sec. 2.3 may be employed. The chopper-stabilized d-c ampli¬
fier of Fig. 2.37 may be used as the high-gain, driftless preamplifier. Since
high gain is required only to combat drift, the high-frequency channel
shown in Fig. 2.37 can be omitted and replaced by a simple bypass con¬
nection of unity gain. The summing amplifier may then be followed
by several stages of d-c amplification to provide a sufficiently high loop
gain at higher frequencies to ensure that the over-all gain of the amplifier
102 CONTROL SYSTEM COMPONENTS [Chap. 2

is determined by the constants of the feedback network. A simplified

schematic of this arrangement is shown in Fig. 2.38, and a practical form
of the circuit is shown in Fig. 2.39.1 The bypass connection from e\ to the
summing circuit is required so that the low-pass filters in the chopper-
amplifier channel do not reduce the over-all gain to very low values at

relatively low frequencies. It ensures that the gain is at least as large

as (x2 for all frequencies. The over-all forward gain is

Gi
%-Q 1 + (RiCis T 1)(J?2C2S T 1)_
G2[(RlCiS + 1) (R^C2.S + 1) -f~ Cj]
(2.80)
{RiCis T ^{RzCzs T 1)
When the numerator is solved for its roots, it is found that, unless the
ratio of R\C\ to R2C2 is greater than G\ (or less than 1/Gfi), the roots have
large imaginary components, indicating a pronounced dip in the frequency
response. Since this is undesirable, the ratio of the two time constants
should be kept of the same order as the gain. The adjustment is not,
however, very critical. The open-loop frequency response will then have
the appearance shown in Fig. 2.40. Note that the gain is very high for

1 E. A. Goldberg, Stabilization of a Wide-band D-C Amplifier for Zero and Gain,

RCA Rev., June, 1950, pp.’ 297-300.
Sec. 2.25] D-C AMPLIFIERS 103
direct current but drops to a more moderate value at intermediate fre¬
quencies. At very high frequencies the gain drops again; this is due to
capacitors Ca and Cb in the d-c amplifier. These capacitors are required
to make the entire feedback loop, of which the amplifier is only a part,
stable. The reader is referred to standard texts on servomechanism
theory1 for details on the design of such components. The effect of these
capacitors is not included in Eq. (2.80), which therefore applies only at
low frequencies.

0.0001 0.001 0.01 0.1 1 10 100 1000 104 105 106 107
co radians/sec
Fig. 2.40. Loop gain of Goldberg amplifier.

2.24. Design Considerations. Many design methods that are quite

well established and straightforward for audio amplifiers break down when
the attempt is made to apply them to d-c amplifiers. They must, there¬
fore, often be replaced by relatively cumbersome trial-and-error methods.
Thus, for instance, no really simple method is available for the optimum
design of a simple triode amplifying stage having a cathode-bias resistor.
Hence the following remarks have been set down chiefly to point out some
of the special problems encountered in d-c-amplifier design and some of
the means used to solve them. Lest the reader be discouraged by the lack
of definite design procedures, it should be noted that d-c amplifiers are no
more critical than audio amplifiers in their tolerance requirements, and it
is usually found that large variations in component values result in only
relatively minor performance changes. Hence certain standard circuits
have been developed in most cases and are usually adequate.
2.25. Triode Amplifier with Resistance-coupling Network. When a
simple triode amplifier is coupled to another stage, the nature of the
1 Chestnut and Mayer, '‘Servomechanisms and Regulating System Design,” John
Wiley & Sons, Inc., New York, 1951, vol. 1.
104 CONTROL SYSTEM COMPONENTS [Chap. 2

coupling network must be considered in the design of the amplifier. Con¬

sider first the resistance coupling shown in Fig. 2.41a. The effect of this
coupling on the load line for the amplifier will be examined first. In Fig.
2.416 load line 1 has been drawn for an amplifier for which Ri + R2 is
infinite; the slope of the line is —I/Rl. The line is extended for values
of negative plate voltage, even though the tube cannot operate in that
region for reasons that will become apparent below.

(a) Circuit (b) Plate characteristic with load lines

Fig. 2.41. Design of amplifier with resistance coupling: (a) schematic circuit; (b) plate
characteristic with load lines.

If the resistance of the coupling network becomes finite, both the effec¬
tive value of plate-load resistance seen by the tube and the effective posi¬
tive supply voltage will be reduced. The plate load becomes simply the
parallel combination of Rl and Ri + R2, or

/ _ Rl(Ri + R2)
(2.81)
Rl T R\ T R2

The effective positive supply voltage is the voltage appearing at the

plate of the tube when the plate current is zero. The current flowing
through Rl into the coupling network causes this voltage to be less than
the actual supply; thus

p _ (Ebb — Ekk)RL
Rl H- Ri H- R2
(Ri -f R2)Ehb T RlEicIc
(2.82)
Rl T R1 ~b R2

Both of these effects are indicated by line 2 in Fig. 2.416, which repre¬
sents the load line resulting from a finite R1 + R2. All the load lines
must meet at the point ep = Ekk, since there is then no voltage across R\
and R2. It is clear that, if the level of impedance of the coupling network
is not high enough, the tube will operate in an undesirable region with
relatively small gain and be able to handle only relatively small signal
swings. On the other hand, very high coupling impedance results in poor
high-frequency response and increased sensitivity to noise pickup. A
Sec. 2.25] D-C AMPLIFIERS 105
compromise is, therefore, usually necessary, and the levels of Ri and R2
are usually taken so as to keep line 2 in Fig. 2.416 fairly close to line 1.
The design procedure in detail is as follows:
1. Pick values of Rl and R\ + R2 giving sufficient gain and signal¬
handling capacity.
2. Decide on operating point (usually in the middle of the linear range).
3. Specify operating voltage of next grid (usually near zero volts).
4. Find Ri and R2 from relation

R1 _ Rkk
(2.83)
R1 + R2 £b — Ekk

When an optimum design is required, a number of trials are usually

necessary to obtain the best combination of Rl, Ri, and R2.
It is theoretically possible to compensate for the poor high-frequency
response of resistance-coupling networks by bridging Ri of the coupling
network with a capacitor, as shown in Fig. 2.42a. The optimum value

(a) (b)

Fig. 2.42. Use of capacitor to improve frequency response of coupling network.

for this capacitor may be computed if the input capacitance of the follow¬
ing stage is known. This capacitance can be found from the published
values of interelectrode capacitances by methods given in standard texts.1
If this input capacitance is designated by C2, then the equivalent circuit
of the coupling network takes the form shown in Fig. 2.426, where Ra is
the output impedance of the previous stage and C1 the bridging capacitor.
The transfer function of the coupling network is found by the methods
discussed in Chap. 1. It is given by

c2 / R2
ex \R0 T Ri R2
R1C1S + 1
R0RiC\R2C2 9 R0RiCi -f- R0R2C2 T R2R\C\-b R\R2C2 . ^
R0 -b Ri ~b R2 Ro ~b Ri ~b R2
It may be shown that the best value of C\ is the one that makes
R1C1 = R2C2
1 See, for instance, Reich, “Theory and Applications of Electron Tubes,” McGraw-
Hill Book Company, Inc., New York, 1944, pp. 93-96.
10G CONTROL SYSTEM COMPONENTS [Chap. 2

With this value of C\ the transfer function of the coupling; network

becomes

e2 Ri 1
(2.85)
Ci \R0 Ri -\r R2/ \[RoR2C2/(Ro + R\ + R2)]s + 1

i.e., the zero in the transfer function exactly cancels one of the poles, and
the remaining pole is moved to a relatively high frequency. The transfer
function of the coupling network without the compensating capacitor C1
may be obtained by setting C1 = 0 in Eq. (2.84). Comparison of the
result of this with Eq. (2.85) indicates that the passband of the network is
increased by a factor of (R1 + R0)/R0 by use of the correct value of Ci.
2.26. Design of Triode Amplifiers with Neon-tube Type Interstage
Coupling Network. When the neon-tube coupling networks are used, the
design procedure is similar to that described in the previous section.

(a) Circuit [b) Plate characteristic with load line

Fig. 2.43. Neon-tube coupling.

Since the operating current for neon tubes must fall within rather narrow
limits, one variable, namely, the impedance level of the coupling network,
is specified in advance. The effect of the coupling on the effective plate
voltage and plate-load resistance is similar to that found in the resistance-
coupled amplifier. The effective supply voltage is

E'bb = Ebb ~ ineR'L (2.86)

The effective plate-load resistance is

R'l = * (2.87)
Kl T riff

where ine is the neon-tube current and Rg the grid-leak resistance of the
following stage (see Fig. 2.43a). The detailed design steps are then as
follows:
1. Specify the quiescent value of the voltage at the grid of the following
stage. If Ekk is given, this immediately specifies the value of Rg, since
Sec. 2.26] D-C AMPLIFIERS 107
2. Select a value of Rl, which, together with the value of Rg decided on
in (1) above, will result in an effective load line giving sufficient gain and
signal range.
3. The proper number of neon tubes to use between stages is now
determined by noting that the sum of the quiescent grid voltage deter¬
mined in (1) plus the neon-tube drop must result in an operating point for
the first stage that will permit the required signal swing in the first tube.
This requirement is complicated by the fact that the voltage drop of neon
tubes picked at random may vary from about 49 to 74 volts. Hence, in
order to avoid selection of neon tubes, it is desirable to pick Rl in such a
way that the full range of neon-tube voltages will not force the operating
point of the first stage into undesirable regions.
This procedure is best illustrated by an example. Assume that ec2, the
quiescent second-grid voltage, is —15 volts and that a plate-signal swing
of 10 volts peak is present. Let Ebb = 300 volts. The effective plate
supply voltage E'bh is less than this,
the amount depending on the value
of R'l. Suppose two neon tubes are
used. The operating-point voltage
for the plate of the first stage may
then vary from 83 to 133 volts with
variations in neon-tube voltages
(see points A and B, Fig. 2.44).
The signal swing will cause the plate
Fig. 2.44. Effect of neon-tube tolerances
voltage to vary from 73 to 93 volts
on operating point.
or from 123 to 143 volts around
these two operating points. It is clear from the diagram that the
lowest value of plate voltage (73 volts) wou'd require a positive grid
voltage if the lower plate-load resistance R'Ll were used. Hence the
larger value shown in the figure, RfL2, would be indicated. Another
possibility is to use three neon tubes. Considerations similar to those
discussed would then indicate a possible plate-voltage variation between
122 and 217 volts. This would probably be satisfactory if the plate-load
resistance were R'L1, but the tube would be close to cutoff with the larger
plate-load resistance.
It should be noted, incidentally, that the variation of plate operating
point required to accommodate random values of neon-tube voltages is
accomplished by the zero balance adjustment required in every d-c ampli¬
fier. This adjustment must be designed with the proper range to handle
the variations discussed here, and other possible tolerance effects. The
adjustment should usually be placed in the first stage of the amplifier, so
that all stages may be adjusted together.
The question of neon-tube tolerance is one of the simpler examples of
108 CONTROL SYSTEM COMPONENTS [Chap. 2

the more general problem of designing a circuit to make use of components

with standard tolerances. A complete design would require a check of
the effect of the tolerances of all components and a redesign of sections
of the circuit which cause unsatisfactory operation under one or several
extremes of component variations. Occasionally, particularly in very
critical applications, it is necessary to select components specially; ordi¬
narily, however, this should be avoided if possible.
As was pointed out in Sec. 2.17, neon-tube coupling networks have poor
high-frequency response owing to the “inductive” effect of ionization
time in the neon tubes. It might seem that a capacitor bridging the neon
tube would improve the response here as it did with the resistance¬
coupling network. This procedure must, however, be applied with cau¬
tion, since inspection of the circuit reveals that it has a form almost indis¬
tinguishable from the classic gas-tube relaxation oscillator. Hence,
unless special precautions are taken to prevent the oscillations, capacitors
across the neon tubes are not ordinarily desirable.
2.27. Design of Cathode-follower Circuits. We assume that the cath¬
ode follower operates into a specified load resistance R and that the maxi¬
mum output-signal swing is to be obtained (see Fig. 2.45a for the circuit).

For a particular value of Rk, a “static” load line (i.e., the line for R = *>)
can be drawn for the cathode follower from considerations similar to those
employed in amplifier stages. When the plate current is zero, the plate-
to-cathode voltage is Ebb — Ekk. This locates one point on the static load
line. When the plate-to-cathode voltage is zero, the plate current is
(Ebb — Ekk)/Rk• This locates another point and results in a line such as
the one shown in Fig. 2.455. This load line may be used to analyze the
no-load behavior of the cathode follower. When the output voltage
ek = 0, the plate-to-cathode voltage is equal to Ebb. This point may
therefore be considered.to be the quiescent operating point.
Sec. 2.28] D-C AMPLIFIERS 109

If the load R is connected to the cathode follower, the tube operates

along the dynamic load line shown in the figure. This line may be con¬
structed as follows: when ek = 0, the dynamic load line and static load
line must intersect, since there is no current in the load. For other values
of ek the slope of the dynamic load line is determined by the parallel com¬
bination of Rk and R and is given by — (R + Rk)/RRk.
The optimum design for the cathode resistor Rk requires a knowledge
of the maximum plate dissipation of the tube. A hyperbola eviv = IFmax
(the maximum plate dissipation) is drawn on the characteristic curves for
the tube. The exact value of the maximum plate dissipation depends on
the duty cycle to which the circuit will finally be subjected. Thus, if the
output voltage is generally small and if peak voltages are required only
for short times, a higher value of IFmax may be assumed than if the tube
must deliver maximum voltages for extended periods of time. However,
assuming that the maximum dissipation curve has been drawn, the opti¬
mum design is one permitting equal swings from ek in both directions, as
shown in Fig. 2.456.
The design proceeds as follows:
1. Assuming Ebb, Ekk, and R are known, estimate an approximate value
for Rk so that the static load line passes about halfway between the maxi¬
mum dissipation line and the iv = 0 axis. This makes possible a tenta¬
tive choice for the dynamic load line, which has the slope — (R + Rk)/RRk.
2. Draw the tentative dynamic load line such that equal swings on both
sides of the quiescent operating point are possible (see Fig. 2.456).
3. The static load line should bisect the dynamic load line at the oper¬
ating point. If it does not, a correction must be applied to the estimated
value used in step 1 and the procedure repeated until it does.
While the above procedure will result in an optimum design, it might
be well to point out that optimum design is seldom justified, particularly
if tube and component variations are ignored. In this case the design is
much simpler, consisting in fact of step 1 above.
2.28. Design of Difference Amplifiers. Since the difference amplifier
is a relatively complicated circuit, the graphical design technique using
the load line cannot be applied exactly; however an approximate design
that is adequate for determining limits of linear operation and quiescent
operating points can be made fairly easily.
In order to establish some rough criteria for optimum values for RL and
Rk (see circuit, Fig. 2.46a), we first consider the operation of the cathode-
follower section (section 2 of Fig. 2.46a), as its grid voltage is varied from
a large positive value through zero to large negative values. The grid of
section 1 is assumed to be grounded. If Ec2 has a large positive value,
both cathodes in following Ec2 also take on large positive values, in fact,
slightly above Ec2 to maintain proper bias of tube 2. Tube 1 is then cut
110 CONTROL SYSTEM COMPONENTS [CHAP. 2

off and presents a load of infinite impedance to the cathode follower.

This mode of operation is indicated in Fig. 2.466 by the part of the curve
marked 1. As the grid and the cathodes become less positive, the point is
eventually reached where tube 1 begins to conduct. At this point it
changes fairly abruptly from an infinite impedance to the relatively low
value (r„i + Rl)/(hi + 1)- The operating line of the cathode-follower
section leaves the static load line and switches to the dynamic load line
indicated by 2. Note that the dynamic load line is slightly curved at the
top; this is due to the larger value of rvX when tube 1 is just beginning to
conduct. As Ec2 is decreased further, the point is finally reached where
the cathode follower cuts off, and the operating line continues along the
ip = 0 axis. If grid 1 had not been grounded, line 2 would have shifted
either left or right, depending on whether grid 1 was positive or negative,
respectively.

Fig. 2.46. Design of difference amplifier: (a) schematic; (6) load lines.

It should be clear that linear operation of both halves of the circuit is

represented only by the line segment marked 2 in Fig. 2.466. If maxi¬
mum linear range is an important consideration, we should endeavor to
make this line segment as long as possible. Assuming that the supply
voltages are fixed, both a reduction in Rk and an increase in Rl will have
this effect. Hence Rk should be just large enough to prevent excessive
plate dissipation in the cathode-follower section, while Rl should be made
as large as possible consistent with linear operation of the amplifier sec¬
tion. The operating point should then be located somewhere near the
center of line segment 2 if possible, although other circuit requirements,
e.g., the need to accommodate component tolerances, will usually set the
operating point at some less-than-optimum value.
The design of the interstage coupling networks to follow the amplifier
is not so straightforward as it was for the simple triode with grounded
cathode. The reason for this is that the amplifier section of the difference
amplifier has an effective cathode bias resistance equal to the output
resistance of the cathode-follower section. Strictly speaking, therefore,
the characteristic curves should not have the slope 1 /rv but rather the
Sec. 2.28] D-C AMPLIFIERS 111

slope l/rp = 1 /[rp + (fj. + 1 )R0], Ro being the cathode-follower output

resistance. Generally, however, the labor involved in replotting the
characteristics from the published curves is not necessary if approximate
results are sufficient, particularly since the correction is not very large.
Thus an approximate design for Rl and the interstage coupling is usually
carried out in the same way as that described in Secs. 2.25 and 2.26 for
the simple triode. It should be noted, however, that the correct value of
gain cannot be deduced directly from the construction of a load line on
the published tube characteristics. The procedure should rather be to
determine values for /jl and rp at the operating point and to compute the
gain from Eq. (2.37).
One final problem that should be considered in connection with this cir¬
cuit is the determination of the proper value of voltage on grid 2, once an
operating point for plate 1 has been specified. The reader should be
cautioned here, incidentally, against the fallacy of assuming that the two
grid voltages will have approximately equal quiescent values simply
because the circuit amplifies the voltage difference existing between
the grids. Such is not the case, primarily because the circuit is not
symmetrical.
When the quiescent value of epl is given, the procedure for finding the
corresponding value of Ec2 is as follows: (We assume Ecl to be zero for the
sake of simplicity, but the method is easily extended to other values of
Eci.)
1. With epi and Rl known, the plate current ipi and the required grid
bias of section 1 may be read from the published plate characteristics of
the tube. Hence, if the grid is grounded, the cathode must be positive by
the amount of the bias.
2. Once the cathode voltage has been determined, the voltage across
the cathode resistor is known. Hence the current in the cathode resistor
) can be computed.
3. The plate current in the cathode follower is equal to the difference
between the current in the cathode resistor and the plate current of
section 1.
4. Since the plate of the cathode follower is at Ehh and since the cathode
voltage is known from step 1, the plate-to-cathode voltage is known.
The plate current is known from step 3. These two ordinates establish
the required grid bias in the cathode follower.
5. The desired voltage on the cathode-follower grid is then the sum of
the cathode voltage (from step 1) and the (negative) grid bias from step 4.
Usually this voltage is negative with respect to the voltage on grid 1.
In addition to clarifying circuit operation these steps make possible the
determination of the size of the potentiometer required to adjust the cir¬
cuit to accommodate various component tolerances.
112 CONTROL SYSTEM COMPONENTS [Chap. 2

PROBLEMS

2.1. A cathode follower using a 12AX7 vacuum tube has a cathode resistor of
200 kilohms; Ebb = 300 volts; Ekk = —300 volts. Vibrations of the grid cause n to
change by ±5 per cent; the change of plate resistance is negligible. With grid
grounded, find the change of output voltage caused by the vibrations.
2.2. Compute the drift due to variations of Ebb and Ekk for the difference amplifier
(Fig. 2.11). After obtaining the exact expression, simplify the result by assuming
that the fi’s and rp’s of the two tube sections are identical and that Rk » rp.
2.3. Show that the drift in a properly adjusted Miller circuit (Fig. 2.13) due to
variations of Ebb is given by Eq. (2.50) if the assumption is made that the yu’s of the
two tube sections are identical.
2.4. Figure 2.47 shows a circuit that may be used to add three signals. Find the
expression for e0 as a function of ei, e2, ez. Assume that all three tubes have the same
n and rp and that Rk rp.

2.5. (a) Determine the drift caused by a variation of Ebb on the output of the cir¬
cuit in Prob. 2.4. (b) Repeat for a variation of Ekk- (c) Repeat for a change in
heater voltage.
2.6. (a) Find the gain and output impedance of the circuit of Fig. 2.48. Assume
the two tubes are identical. (6) Find the drift due to change of Ebb. (c) Find the
drift due to change in heater voltage.

2.7. The circuit (Fig. 2.49) shows a series-type amplifier which is supposed to have
the feature that drift due to heater-voltage variation is canceled in it if R is correctly
adjusted, (a) Find the value of R that provides perfect cancellation of filament-
Problems] D-C AMPLIFIERS 113

voltage variation effect. (6) Determine the drift due to changes in Ebb- (c) Deter¬
mine the gain e0/ei. Assume that the tubes are identical.

Fig. 2.49

2.8. When transistor circuits are to be analyzed by means of the node method the
transistor equations are most conveniently expressed in terms of y parameters as
follows:
ii = y litfi + 2/12^2
ii = 2/21^1 + 2/22^2

Express the four y parameters in terms of the h parameters.

2.9. Show that the equivalent circuit shown in Fig. 2.50 approximately represents
the effect of Ico in the grounded-emitter circuit.

Fig. 2.50. Equivalent circuit of common-emitter transistor amplifier showing effect

of Ico»

2.10. Obtain the h parameters for the grounded-base transistor amplifier in terms
of the common-emitter parameters [see Eq. (2.79)]. Use the approximations that
hre <K 1 and that hiehoe <£ h/e.
2.11. Show that the approximate expressions for the voltage gain, input impedance,
and output impedance of the grounded-collector transistor amplifier as given in
Table 2.1 are correct.
2.12. Compute the modified h parameters for the common-emitter transistor ampli¬
fier with a resistance Re connected between emitter and ground. Find the effect
of Re on the input impedance voltage gain and current gain. Assume that Rehoe <$C 1.
2.13. Repeat Prob. 2.12, but consider a resistance Rf connected between collector
and base. Assume that Rf » Ke.
2.14. Find the gain v2/(vn — v;2) for the difference amplifier of Fig. 2.31. Also
find the input impedance seen by vn.
2.16. Find the drift in v2 due to change in Ico for the difference amplifier of Fig. 2.31.
114 CONTROL SYSTEM COMPONENTS [CHAP. 2

2.16. The circuit of Fig. 2.51 shows a resistance-coupled d-c amplifier with the
following characteristics:

Tube type = 6SL7

Rl — 300 kilohms
Ebb = 300 volts
Ekk = —300 volts

It is desired to optimize the gain and the gain-bandwidth product as a function of the
resistance level of the coupling network. For this purpose the gain and gain-band¬
width product are to be computed for R\ + R2 = 2 megohms, Ri + R2 = 1 megohm,

and Ri + R2 = 0.5 megohm. For the purpose of this problem the bandwidth is
defined as the frequency in radians per second for which the gain of the amplifier
is 0.707 times the value at direct current. The input impedance of the second stage
of the amplifier may be assumed purely capacitive with a value of 150 nnf. Assume
further that the quiescent value of the grid voltage for both tubes is —1 volt, that
the output impedance of e; is zero, and that the load impedance seen by the second
stage is infinite.
Sketch a curve of the behavior of the gain and the gain-bandwidth product as a
function of R\ + R2. Comment on the result.
2.17. The d-c amplifier shown in Fig. 2.52 is to be designed to permit a maximum
load-voltage swing of ±50 volts into the 10-kilohm load. Quiescent value of input
and output is 0 volt. Find values for Ri, R2, Rz, and fib; determine the number of
neon tubes required between the stages and the required adjustment range for en2.

Fig. 2.52
 CHAPTER 3

POWER AMPLIFIERS

3.1. Introduction. Usually the power element or output stage of a

closed-loop system requires a driving source capable of supplying con¬
siderable power itself. In this chapter two convenient types of power
amplifiers, the thyratron amplifier and the relay amplifier, are considered
in some detail. Also, the increasingly popular transistor power amplifier
and the magnetic amplifier are discussed.
Rotating amplifiers will be considered in Chap. 4 of this text. The
design of conventional a-c amplifiers has been the subject of exhaustive
discussion in the literature for many years and will not be covered here in
any detail. A paragraph on the subject has been included in this chapter
essentially as an introduction to the literature. D-c amplifiers have been
discussed in Chap. 2 of this text, and the paragraph in this chapter extends
this material to d-c power amplifiers.
3.2. A-C Power Amplifiers. Most textbooks on electronics consider
the conventional power amplifier in considerable detail.1 Class A, class
AB, and class B single-ended and push-pull vacuum-tube-amplifier design
methods are straightforward and may be as analytical as the designer
desires. Class C amplifiers are usually designed by graphical methods2
because of the difficulty of analytical techniques. Plate efficiency and
power output for a given tube increase, in order, as it is operated as class
A, AB, B, or C. Single-tube operation is practical with class A, but
push-pull operation is desirable if the power level is appreciable. With
the other classes, push-pull operation is required, unless the load is a tuned
circuit. An example of a single-ended class A application in a control
system might be the driving amplifier for an instrument synchro or
resolver.3
It will be remembered by the reader that the various classes of oper-

1 See, for example, Reich, “Theory and Applications of Electron Tubes,” McGraw-
Hill Book Company, Inc., New York, 1944, chaps. 7, 8.
2 See, for example, Members of the staff of the Department of Electrical Engineer¬
ing, Massachusetts Institute of Technology, “Applied Electronics,” John Wiley &
Sons, Inc., New York, 1943, chap. 10, art. 4.
3 Valley and Wallman, “Vacuum Tube Amplifiers,” Radiation Laboratory Series,
vol. 18, McGraw-Hill Book Company, Inc., New York, 1947, chap. 9.
115
116 CONTROL SYSTEM CONPONENTS [Chap. 3

ation of vacuum-tube amplifiers are defined by the grid bias. For class A
operation the grid is biased so that plate current flows throughout the
input-signal voltage cycle. In class B operation the grid is biased to cut¬
off so that plate current flows only when the grid is driven in the positive
direction by the grid signal. Thus if the input signal is a sine wave, the
plate current flows for one half the cycle. Class AB operation is inter¬
mediate between these first two. In class C operation the grid is biased
beyond cutoff, usually to at least twice the cutoff value. As is readily
seen, the transfer from signal voltage at the grid to plate current is grossly
nonlinear except in class A operation. If desired, the subscript 1 or 2
may be added to the designation of the class of operation to denote the
magnitude of grid signal anticipated. If the grid is driven positive with
respect to the cathode, thus drawing grid current, the subscript 2 is
added; otherwise the subscript 1 is used.

Fig. 3.1. Class B operation as shown on a typical transfer characteristic, which shows
plate current plotted against grid voltage, with the voltage from plate to cathode of the
tube held constant. The grid-signal amplitude shown is causing class Bi operation.

In 'push-pull operation, two tubes are required. The two tubes are
driven by signal voltages that are 180° out of phase. Thus when one
tube is nearing cutoff, the other tube is entering its operating range. The
resultant composite transfer is linearized, and distortion is reduced.
Although the theoretical values are somewhat higher, under actual oper¬
ating conditions the plate efficiency of class A amplifiers is about 30 per
cent; of class B amplifiers, about 60 per cent; and of class C amplifiers, 70
to 80 per cent. Thus it may be seen that it is usually worth while to
employ class B operation, even though the design procedure is somewhat
more involved than that for class A amplifiers.
The definition of class B operation requires that the tube be biased to
cutoff, as shown on the transfer characteristic in Fig. 3.1. Since plate
current flows for only one half of the grid-voltage cycle, the resultant
plate-current wave is seriously distorted. A second tube is added to the
circuit that will operate on the other half of the signal-voltage wave, and
the resultant plate currents are added in the load. In a-c applications
Sec. 3.2] POWER AMPLIFIERS 117

the load is usually coupled to the amplifier by means of a transformer.

The transformer changes the impedance of the load as seen by the tubes
in such a way as to allow a maximum transfer of power. It is conceivable
that, in servomotor applications, a single tube could be used in class B
operation, since the motor will respond primarily to the basic carrier fre¬
quency. However, this type of operation would cause saturation of the
coupling transformer by the d-c component of current flowing in its pri¬
mary. In addition, harmonics of the carrier would flow in the motor,
resulting in power losses and possible spurious torques and saturation
effects. Figure 3.2 shows a simplified push-pull power-amplifier circuit.

Ebb

■/Vr- ■ Level
0 -

0-

Input
from
preceeding
stage

Fig. 3.2. Push-pull amplifier circuit.

A signal on the input transformer drives one grid positive and the other
grid negative. The plate currents flow in opposite directions in the out¬
put transformer, thus canceling any d-c component of flux in the trans¬
former core. In Fig. 3.3 the composite plate characteristics of the tubes
are shown. Composite plate characteristics are commercially available
for tubes such as the 6L6 and 807 that are commonly used in push-pull
circuits. The proper plate-battery voltage and grid-bias voltage must
be carefully chosen to establish the proper operating point. In Fig. 3.3
the curvature of the characteristics at low plate current has been exagger¬
ated. We note that, strictly speaking, this is not class B operation, since
both tubes conduct for small signal voltages. If the characteristics are
adjusted so that the tubes are completely cut off, the composite character¬
istics are no longer straight lines and a certain amount of distortion is
introduced.1 If we assume that the characteristics are essentially
straight lines, it is possible to derive an equivalent circuit for the oper-
1 Reich, op. cit.} art. 8.1.
118 CONTROL SYSTEM COMPONENTS [Chap. 3

Fig. 3.3. Composite plate characteristics for push-pull operation. Strictly speaking,
this is class AB operation.

ation in the class B mode. Essentially one tube is open-circuited, while

the other tube is conducting. We may represent this as an a-c equivalent
circuit, as shown in Fig. 3.4. As usual, /j. is defined as the ratio of a change
of grid voltage to the change of plate
n> voltage required to maintain constant
-AAAAA
plate current, and rp is the ratio of Aep
to Aip at a constant grid voltage. Note
that only one half of the turns of the
primary of the output transformer are
Fig. 3.4. A-c equivalent circuit for
utilized by the plate current of either
class B push-pull amplifier. tube. Thus the equivalent turns ratio of
the transformer must account for this
fact. The load impedance may be transferred to the primary by the
turns ratio squared; thus
At
(3.1)
6g Tp | CL“Rl
where a is defined as
„ ± nl
(JL — (3.2)
2n-
The power dissipated in the load is

fl6g
Pi oad = PR = a2Rh (3-3)
jv + a2RL,
Sec. 3.2] POWER AMPLIFIERS 119
For eg, g, and Rl constant we may determine the optimum turns ratio by
differentiating load power with respect to a and setting t he result to zero.
The result is

a2 (3.4)

which is the familiar statement that the load impedance must be matched
to the impedance of the generator for maximum power transfer. Occa¬
sionally one is led by this familiar result into assuming that the converse is
true. The reader may show that, if the internal impedance rp of the
generator can be adjusted by choosing the
proper tube, it should not be made equal to -ww—
the load impedance but rather should be made lp\
■o2Rl
as small as possible, for maximum power
transfer. O
~ Peg
\02ja>L
In general, the load is not purely resistive.
If the load consists of a servomotor, for
Fig. 3.5. Equivalent circuit
instance, it will have a large inductive react¬ with reactive load.
ance. Under these circumstances the equiva¬
lent circuit will be as shown in Fig. 3.5. The load impedance is referred
to the primary of the output transformer. For this type of load the plate
current is
dip
— ixeg = rpip + oPRiiv + cl2L (3.5)
dt

The plate current is no longer in time phase with the generator voltage,
and no longer does the simple load
line describe the operation of the cir¬
cuit. From Eq. (3.5) we see that
only when the rate of change of plate
current is zero will the operating
point lie on the load line. When the
plate current is increasing, the volt¬
age across the inductance opposes
the increase, and when the current is
decreasing, the inductance tends to
lpZ
maintain the flow. It can be shown1
Fig. 3.6 Operating locus for inductive that, for a sinusoidal input signal,
load. the operating point follows an ellipse
whose major axis is inclined some¬
what to the load line, as shown in Fig. 3.6. The ratio of the distance
along the major axis to the minor axis is equal to the ratio of the load
1 Dow, “Fundamentals of Engineering Electronics,” John Wiley & Sons, Inc.,
New York, 1945, p. 274.
120 CONTROL SYSTEM COMPONENTS [Chap. 3

resistance to the load reactance. The presence of inductive reactance in

the load reduces the maximum power that can be delivered to the load,
while at the same time the required volt-ampere capacity of the driving
amplifier remains high.
In a-c servo applications the frequency of the input signal is normally
a narrow band about the carrier frequency of the system. It is therefore
possible to correct the power factor of the load by placing a capacitor
across it. The tuned load will absorb more power, thus more effectively
utilizing the capacity of the driving amplifier. At the same time, since
the loads encountered in servo applications normally have a large resistive
component, there is little possibility of making the tuning so sharp as to
reject the sidebands of the carrier. The resistive component also limits
the usefulness of class C amplifiers for this application. Usually the
effect on the phase lag of the control loop by filters at the carrier frequency
must be considered.1
Power transistors of the junction type have been under extensive
development, and units capable of handling currents of over 10 amp at
collector voltages of about 100 volts are presently available commercially.
In addition to the well-known general advantages of transistors over
tubes, transistor power amplifiers have several special advantages.
First, practical transistor amplifiers can achieve almost the theoretical
maximum power efficiency for a given class operation. Efficiencies of
48 per cent are achieved in class A operation and 70 per cent in class B
operation.2 Second, the power output of a transistor is typically in
the form of low voltage and high current. Since many control-system
devices such as solenoids, relays, motors, magnetic amplifiers, etc., require
current for torque or flux, this leads to an excellent match between the
transistor and its load in many control-system applications.
The power-handling capacity of a transistor is limited by the tempera¬
ture at which the collector junction can operate and by the breakdown
voltage of the collector. In certain connections a condition of tempera¬
ture runaway can occur. The quiescent collector current raises the tem¬
perature of the collector junction; this causes an increase in the collector
current. Under normal loads and ambient temperatures the effect is not
cumulative, but under heavy currents and increased ambient tempera¬
ture, compensating circuits must be considered.
As with vacuum tubes, the class B push-pull transistor power amplifier
is most common. Its high efficiency and low distortion are its main
advantages, since even harmonics are canceled. Of the three possible

1 See Truxal, “Automatic Feedback Control System Synthesis,” McGraw-Hill

Book Company, Inc., New York, 1955, sec. 6.7.
2 R. A. Hilbourne and D. D. Jones, Transistor Power Amplifiers, Proc. Inst. Elec.
Eng., vol. 102, part B, pp.- 763-774, 1955.
Sec. 3.2] POWER AMPLIFIERS 121

symmetrical connections, the common-base connection is least used. Its

power gain is low and decreases for large signals. The circuit’s main
advantage is its low distortion. The power gain of the common-collector
circuit is typically several times higher than the grounded-base connec¬
tion, and its distortion is low. This connection has some application in
high-quality audio amplifier circuits and in circuits in which maximum
power output is more important than power gain.
The common-emitter connection shown in Fig. 3.7 will typically have
16 to 40 times more power gain than the other circuits, and its somewhat

Fig. 3.7. Basic circuit of the grounded-emitter push-pull amplifier with p-n-p junction
transistors.

greater inherent distortion is not usually an important drawback in con¬

trol systems. In many control-system applications the load is center-
tapped, and the output transformer may be eliminated.
The power gain of the common-emitter circuit is1

1 4 Rl
(3.6)
1 + 3B rs

where a0 = current gain for zero emitter-current bias

B = fractional third-harmonic distortion
Rl = reflected load resistance (m2R0)
rs = reflected input resistance (n2Rg)
where m, n, Rg, and R0 are defined in Fig. 3.7. It will be remembered
that the theoretical maximum for the current gain a of a junction transis¬
tor is unity. Typical operating values are between 0.94 and 0.98. B, the
third-harmonic distortion, is usually between 0.1 and 0.2.
The design of transistor circuits cannot rely heavily on the load-line
techniques used in vacuum-tube amplifiers, since the collector character¬
istic, like that shown in Fig. 3.8, depends on the source impedance. From
the general four-terminal equations given by Shea2 and others,

Vl = T\\i\ + T \2X2 /o

v2 = r21ii + r22i2

1 Ibid.
2 Shea, “Principles of Transistor Circuits,” John Wiley & Sons, Inc., New York,
1953, p. 34.
 vT*.

122
o.A.

Fig. 3.8. Typical static collector characteristic of a junction transistor.

Sec. 3.3] power amplifiers 123

we may find the output impedance

T12^21
Ho — f 22 (3.8)
r n + rs

Thus rs has a very definite effect on the characteristic. Fortunately the

typical junction-transistor characteristic is so nearly linear that good
results are obtained by determining
nzRg rb rc[/-a)
the small-signal parameters and
—Wv—o—AW—1
employing, even for large signals,
the small-signal equivalent circuit
shown in Fig. 3.9. The character¬ Input
voltage
1
_
>re

-O- •-O-
rm^b <
cp <

istic curves are used principally to

Fig. 3.9. Equivalent circuit of class B
check operating points and voltage grounded-emitter push-pull amplifier.
and power limits. In Fig. 3.10 are
shown the composite collector characteristics of the 2N57 p-n-p junction
transistor with the operating limits set by the manufacturer.
3.3. D-C Power Amplifiers. The problems involved in d-c power
amplifiers are the same problems discussed in Chap. 2. However, since
the voltage gain of a power amplifier is usually relatively low, noise and

Collector current, Ic amperes

1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

Fig. 3.10. Push-pull composite characteristics for class B operation. The two curves
are matched at the operating point Vc = —60 volts.

drift introduced in this stage do not have so serious an effect as noise and
drift introduced at lower signal levels. The d-c power amplifier differs
from the a-c power amplifier in several respects. First, transformers
cannot be used for phase inversion at the input or for load matching at
124 CONTROL SYSTEM COMPONENTS [Chap. 3

the output. Second, both B+ and B~ supplies must be employed if the

output is to be balanced about ground. And third, if several stages of
push-pull amplification are employed, the interstage coupling networks
must pass direct current.
The electronic phase inverter has been discussed in Secs. 2.13 to 2.16.
In that development both direct-coupled and a-c phase inverters were
considered. Quite often, even in power amplifiers designed only for a-c
service, phase inverters are employed in order to eliminate the weight and
expense of the input transformer.
If the output of the amplifier is to be direct-coupled to the load, the
output impedance of the amplifier should be made as low as possible in
order to permit maximum power transfer. The output impedance of the
amplifier may be made low by choosing tubes such as the 6AS7 that have

Fig. 3.11. Single-ended push-pull electronic amplifier with feedback.

low rp. It is also common practice to employ a cathode-follower output

stage to improve the load matching.
In addition to producing higher efficiency and lower distortion, push-
pull operation of d-c amplifiers minimizes the effect of both plate- and
filament-supply-voltage variation.
There are several interesting variations of the conventional electronic
push-pull amplifier. Figure 3.11 shows a direct-coupled version of the
so-called single-ended push-pull amplifier.1 The input half of the double
triode is connected as an amplifier with unbypassed cathode resistance.
This acts as negative feedback for the input stage and also provides an
input for feedback around the whole amplifier. The second half of the
double triode is a phase inverter of the cathode-follower type discussed
in Sec. 2.14. The phase inverter is direct-coupled to the power-output
tubes. The upper output tube is driven from grid to cathode, not grid to
ground. The power-tube currents add in the load, and their grid voltages
are in phase opposition; thus this is a true push-pull amplifier. Another

1 A. Peterson, and D. Sinclair, A Single-ended Push-pull Audio Amplifier, Proc.

IRE, vol. 40, p. 7, 1952.
Sec. 3.3] POWER AMPLIFIERS 125
way to look at the circuit is to consider the upper output tube as a cathode
follower and the lower output tube as a variable cathode resistor.
In Fig. 3.11 one terminal of the output is at ground potential, which is
the reason for calling this a single-ended amplifier. The one disadvantage
of the circuit as shown in the figure is that the quiescent current in the
load is not zero. By the addition of a B~ supply the quiescent load cur¬
rent can be made zero. In this case the lower load terminal will be
returned to ground rather than to the cathode of the lower output tube.
It will be observed that the plate current of the phase-inverter stage
must flow through the lower output tube; thus the design of the output
stage and the phase inverter are interdependent. It will be found that
the choice of tube types is restricted by the fact that the quiescent phase-
inverter current establishes the operating points of the output tubes. In
order to maintain balanced gain through both halves of the phase
inverter, the cathode resistor and plate resistor must be kept essentially
equal; thus the grid bias on the output tubes will be equal. However,
because of the additional phase-inverter current carried by the lower out¬
put tube, its operating point will differ somewhat from that of the upper
tube. The design procedure is as follows:
1. Choose a phase-inverter tube that will permit an adequate plate-
current swing to handle the required signal amplitude. Set its approxi¬
mate operating point by choosing its resistors and supply voltage.
2. Choose an output-tube type that will handle the required load swing
with the plate voltage and grid bias established in 1.
3. Modify 1 in order to conform more closely to the requirements in 2.
This may entail changing tube types as well as supply voltages and
resistors.
4. The restrictions placed on the circuit are extreme, and it may not be
possible to satisfy the operating requirements for all three tubes in an
optimum manner. Then there are several additional possibilities. First,
it is not necessary that the B+ supply and the B~ supply be equal.
Second, a small, 200- to 500-ohm variable resistor may be placed in the
cathode lead of the lower output tube in order to provide an additional
adjustment. In any case the design is essentially a trial-and-error process.
Push-pull transistor amplifiers can also be adapted for d-c applications.
The transistor power amplifier shown in Fig. 3.7 can be modified for d-c
operation by using a d-c phase inverter and operating the amplifier
directly into a split-phase load without a matching transformer. For
maximum power transfer the designer must have the input-impedance
specification of the load at his direction. This solution of the problem is
of the brute-force type and does not take advantage of the unique potenti¬
alities of the transistor. Full use of these potentialities is made in the
complementary push-pull amplifier.
126 CONTROL SYSTEM COMPONENTS [Chap. 3

The principle of complementary operation was first discussed by

Sziklai.1 The Sziklai amplifier employs a matched pair of n-p-n and
p-n-p junction transistors, as shown in Fig. 3.12. Since the transistors
have characteristics each of which is the negative image of the other, they
operate without the requirement of external phase reversal at the input
and output.
In practically every sense this is the ideal push-pull amplifier. Both
the input and output are single-ended and operate at ground potential.
No expensive and complicated coupling
circuits are required, and the circuit is
simplicity itself. Matched p-n-p, n-p-n
junction pairs are essential to the success¬
ful operation of this circuit, and several
such pairs are currently available. For
instance, the 2N68-2N95 and the
2N101-2N102 pairs are available for am¬
Fig. 3.12. The Sziklai amplifier
employing complementary junc¬ plifiers in the 10-watt range, and the list
tion transistors. is soon to be extended. The conven¬
tional class B analysis holds here.
3.4. Thyratron Amplifiers. For many medium- and low-power appli¬
cations thyratron control represents the most economical and lightweight
solution. Thyratrons are essentially instantaneous in operation com¬
pared to the usual servo frequencies. Mercury-vapor thyratrons are
somewhat sensitive to their environment, but inert-gas tubes have been
developed that are as rugged as high-vacuum tubes.
The operation of a thyratron, or gas-filled tube, is fundamentally differ¬
ent from the operation of a vacuum tube. The gas tube acts as a relay
or a switch rather than as a proportional device. Conduction cannot
take place as long as the grid is held negative with respect to the cathode.
If the grid-to-cathode voltage is gradually made less negative, the tube
will suddenly begin conduction if the anode is positive with respect to the
cathode. The grid-to-cathode potential at which conduction begins is a
function of temperature and anode-to-cathode voltage. Figure 3.13
shows the grid potential for initiation of conduction plotted against posi¬
tive anode potential for a typical thyratron.
Conduction is made up of electron flow from cathode to anode and by
positive-ion flow from anode to cathode. Immediately after the start of
conduction a sheath of positive ions builds up surrounding the grid and
effectively insulates it from the remainder of the gas; thus the grid has no
further effect on conduction2 once the gas is ionized. The current flow is

1 G. C. Sziklai, Symmetrical Properties of Transistors and Their Applications,

Proc. IRE, vol. 41, p. 717, 1953.
2 Dow, op. cit., chap. 21.
Sec. 3.5] POWER AMPLIFIERS 127

not limited by the grid potential, and the conducting thyratron is effec¬
tively a short circuit from anode to cathode. If the anode current is not
limited by the load, the tube will be damaged by the excessive current
flow. There is a small voltage of 10 to 20 volts, depending on the gas,
across the tube that is approximately independent of current, owing to
the mechanism of gas conduction.1 In order to stop conduction, the

Fig. 3.13. Typical thyratron firing characteristics. Grid potential for initiation of
conduction is plotted against positive anode potential for a typical thyratron.

positive anode voltage must be removed. This is usually accomplished

by supplying the anode with an a-c anode voltage, thus operating the
thyratron as a rectifier in which the average current is controlled by the
grid potential.
3.5. Control of Firing Angle. Although there are a number of methods
for controlling the point at which the thyratron begins to conduct, the

[b] Firing angle <f>a

Fig. 3.14. Relation between anode and bias voltages for combination of a-c and d-c
bias.

method most commonly used in control systems utilizes a combination of

a-c and d-c grid bias together with a d-c signal voltage. As shown in
Fig. 3.14, the a-c component of the grid bias consists of a voltage at line
frequency, which lags the voltage applied to the anodes by 90°. The d-c
1 Ibid.
128 CONTROL SYSTEM COMPONENTS [Chap. 3

component of the bias is approximately equal to the amplitude of the a-c

component; hence with no signal the grid voltage just misses crossing the
curve representing the critical value for which conduction starts. A posi¬
tive signal applied in addition to the two bias voltages moves the total-
grid-voltage wave up so that it now intersects the critical line at the angle
4>a shown in Fig. 3.146. The angle <f>a is the firing angle, and the tube
conducts during the interval shown shaded in Fig. 3.146. Note that an
increase in signal voltage results in a smooth advance of the firing angle.
If the amplitude of the a-c component of the bias is very much larger than
the critical grid potential, then the firing angle is approximately

| + sin- (i _ g) (3.9)

where E{ is the signal voltage and Edc is the d-c bias, which is assumed to
be approximately equal to the amplitude of the a-c bias. For full control
of the firing angle from 180 to 0°, E{ varies from zero to 2Edc. Thus the
required signal range for full control is determined by the amplitude of
the a-c bias. This can also be shown by computing the gain, or the
change in firing angle degrees per volt of input signal, by differentiating
Eq. (3.9) with respect to E{. This gives

d(f>a 1
(3.10)
dEi ■\Z2EiEdc E,2
If this is evaluated for E{ = Edc, that is, for <f>a = 90°, the gain becomes

d<t>a
(3.11)
dEi 4>a = 90° Edc

and is thus shown to be inversely proportional to the d-c bias. Thus the
gain can be increased by decreasing the d-c bias and therefore the peak
value of the a-c bias. Beyond a certain point, however, this reduction
leads to difficulties. One of the important advantages of the control
method using a-c bias is that the intersection of the grid voltage with the
critical control line is sharp and not greatly affected by small variations
of the critical voltage due to temperature or random effects. This advan¬
tage is largely lost if the amplitude of the a-c bias is reduced too greatly,
and erratic firing may result.
The firing angle can also be controlled by shifting the phase of the a-c
bias relative to the phase of the anode. In this method a fairly large a-c
grid voltage is used, usually without d-c bias. The major disadvantage
of this method is the difficulty of shifting the phase of the grid voltage.
Although a number of circuits are available for this purpose,1 they usually
1 Reich, “Theory and Applications of Electron Tubes,” McGraw-Hill Book Com¬
pany, Inc., New York, 1944, pp. 510-515.
Sec. 3.6] POWER AMPLIFIERS 129

require a change of resistance or other circuit parameter. Hence the

phase-shift method may occasionally be used when the control signal is in
the form of a shaft position which can be used to vary an adjustable
resistor. It is also possible to shift the phase electronically by the use of
a reactance-tube circuit1 or a saturable reactor. However, these methods
are not so commonly used in servo systems as the control method using
a-c bias described at the beginning of this section.
Extremely accurate control of the firing angle is possible by applying
sharp voltage pulses to the grid.2 In this method the signal is applied to
a pulse-forming circuit in such a way that the time of the pulse is con¬
trolled by the signal. Although the firing angle can be regulated very
precisely by this method, the circuitry required to produce the necessary
pulses is usually too complicated to justify its use.
Direct application of the d-c signal to the grid without any a-c bias also
permits a certain control of the firing angle because of the slight curvature
of the critical-grid-voltage line. Despite its apparent simplicity, this
method is not used in general, because it permits control of the firing
angle only between 0 and 90°. Furthermore, since it depends on the
curvature of the critical-grid-voltage line, it is very sensitive to any vari¬
ations of this line, and firing tends tq be erratic.
3.6. The Thyratron Amplifier with an Inductive Load. A simple thy-
ratron control circuit which might be used to control the current flowing
in the field of a d-c motor is shown in
A-c supply
Fig. 3.15. The relative magnitudes
of the load inductance and resistance
tm!
in this circuit have an important
h— es —H T
effect on the waveshape of the cur- ' er <>/?
rent that flows during periods of tube 1
t Vr- /• J t k

conduction. If the inductance is Control Control ‘ V

£'L §
signal
relatively small, the current and volt- s,9w!
age waveshapes are identical. How¬ l i

ever, if the inductive reactance is

appreciable, the build-up and decay of *IG' Half-wave thyratron circuit
.. , • i i i T.i . with inductive load,
the current is delayed, and the current
waveshape departs from that of the voltage. The determination of the
average load current as a function of firing angle becomes a rather involved
task under these conditions. Both graphical and analytical techniques
are used. The graphical method has the advantage of displaying the
fundamental processes taking place in the circuit somewhat more clearly,
but when accurate numerical results are required, the analytical method is
usually preferable.
1 Ibid.
2 J. G. Skalnik, Pulse Control Thyratron, Electronics, December, 1949, p. 120.
130 CONTROL SYSTEM COMPONENTS [Chap. 3

Typical voltage and current waveshapes found in the half-wave circuit

of Fig. 3.15 are shown in Fig. 3.16. The components of voltage are
labeled in Fig. 3.15. At point a the tube ionizes, and eT drops to a very
small value that is essentially independent of anode current. The
remainder of the voltage es divides across R and L. The voltage ez, is
equal to L di/dt, and er is equal to Ri.
The voltages eL and er may be found
by point-by-point calculations. Cal¬
culations should be made in a logical
sequence, e.g., that given by Chin
and Moyer.1 At point a the current
is zero; therefore iR is zero, and

eL = es — eT (3.12)
Fig. 3.16. Voltage and current wave¬
where er is a constant, usually taken
shapes for circuit in Fig. 3.15.
as 15 volts or perhaps neglected,
Having found cl at point (a), we write

di _ eL
(3.13)
dt L

thus establishing the slope of the current wave. The initial slope of er is
found by multiplying the initial value of di/dt by R. The increment over
which this initial slope is extended depends on the accuracy required for
the results. If it is sufficient to find the initial slope, the maximum value,
and the point at which er once more becomes zero, the calculation is fairly
simple. We know that the maximum value (or zero slope) of the iR volt¬
age wave falls at point b, since at that point eL and therefore di/dt are
zero, and
er — es — er (3.14)

As yet, we have not determined point b. We pick a point b such that the
integral of cl from a to 6 times R/L is equal to the value of er at point b.
This may be seen by noting that
r di
eL = (3.15)
LJt
Therefore Jb eL dt = fib Ldi
=0Jla
(3.16)

and Jb eL dt = Lib (3.17)

@rb
Now ib = (3.18)
R

1 P. T. Chin and E. E. Moyer, Controlled Rectifier Circuits, Trans. AIEE, vol. 63,
p. 501, 1944.
Sec. 3.6] power amplifiers 131

Substituting (3.18) into (3.17), we have

(3.19)

The integral is proportional to the area shown crosshatched in Fig. 3.16.

The argument used to obtain the build-up of current to point 6 also
holds during the decay of the current. The point c at which i becomes
zero, the so-called extinction point, may be found in the same manner as
point b. A point c is chosen such that the shaded area between points b
and c is equal to the crosshatched area between points a and b found above.
Since an inductor must return as much energy to the line as it receives and
since the two areas represent positive and negative energy, respectively,
we see that the areas must be equal.
In general, analytic methods are preferred over graphical methods.
We shall apply analytic methods to the same circuits as an example.
Note that the voltage applied to the load is

e(t) = Es sin (tot + 0a) — ET (3.20)

where Es is the amplitude of the supply voltage, ET the tube drop, co the
line frequency in radians per second, and 0a the firing angle. The time
origin (t = 0) is the time at which the tube fires.
The instantaneous load current is best found as a function of time by
use of Laplace transform methods. The Laplace transform of the voltage
of Eq. (3.20) is
Es(co cos 4>a + s sin 0a)
(3.21)
s2 + CO2

The impedance of the circuit in terms of the complex variable s is

Z R + Ls — L (3.22)

Therefore the transform of the load current is

Es co cos 0a + s sin 0a _ ET 1
(3.23)
L (s2 T w2) (s T R/L) L s(s T R/L)

To find the current as a function of time, we perform an inverse transfor¬

mation on this expression. This results in

Y/\ _
sin (cot + 0a — 0) — sin (0a — 6) exp ( — ^ t

Et R ,
1 — exp (| — j-1
i (3.24)
~R

In this equation Z = \/R2 + co2L2 and 6 = tan-1 coL/R. Although this

132 CONTROL SYSTEM COMPONENTS [Chap. 3

expression can be used to find the instantaneous value of the load current
at any time t, it cannot be solved directly for the time at which the current
goes through zero, i.e., the extinction point, because of the simultaneous
presence of exponential and trigonometric functions of time. It is, how¬
ever, possible to obtain the same information indirectly by assuming an
arbitrary value of cct and finding the value of firing angle <f>a required to
make the current go through zero at this point. If i(t) in Eq. (3.24) is
set equal to zero, the equation becomes

ErZ
sin (cd + (f>a — 6) — sin (0„ — S) exp ( — 1
) ESR
(3.25)
This equation may be rearranged to give

<t>a — 0 + 7T

_EtZ[ 1 - exp (~(R/L)t)\_

— sin-1
ESR -\/l — 2 [exp ( — (R/L)t)] coseot + exp ( — 2(R/L)t)_
sin cot
tan —i
(3.26)
cosect — exp ( —(R/L)t)

Some care must be used in evaluating the inverse trigonometric functions

in this expression, since these functions are multivalued. If the result is
examined carefully, however, it is clear that the sin-1 function represents
angles from 0 to 90° while the tan-1 function represents angles from 0 to
180°. The time t in Eq. (3.26) is the time at which the tube conduction
ceases, measured from the firing instant. Hence if (f>c is the extinction
angle, we have
<t>c — <f>a — wt (3.27)

where both 0a and 0C are measured from point 0 in Fig. 3.16. Sub¬
stituting for cot in Eq. (3.26), we obtain

0a = 6 + 7T — sin-1
1_1
1

^ S3

EtZ
■e-

1 — exp
1

1
1

1_1

1-1

ESR -\Jl 2 exp _ R (0c

-e-

0a) COS (0c — 0a) + exp

1

1
o

coL
3

sin (0C - 0a)

— tan —i (3.28)
R
COS (0c — 0a) — exp ..T (0c 0a)

Thus we find that the firing angle 0a can be determined analytically in

terms of the difference between extinction and firing angles. Note that a
considerable simplification results if the tube drop ET is neglected, since
an entire term of ,Eq. (3.28) will then disappear. Also note that, if the
Sec. 3.6] POWER AMPLIFIERS 133

term (R/coL)((f)c — <f>a) is large (greater than about 3), all the exponential
terms become negligibly small. Then, approximately,

0a 6 + 7r — sin-1 — 0c + 0a

EtZ
or 0c 6 + 7r — sin-1 (3.29)
ESR

This indicates that, if the inductance is sufficiently small and if the differ¬
ence between extinction and firing angles is sufficiently large, the extinc¬
tion point becomes independent of the firing point. A typical set of

Fig. 3.17. Relation between firing angle and extinction angle for half-wave circuit with
inductive load. Tube drop is neglected.

curves of 0a versus 0C for various values of ooL/R is given in Fig. 3.17.

The tube drop has been neglected in drawing these curves.
Once the extinction point has been found, the determination of the
average load current becomes relatively simple. Although it is possible
to find the average current directly by integrating Eq. (3.24) over the
proper limits, it is easier first to find the average voltage and then to
divide by the load resistance. For the half-wave circuit considered here,
134 CONTROL SYSTEM COMPONENTS [Chap. 3

the average voltage applied to the load is

E„ = 4 / C®. sin * - et) d<t>

J <t>a

= [Xs (cos <t>a - COS 0C) — ET(<t>c - 0a)] (3.30)

The average load current is now simply E&y/R or

hv = 2~R (C0S 0a ~ C0S 0c) - ET(4>c - 0a)] (3.31)

If the inductance is sufficiently small and if the firing point comes suffi¬
ciently early so that (i2/o>L)(0c — 0O) > 3, the extinction angle may be
found from Eq. (3.29). Under these conditions the average current
becomes

J- n.v Es cos 0a + ^ VEW Et2Z2

2tR
, „ ,x , ^ . .x — tt +, sin-1
. xetz'
+ Et 1 v + cos 0a — tan-1 „ „ (3.32)
it iLsK

where X = coL, Z = (jR2 + X2)1^, and where we have used the fact
that 6 = tan-1 uL/R. If the tube drop ET is negligible, this becomes

+ S)

Here it is clear that an increase of inductance, which increases Z, will

result in less average current.
When the inductance is very large, so that R/ccL ~ 0, Eq. (3.28) may
be solved approximately to give

0C ~ 7T + 26 — 0O
But if the inductance is very large, d « tt/2. Hence approximately

0c ~ 2tT 0a (3.34)
This result implies that for very large inductance the conduction period
for negative values of instantaneous supply voltage is approximately
equal to the conduction period for positive voltage; i.e., the shaded area
in Fig. 3.16 is equal to the crosshatched area. The average current is
therefore approximately equal to zero. This is borne out analytically
by substituting Eq. (3.34) into Eq. (3.31). If the tube drop is neglected,
the result will be seen to be zero. In practice the load current is, of
course, not zero, since Eq. (3.34) is only approximate; however the effect
of large inductance in severely reducing the average load current should
be evident.
Sec. 3.6] POWER AMPLIFIERS 135
Curves of average current as a function of firing angle for several values
of ooL/R are given in Fig. 3.18. These curves are obtained as follows:
First, a number of values of </>c — </>a are chosen and used in Eq. (3.28) to
determine <f>a. Then (f>c can also be found, and the current is then
obtained from Eq. (3.31). The tube drop has been neglected, and the

0 20 40 60 80 100 120 140 160 180

<f>Qy degrees

Fig. 3.18. Normalized average load current versus firing angle for half-wave thyratron
circuit. Tube drop is neglected.

curves have been normalized with respect to the maximum current in a

purely resistive load, Es/ttRa.
In order to increase the load current in half-wave circuits driving highly
inductive loads, use is sometimes made of the so-called freewheeling
circuit, shown in Fig. 3.19a. In this circuit the diode across the load

A-c

[o) Schematic circuit [£) Typical wave forms

Fig. 3.19. Freewheeling circuit.

conducts during the negative half cycle of a-c supply voltage and provides
a low-resistance discharge path for the load current. Thus the load cur¬
rent is not forced down by the negative supply voltage, as is the case in
the half-wave thyratron circuit without a diode, but instead it follows a
more slowly decreasing, exponential discharge curve. This is shown in
136 CONTROL SYSTEM COMPONENTS [Chap. 3

Fig. 3.196, which shows both the voltage and current waveshapes. If the
tube drop of the diode is negligible, the load voltage during negative half
cycles of a-c supply voltage is zero, and the waveshape is identical with
that observed on a half-wave circuit with a pure resistance load. Hence
the average load current is equal to the average current found for pure
resistance, (E8/2ttR){\ + cos 0O).
3.7. The Full-wave Thyratron Amplifier with Inductive Load. The
full-wave circuit shown in Fig. 3.20 is more commonly used than the half¬
wave circuit considered in Sec. 3.6. Since there are two voltage pulses
per cycle of supply voltage, a smoother current flow to the load results.
The analysis of the full-wave cir¬
cuit is complicated somewhat by
the fact that under certain condi¬
tions the load current may be con¬
tinuous. This happens, in fact,

A-c
<? supply <?

Fig. 3.20. Full-wave thyratron circuit. Fig. 3.21. Voltage and current wave-
Rg limits the current drawn by the con- shapes in full-wave thyratron circuit,
trol grid.

whenever the extinction angle of one tube as given by Eq. (3.28) occurs
later than the firing point of the other tube. Typical waveshapes of
voltage and current, for both the continuous and discontinuous case, are
shown in Fig. 3.21. Since in normal operation the firing points of the two
tubes are 180° apart, the condition for continuous current flow is
0c 0a ^ TT

where both 0C and 0a are measured with respect to the same origin and on
the same tube. Substituting this relation into Eq. (3.28) gives
EtZ 1 — exp ( — Rtt/uL)
0a, < 0. — sin-1 (3.35)
ESR 1 + exp ( — Rtt/uL)
Sec. 3.7] POWER AMPLIFIERS 137
for the condition for continuous current flow. In making this substitu¬
tion, it should be noted that the last term of Eq. (3.28) must have the
value 7r rather than zero.
The determination of average voltage and current is now reasonably
straightforward. If the circuit operates in the discontinuous mode, the
voltage and current are found exactly as for the half-wave circuit except
that the averages must be taken over one half cycle. Thus

1 Pc
E&v = - / (Es sin $ — Et) d<f)
^ J <t>a

= - [E8 (cos 0a — cos <f>c) — ET(<t>c — 4>a)] (3.36)

7T

while the average load current is

hv = ^ [Es (cos </>a — cos 4>c) — Et(4>c — <f>a)] (3.37)

If the inductance is small and if the firing point comes early enough,
expressions similar to Eqs. (3.32) and (3.33) can be derived for the full-
wave circuit; in fact the current is simply twice the value given in these
equations.
When the circuit operates in the continuous mode, the average voltage
and current are obtained more easily, since it is only necessary to integrate
from one firing point to the next. This is made clear by inspection of
Fig. 3.216. Thus, for the continuous mode the average voltage is given
by
-j r <£a+ir
Eav = - / (Es sin </> - Et) d(f>
. 4>a

= - Es cos <f>a — Et (3.38)

7r

and the average current becomes

7.v = ^ E, cos (3.39)

Although the inductance no longer appears explicitly in Eq. (3.39), the

load current is still affected by it, since the inductance is responsible
for the continuous conduction of the tubes. This causes negative voltage
to be applied to the load for part of the cycle and results in a reduction
of average current (see Fig. 3.21). The rate of increase of current with
firing angle is, however, more rapid in the continuous mode than in the
discontinuous mode. This is indicated by the sharp change in slope of
the curve of current versus firing angle. Typical curves of this type are
shown in Fig. 3.22.
138 CONTROL SYSTEM COMPONENTS [Chap. 3

3.8. Thyratron-amplifier Transfer Function. Up to this point only

the static performance of the thyratron amplifier has been considered.
In control applications one is, however, also very much interested in the
transient performance or, in general, in something equivalent to a transfer
function. The transient behavior of the circuit depends on whether it is
operating in the discontinuous or the continuous mode. If the circuit
operates in the discontinuous mode, each current pulse represents an
entirely new transient, which is completely independent of the previous
current pulse. Thus in the discontinuous mode the thyratron amplifier
may be characterized by a one-cycle response time; i.e., the current
reaches a new average value within one cycle of the change of firing angle.

Since the current in the half-wave circuit is always discontinuous if no

freewheeling diode is used, this circuit always has a half-cycle response
time. However, when the full-wave circuit enters the continuous mode,
its behavior changes. The average voltage still has a one-cycle response,
but the current now increases exponentially with the time constant L/R.
It should be noted that, since the thyratron amplifier is a nonlinear
circuit, a transfer function in the usual sense does not exist. However,
in the continuous case one may define a “quasi-linear” transfer function,
which relates small variations of load current to small variations of firing
angle. Such a transfer function has the form

A/ = _
(3.40)
A$a (L/R)s + 1

In this equation k is the slope of the curve of current versus firing angle
and is a function of the operating point around which the small changes
AI and A<£a are taken. •
Sec. 3.9] POWER AMPLIFIERS 139
3.9. The Thyratron Amplifier with a D-C-motor Load. Thyratron
amplifiers are quite often used to control low- and medium-power d-c
motors. A simple half-wave circuit for unidirectional speed control is
shown in Fig. 3.23 with all voltages and currents labeled. The motor
field is assumed to be excited by a separate d-c supply and is assumed to
be fixed.
A d-c-motor load differs from loads
considered up to now in that it gen¬
tmJ
erates a counter emf Eg = kv f2, where
12 is the speed and kv the velocity volt¬
age coefficient (see Sec. 4.15). When
the speed is constant, this counter
emf is a fixed voltage and may be rep¬
resented by a battery in the load cir¬ Fig. 3.23. Half-wave thyratron circuit
with motor load.
cuit. As a result of the presence of the
counter emf, the thyratron may not fire when the grid potential passes the
critical value, and it is convenient, therefore, to define three modes of
operation for the circuit. These modes are defined by reference to Fig.
3.24.
When the grid passes the critical value in region 1 of Fig. 3.24, the tube
cannot conduct, since the voltage on its plate is negative. However,
unless the grid signal consists of short pulses, the tube will conduct as
soon as the instantaneous a-c supply
voltage exceeds the counter emf Eg (point
a in Fig. 3.24). Thus in mode 1 changes
in firing angle that do not go beyond
point a have no effect on the current
flowing in the motor. A firing angle
occurring anywhere between points a and
b in Fig. 3.24 puts the circuit into oper¬
Fig. 3.24. Operating modes of ating mode 2. In this mode the circuit
thyratron amplifier with motor
operates normally, with the tube firing
load.
as soon as the critical grid potential
is exceeded. Mode 3 is defined by a firing point occurring in region 3.
The armature current is always zero in this mode. The three modes are
defined mathematically by the following three equations:

For mode 1: 0 < 4>a < sin-1

s

For mode 2: sm-11 ~

• -^0 s A
< (f>a
' s
< ‘1
7T - sin-1
. ' ^0
(3.41)
-t-J S -t-J s

For mode 3: 7T > </>a > 7T - Sin'1 ~

hl S>
 140 CONTROL SYSTEM COMPONENTS [Chap. 3

where </>« is the firing angle. Note that the sin-1 functions used in these
equations are the so-called principal values. Note also that, for a given
value of (f)a, the circuit can be in only one of two modes. Thus for
4>a < tt/2 the circuit goes from mode 2 to mode 1 as Eg is increased. For
<t>a > 7r/2 the circuit goes from mode 2 to mode 3 as Eg is increased.
The analysis of the operation of the motor is simplified considerably if
it is assumed that the inductance of the armature circuit is negligible.
This is true for most d-c motors, particularly those designed for use with
thyratrons, because, as will be shown later, the presence of inductance
reduces the maximum current and developed torque. Thus, in the fol¬
lowing analysis we assume that the inductance is zero. Also, it is con¬
venient, although not necessary, to neglect the tube drop Et-
If a firing angle and the speed of the motor are given, the average arma¬
ture current can be found as follows: If the circuit is in mode 1, the tube
conducts during the time interval bounded by points a and b in Fig. 3.24.
Hence the average current is

r 4>b
I = (Es sin </> — Eg) d<t>
2ir Ra <t>a

1
[Es (cos 4>a — cos (t>b) — Eg(4>h — </>«)] (3.42)
2 7rRa

where Ra is the armature resistance. But 0a and 0b are defined by

(3.43)

Hence (3.44)

If the circuit is in mode 2, the thyratron fires at the firing point <f>a,
which is now no longer determined by Eg but may have any arbitrary
value. Hence

V —sin"1 (Eg/Ea)

{Es sin 6 — Eg) dd (3.45)

2tt Ra <t>a
Es E Eg • .Eg
cos 0a + A /1 (3.46)
2irR a Eh2 + § (*. - x) + ^ j?sm
■t-J .9
W
-tij o

Since the developed torque of a motor is given by Qd = kj [see Eq. (4.56)]

and since Eg = kv£l, Eqs. (3.45) and (3.46) may be rewritten in terms of
torque, speed, and-firing angle as follows:
Sec. 3.9] power amplifiers 141

For mode 1:

s -J

(3.48)

These two equations can be used to plot curves of developed torque versus
speed for different firing angles. Such a set of curves is shown in Fig.
3.25. These curves have been normalized with respect to maximum

Fig. 3.25. Speed-torque curves for thyratron-controlled motor.

torque and speed of the motor. The points of transition between modes
are also indicated in this figure. Note that the transition from mode 1 to
mode 2 is not abrupt; in fact it can be shown by differentiating Eqs. (3.47)
and (3.48) that the slopes of the two curves are the same at the transition
point. The same is true for the transition from mode 2 to mode 3.
A number of interesting points can be deduced from Fig. 3.25. Note
that, for firing angles less than 90° and for small developed torque, the
motor operates almost entirely in mode 1. Its speed is therefore not con-
142 CONTROL SYSTEM COMPONENTS [CHAP. 3

trolled by the firing angle; in fact it appears from the curves that the
no-load speed for all firing angles less than 90° is equal to the maximum
speed. If the no-load speed of a motor is to be controlled, firing angles
of more than 90° must be employed. At these firing angles the circuit
operates, however, in a highly nonlinear on-off region, i.e., the boundary
between mode 2 and mode 3. Under certain conditions this may give
rise to a nonlinear type of oscillation. Thus, consider the circuit shown
in Fig. 3.26. Suppose that for some reason the motor is running too fast.
The feedback circuit acts to retard the firing angle. If the motor is
running at no load, the circuit goes into mode 3, and the armature current
is cut off. The motor now coasts until its speed is less than the reference
speed. If there is a relatively large time lag in the controller, the motor

Fig. 3.26. Block diagram of typical thyratron motor-control system.

may slow down too much before the thyratron fires again and may then
again overspeed and coast. The action resembles the bouncing of a ball
being pushed along a horizontal plane by a block moving at uniform
speed.
By differentiation of Eqs. (3.47) and (3.48) it can be shown that the
slope of the curves for zero developed torque is zero. This means that
the speed regulation near zero torque is very great; i.e., the motor must
slow down considerably to support a relatively small torque change. On
the other hand, near full torque the curves are all approximately linear.
They may, therefore, be used in determining a quasi-linear transfer func¬
tion, as described in Sec. 4.16. In general, the effect of the thyratron
amplifier can be approximated by a series resistance which tends to
increase the motor time constant.
3.10. The Effect of Armature Inductance. In the preceding section
we have considered the armature inductance of the motor to be negligible.
However, since the ratio of inductive reactance at line frequency to the
resistance is the significant parameter determining the effect of the
inductance, this assumption is not always completely justified. In
Sec. 3.10] POWER AMPLIFIERS 143
general, however, it is true that the inductance, even if not completely
negligible, is small.
An accurate analysis of the effect of inductance was shown in Sec. 3.6
to be a rather involved task. However, if the inductance is small, its
effect may be taken into account approximately without too much diffi¬
culty. We assume, therefore, that the inductance is sufficiently small
so that 2ir,}La <<C Ra, where / is the line frequency.
During the period of tube conduction, the armature current may be
thought of as consisting of two components, an a-c component due to the
a-c supply voltage and a d-c component. If the inductance is small, the
a-c component will be approximately sinusoidal in shape, particularly
toward the end of the conducting period when the initial switching
transient can be assumed to have decayed. Also, if the inductive reac¬
tance is small compared to the resistance, the amplitude of the a-c com¬
ponent is approximately Es/Ra. Thus, the only effect of the inductance
on the a-c component of the armature current is to delay it, so that its
cycle is behind the a-c supply voltage by the angle tan-1 (2TrfLa/Ra)- The
d-c component of the current cannot be affected by the inductance at all.
Hence, we find that the instantaneous armature current during the period
of tube conduction is approximately

_ tan-1 J (3.49)
‘ti'a / -tv a

Once the thyratron has been fired, it continues to conduct until the plate
current goes through zero. By setting the instantaneous current equal
to zero in Eq. (3.49), we find that
the extinction point is given by

2trft = <t>'b = sin-1 ^ + tan-1

J-J s -tv a

(3.50)

Thus note that to a first approxima¬

tion the effect of a small amount of
inductance in the armature circuit
is to delay the extinction point by Fig. 3.27. Waveshape of motor-terminal
voltage when inductance is present in the
the angle tan-1 (27rfLa/Ra)- Since
armature.
the motor-terminal voltage is equal
to the applied a-c voltage during tube conduction and is equal to Eg
during periods of nonconduction, the waveshape of the terminal voltage
has the form shown in Fig. 3.27, which shows operation in mode 2.
As can be seen from the figure, the delay in the extinction point reduces
the area under the applied-voltage curve, and it therefore reduces both
the average voltage and current. The reduction in area is equal to the
144 CONTROL SYSTEM COMPONENTS [Chap. 3

section shown shaded in Fig. 3.27. If tan-1 (2xfLa/Ra) is small, this

shaded area is approximately triangular in shape and is therefore approxi¬
mately equal to
d_
AA = K (Es sin <f>) (3.51)
d(f> <t> — <t>b

But since <f>b = x — sin-1 E0/Es,

d_
(Es sin </>) = Es cos
d($> <t> = 4>b

= - VE'2 - E„2 (3.52)

Also for small values of the argument, tan-1 x « x. Therefore, the

shaded area in Fig. 3.27 is approximately equal to

1 /2tr/LaV
AA = - ? 2 (3.53)
2 V Ra ) Ve;- - e

and the reduction in the average current is therefore

A/
AA J_ /2t/L0V (3.54)
Ve.2 E 2
2ttR 47T^a \ Ra /

The reduction in current will, of course, result in reduced torque. It is

clear from Eq. (3.54) that the reduction depends on Eg and therefore on
the speed. It is zero when the motor runs at maximum speed and is
largest at standstill.
3.11. D-C Motor Driven by Full-wave Circuit. For smoother opera¬
tion of the motor, the full-wave circuit shown in Fig. 3.28 is often pre-
<? A~C 9
supply

Fig. 3.28. Full-wave thyratron circuit with motor load.

ferred over the half-wave circuit discussed in Sec. 3.10. The operation of
the circuit is quite similar to the half-wave circuit as long as the current
is not continuous. In fact, all that has been said concerning operating
modes, speed-torque curves, effect of inductance, etc., applies to the full-
wave circuit, except that, for a given firing angle and speed, the armature
Sec. 3.12] POWER AMPLIFIERS 145

current and therefore the torque are twice as great as in the half-wave
circuit. However, as was pointed out in Sec. 3.7, under certain condi¬
tions the current in the full-wave circuit may be continuous. If the
inductance is negligible, this can happen only for negative speeds and is
therefore rather unlikely during normal operation. However, if the
armature inductance is sufficiently large, the current may become con¬
tinuous if the speed is low enough and the firing angle small enough. The
extension of the theory to this case is straightforward and is left to the
reader.
3.12. Thyratron Circuits for Reversing Control. Two commonly used
circuits permitting bidirectional motor-speed control are shown in Fig.

Control
signal 9 9

Fig. 3.29. Thyratron circuits for bidirectional motor-speed cbntrol. Both are half¬
wave circuits.

3.29. Both of these circuits are half-wave circuits in which one tube con¬
trols the motor when it is running in one direction while the other tube
controls it in the opposite direction. The transformer connections in the
circuit of Fig. 3.296 are such that the two tubes conduct on alternate half
cycles. This is necessary, since there would otherwise be a short circuit
through the tubes and transformer windings whenever both tubes con¬
ducted simultaneously. This problem does not exist in the simpler cir¬
cuit of Fig. 3.29a; however, in it also the two tubes can conduct only on
alternate half cycles. For smooth transfer of control from one tube to
the other near zero motor speed, the grid bias is usually so adjusted that
for zero signal input both tubes fire for a short part of their respective
half cycle. An increase of signal in the positive direction advances the
firing angle of one tube, say, tube 1, and retards the angle of the other
tube (tube 2), until eventually only tube 1 conducts. For negative
signal the action is reversed.
When only one tube conducts, the analysis of the circuit is exactly the
same as that carried through in Sec. 3.9. However, in the interval near
zero speed the effect of the two tubes operating together must be con-
146 CONTROL SYSTEM COMPONENTS [Chap. 3

sidered. For the sake of simplicity we assume that the armature induct¬
ance and the tube drop are negligible.
The waveshape of the armature current is shown in Fig. 3.30 for oper¬
ation in mode 2. The average armature current is found by integrating
over the two pulses of current occurring per cycle. We have, therefore*

'ir — sin 1 Eg/Es

I AV (Es sin 0 — Eg) d4>
27rRa <t>al
'2ir+ sin-1-Z?c(/.Z£a
+ Es(sin 0 Eg) d(f)
</>o2

[Es (COS <f)al + COS </>a2) - Eg(3TT — 4>al - 0a2)] (3.55)

27rjRa

Suppose now that 0ai = </>«o — A0a and 0a2 = tt + 0aO + A</>a. This
implies that for zero signal the firing angle of tube 1 is </>ao and that the

Fig. 3.30. Armature-current waveshape for the circuit of Fig. 3.29.

firing angle of tube 2 comes 180° later. Also the signal A</>a advances the
firing angle of one tube by the same amount that it retards the firing
angle of the other. Substituting into Eq. (3.55), we obtain after some
reduction

[Es sin <f>ao sin A<f>a 4* a 0)] (3.56)

7T E a

Since the average developed torque Qd = ktI*v and since the counter emf
Eg = kvtt, as in Sec. 3.9, we convert Eq. (3.56) into a relation between
speed and torque:

A [E8 sin </>a0 sin A</>a — kvQ(ir 0a 0) ] (3.57)

TcEa

This relation holds only when both tubes are conducting and are in
mode 2. If it is assumed that the grid-control circuit limits variations
of firing angles to the range between 0 and 180° on each tube, then
Eq. (3.57) applies'only if A0« meets all of the following conditions:
Sec. 3.12] POWER AMPLIFIERS 147

0 < 0aO — A(f>a < 7T

0 < (f)aO + A <{>a <C 7T

fc„12 (3.58)
sin-1 < 0oO — A <t>a <7r — sin-1
X
j Jcv 12
— sin < 0ao + A<t>a <7r + sin-
X
A composite set of speed-torque curves for the motor can now be plotted
by use of Eqs. (3.47), (3.48), and (3.57), keeping in mind the limitations

Fig. 3.31. Speed-torque curves for bidirectional thyratron control circuit. Armature
inductance and tube drop are neglected; <f)ao = 135°.

on speed and firing angles given by inequalities (3.58). Such a set of

curves for a quiescent firing angle <f>ao = 135° is given in Fig. 3.31. Of
particular interest in this figure is the fact that, in the region of zero speed
and small A<f>a, the speed-torque curves are parallel straight lines. The
slope of the curves in this region can be obtained by differentiating Eq.
(3.57) with respect to 12; the result is

dJQa (3.59)
d!2
148 CONTROL SYSTEM COMPONENTS [Chap. 3

Comparison of the speed-torque curves of a d-c motor driven by a linear

source (Sec. 4.16) indicates that the thyratron-controlled motor in its
linear region is equivalent to a linear d-c motor with a resistance equal to
RJ{ 1 — </>ao/7r). Hence it has a time constant

JRg
(3.60)
kvkt(l 0ao/7r)

It is clear that the time constant is reduced by making </>a0 as small as

possible. A small value of </>a0 also has the advantage of extending the
region of linearity of the circuit. However, </>o0 should not be made less
than 90°, since this would reduce the maximum speed available from the
motor. A disadvantage of small </>a0 is that the efficiency is reduced and
the motor heating is increased. Thus, if efficiency and heating are impor¬
tant factors, as they would be in large power installations, 0aO must be
kept as large as possible.
When (f)a0 is near 90°, the circuit may operate in mode 1, as described
in Sec. 3.9, and become nonlinear from that cause. Since such small
values of <£a0 are used relatively rarely, the analysis of this type of oper¬
ation will not be carried through here. It offers no particular difficulty.
The effect of armature inductance may be considered for the bidirec¬
tional circuit in the same way as for the unidirectional circuit. If the
inductance is small, then the reduction of current in the positive pulse is
approximately canceled by the reduction in the negative pulse, particu¬
larly at low speeds. Thus the effect tends to be even less for the bidirec¬
tional circuit than for the unidirectional circuit. I

Full-wave bidirectional circuits are seldom used because of the diffi¬

culties of avoiding short-circuit paths. Unless special precautions, e.g.,
current-limiting resistances, are employed to limit the short-circuit cur¬
rent, a full-wave circuit could never be operated with the firing points of
the forward and backward halves of the circuit overlapping. The circuit
would therefore be required to have a deadband region in which none of
the tubes conduct. This, however, is very undesirable in a servomecha¬
nism, in which operation about the null is the rule.
It should be noted that all the foregoing analyses of thyratron circuits
have considered only average behavior of the circuits. In control sys¬
tems designed for a sufficiently wide frequency band to require significant
amplification at frequencies near line frequency, the pulsed nature of the
thyratron current must be considered. The thyratron amplifier is then
best analyzed on the sampled-data basis.1
3.13. Other Thyratron Applications. Although the control circuits
discussed in detail up to this point are of fundamental importance,
1 Truxal, “Automatic Feedback Control System Synthesis,” McGraw-Hill Book .
Company, Inc., New‘York, 1955, chap. 9.
Sec. 3.13] POWER AMPLIFIERS 149
there are a few other circuits using thyratrons which deserve brief
mention.
The split-field series motor discussed in Sec. 4.19 is readily adapted to
thyratron drive. A simple half-wave control circuit for this service is
shown in Fig. 3.32. When thyratron 1 conducts for a longer period than

Fig. 3.32. Thyratron control circuit for split-field series motor.

thyratron 2, the motor turns in one direction, whereas if thyratron 2 con¬

ducts for the longer period, the motor reverses. At the expense of some¬
what greater complexity of the control circuits, the circuit can be con¬
verted to full-wave operation.
Thyratrons may be used to control the operation of a-c motors of both
the induction type and the com¬
Forward Backward
mutator type.1 In Fig. 3.33 is
shown a two-phase induction ser¬
vomotor driven from a single¬
phase supply. When the forward
thyratrons are allowed to conduct,
phase 1 is driven directly and phase
2 is driven through the phase-split¬
ting capacitor C. Allowing the
reverse thyratrons to conduct re¬
verses the phase excitation and thus
the direction of rotation. As the Fig. 3.33. Thyratron control circuit for
firing angle of the thyratron pair is two-phase induction servomotor.
retarded, the rms voltage applied
to the motor is smoothly reduced to essentially zero.
An example of control of a commutator type of motor by means of
thyratrons is given in Fig. 3.34. An a-c repulsion-type motor is shown,
in which one or the other set of brushes is short-circuited by a conducting
thyratron, depending on the direction of rotation desired.
1 J. H. Burnett, Driving Servomotors with Grid-controlled Thyratrons, Elec. Mfg.,
vol. 50, p. 138, October, 1952.
 150 CONTROL SYSTEM COMPONENTS [CHAP. 3

In applications of thyratrons a number of practical points must be kept,

in mind. Some of these, such as the maximum current and voltage rat¬
ings of the various electrodes, are assumed to be familiar to the reader.1
Other points of this sort, such as electrical noise generated by thyratrons,
etc., are discussed in a number of books on industrial electronics and
practical control.2 An interesting cause of oscillation in thyratron cir¬
cuits is discussed by Greenwood, Holdam, and MacRae.3
Since grid current is drawn by a conducting tube, the grid bias of the
nonconducting tube is changed by the resulting voltage drop in the com¬
mon grid circuit. As the conducting tube cuts off, the operating points
shift, and this may result in a spuri¬
ous error signal and continuous
oscillation of the output device.
This might occur in the circuit of
Fig. 3.29b, for instance, if the pre¬
ceding amplifier has a significant
output impedance. The remedy is
to isolate the control circuits for the
two tubes by decreasing the output
impedance of the preceding stage or
Fig. 3.34. An example of thyratron con¬
trol of commutator-type motor.
by providing separate tubes to drive
the thyratron grids.
3.14. Relay Amplifiers.4 The electromagnetic relay is one of the sim¬
plest and least expensive power amplifiers available. Its chief disadvan¬
tage is that it represents a highly nonlinear circuit element. By use of
negative feedback, it is, however, possible to obtain an almost perfectly
linear power amplifier still retaining many of the advantages of simple
relays.
The basic configuration is given in Fig. 3.35a. The operation may be
described by assuming that a step of voltage is applied to the input at
time zero. Before this time the output of the relay was zero. The input
step begins to charge the capacitor, and at t\ in Fig. 3.36 the voltage er has
reached a large enough value to close the relay. At fa, then, em suddenly
becomes Eb; that is, the contact of the relay closes, and the output is con¬
nected to the positive supply. Now the capacitor must charge to e,- — Eb,

1 However, see Dow, “Fundamentals of Engineering Electronics,” John Wiley &

Sons, Inc., New York, 1945, and Reich, “Theory and Applications of Electron Tubes,”
McGraw-Hill Book Company, Inc., New York, 1944, and other texts cited earlier.
2 Ahrendt, “Servomechanism Practice,” McGraw-Hill Book Company, Inc., New
York, 1954, p. 114.
3 Greenwood, Holdam, and MacRae, “Electronic Instruments,” Radiation Labora¬
tory Series, vol. 21, McGraw-Hill Book Company, Inc., New York, 1948, p. 411.
4 J. Gibson and F. Tuteur, The Response of Relay Amplifiers with Feedback, Trans.
AIEE, vol. 76, part. II, pp. 303-307, 1957.
Sec. 3.14] POWER AMPLIFIERS 151

and a downward exponential begins. At U the capacitor voltage has

become so small that the relay cannot hold in, and the contact falls open.
The capacitor once more charges to a value large enough to close the
relay, and the process continues. The relay chatters, and the output is a
chain of pulses. This is certainly not a faithful reproduction of the input,

Low-pass
fitter ' \
er Output
VWV-1- Retoy \ \ er
Input
lut V-/
R C.1 -m

I o
o h-oOut

(o) Schematic (I) Relay characteristic

Fig. 3.35. A relay amplifier with feedback.

but if the load acts as a low-pass filter and is essentially sensitive only to
the average value of the pulse chain, it is shown below that the amplifier is
quite satisfactory and in fact “linear” over most of the operating range.
If the input-signal amplitude is large, the pulses are long and the average
value of the output is high. The limit on input-signal amplitude is
reached at the point at which the relay remains closed for the entire
period.

Fig. 3.36. Waveshapes for relay amplifier with step-function input.

The relay used in the amplifier must be a polarized relay, i.e., one that
will close to one contact for a positive voltage applied to the coil and to
the other contact for a negative voltage applied. The same effect may,
however, be obtained by using two unpolarized relays which are in con¬
junction with diodes. In any physical relay there is a deadband, or
152 CONTROL SYSTEM COMPONENTS

range of coil voltage that is too small to activate the relay. All practical
relays also display hysteresis; i.e., a larger value of coil voltage is required
to close the relay than that required as a minimum to hold in the
relay.
The major advantages of relay amplifiers are simplicity, ruggedness,
and economy. The major disadvantage is the fact that the load must
supply a good deal of smoothing if the operation is to be acceptable.
However, it is shown below that, for a typical motor with an integration
and a single time constant, a good replica of the input to the relay ampli¬
fier is possible at the shaft output.
3.15. Relay-amplifier Stability. Before discussing the input-output
relations of a relay amplifier with feedback, we consider the stability of
such a device. The stability of feed¬
Cm back systems containing relays is perhaps
h <-
most easily investigated by use of the
describing-function method of Kochen-
>f j burger.1 In this method the highly non¬
linear input-output relation of the relay
ir is replaced by a quasi-linear describing
A >
>f i\ ■ function. This describing function is
c. the relation between the magnitude of
i
r
i

-> h <-
a sinusoidal input signal and the magni¬
Fig. 3.37. Relay characteristic. tude and phase angle of the fundamental
component of the output. Kochen-
burger has shown that a relay with the input-output characteristics shown
in Fig. 3.37 has the describing function

4 sin
h2 /—a (3.61)
7T Ir

where Emi is the complex magnitude of the fundamental component of

the output voltage, Ir is the input amplitude, and

A — h — . A T h
COS —l _ COS" 1_
2L 2Ir
(3.62)
A — h _ j A + h\
cos —i
P — /4
~2i^ + cos -2ir)
In Eqs. (3.62) A and h are the deadband and hysteresis, respectively, of
the relay, as indicated in Fig. 3.37.
The remainder of the feedback loop making up the system is assumed
to be linear and has the transfer function Hlm The system is stable if

1 R. J. Kochenburger, A Frequency Response Method for Analyzing and Synthesiz¬

ing Contactor Servomechanisms, Trans. AIEE, vol. 69, p. 270, 1950.
Sec. 3.15] POWER AMPLIFIERS 153
the roots of the characteristic equation 1 + HiH2 = 0 lie in the left half
plane for all possible values of H2. Stability can be determined by a
modification of the usual Nyquist method. In its more commonly used
form, this modification takes the form of a complex plot of — H2 and 1/Hi,
as shown in Fig. 3.38 for the case of a relay feedback system in which Hi

Fig. 3.38. Typical inverse Nyquist diagrams for a feedback system containing a relay.

has at least two lags. Note that H2 is a function of Ir, while Hi is a func¬
tion of frequency. Intersection of the two curves indicates instability.
For the relay amplifier with feedback shown in Fig. 3.39, Hi is given by

kRi k'
Hi = (3.63)
RiR2Cs 1 + Ts
(R1 + R2) 1 + Ri T R2

where k is the amplifier gain. Therefore the Nyquist plot will have the
form shown in Fig. 3.40. Since the curve of l/Hi does not intersect that

Fig. 3.39. Typical relay-amplifier circuit. Fig. 3.40. Inverse Nyquist diagram for
relay amplifier of Fig. 3.39.

for — H2, the implication is that the system is stable no matter how large
the gain k is. However, at high gains, previously neglected lags, e.g., the
time required for the relay to close after the coil is energized, may cause
the frequency locus to bend around as shown in Fig. 3.38, thus possibly
causing instability. Similarly, if the filter break frequency is raised to a
154 CONTROL SYSTEM COMPONENTS [Chap. 3
value such that other lags in the loop, such as the mechanical and/or elec¬
trical lags of the relay itself, become important, the frequency locus bends
to the left around the origin, as in the case of increased gain. If it
intersects the amplitude locus, instability is indicated.
3.16. Relay-amplifier Static Characteristic. The output of a relay
amplifier with feedback is a series of pulses, positive or negative, of ampli¬
tude Eb and with length and repetition rate determined by the input.
Usually the relay amplifier is em¬
'm ployed in an application where only
Eh —- the average value of this train of
" A pulses is important.
~e2~e\ In order to determine the static
*1 *2
relation between the input voltage
V A
ein and the average output voltage
-Ef
Em, we shall consider the circuit in
Fig. 3.39, with the static character¬
Fig. 3.41. Pull-in and drop-out voltages
istic between e and em as shown in
defined on the relay characteristic.
Fig. 3.41. Let us suppose that, at
zero time, to, e = 0. Thus the relay is open, and em = 0. Let a step of
voltage, Exn in magnitude, be applied, as shown in Fig. 3.36. Then

e = Ein( 1 - e-</r0 (3.64)

where 7\ = RiC. When e reaches e2, the relay closes. Substituting e2

for e in Eq. (3.64), we may solve for ti, the time of closing:

Exa - e2 = (3.65)
or t\ = —Ti In (Ein — e2) (3.66)

After the relay closes,

R2 Ri
e [1 Eh [1 e-(t-tl)/T21
Ri + R2 Ri T R2
+ e2e-(t-t')/T> (3.67)

where T2 = [RiR2/(Ri + R2)]C. When e reaches ei, the relay opens.

The time of opening, t2, may be found by setting e = ei and solving for
t = t2:
Ri Ri
&i ~ Ein + Eb
e-«2-n)/r2 — Ri H- R2 Ri + R2
(3.68)
Ri Ri
e2 - Ei
Ri H- R2
+ Eb
Ri T R2
R2 . 7-, R\
e\ — Eh +
R i T R2
Eb
Thus t2 — t\ — — T2 In R\ + R 2
(3.69)
Ri Ri
e2 — Eu + Eb
Ri T R2 Ri T R2
Sec. 3.16] POWER AMPLIFIERS 155

With the relay open,

e = Ein[l - e-«-t*'>/T'\ + elC-«-‘»>/rx (3.70)

It will be noted that Eq. (3.70) is the same as Eq. (3.64) plus the term
due to the initial voltage on the capacitor. All the segments of the volt¬
age curve will be exactly the same as those described by Eqs. (3.67) and
(3.70). Equation (3.64) is the special case when no charge is on the
capacitor. The relay closes again at £3, when e = e2; this time is given by

h-U = -T1 In --(3.71)

During the period of time U — U the output voltage is — Eb, and during
the period t2 — h the output is zero. These two lengths of time represent
the cycle of the output-voltage wave. The average output is therefore

— Eb(t 2 — t\)
Em = (3.72)
h - tl

ei - Eiu „■ + Eb Rl
Ri T R2 R i + R2
Eb In
777
e2 ~ Ain
R2 I
r>—;—T-,—h Eb
77TR 1

R1 T R2 R1 T R2
or Em =
p _ rp -^2 I jp
Ti ln e2 - Ein + ln 1 iD^i + R2 ^ b Ri + R2
T2 e\ — Ex R2 , T7, R1
62 — Ex + Ei
R1 T i?2 R1 T R2,
(3.73)

By a relatively minor change in the circuit, R2 can be grounded when¬

ever the relay is open. This will make Ti = T2 and will contribute to
the linearity of the input-output relation. It is also usual to make
Ri = R2. In this case Eq. (3.73) reduces to

1 + 2ei/(Eb — E in)
1 + 2e2/ (Eb — Ein)
(3.74)
1 - Eb/(Ehl -2d)
1 - Eb/(Ein - 2e2)

In certain applications an electronic difference amplifier is used rather

than the resistor-adding network. The low-pass filter may be incorpo¬
rated as a part of the interstage coupling network of the electronic ampli¬
fier. When the difference amplifier rather than the resistor-adding net¬
work is employed, the factors representing the voltage-divider action do
 156 CONTROL SYSTEM COMPONENTS [Chap. 3
not appear, and Eq. (3.73) can be simplified to yield

jyT i 1 + e\/ (Eh — Ein)

6 1 + e2/(Eb - Em)
(3.75)
1 - &/(g,. - eQ
1 - Eb/(Ein - e2)

By examination of this relation we can see that it gives the required nega¬
tive real result only for the region
e2 < Ein < Eb + gi. For Ein < e2
the relay does not close; the output
is zero. For Ein > Eb + ei, the relay
remains closed and the output is Eb.
It may be shown that, if e\ and e2
are small with respect to the battery
voltage, the slope of Em versus Ein is
essentially constant over the operat¬
ing range. The exact relation be¬
tween input-voltage magnitude and
the average of the output voltage can
Fig. 3.42. Static characteristic of the
be determined by the use of Eq.
relay amplifier, normalized to the relay¬
(3.75). In Fig. 3.42 is shown Em
closing voltage Eb. The contact volt¬
age was ten times the closing voltage in versus Ein normalized to Eb.
this normalization. 3.17. The Frequency Response of
a Relay Amplifier. The limitation
on the frequency response for a sinusoidal input of a relay amplifier is
determined to a first approximation by the time constant T of the
filter or the chatter rate of the output pulses. Actually, however, this
chatter rate or output-pulse repetition rate depends on the input-signal
magnitude, and in the limit, where Ein is either greater than Eb or zero,
the relay remains always closed or always open. Since the relay amplifier
is a nonlinear device, the entire concept of frequency response is con¬
trived, and a measure for one type of input signal will not hold for
another. For a step-function input, the relay amplifier is particularly
fast and essentially establishes the required average value of em in one
opening and closing.
The response of a relay amplifier to sinusoidal signals depends on the
amplitude and the frequency of the input signal. For signals whose peak
amplitude at the input of the relay is less than the value required to close
the relay, there is no output. For instance, if the closing voltage of the
relay is 10 volts, a signal whose peak amplitude is less than 10 volts will
not close the relay. Since the low-pass filter attenuates the amplitude
of the input signal, the peak amplitude must be computed at the input
to the relay. At-the break frequency of the filter, for instance, the filter
 Sec. 3.17] POWER AMPLIFIERS 157

reduces the input amplitude by a factor of 0.707; thus the minimum input
signal amplitude must be 14.14 volts peak. Essentially a relay amplifier
will operate properly for normal-amplitude signals up to the break fre¬
quency of the low-pass filter. In Fig. 3.43 are shown the amplitude
reduction and phase shift for a typical relay amplifier whose relay-closing
voltage is 10 per cent of the battery voltage applied to the relay terminals.
The frequency response shown in Fig. 3.43 was obtained by smoothing
the pulse-train output in order to recover the input-signal frequency com¬
ponent. When the input-signal frequency is equal to or greater than the

Fig. 3.43. Frequency response of relay amplifier normalized to relay-closing voltage.

break frequency of the filter, especially at low amplitudes, the number of

output pulses per cycle becomes small, and even the integrated output
becomes rather steplike. In Fig. 3.44 the various waveshapes for a
particular signal are shown. Fig. 3.44a shows the sine-wave input to the
amplifier; its normalized peak amplitude is 5.0; its frequency is the same as
the break frequency of the amplifier low-pass filter. In Fig. 3.446 is
shown the waveform at the output of the low-pass filter, which illustrates
the deadband and hysteresis of the relay. In Fig. 3.44c is given the out¬
put of the relay, and Fig. 3.44d shows the integrated output. In a typical
power-amplifier application the relay output drives a d-c motor, which
would represent an integration plus a low-pass filter. Thus it could be
158 CONTROL SYSTEM COMPONENTS [Chap. 3

expected that the waveshape in d would be smoothed by the filter before

appearing at the output shaft.
It is difficult to establish an equivalent loop gain constant for a relay
amplifier with feedback without making several rather arbitrary assump¬
tions. However, from Fig. 3.43 it can be seen that such an equivalent
gain depends on the input-signal amplitude. At maximum-amplitude
signal the break frequency of the closed-loop response of the amplifier
occurs at about 4/T, and for minimum-amplitude signals the break
approaches 1/T.

(c) Output of relay

(d) Integrated output of relay

Fig. 3.44. Relay-amplifier waveshapes.

The amount of phase shift contributed by the filter to the over-all

phase shift shown in Fig. 3.43 is inversely related to the gain of the ampli¬
fier, as a result of the negative feedback. In addition to decreasing the
closed-loop phase shift, an increase in amplifier gain will act to reduce
the effective deadband, thus increasing the usefulness of the amplifier at
low signal amplitudes. As the gain is increased, the loop becomes less
stable from the increased effect of previously neglected lags in the ampli¬
fier and relay itself, until finally the amplifier goes into oscillation, as
shown in the preceding section. That is, its output for no input is a con¬
tinuous string of. alternate positive and negative pulses. As a signal is
Sec. 3.18] POWER AMPLIFIERS 159
placed on the input terminals, the length of the positive pulses grows
longer and that of the negative pulses shorter, or vice versa; thus the aver¬
aged output follows the input. This type of operation is more sensitive
to low-amplitude signals than is stable operation and will have minimum
phase shift from input to output. However the continuous vibration of
the power element under quiescent conditions may be undesirable.
3.18. Magnetic Amplifiers. Introduction. Magnetic amplifiers have
become more and more widely used in recent years for both high- and low-
power applications. The simplest form of magnetic amplifier requires a
d-c input signal and delivers an a-c output. This type may be used when
the load is one that requires alternating current, such as the a-c servo¬
motor (see Chap. 7). In conjunction with rectifiers, magnetic amplifiers
also can deliver a d-c output signal and may then be used to drive the field
of a d-c generator, the solenoid of a hydraulic valve, etc. In power¬
handling capacity, magnetic amplifiers range from very small sizes capable
of delivering a few milliwatts to very large sizes designed for outputs of
many kilowatts.
The primary advantages of magnetic amplifiers over electron-tube
amplifiers are ruggedness and the absence of filaments that require
warm-up time and constitute an important source of power loss. Mag¬
netic-amplifier circuits can be designed for a wide range of input and out¬
put impedances, and since the control and load circuits are not conduc-
tively coupled, they provide somewhat more flexibility than vacuum-tube
circuits. On the other hand, magnetic amplifiers have a limited fre¬
quency response and produce an output that is a modulated a-c carrier.
They have a much lower input impedance than vacuum tubes and provide
only a finite power gain, whereas vacuum tubes may, in theory at least,
give an infinite power gain. A practical disadvantage is that magnetic
amplifiers must in general be designed specially for each application, since
both the supply voltage and the load impedance critically affect the opera¬
tion. Hence magnetic amplifiers are usually considerably more expensive
than vacuum tubes, particularly when they are used in applications where
mass production economies are not applicable.
A simple magnetic-amplifier circuit is shown in Fig. 3.45a. This circuit
is the series-connected type and consists of two iron cores wound with two
windings each. The iron cores are made of special alloys such as Delta-
max or Supermalloy which saturate sharply and with a relatively low
level of magnetizing force H. A typical hysteresis loop of a standard core
material is shown in Fig. 3.455. One of the windings on each of the cores
is referred to as the load or gate winding. In the series amplifier the two
gate windings are connected in series with the load to the a-c source. The
other two windings, the control windings, are also connected in series to
the control source; however they are connected in opposition, so that the
100 CONTROL SYSTEM COMPONENTS [Chap. 3

Load
m-
Gate B
^ A-c supply
winding

Control Q
voltage \~)

Gate
Control winding
winding

(a) U>)
Fig. 3.45. Series-connected magnetic amplifier, and hysteresis loop of core material.

fundamental component of a-c voltage induced in the control windings

is canceled and does not affect the control source.
A number of other core and coil arrangements are also used in practice,
depending on the application.
Thus the two gate windings may
be connected in parallel, as shown
in Fig. 3.53. Quite often a three-
legged core is used, as shown in Fig.
3.46, and frequently the cores are
wound with more than one control
winding. The basic operating prin¬
ciple is, however, essentially the
Fig. 3.46. Magnetic amplifier using three-
legged core. same for all of these types, and it
will therefore be illustrated by using
the series-connected circuit as an example.
3.19. Operation of the Series-connected Magnetic Amplifier. In
order to simplify the explanation of the operation of a series-connected
circuit as much as possible, the following assumptions are usually made:
1. The core has either zero or infinite inductance,
depending on the state of the flux in the core. B
2. The resistance of the control circuit including the
control source is very small, although not always H
negligible.
3. The gate windings have zero resistance.
4. The load is resistive.
Assumption 1 amounts to a simplification of the Fig. 3.47. Ideal¬
hysteresis loop of the core material to three straight ized hysteresis
loop.
lines at right angles to each other, as shown in Fig.
3.47, and corresponds fairly closely to the hysteresis curves found
in cores made of high-quality iron. Assumption 4 is made in this neces¬
sarily elementary treatment because the operation of magnetic amplifiers
Sec. 3.19] POWER AMPLIFIERS 161
with reactive loads is fairly complicated and beyond the scope of this
text.1 Since the first three assumptions are obviously never exactly met
in practice, it may be expected that the following analysis will not be
exact; however it is found that the results obtained correspond reasonably
well to those obtained in actual operation.

Fig. 3.48. Schematic diagram of a magnetic amplifier.

We consider first the operation of a magnetic amplifier with zero con¬

trol current. The alternating supply voltage applied to the two gate
windings sets up an alternating flux in the cores, which in turn generates a
counter voltage of self-induction, as given by Faraday’s law:
d(f>
a (3.76)
dt
where eg is the total voltage appearing across both gate windings and Ng
is the number of turns in each gate winding. By assumption 1 no current
is required in the winding to establish
the flux. Hence there is no current
in the load and no voltage drop across
the load, so that the reactor voltage
eg is at all times exactly equal and
opposite to the applied voltage es
(see Fig. 3.48 for the notation used).
A reactor is said to be normally excited
when the maximum flux set up in Fig. 3.49. Flux and voltage in a nor¬
the core by the applied a-c supply mally excited reactor with zero control
voltage is exactly equal to the satura¬ current.
tion value <f>s. A normally excited reactor never saturates if the control
current is zero, and the load current remains zero during the entire a-c
voltage cycle. The flux and voltage cycles obtained under these condi¬
tions are shown in Fig. 3.49.
When control current is passed through the control windings, an addi-
1 See, however, Storm, ‘‘Magnetic Amplifiers,” John Wiley & Sons, Inc., New York,
1955.
162 CONTROL SYSTEM COMPONENTS [Chap. 3

tional magnetomotive force is added to each core and causes the cores to
become saturated for some part of the a-c cycle. Suppose that, at a cer¬
tain point in the cycle, reactor 1 of Fig. 3.48 saturates. When this hap¬
pens, the flux can no longer change, or

t =0 (3-77)

Hence the reactor can no longer generate any counter emf and, by
assumptions 1 and 3, now represents a short circuit, not only on its gate
winding, but also on its control winding. As a result the control winding
of reactor 2 is now connected only across the resistance of the control cir¬
cuit, which, by assumption 2, is negligibly small. The entire control
circuit seen by reactor 2 therefore represents a short circuit, and even
though this reactor is not saturated, the voltage across both of its wind¬
ings must also be zero. The flux in the core must therefore be constant.
Hence all the supply voltage appears across the load. During the next
half cycle the process repeats, with the roles of the two reactors reversed.
In Fig. 3.50 are shown the resulting waveforms of voltage, flux, and cur¬
rent. In drawing these curves, we assume that the circuit is operating
in the steady state and that the reactors are normally excited. At time
t — 0, reactor 2 is just coming out of saturation, and the flux in reactor 1
has the value (/>a. This value is chosen to meet the requirement that the
circuit is in steady-state conditions, i.e., that 4>i = <t>a for cot = 2ir,
47r, ....
As time increases from zero, the flux in both reactors increases, with
d(f>/dt exactly equal to the value required to make the reactor counter
voltage equal and opposite to the applied voltage. This state of affairs
continues until cot — a, at which point reactor 1 saturates. As indicated
in the preceding paragraph, the flux in both reactors now remains con¬
stant, and the voltage drop across the reactors is zero. The supply volt¬
age therefore appears across the load, and load current flows. As soon as
the supply voltage passes through zero, d<t>/dt can again be supplied by
both reactors. The reactors therefore are again able to generate counter
emf, and the load voltage and current go to zero, until at cot = t + a, reac¬
tor 2 saturates again and the cycle repeats.
The waveshape of the load-current pulses is the same as that of the
load-voltage pulses, since the load is assumed to be resistive. During the
first pulse, i.e., for a < cot < tt, reactor 1 is saturated, and its load and
control circuits are therefore completely decoupled. The fact that the
load current flowing in its gate winding produces mmf has no effect on its
control winding. In reactor 2, however, the flux is constant at a value
that is less than the saturation value. According to the ideal form of the
hysteresis loop assumed for the core material (Fig. 3.47), there can there-
Sec. 3.19] POWER AMPLIFIERS 163
fore be no net magnetomotive force in this reactor. Hence the mmf of
the gate winding is opposed by an equal and opposite mmf produced by
the control winding, or

Ndc = NgiL (3.78)

During the next half cycle the load current flows in the opposite direction,
but since reactor 1 is connected in the opposite direction to the control

circuit, the control-current pulse is in the same direction as the previous

pulse. Hence the control current consists of unidirectional pulses, as
shown in Fig. 3.50c.
The control current has an average component 7C. Since the wave¬
shapes of the load- and control-current pulses are the same,

NCIC = NgIL (3.79)

where II is an average over one half cycle. In Fig. 3.48 is shown one
164 CONTROL SYSTEM COMPONENTS [Chap. 3

method of connecting a d-c ammeter and a bridge rectifier to read / /,.

Since Ic is a d-c current, which cannot be induced in the control circuit
by transformer action, it must actually be the current supplied by the
control voltage Ec\ i.e.,

Ic = Is (3.80)
■Kc

Thus we find that the current gain of the series magnetic reactor is given
by
II = Nc
Kt (3.81)
Ic Ng

where Ic and II are average quantities. It is interesting to note that

neither the supply voltage nor the various resistances of the circuit enter
into this result, and the implication is, therefore, that a magnetic amplifier
acts as a controllable constant-current device. This is found to be closely
true in actual practice for load currents less than the value requiring con¬
tinuous conduction of the reactor,
i.e., for II < ES/RL. This is shown
in Fig. 3.51, where the solid line
shows the theoretical relation and
the dotted curve shows a typical
measured response. Note that the
chief difference is the rounding off
at the top and the small current
Fig. 3.51. Load current versus control
current. existing for zero control current.
These differences are due to the fact
that in actual practice the reactor requires some magnetizing current to
establish the flux in the core.
The voltage gain of the series-connected amplifier may be derived from
Eq. (3.81):
El _ RlIl _ NcRl
(3.82)
Wc ~ RTc ~ Njfe

Here again, all quantities are average values. The d-c power gain is

ElIl Nc2Rl
(3.83)
ECIC Ng2RC

Since EL and II are average quantities, the power output given by this
expression is that obtained after rectifying and filtering the current out¬
put and then passing it through the load. If the power content of the
a-c current pulses’coming from the magnetic amplifier is of interest (as it
Sec. 3.20] power amplifiers 165

might be in a heating or lighting application), the power gain would be

K- - W.F' 13 84)

where F is the form factor of the current wave shown in Fig. 3.50d.
In the preceding discussion, use has been made of the approximation
that Rc, the control-circuit resistance, is negligibly small. Since in prac¬
tical circuits there is always some resistance in the control circuit, the
question arises of how large Rc may become, while still being considered
negligibly small. As has already been demonstrated, only one reactor
saturates at any one time, and the other one is unsaturated. The unsatu¬
rated reactor acts exactly like a transformer. The small control-circuit
resistance is therefore reflected to the gate winding by means of the
standard transformer relation:

R'c = ^4rc (3.85)

As long as this reflected resistance is very much less than the load resist¬
ance Rl, the voltage drop across the reactors will be negligibly small com¬
pared to the load voltage during periods of conduction. A ratio of Rl to
R'c of 10 or more is usually considered adequate to ensure that the assump¬
tion of negligible Rc does not result in excessive errors. If R'c is greater
than Rl, the reactors operate in a different mode, which is not considered
here.1
3.20. Time Constant of the Series-connected Amplifier. The rela¬
tions derived in the previous section were obtained on the basis of steady-
state operation. When the control voltage is varied, however, the
dynamic performance of the amplifier must also be considered. This
performance is affected chiefly by the time constant of the control wind¬
ing. There is a slight residual time lag in the load winding which is not
considered here.
In order to find the time constant of the control winding, we must
determine its effective inductance Lc. We may write

Ee = Rcic + Ne - Nc

d
= Rcic + Nc (</>i — ^2) (3.86)

This may be rewritten as

d(<£i — <t>2) dic
Fc — Rcic d- N,• (3.87)
dir dt
1 Ibid., and Geyger, “Magnetic Amplifier Circuits,” McGraw-Hill Book Company,
Inc., New York, 1954.
166 CONTROL SYSTEM COMPONENTS [Chap. 3
We note by reference to Fig. 3.506 that, for constant control current,
0i — 02 = 0d is constant. Hence we may define

d<f>d
(3.88)
die

the effective inductance of the control circuit. The problem now is to

determine 0d as a function of ic and the amplifier constants. We proceed
by showing first that the average load voltage is directly proportional
to 0<f.
From the previous discussion,

El = Es - Eg (3.89)

where El, Es, and Eg are the load,

supply, and gate voltages, all aver¬
aged over a half cycle of the supply
frequency. The instantaneous gate
voltage is given by

Fig. 3.52. Flux in series-connected \ = N> ft ^

reactors.
where 0 is the instantaneous flux in
one of the cores. Suppose the flux in core 2 is as shown in Fig. 3.52. If
the core is normally excited, then by inspection of Fig. 3.52 we have
02 = — 0s COS Cot 0 < cot < a
02 — — 0S COS a a < cot < 7T (3.90)

Hence 3g = 2Ng ^ ( — 0s COS cot)

= 2NgCo<j)s sin cot 0 < cot < a

eg = 0 a < cot < 7r (3.91)
The average of eg is found by integration over a half cycle:
a
Eg = 2Ngco(f)s sin cot d(oot)
7T JO
CO
2Ng<J>8 — (1 — cos a) = 4i\ro0s/(l — cos a) (3.92)
IT

where co = 2x/. The average supply voltage Es is equal to Eg for the case
where the reactor does not fire, i.e., where a = ir. Hence
Es = 8Ng(t>sf (3.93)
From Eqs. (3.89), (3.92), and (3.93) we have now
El = SNg(f)sf — 4:Ng<f)sf(l — cos a)
= 4W„/0S(1 + cos a) (3.94)
But 0i( 1 + COS a) = 0S — ( — 0S cos a) = <j>d (3.95)
Sec. 3.21] power amplifiers 167
from Fig. 3.52, so that, finally, the desired result is

El — 4 Ngf<t>d (3.96)

All that is required now to find Lc is to relate (f)d to ic. By use of Eqs.
(3.85) and (3.86), we have

El _ IlRl _ NcRl
(3.97)
Ic Ic Ng
Hence, from Eqs. (3.86) and (3.96)

_ Nc4>d _ NcEl _ Nc2Rl

(3.98)
c~ ~77 " WjTc “ W7f
Thus the time constant is given by

Lc _ Nc2Rl
Rc MN^Rc
Note that the time constant is inversely proportional to the supply fre¬
quency. This indicates that, in order to obtain high response speeds,
magnetic amplifiers should be operated from high-frequency supplies.
A commonly used figure of merit applied to power-amplifying equip¬
ment is the product of power gain and bandwidth. This figure of merit
reduces to a very simple result for magnetic amplifiers if the bandwidth is
defined as the reciprocal of the time constant. Using the relation for d-c
power gain developed in the previous section [Eq. (3.83)], we find that

K„ ~ = 4/ (3.100)

Here again the advantage of high-frequency operation is apparent. It is

also clear that, for a given supply frequency, an increase in gain must be
paid for by a decrease in bandwidth. This is a result found quite gener¬
ally in amplifying equipment.
3.21. Reactors Connected in Parallel. The parallel connection is
used to supply relatively low voltage, high-current loads, and its operation
differs in some respects from that of the series-connected amplifier dis¬
cussed above. The circuit is shown in Fig. 3.53. In analyzing this cir¬
cuit, the same assumptions that were made in connection with the series
amplifier are employed, except that the assumption of negligible control-
circuit resistance is not required. The cores are assumed to be normally
excited.
In the parallel connection only the saturated reactor conducts load
current. This is true because this reactor represents a short circuit if the
gate-winding resistance is assumed to be zero, while the other reactor
represents the finite resistance R'c = (Ng2/Nc2)Rc• Hence the load-cur-
J 68 CONTROL SYSTEM COMPONENTS [Chap. 3

rent pulses do not induce pulses in the control circuit, as m the series-
connected amplifier, and the control current consists of a steady d-c cur¬
rent. However, by assumption 1 of Sec. 3.19, the net mmf on any core
must be zero during periods when it
is not saturated. Hence during
these periods there must be a current
in the gate winding such that
Ngig = Ncic (3.101)
When both reactors are unsaturated,
this current flows in opposite direc¬
tions through the two gate windings.
The load current is therefore zero
during this time, and only ig circu¬
lates between the two gate windings.
The waveshape of the two gate¬
winding currents is shown in Fig.
3.54c. The shaded portion repre¬
sents the load current. The average
or d-c value of gate-winding currents
must be zero, since there are no rec¬
tifiers in the circuit. Hence the
average value of a single load-cur¬
rent pulse must be equal to the value
of gate-winding current during the
period when the reactor is not satu¬
rated. For each reactor, therefore,
Fig. 3.54. Waveshapes pertaining to N I
parallel-connected magnetic amplifier. II = (3.102)

where II is the average value of the load current over half a cycle.
Since each reacfor furnishes only half the load-current pulses, the total
Sec. 3.21] POWER AMPLIFIERS 169

load current is given by

2NCIC
II = (3.103)
N,
Thus the current gain is given by

j, _ II _ 2Ne
(3.104)
Kl L Nn
The voltage gain is
El _ IlRl _ 2NcRl
Ke = (3.105)
Ec IcRc
The average or d-c power gain is
-tt' _ 4Nc2Rl (3.106)
p ~ ~n7rc
and the rms or a-c power gain is
tv- 4lNc2Rl p<>
_

(3.107)
^ ~ Ng2RC ^
where F is the form factor of the load-current pulses.
The time constant of the parallel-connected amplifier can be obtained
by methods similar to those used in connection with the series amplifier.

Fig. 3.55. Circulating current in parallel amplifier.

An important difference is that in the parallel circuit the two gate wind¬
ings permit the flow of a circulating current which increases the total time
constant. The path of the circulating current is shown in Fig. 3.55a and
may be seen to be equivalent to a short circuit coupled to the control
winding. This equivalence is made somewhat clearer in Fig. 3.555, in
which the circuit of Fig. 3.55a has been redrawn with the lower reactors
turned upside down and the connections to the load removed.
By means of straightforward circuit analysis1 it can be shown that, if
the coefficient of coupling between the two circuits is unity, the effective
time constant is given by

r=b ■tvc
+ wQ (-3-108)

1 See Sec. 4.3 and Eq. (4.18), where this same problem is discussed in a different
connection.
170 CONTROL SYSTEM COMPONENTS [Chap. 3

where Lc and Lg are the effective inductances of the control and gate wind¬
ings, respectively, while Rc and Ra are the corresponding resistances.
The inductance of the control circuit may again be defined as in Eq.
(3.88)
d<f>d
Lc = N (3.88)
c dic

since <f>d is a constant for constant ic as before. By reasoning that is

entirely analogous to that used for the series amplifier, we can show that

El 2N (if (t>d
El
or 4>d (3.109)
2NJ
Hence Eq. (3.88) becomes

= „ d_(EL/2Nsf) _ Nt dEi NcRc dEL

(3.110)
dic 2Naf d(Ec/Rc) 2Ngf dEc

The voltage gain is given by Eq. (3.105) and may be substituted. Hence

NcRc 2NcRl N*Rl

(3.111)
Lc 2NJ NgRc N/f
In a like manner Lg may be defined as

r — af dffrd_ \j- d(EL/2Ngf) _ NgRc dEL

(3.112)
0 di' 9 d[(Nc/Ng)ic\ 2fNc dEc
and, again employing the voltage gain,

T — ^L
(3.113)

Thus from Eq. (3.108)

rp _ Nc~Rl i Rl
(3.114)
~ jNjRc ^ jRg

If the windings are designed according to the usual transformer practice,

the resistances of the windings are proportional to the square of the num¬
ber of turns. If it is then also assumed that the resistance of the control
circuit is due only to the resistance of the control winding, we may write

R N°* (3.115)
c Nc2
and the time constant becomes

2Nc2Rl
T = (3.116)
NgtRJ
In order to compare the performance of the parallel- and series-connected
Sec. 3.22] POWER AMPLIFIERS 171
amplifiers, it is again instructive to determine the ratio of d-c power gain
to time constant. From Eqs. (3.106) and (3.116) this ratio is found to be

Kv 4:Nc2RL/Ng2Rc
= 2/ (3.117)
T 2N c2Rl/N g2Rcf
Thus it appears that the parallel-connected amplifier has a somewhat
poorer performance than the series circuit. It is clear from the preceding
analysis that this is due entirely to the short-circuit path existing in the
gate-winding circuit.
3.22. Magnetic Amplifiers with Feedback. The performance of mag¬
netic amplifiers can be improved greatly by the use of positive feedback.
The feedback can be applied in a
number of ways. By way of illus¬
tration we consider the series-con¬
nected amplifier with feedback ap¬
plied through a special feedback
winding. In a typical circuit such
as the one shown in Fig. 3.56, the
alternating load current is rectified
and applied to feedback windings
N/ on each of the cores. The feed¬
back windings are connected in op¬
position just as the control windings
are. and the effect of the current Fig. 3.56. Magnetic amplifier with
feedback.
flowing in these windings is the
same as that of the current flowing in the control windings. The equal
ampere-turn relationship shown to exist in the simple reactor must hold
here as well, and we have, therefore,

IlNa — ICNC ~b IgN/ (3.118)

Nc
or II = I, (3.119)
Na- N,

If we define a feedback ratio B as

B ± Nj (3.120)
Nf
Nc 1
then IL = It (3.121)
Na 1 - B
We may also write that
NcRl 1 Ec Nt
El = IlRl = /< Ri (3.122)
Na 1 - B Re Ng 1 - B

The current gain is available from Eq. (3.121) and the voltage gain from
172 CONTROL SYSTEM COMPONENTS [Chap. 3

Eq. (3.122). The average d-c power gain is therefore

K ElIl _RlNc 1 Nc 1 _ RlN2 1

p Ec Ic RcNg 1 - B N0 1 - B RcNg2 (1 - B)2 {6'LZ6)
and for the rms power gain we multiply by the square of the form factor,
as before.
The time constant of the circuit depends on whether the load current is
increasing or decreasing. Inspection of the circuit shows that the rectifier
bridge forces the current in the feedback windings to be equal to the load
current only when the load current is constant or increasing. When the
load current decreases, a larger current may persist in the feedback wind¬
ings and flow through the rectifier. Thus, for decreasing current the
feedback winding constitutes a short-circuited winding which has the
same effect on the time constant as the short-circuited gate windings in
the parallel amplifier. For rising transients the time constant is therefore
Lc/Rc, while for falling transients it is Lc/Rc + Lf/Rf.
The inductance Lc may be found as before:

Le ~ Nc w (3-88)

The relation between <f>d and EL [Eq. (3.96)] is unchanged by the feedback
winding; hence, by the same reasoning as that employed in Sec. 3.20,

N2Rl
(3.124)
4iV/(l - B)

The time constant for rising transients is therefore

Lc _ Nc2Rl
(3.125)
Wc ~ ±fNg2Rc( 1 - B)
Since inductance varies as the square of the number of turns, we have

Nl Lc
(3.126)
N2
so that for falling transients the time constant is given by

Lc Nf2 Lc _ N2Rl ( 1 Nf2\

(3.127)
Rc ^ Nc2 Rf 4/iV7(l - B) \RC ^ N2Rf)

The performance of the circuit may again be judged by the figure of merit
defined above. From Eqs. (3.123) and (3.125) we have that the ratio of
power gain to time constant for rising transients is given by

Kv 4/
(3.128)
T (1 - B)
Sec. 3.23] POWER AMPLIFIERS 173

Thus it is clear that, although the gain and time constant are increased by
positive feedback, there is a net advantage in the use of feedback.
The expressions obtained indicate that the gains and time constant
become infinitely large as B —> 1. This is not found to be true in practice,
primarily because the accuracy of the assumptions made in the derivations
begins to diminish. In particular, the assumption of the square hysteresis
loop (assumption 1 of Sec. 3.19) becomes less and less tenable, and the
actual shape of the hysteresis loop must be considered as the feedback
factor approaches unity.
It should also be noted that a tacit assumption made in the above
analysis was that the rectifiers are perfect, i.e., that they do not permit
any reverse current to flow. In practice, some reverse current always
does flow and has the effect of slightly reducing the feedback factor B.
3.23. The Self-saturated Amplifier. The most commonly used method
of providing feedback in magnetic amplifiers is the method of self-satura¬
tion. A number of circuits employing this principle are shown in Fig.
ftr

/wv*h
§ Load
o—A/W—
o
°

[W-

io) Doubler circuit, [d) Single-phase bridge, (^) Full-wave circuit,

a-c output d-c output d-c output
Fig. 3.57. Typical self-saturated magnetic-amplifier circuits.

3.57. In all of these circuits, rectifiers in series with the gate windings
produce a d-c component in the gate-winding current which is equal to the
average (over one half cycle) of the load current. It can be shown1 that,
if the rectifiers are ideal, i.e., if they do not permit any reverse current
flow, the self-saturated amplifier is equivalent to a feedback amplifier in
which the feedback factor B is equal to unity. Hence, as indicated above,
the simplified analysis of the previous paragraphs no longer yields reason¬
able results, and a different approach must be used. We present here
only an empirical point of view, which takes the experimental control
characteristic of the amplifier as its starting point and deduces all further
results from it.
A typical self-saturated-amplifier control characteristic is shown in
Fig. 3.58a. Note that the load current is almost maximum when the
control current is zero and that a “bias’’ value of negative control current
is therefore required to produce minimum output. This is due to the fact
that the rectifiers in the circuit cause the amplifier to saturate itself when
1 W. J. Dornhoefer and V. H. Krummenacher, Applying Magnetic Amplifiers,
Elec. Mfg., March, 1951, vol. 47, part I, pp. 94 ff., April, 1951, part II, pp. 112 ff.
174 CONTROL SYSTEM COMPONENTS [Chap. 3

there is no control current and that the action of the control current is
actually to “desaturate” the cores. It can be shown1 that the sloping
portion of the characteristic from A to B in Fig. 3.58a is identical to the
back, or demagnetizing, portion of the hysteresis loop of the core material.
In order to obtain approximate equations for the gain and time constant
of a self-saturated magnetic amplifier, it is convenient to approximate the
empirically determined curve of Fig. 3.58a by three straight lines, as

(o) Actual characteristic (/?) Straight-line approximation

Fig. 3.58. Control characteristic of self-saturated amplifier.

shown in Fig. 3.58b. If this approximation is valid, we obtain by direct

proportion
Es \ NC(IC - Ich)
II = (3.129)
Rl Lmiy Nc(Ict - I a) + L
so that the current gain is given by

K- - w. - mm NJ- (3130)
In these equations Icb is the “bias” value of control current which results
in the minimum output II min, Es is the control current resulting in full out¬
put, and NI = Nc(ICs — Eb) is the number of control ampere-turns required
to vary the output from maximum to minimum. Note that NI is a con¬
stant that depends only on the back slope of the hysteresis loop of the core
material and becomes smaller as the quality of the core material is
improved. The voltage gain is given by

OEl _ Rl dll _ ESNC

(3.131)
dEc Rc dlc RC(NI)
Thus for a given amplifier operating at a fixed supply voltage Es, the volt¬
age gain is a constant and is independent of the load, whereas the current
gain is inversely proportional to the load resistance. Self-saturation
therefore is seen to change the magnetic amplifier from an adjustable con-
1 Ibid.
Sec. 3.24] POWER AMPLIFIERS 175
stant-current generator [cf. Eq. (3.81)] to an adjustable constant-voltage
generator.
From Eqs. (3.130) and (3.131) the power gain for direct current is given
by
_E,%*
(3.132)
RlRcIni)2

The time constant of the self-saturated amplifier can be obtained by

the type of argument employed in the previous analysis, since the relation¬
ship between <j>d and El derived in Eq. (3.109) is not affected by the pres¬
ence of the rectifiers in the gate circuit. Thus, from Eq. (3.110)

j -kj d*Pd KeRcNc

(3.110a)
° c dR ~ ~2NT
we obtain by substitution of Eq. (3.131) for Ke

Lc = E'NC2
(3.133)
Rc 2 Rc(NI)N0f

This represents the over-all time constant for rising transients when the
rectifiers are blocked. For falling transients, where circulating paths may
exist through the rectifiers (depending on the exact circuit used), the time
constant will be increased as discussed previously [cf. Eq. (3.108)] and
may be written in general:

(Nc> N/\
(3.134)
2NgfRc(NI) \RC ^ Rj
where Rg is the resistance of the circulating path in the gate winding.
3.24. The Ramey Magnetic Amplifier. A magnetic amplifier operating
on a somewhat different principle
from the ones described above has Dy D2 R,
"L
«- H—ww
been described by Ramey.1 A cir¬
cuit diagram for a single-ended am¬
plifier of this type is shown in Fig. © o<
N,c o! No
0,
3.59. Note that rectifiers are em¬ L0- Control 3
Gate
ployed both in the control and load winding winding
winding, and that on the control side Fig. 3.59. The Ramey magnetic ampli¬
an a-c bias voltage is inserted in ad¬ fier-basic circuit.
dition to the control signal ec.
The operation of the amplifier may be explained most easily by assum¬
ing that ec is a d-c voltage. The direction of this voltage is such as to
apply reverse bias to the rectifier D i. The a-c bias voltage is in phase
1 R. A. Ramey, On the Mechanics of Magnetic Amplifier Operation, Trans. AIEE,
vol. 70, part II, pp. 1214-1223, 1951.
176 CONTROL SYSTEM COMPONENTS [Chap. 3

with the supply voltage es. Thus, if eb = Eb sin cot, es = Es sin cot, where
Eb and Es are maximum values. Suppose now that at time t = 0 the core
is saturated so that the flux 0 = 0S. During the first half cycle the bias
voltage eb is in a direction opposing ec, and the direction of es is such that
the rectifier Z>2 is biased in the reverse direction. No current flows in
either winding until the instantaneous amplitude of eb exceeds ec. At this
time the voltage difference eb — ec is applied to the control winding, and
the directions of the voltages are such that the core is demagnetized.
The flux therefore decreases from the saturated value 0S until the bias
voltage eb is again less than ec. At the end of the first half cycle the flux is
given by the expression
1 r ait = ir — sin-1

(P = (f)s — b (Eb sin cot — ec) dt = <f>s-- (3.135)

A cj ut= sin-1 -=p COW c

where A x is the area between the

bias and control voltages as shown
in Fig. 3.60. Ideally this amount of
flux remains in the core until the
end of the first half cycle. During
the second half cycle, rectifier Z)2
permits current flow in the gate
circuits, but the control circuit is
blocked by rectifier D\. If we as¬
sume that the voltage drop in the
load due to the magnetizing cur¬
rent is negligible, the full supply
voltage is applied the gate winding.
The winding direction is such that
the flux again increases. The flux
Fig. 3.60. Voltage and flux variations in continues to increase until the core
the Ramey amplifier. saturates again, and at this time
the supply voltage appears across
the load. The flux during the second half cycle is given by
’cot
Ax
0 0s Em sin cot dt (3.136)
0?AC Nc ait = tt

When the core again saturates 0 = 0., and

'ait = a
Ax A2
Em sin cot dt (3.137)
Nt ait = ir U)Ng

where a is the firing angle and where A2 is the area under the supply volt¬
age as shown in Fig. 3.60. The output voltage consists of pulses, and its
average value is proportional to the area shown in Fig. 3.60 as A3. It
Sec. 3.24] POWER AMPLIFIERS 177

should be clear that although a d-c control signal has been assumed for
this analysis, an a-c signal at supply frequency will work just as well.
It will be noted that the amplifier operates on a shared-time principle.
During the first half cycle the control voltage charges the core, and during
the second half cycle the supply voltage discharges it. For this reason
the response time of this amplifier is less than one cycle, and it is therefore
much faster than conventional amplifiers.
A disadvantage of the amplifier is that the control signal must be able
to furnish sufficient power to magnetize the core completely in one half of
a cycle of the supply voltage. Thus the input power required tends to be
larger than for the self-saturated amplifier described in Sec. 3.23, in which
charging of the core takes several cycles. With modern high-grade cores
the required magnetizing current is, however, quite small, and typical

Reactor

Fig. 3.61. Bridge-type Ramey circuit for full-wave output.

amplifiers of this type may have an input impedance greater that 10,000
ohms.
The basic circuit shown in Fig. 3.59 can produce only a half-wave,
unidirectional load current. A somewhat more complex circuit using two
reactor cores is shown in Fig. 3.61. As shown, this circuit produces a
full-wave, unidirectional current in the load, but by reversing the rectifiers
in the lower section the circuit is converted to furnish a half-wave, reversi¬
ble current to the load.
It has been the purpose of these sections to introduce the reader to the
analytic assumptions usually made in the treatment of magnetic ampli¬
fiers and to derive by their use the gains and time constants of several sim¬
ple representative types. This will enable the reader to assess the oper¬
ation of these devices in a control system. For design procedures and
information on problems of cascading, inductive loads, and the like, the
reader is referred to the texts by Storm1 and Geyger2 and to the current
periodical literature.
1 Storm, “Magnetic Amplifiers,” John Wiley & Sons, Inc., New York, 1955.
2 Geyger, “Magnetic Amplifier Circuits,” McGraw-Hill Book Company, Inc., New
York, 1954.
178 CONTROL SYSTEM COMPONENTS [Chap. 3

PROBLEMS
3.1. A half-wave thyratron amplifier is used to control the current in a pure-resist-
ance load of 25 ohms. The supply voltage is 220 volts, and the tube drop is negligible,
(a) Obtain the “small-signal” gain dl/d(j>a for <f>a = 90°. (b) The firing angle of the
circuit is controlled by use of combined a-c and d-c bias with the a-c bias shifted 90°
from the a-c plate supply. If the a-c bias is 55 volts rms and if the d-c signal voltage
is Ein, find the over-all “small-signal” gain dl/dElu for an operating point defined by
<f>a = 90°. Assume that the critical firing potential is very small compared to 50 volts.
3.2. A half-wave thyratron amplifier of the form shown in Fig. 3.15 is used to con¬
trol the current in a generator field. The field has a resistance of 50 ohms and an
inductance of 0.5 henry, and the thyratron amplifier is connected to a 220-volt,
60-cps single-phase supply. It is desired to vary the field current from 0.5 to 1 amp.
(a) Find the range of firing angles required. (6) Find the approximate gain A//A<£a
at the mid-point of the required current range, i.e., for a current of 0.75 amp. Assume
that the tube drop is negligible.
. .
3 3 The firing angle for the thyratron of Prob. 3.2 is controlled by use of combined
a-c and d-c bias, with the a-c bias shifted 90° from the a-c plate supply. If the a-c
bias is 110 volts rms, find the over-all gain of the circuit, dI/dEs, where Es is the d-c
signal voltage, for the case I = 0.75 amp. Assume very small critical grid potential.
3.4. Find the average load current in the half-wave thyratron circuit of Prob. 3.2
if the load resistance is 100 ohms, if the inductance is 0.1 henry, and if the firing angle
is 45°. Assume that Eq. (3.29) applies. The tube drop is 10 volts.
3.5. The load of the full-wave thyratron circuit of Fig. 3.20 consists of a 10-ohm
resistance and an inductance of 0.1 henry. Plate voltage is 110 volts rms, 60 cps,
and tube drop is 15 volts, (a) Find the firing angle 4>a for which the load current is
just continuous. (6) Find the load current for firing angles <£a = 30° and <£a = 150°.
3.6. In connection with the thyratron-controlled motor, show that the change-over
from mode 1 to mode 2 operation and from mode 2 to mode 3 operation is not
abrupt.
3.7. A half-wave thyratron circuit used to drive a d-c motor has a supply voltage
of 220 volts, 60 cps. The thyratron tube drop is negligible. One of two motors is
used with this circuit. One of them has an armature resistance of 0.6 ohm, an induct¬
ance of 0.001, and a Kt of 1 ft-lb/amp, while the other has a resistance of 0.7 ohm,
negligible inductance, and a Kt of 0.9 ft-lb/amp. In all other respects the motors
are identical. Find the ratio of maximum stall torques available from the two motors.
3.8. A relay amplifier with feedback of the sort shown schematically in Fig. 3.35
uses relays which close when the voltage across their coils exceeds 40 volts; they open
when the voltage drops to 25 volts. The amplifier used between the RC filter and
the relay coils has a gain of 20, and the time constant of the filter is 0.1 sec. For a
10-volt peak input signal, find the maximum frequency to which the amplifier can
respond.
3.9. A series-connected magnetic amplifier like that shown in Fig. 3.45a has 5,000
turns on its control winding and 2,000 turns on its gate winding. The control resist¬
ance is 10 ohms, and the frequency of the supply is 60 cps. A load resistance of 100
ohms is connected to the amplifier, and the control voltage is adjusted to produce an
average load current of 0.5 amp. At time t = 0, a 400-ohm resistance is shunted
across the 100-ohm load, (a) Find the control current for t < 0. (b) Find the load
and control currents at the instant after the 400-ohm resistance is added, (c) Sketch
the behavior of the average load current as a function of time after t = 0. Neglect
the pulsed nature of the output, (d) Draw an equivalent circuit that, on the average,
behaves like this magnetic-amplifier circuit.
 CHAPTER 4

D-C MACHINES

4.1. Introduction. D-c machines are used in the power-output stage

of a large class of electromechanical servomechanisms. D-c machines are
preferred over a-c machines in high-power applications because of the
ease of control of the speed and direction of rotation of large d-c motors.
A widely used arrangement of d-c machines in servomechanisms is the
Ward-Leonard system, in which a d-c generator drives a d-c motor. A
common form of this system is shown in Fig. 4.1. The generator is driven
at an essentially constant speed by a prime mover, and the field of the
motor is separately excited by a constant-voltage source. (In low-power
servos, the motor field may be established by a permanent magnet.) In
the figure, the generator field is shown with a center tap and is driven by a

Fig. 4.1. A common form of Ward-Leonard system.

push-pull amplifier. The currents flowing in the two halves of the field
winding are therefore in opposition, and their effects tend to cancel; hence
the mmf produced in the air gap of the generator is proportional to the
difference of the two currents. Although this method of exciting the
generator field is by no means the only one used, it has the advantage of
permitting reversal of the mmf, and therefore reversal of the d-c motor,
without the need of actually reversing the field currents. It is therefore
almost universally used when the generator field is driven by an electronic
control amplifier.
The generator used in the, Ward-Leonard system may be any one of
several types. In low-power applications a simple d-c generator may
be used, but for power levels above a few hundred watts the power ampli-
179
180 CONTROL SYSTEM COMPONENTS [Chap. 4

fication afforded by the ordinary generator is no longer sufficient to raise

the low-power level of the electronic amplifier up to the level required by
the motor. In that case one of the more elaborate, two-stage generators,
such as the Rototrol, Regulex, or Amplidyne, is used. These generators
will be described in detail in succeeding paragraphs.
4.2. Generated Voltage in a Simple D-C Generator. The basic rela¬
tion for the open-circuit voltage produced by a d-c generator is derived
from Faraday’s law and may be written in the form1

NgP(t> 12
E0 = (4.1)
V
where Eg = generated voltage, volts
Na = total number of armature conductors
P = number of field poles
0 = total magnetic flux per pole, webers
12 = speed of rotation, radians/sec
p = number of parallel paths in armature winding
For a particular machine, Na, P, and p are constants, and therefore Eq.
(4.1) may be written in the simpler form

Eg = /ci0£2 (4.2)

The equation indicates that, if 0 is held constant, the generated voltage

is directly proportional to the speed 12. This principle is used in the
d-c tachometer (also called rate generator). Such a machine is shown in
Fig. 4.2 and is seen to be a small generator, usually built with a stable,
permanent magnet to provide the field excitation. The commutator of
a d-c-tachometer armature usually has a relatively large number of seg¬
ments so as to provide as smooth an output as possible. Nevertheless
the output contains a ripple component, commonly of the order of 5 to
10 per cent of the d-c voltage; this sometimes causes difficulty in very
high performance loops. Equation (4.2) indicates that the d-c output
voltage is accurately proportional to the speed, with no time lag; however,
at very high speeds, brush bounce has the effect of reducing the accuracy
and at the same time increasing the ripple and noise output. If calibra¬
tion accuracy is an important requirement, care must be taken never to
short-circuit or load the tachometer heavily, since the resulting mmf of
armature reaction2 may demagnetize the field magnets to some extent and
thus change the calibration.

1 See, for instance, Bull, ‘‘DC Machinery,” John Wiley & Sons, Inc., New York,
1947, p. 46.
2 Ibid,., chap. 11. Also Dawes, “A Course in Electrical Engineering,” vol. 1,
“Direct Currents,” 4th ed., McGraw-Hill Book Company, Inc., New York, 1952,
pp. 415ff.
Sec. 4.2] D-C MACHINES 181
When a generator is used for power amplification, as in the Ward-
Leonard system, it is driven at constant speed. Hence SI in Eq. (4.2) is a
constant that may be combined with ki, so that only cf> and Eg are variable,
0 being the independent variable. When the generator is open-circuited,
</> depends only on the field current.
The relationship between the flux and the field current is ordinarily
expressed graphically in the form of a hysteresis loop or saturation

Fig. 4.2. Stator, armature, and brush rigging of a typical d-c tachometer.

curve of the magnetic path in which the flux is established. In most

designs for generators, a steel of low retentivity is used. The hysteresis
loop is thus narrow enough for the hysteresis effect to be ignored, at least
in a first approximation, and an “average’’ saturation curve is used to
describe the characteristics of the magnetic path (see Fig. 4.3). A further
simplification is possible in many cases by virtue of the fact that in normal
operation the voltage of the generator will be near zero,1 where the
1 This statement is true for position servos, where the motor normally does not turn
or turns only slowly. In speed-control systems, where the motor normally runs at a
relatively high speed, a linear approximation to the magnetization curve can also be
made; however the slope of the straight-line approximation then depends on the
operating point.
182 CONTROL SYSTEM COMPONENTS [Chap. 4

magnetization curve is approximately a straight line. In this case we

may say that the flux is linearly proportional to the ampere-turns, or
N f if
(4.3)
~(R~

where Nf is the number of turns of the field and (R is the reluctance of the
magnetic path expressed in suitable units. A combination of Eqs. (4.2)

Fig. 4.3. Hysteresis loop and “average saturation curve” of a generator.

and (4.3) results then in a linear relation between field current and gener¬
ated voltage:
Eg = kgif (4.4)

The field circuits of almost all d-c generators are highly inductive, since
a large number of turns are used to obtain the required magnetomotive
force with a minimum of field current.
Hence, the inductance must be con¬
sidered, as in the circuit shown in Fig.
4.4. We have

E, = R/if + L, (4.5)
Fig. 4.4. The equivalent circuit of
a d-c generator.
if the armature current is zero or con¬
stant. By use of Laplace transforms we can solve for the field current

Sf WRf)ti, (4.6)
Rf -T LfS 1 -j- TfS
where Tf = Lf/Rf is the field time constant. In this equation the resist¬
ance Rf is the total resistance of the field circuit together with the output
resistance of the driving source considered as a Thevenin generator;1 Ef is
the Thevenin voltage of the source.
1 Everitt and Anner, “Communication Engineering,” 3d ed., McGraw-Hill Book
Company, Inc., 1956, p. 119.
Sec. 4.2] D-C MACHINES 183
When a center-tapped field is used with a push-pull circuit, as shown in
Fig. 4.1, an equation very similar to Eq. (4.6) can he derived. If the
push-pull amplifier stage consists of two vacuum tubes with an amplifica¬
tion factor fj, and a plate resistance rp and if the voltages applied to the
grids of these tubes are eg and — eg, respectively, then an equivalent circuit
for the push-pull arrangement takes
rP
the form of Fig. 4.5. Here the
arrows indicate the assumed posi¬
tive direction of the voltages and
currents. With the current as¬
sumed as shown, the effective mmf
is Nf(ifi — if 2), if both halves of the
field winding have the same number
of turns, N/. R'f represents the
Fig. 4.5. Equivalent circuit of a d-c gen¬
resistance of each field winding erator with push-pull field.
alone; M is the mutual inductance
between the two halves; Ebb and Zb are the voltage and output impedance,
respectively, of the amplifier power supply.
The mmf produced by the two fields may be found as a function of the
control voltage eg by use of Kirchhoff’s laws. In writing the mesh equa¬
tions for the circuit, it is convenient to let

Z = rp + Rrf + LfS + Zb . -v
and ZM = Zh-Ms 1 ]

Then the two mesh equations take the form

~ V&g + Ebb = Zlfi + Zjuif2

(4.8)
/j.eg + Ebb = Zntifi -f- Zif2
— 2 fxeg — 2 neg
whence V1 — V2 (4.9)
Z — Zm r p + R'f + (Lf + M)s
where in the second step we substitute for Z and Zm from Eq. (4.7). The
mmf is therefore
2Nffieg — 2N f/ieg
8= Nffoi ~ t/i) = rP + R'f + (Lf T M)s (rp + Rf)( 1 + Tfs)
(4.10)

where Tf = (Lf + M)/(rp + R'f). If Lf + M and rp + R'f are thought

of as the equivalent inductance and resistance, respectively, of the push-
pull field winding and if we let — 2gcff be equal to the driving voltage Ef of
the circuit of Fig. 4.4, this result is seen to be identical with the one
obtained in Eq. (4.6). Hence Eqs. (4.2), (4.3), and (4.6) may be com¬
bined to obtain the expression for the open-circuit voltage:

k\£lk(f)Nf/Rf a (kg/Rf)Ef
(4.11)
Tfs+l Ef = Tfs + 1
184 CONTROL SYSTEM COMPONENTS [Chap. 4

In later discussions we shall make use of the equivalence of push-pull and

single-ended field coils and ordinarily carry through the various analyses
on the basis of a single-ended field, even though the push-pull type is more
common.
4.3. The Effects of Hysteresis and Eddy Currents. In the previous
discussion the properties of the iron making up the magnetic paths of the
generator were not considered. However, in practice, the imperfections
in the iron will affect the generator performance. The fact that the iron
saturates results in a fundamental nonlinearity between generated voltage
and applied voltage. Once the generated voltage reaches the saturation

Fig. 4.6. Phase lag produced by hysteresis. Fig. 4.7. Gain change produced by
hysteresis.

value, the steady-state gain of the machine (defined as dEg/dEf) decreases

rapidly and approaches zero for very large values of Ef.
Hysteresis of the iron causes a phase lag between input and output of
the generator that is a function of the amplitude of the input but is inde¬
pendent of frequency. This effect is best demonstrated by assuming that
the input voltage is sinusoidal and plotting the resulting output voltage
through one complete cycle as in Fig. 4.6. This construction also indi¬
cates that the output wave contains harmonic components not present in
the input. The extra phase lag of the fundamental component, which is
not predicted by the linear approximation, may result in poorer stability
of the system of which the generator is a part. Furthermore, comparison
of the hysteresis loops obtained on a given piece of iron for various ampli¬
tudes of excitation indicates that the average slope for small amplitudes,
and therefore the gain, is less than that for larger amplitudes (Fig. 4.7).
Thus the combined effect of hysteresis and saturation on the gain is to
reduce the gain, for both very small and very large amplitudes, compared
Sec. 4.3] D-C MACHINES 185

to its value at intermediate amplitudes. A complete mathematical

description of all of these effects results in nonlinear differential equations
of considerable complexity, which cannot, at the present time, be solved
completely. Hence many of these effects are often neglected in practical
analyses, for the sake of simplicity.
An additional inperfection of the iron is the generation of eddy currents
in the field poles whenever the field current changes. This effect may be
considered to be equivalent to having a large number of closed circuits
in the iron mutually coupled to the field circuit and to each other. Since
these circuits must be chosen somewhat arbitrarily, a simplifying assump¬
tion is made that yields an approximate answer sufficiently accurate for
many purposes. The multiplicity of mutually coupled circuits is replaced
by an equivalent single eddy-current circuit that is mutually coupled to
the field winding. This situation
is diagramed in Fig. 4.8, where Le
and Re represent the effective in¬
ductance and resistance, respec¬
tively, of the eddy-current path,
and where M is the mutual induct¬
ance between the eddy current and
field circuits. If the number of
Fig. 4.8. Equivalent circuit accounting
turns of the field winding is Nf and
for eddy currents.
if the number of effective turns for
the eddy-current circuit is Ne, then the following three simultaneous
equations (written in Laplace transform notation) describe this situation:

Ef — (Rf + Lfs)if — Msie (4.12)

0 = —J\lsif + (Re + Les)te (4.13)
Eg = k(Nfif - Nete) (4.14)

The signs of M in Eqs. (4.12) and (4.13) must be the same but can be
either positive or negative, depending on the sense assumed for the wind¬
ing of the coil representing the eddy-current circuit. The sign of Ne in
Eq. (4.14) must, however, correspond to the signs chosen for M in such a
way that the effect of the eddy current ie is to oppose the field current if.
This is required by Lenz’s law. Simultaneous solution of the three equa¬
tions then vields
7 T i Af
77 A fLe NJI
kEf l 1lf + RjRe RfRe ■) (4.15)
LfLe - M*
5+1
RfRp.
This equation can be simplified if we assume that the magnetic paths for
both the eddy-current circuit and the field winding are identical. With
186 CONTROL SYSTEM COMPONENTS [CHAP. 4

this assumption we have

Le = ^L, (4.10)

This assumption also implies perfect coupling of the two circuits, so that

M = VlZ} = L, |r (4.17)

Substitution of these relations in Eq. (4.15) gives

p = m,/Rf)Sf (kNf/MJk U ic\

0 [(.Lf/Rf) + (Le/Re)]s +1 (J7/+T'e)s+ 1

where Te is the “eddy-current-circuit” time constant. We note that, at

least to the accuracy of the assumptions made in this development, the
effect of eddy currents is to increase the time constant of the generator.
This additional time lag may become particularly serious when, in order
to obtain rapid response, Rf is made large. A pentode amplifier driving
the generator field has this effect. Equation (4.18) shows that, even if
Tf can be made negligibly small, the time constant Te will still cause a lag
in the transfer function relating Ef to Eg. It is clear that the suppression
of eddy currents in the field of a generator is of great importance where
maximum response speed is desired; and for this reason extra-thin punch-
ings are used in the laminations of the field poles of generators designed
for service in servo control systems.
4.4. Factors Affecting Terminal Voltage of a Loaded Generator. Up
to this point we have been concerned primarily with the open-circuit volt¬
age produced by a generator and its relation to the voltage applied to the
field. The implicit justification for this preoccupation with open-circuit
voltage has been Thevenin’s theorem.1 This theorem, which strictly
applies only to linear systems, may be extended to apply even to some
nonlinear systems by supposing the output impedance to be in part a
function of the load current. The knowledge of the transfer function
from generator input to open-circuit voltage is not a complete description
of the generator under load; the output impedance is also required.
When a generator delivers current to a load, the output voltage drops
because of the effects of the resistance and inductance of the armature
winding, armature reaction, and mutual inductance between armature
and field circuits. The armature resistance usually includes the brush
resistance and is therefore not completely independent of the armature
current. Nevertheless, a linear relation between armature resistance
drop and armature current is almost always assumed, because in most

1 Everitt, Joe. cit.

Sec. 4.4] D-C MACHINES 187
generators the voltage drop due to the nonlinear brush contact is quite
small compared to the total voltage.
Armature reaction, defined as the mmf set up by the armature current,
may decrease the air-gap flux and thereby reduce the output voltage.
The decrease of air-gap flux may be due either to a demagnetizing compo¬
nent of the armature reaction or to a distortion of the air-gap flux pattern.1
Since the mmf of armature reaction is directly proportional to the arma¬
ture current, it may be assumed that the reduction in flux, and therefore
the reduction in output voltage, is proportional to the current. This
means that the armature-reaction effect is similar to the resistance effect,
and the two may therefore be lumped together as an “effective” resist¬
ance for purposes of analysis. This “effective” resistance cannot, of
course, be measured by simple voltmeter-ammeter tests but must be
determined by some form of load or short-circuit test.
Since armature reaction also causes sparking at the brushes,2 consider¬
able effort is often expended to reduce it. This is particularly true in
servo generators, which are often called upon to deliver heavy accelerating
or braking currents, and which would spark badly if armature reaction
were permitted to go unchecked. Both commutating poles and pole-face
windings3 are therefore used extensively in these generators to improve
commutation. When such compensation of armature reaction is used to
reduce sparking, it will also diminish the other effects of armature reac¬
tion. Hence armature reaction is often neglected altogether in the
analysis of servo generators. However, in certain types of generators,
particularly in Amplidynes, armature reaction is essential to the operation.
The self-inductance of the armature circuit includes the inductance of
the armature winding together with any series windings used on commu¬
tating poles and pole-face compensating windings. If compensating
windings are used, the mutual inductance between the compensating
winding and the armature winding subtracts from the total inductance.
Thus, if La is the inductance of the armature, Lc the inductance of the
compensating winding, and M the mutual inductance between them, then
the effective inductance4 becomes La + Lc — 2M. Hence, a compensat¬
ing winding, in addition to reducing armature reaction, also reduces the
inductance. This is desirable, since inductance causes a time lag between
the open-circuit voltage and the load current.
Mutual inductance between armature and field circuits is important
only in compound generators using a series field, since otherwise the flux

1 Dawes, loc. cit.

2 Ibid., pp. 426-428.
3 Ibid.
4 This equation is derived in a number of elementary texts; see, for instance, Dawes,
ibid., p. 319.
188 CONTROL SYSTEM COMPONENTS [Chap. 4

paths of the armature and field circuits are essentially at right angles to
each other and the coupling is very small. Compound generators are not
ordinarily used in servomechanisms, but we mention mutual inductance
here in anticipation of the discussion of the more elaborate generators
such as Rototrols and Amplidynes, where the effect is not always
negligible.
In order to assess the general importance of mutual inductance in d-c
generators, we compute the relative magnitudes of the voltages induced
in the armature by speed action and
those induced by mutual induct¬
ance. For this purpose, consider
an armature winding rotating in an
alternating field, as shown in Fig.
4.9. The armature voltage gener¬
ated in each turn of the armature
winding is given by
Fig. 4.9. Simplified schematic of arma¬
ture winding. (4.19)

where </> is the flux in webers linking the particular turn in question. The
flux threading a particular turn is a periodic function of the space position
of that turn. Thus the turn made up of the conductors A and B is linked
by the maximum amount of flux for the armature position shown in the
figure. This position is defined as 6 = 0. As the armature rotates, the
flux decreases to zero as 6 becomes 90°, reverses sign (relative to the turn
under discussion), reaches a maximum for 6 = 180°, goes to zero again,
and again reaches a maximum when 6 = 360°. With the usual shape of
the field poles the function of flux versus 6 is a flat-topped wave, but for
purposes of analysis it is convenient to assume a sinusoidal wave. If the
flux at any particular position of the armature also varies as a sinusoidal
function of time, we may write

4> = 4>m cos 6 cos tot = k+If max cos tct cos 6 (4.20)

where to and If max are the frequency and amplitude, respectively, of the
field current and k$ is a constant, linearity of the magnetic path being
assumed here. Hence, the differentiation indicated in (4.19) gives

dd
= —k^If max( to cos d sin ut + cos ut sin d (4.21)

The effective value of eg may then be written in the standard complex

notation
'Eg = — Qkflf sin 6 + juk^If cos 6 (4.22)
Sec. 4.4] D-C MACHINES 189
where we have substituted the speed of rotation, 12, for dO/dt in accordance
with the nomenclature used in Eq. (4.1). Note that the first term on the
right is proportional to speed (0) and independent of frequency. This
term represents the voltage induced by speed action and may be desig¬
nated as Eq. It is a maximum in those armature turns for which 6 = 90°.
The second term on the right is independent of speed but proportional to
frequency and is maximum in those turns for which 6 = 0°. This is the
voltage due to the effect of mutual inductance and is designated as Em.
Inspection of Eq. (4.22) reveals that

Em max — -qE a max (4.23)

i.e., the ratio of mutually induced voltage to voltage generated by speed

action is equal to the ratio of the frequency of the field current to the
speed of rotation of the armature, both expressed in radians per second.
(In a generator with more than two poles, 12 is the speed in electrical
radians per second, i.e., the actual speed multiplied by the number of
pairs of poles.) In most servo systems the ratio co/12 is very much less
than unity, and mutual-inductance effects are therefore usually negligible
compared to speed effects.
The complete behavior of a generator under load may now be expressed
in the form of the two simultaneous equations given below. These equa¬
tions include all the factors discussed in this section, even though the
reader may have formed the impression (correct in many cases) that the
only factor not normally neglected is the armature resistance. We do not
include the effect of eddy currents and hysteresis in the iron, however,
since inclusion of these effects only complicates the final expressions with¬
out adding much in the way of new information.

Ef = (Rf + Lfs)if + Mafsia (4.24)

Eg — (Ra + Las)la — MfaSlf = V0 (4.25)

In these equations Ef is the voltage applied to the field; Lf and Rf are

the inductance and total resistance, respectively, of the field circuit;
Maf = Mfa is the mutual inductance between armature and field circuits;
Eg is the open-circuit voltage produced by the generator; Ra is the “effec¬
tive” armature resistance, including armature reaction and brush drop;
La is the armature inductance; and V0 is the terminal voltage existing
under load conditions. Eg and if are related by Eq. (4.4):

Eg = kgtf (4.4)

Since we are interested in the output impedance of the generator, it is

190 CONTROL SYSTEM COMPONENTS [Chap. 4
\

convenient to rewrite Eqs. (4.24) and (4.25) in such a form that Ef and ia
are the independent variables and V0 and if the dependent variables:

Ef — MafSla — (Rf + LfS)lf (4.26)

~ (Ra + Las)ia — ( — kg + Mfas) If + V0 (4.27)

Solving for V0, we obtain

Ef(kg - Mfas) (Rf + Lfs) (Ra + Las) — MlfS2 -+■ MafkgsA

Rf Lfs Rf + LfS la
kg/Rf ~~ (Mfg/Rf) (4.28)
- Ef - (r„ + Las + MafS kR~+M£f)
Tfs + 1

The first term in this equation represents the open-circuit voltage, and
except for the very small mutual-inductance term in the numerator, this
term is identical with the expression for the open-circuit voltage already
derived [Eq. (4.11)]. The second term shows the effect of the load cur¬
rent; hence the quantity in brackets multiplying the load current repre¬
sents the output impedance. It is seen that the output impedance con¬
sists primarily of Ra + Las, the resistance and self-inductance of the
armature. The effect of mutual inductance can be estimated approxi¬
mately if we assume that kg y> Mafs and Rfy>Lfs. This can, of course,
be true only at low frequencies; however, if these relations do hold, the
mutual-inductance term reduces approximately to Mafskg/Rf and has the
form of an additional inductance in the armature circuit, which either
subtracts from or adds to the self-inductance La (depending on the sign of
Maf relative to the sign of kg). In a noncompound generator the entire
effect is usually extremely small.
4.5. Figure of Merit. A figure of merit used very commonly for all
types of power-amplifying equipment is the product of power gain and
bandwidth. In order to compute this figure for the generator as a func¬
tion of some of the design features, it is first necessary to give precise
definitions for speed of response and power gain. Both of these quantities
will be defined for a generator connected to a pure resistance load; and
since the power gain is a function of the value of the load resistance, the
definitions are based on the use of a standard load resistance equal to the
generator output resistance. The response speed is then defined as the
frequency in radians per second of the sinusoidally varying control voltage
Ef which results in a reduction of the amplitude of the load voltage by a
factor of 1/\/2 from the very low frequency value. This is the commonly
used half-power-point definition. The power gain is defined as the ratio
of the power dissipated in the standard load resistance to the power
supplied to the field by the control voltage, both powers to be measured
at direct current or at very low frequency.
For a generator with.an output impedance that is purely resistive (i.e.,
Sec. 4.5] d-c machines 191

for which the effects of the mutual and self-inductances in the armature
circuit are negligible), the speed of response is given by

1 _ Rf
(4.29)
Tf + Te L/(l + TV TV)

where Rf and Lf are the resistance and inductance of the field circuit,
respectively, and Te is the eddy-current time constant (see Sec. 4.3).
The power input is
El
(4.30)
Rf
and the power output is
Et
Po = Ri (4.31)
.Ra T Rl,

where Ra is the armature resistance, including the effects of armature

reaction and brush resistance, and Rl is the load resistance. If Rl = Ra,
this becomes
El = kg*Ef*
(4.32)
4 Ra 4 Rf2Ra

Hence the figure of merit becomes

(4.33)
4:RaLf(l + T e/ Tf)

The quantities entering into this expression are still related to some
extent. Thus, kg is proportional to the number of turns on the field-coil
winding [see Eqs. (4.1), (4.3), and (4.4)], while Lf is proportional to the
square of the number of field turns. Hence, increasing the number of
turns will not affect the figure of merit. By similar reasoning it may be
shown that the number of turns of the armature, the number of poles, the
type of armature winding, etc., all have a negligible effect on the figure of
merit. On the other hand, since both kg and Lf are inversely proportional
to the magnetic reluctance, any decrease of reluctance will result in an
over-all improvement. Such a decrease might be produced by a reduction
of the air gap or the use of a higher grade of iron. The figure of merit is
proportional to the square of the speed, since kg is proportional to speed.
This indicates that higher performance can be obtained from a given
machine by running it at higher speed. This fact is used in some aircraft
generators.
The particular figure of merit computed here gives an indication of
which of two generators is “ better,” if the only criteria are power gain and
bandwidth. However there are other criteria, and therefore other figures
of merit, that may be more pertinent in a given situation, and hence no
single figure of this sort is ever adequate to give a complete evaluation of
192 CONTROL SYSTEM COMPONENTS [Chap. 4

the worth of a machine in a given situation. In some cases the product of

voltage gain and bandwidth is more useful, and in most cases, factors such
as cost, overload capacity, size and weight, etc., are very important and
must be carefully considered in an over-all design.
4.6. The Rototrol Generator. A single generator of the form described
in the preceding paragraphs is capable of power gains of the order of 10 to
20. When larger power gains are required, it becomes necessary to cas¬
cade two or more generators, i.e., to use one generator to excite the field of
the second, etc. The Rototrol1 is essentially such a cascade arrangement
of two generators, but since the first, or pilot, generator is partially self-
excited, the power gain can theoretically be made infinite. A schematic
diagram is shown in Fig. 4.10.
It

Pilot generator Main generotor

Fig. 4.10. Schematic of the Rototrol generator.

Both the pilot generator and the main generator are driven at constant
speed by a prime mover not shown in the figure. The pilot generator has
at least two fields, one of which is a series field L i and the other the control
field. In the figure a pilot generator with a single push-pull control field
is shown; however many pilot generators, particularly those designed for
heavy-duty industrial service, may have four or five separate control
fields in addition to the series field. This multiplicity of control fields
makes it possible to build a control system for fairly high performance
without the use of vacuum tubes. Such a system is shown in Fig. 4.13.
Since the basic operation of a Rototrol is not affected by the number of
control fields used, we shall concentrate on the single-control-field type in
the following paragraphs and show later (Sec. 4.8) what changes are
necessary in the theory when additional control fields are used.
4.7. Analysis of the Rototrol Generator. The mmf responsible for the
air-gap flux in the pilot generator is the sum of the ampere-turns of the
control field and the series field. For a given value of control-field current
and series-field current the generated voltage of the generator may be
determined from an average magnetization curve of the sort described
1 W. H. Formhals, Rototrol, a Versatile Electric Regulator, Westinghouse Engr.,
vol. 2, pp. 51-54, Mdy, 1942.
Sec. 4.7] D-C MACHINES 193

earlier in connection with the simple generator. Such a magnetization

curve is shown as the solid line of Fig. 4.11. Since the armature circuit of
the pilot generator is closed through the pilot-generator series field, tuning
resistance, and main generator field (see Fig. 4.10), the current and there¬
fore the magnetomotive force of the series field is determined in the steady
state by the relation
NiE
N i/i = ^ (4.34)

E is the generated voltage of the pilot generator, as determined by the

magnetization curve, and R is the total resistance of the pilot armature

Fig. 4.11. Points of equilibrium on Rototrol operating curves.

circuit. Equation (4.34) is a linear relationship and is plotted for three

representative values of R on Fig. 4.11 (the dotted lines). In the absence
of control-field current the series-field magnetomotive force is the only
one present; and since the generator must simultaneously satisfy the rela¬
tions imposed upon it by the magnetization curve and Eq. (4.34), it can
operate in equilibrium only at the points where these two relations
intersect.
We see in Fig. 4.11 that, for one adjustment of the armature-circuit
resistance Rc, the straight part of the magnetization curve and the resist¬
ance line coincide; hence for this resistance there are an infinite number of
intersection points, any of which represents an equilibrium point for the
generator. The value of resistance that achieves this adjustment is
referred to as the critical resistance, and the Rototrol is said to be criti-
194 CONTROL SYSTEM COMPONENTS [CHAP. 4

cally tuned when the resistance is adjusted to this value. A critically

tuned Rototrol is capable of delivering a constant output voltage, even
though the control-field current is zero. It should be noted that in prac¬
tice it is never possible to get perfect coincidence between the resistance
line and the magnetization curve, since the magnetization curve does not
possess a true straight-line portion. Hence one finds that the output
voltage always drifts slowly to an equilibrium point, even with the best
adjustment of the resistance. This imperfection is not, however, of any
great consequence in a closed feedback loop.
When the control-field current changes from zero to a finite value, the
Critical machine will presumably leave its
resistance steady-state condition. Then Eq.
(4.34) is no longer correct because
of the inductance in the armature
circuit, unless E is redefined as the
resistive component of the generated
voltage. If this is done, the total
mmf acting on the air gap of the
pilot generator may be written as

(4.35)

Fig. 4.12. Effect of control-field current

where gc is the mmf supplied by the
on critical-resistance line.
control field and Er has been sub¬
stituted for E in Eq. (4.34) to emphasize the fact that it is the resistive
component. This new situation may now be diagramed as in Fig. 4.12.
The critical-resistance line is shown shifted to the right by the amount of
the additional mmf supplied by the control field (gc). The straight-line
part of the magnetization curve is parallel to the critical-resistance line,
and there is a fixed voltage difference El between them. For any assigned
mmf the voltage read from the magnetization curve is Eg, the generated
voltage. The voltage read from the resistance line is equal to the total
resistance drop in the armature. The difference El must therefore be the
voltage drop due to the total inductance of the armature circuit. The
rate of change of the armature current is directly proportional to this
voltage and is constant as long as the resistance line and magnet¬
ization curve are parallel. El is proportional to gc in the linear region of
the magnetization curve; therefore the rate of change of main-generator
field current and the output voltage are proportional to the mmf of the
control field. In other words, the output voltage is proportional to the
integral of the control-field current; hence a critically tuned Rototrol acts
as an integrating device.
It is now a relatively simple matter to derive the transfer function of
Sec. 4.7] D-C MACHINES 195
the Rototrol. If the magnetization curve of the pilot generator is taken
to be a straight line, the internally generated voltage of the pilot generator
is

Eg kg
* ,N i,
(4.36)
tf + m11

where kg is the generator gain constant, as defined in Eq. (4.4), Ni is the

number of turns on the series field, and Nf is the effective number of con¬
trol-field turns. This expression ignores mutual inductance between the
two fields and the armature and also the effect of eddy currents in the iron.
The armature current i\ is now given by

Ec
^l (4.37)
R 4“ Ls

where R is the total resistance and L the total inductance of the armature
circuit. Note that R may include armature reaction and L mutual-
inductance effects, as explained in Sec. 4.4. Combining Eqs. (4.36) and
(4.37) gives then
kglf
ii = (4.38)
R - (kgN i/N/) + Ls

The current i\ is the field current for the main generator. If this gener¬
ator is linear, then its open-circuit output voltage is

E0 = k2ii (4.39)

where k2 is the main-generator voltage constant. Also the control-field

current if may be related to the control voltage Ef by Eq. (4.6). Com¬
bining Eqs. (4.38), (4.39), and (4.6), we obtain

Eo (kg/Rf) k2
(4.40)
Ef (Tfs + 1 )[R- (kgNi/Nf) + Ls]

The critical value for R is the value that makes

kgN!
R = Rr = (4.41)
N,
Hence the transfer function for a critically tuned Rototrol becomes

E0 kgk2 1
(4.42)
Sf RfL S(Tfs + 1)
The free $ in the denominator substantiates the statement made above
that the critically tuned Rototrol is an integrating device.
The effect of eddy currents in the iron of either the pilot generator or
the main generator is fundamentally the same as for the simple generator.
196 CONTROL SYSTEM COMPONENTS [Chap. 4

For the pilot generator, Eq. (4.18) applies without change, and eddy cur¬
rents simply increase the effective field time constant. For the main
generator, it can be shown by arguments similar to those employed in
connection with Eq. (4.18) that eddy currents simply decrease the total
inductance L of the armature circuit. Demonstration is left to the reader.
Mutual inductance between the control circuit and the armature cir¬
cuit also has essentially the same effect here as in the simple generator,

Fig. 4.13. A speed-control system using a Rototrol and d-c motor.

and an analysis similar to that carried out in connection with Eq. (4.28)
leads to the result that Eq. (4.40) must be changed to read

Eo (k2/Rf)(kg - Mafs)
(4.43)
Ef kgNi T (kg — Mafs)Mafs
(T fS + 1 )R —
~N7 + Ls+ R, + Lfs

when mutual inductance is taken into account. A check of polarities

indicates that Maf in the above equation must be positive if the series
winding is wound in a direction to aid the control field. Hence it appears
that the Rototrol generator has a non-minimum-phase type of transfer
function;1 however, since the mutual-inductance effect is small, the non¬
minimum-phase root occurs at very high frequencies and is usually
negligible. Equation (4.43) also indicates that mutual inductance does
not affect the critical value of R, nor does it affect the integrating action
of a critically tuned Rototrol.
1 Bode, “Network Analysis and Feedback Amplifier Design,” D. Van Nostrand
Company, Inc., Prirtceton, N.J., 1945, pp. 242-244.
Sec. 4.9] D-C MACHINES 197
4.8. Pilot Generators with Several Control Windings. Pilot genera¬
tors with more than one control winding are sometimes used in control
systems where an electronic amplifier is undesirable. A typical system
of this sort is shown in Fig. 4.13, where a Rototrol and d-c motor are used
in a speed-regulating system.
To analyze the effect of a number of control fields on a generator, we
consider for the moment a generator with two control fields, as shown in
Fig. 4.14. The analysis is facilitated by
the use of the superposition theorem,
which permits us to replace first one and
then the other of the two voltages by its
internal impedance and to add the results
of the individual computations. When
one of the two voltages in Fig. 4.14 is
replaced by its internal impedance, the
resulting circuit is identical to Fig. 4.8.
Fig. 4.14. Rototrol with two con¬
Hence the expression giving E0 as a func¬ trol fields.
tion of the remaining input voltage will
have the same form as Eq. (4.15). If the mutual coupling between the
two fields is unity, we may then adapt Eq. (4.18) to obtain

(ki/Ri)Ei + (k2/R2)E <

E0 = (4.44)
KWRO + (L2/R2)]s + 1

Here ki and k2 are voltage constants defined by

Eo k\t i if i2 = 0
Eo k2t2 if ii = 0

and R\ and R2 represent the total resistance, including that of the voltage
source in each of the two control circuits. This result may be generalized
to more than two control circuits.
4.9. The Regulex Generator.
Instead of using a series field for
the self excitation of the pilot
generator, as in the Rototrol, it is
also possible to use a shunt field.
The resulting two-stage generator
is called the Regulex and has been
Fig. 4.15. A schematic of the Regulex gen¬
used in systems built by the Allis-
erator.
Chalmers Company. A schema¬
tic diagram of this machine is shown in Fig. 4.15. Here again the
pilot generator is shown with a push-pull field designed for excitation
by an electronic amplifier; however this generator, like the Rototrol,
is often built with several control fields. The theory of operation is
198 CONTROL SYSTEM COMPONENTS [Chap. 4
very similar to that of the Rototrol, and by proper adjustment of
the shunt-field resistance Ri, it is possible to obtain integrating action from
this system. The complete transfer function has a slightly different form
from that of the Rototrol; its computation is left to the reader.
4.10. The Amplidyne Generator. A generator often used in control
systems is the Amplidyne, first described by Alexanderson, Edwards, and
Bowman.1 The Amplidyne is essentially a two-stage generator, like the

Fig. 4.16. Amplidyne principles: (a) <£/ is the flux set up by the control- field current I/;
4>Q is caused by current Iq. (b) Quadrature brushes short-circuited; </>d opposes <£/,
creating a constant-current generator, (c) Addition of compensating winding con¬
verts machine to Amplidyne.

Rototrol described in the previous sections; however the two stages are
combined in a single machine with a single armature winding. The
Amplidyne therefore tends to be somewhat smaller in size than a Rototrol
of the same rating.
The principle of operation is as follows: Consider the simple d-c gener¬
ator of Fig. 4.16a with a control-field current I/. This current produces a
flux <f)f, and if the machine rotates in the direction shown, the voltage eq
induced in the conductors will be in the direction indicated by the con-
1 E. F. W. Alexanderson, M. A. Edwards, and K. K. Bowman, Dynamoelectric
Amplifier for Power Control, Trans. AIEE, vol. 59, pp. 937-939, 1940.
Sec. 4.10] D-C MACHINES 199
ventional system of dots and crosses. If a current is taken from the
brushes, an armature-reaction flux <j>q due to this current appears at right
angles to <£/. In conventional generators every attempt is made to sup¬
press the armature reaction, but in the Amplidyne a magnetic path is pro¬
vided to encourage it. Furthermore the brushes are short-circuited so
that the maximum amount of 4>q is produced for a given control-field cur¬
rent. The armature-reaction flux will then also induce voltage in the
armature conductors. This voltage is in a direction indicated by the dots
and crosses of Fig. 4.166. The second set of brushes shown in that figure
is used to connect this voltage to the load. We refer to this voltage as the
direct-axis voltage ea. Since the quadrature flux (f>q is very much larger
than the control flux <f>f, the direct-axis voltage is very much larger than
the quadrature-axis voltage eq. The additional amplification takes place
in the magnetic circuit of the machine. The armature-reaction flux is
somewhat analogous to the pilot-generator armature current of the
Rototrol system.
When the output brushes of the machine shown in Fig. 4.166 are con¬
nected to a load, the resulting current flowing in the armature will again
result in an armature-reaction effect, and the reader may convince himself
that this effect opposes the control-field flux. The load current is there¬
fore limited to a value such that the mmf of this armature reaction is
equal to the magnetomotive force produced by the control field If. In
other words the machine acts as a constant-current generator. Its load
current is proportional to the control-field current and essentially inde¬
pendent of speed and direction of rotation of the armature. Referred to
as a Rosenberg generator, this type of machine has been used for railroad-
train lighting service since 1905, and as the Metadyne it has been used in
diesel-electric locomotive drives.1
In servo systems the constant-current feature is usually undesirable
and is eliminated by passing the load current through a compensating
field so located that the armature reaction of the load current is canceled
by the mmf of the compensating field. This is shown in Fig. 4.16c. The
addition of the compensating field converts the machine to an Amplidyne,
a machine delivering a voltage that is only moderately affected by the load
current. Amplidynes are occasionally built with a number of different
control fields for purposes similar to those already discussed in connection
with the Rototrol. Sometimes the quadrature brushes are not short-
circuited directly; instead they are connected through field windings that
may be located in either the quadrature or direct axis to obtain certain
special operating characteristics such as, for instance, integrating action.
A major problem in Amplidynes is sparking at the brushes, due to the
armature reaction on which the operation of the machine depends.
1 Pestarini, “ Metadyne Statics,” John Wiley & Sons, Inc., New York, 1952.
200 CONTROL SYSTEM COMPONENTS [Chap. 4

Special brushes and construction of the field poles are used to alleviate
this difficulty. Nevertheless Amplidynes often spark very severely at
the quadrature brushes when delivering large output voltages.
4.11. Analysis of the Amplidyne. The following analysis of the Ampli-
dyne is based on a paper by Bower1 and assumes complete linearity of
the electric and magnetic circuits, neglects effects of eddy currents in the
iron, and assumes the machine to be running at constant speed. The cir¬
cuit under discussion is shown in Fig. 4.17, and the resistances and
inductances indicated in that figure in all cases represent the entire resist¬
ance and inductance of each of the three circuits. We shall also be con¬
cerned with mutual effects between the circuits and shall distinguish

Rf
-vWf
T Control
Ef /) Lf field

between two types. The first, designated by M, is the mutual inductance

and relates voltage induced in one circuit to the rate of change of current
in another circuit. The symbol N will be used to designate a speed-volt¬
age parameter relating voltage generated in one circuit to the current in
another circuit. To identify the various circuits we make use of a double¬
subscript notation using the subscripts /, q, and d to designate the control
circuit, quadrature circuit, and direct circuit, respectively. Thus the
symbol Mfq is the ratio of voltage in the quadrature circuit to the rate of
change of control-field current causing it.
Before proceeding further with the analysis, we wish to show that,
except for the mutual effects discussed above, the quadrature and direct
circuits may be considered independent even though they use the same
armature winding. The armature circuit has the form shown schemati¬
cally in Fig. 4.18, where the brushes need not be exactly 90° apart. It is
required only that each pair of brushes be exactly 180° apart. Under
these conditions Z\ = Z3 and Z2 = Z4, and therefore Zi/Z2 = Z3/Z4.

1 J. L. Bower, Fundamentals of the Amplidyne Generator, Trans. AIEE, vol. 64,

pp. 873-81, 1945,
Sec. 4.11] D-C MACHINES 201
The circuit represents a balanced bridge, and therefore the voltage Ed is
unaffected by iq, and vice versa.
We now write the Kirchhoff mesh equation for each of the three circuits,
using Laplace transform notation. For the control field we have

Ef — (Rf -f- Lfs)if — M qfSiq — MdfSid (4.45)

'o
In this equation Mqf is the mutual inductance between the control field
and quadrature circuit and should be zero if the brushes are located
exactly on the neutral axis and if the quadrature circuit does not have
field windings collinear with the control circuit. In order to consider the
possibility of shifted brushes, however, we retain this term and adopt the
convention that a brush shift in the direction of rotation will make M qf
more positive. The situation is different for Mdf, which is the mutual
inductance between normally collinear
fields and which is affected only
slightly by brush shift. Mdf is, how¬
ever, a function of the effectiveness of
the direct-axis compensating winding.
Theoretically Mdf would be zero if the
compensation of the direct-axis arma¬
ture reaction were perfect. In order
Fig. 4.18. Schematic of Amplidyne
to include the effect of imperfect com¬
armature.
pensation in the analysis, we adopt the
convention that Mdf is positive in an undercompensated machine, i.e., a
machine for which the output voltage drops as load current is taken
from it.
For the quadrature circuit we have O

Eq — Nfqif — —MfqSif T- (Rq + Nqq + Lqs)iq -f- (Ndq — Mdqs)id (4.46)

In this equation we assume (rather arbitrarily) that Eq = Nfqif is the

“driving voltage’’ in the quadrature circuit which is opposed by the self-
and mutual-impedance drops listed on the right side of the equation.
Here again the mutual-inductance terms Mfq and Mdq are theoretically
zero if the brushes are in the neutral axis and exactly 90° apart. Mfq
must be the same as Mqf discussed above and becomes more positive as
the brushes are shifted in the direction of rotation. However, since Mdq
remains zero as long as the brushes are 90° apart, we adopt here the fur¬
ther convention that Mdq becomes positive if the output brushes are
shifted relative to the quadrature-circuit brushes in the direction of
rotation. The parameter Nqq represents speed voltage induced in the
quadrature circuit by the quadrature current itself. This may be due to
mmf produced in the direct axis by the quadrature current because of a
series winding, or else it may simply be the effect of armature reaction in
202 CONTROL SYSTEM COMPONENTS [Chap. 4

the quadrature circuit. This parameter could be lumped with the resist¬
ance Rq to form an “effective” resistance R'q, as discussed in connection
with the simple generator (Sec. 4.4). Ndq represents the speed voltage
in the quadrature circuit resulting from load current in the direct circuit.
It is not much affected by brush shift but is a function of the amount of
compensation supplied by the direct-circuit compensating coil. It is
assumed to be positive for an undercompensated machine. Therefore, if
Ndq is positive an increase of load current causes a decrease of the quadra¬
ture current and hence the output voltage.
For the direct circuit we have

Ed — Nqdt {N fd ■— Mfds)if — Mqdsiq + (Rd + Ndd + Lds)id + Vo

(4.47)

This equation has the form of a generated voltage, Nqdiq, set equal to the
sum of the impedance drop and the terminal voltage. All the remarks
made previously concerning the signs of the mutual terms apply here
without change. Ndd is the same type of parameter as Nqq in Eq. (4.46).
Nfdif is the speed voltage generated in the direct circuit by direct cou¬
pling from the control field. It is zero if the brushes are on the neutral
axis. Nfd is positive when the brushes are shifted in the direction of
rotation.
The primary purpose of this analysis is the determination of the
Thevenin equivalent of an Amplidyne; hence we wish to solve Eqs. (4.45)
to (4.47) in such a way that the open-circuit voltage and output imped¬
ance become evident. This is most easily done by rearranging these
equations in the form
(U-v ,/>
Ef — idZdf = Zffif + Zqfiq
— $dZdq — Zfqif -fi Z„„% qqvq (4.48)
— idZdd — Zfdtf + Zqdlq + Vo

where, by comparison with Eqs. (4.45) to (4.47), we have

Zff Rf L/S
Zfq — Nfq — MfqS

Zfd Nfd — Mfds

Zqf -MqfS

Zqq Rq *T Nqq + L qS (4.49)

Zqd — N qd — M qdS
Zdf — MdfS

Zd q Ndq — MdqS
Z dd Rd + Ndd + LdS
Sec. 4.11] d-c machines 203

The solution of Eq. (4.48) is shown below in determinant form:

z,f Zqf 1 Zff Zqf zdf

Zfq Zqq 0 Zfq Zqq Zd q
Zfd Z qd 0 Zfd Z qd Zdd
Hif (4.50)
Zff Zqf 0 Zff Zqf 0
Zfq Zqq 0 Zfq Zqq 0
Zfd Z qd 1 Zfd Z qd 1

and is of the form

Vo = HEf - Z0U

where H is the no-load transfer function and Z0 is the output impedance

of the Amplidyne. HEf may be considered the Thevenin open-circuit
voltage.
Although Eq. (4.50) is a complete description of Amplidyne perform¬
ance, subject to the assumptions made at the beginning of this paragraph,
it is too cumbersome to be of much practical use. It is therefore neces¬
sary to make a number of simplifications. These simplifications are made
on the following bases: (1) In accordance with the discussion on mutual
inductance in Sec. 4.4, where it was shown that mutual-inductance effects
are often negligible compared to speed effects in generators, we may
neglect a number of mutual-inductance terms in Eq. (4.50) where this is
desirable. (2) A number of coupling terms in Eq. (4.50) are zero when
the brushes are located properly in the neutral axis and can, therefore, be
considered negligibly small if only small shifts are permitted.
The expression for the open-circuit voltage becomes much more man¬
ageable if we let Zqf = —Mqfs = 0; this simplification is justified on both
the grounds given above. Then the open-circuit voltage becomes

N fqN qd/Rf(Rq + N qq) Afd — M fds

(4.51)
_ (T/s + 1 )(Tqs + 1) R^~+TJs).

where Tf — the control-circuit time constant, and Tq = —r->

Ef \Eq + A qq)
the quadrature-circuit time constant. In this expression the term of
primary importance, and the only one usually considered, is the first.
It indicates that the output voltage is a function of the square of the
speed (since both N’s are proportional to speed) and that the open-circuit
transfer function is characterized by two time constants. Brush shift
has an effect on this term through the parameter Nqq, which increases
with brush shift in the direction of rotation and which may become nega¬
tive for brush shifts in the opposite direction. Nqq can also be made nega¬
tive by a series coil in the quadrature circuit so arranged that its mmf is in
a direction to aid the mmf of the control field. In this way Nqq can be
204 CONTROL SYSTEM COMPONENTS [CHAP. 4

made exactly equal and opposite to Rq, and the open-circuit output volt¬
age becomes approximately

p __ p N f qN gd/RfL q
(4.52)
tj° s(Tfs + )
1

indicating that the output voltage (at low frequencies) is the integral of
the input voltage. If an Amplidyne is to be used in this way, a series coil
with an adjustable shunting resistor may be used to adjust Nqq.
The second term of Eq. (4.51) represents the component of the output
voltage resulting from direct coupling between the control field and the
output circuit. If the brushes are not shifted from their neutral position,
N/d is theoretically zero; hence the only coupling remaining is that caused
by mutual inductance. This will also be quite small in an Amplidyne
that is properly compensated. Thus, the entire term is usually neglected.
It should be pointed out, however, that while the resulting low-frequency
approximation to the Amplidyne transfer function is usually satisfac¬
tory, one occasionally finds an unusual system in which there is a minor
loop or a parasitic loop with a wide passband. In such a system the
presence of a term like the second term of Eq. (4.51) that does not vanish
at high frequencies may make the difference between a stable system and
one that oscillates at high frequencies.
The output impedance is given by a much more complicated expression
than the open-circuit voltage, and our simplifications must therefore
become more extensive. We assume that the brushes are located exactly
on the neutral axis, so that

Zfq = —Nfq Zqd = — N qd

Zfd = — M fdS Zdq = Ndq (4.53)
Zqf = 0

with the other Z’s as in Eq. (4.49). With these simplifications

NdqNqd NdfNfqNqd
Rq T~ Nqq R,(Bq + N„) s
Z0 — Rd + Ndd T* RdS +
TqS + 1 (TfS+ + )
1 )(Tqs 1

_ R, (4.54)
Tfs + 1

In this expression the first three terms represent simply the self-imped¬
ance of the load circuit, and it is almost obvious that it would be a compo¬
nent of the output impedance. The first fractional term represents the
effect of imperfect compensation on the output impedance. It is positive
in an undercompensated Amplidyne, and in many commercial machines it
Sec. 4.12] D-d Machines 205
is several times larger than the first term at low frequencies. This term
may be represented as the impedance of a resistance in parallel with a
capacitance, as shown in Fig. 4.19, and it may be so represented in an

NdaN
dq+V qd R
R VAAAAA-
Rq + N qq
RC = T, C

equivalent circuit of the Amplidyne. Ndg Ng/j

The last two terms are usually omitted R-
Rq Nqq
because they are primarily due to mutual
RC - Tq
inductance and therefore are negligible
Fig. 4.19. Equivalent circuit of
compared to the other two terms. How¬ output impedance of Amplidyne.
ever, here again these terms must some¬
times be retained in the “unusual’’ system.
4.12. The Equivalent Circuit of the Amplidyne. If we consider only
the first term of Eq. (4.51) for the open-circuit voltage and only the first
four terms of Eq. (4.54) for the output impedance, the equivalent circuit
shown in Fig. 4.20 results. Typical values for the time constants are:

Tf = Ll between 0.1 and 0.2 sec. May however be reduced to a few

Rt
milliseconds by use of high-resistance drive, such as pentode
vacuum tube

T„ = t between 0.02 and 0.07 sec

Rq + Nqq
Ld
T, = ) between 0.005 and 0.05 sec
Rd + Ndd

C=
Ndq Nqd
Rd+ Ndd Ld
■K
-wwvw-
-\MAA-
Nfq Nqd
_ Ndq Nqd
Co ~ Rf^Rq+Nqq^ Ef
Rq + Nqq Vo
(Tf s + 7)(Tq s + /)

Fig. 4.20. Approximate equivalent circuit of the Amplidyne.

The values of gain and output resistance depend, of course, on the size
and other construction features.
The reader should note that one effect of the parallel RC circuit in the
equivalent circuit is that the quadrature-circuit time constant Tq does not
necessarily appear in the transfer function of a loaded machine. The
demonstration of this fact is left as an exercise (see Probs. 4.3 and 4.4).
206 CONTROL SYSTEM COMPONENTS [Chap. 4

4.13 Inaccuracies in the Amplidyne Analysis. One of the assumptions

made at the beginning of the analysis was that both the electric and mag¬
netic circuits of the Amplidyne were linear. Without this assumption the
analysis could not have been carried through. However, such an assump¬
tion is at considerable variance with observed facts. In the Amplidyne
both magnetic hysteresis and the nonlinearities caused by the brush con¬
tact serve to alter the performance from that expected by a linear analysis.
We wish to consider particularly the latter effect, since it is somewhat
unusual.
In most generators it is perfectly proper to neglect the nonlinear brush
resistance in comparison with the usually very much larger resistance
encountered in the load circuit. In the Amplidyne quadrature circuit,

Fig. 4.21(a) Brush voltage-current characteristic; (b) actual Amplidyne hysteresis

curve.

however, the brush-contact drop is a very large part of the total voltage
drop, since the brushes are short-circuited. The exact nature of the
brush-contact voltage depends on the brush material and contact pres¬
sure. The general shape of the voltage as a function of current is similar,
however, to that found in electric arcs: a high voltage at small currents,
decreasing as the current increases (see Fig. 4.21a). The result is that a
small quadrature voltage can produce only a very small quadrature cur¬
rent, but as the quadrature voltage increases, the current becomes larger
much more rapidly, sometimes showing a sort of trigger action due to the
brush-voltage function. As a result of this action, the over-all hysteresis
curve (output voltage versus control-field current) characteristically has
a shape similar to that shown in Fig. 4.216, with a deadband effect at the
center. When an Amplidyne is used in a control system, this deadband
tends to reduce the loop gain and the accuracy as the input and output
approach correspondence. There may also be a tendency to hunt, if the
deadband is very severe.
Sec. 4.15] D-C MACHINES 207
4.14. Comparison of the Various Generator Types. A question of
considerable practical interest is which of the three generators, the Ampli-
dyne, the Rototrol, or the Regulex, is “best” for the application at hand.
While we can give no complete answer to this question, a short discussion
relating the various features of the machines may be helpful.
The Rototrol and Regulex are very similar. They are essentially two
simple generators operated in a cascade connection, equipped with a
special feedback connection that makes the output proportional to the
integral of the input. In other respects their characteristics do not differ
from those of two simple generators designed to be used together. They
would have the same maximum power output, the same overload capac¬
ity, the same size and weight, and about the same speed of response. The
integrating feature is of interest primarily in control systems that do not
use electronic amplifiers to drive the generator fields. If electronic con¬
trol amplifiers are used, the integration can be performed more easily and
probably more accurately within the control amplifier. Thus, if a given
power output is required, there seems little reason to choose either the
Rototrol or the Regulex in preference to two simple generators.
The Amplidyne is also essentially a two-stage generator; however, the
two stages are combined on a single armature. In performance the
Amplidyne is probably somewhat poorer than two separate generators.
This is due to the various interactions taking place inside the machine.
The problem of rather heavy sparking at the brushes has already been
mentioned. Furthermore, although a number of different characteristics
may be obtained from the Amplidyne by brush shifting or different wind¬
ing arrangements, it seems clear that two separate generators driven by
an electronic amplifier should permit much greater flexibility in the adjust¬
ment of the characteristics to meet a particular requirement. The chief
advantage of the Amplidyne is undoubtedly its compactness. Since it
uses only a single yoke and armature, it occupies less space and weighs less
than two equivalent cascaded generators would. Hence Amplidynes
are often preferred in aircraft applications.
4.15. D-C Motors with Separate Excitation. When one of the d-c
generators discussed in the previous section is used to control the speed of
a d-c motor, the arrangement used is usually the familiar Ward-Leonard
system shown in Fig. 4.1, where the motor field is separately excited
by a fixed supply and the two armatures are connected. The transfer
function relating the speed or position of the motor shaft to the control
signal applied to the generator will depend on the type of generator
used. Since all the generators discussed thus far may be replaced by
equivalent circuits of the Thevenin type, it is only necessary to find the
general transfer function of a motor connected to such an equivalent
generator.
208 CONTROL SYSTEM COMPONENTS [Chap. 4
The linear analysis of motors requires (1) the assumption of viscous
friction at the output shaft, (2) linearity of the electric and magnetic
circuits, (3) negligibly small armature reaction. For the moment, no
attempt will be made to justify these assumptions, but the effects of the
nonlinearities and armature reaction will be considered later.
The system shown in Fig. 4.22 will be considered. The motor is con¬
nected to a generator with open-circuit voltage E0 and output impedance
Z0. It has an armature resistance Rm and an inductance Lm. The arma¬
ture has a moment of inertia J, and the coefficient of viscous friction
between the armature shaft and a stationary reference system is B. The
armature current is ia, and the motor shaft position is 6. The motor is
connected to a load requiring a load torque Ql.

Fig. 4.22. D-c motor with armature control.

The counter emf is given by the same equation as the open-circuit

generated voltage of a generator, Eq. (4.1); however, here Na, P, </>, and p
are all constant, and 12 = 6 is the independent variable. Therefore this
equation is written in the form

Ec = kv§ = kvsO (4.55)

The torque developed by a motor is given in electric machinery' texts1
as Qd = k<t>ia, where k is a constant containing the same sort of parameters
as are contained in Eq. (4.1). Again, if the magnetic flux <j> is constant,
this can be written as
Qd = kda (4.56)
A fixed relationship exists between kt and kv, as is indicated by the follow¬
ing argument: In electrical terms the power developed is the product of
counter emf and armature current. In mechanical terms it is the product
of developed torque and speed. These two expressions must be equal,
but they are usually expressed in different units. Hence we have
Ecia = CQdd (4.57)
where C is the factor converting electrical to mechanical units of power.
Using Eqs. (4.55) and (4.56), this gives
kv = Ckt (4.58)
1 For instance, Dawes, “A Course in Electrical Engineering,” vol. I, “Direct
Currents,” 4th ed., McGraw-Hill Book Company, Inc., New York, 1952, p. 485.
Sec. 4.15] P-C MACHINES 209
The value of C is unity if a consistent set of units such as the inks system
is used.
The armature current is given by

A E0 - Ec
^a (4.59)
Zo + Rm + LmS

Finally the sum of all torques applied to the motor shaft must be zero, or

Qd = k$a = J4“ Bsd -f- Ql (4.60)

The three equations, (4.55), (4.59), and (4.60), contain five unknowns;
hence we can solve for three as a function of the other two. We consider
E0 and Ql the independent variables and solve for 0. The solution is

(Eo/kv) - (QLZ/kvkt)
(4.61)
s[(JZ/kvkt)s + (BZ/kvkt) d- 1]

where Z is the total armature-circuit impedance: Z = ZQ + Rm + Lms.

Note the free s in the denominator of this expression; its presence indi¬
cates that a motor acts as an integrating device for the conditions given.
This is simply due to the fact that the shaft position is the integral of
shaft speed.
It is worth pointing out that Eq. (4.61) has the form

§ = HE0 - ZmQL (4.62)

in which HE0 is the no-load speed and Zm might be referred to as a mechani¬

cal output impedance. This means that a motor may be represented by
a mechanical Thevenin equivalent, a concept that is occasionally useful.
In many cases of practical importance the armature-circuit impedance
may be considered to be resistive. This is a very good assumption when
a simple generator or a Rototrol is used to drive the motor, and even with
an Amplidyne drive the results are usually not far in error. If this simpli¬
fication is made and if the total resistance of the armature circuit is R,
Eq. (4.61) indicates that for E0 = 0 and Ql constant, the motor speed will
reach the value

_Ql_ _ _Ql_ (4.63)

B T- (kvkt/R) {kvkt/R)[ 1 -f- (BR/kvki)]

The denominator term B + kvkt/R represents the effective coefficient of

viscous damping of the motor. It is made up of the coefficient of mechan¬
ical friction, B, and the term kvkt/R, which represents the electrical damp¬
ing. In most motors, the electrical damping is very much larger than the
mechanical friction, typical values of the ratio BR/kvkt being of the order
of 0.01. Hence the term BZ/kvkt in Eq. (4.61) may be omitted, and the
210 CONTROL SYSTEM COMPONENTS [Chap. 4
transfer function takes the approximate form

a_ K/kv QhR/kykt
(4.64)
s[(JR/kvkt)s + 1] s[(JR/kvkt)s + 1]

In this equation the factor JR/kvkt is a time constant referred to as the

motor-inertia time constant, and it has a value ranging from about 0.01
to 0.1 sec in motors of standard design connected to a zero-impedance
source.
4.16. Relation between the Transfer Function and the Speed-Torque
Curves. The equation for the speed-torque curves of a d-c motor may
be obtained from the complete motor transfer function [Eq. (4.61)]. In
order to do this, first multiply Eq. (4.61) through by 5 to convert the out¬
put to speed rather than position and then set s equal to zero. The last
step is required because speed-torque curves are static characteristics,
and it changes the Z in Eq. (4.61) into Ra, the total armature-loop resist-
ance. The equation for the speed-torque curves becomes
0 = kiE0 — k2QL (4.65)

where /, _ 1 (4.66)
1 fc„( 1 + BRa/kA)
7 _ Ra (4.67)
and
BRa + kvkt
Since all the parameters defining k\ and k2 are constant in a linear motor,
the speed-torque curves are parallel straight lines, and they are equidis¬
tant for equal increments of E0. A typical set of such speed-torque
curves is shown in Fig. 4.23.
Of greater practical interest than the determination of the speed-torque
curves from the transfer function is the reverse process, namely, obtaining
the transfer function from the speed-torque curves. This is often a con¬
venient procedure in practice when the speed-torque curves have been
measured or are otherwise available. The moment of inertia of the motor
armature must be known. Suppose that it has the value J and suppose
an external torque Q0 to be applied to the motor. Setting QL in Eq. (4.65)
equal to Q0 + J dO/dt and using Laplace transform notation, we obtain
sd = kiE0 — k2Js2d — k2Q„
A

z _ kiE0 k2Q0
or (4.68)
s(k2Js -j- 1)
If the values of Aq and k2 from Eqs. (4.66) and (4.67) are substituted into
Eq. (4.68), it will be found that an expression almost identical in form
with Eq. (4.61) is obtained. The only difference is that the armature
impedance Z in Eq. (4.61) has been replaced by the resistance Ra. Thus
the transfer function that has been obtained from the speed-torque curves
is correct if the armature inductance is negligible.
Sec. 4.16] D-C MACHINES 211

The procedure outlined above may be used to determine the approxi¬

mate transfer function of any power element for which the speed-torque
curves are known. It is particularly useful in obtaining the quasi-
linear transfer function of nonlinear power units. In defining this trans¬
fer function, the assumption is made that the characteristics of the unit
are sufficiently regular that, for small variations of the input and output
variables, they may be replaced by straight lines. The quasi-linear
transfer function is usually a function of the operating point about which
the small variations take place.
To illustrate the method of obtaining the transfer function, consider the
speed-torque curves of a typical nonlinear device, say, a d-c motor, driven

Fig. 4.23. Speed-torque curves of linear d-c motor.

by a nonlinear source of some sort (see Fig. 4.24). The curves are
assumed to be plotted for integral increments of some input variable x
corresponding to the generated voltage E0 in Eq. (4.65). At the oper¬
ating point P the average vertical distance between the curves is equal to
the parameter kh while the slope of the particular curve passing through
P is — fc2. By direct analogy with Eq. (4.68) the transfer function at the
operating point P is then given by

kix — k2Q0
(4.69)
s(/c2*/s -f 1)

Since the speed-torque curves are not linear and not equidistant, the
parameters ki and k2 will vary with the operating point.
It should be clear from the above discussion that the transfer function
obtained from the speed-torque curves can show only a single time con-
212 CONTROL SYSTEM COMPONENTS [Chap. 4

Fig. 4.24. Nonlinear speed-torque curves for d-c motor.

stant, namely, the inertia time constant. Its accuracy depends, there¬
fore, on the assumption that other time constants are negligible.
4.17. An Equivalent Circuit for the D-C Motor. It is sometimes con¬
venient to replace the motor by an equivalent electric circuit element.
Such an element can be found if Eqs. (4.55), (4.59), and (4.60) are solved
for the armature current in terms of E0 and Ql. The result may be put
in the form
Ql/kt
E0 +
(J/kvkt)s + B/kvkt
(4.70)
1
Z0+ Rm + LmS + (.J/kvkt)s + (B/kvkt)

If for the moment we ignore the load torque, it is seen that the denomina¬
tor represents an impedance consisting of the elements ZQ, Rm, Lms, all
connected in series with a capacitor
Rm <-m
J/kvkt in parallel with a resistance
-VWNA-
B/kvkt. Furthermore, it may be
Motor
termino/s
J B__ demonstrated that a current source
kvkT kykj
Qh/kt connected in parallel with
the combination J/kvkt and B/kvkc
Fig. 4.25. An equivalent circuit of a d-c
motor.
properly represents the load. The
circuit therefore takes the form of
Fig. 4.25, where we have omitted the output impedance Z0 of the gener¬
ator. It can be shown that the voltage across the capacitor is the counter
emf of the motor, and the speed can therefore be obtained by dividing the
capacitor voltage by lcv.
Sec. 4.19] D-C MACHINES 213
If friction is neglected and if there is no shaft load, the equivalent cir¬
cuit is a simple RLC series circuit, and all equations pertaining to the
motor will have the form of the equations of this circuit.
4.18. Inaccuracies in the Motor Analysis. The assumptions of linear¬
ity, etc., that were made to facilitate the analysis will now be examined
briefly. Nonlinearity of the electric circuit is confined primarily to the
brush-contact resistance and will cause a slight deadband effect in the
motor’s curve of speed versus input voltage. This effect was discussed
in connection with the Amplidyne (Sec. 4.13) but is so small in a motor
that it is usually neglected. A much more pronounced deadband effect is
produced by the static friction between the rotating and stationary parts
of the motor; it is sometimes serious enough, particularly in small motors,
to cause instability of the servo of which the motor is a part. Aside from
this effect, however, the fact that the friction drag is rarely proportional
to speed is of no great consequence, since it was shown that the friction
term can usually be neglected in the motor transfer function.
Armature reaction in motors results in a reduction of flux, just as it
does in generators; hence both kt and kv are reduced slightly when large
armature currents flow. It is seen therefore that armature reaction in a
motor cannot simply be absorbed in an equivalent resistance as it is in
the generator. However, since the field is usually operated in saturation
and since compensating windings and commutating poles are often used
on motors to improve commutation, armature reaction is not usually a
major factor in motor performance.
Nonlinearity of the magnetic circuit can have no effect on the perform¬
ance, since the motor field is maintained (in the absence of armature reac¬
tion) at a constant strength.
4.19. Other D-C Motor Types. A motor that is fairly often used in
low-power servo applications is the split-field series motor, a circuit of
\vhich is shown in Fig. 4.26. These
motors are usually driven directly from
a push-pull electronic amplifier (see Sec.
3.13), or by a polarized relay, as shown
in Fig. 4.28, rather than from a gener¬
ator. For this reason their power out¬
put is limited to about 50 watts. The
Fig. 4.26. The split-field series
two series field coils are wound in such motor.
a way that, for the current directions
shown in Fig. 4.26, the mmf’s oppose; hence the flux is approximately
proportional to the current difference, if the magnetic circuit is assumed
to be linear. Since the armature current is the sum of the two field
currents, the developed torque is
Qd = kt(Ji + /2)(/1 - U) (4.71)
214 CONTROL SYSTEM COMPONENTS [Chap. 4

and the counter emf may be written as

. Ec = kj(l i - 72) (4.72)

The two currents 7i and 72 may be found by application of Kirchhoff’s

laws. In order to simplify the expressions, let us define a self-impedance
Zs as the impedance measured between one of the outside terminals and
the center terminal with the motor not running and the remaining termi¬
nal open-circuited. This self-impedance should be the same for both
windings and is of the form R + jcoL. R and L are the sums of the resist¬
ances and inductances, respectively, of the armature and one of the field
windings. Let us also define the mutual impedance

dE i dE 2
(4.73)
dlo Ii, 0=0 dh 12, 0 = 0

The mutual impedance will be of the form Ra + jcoLa + jcoM, where Ra

and La are armature resistance and inductance, respectively, and M is the
mutual inductance between the two windings. With these definitions
and the use of Eq. (4.72), we have

E\ — ZSI i + ZmI% 4~ M(/i — ^2)

(4.74)
E2 — Zuh 4“ ZSI2 4- kvB(I1 — 12)

Note that these equations are not written in Laplace transform notation,
since they contain products of the variables. Solving (4.74) for 7i and
72, we get
E\ZS — E%Zm — kv0(E\ — E2)
Z/- Z.iR
(4.75)
E<iZs — E\Zm — kvd(E\ — E2)
Zs2 - Zm2

Let us suppose now that E1 and 772 vary an equal and opposite amount
from a quiescent value E0; i.e.,

Ei = E0 4- AE (4.76)
E2 = Eq — A E
Then the substitution of (4.76) and (4.75) in (4.71) gives for the developed
torque
q _ 4.kfEo AE $ktkv0(AE)2 ,
^ “ (Zs2 - Zm2) (Zs2 - Zm2)(Z8 - ZM) 1 j

It may be observed that, at standstill, the developed torque is directly

proportional to AE and that, for any fixed value of AE, torque and speed
are related by a straight-line function. Speed-torque curves for various
values of AE are shown in Fig. 4.27.
Sec. 4.19] D-C MACHINES 215

Since the motor is seen to be quite nonlinear, it is impossible to find an

exact linear transfer function for it. If, however, operation is confined to
low speeds, as may often be the case in servo systems, and if E0 ^>> AE, an
approximate transfer function can be derived by neglecting the second
term of Eq. (4.77) and equating the developed torque to the sum of all
other torques acting on the motor; i.e.,

Qd, — Ql ~b Js26 -f- Bsd (4.78)

For operation at low frequencies, Zs T Zm ~ Rf 2Ra and Zs — Zm ~ Rf)

where Rf and Ra are the resistances of one field coil and the armature,

respectively. Inserting these approximations into Eq. (4.77) and equat¬

ing to Eq. (4.78) yield the desired result:

[4/c^Eo AE/Rf{Rf + 2Ra)] — Ql

(4.79)
s(Js T B)

We see that, under the assumptions made, the motor does not supply
electrical damping and that the only damping is due to mechanical fric¬
tion. As a matter of fact, in the absence of friction the motor shaft posi¬
tion is the double integral of AE, the control voltage. This behavior
differs quite markedly from that observed in the separately excited motor
discussed in the previous sections.
It should be noted, however, that in general the damping supplied is
proportional to the negative of the slope of the speed-torque curve (see
Sec. 4.16). Hence, at large speeds and large control voltages, the split-
field series motor is very heavily damped. The implications of this state¬
ment and the effect of this nonlinear damping on a servo loop in which a
split-field series motor is used will not, however, be pursued here. As
216 CONTROL SYSTEM COMPONENTS [Chap. 4

has already been mentioned, the motor is often used with a polarized
relay in a circuit such as is shown in Fig. 4.28, where either one field or
the other is excited. The combined nonlinearities of the relay and the
motor are then usually such that only
the most approximate analysis is pos¬
sible,1 and the servo must be designed
more by experience and trial and error
than by any sort of exact analysis.
4.20. The Field-controlled Motor.
Still another motor-control method
Fig. 4.28. Relay-operated split-field
used in servo systems employs the
series motor.
field-controlled motor, in which the
armature is supplied with a constant current. Since the control amplifier
needs to supply only the relatively small field current, this drive may be
used in moderately large power applications without including a generator
in the loop to provide additional power amplification. In a sense the
motor acts as a power amplifier itself. A generator is, however, needed to
supply the required constant armature current; hence the amount of
equipment needed is still about the same as for a Ward-Leonard system
of similar power rating. The only advantage from the point of view of
installed power capacity is that the constant current might possibly be
furnished directly from the a-c line by a judicious arrangement of trans¬
formers and rectifiers. This equip¬
ment might be lighter, less expen¬
sive, and easier to maintain than a
rotating generator.
The motor is commonly built with
a center-tapped field coil to permit
Fig. 4.29. A schematic diagram for the
direct operation by a push-pull am¬ field-controlled d-c motor.
plifier. Since it has already been
shown in Sec. 4.2 that a push-pull field can always be replaced by an
equivalent single-ended field, we consider this type here, as shown in Fig.
4.29. If linearity of the magnetic path is assumed, the developed torque
is given by
Qd = kJatf (4.80)
The field current is given by
Et
V (4.81)
Rf(l + Tfs)
where Tf = Lf/Rf.
If the armature has a moment of inertia J and a coefficient of viscous

1 R. J. Kochenburger, A Frequency Response Method for Analyzing and Synthesiz¬

ing Contactor Servomechanisms, Trans. AIEE, vol. 69, pp. 270-284, 1950.
Sec. 4.20] D-C MACHINES 217
friction B and is connected to a load requiring a torque Ql, the torque bal¬
ance shown in Eq. (4.78) applies. Hence, the transfer function is

ft — (kjg/Rf)Ef _Ql
s(Tfs + 1 )(Js + B) s(Js + B) V y

Here again we note that the motor does not provide any electrical damp¬
ing; the only damping is due to mechanical friction. As in the split-field
series motor, if friction is negligible, the motor acts as a double integrator.
Very often the armature current for this type of motor is not obtained
from a true constant-current source, but rather an approximately constant
current is obtained by connecting the armature to a constant-voltage sup¬
ply through a high resistance. If the motor is permitted to run only at
speeds low enough that the counter emf remains negligibly small com¬
pared to the supply voltage, the armature current will remain almost
constant. This method of supplying the armature has the advantage of
doing away with an expensive and complicated constant-current supply.
A large amount of power is, of course, wasted in the resistance, but since
the drive can be used only for power outputs of about 500 watts or less,
this is a relatively minor disadvantage.
If the voltage supplying the armature is V and if the total impedance in
the armature circuit (including series resistance, armature resistance, and
inductance) is R + jwL, the armature current is given by

. _ V - Ec
(4.83)
u R + jo>L
where Ec, the counter emf, is given by
Ec = kvifd (4.84)
The developed torque is still given by

Qd ktialf (4.85)
Note that Eqs. (4.83) through (4.85) are not written in Laplace transform
notation, since they involve, on the right side, a product of two variables.
In order to get an approximate result, we make a linear approximation of
the two nonlinear equations as follows:
For small variations of if and d around some operating value, Eq. (4.84)
may be approximated by the linear term of its Taylor expansion
dEc 6E,
A Er. = Aif + Ad
dif Ec = Ec 66 Ec = Ec
\
= kv60 Aif + kvif0 A6 (4.86)
where Ec0, 6a, and i/a are the operating-point values of the quantities in
question. Similarly, Eq. (4.85) becomes
A Qd kitao Aif —h kiZfo Ata (4.87)
218 CONTROL SYSTEM COMPONENTS [CHAP. 4

If we consider only small variations of ia and Ec, Eq. (4.83) becomes

— AEC
At a (4.88)
R d- Ls

since V is a constant. If now Eqs. (4.86), (4.87), and (4.88) are combined
with the torque balance [Eq. (4.78)] we obtain

\n = Aif[iao(R + Ls)/kvif02 - S0/if0] - AQL(R + Ls)/kvktif02 ( 0

s[Js(R + Ls)/kvktif02 + B(R + Ls)/kvktifo2 + 1] 1 j

If we let R + Ls = Z we note that this expression is not much different

from the expression for 6 in the armature-controlled motor [Eq (4.61)].
In fact, except for the “gain constant” term multiplying Aif, the differ¬
ence consists in the substitution of kvktif02 for kvkt. Thus when the
motor operates around an operating point other than the standstill
point where ifQ and 60 are zero, its characteristics would be expected to
be quite similar to those of the armature-controlled motor. Note, how¬
ever, that if if0 = d0 = 0, the transfer function becomes
k^aoAXf AQl
A6 = (4.90)
s{Js 4“ B)
Under this operating condition the motor loses all electrical damping
and becomes a double integrator if B, the coefficient of friction, is negli¬
gibly small. Note that Eq. (4.90) is essentially the same as Eq. (4.82);
therefore one may conclude that the motor characteristics for operation
about zero field current are not affected by the method used for exciting
the armature. It is also interesting to note that the speed-torque curves
of this motor have exactly the same form as those plotted in Fig. 4.27
for the split-field series motor. Thus the characteristics of the two motor
types should be identical, and conclusions reached by analyzing one
should be applicable to the other.

PROBLEMS
4.1. Obtain the transfer function of the Regulex generator system (Fig. 4.15) and
determine the value of Ri required for critical tuning. The slope of the magnetiza¬
tion curve of the pilot generator is ki volts/amp of field current, while the slope of the
magnetization curve of the main generator is k2 volts/amp.
4.2. Show in what way the transfer functions of the Amplidyne are affected by
the removal of the compensating winding in the direct-axis circuit. Indicate in
what way the machine becomes a constant-current Rosenberg generator, and find
the transfer function relating control-field current to load current.
4.3. The Amplidyne shown in the equivalent circuit of Fig. 4.20 is loaded by a pure
resistance R0. Assuming that the inductance Ld is negligible, find the transfer func¬
tion V0/fif. Note that the time constant Tq does not appear in this transfer function.
4.4. The Amplidyne shown in the equivalent circuit of Fig. 4.20 drives a d-c motor
having a moment of inertia ./, armature resistance Rm, a counter-emf constant kv,
and a torque constant kt. The inductance of the armature and the mechanical fric-
Problems] D-C MACHINES 219
lion of the armature are negligible. Also neglect the inductance La of the Amplidyne.
Find the transfer function 6/Ef, where 6 is the motor shaft position. Note that the
time constant Tq does not appear in this transfer function.
4.5. The following two tests are performed on an Amplidyne to determine its
characteristics:
a. Open-circuit step-function test: For a step function of field current of 1 ma the
open-circuit output voltage is E0 = 10(1 — e~50t) volts.
b. Short-circuit step-function test: For a step function of field current of 1 ma the
short-circuit output current is 70 = 1(1 — e_500<) amp.
This Amplidyne is used to drive a d-c shunt motor which has an armature resist¬
ance of 1 ohm, kv = 0.1, kt = 0.1, J = 0.0002 in a consistent set of units; friction,
armature inductance, etc., are negligible.
Assume the simplest possible equivalent circuit for the Amplidyne; i.e., all mutual
inductances are zero, and the inductance of the direct axis is negligible. Find a rela¬
tion for the transfer function from Amplidyne field current to motor speed.
4.6. A d-c motor has the following ratings:

No-load speed = UNL radians/sec

Full-load speed = COFL radians/sec
No-load current = Inl amp
Full-load current = Ifl amp

a. Find the ratio BR/kvkt in terms of these ratings, where B = coefficient of viscous
friction, R = armature resistance, kv = voltage constant, kt = torque constant.
b. Determine this ratio for a motor for which

No-load speed = 1,020 rpm

Full-load speed = 1,000 rpm
No-load current = 0.2 amp
Full-load current = 4.5 amp

4.7. Show that the characteristic speed-torque curves of the field-controlled d-c
motor shown in Fig. 4.30 are similar to the speed-torque curves of the split-field series
motor (Fig. 4.27). Show that the quasi-linear transfer function of the motor is,
therefore, similar to that of the split-field series motor.

Fig. 4.30
 CHAPTER 5

SYNCHROS AND RELATED DEVICES

5.1. Accuracy and Precision. An important component of almost all

control systems is the device that converts the input and output variables
from their original physical form to a form that is more convenient to use.
Devices of this type are referred to as pickup devices, or transducers. In
an electromechanical system such transducers convert the position or
velocity of the output into electric signals, which are more conveniently
handled in amplifiers and networks. Since servo systems are used to
control all sorts of physical quantities, there is a large variety of trans¬
ducers of different designs and for different applications, and it would be
possible to write several volumes on the subject.1 Space limitations here
permit us, however, to discuss only a small segment of this large topic,
and we shall confine our attention primarily to a small number of the more
important electromechanical transducers, since these illustrate the basic
problems involved and the methods
of attack.
The primary requirement to be
fulfilled by a transducer is accuracy.
Since the transducer always functions
outside of the control loop, its inac¬
curacies cannot be compensated by
means of feedback. A servo cannot
be any more accurate than its trans¬
ducer. The error in a transducer
may be thought of as consisting of
two components: a repeatable com¬
ponent and a random component. In the absence of any direct knowledge
concerning the randomness of the error, it is usual to assume that the
probability of the error follows the so-called normal law:

p(E) = —L= «-<*-*.)•/*-* (5.1)

■\Z2tto-

This is a bell-shaped function, shown in Fig. 5.1. We define the repeat-

1 See, for instance, Draper, McKay, and Lees, “Instrument Engineering,” vols. 1-3,
McGraw-Hill Book Company, Inc., New York, 1952-1955.
220
Sec. 5.2] SYNCHROS AND RELATED DEVICES 221

able component of the error as the average value Em, and although the
probability curve indicates that the random error extends over a very
large range (theoretically infinite), the randomness is usually measured by
a, the standard deviation defined in Eq. (5.1). It is convenient to normal¬
ize these errors with respect to the range of the measured variable and to
agree on the following definitions:

. . range of variable
Accuracy = ---=-- (5.2)
m
t. . . A range of variable
.Precision = --- (5.3)
<7

The accuracy and precision as defined in these equations increase for the
better instrument, which is in accordance with normal usage of these
words. Note that the error contributing to poor accuracy is in the nature
of an error in calibration. It is normally desirable that transducers be
linear, i.e., that the output variable be directly proportional to the input
variable. Any departure from linearity would therefore give rise to a
repeatable error in a nominally linear transducer and thus reduce the
accuracy as here defined. Also, in many servomechanism applications,
two transducers with closely matched characteristics are used, one at the
input and the other at the output. Any mismatch between these trans¬
ducers would result in a repeatable error and adversely affect the accuracy
of the system. In the following discussion of errors in transducers we
shall be concerned primarily with repeatable errors, since the random
errors contributing to lack of precision are primarily a function of the
workmanship and care expended in the construction of the transducer.
In addition to accuracy and precision, another desirable characteristic
of transducers is high input impedance; i.e., the transducer should require
a minimum amount of power for its operation. This is particularly
important in transducers operating at the input to a servomechanism,
since one of the primary functions of an automatic control system is
accurate power amplification between input and output. In some cases,
it is required that only an infinitesimally small amount of power be
extracted from the input; photoelectric devices are often used in such
applications. Other desirable characteristics are rapidity of response
and low noise output.
5.2. Elementary Operation of Synchros. The most common form of
electromagnetic transducer used to convert an angular position of a
rotating shaft into an electric signal is the synchro (also called selsyn or
autosyn). It is typical of the class of induction transducers commonly
employed in control and measurement applications, and the results of the
discussion on synchros are applicable with some modifications to these
other transducers also.
222 CONTROL SYSTEM COMPONENTS [Chap. 5

In servo applications, synchros are commonly employed in pairs in an

arrangement similar to the one shown in Fig. 5.2. Two different types
of synchros are used, the generator (also referred to as transmitter) and
the control transformer. Other synchro devices include motors (receivers
or repeaters) and differential units. These will be discussed later. In
the system shown, the output voltage from the control transformer is used
as an indication of the relative shaft positions 6 and a of the two machines.
Thus, the system acts as a comparator of the two shaft positions and as a
transducer converting this difference of shaft positions into an electric
signal.
A-c o- e0 - Output
supply °~
The construction of a synchro is
vottoge
very similar to that of a miniature,
r\^9 cffoc three-phase synchronous alternator.
Generotor Control
transformer The stator is constructed with
Fig. 5.2. Schematic of a synchro pair. standard slotted steel punchings con¬
taining a three-phase winding that is
usually Y-connected. In order to minimize slot effects, the stator slots
are often skewed. There are three different types of rotors in common
use: the salient-pole type, the umbrella type, and the cylindrical, or
drum, type (see Fig. 5.3). All of these rotors are wound for two electric
poles, and the windings are brought out to slip rings so that continuous
rotation is possible. Salient-pole rotors are used for synchro generators
or motors. Both the other types of rotors are used in control trans¬
formers, where it is desirable to have uniform reluctance all the way
around the air gap. Both the stator and rotor windings of control trans-

(o) Solient pole (6) Umbrella (c) Drum

Fig. 5.3. Typical synchro-rotor cross sections.

formers usually have a higher impedance than those of the generator

so that it is possible to excite several control transformers from one
generator.
An elementary explanation of the operation of a synchro pair consisting
of a generator and control transformer may be obtained by reference to
Fig. 5.4. The alternating current flowing in the generator rotor sets up
an alternating flux, as shown in the figure, and produces alternating volt¬
ages in the stator by ordinary transformer action. These voltages cause
currents, all of the same phase but of different magnitudes, to circulate
Sec. 5.2] SYNCHROS AND RELATED DEVICES 223
in the stator windings. The result is that an alternating flux pattern is
set up in the stator of the control transformer; this pattern has ideally the
same space position as that in the generator. If the rotor of the control
transformer is turned so that its pole axis is at right angles to this flux, the
output voltage e0 is zero. However, in general, if the reluctance of the
air gap of the control transformer is independent of rotor position, the
output voltage is an a-c voltage proportional to the cosine of the angular
difference between the pole axes of the rotors of the generator and control

Fig. 5.4. Schematic of synchro pair showing stator connections.

transformer. Thus, if the voltage applied to the generator rotor is E sin

cct, then the control-transformer output voltage is (see Fig. 56)

e0 = Eom cos (6 — a) sin (cot — (3) (5.4)

where 6 and a are the angles, measured from the same reference, of the
pole axes of the generator and control-transformer rotors, respectively.
Eom is the maximum value of the output voltage. It normally is equal to
\/2 X 57 volts in 115-volt synchros. The phase shift (3 between input
and output voltages is the sort of phase shift found between primary and
secondary voltages in a transformer and is due to the resistance of the
windings. Since the ratio of resistance to reactance is usually quite
small, /3 also is usually of the order of only a few degrees and is often
neglected.
In operation in a servo system the rotor of one of the synchros, usually
the control transformer, is connected to the output shaft, and the input
is applied to the other rotor. The voltage e0 is applied to the amplifier,
which feeds the motor and turns the output shaft. Hence the servo
operates to make the output voltage from the control transformer zero, so
that, under quiescent conditions, 6 — a = 90°. It is convenient, there¬
fore, to define an angle 5 = a + 90°, in order that the condition for zero
output voltage may be h = 6. In this case

e0 = Eom sin (5 — 0) sin (cct — (3) ^ Eom(& — 6) sin (cct — (3) (5.5)

if 5 — d is a small angle. The output voltage is seen to be an alternating

voltage whose amplitude is proportional to the angular difference between
the rotor shafts of the two synchros, while the output phase angle depends
224 CONTROL SYSTEM COMPONENTS [Chap. 5
on the sign of this difference. Thus, in order to recover the complete
information contained in the output signal, it must be passed through a
phase-sensitive detector, or discriminator. Devices of this sort are dis¬
cussed in the next chapter.
It should be noted that, despite its appearance, a synchro is a single-
phase device, and all voltages and currents measured throughout a
synchro system are essentially in phase except for small, incidental phase
shifts such as the angle /3 mentioned above.
Thus far we have tacitly assumed 5 and 0 to be constant in time. Sup¬
pose now, however, that either <5 or 0 or both are oscillated sinusoidally
with a relatively low frequency of cos radians/sec. Thus

5 — 6 = a m sin cost

If crm is small, the approximate form of (5.5) applies, and

e0 « EomaM sin ccst sin (cot — (3) (5.6)

This output voltage is shown in Fig. 5.5 together with the voltage applied
to the generator, the so-called reference voltage. The phase angle (3

Control-
tronsformer
output
voltage

Reterence
voltage
wvwww
Fig. 5.5. Control-transformer output-voltage waveforms.

is taken to be zero. The figure shows the 180° phase reversal of the out¬
put voltage relative to the reference when the difference 5 — 0 becomes
negative. The output voltage is an example of a suppressed-carrier
amplitude modulation. This may be shown by writing equation (5.6) in
the equivalent form:

e„ = {cos [(co - us)t - 0}- cos [(« + o».)t - 0]\ (5.7)

The voltage is seen to be composed of two sidebands having the fre¬

quencies co — cos and co + cos, but there is no carrier term (co) present. The
form of this voltage is typical of the outputs of a large number of trans¬
ducers used in servomechanisms in which the input variable is used to
modulate the amplitude of an a-c carrier.
i
Sec. 5.3] synchros and related devices 225

5.3. Static Errors and Residual Voltages. The relationship between

the rotor positions and control-transformer output voltage as given in
the previous section describes an ideal situation, from which all synchros
depart to a certain, even if usually very small, extent. It is found, for
instance, that, as the rotors of the two synchros approach correspondence,
the output voltage does not go to zero, as Eq. (5.5) indicates, but instead
only goes to a minimum. The residual voltage remaining at the minimum
consists primarily of a component at carrier frequency co but 90° out of
phase with the predominant phase of the output signal for large differ¬
ences in rotor displacement. It is therefore referred to as the quadrature
voltage, or quadrature error.
It can be shown that one reason for the existence of the quadrature
error is that the impedances making up the stator circuit do not all have
the same phase angle. Thus consider the system shown in Fig. 5.6. The

Fig. 5.6. Currents and voltages in a synchro system.

two currents ia and % are alternating at the carrier frequency to. Their
magnitude depends on 6, the rotor angle of the generator, and although
ideally they are in time phase, in practice a small difference in the phase
angles of the impedances of the stator results in a small phase angle 7
between the currents. Thus, let

ia = Ia sin Oit
ib = h sin (a)t — 7) (5.8)

The control-transformer output voltage is produced by the mutual induct¬

ance between the control-transformer stator and rotor windings. This
mutual inductance is a periodic function of a, the rotor angle. For the
usual cylindrical rotor construction this inductance may be assumed to
be a sinusoidal function of a. In order to determine the exact form of
this inductance, we note from Fig. 5.6 that, when a = 30°, the coupling
between the stator circuit traversed by current ib and the rotor is a maxi¬
mum. Similarly, when a = —30°, the coupling between ia and the rotor
is maximum. Hence, if we let Mc be the maximum value of the mutual
inductance between either ia or 4 and the rotor, then the output voltage
may be written

e„ = Mc cos (a + 30°) ^ + Me cos (a - 30°) ~ (5.9)

dt at
226 CONTROL SYSTEM COMPONENTS [CHAP. 5

By the use of Eq. (5.8), Eq. (5.9) becomes

e0 = coMc[Ia cos (a + 30°) cos cot + Ib cos (« — 30°) cos (cot — 7)]

(5.10)
By expanding the cos (cot — 7) term, this becomes

eQ = a)Mc[[Ia cos (a + 30°) + h cos 7 cos (a — 30°)] cos cot

+ [Ib sin 7 cos (a — 30°)] sin cot] (5.11)
This equation indicates that the rms value of the control-transformer out¬
put cannot go to zero for any value of a, unless Ib is zero, a condition
occurring for only two positions of the generator rotor. The second term
in the equation is the quadrature error referred to above, and for small 7
it is approximately proportional to 7, the phase difference between the
stator currents. It is clear that this voltage can be kept to a minimum
by making certain that the impedances of the stator circuit all have the
same phase angle. Usually synchros are wound quite carefully, and the
windings are almost identical, but unless the connecting cable is per¬
fectly constructed, it may introduce unbalancing impedance that will give
rise to considerable error.
In addition to the effect discussed above, one usually finds harmonic
voltage components near the synchro nulls. Some of these are due to
harmonics present in the supply voltage, but most of them are due to
nonlinearities in the iron, etc. Quadrature voltages of fundamental fre¬
quency are also sometimes caused by unbalanced capacitive coupling
between stator and rotor. This effect is usually noticeable only at higher
carrier frequencies, and although not important in synchros, it is an
important cause of quadrature error in many other suppressed-carrier
types of transducers.
The phase discriminator, which in one form or another is always used
to recover the actual error information from the synchro signal, can, by
proper adjustment, be made insensitive to all quadrature and harmonic
voltages. Hence these residual effects do not actually result in any
appreciable servo error. However, if it is desired to amplify the synchro
error signal in an a-c amplifier, this amplifier must be designed to handle
input amplitudes that are at least as large as the maximum quadrature
signal, and preferably somewhat larger. Hence a large amount of quad¬
rature voltage limits the amount of a-c gain that can be used ahead of the
discriminator. If more gain is needed, a d-c amplifier can be used after
the discriminator. However, since d-c amplifiers tend to drift, it is
usually desirable to have as much of the required gain as possible in the
a-c amplifier. For this reason it is important to reduce the quadrature
error to a minimum. Another disadvantage of a large quadrature signal
is that it results in a considerably larger ripple output from the dis¬
criminator (see Chap: 6) and therefore requires more thorough filtering.
Sec. 5.3] SYNCHROS AND RELATED DEVICES 227
The component of primary importance in the control-transformer out¬
put voltage is the inphase component given by the coefficient of cos wt in
Eq. (5.11). When a synchro pair is used in a servo, the servo system acts
to make this term go to zero. Thus, if zero voltage does not correspond
to exactly the same angular difference between the rotors under all oper¬
ating conditions, an error is introduced into the system. In practice, two
types of errors are found: (1) static errors depending on the position of the
input and output shafts and (2) velocity errors, which are a function of the
speed of the rotors.
A typical test record of the static error of a generator-control-trans¬
former system as a function of generator rotor position is shown in Fig.
5.7. It shows that the error consists predominantly of a second- and

sixth-harmonic function of rotor position. It can be shown that the sec¬

ond harmonic is caused by unbalance of the stator impedances, eccen¬
tricities of the generator rotor relative to the stator bore, or ellipticity of
either the stator bore or the rotor surface. In fact, any sort of stator-
circuit unbalance that might cause the maximum values or the phases of
the stator currents Ia and Ib to be different can be the cause. For zero
inphase output voltage we have from Eq. (5.11)

Ia cos (a + 30°) + lb cos 7 cos (a — 30°) = 0 (5.12)

Ia and Ib, the amplitudes of the two stator currents ia and % shown in Fig.
5.6, are periodic functions of 0 (a “negative amplitude” refers to a voltage
180° out of phase with one having a positive amplitude). The salient
rotor construction commonly used in synchro generators may be expected
to make Ia and Ib nonsinusoidal functions of 6, but for the present it is
convenient to ignore this fact and to assume a sinusoidal variation.
Inspection of Fig. 5.6 indicates that, with 6 = 30°, Ia = 0 and, with
6 = —30°, Ib = 0. Furthermore both Ia and Ib are positive as shown,
when 6 = 0°. Hence, if we let Iam and Ibm be the maximum values of Ia
228 CONTROL SYSTEM COMPONENTS [Chap. 5

and Ib, respectively, then on the assumption of a sinusoidal variation we

have
la COS (6 + 60°)
= lam
(5.13)
Ib = Ibm cos (6 — 60°)
If these values of Ia and Ib are inserted into Eq. (5.12), the condition for
zero output voltage becomes, after some trigonometric and algebraic
reduction,

bm
cos 7 — 1 cos 26
/ am 2 \/S y/Z
cos \6 — a — tarn (5.14)
1 + (ir~ cos
/
7—I i+ sin 26
am
V3
In arriving at this result, we have used the fact that sin (6 + a) may be
written as sin [26 — (6 — a)]. Note that, if the stator circuits are
balanced, Iam = hm and 7 = 0; hence the relation reduces to the ideal
COS (6 — a) =0
(5.15)
or 6 - a = 90°

However, if for any reason the stator circuits are not exactly balanced,
either Iam differs from Ibm or 7 has a nonzero value, or both, so that an
additional angle, which by Eq. (5.14) is a function of 26, must be added in
Eq. (5.15) to reduce the inphase voltage output of the control transformer
to zero. This additional angle represents an error in the synchro system.
The sixth-harmonic error component indicated in Fig. 5.7 can be shown
by a similar argument to be caused by the salient-pole construction of the
generator rotor. This results in a nonsinusoidal variation of the currents
Ia and h with rotor angle that may be expressed in a Fourier series.
Owing to symmetry, only the odd harmonics will be present, and it can
be shown1 that the third, ninth, etc., harmonics cannot exist because of the
balanced three-phase arrangement of the stator. Hence, if we consider
harmonics up to and including the ninth, there will be present, in addition
to the fundamental, only the fifth and seventh harmonics of 6. Thus, if
it is assumed that the stator impedances are all balanced so that Iam = Ibm,
we have
I a — 11 cos (6 -f- 60°) + h cos (56 — 60°) T Ii cos (/ 6 T 60°) ,_
h = I\ cos {6 — 60°) + h cos (56 + 60°) + 17 cos (76 — 60°)

Here Ii, Ib, and 17 are the maximum amplitudes of the fundamental, fifth,
and seventh harmonics, respectively, and are assumed to be the same for
1 See, for instance, Kerchner and Corcoran, “Alternating Current Circuits,” John
Wiley & Sons, Inc., New York, 1938, pp. 268ff. The discussion given there is for a
polyphase circuit, but a similar argument can be used in the single-phase synchro
system to show that the third space harmonic of stator current cannot exist in the
lines. •
Sec. 5.3] SYNCHROS AND RELATED DEVICES 229

Ia and lb- The phase angles for the fifth- and seventh-harmonic terms
can be deduced from inspection of Fig. 5.6 and the facts that, for 6 = 30°,
Ia — 0 and, for 6 = —30°, h = 0. Both I5 and may be negative. If
it is assumed that y in Eq. (5.12) is zero and if Eq. (5.16) is then sub¬
stituted into (5.12), there results, after some reduction, the new relation
for zero inphase output:

/i cos (6 — a) -f- /5 cos (56 + a) + /7 cos (76 — a) = 0 (5.17)

But cos (56 + a) = cos [66 — (6 — a)], and

cos (76 — a) = cos [66 + (6 — a)]

Using these results, we obtain after further manipulation

[(/,//1) - (/7//1)] sin 60

cos 6 0
1 + [(J5//1 + (/7//1)] cos 60,
or e _ „ = 90° + tan-1 [(WL) - (L//01 sin 60 (5.18)
o a uu q- ran \(t_/t\ _i_ /r /r m

Thus the presence of a sixth-harmonic error is shown. In modern high-

quality synchros this error may be made very small by so arranging the
pitch and distribution factors of the stator winding that the fifth and
seventh harmonics tend to be canceled. For details of this procedure the
reader is referred to standard texts on a-c machinery.1
It should be noted that the developments just carried out are intended
primarily to furnish qualitative insight into the possible causes of synchro
error, and the results obtained should be regarded as approximate. For
more accurate consideration of these errors and for analyses leading to
improved synchro design methods, the interested reader is referred to
papers by Chestnut,2 Kronacher,3 and Rosenbloom, Weiss, and Fried4 in
which the causes of synchro errors are treated in considerable detail.
The additional space harmonics apparently present in the error curve
shown in Fig. 5.7 are due primarily to winding irregularities and slot
effects. The maximum error is seen to be less than 0.2°, and in some
synchros designed for particularly high accuracy, errors have been reduced
to less than 0.1°.
1 Puchstein, Lloyd, and Conrad, “Alternating Current Machines,” John Wiley &
Sons, Inc., New York, 1954, pp. 133-141.
2 H. Chestnut, Electrical Accuracy of Selsyn Generator-Control-transformer Sys¬
tem, Trans. AIEE, vol. 65, pp. 570-576, 1946.
3 G. Kronacher, Static Accuracy Performance of the Selsyn Generator-Control-
transformer System, Trans. AIEE, vol. 69, pp. 645-653, 1950.
4 J. H. Rosenbloom, G. H. Weiss, and B. D. Fried, “Linear Lumped Parameter
Analysis of Synchros,” U.S. Naval Ordnance Laboratory, White Oak, Md. This
study is in nine parts, NAVORD Reports 1710, 2260, 2172, 2173, 2570, 2346, 2829,
2569, 3633, and is summarized in a final volume, “A Handbook for Synchro Systems,”
by G. H. Weiss and G. L. Beyer, Jr., NAVORD Rept. 3600, December, 1953.
230 CONTROL SYSTEM COMPONENTS [Chap. 5

5.4. Velocity Errors. In addition to the static errors discussed in the

previous section, synchros are subject to an error that is due to rotational
velocity. Thus, if the rotors of a generator and control transformer are
aligned to produce essentially zero output and if the rotors are then rigidly
coupled together, so that there can be no relative motion between them,
and if then the two synchros are rotated together, a voltage will appear at
the control-transformer terminals. This voltage will consist in part of
components due to the static errors discussed above, but there is also a
steady component that increases with speed. Hence if the two synchros
are used in a servo system that tends to null the inphase component of
output voltage, the output tends to run slightly behind the input when a
constant-velocity input signal is applied. This velocity error is in
addition to the error normally found in servo systems having only one
integration.
Qualitatively speaking, it is not unreasonable that such an error should
exist, for whenever one electric circuit moves with respect to another,
speed voltages are induced. This effect is discussed in some detail in
Sec. 4.4. A quantitative analysis of the phenomenon in synchros
requires, however, a considerable background of advanced synchronous-
motor theory, and even an approximate treatment of it is beyond the
scope of this text. We present here, therefore, only the results of some
of the quantitative studies that have been made.1’2 The theoretical
results given here are derived on the basis that the static errors and null
voltages considered in the previous sections are zero. The further impor¬
tant assumption is made that the impedance of the control transformer is
very much higher than that of the generator. Under these conditions the
magnitude of the control-transformer output voltage is given by

1 + Q2
2 {[y cos a — (1 — v2)Q sin a]2 + sin2o-j
[1 - (1 - v2)Q2]2 + 4Q
(5.19)

and the phase angle of the output voltage relative to /30, the phase angle
at standstill, is given by

1 + (1 + v2)Q2\ 1
(3 — /30 = tan-1 + tan-
1 + (1 - v2)Q2J KV cot a — (1 — v2)Q,
(5.20)

where ke — synchro gain constant, equal to 0.707Eom if E0 represents rms

value of output voltage

1 G. H. Weiss, “Linear Lumped Parameter Analysis of Synchros, IX, Effects of

Angular Velocity on a Simple Control System,” NAVORD Rept. 3633, U.S. Naval
Ordnance Laboratory, White Oak, Md., Feb. 8, 1954.
2 Chestnut, op. cit'.
Sec. 5.4] synchros and related devices 231

v = ratio of rotational velocity, radians/sec, to excitation fre¬

quency, radians/sec
Q = ratio of reactance to resistance in stator of control transformer
a = angular displacement between the two rotor positions, i.e.,
a = 6 — 5 = 6 — a — 90°
We note that the rotational velocity of the two synchros enters the two
expressions only in the form of v, the ratio between the velocity and the
frequency. Hence it appears that the velocity error for a given speed
may be reduced by operating the synchros at a higher frequency. The
fact that the phase angle of the output voltage changes with velocity
shows that both the inphase and quadrature components of the output are
functions of velocity. However, for low velocities such that 1 ± v2 « 1,

-90 -60 -30 0 30 60 90

cr, degrees

Fig. 5.8. Output voltage as a function of rotor displacement and velocity.

it may be seen that the phase shift is practically independent of velocity,

and hence most of the output voltage consists of the inphase component.
The characteristics of Eqs. (5.19) and (5.20) are best displayed graphi¬
cally,1 as in Figs. 5.8 and 5.9. Q is taken equal to 3 and Eom to 57 volts.
It is noted that for zero velocity the output voltage comes to a sharp null,
and the phase angle changes abruptly from 0 to 180° at the point where
the rotors are 90° apart, i.e., where cr = 0. As the velocity increases, the
output voltage no longer goes to zero for any value of a but only reaches
a minimum, and this minimum shifts to positive values of a. Of greater
significance in a servo system using a discriminator that rejects the out-of¬
phase component of the output signal is the fact that the point at which
the phase shift is 90° also shifts to positive values of a with increased
velocity. At this point the inphase component is zero, and hence the
servo will be in error by at least the amount of this shift.
1 Weiss, op. cit.
232 CONTROL SYSTEM COMPONENTS [Chap. 5

Comparison of Figs. 5.8 and 5.9 indicates that the point of minimum
output voltage coincides quite closely with the point at which the phase
shift is 90°, at least for relatively low velocity. Hence the error intro¬
duced into the servo system by the velocity effect is approximately equal

-90 -60 -30 0 30 60 90

a, degrees
Fig. 5.9. Phase of output as a function of rotor displacement and velocity.

to the value of a for which the output is minimum. This value, desig¬
nated as crm, may be obtained by differentiating Eq. (5.19) with respect to
(t and setting the derivative equal to zero. The result is given by

2 vQ
tan 2dm (5.21)
1 + (1 - v2)Q2

and is plotted in Fig. 5.10 for various values of Q. For the relatively low
velocities usually encountered, 1 — v2 « 1, and the error is small enough
so that tan 2am « 2crTO. Hence for low velocities Eq. (5.21) may be
reduced to
_ 57Q
degrees (5.22)
v 1 + Q2

Thus in a 60-cps synchro system having Q = 4, the error is about l°/300

rpm. Note that, by the definition of a, a = 6 — 8. A positive a means
that 6 is greater than 8; hence for positive velocity the output shaft lags
behind the input shaft.
The velocity error is, of course, present also whenever the synchro sys¬
tem is subjected tb dynamic inputs of any sort and might be expected to
Sec. 5.4] SYNCHROS AND RELATED DEVICES 233

affect the synchro transfer function at high input frequencies. This

transfer function cannot be obtained directly from Eqs. (5.19) and (5.20),
since in deriving these equations steady-state conditions were assumed.
In fact, a mathematical description of the complete dynamic situation,
where both 6 and a are allowed to vary in an arbitrary fashion, results in
a set of differential equations with variable coefficients, and a simple
transfer function in the usual sense does not, therefore, exist. A number
of relatively simple results may, however, be obtained if the velocity of
the generator rotor is assumed to be constant. Although no derivations

0 0.1 0.2 0.3 0.4 0.5. 0.6 0.7 0.8 0.9 1.0
v — @/(jj
Fig. 5.10. Variation of <rm as a function of velocity and Q [see Eq. (5.21)].

will be given here, it is possible to show1 that under these conditions the
output voltage is given by

En =
UQ - j) - [cos <7 + jQ(l — v2) cos <r + jv sin a]
1 - (1 - V2)Q2 + 2jQ CO

+ v cos <j — Q(1 — v2) sin a + j sin od (5.23)

where co is the line frequency in radians per second, the other symbols are
as defined in connection with Eqs. (5.19) and (5.20), and a is the difference
between the velocities of the generator and control transformer. If a is
1 Ibid. Equation (5.23) is not given in Weiss’s work but may be derived from the
results given there.
234 CONTROL SYSTEM COMPONENTS [Chap. 5

zero, Eq. (5.23) can be reduced to a form equivalent to Eqs. (5.19) and
(5.20). If p = 0, i.e., if the generator rotor does not rotate, Eq. (5.23)
reduces to the simple expression

E0 = ke[ sin a — j— cos a (5.24)

co

For small values of a this may be written approximately as

(5.25)

It is clear that, although the velocity of the control-transformer rotor does

contribute a component to the output, this component is 90° out of phase
with the dominant term and would, therefore, be rejected by a properly
adjusted phase discriminator following the control transformer. Hence,
when the generator is not rotating, the transfer function relating discrimi¬
nator output voltage to control-transformer rotor position is independent
of frequency.
In general, the inphase component of the control-transformer output
voltage does increase somewhat as the frequency of oscillation of either
of the two rotors is increased. Equation (5.23) indicates, however, that
the magnitude of the term contributed by the velocity becomes com¬
parable to the magnitude of the term due to position only for velocities
approaching synchronous velocity. Hence we infer that any time con¬
stants contributed by the velocity error to the synchro transfer function
will be of the order of one period of the line frequency and will therefore
be negligible.
An additional result deducible from Eqs. (5.23) and (5.24) is that, when
a is varied sinusoidally at relatively high frequencies, the output voltage
from the control transformer will
not have the idealized form shown
in Fig. 5.5. The sharp null shown
in that figure at points where a = 0
will disappear, since d is a maximum
at these points. Hence the output
Fig. 5.11. Control-transformer output voltage tends to be as shown in Fig.
voltage for high-frequency signals. _ 11
. .
0 11

It is evident from Fig. 5.8 that the separation between maximum and
minimum values of output voltage decreases with velocity. This may
be interpreted as a reduction in gain dEa/da of the synchro system with
velocity. Since we are again primarily concerned with the inphase com¬
ponent of the output, it is instructive to compute the ratio of a small
change of inphas^ output voltage as a function of a small change in a.
Sec. 5.5] synchros and related devices 235

This result can be derived by use of Eqs. (5.19) and (5.20), and a simple
expression results if this is evaluated for a = 0:

d[Re(E0)] QKi - v2)2 + 2Q2 + l

(5.26)
da Q4(l - v2)2 + 2Q2(1 + V2) + 1

This expression is plotted in Fig. 5.12 for Q = 3; note that the change in
gain is not very great until the
velocity becomes greater than one-
third of synchronous speed.
5.5. Two-speed Systems. The
static accuracy of servomechanisms
using synchros can be made much
higher than the static accuracy of
the synchros themselves by the
simple expedient of gearing up the Fig. 5.12. Variation of gain as a function
rotor of the control transformer of speed.
relative to the actual output. The
generator must, of course, be geared up in the same amount relative to the
true input. Then if the static synchro error is e, and if both synchros
make N revolutions for each revolution of the output or input, the error
at the output will be reduced to e/N. This assumes, of course, that the
gears do not introduce an additional error; but if precision gears are used,
a considerable improvement in accuracy is possible. Commonly used
values of N are 36, 31, and 25.
Unfortunately, a system in which the synchros are geared up by a factor
of N will have N — 1 false points of synchronization. Thus, for instance,
assume N = 36. Then for every 10° rotation of the output the control
transformer rotates through 360°. Hence, if the system is in equilibrium
for a given value of the input, it will also be in equilibrium when the out¬
put is displaced from its correct value by 10°, 20°, etc. This difficulty
may be overcome by using a second set of synchros arranged to make one
revolution for each revolution of the output or input. These synchros
are referred to as the one-speed synchros and serve to bring the servo
into approximate alignment. When the error in the one-speed system is
sufficiently small, the one-speed system may be switched off and the
N-speed system allowed to take over to complete the final accurate align¬
ment of the over-all system. A system of this sort is shown schematically
in Fig. 5.13. For the system to work properly it is, of course, necessary
for the stators of the two pairs of synchros to be clamped into their sup¬
porting frames in such a way that the two synchros tend to line up the
output to exactly the same point.
It should be noted that, in all cases where a two-speed synchro trans¬
mission is employed, the one-speed transmission is used only for approxi-
236 CONTROL SYSTEM COMPONENTS [Chap. 5
mate synchronization and is not operative during normal operation of the
servo. Hence the performance characteristics of a servo considered in the
design are always those observed with the high-speed system in control.
A number of circuits are available to perform the required switching
between the high-speed and one-speed systems. One simple circuit
utilizing neon tubes as switches is shown in Fig. 5.14. Its operation

N-speed synchros

Fig. 5.13. Servo using two-speed synchro transmission.

depends on the fact that the amplifier has an infinite input impedance and
may be explained as follows: The signal coming from the one-speed con¬
trol transformer is amplified in Vi and V2 and is applied through the trans¬
former T2 to the neon tubes. If the signal is large enough, the neon tubes
fire and provide a low-impedance path for the one-speed signal to the
amplifier input. At the same time the signal from the high-speed control

Fig. 5.14. Neon-tube switching circuit for two-speed synchro.

transformer is short-circuited by the neon tubes, since the resistors R have

a very much higher resistance than the circuit containing the neon tubes.
As the servo approaches synchronism, the signal from the one-speed con¬
trol transformer decreases to such a value that the neon tubes go out.
They now offer essentially infinite resistance and therefore disconnect the
one-speed signal from the amplifier. The high-speed signal, however,
now passes through the amplifier, since the resistors R are very much
Sec. 5.5] SYNCHROS AND RELATED DEVICES 237

smaller than the input impedance of the amplifier and the resistance of
the neon-tube circuit. The turns ratio of the high-speed input trans¬
former must be so chosen that the high-speed signal is never sufficiently
large to fire the neon tubes.
For a given small rotation of the output away from synchronism, the
output voltage of the iV-speed synchro is N times as high as that of the
one-speed synchro. Therefore, the loop gain with the high-speed channel
operating is inherently N times as large as with the one-speed system
switched on. An advantage of the circuit described here is that the extra
amplification included in the one-speed channel permits the loop gain to
remain approximately constant as the system is switched from one chan¬
nel to the other. This contributes to smooth operation.
R
Amplitude of

Error

(o) Switching circuit using clipper (b) Wave shape of amplifier input signal

Fig. 5.15. A simple switching circuit for an JV-speed synchro.

A switching method not having this advantage but employing a simpler

circuit is shown in Fig. 5.15a. The resistor and the rectifiers and bat¬
teries constitute a simple clipping circuit, which prevents the signal from
the high-speed channel from exceeding a certain level. The one-speed
signal is then simply added to the clipped high-speed signal. The form
of the composite signal is shown in Fig. 5.155; note that, if the clipping is
sufficiently great, it permits the voltage to go through zero only once in
the neighborhood of synchronization. As a result of the sudden change
in gain produced by the clipper, a servo using this circuit tends to have
relatively poor synchronizing characteristics when slewing, i.e., when
large and rapid changes of the input are made. Under these conditions
the motor may be running at such a high speed as it approaches corre¬
spondence that it passes entirely through the region of high-speed control
on the first approach, and possibly on several succeeding ones. The sys¬
tem behaves, therefore, as though it were poorly damped, even though
the response to small disturbances, which do not require operation beyond
the high-speed control zone, is quite well damped. This type of perform¬
ance is typical of systems in which some component in the loop saturates
for relatively small errors, i.e., where the zone of linear operation is small.
238 CONTROL SYSTEM COMPONENTS [Chap. 5

Hence there is no objection to using the circuit of Fig. 5.15a if amplifier

saturation takes place for smaller error values than those for which
clipping occurs.
The two circuits described here are typical but do not exhaust the list
by any means. The reader is referred to the literature for a number of
other circuits of this sort.1
An important problem in the design of switching circuits is the determi¬
nation of the point at which switching is to occur. A criterion for proper
operation of the switching circuit may be developed by reference to Fig.
5.16, which shows the variation of error voltage as a function of servo
displacement error for the two synchros. Suppose first that the synchros
have no static error. This is indicated by the heavy lines in Fig. 5.16.
Switching from the one-speed to the .A-speed channel must occur for
error values on the one-speed synchro between e\ and — ei, for if switching
voltage
amplitude

occurs outside this zone, the A-speed synchro will cause the system to
move away from the correct correspondence point at 0 to a false point
such as A. The switching circuit normally has hysteresis; i.e., it takes a
larger voltage to switch the system to the one-speed channel than it does
to switch to the A-speed channel. If the hysteresis is so great that it
takes a voltage greater than e2 to reconnect the one-speed channel, then
the system will simply lock in at the false point A. More probably, how¬
ever, the system switches back to the one-speed channel at a voltage less
than e2, so that the servo oscillates between the voltages e\ and e2.
Neither mode of operation is, of course, desirable.
If es is the voltage output of the one-speed control transformer at the
point at which the switch connects the system to the A-speed channel,
we see from Fig. 5.16 that for proper operation es < ke sin (tt/N) , or approx¬
imately es < keTr/A, where ke is the synchro constant. If this inequality
1 For instance, Ahrendt, “ Servomechanism Practice,” McGraw-Hill Book Com¬
pany, Inc., New York, 1954, pp. 59-64; James, Nichols, and Phillips, “Theory of
Servomechanisms,” Radiation Laboratory Series, vol. 25, McGraw-Hill Book Com¬
pany, Inc., New York, 1947, p. 84.
Sec. 5.5] SYNCHROS AND RELATED DEVICES 239

is only barely satisfied, then, although the system will lock in at the right
place, it may hesitate slightly immediately after the switch has operated,
since the voltage from the V-speed synchro is very small. Hence in prac¬
tice es should be about )^/ce7r /N.
Actually, errors exist in both synchros; in fact it is the existence of
these errors that makes the extra complication of the two-speed system
necessary. A possible range of one-speed and V-speed error is shown as
the dotted lines in Fig. 5.16. Suppose we let ei and be the maximum
error of the one-speed and V-speed synchro systems, respectively. For
identical synchros, ei will be approximately equal to en- Then, by inspec¬
tion of Fig. 5.16 and by use of the sort of reasoning used above, we find
that for proper operation

es < ke (T LI
\N ~ N “
6N

V
(5.27)

or, if ei = €n,

e,<^[^-(N + Die,!] (5.28)

The absolute-value bars are used because ei and may be either positive
or negative. As for the case of zero synchro error, it is desirable to do
more than to barely satisfy this inequality.
Since only the magnitude of es, not its phase, is usually used to actuate
the switching circuit, es cannot be negative. Hence inequality (5.28)
gives us an upper limit on the gear ratio V that may be employed between
the true output and the V-speed selsyn. Specifically

N + 1 < A (5.29)
Nil
Thus for an error of, say, 0.5° (this is rather larger than usually observed,
but serves to illustrate the principle), V would have to be less than 360.
Actually V seldom exceeds 100 in practice, since the switch cannot be
expected to operate always at exactly the same voltage and a safety factor
is required. Furthermore es cannot actually be zero but must have a
finite value. Very large values of V are not justified if the gearing error
becomes comparable to the synchro error or if other components of the
servo limit the maximum accuracy obtainable. Another factor limiting
maximum gear ratios is the fact that the moment of inertia of the control-
transformer rotor is reflected by the square of the gear ratio to the motor
shaft (see Chap. 8), and with small motors and large ratios this may result
in a considerable increase of the effective moment of inertia seen by the
motor. Finally, the top design speed of the synchros may limit the maxi¬
mum gear ratio that should be employed.
One further problem of two-speed synchro operation must be discussed
240 CONTROL SYSTEM COMPONENTS [Chap. 5

here: the problem of the 180° ambiguity found in two-speed systems hav¬
ing an even gear ratio. In Fig. 5.17 is shown the variation of output
voltage of the two synchros near 0° and 180° for systems having odd and
even gear ratios. Considering the odd gear ratio first, we see that the
slope of the curves at 180° for both synchros is negative; hence 180° is not
a point of stable equilibrium for the servo system. If the servo happened
to be energized, with the error exactly 180°, any small disturbance would
cause the A-speed voltage to increase and make the system drive away

la) Output voltage with lb) Output voltage with

odd-gear ratio even -gear ratio

Fig. 5.17. Variation of output voltage on two-speed synchros.

from the 180° position. As the voltage increases, the one-speed system
takes over and brings the servo to the proper position at 0° error. For
the even ratio, however, the slope of the A-speed curve is positive both at
0° and at 180°. Therefore, since at 180° the one-speed system is switched
off, the 180° point is a point of stable equilibrium at which the servo may
come to rest.
It is often desirable to use even gear ratios, particularly the ratio 36,’
since this permits dials calibrated for 360° per revolution and 10° per
revolution to be attached directly to the one-speed and 36-speed synchro
shafts, respectively. To overcome the difficulty discussed above, a
Total one-speed
One-speed voltoge with
stick-off voltage signal after
stator shift

(a) Addition of stick-off lb) Stator of one-speed

voltage Selsyn shifted
Fig. 5.18. Elimination of false zero at 180° in even-gear-ratio synchro.

constant voltage, the so-called stick-off voltage, is added to the output

of the one-speed generator. The result of adding this voltage is shown
in Fig. 5.18a. The dotted line shows the resultant of adding the stick-off
voltage to the one-speed selsyn output. Note that the first effect is that
the one-speed and A-speed signals no longer go to zero together at either
0° or 180°. It is necessary that the two signals both be zero when the
error is zero, and this may be accomplished by resetting the zero of the
one-speed synchro, i.e., by so shifting its stator that the total one-speed
Sec. 5.6] SYNCHROS AND RELATED DEVICES 241

voltage goes to zero at the same point as the iV-speed voltage. The effect
is shown in Fig. 5.186. Note that with the proper amount of stick-off
voltage the signal of both synchros goes to zero near 180° with a negative
slope.
It should be noted that the voltage waves of Fig. 5.18 represent the
amplitude of a-c signals, with negative amplitude referring to a signal
180° out of phase with one having positive amplitude. Hence the con¬
stant stick-off voltage must be a constant a-c voltage, in phase with the
synchro output. This voltage is easily added to the one-speed synchro
output by means of a small transformer in a circuit, such as the one shown
in Fig. 5.19.

A-c
supply

One-speed synchro One -speed

generator QT

Fig. 5.19. Method of adding stick-off voltage.

5.6. The Differential Generator. When it is necessary to have a servo

respond to two or more different shaft-position inputs at various remote
locations, differential synchro generators provide a convenient means for
adding the additional inputs. Differential synchros are very similar in
construction features to synchro generators or control transformers. The
primary difference is that the rotor of a differential unit is wound for three
phases rather than for single phase, and three slip rings and brushes are
required. The rotor is of the slotted-drum type constructed of punched
laminations of the sort shown in Fig. 5.3c. The windings of both rotor
and stator are of the standard, distributed type. Electrically a differ¬
ential generator, therefore, resembles most closely a wound-rotor induc¬
tion motor. Despite its constructional similarity to a three-phase
machine, it operates on single phase, like the other synchro devices dis¬
cussed thus far.
A typical circuit showing a differential synchro is shown schematically
in Fig. 5.20. Note that the differential synchro is inserted between the
generator and control transformer. Its operation is best explained by
considering it as a variable mutual inductance inserted between the
generator and control transformer. For simplicity, we assume the
synchro generator to be perfect, free from errors and unbalances, so that
Eqs. (5.13) become

Ia — Im cos (6 + 60°)
(5.30)
h = Im cos (6 — 60°)
9d9 CONTROL SYSTEM COMPONENTS [Chap. 5

where Im is the maximum value of Ia and Ib. The discussion is also sim¬
plified by assuming that the differential generator has unity turns ratio
(this is standard practice) and that the coefficient of coupling between
stator and rotor circuits is unity. Under these conditions, standard
transformer theory shows1 that the ampere-turns acting along any axis
in the differential-generator stator must be equal and opposite to the
ampere-turns acting along the same axis in the rotor. It is convenient
here to choose the axis passing through the leg 0'2' of the rotor (see Fig.
5.20). The ampere-turns contributed by the rotor along this axis are

NI'a + N(I'a + I'b) cos 60° + NI'b cos 120° (5.31)

where N is the effective number of turns on each winding. In writing

this expression, we assume the winding to be so distributed that the
ampere-turns measured in any axis are proportional to the cosine of the

Control
transformer

Fig. 5.20. Synchro system with differential generator.

angle between the axis under consideration and the axis of the winding
contributing the mmf. Thus, the angle between the reference axis
and the leg l'O', carrying current I'a + I'b, is 60°, and the contribution
to the mmf along the reference axis must be as given by the second term
in (5.31). The angles between the three stator windings and the axis
0'2' are (60° — </>), ( — 0), and (120° — 0) for the legs 10, 02, and 03,
respectively. The rotor angle 0 is measured as shown in Fig. 5.20. Note
that the direction in which the current is assumed to flow in the winding
is important. The stator ampere-turns contributed to axis 0'2' are

N(Ia + Ib) cos (60° — 0) + NIa cos 0 + NIb cos (120° — 0) (5.32)

Equating expressions (5.31) and (5.32) yields, after some reduction,

%I’a — [Ia cos (30° — 0) + Ib cos (90° — 0)] (5.33)

If now we insert expressions (5.30) for Ia and Ib, there results

I'a = Im cos (6 — 0 + 60°) (5.34)

1 See, for instance, Langsdorf, “Theory of Alternating-current Machinery,” 2d ed.,

McGraw-Hill Book Company, Inc., New York, 1955, p. 10.
Sec. 5.7] synchros and related devices 243

In a similar fashion, and by using leg 0'3' as the reference axis, we find

I'b = Im cos (6 — <j) — 60°) (5.35)

We see that under these simplifying assumptions the form of /' and I'b is
the same as that of Ia and Ib, except that the angle 6 — <j> has been sub¬
stituted for the angle 6. Hence, the effect of the differential generator is
to change the control-transformer angle for zero output voltage to
8 = 6 — <f> rather than 8 = 6, the value
obtained when the differential generator
is not present. Thus the shaft angle of 9
the differential generator is added (or
<p 8 Servo
subtracted) from the angle of the input.
This may be represented schematically ^IG' °’21' An eci^RaJfnt^ block
. diagram showing the effect ot the
as m Fig. 5.21. It is, of course, possible dilierential generator.
to use more than one differential genera¬
tor when more than one additional input must be inserted into the system.
In practice, differential generators are subject to the same sort of errors
as other synchros.1 Also, the coefficient of coupling between stator and
rotor circuits is less than unity, and therefore the differential generator
requires a fairly large magnetizing current, which must be supplied by
the synchro generator. This current produces heating in the generator,
and it places a limit on the number of differential generators that can be
excited by a synchro generator of a given size. The magnetizing current
may be reduced quite considerably, at least as far as the synchro generator
is concerned, by the use of a synchro capacitor. This is a set of three
capacitors, usually connected in delta, and “potted” in a single can.
They are ordinarily connected across the three lines connecting the
synchro generator and differential generator. Synchro capacitors are also
sometimes used in simple generator-control-transformer systems, particu¬
larly when one generator supplies several control transformers. Since
the impedance of the control transformer is usually quite high, a smaller
capacitor is required in this application.
5.7. Synchro Repeaters. A synchro repeater (also called synchro
motor) is externally and in electrical characteristics identical to a syn¬
chro generator. The chief constructional difference is that the rotor of a
synchro repeater usually has a vibration damper (see Chap. 8) on it. In
typical systems the repeater is connected to the generator as shown in
Fig. 5.22. The operation may be explained qualitatively by thinking of
the generator as transmitting a magnetic-field pattern to the stator of the

1 G. H. Weiss and J. H. Rosenbloom, “Linear Lumped Parameter Analysis of

Synchros, VIII, Simple Differential Control Systems,” NAVORD Rept. 2569, U.S.
Naval Ordnance Laboratory, White Oak, Md., December, 1953.
244 CONTROL SYSTEM COMPONENTS [Chap. 5

repeater in a way similar to that explained in connection with the gener¬

ator-control-transformer system (Sec. 5.2). Since the rotor of the
repeater is also excited, it produces another magnetic field, alternating in
step with the field established in the stator. Owing to magnetic attrac¬
tion, the rotor of the repeater will experience a torque tending to line up
these two fields, and equilibrium for the system is reached only when the
two rotors are parallel. Since the repeater is electrically identical with
the generator, the generator rotor experiences the same torque as the
rotor of the repeater. The system acts, therefore, as a direct torque
transmission. The torque is actually a sinusoidal function of the angular
difference between the two rotor positions, but for the small angular
differences that are usually of interest the torque may be assumed to be
proportional to the angular difference. The
LB LB two synchros act, therefore, as though they
A-c o- were coupled together mechanically by a
supply t GEa3=n
Generator Repeater
spring. The spring constant of this spring
may be referred to as the torque gradient
Fig. 5.22 A generator-repeater
of the system and may be computed as a
system. function of the impedances and other pa¬
rameters of the two machines.1
Repeaters are commonly used for remote indication of position, in
which case their load is only a pointer or a dial. Their accuracy in this
application is improved by mounting the rotor on ball bearings and by
reducing all sources of friction to an absolute minimum. Since the elec¬
trical damping is also inherently very small, a synchro repeater would
oscillate for many cycles as a result of any disturbance if it were not
equipped with the vibration damper already referred to. Since a machine
designed as a repeater is, however, otherwise identical with a generator, it
may be used in place of the generator if the presence of the vibration
damper is not objectionable.
The generator-repeater combination shown in Fig. 5.22 is only one
example of a synchro torque system. Other torque systems are occasion¬
ally used. In one of these, two synchro generators and a differential
generator (equipped with a vibration damper) are employed. One of
the generators feeds the stator and the other the rotor of the differential
unit, and the effect is that the rotor of the differential generator tends to
take on a position that is equal to the difference between the two synchro¬
generator rotor positions. The system, therefore, acts like a mechanical
differential gear, but since such a system is not very stiff, its characteristics
tend to be inferior to those of a mechanical differential.
1 B. D. Fried and J. H. Rosenbloom, “Linear Lumped Parameter Analysis of
Synchros, VII, Simple Torque Systems,” NAVORD Rept. 2829, U.S. Naval
Ordnance Laboratory, White Oak, Md., July, 1953.
 5.8] SYNCHROS AND RELATED DEVICES 245

5.8. Variable-reluctance Transducers. All synchros suffer from the

disadvantage that an electric connection must be made to the rotor and
that brushes and slip rings are required. This disadvantage does not
exist in variable-reluctance transducers, and it is therefore possible to
reduce friction in them to very low values. On the other hand, variable-
reluctance transducers usually can be used only over a limited range of
the input position. One of the simplest examples of this type of device
is the E pick-off, shown schematically in Fig. 5.23. In its most common
form it consists of an E-shaped
Armature
stator made by stacking a number
of E-shaped iron laminations. A
coil is wound on each one of the
Stator
three legs, and the two outer coils
are connected in opposition. The t
center coil is normally excited by
ft -o A-c
~° supply
a constant a-c voltage. When the
Fig. 5.23. An E pick-off.
movable armature, also made of
iron laminations, is symmetrically located with respect to the two
outer legs, the mutual inductance between each of the outer legs
and the central leg is the same, and since the outer coils are wound in
opposition, no output voltage results. This position of the armature is
referred to as the null position. If the armature is displaced to the
right from the null position, the voltage induced in the right-hand coil is
larger than that induced in the left-hand coil; hence an output voltage is
produced that is ideally in phase with the applied voltage. Motion of
the armature to the left produces an output that is 180° out of phase with
the voltage produced by motion to the right. Within limits the output
voltage is approximately proportional to the displacement; hence we have
a device that modulates the a-c supply voltage in accordance with the
displacement of the armature. The modulation provided is of the sup-
pressed-carrier type. If the armature position is made to oscillate
sinusoidally about the null position, an output-voltage waveform of the
sort shown in Fig. 5.5 results.
In practice, E pick-offs depart in a number of ways from the ideal
behavior described above. For example, they usually produce a fairly
large quadrature voltage at the point where the output voltage is theo¬
retically zero. Qualitatively, the reasons for the existence of this voltage
are similar to those already discussed in relation to synchros (Sec. 5.3).
The quadrature voltage can often be reduced markedly by connecting a
small variable capacitance between one of the leads to the center coil and
one of the output leads (see Fig. 5.24). The purpose of the isolating
transformer shown in Fig. 5.24 is to permit connecting the capacitor to
either side of the input coil. Occasionally, no adjustment of the capaci-
246 CONTROL SYSTEM COMPONENTS [CHAP. 5

tor completely removes the null voltage. In this case a high-resistance

potentiometer may be connected across the center coil, with the tap going
to the ungrounded lead of the output. This provides a second adjustment,
and by adjusting both the potentiometer and the capacitor simultane¬
ously, the voltage existing at the null position can be reduced to a residue
consisting only of harmonics of the supply voltage. If these are still
objectionable (as they sometimes are in very high gain systems), the out¬
put signal may be passed through a bandpass filter adjusted to pass only
the fundamental frequency. E pick-offs are also subject to static and
velocity errors just as synchros are. However, since E pick-offs are
usually used singly, the type of error in synchros arising from the fact
that these units are used in pairs is of no consequence. Also, since E
pick-offs have only a limited range of displacements, the velocity error is
unimportant. The primary accuracy
consideration of importance with E
pick-offs is, therefore, linearity. Line¬
AJ-C
arity depends on the way in which the
supply reluctance changes between the center
O-

Isolating leg and the outer legs, and the most

transformer £ p!ckoff
linear system will result if the reluc¬
Fig. 5.24. Schematic of E pick-off tance decreases as much on one side as
showing capacitor connected to re¬
it increases on the other. This can be
duce quadrature voltage.
accomplished by making the length of
the armature such that at the null position it covers approximately
half of each of the outer legs. It is also important that the reluctance
of the iron, which is a nonlinear function of flux density, be negligibly
small compared to the air-gap reluctance; hence a relatively long air
gap is desirable if linearity is important. Greater sensitivity at the
expense of reduced linearity can, however, be obtained by reducing the
air gap to a minimum and by shortening the armature so that at the null
position it just barely covers the two outer legs (as in Fig. 5.23).
Although the E pick-off described here is designed to convert linear
motion into an electric signal, it can be used also for rotary motion by
shaping the moving armature and the E-shaped stator as shown in Fig.
5.25. It is clear, of course, that only a limited range of angular displace¬
ment can be handled by this pickup, but in many null-type systems,
where the stator and armature are moved directly by the input and out¬
put, respectively, and where relative displacement between stator and
armature represents the error of the feedback loop, this is no disadvantage.
A somewhat different form of variable-reluctance transducer that is
very closely related to the E pick-off is shown in cross section in Fig. 5.26.
Three coils are used, just as in the E pick-off, but they are wound on a
fiber or other nonmagnetic and nonconducting core. The two outer coils
Sec. 5.0] SYNCHROS AND RELATED DEVICES 247

are connected in opposition. A small rod, carrying a short iron section,

passes through the center of the coils as shown in the figure and performs
the same function as the armature in the E pick-off. Normally the cen¬
ter coil is connected to the supply and acts as the primary, and the signal
is taken from the two outer coils. The operation is exactly the same as
that shown for the E pick-off. This transducer can be kept very small in

■Primory

■■
winding
Output

Magnetic Non-magnetic \ ■<—>

■III
-Secondary-
windings
Fig. 5.25. A schematic of an E pick- Fig. 5.26. Schematic of a variable-
off designed for rotary motion. reluctance transformer.

size and is, therefore, often used in miniature equipment. It is commonly

used in accelerometers and pressure pickup devices as the element that
converts the minute displacements obtained in these devices into an
electric signal.
5.9. The Microsyn. Another device very closely related to the E pick-
off is the microsyn, shown schematically in Fig. 5.27. Both the stator
and rotor are made of iron laminations, the rotor being specially shaped to
vary the reluctance of the stator
magnetic circuit. The stator has
four poles, and each pole carries two
coils, a primary and a secondary
coil. All the primaries are con¬
nected in series and are so wound
on the poles that, for a particular
direction of current in the winding^
the flux vector is inward on poles 1
and 2 and outward on poles 3 and 4.
The secondaries are also all in series
but are connected in such a way that the voltages induced in the
windings on poles 1 and 3 oppose the voltages induced on poles
2 and 4. Hence, when the rotor is in the neutral position as shown
in the figure, the flux in all four poles is the same, and no output voltage
is produced. However, motion of the rotor away from neutral results in
a voltage output, which over a limited range of rotation (about 7°) can be
248 CONTROL SYSTEM COMPONENTS [CHAP. 5

made quite accurately proportional to the angle. The microsyn is used

normally only for limited angular rotations.
There are a number of other devices of the variable-reluctance type
such as the telegon, the magnesyn, and others. The reader is referred to
the literature1 for descriptions of these transducers.

PROBLEMS

5.1. The inductive reactance of each leg of the stator winding of a type 1 CT con¬
trol transformer is 1,000 ohms and that of the type 1 G generator used to excite it is
100 ohms. Nominally the ratio of resistance to reactance in both machines is 1:4,
but as a result of a bad connection in one stator lead, the resistance ratio in this one
leg is increased to 1:3.9. Find the maximum quadrature voltage from the control
transformer if the maximum inphase output voltage is 57 volts rms.
6.2. Determine the maximum second-harmonic error in a synchro transmission in
which the ratio of Iam/hm = 0.99. Assume that all other errors, including the quad¬
rature error, are zero.
6.3. Find the maximum value of sixth-harmonic error in a synchro transmission in
which the fifth space harmonic is 20 per cent of the fundamental and the seventh
10 per cent of the fundamental. Assume all other errors negligible.
5.4. A servo system using a two-speed synchro transmission is supposed to have
an effective velocity-error constant of at least 1,000. The ratio of inductive react¬
ance to resistance in the control transformers is 4. Assuming that the servo gain
constant cannot exceed 5,000, find the maximum gear ratio permissible between the
output shaft and the high-speed synchro.
5.5. A servo system uses a two-speed synchro transmission with a 36:1 ratio
between high- and low-speed synchros. The error in both synchros may be as large
as 0.3°; the synchro constant for both is 1 volt /deg. The switching circuit used to
switch the system between the high- and low-speed synchros has the form shown in
Fig. 5.14. The neon tubes fire when an rms voltage of 60 volts is applied to them;
they go out when the voltage is 40 volts. Assume that the transformers T\ and T%
have unity turns ratio; find the extreme values of the combined amplifier gain and
transformer T2 turns ratio that will permit reliable switching.
5.6. What is the value of the stick-off voltage required to prevent ambiguity in a
system using a two-speed synchro system with 36:1 gear ratio? The synchro con¬
stant for both synchros is 1 volt/deg. Neglect synchro error.

1 See particularly Greenwood, Holdam, and MacRae, “Electronic Instruments,”

Radiation Laboratory Series, vol. 21, McGraw-Hill Book Company, Inc., New York,
1948, pp. 362-370. Also Batcher and Moulic, “Electronic Control Handbook,”
Caldwell Clements, Inc., New York, 1947.
 CHAPTER 6

DEMODULATORS AND MODULATORS

6.1. Fundamental Concepts. The output signal delivered by a syn¬

chro control transformer or by any of the electromagnetic transducers
described in the previous chapter is in the form of a relatively high fre¬
quency carrier modulated by signal information that is normally confined
to a much lower frequency spectrum. Although there are some servo
power units, such as the a-c motors considered in the next chapter, that
can utilize this modulated signal directly, it is usually necessary to recover
the low-frequency envelope by use of a demodulator or discriminator.
Since the output signal delivered by most of the transducers used in
servomechanisms is of the suppressed-carrier, amplitude-modulated type,
demodulators used with these transducers must be phase-sensitive; i.e.,
they should put out a positive voltage when the input is in phase with a
given reference and a negative voltage when the input is 180° out of phase.
A simple circuit to accomplish this type of demodulation is shown in
Fig. 6.1. This circuit is essentially a push-pull amplifier of standard

design with plate load Rl and cathode-biasing resistor RK• The feature
that makes the circuit act as a phase-sensitive demodulator is the a-c
plate-supply voltage. This supply is referred to as the reference volt¬
age, and it must be of the same frequency as the input signal. As a
result of the a-c supply, the amplifier is cut off every other half cycle,
operating only during the half cycle in which the plate voltage is positive.
If the phase of the input signal is such that during the on period of the
amplifier the upper grid in Fig. 6.1 is positive and the lower one negative,
more current flows in the upper triode than in the lower, and hence the
249
250 CONTROL SYSTEM COMPONENTS [Chap. 6

plate of the upper triode is more negative than that of the lower. If the
phase of the input is reversed, the polarity of the output reverses. Typi¬
cal output waveshapes obtained from the circuit for various phase rela¬
tions between input and reference are shown in Fig. 6.2. In drawing
these figures, it is assumed that the gain of the tubes is independent of the
magnitude of the plate voltage; this is a good assumption during most of
the time that the voltage is positive. Figure 6.2c shows the effect of an
input signal 90° out of phase with the reference; note that the average, or
d-c, output is zero. The discriminator is therefore seen to reject quad¬
rature signals. In general, if the input is less than 90° out of phase with
the reference, the d-c component of the output is reduced from the value
it would have if the signal and reference were exactly in phase. A typical
output waveform is shown in Fig. 6.2d.

Fig. 6.2. Demodulator output waveshapes.

In all cases shown, the output contains a large a-c ripple component.
This ripple is normally highly objectionable because it results in satura¬
tion, heating, and general malfunctioning of amplifiers and power devices
following the discriminator. Hence low-pass filters are usually used in
the discriminator output. For most applications these are simple RC
networks of the type described in Chap. 1.
The output waveforms shown in Fig. 6.2a and 6.2b are those obtained
from a standard half-wave rectifier. Hence the discriminator circuit
shown in Fig. 6.1 is referred to as a half-wave discriminator. Some of
its other features are that it provides gain between input and output and
that it delivers a push-pull output signal. If the two halves of the circuit
are exactly identical, zero input results in both zero d-c and zero ripple
output. In practice, however, perfect balance of the circuit is difficult to
achieve, and there is usually some output for zero input. Since the cir¬
cuit unbalance may vary as a result of temperature and supply-voltage
changes, the output produced by the unbalance may also vary; i.e., the
circuit may drifts This is not unreasonable since the circuit is in part a
Sec. 6.2] DEMODULATORS AND MODULATORS 251

d-c amplifier, but it should be noted that since it is a push-pull circuit,

the drift will be quite small.
6.2. Analysis of the Half-wave Discriminator.1 In analyzing the
operation of the discriminator circuit considered in the preceding para¬
graph, it is convenient to assume that the output voltage delivered during
the on period of the amplifier is equal only to the amplified input signal
and is independent of the plate voltage. While this is not quite true, it is
approximately correct during the major part of the on period if the plate
characteristics of the tube in question are fairly linear, that is, if rv and tx are
approximately independent of the operating point. Also, it will be found
that the exact form of the output signal will affect the analysis only in
detail, not in principle.
If we let the gain during the on period q- - -
be G, then under the above assumption
the operation of the circuit is simply
explained by considering it to multiply
the input signal by the square wave Fig „ 3 Approximate demodu.
shown in Fig. 6.3. This is a square wave lating function,
having an amplitude of G/2 plus a d-c
component of G/2. It may be expanded into the Fourier series:

fd(t) = % + — (sin cot + sin Scot + sin Scot + • • •) (6.1)

2 7r

where co is the reference or carrier frequency in radians per second.

The input signal applied to the discriminator is assumed to be sup-
pressed-carrier amplitude-modulated, and for the moment we assume that
the carrier is a sinusoidal function of time. If the modulating function is
F(t), then the input signal to the discriminator is given by

Vi(t) = F(t) sin (cot — (3) (6.2)

where co is the carrier frequency and (3 is the phase shift that might exist
between the carrier and the reference. The output of the discriminator is
the product of the input and the demodulating gain function of Eq. (6.1):

v0(t) = GF(t) sin (cot — (3) ^ + - ( sin cot + i sin Scot + \ sin Scot +
2 7T \ o 5

In expanding this expression, it is convenient to expand first the factor

sin (cot — (3) into sin cot cos (3 — cos cot sin |8. Multiplying this into the
series and simplifying results in

1 J. L. Bower, “ Discriminators and Modulators for Servomechanisms,” unpublished

report, 1945.
252 CONTROL SYSTEM COMPONENTS [Chap. 6

1 . 1 . , 2/1 0 , . 1 . ,
v0{t) = GF(t) cos /3 —\- - sm cot-l ^ cos 2cot + «—= cos 4oot
7T 2 7T \o 0-5

+ ^—y cos 6co£ + • •

— sin jS ^ cos w/ + - (l. sin 2co£ + sin 4co2 + sin 6oot + • • •) |

2 7t\3 3-5 5-7 /J J
(6.4)

The useful part of the output is provided by the first term in the first
bracket.

Useful vQ(t) = — (cos /3)F(t) (6.5)

7r

All other terms represent ripple. We note that the original modulation
F(t) has emerged from the discriminator multiplied only by the constant
factor (G/tt) cos (3, without any phase lags or amplitude changes depending
on frequency. We conclude from this that the processes of modulation
and demodulation do not inherently introduce any phase lag (or lead) into
the signal channel and that the phase lag normally found is entirely due
to the ripple filters. As mentioned previously, these are almost always
required to eliminate the objectionable effects of the ripple on the circuits
following the discriminator.
Equation (6.5) shows that the gain between input and useful output is
proportional to the cosine of the phase shift between the carrier and the
reference. The desirability of keeping this phase angle small if maximum
gain is to be secured is therefore evident. However it is also clear that
phase shifts of the order of less than 10° are not of any great consequence.
Phase shift also has some effect on the magnitude of the ripple. This can
be appreciated qualitatively by inspection of Eq. (6.4), since the second
bracket, which contributes a large ripple component, is proportional to
the sine of the phase angle and therefore vanishes if the carrier and refer¬
ence are in phase. A somewhat more quantitative idea of the effect of
phase angle on output ripple is obtained by computing the magnitude of
the harmonics from Eq. (6.4) as a function of |8. The results of such a
computation are given in Table 6.1 for the case of GF(t) — 1. The table

Table 6.1. The Effect of Phase Angle on the Magnitude of

Output Ripple

Harmonic Magnitude
0 0.318 cos j8
1 0.5
2 0.21 Vl + 3 sin2 /3
4 0.043 Vl +15 sin2 0
6 0.018 Vl + 35 sin2 /3
8 0.01 Vl +63 sin2 /3
Sec. 6.2] DEMODULATORS AND MODULATORS 253

shows clearly that the ripple voltage increases with /?, but since the largest
ripple-frequency component, the fundamental, is independent of /3, one
would not expect the rms value of the ripple to increase very much as /3
changes from 0 to 90°. It is instructive in this connection to compute the
rms value of the ripple for these two extreme values of (3. Using only the
harmonics listed in Table 6.1, we find approximately that for (3 = 0° the
rms value of the ripple is 0.544, while for (3 = 90° the ripple increases to
0.70. Thus it is clear that, although phase shift does increase the ripple
somewhat, the effect is not of any great importance in a half-wave
discriminator.
In order to find the maximum frequency of the input modulation F(t)
that can be handled without ambiguity by a half-wave discriminator, we
let
F(t) = V sin ojst (6.6)

Substitution of this expression into (6.4) results, after some simplifica¬

tion, in

v0{t) = GV cos jS - sin cost + x/\ cos (co — oos)t — Iq cos (co + cos)t —
7T

— sin (3 34 sin (<*> ~ Cx}s)t + 34 sin (w + cos)£ + • (6.7)

The term yielding the useful part of the output is still the first term of the
first bracket, GV (cos (3)(1/t) sin cost, with all else constituting ripple.
Thus the lowest harmonic of the ripple is shifted downward by the signal
frequency and becomes co — cos. It is clear that, when

CO — cos = cos
or cos = )q^co (6.8)

the signal frequency and the frequency of the lowest harmonic of the rip¬
ple are the same, and it is no longer possible to distinguish between signal
and ripple. Hence we have the important and quite universal result that
the absolute upper frequency limit of a half-wave discriminator is equal to
one-half the carrier frequency. Practical reasons such as the difficulty of
constructing a ripple filter with a very sharp cutoff characteristic usually
restrict the maximum permissible signal frequencies to values considerably
below this absolute limit. Thus a commonly observed rule of thumb
applied to the design of feedback systems is that the loop-gain crossover
frequency should be no more than one-tenth of the carrier frequency if a
half-wave discriminator is used.
Equation (6.7) contains additional information that is useful in the
design of ripple filters. Usually these are simple, low-pass RC networks
254 CONTROL SYSTEM COMPONENTS [Chap. 6

of the type described in Chap. 1. However it is sometimes found that a

simple filter of this sort designed to attenuate the ripple to a satisfactory
degree also results in excessive attenuation and phase shift in the signal
spectrum. In such cases, advantage may be taken of the fact that the
ripple is concentrated in frequency bands near the fundamental, second,
fourth, etc., harmonics of the carrier. This fact permits the use of a series
of sharply tuned band-rejection filters, such as the bridged-T or twin-T
circuits discussed in Chap. 1, which produce much less phase lag and
attenuation at lower frequencies than the simple low-pass filter does. In
fact, it can be shown that the phase shift produced by these filters can
be made to approach zero as the rejection band is reduced to zero width.1
It would appear, therefore, that this might be the ideal adjustment.
However, if more terms of the series indicated in Eq. (6.7) are written,
it will be found that the frequency components found in the ripple are not
simply the fundamental, second, fourth, etc., harmonics of the carrier,
but are actually co ± ws, 2co + cos, 4w ± cos, etc. Hence, if the input F(t)
is confined to a spectrum ranging between d-c and some maximum fre¬
quency cosm, the ripple will be found to be concentrated in bands of width
2ojsm centered at the fundamental, second harmonic, etc. The ripple
filters should attenuate all frequencies in this band, and hence they cannot
be of zero width. We conclude, therefore, that in practice there is a cer¬
tain minimum amount of phase lag contributed by a frequency discrimi¬
nator even if the most sophisticated type of ripple filter is employed.
According to Table 6.1 the fundamental and second-harmonic fre¬
quency components contribute the major share of the ripple. Hence, if
a band-rejection type of filter is used to minimize the low-frequency phase
lag, it is usually sufficient to attenuate only the fundamental and the
second harmonic by this type of network, with a simple low-pass structure
to attenuate all the higher harmonics. Since this low-pass filter must
attenuate only the fourth and higher harmonic components, its passband
may be relatively wide. Thus the phase shift contributed by this part of
the filter at signal frequencies can also be reduced to a minimum. In this
way it is possible to obtain good attenuation of all the ripple with rela¬
tively small effect on the signal band.
6.3. Full -wave-discriminator Circuits. The analysis of the half-wave
discriminator just concluded has shown that its upper frequency limit is
equal to one-half the carrier frequency and that the lowest harmonic of
the ripple frequency is the fundamental of the carrier. Intuitively, it
appears that the full-wave discriminator, for which there is an output
every half cycle of the carrier, should be able to improve on this perform¬
ance, and it will be shown presently that this is indeed the case. We con-
1 Valley and Wallman, “Vacuum Tube Amplifiers,” Radiation Laboratory Series,
vol. 18, McGraw-Hill Book Company, Inc., New York, 1948.
Sec. 6.3] DEMODULATORS AND MODULATORS 255

sider first a typical circuit, operating on the same principle as the one
shown in Fig. 6.1, and discuss its operation qualitatively.
The circuit of a triode full-wave discriminator is shown in Fig. 6.4. The
dots shown on the transformers are the conventional polarity markings;
hence, when the upper two triodes conduct, the lower two are disabled by
negative plate supply, and vice versa. Suppose now that the phase of the
input is such that the grid of the uppermost tube is positive when the
upper two triodes conduct. Then the output signal e0 is positive in the
direction shown by the arrow. During the next half cycle the lower two
triodes conduct, and the connection of the input transformer is such that
the lowermost grid is now negative. Hence the output signal is again

positive in the direction of the arrow. Thus for each cycle of the refer¬
ence we get two pulses of the output, or a full-wave signal. Had the
phase of the input relative to the reference been inverted, so that the
uppermost grid was negative during the conduction period of the upper
two tubes, we would have found the output to consist of negative pulses.
If the phase of the input were such that during the conducting period of
the upper triodes the uppermost grid were first positive and then negative,
the output would consist of a pulse first positive and then negative. By
this sort of reasoning one can obtain the output waveshapes shown in
Fig. 6.5. We note again that a signal 90° out of phase with the reference
will result in zero output, and if the phase shift between signal and refer¬
ence is somewhere between 0 and 90°, a reduced d-c output results.
In analyzing this circuit we make the same assumption we made in the
analysis of the half-wave discriminator, namely, that the gain of the cir¬
cuit is a constant during the conduction period of either set of triodes.
Also, as has already been noted, the operation of the circuit is such that,
256 CONTROL SYSTEM COMPONENTS [Chap. 6

Fig. 6.5. Output waveforms of full-wave discriminator.

if during one half cycle of the reference the output is positive for positive
input, then for the next half cycle the output is positive for negative input.
Hence, the circuit can be assumed to
multiply the input by the square
wave shown in Fig. 6.6; this differs
from the square wave considered in
connection with the half-wave dis¬
criminator primarily in that its d-c
level is zero. The magnitude is arbi¬
Fig. 6.6. Approximate demodulating
trarily taken as G, the gain of the am¬
function of full-wave discriminator.
plifier in one of the switching modes.
The discriminator gain function may be expanded in the Fourier series:

fd(t) = — (sin cd + % sin 3cot + V5 sin 5cot + • • •) (6.9)

7r

If the input to the discriminator is again

Vi(t) = F(t) sin (cot — 0) (6.2)

then the output given by the product of (6.2) and (6.9) may be expanded
into the form

2 4/1 0 . , 1
v0(t) = GF(t) cos (3 -I U COS 2bit + -—- COS 4co£
7T 7T \6 6 ‘ 5
1
+ -—= cos 6cot + • •

5 • 7

— (sin (3) - sin 2cd + sin 4cot + -=^-= sin 6cot + (6.10)
7r \3 3-5 5-7

The useful part *of the output is F(t) multiplied by the time-invariant part
Sec. 6.3] demodulators and modulators 257

of the series,

Useful v0(t) = - (r(cos (3)F(t) (6.11)

7T

with all other terms contributing only to the ripple. Here again it may be
seen that the input modulation F(t) emerges unchanged by the processes
of modulation and demodulation. Also the effect of (3 on the gain is
identical with that found already for the half-wave discriminator. In
fact, comparison of Eqs. (6.4) and (6.10) shows that, except for a factor
of 2, which can be absorbed into the gain factor G, the only difference
between the outputs of the full-wave and half-wave circuits is that in the
former, the fundamental frequency component of the ripple is absent.
Hence the table of harmonic magnitudes constructed for the half-wave
discriminator (Table 6.1) applies also to the full-wave discriminator if we
set GF(t) = Yl and let the first harmonic be zero.
The table may again be used to find the approximate rms value of the
ripple voltage and to estimate the effect of d on it. Using only the har¬
monics up to and including the eighth, we find that for (3 = 0° the rms
value of the ripple is 0.216, while for /? = 90° it goes up to 0.473. (Note
that these values are computed for GF{t) = y in order to effect a direct
comparison with the half-wave circuit.) It is clear that for d = 0 the
ratio of rms ripple voltage to useful output is less than half as large in the
full-wave as it is in the half-wave circuit. This is due to the absence of
the fundamental, which is responsible for the major share of ripple voltage
in the half-wave discriminator. It is also apparent that the effect of
phase shift between signal carrier and reference is more serious in increas¬
ing the ripple, and it is therefore somewhat more important here to keep
this phase angle low.
We note that the amount of filtering required to reduce the ripple out¬
put following a full-wave discriminator to a permissible amount is much
less than that needed to produce an equal attenuation of the ripple pro¬
duced by a half-wave circuit. This is due both to the smaller rms value
of the ripple and to the fact that the lowest ripple harmonic is the second
harmonic rather than the fundamental of the carrier frequency. Specifi¬
cally, let us assume that a double-section RC filter, producing an attenu¬
ation proportional to the square of the frequency, is employed as a filter.
Then, for the same reduction in ripple relative to d-c output, the filter
used with the full-wave circuit may have about three times the passband
(as defined by the reciprocal of the RC time constant of the two filter sec¬
tions) of the filter used with the half-wave circuit. The superiority of the
full-wave discriminator in causing reduced attenuation and phase lag of
desired signal information is, therefore, evident.
In order to find the maximum frequency of the input modulation F(t)
258 CONTROL SYSTEM COMPONENTS [CHAP. 6

that can be handled without ambiguity by a full-wave discriminator, we

again let
F(t) = V sin oost (6.12)
Then Eq. (6.10) becomes

2.2 2
v0(t) = GV cos (3 - sin oost + — sin (2co — aos)t — — sin (2oo + a>s)£ +
7T 07T 07T

4
— (sin (3) - cos (2co — C0S)£ — - COS (2oj + Ws)£ + • • • (6.13)
7T o o

As before, the first term of the series, GV (cos (3)(2/ir) sin oost, represents
the useful signal, with all other factors being ripple, and it can be seen
that the lowest harmonic of the ripple is 2co — oos. The upper limit on
cos is reached when this lowest ripple frequency coincides with the signal
frequency, i.e., when
oos = 2co — 00s

or oo s — oo (6.14)
Thus the absolute upper limit is equal to the carrier frequency, a limit
twice that of the half-wave discriminator. As was mentioned in connec¬
tion with the half-wave circuit, the actual useful upper limit is normally
considerably less than the theoretical maximum because of practical
difficulties of filter design. However, if these difficulties are assumed to
be identical in the two cases, the fact that the practical upper frequency
limit of the full-wave discriminator is twice as high as that of the half¬
wave discriminator still holds true.
It should be pointed out that the result concerning the maximum theo¬
retical frequency usable with a full-wave discriminator is not universally
true but depends on the fact that the input signal was assumed to have a
sinusoidal carrier. If the carrier were an accurate square wave, identical
with the one assumed for the discriminator demodulating function (Fig.
6.6), the processes of modulation and demodulation would have subjected
the signal to two successive multiplications by a square wave. This, at
least theoretically, would be equivalent to multiplication by a constant,
so that there would be no upper frequency limit and, incidentally, no
ripple in the output. While this result is primarily of theoretical interest,
owing to the difficulty of generating the required accurate square waves,
it does point out the desirability of square-wave modulation and demodu¬
lation in general.
As discussed in connection with the half-wave discriminator, it is possi¬
ble to take advantage of the fact that the ripple is concentrated in rela¬
tively narrow bands around the harmonics of the carrier to design an
improved type of ripple filter incorporating band-rejection filter sections.
The use of this type of filter is of considerably greater practical importance
Sec. 6.4] DEMODULATORS AND MODULATORS 259

with the full-wave than with the half-wave discriminator. For the half¬
wave discriminator the extra complication involved hardly seems justi¬
fied, since the discriminator itself is not optimum and there is little point
in trying to improve an inferior circuit when a better circuit is available.
This argument does not apply, however, to the full-wave discriminator,
for it represents the best that can be achieved without increasing the cir¬
cuit complexity manyfold. Since the lowest harmonic of the ripple is the
second and since the next one (the fourth) is almost five times smaller (see
Table 6.1), a relatively large improvement in filtering efficiency is obtained
by use of a single rejection filter to remove the band of frequencies around
the second harmonic. All other harmonics can be removed easily by a
standard RC low-pass filter, and, as mentioned in connection with the
half-wave discriminator, this part of the filter attenuates only the higher
frequencies and has therefore a minimal effect on the signal band. Here
again it must be kept in mind that the rejection filter must attenuate
effectively all the frequencies in a band that is centered around the second
harmonic of the carrier and has a width equal to twice the maximum fre¬
quency expected in the signal envelope.
The improved performance possible with the full-wave discriminator is
obtained at the expense of considerable additional circuit complexity.
This is easily appreciated by comparing Figs. 6.4 and 6.1. In order to
get accurate full-wave output, the two halves of the circuit of Fig. 6.4
must be carefully balanced; i.e., the gain must be the same no matter
which triode pair is conducting. If this is not the case, then in Fig. 6.5a,
for instance, every other pulse will be larger than the two adjacent pulses
(see Fig. 6.10a), and the output will then contain a component at the
fundamental of the carrier frequency. <? Ref 9

6.4. Other Triode Discriminator

Circuits. The circuits shown in
Figs. 6.1 and 6.4 have been presented
primarily as examples of practical
circuits giving a half-wave or full-
wave output. Other circuits operat¬
ing on the same principles may,
however, be devised and may be pref¬ Fig. 6.7. Single-ended triode discrimi¬
nator.
erable, depending on the applica¬
tion. We shall consider briefly three of these circuits; others may be
found in the literature.1
Probably the simplest triode circuit is the one shown in Fig. 6.7. This
circuit represents essentially one half of the push-pull half-wave circuit
1 A fairly complete list and references to other circuits are given in Greenwood,
Holdam, and MacRae, “Electronic Instruments,” Radiation Laboratory Series, vol.
21, McGraw-Hill Book Company, Inc., New York, 1948, pp. 383-386.
260 CONTROL SYSTEM COMPONENTS [Chap. 6

shown in Fig. 6.1. As in the other two circuits discussed, the important
feature making this circuit a discriminator is the a-c plate supply, which
disables the circuit every other half cycle. The output is a half-wave
signal similar to that shown in Fig. 6.2, but since this is a single-ended
circuit, the output voltage cannot reverse sign and is always negative
during periods of tube conduction. The magnitude of the d-c component
of the output varies continuously from a large negative value for a large,
inphase signal input to smaller negative values as the input decreases to
zero, reverses in phase, and increases again in the opposite direction.
Hence, if it is necessary that the d-c output be zero for zero input, some
sort of d-c potential-shifting device of the type described in Sec. 2.17 must
be employed in the output. The output contains a large ripple compo¬
nent when the input signal is zero, and on the average the ripple is con¬
siderably larger in this circuit than in the push-pull circuit considered in
Sec. 6.1. Thus a much larger amount of filtering is required to reduce the
a-c component of the output to a permissible value. A further disadvan¬
tage of a single-ended circuit is that it has no drift compensation.
The analysis of this circuit proceeds in most essentials like that of the
push-pull half-wave discriminator given in Sec. 6.2. The lowest ripple
harmonic is the fundamental of the carrier, and therefore the upper limit
on the envelope frequency is one-half the carrier frequency. The circuit
operation cannot, however, be explained in terms of multiplication of the
input signal by a square wave, since this would not account for the half¬
wave output obtained for zero input. A more reasonable assumption
here is that the sum of the input and the sinusoidal reference is multiplied
by the square wave. This assumption is justified by noting that the
plate current of a triode may be expressed by

Zp —l-
Tp

When the analysis is carried out on this basis, it will be found that both
the quiescent d-c value and the magnitude of the ripple are determined
primarily by the magnitude of the added reference voltage. This voltage
should, therefore, be kept as small as possible, but on the other hand it
must be large enough so that the sum of input and reference is always
positive during the on period of the circuit. It is clear in summary that,
although the circuit appears to be simpler than the one shown in Fig. 6.1,
the extra filtering and d-c potential-shifting circuits that must be used
with it largely nullify this apparent simplicity.
Both the circuits shown in Figs. 6.1 and 6.4 can be converted to cathode-
follower circuits if it is not required that the signal be amplified and if the
low output impedance and low drift afforded by cathode followers is
desirable. In Fig. 6.8 is shown a half-wave cathode-follower circuit cor-
Sec. 6.5] DEMODULATORS AND MODULATORS 261
responding to the circuit of Fig. 6.1. Again its operation depends on the
fact that the tubes are disabled by the reference signal every other half
cycle. It can, therefore, be analyzed exactly as the circuit of Fig. 6.1
was, and its performance is identical except that the gain is slightly less
than unity. The reason for grounding the reference transformer at the
midtap of the winding is to provide a negative voltage for the cathode

Fig. 6.8. Half-wave cathode-follower discriminator.

return during the operating half cycle. The placement of the tap need
not be accurate. The full-wave counterpart of this circuit can be built
up just as the full-wave circuit of Fig. 6.4 was built up from the half-wave
circuit of Fig. 6.1. Construction of a detailed circuit is left as an exercise
for the reader.
6.5. Diode Discriminators. All the triode discriminator circuits dis¬
cussed thus far drift to a greater or lesser degree because they combine the
functions of discriminator and d-c amplifier. For this reason diode dis¬
criminators are often preferred, particularly in systems having sufficient

Input

Fig. 6.9. Ring demodulator.

a-c amplification without additional d-c amplifiers. A practical advan¬

tage of diode discriminators is that the use of crystal diodes permits very
simple and compact circuitry.
A popular diode discriminator circuit is the so-called ring demodulator,
a schematic diagram of which is shown in Fig. 6.9. This circuit is a full-
wave discriminator, and its operation, which is typical of all diode dis-
262 CONTROL SYSTEM COMPONENTS [Chap. 6

criminators, may be explained as follows: The reference signal is assumed

to be very much larger than the input signal. Therefore, for the half
cycle during which point b is positive and d negative, the lower two diodes
conduct and the upper two are blocked. Hence there is an electric con¬
nection between the output terminal and point c, the lower end of the
input transformer. If the secondary of the reference transformer is
accurately center-tapped and if the impedances of the circuits be and cd
are exactly the same, there will be no voltage between point c and the
output. The output voltage will therefore be the same as that of point c.
During the next half cycle, point b is negative and d is positive. Thus the
upper two diodes conduct, and the output is connected to point a. The
circuit acts as a switch, connecting the output to either point a or point c
in synchronism with the reference signal. The resistances R are required
to limit the current flowing in the diodes and to equalize the impedances
of the four arms.
From the description it should be clear that the operation of the circuit
consists of multiplying the input signal by a square wave of unit ampli¬
tude (if the turns ratio of the input transformer is unity). Hence the
analysis of the full-wave discriminator (Sec. 6.3) is directly applicable,
and the output waveshapes observed under various operating conditions
are ideally as shown in Fig. 6.5. It follows that the lowest ripple har¬
monic is the second harmonic of the carrier and that the maximum enve¬
lope frequency of the input is the carrier frequency.
In practice, unbalances between the bridge arms and inaccuracies in
the location of the taps of the transformers result in output waveshapes
that depart considerably from the ideal forms shown in Fig. 6.5. The
effect of these unbalances is that the output contains a component at
fundamental frequency, so that a filter with a reduced passband is
required to smooth out the ripple. Since this to some extent defeats
the purpose of the full-wave discriminator, some effort to balance the cir¬
cuit properly is usually justified. The various adjustments required for
this end are best understood by considering the various output waveforms
in some detail.
Typical waveforms observed when the signal is in phase with the refer¬
ence are shown in Fig. 6.10. The voltage of Fig. 6.10a may be obtained
by adding a sine wave, at fundamental frequency and in phase with the
reference, to a true full-wave-rectified signal like that shown in Fig. 6.5a.
It can be shown by means of elementary circuit analysis that this type
of output results from inaccurate midtap location on either of the two
transformers or from inequalities in the resistors used in the bridge arms.
If the center tap of the input transformer is incorrectly located, then no
compensating adjustment of either the resistances or the reference center
tap can be made,to reduce this unbalance except at one signal amplitude.
Sec. 6.5] DEMODULATORS AND MODULATORS 263

On the other hand, if the signal transformer is not at fault, then it should
be possible by the adjustment of any one of the four resistors to produce a
true full-wave-rectified output. However, it may then be found that there
is some d-c output for zero input. If this is undesirable, then another
adjustment must be provided. Assuming that the transformer taps
cannot be moved, one of the resistors in circuit bad and one in circuit bed
of Fig. 6.9 must be made adjustable to permit both zero d-c output for no
input and a balanced full-wave output to be achieved.
The waveform shown in Fig. 6.105 results from the addition of a full-
wave-rectified voltage and a sine wave at fundamental frequency but 90°
out of phase with the reference. This effect cannot be produced by
resistance unbalance but may be caused by unequal capacitances between
the two halves of the reference winding and ground. This in turn may be
caused by a reference transformer that does not have an electrostatic

AYV
(a) id) (c)
Fig. 6.10. Output waveforms resulting from circuit unbalance.

shield between its primary and secondary windings. The effect may be
simulated by a small capacitor connected from either point 5 or point d
to ground (Fig. 6.9). The remedy is to use a transformer with a Faraday
shield or to add compensating capacitors between points 5 or d and ground.
In Fig. 6.10c is shown the result when both the effects described occur
simultaneously. Both resistive and capacitive adjustments must then
be made to convert the output to the balanced full-wave form.
In addition to the balancing problem, another point of interest in
this circuit is the maximum input signal for which the circuit will still
function properly. To determine this, assume that the circuit is properly
balanced and that the lower two diodes (Fig. 6.9) are conducting. The
voltage across either one of the upper diodes can then be found by
KirchhofFs laws. Let e0 be the output voltage and ea, eb, etc., the voltage
of points a, 5, etc., all with respect to ground, and let eR be the reference
voltage across one half of the reference-transformer secondary [that is,
eR — A(eb — ed)]', then the voltage across the rectifier in arm ab is given
by ea — (e0 + eR). However, during the period when the lower set of
rectifiers conducts, eQ = ec = — ea; hence the rectifier voltage is 2ea — eR.
Similarly the rectifier voltage in leg ad is given by 2ea + eR. The refer¬
ence voltage eR is positive, since the lower rectifiers conduct. Hence, if
ea is positive, the voltage across the rectifier in arm ab becomes positive
when ea exceeds eR/2, and the rectifier conducts. For negative ea the
264 CONTROL SYSTEM COMPONENTS [CHAP. 6

voltage across the rectifier in arm ad becomes negative when ea exceeds

e/e/2, and this rectifier conducts. Since neither of these rectifiers should
conduct when the lower set conducts, this represents improper operation.
Letting ea = es, the signal voltage, we may therefore say that for proper
operation
es < (6.15)
Note that this computation has also given us the maximum back voltage
on the rectifiers: for the maximum permissible signal this is equal to 2e#.
In applications permitting a half-wave-rectified output signal, simpler
discriminator circuits like those shown in Fig. 6.11 may be used. The
circuit shown in Fig. 6.11a is essentially half of the ring-modulator circuit

9 Ref
UmJ

Fig. 6.11. Half-wave diode discriminators.

and functions the same way. During the half cycle when the diodes con¬
duct, the output and input are connected together, and during the next
half cycle, the output is zero. A slightly different version of this circuit
is shown in Fig. 6.115. Here R2 Ri, so that, during the periods when
the diodes conduct, the signal is essentially shorted to ground. Both of
these circuits when properly adjusted will have zero d-c and ripple output
when the input is zero. This is not true in the circuit shown in Fig. 6.11c,
whose output is always a half-wave-rectified wave. The amplitude of the
output signal is increased by an input that is in phase with the reference
and decreased by a signal that is out of phase with the reference. Thus,
although this circuit is exceedingly simple, it requires a d-c potential-
shifting circuit to make the average value of the output wave zero when
the input is zero, and it usually requires more extensive filtering. Placing
two circuits of the form of Fig. 6.11c back to back results in the push-pull
circuit shown in Fig. 6.lid. The reader will recognize that the circuits of
Fig. 6.11c and d are the diode counterparts of the triode circuits of Fig.
6.7 and Fig. 6.1, -respectively.
Sec. 6.6] demodulators and modulators 265

A full-wave diode discriminator capable of handling relatively large

currents, which may therefore be used to couple an a-c power amplifier to
a load requiring a d-c signal, is shown in Fig. 6.12. The reference-trans-
former windings are connected in such a way that during one half cycle
all the diodes in the upper diamond conduct and the ones in the lower
diamond are blocked. During the next half cycle the situation is
reversed. The circuit therefore acts as a switch that connects first the
upper and then the lower end of the
input transformer to the load in
synchronism with the reference.
The advantage of this circuit over
some of the others that have been
shown in this chapter is that the
load current must flow only through
the rectifiers and therefore encoun¬
ters only their forward resistance.
Hence this circuit can be built with
signal efficiencies (ratio of output
power to power supplied by signal) Fig. 6.12. Diode power discriminator.
of approximately 80 per cent. The
current rating is a function primarily of the current rating rectifiers and of
the reference transformer, but it can easily reach the order of amperes.
6.6. Electromechanical Demodulators. Most of the diode circuits
described in the previous section were shown to operate essentially as
synchronous switches. Hence it is clear that a mechanical synchronous
switch could act as a demodulator in the same way as the electronic circuits
described. Switches of this sort, referred to as electromechanical vibra¬
tors, or choppers, are available commercially and are able to handle fre¬
quencies of up to about 1,000 cps. One application of such a device has
already been described in Sec. 2.23
and the reader is referred to that sec¬
Input Output tion for a short description. Figure
:oTl_
o
o
2.34 shows the external appearance
and internal connections of a typical
commercially available unit. A typi¬
Fig. 6.13. Demodulator using mecha:
ical chopper.
cal demodulator circuit using choppers
is shown in Fig. 6.13; more elaborate
circuits can be devised if the application warrants it.1 The output of
the circuit shown is a full-wave signal; hence the output waveshapes
are of the sort shown in Fig. 6.5. In order to minimize the amount
of driving power required, the central reed is usually constructed to have
1 James, Nichols, and Phillips, “Theory of Servomechanisms,” Radiation Labora¬
tory Series, vol. 25, McGraw-Hill Book Company, Inc., New York, 1947, p. 109.
266 CONTROL SYSTEM COMPONENTS [Chap. 6

its resonant frequency near the carrier frequency for which the chopper
is designed. However, the resonant frequency of the reed should not be
exactly equal to the carrier frequency since very large changes in phase of
the oscillation of the reed would result from relatively small frequency
changes. A compromise is therefore necessary; in a typical design the
resonant frequency of the reed is 80 cps for a chopper designed to operate
at 60 cps.1 There is, of course, still some phase shift between the reed
and the reference signal, and in order to obtain maximum output, it is
necessary to shift the phase of either the voltage applied to the vibrator
coil or the signal voltage. A disadvantage of a vibrator constructed in
this way is that it can be used for only one carrier frequency, and, in
general, vibrators are not very practical for frequencies in the upper audio
range. On the other hand, vibrators possess a number of important
advantages over electronic demodulators: they are practically free of drift,
their noise level is very low, and the balancing problem discussed in con¬
nection with the ring demodulator is absent. They are, furthermore,
very compact, simple to use in a circuit, and, despite the fact that there
is a moving part and electric contacts, their life is quite acceptable.
Hence they are replacing electronic demodulators in many applications.
6.7. The Clamping Discriminator. A discriminator operating on a
principle not found in the discriminators described thus far is shown in the

? Ref ?

circuit of Fig. 6.14. In this arrangement the RC circuit in series with the
grid of each of the triodes is a grid-leak biasing circuit. It keeps the
triodes cut off except during a short time when the reference goes through
its positive peak. At this time the grids are driven positive, the capaci¬
tors Ci are recharged, and the output capacitor C2 is connected to the
input and charges to the value of the input signal. Two triodes are
1 Ibid., p. 108. The chopper referred to in this reference is the Synchronous Con¬
verter No. 75829-1 made by Brown Instrument Company, Philadelphia, Pa.
Sec. 6.7] DEMODULATORS AND MODULATORS 267

needed in order that the charging current may flow in either direction. If
the amplitude of the reference signal is very large compared to the cutoff
bias voltage of the triodes, the period of tube conduction is very short,
and it may be assumed that the charge placed on the capacitor C2 repre¬
sents the value of the input signal at the instant that the reference passes
through its peak; i.e., the circuit samples the voltage existing at the input
at this instant. The load impedance connected to the circuit is assumed
to be infinite. Hence the capacitor holds, or ^clamps,” the signal at this
level until the application of the next positive reference-voltage peak. A
typical form of the output for a signal in which the carrier is in phase with
the reference is shown in Fig. 6.15. Note that a constant-amplitude

Input signal Output

Fig. 6.15. Output waveshapes for typical input.

input ideally results in a d-c output, while for a varying input the output
has a steplike appearance. The ripple content of the output is much less
than for more conventional discriminator circuits: To appreciate this
fully, the reader should keep in mind that the circuit described here is a
half-wave circuit; i.e., the signal is sampled only once per cycle.
In the circuit design, the RC time constant of the grid-leak circuits
should be of an order of magnitude larger than the period of the reference
signal in order that the capacitor shall not discharge significantly during
one period. The grid-leak resistor must be large enough to prevent
excessive grid current from being drawn at any time. The output
capacitor C2 should have such a value that the RC time constant of this
capacitor in combination with the resistance of the conducting tubes will
be very short compared to one period of the reference signal so that the
capacitor will charge up to the full signal level during the short time that
the tubes conduct. On the other hand the time constant with the tube
not conducting should be very much greater than a period of the reference
to ensure that the signal is properly held during the time between sam¬
pling instants. Hence the load impedance should be several orders of
magnitude greater than the plate resistance of the triodes, and a cathode
follower is preferably used here. It is clear that the improvement in
performance is obtained at the expense of a somewhat more complicated
circuit. This is particularly true if, as is commonly done, the reference is
268 CONTROL SYSTEM COMPONENTS [CHAP. 6

passed through peaking circuits and applied to the discriminator in the

form of sharp pulses so as to shorten the sampling time.
If the carrier component of the input signal is 90° out of phase with the
reference, the input signal at the sampling instants is always zero, and
hence the output is zero. In general, if the phase difference between
input and reference is greater than 0° but less than 90°, the output ampli¬
tude is reduced in proportion to cos /3, where (3 is the phase difference. In
this respect, therefore, the circuit acts like a standard phase-sensitive
discriminator.
Since the circuit is a true sampler, the limiting frequency of the enve¬
lope of the input is equal to one-half the number of samples taken per

second.1 Hence an increase in the sampling rate makes the circuit useful
for higher-frequency signals, and it also reduces the amount of ripple at a
given input frequency. One way to double the sampling rate is to use
the full-wave equivalent of the clamping discriminator, shown in Fig.
6.16. The connection of the reference transformer is such that when the
upper pair of triodes conduct, the lower pair is cut off, and vice versa.
The operation of the triodes is the same as already described. If we sup¬
pose the upper end of the input transformer to be positive when the upper
pair of triodes conduct, a positive charge is placed on the output capacitor;
during the next half cycle the lower half of the transformer is positive and
the lower pair of tubes conduct, placing another positive charge on the
capacitor. Thus two samples per reference cycle are obtained.
By modifying the reference signal, the circuit of Fig. 6.16 can be
1 Truxal, “Automatic Feedback Control System Synthesis,” McGraw-Hill Book
Company, Inc., Netv York, 1955, pp. 500ff.
Sec. 6.7] DEMODULATORS AND MODULATORS 269

arranged to yield four samples per reference cycle. This is done by apply¬
ing a reference pulse to the tubes at the instants that the carrier, assumed
to be in phase with the reference, passes through the 45° and 135° points
(see Fig. 6.17). Obviously a fairly elaborate shaping circuit would be
required to obtain the required reference signal, and the additional com¬
plication would probably be worth while only if several discriminators
could all be operated from one reference source in a particular system.
A very simple method for deter¬
mining the frequency response of a
clamping discriminator is given by
Truxal,1 who shows that the transfer
function may be expressed as

1 — e -Ts
H(s) = (6.16)
Ts

where T is the period between sam¬

ples. The gain and phase shift as a
function of frequency of the input Fig. 6.17. Relation of reference to car¬
rier to provide four samples per refer¬
envelope, cos, can be obtained from ence period.
this expression by setting s = jcos and
simplifying the resulting complex function. The result may be put in the
form
sin (Toj8/2) ? J T cv s/ 2
fftfM.) = ?W 2
(6.17)

or, if we let the sampling frequency be co = 2ir/ T, this may be written in

the equivalent form
sin 7r(cos/co) g—; s/co
H(jUs) TTCO
(6.18)
7TC0S / CO

The absolute values of the gain and the phase shift are sketched in
Fig. 6.18. In practice, only the portion below co/2 is of interest, since co/2
is the limit of unambiguous transmission. The gain at this limit is 2/tt,
and the phase lag ir/2 radians. Note that the sampling frequency co is
equal to the reference frequency for a half-wave clamping discriminator
and to twice the reference frequency for a full-wave discriminator. Thus
the results for the upper limit of transmission obtained here are identical
with those obtained with the simpler discriminator circuits discussed at
the beginning of this chapter.
In order to find the magnitude and frequency of the ripple harmonics
in the output, a Fourier analysis must be made on a typical output, such
as that obtained, for instance, by an input with a sinusoidal modulation.
1 Ibid., p. 507. Note that Eq. (6.16) is Truxal’s expression multiplied by 1/T.
Thisois done to make the gain unity for zero frequency.
270 CONTROL SYSTEM COMPONENTS [Chap. 6

Such a Fourier analysis can be made most simply by first finding the
Laplace transform of the output and determining the harmonics from this
by partial-fraction expansion.
The Laplace transform of the output signal obtained for a sinusoidal
input modulation of frequency cos is derived by Gardner and Barnes1 in
connection with their discussion of the solution of difference equations.

Fig. 6.18. Gain and phase shift of clamping discriminator.

It may also be obtained by application of the theory of sampled data

systems.2 In either case it is given by

(sin cosT)(esT — 1)
F(s) (6.19)
s(e2sT — 2esT cos wsT + 1)

The poles of this function are found by solving for the zeros of the denomi¬
nator; they occur for s = 0 and s = j(27rn/T ± cos), n = 0, ± 1, ±2, . . . .
Hence F(s) may be written in the equivalent form
CO

F{s) = - +
s X
71 = — oo
s
kn
j[(2Tvn/T) + C0S]
k'n
+ s + j[(2Trn/T) — cos]
(6.20)

Since 2t/T = co, the sampling frequency, it may be seen from Eq.
(6.20) that the harmonic frequencies are ruo ± ccs, a result completely
analogous to that obtained previously for the simple half-wave discrimi¬
nator [Eq. (6.7)]. The kn may be evaluated by standard techniques
which, although fairly cumbersome, offer no difficulties in principle. The
result of Eq. (6.20) may then be put into the form

|( sin o)sT \ / (2?m/T) + cog \

| \27T7i wsT) \s2 + [(27m/T) + ws]2/

/COS <J0ST — l\ / s
\ 2Ten + o)sT ) \s2 + [(27m/T) + cos]2

1Gardner and Barnes, “ Transients in Linear Systems,” John Wiley & Sons, Inc.,
New York, 1947, p. 299.
2 Truxal, op. cit., pp. 507-517.
Sec. 6.8] DEMODULATORS AND MODULATORS 271

The second fractions of the two products are seen to be the transforms,
respectively, of sin [(27m/T) + ws]t and cos [(27m/T + us]t. Hence the
amplitude of the ripple component at (27m/T) + cos is obtained by taking
the square root of the sum of the squares of the coefficients of these two
factors, with the final result for the amplitude of the nth harmonic:

a _ sin (cosT/2)
(6.22)
n irn + (cosT/2)

for the harmonic frequency (nco -f- cos); and

, sin (to.r/2)
(6.23)
n irn — (a.T/ 2)

for the harmonic frequency (nco — cos).

Note that for n = 0 this result reduces to the amplitude of the desired
output, obtained already by a different argument [Eq. (6.18)]. Note also
that, as cos —> 0, the harmonics go to zero; this agrees with our previous
finding that a constant-amplitude input yields no ripple output.
6.8. Modulators. Modulators are used in servo systems when it is
desirable to convert the low-frequency signals to a higher frequency. A
common example is a servo system in which an a-c motor is used in the
power stage, yet where the error signal is a d-c (low-frequency) voltage.
In such a system the error is usually modulated at a low signal level, and
the resulting high-frequency signal is amplified in a-c amplifiers and
applied to the motor. In this way the drift problems encountered with
d-c amplifiers are avoided. Sometimes both demodulators and modu¬
lators are used in a system, even though both the error signal and the volt¬
age required by the power unit are alternating current. This is done
because the loop-transfer shaping networks or equalizers that are required
to obtain optimum system performance can be given much more satis¬
factory characteristics when they are designed to operate in the low-
frequency-signal spectrum than when they must operate on the modu¬
lated signal directly. Hence, where such networks are to be used, the
signal coming from the transducer is first passed through a demodulator,
then operated upon by the network, then remodulated, amplified, and
applied to the motor. A further example of the use of modulators and
demodulators is the drift-free d-c amplifier described in Chap. 2 (Sec.
2.23). Since a-c servomotors and most of the demodulators used in
servomechanisms require a suppressed-carrier modulation and also since
the suppressed-carrier modulation is more symmetrical than standard
amplitude modulation (i.e., the amplitude of the a-c output is independent
of the sign of the input), almost all modulators designed for use in servo
systems yield suppressed-carrier type of modulation.
All the demodulator circuits discussed in the preceding sections, except
272 CONTROL SYSTEM COMPONENTS [Chap. 6

for the clamping discriminator, will function as modulators if the input

is a low-frequency signal rather than a modulated carrier. This is so
because, as has been explained previously, all of these circuits operate
to multiply the input by a square wave. Hence, when a low-frequency
input signal is applied, for instance, to the ring-demodulator circuit (Fig.
6.9), the modulated output has the appearance shown in Fig. 6.19. The
carrier is a square wave; this is advantageous if the signal is to be demodu¬
lated by another square-wave multiplication, since the ripple output is
much less than for a sine-wave carrier. On the other hand, if the signal

/ 1 Is /
A h
y X
bl
nI V
V

Output
Fig. 6.19. Output of full-wave modulator.

is to be applied to the control phase of an a-c motor, the square-wave

carrier is not desirable; the higher harmonics may, however, be removed
by low-pass filters.
An important modification is necessary on circuits such as that of Fig.
6.9, if, as is usually the case in servo applications, the modulator must be
able to handle input frequencies down to zero frequency. An input
transformer cannot be used under these conditions to provide the push-
pull input required in the operation of the circuit, and a direct-coupled
phase inverter (see Chap. 2, Secs. 2.15 or 2.16) would be required instead.
This complication is, however,
avoided with the circuit of Fig. 6.9
by operating it backwards, i.e., by
Output
-O applying the low-frequency input
at the terminal used for the output
when the circuit is used as a demod¬
Fig. 6.20. A simple transistor chopper. ulator and taking the a-c output
(Bright, Kruper)
from the transformer terminals.
The reader may show that the operation of the circuit is not affected by
this procedure.
Very successful low-level square-wave modulators can be constructed
with transistors.1 Typically, n-p-n fused-junction transistors have been
used, but p-n-p types could also be employed. The major advantages of
transistor choppers are (1) no appreciable phase shift even with chopping
1 A. P. Kruper, Switching Transistors Used as a Substitute for Mechanical Low
Level Choppers, Trans. AIEE, vol. 74, part 1, pp. 141-144, 1955. R. L. Bright,
Junction Transistors Used as Switches, Trans. AIEE, vol. 74, pp. 111-121, 1955.
Sec. 6.9] DEMODULATORS
/
AND MODULATORS 273

rates as high as 10 kc and (2) excellent linearity for signals down to 0.1 milli¬
volts in amplitude. There is little temperature drift so long as the input
impedance is kept low. Figure 6.20 shows a simple transistor chopper.1
The square-wave drive applied to the bases of the transistors alternately
switches one transistor on and the other off. For the half cycle in which
A conducts, B is cut off; thus the input is connected to the output through
the output transformer. For the other half of the cycle, A is cut off and
the path from the input is open-circuited. At the same time B conducts
and shorts the output through resistor R. While a common emitter cir¬
cuit could be used here, the common collector was chosen because the
static collector voltage for zero collector current is only about 1 mv. This
voltage appears across the output transformer in the same polarity for
both transistors, and if the transistors are matched, this is a constant d-c
value and thus does not appear at the output terminals.

[o) Input (b) Demodulator (c) Demodulator

output output otter passage
through high-pass filter

Fig. 6.21. Waveshapes obtained from half-wave modulator.

When a half-wave demodulator is used as a modulator, the low-fre¬

quency signal is multiplied by a square wave having a d-c value equal to
one-half the peak-to-peak square-wave amplitude (see Sec. 6.2). Hence
the output is as shown in Fig. 6.216, and it contains a low frequency as a
harmonic component. If this signal is amplified, however, in an audio
amplifier having the usual high-pass coupling networks, then the low-
frequency component is removed from the output, and the signal shown
in Fig. 6.21c results. It is seen to be identical with that produced by the
full-wave modulator (Fig. 6.19). Thus, except that the gain of the half¬
wave modulator is only one-half as large as for an equivalent full-wave
modulator, there is no particular advantage in the use of a full-wave
rather than a half-wave modulator.
6.9. Drift in Modulators. All electronic modulators drift a certain
amount, with triode circuits being generally worse than diode circuits.
The reasons for drift in triode circuits are simply shown. Consider the
circuit shown in Fig. 6.22, which is essentially the same circuit as the
demodulator of Fig. 6.1 except that it also acts as a phase inverter for the
single-ended input signal (see Sec. 2.16). The operation of this circuit
has been explained as multiplication by a square wave of amplitude G/2,
1 R. L. Bright and A. P. Kruper, Transistor Choppers for Stable DC Amplifiers,
Electronics, April, 1955, pp. 135-137.
274 CONTROL SYSTEM COMPONENTS [Chap. 6

where G is the gain during the conducting period. Clearly if, because of
variation of tube parameters, 0 changes, the output for a given input
changes. In particular, if the input is zero and the tube characteristics
are not exactly identical, some square-wave output will result. Thus if
the characteristics of one tube change relative to those of the other, the
amplitude of this square-wave output can change; this constitutes drift.
For the diode circuit the cause of drift is somewhat different and
depends on the fact that a diode will conduct some current when reverse
voltage is applied. Hence, if we consider the half-wave circuit of Fig.
6.11a, we find that during the period of nominal nonconduction some signal
may get to the output, which instead of oscillating between the signal
amplitude and zero will instead oscillate between full-input amplitude
and reduced-input amplitude. It is clear that, if temperature changes or
other factors cause the back current of the diodes to change, then the

magnitude of the output square wave changes for a given input, and we
again have drift.
Since a major reason for the use of a modulator in a servo is to prevent
the drift that is caused by d-c amplifiers, a modulator that drifts is of
relatively little usefulness. For this reason the most commonly used
modulator, particularly for systems using a 60- or 400-cps carrier, is the
electromechanical chopper, already discussed previously (Secs. 2.23 and
6.6). It may be connected as shown in Fig. 6.13, but with the input and
output terminals reversed, or it may be connected as a half-wave modu¬
lator, as shown in Fig. 2.35. In either case its good square-wave output
and lack of drift are the primary advantages.

PROBLEMS

6.1. The output from a full-wave discriminator is passed through a notch-type

rejection filter which completely removes the second harmonic of the reference fre¬
quency but has negligible effect on any other harmonics. Assume that the input
s’gnal is in phase with the reference. Plot the form of the output signal delivered by
the discriminator and the notch network.
6.2. Repeat Prob. 6.1 for the case of (a) an input signal 90° out of phase with the
reference, and (b) an input signal 45° out of phase with the reference.
Problems] DEMODULATORS AND MODULATORS 275

6.3. A two-section low-pass filter used to smooth out the ripple generated by a half¬
wave discriminator operating on a 60-cps carrier has the transfer function

It is desired to reduce the rms value of ripple output to less than 10 per cent of the
useful d-c output. Assume the input signal is in phase with the reference and find the
time constant T required.
6.4. Repeat Prob. 6.3 for a full-wave discriminator. Find the ratio of time con¬
stants required in the two discriminator types.
6.6. Design a ring demodulator like that in Fig. 6.9. Assume both transformers
have unity ratio. The maximum expected value of the input signal is 20 volts rms.
The diodes are 1N34, having a maximum inverse voltage rating of 60 volts and a
maximum current rating of 50 ma. The circuit operates into a load resistance of
10,000 ohms, and it is desirable to keep the output impedance of the discriminator to a
minimum. Find the values of the resistors R required.
6.6. Find the current rating of the secondaries of the reference transformer and
of the rectifiers required in the power discriminator of Fig. 6.12. Express the result
as a ratio of maximum load current.
6.7. In the clamping discriminator shown in Fig. 6.14 the triodes have a plate resist¬
ance of 1,000 ohms when they conduct. The pulses applied to the grids cause the
tubes to conduct for one-twentieth of a cycle. During this time the capacitor should
charge to 95 per cent of the correct input value. The capacitor should hold the cor¬
rect value with a loss of less than 5 per cent during the period that the tubes do not
conduct. The pulse-repetition frequency is 400 cps. Find the value of C2 required,
and also the minimum value of load resistance.
 CHAPTER 7

A-C MOTORS

7.1. Introduction. The motor used most commonly in low-power

applications is the a-c two-phase induction motor. When it is used as a
servomotor, one of the phase windings is connected to a fixed a-c voltage,
the reference voltage, and the other winding is supplied by a variable con¬
trol voltage, 90° out of phase with the fixed voltage. When the control
voltage leads the fixed voltage, rotation is in one direction; when it lags the
fixed voltage, the direction of rotation is reversed. The speed and torque
depend on the magnitude of the control voltage, and although the relation
is not linear, it is sufficiently regular so that approximately proportional
control may be achieved.
One of the reasons for the popularity of this type of motor is that it
requires only the simplest type of control amplifier. A typical arrange¬
ment is shown in Fig. 7.1, where a synchro pair is used to indicate the

--o A-c
Fixed
phase supply
lM)J
A-c
omplifier
-Sj

Control
phase
C1
1
) Motor

Gears
-1- Synchro
|
generator
A-c
£- KfJ-0 supply
Control
transformer
Input
Fig. 7.1. All a-c servo using a two-phase motor.

error between input and output. The error voltage from the control
transformer is amplified in a standard a-c amplifier and applied to the
control winding of the motor. The 90° phase shift between the control
and reference voltages may be obtained either by connecting a capacitor
in series with the reference winding or by incorporating a simple phase-
shift network in the amplifier. Neither method offers any particular
difficulty, but the second has the advantage that the phase shift is not
affected by the motor operating conditions. Another small departure
Sec. 7.2] A-C MOTORS 277

from conventional amplifier design results from the highly inductive input
impedance of a-c motors. To improve the effective load power factor, a
resonating capacitor is usually connected across the amplifier output
terminals.
It should be noted that it is not necessary to pass the signal from the
control transformer through a phase discriminator; the motor acts as its
own discriminator. Hence many simple control systems require no d-c
amplification at any point in the loop, and therefore the problem of drift
associated with d-c amplifiers is eliminated.
Another desirable feature is that the motor, in common with all induc¬
tion motors, requires no electric connections, such as brushes or slip
rings, between its stator and rotor, and friction is thus reduced to an
absolute minimum. This results in very smooth operation and is one
of the reasons why the induction motor is used even in systems where the
error signal is d-c and must be passed through a chopper or other modu¬
lator to actuate the motor.
A-c servomotors are built for power outputs ranging from x/i to 1,000
watts; however they are most commonly found in applications requiring
less than 10 watts. The efficiency is usually quite low, 5 to 20 per cent
being typical. This, in addition to the fact that the motors often must
operate near zero speed for extended periods of time, necessitates the use
of an external blower on units of more than 10-watt output rating to pro¬
vide proper cooling. Such a blower is usually driven by a separate
single-phase induction motor.
7.2. Construction Features. The stator frames of a-c servomotors are
all constructed in essentially the same way although motors of different
manufacturers have external differences, and various mounting styles are
available. Several types of commercially available servomotors are
shown in Fig. 7.2. The stators have a standard distributed winding to
improve the space distribution of flux density around the air gap. It
is desirable to have a flux distribution as close to sinusoidal as possible,
because space harmonics in the flux tend to produce notches and other
irregularities in the speed-torque curve of the motor. Most 60-cps
motors are wound for two or four poles, but 400-cps motors may be
wound for as many as 10 or more poles, to reduce the operating speed.
The two phases of the winding may be identical; however, in some
motors the control winding is designed for a much higher voltage than
the reference winding, and in some cases the control winding is center-
tapped. These special features make it possible to connect the motor
directly to the output stage of the amplifier without the need of an
output transformer. It is also possible, particularly in the smaller
motors, to design the stator winding in such a way that only a small frac¬
tion of the power input to the motor is supplied by the control amplifier,
278 CONTROL SYSTEM COMPONENTS [CHAP. 7

the major share of the power coming from the reference supply. All of
these special design features serve to reduce the size and weight of the
control amplifier.
There are three basically different rotor designs: the squirrel-cage rotor,
the solid-iron rotor, and the drag-cup rotor. Squirrel-cage rotors are
quite similar to those used in standard induction motors except that, in
order to reduce the moment of inertia, the diameter is usually much
smaller than the length (see Fig. 7.3). Squirrel-cage windings are subject

Fig. 7.2. Various types of servomotors. (Courtesy Ford Instrument Co.)

to cogging, or slot lock, a magnetic attraction between stator and rotor

which keeps the rotor from turning until the control voltage reaches
a certain minimum breakaway value. The winding is therefore usually
skewed (see Fig. 7.3), to minimize this effect.
The solid-iron type of rotor is very similar in external appearance to
the squirrel-cage type, being a slender cylinder. The iron used in its
construction, in addition to having good magnetic qualities, must have
a relatively high conductivity so that sufficient current can be made to
flow in it. There is no slot-lock problem with this type of rotor; however,
the torque developed is somewhat less than for an equivalent squirrel-cage
winding.
Sec. 7.3] A-C MOTORS 279

Where the lowest possible moment of inertia is desired, a motor with

a drag-cup type of rotor is used (see Fig. 7.4). The stator of a motor
designed for this rotor has, in addition to the usual stator punchings and
winding, a central cylinder also made of steel punchings to complete the
magnetic circuit. The drag cup fits into the air space between the wind¬
ings and the stationary cylinder, and the clearances are kept as small as
possible to reduce the length of the air gap. Despite this, the air gap in
a drag-cup motor is always very much longer than in motors using either
of the other two rotors, and for this reason the torque developed by a

rotor
S • (b)
Fig. 7.3. Cutaway view of a-c servomotor with squirrel-cage rotor. Note long slim
design and skewed rotor slots.

drag-cup motor is relatively small. Since, however, the inertia of the

drag cup is extremely low, both the ratios of torque to inertia and of
torque squared to inertia in the drag-cup motor compare favorably with
those of the other rotors.
7.3. Theory of Operation. The theory of operation of the two-phase
servomotor depends on the theory of the polyphase induction motor. We
review this theory here briefly for the benefit of any reader not familiar
with it. For a more detailed discussion of the theory of polyphase induc¬
tion motors, the reader is referred to any one of the standard texts on a-c
machinery.1
1 For instance, Puchstein, Lloyd, and Conrad, “AC Machines,” 3d ed., John Wiley
& Sons, Inc., New York, 1954, pp. 252-324.
280 CONTROL SYSTEM COMPONENTS [Chap. 7

A schematic diagram of a two-phase motor is shown in Fig. 7.5. The

rotor is assumed to have a squirrel-cage winding, and there are two stator
windings displaced in space by 90 electrical degrees. (The physical
displacement between poles on the stator frame is 180°/p for a machine

Fig. 7.4. (a) Exploded view of a drag-cup a-c servomotor; (6) internal construction of
a drag-cup motor.

wound for p poles.) These windings are excited by an alternating volt¬

age. In the standard polyphase machine the two voltages have the same
rms value but are 90° out of phase. The resulting currents set up a
rotating magnetic field in the air gap of the machine, and if the windings
are properly distributed in the slots, this magnetic field is approximately
Sec. 7.3] A-C MOTORS 281

uniform in time and a sinusoidal function of angle. The magnetic flux

cutting the conductors of the rotor winding set up currents in the rotor, and
the interaction of these currents with the flux results in a torque tending
to turn the rotor in the same direction as the magnetic field.
A more quantitative insight into motor operation is obtained by the use
of an equivalent circuit. To obtain this equivalent circuit, we assume,
for the moment, that the rotor is not turning;
the rotor may then be thought of as a short-cir¬
cuited secondary winding of a transformer whose
primary is the stator winding. In a two-phase
motor there are two such transformers, but if a
balanced set of polyphase voltages is applied to the
motor, they are identical, and we need to consider * Reference^
phose
only one. The voltage induced in the rotor, or
Fig. 7.5. Schematic of
transformer secondary, is proportional to the two-phase motor.
primary self-induced voltage, the constant of pro¬
portionality being the effective turns ratio between primary and secondary.
The current per phase flowing in the rotor is therefore

ho =
E 2o N, Ei
(7.1)
R2 + j'^fE2 Ni R2 + flirfLi

In these equations Ei and E2 are induced voltages per phase, and Ni and
N2 are the effective turns of the stator and rotor, respectively. R2 and
L2 are the resistance and inductance per phase, respectively, of the rotor,
and / is the line frequency. The additional subscript o indicates that the
quantities so marked are measured at standstill. If the rotor is permitted
to rotate, the relative velocity between the rotating magnetic field and the
rotor conductors decreases. Hence both the voltage induced in the rotor
and the rotor frequency decrease, becoming zero when the rotor speed
becomes synchronous, i.e., equal to the speed of the magnetic field. It is
easily demonstrated that the induced voltage and frequency are propor¬
tional to the slip S, defined as

synchronous speed — rotor speed

S = (7.2)
synchronous speed

Hence, in general, the rotor current per phase becomes

j SE2o _ A"2 SEi

(7.3)
2 = Ri + WfSLt ~ Nl i?2 + JSX7o
where X2o is the rotor reactance per phase at standstill, Equation (7.3)
may be written in the equivalent form
N2 Ah N2 Ei
12 = (7.4)
Ni (R2/S) + jX2o Ni R2 + jX2o + R2[( 1 - S)/S\
282 CONTROL SYSTEM COMPONENTS [Chap. 7

In the second form of Eq. (7.4) the effective rotor resistance R2/S has
been broken up into two components, R2 and i?2[(l — S)/S\. This is
done to indicate two components of rotor power. The component due to
the currents flowing through R2, that is, \I2\2R2, is the rotor copper loss;
hence the remaining power, \I2\2R2[(1 — S)/S]f is proportional to the
mechanical power developed by the rotor.
In order to determine the stator current, we note that, whether or not
the rotor turns, the points of both maximum induced voltage and maxi¬
mum rotor current must rotate around the rotor in synchronism with
the stator magnetic field. The rotor current therefore causes a demag¬
netizing mmf N2I2 rotating at synchronous speed (note that a squirrel-
cage winding is effectively a single turn; hence the mmf is numerically

X* R'z 4
■ww^ ww—ORHP—
-►/t

fi
R,m \X,m

Fig. 7.6. Induction-motor equivalent circuit.

equal to I2). Since the magnetic flux in the air gap is fixed if the applied
voltage is fixed, a current must flow in the stator to produce a magnetizing
force exactly equal to the demagnetizing N2I2\ i.e.,

NJ[ = N2I2 (7.5)

We use the symbol /{, since the current required to cancel the mmf N2I2
is only the power-producing component of the stator current. There is
also a magnetizing component IM, which is required to set up the rotating
magnetic field. From Eqs. (7.5) and (7.4), we get

_ n ft)
1 R'2 + jX'2o + R'2[( 1 -S)/S\ 1 ;

where R2 = R2(Ni/N2)2 and X2o = X2o(Ni/N2)2. The primary induced

voltage Ei may be obtained in terms of the applied voltage and the stator
current by writing
Ei = Vx — h(Ri + jXi) (7.7)

where V\ is the applied voltage per phase, I\ = I[ + Im is the stator cur¬

rent, and R\ and X\ are the resistance and leakage reactance of the stator,
respectively. Equations (7.6) and (7.7) form the basis of the equivalent
circuit, which takes the form shown in Fig. 7.6. In this circuit XM is the
magnetizing reactance, and the current flowing in it is the magnetizing
component of the primary current. The resistance Rm is added to
account for the fact that the magnetizing current contains a small inphase
Sec. 7.41 A-C MOTORS 283

component to supply the iron losses. In most servomotors R,\r is so

much larger than Xm that it may often be ignored.
The total power going into the rotor is \I[\2R'2/S. Note that

\I[\2R2 = |/2|2E2 = rotor copper loss

so that \I[\2R'2[(l — S)/S] must therefore be the developed mechanical

power.
To find the developed torque, use may be made of the fact that, if a
consistent set of units is used, the product of torque and speed is power.
Using Q for torque and 12 for speed, we have

QQ = 2|/{|

The factor 2 arises from the fact that the analysis has been made on a
per phase basis. Since there are two phases, the actual power is therefore
twice the power per phase. From Eq. (7.2) we see that 1 — S = 12/12s,
where 12s is the synchronous speed. Hence

q = -||/:p§ (7.8)

Since 12s is a constant, the developed torque is seen to be proportional to

the total power dissipated in the rotor.
We can now obtain an expression for the torque as a function of the
input voltage Vi and the slip. To simplify the algebra, we use the
notation
Zx R1 + jXi
J_
1 + 1
Zm Rm jXm
Z2(S) §+JX'
Ex
Then n Z2(S)
Z2(S)ZM/[Z2(S) + Zm]
and Ei = V
xZx + 1 Z2(S)ZM/[Z2(S) + Zm]}
ZM
V\
2_ Z2(S) + Zm Ro
Therefore Q = f = F(S) | Up (7.9)
12., Z2(S)Z m
Z\ +
Z2(S) + Zm
7-4. Theory of Operation of A-C Servomotors. The operation of a-c
servomotors differs from that of conventional induction motors primarily
because the voltages applied to the servomotor usually do not constitute
a balanced polyphase set. Hence we must investigate the behavior of
induction motors supplied by unbalanced voltages.
284 CONTROL SYSTEM COMPONENTS [CHAP. 7

In the following1 analysis we assume that the two stator windings of

the servomotor are identical and that both the reference voltage and the
control voltage are supplied from sources having zero output impedance.
Extensions of the theory to cases in which these assumptions do not hold
are not difficult and are indicated
Reference below. The nomenclature to be
vM winding
VM used is indicated in Fig. 7.7. The
.Ir
IMJ
reference voltage V M is assumed to
Control j/ §
winding c %
*0^jfVc be constant; the control voltage Vc
rv -S
is variable. The vector diagram
(o) Schematic id) Vector relations shows the general relation existing
of two-phase between reference
servo motor and control voltages
between VM and Vc. The angle 9
Fig. 7.7. Nomenclature to be used in
is normally 90°, but for the sake of
analysis of two-phase servomotor. generality it is taken to have any
arbitrary fixed value.
Since the analysis of polyphase motors with balanced voltages applied
to the windings is relatively simple, it is convenient to use the method of
symmetrical components to convert the unbalanced set of voltages shown
in Fig. 7.7 into two balanced sets of voltages of opposite phase rotation.2
The two symmetrical components are shown in Fig. 7.8 for a particular
set of values of VM, Vc, and 9. Note that Vc\ lags Vmi by 90°, and Vc2
leads Vm2 by 90°. The magnitude of the two symmetrical components

[a) Applied voltoges [b) Positive sequence (c) Negative sequence

Fig. 7.8. Voltages applied to motor and their symmetrical components.

and their relative phase position depends on the magnitudes of Vm and

Vc and 9.
To make the transformation we write
Vc = V el + vc2 (7.10)
VM = V Ml + V M2 (7.11)
But V cl = —jV Ml (7.12)
vc2 = jVM 2 (7.13)
so that Vc = -jVM 1 + JVm2 (7.14)
jVc = VmI - Fm2 (7.15)
1 This section essentially follows Koopman, Operating Characteristics of 2-phase
Servomotors, Trans. AIEE, vol. 68, pp. 319-329, 1949.
2 Lyon, “Application of the Method of Symmetrical Components,” McGraw-Hill
Book Company, Inc., New York, 1937, pp. 96-110.
Sec. 7.4] a-c motors 285

Then a simultaneous solution of Eqs. (7.11) and (7.15) gives

VM1 = Yjl+JXi (7.10)

V„2 = Vm ~ jVc (7.17)

Now suppose that the ratio of Vc to VM is k and that Vc lags V M by an

angle 6. Then
Vc = kVM/ — 0 (7.18)
and therefore

VMi = 1/ (1 + k/90° - 6) = + (1 + k sin $ + jk cos 6) (7.19)

z z
Vm2 = -tw (1 — /c/90° — 0) = (1 — k sin 0 — jk cos 6) (7.20)
Z z

so that |Fmi|2 = (1 + 2k sin d + k2) (7.21)

|7M2|2 = V~~ (1 - 2k sin 6 + k2) (7.22)

When the unbalanced set of voltages is applied to the motor, the theory
of symmetrical components indicates that the motor acts as though both
symmetrical sets of voltage were applied simultaneously. Hence, if we
assume that the electric and magnetic circuits making up the motor are
linear, we may say that the torque developed by the motor is equal to the
difference between the torque produced by the positive-sequence voltage
and that produced by the negative sequence.
Equation (7.9) gives an expression for the torque produced by a bal¬
anced set of voltages; suppose this set to be the positive sequence. The
expression for torque produced by the negative sequence must have the
same form as (7.9) except that, if the slip is to retain its zero reference at
positive synchronous speed, (2 — S) must be substituted for S. Since
Eq. (7.9) shows the torque to be proportional to the square of the applied
voltage we find that in general

Q - F(S)\Vmi\2 - F{2 - £)|Fm2|2

or, if we let |Fm|2E(*S) = Qb(S)} the torque produced with balanced input
voltages, then is
VM i V M2
Q = Qb(S) Qb(2 - *8) (7.23)
VM
If we substitute for VMi and Vmi their values as obtained in Eqs. (7.21)
and (7.22), we find that

Q = lA[Qb(S)(l + 2k sin 6 + k2) - Qb(2 - S)( 1 - 2k sin d + k2)] (7.24)

286 CONTROL SYSTEM COMPONENTS [Chap. 7
Equation (7.24) indicates the motor to be generally nonlinear, since in
addition to the squared term in k, Qb(S) is not a linear function of S. It
can, however, be shown that the starting torque of the motor is a linear
function of the applied voltage. When the motor is not running, S = 1,
Qb(S) = Qb(2 — aS') = Qfe(l) = Qbo, and therefore
^starting = Qbok sin 9 (7.25)

The presence of the sin 9 term indicates that the developed stall torque
is a function of the phase angle between the control and reference voltages,
so that the motor may be thought of as a phase discriminator. Normally,
of course, 6 is fixed at 90°, since this results in maximum torque. Equa¬
tion (7.25) also implies that the motor is approximately linear as long as
the speed is low, i.e., as long as Qb(S) ~ Qb(2 — S).
Equation (7.24) may be used to plot a complete set of speed-torque
curves for the motor, provided that Qb(S) is known for all S of interest.
This function is defined in terms of the impedances of the motor as in
Eq. (7.9) and can, therefore, be computed if these impedances are known.
The measurement of the impedances is, however, always rather difficult,
and in particular it is almost impossible to separate rotor and stator
impedances by means of simple measurements. In large motors the
approximation is therefore made that ZM is very large, so that its effect is
negligible; then we see from Eq. (7.9) that Qb(S) becomes approximately

q /m _ 2 |Em|2-K2$
M } ~ Tf. i^ + ^i+y^Xx + x^i2
If this approximation is valid, then all that needs to be measured is R\, R2,
and Xi + X2. The stator resistance Ri can be measured by a d-c test;
Ri + R2 and Xi + X2 are obtainable from a measurement of the motor
input impedance at standstill (S — 1); and if the further assumption is
then made that R2 is independent of speed, Qb(S) can be computed for
any desired value of S.
In small servomotors, particularly in those with a drag-cup rotor, the
assumption that ZM is very large is not particularly valid. Furthermore,
R2 depends to some extent on the rotor frequency because of skin effect
and eddy currents. Hence a computation of Qb(S), in addition to being
rather laborious, will not give very accurate results. The problem is
usually, therefore, completely bypassed by measuring the speed-torque
curve of the motor with balanced input, that is, Qb(S), rather than the
impedances. This is quite easy to do for speeds between zero and syn¬
chronous, and this type of curve is usually furnished by the motor manu¬
facturer as part of the motor characteristics. The only difficulty is that
Qb(S) must be known for values of S ranging between 0 and 2; i.e., a
i

speed-torque curve for speeds from synchronous forward to synchronous

Sec. 7.4] A-C MOTORS 287

backward is necessary. Since the speed-torque curve for negative speed

requires special measuring equipment and is usually not supplied by the
manufacturer, it becomes of interest to be able to determine Qb(S) for all
S given its value for 0 < S < 1. This may be done as follows:
By means of a simple algebraic manipulation, Qb(S) as given in Eq.
(7.9) may be written in the equivalent form

ZM V M 1 2 R'2S
Qt(S) (7.26)
Z\ + Zm R2 + jSX2 + &[Z\Zm/{Z\ + Zm)\ ils

The term ZiZM/(Zi + ZM) is an equivalent impedance, which may be

i,
written as Ro + jX0. Since ZM is usually much larger than Z \R0 -f- jX0|
is approximately equal to \Ri + jXi|, but since ZM ~ jXM, the phase angle
of Rq + jXo is always greater than that of R\ + jXi. Hence Ro is usually
a very small resistance. Also in most servomotors, R2 is quite large.
Hence
Z,Z M
R'2 + jSXi + S |R'2 + SR0 + jS(X'2 + Xo)|2 - X'2
Z i + Zm
+ *S2(X'2 + X0)2 (7.27)

The term \ZmVm/Zi + ZM|2 is a constant, independent of S; hence Qb(S)

has the general form
S
Qb(S) (7.28)
Cl + C2*S2
In writing Eq. (7.28), we make the further assumption that R2 is inde¬
pendent of S. Strictly speaking, this is not true, since skin effect causes
R2 to increase with frequency and, therefore, with S. This effect should
be particularly noticeable in motors designed for operation at 400 cps and
having either solid-iron rotors or squirrel-cage rotors with very deep rotor
bars. The skin effect is relatively negligible in lower-frequency motors,
such as 60 cps, and in motors having drag-cup rotors. At any rate, it is
clear that Eq. (7.28) is an approximation, but it is a very simple equation
and sufficiently accurate in most cases. Since there are only two con¬
stants, Ci and C2, to be evaluated, it is only necessary to know the value
of Qb(S) for two values of S other than zero to be able to compute Qb(S)
for all other values of S. This means, of course, that the knowledge of
Qb(S) for 0 < S < 1 is quite sufficient, if extreme accuracy is not required.
A typical set of motor characteristics is shown in Fig. 7.9; the curve
for k = 1 was taken from the data supplied by the manufacturer for
0 < S < 1; Eq. (7.28) was used to extend this curve into the region
1 < S < 2 by using the given data for *8=1 and S = and Eq. (7.24)
was then used to compute the curves for other values of k. The phase
angle between reference and control voltage was taken as 90°. Note that
near zero speed the curves are very nearly parallel and straight.
288 CONTROL SYSTEM COMPONENTS [CHAP. 7

7.5. Approximate Transfer Function of the A-C Servomotor. Strictly

speaking, the term “transfer function” has meaning only for linear sys¬
tems; however an approximate transfer function may often be defined for
a nonlinear device, such as the a-c servomotor, if an operating point is
specified and if the range of input and output variables is small enough to

permit the characteristics to be approximated to a sufficient degree of

accuracy by the first-order term of the Taylor expansion. In order to
obtain such a transfer function for the a-c servomotor, we first express its
speed-torque characteristics in the form
AQ = A AVC - B AQ (7.29)

where A = dQ/dVc and B = — dQ/dQ, evaluated at the desired operating

point, and A12, ATfc, and AQ are small variations of the speed, applied con-
Sec. 7.5] A-C MOTORS 289

trol voltage, and developed torque, respectively, from their operating-

point values. By analogy with the d-c motor (see Chap. 4), A will be
recognized as the motor gain constant, and 1/B as the effective coefficient
of viscous damping.
Given Eq. (7.29), a transfer function can be found. The small elec¬
trical time lag between the application of the control voltage and the
development of torque is neglected for the moment, and the mechanical
friction is assumed negligible. Then the developed torque is opposed
solely by the torque required to accelerate the moment of inertia and by a
possible arbitrary load torque Qa. If the moment of inertia is J, then, in
Laplace transform notation,

12 = AVc - B(Js£i + Qa)

A _ AVC - BQa
or (7.30)
JBs + 1

In these equations the symbol 12 stands for the transform of A12, etc. If
the small electrical time lag due to the inductance of the stator and rotor
circuits is not negligible, then, approximately,

AVC- BQa
(7.31)
(TlS + 1 )(JBs + 1)

where Ti is the stator-inductance time constant neglected in Eq. (7.30);

it is typically about ten times smaller than the inertia time constant JB.*
To evaluate A and B from the speed-torque curves, we note first that,
since
Vc = kV M
and 12 = 0,(1 - S)
d 12 12s dS
therefore
dVc Vm dk
d!2 dS
and
dQ sdQ

Both dS/dk and dS/dQ may be found by partial differentiation of Eq.

(7.24). Thus

. ai2
A ~ dVc
_ 21L_(sin 6 + k)Qb(S) + (sin 6 — k)Qb(2 — S)__
VM (1 + 2k sin 0 + fcj) _ dQt(2 - S) _ 2k gin e + /c2)
do do
(7.32)

* L. O. Brown, Jr., Transfer Function for a Two-phase Induction Servo Motor,

Trans. AIEE, vol. 70, pp. 1890-1893, 1951.
290 CONTROL SYSTEM COMPONENTS [Chap. 7

and

B = -
<912 _4Q,_
dQ <9Qb($) , 07 • n I 7 o\ dQb(2 — *S) 07 • /) | 7
—(1 + 2k sin 8 + k2)-' -- (1 — 2k sin 0 + k2)
do do
(7.33)

As an example of the application of Eqs. (7.32) and (7.33), let us use the
simplified form of the expression for Qb(S) developed in the previous sec¬
tion [see Eq. (7.28)],
s
Qb(S) (7.34)
Cl + C2S2
and use this to find A and B at the operating point ordinarily of greatest
interest in servo applications, i.e., the point at which S = 1, k = 0. We
obtain
Us Cl "f ^2 •
A = 2
V~u Ci - C SU1
(Cx + C2)2
and B = 212s (7.35)
Cl - c2
In many servomotors, Qb(S) is so close to being a linear function of S that
C2 in Eq. (7.34) becomes negligible. Under these conditions, 1/Ci
becomes Qb(l) = Qbo, the blocked-rotor torque, and thus

A = 2 sin 6 (7.36)
VM
B = 2^ (7.37)
Qb 0
If, on the other hand, C2 is larger than Ci, both A and B become negative.
This represents a form of unstable behavior referred to as single-phas¬
ingwhich is discussed in more detail in a later section. By differ¬
entiating Eq. (7.34) it can be shown that, when C2 is greater than Ci,
Qb(S) peaks for 1 > S > 0, i.e., for positive speeds.
The process of partial differentiation described above is relatively
rigorous, but it is rather involved. Hence a simpler, although less accu¬
rate, method is often employed to obtain A and B. In this method the
speed-torque curves of the motor are replaced by parallel straight lines,
as shown in Fig. 7.10. The line for k = 1, 6 = 90° is drawn through the
points 17 = 17s, Q = 0 and 12 = 0, Q = Qbo, and the lines for various k’s
and d’s are drawn parallel to the first one and such that the blocked-rotor
torque is proportional to k sin 8. It is clear that, since this method con¬
siders only the end points of the actual speed-torque curve of the motor,
it cannot take into account any curvature in this characteristic. Also it
is apparent that-it does not yield the same answer as the partial-differ-
Sec. 7.6] A-C MOTORS 291

entiation method, even when Qb(S) is assumed to be linear. For from

the linear characteristics of Fig. 7.10 we get

A = -=p^- sin 6 (7.38)

B = £ (7-39)

which is half as much as was obtained by the partial-differentiation

method [Eqs. (7.36) and (7.37)]. However, since the method yields the
gain and time constant almost directly by inspection, with the only data
needed being the blocked torque
and synchronous speed, it is very
convenient when only a rough esti¬
mate of the motor transfer function
is required.
7.6. Figures of Merit for A-C
Motors. At this point it is perhaps
appropriate to discuss two figures
of merit for servomotors that are
commonly used: the torque-to-inertia
ratio and the torque-squared-to- Fig. 7.10. Linearized speed-torque curves
for a-c servomotor.
inertia ratio. In both figures of
merit the torque in question is Qbo, the stalled torque obtained
when balanced polyphase voltages are applied to tlie motor. The
torque-to-inertia ratio gives an indication of the acceleration capa¬
bilities of the motor without load and can be shown to be equivalent
to the product of bandwidth and maximum speed, which, in turn, is some¬
what analogous to the voltage-gain-bandwidth product often used as a
figure of merit in amplifiers. For, if the bandwidth is defined as the
reciprocal of the motor-inertia time constant, the product of speed and
bandwidth is &S/JB, and if the simplified definition of B given in Eq.
(7.39) is used, this becomes Qbo/J, the torque-to-inertia ratio. The
torque-squared-to-inertia ratio can be shown to be equivalent to the prod¬
uct of bandwidth and power output, which is a criterion similar to the
power-gain-bandwidth product used for power amplifiers. If the motor
speed-torque curve is a straight line, the torque is given by

Q — QbO

and therefore the developed power is

Q& — Qbo
292 CONTROL SYSTEM COMPONENTS [Chap. 7

The maximum power is found by differentiation to be equal to Qbo%/4,

and therefore the power-bandwidth product becomes j^Qbo^s/JB. If the
simplified definition for B from Eq. (7.39) is again used, this becomes
yiQb^/J- The power-bandwidth product is a somewhat more universal
criterion than the speed-bandwidth product and is not affected by gearing
interposed between motor and load. Hence, when the gearing between
motor and load is such that the load inertia has an appreciable effect on
the motor, the power-bandwidth product or torque-squared-to-inertia
ratio gives a somewhat better indication of motor merit than the torque-
to-inertia ratio.
The torque-to-inertia ratio is increased by decreasing the diameter of
the rotor. This is due to the fact that for constant rotor length, the
inertia is proportional to the fourth power of the diameter, whereas the
torque decreases only at something between the second and third power
of the diameter. The torque-squared-to-inertia ratio is seen by the same
reasoning to be decreased by a decrease of the diameter; hence an optimum
ratio of diameter to length exists, its value depending on how the motor
is to be used.
Torque-to-inertia ratios of motors available commercially range from
about 40 to 120 in.-1 with the ratio decreasing for motors of larger power
rating. Motors designed for 400 cps generally have a somewhat lower
figure than 60-cps motors. Torque-squared-to-inertia ratios vary from
about 10 to over 60 lb. It should be pointed out that, although figures of
merit are convenient means for comparing motors, no single figure of this
sort can be expected to express the merit of the motor adequately in all
situations. Thus, figures such as ratio of stall torque to watt input, etc.,
are sometimes more important, and, of course, in commercial applications
cost is a very important consideration and may often dictate the use of a
motor whose characteristics, as measured by the figures of merit described
here, are quite inferior.
7.7. Motor Characteristics in the
Presence of Finite Control Imped¬
ance. In the development carried
out thus far it has been assumed
Fig. 7.11. Servomotor controlled by
that the source of control voltage
source having finite output impedance. has zero output impedance. In
general, when a motor is controlled
by an electronic amplifier, the output impedance is not zero, and the con¬
trol voltage Vc is then a function of the current Ic in the control winding.
To extend the theory to embrace this possibility, let it be supposed that
the internal voltage of the amplifier is Ec = jkVM and that the amplifier
has an output impedance Zc (see Fig. 7.11). In order to keep the discus¬
sion as simple as possible, we shall assume that Ec and VM are 90° out of
Sec. 7.7] A-C MOTORS 293

phase, although in principle there is no difficulty in considering other phase

angles between the two voltages. Since the current in the windings plays
a role under the conditions assumed here, it will be found convenient to
consider the symmetrical components of the current as well as those of
the voltage. Thus, let
Ic = hi + I c2 (7.40)

The input impedance per phase presented by the motor to the positive
sequence is essentially the impedance of the equivalent circuit of Fig. 7.6.
This impedance is
Zm[(R'2/S + jx']
Z = Zi + = Z{S) (7.41)
Zm + (R'2/S + jX'2)
jRMXM
where Z\ = Ri -\- jXi and Zm =
Rm + jX M

Then the impedance presented to the negative sequence is Z(2 — S).

Hence

lcl = Z(S) (-7'42)

and Ic2 = g (7.43)

Also, from the circuit of Fig. 7.11,

Vc = Ec — IcZc — jkVm — ICZC (7.44)

From previous developments [Eqs. (7.12) to (7.14)] we have

Vd = -jVM1
Vc2 = jV M2
yc = b cl + Vc2 — ~j(VMl — VM2)
Hence Eq. (7.44) becomes

F M2 V Ml
— kV M
m — Zc — y M2 — Vmi (7.45)
[Z(2 - S) Z(S)_
From Eq. (7.11)

VM2 — Vm — Vmi

so that finally
1 + k + [Ze/Z(2 - S)]
Vmi = V (7.46)
2 + [ZC/Z(S)] + [Zc/Z(2 - S)]
and similarly
1 - k + [Ze/Z(S)]
VM2 ~ Vm (7.47)
t
2 + [ZC/Z(S)] + [Zc/Z{2 - S)]
294 CONTROL SYSTEM COMPONENTS [CHAP. 7

The developed torque can now be found from Eq. (7.23) and is given by
the expression
1
Q = Qb(S) l + k +
Z(2 - S)
2 + Z(S) 1 Z{2 - S)

- Qb(2 - g) 1 - k + (7.48)
Z(S)
It is apparent that the presence of Zc results in a rather complicated
expression, especially since it is now necessary to know the input imped¬
ance of the motor in addition to the speed-torque curve for balanced volt¬
ages. The remarks made concerning the difficulty of computing Qb(S)
apply with even greater force to the computation of Z(S), since it is
necessary to find not only the magnitude but also the phase angle as a
function of S. This information is not ordinarily supplied by the motor
manufacturers. Thus, unless a number of rather drastic simplifying
assumptions are made or unless Z(S) is measured directly, Eq. (7.48) is
difficult to use directly. Figure 7.12 gives a typical set of speed-torque
curves for a motor assumed to have the very much simplified equivalent
circuit shown; note that the curves for Zc = 200 + jO are considerably
less regular than those for Zc — 0.*
It is interesting to note that the addition of Zc does not change the fact
established previously that the blocked-rotor torque is a linear function
of k. If in Eq. (7.48) we let S = 1, we obtain, after some simplification,

1 T R
Q — kQb o (7.49)
1 + (Zc/Z) I*

where R is the real part of Zc/Z. For speeds close to zero the approxima¬
tion holds that Z(S) ~ Z(2 — S) « Z( 1), so that Eq. (7.48) can be
simplified somewhat. The transfer function of the motor, although basi¬
cally unchanged, is, of course, also more difficult to find. In particular,
the determination of A and B by the “exact” method becomes consider¬
ably more cumbersome. It is, however, apparent from the appearance
of the curves shown in Fig. 7.12 that both the gain and damping for the
operating point k = 0, S = 1 will be reduced by the presence of Zc in the
control circuit.
7.8. Unbalanced Stator Windings. Quite commonly, motors are built
with more turns on the control winding than on the reference winding.
The advantage of this construction is that the voltage rating of the con¬
trol winding may be made high enough to make it possible to connect the
motor directly to the output tubes of an electronic power amplifier with-
* The equivalent circuit and the curves are adapted from Koopman’s paper referred
to earlier.
Sec. 7.8] A-C MOTORS 295

out an output transformer. The analysis of such a motor follows directly

from the analysis of the motor with equal windings.
Suppose that the ratio of reference current to control current required
to set up a uniform rotating magnetic field in the motor is N. N will be
equal to the turns ratio of the windings if the windings differ only in the
number of turns and are in all other respects the same. In any case, N

Fig. 7.12. Speed-torque curves for simplified equivalent circuit. (From Koopman)

may be thought of as an equivalent turns ratio. If now the ratio of input

impedance of the control winding at any given speed to input impedance
of the reference winding for the same speed is equal to N2 for all S, then
the motor behaves exactly as though it had equal windings, but with a
transformer of turns ratio N connected between the control winding and
the control-voltage source. Such a motor may, therefore, be treated like
a motor with equal windings with the equivalent control voltage Vc/N
applied to it. If the ratio of control and reference impedance differs from
N2, we may consider the difference due to a series impedance (which may
296 CONTROL SYSTEM COMPONENTS [Chap. 7

turn out to be negative) in the control winding, and the motor may then
be analyzed as if it were a motor with equal windings but with an equiv¬
alent Zc in the control winding.
7.9. Single-phasing of an Induction Motor. Single-phasing refers to
operation of a polyphase induction motor on a single-phase supply. A
motor will not single-phase if, when driven by an external torque, it
always develops an opposing torque, i.e., if it has a positive damping
coefficient. Conversely, a motor that develops torque in the direction of
rotation may single-phase if the developed torque exceeds the friction
torque; such a motor exhibits negative damping. Although it is theoreti¬
cally possible to stabilize a servo with a motor having negative damping,
it is difficult and requires relatively complicated control circuits. Single¬
phasing is therefore considered undesirable in a servomotor. Since an
ordinary induction motor will run on single-phase, once brought up to
speed, it is necessary to incorporate special design features in motors
intended for servo use to prevent them from single-phasing. These
features are discussed in the following paragraphs. Two cases are con¬
sidered: (1) the control phase is short-circuited, and (2) the control phase
is open-circuited.
If the control phase is short-circuited, Vc = 0; hence k = 0, and by
Eq. (7.24) the torque developed by the motor becomes

Q = 34lQb(S) - Qb(2 - S)] (7.50)

If the motor is not to single-phase, it is necessary that the torque be nega¬
tive when the speed is positive, and vice versa. Since for positive speed
0 < S < 1, we require
Qb(S) - CM2 - S) < 0 for 0 < S < 1 (7.51)
But if $ < 1, 2 — S > 1; hence the requirement is met if the speed-torque
curve for balanced input is such that the torque for all negative speeds
exceeds the torque for corresponding positive speeds. In other words,
the peak in the speed-torque curve must occur for negative speeds. Fig¬
ure 7.13a shows such a speed-torque curve and the corresponding speed-
torque curve for k = 0; in Fig. 7.135 are shown the speed-torque curves
for a standard induction motor to illustrate the difference. The speed-
torque curve in Fig. 7.13a is characteristic of a motor with a high rotor
resistance. This may be demonstrated by differentiating Eq. (7.26) to
find the value of S for which Qb(S) has a peak. It is found that at the
peak

S2 = (Rjy_ (7.52)
7tV + (X£ -T Xo)2
so that, if the peak is to occur for S > 1,
(R'2)2 > Ro2 + (X' + X0)2 (7.53)
Sec. 7.9] A-C MOTORS 297

The large rotor resistance required to prevent single-phase operation is

one of the main reasons for the low efficiency of servomotors.
When the control winding of the motor is open-circuited, the voltage
Vc is not, in general, equal to zero, but Zc = oo. Hence, using Eqs.
(7.46) and (7.47), we obtain
Z(S)
VMi vM (7.54)
Z(S) + Z(2 - S)
Z(2 - S)
and VM2 Vm (7.55)
Z(S) + Z(2 - S)
and therefore, by Eq. (7.23),

Q = + 2,(2 - gyp KM«>IW ~ Q»(2 “ S^Z{-2 ~ ^ (7-56)

The requirement here becomes, therefore, that

Q6(£)|Z(£)|2 < Qb(2 - S) |Z(2 - £)|2 0<S<1 (7.57)

if the motor is not to run on single-phase. Z(S) is defined in Eq. (7.41)

and increases as S decreases. Hence, even though Qb(S) < Qh(2 — S),

Torque

(^7) Speed-torque curves of servo motor (b) Speed-torque curves of standard

showing positive damping induction motor showing negative
damping
Fig. 7.13. Speed-torque curves of a-c motors.

inequality (7.57) is not necessarily satisfied. A motor that will not

single-phase when the control phase is short-circuited may therefore sin¬
gle-phase when it is open-circuited. The tendency to single-phase is
reduced by making Qb(S) as much smaller than Qb(2 — S) as possible, i.e.,
by using a large rotor resistance. Another technique is to reduce the
dependence of Z(S) on S. By inspection of the equivalent circuit (Fig.
7.6) it is seen that a relatively small ZM and a relatively large R2 will have
this effect. In other words, if most of the current going into the motor is
magnetizing current, changes in the load component of input current due
to changes in speed will be relatively negligible. Since a large R2 also
serves to make the speed-torque curves more nearly linear, it is found
that most servomotors, particularly the smaller ones, where the power
298 CONTROL SYSTEM COMPONENTS [CHAP. 7

loss is of no consequence, employ rotors with large resistances, and the

magnetizing impedance is usually considerably less than in conventional
induction motors.
When a motor is operated at a frequency higher than the design fre¬
quency, all reactances are increased by the ratio of frequencies. Inspec¬
tion of Eq. (7.52) indicates that this causes the slip for which maximum
torque is developed to become smaller; and if the frequency ratio is suffi¬
ciently high, the peak may be moved into the positive-speed region.
Also, an increase in magnetizing reactance increases the dependence of
Z(S) on (S). Therefore, a motor that does not single-phase at its design
frequency may very well single-phase if operated at substantially higher
frequencies. It is therefore not good practice to operate a servomotor at
frequencies higher than the design frequency, even though the windings
would not be damaged.
7.10. The Induction Motor as a Tachometer. When the reference
phase of a two-phase induction motor is excited and when the rotor is
turned by some outside source, an a-c voltage appears at the control-field
terminals. This voltage may be used as an indication of speed, and a
motor so used becomes a tachometer. When a motor is designed specifi¬
cally for use as a tachometer, its rotor is of light construction in order to
hold the added inertia to a minimum. For this reason, tachometers are
usually built with drag-cup rotors. These have the additional advantage
of having no slot ripple in the output, such as is observed with squirrel-
cage rotors. The large air gap necessitated by the drag-cup construction
is also an advantage, as will be seen presently.
The voltage generated may be obtained from the theory developed in
the last section.1 If the control winding is open-circuited, Zc = <x>. As
has already been demonstrated,

_ Z(S)
V Ml = VM (7.54)
Z(S) + Z(2 - S)
Z(2 - S)
V M 2 = VM (7.55)
Z(S) + Z(2 - S)

The generated voltage is Vc and is given by

V c — Vcl + V C2 — — jV Ml + jVM2 (7.14)

Z(S) - Z(2 - S)
or Vc = -jV M (7.58)
Z(S) + Z(2 - S)

Equation (7.58) can be evaluated by making use of the expression for

1 For an extensive treatment of the a-c tachometer, see R. H. Frazier, Analysis of

the AC Tachometer by Means of Two-phase Symmetrical Components, Trans. AIEE,
vol. 70, pp. 1894-1906, 1951.
Sec. 7.10] a-c motors :299

Z(S) given in Eq. (7.41). The result, after some reduction, becomes

i - s
yc — -jvmKZm2 (7.59)
Ki + K2S(2 - S)
where

k, ± Ro[(Zi + zM)(zM + K +jX't) + zyzM + jx'2)]

K. 2 = [Z.Zm* - Z.X’fi - ZmX? + jZMX'2(ZM + 2Zi)]
^1 = ^1+ jXx
Zm — JRmX m/ (Rm + jX m)
By Eq. (7.2) 1 — S = ft/0,; hence Vc is seen to be proportional to
speed provided K2 is negligible.
Since Eq. (7.59), as it stands, is rather complicated, we make use of the
fact that most tachometers use drag-cup rotors, for which the rotor react¬
ance is negligible1 and the exciting impedance ZM is relatively small and
almost purely reactive. We suppose, therefore, that

X' = 0
ZM jXm R2 (7.60)
Z\ -f- Zm Z\ = R\ + jX i
Then Eq. (7.59) becomes approximately

XM2 (fi/«.)
Vc - jV M (7.61)
(Ri+jXJRUl + [(XM/R'2)Wtt8)]*}
Note that the nonlinearity is of the “saturating” type; i.e., the voltage
at high speeds is less than the value that would be obtained if the tachome¬
ter were linear. Note also that the phase angle
between Vc and Vm depends on the power-factor
angle of the stator and becomes 90° as the power-
factor angle approaches zero.
The explanation of the operation of an a-c
tachometer given above is based essentially on
the so-called double-revolving-field theory of the
operation of single-phase induction machines.
Fig. 7.14. Flux patterns
While the results obtained are correct, a somewhat in the a-c tachometer.
clearer physical picture of the operation is
obtained by use of the cross-field theory. According to this theory, the
currents flowing in the main stator winding set up an alternating flux
<j>M in the air gap (see Fig. 7.14). When the rotor rotates, two types of
voltage are induced in it by the main flux, a transformer voltage Em and
a speed voltage Eq. This effect is explained in some detail in Sec. 4.4,
and it is shown there that the speed voltage Eq is in time phase with the
1 Koopman, op. cit., p. 322.
300 CONTROL SYSTEM COMPONENTS [Chap. 7

flux 4>m, is proportional to the speed of rotation, and is a maximum in the

rotor conductors which are shown shaded in Fig. 7.14. The speed
voltage results in a current flowing in the rotor, and this current, in turn,
results in an alternating magnetic flux, called the cross field, <f>c. The
cross-field flux then induces the output voltage in the turns of the output
winding by ordinary transformer action. The speed voltage in the rotor,
Eq, is directly proportional to speed if <}>m is constant. Thus the resulting
current, the cross flux, and the output voltage should also be directly
proportional to speed. The main flux (f>M may be assumed to be constant
if the magnetic reaction of the rotor on the stator is negligible. This
would correspond to a very small XM and a large R2, and it is seen in Eqs.
(7.59) and (7.61) that these are indeed the conditions for linearity.
Although a sudden change in speed results in an instantaneous increase
of the speed voltage, it requires a short time for the resulting rotor current
to be established unless the rotor reactance is negligible. This time delay
results in a small time lag between the speed change and the resulting
change of output voltage. It can be shown, however, that the resulting
time constant is of the order of one period of the exciting frequency and
is, therefore, normally negligible.
The primary advantage of the a-c tachometer over its d-c counterpart
is the almost complete absence of commutator ripple. It is true that the
output is an a-c voltage and must, therefore, usually be passed through a
phase discriminator. However, for any speed, the output frequency is
constant and thus can be filtered as much as is desirable. This is not
possible with commutator ripple, since its frequency is proportional to
speed, and a filter which works satisfactorily at a relatively high speed will
be unsatisfactory at lower speeds. Additional advantages are the absence
of brush friction and brush bounce. On the other hand, it is easier to
keep a permanent-magnet d-c tachometer calibrated, particularly if the
magnet has been properly aged. With a-c tachometers the output volt¬
age is directly proportional to the reference voltage, and it therefore
requires a closely regulated supply to maintain calibration. Furthermore,
both R i and R2 vary with changes of temperature, with immediate effect
on calibration. Thus, when a tachometer is used to indicate absolute
speed to a high accuracy, the d-c type is probably preferable, but for use
in subsidiary stabilizing loops of servomechanisms, where calibration
accuracy is relatively unimportant but where commutator ripple is highly
objectionable, the a-c tachometer seems preferable.
7.11. Other A-C Servomotors. The Shaded-pole Induction Motor.
Shaded-pole induction motors are used quite extensively to drive desk
fans, phonograph turntables, and many other small-power devices in
common use. Motors designed for these applications usually have a
single, fixed shading coil per pole, often consisting simply of a heavy cop-
Sec. 7.11] A-C MOTORS 301

per ring. Neither their direction of rotation nor the developed torque
can, therefore, be changed easily. To make shaded-pole motors reversi¬
ble, they are built with two shading coils per pole. These coils are wound
with a relatively large number of turns, and their ends are brought out to
external terminals so that the current in the shading coils can be con¬
trolled. An exploded view of such a motor is shown in Fig. 7.15. Note
that a standard squirrel-cage rotor is used.

Fig. 7.15. Shaded-pole servomotor. (Barber Colman Co.)

The primary advantage of this motor is its very much lower cost com¬
pared to standard servomotors. This makes the motor useful in many
noncritical commercial applications where low cost is a primary consider¬
ation. In such applications the motor is often used with relay control of
the shading coils, and in this way it is possible to design simple follow-up
systems employing no amplifiers and only the simplest and most rugged
components. Somewhat higher performance can be obtained by elec¬
tronic control of the shading-coil current. For best results the shading-
coil current should be 90° out of phase with the current in the main field
winding, and under these conditions the operation of the motor resembles
that of a two-phase motor.
302 CONTROL SYSTEM COMPONENTS [Chap. 7

The performance of commercially available shaded-pole motors as

measured by such criteria as the torque-to-inertia ratio or the torque-
squared-to-inertia ratio is considerably poorer than that of conventional
motors. Typically the torque-to-inertia ratio of a Kearfott type Rill
motor is more than four times as high as that of a shaded-pole motor of
comparable rating. Also, owing to the salient-pole construction of
shaded-pole motors, the speed-torque curve tends to have irregularities
not found in motors with well-distributed windings. Thus the shaded-
pole motor is not ordinarily used in high-performance systems.
Qualitatively the operation of the motor is usually explained by noting
that the action of the short-circuited
shading coil delays the flux passing
through it relative to the flux passing
through the unshaded portion of the pole.
This delay causes the peak of the flux
wave to move from the unshaded to the
shaded part of the pole, and this motion
of the flux results in torque on the
squirrel-cage winding as in the polyphase
induction motor. Hence, if in Fig. 7.16
coils a and d are short-circuited, rotation
Fig. 7.16. Shaded-pole servomotor will be in the clockwise direction; if b
showing control coils. and c are short-circuited, the direction is
reversed.
Quantitative analyses of the shaded-pole motor have been made by
Trickey, Kron, and Chang.1 Most of these analyses are rather complex,
primarily because the salient-pole construction of shaded-pole motors
makes an analysis based on sinusoidal air-gap flux distribution rather
unrealistic. It is found empirically, however, that the speed-torque
curves of many commercially available motors, when controlled by an
electronic circuit, such as the one shown in Fig. 7.17, are quite regular and
are similar to the characteristics of standard two-phase induction motors
(see Fig. 7.18). Hence approximate transfer functions may be derived
for shaded-pole motors in the same way as for two-phase motors (see
Sec. 7.5). Since the applications in which these motors are used are not
apt to be very critical, only the most approximate method seems justi¬
fied. Thus, assume a straight-line relation for the speed-torque curve,

1 P. H. Trickey, An Analysis of the Shaded-pole Motor, Elec. Eng., September,

1936, pp. 1007-1014; Performance Calculation of Shaded-pole Motors, Trans. AIEE,
vol. 66, pp. 1431-1438, 1947. G. Kron, Equivalent Circuits of the Shaded Pole
Motor with Space Harmonics, Trans. AIEE, vol. 69, pp. 720-727, 1950. Chang,
Equivalent Circuits and Their Applications in Designing Shaded-pole Motors, Trans.
AIEE, vol. 70, pp.‘690-698, 1951.
Sec. 7.11] A-C MOTORS 303

such as is shown by the dotted line in Fig. 7.18a; then by analogy with
Eqs. (7.30), (7.38), and (7.39), we have

(&m/Vc max) Vc ~ {&m/Qo)Q

(7.62)
J (Qm/Qo)s + 1
where Vc max is the maximum voltage applied to the shading coil, J is the
moment of inertia, Vc is the voltage applied to the shading coil, and Q is

Fig. 7.17. Electronic control circuit used with shaded-pole motor.

Torque Torque

io) Speed-torque for maximum {£>) Torque versus voltage

voltage
Fig. 7.18. Typical characteristics of shaded-pole motors.

the applied torque. Q0 and SlM are the stalled torque and maximum
speed, respectively, as determined by the straight-line approximation of
Fig. 7.18a.
PROBLEMS
7.1. Plot the speed-torque curve for a two-phase induction motor having the equiva¬
lent circuit shown in Fig. 7.19. The voltage per phase is 50 volts, and the two voltages
are 90° out of phase. The frequency is 60 cps, and the motor is wound for two poles.

40sl j 120 sv.

o-WWV---

Fig. 7.19
304 CONTROL SYSTEM COMPONENTS [Chap. 7

7.2. Assume that one phase of the motor of Prob. 7.1 is supplied with a fixed 60-cps
voltage of 50 volts and that the variable voltage applied to the other phase is 90°
out of phase with the fixed voltage. Plot speed-torque curves for variable voltages of
0, 10, 20, 30, and 40 volts.
7.3. Find the constants Ci and C2 [see Eq. (7.28)] for the motor of Prob. 7.1.
7.4. Find the approximate motor transfer function for the motor of Prob. 7.1 on
the assumption that the moment of inertia of the rotor is 0.1 oz-in.2 Consider two
operating points: (a) speed = zero; variable voltage = zero; (b) speed = zero,
variable voltage = fixed voltage = 50 volts. Variable voltage lags fixed voltage
by 90°.
. .
7 6 The following data are available on the Diehl FPE-25-11 motor:

Output, watts. 5
Poles. 2
Reference volts. 115
Control volts. 115
Frequency, cps. 60
Locked torque, oz-in. 5.5
Torque at 1,800 rpm, oz-in. 3.8
Moment of inertia, oz-in.2. 0.098

Assume that the torque equation for 115 volts on both phases is given by

T = Ci + C& s ~ slip

(a) Find the quasi-linear transfer function for the operating point of zero speed, zero
control voltage. (6) Find the quasi-linear transfer function for the operating point
of zero speed, full control voltage, (c) Find the approximate transfer function using
the straight-line approximation to the speed-torque curves. The control voltage is
90° out of phase with the reference voltage in all cases.
7.6. Assume that the motor of Prob. 7.1 is used as an a-c tachometer. If the
reference voltage is 50 volts, find the numerical relation between velocity and output
voltage.
 CHAPTER 8

MECHANICAL NETWORKS AND GEARS

8.1. Introduction. Most feedback control systems have some

mechanical components, as distinguished from electric and hydraulic
components, etc. However in this discussion the classification mechani¬
cal element will be restricted to elements that actually perform a func¬
tional service in the loop such as sensing speed or performing addition or
subtraction. The advantages of mechanical components are reliability,
environmental stability, and, surprisingly, sometimes size. Certain
mechanical filters have been constructed that are smaller than the equiv¬
alent electric filter and superior to the electric filters in performance. In
general, however, mechanical elements are heavier and bulkier than other
types of elements, although if the power supplies required for pneumatic,
electric, and hydraulic systems are included in the weight and size calcu¬
lations, mechanical systems appear to better advantage. We shall con¬
sider in this chapter simple arrangements of mechanical elements and in
the following chapter somewhat more elaborate configurations.
8.2. Springs, Masses, and Dashpots. The equation that relates force
and mass may be written as

S = Ma = M~ (8.1)

where a is acceleration in the x direction. The Laplace transform of this

relation is
3 = Ms2x (8.2)

The simple spring is defined as a device in which the force is linearly

related to the deflection:
% = Kx (8.3)

In the simple dashpot, the reaction force is proportional to relative veloc¬

ity between its terminals:
% = Bsx (8.4)

Figure 8.1 gives the conventional representation for these devices.

While mechanical systems may be analyzed directly, it is occasionally
useful to employ a mechanical-to-electrical analogue for ease in solution
305
306 CONTROL SYSTEM COMPONENTS [Chap. 8

of mechanical systems. Analyzing mechanical systems in terms of elec¬

trical quantities permits use of the extensive work done on electric filters
and is convenient if a computer is available to implement the solution.
In Fig. 8.2 are shown three systems, two electric and one mechanical,
with their describing equations. Since all the equations are of the same
form, their solutions are identical.
The mechanical system may be compared with either of the electric
networks. The series or impedance analogue compares the mechanical

■7777777/

(O) ib) lc)

Fig. 8.1. Conventional representation of mass, spring, and dashpot.

system with the first of the electric networks. Mechanical force is then
analogous to voltage. The second, and more modern, analogue is the
mobility analogue, in which mechanical force is analogous to current.
As Firestone1 points out, the mobility analogue is the more convenient of
the two because the schematic diagrams look the same. Furthermore,
current is more closely allied to force than is voltage, since the methods of
measurement are similar. In Table 8.1 is shown a comparison of elec¬
trical quantities and analogous mechanical quantities.
Electric Electric Mechanical
11 Force

M
I tzJ
777777,
O
O
O
O

e = L^+Ri +■£■fidt i=cft + Ge+£/edt Fm=Mdft ++k/vdt

Fig. 8.2. Three analogous circuits with their equations.

By analogy to electric circuits or directly from the law of conservation

of energy, we may write the relations for mechanical elements. Mechani¬
cal elements connected in parallel must have the same velocity and posi¬
tion across them, and elements in series must have the same force through
1 F. A. Firestone, The Mobility Method of Computing the Vibration of Linear
Mechanical and Acoustical Systems: Mechanical-Electrical Analogies, J. App. Phys.,
vol. 9, p. 373, 1938. •
Sec. 8. 2] MECHANICAL NETWORKS AND GEARS 307
Table 8.1. A Comparison of Electrical and Mechanical Quantities

Electrical Electrical
Mechanical (mobility analogue) (series analogue)

Force, J. Current, i Voltage, e

Velocity, v. Voltage, e Current, i
Mass, M. Capacitance, C Inductance, L

Dashpot (viscous damping), B. Resistance, G = -jj Resistance, R

Spring, S = 1 /K. Inductance, L Capacitance, C

them. We may add the velocities of components in series in order to

find the total velocity. As an example, consider the transfer function
from the input to output of the elements shown in
Fig. 8.3. Since the elements are in series the same jx input

force must be transmitted through each, if they are

to be in equilibrium.
/ output
Thus:
S total = SF.dashpot = Tspring is ° °
n o o
or
3wi = Bs(x — y) = Ky (8.5) 77777777

Fig. 8.3. Dashpot and

Since the two elements are in series, their relative
spring in series.
velocities may be added to obtain the total velocity
of the point x. By relative velocities is meant the velocity across each
element, or the velocity that one end of the element has compared with
its own other end.

Vjo-
4'
M
lT
o o
M y o o
O o is
£3
O O'* 77//////.

m
B V/////////7////////.

7Z//7.

(o) U)
Fig. 8.4. Series mass, spring, and dashpot.

The transfer function of the system shown in Fig. 8.4 is most easily
obtained by rearranging Eq. (8.5) as follows:

Bsx ~ (Bs + K)y (8.6)

308 CONTROL SYSTEM COMPONENTS [Chap. 8

so that
y _ Bs _ (B/K)s
(8.7)
x Bs + K (B/K)s + 1

One more example may clarify a point concerning the handling of mass.
In Fig. 8.4a is shown a series-connected mass, spring, and dashpot. It
should be apparent that there can be no relative velocity across the mass.
One end of the mass is moving at exactly the same velocity as the other
end. The mass moves only relative to ground; this is the implication of
the schematic representation of mass shown in Fig. 8.46, which is identical
to Fig. 8.4a. Thus the mass in no way affects the transfer function of the
system for position or velocity.1 This may be more clearly seen by
redrawing the circuit as shown in Fig. 8.46. We may find the transfer
function in exactly the same manner as above.

I = (B/K)s + 1 (8'8)

If the relationship between input force and output force is desired, a

load must be assumed and the mass must then be considered.
8.3. Mechanical Equalizers. In addition to their use as the usual
system elements, springs and dashpots are also sometimes installed in
systems in order to modify the system transfer function to improve
closed-loop operation. These combinations are called controllers, or
equalizers, and are analogous to their electrical counterparts.
Figure 8.5 is a table of transfer functions of such equalizers using only
springs and dashpots.2 Note that several of the configurations result in
the same transfer function. This adds flexibility to the design, since it
may be possible to utilize a portion of the original system in the control
network and several choices of configuration increase this possibility.
There is no theoretical reason for not extending the use of mechanical
elements to the more complicated networks such as the bridged T and
twin T used in carrier servos. Figure 8.6 shows a bridged-T network.
Figure 8.6a is the electric network, and Fig. 8.66 is its mechanical analogue.
The twin T can likewise be constructed with mechanical components.
Figure 8.7 shows this configuration.3 Since complete design data are
available for the electric networks (see Chap. 1) used in carrier systems,

1 Electrical engineers who wish to compare this circuit to an electrical analogue

should remember that since the mass can show relative velocity only with respect to
ground, it is equivalent to a capacitor connected to ground. The electrical analogue
of the circuit is a capacitor in parallel with a coil and resistor in series.
2 J. E. Gibson, Fourteen Ways to Construct Control Functions Mechanically,
Control Eng., vol. 2, pp. 05-69, May, 1955.
3 Truxal, “Automatic Feedback Control System Synthesis,” McGraw-Hill Book
Company, Inc., New York, 1955, p. 390.
Sec. 8.3J MECHANICAL NETWORKS AND GEARS 309

a diagram, log gain

Schematic magnitude vs. log fre¬
diagram Transfer function* quency in radians/sec
(asymptotic)

(1) Y Ts
J “ 1 + Ts

(2) 1
1 + Ts
B
K

(3) T2 1 + TlS
X T i 1 + T#
A B\ A B\
1 Ki 2 Kx + A2

(4) Y Ki
X Ai + K 2 1 + Ts

rji _A_ B2
Ai + A2

Y 1 + T 2s
X 1 -T T\S
A B i -T B 2 Bi
T2 =
AT

(6) Y T is
J ~ 1 + 7Ts
Ai + A2
2 irT~

Fig. 8.5. Table of mechanical equalizers.

310 CONTROL SYSTEM COMPONENTS [Chap. 8

a diagram, log gain

Schematic magnitude vs. log fre¬
Transfer function*
diagram quency in radians/sec
(asymptotic)

(7) F 1 + TV
1 ~ 1 + Tl8 log &<
A b2 b2 A B,
Tx = — + — t2
Ai A2 K2

(8) Y T is
~ 1 4 T2s -

A Bl
T2 = — + —
Kx K2

(9) Y T2 1 + TV
J ~ Til + T2s

qi
A b2
JL _“ qi
a
A_ _1
BxB2_
' K2 2 A2(£x + 5a)

(10) F Bx 1
X B i -f- B2 1 T 7’s
A
qi .A. BxBo
1
Ki(Bi + B2)

F Bx 1 + TV
X B\ + B2 1 + T2s
^ a #2 ni (Ax + K2)B\B2
l 1 — l2
k2 A\K2{B\ -\- B2)
See Note 1

(12) F AT 1 + TV
I

2 1 + TV

rji A Bx1 rp ± Bx + B2
'_ if, * K, + K,
See Note 1

i
Fig. 8.5 (Continued)
Sec. 8.3] MECHANICAL NETWORKS AND GEARS 311

a diagram, log gain

Schcmat ic magnitude vs. log fre¬
Transfer function*
diagram quency in radians/sec
(asymptotic)

(13) Approximate (see Note 2):

Y_Tis
£ J ~ (1 + 7V)( 1 + T2s)
Exact:
Bz^
Y_Tzs_
^ .
77 77
J ~ 1 + (7\ + 7\)s + TXT#* When > r2 breaks ore as shown.
If Ti>T\ reverse labels on breaks.

(14) Approximate (see Note 3):

Y = (1 + 7V)(1 + T2s)
X (1 4 T3s)(1 + Ths)
£2
X2
+ B2 B1B2

3 xr (Bi + B2)K2
* In the steady state s = job.
Note 1. By choice of parameters T2 can be made larger or smaller than T\, and
thus either a —1 slope or a +1 slope can be synthesized.
. Note 2. The approximation is good for Tx » T2. To find the approximate rela¬
tion B2 is considered open at low frequencies, and Bx is considered shorted at high
frequencies. See Chap. 1 for discussion of approximate design methods.
Note 3. The approximate relation is obtained by considering X2 shorted at low
frequencies and K1 open at high frequencies. Multiplied out, the approximate rela¬
tion is

Y
X

while the exact relation is

Y
X

The approximation is therefore good as long as T3 » T2.

Fig. 8.5 (Continued)
312 CONTROL SYSTEM COMPONENTS [CHAP. 8

it is probably simplest to design the mechanical networks by analogy to

the electric networks.

ia) Electric analogue U>) Mechanical analogue

Fig. 8.6. Bridged T.

(a) Electrical system

(T = -j-. T ond G in inverse henrys and mhos)

AiKo
-AW- -AW-

A3
~A/VV—^
-« ■ ^a- -

(h) Analogous mechanical system (translational motion only)

ic) Possible mechanical unit for circuit of (h)

(system filled with hydraulic fluid)
Fig. 8.7. Mechanical twin-T network. (Truxal)

8.4. Electromechanical Networks for Carrier Servos. A-c servomech¬

anisms or carrier-type systems are simple to manufacture and usually
require smaller, lighter components than do d-c systems. A-c systems,
however, usually cannot match the high performance of d-c systems
because of the difficulty of equalizing a-c systems. We have discussed
lead networks for a-c systems in Chap. 1. It will be recalled that these
networks do not produce as much phase lead as a conventional lead
equalizer in an equivalent d-c system. In addition to the smaller amount
Sec. 8.4] MECHANICAL NETWORKS AND GEARS 313

of lead available, the conventional carrier networks are quite sensitive to

shifts in the carrier frequency. A shift in carrier frequency will reduce
the theoretical phase lead and may, in fact, produce phase lag. This
sensitivity to carrier shift is on an absolute basis rather than on a per¬
centage basis. Thus in aircraft installations, where the simplicity of an
a-c servo is most appreciated, a given percentage shift in frequency of the

(a)

Hex nut
inertia weights

Ball bearing
Torque motor
rotor and
stator

Pickoff electric
spring stator
ond rotor

Magnetic flux
return
Damping rotor
Magnet
Magnet holder

(b)

Fig. 8.8. Electromechanical lead network. (From McDonald)

high-frequency carrier (400 to 800 cps) is more damaging than the same
percentage shift would be in a 60-cycle carrier system.
McDonald1 suggests that a compact electromechanical unit may be
used to demodulate the a-c signal and apply it to a mechanical d-c lead
network and then convert the signal back into a-c form. The block
diagram of such a device is shown in Fig. 8.8a, and in Fig. 8.8b is
shown a cutaway drawing of the actual device. The torque motor acts

1 1). McDonald, Electromechanical Lead Networks for A-C Servomechanisms, Rev.

Sci. Instr., vol. 20, p. 775, 1949.
314 CONTROL SYSTEM COMPONENTS [CHAP. 8

as the demodulator, and the induction pick-off acts as the modulator.

It will be noted that such a unit is light, compact, and completely free
from effects of carrier-frequency shift. Furthermore the full phase lead
available from d-c networks may be obtained. It should be noted that
this is simply an electromechanical form of the demodulator-d-c-equalizer-
modulator chain discussed in Chap. 6.
Constont field
excitation 8.5. The Lancaster Damper. A
rather common form of mechanical
equalizer is known as the Lancaster
damper. It consists of a damper plate
mounted on the shaft of a motor and a
second plate which is free to turn with
respect to the first plate, as shown in
Fig. 8.9. The intervening space is filled
with damping fluid, usually hydraulic
Fig, 8.9. The Lancaster damper.
oil.
A second type of damper, shown in Fig. 8.10, works on an electro¬
magnetic principle. The copper cup fixed to the motor shaft is quite
light and adds a negligible amount to the original inertia of the motor.
The toothed steel wheel is magnetized. The flux passes from a north
pole across the air gap to the steel ring and back across the air gap to the
adjacent south pole. When the copper cup moves in the air gap with

Fig. 8.10. A magnetic form of Lancaster damper.

respect to the flux, eddy currents are set up in the copper, which in turn
produce a retarding torque.
The frequency response of a d-c motor with an inertia load and with no
damper has been discussed in Chap. 4 and its asymptotic a diagram is
shown in Fig. 8.11a. The break occurs at a frequency of kvkt/J\Ra
radians/sec, where kv is the generated voltage constant of the motor, kt is
Sec. 8.6] MECHANICAL NETWORKS AND GEARS 315

the torque constant of the motor, Ji is the moment of inertia of the rotor
and load, and Ra is the armature resistance. The reader may show that,
with the coefficients of the damper properly selected, the new a diagram
will be as shown in Fig. 8.116. The response is thus well damped. The
empirical design of these dampers long preceded the control-system
approach to the problem and has been used with equal success with a-c
motors. The damper is used in many simple a-c or d-c systems as the
only equalizer in the loop. In certain cases the use of this device makes
possible a satisfactory response in a very simple a-c control system, where
without the damper a more complex d-c system with an electronic
equalizer would be required in order to meet specifications.
The Lancaster damper has several disadvantages. Since it is at a high-
power-level point in the loop, it loads
the motor and may absorb an ap¬
preciable portion of its rated torque.
Another disadvantage is that, after
the mass J2 (see Fig. 8.9) has reached
an appreciable speed, it provides a
torque that tends to cause the motor
to overshoot the zero-error position;
this may cause poor synchronizing
performance. Also, in the hydraulic-
type damper the effect of temperature
on the coefficient of viscous damping
will change the frequency of the
breaks on the a diagram and thus can
decrease the effective damping.
A modification of the Lancaster,
or untuned, damper is the so-called
tuned damper.1 When, in addition
to the Viscous-fluid coupling, the free plate is coupled to the shaft by a
spring, the damper is said to be tuned.
8.6. Gears. Gears are used to reverse the direction of rotation of a
shaft, to provide self-locking action, to provide a right-angle drive, etc.
The most common use is to provide a change in shaft speed. For
instance, most electric motors are designed to operate at relatively high
speeds and low torque. The typical control application for motors is just
the reverse of this. A reduction in shaft speed and an increase in shaft
torque are the usual application for gearing between an electric motor and
its shaft load.
The spur gear and pinion is the most common type of gearing in control
1 Greenwood, Holdam, and MacRae, “Electronic Instruments,” Radiation Labora¬
tory Series, vol. 21, McGraw-Hill Book Company, Inc., New York, 1948, Sec. 11.6.
316 CONTROL SYSTEM COMPONENTS [Chap. 8

Fig. 8.12. (a) Spur gear and pinion; (6) helical gear; (c) bevel gear and pinion; (d) worm
and gear.

applications.1 The small gear is called the pinion (see Fig. 8.12). This
type of gearing is simple to manufacture, since the gear teeth are at right
1 For more complete information on the design of gears see Berard, Waters, and
Phelps, “ Principles of Machine Design,” The Ronald Press Company, New York,
1955.
Sec. 8.6] MECHANICAL NETWORKS AND GEARS 317
angles with the body, or gear disk. The spur gear is a high-efficiency gear
and is thus reversible. Generally the highest gear ratio that is practical
with spur gears is about 10:1 for one mesh. Figure 8.13 gives a compari¬
son of proportions for the American Standard 20°-involute fine-pitch
system and the 14^°-pressure-angle system as modified for fine-pitch
service.1
For many years gear design has been standardized on these two pressure
angles, 14^° and 20°. The 14^° design is the older of the two standards
and was the product of practical considerations rather than engineering
research.2 As modified for fine-pitch service, the 14^° design has an
undercut pinion, which decreases the gear strength. The smaller pres¬
sure-angle design theoretically provides smaller friction and less backlash,
but in practice the improvement is negligible. In addition, tests show

20 deg 14'/2 deg

1.000 1.000 Wnrkinn

Addendum
Pd Pd
1.200 1.200
+ 0.002 Dedendum + 0.002
Pd
2.000 2.000
Working depth
Pd Pd
2.157
+ 0.002 Whole depth + 0.002
rd
0.157
—+ 0.002 Cleoronce + 0.002
Pd
Clearance A—* - Pressure angle 3.1416 Clearance u Pressure angle
--■p1" + 0.002 Circulor pitch
- 20 deg hd Pd =! 41/2 deg

(a) 20° pressure ongle (£) 14V2°pressure angle

Fig. 8.13. Comparison of the standard pressure-angle designs for fine-pitch service.
{Martin)

that scoring of the gear teeth can be eliminated by increasing the pressure
angle to 25°. Especially for small pinions the higher pressure ratio pro¬
vides higher surface durability and greater beam strength. It is possible
to design high-pressure-ratio gears with stubbed teeth to increase beam
strength and to yield a more favorable contact ratio.
1 Contact ratio is defined as the length of the path of contact between
two teeth divided by the tooth pitch (circular pitch), along the path of
contact, and it gives the average number of teeth in contact at any time.
For pinions and small gears the contact ratio for 20°-pressure-angle gears
may be more than twice that of 14^°-pressure-angle gears.3 This allows
use of a thinner gear or one made of lighter material. Furthermore,
owing to the undercut pinions in 14^° gears, the gear action is not con¬
tinuous, and the composite error is greater.4 The 20°-involute fine-pitch
system is the present American Gear Manufacturers7 Association (AGMA)
1 L. D. Martin, Instrument Gears, Machine Design, vol. 26, no. 2, p. 129, February,
1954.
2 Ibid.
3 Davison, Practical Considerations in Instrument Gear Design, Product Eng.,
vol. 22, no. 9, pp. 183-187, 1951.
4 Ibid.
318 CONTROL SYSTEM COMPONENTS [Chap. 8

standard, and indications are that even higher pressure angles may be
adopted in the future.
Gears are usually manufactured by cutting or hobbing the teeth in the
blank with special machine tools. With small pinion gears there are two
other methods that may also be used. Small pinions may be extruded,
or they may be cold-drawn. Only the softer nonferrous metals such as
bronze and aluminum lend themselves to extrusion, but any metal with
good cold-working properties may be cold-drawn.
Cold-drawing has several advantages over the two other methods.
The working strength of the teeth is increased. The method produces a
smooth, hardened surface, and the dimensional accuracy of the gear is
greater than with the hot extrusion process. Surface hardness up to
50 per cent greater and durability of 200 per cent more than that obtained
in the other processes are possible.1
The helical gear (Fig. 8.125) and the herringbone gear are modifications
of the spur gear. In the helical gear the gear teeth are slanted or helical
with respect to the gear disk. The helical gear is a smoother-running
gear than the spur gear, since the entire tooth does not make contact at
the same instant with its mate. A disadvantage of the helical gear is the
side thrust due to pressure of the mating teeth. This may be overcome
by combining a right-handed and a left-handed helical gear on the same
shaft. If the two gears are joined as a unit, the result is called a herring¬
bone gear. The herringbone gear is used only in very high power applica¬
tions such as nautical propulsion units, and is of little interest, therefore,
for control systems.
Both the spur gear and the helical gear may be convoluted into various
shapes. If the large gear is laid out flat, the result is called a rack and
pinion. If the gear shafts are not held parallel but placed at an angle,
the gear teeth must be modified, and the result is a bevel gear and pinion
(see Fig. 8.12c).
The pitch of a gear is defined as the number of teeth on a disk with a
1-in. diameter and is thus a measure of the fineness of the gear teeth. For
fractional-horsepower applications the standard production gear pitches
are 32, 48, and 64. The 48 pitch is recommended for standard applica¬
tions. For applications in which extreme smoothness is desired, the 64
pitch should be used. Where smoothness is not important and long gear
life under heavy loads is the prime consideration, the 32 pitch may be used.
A gear that is somewhat different from those described above is the
worm and gear (see Fig. 8.12d). The worm is a form of screw thread
which bears on a large gear set with its axis at right angles to the axis of
the worm. The gear is turned by rotation of the worm. The main
1 E. H. Rathbone, Advantages of Cold-drawn Pinions, Product Eng., vol. 21, no. 12,
p. 114, 1950.
Sec. 8.6] MECHANICAL NETWORKS AND GEARS 319
advantage of the worm-gear drive is that ratios of 100:1 or more may be
obtained with one set of gears. Usually worm gears are self-locking; i.e.,
the gear cannot drive the worm. In some applications this is desirable,
but in a servo control system the nonlinear action may cause stability
problems. It can be shown, however, that worm gearing can be made
fully as efficient as spur gearing1 and thus may be made reversible.
Self-locking of a worm-gear drive occurs under conditions of heavy
friction or small lead angle. The efficiency of a worm and gear may be

Fig. 8.14. Worm gearing with angles defined.

derived by considering the wedge, as shown in Fig. 8.14. The forces act¬
ing on the wedge may be expressed as components.

&z = $ cos <f> sin X + cos X (8.9)

$y — JF cos 0 cos X — /SF sin X (8.10)

Here $ is the total force between the bodies, X is the lead angle of the
worm, at the pitch diameter, 0 is the pressure angle, or the slope of the
tooth, and / is the coefficient of friction. There is also a component of
force in the z direction which tends to move the worm and gear apart, but
since this motion is restrained by the shafts, it does not enter into the
efficiency calculation. Solving Eqs. (8.9) and (8.10) simultaneously
gives
cos 0 sin X + / cos X
(8.11)
cos 0 cos X — / sin X

(8.11)

(8.12)

We shall define the efficiency of the gear train as the ratio of Eq. (8.12)
to Eq. (8.11). The result is

tan X (cos 0 cos X — / sin X)

(8.13)
cos 0 sin X + / cos X

1 Berard, Waters, and Phelps, op. cit.

.320 CONTROL SYSTEM COMPONENTS [Chap. 8

Equation (8.13) can be differentiated with respect to X and set equal to

zero to find the value of X for maximum efficiency. The result is

x = 45° + tan-1 —C (8.14)

COS <f>

The coefficient of friction, /, depends on the material and lubrication

conditions; values of/ are available in the handbooks. For typical cases
the lead angle for maximum efficiency is in the neighborhood of 45°.
For the case of the gear driving the worm, Eqs. (8.9) and (8.10) are
solved for Fy, and the coefficient / changes sign. The resultant efficiency is

cos </> sin X — / cos X

V = (8.15)
tan X (cos </> cos X + / sin X)

If the worm is self-locking, rj' will be less than zero, or

tan X < / (8.16)

cos </>

Thus self-locking is dependent on friction and/or a small lead angle. Of

course, a small lead angle is desirable since it makes the installation more
compact and permits a higher over-all ratio. The high ratio is one of the
principal reasons for choosing a worm and gear in the first place. Nichols1
states empirically that self-locking gears should be avoided in closed-loop
systems because of their unstabilizing influence. However, the objection
is not to the irreversibility but rather to the possibility of a load or
accelerating torque wedging the gears so that the worm could not turn
when driven in the normal manner. Nonlinear oscillation can occur if
the gears wedge. The drive motor may exert enough torque to break
the gears free, but the load then is accelerated and may overshoot. If
this overshooting occurs, the gears will wedge again and the oscillation
will continue.
'1 (V i 8.7. Design of Gear Trains for Minimum
Motor t f Inertia. The moment of inertia placed on
H—
-(-
J
the shaft of a motor is a factor in its time
Nz
constant. We now wish to determine how
U>2
a gear ratio affects the load moment of
Fig. 8.15. A motor driving an
inertia reflected to the motor shaft. It will
inertia load through a set of
gears. then be possible to adjust the gear ratios to
minimize this time constant. Figure 8.15
shows the configuration under consideration. Two relations may be
written for the set of gears. First from the power consideration, in a
Barnes, Nichols, and Phillips, “Theory of Servomechanisms,” Radiation Labora¬
tory Series, vol. 25, McGraw-Hill Book Company, Inc., New York, 1947, p. 130.
Sec. 8.7] mechanical networks and gears 321

perfect set of gears,

7>1 = T 2^02 (8.10

where T is torque and o> is speed. The second relation is that the speeds
are inversely proportional to the gear ratio,

COi N2
(8.18)
0J2 N1

The Newton’s law relation may be written for the output shaft load as

J = L2 = .. (8.19)
CL 2 S(ji2

Substituting for the torque and speed from Eqs. (8.17) and (8.18), we have

T \&\/&2 T1N2/N]
J = (8.20)
sojiN1/'N2 so^iN 1/N 2

The quantity Ti/s&i represents a moment of inertia at the motor shaft,

and it is related to the moment of inertia on the load shaft by

S&i
Ti
J
m
Thus the load moment of inertia reflected to the motor shaft is reduced
(8.21)

by the gear ratio squared. Therefore with only a moderate gear reduc¬
tion the reflected load moment of inertia
becomes negligible. The moment of the Motor
gears themselves is quite often not negli¬
gible, however.
r2
In order to minimize the inertia of the
gear train, the designer is tempted to place
the maximum practical ratio in the first
set of gears, thus making the effect of the
Fig. 8.16. Motor and a two-mesh
following sets negligible. Let us examine gear train.
the situation to see if this procedure is
correct. Figure 8.16 shows the configuration under consideration. The
total inertia seen at the motor shaft is

Jm -Ji + {Ji + Js) (?) +Ji (s) (s) + Jl (?) (a) (8-22)
where the J’s are the inertias of the gear disks and the r’s are the radii of
the disks. We shall assume that the reflected load inertia is negligible;
the inertia of r4 would then also be negligible, but it must be included in
the computation in order to solve for the individual gear ratios. Nor¬
malizing (8.22) to the inertia of the pinion ./1 and assuming that the
 322 CONTROL SYSTEM COMPONENTS [Chap. 8

pinion inertias Ji and J3 and the radii r4 and r3 are identical,1 we have

f (ulY (8.23)
7f = 1 + (£ + 1)fe) +7 1 v2r4/

The moment of inertia is proportional to the fourth power of the disk

radius if we assume that all the disks are of equal thickness. Hence we
may write
Jy
J1 = +l^+ ) +
=
1

+ :*! +
1 3

^
ri
ri1 \r2r4,
(8.24)

1 (8.25)
rp ?Y r2

For a given over-all ratio, rir3/r2r4 is constant. Thus, since r 1 and r3 are
constant, r2r4 is constant:

M* = = Kr
\7V over-all/

Letting
r i2 = r32 = K2
we have
Jr
J1
= 1 + K, + ^r2
r22
+ T27fl (8.26)

Differentiating Eq. (8.26) with respect to r2 and setting to zero will give
the r2 for minimum inertia:

d Jm 2r2 2 K2 4Ki
= 0 (8.27)
dr 2 \J 1 Kl rV r2i
r26 - K2W ~ 2KiK2 = 0 (8.28)

As an example, take the over-all ratio as 10 and let the pinions have a
pitch diameter of in. Then

K2 = (0.25)2 = 0.0625
(0.25) (0.25)12
and K1 = 0.39
Vio
As an approximation, neglect the r22 term, which will be about one-tenth
the constant term and will make only a 1 per cent change in r2. Then

r2 = [2(0.0625) (0.39)]* = (0.049)^ = 0.605

giving a ratio of r2/r4 = 0.605/0.25 = 2.42. The second mesh would

then make up the rest of the over-all ratio, 4.14 in this case. This

1 The pinions are chosen as small as practical and in instrument gearing are usually
identical. The more general case of power gearing, in which load capacity must be
included, is considered below.
Sec. 8.7] mechanical networks and gears 323

arrangement results in a reduction of inertia by a factor of 10 over the

case of taking the whole 10:1 ratio in one step.
Figure 8.17 is a plot which shows the reflected inertia for any ratio for
up to five meshes. Figure 8.18 is a nomograph which solves the equations
given above for any over-all ratio. It may be seen from Fig. 8.17
that the reduction of inertia is relatively slight for four and five meshes

Overall gear ratio

Fig. 8.17. The reflected inertia as a function of over-all gear ratio with the number of
meshes as a parameter. The multimesh curves are computed for optimum apportion¬
ing of the ratio between the meshes. Note the relatively small improvement of four
and five meshes over three meshes. (Courtesy Reeves Instrument Cory.)

compared with three meshes. For any over-all ratio up to 100 it is

probably desirable to use a gear train of not more than three meshes.
The number of meshes should be kept at a minimum to minimize the
effect of backlash (Sec. 8.9).
In power gear trains, as differentiated from instrument gear trains, the
design of the gear train is influenced by the strength of the materials used.
If the power transmitted through a gear train is assumed constant, the
torque increases in direct proportion to the speed reduction or gear-ratio
324 CONTROL SYSTEM COMPONENTS [Chap. 8

r-1,000

-800
-700
r 600
f-500
T400

r 300

:-200

x
4
r-100
-80 c
-70 £
“60 r
r~50
i S
cn
r-40 ^
o
=-30 'o
Z L_

r-20 >
x o
2 r n

-10

4-

5-

6—1
Fig. 8.18. A nomograph that solves the equations given in the text. Place a straight¬
edge on the right-hand scale at the over-all gear ratio. The straightedge must also lie
on the point giving the number of meshes. The straightedge then intersects the left-
hand scale at the optimum ratio for the first mesh. The process may be repeated for
the subsequent meshes. The example in the text may be checked using these curves.
('Courtesy Reeves Instrument Cory.)

reduction. Thus the pinion gears in the second and following meshes
must be made wider in order to support the increased forces. It will be
shown that the calculation for power gearing can be significantly changed
when this requirement is included.
There are two standard empirical relations that are used to calculate
Sec. 8.7] MECHANICAL NETWORKS AND GEARS 325
the required size of power gears, the Lewis beam-strength formula and
the Buckingham wear-load formula. In a first approximation1 both of
these relations can be reduced to

WDP2 = Km (8.29)

where W is the face width of the meshing gears, Dp is the pitch diameter of
the driving pinion, m is the gear ratio from motor to given pinion, and K
is a constant of proportionality. Thus it is seen that the face width
and/or the pitch diameter of the pinions must be increased as we progress
through the gear train. This increase must be made when the forces in
the gear train are an appreciable portion of the maximum allowable force
for the material and gear design used. When over-all gear ratios are
greater than 100 or when four or more pinions are used, this increase in
pinion size will result in a considerable change in the gear design.
From practical fabrication and utilization considerations, Peterson2
suggests that face width and pitch diameter of successive pinions be
chosen by the empiric relation

W_=Dp=Po (8.30)
Wo Dpo P

where P is the diametral pitch of the gear and the subscript 0 refers to the
motor pinion.
The solution for minimum inertia employing the criterion in Eq. (8.30)
is displayed graphically for various meshes in Fig. 8.19. It will be noted
that, for small over-all ratio and low number of total meshes, the results
from Fig. 8.19 are identical with those from Fig. 8.17, which was computed
for constant-sized pinions. For higher ratios, however, the use of Fig.
8.19 will result in a reduction in inertia of greater than 2 to 1.
The advantages of constant-sized pinions are, of course, all the advan¬
tages of standardization. It should be borne in mind also that, where the
force transmitted is negligible with respect to the strength of the gears,
all the gears may be reduced in thickness and Fig. 8.19 will still yield
optimum results. Figure 8.20 gives curves that allow calculation of the
individual gear meshes after the total number of meshes has been chosen
from Fig. 8.19.
A possible compromise between the two extremes of complete standard¬
ization and individual design exists for medium- and light-duty gear
trains. It is possible at the slow-speed end of the train to use gears of
lower pitch than those used at the high-speed end. The low-pitch gear
teeth are larger and stronger and are more suited for heavy duty. The
over-all cost of the gear train is kept low by employing standard gear
1 D. Peterson, Power Gear Trains, Machine Design, vol. 26, no. 6, p. 161, 1954.
2 Ibid.
326 CONTROL SYSTEM COMPONENTS [Chap. 8

f __1__1______i___L__J___III
1 2 3 4 6 8 10 20 30 40 60 80100 200 400 600 1,000
Total reduction ratio,m
Fig. 8.19. Reflected inertia as a function of gear meshes for power gears. {From
Peterson)

Fig. 8.20. Choice of individual meshes given over-all ratio and number of meshes.
Subsequent meshes are chosen by reducing ratio and number of meshes by the first
mesh choice. {From Peterson)

pitches; yet the strength requirements are met. This solution is widely
used in practice.
Quite often, the inertia of the gear train can be reduced by measures
other than optimizing the ratios. By using punched gear disks rather
than solid disks; the' inertia can be reduced to about 75 per cent of the
Sec. 8.8] MECHANICAL NETWORKS AND GEARS 327
solid-disk value. A punched gear disk is one in which much of the
material has been removed from the otherwise solid body of the disk.
Substitution of materials can work an even greater reduction. The
substitution of aluminum gears for steel reduces the inertia to 33 per cent
of its original value, and the substitution of aluminum for brass gears
reduces the inertia to 25 per cent of its original value. Of course, factors
such as tensile strength and durability must be considered when sub¬
stituting materials.
The use of nylon for medium- and light-duty gears is also a possibility.
The extreme lightness of nylon (specific gravity, 1.14) is its main attrac¬
tion, although it is also easy to fabricate, resilient, and corrosion-resistant.
Nylon has a low coefficient of dry friction, which in effect makes it self-
lubricating, and it resists wear and abrasion. Nylon’s ultimate tensile
strength is about 12,000 psi, and it should not be subjected to environ¬
mental temperatures of greater than 100 to 120°F because it rapidly
loses its tensile strength and form at high temperatures.1
8.8. Gear Ratio for Load Matching. In instrument gearing the load
is usually negligible, and the inertia of the gear train is the only factor
that must be considered. The gear ratio is chosen on the basis of accu¬
racy and/or loop gain. The motor is chosen for its time constant, includ¬
ing the inertia of the gear train.
In power applications, however,
the power requirements of the load
and the power capacity of the motor
become important. It is usually
necessary to choose the smallest
motor that is capable of supplying
the load in order to minimize size,
weight, and cost of the power pack¬
age. The over-all gear ratio will thus Fig. 8.21. (a) Torque versus speed and
(b) horsepower versus speed of typical
be chosen with this in mind. The
a-c servomotor.
number of meshes and ratios of each
mesh should be designed by the methods discussed in Sec. 8.7 for power
gear trains.
In Fig. 8.21 is given, as an example, the approximate torque and horse¬
power plotted against speed for a typical a-c servomotor. In this particu¬
lar case, the motor develops maximum power at Ns/2, but whatever the
shape of the motor speed-torque curve, the speed for maximum power
may be found. Assuming that the required load velocity is known, the
over-all gear ratio should be chosen so that the motor is operating at the
speed at which it develops maximum horsepower when the load is oper-
1 R. Zimmerli, Designing Fabricated Nylon Parts, Machine Design, vol. 26, no. 3,
March, 1954, pp. 153-159.
328 CONTROL SYSTEM COMPONENTS [Chap. 8

ating at its design velocity. The horsepower rating of the motor can be
determined by determining the speed-torque characteristic of the load
and reflecting it through the gear train to the motor shaft. For instance,
in Fig. 8.22 are shown the horsepower and torque curves for a load con¬
sisting of viscous damping and inertia.
The maximum velocity of the load and
the maximum acceleration are specified.
When the reflected load horsepower-speed
Design
horsepower curve and the torque-speed curve are
superimposed on the motor curves, the
motor output must exceed the load by a
given design factor, usually 1.5 or 2 to 1.
8.9. Backlash in Gears. Backlash in
gears is the looseness or play between the
input and the output of the train. Back¬
lash may be measured as the angular dis¬
placement through which the input can be
moved with the output fixed, or it may be
Design Load speed given as the linear distance along the pitch
velocity
circle through which a gear may move with
Fig. 8.22. Horsepower and
respect to its fixed mating gear. AGMA
versus N for typical load.
standards are given in inches and can be
converted to angles if the pitch diameter is known. The AGMA standard
backlash classifications are shown in Table 8.2, along with composite error
limits.
The composite error referred to in Table 8.2 is an effect which may be
considered separately from backlash. In the cutting of the gear teeth
and the locating of the center bore for the gear shaft, there will inevitably
develop a certain amount of eccentricity between the shaft and the pitch
circle as well as distortions of the pitch circle itself. The eccentricity
that these errors lend to the gear-tooth motion as the gear is turned is
called composite error. The distance between the extremes of this
eccentricity is the total composite error, while the variation over the
angle covered by one tooth is called the tooth-to-tooth composite error.
Backlash results in a nonlinear relation between input and output of
the gear train and has received considerable study in the past few years.1
For large-amplitude signals, backlash has little effect, but as the signal
amplitude is reduced, backlash becomes a more important factor.
If a frequency-response characteristic of a portion of a system is to be
taken, the standard procedure is to maintain a constant input magnitude
or even reduce the magnitude as the frequency is increased. Let us
1 Chestnut and Mayer, “Servomechanisms and Regulating System Design,” John
Wiley & Sons, Inc., New York, 1955, vol. II, sec. 8.1.
Sec. 8.9] MECHANICAL NETWORKS AND GEARS 329
assume that the system contains a gear train with some backlash. Since,
in most systems, the response drops off at high frequencies, the amplitude
of the motion of the input gear will be reduced as the frequency is
increased. On account of backlash, the output of the gear train will be
reduced over and above the reduction that linear theory would predict at
•

Table 8.2. AGMA Standard 236.04 for Allowable Backlash

and Composite Error in Gears
Standard Specified Backlash

Diametral 'pitch Backlash,* in.

Class A
20-45. 0.004 -0.006
46-70. 0.003 -0.005
71-90. 0.002 -0.0035
Class B
20-60. 0.002 -0.004
61-120. 0.0015-0.003
121 and finer. 0.001 -0.002
Class C
20-60. 0.001 -0.002
61-120. 0.0007-0.0015
121 and finer. 0.0005-0.001
Class D
Any pitch. No measurable backlash

Total and Tooth-to-tooth Composite Error Limits

for Spur, Helical, Worm, and Bevel Gearing

Total composite Tooth-to-tooth

Class
error, in. composite error, in.

Commercial 1. 0.006 0.002

Commercial 2. 0.004 0.0015
Commercial 3. 0.002 0.001
Commercial 4. 0.0015 0.0007

Precision 1. 0.001 0.0004

Precision 2. 0.0005 0.0003
Precision 3. 0.00025 0.0002

* Between two assembled gears at their tightest point of mesh. Backlash will be
increased when the low points of runout are in contact.

the high frequencies. Figure 8.23 shows an example of this phenomenon.

If a constant output magnitude which is large enough to eliminate the
effect of backlash is maintained throughout any tests, the effect of back¬
lash will not appear. The effect of backlash may be important to the
stability of a closed-loop system, especially if it is a high-performance
system with a rather narrow range of stability.
330 CONTROL SYSTEM COMPONENTS [CHAP. 8

Backlash can be reduced by several means. One method is to use

spring-loaded gears. A spring-loaded gear is a set of two identical spur
gears placed side by side, as shown in Fig. 8.24. One of the pair is fixed
to the shaft as usual, but the other gear is loose on the shaft. The loose
gear is attached to the fixed gear by springs. Thus the two spur gears
are set to bear on the pinion with spring tension, so that the gears are in
contact regardless of the rotation; in this manner the backlash is reduced.
Spring loading increases the rate of wear of the gears, because the gears
are always fully loaded. Further, if the springs are not stiff enough,
there will be some “give” when the train is reversed. This may lead to

Fig. 8.23. Distortion of experimental frequency response due to backlash in a gear

train.

instability. Finally the double gear increases the train inertia and
requires a longer pinion.1
A second possibility is to reduce the number of gears in the train. If
the over-all ratio is fixed, this requires a larger ratio in the first stages than
would be dictated by minimum-inertia considerations. The final choice
of the number of gear meshes is thus a compromise between reduction of
the effect of backlash and reduction of inertia. The reduction of inertia
must usually, therefore, be achieved by other means than increasing the
number of gear meshes.
Another possibility for reducing backlash is to use a gear train with a
very fine pitch and to set the shafts so that the gears bear rather tightly.
This increases bearing and rubbing friction, however, and thus cannot be
carried too far. While it is theoretically possible to adjust the operating
1 G. W. Michalec, Precision Gearing, Machine Design, vol. 27, p. 202, February,
1955.
Sec. 8.9] MECHANICAL NETWORKS AND GEARS 331

centers of conventional circular-involute gears and still obtain proper gear

action, arrangements to force the gear centers together by screw adjust¬
ments on the bearings, etc., result in rapid wear and thus reduce the back¬
lash only temporarily.
A final solution to the problem of backlash is to adopt an entirely
different gear design based on the conical involute rather than the con¬
ventional circular involute. This design is called tapered-tooth involute
gearing, or beveloid gearing. The cross section of a gear tooth developed

Outside gear has

slightly larger
diameter hole
for shaft

Totally enclosed
C spring

Fig. 8.24. Spring-loaded spur gears. Two forms of tension springs are shown.

by conventional design is constant throughout the tooth thickness, while

the gear tooth resulting from the conical involute is tapered, hence the
name. Figure 8.25 illustrates this. The conical-involute gear has
tapered tooth thickness, tapered root, and, in most cases, tapered outside
diameter.
The advantages of beveloid gearing1 are ease in holding manufacturing
accuracy, since precision tolerances in the range of 0.0002 in. total com¬
posite error are possible, unlimited meshing combinations, and freedom
of gear arrangements. This last advantage means that beveloid gears are
not designed for a given mounting angle or mounting center. Beveloid
1 A. Beam, Beveloid Gearing, Machine Design, vol. 26, no. 12, p. 220, 1954.
332 CONTROL SYSTEM COMPONENTS [Chap. 8

gears allow adjustment of mounting angle, mounting distance, and direc¬

tion of axes. This means that backlash can always be eliminated by
sliding the gears on their shafts or by adjusting the shafts. Beveloid
gears have two major disadvantages: First, they are somewhat more
expensive to manufacture than conventional gears, since automatic
machines are not used. Second, with gears with intersecting or skew
shafts, the contact between gears is theoretically a point, a fact which

(a) Conventional gear tooth

(6) Conical involute gear tooth

Fig. 8.25. Cross section of conical-involute gear tooth.

limits the allowable loads. When the shafts are parallel, however,
beveloid gear teeth have line contact and are thus equivalent in wear and
load-carrying ability to conventional gears.1 For small skew angles,
beveloid gears approach the loading potential of conventional gearing.
8.10. Ball -screw Actuator. Several possible ways of converting rotary
motion into linear motion exist. The rack and pinion and the worm and
pinion are two examples. Another example is the simple screw thread
and nut. If the nut is prevented from turning as the screw is turned, the
nut travels linearly along the screw. The ball-screw actuator differs from
1 Ibid,
Sec. 8.10] MECHANICAL NETWORKS AND GEARS 333
the nut and screw only in that the operating surfaces rest on ball bearings.
Figure 8.26 shows a ball-screw actuator. As the actuator moves along
the screw, the balls are left behind. The tube shown in the figure carries
the balls to the front of the actuator, where they reenter it. The advan¬
tage of the ball-screw actuator over conventional threads is the greatly

Fig. 8.26. The ball-screw actuator, (a) Typical ball-bearing screw assembly employ¬
ing threaded nut with single ball circuit. Projecting guide finger on return tube
deflects balls into tube for recirculation. (6) Heavy-duty ball-bearing screw assembly
which has three ball circuits and yoke-type ball deflectors. (Courtesy Saginaw
Steering Gear Division, General Motors Cory.)

increased efficiency of the former. Ball-bearing screws operate at effi¬

ciencies of over 90 per cent,1 which is more than twice as efficient as a
typical screw and nut.
PROBLEMS
8.1. Design an instrument gear train with an over-all reduction of 50 and three
meshes. What are the relative decreases in inertia with two, three, four, and five
meshes?
8.2. Design a power gear train with an over-all reduction of 50 and three meshes.
What are the relative decreases in inertia with two, three, four, and five meshes?
Compare these results with those in Prob. 8.1.
8.3. Consider the Lancaster damper shown in Fig. 8.9. Suppose that the inertia
of the motor armature and the rigidly attached damping disk is Ji, the inertia of the
floating disk Jo, and the coefficient of viscous friction B. The motor has a generated
voltage coefficient kv, a torque constant kt, armature resistance Ra, and negligible
armature inductance, (a) Find the transfer from E to 9 without the damper.
(6) Find the transfer from E to 6 with the damper, (c) Letting J\ =0.1, Jo, =
0.05, B = 0.5, Ra = 0.2, kv = 0.3, and kt = 0.3 in a consistent set of units, plot the
asymptotic a diagram for the device.
8.4. Find the transfer from E to 9 for a tuned damper. Let the spring constant be
K. Compare result to (b) in the previous problem.

1 D. A. Galonska, Ball-bearing Screws, Machine Design, vol. 27, p. 201, October,

1955.
 CHAPTER 9

MECHANICAL COMPONENTS

9.1. Introduction. In this chapter a number of mechanical compo¬

nents widely used in control systems will be considered. No attempt
will be made to include material that is available in standard works on
machine design and mechanisms; only the requirements that must be met
by these components that are peculiar to their use in control systems will
be discussed. The use of mechanical elements in computers has also been
adequately discussed elsewhere1 and will therefore not be considered here.
9.2. Differential Gears. Differential gears are used in control systems
to add or subtract the positions or velocities of two shafts. The most
common type is the bevel-gear differential shown in Fig. 9.1. Gear 1 is

^ Keyed
Ball bearing =a Input No.l

Ball
bearing
T.M
2- 4
::c.- j^-Boll bearing
Output Z?
(B =mm i N\\\m=
t 5
Keyed =2 f Input No. 2
~Ball bearing
Fig. 9.1. Bevel-gear differential.

integral with gear 2, and gear 3 is integral with gear 4. Both of these two
pairs of gears are arranged to rotate freely about the output shaft. The
bevel gears 2 and 3 mesh with gears 5 and 6. Gears 5 and 6 also ride free
of their shaft on ball bearings. Consider first that input 1 is fixed; then
gears 3 and 4 cannot turn. If input 2 is turned in the direction of the
arrow, gears 1 and 2 will also turn, as will gears 5 and 6, all in the arrow
direction. Gears 5 and 6 will thus “walk” around on the stationary gear
3, and they will carry the output shaft with them, in the arrow direction.
Now if input 2 is fixed and input 1 is turned, the output shaft will still
turn in the direction of the arrow. When both input shafts are rotated
1 Soroka, “ Analog Methods in Computation and Simulation,” McGraw-Hill Book
Company, Inc., New York, 1954, Chap. I. Svoboda, “Computing Mechanisms and
Linkages,” McGraw-Hill Book Company, Inc., New York, 1948.
334
Sec. 9.2] MECHANICAL COMPONENTS 335

simultaneously, these two actions take place simultaneously, so that the

output motion is the sum of the motions generated by input 1 and input
2, separately. Subtraction may be accomplished by defining one of the
inputs in the opposite sense or by reversing its direction with a 1:1 spur
gear. In a symmetrical differential gear train like that shown in Fig. 9.1,
there is a 2:1 reduction from either input to output because of the “ walk¬
ing ” action discussed above. The output velocity or position is therefore
one-half the sum of the two input velocities or positions.
A common application of differential gears in a control system is as the
device that subtracts the output of a servo from the input to yield the
error. Differential gears have also been used in mechanical analogue

computers as adders and subtractors. A somewhat different application

is in the electromechanical clutch that was used as the power element of
the C-l autopilot. The unit is shown in a simplified schematic in Fig. 9.2.
When power is applied to the energizing circuit, solenoids pull the clutches
into contact with fixed bearing plates, thus locking the differential and
the output shaft. When an input signal is fed into the component, one
or the other of the clutch-plate brakes is released, and that plate is
pulled into contact with the moving disk. The four solenoids are
electrically interlocked so that only proper combinations may be energized
at any one time.
If the power fails or is turned off, the clutches return to the neutral posi¬
tion, and the controls are free so that the human pilot may manipulate
them in a normal manner. Should one of the clutches fail to open during
operation, the output shaft will still be free, and finally, if one clutch
fails by sticking in either closed position, the shaft may be freed by turning
the power off. The fact that with a differential gear both clutches must
336 CONTROL SYSTEM COMPONENTS [CHAP. 9

operate for any force to be transmitted to the output shaft is considered a

safety feature.
The spur-gear differential shown in Fig. 9.3a differs in several respects
from the more common bevel-gear differential. It is quite compact, with
a width of only three gears and the necessary supporting brackets.

^B

Planet carrier

Plonet
gears

>B
Section A-A

(b) A device used in on aircraft-gun computer

Fig. 9.3. The spur-gear differential.

More important, the transfers from the two inputs to the output are not,
in general, the same, and they are not independent of the gear ratios as in
the bevel-gear differential.1 Figure 9.36 shows a more symmetrical form of
spur-gear differential used in an airborne computer. In Fig. 9.3a gear 3 is
integral with input shaft 1. Gear pair 2 and 5 and gear pair 4 and 7 are
integral. Let us assume that input 2 is fixed and that input 1 is moved
1 Svoboda, ibid.
Sec. 9.3] MECHANICAL COMPONENTS 337

in the direction shown by the arrow. Gear 1 then cannot move, and gear
3 is moved by the input shaft in the direction shown. The two gear pairs
are thus driven, and they in turn drive gear 6, which is integral with the
output shaft. If input 1 is fixed, input 2 turns gear 1, which walks the
gear pairs around the stationary gear 3. We note that the output shaft
will only move if the ratio r3/r2 is not the same as r6/r6. We have the
further constraint that, if the input 1 shaft and the output shaft are to be
collinear, as shown in Fig. 9.3a, it must be true that

r3 + r2 = rb + r6 (9.1)

From input 1 we have the relation

Output = r3 rb
Inputx r2 r6

while from input 2, in terms of the radius of gear 1, we have

Output _ r3 rb
(9.3)
Input2 r2 r6

where the positive sense for input 2 is the same as for input 1. The net
output is thus

Output = (inputi) + (— — — ) (input2) radians (9.4)

y’2r6/ v* 2 r 6/

Given either r3/r2 or rb/rb, we may then select the other ratio to give any
desired ratio of the two input-to-output ratios. While this flexibility is
useful in adjusting gains, the most common arrangement is an input-out¬
put ratio of 0.5 for both inputs. In this case, r3/r2 and rb/rb can equal 1.0
and 0.5, respectively.
9.3. The Universal Joint. The Hooke joint, or universal joint, is a
device used to couple two shafts that meet at an angle. In Fig. 9.4 is
shown a primitive form of the joint. The points Ai and A2 are the ends
of the yoke on shaft A, and the points Bi and B2 are the ends of the yoke
on shaft B. We are interested in the distortion of the angular motion
produced by this device. Let us erect a plane normal to the shaft A and
a plane normal to the shaft B, both passing through the point 0. Then
these two planes will meet at the angle a, the angle between the two
shafts. The two planes will contain, respectively, the locus of the point
A i and the locus of point B\. In Fig. 9 Ad are represented the two planes
and the loci of the two points Ai and Bx. If shaft A is moved through
an angle 6, Ai will move to A[ and Bi will move to B[. The angle AiOB1
and A[OB[ is always a right angle because it is held so by the physical
cruciform. Furthermore the projection of A[OB[ on the plane normal to
A is a right angle. This can be shown from the theorem of solid geometry
338 CONTROL SYSTEM COMPONENTS [Chap. 9

that the projected angle of two intersecting lines at right angles on a plane
parallel to one of the lines is always a right angle. The plane normal to A
is parallel to line 0A\, in fact it contains it. We have before us the prob¬
lem of finding the angle </>, or BiOB[, which the shaft B has actually
turned through. The point /3 on the line of intersection of the two planes

By

(£>) End

Fig. 9.4. The universal joint. View (c) shows the joint 90° later than (a); view (b) is
the end view of the joint with angular position as in (a).

may be defined as having the same vertical height as B[ and b[. Thus
b[(30 and B[(30 are right triangles, and we may write

b[(3
tan 6 = (9.5)
0(3

and tan </> = (9.6)

Of}
Furthermore the angle is a right angle, and we may write
Sec. 9.4] mechanical components 339

Taking the ratio of Eq. (9.6) to Eq. (9.5), we have

tan (f) _ B[(3 _ 1

(9.8)
tan 6 b[(3 cos a

In Fig. 9.5 is shown the variation of </> about its proper value as a function
of 6 for a = 45°. The maximum error of about 10° occurs at 6 = 38°.

In Fig. 9.6 is shown the maximum error in 4> as a function of a and also
the value of 6 for which this error occurs. For small shaft angles this
error is negligible. For large shaft angles two universal joints, each pro¬
viding one half the total angle, should be
used. If the yokes connected to the cen¬
tral shaft between the two universals are
coplanar, the errors of the two joints can
be made to cancel.
In a velocity system the angular varia¬
tion in a universal joint produces spurious
frequency components in the output ve¬
locity which must be considered if a care¬
ful analysis is being conducted.
9.4. Strain Gauges. A strain gauge is
a device that measures minute motions of Fig. 9.6. Maximum of <j> — 6 as a

a body under applied stresses.1 While function of a. and the value of d

for which this occurs.
strain is not usually of direct interest in a
control system, a strain gauge may be used to measure pressure or
flow in hydraulic lines, for example, by measuring the change in
pipe dimensions. While completely mechanical strain gauges exist,
such as extensometers, for example, strain gauges used in control applica¬
tions are usually required to deliver an electric output signal. One reason

1 Hetenyi, “Handbook of Experimental Stress Analysis,” John Wiley & Sons, Inc.,
New York, 1950.
340 CONTROL SYSTEM COMPONENTS [Chap. 9

for this is that the signal derived from a strain gauge is usually at a very
low level and requires considerable amplification before it is useful. One
such strain gauge is the capacitor type. It consists essentially of two
plates so mounted on the object undergoing strain that they move relative
to each other as a result of the strain.
/////;
The relative motion of the two plates
Specimen causes a small change in the value of
capacitance between them. This is best
detected by the use of a Wheatstone
bridge, as shown in Fig. 9.7. A constant
Fig. 9.7. Typical capacitor strain value of a-c voltage is applied to the input
gauge showing bridge type of
terminals. A deflection of the observed
sensing circuit.
body will change the spacing of the
capacitor and unbalance the bridge, thus causing a voltage to appear at
the output terminals. The voltage is a function of C' and thus of strain;
the relationship is given by

e = ejS: — e. __ = e. (l_— ) = e. (\_1_^ (9 9)

2 111C + C' m \2 C + C'J in \2 1 + C'/Cj v ;
The variation of output voltage as a function of capacitance C' is
sketched in Fig. 9.8. Although
the relationship between output-
voltage magnitude and capacitance
is nonlinear in general, the section
near zero output is approximately
linear. The variation of C’ is usu¬
ally so small that the assumption
of linearity gives very accurate
results. The output in Fig. 9.8 is
shown positive and negative to in¬
dicate a phase reversal at null. A Fig. 9.8. Variation of output voltage with
phase-sensitive detector is required capacitance C' in the circuit of Fig. 9.7b.
to recover this information.
A more common type of strain gauge is shown in Fig. 9.9a. It consists
of a very fine wire, commonly nichrome, bonded to the specimen with
cement and encased in plastic. As the specimen yields under stress, the
nichrome wire is stretched, and its resistance increases. As in the
capacitor case, the gauge is usually employed in a bridge circuit as shown
in Fig. 9.9c. Although with the resistance-wire type of gauge, it is pos¬
sible to use a d-c input, alternating current is usually preferred in order to
simplify amplification. Whenever alternating current is used, quadra¬
ture error must be guarded against by constructing the circuit to be as
symmetrical as possible with respect to ground. As in the capacitor-
Sec. 9.51 mechanical components 341

bridge ease, the operation may be considered linear over the normal
operating range.
Symmetry can be improved, thus decreasing quadrature error, and
nonlinearity decreased, as shown in Fig. 9.10, by arranging resistor a to
Cross section A-A

(a)
Fig. 9.9. A nichrome-wire strain gauge and bridge circuit. (■Courtesy Baldwin-Lima)

stretch as resistor b compresses. The most serious disadvantage of

nichrome-wire gauges is that they are temperature-sensitive. This
sensitivity is reduced by using identical gauges in all four arms of the
bridge. However, if one or two of the gauges are bonded to the specimen
while the others are isolated from
the specimen, temperature differ¬
entials will still occur. A possible
solution is to bond all the gauges to
the specimen. If two gauges are
arranged to expand as the others
contract, gain is increased, but the
nonlinearity is also emphasized.
It is sometimes possible to orient
two or three of the gauges at right
angles to the strain, thus immobiliz¬ Fig. 9.10. Canceling the effect of strain-
gauge nonlinearity by using two gauges.
ing them.
Care must be taken that the strain gauge measures only that quantity
desired. Thus, if pressure is to be measured in a pipe, vibration of the
pipe must be considered because the strain gauge will also measure vibra¬
tion if the vibration is allowed to reach the gauge.
9.5. The Flyweight Tachometer. One of the most reliable methods of
sensing angular velocity is the flyweight tachometer. Flyweights were
342 CONTROL SYSTEM COMPONENTS [Chap. 9

the basic speed-sensing element in the flyball steam-engine governor,

which is considered to be one of the first closed-loop control systems.
The essential principle is still used in control systems where the require¬
ment of absolute reliability forbids the use of electronic components.1
Figure 9.11 shows a modern version of the flyweight tachometer. The
flyweights whirl about the vertical center line and bear on the actuator,
which does not turn. The spring-restrained flyweight tachometer is
superior to the free tachometer because it can be made relatively free of
the effects of gravity and because the weights move through a smaller
arc, thus making the unit more compact. The tachometer is arranged to
operate an actuator of some sort, e.g., a hydraulic valve. The position of
the actuator is approximately proportional to the square of the velocity
Restraining

Fig. 9.11. A flyweight tachometer.

since the centrifugal force of the rotating mass increases as the square of
the velocity. For the configuration shown in Fig. 9.11 the torques about
the pivot point may be summed as
M/co2(a + l sin a)l COS a = (Kx + Bx + Max)b cos a (9.10)

centrifugal lever reaction lever

force arm force arm

The various terms are defined in Fig. 9.116:

Mf = mass of flyweights
Ma — mass of stem, etc.
co = speed of rotation
a = distance between pivot point and center line
l = distance from pivot point to center of mass
b = distance from actuator to pivot point
K = spring constant
B = coefficient of viscous damping
1 For instance, in the usual hydraulic airplane-propellor pitch control.
Sec. 9.6] MECHANICAL COMPONENTS 343

This equation is nonlinear, not only because of the co2 term but also
because of the presence of the sin a term on the left-hand side. This term
can cause nonsinusoidal oscillations under certain conditions. We shall
neglect this possibility and assume that the terms in the first parenthetical
expression may be represented by a “constant” K. Then taking the
transform of the simplified equation,
A

X MflK/b
A O
(9.11)
or M „s2 + Bs + K

The denominator of this expression is the familiar relation for a damped

spring-mass system. In practice, the system is overdamped, and the
natural frequency is usually quite
high compared with the response of
the system in which the tachometer
is installed, and hence it may be
neglected. Equation (9.11) is a
linear transfer function in terms of
co2, not co. Thus the variation of x
with co follows a square-law relation.
The restraining spring may be
fashioned to counteract this effect
(a so-called “hard” spring), or the
operation may be assumed linear in Fig. 9.12. Actuator position versus fly¬
some small operating range. Fig¬ weight velocity for flyweight tachometer
ure 9.12 shows a plot for an actual with nonlinear spring to give flatter char¬
acteristic. Square-law curve shown for
governor using a nonlinear spring.
comparison.
Actuator position is plotted against
flyweight velocity, in the steady state. Superimposed on the curve is a
square law characteristic for comparison.
9.6. The Gyroscope. Introduction. The gyroscope is a practical
embodiment of the spinning top. Basically the gyroscope is a disk or
wheel mounted on an axle and supported by a framework, or gimbals, so
that the disk may turn in any direction. Like the top, a gyroscope
resists changes in the position of the axis about which it is spinning; an
attempt to twist its axis in one direction results in a motion of the spin
axis in another direction. This is called precession and is discussed
quantitatively below.
Precession has been employed to stabilize ships and aircraft. A gyro
may be mounted in such a manner that it is allowed to precess in a given
plane. This precession provides a torque in a plane at right angles to the
precession plane, and this torque may be used to combat the forces that
cause an aircraft or ship to pitch or toss or roll. Naturally a gyro used to
stabilize an ocean-going vessel or an aircraft would be quite large. For
344 CONTROL SYSTEM COMPONENTS [Chap. 9

this reason modern practice is to use the gyroscope merely as an indicating

instrument or stable reference and to employ a feedback system to
actuate stabilizing elements in the vehicle.
9.7. The Gyroscope. Precession. The operation of a gyroscope may
be demonstrated by considering a disk spinning in free space. The disk
will not change its initial angular momentum unless acted upon by outside
forces. The angular velocity of the disk will be taken as cos. The sub¬
script s stands for spin axis, i.e., the axis about which the disk is initially
turning, as shown in Fig. 9.13.
By convention, angular momen¬
tum is shown in vector form as an
arrow pointing along the axis of
rotation. The arrow points in the
direction of advance of a right-hand
screw that is turned in the same
direction as the disk is turning.
Now let us suppose that a torque is
applied to the disk in such a direc¬
Fig. 9.13. The primitive gyroscope. tion as to attempt to turn its flat
side horizontal. This torque is rep¬
resented by a vector along the z axis, as shown in Fig. 9.13. By New¬
ton’s law, torque is the time rate of change of angular momentum, or

(9.12)
J — const.

This equation is a vector equation. Since the torque is a vector pointing

along the positive z axis, the change in angular velocity, du/dt, must also
point in the positive z direction, since J is not a vector quantity. This is
shown in the vector diagram of Fig.
■ Y axis
9.14, in which the y axis is taken to be
perpendicular to the plane of the
paper. We see that the effect of the
torque Tz is to rotate the spin axis ' New position of the
Z axis spin ox/s
of the gyroscope about the y axis.
Fig. 9.14. Gyroscope vector diagram.
This rotation is the gyroscopic pre¬
cession. Note that torque applied along the positive z axis causes the
spin axis of the gyroscope to rotate in the negative y direction. Thus the
spin axis tends to line itself up with the torque vector.
Figure 9.14 can be used to derive the quantitative relation between the
torque and the resulting velocity of precession. Thus let the torque Tz
act for a short increment of time, dt. Its effect is to rotate the spin axis
through the small angle —ddy. If the angle is very small, it is equal to its
Sec. 9.7] MECHANICAL COMPONENTS 345

tangent, or
T dt
— ddy = tan ( — dOy) = -J -— (9.13)
J s^s
so that
ddy Tz
(9.14)
dt Wy Jsws
where is the velocity of precession.
By similar reasoning it can be shown that the gyro precesses about the
z axis when a torque is applied along the y axis, or

dez = Ty
(9.15)
dt J §co §

Again the direction of precession is such as to line up the spin axis with
the direction of the torque vector.
Equations (9.14) and (9.15) are only approximations, because they do
not consider the fact that a change in precession velocity, multiplied by
the rest inertia of the gyro, also represents a change in angular momen¬
tum, which requires torque. Also they neglect the effects of gimbal fric¬
tion, which result in braking torques when the precession velocities
become large. The equations are therefore reasonably accurate only as
long as the precession velocity is low and the torque variations small.
There are several interesting things about the results obtained thus far.
First, the precession is along an axis perpendicular to that of the applied
torque; yet had not the disk a velocity about the spin axis, rotation would
have taken place along the axis of the applied torque. Second, the
applied torque results in a velocity rather than an acceleration, and
finally the moment of inertia about the spin axis enters into the relation
rather than the moment of inertia about the axis of the applied torque.
All of these facts make the gyroscope somewhat puzzling on first
examination.
The phenomenon of precession may be examined in another manner
that is perhaps more closely related to familiar concepts. To show the
force required for precession, the velocity of precession will be assumed,
and the required force will be found. The coordinate system fixed in
space is as shown in Fig. 9.15. Assume that the disk is spinning with an
angular velocity cos and also that the disk is precessing with a velocity
cop. Consider a particle of the disk near the rim.1 This particle has a
linear velocity parallel to the x axis, which is due only to the precession.
This linear velocity of the particle varies as the particle is carried around
by the spin velocity. At point A in space the component of velocity in
the x direction has a maximum negative magnitude, and at point C the
1 S. T. Preston, The Mechanics of the Gyroscope, Sci. American Supplement, vol. 58,
no. 1501, pp. 24,057-24,058, Oct. 8, 1904.
346 CONTROL SYSTEM COMPONENTS [Chap. 9

velocity has a maximum positive magnitude. At points B and D the

linear velocity is zero. Figure 9.156 shows a pictorial representation of
the linear velocity in the x direction of a particle of the disk. If we now
consider the linear acceleration of the particle in the x direction, we find
that at point B the velocity is changing from negative to positive and the
acceleration has a maximum positive magnitude. Throughout the semi¬
circle ABC the acceleration is positive, and on the upper half of the disk,

Fig. 9.15. (a) The disk with respect to coordinates fixed in space. (6) The linear
velocity of a particle of the disk in the x direction, (c) The linear acceleration in the
x direction of a particle of the disk.

the acceleration is negative, as shown in Fig. 9.15c. By Newton’s second

law a force must be applied to the disk in order to cause this acceleration,
or more conveniently it may be applied to the axle, as was shown in
Fig. 9.13.
This explanation of precession has the advantage of utilizing only
Newton’s second law in its simplest form. Furthermore the acceleration
is in the direction of the force, which appeals to reason, and finally the
applied force results in an acceleration.
It is possible to sum up the forces on all the particles to obtain the total
torque required to produce a given velocity of precession. Den Hartog1
1 Den Hartog,•“ Mechanics/’ McGraw-Hill Book Company, Inc., New York, 1948.
Sec. 9.8] mechanical components 347

has shown that this process gives results identical to those already
obtained.
9.8. Gyroscope. Equations of Motion. When the gyroscope is used
as a control-system component, the elementary relations between torque
and precession velocity [Eqs. (9.14) and (9.15)] are not sufficiently accu¬
rate, and the effects of gyro rest inertia and gimbal friction must be con¬
sidered. This may be done rather easily if the arrangement of the gyro
disk, gimbals, and torque motors is such that during precession the spin
axis always remains in a plane perpendicular to the torque causing the
precession. A commonly used configuration meeting this requirement
is shown in Fig. 9.16. Note that the torque of one of the torque motors

is applied to the gyro through a bail-ring having a slot in which the gyro
spin axis is free to slide. This arrangement permits both torque-motor
stators to be fixed to the outer frame. In this type of system we may
consider the torques required to accelerate the gyro rest inertia and to
overcome gimbal friction as subtracting from the “applied” torque,
leaving an “effective” torque resulting in precession only. If the gimbal
friction is assumed to be viscous, then this reasoning modifies Eqs. (9.14)
and (9.15) as follows:
rp j d~dz ddz j dOy ^ i^
T‘ - W - B° Tt ^ Hi (9J6)
rp j d Qy dOy _ j ddz
y Jy dt2 v dt ~ sC°s dt K' }
348 CONTROL SYSTEM COMPONENTS [Chap. 9

where J z and Jy are the rest inertias of the gyro and gimbals measured in
the z and y axes, and Bz and By are the coefficients of friction in these
axes. The torques Tz and Ty are the components of torque perpendicular
to the spin axis, since components parallel to the spin axis do not result
in precession. Hence if the actual torques applied by the torque motors
are T'z and T'y and if dz and 6y are measured between the spin axis and a
line perpendicular to the two torque axes, then Eqs. (9.16) and (9.17)
become
d2dz dSz j ddy
T[ cos 6y — Jz - B, (9.18)
w dt ~Js0)S~dt

d2dy r> ddy dOz

T'y cos Qz — Jy J, CO, (9.19)
dt2 - B«Hi dt
The trigonometric functions multiplying the applied torque clearly make
this a nonlinear set of differential equations which is difficult to solve in
general. In particular, it is not
Direct
possible to obtain transfer func¬
Oz tions relating the torque applied by
a torque motor to displacement of
the gyro spin axis.
In many cases of practical inter¬
est, it is true, however, that 6y and
6Z are approximately zero; i.e., the

-e.y gimbals of the gyro of Fig. 9.16 are

at right angles to each other and to
the frame. In that case the cosine
Fig. 9.17. An approximate block-diagram
functions appearing in Eqs. (9.18)
representation of a gyroscope.
and (9.19) are approximately equal
to unity, and the equations become linear. They can then be transformed
and rewritten as follows:

(J zs2 + Bzs)dz — J gCOgSOy = Tz (9.20)

{J ys2 + Bys)dy + JsussOz = Ty (9-21)
A block diagram may be derived from these equations; one form is shown
in Fig. 9.17.
When gimbal arrangements differing from that shown in Fig. 9.16 are
used, the behavior of the gyroscope may become more complicated.
Thus, consider the arrangement shown in Fig. 9.18. Here the stator of
the torque motor driving the inner gimbal is mounted on the outer gimbal.
Careful inspection of this figure will indicate that the gyro may precess in
such a way that the spin axis traces out a conical surface rather than a
plane. This happens, in fact, whenever torque is applied to the inner
gimbal by the torque motor mounted on the outer gimbal, and when the
two gimbals are not at right angles to each other.
Sec. 9.8] MECHANICAL COMPONENTS 349

The differential equations describing the motion of this type of gyro¬

scope cannot be obtained by the simple reasoning employed in obtaining
Eqs. (9.18) and (9.19). They are nonlinear equations and contain terms
involving the product of the two pre¬
cession angles in addition to trigono¬
metric terms. For a detailed dis¬
cussion of gyroscopes operating in
this way, the reader is referred to
texts on advanced dynamics.1 Here
again, if the assumption is made that
the gimbals are usually approxi¬
mately perpendicular to each other
and to the frame, no difficulty is ex¬
perienced in linearizing the differen¬
tial equations; in fact they become
identical with Eqs. (9.20) and (9.21)
derived above.
The transfer function relating pre¬ Fig. 9.18. Gyro capable of precessing
along a conical surface.
cession angle to applied torque can be
obtained by inspection of the block diagram of Fig. 9.17, or by a simulta¬
neous solution of Eqs. (9.20) and (9.21):

e2 J s^s
(9.22)
Ty s[J yj zs2 + (JyBz + JzBy)s + (JsCOs)2 + ByBz]

This transfer function is, of course, accurate only if the conditions under
which Eqs. (9.20) and (9.21) were derived are met. Normally, friction
is sufficiently small to make ByBz « Jsws; hence Eq. (9.22) is approxi¬
mately equivalent to

80, i u s^s (9.23)

T'
xy
JyJz , , JyBz ~ b JzBy
S2 + - -r-o
T 2., 9--
2 5+1
J *'-•
NO, 2 U s Ws

We note that the gyro response has the standard quadratic form, and if
friction is small, the response to step or other shock inputs is oscillatory.
The transient oscillations observed in a gyro are referred to as nutations,
since the gyro spin axis characteristically traces out a conical surface dur¬
ing these oscillations. The frequency of nutation, i.e., the undamped
natural frequency of Eq. (9.23), is seen to be

con (9.24)

1 See, for instance, Page, “Introduction to Theoretical Physics,” D. Van Nostrand

Company, Inc., Princeton, N.J., 1935, pp. 137-147.
350 CONTROL SYSTEM COMPONENTS [Chap. 9

Since Jy and Jz are usually larger than J8, the nutating frequency is
usually less than the spin velocity. In fact it can be demonstrated that
the nutation frequency of a thin disk spinning in space without gimbals is
equal to one-half the spin velocity.
The transfer function relating precession velocity to precessing torque
[Eq. (9.23)] is referred to as the direct transfer function of the gyro. How¬
ever, owing to the gyro rest inertia and gimbal friction, a torque applied to
one set of axes of a gyro also results in some gyro motion in the same axis.
The relation between torque and angular displacement along the same
axis is referred to as the cross-coupling transfer function and can be

Fig. 9.19. A more complete gyro block diagram.

derived from Eqs. (9.20) and (9.21). If gimbal friction is small, it is

given approximately by

By[(Jy/By)S + 1]
(9.25)
J zJ y j zBy —I- J yB z
J 2 C0S 5
T 2., 2 s2 + T 2., 2 S + 1
K'J s ds

A similar expression may be derived for the y axis. A large cross coupling
is usually undesirable in a gyro, and Eq. (9.25) indicates that it is mini¬
mized by reducing friction and by making the spin velocity as large as
possible.
The complete set of gyro transfer functions may be summarized in the
block diagram shown in Fig. 9.19. In this figure con is the nutation fre¬
quency given by Eq. (9.24), and t; is the damping coefficient, which, by
Eq. (9.23), is equal-to (l/2Jscos)(Bz \/Jy/Jz + By \0~JTy).
Sec. 9.9] MECHANICAL COMPONENTS 351

9.9. Practical Gyroscopes. The rotor of a practical gyroscope is not

usually a disk but instead consists of the rotor of a polyphase induction
motor designed to operate at speeds as high as 20,000 rpm. Some of the
early aircraft instruments were, however, disks driven by a jet of air.
Some of the modern aircraft gyros are enclosed in a temperature-stabilized
environment and may be hermetically sealed. The sealed case may be
supported or floated in a damping fluid if demanded by the application
(see Sec. 9.11). Usually there is provision made in aircraft gyros to lock
or cage the gyro so that it cannot precess or move in any direction. This is
necessary so that, when the craft undertakes violent maneuvers, the gyro
will not be driven into its stops or wound up in its flexible connections.
The present manufacturing trend in aircraft gyros is to employ a separate
gyro in a single gimbal for each coordinate to be sensed. This eases the
problems of erection, or placing the gyro spin axis at its desired setting,
and the problem of providing precession torque and of sensing gyro
position.
A major problem in the construction of practical gyros is the minimiza¬
tion of drift. Drift is caused by a number of factors, but the primary
causes are unbalance of the gimbals and bearing friction. Unbalance of
the gimbals results in gravitational forces, which tend to precess the gyro.
The vibration that is always present when the gyro is spinning acts to
nullify the small static friction present even in the finest bearings, so that
mass unbalances which would not be detectable when the gyro disk is not
running will produce small rates of precession when the gyro is spinning.
Drift from this source can be minimized by fitting the gimbals with
adjustable weights so that any slight unbalance may be compensated for.
Bearing friction results in precessing torques whenever the external
mounting of the gyro is shifted. Bearing friction is caused by small
irregularities in the bearing surfaces and is largely random in character.
Hence it results in random precession, or drift, which usually has a non¬
zero average value. The average amount of drift varies with the installa¬
tion. With small gyros such as those used in artificial horizons of light
airplanes, the drift is often large enough to require manual resetting of the
gyro every 10 minutes. On the other hand, very high quality gyroscopes
can be built with average drift rates of less than 1° in 12 hours. Prior to
World War II, such gyros required very heavy rotors, a typical model
designed for shipboard installation having a rotor weighing 54 lb. Pres¬
ent manufacturing procedures have, however, developed to the point
where a single-degree-of-freedom gyro, weighing about 5 lb complete with
torque motors and case, can now be obtained with approximately the
same drift.
Probably the most critical example of gyroscope applications occurs in
inertial navigation systems. In these systems the gyroscope is usually
352 CONTROL SYSTEM COMPONENTS [Chap. 9

a primary reference although systems that are corrected by optical star-

tracking methods have been proposed. In any case the simpler magnetic
and gravity references cannot be used. A magnetic reference is useless
in transpolar flights, and gravity is undesirable as a reference since it is
not constant around the globe. Obviously both gravity and terrestial
magnetism would be uselss as references in interplanetary flight.
To minimize drift, gyros must be constructed with extremely fine bear¬
ings and have, therefore, almost no inherent damping. For this reason,
when a free gyro is used as part of a feedback loop, as in the gyroscopic
velocimeter described in Sec. 9.10, it contributes roots to the characteristic
equation of the system that are almost on the imaginary axis. Some
form of equalization is therefore always required in such feedback circuits
to produce a stable and well-damped system. In rate gyros, the lack of
inherent damping results in a highly oscillatory response, as indicated by
Eq. (9.27). For this reason some form of external damping must usually
be added. In some models electromagnetic damping is used; others are
damped by enclosing the gyro in a damping fluid; this latter scheme
simultaneously provides a shock mounting for the device. It should be
noted that purely viscous damping should theoretically not result in any
average drift, since it is nonrandom and perfectly symmetrical. Further¬
more, damping devices may be constructed (somewhat on the order of
the Lancaster damper described in Sec. 8.5) which dissipate energy when¬
ever the member to which they are attached moves, but which add no
friction between the gyro and the stationary frame.
Proper damping is particularly important in marine gyrocompasses.
In addition to being built with the finest bearings available, these gyros
are usually constructed to have an undamped natural period of the order
of 85 minutes. This long period, which incidentally is the period of a
simple pendulum having a length equal to the radius of the earth, has
been shown by Schuler1 to result in a compass whose north-south indica¬
tion is not affected by accelerations of the ship or airplane on which it is
mounted. A gyro having this period is said to be Schuler tuned, and
it is clear that any device having such a long period should be damped so
as to cause the oscillations to die out as quickly as possible. The damping
problem is, however, complicated by the fact that some of the more com¬
mon damping methods may cause north-seeking errors,2 and a number of
special methods are therefore used. For detailed description of these
methods the reader is referred to textbooks dealing with gyrocompasses.3
1 M. Schuler, Die Storung von Pendel und Kreisel apparaten durch die Besch-
leunigung des Fahrzeuges, Physik. Z., vol. 24, p. 344, 1923.
2 Rawlings, “The Theory of the Gyroscopic Compass,” The Macmillan Company,
New York, 1944.
3 Richardson, “The Gyroscope Applied,” Philosophical Library, Inc., New York,
1954.
Sec. 9.10] MECHANICAL COMPONENTS 353

9.10. Gyroscope Applications. Two basic gyro types are used in prac¬
tice: the free gyro and the restrained gyro. A free gyro is mounted in a
set of gimbals so that angular motion of the supporting frame is not trans¬
mitted to it. Ideally the direction of the spin axis of a free gyro is there¬
fore fixed in space. Such gyros are used to provide a fixed reference
position, for instance, in the vertical and directional gyros found in auto¬
matic pilots and in inertial navigation equipment. Other uses of this
sort are in artificial horizons, in roll-stabilizing equipment, and so forth.
Since even the most carefully constructed gyros are subject to a certain
amount of random drift, the accuracy of the reference direction estab¬
lished by the gyro tends to deteriorate with time. In some cases this is
of no consequence. Thus, a gyro used to establish the vertical or direc¬
tional references in a guided missile having a life of only a few minutes
may drift a few degrees per hour without any adverse effect on the accu¬
racy of the missile. Drift in the gyro used as the artificial horizon in an
airplane may not be important if the gyro can be reset occasionally.
However, in more critical applications the gyro is not completely free but
is equipped with small torque motors so that it can be precessed. In this
way the gyro can be automatically slaved to a more accurate primary
reference. Thus, a gyro used to establish a vertical may be slaved to a
pendulum of some sort. The position of the gyro is continuously com¬
pared to that of the pendulum, and any difference in the positions is used
to precess the gyro to make its position correspond to that of the pendu¬
lum. The precessing torque applied to the gyro is kept so small that the
maximum precession rate is only a little larger than the maximum drift
expected. In this way the gyro cannot follow rapid fluctuations of the
pendulum position, and the system acts merely to keep the gyro spin axis
aligned with the average pendulum position. Thus the gyro acts not so
much as a primary reference but as an integrating or smoothing device
applied to the actual reference. In a similar way directional gyros are
often slaved to a magnetic compass which provides the primary north-
south reference.
A somewhat different principle is used in the gyrocompass used on
ships. This is essentially a free gyro equipped with a weight that tends
to keep the spin axis in a horizontal plane. The weight hangs, supported
by two bearings on the spin axle, centered below the gyro disk. This
weight, together with the rotation of the earth, acts to precess the gyro
until the spin axis takes on a north-south position.
An application of the gyroscope that differs fundamentally from those
described above is the measurement of low angular velocity. In this
application, use is made of the fact that the precession velocity is directly
proportional to the precessing torque. Thus in one form of gyroscopic
velocimeter the gyroscope is equipped with synchro devices (see Chap. 5)
354 CONTROL SYSTEM COMPONENTS [CHAP. 9

that sense the relative displacement between the gimbals and the outer
frame. The output of the synchro is amplified and applied to the proper
torque motor in such a way that the resulting gyro precession returns the
output of the synchro to zero. This feedback system therefore acts to
keep the gyroscope aligned with the
outer frame. The torque devel¬
oped by the torque motors to ac¬
complish this result is, however, by
Eq. (9.14) directly proportional to
the precession velocity and there¬
fore to the velocity of the outer
frame. Commonly, the torque
motor is an electric motor, and the
torque is proportional to the voltage
Fig. 9.20. Gyroscopic velocimeter.
applied across the windings (see
Chap. 7). This voltage may there¬
fore be used as an indication of the velocity. A simplified schematic
diagram of such a system arranged to measure velocity in two axes is
shown in Fig. 9.20.
A similar, although somewhat simpler, device working on essentially
the same principle is the restrained, or rate gyro, shown in Fig. 9.21. In

Z axis

Fig. 9.21. A simplified schematic of a rate gyro.

this figure the gyro wheel is mounted in a gimbal which is pivoted so as

to be able to rotate in an outer frame. The rotation of the gimbal relative
to the outer frame is, however, checked by means of the restraining
Sec. 9.11] MECHANICAL COMPONENTS 355

springs shown. Suppose now that the entire assembly is rotated about
the y axis. F-axis torque is applied to the gyro through the gimbal, and
the gyro begins to precess about the 2 axis. This precession is arrested by
the springs, which act to produce a torque on the gyro about the z axis,
opposing the z-axis precession. The torque generated by the springs
results in precession about the y axis, so that the gyro is able to follow
the rotation of the outer frame. The torque supplied by the springs
must, therefore, be precisely sufficient to precess the gyro at the rate at
which the outer frame is being rotated. Hence this torque is a direct
measure of the speed of the outer frame. The simplest way to measure
the torque is to measure the change in displacement of the gimbal relative
to the outer frame, since if the restraining springs are linear, this displace¬
ment is proportional to the torque. For maximum accuracy the spring
should not deflect appreciably, since the torque required to precess the
gyro about the y axis (fixed with respect to the outer frame) becomes
smaller as the spin axis of the gyro approaches the y axis.
If the angular momentum of the gyro is large," a large torque is gener¬
ated by relatively small velocities. Hence rate gyros can be made sensi¬
tive to very low angular velocities. With reasonable care a rate gyro can
be constructed that indicates the angular velocity of the earth, a velocity
of less than 0.0007 rpm. It should be noted, incidentally, that the
velocity measured by a rate gyro is always with respect to an inertial
reference fixed in space; furthermore since velocities about the x and z
axes of Fig. 9.21 do not have the action described, the rate gyro is in
general sensitive only to the y component of the angular velocity applied
to it.
The transfer function of the rate gyro can be obtained from Eq. (9.20).
The spring acts to make Tz proportional to dz; hence Eq. (9.20) may be
rewritten as follows:

KZ6Z T Bzsdz T Jzs~9z = Jscx)ssdy (9.26)

where Kz is the spring constant of the restraining spring. Hence

J, o),
6, Jzs2 -f- Bzs -f- Kz
sdr (9.27)

Note that some damping is necessary in the 0 axis to prevent an oscillatory

response.
9.11. Application of Gyroscopes to Inertial Navigation. Inertial
navigation is a rather recent development which requires gyroscopes
with accuracies that would have been undreamed of a few years ago.
Although gyros of many types are used in this application, one of the
most frequently used types is a single-degree-of-freedom type developed
by C. S. Draper and associates. This is the so-called HIG gyro (hermetic
356 CONTROL SYSTEM COMPONENTS [ChAP. 9

integrating gyro).1 A schematic diagram of this gyro is shown in Fig.

9.22. In principle the construction of this gyro is quite similar to that of
the simple rate gyro shown in Fig. 9.21; however, no restraining springs
are used. Also, the inner gimbal, which holds the spinning gyro wheel,
is built in the form of a hermetically sealed cylindrical can, or float, with
shaft extensions. The can is filled with helium which acts as a neutral
atmosphere. The outer case of the gyro is completely filled with a viscous
fluid whose density is just sufficient so that the gimbal and gyro wheel
are in neutral buoyancy; i.e., in a large quantity of the fluid the inner can
would neither sink nor rise to the surface. Thus the load on the jewelled
bearings that support the gimbal to the outer case is negligibly small, and
there is therefore practically no friction between the outer case and the
gimbal. Attached to one of the shafts extending from the gimbal can is
the rotor of a microsyn generator (see Sec. 5.9), and a torque motor is
attached to the other shaft.
The operation of this gyro may be explained as follows. Suppose that
the outer case is rotated about the y axis (see Fig. 9.22). As has been
explained previously, the gyro will then tend to precess about the 2 axis.
There are no springs to arrest the precession as in the rate gyro, but since
the inner gimbal is floated in a highly viscous fluid, a viscous shear torque
is developed which limits the precession velocity. Thus in Eq. (9.27)
the spring constant term vanishes, and we obtain for the transfer function:

Q j s&s/Bz *
z “ (J'/B')8 + 1 Uy

The output angle 8Z is detected by the microsyn generator mentioned

above, and since in most practical applications the variation of the input
angle is quite slow the time constant Jz/Bz is usually sufficiently small as
to be negligible. Hence for small input angles 0y the output signal is
proportional to the input angle rather than to the rate of change of input
angle. This is the reason for the name “integrating gyro.”
Note that the HIG gyro furnishes the same sort of information as a
free gyro, i.e., attitude information. However, the special construction
features, in particular the fact that it is a single-axis device, make this
gyro much more accurate and rugged than the usual free gyros.
It should be noted that although the primary input axis to the gyro is
the y axis of Fig. 9.22, the gyro will also deliver an output signal when the
case is accelerated about the 2 axis. This type of output is usually unde¬
sirable and is minimized by using a highly viscous hydraulic fluid. The

1 C. S. Draper, W. Wrigley, and L. R. Grohe, The Floating Integrating Gyro and

Its Application to Geometrical Stabilization Problems on Moving Bases, Aeronaut.
Eng. Rev., vol 15, no. 6, June, 1956.
Sec. 9.11] mechanical components 357

developers of the device state that for the usual accelerations of the z
axis encountered in practice, this output is negligible.
In most applications to inertial navigation three HIG gyros are
mounted on a stable platform to detect rotation about the x, y, and z
axes. The output signal from each gyro is amplified and applied in the
proper fashion to servomotors which control the rotation of the stable
platform about the three axes. Thus, the platform is maintained at a
fixed attitude with respect to inertial space.
The torque motors mounted on the shaft extension of the inner gimbal
of the HIG gyro may be used to apply an additional precessing torque to
the gyro. This torque will produce an output signal even when the input

Fig. 9.22. Simplified diagram of the HIG gyro.

angle 6y is zero. This output will be amplified and applied to the servo¬
motor with the result that the stable platform tilts, or turns, at a rate pro¬
portional to the torque input. Thus the platform position can be con¬
trolled by currents applied to the torque motor. It is necessary to do
this since the gyros will tend to maintain the platform fixed with respect
to an inertial reference in space. It is, however, normally desirable that
the platform be maintained at a fixed attitude (usually horizontal) with
respect to the earth. Since the earth rotates in space a platform mounted
on a fixed point on the earth’s surface would therefore have to be rotated
at the same rate as that of the earth. In a moving vehicle the rotation of
the platform would have to be at a greater or lesser rate depending on the
direction of motion of the vehicle.
In addition to the stable platform, a complete inertial navigation
system makes use of accelerometers. In its simplest form an accelerom¬
eter consists of a spring-mass system as shown in Fig. 9.23. By Newton’s
law a force is needed to accelerate the mass. Whenever the frame of the
358 CONTROL SYSTEM COMPONENTS [Chap. 9

device is accelerated therefore, the spring deflects until it generates

enough force to accelerate the mass at the same rate as the frame. The
deflection of the spring, which may be measured by one of the standard
pickup devices discussed in Chap. 5, is a direct measure of acceleration.
More sensitive accelerometers may be constructed by mounting a gyro¬
scope such that any acceleration results in a torque on its input axis (see
Fig. 9.27 for a possible arrangement). No matter what form the acceler¬
ometer takes, it should be clear that it measures acceleration in only one
axis. This is really a consequence of the fact that acceleration is a vector
quantity. It should also be clear
that an accelerometer cannot very
well distinguish between the accel¬
eration of gravity and other accel¬
erations. Thus, if an accelerometer
is to be made insensitive to gravity
Fig. 9.23. Simplified diagram of an
it must be mounted on an accurately
accelerometer.
horizontal platform.
Suppose now that we mount three accelerometers on a stable platform
on a moving vehicle. The stable platform must be able to maintain
accurately the north-south direction, and it should be exactly horizontal.
The three accelerometers may then be mounted at right angles to each
other such that one of them measures vehicle acceleration along a north-
south axis, another measures accelerations along the east-west axis, and
the third measures accelerations in the up-and-down direction. If the
output of each accelerometer is integrated, the three component velocities
are obtained, and if these velocities are again integrated, the position of
the vehicle relative to the starting point is obtained. It is clear that if
any accuracy is to be realized by this system all parts must be extremely
precise. In particular, if the stable platform is not exactly horizontal,
the north-south and the east-west accelerometers detect a component of
the acceleration due to gravity and the output will therefore be erroneous.
The accelerometer output may be used to maintain the platform hori¬
zontal. Qualitatively it is easy to see why this should be so. Assuming
that the platform was originally horizontal, and that the coordinates of
the vehicle relative to the earth were known, the integrated accelerometer
output should indicate exactly where the vehicle is at any later time.
Hence it should be possible to compute what the correct platform posi¬
tion should be, and to apply the proper corrections to the torque motors of
the gyros.
To illustrate the process of maintaining the platform horizontal some¬
what more quantitatively, consider the simplified one-dimensional con¬
figuration of Fig. 9.24a. The angle 0V represents the angle made by a
true vertical with,an inertial reference. The indicated vertical, which is a
Sec. 9.11] MECHANICAL COMPONENTS 359

line drawn perpendicular to the table, makes an angle 6t with the reference
line. Suppose first that there is initially no error, i.e., 0V = dt, and let
the vehicle accelerate with a linear acceleration a. If the accelerometer has
a gain constant Ka, then it will deliver an output voltage Kaa. As a
result of the acceleration the vehicle moves along the surface, and there¬
fore the angle of the true vertical with respect to the reference changes.

Pickup

Fig. 9.24. The inertial navigation problem: (a) coordinate system; (6) single-axis
block diagram.

If R is the radius of the earth, the acceleration of the true vertical with
respect to the inertial reference is

6 = a/R (9.28)

A torque must therefore be applied to the gyro mounted on the table such
that the table angle 6t also accelerates this amount. If HIG gyros are
used, a fixed voltage applied to the torque motor results in a fixed velocity
of rotation of the table. Therefore, we must integrate the accelerometer
output once to obtain the angular velocity of the true vertical and apply
the output from the integrator to the torque motor. If the gain of the
integrator is Ki and if the ratio of table angular velocity to gyro input
360 CONTKOL SYSTEM COMPONENTS [Chap. 9

voltage is Kt,
0t = KtKifKa a dt
or
dt = RtKiKa a (9.29)
Using Eq. (9.28) gives
dt = KtKiKaRdv (9.30)

Since the object of the torquing is to keep the table horizontal we desire
that d t = dv and therefore
KtKiKaR = 1 (9.31)

This is the so-called Schuler tuning condition.1

It is interesting to observe now what happens as the result of an initial
misalignment of the table. Let the angle between the table and the hori¬
zontal be 4> = dt — dv; 4> is the error angle. Since the table is not hori¬
zontal, the accelerometer reads a component of g; the acceleration of
gravity and its output becomes

ea = Ka(a cos 4> — g sin <j>)

~ Ka{a — g<t>) (9.32)

if 4> is small. Therefore Eq. (9.30) becomes

dt = KtKiKaR(dv - g/R <f>) (9.33)

Substituting dt = 4> + dv, and applying the Schuler tuning condition

[Eq. (9.31)] gives
dv T 4> = dv — g/ R 4>
or
$ + g/R4> = 0 (9.34)

The solution of this differential equation is oscillatory with zero average

value and peak amplitude <f>. Thus the stable platform oscillates about
the correct value with a maximum excursion equal to the initial misalign¬
ment. Similarly the output from the accelerometer, the velocity inte¬
grator, and the distance integrator will oscillate about their correct out¬
puts. We have demonstrated therefore that an initial misalignment does
not give rise to an ever-increasing error, but rather to an oscillatory, and
therefore bounded, error. Similar results are obtained by considering
errors resulting from accelerometer offset, etc.
The angular frequency of oscillation of the table is seen to be a? = Vg/R
and the period is 2ir/co = 84.4 min. This is the Schuler period. It is
easy to show2 that a compound pendulum having this period continues to
point to the center of the earth (i.e., will maintain a true vertical) in spite
1 M. Schuler, loc. cit.
2 Ibid.
Sec. 9.11] mechanical components 361

of accelerations along the surface. The stable platform acts very much
like such a pendulum.
The discussion of the inertial navigation problem given here has per¬
force been quite brief. Nothing has been said about the three-axis prob¬
lem, and about the necessity of feeding corrections for earth rotation and
other factors into the system. For further information on this subj ect the
reader is referred to the literature.1

PROBLEMS
9.1. A worm-gear differential is shown in Fig. 9.25. One input is the rotation of the
shaft Xi. The second input is the linear displacement of the worm by moving the

Fig. 9.25. Differential worm gearing.

bracket X2. The worm may slide along the spline on Xi, but it cannot turn with
respect to Xi. The rotation of the gear X3 is the output. Discuss the possibilities
of such a device with respect to limitations on the magnitude of input signals and the
possible variations in transfer functions from each input to the output.
9.2. A spiral- and spur-gear differential is shown in Fig. 9.26. Rotation of the
shaft Xi forms one input; the second input is the linear motion of shaft X2. The

Fig. 9.26. Differential with exactly displaced spiral gear.

output is the angular rotation of shaft X3. Discuss the possibilities of such a device
with respect to the limitations on input-signal magnitudes and the possible limitations
on the transfer functions from each input to output.

1 N. F. Parker and C. P. Greening, Inertial Navigation, AGARD, Second Guided

Missiles Seminar—Guidance and Control, pp. 67-86, September, 1956. W. T.
Russell, Inertial Guidance for Rocket-propelled Missiles, Jet Propulsion, vol. 28,
no. 1, p. 17, January, 1958. (This article contains a list of references to other
papers.)
362 CONTROL SYSTEM COMPONENTS [CHAP. 9

9.3. Demonstrate that the nutating frequency of a thin disk spinning in free space
is equal to one-half the spin velocity of the disk.
9.4. Write the transfer function of a free gyro from Tz to 6y.
9.6. Write the transfer function of a rate gyro from Ty to 0Z and construct a block
diagram. *
9.6. In Fig. 9.27 is shown a simplified diagram of a possible gyroscopic accelerom¬
eter. Derive the transfer function.

Fig. 9.27. Simplified sketch of a gyroscopic accelerometer.

9.7. In Fig. 9.28 is shown a simplified diagram of a monorail car. Derive the equa¬
tions of motion and investigate stability.

oc axis

Fig. 9.28. Sketch of monorail car.

 CHAPTER 10

PUMP-CONTROLLED HYDRAULIC SYSTEMS

10.1. Introduction. In the general engineering sense, hydraulics is the

study of the statics and dynamics of fluids, both liquid and gas. For the
control engineer the term has a more restricted meaning. Hydraulic
is reserved for incompressible (or almost) fluids or liquids, and usually
the liquids are restricted to the various types of hydraulic oil, although
occasionally it is convenient to use a liquid being processed in the control
system itself. The qualification “almost” on the description incom¬
pressible is necessary because under certain conditions the compressibility
of the liquid must be considered. The intent, however, is to differentiate
“hydraulics” from “pneumatics,” in which compressibility is a major
consideration.
Hydraulic control systems have several unique advantages. A
hydraulic system may be designed to deliver high power to elements
separated by some distance, such as throughout an aircraft. Hydraulic
systems are thus better for some purposes than mechanical systems, which
must be localized; of course, electric systems are superior to all other
methods from this aspect. The actual power element or hydraulic motor
can be made much smaller physically than an equivalent electric motor.
A volume factor of 20:1 for units of about 5 hp may be cited. Owing to
this size reduction, the time constant of the hydraulic motor will be much
smaller than the time constant of the equivalent electric motor. Hydrau¬
lic elements are also more rugged than electric components and are not
usually susceptible to noise pickup from shocks and vibration as are
vacuum tubes.
Hydraulic systems have their disadvantages as well, of course.
Hydraulic lines are not so flexible as electric wire. Hydraulic systems
are somewhat more sensitive to environmental temperature variation
than electric systems. Hydraulic systems require fairly elaborate power
supply and pressure-regulating devices, and finally, in case of damage to
one part of the system, there is danger of the entire supply of hydraulic
oil being lost.
10.2. Basic Types of Hydraulic Control Systems. There are two basic
classes of hydraulic control systems. The first of these classes is the
364 CONTROL SYSTEM COMPONENTS [Chap. 10

pump-controlled system, which usually consists of a single variable-stroke

pump and a fixed-stroke motor. Control of the motor is exercised by
varying the amount of oil delivered by the pump. This is accomplished
by mechanically changing the pump volume or stroke, as shown in Fig.
10.1. The pump-controlled system is quite simple and is used when there
is only one motor to be controlled. As shown in Fig. 10.1 the motor is the
usual rotary form. In certain applications a linear actuator is required.

Prime
mover

Fig. 10.1. A pump-controlled hydraulic system.

For installations with several motors or actuators that must be inde¬

pendently controlled it is usually more economical and compact to employ
one large pump and several motors that are individually controlled by
valves. This is the second basic class of hydraulic control systems, the
valve-controlled system. Although fast-acting, accurate servo valves
are expensive, they result in a lighter, more flexible system, and when the
relatively inexpensive fixed-stroke pump plus pressure regulator is com¬
pared to the several expensive variable-stroke pumps required by inde¬
pendent pump-controlled systems, the total cost of valve-controlled oper¬
ation is usually less.
Other valves
and actuators

Control
input
Fig. 10.2. A valve-controlled constant-pressure hydraulic system.

There are two basic types of valve-operated systems. In the first

type, the constant-pressure system shown in Fig. 10.2, some form of pres¬
sure-regulating device is used with the pump to provide a constant pres¬
sure supply. When the valve is in the closed position, there is no oil flow
from the supply, or, at most, only a small leakage flow. Valves used in
this type of operation are referred to as closed-center valves. The second
type of system 'is the constant-flow type. In this type, an open-center
Sec. 10.3] pump-controlled hydraulic systems 365

valve is used which permits the hydraulic fluid to pass directly to the
sump when the valve is in the off position. Operation of the valve diverts
part of the fluid away from the sump to the actuator, and in the fully on
position all the fluid goes to the actuator. In this type of system the
supply pressure is low whenever the valve is in the off position but rises
as oil is diverted to the actuator. No pressure regulator is required
except possibly as an emergency relief valve, and since fixed-stroke
hydraulic pumps are inherently constant-flow devices, neither is any flow
regulator required.
If a fixed-stroke pump is used with a pressure regulator in the constant-
pressure system, the power delivered by the pump must be essentially
constant, since both the flow and pressure are constant. Hence this sys¬
tem is very inefficient, particularly if the duty cycle is such that relatively
large peak loads occur only occasionally. Except for the power that is
delivered by the hydraulic motor or actuator to its load, all the power
delivered by the pump is eventually converted into heat, which raises the
oil temperature. Hence fairly elaborate cooling is usually required in
this type of system to maintain the oil temperature below its permissible
maximum.
Although the constant-flow system does not suffer from this particular
disadvantage, it is not used as commonly as the constant-pressure system.
This is due to the fact that the control of the actuator is not so positive
with this type of system as with the constant-pressure type. It might
be noted incidentally that in the constant-flow system the load is not
locked rigidly when the valve is in the centered position. Depending on
the application, this may or may not be an advantage.
In this chapter we shall consider the pump-controlled hydraulic control
system and the various components employed. In the following chapter
the valve-controlled system and its components will be discussed.
10.3. The Pump-controlled Hydraulic System. The basic diagram of
the pump-controlled system is shown in Fig. 10.1. The most common
type of variable-stroke pump used in a system of this type is the piston
pump, shown in Fig. 10.3. The prime mover turns the shaft to which the
cylinder block is attached. The piston-return plate, or wobble plate,
is fixed. As the cylinder block rotates, it carries with it the pistons which
bear on the wobble plate. The fluid-valving plate is positioned in such a
manner that the cylinder volume is increasing as the piston passes the
inlet port and decreasing as the outlet port is passed. Thus the fluid that
was taken in through the inlet port is expelled through the outlet port.
The flow rate of the piston pump can be changed by changing the angle
of the wobble plate with respect to the cylinder block. This adjustment
can be made mechanically, hydraulically or electrically. The ease of this
adjustment and the fact that maximum flow rate is not affected by pro-
3C6 CONTROL SYSTEM COMPONENTS [Chap. 10
viding for this adjustment make the piston pump the most widely used
high-performance variable-stroke pump. In standard piston pumps,
considerable reaction force is experienced at the lever that controls the
wobble-plate angle. This reaction force is usually in such a direction as
to reduce the flow to zero, although it is possible by offsetting the pivot to
neutralize this force for a particular load and indeed to reverse the net
force. Owing to this reaction force, it is necessary to provide a stage of
power amplification to operate the pump stroke control if the stroke is to
be controlled electrically. This stroke control could be of the type dis¬
cussed in Sec. 11.1, for instance.
The fluid is passed from the outlet port of the variable-stroke pump
through the high-pressure line to the motor. Any of the various types of
hydraulic pumps theoretically work equally as well when operated as

Fig. 10.3. Basic variable-stroke piston pump.

motors; however, motors are almost always built with a fixed stroke.
Thus, a piston type of motor has a wobble plate fixed at a particular angle
(usually 30 to 40°). The high-pressure fluid is admitted by the valving
plate to a cylinder and forces the piston back. This can be accomplished
only by the cylinder’s turning with respect to the stationary wobble plate,
thus turning the output shaft. The cylinder which has been filled by the
fluid is then carried around to the exhaust port and the fluid expelled to
the low-pressure line by the returning piston. Other types of pumps and
motors that are discussed in a later section can also be employed in such a
system with varied degrees of success.
All hydraulic transmissions must have some sort of make-up system,
to replenish the fluid lost by leakage, and must also have an overload
relief valve, for protection in case of load seizures, etc. A typical replen¬
ishing system and relief protection are shown in Fig. 10.4. In the usual
case of bidirectional operation, either line may be the high-pressure line;
thus a relief valve should be placed in both lines. Similarly, replenishing
is usually provided in both lines to provide for the possibility of operation
in either direction for long periods of time. Care must be taken that the
Sec. 10.4] pump-controlled hydraulic systems 367
fluid picked up from the sump is not aerated. Defoaming agents are
sometimes added to the hydraulic fluid in order to reduce its compressi¬
bility and to improve hydraulic efficiency. The replenishing feature adds
a nonlinearity to the operation of the system since it places a lower limit
below which the return-line pressure may not drop.
The relief valve also contributes a nonlinearity to system operation.
The pressure in the high-pressure line is proportional to the developed
torque on the motor. Thus a relief valve sets an upper limit on the
torque that can be produced by the motor. In the case of a pure-inertia
load this sets an upper limit on the motor acceleration and, in the case of a
viscous-damping load, an upper limit on the motor velocity.

Fig. 10.4. Hydraulic transmission with replenishing valve and overload relief valve.

10.4. Analysis of the Pump-controlled System. The delivery rate of

an ideal piston pump can be calculated as the product of the volume
swept out by a piston per revolution times the number of pistons times
the speed of revolution.

Fp = xAcn = kpx (10-1)

where Fp — pump-flow rate, in.3/sec

x = length of stroke, in.
A = cross-sectional area of piston, in.2
c = number of cylinders
n = speed, rps
Ideally all the pump flow passes through the motor and is converted to
output shaft speed. Hence the speed of the motor shaft is theoretically
directly proportional to the pump-stroke position x.
When there is no load on the motor shaft, there is no back pressure in
the motor, and all the fluid flows through the motor in the proper chan¬
nels. When there is a load on the motor shaft, however, there must be a
pressure built up by the hydraulic flow to overcome this resistance to free
368 CONTROL SYSTEM COMPONENTS [Chap. 10

movement. The free flow of fluid in the normal motor channels is

impeded, and fluid leaks through gaskets and around end plates. It is
usually assumed that the leakage flow is proportional to motor pressure
or, since it will be shown that pressure is proportional to developed
torque, to developed torque in the motor.

Fl = KLTd (10.2)

Thus the pump flow does not all go into useful work in the motor, and as
we shall see below, an additional portion of flow must be accounted for.
To show that the pressure across the motor is proportional to developed
torque, we recognize that the angular displacement of the motor shaft is
directly related to the amount of oil (qm) forced through the proper chan¬
nels in the motor. Thus
qm = Kj (10.3)

where Km is the motor displacement constant, and 6 is the angular posi¬

tion of the motor shaft. Differentiating Eq. (10.3) gives

sqm = Fm = Kmsd (10.4)

By the law of conservation of energy the hydraulic power must be

equal to the mechanical power developed by the motor. Thus,

lb in 3 in.-lb
V in. 21 X F sec = TJ (10.5)
sec

Since from Eq. (10.4) we have

Fm = KJ (10.6)
therefore Td = (10.7)

The back pressure built up by the load not only causes leakage flow in
the motor and in the pump as well, which is accounted for by the term Fl,
but also causes the fluid in the lines to compress. The compressibility
of the hydraulic fluid is partially due to the minute quantities of air
entrained in it and to the expansion of the tubing with pressure, but the
major factor is the oil itself, which is compressible, as is any solid or fluid.
Usually the compressibility is a second-order effect except when the sys¬
tem operates at very high pressures and when there is a fairly large
amount of oil in the system. Volume compressibility is essentially pro¬
portional to pressure. Thus the compressibility flow is proportional to
the rate of change of pressure
Fc = Kcsp (10.8)

We can calculate the coefficient of compressibility, Kc, by assuming that

the motor is stalled, so that Fm is zero, and that the leakage flow Fl is also
zero. The pump pulls a volume of oil out of the low-pressure line
Sec. 10.4] pump-controlled hydraulic systems 369

and forces it into the high-pressure line. The compression volume in the
high-pressure line is
Vi Api
Agi = (10.9)
7
where q = oil volume, in.3
p = pressure, psi
V = volume of pipe, in.3
7 = bulk modulus of fluid, psi
Ap = change in pressure in line, psi
and where the subscript 1 refers to the high-pressure line.
Similarly the compression volume in the low-pressure line is

V2 Ap2
Aq2 = (10.10)
7

The two compression volumes will be equal in magnitude and opposite in

sign, because the fluid pumped from one line is identical to the fluid
pumped into the other. Hence

Aqi = —A q2 = A qc (10.11)
Then if
V
V,= V2 = ^ (10.12)
Lj

where V is the total oil volume under consideration, we may rewrite Eq.
(10.9) as

Aqr = ~ Ap! (10.13)

"7

Since the two volumes are the same, and if the replenishing valve does not
open, we have also that

Ap\ = -Ap2 - ~ (10.14)

where Ap is the pressure difference between the two lines. Then

^ = A- = Ke in.5/lb (10.15)

where Kc is the compressibility coefficient. The bulk modulus of stand¬

ard hydraulic fluids is usually about 2.5 X 105 psi; hence the compressi¬
bility coefficient is about 10~6 in.5/lb per cubic inch of fluid. Newton1
points out that, at atmospheric pressure, air makes up about 0.2 per cent
by volume of the fluid bulk, causing a considerable increase in Kc.

1 G. C. Newton, Speed Transmissions as Servo Motors, J. Franklin Inst., vol. 243,

p. 458, 1947.
370 CONTROL SYSTEM COMPONENTS [Chap. 10

Zweig1 has shown that the compressibility coefficient is increased about

25 per cent in a standard V£-in. copper hydraulic line by expansion of the
tube walls under pressure.
The theoretical pump flow Fv is absorbed in these three component
flows and may thus be expressed by

Fp — Fm + Fc + F l (10.16)

Using the flow relations above and given a knowledge of the load, it is
possible to derive a transfer function from pump stroke x to output shaft
position 6.
10.5. Transfer Function of the Pump-controlled System. In order to
derive the transfer function of the pump-controlled system, it is necessary
to determine the relation for the shaft load. Let us assume for generality
that the load consists of inertia, viscous damping, and an arbitrary load
torque. The relation in terms of developed torque and shaft position
will then be
Td = (,Js2 + Bs)6 + TL (10.17)

where J is the moment of inertia, B is the coefficient of viscous damping,

and Tl is the arbitrary load torque.
Substituting the developed relations for the various flows into Eq.
(10.16) and then eliminating the extraneous variables by the use of Eqs.
(10.17) and (10.8) yields

Kpx - [KL/Km + (Kc/Km)s\TL

s[(JKc/Km)s2 + (,JKL/Km + BKC/Km)s + Km + BKL/Km]
(10.18)

As we would expect, the gain of the system is determined primarily by the

pump constant Kv. The free s in the denominator indicates an integra¬
tion. The presence of compressibility combined with other parameters
provides an additional degree of freedom to the system and thus allows
the possibility of underdamped roots or resonance at some critical fre¬
quency. In small, localized systems, Kc approaches zero, and the Kc
terms may be neglected. In systems with long lines between pump and
motor, it may not be possible to neglect the pressure drop in the line due
to flow, as was done in this analysis. The problem of the hydraulic
transmission line is discussed in some detail in a later section (Sec. 10.11).
10.6. Other Pump and Motor Types. Hydraulic pumps may be
divided into two classes, positive-displacement pumps and turbopumps.
Generally speaking, only positive-displacement pumps are used in con¬
trol-system applications. Pumps are also classified as radial flow or axial

1 F. Zweig, “Compressibility Effects in Hydraulic Transmission Lines,” Yale Univ.

Dept. Elec. Eng. Tech. Rept., June, 1950.
Sec. 10.6] PUMP-CONTROLLED HYDRAULIC SYSTEMS 371

flow for purposes of description. Actually there seems to be no limit to

the configurations proposed for pumps and motor designs, and those dis¬
cussed here are merely representative. The various problems of leakage,
wear, etc., appear in all designs, and in general a device may be used inter¬
changeably as a motor or a pump.
J00
c Comparison of gear, vane and piston type pumps
CD
O Pumps of approximately 3-gpm rating
£90
Piston pump
O
S 80 gpp-Vane pump

Geor pump
OI 70
L_ Pump speed, 1,200 rpm
CD
>

0 1,000 2,000 3,000

Discharge pressure, psi
Fig. 10.5. Over-all efficiency versus discharge pressure for the three most common
types of pumps. (From Kaay et al.)

The three most common types of positive-displacement pumps are the

piston pump, which was used as an example in the pump-controlled system,
the vane pump, and the gear pump. Piston pumps may be adapted for
variable delivery with no loss in efficiency and delivery, while gear pumps
are essentially fixed-displacement devices. Vane pumps lie between
these two extremes. Figures 10.5 and 10.6 show sample test data for the

Fig. 10.6. Efficiency versus operating speed for the three most common types of
pumps. {From Kaay et al.)

three types of pumps at various pressures and operating speeds.1 The

pumps used to obtain the test data were rated at 3 gpm and designed for
the same type of service. The three types of pumps will be discussed in
detail in the following sections.
1 H. A. Vander Kaay, R. J. Murphy, and W. Traut, Hydraulic Pumps, Product
Eng., vol. 26, no. 7, p. 132, 1955.
372 CONTROL SYSTEM COMPONENTS [Chap. 10

Basically, all hydraulic pumps may be used as hydraulic motors, and

usually a hydraulic system consists of a matched set, since the factors
such as speed and pressure that dictated the choice of the pump type
are equally operative in the choice of the motor. Usually the stroke of a
hydraulic motor is fixed, and the motor is controlled by controlling the
fluid flow. Surprisingly, the simplest of the pumps, the gear pump, has
in the past been unsuccessful as a motor. The hydraulic unbalance
wedges the gears at standstill and reduces the available starting torque of
the standard gear motor to an unacceptable level.
10.7. Piston Pumps. Piston pumps are the most complex and there¬
fore the most expensive of the three main types of positive-displacement
pumps. The basic piston pump was shown in Fig. 10.3, and the operation
has been described.
The pistons and cylinder block, while they must be held to very close
tolerances, are not the source of most of the leakage flow or of the manu¬
facturing difficulty. The clearance between cylinder block and valving
plate is the major source of design problems in most models.
Early piston pumps had only a few pistons, and this resulted in pul¬
sating fluid delivery. By increasing the number of pistons and decreasing
their diameter, the same rate of flow can be maintained while the pulsa¬
tions are minimized.
The power delivered by a hydraulic pump is the product of pressure and
flow. The flow is equal to the product of the displacement of the pump
per revolution times the speed of the pump. The weight and size of the
pump are primarily functions of the pump displacement. Thus, if the
operating pressure and the pump speed are increased and the displace¬
ment decreased, the output power can be maintained while the size and
weight of the pump are reduced. These factors also reduce the size and
weight of other components in the system. Pressures and pump speeds
have thus shown continuous increases in applications such as use in air¬
craft, where weight is of primary importance. Piston pumps have been
used at speeds as high as 6,000 rpm and pressures up to 8,000 psi without
exceeding critical fluid velocities, etc.1 It will be noted from Figs. 10.5
and 10.6 that the piston pump is superior to the other pump types at the
higher operating speeds and pressures.
10.8. Vane Pumps. The vane pump is a considerably simpler mechan¬
ical device than the piston pump. In the fixed-displacement vane pump
with two inlet and two outlet ports, the hydraulic forces are balanced, and
the bearings experience essentially no hydraulic reaction force. A fixed-
displacement vane pump is shown in Fig. 10.7. Low-pressure fluid is
picked up by a vane as it passes an inlet port and is forced out the outlet
port at high pressure. By proper design the displacement of the vane
1 Ibid.
Sec. 10.8] PUMP-CONTROLLED HYDRAULIC SYSTEMS 373

slots can be made to contribute to the output. This may represent an

increase of 10 to 15 per cent in the displacement of the pump.1 It is only
in the larger sizes that the additional porting required for the inlet and
the outlet flow is worth while, however. It should be noted that it is not
leakage flow that is displaced but fluid introduced into the vane slot
specifically for this purpose.
Centrifugal force is usually sufficient to maintain contact between the
vanes and the stator, although the vane slots are sometimes pressurized
to eliminate an adverse pressure differential, or the vane may be spring-
loaded. Ideally the vane is in line contact with the stator, which is not
so desirable as the surface contact of the pistons in the piston pump. The

Fig. 10.7. Fixed-stroke vane pump with two inlet ports and two outlet ports.

vanes are the weak point in vane pumps, and the problem of chipping and
breaking has not been completely solved. However, it will be noted here
that the gradual wear of the vane edges during operation does not increase
the leakage flow as wear on the pistons would.
The reciprocating mass in the vane pump is quite small as compared
with the piston pump, and the flow of fluid is smoother. Along with
these factors the simplicity of construction contributes to the quiet,
smooth operation and the long life of the vane pump.
Vane pumps can be constructed with a variable stroke. A simplified
sketch of a variable-stroke vane pump is given in Fig. 10.8. When a
variable stroke is provided, only one inlet and outlet port may be used.
Because of the cam ring and end clearances required for the variable-
stroke pump, its volumetric efficiency will be lower than a fixed-displace¬
ment pump of the same size. In addition, the delivery volume of a
variable-displacement pump is one-half that of the fixed-stroke type.
1 Ibid.
374 CONTROL SYSTEM COMPONENTS [CHAP. 10

The variation in displacement may be made by changing the eccentricity,

by sliding the cam ring, or by rotating the cam ring with respect to the
ports. The variable-displacement pump is unbalanced hydraulically,
and the bearings and cam ring must be designed to support the hydraulic
force produced when the pump op¬
Zero stroke
position of erates against pressure. For these
reasons the variable-displacement
vane pump is not usually designed
for a flow of more than 70 gpm or a
pressure of over 1,000 psi.1
10.9. Gear Pumps. Gear pumps
Movoble have fewer moving parts than the
com ring
other two common types of hydrau¬
Zero stroke lic pumps and are the easiest to
_ Maximum manufacture. They are simple,
stroke
rugged, efficient, and low in cost.
Figure 10.9 shows a typical gear
pump and the direction of fluid
Fig. 10.8. Variable-stroke vane pump.
flow. One of the gears is driven
Note that only one inlet and one outlet
from the prime mover, and the
port may be used.
other gear idles. The fluid is caught
up in the space between the rotating teeth and the housing and is carried
to the outlet under pressure. It will be seen that the gear pump
is intrinsically a fixed-displacement device.
Losses in the gear pump may occur in several ways. Leakage can
occur between the gears and the housing. This can be held to a minimum
by proper bearing design which permits the
clearance between housing and gears to be held
within close limits. Another source of leakage is
the space between the gears and the end plates.
Usually these end plates are fixed, with a clearance
between them and the gears of several ten-thou¬
sandths of an inch. Occasionally the end plates
are made free to slide axially and are held tightly
to the gears with pressure springs. This results
in higher hydraulic efficiency but also higher
mechanical losses. A somewhat more elaborate
method of holding the end plates against the
gears is by hydraulic pressure. In one design the pressure varies with
load flow, thus maintaining minimum leakage with minimum loss in
efficiency.
Simple gear pumps of the type shown in Fig. 10.9 are hydraulically
1 Ibid.
Sec. 10.9] pump-controlled hydraulic systems 375

unbalanced, and the gear bearings must be designed to withstand the

forces involved. It is possible, however, to design a gear pump with two
inlets and two outlets, thus balancing the hydraulic forces. A disadvan¬
tage of the gear pump is its operational noise. Gear pumps are made
in wide ranges of displacements and speeds. They may be operated at
speeds up to 4,000 rpm and occasionally
as high as 9,000 rpm; however they are
somewhat more efficient at low speeds,
as may be seen from Fig. 10.6.
Several problems are encountered in
operating gear pumps at high speeds.
First, it is difficult to fill the displace¬
ment volume between the teeth and
housing completely. Some gear-pump
models include a nozzle in the inlet port
to increase the fluid velocity as it enters,
thus forcing the fluid into the displace¬
Fig. 10.10. A gear-pump design
ment volume. An additional problem in that recovers the oil trapped be¬
gear-pump operation is the reduction in tween the two gears.
volumetric efficiency that occurs when
fluid is trapped between the gear teeth at the outlet port and carried back to
the inlet. One ingenious solution to this problem is shown in Fig. 10.10.
The driven gear rides on a stationary shaft. Small ports in the driven-
gear teeth allow the trapped fluid to flow into the passage in the stationary
shaft. The fluid is then added to the output flow. In addition to the
increased flow, this method also re¬
Force on ge>
and bearing duces the fluid pressure between the
due to hydra\ teeth that tends to separate the teeth
pressure
and cause bearing wear.
While dirt should not be allowed
High Low to enter any hydraulic system, the
pressure pressure
gear pump will experience less dam¬
age from contamination than the
other forms of pumps, since the par¬
ticles will be thrown to the outer rim
of the gears and be expelled to the
Fig. 10.11. Hydraulic-force compensa¬
system. The standard gear pump is
tion of a gear motor. (Courtesy Eastern
Industries, Inc.)
unsatisfactory as a hydraulic motor,
because wedging is caused by the
unbalanced force from the high-pressure inlet. However, an interesting
modification of the gear pump can balance this hydraulic force. In Fig.
10.11 is shown a gear motor with hydraulic balancing. Passages in the
case bring high-pressure fluid to bear on the gear shafts at a point designed
376 CONTROL SYSTEM COMPONENTS [Chap. 10

to cancel the hydraulic force oil the gears. The method is successful in
reducing the breakaway pressure required by the motor at starting to 10
to 20 psi, a negligible value.
10.10. The Ball Pump. The ball pump, shown in Fig. 10.12, is basi¬
cally a modification of the piston design. The balls are forced in and out
radially by the cam plate and hydraulic pressure, thus displacing fluid, in
the case of a pump, or forcing rotation, in the case of a motor. The inlet
and outlet ports are inside the ball race. The stroke may be made vari¬
able by allowing the cam-plate eccen¬
Eccentric
Block cam plate tricity to be adjusted. In motor oper¬
(.stationary)
[rotates) ation, high-pressure fluid is admitted at
the port marked in and forces the balls
radially outward. This may be accom¬
plished only by rotary motion of the
ball race. The ball pump provides a
line contact between ball and race, while
the conventional piston pump provides
'Stationary
a surface contact between piston and
Fig. 10.12. Ball pump.
cylinder. Thus the problem of leakage
due to ball wear or eccentricity is severe in the ball pump. The radial
design has the advantage over the conventional axial piston pump of
simplicity and ease of manufacture.
10.11. Hydraulic Transmission Lines. While the effects of very short
transmission lines, if they provide adequate cross-sectional area, may be
neglected, it is often necessary to consider the pressure drop in a line of
moderate length; and in long lines the effect of the dynamic flow charac-

Parabolic velocity Approximately ellipsoidal

[a) Laminar flow [b] Turbulent flow

Fig. 10.13. (a) Laminar flow with a parabolic velocity distribution; (b) turbulent flow
with approximately ellipsoidal velocity distribution.

teristics upon the transfer function of the control system must also be
considered.
Hydraulic flow may be divided into two types: first, viscous flow, or
laminar flow, and, second, turbulent flow. In laminar flow the particles
of fluid flow smoothly without jumbling. The flow may be thought of as
concentric tubes of fluid sliding on one another, after the manner of an
expanding telescope. The innermost tube has the highest velocity, while
the tube nearest the boundary is stationary. The tubes of flow do not
Sec. 10.11] PUMP-CONTROLLED HYDRAULIC SYSTEMS 377
exist in turbulent flow. The individual oil particles are constantly inter¬
mixing. The two types of flow are shown diagrammatically in Fig. 10.13.
Under low pressures and for fluids of moderate to high viscosity the flow
may be laminar, but as the pressure is increased, the flow becomes turbu¬
lent. Flow may be characterized by a dimensionless constant Nr, the
so-called Reynolds number, after Osborne Reynolds, whose classic experi¬
ments in 1883 defined the types of flow. The Reynolds number is defined
as
DV
Nr — — (10.19)

where D = diameter of the fluid section1, in.

V = mean velocity of flow, in./sec
v = kinematic viscosity of fluid, in.2/sec
This dimensionless ratio occurs throughout fluid mechanics and is essen¬
tially independent of the particular fluid and configuration. Thus it is
possible to say that, for a given Reynolds number, flow will be laminar for
oil in a pipe or for air around an aircraft wing.
When the flow is increased slowly in a smooth straight pipe through a
carefully rounded entrance, laminar flow has been maintained for values
of the Reynolds number as high as 50,000. Under normal conditions,
however, turbulence commences at Reynolds numbers of 2,400 to 3,000,
and if the installation has many bends, fittings, and valves, the flow may
be turbulent at Nr’s as low as 1,000.* The pressure drop and resultant
energy loss are significantly higher in turbulent flow than in laminar flow.
The relationship for flow through a pipe may be given by the Chezy flow
formula

Pi - P°- = gj (10.20)

where p = pressure, psi

€ coefficient of energy loss
-=

P unit density of fluid, lb/in.3

F volumetric flow, in.3/sec
9 acceleration of gravity, in./sec2
A cross-sectional area of pipe, in.2
The Chezy relation was developed empirically in 1775, but it may be
derived from the law of conservation of energy.2 The coefficient of

1 The concept of the Reynolds number has been found useful throughout the field
of hydrodynamics, and the parameter D is in general some characteristic dimension
of the system.
* N. M. Sverdrup, Theory of Hydraulic Flow Control, Product Eng., vol. 26, no. 4,
p. 165, 1955.
2 See any text on fluid mechanics, e.g., Dodge and Thompson, “Fluid Mechanics,’’
McGraw-Hill Book Company, Inc., New York, 1937.
378 CONTROL SYSTEM COMPONENTS [CHAP. 10

energy loss, e, depends on the coefficient of friction and the normalized

length of the pipe:
e=/^ (10.21)

Empirical values of / for smooth pipe as a function of the Reynolds num¬

ber are given in Fig. 10.14. It will be noted that for laminar flow the

102 103 104 105

Reynolds number, R

Reynolds number, R
Fig. 10.14. Friction coefficient versus Nr for pipes.

variation of / is linear, and as might be assumed, a relation may be devel¬

oped for this region that eliminates the need of the diagram and the
Chezy formula. This relation is called the Hagen-Poiseuille law for
laminar flow and is derived in texts on fluid mechanics.
Sec. 10.11] PUMP-CONTROLLED HYDRAULIC SYSTEMS 379

Tr(pi — Pz)DA
(10.22)
128juL
where p is the absolute viscosity of the fluid. The relationship for flow
shapes other than circular pipes may also be derived. The absolute
viscosity p in Eq. (10.22) is related to the kinematic viscosity v in the
definition of the Reynolds number by

(10.23)

where p = absolute viscosity, lb-sec/in.2

p = density of fluid, lb/in.3
Both the absolute viscosity and the density of hydraulic fluids vary with
temperature; thus the kinematic viscosity is also a function of tempera¬
ture. Figure 10.15 shows the density and the kinematic viscosity of
xicr4

Fig. 10.15. Density (p) and kinematic viscosity (v) of Univis J43 versus temperature.

Univis J43 AN-VVO-366, a standard hydraulic fluid, plotted against

temperature.
When the hydraulic transmission lines become very long, wave effects
and related phenomena must be considered, and the line will then have a
380 CONTROL SYSTEM COMPONENTS [Chap. 10
i
very pronounced effect on the transfer function relating pump stroke to
motor speed. Zweig1 has shown that a hydraulic transmission line
behaves in the same general way as an electric transmission line, and that
the techniques used for many years in analyzing electric lines may be
used also for hydraulic lines. In particular, he has shown that the well-
known Smith chart may be used to simplify computations connected
with such a line. The criterion for the length of a transmission line is
the velocity of propagation of the line. If the time required for a signal
to pass down the line is short with respect to the period of the highest-
frequency wave that is to be transmitted, the transmission effects may
be ignored. As will be shown below, the velocity of propagation in a
dissipationless line is given by

where y is the bulk modulus of the oil, psi; p is the density in lbs/in.3, and g
is the acceleration of gravity. For most hydraulic fluids the velocity is in
the neighborhood of 4,200 ft/sec. Hence a 100-ft line is equivalent to one
wavelength for a signal frequency of 42 cps and would be considered long
at this frequency. It could be considered short at frequencies of less than
4 cps.
The parameters determining the behavior of a transmission line are
the surge, or characteristic impedance ZQ) and the propagation constant
a + j(3. The characteristic impedance is the complex ratio of the sinus¬
oidal voltage to current in a wave passing down the line without reflection
from the end. The propagation constant is a complex number a
where a is the attenuation constant expressed in nepers per unit length,
and (3 is the phase shift constant expressed in radians per unit length. It
is shown in standard texts on transmission line theory that the velocity of
propagation is co//5, where oo is the signal frequency in radians per second.2
For electrical lines it has been shown that the characteristic impedance
is given by

w + W? / _ I tan-i A
(uC)2 / 2 a>L

The attenuation constant a is given by

co 2LC V2
a

1 Zweig, op. cit.

2 See, for instance, Everitt and Anner, “Communication Engineering,” 3d ed.,
McGraw-Hill Book Company, Inc., New York, 1956, p. 231.
Sec. 10.11] PUMP-CONTROLLED HYDRAULIC SYSTEMS 381

and the phase shift constant is

(c02LC)2 , (ccRCy . « 2LC

+ +
where co is the frequency in radians per seconds, while R, L, and C are the
resistance, inductance, and capacitance of the line per unit length.
For a hydraulic line in which the flow is assumed to be laminar, the
resistance per unit length can be obtained from the Hagen-Poiseuille law
[Eq. (10.22)] and becomes

Rh tr£>4

The hydraulic inductance per unit length is, by direct analogy with
electric circuits, the ratio of pressure drop per unit length to the resulting
rate of change of flow. If the pipe has a cross section of A in.2 and the
fluid has a density of p lbs/in.3, then the force exerted on a unit length of
fluid by the pressure drop pi along this unit length is pi A lbs. Also, the
unit mass of the fluid is p A lbs. Hence by Newton’s law

. pA 1 dF p dF
PlA = Y A~dt = ~gTt

where g is the acceleration of gravity, and F is the flow. Hence,

T = Vl = JL
h dF/dt Ag

Finally the capacitance per unit length of line is the ratio of a change of
oil volume to a change of pressure in the unit length. For a fluid having
a bulk modulus of 7 psi the change in volume [see Eq. (10.9)] is

Aq = ^
T

Therefore Ch = ~ = —
Ap 7

Thus, for the hydraulic transmission line we obtain

radians/in.
382 CONTROL SYSTEM COMPONENTS [CHAP. 10

For the special case of the dissipationless line these expressions reduce to

the simpler forms

F A-
oc+J/9 k ZL
PY lbs-sec
Load Zo =
A2g in.'
Pl Zl a = 0 (10.24)
1 Z„ Y
ZL P_
F oc+j/3
-F, /3 = co radians/in.
gy
Fig. 10.16. Hydraulic transmission line. .
r or
a typical hydraulic fluid
p = 0.03 lbs/in.3, 7 = 2 X 105 psi so that Z0 for a standard (ID)
copper tube is 26.3 lbs-sec/in.5 while 13 is 1.98 X 10-5 co radians/in. The
velocity of propagation is co//3, or approximately 50,000 in./sec.
Zweig1 has shown that for a symmetrical transmission line such as the
one shown in Fig. 10.16 the pressure p across the input terminals is given
by
p = Fl[Zl cosh (a + j(3)l + 2Z0 sinh (a + j$)l] (10.25)

and the input flow is

F = FL[cosh (a + j$)l + (Zl/2Z0) sinh (a + j$)l] (10.26)

In these equations:
p = pressure difference across the input terminals of the transmission
line, psi.
Pl = pressure difference across the load, psi.
F = flow at the input terminals of the line, in.3/sec.
Fl = flow at the load, in.3/sec.
Zl = Pl/ql = hydraulic impedance of the load, lbs-sec/in.5
I = length in line, in.
Z0 and a + j(3 are the characteristic impedance and propagation con¬
stant of one line.
The input impedance of the line is obtained from Eqs. (10.25) and (10.26)

V _ P _ oV (Zl/2Zo) cosh (a + j(3)l + sinh (a + j/3)l ( ,

* F 0 cosh (a + jf3)l + (.Zl/2Z0) sinh (a + jp)l K }

It should be borne in mind that a single hydraulic pipe is equivalent to

two electric wires, and that therefore the configuration shown in Fig.
10.16 is not exactly equivalent to an electric line connecting a generator
to a load. For this reason Eqs. (10.25) to (10.27) are not in exactly the
same form as that given in texts on electrical transmission lines. In par¬
ticular, the factor 2 appearing in several places in the equations arises

1 Zweig, op. cit.

Sec. 10.11] PUMP-CONTROLLED HYDRAULIC SYSTEMS 383

from the fact that there are two lines, each with a characteristic imped¬
ance ZQ, and a propagation factor a + jfi-
In practice we are usually interested in obtaining the transfer function
of a system including a transmission line. Thus, suppose that the load is
a motor of the type discussed in Sec. 10.4, and the source is a variable
stroke pump. The first step is to find the hydraulic impedance ZL pre¬
sented by the motor. This is done by assuming the motor to have inertia,
viscous friction, and possibly leakage. The flow for a given pressure may
then be obtained by the methods outlined in Sec. 10.4. The parameters
Z0, a, and /6 of the hydraulic transmission line are assumed to be known,
and thus the input impedance Zt presented by the combination of motor
and transmission line may be computed by Eq. (10.27). This is the load
impedance that the pump sees, and therefore, given the characteristics of
the pump, F and p can be computed as a function of pump stroke.
Finally, Eqs. (10.25) and (10.26) may be used to determine pl and FL, and
thus the motor output.
When the product (a + j(3)l is small it is possible to simplify Eqs.
(10.25) and (10.26) by using only the first two terms of the Taylor expan¬
sion for the hyperbolic functions. The approximate expressions for the
equation are, then,
2/ 1
p = FlZl 1 + ~Z^S°L 2 (10.28)

and F = FL 1 + (« + jP)l + + jfiW (10.29)

If the constants are such that the second and third terms in the brackets
are small compared to unity, then a solution assuming compressionless,
inertialess flow is adequate, and the transmission line has negligible effect
on the system. Even if this is not possible, it is usually permissible to
neglect the power dissipated in the lines, and to use the expressions for
Z0, a, and d given by Eq. (10.24). This simplification gives

2p
p = flz i +y Zz^Ag ccl - 2gy CO n2 (10.30)

.zla
F = Fl i +j cd - co2Z2 (10.31)
2t 2gy

Equations (10.30) and (10.31) indicate that changes in pressure and flow
will be chiefly in phase angle if the load is essentially resistive, and will be
chiefly in magnitude if the load is reactive.
The expressions obtained here for hydraulic transmission lines may be
used also with pneumatic transmission lines, provided that the magnitude
of the pressure wave along the lines is small enough so that one can reason¬
ably define a constant bulk modulus. The reader is, however, referred
384 CONTROL SYSTEM COMPONENTS [CHAP. 10

also to Sec. 12.6 for an empiric relation of a somewhat different form that
has been obtained for pneumatic transmission lines.

PROBLEMS
10.1. In a particular pump-controlled hydraulic system like that shown in Fig.
10.1 the pump is rated at 0.738 in.3 per revolution, 3,600 rpm, and 1,000 psi at full
stroke. It may be assumed that the leakage flow is 5 per cent of the load flow at
full pressure. The fluid is Univis J43 at 70°F, and the pipe is %-in. copper with the
motor located 50 ft from the pump. Derive the transfer function from the percentage
stroke to motor-shaft-position output for a motor having a displacement of 0.378
in.3 revolution, a leakage of 5 per cent of full load flow at 1,000 psi, an effective
inertia of 50 ft-lb-sec2, and a coefficient of viscous damping of 10-ft-lb-sec. Neglect
all transmission-line effects except compressibility.
10.2. What is the Reynolds number at full load in Prob. 10.1?
10.3. For a coefficient of energy loss of 0.1 what is the pressure drop in the line in
Prob. 10.1?
10.4. Establish the input impedance at the pump of the transmission line and load
in Prob. 10.1 as a function of the driving frequency. Assume the line is symmetric.
10.5. Find the transfer function of the system of Prob. 10.1 (a) considering the
transmission line effects exactly, (b) using the approximate-transmission-line equations
for short lines [Eqs. (10.28) and 10.29)], and (c) using the approximate-transmission-
line equations for dissipationless lines [Eqs. (10.30) and (10.31)].
10.6. Figure 10.17 shows a simplified diagram of a common type of hydraulic trans¬
mission. The motion of the solenoid armature is amplified by the “stroke servo”
consisting of the pilot valve, power piston, and feedback link abc. The length of
ab of the feedback link is 0.5 cm, and the length be is 4.5 cm. The output of the

Fig. 10.17. Simplified diagram of a common type of hydraulic transmission.

Problems] pump-controlled hydraulic systems 385

power piston is applied to the stroke arm of the hydraulic pump and varies the oil
delivery to the hydraulic motor. The motor is of the fixed-displacement type; hence
its speed is directly proportional to the amount of oil flowing through it.
In order to find the significant data on the stroke servo and solenoid, the linkage
is removed from the power piston at point c, and pivot c is then connected to
a fixed point so as to act as a fulcrum for the lever abc. Under these conditions it is
found that
1. The power piston does not move if the currents in the two halves of the solenoid
are equal.
2. The pilot valve moves down 0.002 in. and the piston moves down at the rate of
0.2 in./sec when the currents flowing into the solenoid differ by 1 ma.
3. When the valve is moved down 0.002 in. very suddenly, i.e., if a step function of
0.002 in. is applied to the valve, the initial acceleration of the power piston is 1,000
in. /sec2.
4. When an alternating current of constant amplitude but varying frequency is
forced into the solenoid, a strong resonance is observed in the solenoid response at
40 cps, and at higher frequencies the response drops off as the second power of fre¬
quency. At the resonant frequency the amplitude of the xs swing is three times that
observed at very low frequencies.
The following data are available on the pump-and-motor part of the transmission:
1. Pump speed is 4,000 rpm.
2. The pump delivery is proportional to y, being zero when y is neutral (as shown
in the figure) and 0.34 in.3 per revolution when y is 0.5 in. from the neutral.
3. The oil displacement of the motor is fixed and equal to 0.38 in.3 per revolution.
4. The motor has an effective moment of inertia of 0.015 lb-in.-sec2 and a coefficient
of viscous friction of 0.1 lb-in.-sec.
5. Owing to oil leakage from the high- to the low-pressure lines, the motor speed
will decrease by 3 rpm per inch-pound of torque applied to the output shaft.
6. Total oil volume in both oil lines is 12 in.3.
7. The effective compressibility of the oil used (this includes the stretch in the tube
walls) is 4.4 X 10~6 in.2/lb.
Assume linearity throughout the analysis of this system. Also it is permissible
to assume that the reaction force of the pump on the power piston and the hydraulic
reaction force of the pilot valve are negligible. Further, the volume of oil in the
stroke servo is so small that compressibility effects in that part of the transmission
may be neglected.
Problem: Find the complete expression for the transfer function 6/is, where 6 is
the position of the output shaft of the hydraulic motor, and is is the difference in
currents in the two halves of the solenoid. Sketch the asymptotic and exact fre¬
quency-response curve, and the phase-shift curve for the transmission. Neglect
effects above frequencies of 1,000 radians/sec.
 CHAPTER 11

VALVE-CONTROLLED HYDRAULIC SYSTEMS

11.1. Introduction. The use of a variable-stroke pump to control a

hydraulic motor as discussed in Chap. 10 is unsatisfactory if there are
several hydraulic actuators that are supplied from the same pump and
that must be independently controlled. Usually there is one hydraulic
power supply that operates at constant pressure in this type of system,
and the actuators are individually
controlled by pilot valves.
In this chapter we shall consider
the valve-controlled system and the
components that are unique to it.
The motors and pumps have been
considered in Chap. 10. Control
valves and pressure regulators will be
discussed here. As in Chap. 10, the
devices will be introduced by a dis¬
cussion of a simplified system.
In Fig. 11.1 is shown a hydraulic
actuator or stroke amplifier such as
might be used to operate the vari¬
able-stroke pump in Sec. 10.3. The
valve is a standard three-land spool
type and is actuated from the input
through the linkage with a pivot at
point W. Mathematically Y and W
Fig. 11.1. A simple hydraulic valve and
actuator.
are the same point and are named
separately only for clarity. If the
hydraulic fluid is considered incompressible, point W cannot move with
the valve closed; thus W may be considered a pivot. If X moves down,
Z moves down, and high-pressure fluid is ported to the top of the actuator.
At the same time the bottom of the actuator is opened to sump. The
actuator moves down, following the input. As W moves down, the link¬
age moves about X as a pivot, carrying Z up and closing the valve. Let
us assume that the vaflvQ flow is proportional to valve opening Z; then
386
Sec. 11.2] VALVE-CONTROLLED HYDRAULIC SYSTEMS 387

Z = ~~ X in. (11.1)

and F = KVZ in.3/sec (11.2)

where Kv is the constant relating motion to now and has the dimensions
of (in.3/sec)/in. Flow may be con¬
verted to velocity of the actuator by
dividing by the area A of the piston.
Thus the transfer function from valve
position Z to Y is

Y Kv Fig. 11.2. Block diagram for the

? = X (11-3) hydraulic actuator shown in Fig. 11.1.

The transfer function through the feedback link is the lever ratio

(11.4)

where the minus sign shows the change in direction. Figure 11.2 shows
a block diagram for the complete unit. The closed-loop transfer function
from X to Y is thus

Y _ a + b Kv/As _ (a + b)/a
J 1 + (aKv/bAs]) ~ JbAjaTQV+1 (1L5)

The valve in this example primarily controls the hydraulic flow. It is

also possible to design a valve that primarily controls pressure. Figure
11.30 shows an example of a pressure-control valve of the flapper type
discussed in Sec. 11.9. The position of the flapper controls p2, the cham¬
ber pressure which accelerates the actuator if it is unloaded. The trans¬
fer function from flapper position to unloaded actuator position thus
contains two integrations.
11.2. Spool-type Pilot Valves. The spool valve shown in Fig. 11.1 is
one of several spool-type valves that are classified as four-way valves.
In the four-way valve there are four orifices that are critically spaced with
respect to one another. In Fig. 11.3 are shown two more four-way valves.
It will be noted that these valves involve no basic modification in oper¬
ation nor simplification of manufacturing. The number of critical
dimensions remains the same as with the valve in Fig. 11.1.
A somewhat different type of operation is involved in the two-way
spool valves shown in Fig. 11.4. Here there are only two orifices, and
the valves are restricted to operation with unequal-area actuators. The
high-pressure supply is connected to the smaller-area side of the actuator,
and the valve controls the flow and pressure to the larger-area side. The
388 CONTROL SYSTEM COMPONENTS [Chap. 11

derivation of the transfer function from valve position to actuator position

is left as an exercise for the reader.
The clearances and tolerances in pilot valves are held to as close as
0.0001 in. to minimize leakage. Proper fitting of the valve spool is a

r
■/////.
1 /yAFT
'/////
Sump
7777
y
'/////
Sump
77777
I
Y/y/yy
LI
'7/7 7777, ^
Pump To actuator Pum/^ y. To actuator
'77777a. ryyyy.
7Q 7777
'/////.
Sump Sump
77777
u
[a) ib)
Fir;. 11.3. Two types of four-way spool valves.

Fig. 11.4. Two-way spool valve.

difficult manufacturing problem. Exact alignment of the valve lands and

ports is also a major problem. A valve in which the edges of the lands
exactly meet the edges of the ports is called a zero-lap valve. It is
impossible to manufacture a valve with exactly zero lap, and if the valve
is to be constructed at all, the de-
u Votve
spool
Li i 1 signer must allow at least a certain
land -Bevel -—^ minimum underlap or overlap (see
77,77, 77/77/ ^7/777, Fig. 11.5). Even if the valve spools

\ f
Valve yy.
port
1 I could be manufactured with zero lap
and no bevel, the initial knife-edge
would rapidly round off in use.
Overlap Underlap
A valve with overlap has a dead
Fig. 11.5. Overlap and underlap on
lands of valve spool.
zone equal to the amount of overlap.
This nonlinearity results in loss in
sensitivity near the center and may cause instability in control systems.
Usually, therefore, valves with a small amount of underlap are preferred
for control-system applications. The disadvantage of the underlapped
valve is the leakage that exists even when the valve is centered.
Sec. 11.3] valve-controlled hydraulic systems 389

In the manufacture of a four-way valve spool, there are three critical

dimensions that must be controlled, without including the diameter of the
lands. These are the distances between the four orifices. These critical
dimensions cannot be adjusted independently. One solution to this
difficult manufacturing problem is to build a valve with two separate
spools and to connect the two spools together by an adjustable linkage.
While eliminating the interdependence between dimensions, this method
introduces several other problems. The resulting valve is bulkier, and
care must be exercised if the play in the link is to be smaller than the
manufacturing inaccuracies of the original valve.
Although round valve ports are easier to manufacture than rectangular
ports, the latter are preferred in high-performance systems because the
variation of port area is then a linear function of stroke.
Several proposals have been advanced that avoid, to some extent at
least, the extremely close tolerances required by high-gain spool-type
valves. For instance, valves that operate on entirely different principles,
such as slide valves and nozzle valves, are used. The tolerances of slide
valves and nozzle valves, as discussed below, are not quite so critical as
those of spool valves. The nozzle valve (Sec. 11.10), however, is charac¬
terized by relatively high leakage, and the decrease in manufacturing
costs of the slide valve (Sec. 11.11) is largely canceled by the relatively
expensive linear torque motor required.
11.3. Pulsed Operation of Hydraulic Valves. A possibility for the
reduction of required manufacturing accuracy in spool valves lies in a very
interesting mode of operation proposed by Jackson1 and Chubbuck.2
The input signal to the valve is in the form of a pulse-length-modulated
wave rather than a continuous d-c value. That is, the input is a train of
pulses, each of constant amplitude and large enough to cause the valve
spool to move from full open in one direction to full open in the other.
The waveshapes are shown in Fig. 11.6. In Fig. 11.6a is shown the ideal
characteristic of the conventional proportional valve; this ideal is only
approximated in practice. In Fig. 11.66 is shown the input to the valve
for no signal in the pulse operation mode. The signal is converted into
the form of pulse-length modulation (PLM) by a modulator preceding the
valve. In Fig. 11.6c is shown the pulse wave train for a signal of maxi¬
mum amplitude applied to the PLM valve driver. The advantages of
the PLM mode appear to be several:3
1. The torque motor or electric actuator of the valve can be optimized

1 K. R. Jackson, U.S. Patent No. 2,655,940, 1950.

2 J. G. Chubbuck, “Acceleration Switching Hydraulic Servo,” Johns Hopkins
Univ. Appl. Phys. Lab. Tech. Rept., May, 1955, and Control Eng., vol. 4, no. 3,
March, 1957.
3 J. E. Gibson, F. B. Tuteur, and T. Mapes, A New Hydraulic Servo Valve for
Pulse-length-modulation Operation, ASME paper no. 57-A-128.
390 CONTROL SYSTEM COMPONENTS [CHAP. 11

for maximum force per ampere. This is not so in the conventional valve,
since efficiency must be sacrificed in order to obtain linearity. Thus in
the PLM mode the actuator may be made smaller and lighter.
2. In conventional valves, the flow gain is usually reduced at small sig¬
nal amplitudes. This does not occur in PLM mode operation.
3. Small-signal sticking is eliminated since in a sense the PLM mode
provides the spool with a massive
dither (see Sec. 11.6).
4. Small-opening Bernoulli forces,
discussed in Sec. 11.5, become
unimportant.
5. Valve lock, discussed in Sec.
11.6, is a less serious problem because
(o) Conventional input signal for ideal the valve is kept in motion and be¬
proportional valve
cause it is not necessary to design for
Valve
exceedingly small tolerances, as is
opening required in the linear mode. Fur¬
thermore, the emphasis in the design
0
of the torque motor is on maximum
force rather than on linearity. The
increased force tends to override the
(b) Pulse train for no input signal; the normal small-opening effects.
overage flow is zero
6. Manufacturing tolerances may
be relaxed, since the flow need not be
linear with valve stroke. Round
valve ports, for instance, are per¬
fectly satisfactory.
7. Since the torque motor for the
PLM mode can be designed effi¬
(c) Pulse train for maximum input signal
ciently, a single-stage valve becomes
Fig. 11.6. Input waveshapes for pulse
operation of hydraulic valves.
a possibility in many applications
where two-stage valves were formerly
required. The single-stage PLM valve with its PLM modulator will
usually be more economical than the precision two-stage linear valve and
its driver.
Naturally the pulsed valve has several disadvantages compared to
linear valve operation:
1. The output flow is in the form of pulses of fluid; thus the load must
be such that its low-pass filtering effect is sufficient to smooth the flow
adequately. Fortunately, many loads do provide adequate smoothing.
One case where pulsed operation would not be desirable would be where
the load was driven, through a multimesh gear train, by a rather light
hydraulic motor. t The motor could follow the high-frequency pulsing of
Sec. 11.3] valve-controlled hydraulic systems 391

the fluid and destroy the gear train by vibrating within its initial backlash
limits.
2. An original disadvantage of the PLM mode was the rather complex
electronic pulse-length modulator required. When it is realized, however,
that the magnetic amplifier is almost ideally suited to the PLM mode, this
problem is eliminated.1 The normal output waveshapes of magnetic
amplifiers are portions of the sine wave of the a-c supply frequency.
Thus the magnetic amplifier may drive the valve directly by controlling
the instant of conduction.
Tests have shown that pulse operation will also improve the perform¬
ance of high-precision valves originally designed for linear operation.
The frequency response will usually be improved, and resonances and
dips due to deadband and friction will be eliminated.
The pulsed mode of operation is by no means limited to spool valves.
It may be applied to all the common designs. Indeed, the concept of
pulse operation opens up possibilities for entirely new types of valves and
hydraulic systems.
In a sense, the pulse mode of operation removes the burden of high
precision from the mechanical design of the valves and increases the
requirements on the electric drive for the valve. Fortunately, the prob¬
lems involved in designing electronic circuits to convert a continuous
signal into the form of a pulse-length-modulated wave are relatively easy
to solve. The circuit can be a modified form of the Eccles-Jordan trigger
circuit and can be built with six or seven transistors including the power
stage, or a compact magnetic amplifier can be designed to provide the
required drive.
The basic pulse repetition rate must be at least twice2 the highest fre¬
quency that the valve must follow, and for negligible phase shift due to the
pulsing, the pulse repetition rate should be about ten times the highest
signal frequency. In practice, a factor of three to five times the highest
signal frequency is satisfactory. If the repetition rate is too high, the
valve will be unable to follow it, and if the repetition rate is too low, the
frequency response of the valve is limited.
A single-stage servo valve designed for PLM operation with a 400-cps
repetition rate provided a frequency response that was flat to above 100
cps.3 The repetition rate of 400 cps was chosen so that the magnetic-
amplifier driver could operate from a 400-cps supply. The maximum
theoretical operating frequency would be one-half this value, or 200 cps.
Actual operation was successful up to about 150 cps.

1 Ibid.
2 B. M. Oliver, J. R. Pierce, and C. E. Shannon, The Philosophy of PCM, Proc.
IRE, vol. 36, no. 11, pp. 1324-1331, 1948.
3 Gibson, Tuteur, and Mapes, op. cit.
392 CONTROL SYSTEM COMPONENTS [Chap. 11

As was noted above, the output flow of the PLM-operated valve must
be smoothed by the load. In this sense, the PLM mode of operation may
be considered to be the hydraulic equivalent of the relay amplifier (see
Chap. 3) operated in the stable oscillation mode. In fact, the static
characteristic found experimentally by Chubbuck for the pulsed valve is
identical in form to the static characteristic of the relay amplifier derived
in Chap. 3 and shown in Fig. 3.42.
Indeed analogues of on-off electronic circuits and devices will provide
the hydraulic-servo designer with many interesting possibilities for
investigation.
11.4. Flow-Pressure Relations in Orifices and Valves. In Sec. 11.1
it was assumed that the flow through the pilot valve was directly propor¬
tional to valve opening. Actually this is true only if the pressure drop
across the valve is ignored or assumed constant. This assumption is not
supported by the actual physical facts. Let us consider the actual
relationships. After defining certain general flow relations in orifices we
shall discuss more specific relations in spool-type valves.
By definition, a valve or orifice is a portion of the flow path that has a
restricted flow area and across which a large pressure drop may be
developed. It may be shown by the law of conservation of energy1 that
theoretical volumetric flow through an
orifice is

F = A (11.6)

where p is the pressure drop across the

orifice, A is the actual cross section of
the orifice, g is the acceleration of
gravity, and p is the density of the fluid.
Actually, the flow from an orifice is
less than the theoretical flow for two
reasons. First, owing to friction in the
flow of the fluid through the orifice,
Fig. 11.7. Orifice in a thin plate, the flow is reduced by a factor Cv, the
showing the contraction effect. velocity coefficient. Second, the stream
is contracted as it moves through the
orifice because most of the particles of fluid are moving in a curved
path, as shown in Fig. 11.7. The effective minimum cross section of the
stream is therefore less than the orifice cross section. The ratio of effec¬
tive area to orifice area is called Ca, the contraction coefficient. The
actual flow through an orifice is thus

1 Dodge and Thompson, “Fluid Mechanics,” McGraw-Hill Book Company, Inc.,

New York, 1937. ,
Sec. 11.4] VALVE-CONTROLLED HYDRAULIC SYSTEMS 393

F = CvCaA = CA (11.7)

where C is the discharge coefficient of the orifice. The value of C depends

on the shape of the orifice as well as on the area of the orifice and the
operating pressure. For usual operating conditions, pilot valves in
control systems have been found to have discharge coefficients between
0.6 and 0.8. The higher value applies for orifices with rounded edges, a
shape which reduces separation and turbulence in the flow.
Flow separation occurs at sharp bends or edges around which the fluid
must pass. Flow separation is the curling back or eddying that occurs
following such bends or edges; Fig. 11.8 shows several examples of flow

%&£/////// I
V77777777.

V
/

Pilot valve Orifice

Sharp bend
in pipe
Fig. 11.8. Examples of flow separation in typical control-system applications.

separation. Flow separation causes abnormally high pressure drops and

energy losses in the system. Smoothly rounded corners will decrease or
eliminate separation. In most control valves it is necessary to tolerate
separation, since the edges must be sharp.
Cavitation, another hydraulic phenomenon similar in certain of its
effects to separation, occurs in a hydraulic device whenever the fluid pres¬
sure is reduced to the vapor pressure of the fluid. The fluid then passes
into its vapor state. The small vapor cavities or bubbles are carried into
regions of higher fluid pressure, where they collapse. Cavitation causes
a higher pressure drop than would otherwise be expected and resultant
energy losses. What may be equally important, the collapse of the
cavities is accompanied by sudden shock forces that may reach high
magnitude. These shock forces may induce metal fatigue and pitting of
the surfaces exposed to the cavitation. There also can be an erosion
effect from bubbles forming in the pores of the metal and bursting when
the metal is carried into a high-pressure region.
Cavitation is occasionally useful in hydraulic-system design. Certain
devices must be protected by limiting the maximum flow, should the
 394 CONTROL SYSTEM COMPONENTS [CHAP. 11

downstream pressure drop; for example, the fuel flow must be limited in
case of combustion failure in a rocket.1 A smooth decrease in tube
diameter, called a venturi tube, is placed in the fuel line, as shown in
Fig. 11.9. The fluid velocity must increase through the constricted area.
From the law of conservation of energy the pressure head must decrease
as a result of the increase in velocity head. If the pressure drop is suffi¬
cient to result in a pressure below the vapor pressure of the fluid, cavita¬
tion will occur, as shown in the figure. Thus the equivalent cross section
of the tube is reduced, and the flow is limited at some upper limiting flow
rate. At normal flow rates the constriction of the venturi tube is so
slight as to cause only a negligible pressure drop.

The pressure-flow relations, even in an ideal valve, are nonlinear; how¬

ever this does not make analysis impossible. For large-signal operation
the characteristic curves can be employed in a graphical solution, and for
small-signal analysis, linearizing assumptions can be made about an
operating point to allow an equivalent linear transfer function to be
developed.
We may apply the pressure-flow relations in an orifice to derive the
pressure-flow characteristics of the spool-type valves discussed in the
previous section. Before it is possible to employ Eq. (11.7), however, it
is necessary to relate the stroke x of the valve spool to the cross-sectional
area A in the equation. If the valve has square ports, this relation will
be linear, and it is merely required that the width of the port be placed
1 N. M. Sverdrup, Theory of Hydraulic Flow Control, Product Eng., vol. 26, no. 4,
p. 165, 1955.
Sec. 11.4] VALVE-CONTROLLED HYDRAULIC SYSTEMS 395
in the relation. However, if the port is circular, the area as a function of
stroke is as shown in Fig. 11.10. Even in this case the cross-sectional area
is approximately proportional to stroke.
Let us now consider the effect of four orifices in a bridge-type circuit of
the sort found in the typical spool-type valve (see Fig. 11.11). We shall

P-0 Fc^
1 I
V77^\

'<£22}
P=Ps
Fa a I
Fb-A
7F77^\
Fz,Pz
///// W'
p — 0 Fd^d
V777}

0 20 40 60 80 100 120
Stroke x os a percent of diameter D

Fig. 11.10. Cross-sectional area of a cir¬

140
I I.
Fig. 11.11. A typical spool-
cular valve port as a function of stroke, type valve with underlap
normalized to total area and full stroke. x0.

assume that the orifice area is proportional to the displacement x, and

that the valve is symmetric and has an underlap x0. Also, it will be
assumed that the valve operates with a constant supply pressure ps and
that the pressure in the return line is zero. From Eq. (11.7) we may
write for the four orifices

Fa = k(x0 + x) \Zps — Vl x0 + x > 0; otherwise Fa = 0

Fb = k{x0 — x) y/p8 — P2 x0 — x > 0; otherwise Fb = 0
(11.8)
Fc = k(x0 — x) -%/pi x0 — x > 0; otherwise Fc = 0
Fd = k(x0 + x) VP2 x0 + x > 0; otherwise Fd = 0

and the load flow will be

F = 1Fa — 1F c (11.9)

Thus we may write, combining Eqs. (11.8) and (11.9), that

Fl = k(x0 + x) \/ps — pi — k{x0 ~ x) y/p[ (11.10)

Equation (11.10) can be simplified if we assume that the valve is sym¬
metrical. This implies that ps = pi + pz- Also, in general, the load
396 CONTROL SYSTEM COMPONENTS [Chap. 11

pressure Pl = p 1 — p2. Equation (11.10) may therefore be rewritten as

Fl = k(x<, + x) JP’ ~ - k(x0 - x) JP’ + Pl (11.11)

or if we make the natural definition pv = ps — Pl, where pv is the valve

pressure, then Eq. (11.11) may be written

Fl = k(x0 + x) — k(x0 — x) yjps — ^ (11.12)

Equation (11.12) holds in the underlap region, where |rc| < x0. Outside
of the underlap region one term of the equation falls out, and only the
first or second term remains, depending on whether x > x0 or x < — xa.

Fig. 11.12. Flow plotted against pressure for an underlapped valve operating in the
underlapped region and with constant supply pressure.

The reason for manipulating the relations into the form of Eq. (11.12)
is that characteristic curves for the valve may be calculated and employed
in an analysis, as shown in Sec. 11.7. In Fig. 11.12 is shown the charac¬
teristic calculated from Eq. (11.12) of an underlapped valve operating in
the underlapped region with a constant supply pressure. It is assumed
here that area of the valve orifices is proportional to stroke. The orifice
coefficient used was 0.7, which is a reasonable assumption. It will be
noted that the characteristic is essentially linear in this region. This is
one of the reasons that the underlapped valve is preferred to the over¬
lapped valve. •
Sec. 11.5] valve-controlled hydraulic systems 397

In Fig. 11.13 is shown the characteristic of the valve outside of the

underlapped region. In this case one of the terms of Eq. (11.12) was
dropped, as mentioned above, in order to calculate the characteristic.
The characteristics for positive and negative stroke are superimposed as
shown, for convenience in graphical analysis. Figure 11.13 may also be
used for a zero-lapped valve directly and for an overlapped valve if the
portion of the stroke in the overlap region is accounted for separately.
The calculation of valve characteristics for a constant supply flow rather
than constant supply pressure is left as an exercise for the reader.

Fig. 11.13. The flow-pressure characteristic of an underlapped, spool-type valve for

operation beyond the underlap region.

11.5. Axial Hydraulic Reaction Forces in Spool-type Valves. The flow

of fluid through the orifices of a control valve causes a hydraulic reaction
force, also referred to as Bernoulli force. This force has two components,
a radial and an axial component. The radial component tends to push
the valve spool sideways against the sleeve and causes sticking. It is
usually minimized by locating the valve ports symmetrically about the
spool, but a component may remain if the leakage paths between spool
and sleeve are not symmetrical. The radial component of force is con¬
sidered in more detail in Sec. 11.6.
The axial force is usually in such a direction as to close the valve. This
force has been considered in some detail by Lee and Blackburn.1 The
1 S. Y. Lee and J. F. Blackburn, Contributions to Hydraulic Control I: Steady-
state Axial Forces on Control-valve Pistons, Trans. ASME, vol. 74, p. 1005, 1952.
398 CONTROL SYSTEM COMPONENTS [Chap. 11

valve configuration studied is shown in Fig. 11.14. The flow into the
chamber through de is essentially unrestricted if the opening x is small
compared to the circumferential length of the orifice; and if the flow is
assumed to be irrotational, nonviscous, and incompressible, the solution
of the flow pattern in the chamber is the solution of Laplace's equation
for the configuration. This solution1 yields a 6 angle of 69° when the
valve is square (</> = 90°) and when there is no radial clearance. With 6
known, the force on the valve spool can be found by conservation of
momentum of the flow in the valve chamber. Since the area de is large
with respect to the orifice area, the velocity of flow through de is negligible
as compared with the velocity of flow through the orifice. Axial pres¬
sures on the ends of the chamber are
In
/ equal. Thus the flow through the
Valve
body

777777a f
\
L // /
C\ Yi///f'/////7
F ,<t>
orifice produces a change in momentum
that is balanced only by an axial piston
force. This change of momentum may
7V h Yh be expressed in terms of the fluid flow by
YW7777777777777777777TYA7

Spool iI X

Fig. 11.14. Configuration for deriva¬ 7, <J,’> - T

tion of axial reaction force.
where Mv is the momentum, F the flow
rate, p the density, v the velocity of the fluid at the vena contracta (area
ab in Fig. 11.14), and g the acceleration of gravity. Note that F is con¬
sidered positive when the flow is out of the controlled orifice. The axial
component of the force is therefore given by

*5"axial = — — COS 6 (11.13)

— u

The minus sign arises from the fact that the expression is for the force by
the fluid on the container; this is the negative of the force required to
accelerate the fluid. By Bernoulli's equation the velocity v is given by

where p is the pressure drop across the orifice. Also, the flow rate F may
be expressed in terms of p by Eq. (11.7). Hence the axial force becomes

cose
\ P 9 \ P
= —2CAp cos 0 (11.14)

1 R. Von Mises, “Berechnung von Ausfluss-und Ueberfallzahlen,” Z. Ver. deut.

Ing., vol. 61, p. 494, 1917.
Sec. 11.5] valve-controlled hydraulic systems 399

where C is the orifice coefficient and A is the cross-sectional area of the

orifice. For an orifice coefficient of 0.6, the axial force per orifice is

Saxial = 0.43 Ap lbs

if the pressure is expressed in psi and the area in square inches. In a

symmetrical valve the inlet and outlet orifices are identical and are in
series. Therefore, the total force is twice the force per orifice. However,
the total pressure drop across the valve is also twice the pressure per
orifice. Thus, the total axial force may be written

$T = 0A3pv

where pv is now the total valve-pressure drop.

Note that in Fig. 11.14 both F and v cos d are positive; hence the force
is negative (to the left) and tends to close the valve. If the flow is
reversed, both F and v cos 6 would be negative, and the force would still
be in the same direction.
Lee and Blackburn1 have investigated the more realistic case of a valve
with radial clearance, and with land edges of finite radius. Both of these
imperfections of the valve result in an increased orifice area and a
decreased angle 6 of the jet. Hence they both tend to increase the force
generated per unit displacement of the valve spool, particularly for small
spool displacements. A typical set of curves of force versus displacement
for different clearances and radii is shown in Fig. 11.15. The pressure
drop across the orifice is constant in this figure.
The hydraulic reaction force given in Eq. (11.14) is that obtained for
steady flow. When the flow is changed by the opening or closing of the
valve, an additional force is generated. It is not unreasonable that such
a force should exist. Thus, consider the valve configuration shown in
Fig. 11.14 and assume that the valve spool is suddenly moved slightly to
the left so as to reduce the orifice opening. The average velocity of the
oil volume in the chamber must then be reduced, and the resulting
momentum change would be expected to resist the leftward motion of the
valve. The same result is obtained if the valve spool is moved to the
right: the momentum change of the fluid will resist the spool motion.
On the other hand, suppose that the fluid flow is reversed: into the cham¬
ber at ab and out at ed. Again assume that the spool moves to the left.
Now the momentum of the fluid is in the opposite direction, and tends to
pull the spool along with it. Hence the transient force is in a direction to
aid the spool motion, Lee and Blackburn2 have shown that the total

1 Lee and Blackburn, loc. cit.

2 S. Y. Lee and J. F. Blackburn, Contributions to Hydraulic Control II: Transient-
flow Forces and Valve Instability, Trans. ASME, vol. 74, pp. 1013-1016, 1952.
400 CONTROL SYSTEM COMPONENTS [CHAP. 11

reaction force at each orifice may be written

= - p- Fv cos e + p- L ^ (11.15)
g g ai

where L is the axial distance between centers of incoming and outgoing

flows (see Fig. 11.14), and where all other quantities are as defined previ¬
ously. The distance L is negative when the flow leaves the chamber
through the controlled orifice (as in Fig. 11.14), and it is positive when the
flow enters through the controlled orifice. The flow rate F is again
assumed to be positive for flow out of the chamber. The first term of
Eq. (11.15) is identical with Eq. (11.13) and represents the steady-state

Fig. 11.15. Hydraulic-reaction force curves: (1) Ideal case, clearance = 0, radius = 0;
(2) clearance = 0.0001 in., radius =0; (3) clearance = 0.0005 in., radius = 0.0003
in.; (4) clearance = 0.0003 in., radius = 0. {From Lee, Blackburn)

component of the force. The second term is the transient force. For the
situation shown in Fig. 11.14 the transient force is negative, i.e., in a
direction to close the valve, when the valve is being opened and the flow
is increased. This is in accordance with the qualitative discussion of the
transient-force phenomena given above.
It is again convenient to convert Eq. (11.15) to a form such that the
force is a function of pressure, rather than flow. Using Eq. (11.7) we
obtain

= ~2CAV cos e + CL4vpW + CLA Tt

If, as in Sec. 11.4, we make the further assumption that A is proportional

Sec. 11.5] valve-controlled hydraulic systems 401

to x, we can set A = wx where w is the effective width of the orifice. Then

the axial force becomes

ffaxiai = —2Cwxp cos 6 + CLw p ^ + CLwx ^ (11-16)

The approximation is often made that the pressure drop across the valve
is constant. If this assumption is made, dp/dt = 0, and the last term in
(11.16) above vanishes. Equation (11.16) can then be written in the sim¬
ple form

= -K& - K,jt (11.17)

where Ki and K2 are constants. Thus, to a first approximation the

Bernoulli force may be represented by an equivalent spring and coefficient
of viscous damping as far as the valve-spool dynamics are concerned.

Supply Supply

Fig. 11.16. Damping in hydraulic valves: (a) valve with positive damping; (6) valve
with negative damping.

Since the steady-state force acts to close the valve the equivalent Ber¬
noulli spring aids the centering springs usually employed with valves.
The damping term may, however, be positive or negative depending on
the sign of L. Usually a valve consists of several orifices in series, and the
Us for different orifices usually have opposite signs. Thus, consider the
valve shown in Fig. 11.16a. In this valve L2 is negative and Li is positive.
Orifice a produces negative damping proportional to Li and orifice d pro¬
duces positive damping proportional to L2. The net damping in the valve
is therefore positive, since L2 is greater than L\. In Fig. 11.165 is shown
a valve for which L2 is less than Lh and in this valve the damping from the
hydraulic reaction force subtracts from the viscous damping. Since the
hydraulic reaction force is a function of the pressure, it is therefore possi¬
ble for the valve of Fig. 11.166 to exhibit negative damping for sufficiently
high pressures. This may, of course, result in instability and violent
oscillation of the valve.
402 CONTROL SYSTEM COMPONENTS [Chap. 11

Valves have been designed that counteract the effect of the steady-state
axial force developed by the orifice.1 Figure 11.17 shows the configura¬
tion of a force-compensated valve and a detail of one of the ports. In this
device the inlet line to the chamber is constricted. The flow, entering at
an angle 6i, tends to close the valve as usual. The chamber of this valve
is shaped like a turbine bucket, and the flow leaving at an angle 02 is in
such a direction as to generate a reaction force that tends to open the
valve. In addition, the circulating flow entering the chamber at 03 also
aids in keeping the valve open. While the design of this valve is essen¬
tially empirical, Lee and Blackburn report a reduction in the steady-state
axial force by a factor of 50 compared with a conventional valve, and they
state that the redesigned valve is essentially perfectly compensated
throughout its operating range. The small notch shown dotted in Fig.

In Out

(A)
Fig. 11.17. Force-compensated valve. {From Lee and Blackburn)

11.17a simplifies the manufacture of the spool and does not appreciably
affect the compensation.
In addition to special shaping of the valve lands, there are several other
solutions to the problem of reducing axial forces on the valve spool. A
two-stage valve may be designed to solve this problem, or an entirely
different operating principle may be employed. Both of these possi¬
bilities are discussed below.
11.6. Radial Hydraulic Forces in Spool-type Valves. Radial or lateral
forces on valve spools, if not symmetrical around the periphery of the
spool, can cause the spool to move radially and lock or freeze against the
cylinder wall. This problem of hydraulic lock has become more acute
with the high pressures and low clearances of modern hydraulic systems.
Hydraulic lock may arise from several sources. One possible cause of
hydraulic lock is the radial component of the Bernoulli force discussed in
the previous section. In order to eliminate this possibility, the valve
ports must be made symmetric around the valve periphery. A second
possibility is dirt or metal chips that lodge in the radial clearance between
piston and cylinder, thus wedging the piston. This possibility can be
minimized by filtering the oil in the system.
1 Ibid.
Sec. 11.6] VALVE-CONTROLLED HYDRAULIC SYSTEMS 403

The most common source of hydraulic lock is a radial force produced

hydrodynamically. Figure 11.18 shows a simplified view of a piston and
cylinder. If the piston is a perfect right-circular cylinder and if the walls
of the cylinder are parallel to the piston, there will be no net radial force
due to the leakage flow. For any portion of the leakage flow, the clear¬
ance is constant, and the pressure gradient is uniform for the length of

Fig. 11.18. Perfect right-circular cylindrical piston and cylinder, eccentric to one
another.

the leakage path if we assume that the flow is laminar and that the veloc¬
ity initially is zero. Thus even for an eccentric right-circular cylinder the
radial pressures on the piston balance around the periphery of the valve.
The question naturally arises as to the forces involved if this perfect
piston is somehow canted in the perfect cylinder, as shown in Fig. 11.19.
If we consider two leakage paths l\ and U on opposite sides of the piston,
the pressure distribution along the leakage paths will be as shown in Fig.
11.196 as a result of the interrelation of pressure head and velocity head.
Thus throughout the length of the leakage path there will be a net pressure

Pressure difference
tending to allign
piston axially

Distance along leakage path

[b)
Fig. 11.19. Perfect right-circular cylindrical piston canted lengthwise in perfect
cylinder.

difference tending to drive the piston to the right in Fig. 11.19a or, in
general, toward the side with the constricted upstream flow. When the
piston arrives at the cylinder wall, it will align itself with the wall, and the
pressure drops along any leakage path will become equal. There will
thus be no net force holding the piston against the cylinder, and hydraulic
lock will not occur.
404 CONTROL SYSTEM COMPONENTS [Chap. 11

A somewhat more practical case than that of the perfect piston and
cylinder is the case of a tapered piston in a perfect cylinder. Figure 11.20
shows an eccentric tapered piston. If the clearance is small, the leakage
flow may be considered two-dimensional, and the relationship for laminar
flow between two flat plates may be employed with little error to find the
relationship between leakage flow and pressure.1

Fig. 11.20. An eccentric tapered piston.

By the same process2 that is used to determine the Hagen-Poiseuille

laminar-flow relation in a pipe, the average velocity of flow at any cross
section of a leakage path may be determined as

1 dp y2
(11.18)
3/x dx 4

where y is the distance between the plates. For an element of unit width
the volumetric flow will thus be

Fx =_1._dpy^
(11.19)
z 3^t dx 4
where 2 is the width.
For a linear taper the relationship between y and x along a leakage path is

y = c -f- kx (11.20)
where c is the clearance between piston and wall, as shown in Fig. 11.20,
and k is the taper of the piston. Solving for p by integration,
6 nFxk
V = z(c + kx)2 + Ci (11.21)

The constant of integration, c1} can be evaluated by setting the pressure

equal to pi where x equals zero. Then

6 \xFxk 1
V = (11.22)
z (c + kx)2
Thus pressure is a parabolic function of x and varies with the taper k and
clearance c. A smaller c results in a lower pressure at any station along
1 I). C. Sweeney, Preliminary Investigation of Hydraulic Lock, Engineering, vol.
172, pp. 513-516, 580-582, 1951.
2 Dodge and Thompson, op. cit.
Sec. 11.7] VALVE-CONTROLLED HYDRAULIC SYSTEMS 405
the leakage path, as shown in Fig. 11.21. Thus there is a net radial pres¬
sure along the path in such a direction as to reduce the clearance. The
result is hydraulic lock.
It may be seen from Eq. (11.22) that, if the taper k is negative, the net
force will be in such a direction as to center the piston. For this reason,
since some tolerance must necessarily be accepted in any manufacturing
process, pistons are sometimes deliberately manufactured with a slight
negative taper in order to eliminate the possibility of a positive-tapered
piston being produced from a nominally zero-taper design. Quite often,
however, this solution to the problem of hydraulic lock results in excessive
leakage flow.
It has been found by Sweeny1 and others that, if several radial grooves
are cut into the surface of the piston land, to eliminate the pressure differ¬
ential around the spool, hydraulic lock can
be eliminated. The size of the grooving
should be large with respect to the clear¬
ances involved. Both the depth and
width of the grooves should be at least ten
times larger than the clearance if the clear¬
ance is of the order of a few ten-thou¬
Fig. 11.21. Pressure drops in leak¬
sandths of an inch, in order to permit free
age paths of tapered piston.
flow around the periphery. Sweeny found
that five or six of these grooves reduce the locking force to about 1 per
cent of its previous value in a typical spool.
Forces due to factors other than the pressure gradient caused by leak¬
age flow are also sometimes important contributors to hydraulic lock.
These factors include static friction, collection of dirt, and metal-to-metal
contact. Sweeny has found that these forces including hydrodynamic
lock itself build up rather slowly, taking 4 to 5 minutes to reach maximum
in some cases. A common solution to the static sticking problem is to
agitate the pilot valve at some high frequency either mechanically or by
superimposing an a-c component on the valve-driving signal. This
agitation is called dither and must be at a frequency high enough to
assure that the controlled elements are unable to follow the rapid motion
of the pilot valve.
11.7. Graphical Analysis of Control Valves. For large input signals
the nonlinearity of valve characteristics cannot be ignored. Analytical
methods of treating the problem are possible, but graphical techniques
appear more convenient. The similarity between the characteristic
curves has led to the extension of vacuum-tube analysis techniques to
hydraulic valves. In this section we shall consider several of these tech¬
niques in detail. It should be noted that these nonlinear analyses do not
1 Sweeny, op. cit.
406 CONTROL SYSTEM COMPONENTS [Chap. 11

permit the evaluation of equivalent gains or time constants for the ele¬
ments; thus the approximations considered in the following section, which
do permit this, are more often employed than these more exact methods.
It is possible to construct load lines on the valve characteristics that
relate flow and pressure, such as those shown in Sec. 11.4. A load line is
the pressure-flow relation for the load that is driven by the valve. Since
the valve and the load are in series, the same flow must exist in each, and
their two pressure drops must sum to the supply pressure. Since the load
line and the valve characteristic represent relations that must be simulta¬
neously satisfied, the point of operation must occur where the two curves

Fig. 11.22. A load line for a resistive load on an ideal valve characteristic. (Cunning¬
ham])

intersect. The load can be assumed to be so close to the valve that trans¬
mission-line effects can be ignored, or the effects of the line may be
included in the load. Two load lines of this type are illustrated in Fig.
11.22. The valve characteristic curves on which these load lines are
drawn are those of an ideal valve which has zero lap and square ports.
The relation between pressure across the load {p£) and flow through it is
assumed linear for the solid line. The load line is constructed relative
to the assumed supply pressure (ps). The actual operating point of the
system must lie on the load line at a point determined by the valve
opening (x). If x = 20, the operating point Q will be as shown in the fig¬
ure. Quite often the load presented to the valve has a nonlinear relation
between pressure and flow. We might refer to this as a nonlinear
hydraulic resistance. For instance, in Fig. 11.22 is shown dotted a
parabolic load1 line such as would be presented by an orifice. If the
Sec. 11.7] VALVE-CONTROLLED HYDRAULIC SYSTEMS 407
parabola passes through the operating point Q with the same slope as
the linear resistance, it will have an equation

pv = 1,270 - 100F2 (11.23)

It is possible to construct on the characteristic curves, lines of constant

delivered power.1 The power delivered to the load is proportional to the
product of load pressure pL and flow F.

(JlA F (o'!i
\in.2/ \min/
finA J_ (lit L hP
\gal / 12 \in./ 33,000 ft/min
(11.24)

or W — 5.83 X 10~4plF hp (11.25)

where pressure is in pounds per square inch and flow is in gallons per min¬
ute. For a given supply pressure, curves of constant load horsepower

Fig. 11.23. Lines of constant delivered power superimposed on a valve characteristic.

('Cunningham)

may be calculated from Eq. (11.25). These curves are hyperbolas and
are shown plotted for ps at 1,400 psi in Fig. 11.23. For a given value of
x we can find the maximum power that can be delivered to the load and
the pressure at which this occurs. From Eq. (11.12) we may retain the
first term for a zero-lap valve and write for a constant x that

pv = kQF2 (11.26)

and the load power may be written as

W = k2(ps - pv)F = k2(ps - k0F2)F (11.27)
1 W. J. Cunningham, “The Hydraulic Control Valve: An Analysis of Its Perform¬
ance,” Yale Univ. Dept. Elec. Eng. Tech. Rept., June, 1950.
408 CONTROL SYSTEM COMPONENTS [Chap. 11

Maximizing by setting dW/dF to zero yields pv = pj3 and

2 ps
Vl = (11.28)
3

This contour or straight line of maximum power is shown in Fig. 11.23.

In general the operating point should lie to the right of this boundary.
All of these constructions that have been made above allow the designer to
set the optimum region of operation for the valve and motor and to establish
the response to a given input. A dynamic analysis may be made quite
easily by moving the operating point along the load line in any manner
prescribed by the input signal and picking off the required values of pres¬
sure and flow. A somewhat more elaborate procedure is required if the
load is not a pure resistance.
In actual practice the pure-resistance load is quite rare. More often
the load consists of resistance plus the inertia of the moving parts.
Under these conditions the flow equation may be written

dF
ps = Pv + RlF + Jl (11.29)

where JL is the equivalent hydraulic inertia of the load. A methodical

graphical technique may be employed to solve for the response of a system
to various loads.1
In this construction x is the actual piston displacement; thus reaction
forces and time lags between the input to the torque motor or solenoid are
not included. They must be taken care of in the analysis of the remainder
of the control system. Let us assume that a step-function input signal is
impressed on the valve and that we desire to determine the output as a
function of time. If at some instant the flow has a value F, and an
incremental change AF occurs, Eq. (11.29) becomes

ps — RlF, — pv = ^Rl + AF

or AF = V* ~ R'~ f v- (11.30)

The numerator of this equation is the distance on the characteristic

between the load line and the valve characteristic.
The construction necessary is shown in Fig. 11.24. It will be assumed
that ps = 1,400 psi and that, at t = 0, x is suddenly changed from 0 to 40.
Assume that the hydraulic load has a resistance of 220 psi/gpm and an

1 Cunningham, op. cit., from which this example is adapted. Preisman, “ Graphical
Constructions for Vacuum Tube Circuits,” McGraw-Hill Book Company, Inc., New
York, 1943.
Sec. 11.7] valve-controlled hydraulic systems 409

inertia of 3.11 (sec-lb/in.^/gpm.1 The load line of the hydraulic resist¬

ance is drawn on the static characteristic as before. At is chosen as 0.001
sec for convenience, and the operator RL + JL/At is evaluated as 3,330
psi/gpm. At zero time there is no flow, even though the valve is open;
thus RlFo and pv are zero. Thus from Eq. (11.11), the full supply pres¬
sure is effective in causing an increment in F. Next, from the point
pv = ps on the horizontal axis, a line is drawn with a slope of Rl + Jl/At.
The intersection of this line with the static characteristic gives the flow Fi
at the end of the first interval of time. This point is located on the Rl

Fig. 11.24. Construction for valve supplying a load with inertia. (From Cunningham)

curve, and a new line at the same slope locates the flow at the second
interval. The major approximation implied here is that the rate of
change of flow is constant throughout the time interval chosen. Accuracy
is improved by decreasing the time interval, but of course this requires
more computation. With our choice of 0.001 sec for the time interval,
about 12 construction steps are required to arrive at the operating point
at the intersection of the load line and the characteristic curve. In Fig.
11.25 is shown this response plotted against time. The response for a
linear approximation for the valve characteristic is shown dotted in the
same figure for comparison.

1 This corresponds to a hydraulic motor with a displacement of 0.38 in.3 per revolu¬
tion with a mechanical load consisting of a resistance of 0.0174 ft-lb-sec and an
inertia of 2.5 X 10-4 slug-in.3 per revolution.
410 CONTROL SYSTEM COMPONENTS [Chap. 11

The linear approximation was calculated from the equation

dF
ps — (Rv + Rl)F + J (11.31)
dt
which has a solution

Ps f Rv + Rl A
F = - 1 — exp
xp (- -J t) (11.32)
Rv + Rl
and Pl = ps — FRV (11.33)

In these equations Rv represents an equivalent resistance for the valve.

It seems reasonable to choose Rv so that the final value of flow will be the

Fig. 11.25. Response of valve and load with inertia to a step input. (From Cunning¬
ham)

same for the linear approximation and the more accurate representation.
The value of Rv is 482 psi/gpm under this assumption. We see from the
figure that the flow increases more rapidly for the nonlinear construction,
thus implying that the effective resistance of the valve orifice is less at
small openings than would be assumed in the linear analysis.
It is possible to extend this construction technique to more involved
inputs. For example, consider a sinusoidal driving signal in x. Let

x = sin (27r 100 (11.34)

 Sec. 11.7] VALVE-CONTROLLED HYDRAULIC SYSTEMS 411

Since the values of x will be both positive and negative, we must employ
a composite characteristic curve. This second family of static curves is
plotted below the horizontal axis with the pv scale reversed and the ps
points coinciding.
Let us suppose that it is desired to calculate the values of flow at 10°
intervals in the sinusoidal variation; thus

At = (360 X 10) = °'00278 sec

We must then construct characteristic curves for values of x at 10° inter¬

vals in cot. The solution is begun as before by erecting the Rl line, as
shown in Fig. 11.26. The value of the operator RL + J l/At with the
Fy Qol/min

Fig. 11.26. Construction for inertial load and sine-wave driving function. (From
Cunningham)

assumed value of At is 1,340 psi/gpm. A line is drawn from ps at this

slope, intersecting the static characteristic that applies at the end of the
first time interval. The construction continues as before, employing,
however, the particular static characteristic of interest at the particular
moment of time. A curve may be drawn through the successive points,
defining the operating path for the system. This type of solution is per¬
fectly general and will include both the transient and the steady state.
In this particular example, the path closes at each cycle, and the two
parts of the solution cannot be distinguished. The solution may be
412 CONTROL SYSTEM COMPONENTS [Chap. 11

considered as a transient, repeated each half cycle. This is so, since in a

zero-lap valve the flow must be zero whenever the stroke is zero. The
plot of the various quantities of interest in this example is shown in Fig.
11.27 as functions of time. The flow is a distorted sine wave which lags
the wave of pressure, as would be expected in a load with inertia. The
small cusp in the pressure wave, which is due to the zero lap, would proba¬
bly not be observed in practice because of practical effects such as leakage.
It has been pointed out that the nonlinear analysis does not yield a
time constant or gain constant for the hydraulic valve. Indeed, by the

C7»

^ kT
g 600
E
h 2

40 400

1
20 200

0 0 0

-20 -200
-1

-40 -400

-2
-600
Fig. 11.27. Time response of a system with inertia for a sinusoidal driving function.
{From Cunningham)

very definition of a nonlinear element, its transfer function depends on the

input. However, it would be instructive to compare particular values
obtained in the examples above, for any possible correlation. From the
step-function analysis shown in Fig. 11.25 one can deduce a time constant
of 4 msec and a gain of approximately 0.05 gpm per unit of stroke.
Figures from the sinusoidal analysis do not compare well with these
because of the rather unrealistic assumption of zero lap. The reader may
show that at 20 cps the peak of the flow curve lags the x curve by about
15°, whereas a linear system with a time constant of 4 msec would have a
phase lag of about 26°. If a more elaborate analysis were undertaken
which included underlap, the correlation would be improved.
11.8. Linearized Small-signal Analysis. While the graphical analyses
considered in the previous section yield more accurate results under given
Sec. 11.8] VALVE-CONTROLLED HYDRAULIC SYSTEMS 413

conditions than a linear approximation, they are rather long and compli¬
cated, and they do not permit extension to general inputs. Under condi¬
tions of small driving signals it is possible to make good linear approxi¬
mations with resultant simplification of the analysis.
The first linear approximation is suggested by an examination of Fig.
11.12, the flow-pressure relations for an underlapped valve in the under¬
lapped region. The variation is almost linear in this region. The sys¬
tem is analogous to an electric bridge circuit, as shown in Fig. 11.28.

Fig/11.28. Bridge analogue for underlapped valve in the underlapped region.

If the lapping is symmetric, the bridge is balanced when the valve is

centered, and no fluid flows in the load. Motion of the spool down
increases resistances Ra and Rd and decreases Rc and Rb. We can repre¬
sent this by writing
Rc — R — x
Ra — R x
(11.35)
Rb = R — x
Rd — R T x
where R is the orifice resistance when the spool is centered and x is a
parameter proportional to stroke. From Fig. 11.28 it may be seen that

R = jU (11.36)
X leak

where ps is the supply pressure and Fleak is the total leakage flow with the
piston centered. The reader may show that

_xps
(11.37)
RRl + R2 — x1
Thus for small values of x we have the linear transfer function

FL _ Ps
(11.38)
x (RRl H- R2)

A more important application of linearization procedures may now be

made for operation of a valve in general. The reader may recall that
414 CONTROL SYSTEM COMPONENTS [Chap. 11

vacuum tubes are also basically nonlinear devices, since the plate current
as a function of plate or grid voltage is given by Child’s law:

ip = c(aeg + ep)^ (11.39)

However the analysis of vacuum-tube circuits proceeds with little diffi¬

culty. As is the case with vacuum tubes, hydraulic valves may be
analyzed either graphically as above, when large-signal behavior is
required, or on a small-signal “linearized” basis.
To simplify the linearization procedure, we assume that the valve is
symmetrical, that there is no leakage in the load, that supply pressure ps
is constant, and that the sump pressure p0 is zero; i.e.,

Fi = F2 = Fl
(11.40)
P1 — P‘2 = PL
Pi + P2 = Ps (11.41)
Po = 0 (11.42)

where FL is the load flow and Pl is the pressure across the load, and where
other symbols are defined as in Fig. 11.11. Equation (11.41) shows that,
if pi goes up a certain amount, p2 goes down by the same amount.
The procedure is based on the assumption that, for small variations of
the variables about an operating point, the characteristics are essentially
linear. Thus, consider the nonlinear function between flow, pressure, and
valve displacement:
Fl = FL(pL,x) (11.43)
Differentiation gives
0F t dF t
dFL = —F dx + —F dpL (11.44)
dx dpL

If the differentials are replaced by small increments of the variables, this

may be written as
A Fl = G Ax — Y A pL (11.45)
or, dropping the A’s,
Fl = Gx — YpL (11.46)

where G = SFl/^x and Y = —dFL/dpl. G and Y will be approximately

constant if the variation of x and pL is not excessive. This procedure is
exactly analogous to the process of differentiating Child’s law [Eq. (11.39)]
with the result

= 8AC(<*eg + ep)» A i (11.47)

dep rp
dip
and % Ca(aeu + ep)^ — gm (11.48)
dea
Sec. 11.8] valve-controlled hydraulic systems 415

As an example of a typical G, take the relation for an underlapped valve

given in Eq. (11.11) and find dFL/dx.

Thus G = V2 k [\/ps ~ Pl + \/ps + Pl] (11.50)

Similarly the reader may compute Y and for given parameters establish
the transfer function for the valve. Equation (11.50) holds only in the
underlap region, and when the relation for the valve outside the underlap
region is found, it will be noted that the gain drops by a factor of 2. For
large-signal operation the high-gain region about center may sometimes be
ignored.
Once having accepted the restrictions implied in the small-signal
analysis, the operation of the valve with reactive loads of all sorts may be
considered. As a matter of fact, we may go further and consider the
effect of the mass and friction in the valve stem itself. To do this, we
assume that the input to the valve is a force source of finite stiffness. In
order to get a complete expression of the various reactions taking place
in the valve, the hydraulic reaction force must be considered. This force
has been discussed in Sec. 11.5 and was shown to be a function of pressure,
displacement, rate of change of pressure, and rate of change of displace¬
ment [see Eq. (11.16)]. Thus using the symbol for reaction force, we
can say

$h = $(pl,Pl,x,x) (11.51)

where p = dp/dt, and x = dx/dt. For small variations of the variables

Eq. (11.51) may then be written in the form

d5 . d& . . d& . d$ .
3* = Vl + ^Vl+^x + ~x
dpL^ ' dpL
— Ai Pl + X2 Pl + KhX + B (11.52)

The parameters Ai, X2, Kh, and Bh are considered constant for small varia¬
tions of the variables, and may be obtained by differentiation of Eq.
(11.16). If measured characteristics relating the reaction force to pres¬
sure and displacement are available, the parameters Ai and Kh may be
obtained from them. This procedure has the advantage that the actual
force characteristics rather than characteristics for an ideal valve are
used, but the parameters X2 and Bh cannot very well be obtained this way.
A compromise solution might therefore be used by which Ai and Kh are
determined from experimental curves and X2 and Bh from the ideal
relation.
416 CONTROL SYSTEM COMPONENTS [Chap. 11

To determine the equation of motion of the valve stem, it is assumed

that the applied force is and that this force is opposed by the hydraulic
reaction fo, the force of accelerating the mass of the valve stem, that is,
M d2x/dt2, the force of friction between the stem and valve body, that is,
B dx/dt, and the spring force Kx. Hence the equation of motion may be
written in Laplace transform form

— \ipL — X2sp£ ~ KhX — Bhsx — (Ms2 + Bs + K)x (11.53)

We also have the relation

PL = Gx - YfL (11.54)

Furthermore there is a relation between pL and FL through the load.

The hydraulic impedance ZL} which may be reactive, may be defined as

Vl = ZlFl (11.55)

These three expressions may be solved directly to yield FL or pL as a

function of but a greater insight into the operation of the valve is
afforded by constructing an equivalent circuit for the valve to satisfy
these equations. Such a circuit is given in Fig. 11.29. Equation (11.54)

Displacement-
force feedback

Fig. 11.29. Linearized equivalent of hydraulic-flow-control valve.

is represented by a perfect flow source yielding a flow F for a given input x.

The fact that the actual valve is not a perfect flow source is taken into
account by the leakage Y, which causes the flow to the load to be dimin¬
ished when pressure exists. Reaction force is fed back from x and pl as
per Eq. (11.52) and results in a net force 25^ tending to move the valve
piston.
In most hydraulic actuators or motors the leakage and effect of com¬
pressibility are so small that the oil flow into them is a reliable index of
output displacement. Hence, we may solve Eqs. (11.53), (11.54), and
(11.55) for Fl, or we may obtain FL as a function of 9rf from the equivalent
Sec. 11.8] valve-controlled hydraulic systems 417
circuit. The result in either case is

1 l/r + zL
Ms2 + B's + K'
(11.56)
1 | G(\i + X2 s) Zl/T
“h Ms2 + B's + K' ZL + l/Y
SiG
(11.57)
(.Ms2 + B's + K')( 1 + YZl) + GZL(\, + X2s)

where B' = B + Bh, and K' = K + Kh. It was shown in Sec. 11.5 that
Bh is negative for valves for which the distance between the load outlets
and the sump orifices is less than the distance between load outlets and
pressure orifices (see Fig. 11.16). Therefore B' may be a negative quan¬
tity and the valve may be unstable and oscillate no matter what load is
connected to it. However, if the load outlets are midway between the
pressure and sump orifices, i.e., if Li = L2 in Fig. 11.16, Bh and X2 will be
zero. Under these conditions the valve may still be unstable if the load
contains an inertia component so that ZL — R + Ls. With this type of
load the denominator of Eq. (11.57) will be a cubic polynomial of s. In
order to determine whether or not the valve is stable the standard Routh
test1 may be applied to the denominator. For the case of ZL = R + Ls,
this test requires that for stability

B'M( 1 + YR)2 + B'K'Y2L2 + {B')2YL{ 1 + YR) + B'YGXiL2

+ il/ULXi > 0 (11.58)

The inequality shows that if Xi = 0 and B' > 0, the valve is stable. On
the other hand, if Xi <0, and B' = 0, the valve is always unstable.
Since the parameter Xi is negative in a valve in which the hydraulic reac¬
tion force is not compensated it is apparent that a certain amount of posi¬
tive damping (B' > 0) is required for stability.
We have seen here how a linear analysis can reveal valve instability.
By examining Eq. (11.58) the designer may adjust the various parameters
of the system to eliminate this possibility. The linear analysis has the
advantage not only of relative simplicity but also of revealing the approxi¬
mate transfer function of the unit. The accuracy of the analysis is
reduced, however, if large input signals destroy the assumed linear rela¬
tions about the quiescent operating point.
In this connection it should be noted that certain types of nonlinearities
will cause oscillations in spool-type pilot valves under conditions which
would allow the valve to be perfectly stable if the system were linear. To
take one example, there is evidence to indicate that static friction is
1 Chestnut and Mayer, “Servomechanisms and Regulating Systems Design,”
John Wiley & Sons, Inc., New York, 1951, Vol. I.
418 CONTROL SYSTEM COMPONENTS [Chap. 11

responsible for a mode of oscillation that is commonly observed in

hydraulic and pneumatic valves. Another serious cause of instability
is that in many cases the valve stem is not held securely by the valve¬
centering springs. Under these conditions the axial hydraulic reaction
forces cause the stem to bounce from one spring to the other at a high
frequency, giving rise to a chattering type of oscillation that will rapidly
destroy motors and gearing connected to the valve.1
11.9. Flapper Valves. One of the attempts to overcome the axial
reaction force developed in piston valves has resulted in a modified design
for the piston chamber, as discussed above. Further attempts to reduce
this force and to reduce the high manufacturing costs of control valves
have resulted in valves that operate on altogether different principles.
Figure 11.30 shows one such device.
The figure shows an elementary
flapper valve. The high-pressure
supply is connected to one side
of the unequal-area valve piston.
The pressure on the other side of
the piston is controlled by varying
the position of a flapper which par¬
tially closes a second orifice. The
force that is developed on each side
of the piston is the product of the
Fig. 11.30. Primitive flapper valve with

unequal-area actuator. pressure and the area across which

the pressure is exerted. It will be
remembered that, although the pressure on the left-hand side of the piston
is always higher than that on the right-hand side, the area of the left-hand
side of the piston is smaller. Thus by properly adjusting the flapper, the
forces may be balanced.
By application of the equations of flow through orifices the relationship
between p2 and flapper position may be found. From Eq. (11.7) the
flow through the upstream orifice having a diameter D i is

Fl = (t/4)CDf Vpi ~ Pi (11-59)

When the distance Y between the flapper and the orifice is very small the
flow at the flapper is controlled approximately by the cylindrical area
formed by extending the inner surface of the orifice pipe to the flapper.
This area is tD2Y, where D2 is the orifice diameter. Therefore, for very
small Y we have approximately

F2 = 7tC'D2Y Vp2 (11.60)

!J. L. Bower and F. B. Tuteur, Dynamic Operation of a Force Compensated

Hydraulic Throttling Valve, Trans. ASME, vol. 75, p. 1395, 1953.
Sec. 11.9] valve-controlled hydraulic systems 419

The discharge coefficient C' is probably different from the coefficient C

used in Eq. (11.59) since the shapes of the orifices are different.
The flapper ceases to control the orifice when the cylindrical area
ttF)2Y referred to above approaches in magnitude the orifice area (7r /4 )D,\
or when F = 0.257)2. For very much larger Y the flow is controlled
solely by the orifice and

F2 = (tt/4)CI>22 Vp~2 (11.61)

From Eqs. (11.59), (11.60), and (11.61) it is possible to obtain a relation

for the chamber pressure p2 as a function of flapper displacement F. For
this purpose we ignore the difference in discharge coefficients and let
C' = C. This is permissible if only an approximate relation is desired.

Fig. 11.31. Chamber pressure versus flapper travel, normalized. (From Nightingale)

Also, suppose that the piston is blocked so that the flow to the piston is
zero. Under these conditions F2 = F1 and therefore for F/7)2 < 0.25 we
have
(tt/4)C7)i2 \/pi — P2 = tCD2Y yjpi
P2 _ (Di/D2)4
or (11.62)
Pi 16(F/7)2)2 + (Di/7) 2)4

For F > 0.257)2 we equate Eqs. (11.59) and (11.61) giving

P2 = (7)i/7)2)4
(11.63)
Pi 1 H~ (7) 1/7)2)4

A plot of chamber pressure normalized with respect to supply pressure

versus flapper position normalized to orifice diameter is shown in Fig.
11.31.1 Despite the nonlinear equation relating these variables it is seen
that in the operating range the pressure is approximately a linear function
of flapper position.
By equating forces on the piston and by performing the usual partial

1 J. M. Nightingale, Hydraulic Servo-valve Design, Machine Design, vol. 27, no. 1,

p. 191, 1955.
420 CONTROL SYSTEM COMPONENTS [Chap. 11

differentiations as in Sec. 11.8, it is possible to obtain a small-signal linear

approximation for the flapper-control system. By equating flows in and
out of the control chamber and writing a force balance for the piston, a
transfer function x/y may then be obtained. The details of this computa¬
tion, however, are left to the reader (see Probs. 11.10 and 11.11).
The hydraulic reaction force on the flapper may be approximated, if
the stream is assumed to strike the flapper at right angles, by

Force = ^ (11.64)
g
where F is the volumetric flow, v is the velocity of discharge, p is the den¬
sity of the oil, and g the acceleration of gravity. The flow and discharge
velocity can be found by the relations given above. This force can be
considerably reduced by arranging two nozzles, opposing one another on
either side of the flapper, as in the Moog valve discussed below.
The major limitation on the flapper valve is that the controlled-pressure
chamber must be kept rather small if the time constant of the valve,
which consists of the resistance of the upstream orifice and the capacity
of the controlled-pressure chamber, is to be kept small. Enlarging the
flapper orifice will improve the response of the valve, but at the expense
of increased flapper reaction force and leakage flow. These restrictions
place a limitation on the load that can be controlled by a flapper valve,
and as a result the flapper valve is most often employed as the first stage
of a two-stage valve.
11.10. Nozzle Valves. The Askania nozzle valve is a valve that has
a very low hydraulic reaction force.1 A schematic of the valve is shown
High in Fig. 11.32. The hydraulic flow is formed
pressure-^ into a jet by the movable nozzle, and the flow
Pivot is directed into one of two ports on either side
,—Jet pipe of the receiving block and conducted to the
'A load. The only force that must be supplied
'Input

r
f L
Centering \
<- by the input is that required to overcome fric¬
spring tion at the pivot and the inertia of the jet pipe.
Receiving
block
The major disadvantage of this device is its
VMsActuator
relatively high leakage. The frequency re¬
sponse of this device with the standard 8-in.-
Fig. 11.32. Askania jet-pipe
long, 1%2"in.-diameter jet pipe peaks at about
valve.
30 cps and has a damping ratio (f) of 0.2.
11.11. Slide Valves. Owing to the high cost of manufacturing spool-
type valves to the very close tolerances required by high-performance
systems and owing also to the high leakage of the flapper-type valves,
1 S. Z. Dushkes and S. L. Cahn, Analysis of Some Hydraulic Components Used in
Regulators and Servomechanisms, Trans. ASME, vol. 74, p. 595, 1952.
SEC. 11.11] VALVE-CONTROLLED HYDRAULIC SYSTEMS 421

efforts have been made to develop a valve with neither of these disadvan¬
tages. The slide valve1 shown in Fig. 11.33 is one possible answer. The
advantage of the slide valve is that the slide and block can be clamped
together, so that the two critical holes may be drilled through the block
and slide at the same time. The spacing between the holes is not critical.
Following this operation, the plugs and bushings are inserted to complete
the critical portion of the device. Exact alignment is thus assured for
all the orifices. The vertical spacing between slide and block may be
established by a milling operation. The slide is suspended by bars and
actuated by a torque motor of some sort, or the slide may be allowed to
rotate with respect to the block. In either case, the variation of area of
the orifices is essentially linear with respect to motion of the slide. The
slide valve can be force-compensated in the same manner as spool-type

Top

Side

Fig. 11.33. A linear slide valve. (From Lee)

valves, or the valve with rotary motion for the slide plate may be designed
so that the Bernoulli force of one nozzle cancels that of another.2 In fact,
in the configuration shown in Fig. 11.33, a component of hydraulic force
acts to open the slide if the slide is displaced from center. Unfortunately,
however, this opening force increases as the valve slide is opened, while
the Bernoulli force decreases. Thus these two components cannot be
relied upon to cancel each other. Figure 11.34 gives a plot of flow versus
displacement of the slider of a rotary-plate valve. Note the linear char¬
acteristic, and note that the leakage flow is only about 1.5 per cent of full
output flow, whereas the usual flapper-valve leakage flow is about 5 to
10 per cent of full-load flow. A practical disadvantage of the slide valve
is the scoring of the bushings and block by chips and dirt that have a
tendency to accumulate in the slide cavities.

1S. Y. Lee, New Valve Configurations for High-performance Hydraulic and

Pneumatic Systems, Trans. ASME, vol. 76, p. 905, 1954.
2 Ibid.
422 CONTROL SYSTEM COMPONENTS [Chap. 11

An actuator that has been constructed1 embodying the slide-valve

principle is shown in Fig. 11.35. The ram could be used to position an
output element mechanically, or it could be the spool of a high-power
pilot valve designed for the same service as the two-stage piston valves
discussed below and the flapper valves described above.

Fig. 11.34. Flow versus displacement for a rotary slide valve. (From Lee)

Fig. 11.35. Actuator employing a slide valve. (From Lee)

The frequency response of this actuator is given in Fig. 11.36. It will

be seen that, in addition to its other advantages, the slide-valve principle
allows a rigid, compact design with very high frequency response.
Rotary slide valves of this type are presently available commercially.

1 S. Y. Lee and J. L. Shearer, A Miniature Electrohydraulic Actuator, Trans.

ASME, vol. 77, p'. 1077, 1955.
Sec. 11.12] valve-controlled hydraulic systems 423

2.0

1.0
0.8
■S 0.6
2 0.5

0.2

0.1
10 20 30 40 60 80100 200 300 400 600 1,000
Cycles per second
Fig. 11.36. Frequency response of actuator and slide valve. (Lee and Shearer)

11.12. Two-stage Valves. There are rather definite practical limita¬

tions on the power-handling capacity and the speed of response at these
high powers for single-stage valves. Various combinations of the valves
already discussed have been used in two-stage valves to overcome these
limitations.
Let us first consider a two-stage spool-type valve, shown in Fig. 11.37.
This valve has an operating characteristic somewhat different from that of
a single-stage valve. A given displacement of the input results in a given
flow into the actuator cylinder. Thus the actuator takes up some con¬
stant velocity. It may be seen, then,
that the main valve opens at a constant \InPut
rate and that the flow to the load con¬
stantly increases. The transfer function
from input position to main fluid flow
thus contains an integration. While this
integration is sometimes desirable, it is
usually removed by placing a feedback
loop around the pilot valve and actuator
cylinder.
Figure 11.38 shows three ways of add¬
ing feedback in a two-stage spool valve.
In Fig. 11.38a the feedback is through
a linkage. When the input moves down,
Fig. 11.37. Two-stage hydraulic
for example, the pilot-valve spool also
valve.
moves down, with the actuator cylinder
as a fulcrum, thus porting high-pressure fluid to the actuator cylinder.
As the cylinder moves down, the feedback link moves about the input as
a fulcrum and moves the pilot-valve piston up, thus cutting off the flow of
oil. In effect, then, the actuator follows the input.
424 CONTROL SYSTEM COMPONENTS [Chap. 11

In Fig. 11.386 the input varies the position of a movable sleeve in the
pilot valve. If the input is moved down, for instance, the high-pressure
oil is ported to the bottom of the actuator cylinder. The actuator moves
up, carrying the pilot-valve piston down by means of the feedback link,
thus stopping the flow of oil.
Figure 11.38c shows a Vickers two-land sleeve valve that operates in
a manner somewhat similar to the action of Fig. 11.386; however, the
principle of differential area is employed in the system. In Fig. 11.38c
the areas are in the ratio of 2:1, which is the most common arrangement.
If the input is moved down, the valve spool ports high pressure to the bot¬
tom of the actuator cylinder, and the top of the actuator cylinder is
opened to sump. The actuator will move up, and the sleeve will be
Input

io) [I) (c)

Fig. 11.38. Feedback links for pilot valve.

pulled down by the feedback link, thus closing off the fluid flow. The
actuator follows the input with a reversal of direction. Now, if the input
is moved up, a rather different effect occurs. The high-pressure fluid is
ported to both the top and bottom of the actuator cylinder. The net
force is down.
It is also possible to construct very compact two-stage valves without
constructing the parts separately and connecting them by linkages.
Figure 11.39 shows a two-stage slide valve used in a hydraulic autopilot
control manufactured by Siemens (Germany) during World War II.
If the force developed in the left-hand chamber, p4A4, is equal to the force
developed in the right-hand chamber, p3A3, the spool is at rest. If the
input is moved to the right, for instance, the orifice is reduced in size
and p4 increases, thus increasing the force in that chamber and moving the
spool to the right. The new position of the spool will be such that the
chamber forces are once more balanced. The motion of the spool
ports oil to the main actuator. While this valve has been described1
1 R. Hadekel, Hydraulic and Pneumatic Servos, Automation, March, 1955.
Sec. 11.12] VALVE-CONTROLLED HYDRAULIC SYSTEMS 425

High Low High

pressure pressure

Fig. 11.39. A Siemens two-stage hydraulic pilot valve.

as analogous to a Wheatstone bridge with the resistance arms Ri,

R2, R3, and R4, it would appear to be more direct to consider it on
the differential-area principle, as above, since in equilibrium
is not equal to p4. The Pegasus valve of this type in which the
input is supplied by a solenisoid is presently available commercially.
A more compact two-stage valve
High pressure
is one that employs a flapper valve P\
as its first stage. An elementary
form of such a valve is shown in Fig.
11.40. The pressure p2, controlled
by flapper position, actuates the
spool against the spool spring. A
disadvantage of this valve is that,
for a given flapper position, a change
in the high pressure will cause a
change in the position of the spool,
since the force developed by p2 is
opposed only by the spring.
The Moog1 valve, shown in Fig.
11.41, is one of the earlier of the
low-reaction-force two-stage hy¬ Actuator
draulic valves and is designed to Fig. 11.40. Primitive form of a two-stage

operate at 1,000 to 3,000 psi. The flapper-spool valve. More practical two-
stage valves of this type have been
arrangement of balanced nozzles at
developed.
the flapper reduces the force re¬
quired by the flapper, and the us of two chambers and two springs
eliminates the direct reliance on the value of the supply pressure
1 R. L. Scrafford, “ Hydraulic Servos Incorporating a High-speed Hydraulic-
amplifier Actuated Valve,” Trans. AIEE, part II, vol. 72, p. 175, 1953.
426 CONTROL SYSTEM COMPONENTS [Chap. 11

characteristic of the primitive type. The force on the flapper is

developed by the differential current flow in the two halves of the split-
coil winding. Eight milliamperes differential current is sufficient to
cause full rated flow of about 8 gpm in the valve. Valves capable of

(a) Schematic

(b)
Fig. 11.41. The Moog valve.

larger outputs are also available. The leakage flow through the nozzles
at zero signal is typically about 2 per cent of full rated flow for the 8-gpm
model. This leakage flow cannot be reduced proportionally for smaller-
flow models. From test characteristics furnished by Moog, it would
appear that f, the damping ratio, is about 0.7, and the critical or
natural frequency of resonance is about 40 cps for the Model 500. Thus
Sec. 11.12] valve-controlled hydraulic systems 427
empirically

F 1 _ 1
ii - i2 _ (M/K)s2 + (B/K)s + 1 “ (s2/6.3 X 104) + (s/28.3) + 1
(11.65)

The amplitude characteristic versus frequency is shown in Fig. 11.42.

Test data for a model designed for a maximum flow of 0.5 gpm show a
f = 0.5 and a natural frequency of about 130 cps. The spring-mass

Fig. 11.42. a diagram for Moog Mode] 500 Fig. 11.43. Equivalent circuit of a
valve. (Courtesy Moog Valve Co.) portion of the Moog valve, where x is
flapper position.

resonant frequency of the second-stage spool is about 1,000 cps for the
typical Moog valve; thus it is well above the frequencies of interest. The
valve may be thought of as the equivalent of a bridge circuit, as shown in
Fig. 11.43. This circuit could not of itself cause the underdamped
response shown in Fig. 11.42. The flapper is flexible, and the interaction
between the nozzle flow and flapper position is perhaps responsible.
The Cadillac Model FC2 flow-control valve, as shown in Fig. 11.44, is
a two-stage valve similar to the Moog valve. Unlike the Moog valve,
however, the piston of the Cadillac valve is not directly restrained by
428 CONTROL SYSTEM COMPONENTS [Chap. 11

centering springs. The spool is operated on the differential-area principle

by flapper-controlled pressure, and the piston position is fed back to the
flapper through a feedback spring. This accomplishes the centering of
the piston as effectively as spring centering. The time constant of the
two actuating springs and the mass of the flapper are small enough to be

Transfer from force to flapper

position includes fiopper moss,

Fig. 11.45. Equivalent block diagram of the Cadillac Model FC2 valve.

negligible in this device. The time constant of the Cadillac FC2 valve is
determined by the gain of the feedback loop. Figure 11.45 gives a block-
diagram representation of the FC2 valve. The block diagram shown is
only one of several possible block diagrams that could be drawn. The
choice of force feedback is arbitrary. The flapper spring-mass system
has a natural or critical frequency that is
quite high and may be neglected. Like¬
wise the transfer function from piston
position to feedback spring force may be
considered a constant. Thus the trans¬
fer function of the closed loop from input
to output is

y _ m __fcife2fc8__
x 1 + fiB as2 + bs + c + kik2kzki
(11.66)
where n is the forward transfer function
and B is the transfer function through
the feedback path of the loop. The
Fig. 11.46. Frequency response of
Cadillac FC2 valve. (Courtesy
transfer function y/x thus has a simple
Cadillac Valve Co.) quadratic denominator, assuming that
the time constants of the several spring-
mass systems are very small. In the FC2 valve the loop-gain constant
/ci/c2/c3/c4 is about 314, from data furnished by the manufacturer, and
the response of the system is as shown in Fig. 11.46.
The frequency responses of the Moog and Cadillac valves are similar
with the exception that the FC2 valve amplitude response falls off more
gradually than the typical Moog characteristic. However, the damping
Sec. 11.13] VALVE-CONTROLLED HYDRAULIC SYSTEMS 429
factor of both valves may be adjusted within a limited range by changing
internal dimensions.
In all the hydraulic valves so far discussed, including the flapper valves,
the transfer is from the input variable to flow. Constant flow will cause
a constant velocity in the load element. If the over-all control system
controls the position of this element, the loop contains one integration.
Occasionally, an extremely high performance control system is required,
and the designer must provide two integrations within the loop to meet
dynamic accuracy specifications. Both the Moog and Cadillac valves
can be adapted to provide two integrations. In the Moog valve the cen¬
tering spring may be removed, while in the Cadillac valve the feedback
spring and lever may be removed. The force set up by the differential
pressure in the chamber will then move the piston at a constant velocity,
and the oil flow will constantly increase. Naturally, such a device must
be part of a feedback loop to operate properly, and the loop must be
properly equalized or compensated if it is to be stable.
In certain applications both the Moog Model 500 valve and the Cadillac
FC2 valve encounter problems with magnetic dirt building up around the
poles of the torque motor and hindering proper operation. Both con¬
cerns have developed dry torque-motor designs to overcome this difficulty.
11.13. Pressure-regulating Devices. Several different methods of
obtaining constant-pressure hydraulic supplies are used, depending on

Fig. 11.47. Elementary form of pressure regulator. Constant-flow pump with

bypass orifice for constant-pressure service.

the type of pump and the accuracy with which the pressure must be
controlled.
One type of pressure control is shown in simplified form in Fig. 11.47.
It consists of a spring-loaded valve connected from the high-pressure line
to sump. This type of pressure control is used with fixed-stroke pumps.
The flow from the pump is essentially constant at full flow, with the valve
porting to sump any oil in excess of that required by the load. There is
a considerable waste of flow and power in the regulating valve unless the
system is always operating at close to full load. The wasted power
results in a temperature rise in the hydraulic fluid and may require
430 CONTROL SYSTEM COMPONENTS [CHAP. 11

special provisions for cooling. Commercial devices operating on this

principle may be more elaborate.
A second method of pressure control consists of a pressure-sensitive
actuator and a variable-stroke pump, as shown in Fig. 11.48. The main
advantage of this method is the saving in power and flow. The pump
supplies only that flow required by the load. The major disadvantage of
this device is the extra expense of the variable-stroke pump.

io) Primitive

Ball Piston return Valving

bearing springs plate
Non rotating
wobble plate

Low
pressure

High
pressure
Pressure
adjust ■

Pressure control spring

lb) Practical
Fig. 11.48. Variable-stroke pump with pressure-controlled stroke for constant-pressure
service.

In constant-pressure supply systems a storage reservoir, or accumula¬

tor, is usually included to smooth variations in pressure due to pump rip¬
ple and to supply the system during periods of peak demand. In the case
of intermittent operation of actuators, the hydraulic pump may be com¬
pletely inadequate to supply peak demand, but it serves to charge the
accumulator during periods of inaction.
The accumulator may be of the bellows type, in which the high-pressure
fluid expands a bellows, or it may be in the form of a cylinder in which the
pressure actuates a piston that compresses a spring. Repeated cycling
of the bellows type causes fatigue, and hence this type is unsatisfactory
in general. The most satisfactory type of accumulator for high-perform¬
ance systems 'is the hydropneumatic type in which the spring action is
obtained from compression of a gas. The inflexible-separator or piston
Sec. 11.14] valve-controlled hydraulic systems 431

type is generally not so desirable as the newer flexible-separator type.

Owing to the inertia of the piston the pressure peaks may actually be
increased in the piston type. The flexible-separator hydropneumatic
accumulator is the better design. It yields more power output and is
more reliable for a minimum weight and size.1
11.14. Choice of Operating Pressure for Hydraulic Systems. One of
the first decisions to be made by the designer of a flow-control hydraulic
servomechanism is the choice of the system operating pressure. Early
hydraulic mechanisms were usually operated in the pressure range of
several hundred pounds per square inch. The operating pressure was
chosen empirically and gradually increased as experience was gained in
the field. Modern aircraft hydraulic systems have been designed for
pressures as high as 4,000 to 5,000 psi.
Several factors enter into the choice of system pressure. As the oper¬
ating pressure for a given load is increased, the following effects are noted:
1. The required flow is decreased, thus allowing the use of smaller dis¬
placement pumps, valves, and actuators, as well as smaller lines.
2. The weight of the fluid required decreases continuously, and the
weight of the components decreases up to a value of about 4,000 psi,* at
which point the strength of the material dictates an increase in size and
weight.
3. The operating temperature of the system steadily increases as a
result of the increased leakage at valves and seals.
4. The danger of fire and explosion at extremely high pressures dictates
the use of nonflammable artificial hydraulic fluid, which is more expensive
by a factor of 10 than petroleum-base products.
5. At the very high pressures manufacturing difficulties prevent taking
full advantage of the reduction in size of valves and actuators. All of
these factors have influenced the choice of 800 to 1,500 psi as standard in
most modern high-performance control systems. In the United States,
3,000 psi is standard in aircraft applications, because weight reduction is
of prime importance and the expense factor is secondary, whereas certain
British aircraft such as the Bristol Britannia operate at 4,000 psi. Pro¬
posals for raising the standard pressure in U.S. aircraft systems from
3,000 to 4,000 psi will result in an improvement of only 2x/i per cent in the
figure of merit and are thus not worth while, according to Cooke.2

PROBLEMS
11.1. Show from Eq. (11.5) the result of making a in Fig. 11.1 equal zero.
11.2. In Fig. 11.49 is shown a typical spring-centered spool valve. Find the trans¬
fer function x0/$l for operation about the point xv = 0, pl = 0 for the hydraulic-

1 E. M. Greer, Hydraulic Accumulators, Machine Design, vol. 26, no. 1, p. 132, 1954.
* C. Cooke, Optimum Pressure for a Hydraulic System, Product Eng., vol. 27, no.
5, p. 162, 1956.
2 Ibid.
432 CONTROL SYSTEM COMPONENTS [Chap. 11

actuator arrangement shown. The valve is completely symmetrical and slightly

underlapped; hence when the valve piston is at the exact center (xv = 0), each orifice
is open. The amount of underlap is such that a motion of the valve piston ±0.001

&
~\xv y
I
Y
Sump- D

High
pressure b
—b

Sump■
-p
Spring

7777,
Fig. 11.49

in. from the neutral position will just close one set of orifices. Assume that the ori¬
fices do not leak at all when they are closed and produce no steady-state hydraulic
reaction force. Also assume that transient hydraulic forces are negligible.
Each orifice is characterized by the two equations

Fi = Cxi Pi and = kxipi

where C, k = constants
g = acceleration of gravity
p = density of oil = 0.03 lb/in.3
Xi = amount of opening of fth orifice, in.
Fi = flow through ith orifice, in.3/sec
Pi = pressure drop across fth orifice, psi
$i = force of hydraulic reaction produced at ith orifice, lb
The following data are given:
Mass of valve piston =0.1 lb
Valve spring constant = 200 lb/in.
Valve coefficient of viscous friction = 0.04 lb-sec/in.
Mass of actuator piston = 1.5 lb
Area of actuator piston = 1 in.2
Friction of actuator piston = 1 lb-sec /in.
When xv = 0.005 in., when supply pressure = 2,000 psi, and when there is no load on
actuator, the piston moves at a rate of 5 in./sec and the total hydraulic reaction force
is 0.9 lb.
11.3. Assume that at each of the valving orifices shown in Fig. 11.49 the relation
between pressure flow and displacement is
Problems VALVE-CONTROLLED HYDRAULIC SYSTEMS 433

where Ft flow through fth orifice, in.3/sec

C a = constants
P density of oil = 0.03 lb/in.3
9 acceleration of gravity
Pi pressure drop across ith orifices, psi
Xi spool displacement at each orifice, in.
Note that, for orifices a and c, Xi = xv, and for orifices b and d, Xi = — xv. When the
valve is centered, xv = Xi = 0. Each orifice produces a hydraulic reaction force given
by the relation

Ti = KFi Vpi
All the forces 5^ are in such a direction as to tend to close the valve.
The following data are given:
Mass of valve piston = 0.1 lb
Valve spring constant = 200 lb/in.
Valve coefficient of viscous friction = 0.04 lb-sec/in.
Mass of actuator piston = 1.5 lb
Area of actuator piston = 1 in.2
Friction of actuator piston = 1 lb-sec/in.
When xv = 0.005 in., when supply pressure = 2,000 psi, and when there is no load on
actuator, the piston moves at a rate of 5 in./sec and the total hydraulic reaction force
is 0.9 lb. When xv = 0, when supply pressure = 2,000 psi, and when a force of 1,000
lb is applied to the piston, the piston moves at a rate of 0.01 in./sec.
Find the transfer function x0/%i for operation about the point xv = 0, pl = 0.
11.4. Find the axial force developed at one port of a spool valve with a piston
diameter of 34 in- and a rectangular port of 0.1-radian width. Assume the stroke is
346 in- and the radial clearance is 0.002 in. Assume an orifice coefficient of 0.6 and an
operating pressure of 3,000 psi.
11.5. The flow relation of a hydraulic valve is given by

F = 25x s/pl
where F = flow, in.3/sec
x = valve-spool displacement, in.
pv = total pressure drop across valve, psi
The valve is connected to a constant-pressure supply of 1,500 psi. The sump pressure
is zero. A hydraulic motor with the following characteristics is connected to the
servo ports of the valve:

Moment of inertia = 0.003 lb-in.-sec2

Viscous friction = 0.2 lb-in.-sec
Hydraulic displacement = 0.38 in.3/rev

a. Find the response of motor speed versus time when x is given a step displacement
from zero to 0.01 in.
b. Find the response of motor speed versus time when x varies sinusoidally at a fre¬
quency of 10 cps with an amplitude of 0.01 in.
11.6. Calculate the orifice diameters for a flapper valve to operate at 4,000 psi.
Allow a maximum flow of 0.05 gpm through the nozzle. Calculate the reaction force
on the flapper when it is in the center of its operating range. The density of the oil is
0.03 lb/in.3.
11.7. Assume that the nozzle in Fig. 11.32 has been imperfectly made and that the
jet emerges at an angle of 10° from the center line of the jet pipe at 1,500-psi supply
pressure and an orifice diameter of 0.1 in. What is the effect?
434 CONTROL SYSTEM COMPONENTS [Chap. 11

11.8. A flapper valve operating at 1,500-psi supply pressure has an upstream orifice
with a diameter of 0.005 in. and a controlled-orifice diameter of 0.012 in. (see Fig.
11.50). The piston has a large area of 0.05 in.2 and a small area of 0.025 in.2 The
piston mass is 0.02 lb and the viscous friction between the piston and cylinder is such
that a force of 0.04 lb is required to move the piston at a rate of 1 in./sec. Obtain a
small-signal transfer function between flapper position and piston position for the oper¬
ating point defined by the fact that the piston is stationary in the middle of its allowed
travel, and there is no external force on it. Assume that the coefficient of discharge for

Upstream Q
orifice v Flapper-
Supply
pressure V7777A V/77777^7777^
^ ^ ^ V, Controlled

^ V/////////////M V,

I
V////////777/////////&.
Fig. 11.50

both orifices is 0.65 and that the compressibility of the fluid in the chamber is negli¬
gible. The density of the fluid is 0.03 lb/in.3
11.9. The flapper of the valve described in Prob. 11.8 is spring-restrained such that
the hydraulic reaction force at the quiescent operating point is balanced. The spring
constant of the spring used for this purpose is 100 lb/in. The flapper is controlled by
an electric torque motor which supplies a force of 0.05 lb/ma. The mass of the flapper
is negligible. Find the small-signal transfer function from input current to output
position of the piston for this system. Consider operation around the operating point
defined in Prob. 11.8.
11.10. A valve of the Moog type (see Fig. 11.41) has a second-stage valve piston
with a diameter of 0.1875 in. and a weight of 0.02 lb. The spring constant of each one
of the springs used to center the piston is 400 lb/in. The control orifices for the inlet
and sump are rectangular in shape and 0.03 in. wide. There are four of these rec¬
tangular holes spaced equally around the circumference of the sleeve at each land of
the valve piston. The first stage is a flapper valve having fixed orifices of 0.005-in.
diameter and variable orifices of 0.012-in. diameter. The flapper is controlled by a
torque motor supplying a force at the flapper of 10 lb/amp. The flapper is spring-
restrained, but has negligible mass or damping. The chamber volume is small enough
so that compressibility of the oil can be neglected.
The flow gain of this valve is measured by connecting the two outlet ports together
with a pipe large enough so as to produce negligible pressure drop. A flow meter is
then placed into the return line and the valve is connected to a 1,500-psi supply.
After subtracting the first-stage leakage flow, the flow gain is found to be 2-in.3/sec/ma
current difference into the torque motor.
Using small-signal methods about the operating point defined by load pressure =
load flow = 0, supply pressure = 1,500 psi, find the transfer function of the valve.
The density of the fluid is 0.03 lb/in.3
11.11. For the slide valve shown in Fig. 11.33, show that the reaction force tends to
open the valve when the slider is displaced from dead center.
 CHAPTER 12

PNEUMATIC SYSTEMS

12.1. Introduction. Control systems that transmit power by means

of fluid flow are subdivided into two classifications. Hydraulic systems
constitute the first class and have been considered in the preceding chap¬
ters. The second class comprises pneumatic systems. The division is
made on the basis of the importance of compressibility of the flow medium.
Roughly, the division can be made on the state of the fluid. The medium
for hydraulic systems is a liquid, while for pneumatic systems the fluid is
a gas.
In general, a pneumatic control is “softer” than a hydraulic system.
The torque of a pneumatic motor builds up slowly with respect to time,
and the system will yield under shock loads. While this contributes to
the long life of the components, compressibility complicates the stability
problem in closed-loop systems. In pneumatic systems there is no possi¬
bility of harmful shock waves when the flow is suddenly stopped; this
problem does exist in hydraulic systems and is sometimes called the
“water-hammer” effect.
Pneumatic systems have certain other advantages over hydraulic sys¬
tems. First, there is no fire hazard. Although artificial hydraulic fluids
that are nonflammable have been developed, petroleum-base fluids are
still preferred for hydraulic systems because of their anticorrosion and
lubricating qualities as well as their lower cost. Even minute hydraulic
leaks in high-pressure systems can fill the surrounding atmosphere with
atomized hydraulic vapor, thus creating an extreme fire and explosion
hazard.
Second, since, in a pneumatic system, the air can be vented to the
atmosphere at the actuator, it is often possible to design the system with
only one line, while in a hydraulic system this is not possible. This cuts
down on the cost of fittings and lines, etc. It also results in a weight
reduction, which is the third advantage.
The weight of pneumatic lines and fittings is less than the weight of
equivalent hydraulic fittings. There is also the further reduction in
weight due to the elimination of the return line, and, finally, the weight of
the hydraulic fluid itself is eliminated in pneumatic systems. In an
435
436 CONTROL SYSTEM COMPONENTS [Chap. 12

interesting comparison between pneumatic and hydraulic systems for the

Convair Model 240 aircraft, Gerwing and Famme1 found that the pneu¬
matic system weighed exactly one-half as much as the hydraulic system.
A more detailed weight comparison of electric, hydraulic, and pneumatic
components is given in the following chapter.
A fourth advantage of pneumatic systems is the ease of maintenance.
When components must be replaced in a hydraulic system, the fluid must
be drained before replacement can proceed. Furthermore, fluid “weep-
age,” or the oil caught behind seals and connections during assembly that
oozes out over a period of time, must be cleaned off periodically in
hydraulic systems.
A fifth advantage that can be important if the control system is sub¬
jected to extremes in temperature is that the viscosity of the fluid in
pneumatic systems is usually negligible. Furthermore the variation in
viscosity with temperature is small; thus, even when viscosity is con¬
sidered, the effect of temperature will be small. This is not so in hydraulic
systems. There viscosity is a first-order effect, and the variation of
viscosity with temperature is a major problem.
There are several disadvantages of pneumatic systems as compared
with hydraulic systems. First, the lubrication of components must be
carefully considered, since this is not automatically taken care of as in
hydraulic systems. Second, more work is required to pump a compressi¬
ble fluid up to a given pressure than is required by a noncompressible
fluid. This work must be charged to the efficiency of the system if the
work is not recovered by the actuator. The third disadvantage of a
pneumatic system is the possibility of explosion if the pneumatic storage
tank, or accumulator, is punctured. It will be seen in Chap. 13 that, in
order for the maximum weight savings to be realized with a pneumatic
system, a large accumulator must be provided. The large amount of
energy that is stored in the compressed gas will be explosively released if
the tank is ruptured. A break in a hydraulic system is messy and a
potential fire hazard, but the same danger of explosive expansion does not
exist. Modern pneumatic-component design attempts to render the
accumulator penetration-proof.
Finally, the compressibility of the fluid adds equivalent springs to the
system in various places, and thus lags and time constants are developed.
This is a fundamental limitation on the high-speed response of a pneu¬
matic system. In addition to the lags added to the system by compressi¬
bility, the decreased density of the fluid results in a slower signal-trans¬
mission rate. A line length that would be considered negligible in a
hydraulic system could have a major effect on the operation of a pneu-
1 H. F. Gerwing and J. H. Famme, Pneumatic versus Hydraulic Systems for Air¬
craft, Product Eng., vol. 21, no. 11, p. 21, 1950.
Sec. 12.2] PNEUMATIC SYSTEMS 437
matic system. Basically a pneumatic system is slower than an equiva¬
lent hydraulic system. Shearer and Lee1 compare a simple hydraulic
positioning system with an equivalent pneumatic system operating at the
same pressure and find the hydraulic system about forty-five times faster.
This disparity could be reduced, however, by operating the pneumatic
system at a considerably higher pressure while still not exceeding the
limits of safety.
Probably the oldest application of pneumatic systems is in low-pressure
process control in the chemical industry. For many years process-con¬
trol engineers were alone in their attempts to analyze closed-loop pneu¬
matic systems, and understandably they have built up a jargon of their
own. Unfortunately their nomenclature differs considerably from that
used in the newer and more widely known control fields. There is evi¬
dence that this lack of communication has prevented the use by the rest
of the field of techniques developed many years ago by process-control
engineers and has hindered progress in the process-control industry more
recently by hampering the use of work done outside that field.
12.2. Pneumatic Flow. Usually in the analysis of a pneumatic system
the viscosity of the medium is ignored, since the viscosity of air is about
10“3 that of hydraulic fluids. While the analysis of the flow relations
proceeds in exactly the same way as the previous analysis of incompressi¬
ble fluids, different factors are important. The general energy equation
from which the Bernoulli equation was developed is
p lVi V 2V2 . work V22 — U12 z2 — z\
9 + U2 U\ -\- (12.1)
778 778 + TuT — —

20(778) + 778
where q is the heat transferred to the fluid, p is the pressure in pounds per
square foot, work is the mechanical work done by the fluid, u is the inter¬
nal energy, V is the velocity in feet per second, z is the elevation, and v is
the specific volume, or volume per unit weight. Usually no heat is trans¬
ferred to the fluid, and thus q is zero. A process for which q is zero is
called an adiabatic process. Furthermore the difference in head or eleva¬
tion is negligible, and in a flow process no work is done by the fluid. Then
P1V1 P2V2 _ , V22 - F12
(12.2)
778 778 "2 "1 + 2g(778)
This equation applies whether the fluid is liquid, gas, or vapor. We shall
assume that our fluid follows the perfect-gas law pv = RT, where R is the
gas constant and T is the absolute temperature. For air, R is 53.3 ft/°R.
It can be shown2 that, for any perfect-gas process, the change in internal
1 S. L. Shearer and S. Y. Lee, Selecting Power Control Valves, Control Eng., vol. 3,
no. 4, pp. 73-76, 1956.
2 Binder, “Fluid Mechanics,” Prentice-Hall, Inc., Englewood Cliffs, N.J., 1949,
p. 197.
438 CONTROL SYSTEM COMPONENTS [Chap. 12

energy is
u2 — U\ = cv(T2 — T i) (12.3)
where cv is the specific heat of the fluid at constant volume. It can be
further shown1 that
1 R
cv = (12.4)
778 k - 1
For a general polytropic process
pvk = constant (12.5)

and for air, k = 1.4, if the process is adiabatic, and k = 1.0, if the process
is isothermal. Thus by combining Eqs. (12.3) to (12.5), Eq. (12.2) may
be written as
(*-!)/*'
V22 - Vi2 k Vi i - (2* (12.0)
2 k — 1 mi

where mi is the initial mass density of the fluid. Thus by definition,

wi 1
mi = (12.7)
9 Vi g

where w\ is the initial weight density and V\ the initial specific volume.
12.3. Pneumatic Flow through an Orifice. Let us consider the case
of adiabatic flow through a nozzle or orifice, as shown in Fig. 12.1. We
shall assume that V i, the velocity on the high-
pressure side, is negligible with respect to V2, the
velocity at the outlet. Thus Eq. (12.6) becomes

(k—l)/k
2k pi P2
P2 1 - (12.8)
k — 1 mi J> L

Let W be the weight of gas flowing per second

and A 2 be the area at the throat. Then

Fig. 12.1. Flow through a2v<

w = (12.9)
orifice. v2

where v2 is the specific volume of the gas at the throat of the nozzle.
Substituting for V2 in Eq. (12.9) and employing Eqs. (12.5) and (12.7)
results in
(k+l)/k
Vi
W = A2 (12.10)
Vi

Figure 12.2 shows a plot of Eq. (12.10), the weight flow through an orifice
as a function of pi/pi. The maximum flow occurs at a ratio p^/px — 0.53.
1 Ibid.
Sec. 12.4] PNEUMATIC SYSTEMS 439

This may be found from Eq. (12.10) by differentiating. For pressure

drops greater than the critical drop, Eq. (12.10) shows the flow decreasing.
Actually this does not occur; rather the flow remains constant, no matter
how low the downstream pressure becomes. This is an effect similar to
the venturi tube effect discussed in Chap. 11. The velocity of flow
becomes supersonic above the critical pressure ratio, and no downstream
condition can influence upstream conditions.
If in Eq. (12.10) we let p2/pi = 0.53 and k = 1.4, the relation for flow
through a converging orifice at greater
than critical pressures may be found to
be
QPi
W = 0.7A2^lg- = 0.7A2
V\ RT/pi
(12.11)
For air, R = 53.3; thus

0.53 A 2P1 Fig. 12.2. Weight flow through

W = (12.12)
Vt 1 orifice.

where T\ is the absolute temperature in degrees Rankine. Equation

(12.12) is sometimes called Fliegner’s equation. Thus for pneumatic
flow, but not for hydraulic flow, there is a linear relation between upstream
pressure and flow in the region above critical pressure drop, and in this
region flow is independent of downstream pressure. Since this relation¬
ship holds so long as the pressure drop across the orifice is more than 1.89
to 1, pneumatic systems in general are designed to operate in this region
to take advantage of this fact. Just as in hydraulic flow, the actual
measured area of the orifice is not the area of the jet flow, nor does this
relation include the effects of turbulence, viscosity, or converging flow.
These effects reduce the mass rate of flow through the orifice and are
usually accounted for by an orifice coefficient of between 0.6 and unity.
The Compressed Air Institute recommends 0.65 for sharp-edged orifices
and 0.77 for well-rounded entrances. Stenning1 reports that values
between 0.8 and unity apply for servo-valve-type orifices.
12.4. Compressible Flow in Pipes. The general equation for com¬
pressible flow in pipes may be derived in the same manner as the equation
for hydraulic flow, which was obtained in Chap. 10. The general differ¬
ential equation for compressible flow is2

/F2 7tD dl =
7tD- dl y dV
(12.13)
8 gv 4 gv dl

1 A. H. Stenning, An Experimental Study of Two-dimensional Gas Flow through

Valve-type Orifices, ASME paper 54-A-45.
2 See, for instance, Binder, op. cit., p. 229.
440 CONTROL SYSTEM COMPONENTS [Chap. 12

where/is the pipe friction factor, which can be expressed as a function of

the Reynolds number and can be determined from the Staunton diagram
in Fig. 10.5. Rearranging (12.13), we obtain

VdV , fV'2dl
v dp + +J 2 g D 0 (12.14)
g
2 gv dV fdl
and dp -\- 2 -y- + 0 (12.15)
Vs D
Usually certain simplifying assumptions concerning the nature of the
flow may be made before integration is attempted. One type of flow is
isothermal, or constant-temperature, flow; the second type of flow is
adiabatic, or no-heat-loss, flow. Actual pneumatic processes lie between
these two extremes. Experimental work on many types of pneumatic
components such as flapper valves, capillary tubes, and tapered-pin
orifices have shown excellent agreement with theoretical isothermal flow.1
For isothermal flow pv = RT, and Eq. (12.15) can be integrated as

P i2 — P 22 =
gv i \ Vi Dt
(12.16)

Figure 12.3 shows normalized pres¬

sure versus the pressure ratio fl/D
for various initial Mach numbers,
where the Mach number is defined
as
Vi
Nni = (12.17)
\/gkRT i
and in this case we take k = 1.4 for
air. For comparison, the plot for
Fig. 12.3. Pressure versus pressure ratio
incompressible fluids under the
for various Mach numbers.
same conditions is also given in Fig.
12.3. It may be seen that, for low Mach numbers and short lengths of
pipe, incompressible flow may be assumed with little error.
Quite often the logarithmic term in Eq. (12.16) can be neglected with
negligible error. Under this assumption (12.16) becomes

Vi2Pi fl (12.18)
pl2 - P22 =
gv i D
For Reynolds numbers less than 2,000 the flow is usually laminar, and /
may be represented by the empirical relation found from the Staunton
1 C. R. Webb, Nonlinearities in a Pneumatic Controller, Trans. Soc. Instrument
Tech., vol. 7, no. 1‘, p. 9, 1955.
Sec. 12.6] PNEUMATIC SYSTEMS 441

diagram of
64
(12.19)
Nr

Substituting the relation for volumetric flow for V\ and rearranging, there
results
tt(pl2 - Pi2)DA
(12.20)
128/z/2pi

Equation (12.20) has been arranged for comparison with Eq. (10.22), the
Hagen-Poiseuille law for incompressible laminar flow in a pipe. While
there are some similarities, the relations are not the same. Figure 12.3
shows the error involved in using the incompressible relations instead of
the compressible relations for various Mach numbers.
12.5. Compressibility as an Equivalent Spring. Even if compressibil¬
ity is ignored in the flow through pipes, it should be considered for its
effect on the transfer function of the various system components. Just
as a viscous fluid adds viscous damping to various components, a com¬
pressible fluid acts as an equivalent spring. For instance, the resonant
frequency of a typical spool-type pilot valve was increased from 700
radians/sec to 950 radians/sec when 1,000-psi air was introduced into the
system; this result was due to the equivalent spring action of the air sup¬
ply. Usually the viscous-damping effect of compressed air is negligible
compared with friction and other sources of damping.
The spring rates of trapped compressed air can be important in
actuator cylinders. Figure 12.4* gives the equivalent spring constant of
compressed air initially at 100 psi trapped behind the piston in cylinders
of various sizes. The relation for the equivalent spring constant is

7TD ~p o
(12.21)
where k = spring rate, lb/in.
D = cylinder diameter, in.
L = cylinder length, in.
po = initial pressure behind piston, psi
12.6. Pneumatic Transmission Lag. Another effect that should be
considered in pneumatic systems is the transmission- or transport-lag
effect. Since air is much less dense than hydraulic fluid, the velocity of
propagation is lower. The approximate figure for the speed of sound in
air may be taken as 1,100 ft/sec, about one-fifth the speed of sound in
hydraulic oil. The actual signal-propagation velocity (group velocity)
will be less than this. If the pneumatic line were a perfect transmission

* H. Levenstein, Pneumatic Servomechanisms, Control Eng., vol. 2, no. 6, p. 67,

1955.
442 CONTROL SYSTEM COMPONENTS [Chap. 12

line, the output would reproduce perfectly any inputs, delayed in time,
however. The Laplace transfer function for a perfect transmission
line may be found by employing the real translation theorem.1 If
£[/(£)] = F(s), it can be shown that

£[/(* “ <o)I = F(s)e~^ (12.22)

Unfortunately actual pneumatic transmission lines are not perfect; they

attenuate and distort the signal. In Fig. 12.5 is given the experimental

1,000
800
700
600
500
400
300
250
200
150
C

£ 100
80
70
60
50
40

30

20

10
0.25 0.4 0.6 0.8 1.0 2 3 4 6 8 10
D cylinder diameter, in.
Fig. 12.4. Equivalent spring rates of compressed air. Trapped air was initially at
100 psi.

response of a long pneumatic line to a step-function input.2 The delay

time appears to be about 3.0 sec, which is more than twice the time that
would be calculated if the theoretical velocity of sound in air were
employed. Furthermore, there is a high-frequency attenuation that
rounds off the sharp edge of the step function.
As a first-order approximation, a transmission line might be represented

1 Gardner and Barnes, “Transients in Linear Systems,” John Wiley & Sons, Inc.,
New York, 1950, p. 236.
2 M. Bradner, Pneumatic Transmission Lag, Instruments, vol. 22, p. 618, 1949.
Sec. 12.7] PNEUMATIC SYSTEMS 443

Actual
'curve

'Delay time
3 sec plus
single time
constant of
14.5 sec

Fig. 12.5. Experimental response of a long pneumatic line to a step-function input.

by a delay time and a single time

constant, or

■} tas
y (12.23)
x 1 + Ts

where U is the delay time and T is

the time constant. Shown in Fig.
12.5 for comparison with the actual
response of the line is the response of
this approximation to a step function rod/sec
[a) The attenuation
of applied pressure. U is taken as
3 sec, and T was taken as 14.5 sec.
The time constant T was taken from
the actual curve at 63 per cent of the
final value. In Fig. 12.6 are shown
the frequency response of the actual
line, as obtained by Bradner, and the
response of the approximation for
comparison. It may be seen from
Figs. 12.5 and 12.6 that the approxi¬
mation compares quite favorably [5) The phase shift
with the actual test data. For more Fig. 12.6. Frequency response of a
accurate results the exact transmis¬ pneumatic transmission line. This is
sion-line equations given in Chap. 10 the line whose response is shown in
Fig. 12.5, and the single-time-constant
should be employed.
approximation is also shown here for
12.7. Pneumatic-Electric Ana¬ comparison.
logues and Pneumatic Control Func¬
tions. Just as it was possible to find electric analogues for mechanical and
hydraulic elements, it is possible for pneumatic elements. Usually pres-
444 CONTROL SYSTEM COMPONENTS [Chap. 12

sure is represented as voltage, and volumetric flow as current. Stolarik1

has developed resistance elements consisting of baffle plates and capillary
tubes that are essentially linear in their behavior. Capacitors are tanks
or reservoirs. Figure 12.7 shows a pneumatic resistance and capacitance
network and its electric equivalent.
Using the pneumatic equivalents of resistance and capacitance, it is
possible to construct pneumatic networks to given frequency-response
specifications, thus making possible the synthesis of stabilizing networks
for closed-loop pneumatic systems.
The particular network to be con¬
structed depends on the control
function that is to be synthesized.
In Chap. 1 is given a list of control
functions and the electric networks
that generate them. The pneu¬
matic device may be constructed
by analogy to these electric net¬
works. The magnitudes of the

10,-3 10'2 10'1 10c

Diometer of orifice, in.
Fig. 12.7. A pneumatic high-pass network Fig. 12.8. Resistance of a round orifice.
and its electric equivalent. (After (Stolarik)
Stolarik)

various pneumatic components may be determined by calculation or by

the use of Figs. 12.8 to 12.13. These curves are also valuable in calculat¬
ing the unavoidable lags due to given pneumatic components.
Figures 12.8 through 12.11 relate pneumatic resistance to variously
shaped constrictions. Pneumatic resistance is defined by analogy to
electric resistance as
AP
R = lb-sec./in.5 (12.24)
F

where AP is the pressure drop and F is the volumetric flow. The pressure-
flow relations have been derived in Secs. 12.3 and 12.4. These relations

1 E. Stolarik, Notes on Pneumatics, Aero Dig., vol. 60, no. 1, p. 45, no. 2, p. 44,
1950.
Sec. 12.7] PNEUMATIC SYSTEMS 445

are merely presented in a convenient form in the following curves. The

resistance is derived for air at standard conditions of 15°C and 14.7 psi
and for an orifice discharge coefficient of 0.62. While the resistance of
capillary tubes is essentially independent of pressure, it does depend on
temperature, and the resistance of orifices depends on both temperature

10~4 10'3 10‘2 10'1

BH, in.2
Fig. 12.9. Resistance of rectangular orifice. (Stolarik)

and pressure. For both Fig. 12.8 and Fig. 12.9 a correction factor for
other than standard conditions may be derived as

(12.25)

For Figs. 12.10 and 12.11 the correction factor

(12.26)

may be applied for other than standard conditions.

The pneumatic capacitance of a volume can likewise be defined by
analogy to electric circuits as

in.5/lb (12.27)

where F is the volumetric flow in cubic inches per second and P is the pres¬
sure in pounds per square inch. The value of capacitance is independent
of temperature but depends upon pressure and on the process of expan¬
sion. If the expansion is rapid, the process may be assumed to be
446 CONTROL SYSTEM COMPONENTS [Chap. 12

10’

10c

CD
CL

m.
c
\o
CD
CO

10 -1
CO
E
x:
o
CD
O
c
o
10 -2
co
CD
cr

10'3
-3
10 10-2 10'1
Diameter of tubes, in.

Fig. 12.10. Resistance of capillary tube. (Stolarik)

For standard conditions

,Fig. -12.11. Resistance of flat plates. (Stolarik)

Sec. 12.7] PNEUMATIC SYSTEMS 447

adiabatic, and if the expansion is very slow, an isothermal process may

be assumed. Usually the actual process is somewhere between the two
limits. Figure 12.12 relates capacitance to volume, and Fig. 12.13 gives
the correction for pressure.
In addition to their use in the synthesis of pneumatic control functions,
the relations between physical configuration and pneumatic resistance
and capacitance permit the determination of time constants of a pneu¬
matic element that performs an operational function in the loop, such as
the bellows in the example in the following section.
A second method of synthesizing control functions is to design a pneu¬
matic or pneumatic-mechanical closed-loop system with the required

Independent of temperature

10"’ 10° 101

Volume, in.3
Fig. 12.12. Capacitance. (Stolarik) Fig. 12.13. Pressure correction for
capacitance. (Stolarik)

input-output relations. Caldwell1 has compiled a table of such units

that is reproduced in Fig. 12.14. Williamson2 has designed small, com¬
pact, adjustable lag networks and lead networks in the modular, or
“plug-in,” form for the same types of control applications. Figure 12.15
shows sketches of these devices.
As an example of the analysis of a pneumatic equalizer, let us consider
the lag-lead network shown in Fig. 12.16.* We desire the transfer func¬
tion from flapper input position 6 to output pressure p. We shall assume

1 W. I. Caldwell, Generating Control Functions Pneumatically, Control Eng., vol.

1, no. 1, p. 58, 1954.
2 H. Williamson, Theory and Design of Compound Action Pneumatic Controllers,
Trans. Soc. Instrument Tech., vol. 6, no. 4, p. 153, 1954.
* A. R. Aikman and C. I. Rutherford, The Characteristics of Air Operated Con¬
trollers, in Tustin (ed.), “Automatic and Manual Control,” Proceedings of Cranfield
Conference, 1951, Butterworth & Co. (Publishers) Ltd., London, 1952, p. 175.
 Plus Rate

448
449
Fig. 12.14. Pneumatic-mechanical equalizers. A indicates supply pressure.
 Plus Reset Plus Rate

450
Fig. 12.14. (Continued)
 <v
03x
o
Fh
Pi
>>
E
PX
CQ
O03
c3

T3

in«o

T5a
ce
•p
o
CO
_
fc-
o
Plus Reset Plus Rate

■G

oa
o
r5

o
-A
>>
-aa>
w
ot-

<
z
'o
t-

o
HH
G
Oo
H
«
OPH
'O
o G
3
« .g
Ph "3
G
O
o
oo.
o
-G

pG
o
3
O£
-G

c«
G
-G
«
3
£

oc
2
o
-p>
G

451
452 CONTROL SYSTEM COMPONENTS [Chap. 12

Timing adjuster

Timing restrictor
Timing
capacity

Negative feedback
pressure

Sensitized diaphragms
providing mid point toad ■

Plug-in connectors
PVC tubes

Proportional pressure Proportional and

integral pressure
Air leak valve
(o) Lag

Timing odjuster

Timing restrictor

Timing capacity

Air leak valve

Sensitized diaphragms

Plug-in connectors Zero spring

PVC tubes

Proportional pressure Proportionol and

derivative pressure
(b) Lead
Fig. 12.15: Plug-in pneumatic equalizers. (Williamson)
Sec. 12.7] pneumatic systems 453

that the flapper is operating in its linear range. Thus

b a
cd — cKp2 (12.28)
a + b a + b
where P = initial output pressure
c = gain constant of flapper, psi/in.
K = spring constant of feedback bellows, in./psi
Following the usual procedure of writing the linear equation about an
operating point, P may be struck out of Eq. (12.28). For the derivative
restrictor we may write

V = (1 + T1s)p1 (12.29)

where the constant T has the dimensions of a time constant. Its value
is determined by the area of the restriction and the volume of air at the
Air
Output supply

Fig. 12.16. A pneumatic lag-lead network, or a so-called type I controller.

pressure pi. Likewise we may relate pi and p2 through the second

restrictor by
AT2sp, = (1 + T2s)p2 (12.30)

where A = -A— at operating point

dp i
T2 = time constant of integral constriction and volume of feedback
bellows
Solving the three equations, we obtain

P _ t, _(Tis + 1)(T2s + 1)_

(12.31)
e T,T2s^ + (T, + T2 + AhT2)s + 1
acKA be
where and k2 =
a + b a + b
454 CONTROL SYSTEM COMPONENTS [Chap. 12

The form of Eq. (12.31) is rather different from that commonly used in
the process-control industries. There, the traditional method of analysis
would relate p and 6 by the differential equation containing the sum of the
derivative, proportional, and integral terms. An attempt is usually made
to cling to a physical interpretation of the separate terms by the introduc¬
tion of interaction terms, which turn out to be various ratios of the time
constants in Eq. (12.31). The interaction terms attempt to relate con¬
troller dial settings, which adjust the two restrictors in the example
above, to the time constants1 of the transfer function. As the advantages
of operational techniques become more widely known, the transfer func¬
tion and the frequency response of elements will themselves become a
part of physical reality for process-control engineers, and manipulation
of the time constants will be acceptable. This linearized analysis may
be used to develop transfer functions for the configurations shown in
Fig. 12.15 and the additional configurations given in the problems at the
end of this chapter.
12.8. A Pneumatic Control System. As an example of a pneumatic
control system that includes several of the more common types of compo¬
nents, we shall consider the pneumatic positioning system shown in Fig.

12.17. This system might be used to actuate the control surfaces of an

aircraft or to perform any other positioning operation. A is a source of
low-pressure air (50 psi) for operating the flapper valve. This low-
pressure supply, in contrast to the high-pressure source, must have its
pressure rather closely regulated, since the spring E is adjusted for a given
pressure. Likewise the spring I is adjusted so that the entire system will
be in the center of its operating range with no signal to the torque motor.
1 J. Janssen, Analysis of Pneumatic Controllers, in Tustin (ed.), ibid., p. 189.
Sec. 12.8] PNEUMATIC SYSTEMS 455

It would be possible to operate the flapper valve on high-pressure air,

but the leakage and the force required from the torque motor would be
unnecessarily high, and furthermore it is difficult to regulate the pressure
properly. B is the upstream orifice, and C is the downstream or con¬
trolled orifice of the flapper valve. D is the torque motor that actuates
the flapper. The spring E balances the force from the air jet. The bel¬
lows F operates the pilot valve G against the spring I. H is the high-
pressure air supply (500 to 1,000 psi). The exhaust port J of the pilot
valve is larger than the inlet port, to allow the expanded low-pressure air
to escape without throttling. This is one of the few particulars in which
pneumatic valves differ from hydraulic valves. The extension of the
pilot-valve spool K may be used if it is desirable to pick off spool position.
This two-stage pneumatic amplifier consisting of the flapper valve and
pilot valve would be referred to as a “relay” in the process-control indus¬
try. It drives the power piston L, which operates on the differential-area
principle. Even if the pressure were the same on both sides of the power
piston, it would move upwards, because the pressure acts over a larger
area on one side of the piston. The output position M actuates the load
and is normally fed back to close the control loop.
From the data in Chap. 13 we shall see that in the operating range there
is a linear relationship between flapper position and bellows pressure.
The bellows force tending to open the valve will be assumed to be propor¬
tional to the chamber gauge pressure. Newton’s law for the mechanical
system is thus
^beiiows = {Ms2 fl- Bs + /bi + /c& + kc)x (12.32)

where M is the mass of the moving parts and B is the coefficient of viscous
damping. k\ is the coefficient of the spring 7; kb is the spring constant of
the bellows; and kc is the equivalent spring due to the Bernoulli force at
the valve orifice. Thus the transfer function from bellows force to valve
position is
x _ 1
(12.33)
ffbeiiows Ms2 + Bs + ki + kb + kc
The relationship between flapper position and bellows force may be
assumed linear in the operating range; however there is a pneumatic time
lag between flapper position and bellows pressure which is proportional to
bellows force
tlbellows _ '^flap
(12.34)
#flap 1 + TbS

The method of finding Tb for a given system is discussed below. The

transfer function from valve opening to pressure on the power piston
depends on the length of the transmission line. If the line is short, the
456 CONTROL SYSTEM COMPONENTS [Chap. 12
transmission-line effects may be ignored, and a single time constant made
up of the resistance of the valve orifice and the capacitance due to the
volume of air in the line and the power piston may be assumed. If the
line is not short, the approximation of a transport lag and the time con¬
stant, as discussed in Sec. 12.6, will usually be adequate.
A simple equivalent circuit may be employed for the pneumatic valve
if the valve may be assumed to be approximately linear. This circuit is
shown in Fig. 12.18. According to this
Pneumatic
resistance circuit the pressure p delivered by the
valve for zero (or constant) flow is pro¬
portional to x, the valve spool position.
The slope dp/dx is referred to as the pres¬
sure sensitivity of the valve, and may be
obtained by direct measurement, or by
Fig. 12.18. Equivalent circuit of
pneumatic valve.
calculations of the sort performed in
Chap. 11 for hydraulic systems. As a
result of fluid flow to the load this pressure is reduced, and this effect is
indicated by the series resistor R. This resistance, which is the slope of
the pressure versus flow curve, may also be obtained by measurement or
by calculation. The reader may recognize this equivalent circuit of the
valve as being in the form of the Thevenin equivalent1 circuit used in
electric circuit analysis.
The time constant Th will now be determined, although it will be found
to be a function of the operating conditions rather than a true constant.2

-o e<

High U
f IB Cf
pressure

V
Fig. 12.19. The flapper valve and its electric analogue.

Figure 12.19 shows the pilot valve and its electric analogue. In the nor¬
mal operating range the ratio of the upstream pressure to the downstream
pressure across orifice B is greater than 2:1; thus the flow through the
orifice is constant. In the electric analogue this is represented by a con¬
stant current source I. The transfer function of interest is from a change
in the flapper orifice resistance Rc to the change in bellows pressure, repre-
1 W. Everitt and G. Anner, “Communication Engineering,” 3d. ed., McGraw-Hill
Book Company, Inc., New York, 1956, p. 231.
2 For a derivation that assumes linearity about an operating point, see H. A. Helm,
Frequency Response Approach to the Design of a Mechanical Servo, Trans. ASME,
vol. 76, p. 1209, 1954, and the appendix by R. Oldenburger.
Sec. 12.8] PNEUMATIC SYSTEMS 457
sented by the voltage eF. Unfortunately, as the bellows expands, its
equivalent capacitance changes; thus CF is a function of x. If the change
in bellows volume is only a small percentage of the total bellows volume,
this variation in C may be neglected. We may write by KirchhofPs law
that
I = iRc + icF (12.35)

or I = + cF (12.36)
rCc CU

Taking the Laplace transform we obtain

T A
— — + C FscF — CFeF(fi) (12.37)
S Kc

where eF(0) is the initial value of the voltage eF. Suppose now that at
time t = 0 the resistance Rc is changed suddenly from the value RCi to a
new value Rc2. If we assume that a static condition exists prior to this
change then deF/dt in Eq. (12.36) is zero, and the initial value of the volt¬
age on the capacitance is
£f(0) = Rcil (12.38)

Substituting this value for e^(0) in Eq. (12.37) and letting Rc take on the
new value Rc2 we obtain

I _ eF
T CFscF — RciCfI (12.39)
s Rc 2

Solving for eF gives

A
6F —
RcJ(T 1 s + 1)
(12.40)
s(T 2S +1)
where Ti = RciC F
(12.41)
t2 = RciC F

Finally if we take the inverse Laplace transform of Eq. (12.40) we find

that
eF(t) = IRc2 H- I(Rci — Rc2)t~t/T2 (12.42)

Thus the voltage eF approaches its final value in the familiar exponential
manner, and the time constant of the circuit is T2. When Rc is varied in
some other manner than stepwise, T2 varies during the process; however,
as an approximation for the time constant, some average value of resist¬
ance may be assumed.

PROBLEMS
12.1. Find the mass weight of flow for a smoothly rounded orifice 4i6 in- in diam¬
eter at 20°C. The upstream pressure is 1,000 psi.
458 CONTROL SYSTEM COMPONENTS [CHAP. 12

12.2. Find the equivalent time constant of a 5-in.-diameter cylinder 50 in. long with
a 1-lb piston.
12.3. Find the physical dimensions of a pneumatic network at 15°C and 1,000 psi
to synthesize the function 1 /(Ts + 1).
12.4. In Fig. 12.20 is shown a schematic of a so-called type II controller, which is
sometimes called a series controller because of the arrangement of the derivative and
integral restrictors.1 Find the transfer function from 6 to p.

Fixed Air
Output restrictor supply

Fig. 12.20

12.5. Find the transfer function of the parallel type III controller shown in Fig.
12.21, which is taken from Aikman and Rutherford.

Fixed Air
Output restrictor supply
<
P

Derivative xh
action Nozzle or
restrictor pilot valve

Integral
action —3 &
restrictor

Fig. 12.21

1 Aikman and Rutherford, loc. cit.

Problems] PNEUMATIC SYSTEMS 459
12.6. Find the transfer function of the type IV controller shown in Fig. 12.22,
taken from Aikman and Rutherford. Note the use of a combination of hydraulic
and pneumatic elements.

Fixed Air
Output restrictor supply

12.7. Find the transfer function of the type V controller shown in Fig. 12.23, taken
from Aikman and Rutherford.

Fixed Air
Output restrictor supply
460 CONTROL SYSTEM COMPONENTS [Chap. 12

12.8. Determine the transfer function of the lag4ead controller shown in Fig.
12.24, which is taken from Janssen.1 Assume that the gain of the flapper-nozzle
pilot valve is very high.

A
T

~m WiT^
K b

^ Pi P»
4 B -Pilot
valve
s\aaAaa.\v
Sn\\\W\W\^
V; Kf
r 1— Supply
J i—,pressure

vrvi
v;El vd:t
vN K
I Rri P+p
output
pressure
Fig. 12.24

12.9. Develop Eq. (12.16) from Eq. (12.15) by assuming constant weight rate of flow
and constant temperature.

1 Janssen, op. cit.

 CHAPTER 13

PNEUMATIC COMPONENTS

13.1. Introduction. In general, all pneumatic control systems are of

the valve-controlled type. There are few, if any, systems in which the
pump is placed in the loop and directly controlled. Quite commonly the
pneumatic pump is incapable of supplying the maximum load of the sys¬
tem by itself. Usually an accumulator, or storage element, and a pres¬
sure-regulating valve are included in the pressure-supply unit. In fact,
in some units that have a limited operation time, e.g., in guided missiles,
the pump is omitted altogether, and the system operates from an accumu¬
lator that is charged from a fixed supply before placing the system in
operation.
Many of the components used in a high-pressure pneumatic system are
identical with their hydraulic counterparts, and the two types may in fact
be used interchangeably when this is true, as it is in several types of
valves and motors, for example. One common difference in pneumatic
components, however, is that the low-pressure outlets must usually be
significantly larger than the inlet ports, so that the increased volume of
gas at low pressure will not be throttled as it passes.
Pneumatic systems are usually slower than equivalent hydraulic sys¬
tems owing to the finite time required for expansion of the gas. It is
unfair, however, to compare the same system operating at the same supply
pressure on the basis of speed of response with hydraulic and pneumatic
fluids. The comparison should actually be made on a weight or cost
basis for given specifications. Pneumatic systems are being designed to
meet faster and faster response requirements, and certain pneumatic con¬
trols on jet aircraft, for example, have frequency responses that extend to
higher than 10 radians/sec.
A typical example of an all-pneumatic control system on a military jet
aircraft is shown in Fig. 13.1.1 The device is employed to maintain a
constant ratio2 between the pressure in the combustion chamber and the
1 W. E. Reed, Pneumatic Control of a Turbojet Variable Nozzle, Control Eng.
vol. 3, p. 93, October, 1956.
2 Actually there are reasons for adjusting this ratio as a function of flight condition
in order to maintain constant turbine-blade temperature, etc.
461
462 CONTROL SYSTEM COMPONENTS [Chap. 13

downstream pressure in the turbojet engine. The differential pressure

bellows controls the pneumatic valve, which in turn operates the actuator.
Changing the nozzle opening changes p2, the downstream pressure.
If the pressure p2 is too large the valve stem is forced down and orifice
2 narrows. As a result p3 increases, and equilibrium is reached when the
force generated by the pressure difference p2 — p3 is equal to the oppos¬
ing force produced by the bellows.
The bellows may be represented by a single lag with a time constant of
about 0.01 sec in the example given, while the equivalent spring of the

Air Nozzle

Actuator

Fig. 13.1. An all-pneumatic servo in a jet-engine application. (Reed)

compressibility of the air in the actuator and the mass of the actuator
plus the mass of the load contributes an underdamped complex-conjugate
pair of roots. There is, in addition, the integration between valve posi¬
tion and actuator position. Reed finds the transfer from pressure ratio
to actuator position from experimental data to be approximately

x k
(13.1)
Ap s(0.01s + 1) (0.0004s2 + 0.008s + 1)

and a possible block diagram is shown in Fig. 13.2. It is seen from the
equation that the system becomes unstable if k is too large. It is possible
to equalize this system pneumatically by installing a compensator of the
form shown in Fig. 13.3. The nozzle-position actuator is connected to
the piston of fhe compensator, and the output of the compensator posi-
Sec. 13.2] PNEUMATIC COMPONENTS 463

tions the orifice 0i. The pressure in the volume C\ is proportional to x,

and the force out is proportional to the rate of change of pressure on the
top of the diaphragm. The actual linear transfer function for the
equalizer is given in the figure. The usual linear design techniques may
be used to set the time constants R\C\ and R2C2 so that the resultant fre¬
quency response of the system with equalizer indicates a satisfactorily
stable system.

Air volve
Pneumatic and Variable
control actuators nozzle Engine

Fig. 13.2. Simplified block diagram of pneumatic control system.

Turbine discharge
pressure

R\

* 9
F_=_ksz
2 (/?,<;, s+/)(R2Czs+/)

Fig. 13.3. A pneumatic equalizer for the system in Fig. 13.2.

The object of discussing this example is to point out that completely

pneumatic control systems including equalizers have a definite place in
modern high-speed applications. It is true, however, that at the present
time the number of hydraulic installations far exceeds the number of
pneumatic installations in high-speed, high-power applications.
13.2. Comparison of Weight of Electric, Hydraulic, and Pneumatic
Systems. In Chap. 12 an aircraft pneumatic system was mentioned
that weighed only one-half as much as an equivalent hydraulic system.
464 CONTROL SYSTEM COMPONENTS [Chap. 13

As might be expected, the matter of the weight of equivalent electric,

hydraulic, and pneumatic systems for aircraft has received careful study,
and the work of Geyer and Treseder1 seems particularly complete. Their
data on actual production components were for open-loop systems but
may be applied directly to control systems also. The systems are broken
into four categories: generation, storage, transmission, and actuator.
In Fig. 13.4 are shown the data on generation units. The production
units for pneumatic systems are usually rated for less than 1 hp continuous

0.01 0.02 0.05 0.1 0.2 0.5 1 2 5 10 20 50 100

Horsepower

Fig. 13.4. Weight-per-horsepower comparison for modern aircraft power supplies.

{Data from Geyer and Treseder)

duty, because storage of energy in pneumatic form is more economical, in

terms of weight, than generation.
In Fig. 13.5 is shown the weight of energy-storage elements for the
various types of systems. The weight of hydraulic accumulators and
pneumatic storage cylinders is seen to be approximately equal, with
the pneumatic having a slight edge. These figures include the weight of
the working fluid. The data are plotted against the length of time that
the storage device must deliver power to the load. It may be seen that
the electric battery is the lightest storage form for loads of duration
greater than 1 minute. The additional curves plotted in Fig. 13.5 for the
weight of generating equipment allow a choice between generation and
storage for loads for various durations. It may be seen that in practically
1 H. M. Geyer and R. C. Treseder, Weight Analysis of Aircraft Actuators, Trans.
AIEE, part II, vol. 71; pp. 118-126, 1952.
Sec. 13.2] PNEUMATIC COMPONENTS 465

100
80
60
40

20

10
T 8
^ 6
-O

i2
i
0.8
0.6
0.4

0.2

0.1
1 2 4 6 8 10 20 40 60 100 200 400 1,000
Time of operations, sec
Fig. 13.5. Weight comparison for storage units as a function of time of operation.
(Geyer and Treseder)

all cases it is more economical to provide adequate generating capacity in

hydraulic and electric systems than to accumulate energy in a storage
container. The hydraulic accumu¬
lator and the electric battery would
be used only for smoothing shocks
and for emergencies. In the case of
the pneumatic system, however,
the high weight cost of compressor
horsepower renders the pneumatic
storage element more economical
for many loads. A major draw¬
back, however, to the storage of
large amounts of energy in pneu¬
matic form is the danger of explo¬
sion should the accumulator be
pierced. Horsepower

It is in the transmission of power Fig. 13.6. Weight cost of the transmission

of power 75 ft. (Geyer and Treseder)
that the pneumatic system gains its
major weight advantage over hydraulic systems. Figure 13.6 gives the
weight cost of transmitting energy 75 ft. The pneumatic system is
lighter because only one line is needed and the weight differential
between the fluids is in its favor. The major drawback to the transmis-
466 CONTROL SYSTEM COMPONENTS [Chap. 13

sion of power by compressed air over long distances is the transmission¬

line effects developed. The lags of pneumatic transmission lines were
discussed in Chap. 12.
In Fig. 13.7 is shown a weight comparison of pneumatic, hydraulic, and
electric actuators. It will be noted that, although the pneumatic actuator
has a slight edge with respect to the hydraulic type, they can be con¬
sidered equal for all practical purposes. Both are superior, it may be
seen, to electric actuators.

Fig. 13.7. Weight per unit of work versus rated work of actual production aircraft
actuators. (Geyer and Treseder)

The pneumatic system, then, appears to be the most economical in

weight in a moderately high peak horsepower system where a moderate
to long transmission line is required and with a duty cycle that allows the
energy to be stored in a large accumulator by a small compressor. This
type of system is quite common in aircraft and guided missiles. In a
system with other than these specifications, the pneumatic system might
still be chosen to obtain one of its other advantages, as discussed in
Chap. 12. The problem of establishing the optimum balance of the
reserve energy supply for peak loads and the weight of stand-by equip¬
ment required to meet these relatively infrequent demands is one that
must be solved for each individual case. It is conceivable that a system
might have a pneumatic compressor capable of supplying less than one-
tenth the peak' load anticipated in a particular installation, if the peak
Sec. 13.4] PNEUMATIC COMPONENTS 467

load were met infrequently. The accumulator could be charged initially

from a fixed supply, and the system pump would act merely as a topping
unit.
13.3. Choice of Operating Pressure. In the process-control industry,
where size and weight are not of ultimate importance, the operating pres¬
sure of pneumatic systems is typically below 50 psig. The system compo¬
nents are then relatively large and easy to manufacture. Operation is
reliable, and there is little danger of components under pressure exploding
from rough handling. As the demands on speed of response are increased
in the more critical processes envisioned in future developments, these
operating pressures will be increased.
In high-speed pneumatic systems, especially air-borne systems, pres¬
sures have been continuously increased, although as yet there does not
seem to be the standardization that there is in hydraulic installations.
Aircraft pneumatic systems may operate from jet-engine compressor dis¬
charge pressure of 100 to 200 psi at sea level to about 50 psi at 50,000 ft
altitude. Typical operating pressures for systems supplied by individual
positive-displacement compressors are 500 psi, 1,000 psi, and 1,500 psi.
The compressor charges an accumulator at about 3,000 psi, and the accu¬
mulator discharges to the system through a reducing valve. Experi¬
mental compressors and accumulators have been designed for 5,000-psi*
storage for systems with 500- to 1,000-psi operating pressures. Schmidlin
states that these 5,000-psi compressors are not appreciably heavier than
the more usual 3,000-psi compressors and that the drive-power increase
is only 15 per cent. The 5,000-psi accumulator is considerably smaller
than an equivalent 3,000-psi element, the volume of the former being
approximately 60 per cent of the volume of the lower-pressure element,
as one would predict theoretically. Furthermore, the 5,000-psi accumu¬
lator requires walls of such a thickness that the container is essentially
nonshatterable, whereas present design for the 3,000-psi unit requires
wire-wrapped reinforcement. No doubt, both operating pressures and
storage pressures will continue their present steady increase in the near
future.
13.4. Pneumatic Power Supplies. The design of air compressors is
beyond the scope of this volume, but the basic features necessary for
intelligent use of compressors in control systems will be discussed. Pneu¬
matic pumps, or compressors, may be of the turbo type or of the positive-
displacement, or piston, type. For operating pressures above 50 psi the
positive-displacement type is used almost exclusively because of its
greater efficiency.
The pv diagram of an ideal reciprocating-compressor cycle is given in
* A. E. Schmidlin, Potential Advantages of a 5,000 psi Pneumatic System, App.
Hydraulics, October, 1954.
468 CONTROL SYSTEM COMPONENTS [CHAP. 13

Fig. 13.8. The intake stroke (1-2) is at inlet pressure. The compression
stroke will lie between the extremes of perfect isothermal compression
(2-3) and adiabatic compression (2-3'). The isothermal process requires
that the heat due to compression be extracted continuously to maintain
the gas at constant temperature, while the adiabatic process requires that
the cylinder walls be perfectly insu¬
lated so that no heat escapes. The
Discharge pressure
area enclosed by the cycle represents
:Isothermal compression
(pv-c)
the work of compression; thus the
isothermal cycle is to be preferred
Adiabatic compression
(pi/1A = c) since it represents the smaller value of
work. Even with isothermal com¬
pression, however, a good deal of
Inlet pressure
energy is required to pump a volume
Volume
of gas to an operating pressure of
Fig. 13.8. The pv diagram of an ideal
compression cycle. 800 or 1,000 psi. Let us say that an
ideal pump is to compress air, at the
rate of 0.001 lb/sec to 1,000 psig. If perfect isothermal compression is
assumed, we may use the familiar relation

WRT. p.
(13.2)

where W = weight of air, lb/sec

R = gas constant = 53.3 ft/° Rankine for air
T = absolute temperature, ° Rankine (530° Rankine = 70°F)
p = absolute pressure, psia

(0.001) (2.47 X 105) (530) (6.94 - 2.71)

Thus (13.3)
(6,600) (386) U'^ *

To compress the same weight of air adiabatically requires

(fc—\)/k
WRT k
~ 1 (13.4)
P ” 6,600<7 k - 1
or
286
, (0.001) (2.47 X 105)(530) 1.4 1,015V
- 1 = 0.417 (13.5)
hp = -(6,600) (386)-Od 15 )

In order to bring the efficiency of practical compressors as near the

isothermal ideal as possible, several stages of compression are used, and
interstage cooling is provided. Two or three stages are used with
1,000-psi units and as many as seven stages for 5,000-psi units. The cycle
of a two-stage compressor is shown in Fig. 13.9. The intake stroke (1-2)
remains the same &s for a single-stage unit. The actual polytropic com-
Sec. 13.4] PNEUMATIC COMPONENTS 469
pression stroke (2-5) for the first stage lies between the extremes of iso¬
thermal and adiabatic compression. Interstage cooling takes place at
constant pressure (5-5'), and here it is assumed that the cooling is perfect.
The second-stage compression stroke (5'-3") then takes place. The
shaded area represents the work saved by interstage cooling. Typical
comparisons for work expended in one-, two-, and three-stage compression
cycles are shown in Fig. 13.10.
In addition to improving efficiency, multistage compressors in which
the compression ratio per stage is held down to a factor of 2 or 3, with
interstage cooling provided, have another advantage. It is rather dan¬
gerous to compress even dry air with ratios of 8 or 10, since even a small
amount of lubricating oil from bearings or oil working back from the

Fig. 13.9. The compression cycle of a Fig. 13.10. Efficiencies of multistage com¬
two-stage compressor with inter¬ pression cycles for pneumatic supplies.
stage cooling. {From Lucke, “ Engineering Thermodynamics,”
Columbia University Press, New York)

aspirator into the compression chamber may cause a diesel type of explo¬
sion at the high temperature and pressure in the chamber.
Much of the work actually done in compressing air cannot be recovered,
because it is impossible to design nozzles and loads to allow the air to
expand most efficiently, i.e., isothermally. Among other reasons this is
due to the time that would be required for such an expansion. Thus
over-all efficiencies of pneumatic systems are typically below 50 per cent,
and figures as low as 30 per cent are sometimes quoted.1 The efficiency
of the compressor alone will usually lie between 70 and 85 per cent.
An exception to the general rule of positive-displacement compressors
is found in jet aircraft, where the pneumatic supply may be the engine-
compressor discharge pressure, as in the example discussed in Sec. 13.1.
While this procedure is simple and convenient, it has at least two dis-
1 J. L. Shearer and S. Y. Lee, Selecting Power Control Valves, Control Eng., vol.
3, p. 72, March, 1956.
470 CONTROL SYSTEM COMPONENTS [Chap. 13

advantages. First, the compressor discharge pressure may vary by a

factor of 5; 1 from sea level to 50,000 ft, thus seriously affecting the per¬
formance of the pneumatic system. The second disadvantage occurs in
pneumatic systems designed for relatively high power applications.
Here the efficiency of the jet engine may be disturbed by heavy drains at
the compressor discharge.
Even when the high-pressure air is supplied by a piston-type compres¬
sor, a change in altitude has an effect. A given compressor will not
produce so large a weight of compressed air nor will its efficiency be con¬
stant as the ambient pressure is reduced.
13.5. Pneumatic Control Valves. The basic similarity between pneu¬
matic valves and hydraulic valves allows us to base this discussion on
Chap. 11 and, in the main, merely point out the differences between the
devices. The basic configurations such as the piston valve and the flapper
valve are quite the same for both liquids and gases. In fact the original
development of the flapper valve seems to have been made for low-
pressure pneumatic systems in the process-control industry and thereafter
adapted for high-pressure hydraulic fluids.
Certain factors assume a different importance in pneumatic valves.
The fluid reaction force, or Bernoulli force, for example, is negligible in
the typical pneumatic control valve because of the very small mass rate
of flow. For example, 1 ft3 of air at 1,000 psi weighs 0.0134 lb, whereas
the weight of 1 ft3 of Univis J43 hydraulic fluid is 52.6 lb.
The viscosity of compressed air is negligible compared with that of
hydraulic fluids, and thus the viscous damping in a pneumatic valve may
usually be neglected. The low viscosity does, however, add one problem.
The leakage flow in the clearances of the valve can be large. Even with
clearances as small as a ten-thousandth of an inch the leakage flow is
almost as large as it would be in an orifice of the same area as the clearance
area;1 i.e., the pressure drop in the leakage path is negligible. The char¬
acteristics of control valves can be calculated in the same manner as for
hydraulic valves. Experience has shown that the flow process through
such devices is very nearly adiabatic. Figure 13.11 shows the measured
characteristic2 of a nominally zero-lapped four-way slide valve of the
type discussed in Chap. 11. The curves will be recognized as quite
similar in form to the hydraulic-flow characteristics considered previously.
Figure 13.12 is the calculated characteristic of an open-centered or under¬
lapped valve; again the curve will be recognized as similar to the corre¬
sponding hydraulic-flow characteristic.
The performance of a typical positioning system using air has been
1 A. Egli, The Leakage of Gases through Narrow Channels, Trans. ASME, vol. 59,
p. A53, 1937.
2 J. L. Shearer, Study, of Pneumatic Processes, Trans. ASME, vol. 78, p. 239, 1956.
Sec. 13.5] PNEUMATIC COMPONENTS 471
compared to the performance of the same system using hydraulic fluid
by Lee and Shearer.1 A schematic diagram of the system is shown in
Fig. 13.13. The supply pressure in both cases is 1,000 psi, and a slide
valve is used to control the flow to a linear actuator. The quiescent

Pa
Fig. 13.11. Measured flow-pressure characteristic of a zero-lap valve as a function of
stroke x. (Shearer)

leakage flow in the valve is 0.005 in.3/sec, and the maximum no-load
velocity of the ram is 14 in./sec. In accordance with the principle dis¬
cussed above, the exhaust port of the valve has a maximum diameter
of % in., while the high-pressure port has a maximum width of only ^ in.
1 S. Y. Lee and J. L. Shearer, Proceedings of the National Conference on Industrial
Hydraulics, vol 8, p. 168, 1954.
472 CONTROL SYSTEM COMPONENTS [Chap. 13

The closed-loop frequency response of this servo is shown in Fig. 13.14,

and for comparison is shown the frequency response of the same system
when 1,000-psi hydraulic fluid is substituted for the compressed air. It
is clear that the frequency response of the pneumatic system is inferior
to that of the hydraulic one.

Fig. 13.12. Pressure-flow characteristic of an underlapped pneumatic valve with the

ratio of stroke x to underlap u as a parameter. (Shearer)

Pneumatic
supply

Position feedback
Fig. 13.13. Diagram of a pneumatic positioning system.

Many pneumatic positioning systems are designed with spring-loaded

actuators that may take the form of a bellows or a Bourdon tube. In
such a case the valve should control pressure rather than flow. Of course,
it is rather artificial to make this distinction between pressure control and
flow control, since any valve is merely an adjustable orifice that actually
controls only the relation between pressure and flow, but we are conform¬
ing here to common usage. An open-center slide valve or spool valve
Sec. 13.5] pneumatic components 473

may be considered as a pressure-control valve in this sense, as may a

flapper valve. In each case it is not necessary that there be load flow
in order to establish a required load pressure. The load pressure is
established by positioning the valve in the underlap region so that the
leakage flow to sump sets up the required pressure drop in the valve.

1 10 100 1,000
cps

Fig. 13.14. The frequency response of the positioning system shown in Fig. 13.13.
(From Lee and Shearer)

High pressure

Fig. 13.15. A two-stage, flapper-controlled pressure valve with differential-area

actuator.

Two pressure-control valves, somewhat more complex, are shown in

Figs. 13.15 and 13.16. The valve in Fig. 13.15 has a free piston that is
positioned by the pressures on its two ends. P'c is controlled by the
flapper stage, and the spool will move to such a position that the flow
from the supply through the high-pressure orifice and then the low-
pressure orifice to sump will make Pc such that
PcAc = P'CA'C
474 CONTROL SYSTEM COMPONENTS [Chap. 13

Since Pc is also applied to one end of the differential-area actuator, a

change in Pc results in an acceleration of the unloaded actuator. So long
as the size of the chambers is small enough to be negligible, the ampli¬
tude ratio varies inversely with the second power of frequency. If the
differential-area actuator is replaced by a bellows, the reader may show
that the position of the flapper controls the position at the output of the
bellows.
The valve shown in Fig. 13.16 is of the so-called nonbleed type and is
widely used in the process-control industry. The bellows limit its use
to the lower range of operating pressures. The valve has two stages, the
first consisting of the input flapper /i and nozzle Ui arrangement that

Fig. 13.16. Another type of two-stage, flapper-controlled pressure valve, called a non¬
bleed relay in the process-control industry.

controls the pressure pi. The second stage consists of the dual bellows
and the two nozzles n2 and n3 with their common flapper /23.
In equilibrium, piAi = p2A2, n2 is closed and n3 is almost closed by
flapper /23. If p 1 increases, the net bellows force carries n2 down, thus
forcing /23 away from n3 and allowing p2 to increase, returning n2 and w3
to equilibrium. If p 1 decreases, n2 moves up, leaving /23 to close n3 and
allowing p2 to vent to atmosphere through orifice 0 until the pressure
ratio is reestablished. The “nonbleed” appellation can be perhaps jus¬
tified since bleeding only takes place when either pi or p2 varies.
13.6. Pneumatic Motors. Rotary motors of any of the types discussed
under hydraulic components should theoretically operate properly under
pneumatic pressure, provided only that they receive proper lubrication.
At the present time., however, the only rotary-motor type commerically
Sec. 13.6] PNEUMATIC COMPONENTS 475

available for high-pressure pneumatic applications is of the vane type.

The clearance volume of this device, shown in Fig. 13.17, is larger than
the volume of a hydraulic motor of the same horsepower rating, but the
operating principle is identical.
The high-pressure fluid is admitted at the inlet port and acts against
the differential area of the vane, developing a torque. The rotor turns
and carries the vane past the outlet port, where the air is expelled to
atmosphere. The motor must be symmetric for reversible operation,
and no energy is extracted from the expanding air, thus wasting the
energy of compression. Although the vanes may be spring-loaded, the
typical vane motor depends on centrifugal force to keep the vane seated
against the housing. The spring-loaded vane is advised when the motor

Suction
Pressure

1,500

1,000
E
Cl
L_

500

01_l_l_I_l
Suction 0 5 10 15 20
'Pressure Torque, ft-lb

Fig. 13.17. Vane-type pneumatic Fig. 13.18. Speed-torque curves with

motor. operating pressure a parameter for a
pneumatic vane motor.

is to be stalled in normal operation. Typical speed-torque curves for a

vane motor as a function of operating pressure are given in Fig. 13.18.
Since torque is proportional to pressure in pneumatic devices as in hydrau¬
lic devices, the ideal characteristic would be a straight vertical line. It
may be seen that actual device corresponds closely to theory.
Up to the present time the gear-type design has not been successful
either as a hydraulic or a pneumatic motor, owing to the high starting
friction that is usually involved. It appears, however, that much of this
friction is due to improper design arising from attempts to produce an
economical unit. A properly engineered and manufactured gear motor
can be made to have as little as 5 lb-in. of static friction and to break free
and run at as little as 20-psi pneumatic pressure. These units will operate
successfully as pneumatic motors if the compressed air is supplied with
an oil spray from an aspirator in the line. The oil suspended in the com-
476 CONTROL SYSTEM COMPONENTS [Chap. 13

pressed air not only acts as a lubricant but also allows an oil film to be
built up which seals the meshed gears and contributes to the volumetric
efficiency of the device by preventing leakage between the gears to sump.
Figure 13.19 shows speed-torque curves for a gear motor designed for
pneumatic service.
The piston-type hydraulic motor should be capable of operation as a
pneumatic motor, but a curious phenomenon is sometimes observed when
this application is attempted. After a short period of proper operation,
a sort of diesel action begins to take place. Apparently, the compressed
oil vapor suspended in the air begins to explode behind the cylinders. At
the same time, however, the motor is
refrigerated by adiabatic expansion
Pressure ■ rpm
730psi ■
Motor of the air in the chambers, and the
motor case remains cool. This diesel
valve
action probably reduces the efficiency
of the device and may damage the
mechanism if allowed to continue.
13.7. Actuators. Linear actuators
2,000 of the piston type used for hydraulic
s 1,800 systems are used also for compressed-
" 1,600 air service. Both spring-loaded and
3
Q.
=)
1,400 differential-area types are also in use.
o /_
1,200 The only construction difference ap¬
\ c>
1,000 On pears in the packing and seals. The
800
>
* “
spring-loaded actuator seems more
600
VC
\ b common in pneumatic systems, and
VP
\r5 it may take the form of a bellows
400
\c
200 VcT' \c or a Bourdon tube as well as the
\cp

0 piston type.
0.2 0.4 0.6 0.8 1.0
Output torque, ft-lb
Bellows and Bourdon tubes are
also useful as temperature-sensitive
Fig. 13.19. Speed-torque curves for a
pneumatic gear motor. and pressure-sensitive devices in
both hydraulic and pneumatic control
systems. When these elements serve as temperature-sensitive devices,
their principle of operation depends on the temperature coefficient of
expansion of the fluid enclosed. In the pressure-sensitive application,
the linear extension of the device depends on the pressure of the fluid
enclosed.
As a spring-loaded linear actuator, the Bourdon tube usually has much
the stiffer spring constant. While it is possible to design a bellows with a
very high spring rate for applications where only a small extension is
necessary, the Bourdon tube will usually exhibit better repeatability in
position versus pressure and will have the longer life. Bourdon tubes
Sec. 13.7] PNEUMATIC COMPONENTS 477
are constructed of steel or brass and typically exhibit excellent linearity
within their operating range. Typical characteristics are given in Fig.
13.20. Neither Bourdon tubes nor
bellows should be allowed to expand
beyond their free length if long life is
to be assured.
While bellows have been constructed
of many types of materials, the metal
bellows must be used in high-per¬
formance, high-pressure applications.
Metal bellows may be employed in
systems with pressures up to and be¬
yond 3,000 psi. Bellows are used in
many aircraft control systems as pres¬
sure-sensitive devices that correct for £=3
to
altitude, e.g., in bombsights and fuel to
03
l_
CL
controls to set the fuel-air ratio. O
Bellows can be constructed to sense o
E
3
differential pressure, as shown in Fig. 03
C
Cl
13.21. The pressures pi and p2 are
introduced to the inside and the out¬
side, respectively, of the large bellows.
The large expanding bellows is sealed
at both ends by the two small bellows
Fig. 13.20. Bourdon tubes: (a) sche¬
contained inside it. The motion of
matic; (b) typical characteristics.
the actuating rod is then a function of
the pressure difference. The unit is more compact and has a higher
sensitivity than two separate bellows connected by mechanical linkage.

a Position depends on
T P\~Pl
Fig. 13.21. Differential-pressure bellows. (Howard)
478 CONTROL SYSTEM COMPONENTS [Chap. 13

Various metals are used in bellows construction depending on the

service factors important in the particular application. Table 13.1 sum¬
marizes present experience with the various common materials. Bellows
are constructed either with standard or extra-flexible convolutions. The
standard convolutions have hysteresis values ranging from 4 per cent for

Table 13.1. Metals Used in Bellows Construction*

Corrosion Max temp, Ease of

Metal Cost
resistance °F fabrication

Brass. Fair 350 Excellent 1

Brass, silver-clad f. Excellent 350 Excellent 3
Bronze. Fair 350 Excellent 2
Monel. Good 900 Fair 4
Stainless steel. Excellent 1100 Fair 5
Inconel. Excellent 1500 Fair 6 (high)

* J. H. Howard, Metal Bellows, Machine Design, vol. 26, no. 1, pp. 137-148, 1954.
f The silver-clad brass is not plated, but rather the solid silver is rolled on.

Table 13.2. Characteristics of 1^-in. Bellows*

Characteristic Regular Extra-flexible

Root diameter, f in. 2/1J2

Convolutions (standard) f. n 11
Effective area,§ in.2. 0.69 0.62

Wall thickness, in. 0.004 0.005 0.006 0.007 0.004 0.005 0.006
Approx, free length, || per convolu-
tion, in. 0.084 0.079 0.074 0.070 0.100 0.087 0.076
Maximum deflection,^ per convolu-
tion, in. 0.029 0.028 0.027 0.023 0.047 0.034 0.029
Spring rate,* * * § ** lb/in. per convolution
172 291 552 687 62 108 226
Flexibility, ff psi per convolution,
in/psi. 0.0040 0.0024 0.0012 0.0010 0.010 0.006 0.003
Maximum internal pressure, psi. 55 80 150 200 55 65 130
Maximum external pressure, psi.... 60 88 165 220 61 72 143

* J. H. Howard, Metal Bellows, Machine Design, vol. 26, no. 1, pp. 137-148, 1954.
f To obtain approximate inside diameter, subtract two times the wall thickness
from root diameter.
t May be decreased in every case, increased in some.
§ To obtain volume, multiply by bellows length.
|| To obtain total free length of bellows, multiply by number of convolutions.
H To obtain total maximum deflection of bellows, multiply by number of convolu¬
tions.
** To obtain spring rate of bellows, divide by number of convolutions,
ft To obtain flexibility of bellows, multiply by number of convolutions.
Sec. 13.7] pneumatic components 479

brass to 1.0 per cent for bronze. These figures are reduced by 2 by heat-
treating and by a factor of about 3 for extra-flexible convolutions.1
Design tables are available from manufacturers for the physical dimen¬
sions of all standard bellows. Table 13.2 is an example of one standard
table for lV^-in. convolutions.

E
3
E
X
O
E

c
CD
(_>

CD
QL

of
3
CO
CO
CD

c
o
CO
c
o
o

Fig. 13.22. Life expectancy of metal bellows as a function of pressure variation and
stroke. (Howard)

When bellows are subjected to large amounts of flexing or large varia¬

tions in pressure, their life expectancy is reduced. This factor must be
considered in the design. Figure 13.22 shows the life expectancy of metal
bellows in cycles as a function of pressure variation and per cent stroke.
1 J. H. Howard, Metal Bellows, Machine Design, vol. 26, no. 1, pp. 137-148, 1954.
480 CONTROL SYSTEM COMPONENTS [Chap. 13

Bellows should never be expanded beyond their free length. The figure
for variable-pressure life must be used when pressure changes, even if the
bellows remains fixed. Internal flexing takes place as the pressure
changes, even if the ends are fixed, thus reducing the life. Several com¬
mon bellows configurations are shown in Fig. 13.23.

(b) Extension-compression
stop assembly

ic) Spring-loaded assembly (d) Adjustable spring-loaded

with stops ossembly
Fig. 13.23. Common types of bellows assemblies. (Howard)

Howard1 sums up the properties of bellows as follows: flexibility, or

inches of stroke per pound per square inch of pressure, varies directly
with the number of convolutions, as the square of the outside diameter,
inversely with the cube of wall thickness, and inversely with the modulus
of elasticity of the bellows material. Spring rate, of course, is the inverse
of flexibility.
PROBLEMS
13.1. Calculate the input impedance of the transmission line and load in Prob.
10.1 for compressed air as a fluid.
13.2. Discuss the advantages and disadvantages of using compressed nitrogen as a
fluid in a pneumatic system.
13.3. A missile manufacturer is investigating the use of high-pressure superheated
steam as the driving fluid in a pneumatic system. The fluid will be stored as water
and passed along inside the skin of the missile for cooling, thus being converted to
steam. Discuss the practicality and implications of such a proposal.

1 Ibid.
 NAME INDEX

Adler, R., 1 Dow, W. G., 119, 150

Ahrendt, W. R., 150, 238 Draper, C. S., 220, 355, 356
Aikman, A. R., 447, 458, 459 Dushkes, S. Z., 420
Alexanderson, E. F. W., 198
Anner, G. E., 182, 380, 456
Edwards, M. A., 198
Egli, A., 470
Barnes, J. L., 11, 270, 442 Everitt, W. L., 182, 186, 380, 456
Beam, A., 331
Berard, S. J., 316, 319
Beyer, G. L., Jr., 229 Famme, J. H., 436
Binder, R. C., 437, 439 Firestone, F. A., 306
Blackburn, J. F., 397, 399 Fleck, J. T., 22, 34
Bode, H. W., 18, 20, 196 Formhals, W. H., 192
Bower, J. L., 21, 22, 34, 200, 251, 418 Frazier, R. H., 298
Bowman, K. K., 198 Fried, D. D., 229, 244
Bradner, M., 442
Bright, R. L., 272, 273
Brown, L. O., Jr., 289 Galonska, D. A., 333
Bull, H. S., 180 Gardner, M. F., 11, 270, 442
Burnett, J. H., 149 Gerwing, H. F., 436
Burns, L., Jr., 1 Geyer, H. M., 464
Geyger, W. A., 165, 177
Gibson, J. E., 150, 308, 389
Cahn, S. L., 420 Goldberg, E. A., 102
Caldwell, W. I., 447 Greening, C. P., 361
Carr, C. C., 8 Greenwood, I. A., Jr., 150, 259, 315
Chang, S. S. L., 302 Greer, E. M., 431
Chestnut, H., 11, 13, 16, 43, 83, 229, 230, Grohe, L. R., 356
328, 417 Guillemin, E., 18, 20, 37, 88
Chin, P. T., 130
Chubbuck, J. G., 389
Conrad, A. G., 229, 279 Hadekel, R., 424
Cook, A. L., 8 Helm, H. A., 456
Cooke, C., 431 Hetenyi, M. I., 339
Corcoran, G. F., 228 Hilbourne, R. A., 120
Cunningham, W. J., 407, 408 Holdam, J. V., Jr., 150, 259, 315
Howard, J. H., 478
Hunter, L. P., 87-89, 94, 95, 97
Davison, L. M., 317
Dawes, C. L., 180, 187, 208
Den Hartog, J. P., 346 Jackson, K. R., 389
Dodge, R. A., 377, 392, 404 James, H. M., 16, 43, 238, 265, 266, 320
Doetsch, G., 11 Janssen, J., 454, 460
Dornhoefer, W. J., 173 Jones, D. D., 120
482 CONTROL SYSTEM COMPONENTS

Kerchner, R. M., 228 Rathbone, E. H., 318

Kochenburger, R. J., 152, 216 Rawlings, A. L., 352
Koopman, R. J. W., 284, 294, 295, 299 Reed, W. E., 461
Korn, G. R., 6 Reich, H. J., 68, 81, 115, 128, 150
Kretzmer, E. R., 95 Richardson, K. I. T., 352
Kron, G., 302 Roberts, W., 1
Kronacher, G., 229 Rosenbloom, J. H., 229, 243, 244
Krummenacher, V. H., 173 Russell, W. T., 361
Kruper, A. P., 272, 273 Rutherford, C. I., 447, 458, 459

Langsdorf, A. S., 242 Schmidlin, A. E., 467

Lee, S. Y., 397, 399, 421, 422, 437, 469, Schuler, M., 352, 360
471 Scrafford, R. L., 425
Lees, S., 220 Shannon, C. E., 391
Levenstein, H., 441 Shea, R., 88, 94, 95, 121
Lloyd, T. C., 229, 279 Shearer, J. L., 422, 437, 469, 470, 471
Lyon, W. V., 284 Sinclair, D., 124
Skalnik, J. G., 129
Slaughter, D. W., 95
McDonald, D., 313 Soroka, W. W., 334
McKay, W., 220 Stanton, L., 43
MacRae, D., Jr., 150, 259, 315 Stenning, A. H., 439
Mapes, T., 389 Stolarik, E., 444
Martin, L. D., 317 Storm, H. F., 161, 177
Mayer, R. W., 11, 13, 16, 43, 83, 328, 417 Sverdrup, N. M., 377, 394
Michalec, G. W., 330 Svoboda, A., 334, 336
Miller, S. E., 74 Sweeney, D. C., 404, 405
Moyer, E. E., 130 Sziklai, G. C., 126
Murphy, R. J., 371

Thompson, M. J., 377, 392

Treseder, R. C., 464
Newton, G. C., 369
Trickey, P. H., 302
Nichols, N. B., 16, 43, 238, 265, 266, 320
Trout, W., 371
Nightingale, J. M., 419
Truxal, J., 120, 148, 268-270, 308, 312
Tustin, A., 447, 454
Tuteur, F. D., 150, 389, 418
Oldenburger, R., 456
Oliver, B. M., 391
Valley, G. E., 43, 52, 53, 115, 254
Vander Kaay, H. A., 371
Page, L., 349 Von Mises, R., 398
Parker, N. F., 361
Pestarini, J. M., 119
Peterson, A., 124 Wallman, H., 43, 52, 53, 115, 254
Peterson, B., 325, 326 Waters, E. O., 316, 319
Phelps, C. W., 316, 319 Webb, R. C., 440
Phillips, R. S., 16, 43, 238, 265, 266, 320 Weiss, G. H., 229-231, 233, 243
Pierce, J. R., 391 Whinnery, J. R., 10
Preston, S. T., 345 Williamson, H., 447
Puchstein, A. F., 229, 279 Wrigley, W., 356

Ramey, R. A., 175 Zemmerli, R., 327

Ramo, S., 10 Zweig, F., 370, 380, 382
 SUBJECT INDEX

A-c servo (see Carrier-frequency servo; Amplifiers, a-c power, plate efficiency, 116
Servomotors) power transistors, 120-123
Accelerometer, 357-359 push-pull operation, 116
Accumulator, 430 transformer coupling, 117
pneumatic, 464, 467 vacuum-tube type, 124-126
Accuracy, 200, 221 d-c, 61-111
Actuator, hydraulic, 386 cathode-temperature variation, 65
pneumatic, 476-480 chopper-stabilized, 97-103
Air compressors, 467-470 definition, 61
Ambiguity in speed synchro systems, design considerations, cathode-fol¬
240-241 lower circuits, 108, 109
American Gear Manufacturers Associa¬ difference amplifier, dynamic load
tion, 317, 328 line, 110
Amplidyne, 180, 198-206 static load line, 110
analysis, 200 triode amplifiers, 103-108
inaccuracies in, 206 with neon-tube-type coupling
armature reaction in, 187, 199 network, 106-108
compensating winding, 187 with resistor coupling network,
deadband effect, 206 103-106
definition, 198 drift, 61-80, 94, 95
direct-axis time constant, 205 ideal coupling network, 84
direct-axis voltage, 199 interstage coupling network, 80-84
effect of nonlinear brush resistance, 206 linear analysis, 64
effectiveness of direct-axis compensat¬ noise, 61
ing winding, 201 power-type, 124-126
equivalent circuit, 205 with RC networks, 52-54
field time constant, 205 transistor, 84-97
hunting, 206 (See also specific types of amplifiers)
mutual inductance, 188, 200 Analogue, mechanical to electrical, 305-
open-circuit voltage expression, 203 307
output-impedance expression, 204 Armature reaction, in Amplidyne, 187
principle of operation, 198 in d-c motors, 213
quadrature-axis brushes, sparking at, definition, 187
200 effect of, in d-c generator, 180, 187
quadrature-axis time constant, 205 in d-c tachometer, 180
quadrature-axis voltage, 199 Artificial horizon, 351, 353
speed voltage parameter, 200 Askania nozzle valve, 420
transfer function, low-frequency ap¬ Aspirator, pneumatic, 475
proximation, 204 Asymptotic diagram, 14, 20
trigger action, 206 Autosyn, 221
Amplifiers, a-c power, 115 (See also Synchros)
classes of operation, 116
equivalent circuit, 118
inductive loads with, 119 Bail ring, 347
maximum power transfer, 119 Ball, bouncing, 142
483
484 CONTROL SYSTEM COMPONENTS

Ball pump, 376 Cathode follower, phase inverter, 77, 78

Ball-screw actuator, 332, 333 with plate-load resistor, 69
Base, transistor, 85 Cathode temperature variation (see
Bellows, pneumatic, 455-457, 462, 476- Heater-voltage effect)
480 Cavitation, hydraulic, 393
common configurations, 480 Characteristic impedance of hydraulic
differential-pressure type, 477 line, 380
life expectancy, 479 Chezy flow formula, 377, 378
material, 477 Child’s law, 415
metals used, 478 Chopper, 98, 99, 265, 274
summary of properties, 480 electromechanical type, 98-99, 265-
Bernoulli force, 401, 470 266, 274
Bias battery, 81 short-circuiting type, 99
Bourdon tube, 472, 474, 476, 477 use in d-c amplifier, 64, 97-103
characteristics, 477 Chopper frequency, 100, 101
Break point, 14 Cogging, 278
Bridge, 21 Collector, transistor, 85
Bridged-T network, 43-46 Commutating poles, 187
with inductance, 53 Complementary operation in transistor
as ripple filter, 254 amplifiers, 126
use of, in carrier servo, 48 Complexity, system, 55, 56
in mechanical networks, 308-312 Compressibility, 435
Brush-contact resistance, 206 as equivalent spring, 441
Brush-shift effect in Amplidyne, 201-203 of hydraulic fluid, 368-369
Compressible flow in pipes, 439-441
Conduction period in thyratrons, 134
Cadillac valve, 427-429 Contact ratio of gears, 317
Capacitance, 20 Contraction coefficient of hydraulic line,
effect on two-terminal-network func¬ 392
tion. 26 Control transformer, synchro, 222-244
pneumatic, 445-447 Critical resistance, 193, 194
Capacitors, 89 Cylindrical rotors, 222
active zone, 26, 27, 58, 59
air dielectric variable, 8
electrolytic, dry, 9 Dashpot, 305
nonpolarized, 9 D-c amplifier (see Amplifiers)
polarized, 9 D-c generator (see Generator)
mica, 9 D-c motor (see Motors)
oil-filled, 9 D-c rate generator, 180, 181
paper, 8 D-c tachometer, 180-181
for silicon dielectric, 9 Deadband, 206, 388
strain gauge, 340 in d-c motors, 213
tubular, 8 in relays, 151, 152
Carrier-frequency servo, 48 Decibels, 16
effect of change in carrier frequency, Deltamax, 159
50-52 Demodulator, 249-270
Cascading of L sections, 24, 25, 42, 43 cathode-follower type, 261
Cathode bias, 70 clamping action, 267
Cathode-coupled inverter, 79, 80 clamping type, 266-268
Cathode follower, 25, 66-70 maximum frequency, 268
design, 108-109 peaking circuits, 268
drift, 67 describing functions, 152-154
due to component variation, 68 diode, 261-265
equivalent circuit, 69, 70 electromechanical, 265-266
gain, 67 full-wave type, 254
impedance, input, 67, 68 diode type, 265
output, 67 maximum frequency, 258
 SUBJECT INDEX 485
Demodulator, full-wave type, output E pick-off, 245-247
component, ripple, 257 Eddy current, 184-186
useful, 256 effect in d-c generator, 184-186
superiority, 257 Efficiency of worm gear, 319, 320, 333
fundamental concepts, 249 Electromechanical chopper, 98-99, 265,
half-wave type, 250-254 266, 274
analysis, 251 Electromechanical networks, 312-314
diode type, 264 Electromechanical vibrator (see Chop¬
harmonics, 252 per)
maximum frequency, 253 Electronic components, popularity, 1
output component, ripple, 252 Emitter, 85
useful, 252 Equivalent circuit, a-c servomotor, 281-
phase shift, 252 283
use of band-rejection filters, 254 cathode follower, 69, 70
inaccuracies, effect, 262 heater-voltage effect, 65
maximum frequency limit, 253, 258, push-pull amplifier, 118
268 triode, 65, 71
ring, 261 Error, random, 220, 221
sampling action, 267 repeatable, 220, 221
single-ended triode circuit, 259 Extinction point in thyratrons, 131-136
triode type, 249-261
unbalance, 262
(See also Discriminator) Faraday’s law, 161, 180
Difference amplifier, 72 Feedback, in chopper-stabilized d-c
design, 109-111 amplifier, 101-103
effect of heater-voltage variation, 73 effect on drift, 62, 63
equivalent form for, 73 in magnetic amplifier, 171-175
transistor type, 95 in relay amplifier, 150-157
vacuum-tube type, 72 Figure of merit, a-c servomotor, 291-292
Differential network, 48 d-c generator, 190-192
Differential synchros, 241-243 magnetic amplifiers, 167, 171, 172
Differentiating gear trains, 334 Fire hazard, 435
bevel-gear differential, 334 Firing angle of thyratrons, control, 127-
spur-gear differential, 336 129
effect of inductive load, 129-138
Diode, 151
Flapper valves, 418-420
free-wheeling, 135
Fliegner’s equation, 439
Discharge coefficient of hydraulic flow,
Flow through valve orifices, 392-397
393, 419
Flow separation, hydraulic, 393
Discriminator, phase-sensitive, 224, 226,
Flyweight tachometer, 341-343
231, 249-271
effect of nonlinearitities, 343
(See also Demodulator)
equations of motion, 342
Distribution factor, 229 nonsinusoidal oscillation, 343
Dither, 390 spring-restrained type, 342
Drag-cup rotor, 278-280, 298 Four-terminal networks, 18, 20-22
Drift, of cathode follower, 67-69 Fourier analysis, 228, 251, 256, 270
causes, 62 Free-wheeling circuit, 135, 136
in difference amplifier, 72-74 Frequency response from transfer func¬
effect of negative feedback, 62, 63 tion, 13
in gyroscopes, 351-353 Friction effect on d-c motor, 209
in Miller circuit, 74-77
in modulators, 273-274
in single-stage triode amplifier, 71 Gear pump, 374-376
in transistor amplifiers, 94, 95 Gear trains, 320-328
Driving-point impedance, 19, 20 backlash, 328, 330
RC, exact synthesis, 37 composite error, 328
(See also Two-terminal network) design for minimum inertia, 320-327
486 CONTROL SYSTEM COMPONENTS

Gear trains, differential gears, 334 Gyroscope, cross-coupling transfer func¬

power, 323-327 tion, 350
ratio for load matching, 327 definition of spin axis, 344
in two-speed synchro systems, 329 direct-transfer function, 350
Gears, 315-333 equation of motion, 347
backlash, 328-332 introduction, 343
ball-screw actuator, 332 linearized equations, 349
bevel gear and pinion, 318 nutation, 349, 350
Beveloid, 331-332 frequency, 349
composite error, 328-329 practical gyros, 351
contact ratio, 317 aircraft, 351
helical, 318 applications, 351
herringbone, 318 bearing friction, 351
pitch, 318 marine gyrocompasses, 352
pressure angle, 317 minimization of drift, 351
punched gear disk, 327 rate gyros, 352
rack and pinion, 318 precession in, 343-347
spring-loaded, 330 in conical surface, 348
spur gear and pinion, 315 Schuler-tuned, 352, 360
standards for fine-pitch service, 317
substitute materials, 327
tapered-tooth involute, 331 h parameters, 88-94
worm-and-gear, 318 Hagen-Poiseuille law, 378, 379, 441
efficiency, 319 Heater voltage, in difference amplifier, 73
maximum, 320 effect on drift, 62
self-locking, 319, 320 Heater-voltage effect, equivalent circuit,
Generator, d-c, 179-207 65
band width, 190 Miller circuit, 74-77
figure of merit, 190, 191 Helical gear, 316, 318
power gain, 190 Herringbone gear, 318
response speed, 190 High-pass filter, 17
Goldberg amplifier, 102, 103 Holes, 85
Grooves in hydraulic control-valve pis¬ Hooke joint, 337-339
tons, 405 Hydraulic capacitance, 381
Gyrocompass, 355-359 Hydraulic control systems, 363-431
Gyroscope, 343-361 actuator, 386
applications, 353 unequal-area type, 387
free gyro, 353 analogy to Child’s law, 414
gyroscopic velocimeter, 353, 354 axial reaction force, 397
to inertial navigation, 355-361 transient, 399, 400
control of stable platform, 357 basic types, 363
east-west errors, 358 bridge analogue for underlapped valve,
effect of initial errors, 360 413
equations of motion of stable plat¬ bulk modulus, 369
form, 360 cavitation, 393
HIG gyro, 355-359 choice of operating pressure for, 431
north-south errors, 358 compressibility coefficient, 369
simplified system block diagram, compressibility flow, 368, 369
359 constant-flow system, 364
transfer function, 349, 350, 355, constant-pressure system, 364
356 control valves, axial reaction force,
use of accelerometer with, 357 397-402
rate gyro, 354 flapper, 418
restrained gyro, 353-355 graphical analysis, 405-412
approximate block-diagram represen¬ lap, 388
tation, 348 maximum power output, 407, 408
basic torque equation, 344 nozzle type, 420
 SUBJECT INDEX 487
Hydraulic control systems, control Hydraulic control systems, two-stage
valves, pressure-flow relations, valves, feedback to remove inte¬
392-397 gration, 423
radial forces, 402-405 feedback links, 424
slide, 420, 422 Moog, 425-427
small-signal linear analysis, 412-418 Siemen’s, 424
spool type, 387-418 valve controlled, 364, 365, 386-434
two-stage, 423-429 Bernoulli force, 390, 401
definition, 343 damping, 401
equation of flow in underlapped region, flow-pressure relations, 392
396 force-compensated, 402
flapper valve, 418 graphical analysis, 406
balanced, 420 introduction to, 386
hydraulic accumulator, 430 pulsed operation, 389
bellows type, 430 advantages, 389-390
hydropneumatic type, 430 disadvantages, 390
hydraulic flow separation, 393 spool-type valves, 387
leakage flow, 368 four-way, 388
linearized equivalent, for flow-control two-way, 387
valve, 416 zero-lap, 388
stability, 417 valve lock, 390, 402
linearized small-signal analysis, 412 Venturi tube, 394
lock, 402-405 Hysteresis, magnetic, 181, 182, 184
maximum load power, 408 effect on phase lag, 184
motor, 366-376 in relays, 152
flow, 368 Hysteresis curve in Amplidynes, 206
pressure-regulating devices, 429-430
pump-controlled, 364, 365-367
analysis 367-370 Impedance, 13, 24, 25
transfer function, 370 characteristic, of hydraulic lines, 380-
pump flow, 368 382
pumps, 364-376 Impedance analogue, 306
ball, 276 Impedance level in networks, 30, 34, 36,
comparison of types, 370-376 41
gear, 374-376 Impedance ratio in cascading of L sec¬
piston, 365-372 tions, 24, 42, 43
vane, 372-374 Inductance, of armature of d-c generator,
variable-stroke, 366-372 187
reaction force, 397-405 d-c motor armature effect in thyratron
radial, 402 amplifiers, 142-144
relief protection, 366 effect, on realizable transfer function of
replenishing system, 366 networks, 21, 22, 53
slide valve, 420-422 in thyratron amplifier, 129-138, 142-
stroke amplifier, 386 144
transmission, 367-370 effective, in magnetic amplifiers, 166
transmission lines, 376 Inductor, 1, 10, 11, 131
absolute viscosity, 379 air-cored, 10
characteristic impedance, 380 iron-cored, characteristics, 10
Chezy flow formula, 377 never use, 11
coefficient, of energy loss, 378 quality factor, 10
of friction, 378 Inertial navigation, 351, 355-361
kinematic viscosity, 379 Infant mortality, 56
Reynolds number, 377 Input impedance of networks, 24, 25
turbulent flow, 377 cathode-follower, 67, 68
viscous or laminar flow, 376 transducers, 221
two-stage valves, 423 transistor-amplifier, 91-93
Cadillac, 427-429 triode-amplifier, 72
488 CONTROL SYSTEM COMPONENTS

Integral controller, 17 Magnetic amplifier, Ramey, 175-177

Interstage coupling networks for d-c shared-time principle, 177
amplifiers, 80-84 self-saturated, 173-175
gain, current, 174
d-c power, 175
Kirchhoff’s law, 183, 214 voltage, 174
Kirchhoff’s mesh equation, 201 time constant, for falling transients
175
for rising transients, 175
L sections, 22-43 series-connected type, 159-167
approximate synthesis procedure, figure of merit, 167
26-33 gain, a-c power, 165
exact synthesis, general method, 34-43 current, 164
simple transfer function, 33, 34 d-c power, 164
number required in ladder network, 25, voltage, 164
26 time constant, 165
Ladder networks, 24 use of special alloy, 159
approximate synthesis, 25-26 Magnetization curve, 182, 193
Lag-lead network, 17 Mass, 305, 308
Lag network, 17 Mechanical filters, 305-315
Lattice network, 21 Metadyne, 199
Lead network, 17, 48 Microsyn, 247, 248
Lenz’s law, 185
Miller circuit, 74-77
Level changer, 83
effect of change in positive supply volt
Linearity, best, 3
age, 77
independent, 3
equivalent circuit for, 75
normal, 3
practical form, 76
of variable resistors, 3
Mobility, analogue, 305-315
zero-based, 3
Modular packing, 57
Load line in hydraulic valves, 406
Logarithmic representation of transfer Moog valve, 425-427
function, 14 Motors, d-c, 207-218
Low-pass filter, 17 analysis inaccuracies, 213
armature reaction, 213
brush-contact resistance, 213
Mach number, 440 deadband, 213
Magnesyn, 248 electrical damping, 209, 215, 217
Magnetic amplifier, 159-177 equivalent circuit, 212
with feedback control, 171-175 field-controlled type, 216-218
gate winding, 159 gain constant, 218
hysteresis loop, 159 mechanical output impedance, 209
normally excited, 161 quasi-linear transfer function, 211
parallel-connected reactors, 167-171 relation, between constants, 208
figure of merit, 171 between transfer function and
gain, a-c power, 169 speed-torque curves, 210
current, 169 separate excitation, 207-213
d-c power, 169 speed regulation, 142
voltage, 169 split-field series type, 213-218
time constant, 169-170 thyratron-controlled, 139-149
with positive feedback, 171 time constant, 210
figure of merit, 172 pneumatic, 474-476
gain, current, 171 gear type, 475
d-c power, 172 piston type, 476
voltage, 171 diesel action in, 476
time constant, for falling transients, vane type, 475
172 Mutual inductance in d-c generators,
for rising transients, 172 187-190
 SUBJECT INDEX 489
Negative feedback effect on drift, 62, 63 Pneumatic components, weight com¬
Neon tubes for interstage coupling in d-c parison with electric and hydraulic,
amplifiers, 81, 106-108 463
Node method, 11 Pneumatic compressor, 467
Nonbleed relay, 474 advantages of multistage compression,
Nonbleed valve, 474 468
Normal excitation, 161 Pneumatic-control functions, 443
Normal law, 220 Pneumatic-control valve, 470-474
Null networks, 43-52 effect of viscosity, 470
Nutation of gyroscope, 349, 350 two-stage flapper-control pressure
Nylon used for gears, 327 valve, 473, 474
Nyquist diagram, 153-154 underlap characteristic, 472
Pneumatic-electric analogues, 443-447
Pneumatic equalizer, 463
Operating pressure of pneumatic sys¬ Pneumatic-equalizer networks, 447-454,
tems, 467 463
Orifice coefficient, 439 Pneumatic flapper-type control valves,
Oscillations in thyratron circuits, 150 453, 454, 456, 470, 473, 474
Output impedance, Amplidyne, 202-205 Pneumatic flow, 437-439
cathode follower, 67 adiabatic, 437
d-c generator, 186-190 compressible, in pipes, 439
mechanical, in d-c motor, 209 Mach number, 440
networks, 24, 25 pipe friction factor, 440
transistor amplifier, 91-93 relation for, 439
triode amplifier, 71 Reynolds number, 440
Overlapped valves, 388 critical-pressure ratio, 439
general energy equation, 437
through orifice, 438
Peaking circuits, 268 Pneumatic jet-engine control servo, 461-
Phase inverter, 77-80, 124, 125 463
balanced, 78 Pneumatic motors (see Motors)
cathode-coupled type, 79 Pneumatic orifices, 438, 439
cathode-follower type, 77 Pneumatic relay, 455
paraphase type, 78 Pneumatic resistance, 444-447
Phase-sensitive detector (see Discrimin¬ Pneumatic spool-type control valves,
ator) 454-456, 470-474
Pickup device, 220, 221 Pneumatic systems, 435-480
Pinion, 315-317, 322 advantage over hydraulic system, 435
Piston pumps, 365-372 choice of operating pressure for, 467
Pitch factor, 229 disadvantages as compared to hy¬
Pneumatic actuator, 476-480 draulic, 436
differential area, 476 explosion hazard, 436, 465, 467
linear, 476 introduction to, 435
spring-loaded, 476 response speeds, 437
(See also Bellows; Bourdon tube) Pneumatic transmission lag, 441
Pneumatic amplifier, 455 first-order approximation for, 442
Pneumatic aspirator, 475 Pneumatic valve, equivalent circuit, 456
Pneumatic bellows, 276-280, 455-457, Polarized relay, 216
462 Pole face windings, 187
Pneumatic capacitance, 444-447 Poles, effect of successive, 32, 33, 36
Pneumatic components, 461-480 of network function, 19-25
interchangeable with hydraulic, 461 networks with complex, 52-54
introduction to, 461 restrictions on, 21-25
power supplies, 467-468 Positive feedback in magnetic amplifiers,
advantages of multistage compres¬ 171-173
sion, 468 Potentiometer (see Resistors, variable)
weight advantage over hydraulic, 465 Potentiometer slide wire, 5
490 CONTROL SYSTEM COMPONENTS

Power amplifiers, d-c, 123-126 Relay amplifiers, frequency response,

single-ended push-pull, 124, 125 156-158
transistor, 125 hysteresis, 152
Precession, 343-347 output pulse-repetition rate, 156
Precision, 220, 221 polarized relay, 151, 213
Pressure-regulating devices, 429, 430 stability, 152-154
Propagation constant of hydraulic line, static characteristic, 154-156
380 Reliability, 54-57
Proportional plus rate network, 17 Relief valve, 366, 367
Proportional plus rate plus integral con¬ Replenishing system, 366, 367
troller, 17 Reset network, 17
Proportional plus reset network, 17 Residual voltages in synchros, 225, 226
Pulse-length modulation, 389-392 Resistance, 20
Pulsed operation of hydraulic valves, as interstage coupling in d-c amplifiers,
389-392 83, 104-106
Push-pull amplifier, ideal, 126 pneumatic, 444-447
strain gauge, 340-341
temperature coefficient, 7, 8
Q in inductor 10, 11 Resistance-capacitance networks (see RC
Quadrature in strain gauges, 341 networks)
Quadrature circuit in Amplidyne, 199- Resistance padding, 6, 7
206 Resistors, 1-8
Quadrature voltage in synchros, 225- carbon, 1
226 color code, 1
high-resistance wire, 1
linearity, 3
Radio Manufacturers Association, 2 precision variable, 3
Ramey magnetic amplifier, 175-177 mechanical characteristic, 5
Rate generator, 180-181 nonlinear, 6, 7
Rate gyro, 354, 355 resistance padding, 6
RC networks, 11-54 voltage padding, 6
amplifiers, 52-54 slide-wire, 5
use of, 52 standard power ratings, 1
bridged-T, 43 standard tolerances, 1
five simple, 17 standard values, 2
four-terminal, 20 variable, 2-7
general properties, 18 gauging, 6
L section, 22 linearity, 3, 4
approximate synthesis, 26 moment of inertia, 5
cascading, 24 resolution, 4
exact-synthesis method, 33-34 starting torque, 5
ladder structure, 22 Resolution, 4
lattice structure, 22 Resonance, 15
synthesis, 17-43 Reynolds number, 377, 378, 440
introduction to, 17 Ripple, 180, 252-262
table of common types, 17 commutator, 181, 300
twin-T, 43 in discriminator output, 252-262
two-terminal, properties, 19 Ripple filters, 253-259
RC rate network, 11, 17 Root form of transfer function, 12
Reference voltage, 224, 249-250, 276 Rosenberg generator, 199
Regulex, 180, 197-198, 207 Rotors for synchros, 222
Rejection amplifier, 53 Rototrol, 180, 192-197, 207
Rejection network (see Twin-T circuits) analysis, 192
Relay amplifiers, 150-159 critical resistance, 193
chatter rate, 156 critically tuned, 194
deadband, 151 definition, 192
describing function, 152 effect of eddy currents in, 195
 SUBJECT INDEX 491
Rototrol, with multiple-control windings, Slide valve, 420-422
197 Slip, 281
nonminimum phase transfer, 196 Slot lock, 278
pilot generator, 192 Smith chart use with hydraulic trans¬
transfer function, 195 mission, 380
Solid iron rotor, 287
Space harmonics, in a-c servomotors, 277
Salient pole rotor in synchros, 222, 227, in synchro air-gap flux, 229
228 Sparking, 187, 199, 200
Sampled-data analysis, 270 Spring, 305
Saturation curve, 181 Spur gear, 315-317
Schuler tuning of gyroscopes, 352, 360 Square-wave modulator, advantages, 258
Self-saturation in magnetic amplifiers, transistor circuit, 272
173-175 Square waves, 251-274
Selsyn, 221 Squirrel-cage rotor, 278
(See also Synchros) Stability of relay amplifiers, 152-154
Separation principle of poles and zeros, 20 Stable platform, 357-359
Servomotors, a-c, 49, 276-303 Stator of synchros, 222
advantages, 277 Stick-off voltage, 240, 241
approximate transfer function, 288 Strain gauge, 339-341
characteristics with finite control im¬ Supermalloy, 159
pedance, 292 Suppressed-carrier amplitude modula¬
cogging, 278 tion, 224, 249, 251, 271
construction features, 277 Symmetrical components in analysis of
control by thyratrons, 149 a-c servomotor, 284-285
cross-field theory, 299 Synchro capacitors, 243
double-revolving-field theory, 299 Synchros, 49, 115, 220-244
effect of unbalanced stator winding, classification, 222
294 construction, 222
efficiency, 277 control transformer, 222-244
equivalent circuit, 282 differential units, 241-243
figures of merit, 29 elementary operation, 221
impedance in control circuit, 292- generator, 222-244
294 harmonic voltage component, 226
as phase discriminator, 277, 286 iV-speed systems, 235
power range, 277 one-speed systems, 235
resonating capacitor, 277 output-signal form, 49, 223, 224
rotor types, 278 quadrature error, 225-226
shaded-pole induction motor, 300- repeater, 222, 243-244
303 static errors, 225, 227-229
single-phasing, 290, 296-298 suppressed-carrier amplitude modula¬
slip, 284 tion, 224
slot lock, 278 synchro capacitor, 243
speed-torque curves, 288-295 synchronous velocity, 234
starting torque, 286 torque gradient, 244
symmetrical components, 284 two-speed systems, 235-241
synchronous speed, 281 diode-clipper switching circuit, 237
theory of operation, 279 false point synchronization, 238
use, 276-277 neon-tube switching circuit, 236
as tachometer, 298 oscillation, 238
Shaded-pole a-c motor, 300-303 slewing characteristics, 237
Shared-time principle, 177 stick-off voltage, 240
Single phasing of a-c servomotors, 290, switching circuits, 236-241
296-298 velocity error, 227, 230-235
Skin effect in a-c servomotor, 286-287 vibration damper, 243
Sleeve valve, 424 Synchronization, 236
Slewing, 237 Sziklai amplifier, 126
492 CONTROL SYSTEM COMPONENTS

Tachometer, a-c type, 298-300 Transducers, 220

armature reaction, 180 accuracy, 221
brush bounce, 180 errors, 220
d-c type, 180-181 precision, 221
flyweight, 341-343 variable reluctance, 245
ripple component, 180 Transfer function, a-c servomotor, 288-
Telegon, 248 291
Temperature coefficient of resistance, 7, 8 all-pass type, 54
Temperature effect, on thynatron con¬ Amplidyne, 203-205
duction, 127 with complex zero, 44-46
on transistors, 84, 87, 94 d-c generator, 189, 190
Thermal runaway, 94 definition, 11
Thermocouple, 97 field-controlled d-c motor, 217, 218
Thevenin circuit, 182, 186, 207 frequency characteristic, 23
of Amplidyne, 202-205 gyroscope, 349, 350, 355, 356
of d-c motor, 207 hydraulic transmission, 370
equivalent mechanical, 209 with imaginary zeros, 46-52
Three-terminal networks, 18, 20 nonminimum phase, 53
Thyratron, conduction, 126 pneumatic bellows, 455-457
control with, of a-c motors, 149 RC networks with amplifiers, 52-54
of firing angle, 127-129 realizable, with four-terminal RC net¬
of split-field series motor, 149 work, 20
grid potential, 126, 127 root form, 12
operation, 126 Rototrol, 194-196
positive ion sheath, 126 separately excited d-c motor, 209-212
tube drop, 127, 132-135 split-field series motor, 215
Thyratron amplifier, 126-150 steady-state frequency response from,
continuous load current, 136-137 13
with d-c motor load, 139-149 synchro, 233-235
bidirectional control, 145-148 thyratron amplifier, 138
extinction angle, 131-143 quasi-linear, 138, 142
firing angle, 128, 133, 134, 135, 140 time-constant form, 12
free-wheeling circuit, 135 Transient hydraulic-reaction force, 399-
full-wave, 136, 137 401
half-cycle response time, 138 Transistor amplifiers, complementary
with inductive loads, 129-138 operation, 126
one-cycle response time, 138 drift, 94, 95
reactance tube circuit, 129 power-type, 120-123
transfer function, 138 Transistor circuits, difference amplifier,
tube drop, 140 95
Time constant, Amplidyne, 205 grounded base, 86, 91-93, 120
d-c motor-controlled thyratron, 148 grounded collector, 91-93, 121
magnetic amplifiers with feedback, grounded emitter, 87-93, 121
172 thermal runaway, 94, 120
motor inertia, 210 Transistors, 84-97, 120-123, 272-273
parallel-connected magnetic amplifier, action defined, 85
169-171 analysis of simple circuits using, 85-90
self-saturated magnetic amplifier, 175 base, 85
series magnetic amplifier, 165-167 collector, 85
Time-constant form of transfer function, complementary amplifier, 125, 126
12 effect of temperature on, 84, 87, 94
Torque, a-c servomotor, 49, 283-298 emitter, 85
Torque gradient in synchro repeater, equivalent circuits, 86
244 h parameters, 88-94
Torque to inertia ratio of a-c servo¬ modified, 93
motors, 279, 291 holes in, 85
Torque motor, hydraulic, 390, 429 junction, 85
 SUBJECT INDEX 493
Transistors, modulators, 272, 273 Vacuum tubes, reliability, 55, 57
multistage amplifiers, 95-97 Valve-controlled hydraulic systems (see
single-stage circuits, 90-94 Hydraulic control systems)
stability factor, 94 Valve lap, 388
temperature dependence of parameters Valves, hydraulic (see Hydraulic control
in, 87 systems, control valves)
temperature-sensitive elements for Vane pumps, 372-374
compensation, 95 Variable-reluctance transducers, 245-248
use of, in a-c power amplifiers, 120-123 Variable-reluctance transformer, 247
in d-c amplifier, 84 Velocity coefficient of hydraulic flow, 392
in d-c power amplifier, 125 Velocity effect on synchro gain, 234, 235
Triode, as cathode resistor, 80 Velocity error, 227, 230-235
characteristic curves, 64 Velocity propagation in hydraulic lines,
equivalent circuit, 65 380
Triode circuits, 64-80 Vena contracta, 298
single-stage amplifier, 70-72 Venturi tube, 394
drift, 71 Vibration damper in synchro repeaters,
gain, 71 243
grounded-grid amplifier, 72 Virtual cathode, 65
output impedance, 71 Viscosity, absolute, 379
Twin-T circuits, use, with amplifiers in kinematic, 379
generation of complex pole-trans¬ Voltage-divider principle of obtaining
fer functions, 52-54 transfer function, 12, 13, 22
in carrier servo, 48 Voltage padding, 6, 7
in mechanical network, 308, 312
as series equalizer in a-c servos, 48-
52 Ward Leonard system, 179, 207, 216
Two-terminal network, 18, 19 Wear-out region, 56
effect of capacitors, 26 Wobble plate, 365
exact synthesis, 37, 38 Worm gear, 318-322
properties, 19, 20
Two-terminal RC network, approximate
synthesis method, 27, 28 Zener diode, 82-83, 97
Zero-lapped hydraulic valves, 388, 406-
412
Umbrella rotor, 222 Zeros, complex, network with, 44-52
Underlapped valves, 388, 395-397 of network function, 19, 20, 22-25
Unequal-area hydraulic actuator, 387 purely imaginary, network with, 44,
Universal joint, 337-339 46-49
velocity error, 339 restrictions on, 22-25