Electronics

Circuit Designer's Casebook

Published by Electronics -1221 Avenue of the Americas, New York, N.Y. 10020
McGraw-Hill Books for Electronics Engineers
CHARACTERIZATION OF SOURCEBOOK OF ELECTRONIC CIRCUITS
SEMICONDUCTOR MATERIALS By JOHN MARKUS, Manager, Information Research,
McGraw-Hill Inc.
By PHILIP F. KANE, Director, and 896 pages, 81/4 x 11, over 3000 illustrations
GRAYDON F. LARRABEE, Manager, Compositional
Methods Branch, both of Central Analysis & Characterization A virtual desk -top retrieval center for engineers, designers,
Laboratory, Texas Instruments, Inc. and technicians in all areas of electronics, this first book
333 pages, 221 illustrations of its kind contains over 3,000 electronic circuits, complete
Requiring no special knowledge of semiconductor technology, with values of all components. A distillation of circuit
this comprehensive one -of -a -kind volume brings you a wealth knowledge for the past ten years, it brings together ma-
of practical information for the evaluation and application terial from many diverse sources. The majority of circuits
of semiconductor materials and characterization techniques. shown are recent semiconductor designs, but important
Geared to meet the needs of all professionals, this unique electron -tube circuits are 41equately represented. Included
guide provides a brief account of semiconductor develop- with each circuit is a concise description of its significant
ment, then presents a concise nonmathematical analysis of features, performance data, and operating characteristics
the band theory. Semiconductor principles, bulk material so you can find the circuit that most closely meets your
characterization, single -crystal growth, chemical and physical needs in any practical, on-the-job situation. It also tells
imperfections in single crystals, epitaxial films, diffusion, and you where to get further information on each circuit.
thin films are among other topics discussed.

HANDBOOK OF SEMICONDUCTOR
THE NEW ELECTRONICS ELECTRONICS, Third Edition
By BRUCE SHORE, Administrator of Scientific Information, Edited by LLOYD P. HUNTER, Professor of Electrical
Radio Corporation of America Engineering, University of Rochester, and prepared by
254 pages, 85 photographs 17 top specialists
Today's solid-state revolution-from the discovery of the 1,100 pages, 963 illustrations and tables
transistor in 1948 - is discussed in this "how -it -works -and - Everything you need to know about semiconductor circuitry
where -it -came -from" guide. It takes you into the laboratories is clearly explained and illustrated in this practical book. A
where the great discoveries were and are being made, and now -famous work, it brings you step-by-step guidance in
explains in everyday language exactly what is happening. device physics, fabrication techniques, evaluation methods,
Completely up to date, the book offers you a rapid under- and circuit design techniques in a single volume. Written
standing of solid-state electronics and the phenomena asso- by specialists who were chosen for their proven knowledge,
ciated with it - shown not only in all their beauty and spirit ability, and experience, the book includes all recent advances
of adventure, but in their growing centrality to industrial and to keep you abreast of today's rapidly developing integrated
individual life. Illustrated with scores of photographs, the circuit techniques. In this new edition, a section has been
book offers full historical background from the ancient added on the design of integrated operational amplifiers, and
Greeks to the dedicated men and women involved in the also a section dealing with the utilization of computers. Page
field today. after page of illustrations and tables offer visual clarity to
every key aspect of semiconductor circuitry, and the entire
book has been completely expanded, updated and revised.

AN ILLUSTRATED GUIDE INTEGRATED CIRCUITS

TO LINEAR PROGRAMMING Design Principles and Application
By SAUL I. GASS, Vice President, World Systems Prepared by THE ENGINEERING STAFF,
Laboratories, inc. MOTOROLA, INC., Semiconductor -Products Division
224 pages, with charts, graphs, and drawings 385 pages, 71/4 x 97/s, 298 illustrations
In this excellent nontechnical introduction to an important With the help of this authoritative guide you can easily
new tool for solving business problems, the author draws on apply the valuable know-how of over twenty outstanding
a wide variety of actual situations to show you just exactly circuit design engineers to your own daily design problems.
what linear programming is, what it can do, how it is applied, It gives a unified treatment to all phases of integrated cir-
and what kinds of problems it can handle for you. The search cuits, clearly and precisely defining every important point
for the best solution - the optimum solution - has intrigued from initial material processing to device packaging and
man throughout the ages, and this new, completely authorita-
tive but thoroughly entertaining new book offers you all reliability testing. You'll find in it detailed information on
essentials of a key element in management today. Moving such subjects as acceptor doping . . hole -electron spin
.

logically from the simple (getting dressed each day) to the assymmetry ... recombination and generation ... the gaus-
most sophisticated levels of linear programming, the book sian distribution . . . diffusion time . sheet resistance
. .

is an admirable springboard for more advanced work and ... and much more.
study in the entire fascinating field.

SOLID-STATE ELECTRONICS HANDBOOK OF THIN FILM TECHNOLOGY

A Basic Course for Engineers and Technicians Edited by LEON I. MAISSEL, Manager, Thin Film Basic
Studies, and REINHARD CLANG, Manager, Device
By ROBERT G. HIBBERD, Serniconductor-Components Process Studies, both IBM Components Division,
Division, Texas Instruments Incorporated East Fishkill, New York
164 pages, 7 x 10, 90 illustrations 1,216 pages, 887 illustrations plus charts and tables
This unique work presents the principles of semicon- This all-inclusive reference work offers you today's latest and
ductors in an unusual way. It starts with a description most dependable source of guidance on thin film theory and
of semiconductors and their properties, the p -n junction practice. The volume serves both as an introduction to the
and the junction transistors, and the characteristics of field and a practical day-to-day guide for scientists engaged
transistors and basic transistor amplifier circuits. It then in thin film work. Profusely illustrated, it gives you easy
access to all information you need, helping you to catch up
describes the manufacture of transistors and other semi- with the fast-moving field. Each of the 23 detailed sections
conductor materials and devices. The book also discusses has been written by a recognized authority who covers the
the whole family of semiconductor devices and examines subject areas he knows best. Divided into four sections: Prep-
the applications of integrated circuits. Down to earth in aration of Thin Films, The Nature of Thin Films, Properties
presentation, it can be used by non -technical readers to of Thin Films, and Applications of Thin Films (this latter
obtain a working familiarity with the subject. section with particular emphasis on thin films in the elec-
tronics industry).

See additional titles and ordering information on back cover

Electronics.
Circuit Designer's
Casebook
Published by
Electronics
A McGraw-Hill Publication

Price $4.00, Printed in U.S.A.

All material copyrighted by
Electronics
McGraw-Hill, Inc.
1221 Avenue of the Americas,
New York, New York 10020
 Table of Contents Series -connected op amps null offset voltage
Op amps multiply RC time constants
21
25
MEASUREMENT Controlling op -amp gain with one potentiometer 48
Nomograph determines aperture time error Op amp with feedback makes full -wave rectifier 56
1

Differentiate and count to find frequency error 14

Complementary output stage improves op -amp
response 60
Logic system checks out analog -to -digital converter 38
Logic probe with LED display checks ECL circuits. 46
Approximating true log output at high frequencies 81
Converter for oscilloscope provides four -channel Doubling op amp summing power 84
displays Instrumentation amplifier conditions computer
46
inputs 119
TTL gates speed up pulse -height analysis 50
Feedback linearizes resistance bridge Dynamic zero -correction method suppresses
64
Soldering iron converts to constant -temperature offset error in op amps 123
probe 65
Precision auto tachometer squelches point bounce 76 CONTROL
SCR reset for integrator provides high speed 2
POWER SUPPLIES Feedback current switch divides rf inputs by 20 6
Stable voltage reference uses single power supply 16 Fast -switching modulator reverses uhf signal phase 16
Filament transformer output drops cost of 400 -Hz Linear signal limiting with feedback multiplier 19
supply 24 Height -to -width converter digitizes analog samples 23
Regulating high -voltage with low -voltage transistors 25 Control one-shot divides frequency by up to 30 28
Pulsed standby battery saves MOS memory data 28 Series resistance improves potentiometer linearity 32
Four -ampere power supply costs just $13 to build 74 Amplitude modulator is highly linear 35
FETs remove transients from audio squelch circuit 36
Diode plus low-cost op amp makes accurate
FILTERS thermostat 37
Tunable active filter has controllable high Q 7 Broadband cutoff limiter is phase -transparent 41
Wien bridge in notch filter gives 60 dB rejection 54 Frequency doubler accepts any waveshape 43
Active filter has separate band and frequency Op amp cancels video switching transients 44
controls 57 Precision integrator resets as it samples 52
Voltage -tuned filter varies center frequency linearly 69 Logic circuit converts synchronous motor to
Active resonators save steps in designing active stepper 58
filters 90 Quasi -matched MOSFETs form filterless squaring
Active filters get more of the action 95 circuit 63
Tables shorten design time for active filters 113 Gated MOSFET acts as multiplexing switch 67
Stepper drive circuit boosts motor torque 68
Temperature -stable decoder for modulated pulse
GENERATION widths 70
MOSFET network minimizes audio oscillator Electronic fuel injection reduces automotive
distortion 3 pollution 104
Bootstrapped capacitor stabilizes UJT oscillator 4
Op amps generate precision staircase 8
Stable square -wave generator provides broad COMPARATORS AND DETECTORS
bandwidth 10 VSWR detector protects class C rf amplifiers 11
ECL gates stretch oscillator range 18 One-shot/flip-flop pairs detect frequency bands 27
Triangular -wave generator spans eight decades 30 Analog voltage sensor controls LED threshold 51
Output comparator enhances versatility of one-shot 31 Photodetector senses motion in noisy surroundings 62
Thumbwheel switches set synthesizer output Agc rf threshold detector provides fast slewing 77
frequency 34 Phase comparator for servo loops 78
Double -duty multivibrator gives complementary
outputs 39
LOGIC
Preset generator produces desired number of
pulses 42 Schottky diodes eliminate two -level NAND gating 9
Pulse generator accuracy is immune to aging 45 Two adders form BCD subtractor 11
Multivibrator clock obeys digital commands 55 Feedback latch reduces memory recovery time 15
Generator independently varies pulse rate and Simple logic circuits compare binary numbers 20
width 61
Unclocked logic element makes quick decisions 22
Exclusive -OR gate makes bidirectional one-shot. 66 Circumventing BCD addition in digital
C-MOS sums up tones for electronic organ 70 phase -locked loops 33
Staircase generator resists output drift 73 Schmitt trigger prevents clock train overlap 40
Binary division produces harmonic frequencies.. 74 Logic driving gates double as d -a converter
One-shot timing performance: don't take it for switches 49
granted 87 Wired -OR gates increase multiplexer input capacity 52
The integrated Schmitt trigger: a versatile design Gray -code generator avoids output glitches 56
component 101
Ring counter eliminates false gating signals 58
Transistor gating circuit cuts signal delay to 100 ps 72
Diode switching matrices make a comeback 79
AMPLIFIERS Take a bit of advice: use 16 -bit converters carefully 109
Transducer preamplifier conserves quiescent power 13
COMPUTER -AIDED DESIGN digital circuits 143
Brief Fortran program for active low-pass filters 4 Computer helps design of complementary MOS
Programs in Getel speak test engineer's language 125 logic 147
Sceptre rules benevolently over computer -aided Aedcap: the circuit designer's computer -assisted
design 130 slide rule 152
Optinet guides electronic products from design Easy -to -use Hypi program makes possible transition
through marketing 136 from nonlinear to linear transistor model 157
Program calculates load effects of high-speed
Foreword
Ask any electronics engineer if he could use another filter
to work with and, after mentally ticking off such types as
constant -k, m -derived, Chebyshev, Butterworth, active, L -C,
and a host of others, he'd probably say: "Thanks, but no
thanks. have enough problems building the ones I know
I

about." But there is another kind of filter every engineer

needs desperately-an information filter.
Probably no one works closer to ground zero in the in-
formation explosion than the electronics engineer. How
can he recognize and save the published material that's
of lasting value, but only note and then discard the current
news information that has a built-in obsolescence factor?
Making these evaluations under the real-time pressures of
completing a design project isn't easy.
Books like this one, we feel, will help him assemble a
readily accessible storehouse of design information with
a long shelf life. In this case, we have done the filtering.
This book is a collection of recently published basic,
how-to articles from the technical article section of Elec-
tronics, combined with brief circuit ideas from the De-
signer's Casebook section. It can be used both as an
up-to-date practical design textbook and as an idea book.

The editors of Electronics

 The nomograph shown is based on this relationship,
and assumes a full-scale input voltage swing at the fre-
Nomograph determines quency of interest. If the signal is less than full scale, the
aperture time error error will be smaller. When the only limit on the input
signal rate of change is slew rate limiting, dv/dt is easily
by Stephen Muth computed and the nomograph is not necessary.
ILC Data Device Corp., Hicksville, N.Y. As an example, suppose the maximum allowable er-
ror for aperture time is 0.5% and the upper frequency
limit is 2.6 megahertz. Use a straight -edge to intersect
A simple nomograph does away with the tedium of cal- the frequency axis on the bottom and the error axis in
culating the error due to aperture time for a sample - the center. Maximum permissible aperture time is read
and -hold circuit. It can also be used to compute the con- from the top axis -300 picoseconds. (The present limit
version error of an analog -to -digital converter which for aperture time in high -accuracy -0.1% linearity-
has a varying dc signal or a low -frequency ac signal as sample-and-hold modules is about 100 ps.)
its input. To find the error of an a/d converter with a conver-
Being a measure of the repeatability of the analog sion time of 22 microseconds and a dc input varying at
sampling switch characteristic, aperture time reflects the 20 hertz maximum, treat conversion time as aperture
uncertainty of when exactly the switch opens. Errors time and signal rate variation as frequency. Use of
due to it vary with the signal rate of change at the straight -edge and nomograph shows the maximum er-
sample point. The voltage that is sampled and held can ror to be 0.3%. It should be noted that the error of the
have an error of: a/d converter can be reduced by two orders of magni-
E = (dv/dt)T tude by employing a sample -and -hold module to keep
where E is the voltage change or error, dv/dt the max- the converter's input constant for its full 22-µs conver-
imum signal rate of change, and T the aperture time. sion time.

Computation aid. Nomograph solves equation for percent -of -full-scale error dye to aperture time of sample -and -hold circuit. Error, E, can
be written in terms of signal rate, dv/dt, and aperture time, T: E = (dv/dt)T. Using straight -edge to intersect all three axes gives the answer.
(Signal rate is represented by frequency axis.) Conversion error of a/d convertor can also be found by using the same technique.

+6V

_L. 0.01 pF 100 S2 - 100 p F

25µF _ 100 p F
TEXAS
7 INSTRUMENTS 12
OUTPUT
SN76001 4 2 V PK-PK
3
- 0.05 p F
CI -
0.5 pF 82 pF
R2 3.3 ki2
25 A+ F

1N914
1 k2
3N128
ADJUST FOR DESIRED
OUTPUT LEVEL
100 k2
 The reset circuit consists of cross -coupled NAND gates
that form an R -S flip-flop and another NAND gate, con-
SCR reset for integrator nected in a 1 -microsecond -delay inverter configuration,
that keeps the flip-flop's Q output normally low. The in-
provides high speed tegrator output is fed through a forward -biased diode to
by Marshall W. Williams the base of a transistor, allowing the transistor to turn
University of Georgia, Athens, Ga. on only when the integrator output reaches approxi-
mately 1.4 volts.
When the transistor conducts, its collector goes low,
A fast -switching three -gate reset circuit allows steady or applying a negative -going transition to the flip-flop's S
varying dc voltage to be converted into TTL-compatible input. The Q output of the flip-flop now goes high, turn-
pulses for driving counters or other data storage or data ing on the SCR gate and producing a negative -going
processing devices. The circuit, actually a digitizing in- TrL-compatible pulse at the flip-flop's 45 output.
tegrator, performs short-term integration on the input As the Q output goes high, a delayed positive pulse is
voltage. Each time an SCR conducts, the integrator sec- fed to the inverter gate, causing its output to go to logic
tion is reset, and a pulse output is obtained. 0 and resetting the flip-flop. The integrating capacitor is
Because of its fast reset action, the circuit exhibits discharged to approximately 0.7 v (the turn-off voltage
good linearity and accuracy. An output pulse rate of of the scR) to initiate the next integration period.
10,000 counts per second injects an error of only 1%, With the input resistor values shown, integrator reset
which decreases to 0.001% at 10 counts per second (cor- rate may be adjusted to around 0.25 millisecond when
responding to a 25 -millivolt input to the integrator.) Er- -10 v is applied to the input; lower input voltages give
ror, in this case, refers to the deviation from linearity of proportionately longer integration periods. A chopper -
input voltage versus output frequency. type operational amplifier is indicated in the diagram,
To avoid elaborate isolation circuitry, the SCR'S cath- but almost any op amp will work if larger drift errors
ode is connected to the integrator's summing point, per- can be tolerated.
mitting the integrator to accept only negative voltages.
Positive voltages can be accommodated by preceding
the integrator with an inverter stage.

Digitizing integrator. Dc input voltage is integrated and 'converted into TTL-compatible pulses by reset circuit that switches SCR to dis-
charge integrating capacitor. When integrator output turns on transistor, CI output of flip-flop (cross -coupled gates) goes high, switching on
SCR. Logic output pulse at flip-flop's CI output is terminated when inverter gate resets flip-flop and SCR turns off.

1 MS2

DC
0 TO 470 IcS2
- 10 V MAX

2
 and C2 are directly proportional to R1 and C1, respec-
tively: R2 = 10R1 and C2 = C1/10.
MOSFET network minimizes An automatic gain control network at the negative
audio oscillator distortion feedback terminal of the audio amplifier assures a low -
distortion sine -wave output. The MOSFET acts as a vari-
by Glen Coers able resistor. Its resistance is set so that the gain of the
Texas Instruments Incorporated, Dallas, Texas amplifier and the loss of the frequency determining net-
work are equal when the desired output is reached.
Negative half cycles of the output are rectified by a
Because a MOSFET feedback control keeps the output diode and filtered by capacitor C3. Resistor R3 provides
level constant and prevents limiting, a low -distortion a slight discharge path so that the peak level of the rec-
audio oscillator can deliver several volts into a 50 -ohm tified voltage is maintained at the gate of the MOSFET,
load. The circuit uses a 1 -watt audio amplifier as both keeping output level constant.
an oscillating element and a power amplifier, and pro- If amplifier gain increases and output amplitude also
vides a total harmonic distortion of less than 1%. More- rises, more negative bias is applied to the MOSFET'S
over, it operates from a single supply, and holds output gate, increasing MOSFET resistance to reduce amplifier
level change below 0.2 decibels if the supply voltage gain and output amplitude. If the amplitude decreases,
changes by 6 volts. the MOSFET becomes biased in the forward direction so
The oscillator circuit is a Wien -bridge type, in which that its resistance is lowered, and both gain and amp-
resistors R1 and R2, and capacitors Cl and C2 function litude are increased.
as frequency determining elements. First, R1 is chosen The audio amplifier used has an open -loop gain of 70
between 100 ohms and 1 kilohm. Then C1 = 1/2/rfoR1, dB. Attenuation of the frequency determining network
where fo is circuit resonant frequency. The values of R2 is 26 dB.

Power audio oscillator. MOSFET acts as variable resistor to control gain of audio amplifier, which functions as both oscillating element and
power amplifier. If output amplitude becomes larger than desired, MOSFET is back -biased, increasing its resistance and lowering amplifier
gain. When amplitude falls below desired level, MOSFET is forward -biased to decrease its resistance and increase amplifier gain.

APERTURE TIME

10 ps 100 ps 1 ns 10 ns 100 ns 1 10 ps 100 ms

I I 1 1 1 Ilf 1 1 1 1111 1 1 111111 I II I 111 I I 1 1 1111 I I I IIII1 I 1 I 1 1111

ERROR
0.01 % 0.1% 1% 10%
I IIH1111j I I MN I I +MINI

1 1 11111 1 1 111111 1 1111111 I 1 111111 1 1 1 11111 1 1 1 11111 I 1111111

1 Hz 10 Hz 100 Hz 1 kHz 10 kHz 100 kHz 1 MHz 10 MHz
FREQUENCY

3
Bootstrapped capacitor
stabilizes UJT oscillator
by Michael J. Debronsky
KDI Labtron Corp., Dayton, Ohio
R1

A highly accurate low -frequency relaxation oscillator

can be built by making the circuit independent of the
unijunction transistor's interbase resistance. Simply
bootstrapping the voltage of the charging capacitor
through a temperature -compensated zener diode back
to the UJT's base2 does the job. Good frequency stability
can be attained -0.05% over a 0°C -55°C temperature
range and 0.5% for a 100% change in the supply.
There are certain circuit conditions that must be ob-
served, however, for good stability. The value of timing
capacitor C1 must be much greater than that of bypass
capacitor C2. The latter bypasses any zener noise to
avoid output jitter. Moreover, current through timing
resistor R1 must be more than Qi's base current, and
the UJT's voltage must be larger than the zener's.
Ci is discharged by the UJT when its voltage is:
V, = 11V,(1-71)
where is the intrinsic standoff ratio of the UJT, and Vz
n
is zener voltage. Usually, the base -emitter voltages of
Qi and Q2 are small compared to Vz. Then:
Va = Clf i(t) dt = VA/RIC' Tightening frequency stability. Low -frequency oscillator provides
frequency stability of 0.05% over a temperature range of 0°C to
And output pulse duration becomes: 55°C. Bootstrapping voltage of timing capacitor C1 back to base2 of
t = 1/f = 11RiC1/(1-71) unijunction frees oscillator circuit from dependence on UJT's inter -
For 60 -Hz operation, R1 ranges from 10 to 50 kilohms, base resistance, the usual cause of .nstability. Temperature -com-
and C1 must be greater than 0.001 microfarad. pensated zener and bypass capacitor C2 are also used.

a simple iterative procedure to converge on normalized

element values, instead of solving for the roots of a cu-
Brief Fdrtran program for bic equation. This technique simplifies calculating gains
other than unity. Finding capacitor values that are ac-
active low-pass filters curate to one part in 107 (0.1 ppm) requires about 30
by Irvine P. Stapp, Jr. traversals of the iteration loop.
University of Kentucky, Lexington, Ky. A filter designed with the program shown in (b) pro-
vides a 15 -hertz cutoff frequency and a gain of 1.20. The
filter is intended to drive a multiplexer input network
A short computer program, written in 1130/1800 For- with a gain of 1/1.20.
tran, makes it possible to design three -pole active low- If a small amount of gain is included in an active fil-
pass Butterworth filters with gains between 1.00 and ter, many scaling amplifiers frequently can be elimi-
2.00. What's more, the program can be easily translated nated from a multiplexed analog -to -digital system
into Basic computer language. where over-all gain must be maintained at some estab-
Unlike previous programs, the one shown in (a) uses lished value. For example, certain process control corn -

4
 puters, like the IBM 1800, require multiplexer input point of 1 radian per second. Besides capacitor values,
points to be preloaded with low impedances, in the or- the program finds the values for R1 and R2 that yield
der of 1 kilohm shunted by 0.05 microfarad. Only a few optimum filter gain.
operational amplifiers can tolerate this much capaci- Other common low-pass filtering functions can be re-
tance with a tight feedback loop. But many will function alized with the program by modifying the iterative
properly if decoupled from the capacitive load by about equations. A data card is needed to indicate the number
100 ohms. of runs, followed by cards specifying, for each run, the
A sample of the program's output, also included in desired cutoff frequency in hertz, the resistance level in
(b), shows the design solution: K is the number of itera- ohms, and amplifier gain. Program statement 23 points
tions required to reach desired accuracy; L, M, and N out where the data should be placed on the card. The
are normalized capacitor values in farads for a 3 -decibel program is easily adapted for remote terminals.

(a)

WRITE (4,21)
21 FORMAT('IELLMEN7 VALUES FOR 3 -POLE LOW PASS BUTTERWORTH FILTER WIT
1H PREASSIGNED GAIN OETWEEN 1.00 AND 2.00.1//' K IS LOOP INDEX WHEN
2 SOLUTION CONVERGED TO 0.1 PPM.'//' L,M,N ARE CAPACITOR VALUES IN
3FARAOS, NORMALIZED TO OHM,
1 RADIAN/SECOND.')
1
26C IF CABS(GL-AE.)AGS(BN-AN).,ABS(BM-AM)-1.E-07) 300,100,200
READ (3,20) NRUN
25C E.F.11-05
20 FORMAT (12)
2E0 CONTINUE
DO 88 J=1,NRUN 300 WRITE (4,22) K,AL,AM,AN ,G
READ (3,23) F,R,G 22 FORMAT(101,11, K=',I3 ,T21,11.=',E13.6,138,111=1,113.6,756,'N=1,E
23 FORMAT (T1,E13.6,T15,E13.6,730,E13.6) 113.6,174,1G=1,11/.6 )
AM =1. Q. 159154.9/(F.R)
AN = 1. CI.ALN
AL . 1.
C2=AM'Q
C=G -1.
C3RK:114-1(1Q.E-03
DO 200 K=1,99
BM=AM R1.3..0.40(
BN=AN R2.RI/E
8L.AL
NRITE(4,24) G,RK,F,C1,C2,C3 ,R1,R2
AL=2.-3../1111+2..AM.E 24 FORMAT ('0','F00 GAIN OF',F6.2,' AND RESISTORS OF',F6.2,' KILOHMS
AN.(2.+4.M.AL.E)/(2.(AL.AM)) 1 A CUTOFF FREQUENCY OF',F9..' HERTZ REQUIRES'/I' ','Cl.',10.6,38
AM.1./(AL.AN)
2,'C2=',F9.6,3X,'C3.',F9.6,1X,'MICROFARADS. OPTIMUM GAIN -SETTING RE
IF CAL*AM+AN-1.E-04) 250,250,260
3SISTORS ARE 7/",'R1=',01.4,38,'R:=',F9.4,1X,'XILOHMS'///' '.119
4(18.))
88 CONTINUE
CALL EXIT
END

(b)
R2 R1

360 k52

20 kS2 20 02 20 1(12

200S2

INPUT C,
TO
1 KE2 0.05µF MULTIPLEXER
I -

ELEMENT VALUES FOR 3 -POLE LOW PASS BUTTERWORTH FILTER WITH PREASSIGNED GAIN BETWEEN 1.00 AND 2.00.

K IS LOOP INDEX WHEN SOLUTION CONVERGED TO 0.1 PPM.

L,M,N ARE CAPACITOR VALUES IN FARADS, NORMALIZED TO 1 OHM, 1 RADIAN/SECOND.

K= 29 L= 0.150211E 01 M= 0.169718E 01 f1= 0.392253E 00 G= 0.120000E 01

FOR GAIN OF 1.20 AND RESISTORS OF 20.00 KILOHMS A CUTOFF FREQUENCY OF 15.00 HERTZ REQUIRES

Cl' 0.796896 C2= 0.900386 C3= 0.208096 MICROFARADS. OPTIMUM GAIN -SETTING RESISTORS ARE

R1= 71.9999 R2= 360.0000 KILOHMS

Filter design. Fortran listing (a) eases design of three -pole active low-pass Butterworth filters. Program finds capacitor values and optimum -
gain resistors. Filter gain is preset between 1.00 and 2.00; the value of ladder resistors is also fixed. Program uses iteration for extremely ac-
curate capacitor determination. Design example for 15 -hertz filter and sample of program output are given in (b).

5
 positive, the base of transistor Q2 is driven negative,
turning Q2 off. As the emitter current of Q2 changes, the
Feedback current switch emitter current of transistor Q3 is forced to increase, and
starts positive feedback action through capacitor CI to
divides rf inputs by 20 the base of Q2. Additional positive feedback is fed from
by Roland J. Turner
the collector of Q2 to the base of Q3 through C2.
RCA Corp., Missile and Surface Radar division, Moorestown, N.J. During the recovery period of Q2, the current switch
formed by Q2 and Q3 acts as a high-speed comparator,
while the base voltage of Q2 decreases toward ground.
In a ripple -carry binary feedback counter, counting As soon as the most positive swing of the input signal
down to 20:1 requires five binary stages, and the time it exceeds the VD bias at the base of Q3, positive feedback
takes for the signal to pass through all five stages limits begins again, and the switch automatically resets itself
counting speed. But an analog counter that employs to the initial state. Transistor Q4 serves as an output
positive feedback around a single current switch stage buffer to drive another analog stage. The output is a
requires only one transition period to establish the square wave with a 2.2 -volt peak -to -peak amplitude.
count. This current switch, which has a transition time Potentiometers R1 and R2 control the initial bias con-
in the order of 1 nanosecond, counts a 1-gigahertz sig- dition of Q2 and Q3, respectively. And the collector bias
nal down to 50 megahertz in one stage. Two switches in of both Q2 and Q3 is well above their saturation voltage.
cascade, then, provide a 400:1 countdown. Moreover, when one of these transistors is in cutoff, it
The usefulness of the counter lies in its ability to pro- still has a 1 -milliampere idling current to assure that it
vide a low -frequency sync signal that is locked to an rf maintains a high fT.
carrier, so that full advantage can be taken of an oscil- The countdown of the circuit can be altered by re-
loscope's vertical bandwidth. The detailed character- turning R2 to a different supply voltage.
istics of each rf cycle of the signal may then be observed
on a scope with a low -frequency sync capability.
When an rf signal at the base of transistor Qi goes
Divide -by -20 counter. Analog circuit counts down rf signal in one transition. For positive input, 02 turns off and Q3 turns on, causing
positive feedback to Q2's base. While 02's base voltage approaches ground, 02 and 03 perform as high-speed comparator. When rf input
exceeds 03's base bias, positive feedback resets 02 and 03 to their initial state. at is buffer stage for output square wave.

+22 V
R1

5 kS2

20 kS2 910 2 20 kS2

C1

I
160 pF

0.1 ;IF
1 k2
220 S2 220 2

5.1 kS2
I 04
2N5841
C2
1µH

160 pF
2N5841 47 pF
02 Q3 VD
2N5841 2N5841 (15 MHz)
e,

1300 MHz) 0.001 1N4376 680 S2

1N4376
}41
1N4376 1N4376
5.1 kS2
22 k 2 1.8 kS2 22 k2

100 pF

-22 V

-22 V

6
 net loop gain below unity. In the conventional Wien -
bridge oscillator, the gain of the amplifier looking into
Tunable active filter its non -inverting input is maintained at 3, and the cir-
has controllable high cuit oscillates. For the modified oscillator, however, am-
plifier gain is less than 3, allowing the Q of the sur-
by Max Artusy rounding RC network to be increased. And, as can be
Stanford Electronics Laboratories, Stanford, Calif. seen from the diagram, the value of resistor R1 is the
same as that for resistor R2, and the value of capacitor
C1 is identical to the value of capacitor C2.
A tunable variable -Q active narrowband filter can be Signal current is introduced into the negative feed-
built from a slightly modified Wien -bridge oscillator back loop by R3, which also determines input imped-
with a net loop gain of less than unity. Not only can Q ance and circuit Q. For oscillation to occur:
be independently controlled, but stable single -fre- 1 + Ri/(R3+ Rs) must be less than 3.
quency Qs as high as 2,000 can be realized. A gain of where Its is the source impedance. Resonant frequency
600 is achieved with a Q of 2,000. Gain becomes ap- becomes:
proximately 140 when Q is 30, and Q remains constant fo = 1/27(RiR2C1C2)1/2
within ±10% over a 10 -to -1 tuning range. (Gain, in this Circuit components should be selected carefully for
case, refers to signal gain, from input to output, rather best tuning range and constant Q. In particular, well -
than amplifier or loop gain.) matched wirewound potentiometers help maintain uni-
Although it is generally known that the effective Q of form Q with changing frequency. All capacitors should
an oscillating tuned circuit is infinite, it is often forgot- be Mylar. Nominal component values shown result in a
ten that stable finite Qs can be obtained by reducing the tuning range of 160 hertz to 1.6 kilohertz for the circuit.
Adjustable 0. Active filter offers Q of up to 2,000 that is stable to within ±10% for moderate Os over 10 -to -1 tuned frequency range. Actually
a modified Wien -bridge oscillator, the circuit operates with a net loop gain of less than 1 so that Q of RC network can be increased. Poten-
tiometers RI and R2 and capacitors CI and C2 determine filter resonant frequency, while potentiometer R3 is an independent Q adjustment.

c,
0.05 pF
,11
TUNE
82 + 15 V
20 k.0.

741 OUTPUT

R4 10k12
C = 0 05 pF
20 k0.
15V

Rs

101(0
0 SET
HE LIPOT

7
 second amplifier through Q1 and Q2. Transistor biasing
is implemented by connecting one input of Al below
Op amps generate ground through resistors R2 and R3.
The staircase output is produced by amplifier A2,
precision staircase which integrates, through capacitor C2, the current sup-
by Jerald Graeme plied by Qi and Q2. The change in output voltage (Deg)
Burr -Brown Research Corp., Tucson, Ariz. becomes:

Pe, =
C2
f i(dt)
A staircase generator with adequate precision for se-
quential control and multiple -level testing produces a where:
staircase output by differentiating and then integrating i = Cide,/dt
only the negative transitions of a square wave. Al- Or, whenever input voltage change (fie;) is less than 0:
though a staircase waveform can be generated precisely Aeo = -C10ei/C2
by a digital -to -analog converter driven by a clock -con- Each negative transition of the input square wave
trolled counter, a simpler, but sufficient, circuit ap- creates a step in the output voltage, generating a stair-
proach is to use operational amplifiers. case waveform. This stepping continues until the output
Amplifier Al differentiates and rectifies input square voltage reaches the trigger level of the reset clamp
wave ei, which is applied to capacitor C1 through re- formed by transistors Q3 and Q4.
sistor R1. For positive input transitions, transistors Qi The clamp triggers when the emitter -base junction of
and Q2 are off, and diode D1 conducts; for negative Q4 breaks down, limiting the output to a peak voltage
transitions, Qi and Q2 conduct. From this rectification, (Vp) of:
only negative -going transitions are transferred to the Vp = VCB - BVEB VHE - Vf

Stepping up. Staircase generator first differentiates and then integrates input square wave to supply precise stepped output. A, performs the
differentiation, and 0, and 0 rectify all negative -going inputs, passing t hem on to integrator A2. Positive input transitions are not used. Every
negative input steps up the output until 03-04 reset clamp triggers when 04's emitter -base junction breaks down.

3
 where VCB is Q4's collector -base voltage, BVEB is Q4's Gain error and nonlinearity for this staircase gener-
emitter -base breakdown voltage, VBE is Q3's base -emit- ator vary with operating frequency and component
ter voltage, and Vf is the voltage across diode D2. Be- characteristics. Operational amplifier input currents re-
cause the collector -base junction of inverted transistor move some of the derivative current pulses, and they
Q4 is forward -biased, its voltage drop is around that of also create output sag by discharging capacitor C2. To
the diode, and: prevent severe error from being introduced by these op
Vert - Vf = 0 amp input currents, amplifier input protection circuitry
-so that Vp is approximately equal to 7 volts: must be avoided since such circuitry will draw high cur-
Vp = VBE - BVEB rent during an input overload.
Since the thermal variations of VBE and BVEB approxi- Slewing rate and overload recovery limit the gener-
mately cancel, Vp is quite stable, with a temperature ator's operating frequency range. Another current
coefficient of around 0.01%/°C. transfer error results from the finite betas of transistors
With Q4's breakdown, base current through Q3 acti- Qi and Q2. However, for the Darlington pair used, the
vates the positive feedback loop formed by Q3 and in- loss is only about 0.002%. Resistors R4 and R5 are in-
verted transistor Q4. Capacitor C2 is then discharged cluded to reduce an output nonlinearity that is intro-
until its voltage reaches VBE, the cutoff for clamp con- duced by leakage of the reset clamp. Circuit gain error
duction. Once output voltage returns to 0 v, the clamp is typically about 0.05%, and nonlinearity is commonly
turns off, and a new staircase begins. around 0.1%.

Diode gating. Pull-up resistor and Schottky diodes replace NAND

gates at input of low -power TTL circuits, improving gate response
Schottky diodes eliminate time and decreasing power consumption. Since Schottky diode drop
two -level NAND gating is only half that of pn junction diodes, Schottky diodes are compa-
tible with logic 0 condition of TTL gates. Pull-up resistor decreases
by Charles J. Huber, gate response time by connecting gate input directly to supply.
Westinghouse Electric Corp., Systems Development division, Baltimore, Md.

To eliminate the need for two -level gating with NAND

gates, a pull-up resistor and Schottky diodes can per- 4.9 V
form the AND function at the input of a low -power TEL 50 kS2 TTL. LOAD
gate. NAND gating causes an unwanted increase in
power consumption and delay time. The diode circuit OUTPUT
can be used for decoding sequence -generator outputs, MOTOROLA
encoding digital signals, and implementing the AND 3- MM80 - 101
SCHOTTKY DIODES
TEXAS INSTRUMENTS
SN 74L00N
function in counter designs.
Conventional pn junction diodes are not suitable for
DELAY TIME
use with low -power TEL gates since diode forward volt- (INPUT TO OUTPUT)
TTL RESISTOR
age drop is comparable to the maximum allowable DIODES
LOADS I kS2)
logic 0 gate input voltage-around 0.7 volt. But RISING EDGE
Ins)
FALLING EDGE
Ins)
Schottky -barrier diodes exhibit only half the voltage
drop of pn junction diodes, permitting them to satisfy 0 1 75 70
the TTL logic 0 condition.
0 5 75
Moreover, the dynamic change from logic 1 to logic 0 ,,0 70
at the output of a TEL gate is limited, since the common - 1 1 .. 70 110
base input stage is not rapidly cut off. Gate response
can be improved considerably by placing a resistor from 1 50 80 65
the supply line to the gate's input pin. 50
2 1 85 65
The test results in the table show that the worst -case
degradation of gate response time from no diodes to 3 1 50 80 65
three diodes is 35 nanoseconds for five-TTL loads. Since
3 2 50 80 80
this time factor is comparable to the propagation delay
of only one TEL gate, an entire gate delay is saved and 3 3 50 80 95
the power consumption is due principally to the pull-up
resistor. When three diodes are used, the dc level at the 3 4 50 85 105
diode -resistor junction is approximately 0.5 v. 3 5 50 85 105
Designer's casebook is a regular feature in Electronics. We invite readers to submit original
and unpublished circuit ideas and solutions to design problems. Explain briefly but thor-
oughly the circuit's operating principle and purpose. We'll pay S50 tor each item published.

9
 ing cycle, its output goes low, Q2 turns off, and C1 be-
gins to charge for the next cycle. Circuit oscillation pe-
Stable square -wave generator riod is approximately:
provides broad bandwidth T = 3.3C1R1/0.7 + one-shot width
One-shot output pulse width should be as narrow as
by William Standke possible for width stability, but wide enough to dis-
Honeywell Inc., Corporate Research Center, Hopkins, Minn. charge fully the timing capacitor. Resistor R2 slows
down this discharge. Although R2 can be eliminated for
small values of capacitance, it must be included for
A high-speed analog comparator and a linear current large capacitances to protect Q2 from excessive currents.
source are the key elements in a stable square -wave The generator's gate input line allows it to be started
generator that can oscillate from a few hertz up to sev- and stopped with a digital signal. It should be noted
eral megahertz. The generator circumvents two com- that the generator always starts in the same phase, and
mon sources of stability error-a nonlinear capacitor - its initial cycle has the same period as the subsequent
charging current and limited op amp bandwidth. ones. If the gating feature is not wanted, the gate line
Current source Q1 linearly charges capacitor C1 until should be connected to a positive supply voltage (Vcc)
capacitor voltage exceeds the reference zener voltage on In addition, the generator can be used as a voltage -con-
one of the comparator inputs. The normally high output trolled oscillator by replacing the zener with a control
of the NAND gate then goes low, starting the one-shot voltage or by controlling the base voltage of Qi.
timing cycle. As the output of the one-shot goes high, The silicon transistors and switching diodes selected,
Q2 turns on and sinks the current from Qi so that C1 is as well as capacitor C1, depend on the operating fre-
quickly discharged. When the one-shot finishes its tim- quency desired.
Controlled pulses. Square -wave generator offers stability and wide operating frequency range. C1 accepts linear charge from Q1 until its
voltage reaches that of zener. Comparator then switches, starting one-shot and turning on 02, which discharges C1. When one-shot timing
cycle is complete, 02 goes off, and C1 charges again. Gate input permits generator to be controlled digitally.

10
Two adders form TEXAS
N2

BCD subtractor INSTRUMENTS

SN74042

by Peter K. Bice 1 8 10 16 4 7 11

Hewlett-Packard Co., Microwave Division, Palo Alto, Calif. CARRY IN

4 4 -BIT ADDER 13
(LOGIC
15 2 9 FOR FIRST
STAGE)
Engineers often avoid using binary -coded -decimal
(BCD) designs because the arithmetic involved seems to TEXAS
be formidable. But BCD arithmetic can be almost as INSTRUMENTS
simple as binary arithmetic. The circuit shown, for in- LOGIC 1 SN7483

stance, does BCD parallel subtraction. All that is needed 1 3 8 10 16 4 1 11

for each digit is two four -bit adders and four inverters. 431T ADDER is LOGIC 1
The first adder does the actual arithmetic, subtracting 15 2
one digit from the other. The second adder converts this
difference to BCD form and, if necessary, generates a CARRY OUT
"borrow" for a subsequent stage. The CARRY OUT is al-
ways logic 1 unless a "borrow" is being generated. The BCD OUTPUT
last subtractor stage will have a logic 1 CARRY OUT if IN, -N21
the difference between the two numbers is positive,or if
the difference is zero. Digital arithmetic. A couple of four -bit adders and four inverters are
The diagram shows one digit of a parallel subtractor. all of the hardware required to build a BCD parallel subtraction cir-
The stages can be cascaded, or one stage can service a cuit. Top adder takes actual difference between input numbers N1
serial shift register. BCD addition, it should be noted, and N2. Second adder then forms BCD number of this difference.
cannot be implemented as easily as BCD subtraction Both "borrow" and CARRY OUT lines are included for use with
since more hardware is required., other stages. Subtractors can be cascaded.

This voltage is then rectified by hot -carrier diode D1,

producing a negative voltage at the junction of resistors
VSWR detector protects R1, R2, and R3. The values of R2 and R4 force transis-
tors Q4 and Q5 into saturation, creating a short circuit
class C rf amplifiers between Q4's collector and ground that allows Qi to op-
by Frederick A. Warren, Jr. erate normally, relatively insensitive to load phase.
Harris-Inlertype Corp., Radiation Systems division, Melbourne, Fla. If the amplifier load is removed or shorted, a high
VSWR results, and the rf voltage across T1 increases. The
rectified voltage at the resistor junction then becomes
An inexpensive protection circuit against open -circuit more negative and pulls the base voltage of transistor
damage to class C rf amplifiers senses and detects volt- Q5 down, driving both Q4 and Q5 into cutoff. Now, Qi's
age standing wave ratio with a single -turn transformer. emitter opens, removing. the drive to transistor Q3 and
In addition, the circuit removes drive from high -power therefore shutting off the amplifier.
stages, has low insertion loss, and suppresses second For adequate circuit reaction time, the R1C1 time
harmonics with a notch filter. constant should be less than one period of the ampli-
Any class C rf power amplifier must be operated with fier's operating frequency. Resistor R3 sets the quiescent
a load, once supply voltage and an rf signal are applied. base current of Q5. And the Q4 -Q5 Darlington pair op-
If the load is removed, a high VSWR is created, and rf erates with only a small drive voltage differential be-
power is reflected back to the output transistor's collec- tween transistor cutoff and saturation regions.
tor. The resulting rf voltage swing can exceed the tran- The protection circuit also offers the advantage of
sistor's collector -base breakdown voltage, causing col- second harmonic suppression (the second harmonic of
lector current to increase in an avalanche manner so an amplifier must be below some maximum power
that the base -collector or base -emitter junction opens. level). Because a class C amplifier is usually operated in
Risking this destruction is normally unavoidable, unless a saturated condition, it often generates high-level har-
the designer is willing to accept a significant insertion monics that require additional filtering after the output
loss or pay a high price for protection. stage, thereby creating undesirable insertion loss or
Figure (a) shows a block diagram of a high -power complexity. But the protection circuit provides an in-
class C vhf amplifier-consisting of transistors Qi, Q2 tegral notch filter whenever its transformer's secondary
and Q3-externally connected to the protection- circuit. winding and capacitor C2 are tuned to the second har-
When the amplifier drives a properly matched load, a monic of the amplifier's fundamental frequency. And
quiescent rf voltage develops across transformer T1. VSWR detection by the protection circuit suffers only

11
negligible degradation because of the filter. Load watchdog. Output of class C rf amplifier (a) goes to protection
Graph (b) depicts the voltage levels at the base of circuit that shuts off amplifier when VSWR exceeds safe operating
transistor Q5 when the amplifier output is loaded, limit. Transformer T1 performs as VSWR sensing and detection ele-
opened. or shorted. The saturation and cutoff voltages ment. Removing or shorting load pulls down Q5's base voltage, cut-
shown are those needed for the Darlington circuit. They ting off 04-05 Darlington pair and opening Q1's emitter to turn off
must be less than the loaded voltage but greater than amplifier. Plots (b) and (c) illustrate amplifier performance.
the open -circuit and short-circuit voltages. Graph (c),
which shows amplitude -versus -frequency curves for the
amplifier with and without the protection circuit, re-
flects the amount of insertion loss. From 170 to 230
megahertz, worst -case insertion loss is 0.15 decibel.

FERRITECORE
ARNOLD EG(80 (MF -34)
MATERIAL I FITI8
ot, 1c OUTPUT
171
00!
CLJSSC t...A.A..!-1 La; CEO
REV! LETT-
PAC ARJ ! 1000 pF
IC HP 5682
HOI.CARRIF.F1
Qy SA1 R ;
DIODE C2
CLAS.; A aL!, DEP
19k2,$e
+ 0 -1%4-- 4 -4- 4 1

r -*P 160 1:0 18') 190 200 210 220 251- 240
FIAECOENC y
INPUT 6I (..t, -2
flE
CLASS A 41

Y11 F
-4
R --._c!" 2 !E' 10K;
"47! Ne.t4INAl_ or'Eti

-6

Q5 +28 V - Ft

2N2222.
-10

AMP L'HER E 1. HOU .1

VSWR PROTECTION

i!,tiPLIF!ER 11

VSWR PRO

/0 !tie. 200 713 22 0

FiiEQUFNCY IiJWz

va.r walle0A,Y11411:16:11-a-
 creased by a factor of 1 + K, the apparent input im-
pedance is RI/(1 +K,.).
Transducer preamplifier On a small -signal basis, then, the preamplifier's input
conserves quiescent power stage is equivalent to a common -source configuration,
while the bias arrangement is that of a source -follower.
by Robert F. Downs For the over-all circuit, the dc input impedance is
Ocean & Atmospheric Science Inc., Santa Ana, Calif. around 1.5 megohms, while the ac input impedance is
about 300 kilohms.
The FET selected for this circuit should have a low
A low -voltage micropower preamplifier holds power pinch -off voltage (Vp) and a low drain current (Inns)
dissipation to approximately 13 microwatts because of when the gate -source junction is shorted. For the device
the low bias current of its two -transistor impedance con- used, Vp is about 0.1 v and IDas approximately 100 µA.
verter output stage. The preamplifier, which is intended Because a FET'S transconductance (gm) depends on
for use with a capacitive transducer, operates at a quies- drain current, Qi's gm is only around 50 micromhos.
cent current of 10 microamperes with a supply voltage Since the FET'S output conductance is negligible, its
of only 1.35 volts. output impedance, like that of a common -source stage,
The gate of field-effect transistor Qi is essentially essentially equals R2. Because this is a high resistance
biased at 0 v through resistor R1. Negative feedback, value, two bipolar transistors, Q2 and Q3, are used as an
provided by resistor R2, maintains Qi's gate -to -source impedance converter.
voltage at approximately -0.4 v, forcing its drain cur- This converter stage operates like a pnp emitter -fol-
rent to less than 4p.A. Resistor R2, therefore, contributes lower, providing very high values of current gain and
significantly to Qi's bias stability. input impedance. Moreover, it realizes greater bias volt-
Preamplifier input impedance depends on both R1 age compatibility between the FET and bipolar stages
and the voltage gain of the field-effect transistor stage. than a conventional Darlington pair could. Converter
Actual FET intrinsic input impedance can be ignored bias current is about 6 µA, input impedance exceeds 2
since it is orders of magnitude larger than R1. megohms, and output impedance is about 4 kilohms.
If ei denotes input signal voltage, the voltage across For the preamplifier, equivalent input broadband
R1 can be expressed as e1(1 + Kv), where lc is the noise is relatively low, about 33 ay from 140 hertz to 20
stage's voltage gain. Since current through R1 is in- kilohertz. And voltage gain is nominally 5 (14 decibels).

Power pincher. Preamplifier for capacitive transducer input dissipates only 13 microwatts and operates from 1.35 -volt supply. Bias current
of impedance converter, composed of bipolar transistors 02 and 03, is only 6 microamperes, keeping total circuit current drain to only 10µR.
FET input stage has source -follower bias arrangement but provides voltage gain of common -source configuration.

,1 35 Vdc

0.33nF

TRANSDUCER
(5000 pF)
0 5 kS2
MINIMUM

13
 applied to the strobe input of a latch, and the variable
square wave is applied to the data input of the latch.
Differentiate and count How many positive or negative transitions the latch
output makes in a given period represents the number
to find frequency error of times the variable frequency has gained or lost a full
by Robert C. Rogers
cycle relative to the reference. Counting these transi-
Texas A&M University, College Station, Texas - tions for a known period, then, yields the total fre-
quency error during that period. A high-level output in-
dicates that the variable is within the limit.
By counting differential pulse transitions for a known The time -base input signal sets the latch to a logic 1 if
period, a simple error -detection circuit measures the fre- the four -bit counter does not reach a value of 10 in the
quency difference between a variable square wave and preceding 0.1 -second timing interval. This signal is also
reference square wave. Usually, frequency error is used to reset the counter for the next counting period,
found by counting both signals for some fixed period which begins when the reset pulse returns to zero.
and then comparing the resultant values, or by mixing There are two drawbacks that should be remem-
the signals and then counting the beat frequency. The bered. The detector fails if the variable frequency be-
first method often requires a considerable amount of comes identical to the reference in both frequency and
digital logic if high -frequency signals are involved, phase, and it becomes ambiguous when the variable
while the second method is generally limited by the frequency is either a higher or lower harmonic of the
passband of the beat -frequency detector. reference. However, even with these limitations, the de-
The frequency error detector shown requires input tector is useful and reliable over the reference range of 5
signals that are compatible with logic circuits. (A zero to 20 megahertz. Error limit for the circuit shown is
crossover detector could be used initially to prepare the -±100 hertz and can be changed by altering the number
inputs.) The reference square wave is differentiated and detected by the counter gate.
Frequency error detector. Differentiated reference square wave drives latch strobe input; variable square wave feeds latch data input.
Counter logs either positive or negative transitions of latch output for specific period to total frequency error between inputs. High circuit out-
put indicates error is within desired limit. Time -base signal sets latch and resets counter. Detector range is 5 to 20 MHz for ±100 Hz.

TEXAS
INSTRUMENTS
SN7493
COUNTER

VARIABLE
INPUT

rut
1

123 S2

Nt
REFERENCE
INPUT
- 10 pF

FAIRCHILD
OUTPUT

an_ 100 S2

TIMEBASE
INPUT
(0.1 Si 100 pF

14
 ampere in 10 nanoseconds for 15 -milliampere open -col-
lector drivers) may require a large pull-up resistor to
Feedback latch reduces keep voltage noise within specifications at the expense
memory recovery time of turn-off speed.
The illustrated latch employs the common cross -
by Joseph McDowell and William Moss couple gate arrangement of G1 and G2. A feedback
Monolithic Memories Inc., Sunnyvale, Calif. gate, G3 actively pulls up the wire-oRed line (like a tri-
state gate) after the cross -coupled gates latch. Diode D1
isolates G3 from node A, allowing low-level signals to
The cycle time of wire-oRed semiconductor memories be sensed at this point. To avoid transistor -transistor
can be improved with the addition of a feedback gate to logic high-level signal problems at node A, D1 should
a NAND gate latch. The resulting three -gate configura- have a low forward -voltage drop.
tion, which also includes a diode and a resistor, provides Resistor R1 is used to limit the current from gate G3
latched data outputs from open -collector memory pack- to the open -collector outputs of memories S1 through
ages with significantly reduced turn-off delay. SN. The value of R1 must be small enough to provide an
For open -collector devices, like S1 through SN in the acceptably low signal level at the latch input. Only the
diagram, the common problem is choosing a pull-up re- output leakage of the wire-oRed memories and the in-
sistor that is small enough for fast turnoff and large put leakage of the feedback latch determine the max-
enough for the current sinking capability of the pack- imum value of pull-up resistor R2. Recommended val-
age's open -collector driver. Turn -on speed is not usually ues for resistor R2 range from to 10 kilohms,
1

a problem, since capacitors C1 through CN are driven depending on power supply load considerations.
from a low -impedance saturated transistor. For 16 wire-oRed memory packages and a 4.7-kilohm
However, turn-off speed or recovery time is deter- pull-up resistor, the feedback latch can reduce turn-off
mined by the RC time constant at the wire-oRed node. If time from 70 to 10 nanoseconds.
many memory outputs turn on at once (for example
when all low -logic signals are stored in a 72 -bit memory
word), the change in the power supply load (about 1

Speeding up wire -oiled memories. Cross -coupled gates G1 and G2, along with feedback gate G3, trim turn-off delay for array of wire-oRed
memories (S1 through SN). Three -gate feedback latch uses diode to isolate G3 so that low-level signals can be detected at wire-oRed node A.
Resistor RE prevents surge currents from G3. Maximum value of pull-up resistor R2 is determined by leakage currents.

DATA OUT
Vcc

C1 -1-
Si
(1)

Li LE t

CHIP ENABLE RESET

(1)

t
S2

CHIP ENABLE
(2)

SN
CN -T-
CHIP ENABLE
(N)

187 n (A) t
(A)
DATA OUT LATCHED
H.1 V
t

IN4531 I Di
co (13
DATA OUT

74H00
TYPICAL
Li t

(CV
DATA OUT LATCHED

RESET ' t

15
Stable voltage reference
uses single power supply
by Mahendra J. Shah
University of Wisconsin, Madison, Wis.

When an integrated circuit differential operational am-

plifier is used in a voltage reference source, two supplies
usually are needed-both positive and negative. More-
over, additional circuitry is generally required to estab-
lish output polarity. But if one supply line is grounded,
only one supply becomes necessary-the one that gives
the desired output polarity.
So long as the non -inverting input of the operational
amplifier is maintained at some positive potential, even
a tiny one, the output of the reference source will also
be positive. Similarly, small negative potentials at the
non -inverting input will result in negative outputs.
The voltage reference source shown employs a single
15 -volt supply to give a positive output. To produce a Eliminating dual supplies. Voltage reference source has either
negative output voltage, the positive supply line is
positive or negative output, depending on polarity of supply used to
grounded, the negative supply line is run to -15 v, and bias op amp. Positive supply results in positive voltage at op amp's
non -inverting input and, therefore, positive output. Negative supply
Zener diode Dz establishes the reference voltage (Vz) produces negative output. In either case, final output voltage is
and the reference current (Iz) for the circuit so that a about 10 volts. Zener diode acts as reference.
constant -current source is provided by resistor R1 and
the buffering action of the op amp. Zener current is: -0.5 v. The negative voltage at the op amp's noninvert-
Iz = VzR2/R1113 ing input causes a negative output:
Due to the configuration of the op amp's output E, = -Vz(I + R2/R3)
stage, circuit output voltage (E0) will always be greater The reference zener voltage of 6.4 v yields a zener
than 0.5 V when a positive 15-v supply is applied and current of about 2 milliamperes. For a positive supply,
the negative supply line is grounded. Therefore, a few output voltage is 9.547 v; temperature coefficient, 1.9
tens of millivolts will always appear at the op amp's parts per million/°C; voltage stability, 9.5 ppm/v; and
non -inverting input, making the output positive: output impedance, 53 milliohms. For a negative supply,
E. = Vz(1 +R2/R3) temperature coefficient remains the same; but output
With the positive supply line grounded and -15 v at voltage becomes -9.560 v; voltage stability, 2.6 ppm/v;
the negative supply input, E. will always be less than and output impedance, 21 milliohms.

turned on or off, and while it is on, the phase can be re-

Fast -switching modulator versed between 0° and 180°. An off signal is attenuated
by at least 35 decibels.
reverses uhf signal phase The modulator (a) consists of a tri-state driver and a
double -balanced mixer, which allows the rf signal to be
by R.N. Assaly controlled by video commands. Three values of control
Massachusetts Institute of Technology, Lexington, Mass.
current- -30 milliamperes, 0 mA, and 30 mA- are gen-
erated by the driver for the three modulator states- 0°
phase, off, and 180° phase, respectively.
In just a few nanoseconds, a modulator for shaping The transition times of the two input logic signals,
transmitted ultrahigh -frequency radar signals can designated as code Q and gate P, are enhanced by a
switch a signal through three states. The signal can be dual four -input AND gate. (The gate -P input turns the rf

16
signal on and off, performing a gating function; the Since Q3 and Q4 do not conduct if their base voltage
code -Q input reverses the phase of an on signal, thereby is less than 0.6 volt, the resistor network holds bias volt-
coding signal phase.) A line driver then produces the age to about 0.5 v during the off state to reduce any de-
outputs labeled A and B: lay during driver state transitions. Also, because a large
A = PQ and B = PQ base voltage must be applied either to Q3 or Q4 for con-
At the tri-state driver output, a difference signal, B -A, is duction to occur, mixer current can be held to a very
used to provide the three desired states, as shown in the low value. If mixer current increases, to even just a few
truth table. microamperes, the attenuation of an off signal becomes
To obtain this difference signal, transistor Qi inverts less than the desired 35 dB. For instance, attenuation
the A output of the line driver to produce signal C. And degrades to about 30 dB when mixer current becomes
output B passes through transistor Q2 without inversion, approximately 10 microamperes.
keeping the circuit balanced and resulting in signal D. Inductors and capacitors in the circuit are simply
A resistor network averages signals C and D, which are used to speed up state transitions. The turn -on of a 430-
then applied to output transistors Q3 and Q4 that drive megahertz signal is illustrated in (b), while (c) shows a
the double -balanced mixer. phase reversal. These transitions took less than 10 ns.
Uhf modulation. Three -state modulator (a) turns off uhf signals or reverses their phase between 0° and 180°. AND gate improves input sig-
-el transitions. To balance circuit, line driver output A is inverted by 01, while B passes through 02 without inversion. Transistors 03 and 04
drive mixer with difference signal of B -A. Turn -on (b) and phase reversal (c) traces are for 430 -megahertz signal.

+6V

50 5052 75012 15 pF
0.01µF 10012 100
0.01 pF

2.2pH
14 1100 pF
1 0.1 pF
A
03
CODE 12
a 33012 2812222

5 1600° OUTPUT
B TO DOUBLE -
s a, 4 68 0 BALANCED
GATE P 10 33012 MIXER
saa 2N3640 LORCH
10 LU 1 k2 1130 pF
ELECTRONICS,
2N2901
7 11 MODEL

0 OlpF 10012 100 2 100 Cl i 0 1 pF

50 Cl

+6 V

I 15mE

(a)
P 0 A B
GATE P L
0 0 1 1 0
0 1 1 1 0 CODE 0
1 0 1 0 -1
0
1 1 1 1

OUTPUT r -L,

17
 2-ns gate, G1. The LC tank circuit tunes the 100 -MHz
crystal overtone, and also acts as a fine -tuner. Another
ECL gates stretch 2-ns gate, G2, from the OR output of G1, buffers the os-
oscillator range cillating section and provides a complementary output.
The frequency doubler section of the circuit consists
by William Blood of two 2-ns gates (03 and G4) performing as phase
Motorola Semiconductor Prou,icts inc., Phoenix, Ariz. shifters, and two high-speed 1-ns NOR gates (G5 and
G6) operating as summers. For a 50% output duty cycle,
the complementary 100 -MHz signals should be delayed
The frequency range of crystal -controlled oscillators can one-fourth of a cycle, or 2.5 ns. This may be done pre-
be extended easily with emitter -coupled logic gates. Se- cisely with delay lines, or approximated with gates G5
lecting the proper crystal and the right number of fre- and G6, as shown. The gating method is easier to imple-
quency doublers results in accurate frequency signals ment and causes only a slight offset in output duty
over the range of 150 to 250 megahertz. Standard crys- cycle. Gates G5 and G6 combine the four -phase 100 -
tals are normally limited to less than 150 MHz because MHz signals, yielding a 200 -MHz output frequency,
of the number of crystal overtones required to achieve when their outputs are wired -OR.
higher frequencies. A third 1-ns gate G7 is used as a bias generator for
A 200 -MHz crystal -controlled oscillator that is built the crystal oscillating section. Tying the output of this
with only two integrated circuit packages is shown in NOR gate back to its input assures that the oscillating
the diagram. Two types of ECL gates, one with a propa- section remains biased in the center of its linear region
gation delay of I nanosecond and the other with a 2-ns over wide temperature and power supply extremes.
delay, are used for best performance.
In the oscillating section of the circuit, the crystal is in
series with a feedback loop from the NOR output of a
Speeding up crystal oscillators. Emitter -coupled logic gates can increase the frequency output of crystal -controlled oscillators to 250
megahertz. For 200 -MHz output, LC tank tunes 100 -MHz overtone of crystal, while gate G2 forms complementary 100 -MHz signals. Phase
shifters G3 and G4 and wired -OR summers G5 and G6 then delay and double these signals. Gate G7 provides buffered bias supply for gate GI.

100 MHz CRYSTAL T

If OSCILLATOR

G1

// 0.15pH
'4(000101
MOTOROLA
G2
RP

7 35 pF
',,MC10101
MOTOROLA

G7
".iMC1662
MOTOROLA
0.001 µF

b 1

G3
%MC 1 0101
MOTOROLA

OUTPUT
(200 MHz)
G6
C 1662

(
G
MOTOROLA
IOM CI 0101
MOTOROLA

FREQUENCY DOUBLER

Rp: TYPICALLY 510 52 TO VE E OR 5052 TO -2 VDC

18
 ey = ex/( I +ex)
For small values of ex, this equation degenerates to:
Linear signal limiting ey = ex
with feedback multiplier which is a linear function representing a 1:I compres-
sion ratio between e and e,. For large values of ex, the
by R.J. Karwoski function becomes asymptotically limiting:
Raytheon Co.. Equipment division. Sudbury, Mass ey = ex/ex = 1
Over-all limiter transfer function can be written as:
e 1-1-lir/[111R0/(R1-i-11.2)]
A signal can he linearly compressed or limited over a
ei
- 1 f-410
wide dynamic range by using a four -quadrant analog
multiplier as a feedback element. Particularly useful for Multiplier output exey/10 and the product of gains of
audio applications, this linear limiting technique does amplifiers A2, A3, and A4 are represented by the factor
away with the signal distortion that occurs with non- 410(O11k. where (ei)pli is the peak input signal ampli-
linear methods. Also, the linear limiter does not require tude. Resistor RF controls both circuit gain and com-
the careful calibration and many trial -and -error adjust- pression. However, for any single value of e1, the limiter
ments needed for a nonlinear limiter. can be set to provide a gain of unity, regardless of RF's
The control section of the linear limiter consists of resistance and how much or how little limiting is
three operational amplifiers and a multiplier. Op amp needed. Additional over-all gain adjustments are then
Al is the throughput amplifier with local feedback unnecessary, even if the compression ratio must be
through resistors RI:. R1. and R2. When RF is shorted, changed.
the control circuitry is bypassed, and the circuit be- With a 100 -millivolt input, the circuit illustrated sup-
comes a linear voltage -follower: plies a 100-mv output, no matter what the setting of RF.
e = e1(1 + RF/ER R.,/( R, + RA) For any value of RF, resistor R1 is selected to keep:
The fundamental feedback equation for the limiter is I + RF/RIR2/(R1 + R2)] = 1 + 410RF(100mv)/R2
based on amplifier A1: Circuit compression ratio can be varied by changing the
e = Aei/( I + A/3) value of RF, but circuit gain remains unity for e, = e =
where gain A is determined by A1's local feedback ar- 100 my.
rangement of RF and R1 in parallel with R.,. Feedback The performance curves show the limiter's transfer
factor /I depends on the control section, whose oper- function for three compression ratios with circuit unity -
ating function resembles the basic feedback equation. gain point at 100 my -1:1 (no compression, RF = 0),
The relationship between multiplier input ey and multi- 8: I (maximum compression, RF = 25 kilohms), and 3:1
plier input ex becomes: (midrange compression).

Variable linear limiter. Compression ratio can be varied by adjusting resistor Rj: without changing circuit gain-limiter's unity -gain point re-
mains e, = e = 100 millivolts. Amplifier Al is controlled by local feedback through resistors RI:, R1, and R., and by additional feedback from
multiplier and amplifiers A2, Al, and A, When Rp = 0. compression ratio is 1:1, when RF = 25 kilohms, compression is 8:1.

Al
74 1
COMPRESSION
AND GAIN
ADJUSTMENT

ex
MULTIPLEXER TELEDYNE/
-exey PHILBRICK
10 4452

C(J%S, A',+PL I UDE

I

125 REGION

1N4606 > 100

A2 E

74 1 01 A 75
47 k1
0
--- 50 1 I COMP RATIO
3 1

25 81
7.5kS2 0.01 pF
135 V)
330 S 0 I I I I I
25 50 75 100 125
mV

19
 Six logic gates (a) can compare two words when the
Simple logic circuits least -significant bit is the initial input. Eight logic gates
(b) are needed when the most -significant bit arrives
compare binary numbers first. The heart of both schemes is a three -state latch
that provides three comparisons for input words A and
by Edward J. Murray B: A is greater than B, A equals B, and A is less than B.
Inter -Computer Electronics Inc., Lansdale, Pa. The timing diagrams in (a) and (b) illustrate circuit
operation when input A is octal number 25 (binary
010101) and input B is octal number 15 (binary
In real-time data acquisition applications, determining 001101). Signals A and B are not limited to a fixed
the relative magnitude of two binary numbers with number of bits per word. Any variable word size can be
hardware, rather than software, now requires only three used if the results are interrogated after the word has
logic modules if data is being transferred serially. Pre- been transmitted and a reset pulse precedes the word
viously, rather complex circuitry was needed. being interrogated.
Only the most -significant difference between two This type of binary comparator is useful in prelimi-
coincident serial data streams defines their relative nary data sorting and number ranging prior to software
magnitude. If the least -significant bit is transmitted processing for multi -channel data acquisition. In the
first, the last difference between coincident word bits timing diagram, the "result available" waveform indi-
determines which is the largest word. When the most - cates the best interrogation periods.
significant bit is the first transmitted, the first difference
establishes relative magnitude, and all other differences
can be ignored.

Checking number size. To find relative magnitude of serial words A and B, only most -significant differences between coincident data bits
must be considered. Transmitting least -significant bit first (a) requires six gates for comparison. Two additional gates are needed (b) if most -
significant bit is transmitted first. In timing diagrams, octal 25 (binary 010101) is compared to octal 15 (binary 001101).

r
A>

A-B

L SN7486N
B
A<B

SN7410N
_J ' SN'7400ft_
TEXAS TEXAS
INSTRUMENTS INSTRUMENTS
RESET RESET

LEAST SIGNIFICANT MOST SIGNIFICANT MOST- SIGNIFICANT LEAST- SIGNIFICANT

BIT BIT BIT
4 d
1

i
A = 258 A 2 i
0 0 0 11- --- _ i-
= 158I
0 1 0 0
B = 15
1-1 o o 1 1 U1 ,1

f
A B A>B
-41.......1
1.4L
A=B A=8
Fir -ir--1 rt--
A B I
A<B ,

ir
RESET

n RESULT
AVAILABLE

WORD N-1 WORD N WORD N+1 WORD N-1 WORD N

20
 floating voltage source, canceling any offset voltage.
As shown in (a). unity -feedback amplifier Al is con-
Series -connected op amps nected to the inverting input of amplifier A.,. providing
null offset voltage a floating offset voltage source between A2's non -invert-
ing input and its output. (If Ai's offset characteristics
by Lawrence Choice were matched with opposite polarity to those' of A2.
Burr -Brown Research Corp., Tucson, Ariz. then Al could be placed at A2's non -inverting input.)
Letting V represent the input offset voltage of A2, V.
the voltage across AI, A the open -loop gain of A". and
The input offset voltage and offset voltage drift of a dif- E1 and E2 the two input signals, then output voltage E,,
ferential operational amplifier can be held essentially to can be written as:
zero by connecting a second amplifier at the inverting E0 = - Ao(VAI-Vos)
input of the first. This auxiliary op amp must have an If VAI = Vs. then:
offset voltage and drift that are matched to the primary E=
op amp. The additional amplifier will then act as a The composite amplifier consisting of AI and A2 can

Getting rid of offset voltage. Letting amplifier AI act as floating voltage source (a) effectively eliminates offset voltage and offset voltage drift
of amplifier A2. As long as AI and A2 are matched, their offset voltages cancel, keeping offset of composite amplifier at zero. Composite am-
plifier of (a) can be used as conventional single operational amplifier for either inverting (b) or non -inverting (c) applications.

COMPOSITE AMPLIFIER

BURR BROWN
3500 MP
Ell

E21

(a)

500 - 2 000 pF

E - E,
lie used as a very -low -drift inverting amplifier, like the
one illustrated in (b). Resistors RI, R2, and R.3 are
added to decrease the effects of amplifier bias current
BURR BROWN
3500 MP on total offset voltage and drift. This particular unity -
gain composite amplifier configuration will be stable in
circuits having feedback factors as large as 50%.
(c) When used as a unity -gain non -inverting buffer, the
composite amplifier should be compensated as indi-
cated in (c). The compensation does not reduce ampli-
fier small -signal bandwidth.

21
 When the Jade is in its quiescent state (DO = 0),
Unclocked logic element XTRUE = XFALSE = 0 and XTRUE = XFALSE = 1. For
the decision state (DO = 1), Jade makes an exclusive
makes quick decisions and singular decision at the rising edge of signal DO -
XTRUE = 1 and XTRUE = 0 if X = 1, or XFALSE = 1
by Leslie K. Torok and XFALSE = 0 if X = 0. The output decision then ac-
University of Toronto, Toronto, Ont., Canada tivates the appropriate task logic.
As long as DO = 1, further changes in X do not affect
the output. Returning DO to logic 0 clears the decision,
A new kind of logic element can make logic decisions causing Jade to assume its quiescent state. It should be
without requiring a clock for synchronization. Called noted that XTRUE = XFALSE only in the quiescent state.
Jade, this asynchronous decision element can operate at For the decision state, XTRUE = XFALSE and XTRUE =
speeds as high as clocked logic blocks, offers easy de- XFALSE, since the outputs are exclusive. Those outputs
bugging, and allows sequential logic systems to be that are not selected remain quiescent.
mechanized directly from flow charts. Moreover, Jade As X drops to logic 0 and DO rises, a spike may ap-
will operate in clocked as well as unclocked systems. pear at the output of gate G1 when the propagation de-
A control signal, DO, functions much like the clock in lay of gate G2 to a logic 1 is less than the propagation
synchronous logic, while input signal X represents the delay of gate G3 to a logic 0. If the spike is wider than
logic condition that must be decided and acted upon. the minimum hold time of gate 04, a double decision is
The Jade element has two states-a quiescent state made. To prevent this, two inverters can be placed be-
when signal DO is logic 0 and a decision state when DO tween Gi and G4 to integrate the spike.
is logic 1. There are four possible outputs: XTRUE (X is Jade can sort decisions at speeds of at least 10
true), XTRUE (not XTRUE), XFALSE (X is false), and megahertz, with signals X and DO having pulse widths
XFALSE (not XFALSE). of about 30 nanoseconds.

Decisions, decisions. Asynchronous decision element named Jade uses control signal DO instead of clock to gate information signal X.
When DO is logic 0, circuit is in quiescent state; when DO is logic 1 , circuit is in decision state and provides single exclusive output out of four
possibilities. Inverters can be added to avoid switching spike that causes erroneous double decision. Truth table shows logic characteristic.

00 P..
)0o `'So 00

XTRUE

X TRUE
X TRUE
ALL GATES:
TEXAS FALSE
INSTRUMENTS
TYPE SN7410
X TRUE X FALSE

DECISION STATE, DO = 1 QUIESCENT STATE, DO = 0

INPUT
X
X FALSE X FALSE X TRUE X TRUE X FALSE X FALSE X TRUE. X TRUE
ii.

0 1 0 0 1 0 1 0 1

1 0 1 1 0 0 1 0 1

22
 verse -biased by a voltage level equal to the amplitude
of the analog sample before its termination. Transistor
Height -to -width converter Q4 is then cut off, and for a period of time that is pro-
portional to the stored analog sample amplitude, a cur-
digitizes analog samples rent source formed by Q3 and the slope -control poten-
by Roland J. Turner
tiomteter linearly discharges capacitor C1.
RCA Corp., Missile and Surface Radar division, Moorestown, N.J. During the time that Q4 is off, the converter generates
a pulse that has a width proportional to the amplitude
of the analog sample. When the stored charge goes to
By controlling the charge on a storage capacitor, a tem- zero, diode DI and transistor Q4 are again turned on by
perature -stabilized height -to -width converter can pro- the current source. After Q4 conducts, a new sample
duce a gray code output from an analog input sample. may be processed. Transistors Q6, Q7, and Q8 act as
The copverter uses a differential diode -transistor ar- pulse shapers to yield the desired output.
rangement to operate over a temperature range of Diodes DI and D2 and transistors Q4 and Q5 are con-
-55°C to +65°C, and its conversion error is less than nected in a differential configuration to keep Q4's con-
0.15 microsecond for a full-scale output pulse width of duction interval independent of temperature variations.
3.25 ,us. The voltage drops of DI and D2 and the base -emitter
During the first half of the input sample, a clear pulse voltage drops of Q4 and Q5 track each other as tempera-
removes all charge from storage capacitor C1. During ture varies.
the second half, a charge proportional to the sampled The converter in the diagram is designed to operate
analog signal is placed on this same capacitor through with a peak -to -peak video input level of 6.5 volts. Max-
transistors Qi and Q2. imum output pulse width is determined by the slope ad-
Current source Q3 keeps diode D1 forward -biased justment, which is set to provide a pulse width of 3.25 us
and transistor Q4 fully on during the sample time. On for a an input video level of 6.5 v. The waveforms
the trailing edge of the analog sample, DI becomes re- shown represent the maximum level of the gray code. 0

Compensating for temperature. Differential hook-up of transistors al and 05 and diodes D1 and D2 maintains temperature stability of height -
to -width converter. Amplitude of analog input sample is converted to gray code output. Second half of input sample charges capacitor C1,
then linear current ramp through transistor 03 discharges C1. During discharge time, D1 and Q. are off, and output pulse is produced.

-12 V
+12 V
1 kS-1.
100 kil
SLOPE
ADJUSTMENT 10 kS2 251118
INPUT
SAMPLE tEl
2.2 at
a(A)
2N1118 0.01µF F--
2N338 200 pf
(D)
4.7 kl?. 1 kS.2
C, 1N995
25706
04
2N706
22 Idt 2N706
270 pF +12 V
151914
+12 V 15914
100 pF

-12 V j_ I (B)
1.2 kS1 1N914
--1 0 lips I.-
65 \I ANALOG
{A1 -12 V
SAMPLE 25706
0V 3 kS2
CLEAR

(B)
0- - -01 0 2 ps
-12 V
CLEAR
-8V
0V
C)

OUTPUT

(0)8 V +
(F)
0V

-1.5 V --11 V

11V

0V
rd--- 3.25 ps -12 V
0V -12 V
(F) rOUTPUT
V

23
 when a conventional filament transformer is used at the
output. With the components shown, an output current
Filament transformer output of about 250 milliamperes is obtainable.
drops cost of 400 -Hz supply The amplitude of the input sine wave depends on the
amount of feedback in the amplifier network. If the
by Glen Coers feedback factor is low, a small signal can drive the am-
Texas Instruments, Components Group, Dallas, Texas plifier, but the output driving impedance becomes high,
possibly causing current limiting in the output stage and
therefore poor voltage regulation. If the feedback is
Power supplies with a 400 -hertz output are often high, a higher level of input voltage will be required,
needed in testing servo systems and aircraft equipment, but the output driving impedance becomes lower and
but they can be expensive to build when their output regulation is improved.
voltage must be on the order of 115 volts, root mean Here, op -amp closed -loop gain (A,,) is 10, making
square. This being the equivalent of a peak -to -peak the required drive voltage around 1 v rms. The value of
voltage of 325 v, the circuit transistors would have to feedback resistor 1Zr is determined by:
have very high operating voltage ratings, and since Rf = Rs(A-1) = 9 kilohms
there are no integrated amplifiers that can handle where Rs is source resistance. Feedback factor /3 is set
± 160 v, a discrete amplifier would be required. by Rf and Rs:
Alternatively, the number of parts and component = Rs/(Rs+Rf) = 0.1
costs can both be considerably reduced by generating while closed -loop output resistance Rout becomes:
the 400 -Hz sine wave at some low voltage level and then Rout = Ro/A4 = 0.01 ohms
stepping it up with a transformer. This approach allows when open -loop output resistance Ro equals 50 ohms,
low-cost transistors and integrated circuit operational and open -loop gain A, is 50,000. This last equation in-
amplifiers to be used, yet it produces enough output dicates that circuit regulation should be adequate be-
power to operate small motors, servos, resolvers, and cause the effective driving impedance is much lower
synchros. Larger output transistors and a larger trans- than the load impedance.
former will, of course, increase output power. The output transistors are connected in a bootstrap
The audio oscillator of (a) provides the sine -wave in- arrangement, eliminating two base -emitter voltage
put for the amplifier of (b). The frequency -determining drops and allowing more ac voltage to be developed. A
components for the oscillator are resistors R1 and R2 Darlington configuration could be substituted, but there
and capacitors C1 and C2. These are returned to the would be a 5% drop in the available output voltage. Ad-
non -inverting input of an op amp that functions as the justing the 400 -Hz drive voltage varies output voltage
circuit's oscillating element. between 0 and 144 v rms.
Voltage gain for the amplifier is supplied by an op
amp, while discrete transistors supply current gain. The
input sine -wave frequency can vary from 60 to 400 Hz

Servo supply. Amplifying low-level high -frequency sine -wave input cuts parts and price of 400 -hertz 115 -volt rms power supply. Audio os-
cillator (a) provides 1-V rms sine wave for amplifier (b). Standard filament transformer delivers output currents of up to 250 milliamperes and
voltages as high as 144 V rms. Bootstrap arrangement of amplifier's output transistors optimizes available output voltage.

+15V
68 812
vw- 150 S2
1pF 1N914 (v. W)
(a) (b) TIP 34
2N3819
910 S2
18 kS2 TIP 29
R
9.1 k
1 MS2 F
+15 V Us WI

2 FILAMENT
1.8 kS?. 680 St TRANSFORMER
1 kS2 SN72 41 6
W) (% WI
3
OUTPUT
6.3 V (115 V rms,
rms
TIP 30 400 Hz)
6.8 812
1 l(S2 (1/4W)
01_ -15 V SINE WAVE
0.04pF 10 k(l OUTPUT SINE WAVE
INPUT 150 sa
1 kS2 1400 Hz) TIP 33
(0 -10 V rms, (14W)
-.- 400 Hz)
0.004 pF 100 812
-15 V
TRANSISTORS AND OP AMPS: TEXAS INSTRUMENTS

24
 age because it is in series with the collector of Qi. This
allows Qi to operate at the difference voltage between
Regulating high voltage with e0 and the voltage across the zener.
low -voltage transistors The output voltage is divided by resistors Ri and R2
and applied to the non -inverting input of an oper-
by Mahendra J. Shah ational amplifier, which functions as a comparator. The
University of Wisconsin, Madison, Wis. op amp compares el with e2, and maintains output volt-
age eo so that ei = e2. Regulated output is a function of
the comparator's inverting input voltage, e2, and the re-
High -voltage regulation usually calls for high -voltage sistors of the voltage divider:
transistors. But, by absorbing the bulk of the output eD = e2(1 +R2/Ri)
voltage with a zener diode, only relatively low -voltage Another diode, D2, protects Qi's base -emitter junc-
devices are needed. The circuit illustrated regulates 250 tion from reverse breakdown. During normal operation,
volts with a 90-v transistor; however, the same concept D1 is forward -biased, allowing Qi to receive base drive
can be applied to regulating voltages in the kilovolt for proper regulation.
range. For the component values shown, voltage regulation
Transistor Qi operates as a shunt regulator, monitor- from no load to full load (25 milliamperes) is less than
ing the output voltage, eo, across the load. Without ze- 0.04%. Unregulated voltage is 330 v, and the voltage
ner diode D1, Qi would be subjected to nearly all the across Qi's collector -emitter junction is approximately
output voltage. But Di absorbs a good part of this volt- 40 v. 0
Putting a zener to work. Zener diode D1 handles most of 250 -volt regulated output, permitting a mere 90-V transistor to be used. Transistor
01 acts as conventional shunt regulator for load resistance. Op -amp comparator maintains output voltage to keep its inputs el and e2 equal,
thereby providing proper base drive for 01. Regulated output voltage is: e0 = e2(1 + R2/R1). Unregulated output of 330 V is also available.

UNREGULATED
+15 V OUTPUT (330 V)

3 kS1

e0

(250 V)
§ 4.7 kL2 0.0022 µ F 'LOAD
DI
2
1 ME2
500 S2 1 (50 V) (200 V)
+15 V
2 8
D2 0,
MOTOROLA 7 LOAD
10i 10 kS7
MC 1439G 40409
4
1N4245 (90 V)
3
1N4736
r
20 kE-2
100 kE2 2722
-15 V

Unusually long time constants can be generated with

Op amps multiply considerable accuracy by combining readily available
low -value resistors and capacitors with a couple of gen-
RC time constants eral-purpose operational amplifiers. Besides being phys-
ically smaller than their higher -value counterparts, low -
by Quentin Bristow value components offer tighter value tolerances and do
Geological Survey of Canada, Ottawa, Ont., Canada not have leakage or polarity problems. Furthermore,
when high -value resistors are used, field-effect-transis-

25
tor -input op amps must be employed. They are more closed -loop gain of N+ 1, since amplifier A2 is simply a
expensive than general-purpose op amps and do not unity -gain inverter.
provide as good input offset and temperature drift spec- The offset null trimmer permits the E1 output to be
ifications. set initially to zero for a zero input. Generally, the trim-
A number of instrumentation applications require mer can be omitted for values of N less than 50. To
time constants in the order of seconds or minutes. Cir- avoid a tedious time lag in circuit output response when
cuit (a), for instance, can be used to stretch one-shot making this adjustment, one end of the capacitor should
output pulses, or as a low-pass insertion filter for moni- be disconnected temporarily.
toring slowly changing meteorological, oceanographic, Gains other than plus or minus unity can be obtained
or other geoscientific phenomena where low -frequency at outputs E1 and E2 by making the input attenuation
noise is undesirable. This network can multiply a basic and feedback ratios unequal; they are both 1/(N+ 1)
RC time constant by a factor as large as 10,000. (For ex- for circuit (a). Also, the inverting gain of amplifier A2
ample, a 100 -second time constant can be realized with can be other than unity. As shown in circuit (b), input
R = 100 kilohms and C = 0.1 microfarad.) attenuation can be controlled by ratio a, feedback by
When V, is a step input, output voltages E1 and E2 ratio y, and inverting gain by ratio /3. The two outputs
rise exponentially to final values -V, and + Vi, respec- become:
tively, with a time constant (taken at 63% of the final El = -V,(7 + 1)[1 - exp(-01/(,8 + ORC))/(a + 1)/3
level) of (N + 2)RC. E2 = Vi(y+ 1)[1 - exp(-t/34/3+ y + 1)RC)[/(a + 1)
E2 = -E1 = - exp(-t/(N + 2)RC)] In applications where desired drift and noise specifi-
The actual values of resistors R1 and R2 are not critical cations cannot be met by a 747 -type op amp, amplifier
because the time constant is determined by ratio N and Al can be stabilized with a temperature -controlled dif-
the values chosen for R and C. The components indi- ferential preamplifier, such as Fairchild's AA727B. This
cated provide a time constant of 50 seconds. integrated circuit has an on -chip proportional tempera-
The drift and noise of either output referred to the ture regulator, affording tight control of chip tempera-
original input V, will be the same as that obtained when ture at about 100°C. The 727 -plus -747 combination pro-
amplifier Al is operated at a closed -loop gain of N+ 1, vides excellent dc stability at high closed -loop gains and
modified of course, by the filtering effect of the time can be treated circuitwise as a single op amp. If a
constant generated. After capacitor C is removed, the preamplifier is added, the null offset trimmer is no
circuit can be seen to be an op amp (A1) connected for a longer effective.

Extending RC time constants. Low -value resistors and capacitors and two op amps can generate time constants that are several minutes
long. Output voltages E1 and E2 exponentially approach level of step input Vi. Time constant, which is 50 seconds for circuit (a), primarily de-
pends on R, C, and ratio N. For circuit (b), there are three controlling ratios: a for input attenutation, y for feedback, and /3 for gain.

R)

(a)

26
 until the input period becomes shorter than that of the
monostable.
One-shot/flip-flop pairs To determine whether an input frequency (f) falls be-
detect frequency bands tween two known frequencies, f1 and f2, two one-
shot/flip-flop combinations are required, as shown. The
by Edward E. Pearson top pair of devices detects an input greater or less than
Opelousas, La. f1, while the bottom pair detects an input greater or less
than f2. The AND gate provides a high output when the
input frequency lies inside the preset band (less than
A retriggerable monostable multivibrator and a type D or greater than f2, if f1 is greater than f2). This detection
flip-flop can form a simple reliable frequency compara- scheme can be expanded to include any desired number
tor that senses if an input frequency is greater than or of segments within the operating passband.
less than a predetermined reference. Connecting addi- The frequency band detector also has an inherent
tional comparators in parallel, together with AND logic, memory function that could be particularly useful in
permits the detection of input frequencies that fall control applications. When the input signal terminates,
within selected bands. for example, with a tone burst, no trigger is available to
Both the one-shot and the flip-flop are wired for posi- the flip-flops, and all outputs remain static until the in-
tive edge triggering. Each input pulse causes the monos- put signal returns.
table's output to go high for the period of its preset tim- Although the detector responds only to the period of
ing interval. The flip-flop is triggered simultaneously, the input signal and does not require the input to main-
but its output is determined by the state of its D input at tain a specific duty cycle, input pulses must have a rapid
the time of trigger threshold. rise time. All trigger thresholds must be reached within
If the period of the input frequency is shorter than an interval that is appreciably less than the mono -
the preset timing of the monostable, a constant high stable's propagation delay time. Circuit speed is limited
level will be present at the D input, forcing the flip- only by the setup and hold performance of the compo-
flop's Q output to remain high. If the input frequency nents being used.
period becomes greater than that of the monostable, the
D input will go low prior to the next incoming trigger.
The flip-flop's Q output then goes low and remains low
Sensing frequency. Retriggerable one-shot and flip-flop compare frequency of input to preset reference frequency. To form frequency -band
detector, two frequency comparators and AND gate are needed. Depending on period of input pulse train, each one-shot output is high or
low. Each flip-flop triggers to level seen by its D input prior to trigger threshold. AND gate output goes high when f falls between fi and f2.

SET FOR
f1 PERIOD -

INPUT
71 r
JULn 2
O
13

FLIP-
If Hz) ONL
SHOT FLOP
3
CP

1
V
SET FOR
f2 PERIOD
ti

10
O D N
ONE- FLIP
SHOT FLOP
9
CP 0 f < f2

TEXAS INSTRUMENTS
y SN74123 SN7474
TO ADDITIONAL SECTIONS
(FOR 13, fa, ... fN

27
 gate G5, which is synchronized by the oscillator.
Control one-shot divides After one pulse of the reference frequency passes to
the output, the one-shot locks out gate G5 for a period
frequency by up to 30 of time determined by the setting of potentiometer Rt.
When the one-shot resets, another single pulse reaches
by Jerome Snaper the output, and the cycle repeats.
Leach Corp., Controls Div., Azusa, Calif. The input frequency, fln, is simply a multiple of the
output frequency, fout:
fin = Nfout
A three -gate control allows precision frequency divi- where N is the division factor. N can have any integral
sions of up to 30 merely by changing a resistance. A value between 2 and 30. The circuit shown divides a
crystal oscillator acts as the frequency source so that all 1.1 -megahertz reference frequency by 11 to yield an
subharmonics of the reference frequency have crystal output frequency of 100 kilohertz.
stability. Additional versatility is possible by substituting a
NAND gates G1 and G2 and the crystal comprise the field-effect transistor or voltage -variable resistor for the
oscillator that generates the reference frequency. The potentiometer. Then, frequency divisions can be elec-
one-shot, consisting of NAND gates G3 and G4, controls tronically swept over a wide range.

Precision division. Crystal oscillator supplies reference frequency for three -gate divider scheme. NAND gates G3 and G4 form one-shot that
controls gate G5. After G5 passes single reference pulse, one-shot inhibits this gate for period selected by adjustment of potentiometer R.
When one-shot resets, another reference pulse passes to output. Crystal frequency can be divided by up to 30 without loss of stability.

560 E2

(fIN =1.1 MHz)

11111111111 (four = 100 kHz)

cs I

N = 11

0.01 pF

0
Cl 1.1 MHz
0.01 mF
-p.
8 - 50 pF 1.5 kit

ALL GATES: 7400

22 pF

one -thousandth of the operational requirement. In

Pulsed standby battery many systems, this power -saving scheme allows inex-
pensive batteries to be used with essentially shelf -life
saves MOS memory data longevity. In some cases, merely a capacitor can supply
enough standby power to sustain memory until an aux-
by K. C. Herrick iliary power source can come on-line.
Fisher Berkeley Corp., Emeryville, Calif. A typical RAM cell consists of a cross -coupled multi -
vibrator with active mos transistor loads. When there is
a power failure, the potential across the multivibrator
A simple pulsed battery supply can reduce standby declines toward zero. As long as its inputs are kept off,
power necessary for mos random-access memories to the cell is isolated from external influence, except for

28
leakages, when the mos transistors cease conducting. plied. The memory's peripheral -circuit supply voltage
The charges remaining on the gates of the cross - VD can become zero and remain zero during the loss of
coupled transistors then begin to leak off exponentially. main power, without any effect on data retention.
If power is reapplied within a short period of time, how- While the memory supply voltage is decreasing, the
ever, sufficient differential charge levels remain on the memory's chip -select input must be held FALSE at the
gates to re-establish flip-flop conduction in the same Vcc ( + 5 v) potential. rrL devices or other elements
state as when power failed. driving this line may change state erratically when the
It is this cell characteristic that can be exploited to 5-v logic supply loses voltage. The presence of these
save standby power. The potential across the multi - stray signals may cause a RAM cell to change state.
vibrator must be reapplied periodically, at a rate fast A circuit for maintaining the chip -select line at the in-
enough to replenish the MOS gate charges before they stantaneous Vcc level is shown in (b). Since the 14-v (in
decline below threshold levels and for long enough to this case) semi -regulated supply is the power source for
re-establish fully the charge levels. Usually, a sufficient the 5-v regulated logic supply, in the event of power
rate is 1,000 hertz with a pulse width of 1 microsecond. failure the 14-v source will lose voltage before the logic
For any given type of memory and upper temperature supply.
limit (leakages increase with temperature), the required Transistor Qi is normally kept on by base current
duty cycle may vary. from the 14-v supply via zener diode Di. Collector cur-
Of course, the potential needed must be reapplied rent from Qi biases Q2 on. Base current from Q2 causes
within the maximum period of time that is allowable only a negligible drop below 5 v in the FALSE voltage
(for example, 1 millisecond). The little -used series mul- level of the memory chip -select line, due to the high re-
tivibrator and a 14 -volt battery (a) will do the job. The sistance ratio of R1 to R2. Whenever a zero chip -select
nominal memory cell potential is 15 v, less one diode voltage is applied to Q2's collector from the control
drop, or about 14.3 v. logic, Q2 conducts, and its emitter voltage changes to
When power fails, diode D1 disconnects the RAM'S the TRUE level (0 v).
VDD line from the main supply. The memory potential Upon power failure, the 14-v and 10-v supplies begin
of Vcc - VDD declines in magnitude until the voltage of to lose voltage. Transistor Qi's base current will become
14 -(Vcc - VDD) is large enough to start the series multi - zero when resistor R3's voltage drops below about 0.7 v.
vibrator. Oscillations usually begin when the voltage At this time, voltage V equals 10.7 v-20.7 v above
across the multivibrator becomes about 0.75 v. The -10 v or 5.7 v more positive than Vcc-a value that still
components shown yield a pulse width of about 1 ps maintains the 5-v Vcc level. When Qi stops conducting,
and a pulse interval of about 500 As. Q2 is cut off, and the memory's chip -select input line is
If a rechargeable battery is used, the diode and re- connected only to Vcc via resistor R2. Therefore, even if
sistor connected with dashed lines may be added to al- Vcc decreases in value, the chip -select line remains at
low the battery to charge while the main power is ap- the Vcc potential.

Preserving memory contents. Series multivibrator and battery (a) allow MOS random-access memory to retain data when power fails. Cir-
cuit takes advantage of MOS charge retention property so that pulsed voltage is sufficient to refresh memory. Dashed components let battery
recharge when main power is restored. Protection circuit (b) for chip -select line prevents erratic logic signals from changing memory state.

+5V
14V
Vcc

IC MEMORY
(RAM) 68 kS1t
X 1N4001
2N3645 - 0.01 F
1052
VD D
10 St
- ---D,
- - 0.01 µF 2N4141

68 kft
10V
1N4001

CHIP SELECT CHIP SELECT FROM LOGIC

TO RAM (0 V, TRUE)

(b)
27 kf2 V
(+14 V)

-10V

29
 constant) until the upper trigger level (Vu) is reached:
Vu = VE1
Triangular -wave generator VBE1 VD5
where VE1 is the voltage at Qi's emitter, VBE1 is Q1'5
spans eight decades base -emitter voltage, and VD5 is the voltage across
diode D5. Once output voltage equals Vu, transistor Qi
by William S. Shaw conducts and Q2 switches off, turning Q3 and Q6 on.
University of Texas, Applied Research Laboratories, Austin, Texas. Since R7 is larger than R6, Q4 and Q5 will switch off.
Due to the symmetry of the circuit, capacitor Ci is
discharged by transistor Q6 at the same constant rate as
Because of its nonsaturating design, a triangular -wave it was charged and by the same current, i2. The capaci-
generator can cover eight decades of frequency-from tor discharges to the lower trigger level (VL):
0.01 hertz to 2 megahertz. Lower and upper frequency VL = VE2 - VBE2 - VD6
limits are set by resistance adjustment. Circuit layout is where VE2 is Q2's emitter voltage, VBE2 is Q2's base -
not critical, and complementary circuitry assures output emitter voltage, and VD6 is the voltage across diode D6.
symmetry and amplitude stability, as well as the ab- When VL is reached, Qi turns off, Q2 turns on, and V2
sence of dc offset. By decreasing output voltage swing becomes larger than V4. This switches on Q4 and Q5,
and increasing current, the generator's frequency range causing Q3 and Q6 to switch off by feedback through ca-
can be made to span 1 to 20 MHz. pacitor C2. The cycle can now repeat.
Transistors Qi and Q2 are constant -current sources Diodes D5 and D6 allow the output to slew above and
whose outputs are switched to produce the charging and below the limits imposed by the emitter -base break-
discharging currents for the output capacitor. Transistor down voltage (about 6 v) of Qi and Q2. The generator
pairs Q3 -Q4 and Q5 -Q6 are differential amplifiers that will oscillate without latch -up as long as VE1 is less than
function as comparators; Q3 and Q6 conduct whenever V1 and VE2 is greater than V2.
Q4 and Q5 switch off, and vice versa. For a 20-v peak -to -peak output, the frequency is:
The series string of diodes D1 and D2 and resistors fout = 0.0425i1/C1
R1, R2, R3, and R4 set comparator voltages Vi and V2. Capacitor C1 determines nominal output frequency,
Diodes D1 through D4 compensate the four -transistor while resistor 113 sets the lower frequency limit and po-
integrated -circuit array so that the generator maintains tentiometer R2 sets the upper limit in addition to pro-
its frequency stability with changing temperature. Com- viding a linear frequency span of 20:1. When R2 = 500
parator input and output currents are: kilohms and C1 = 1,000 microfarads, foot = 0.01 Hz;
i1 = 28.7R1/(Ri + R2 +11,3 ±R4)R5 when R2 = 0 and C1 = 100 picofarads, fout = 2 MHz.
i2 = R7i1/R6
When Q4 and Q5 are on, Q3 and Q6 are off, and out-
put capacitor C1 charges at a constant rate (since i2 is a

High -frequency triangles. Complementary nonsaturating circuitry permits triangular -wave generator to provide 20 -volt output at fre-
quencies as high as 2 megahertz. Output triangle is obtained by charging and discharging C1. Current from 02 charges C1 until output volt-
age reaches threshold Vu, then 03-04 and 05-06 comparators switch, and 01 supplies discharge current until threshold VL is reached.

+15V
3.3kt2

Ri
kl-2
4.1 kit 1

0.1µF D3
1N914 1NI914

V3
C31
D5 0.001 µF
MPS918
C2 R2

500 kS2
10 kS-2, 1N914 OUTPUT

MPS 3640 27 kS2 R3

D6
Q2 VL --
+ 20 V
V2 CI
1N914 100 pF -1,000 Ai F
0.1 p F
D2 R
4.1

3.3 kS2
ALL TRANSISTORS: MOTOROLA
-15V

30
 If VDD = VR and R1 = R2, this equation reduces to:
T = RtCtln(2) = 0.694RtCt
Output comparator enhances The timing equations illustrate that output pulse
versatility of one-shot width is completely independent of any transistor junc-
tion voltage and therefore is independent of junction
by Harvey J. Scherr* temperature dependence. Op -amp offset voltage drift
Westinghouse Corp., Systems Development Div., Baltimore, Md. principally determines the one -shot's temperature per-
formance. And since offset drift is typically as low as 10
microvolts/°C, total variation in the multivibrator's tim-
If an operational amplifier is used as an output com- ing interval is only about 0.002%/°C.
parator, a monostable multivibrator can provide wide Unusually large time constants can be achieved be-
and accurate output pulses over a broad temperature cause the primary restriction on the value of timing re-
range. The one-shot is also retriggerable-that is, its out- sistor Rt is op -amp bias current. Rt must be small
put pulse duration can be extended by reapplying the enough to allow the voltage developed at Ai's inverting
input pulse. In addition, this multivibrator can be reset input to turn the comparator on. Since op -amp bias cur-
to accept a new trigger input within its timing period. rent generally ranges between 0.25 and 0.5 micro-
Each time an input pulse occurs, timing capacitor Ct amperes, resistor Rt can be as large as 10 megohms
is discharged by field-effect transistor Qi, and compara- (when VDD = 10 volts and el = -5 v).
tor Al switches off. In the absence of a trigger input, Cc As illustrated in the timing diagram, the multivibra-
accepts charge from Qi, and the comparator is turned tor can be re -initiated during its timing interval to
on. Switching takes place when Al's input voltages, el stretch output pulse width. This is possible because
and e2, are equal. Because the comparator is off during transistor Qi discharges capacitor Q every time an in-
the timing interval, there are no output errors intro- put pulse occurs. By adding transistor Q2, capacitor C1,
duced by op -amp input offset current. and resistor R3, the circuit can be reset by restoring it to
Output pulse width, T, is determined by the supply its stable state. A fixed -width reset pulse cancels the re-
voltage, VDD, reference voltage, VR, and the timing maining portion of the output pulse.
components, Rt and Q. At the comparator's input: Substituting a resistor for capacitor C2 alters the mul-
el = e2 = VDDexp(-t/RtCt) tivibrator's timing-a time out then occurs with the ab-
which can be rewritten as: sence of a negative voltage from the input terminal. Cir-
exp(-t/RtCt) = Voo/el = VDD/[VRR2/(R1 + R2)] cuit output polarity is easily reversed by interchanging
Output pulse width becomes: the connections to the comparator's inverting and non -
T = RtCt[In(VDD) - ln(VRR2/(Rl +R2))] inverting inputs. The active devices used determine
'Now with Stereo Equipment Sales Inc., Baltimore, Md. component values.

Retriggerable monostable. Input trigger causes Oi to discharge capacitor Ct, turning off comparator Al and producing output. Before tim-
ing period is over, new trigger can be applied to extend output duration. Reset pulse through 02 can terminate output during any part of tim-
ing interval. Output comparator maintains one-shot temperature stability and permits unusually large time constant to be used.

31
 For three points of this equation, the values of Z0 and 0
are the same; or, since there is no linearity error, then
Series resistance improves Ze = O. These points occur at 0 = 0, B = 1, and 0 = Ps.
potentiometer linearity The well-known curve (b) for a loaded potentiometer
can be written in terms of 0 and P.:
by Harry H. Schwartz Zo = 041 + 0(1 - 0)/P.]
Electrodesign Ltd., Ville Lasalle, Otiebee, Canada When the load resistance is very high, P. approaches
infinity, and the pot output is the straight black line. For
finite values of Pm, the nonlinear colored curve is ob-
Adding a series resistance to one end of a loaded linear tained. As P. becomes smaller, linearity error grows.
potentiometer can reduce loading error by a factor of A short computer program (c) that solves the general
five or more. The price for this gain in linearity is a volt- equation for impedance ratio Zo can be used to deter-
age loss across the pot. But losses in amplification or mine the effect of series resistance ratio Ps for various
drive in the order of 3 to 4 decibels can usually be toler- values of load resistance ratio Pm. Generally, the larger
ated in view of the marked improvement in linearity. the magnitude of Ps, the greater is the error reduction.
As shown in (a), voltage Ei is applied to potentiome- The program varies Ps from 0 to 2 in steps of 0.1 and 0
ter Rp and series resistor Rs. The load resistor is R., the from 0 to 1 in steps of 0.05 for P. values of 1, 2, 5, 10,
output voltage is E0, and 0 is the per -unit variation of 20, and 50.
the pot wiper. Letting: The plot of (d) shows that output linearity for P. = 1
Yo = E)/Ei, P. = R./Rp, and Ps = Rs /Rp is improved by a factor of five when Ps is increased
the voltage transfer function can be expressed as: from 0 to 0.6. In (e), linearity error is plotted against 0
Yo = OP. /(P. Ps + P. + OPs +0-62) for P. = 5 and P. = 10 for Ps values between 0 and
If Yo = Yo when 0 = 0 and Yi = Yo when 0 = 1, then 0.6. Again, linearity is greatly improved. Values of Ps
output impedance ratio Z0 is: should be held to 0.5 or less so that the series voltage
Zd = Ye/Y1 = 041 - (1 - 0)(Ps - 0)/(P. +Ps + P.P.)] loss is tolerable.

Reducing loading effects. Output of loaded potentiometer (a) becomes nonlinear, as shown by graph (b), with smaller load resistance ratio
(Pm = Rm/Rp). Increasing series resistance ratio (P. R./Flp) decreases loading error and improves linearity. Computer program (c) finds
output impedance ratio Za for several values of Pm, P,, and wiper position 0. Graphs (d) and (e) show effect of P. on 4 and linearity error.

(a) (b(

0.2 0.4 0.6 0.8

READY
PLIST
10 DATA 1, 2, 5, 10, 20, 50 (dl
30 READ P1
40 FOR P2=0 TO 2 STEP .1
45 PRINT" PM="P1," PS="P2
46 PRINT 5

47 PRINT " A","Z" 0.04 PS -0

48 PRINT P, = 0.2
50 c, 0.02
FOR A=0 TO 1 STEP 5.0 0 0 0 0 E-0 2 P5 = 0.4
60 LET Z=A/(1 - -A) (P2-A))/(P1 + P2 + P1 P2)) cc
0
70 PRINT A, Z
Ps = 0.6
80 NEXT A
85 PRINT
86 PRINT t 0.04 PrIl 10 P, = 0 Ps = 0.2
90 Ps = 0.4
NEXT P2
0.02
95 PRINT Ps = 0.6
96 PRINT 0
100 GOTO 30
-0.02 0-
110 END 0.2 0.4 0.6 0.8 1.2 v
READY
(c) RUN (e)

32
 the next clock pulse and the process repeats.
Let N be the input number; in this case, N = 11815
Circumventing BCD addition and the desired offset is 1070. The number at the out-
in digital phase -locked loops puts of the BCD adders is N + 1070, which becomes
19999 -(N + 1070) at the outputs of the complementing
by Larry Martin circuits. Therefore, the number of counts that occurs be-
Hewlett-Packard Co., Palo Alto, Calif fore the divider resets is the divider state that is sensed.
Or, the input frequency is divided by:
19999 -[19999 -(N + 1070] = N + 1070
Many of the applications for a digital divide -by -N which can be rewritten as:
phase -locked loop require the locked oscillator to be 19999 + 1070 -(19999 - N) = N + 1070, or
offset by a fixed frequency. Usually, the first stage of the 21069 -(19999 - N) = N + 1070
programable binary -coded -decimal divider circuit adds By counting to 21069, instead of 19999, the proper
the input frequency to the desired offset frequency. This number of counts still occurs before the divider resets.
BCD addition, however, can be eliminated by detecting The BCD adders, then, can be removed simply by de-
the proper number at the output of the counter divider tecting a different output state of the counter, as shown
chain. in (b).
If a receiver is tuned to 118.15 megahertz, and its in- Since the divider is an up -counter, detection of the
termediate -frequency stage is at 10.7 MHz, the local os- desired output state should be done when the proper
cillator must then operate at 128.85 wiz. A standard di- outputs are ones. Because there are only six ones in
vide -by -N circuit (a) adds the i-f frequency to the input 21069, as opposed to nine in 19999, one fewer output
frequency, takes the BCD nines complement of the sum, gate is needed.
and then loads the result into a programable counter
divider chain. Detection occurs when the counter out-
puts are all nines. The input number is then reloaded on
Frequency offsetting. Standard divide -by -N circuit (a) sums input fr.equency (118.15 MHz) with offset frequency (10.7 MHz), takes nines
complement of sum, and then detects signal when counter outputs are all nines. BCD adders and one output gate can be omitted, as in (b),
by detecting sum of nines complement and offset frequency (19999 + 1070 = 21069). Correct count is still reached before divider 'esets

(a) DIVIDER
CHAIN
I b DIVIDER
CHAIN
COMPLEMENTING
CIRCUITS Di Di STATE DETECTING
Ci STATE DETECTING C, AND
0 B,
AND COMPLEMENTING LOAD CIRCUITS
13,
BCD LOAD CIRCUITS CIRCUITS
ADDERS A,
L Cp Ao L Cp

Di
Pi 0o
C. Ci
8, Bi
8 A, Ai Aa
2 I Cp A0 L Cp

Di Do Di
C, C,
81 B, tout
0= B AL, AL, 128.85 MHz
C Cp
tout
128.85 MHz
Di
Do
- Ci
- Bi LOAD
7=
C -
- AL,
Cp
LOAD .1111.1M.
S
Di Do Di
Ci Ci
9
Bi B,
A, A,
Cp L Cp

CLOCK IN

CLOCK IN

INPUT OUTPUT

ALL LOGIC: TYPE 7400

33
 and their corresponding decade counters, and a flip-flop
formed by gates G1 and G2. More decades of frequency
Thumbwheel switches set control can be added easily.
synthesizer output frequency The number of oscillator pulses passed by the flip-
flop -controlled NOR gate, G3, is determined by the
by Jerrold L. Foote thumbwheel switch setting (S). Blank spaces occur while
University of Utah College of Medicine, Salt Lake City, Utah the decade counters are counting beyond the switch set-
ting. When these counters reach their maximum count
(M), the flip-flop resets, and the oscillator pulses pass
Two binary-coded thumbwheel switches can set the out- through gate G3 to the binary countdown string. The
put of a two -decade frequency synthesizer that in- output frequency is given by:
cludes the two switches, a single crystal oscillator, and f0 = f,S/MN
two decade counters. Synthesizer output accuracy and where f, is the crystal frequency.
long-term frequency stability are the same as that of the The output of gate G3 is a train of pulses and blank
crystal used as the reference frequency source. What- spaces that must be time -averaged to reduce the pulse -
ever pulse -to -pulse time variation occurs is minimized to -pulse time variation (AT) in the output frequency.
to an acceptable level by a chain of binary counters. This time -averaging is performed by the binary count-
The synthesis technique involves generating a series down string. The closer the tolerance that must be held
of pulses and blank spaces, divided by as many binary on AT, the larger is the number of binary counters re-
counters (N) as needed to yield the desired output fre- quired. Maximum pulse -to -pulse time variation is:
quency (f0). The frequency -selection circuit consists of AT = S/f0N
thumbwheel switches, S1 and S2, two banks of diodes Crystal frequency can be as high as 10 megahertz.

Dialing frequency. Thumbwheel switches determine number of output decades for frequency synthesizer. Single crystal oscillator serves as
frequency source. Decade counters generate pulse train that is divided by binary counter chain to obtain frequency set by switches. When
switch -fixed decade count is reached, flip-flop formed by gates G1 and G2 resets, enabling gate G3 so that pulses pass to binary counters.

CRYSTAL
OSCILLATOR ALL CIRCUITS: MOTOROLA
Vcc

4.7 k2
1N277
1N277 (ALL)

W
s,
0w 2
<H
MC 780 P Z
wD
00 4

14

Y.MC724P

S2
LU CC
'Li 2
MC 780 P z OUTPUT
p0 4

BINARY BINARY BINARY

COUNTER COUNTER COUNTER

TO ADDITIONAL DECADES MC 777 P MC 777 P MC 777 P

34
 +10 Vdc

15 k12 82012
(C)
(A) R Cl 39 pF

MODULATION (B)

1
5.6 kS2
INPUT (Wm) CARRIER
R5 INPUT (c..)c)
a2 6.8 kS-2 1N772
15 kS2
- 2N2222
(E)
2N2222
MODULATED
OUTPUT
R6 15 kS2
(mac ± wc)
R4 C2 39 pF
10 kS2

5.6 kS2 ( 131

R2 D2

820 S2 D3
2N2222 6.8 kS2 1N772

-10 Vdc

Splitting and chopping. Highly linear amplitude modulator theoretically can operate at half the carrier frequency. Transistor 01 splits modu-
lation input into two equal signals that are opposite in polarity and phase. Switch 02 passes positive half -cycles of square -wave carrier, while
switch 03 passes negative half cycles. Chopped modulated signals (points C and D) are then summed by resistors R5 and R6.

occurs at each negative half -cycle of the carrier.

Amplitude modulator Positive and negative chopped modulation signals are
then combined by a simple summing network com-
is highly linear posed of resistors R5 and R6. Signal components of the
chopped outputs that occur at modulation frequencies
by Donald DeKold are summed to zero. Therefore, under perfectly bal-
Santa Fe Junior College, Gainesville, Fla. anced conditions, the modulated output is spectrally de-
void of modulation -frequency components but, of
course, contains the modulation sidebands. Theoret-
Besides offering good linearity, an amplitude -modu- ically, this permits modulating to an upper frequency
lation circuit can operate at modulation signal fre- limit of one-half the carrier frequency without trouble-
quencies ranging from dc to half of the carrier fre- some filtering problems. The modulation envelope, in
quency. At modulation levels as high as 97.5%, the this case, is 180° out of phase with the input modulation
circuit retains its linearity. All signals are directly signal.
coupled without inductive or large capacitive elements. The circuit's output is an amplitude -modulated
Transistor Qi performs as a phase splitter for the square wave containing harmonics principally at odd
modulation signal, which appears at Qi's emitter with multiples of the carrier frequency. (Spectral content is
00 phase shift and somewhat attenuated from its input nw, ± (Om, where co, = carrier frequency, Wm = modu-
level. The dc level of the modulation is approximately lation frequency, and n = 1, 3, 5, . . .) If a sinusoidal
.

-5 v dc at Qi's emitter and +5 v dc at the collector, carrier is wanted, the output must be filtered. Since
where the signal is 180° out of phase with the input. modulation frequency components are absent from the
Transistors Q2 and Q3 are high-speed switches, driven output spectrum, a low-pass filter can be employed to
alternately from saturation to cutoff by the carrier in- select the fundamental carrier frequency and its side -
put. This signal, preferably a square wave, is applied to bands. However, a bandpass filter must be used if the
the bases of Q2 and Q3 through resistors R1 and R2 and output is to be some multiple of wc.
diodes D1 and D2, respectively. The diodes protect the Modulator high -frequency performance depends
transistors from excessive reverse base -emitter voltage largely on the speed of the switching transistors. For the
generated by possible overdrive from the carrier signal. transistors shown, useful modulated output extends to 1
Capacitors C1 and C2 speed up Q2 -Q3 switching times. megahertz. The modulator itself is essentially flat and
The collectors of Q2 and Q3 are coupled to the two linear to 250 kilohertz, with visually apparent distortion
outputs of phase-splitter Qi through resistors R3 and occurring in the modulation envelope above this fre-
R4. These isolate the modulation frequency portion of quency. At a carrier frequency of 100 kHz and modu-
the circuit from the carrier frequency portion. lation frequency of 1 kHz, good linear modulation can
The modulation signal appearing at Qi's collector is be obtained to a modulation depth of 95%.
switched from its average 5-v dc level to ground by Q2 For a modulation input signal of 14 v peak -to -peak,
on each positive half -cycle of the carrier. A chopped the maximum modulated output level will be 7.4 v
version of the modulation signal then appears at Q2's peak -to -peak into an open circuit. The minimum carrier
collector. Similarly, the modulation signal at Qi's emit- input level for a square -wave drive is 2.8 v peak -to -
ter is chopped by Q3; Q3's cutoff -to -saturation transition peak. And overdriving does not produce any unde-

35
sirable effects. The modulation can be any waveform. also drops at higher frequencies, but can be improved
A sine wave can be used as the carrier input, but by using faster switching transistors and lower imped-
chopping action will not be as good. Minimum sine - ance levels throughout the circuit.
wave drive level is 4 v pk-pk. A linear modulation Higher supply voltages offer an alternate method for
depth of 97.5% can be obtained at a carrier frequency of improving circuit output level. The saturation voltage of
10 kHz and a modulation input level of 14 v pk-pk. the chopping transistors represents the theoretical limit
Minimum carrier input drive levels remain essentially of maximum modulation depth; this voltage becomes
unchanged at a lower carrier frequency. Modulator per- less significant when higher supply levels are used. Fur-
formance, however, degrades somewhat at higher out- thermore, using precision resistor pairs (R3 -R4, R5-14,
put frequencies-maximum linear modulation becomes and R7 -R8) assures that positive and negative peak
only 94% at 500 kHz and 88% at 1 MHz. Output level modulation signals are identical at the output.

switching speed of Qi or raising the low end of the am-

FETs remove transients plifier's frequency response. However, this method is
not practical in applications requiring a broad fre-
from audio squelch circuit quency response, such as high-fidelity audio equipment.
Building the circuit with FETs (b) provides a better so-
by Glen Coers lution. Again, transistor Qi is the control, while transis-
Texas Instruments Components Group, Dallas, Texas tor Q2 is the amplifier. Only ac voltage is present at the
gate of Q2; its dc level is at ground potential so that
there is no transient generated when Q2 is switched off.
Using field-effect transistors instead of bipolar transis- When ()I's gate is at zero volts, the device itself con-
tors for an audio squelch circuit eliminates switching ducts, bypassing the audio signal to ground. Transistor
transients without sacrificing switching time. Moreover, Q1 and resistor R1 form a voltage divider that attenu-
because there are no transients, the circuit's frequency ates the signal by about 60 decibels. The "on" resistance
response can be as low as desired. of the FET is approximately 100 ohms.
In a typical audio squelch circuit (a), bipolar transis- To pass the audio input, a negative voltage must be
tor Qi acts as the control device for bipolar transistor applied to Qi's gate to turn the device off. The audio
Q2, which serves as the amplifier. When Qi turns off Q2, signal at the gate of Q2 can then reach the output de-
the base voltage of Q2 switches from a dc level (in this void of any switching transients.
case, 2.8 volts) to ground, causing a large transient out-
put voltage spike to be generated.
The problem can be minimized by slowing the

12V 12V

AUDIO AUDIO
OUTPUT OUTPUT

10 F
AUDIO
IN 11µ

1 k1.2

CONTROL CONTROL
VOLTAGE VOLTAGE

la I

FETs replace bipolars. In bipolar -transistor audio squelch circuit (a , large output transient is generated when Ch turns off, because dc volt-
age at 02's base is grounded. Substituting FET circuit (b) eliminates transients without limiting switching speed. Control FET C), turns off for
negative gate voltage, allowing amplifier 02 to pass audio signal to output. Only ac voltage is present at 02's gate.

36
 heater current. When temperature is too high, diode D2
prevents amplifier output voltage from going negative,
Diode plus low-cost op amp and transistor Qi turns off. Average current then is al-
makes accurate thermostat ways positive and is controlled by the amplifier to com-
pensate for changes in temperature.
by Robert Koss Transistor Q2 and resistor R2 prevent the heater cur-
Adac Inc., Colchester, Vt. rent from exceeding a specified limit. (For this circuit,
maximum heater current is 375 mA.) If the voltage drop
across resistor R2 equals Q2's 0.6-v base -emitter voltage,
Employing a silicon diode as the temperature -sensing transistor Q2 turns on, shorting the amplifier's current -
element allows an inexpensive operational amplifier to limited output to ground, thus turning off transistor Qi.
control temperature to a variation of less than 005°C. A The stability of the control loop depends on several
type 1N4148 diode, for instance, has a typical tempera- thermal time constants that are not easy to calculate.
ture coefficient of -2 millivolts/ °C. Computations can be reduced a little by using a rela-
The resistance bridge in the diagram sets the tem- tively large time constant in the feed -forward direction
perature that is sensed by diode D1 and that is main- for R3C1 and keeping the remaining thermal lags as
tained by the heater element, resistor R1. small as possible. Sense diode D1 should make good
When temperature is too low, control voltage Vi goes thermal contact with the object to be temperature -con-
negative, limiting the op amp's output to 12-14 volts. trolled. It should also be thermally insulated from
Transistor Qi conducts and provides the necessary changes in external conditions. El

Inexpensive thermostat. Tight temperature coefficient of silicon diode permits low-cost op amp to be used for controlling temperature to
within 0.05°C. Resistance bridge sets temperature of heater containing sense diode D1. For low temperatures, amplifier turns on transistor
01, which supplies heater current. For high temperatures, amplifier output remains positive because of diode D2, but 01 stops conducting.

+24 V

r HEATER
1
D, (RESISTANCE
R,

1N4148 88 0 1 WALIRLEdYjE451-,LIFF,
I 19.4 12/FT)
+15 V

02 1N4148

1 33 pF

R3
12.1 kE2 15.8 kE2
32.4 kst
1 k0 Qi
V, 741
2N40322
'NAN
32.4 kl2
500 E2
Q2
2N2369
L _... 1µF
160
TEMPERATURE CONTROL 2
W)

37
 The 10 most significant bits from the latches drive
both a d -a converter and a comparator formed by three
Logic system checks out quad exclusive -OR gates. These gates compare the bits
of the digital ramp to the output bits of the a -d con-
analog -to -digital converter verter, on an individual basis. The analog output of the
by Charles J. Huber
d -a converter corresponds to the input digital code
Westinghouse Electric Corp., Systems Development Div., Baltimore, Md. within ±'/8 the least significant bit.
The delayed clock pulse also passes through another
one-shot delay network before reaching a sample cir-
Testing the conversion accuracy of an analog -to -digital cuit, which strobes a second set of buffer latches. For
converter need not be a laborious and time-consuming zero error at the buffer outputs, the minimum strobe de-
task. The test configuration shown can reduce the job to lay equals the a -d conversion time. An interpolating
a go/no-go operation without undue expense (approxi- voltage applied to the d -a permits continuous voltage
mately $36 for integrated circuits plus the cost of a digi- control of the a -d output over the 1/2 -bit range.
tal -to -analog converter). The test system can accurately determine conversion
This test system produces a 12 -bit digital ramp that is times of 2 microseconds for successive -approximation
converted (by the d -a converter) to a 4,096 -step analog and variable -reference a -d converters. Additionally, the
ramp. The analog ramp is applied to the a -d converter nature and position of other conversion errors can be
under test, and the resulting output from the 10 -bit a -d determined by relating displayed error pulses to the
converter is compared to the 10 most significant bits of digital ramp. For example, small areas of the ramp can
the 12 -bit digital ramp. be investigated by making constants of the 10 most sig-
An input clock is applied simultaneously to a one- nificant bits of the digital input and using the eleventh
shot delay network and to a ripple -through counter con- and twelfth bits as variable controls.
sisting of three four -bit binary counters. The delayed More system flexibility can be obtained by using an
clock becomes the input for six four -bit buffer latches. up/down counter to eliminate the d -a converter's slew
These accept and delay all 12 of the bit outputs from time when the count changes from 111 . 1 to. .

the counter to remove glitches from the digital ramp. 000 . .0. Replacing the counter with a pseudo -ran-
.

The delayed clock now acts as the basic time reference. dom generator allows testing for all input changes.

Verifying converter accuracy. A -d converter test system generates 12 -bit digital ramp with ripple -through binary counter. Buffer latches
smooth out any ramp glitches. D -a converter then develops analog input for a -d converter using only 10 most significant ramp bits. Exclusive -
OR gates compare a -d output with ramp. Errors pass to another buffer for comparison with appropriately delayed clock.

CLOCK DELAY
(ONE SN74123)

12 12 10

11 D -A
BUFFER 10 CONVERTER A- D
COUNTER LATCHES CONVERTER
(ANALOG
(THREE 3 (SIX DEVICES
SN74197) (10 BITS)
2 SN7475) DAC 12-O)

COMPARATOR
(THREE SN7486) ALL ICs TEXAS INSTRUMENTS

BUFFER LATCHES SAMPLE DELAY

(SIX SN7475) (ONE SN74123) (ONE SN74123)

38
 tree time constants, at least two complete output
pulses with a period equal to approximately three gen-
Double -duty multivibrator erator time constants are produced. In this case, the
gives complementary outputs pulses begin and end synchronously. The first output
pulse coincides with the leading edge of the input, and
by Edward Beach the last pulse is a full -width pulse, no matter when the
National Radio Institute, McGraw-Hill Inc., Washington, D.C. input is removed.
Resistor R can range in value from 330 ohms to 1.5
kilohms, while capacitor C can be some value between
Depending on its input signal period, a pulse generator 0.001 microfarad and 1,000 µF. For example, when
operates as a one-shot or as a synchronous astable mul- R = 1,000 ohms and C = 100 [if', the circuit will pro-
tivibrator. In either mode, the circuit provides comple- duce either a 100 -millisecond pulse or a 100 -ms pulse
mentary outputs. An ordinary grounding pushbutton train with a repetition rate of about 3 pulses per second.
switch can serve as the input device. For critical timing applications, a nonpolarized ca-
Gates G1 and G2 form a simple latch that prevents pacitor should be used when C is a high value; for less -
the circuit from operating in the absence of an input critical applications, an ordinary electrolytic or tanta-
(high). When the input goes low, the latch changes lum capacitor will suffice. Since the reverse capacitor
state, allowing gate -G1 to act as an inverter. Gates G1, voltage is only about -0.7 volts, an inexpensive capaci-
G3, and G4 are the pulse -generating portion of the cir- tor can be used.
cuit. The circuit can also function as an inexpensive tran-
If the input remains low for less than three generator sistor -transistor -logic clock simply by replacing gates
time constants, single complementary pulses are pro- G1, G3, and G4 with half a type 7404 circuit. In addi-
duced by gates G1 and G4. The leading edge of each tion, gate G2 could be eliminated and an inverted ver-
output pulse coincides (neglecting gate delays) with the sion of the input signal applied directly to gate GI, but
leading edge of each input pulse. The trailing edge of this could shorten the last pulse in the pulse train.
the output pulse (again, neglecting gate delays) resets
the latch and disables the generator.
When the input signal remains low for longer than
Manual pulser. Pushbutton -operated pulse generator functions either as a monostable or astable multivibrator. Low input signal enables G1 -
G2 latch so that G1 becomes an inverter. Gates G1, G3, and G4 then generate complementary output pulses. When input is low for less than
three RC time constants, circuit is a one-shot; when input period exceeds 3RC, circuit is a synchronous astable multivibrator.

+5V

ALL GATES: TYPE 7400

INPUT
0V -7IF

< 3 RC T, , 3 RC

INPUT
0V
INPUT
0V-
-I
OUTPUT A T = RC OUTPUT A
0V 0V-
OUTPUT B OUTPUT B
0V .11116 0 V --
T= 3 RC -4.1

39
 The Schmitt trigger delays the Q output of the flip-
flop for the time fixed by resistor R and capacitor C; it
Schmitt trigger prevents does not delay the flip-flop's (5 output. The delayed and
clock train overlap undelayed pulse trains then pass through a combination
of NAND gates, producing the two desired non -over-
by R.R. Osborn lapping output clocks.
Roberts Enterprises, Flagstaff, Ariz. Interclock intervals, which occur when both clock
outputs are high, are unequal if resistor R alone sets the
delay, because of the flip-flop's output levels and the
Non -overlapping clock trains are often required in digi- Schmitt trigger's input current. Adding resistor R' and
tal systems, especially when transistor -transistor -logic two diodes, as shown by the dashed lines, allows the in-
circuits must be interfaced with metal -oxide -semicon- terclock intervals to be made equal to each other. The
ductor logic circuits. A single integrated Schmitt trigger, value of R' can range from 0 to 5 kilohms.
using only one RC time constant, can provide the ap- Capacitor C can vary from 0 to 1,000 microfarads,
propriate delay between clock trains. Moreover, the producing interclock intervals of 30 nanoseconds to 1
temperature stability of the IC Schmitt trigger assures second. The time between input clock pulses must al-
that the separation between output clocks remains con- ways be greater than the output clock interval; input
stant, despite changing temperature. clock frequency can be as high as 10 megahertz.

Staggering clock phase. Circuit produces two non -overlapping output clock trains from single input clock to flip-flop. Q output of flip-flop is
delayed by Schmitt trigger for one RC time constant. Delayed clock from Q and undelayed clock from 5 are combined by NAND gates to yield
separate output clock trains. Adding dashed components yields equal interclock intervals. Clock speed can be as fast as 10 MHz.

CLOCK 1
7400
CLOCK FLIP-
IN FLOP CLOCK 2
7473 S QS
A.A.A,

2202

Y, 7413
vs, SCHMITT
1N277 TRIGGER

DELAY

CLOCK IN

(7)

INTERCLOCK
INTERVAL

40
 lable and more clearly defined than transistor satura-
Broadband cutoff limiter tion parameters. And with cutoff signal limiting, the
transistors look like high impedances to low load im-
is phase -transparent pedances, thereby achieving fast limiting action with
controlled passive elements.
by Roland J. Turner Transistors Qi through Q4 perform the limiting func-
RCA Missile & Surface Radar Division, Moorestown, N.J. tion, while transistors Q5 and Q5 operate as a unity -gain
output driver. There are two limiting sections, one con-
sisting of Qi and Q2, and the other of Q3 and Q4. Two
When information is transmitted in the phase domain, transistor stages, then, form each limiting section.
the video or intermediate -frequency processor in a ra- For one polarity of the input signal, one transistor
dar or communications system frequently requires a stage operates as a low -gain broadband amplifier, while
limiter circuit that does not alter the zero crossings of the other limits the section's output by performing as a
the input signal. cutoff isolation amplifier. During the opposite polarity
By using current cutoff limiting, a broadband phase - of the input signal, the transistor stages reverse roles.
transparent (zero phase -shift) limiter can be built that Each limiting section supplies a gain of 3.3. The output
maintains input zero crossings within 14 picoseconds, level of each section is determined by its quiescent oper-
while providing a gain of 20 decibels over its linear ating point.
range. This limiter, which operates from dc to 30 mega- This limiting scheme provides extremely low carrier
hertz, can improve receiver sensitivity, allowing smaller output phase shift for the full dynamic range of the in-
targets to be resolved in a radar system or, in a commu- put signal. Over a 40 -dB input range, from 100 milli-
nications system, reducing level- and frequency -depen- volts peak -to -peak to 10 volts pk-pk, the limiter circuit
dent phase noise so that phase -detection thresholds can is phase -transparent within 0.25°. For example, the out-
be lowered. put is 1.2 v pk-pk for a 120-mv input. For frequencies
The limiter circuit uses microwave transistors that up to 20 MHz, the output impedance of the driver sec-
have a unity -gain crossover frequency (fT) of greater tion is less than 5 ohms.
than 1 gigahertz. The bandwidths of the transistor From dc to 20 MHz, the time displacement of adja-
stages making up the limiter can then exceed 500 MHz cent zero crossings of the output waveform are within
to yield the limiter's over-all wideband performance by 14 picoseconds of the period established by the input
using conventional microstrip techniques. zero crossings. In the phase domain, this means that the
Exceptional signal control is realized by driving the phase of a 20 -MHz input will be shifted less than 0.1° at
transistors into their cutoff regions to achieve limiting the output, making the limiter phase -transparent for all
action. Transistor cutoff parameters are more control- practical purposes. CI

Linear -phase signal limiting. Operating from dc to 30 megahertz, limiter circuit remains phase -transparent within 0.25° over 40 -decibel in-
put dynamic range. Each limiting section contains two transistor stages. Depending on input signal polarity, one stage is low -gain broadband
amplifier, while the other acts as cutoff isolation amplifier. Optimum usage of transistor cutoff parameters achieves desired limiting action.

+12 V

68.1 S! 68.1 13
68.1 12 316 62

0.001 µF 0 0010 0.001 µF j- C

-I
0.001 1.1F

Hf- H
2N5829
L L
INPUT 1N936 1N936 1N936 1N9
L 36

2N285/ 2N2857 2N2857 2N2857

15 mA
2N2857

12.1 kE2 18 S2 12.1 k62 18 62 12.1 kS2

10 mA

C
3.48 kl2 3.48 k62 1.10 k62 1.10 kS2 56212
51-E2
3 3 mA LOAD
20 mA
1µH
-12 V
SUPPLY
0.001 Pr
17 Vdc, 50 mA 0.001 ,F
12V
0.001 µF
-12 V
0.001 µF
-12 V
0.001 /IF
= C: 1µF NON -POLARIZED
MYLAR IN PARALLEL
WITH 1000-pF LOW -
INDUCTANCE BYPASS
LIMITING LIMITING OUTPUT
SECTION 1+1 SECTION DRIVER

41
 FF1's preset function high, as well as one of the inputs
to G4. The output from G5 is now low.
Preset generator produces The first negative edge of a clock pulse triggers the
desired number of pulses flip-flop and makes Q go high. The output of G3 then
goes low, presetting the flip-flop to maintain Q in the
by Glen Coers high condition. When G3's output goes low, one input
Texas Instruments, Components Group, Dallas, Texas to 04 also goes low, enabling this gate so that clock
pulses are passed to its output.
The binary -coded -decimal count that is accumulated
Computer systems and medical instruments are likely in the divide -by -10 arrangement of decade counters is
applications for a digital pulse generator that will de- transferred to the BCD -to -decimal decoders. A de-
liver, on command, any desired number of full -width coder output goes low when that decoder reaches the
pulses from 1 to 999. Three 10 -position switches set the setting of its associated rotary switch. When all three
number wanted. decoder outputs are low, gate G5 is inhibited (its output
Rotary switches Si, Sz, and S3 fix the number of out- goes high). This causes one input to G4 to go high, stop-
put pulses. Si controls the most significant digit, while ping the transfer of pulses.
S3 controls the least significant digit. For example, if Si Placing switch S4 in its reset position drives one input
is set to 3, S2 to 6, and S3 to 8, the number of output to gate G2 low, making its output go high and operating
pulses will be 368. the reset function of the decade counters. Both inputs of
Gates G1 through 04 eliminate any count error Gi are now high, while its output, the clear function of
caused by contact bounce from toggle switch S4. When the flip-flop and one input to G3 are low. This clears Q
S4 is placed in its count position, flip-flop FF1 inhibits to a low condition, causing Gi's output, FF1's preset
gate G4 until the trailing edge of S4's count pulse occurs, function, and one input to G4 to go high. The circuit is
so that even the first output pulse is full width. now ready to start a new count.
Placing S4 in its count position results in a high at Adding more counters, decoders, and switches will, of
both inputs to G2, enabling this gate (its output goes course, increase the number of pulses that can be
low) and resetting the decade counters to zero. The counted. 0
count command also causes one input of G1 to go low,
making its output high. This clears flip-flop FF1 and
keeps its Q output low. The high from Gi also drives
Pulse counter. Three rotary switches control number of output pulses that can be generated; settings may range from 1 to 999. Flip-flop and
gates G1 through G4 assure that all output pulses are full width. Toggling switch S4 to its count position allows gate G4 to transfer clock pulses
to divide -by -10 arrangement of decade counters. Each BCD -to -decimal decoder counts to setting of its rotary switch.

OUTPUT
CLOCK COUNTER

lOs 100s
1/3 SN7427
RO RO RO

04 BD SN7490 BD SN7490 BD SN7490

SN7472 DCBA DCBA DCBA
CLEAR
FF1
DCBA
II
DECODER
10 .11111111
DCBA DCBA
CLOCK SN7442 SN7442
SN7442

PRESET
0123456789 0123456789 0123456789

1/4 SN7400 3 - s, S, -
+5 V
181 (6) (31
4.7 IcEl
1/4 SN7400
1,3 SN7427

S4 COUNT IIIb
T
RESET

1/4 SN7400 CLOCK

4 7 kS-2 START COUNT
STOP
ALL ICs: TEXAS INSTRUMENTS COUNT
Gi OUTPUT
+5 V
G4 OUTPUT
WHEN 5, =0, S2 = 0, S3 = 4

42
 can vary in frequency from 40 to 600 kilohertz. The out-
put is a square wave at twice the input frequency, hav-
Frequency doubler ing a duty cycle of about 38% for the lower -frequency
accepts any waveshape inputs to'around 65% for the higher -frequency inputs.
Comparator Ai operates in its noninverting mode, ac-
by Donald DeKold cepting the input sine wave and producing a square
Santa Fe Junior College, Gainesville, Fla. wave at its output. Capacitors Ci and C2 couple this
square wave to a pair of complementary switches, tran-
sistors Qi and Q2. The switches boost the amplitude of
The frequency of nearly any waveform can be doubled the square wave to 24-v pk-pk. Resistors Ri and R2 pre-
by means of quadrature square waves that drive a bi- vent the bases of Qi and Q2 from being clamped to dc
conditional logic circuit. Only two restrictions must be voltages that would drive them into cutoff.
imposed on the input to this frequency doubler-the Capacitor C3 is alternately charged and discharged
wave form must have a duty cycle of approximately through resistor R3. When Qi is in saturation and Q2 in
50% and a peak -to -peak amplitude of at least 0.5 volt. cutoff, C3 exponentially charges towards 12 v dc; with
For the circuit shown, the input is a sine wave that Qi cut off and Q2 saturated, C3 discharges towards
Frequency times two. Sine -wave input is converted into square wave by comparator Al. Emitter -followers OI and 02 charge and discharge
capacitor C3, producing triangular wave that drives comparator A2. Resulting square -wave output of A2 is in quadrature with square -wave
output of Al. Switches 05 and 06 conduct only when comparator states are different, providing square wave that is twice input frequency.

+12 V dc

OUTPUT
2N2222 68052 SQUARE
2.4 k12 WAVE
Q3

INPUT
01
SINE WAVE
2N3905 1N662 06
(40 - 600 kHz)
91052 0.22p F 04 2N2369A
220 2 CI-
2 0.1 p F R3 HI (C)
2 2N2222
2.41(2
A2
Al (A) ( 8 )41-W0- C4
1 kS2 3 ( Dl 1N662
C2
0.1 p F 2.4 k12 2N2369A
91052
22Q Sl C3 6.8 6.8
k,f2
Q2 0.012µF "2
2N2222
R2
2.2 k12 2.2 kl2
2.4 k2
1N753A
1 kS2

COMPARATORS:
FAIRCHILD pA710
-12 Vdc

INPUT 1 V pk-pk 14 V pk-pk

(C) MAX AT 40 kHz

3.2 V ---
(A)
( DI
-0.5 V -

12 v ---
12 V - - OUTPUT 8.5 V

(B)

-12 V ---

43
 -12 v dc. When the charging or discharging interval is transistor's base -emitter loop prevent false switching
short compared to the R3C3 time constant, the vim' age when both comparators are in the same state, but may
across capacitor C3 approximates a triangular wave that have different output levels.
has its peak value occurring 90° out of phase with the A square wave with a 50% duty cycle can be realized
peak amplitude of the input sine wave. for a nominal input frequency of 120 kHz. At lower fre-
The triangular wave is applied to the noninverting in- quencies, the duty cycle is smaller because the triangu-
put of comparator A2 through capacitor C4. The com- lar wave becomes exponentially rounded. At higher fre-
parator "squares" the triangular wave about its zero quencies, the duty cycle is larger, since the amplitude of
crossings, producing a square wave that is 90° out of the triangular wave decreases, thereby shortening the
phase with the output of the first comparator. duty cycle of the square wave at the output of the sec-
Transistors Q3 and Q4 are emitter -followers that act ond comparator. Also, circuit delays become significant
as buffer amplifiers for both comparators and drive a compared to the period of high -frequency inputs.
set of nonsaturating switches, transistors Q5 and Q6. The frequency doubler can operate below 40 kHz if
When comparator states are the same, Q5 and Q6 are higher capacitor values are used throughout the circuit.
off; when comparator states differ, Q5 or Q6 conducts. To operate at higher frequencies, faster switching de-
The switching action of transistors Q5 and Q6 is vices must be used, and transistors Qi and Q2 must not
equivalent to the biconditional logic function: be allowed to saturate.
XY + XY = 1 Since the maximum input voltage for the type p.A710
which has the effect of doubling the frequency of comparator is ±7 v, the amplitude of the triangular
quadrature square waves. A diode and resistor in each wave must not exceed 14 v pk-pk.

The video gate basically consists of a switching field-

Op amp cancels video effect transistor (Q1), its associated TTL-compatible
driver (bipolar transistor Q2), and an output op amp.
switching transients Transients are generated when the switching waveform
edges pass through the FET'S gate -drain junction.
by Steven E. Holzman By adding a second FET (Q3) and its biasing resistors,
Electromagnetic Systems Laboratories, Sunnyvale, Calif.
the transients can be eliminated by forcing them to can-
cel through the op amp. Since the transients from tran-
The common -mode rejection of an ordinary operational sistor Qi will be identical to those from transistor Q3,
amplifier can help to minimize switching transients in the op -amp's common -mode rejection stops these un-
low-level video gates. The transients are caused by the wanted spikes from reaching the output.
leading and trailing edges of the switching waveform. Capacitor C1 helps maintain equal feedthrough char-
Transient-fres video gate. Leading and trailing edges of gating waveform create transients when passing through first FET, 01. Including a
second FET, 03, duplicates signal path of Oland permits op -amp common -mode rejection to cancel most transients. Capacitor C1 equalizes
feedthrough characteristics of the two FETs so that matched pair is not necessary. Bipolar transistor 02 makes circuit TTL-compatible.

2 kS-2,

0,
2N4860
1 kS2
VIDEO SIGNAL INPUT
GATED
1 kS2 VIDEO
OUTPUT
CI
5 pF NATIONAL

03
1N4148
2N4860
6.2 kS2
-12 V
RI
56 pF
INHIBIT 1

+5V
02
0 ON 4.7 kS2 2N2907
GATING WAVEFORM
1

+5 V

44
acteristics for both Qi and Q3. This eliminates the need In the circuit shown, the amplitude of the switching
for a matched pair of FErs. By simply adjusting resistor transients can be reduced to less than 1 millivolt over an
R1, nearly all of the transients can be cancelled. operating temperature range of -20 to + 60°C.

switches, causing its gate voltage to swing in the nega-

tive direction. This produces a positive -going pulse
Pulse generator accuracy across cathode resistor R5, which turns on transistor Q2
and discharges capacitor C1.
is immune to aging The function of Q2 is twofold. It rapidly discharges C1
by Frank Cicchiello
by passing the discharge current around the anode -
Digilog Systems Inc., Willow Grove, Pa. cathode junction of the PUT. And it also effectively elim-
inates the PUT's valley -current offset. This allows timing
resistor R1 to have a value as low as 2 kilohms, rather
This pulse generator for driving digital circuits won't than a value above 100 kilohms, as normally required.
change frequency, even after its components have aged Output pulse rep rate can then vary over a 250:1 range
or are replaced. And pulse repetition rate remains for a given value of C1 (500 picofarads to 1 microfarad).
stable over a wide temperature range. Output frequency can be up to 1 megahertz.
If any or all of its active devices are replaced or have PUT temperature stability is assured by diode -con-
aged, the generator exhibits only a 1.5% worst -case vari- nected transistor Q3, which matches the temperature
ation in rep rate. Replacing or aging any or all of the characteristic of the Pur's anode -gate junction, even if
passive components causes a worst -case change of 1%. current flow is in the microampere region. Although
Typically, rep rate varies merely 0.5% over a 50°C tem- most standard diodes cannot do this, almost any low -
perature range, and all variations amount to less than leakage diode -connected silicon transistor may be used
2% when taken collectively. for Q3.
When the dc supply voltage (Vcc) is turned on, tim- Negative -going pulses from the PUT are directly
ing capacitor C1 charges exponentially through resistor coupled to the base of transistor Q4 through transistor
R1 until capacitor voltage reaches the threshold voltage Q3; capacitor C2 provides ac coupling. If pulses shorter
(VT) of programable unijunction transistor (PUT) Qi. than 4 to 5 microseconds (typical for this circuit) are re-
This triggering level is established by the voltage divi- quired, the R6C2 time constant of Q4's input coupling
der formed by resistors R2, R3, and R4. For optimum network can be reduced to differentiate the input signal
temperature stability when the supply is greater than 10 and produce narrower positive -going output pulses. The
volts: output pulse train is compatible with both transistor -
VT = 0.5Vcc transistor and diode -transistor logic.
Pulse rep rate is determined by:
rep rate = 1/[0.78R1C1 + (0.5 x 10-12)R1C1]
Once its triggering threshold is reached, the PUT
Stable pulse generator. Power supply charges capacitor C1 until threshold of programable unijunction transistor Q1 is reached, Qi then
switches, turning on transistor 02, which discharges C1. Diode -connected transistor Q3 dc couples resulting pulse to output transistor Q..
Generator's repetition rate changes only 2% for worst -case variation of all its components, even if they age or are replaced.

Vcc

R2
2.00 k,r2

Q3
+5 V
R6
10 kSt
Qi 2.2 kS2
2N6027
PULSE
1.15 kS2 OUTPUT

(1% POLY - 2
C1 = STYRENE
C2
OR MYLAR) 20 kS2
04
85
820 S2 470 pF
"Ro
1.15 kS2

02, 03, 04 PART OF RCA CA3018 '1% METAL FILM RESISTOR

45
 Input resistor R1 and diode D1 protect the circuit
against positive voltages. When an open circuit exists at
Logic probe with LED display the probe tip, resistors R2, R3, and R4 hold point A at a
checks ECL circuits logic low so that gates 01 and G2 are disabled. Since the
LEDs only light for a logic low excitation voltage, both
by William Wilke stay off regardless of the voltage level at point B.
University of Wisconsin, Madison, Wis. When the probe tip is connected to a low or a high
logic level, point B is forced to that level (-1.5 or -0.75
v) and point A is forced to a logic high level. Now gates
A simple logic probe for testing emitter -coupled -logic G1 and 02 are enabled and, depending on the level at
circuits identifies three input conditions-a logic high, a B, either the red or the amber LED lights up.
logic low, and an open circuit. The probe indicates a Resistors R5 and R6 allow the quad ECL NAND gate to
logic high (-0.75 volt) by lighting a red light -emitting switch up to 20 milliamperes through the LEDs for a
diode, a logic low (-1.5 v) by lighting an amber LED, brighter light output. The entire logic probe circuit can
and an open -circuit condition by lighting neither. The be assembled inside an ordinary felt tip pen or other
circuit costs about $4-$ 5 to build. small container. 1=1

ECL logic probe. Red light -emitting diode lights up for logic high input, while amber LED indicates low input. Both LEDs are dark with open
circuit at probe tip. When low or high is present at input, gates G1 and G2 are enabled, point A goes high, and point B goes to logic level at
input, lighting proper LED. For open -circuit input, G1 and G2 are disabled and point A is low, keeping both LEDs off.

-5 V
1110 mA)

R2 82 ks2
MOTOROLA R5 86r2 R6 86x2
MC 1048

R1
PROBE
TIP
1.3 kE2
an D.
(B)

20 - 80 µA
it DI R3 RED
1N914 33 kE2 LED MV 50
MONSANTO
D. j (A)
AMBER
4 33 kg2 L ED MV 1
MONSANTO

verted to four -channel displays. The key element is a

four -channel programable amplifier, the Harris type
Converter for oscilloscope HA -2405. This is an operational amplifier with four
provides four -channel displays identical input stages, any one of which may be elec-
tronically connected to the output stage by two binary
by Grady M. Wood address inputs.
Harris-Intertype Corp., Melbourne, Fla. For the scope converter circuit, each amplifier is
wired in its unity -gain inverting mode. High -value (2
megohm) feedback resistors provide a high input im-
With the help of only two integrated circuits and a pedance for each channel. All four non -inverting ampli-
handful of passive components, conventional single- or fier inputs go to a variable voltage source formed by a
dual -channel oscilloscopes can be economically con- 500-kilohm potentiometer that is between the ±15 -volt

46
power supplies. This arrangement provides an indepen- pulse of approximately 5 v from the 30-v waveform ob-
dent centering control for each channel. Any offset volt- tained at the GATE OUT terminal of a Tektronix type
age resulting from the large feedback resistors is not a 545 scope. Other scopes may require a different passive
problem, since the centering control provides adequate network.
amplifier adjustment range. Important circuit performance characteristics in-
The scope's gate output is divided down by a dual clude: a gain of 1, a bandwidth of dc to 5 megahertz, a
J -K flip-flop to supply binary channel selection signals slew rate of 15 v/microsecond, a maximum input volit-
for the integrated amplifier circuit. This gate signal is age of ± 10 v, an input impedance of 2 megohms, and a
synchronized to the scope's sweep, so that there are no crosstalk figure of 80 decibels.
timing difficulties. After each trace is completed, the Circuit voltage range, bandwidth, and input imped-
negative -going gate signal selects the next channel. The ance can be increased by adding op amps to buffer each
retrace time allows adequate time for channel selection. input. Making the gain of each of these buffers indepen-
The passive network at the input to the flip-flop-ca- dently variable further improves circuit versatility. Ap-
pacitor Cl and resistors R1 and R2-extracts a trigger proximate parts cost for the entire circuit is $25. 0
Scope converter. Monolithic quad operational amplifier provides inexpensive way to increase display capability of standard oscilloscope.
Binary inputs drive IC op amp; dual flip-flop divides scope's gate output to obtain channel selection signals. All channels have centering con-
trols for nulling offset voltage. Negative -going scope gate signal sernts next channel after each trace. Circuit operates out to 5 MHz.

0.001µF
FROM OSCILLOSCOPE GATE OUT-I
13 12 11
5 k.c).
FLIP-FLOP

4 5
1 Id -2

+5 V
TEXAS
INSTRUMENTS
CHANNEL
SN 7473
INPUT
Fr-
16

2 MS2 15
(4)
1µF
DECODE
2 ME2 CONTROL
(31 13
1 µF1 +15 V
1 biF
( 1 ) 12
2 MSH.--MA-7 OUTPUT 15 pF
AMP 11

1 µF 10 TO SCOPE
(2) 11,
INPUT
2 MS2 (2)

HA -2405

2 MS2 15 V

2 MS2

2 MS2
fvV\..
2 MS2

(2)

500 kS2

CENTERING
500 kS2 CONTROL

500 ks2

(4)
500 ks2

47
 = [EiRF/(R1+R1)IRRI/R2) -
(Ri/RF)1
Controlling op amp gain where R1 is the resistor at the op amp's noninverting
with one potentiometer input, and R1_ the resistor at the inverting input.
Varying feedback resistorRF changes the magnitudes of
by T. Frank Ritter both the positive and negative gains without changing
San Antonio, Texas the appearance of the graph; varying resistance ratio
Ri+/ RI_ shifts the null point.
As resistor R1_ approaches infinity, op amp gain
A single potentiometer and a few resistors can control varies from null to positive infinity only, as illustrated in
the gain of an operational amplifier from a selected (b). The equation for output voltage becomes:
negative value, through a null, to its positive open -loop E(, = [ERF/(Ri+ +RI)1[(RI /R2) + (RI/ROI
gain. The variable -gain circuit, which is shown in (a), As can be seen from the figure, the gain curve for this
maintains a high input impedance, even at high ampli- circuit is nearly logarithmic.
fication. It makes a convenient wide -range voltage ref- For a potentiometer rotation of 10% to 90%, amplifier
erence for a voltage regulator because it eliminates the gain can be varied over four decades. The gain at 50%
need to switch the op amp's circuit for above- or below - rotation is the ratio (Rt +R2)/(Ri+ +RI). Any general-
reference operation. purpose differential op amp can be used in the circuit.E
A graph of voltage gain versus potentiometer rotation
is also shown in (a). The equation for output voltage
can be written as:

Wide -range gain adjustment. Potentiometer varies gain of operational amplifier (a) from chosen negative value to positive open -loop value.
Null point can be shifted by changing resistance ratio of noninverting input resistor RI* to inverting input resistor R1+. Removing permits
positive gains to be controlled over wide range, as shown in (b). Circuit's input impedance remains high over full gain range.

RF RF

40 40

30

20 20
POSITIVE GAIN

10

z
NEGATIVE GAIN
0 0
POSITIVE GAIN

-10

--20

30
0 20
I

4C ti
NULL
60

PERCENT ROTATION OF RT
J

80
I

100
-20

30
0
'lit
20 40 60
PERCENT ROTATION Of RT
80 100

48
 Individually packaged gates, however, limit output
Logic driving gates double word length to about two to three bits because of the
marked gate -to -gate variation in logic low and logic
as d -a converter switches high output levels. A TIT high, for instance, is guaran-
teed to be between 2.4 and 5 volts, while a low lies be-
by Amos Wilnai tween 0 and 0.4 V. Output resistance also varies consid-
Monolithic Memories Inc., Sunnyvale, Calif. erably from unit to unit. Excellent voltage -level tracking
can, however, be obtained by using gates that share a
common substrate if they are operated under the same
The design of a weighted -resistor digital -to -analog con- load conditions. A five -bit converter (c) can be built
verter can be simplified by using the gate that supplies with an open collector hex inverter performing the in-
the digital input data as a switch. This approach permits put switching function.
a high -resolution converter to be built with standard When the outputs of the open -collector gates are low,
open -collector logic. With 1% resistors, five -bit resolu- each gate has a load impedance of 500 ohms, and the
tion is possible; tightening the tolerance to 0.1% can collector -emitter saturation voltages of the gate output
provide seven -bit resolution on selected units. transistors are within millivolts of each other. When the
A d -a converter (a) usually contains input switches, a outputs are high, each output transistor is off and the
resistor network, and an operational amplifier that pro- resistor network is referenced to supply voltage Vcc.
vides gain and a low -impedance analog voltage output. Longer output word lengths can be realized in the
The digital input data is generally supplied at standard same way by utilizing two hex inverter packages. There
logic levels-for example, at transistor -transistor -logic may be some variation between packages in the low
levels or diode -transistor -logic levels. Logic gates driv- gate output voltage level, but this can be minimized by
ing the converter can therefore be used as its input selecting the two packages from the same lot (by using
switches (b). the date code on the package).

Let gates do the switching. Basic digital -to -analog converter (a) requires input switching network to interface digital input data. Using indi-
vidual logic gates (b) to drive and switch converter limits resolution. However, up to five -bit word lengths can be obtained with open -collector
hex inverter (c) as driving and switching network. Because inverters have common substrate, is good.

DIGITAL RESISTOR ANALOG

DATA IN SWITCHES
NETWORK OUTPUT

HIGH LOW
REFERENCE VOLTAGES

(al

DIGITAL
DATA IN

ANALOG
GAIN ADJUST OUTPUT

-V
DIGITAL
DATA IN
2

ANALOG
OUTPUT

HIGH

L74051
ALL RESISTORS: 1/4 W, 1",

49
 When input pulse height is greater than Eu, compara-
TTL gates speed up tor Al is fired first, and then upper -level comparator A2
is fired. Once A2 switches on, the pulse stretcher
pulse -height analysis (monostable OS4) produces a positive pulse that turns
off the anti -conic gate for 600 ns. Therefore, when the
by Joseph Laughter pulse from monostable OS2 appears several nanose-
University of Tennessee Medical Units, Memphis, Tenn. conds later, it is blocked by the anti -conic gate and
there is no output. Throwing the window switch to its
"out" positioh disables the anti -conic gate, allowing a
Being used mainly for analyzing nuclear energy, deter- pulse to reach the output each time the input pulse ex-
mining white -noise amplitude, or counting blood cells, ceeds the lower -level limit voltage.
pulse -height analyzers require high-speed performance. This pulse -height analyzer can accept positive pulses
They are complicated to design with discrete transistors.having a maximum rise time of 250 ns and a maximum
But with transistor -transistor logic and integrated com-repetition rate of 500,000 pulses per second. The repeti-
parators, fast operation can be realized at a fraction oftion rate can be increased by using lower -value timing
the usual cost. capacitors for delay line OS1, output pulse shaper OS3,
Positive dc voltage E1, is the lower limit for a pulseand pulse stretcher OS4. However, input rise time re-
passing from input to output, and positive dc voltage Eu quirements become more stringent.
is its upper limit. At the instant the input pulse rises The type CMP-01C comparators can be replaced by
above FL, lower -level comparator Al switches to zero, the more popular type 710 comparators if the supply
triggering monostable multivibrator OS1. The output of voltage is changed from ±15 volts to + 12 and -6 v. Al-
OS1 returns to zero after about 300 nanoseconds and though the type 710 is considerably cheaper, it does not
triggers monostable OS2, sending a 60-ns pulse to the perform as well in critical applications.
anti -conic gate. The anti -conic gate inverts the pulse Timing curves show output wave forms for several
and triggers the output pulse shaper (monostable OS3). important points in the circuit. 0
Examining pulse height. Comparators AI and A2 set lower (EL) and upper (Eu) voltage limits. When pulse height exceeds EL, one -shots OS1
and OS2 slow down pulse from Al and transmit it to anti -conic gate, which fires output pulse shaper OS3. When input exceeds Eu, AI and A2
switch, causing pulse stretcher OS, to turn off gate so that pulse from OS2 cannot reach output. Window switch at "out" disables gate.

UPPER LEVEL COMPARATOR ANTI -CONIC GATE

+5 V
100 NF SHAPER
.1 (10 VI STRETCHER
NI(007 WINDOW

I I-
30 pF

-15 V
kS2 pF SWITCH 1.8 kS2
330 pF

3.9 kS2 111 IT 11

Eu 4 4 5 14 11 10 4 514 11 10
3 74121 6 3 741210S3 6
820 pF
s -tr 0S4 7 7400 OUTPUT
PULSE *-+15 V
INPUT
1.8 kS2
1wv.8
kS2 H1,270 pF 1 8 kS2
PF
PRECISION
MONOLITHICS TT I iTi
CMP-01C 4 4 5 14
3.9 kr2
8 lJ 5 14 11 10 11 10
3 74121 6 3 74121 6
EL 7 OS,
7 7
OS,
15 V
I
30 pF
DELAY LINE
LOWER LEVEL COMPARATOR

Eu- -- - Ell
INPUT EL EL < INPUT < Eu INPUT E, INPUT> EL, > EL

A, -1
A2

OS1
A2 -1
OS1

OS2

GATE
OS4
OS,
n
0S4

OUTPUT GATE
OUTPUT

0 200 400 600 800 1,000 0 200 400 600 800 1,000
TIME Ins) TIME Ins)

50
 the transistors continues to increase with rising input
voltage. Because the transistors are separated by zener
Analog voltage sensor diodes, each succeeding section operates only after in-
put voltage V1 increases by zener voltage Vz. Therefore,
controls LED threshold accurate LED turn -on levels (VD) can be set and, with
by Thomas Mazur the switches providing an abrupt turnoff, incremental
Motorola Semiconductor Products, Phoenix, Ariz. control of each LED can be achieved.
Assuming that all the zeners have identical voltage
ratings:
Most light -emitting diodes are found in alphanumeric (Vi)mix = Vz1 + VEB1 VD1
displays and optically isolated circuits where they are (VI)MAX = NVZ VS
usually controlled, either directly or indirectly, by dig- where VEB is transistor emitter -base junction voltage,
ital logic systems. Analog LED control circuits can also and Vs is the critical voltage level of the preceding uni-
be useful, provided that distinct light/dark LED transi- lateral switch.
tions can be obtained. The scanning circuit in the dia- The LEDS may be operated sequentially or in an over-
gram employs silicon unilateral switches, which func- lapping fashion by varying the type of zener used in
tion like four -layer diodes but have a gate control to each section. For a sequential mode:
produce sharp LED transitions for voltage -level sensing Vs is less than or equal to Vz + VEB VD
applications. For an overlapping mode:
There are N circuit sections, depending on the num- Vs is greater than or equal to Vz + VEB + VD
ber of voltage levels to be sensed. Each section consists In addition, the level of Vs may be reduced by connect-
of a LED, a silicon unilateral switch, a zener diode, a ing a zener diode between a switch's gate and cathode
bipolar transistor, and two biasing resistors. The unilat- terminals.
eral switch begins to conduct when its terminal voltage The scanning circuit may be modified to provide
reaches a critical level, nominally 8 volts. Once the highly discernible visual indication by replacing the
switch is turned on, its terminal voltage decreases to ap- emitter resistors by constant -current sources. This sup-
proximately 1 V. plies the LEDs with uniform current pulses, allowing
While switch voltage is increasing, the LED and the each one to produce a constant light output. Another
transistor's base -emitter junction become forward - modification permits the circuit to serve as a data trans-
biased. LED current, which is limited by the emitter re- fer mechanism-phototransistors can be inserted be-
sistor, rises until the unilateral switch conducts. Since tween the LEDs and the emitter resistors so that the
the transistor junction and the LED require around 2 v LEDs can be optically modulated.
to be forward -biased, the LED shuts off when the switch
voltage drops to 1 v.
Whether the switches are on or off, the voltage across

LED scanning circuit. Silicon unilateral switches SI through SN require 8 volts to trigger, but only 1 V to stay on. Rising input voltage for-
ward -biases transistor 01 and light -emitting diode DI. LED emits light until switch SI conducts; it goes dark abruptly when switch voltage
drops to 1 V. Zener diodes Z1 through ZN establish voltage levels that are sensed by each section of scanning circuit.

10 kr2 2 10 kS2 10kI

O.

D, D2 DN
S2 SN

270 SZ 270 270 a

z, Z2 ZN

SILICON UNILATERAL SWITCH

ALL SEMICONDUCTORS MOTOROLA 01 - ON : MPS 6515

SI - SN : MUS 4987
DI - Dry : MLED 55

51
 LEAST INPUT BITS

Wired -OR DTL gates increase SIGNIFICANT

BIT 0 1 3 4 5 6 7 B9
multiplexer input capacity A II2
ENABLE 10 I, 12 13 14 15 16 7

SO
by Eric G. Breeze 9313
Fairchild Semiconductor, Mountain View, Calif.
o_ Ai S
EIGHT INPUT MULTIPLEXER
A2 S2
c0
A3-+
A 10 -input multiplexer can be built by adding only one TTL
diode -transistor -logic gate package to an eight -input MOST OUTPUT
SIGNIF
multiplexer that has an open -collector output. Decade BIT
multiplexers are not currently available as standard in-
tegrated circuits because of the lead constraints of con- WIRED -OR
POINTS
ventional dual -in -line packages. Either an eight -input
multiplexer in a 16 -lead DIP or a 16 -input multiplexer in CD.
a 24 -lead DIP can be purchased. And modifying the 16 -
DTL
input unit is both an expensive and cumbersome way to OUTPUT
build a decade multiplexer.
Only three DTL gates are needed to add two input bits
to an eight -input multiplexer, provided the multiplexer ED. cs
has an open -collector output that can be wired -OR like
all standard DTL gates. The decade multiplexer illus- 9962 OR 9963

trated makes use of the OR tie facility of the 2 output ALL CIRCUITS: FAIRCHILD
of Fairchild's type 9313 eight -input multiplexer.
The most significant bit of the binary -coded -decimal Two more bits. Binary -coded -decimal input controls selection of in-
control input is connected to the ENABLE (active low) in- put bits to decade multiplexer. When most significant bit A3 is low,
put of the multiplexer and to the inputs of gates 01 and gates G1 and G2 are disabled, and eight operates
G2. When the most significant control bit is low (code normally for input bits 0 through 7. For high A3 bit, eight -input multi-
value of 0 to 7), the multiplexer operates normally, ac- plexer is disabled, but G1 and G2 transfer input bits 8 and 9 to output.
cepting input bits 0 through 7, and Gi and G2 are dis- DTL gates and i output are wired -OR.
abled (their outputs are high).
For BCD input selection codes of 8 or 9, the most sig- plexer to decode input selection code 8.
nificant control bit is high, the multiplexer is disabled, The type 9313 multiplexer contains an inverter stage
and input bits 8 and 9 can pass to the output, since both after its 2 output, making both TRUE (Z, Tm-compat-
G1 and 02 are enabled. Gate G3 performs as an in- ible) and ASSERTION (Z, Dm/TIT-compatible) outputs
verter for the least significant control bit into the multi - available. 0

cuits-while one is sampling, the other is holding the

Precision integrator previous integral value. Reset is also continuous-the
previous value of the integral is fed back so that the cir-
resets as it samples cuit is continuously reset for the full integration time
period. In this way, integration precision is determined
by Dennis J. Knowlton primarily by hardware and not by technique. Adding
University of Wyoming, Laramie, Wyo. an offset adjustment to each of the operational ampli-
fiers permits accuracy to be improved by at least an or-
der of magnitude.
A circuit that continuously samples the integral of an The integrating section is a simple integrator, com-
input and resets itself achieves an accuracy of within posed of an op amp having a low input bias current and
0.1%. Other integrators with this tight an accuracy can a feedback capacitor. The sample -and -hold section con-
be expensive and complicated because they do not inte- tains complementary MOSFETs. One portion of this sec-
grate continuously but rather require some time to tion holds the integral, while the other follows (samples)
sample and reset. the output from the integrating section. When the mos-
The integrator shown uses two sample -and -hold cir- PETS are switched, these roles are interchanged. The

52
741 -type op amps are used in voltage -follower configu- yields the integral of the new data. Two timing dia-
rations to isolate the integrating section from the grams show the circuit's response to a pulse input and to
sample -and -hold section. a step input.
The integrator is a true feedback system. If the inte- Calibration procedure for the integrator is straight-
gration period is 1 second, the integral formed during a forward. After the feedback circuit is unhooked, resistor
1 -second period is fed back to the input during the fol- R1 is adjusted to obtain the desired slope from the inte-
lowing 1 -second period to reset the integrator. Mean- grating section (for example, 1 volt out per second per
while, new data is being integrated. The output, then, is volt in). The feedback circuit is then connected again,
the sum of the new integrated data, plus the old data and resistor R2 is adjusted so that there is no overshoot
that has been integrated back to zero, which simply or undershoot to a step input.

Precision integration. Simple integrator can provide accuracy within 0.1% because it continuously integrates input. Output from integrating
section is sampled by one sample -and -hold circuit, while the other holds previous integral and uses it to reset integrator. Complementary
MOSFETs do the switching. Sample -hold roles reverse every integration period.

INTEGRATING SECTION
=NOM 41. SAMPLE -AND HOLD SECTION
1
5 µF
-t
2N4351 2N4352
741 302

I 180 ks2 40k1.2 308

INPUT I (A) 3µF
(± 10 V)
T-Alvv-r3
180112 200 ka
SQUARE
R2 WAVE
(0.5 Hz,
MIN iMM1 MOM 01111=
1.15 V)

302
2N4352 2N4351
T3 µF

L J
741
OUTPUT

ALL CAPACITORS: POLYCARBONATE

10 10
PULSE STEP
INPUT -4425°' '-50°°-" INPUT

10
w
(A) 5 (A) < 5

0 4- 0 0

10

OUTPUT 5 OUTPUT
5

0
0
0
TIME (s) TIME (s)

53
 The voltage transfer function of the equivalent circuit
Wien bridge in notch filter is:
H(s) = Vo(s)/V,(s) = (s2c2R2+ 1)/(s2c2R2
gives 60 dB rejection +3sCR + 1)
where s = jw, with w representing frequency.
by Donald DeKold This transfer function has a transmission zero at w =
University of Florida, Gainesville, Fla. 1/RC, the center frequency of the notch. At frequencies
above and below the notch frequency, H(s) approaches
unity. Since every R is paired with a C in the expression
A modified phase splitter and Wien bridge network for H(s), the shape of the transfer characteristic cannot
form a notch filter that is capable of providing 60 deci- be changed by varying the ratio of R/C. The filter's Q,
bels of signal rejection. The bridge network, which con- therefore, is constant for any value of R or C, or at any
sists of two capacitors and two resistors, makes this high frequency for which the notch is designed.
rejection possible and allows the filter to be tuned with A practical implementation of the filter is shown in
ganged capacitors or resistors. The three -capacitor, (c), along with its frequency response. Instead of a
three -resistor bridge ordinarily used for the twin -T vari- single collector resistor, a potentiometer and a series re-
ety of notch filter is not as easy to null because more sistor are used so that the filter can be adjusted for max-
components must be trimmed, and maximum notch imum signal rejection.
depth is usually about 45 dB. Employing a standard dual ganged variable capacitor
Because of the wideband frequency response of its for the bridge capacitors allows the notch to be tuned
modified phase splitter, the filter (a) can operate from from 8 to 200 kHz. Notch depth may vary because of
subaudio frequencies up to hundreds of kilohertz. For imperfect tracking of the capacitors, but will never drop
very -low -frequency performance, however, a direct cou- below a minimum of 45 dB. Because the filter operates
pling scheme must be worked out. at a high impedance level, it should be shielded to avoid
Unlike a unity -gain phase splitter, the filter's phase noise pickup at the output node.
splitter has a gain of approximately -2 at its collector. If
collector resistance is small with respect to resistor R of
the bridge, the ac equivalent circuit of (b) can be drawn.

Effective notch. Non -unity -gain phase splitter and four -element Wein bridge make up notch filter (a) capable of suppressing unwanted sig-
nals by 60 decibels. Ganged variable components can be used for bridge R or C, allowing notch to be tuned over broad frequency range.
Filter transfer function can be found from equivalent circuit (b). Practical filter (c) has adjustable collector resistance.

Vi(s)
F-

(a) (b)

FREQUENCY (Hz)
10 102 103 104
0

fo = 174 Hz
Z -20
0
I-

w -40

0.1 µF
-60
(c)

54
 ing the preceding value), the relationship between bi-
Mu ltivibrator clock nary input and frequency output is linear within 8%.
Inverters with open -collector outputs act as input buf-
obeys digital commands fers, providing the necessary pulldown of current from
the timing resistors. Diodes are added to prevent signal
by Patrick L. McGuire interference at the inverter outputs.
General Dynamics, Electrodynamic division, Pomona, Calif. Circuit output is a square wave having a frequency of
half the pulse rate from the multivibrator. Output fre-
quency depends on the value of timing capacitor C:
A simple variable -frequency multivibrator clock source f = 0.049/C
can be made data -dependent by controlling the current where f is in kilohertz and C in microfarads.
into the multivibrator's timing network. When timing The graph shows the multiple of f determined by each
resistor values are selected in increments of two (doubl- input data word.

Programable clock. Controlling current through multivibrator enables binary input to determine frequency of output square wave. In-
put/output relationship is practically linear because values of adjacent timing resistors differ by factor of two. Inverters buffer current from
timing resistors, while diodes guard against signal interaction between bits. The flip-flop halves output pulse rate from the multivibrator.

Vcc

LEAST INPUT OUTPUT

SIGNIFICANT DATA FREQUENCY
BIT WORD
394(12 20 kE2 101<2 4.7 kS-2

0000 15f
0001 14f
0010 13f
0011 12f
0100 11f

BINARY 0101 10f

INPUT 0110 9f
0111 8f
1000 7f
1001 6f
1010 5f
1011 4f
1100 3f
MOST 2/3 SN7405 1N914
SIGNIFICANT (ALL) 1101 2f
BIT 1110 f
1111 0

f(in Hz) = 0.049/C On NF)

ALL ICs: TEXAS INSTRUMENTS

MULTI
OUTPUT
15 14
13 SQUARE
WAVE
1/2 SN74123 FLIP
FLOP

1/2 SN74107

55
Op amp with feedback
makes full -wave rectifier
by Richard Knapp and Roger Melen
Stanford University, Palo Alto, Calif.

A feed -forward resistive element allows one operational

amplifier to do the job of two op amps-perform full -
wave rectification of low-level signals. The resulting cir-
cuit is useful for a wide variety of frequency -doubling
and small -signal rectification applications, such as ac - Full -wave rectifier. Inverting input of operational amplifier remains
to -dc converters, absolute -value detectors, and fre- at virtual ground for both positive and negative input voltages so that
quency multipliers. output voltage is always developed by input current flow through
For both positive and negative inputs, the op amp's feedback resistor R1. Diode DI conducts only during positive inputs,
inverting input is always at virtual ground. And for ei- and diode D2 is on only for negative inputs. Normally, two amplifiers
ther input polarity, output voltage is developed by input are required to perform full -wave rectification.
current flow through feedback resistor R1.
Diode D1 is forward -biased during positive inputs, taining the op amp's inverting input at virtual ground.
while diode 132 remains off. Input current flows primar- Circuit output impedance is approximately equal to
ily through resistor Ri to the output. During negative the resistance of diode DI for positive voltages and to
inputs, diode D2 conducts and diode D1 is off, main - the sum of Ri + R2 for negative voltages.

ing the Gray code can be ANDed without glitches.

Gray -code generator Flip-flops FF1 and FF2 are wired as a Johnson
counter. The Q output or FF1, which is 90° out of phase
avoids output glitches with the Q output of FF2, provides the A Gray -code
output line. Flip-flop FF3 is toggled by the CT output of
by Carl Moser FF2, causing it to produce the B Gray -code output line.
Western Electric Co., Winston-Salem, N.C. The other three flip-flops, FF4 through FF6, form a
modified Johnson counter. FF5 and FF6 are toggled
simultaneously by the Q outputs (labeled X and Y in
When binary signals are ANDed together, undesirable the diagram) of FF2 and FF4. Output lines C and D are
Witches can be generated at a circuit's output if one or generated by FF5 and FF6, respectively.
more signal transitions are simultaneous. A Gray -code Since the circuit is asynchronous, its maximum oper-
generator can be built with six J -K flip-flops that are ar- ating frequency is limited by the delay of the flip-flops.
ranged so that only one signal transition occurs at any For correct output code generation, all the flip-flops
particular time. Therefore, the four output signals form - must be cleared initially.

OUTPUT GRAY CODE

A BCD
CLOCKIIIIIIIIIIIIIIII 1

I I (X)

(Y)
IX)
J Q J Q J
A
-o T FF, --c T FF2 -C T F F5 -0 T FF6
B
K K K K

C I I
CLOCK

(Y)
J Q J Q
Stopping glitches. Output lines of Gray -code generator have no si-
0 T FF4 T FE:
multaneous signal transitions, permitting them to be ANDed together
K without creating glitches. Six flip-flops comprise complete generator.
FFi and FF2 are connected as Johnson counter, while FF5 and FF6
FLIP -F LOPS: TEXAS INSTRUMENTS SN7473 form modified Johnson counter. FF3 supplies one of the output lines;
FF4 is part of simultaneous toggling scheme for FF5 and FF6.

56
 tance (R0) is less than 200 ohms, voltage gain (GO is
Active filter has separate more than 10,000, and output voltage swing (V0s) ex-
ceeds 20 v pk-pk.
band and frequency controls To solve the design equations, let:
K1 = (ReC1R2C41/2 = 1/(2.iff0) = 1.59 X 10-4
by John Jenkins K2 = R1C1 = 1/(2.7;BW) = 3.18 x 10-2
Montgomery, Ala. K3 = (RfCl/R2C2)1/2
[(Vosileomax)2 1]1'2 = 19.98
then the filter's time constants can be computed:
The bandwidth and center frequency of an active band- R1C1 = K2 = 3.18 x 10-2
pass filter can be controlled independently by two sepa- R2C2 = K1/K3 = 7.96 x 10-6
rate resistors. Moreover, the filter's gain remains at RtC1 = K1K3 = 3.18 x 10-3
unity over its full tuning range. Filter Q range is 2 to Rf/Ri = K1K3/K2 = 0.1
200, while center frequency is 1 to 10 kilohertz. For most applications, a few simplified guidelines can
The circuit shown in (a) has these properties, but it be followed to choose component values: resistor R1
requires a variable inductor, which is usually difficult to should be less than 400 kilohms, resistor R2 should lie
tune, can be large, and cannot provide good tempera- between R,2 (about 40 kilohms) and 1 kilohm, the load
ture stability. The transfer function for this LC filter is: resistance should be greater than 1 kilohm, and factor
e0/e; = (s/ReCe)/(g2+s/ReCi + 1/LC1) (1 - R3C3/R2C2) should range between 0 and resistance
Replacing the inductor with an active RC network, as ratio (Re/Re) X 10-2.
illustrated in (b), yields a temperature -stable circuit. If This last constraint requires that time constant R2C2
all the components are ideal and R2C2 = R3C3, the track R3C3 within +0% and -0.1%. Therefore, these re-
equivalent inductance can be expressed as: sistors and capacitors must have closely matched tem-
Leg = R2C2Re henries perature coefficients and operating temperatures.
and the 3 -decibel bandwidth as: Metal -film resistors and NPO-type capacitors that are
BW = 1/(2.7111C1) hertz mounted close together can be used. (The R2C2 and
and the center frequency as: R3C3 time constants can be aligned by first opening the
fo = 1/[277(RfC1R2C2)1/2]Hz filter's input to obtain maximum Q, then increasing R3
A wide range of component values can be used in the until oscillation occurs, and then decreasing R3 until os-
circuit, which is easy to design, once the desired filter cillation just stops.)
specifications are established. As an example, a filter A set of typical component values is noted in (b). As
will be designed with a 5 -Hz bandwidth, a center fre- indicated, resistor R1 tunes filter bandwidth, while re-
quency of 1 kHz, and a maximum output voltage of 1 sistor Re adjusts center frequency. 0
volt peak -to -peak. A few important operational ampli-
fier specifications must also be known. Typically, input
resistance (Ri) is greater than 40 kilohms, output resis-

Active circuit ousts variable Inductor. Bandpass fPter (a) offers independent center frequency and bandwidth adjustments. Hard -to -tune
variable inductor can be replaced by active circuit (b) that provides an equivalent inductance and better temperature stability. Fully active fil-
ter is easy'to design and will operate over a broad range of component values. General-purpose amplifiers can be used.

i41
BANDWIDTH
ADJUST 0.001 p F

BANDWIDTH
ADJUST C2
H2
R 31.8 krz
A2 7.96 kft
741
318 kS2
C3I 0.001 pF

FREQUENCY
C, = ADJUST
0 1 pF 7.96 kSl

(a)

57
Logic circuit converts +5 V +5 V
ks2
synchronous motor to stepper 1 kS2 7473
B CCW
NUM. Q Q

by Michael D. Doering C FFi FF2

Food and Drug Administration, Bureau of Radiological Health, Rockville, Md. INPUT
SQUARE K
CW
WAVE
7473

A simple circuit that is compatible with transistor -tran- +5 V +5 V

sistor -logic circuits can convert a two -coil synchronous
motor to a synchronous stepping motor. Since circuits 7451
X AND -OR
that perform this conversion are not available commer- INVERT
-

cially, the designer is usually forced to use relays or

come up with his own stepper control circuit. C=Ae

The problem of conversion generally arises when a

variable -speed synchronous motor is needed that is ca-
pable of delivering up to 1,800 inch -ounces of torque, or
when a variable -speed adaptor is required for a motor
that is already installed. The converter, therefore, must
be a high -current, high -voltage, variable -frequency cir- +30V 22f2
cuit that can drive a two-phase 115 -volt ac motor.
-1/V r--
e_JYYY
A variable -frequency square wave, which can be sup- 9 10 10 pH
4
plied by Tri, integrated circuits, drives the converter RCA HC -2000 Al 3

and establishes motor stepping speed. Motor -speed ac- +5V

curacy depends on how accurate the input square -wave
frequency is. The size of the step and the maximum -30 V 25011

stepping speed are determined by the motor used.

Flip-flops FF1 and FF2 are connected as a repeating -30V 121.552v22)

two-bit counter, which has its output decoded by an SUPERIOR

AND -OR -INVERT circuit. This arrangement provides the SS150 P1 RC

four states, which are noted as A, B, C, and D in the RCA HC -2000 A2 3

diagram, needed to make the motor step properly. 9 10

4
100-I
The position of the switch at the output of FF2 deter- ADJUST COMPARATOR
SUPPLY VOLTAGE TO
mines the direction of rotation. As can be seen from the MATCH MOTOR
+30V 22 f2
figure, the D state simply represents the switch -selected
B or 5 states.
Output states C and D are fed to the inverting inputs Stepping a synchronous motor. Logic circuit plus high -power corn
of high -power comparators Al and A2, respectively. The parators step two-phase synchronous motor in clockwise or coun
voltage levels of these two states are then compared to terclockwise direction. Frequency of input square wave determines
the reference voltage established by the 5-v supply and motor stepping speed. Two-bit counter formed by flip-flop pair and
zener diode D1. Each comparator drives a separate mo- AND -OR -INVERT circuit produce four logic outputs (A, B, C, and D)
tor coil and can develop a 75-v 100 -watt output. 0 that control motor. Each comparator output supplies one motor coil.

clock in either direction (up or down), the counter can

Ring counter eliminates be reset from any ring output without losing a clock
pulse. Only five dual in -line integrated -circuit packages
false gating signals are needed-a binary counter, a quad latch, a quad
NAND gate, and two BCD -to -decimal decoders.
by Glen Hamilton When the input clock pulse rises to logic 1, the binary
Dickson, Tenn. counter is incremented up or down, depending on the
position of switch S1. The unused clock input to the bi-
nary counter must be held continuously at logic 1. For
A ring counter that performs reliably at speeds up to 7 most applications, switch S1 can be eliminated because
megahertz can be designed to prevent counting errors only one clock direction is required. To switch between
caused by false gating signals. Besides being able to the two clock inputs, two gates must be added so that

58
the unused connection is always held at logic 1. indicated by the waveforms in the timing diagram.
With the transition of the input clock pulse from logic All ring outputs are disabled (in the logic I state)
1 to logic 0, the binary counter outputs are loaded into when the input clock pulse is logic 0. During this time,
the latches. The next clock pulse enables the NAND the ring outputs are inhibited by the NAND gates to pre-
gates, allowing the .45 signals from the latches to pass to vent false gating due to transients. Also, all ring outputs
the decoders, which then select the appropriate ring are always 180° out of phase with the input clock pulse,
output to be enabled. Each succeeding clock pulse se- and a transition from logic 1 to logic 0 is required to re-
quentially steps and enables the decoder outputs. set the ring counter. This means that the ring counter
The number of stages in the ring is determined by the will be reset only when the ring output selected by
ring output at which switch S2 is placed to implement switch S2 is enabled (in the logic 0 state).
the reset function. When the ring output tapped by The circuit can also be used as a frequency divider
switch S2 is enabled, the binary counter is set to the that has its output pulse frequency determined by the
number appearing at its preset input lines. These preset width of the input clock pulse. Any integral divisor,
inputs can be either hardwired or made selectable. from 2 to 15, can easily be selected by switch S2. The di-
For the wiring connections shown, the binary counter vided frequency may be taken from any of the ring out-
has the binary number six continuously held at its pre- puts.
set inputs, and the ring counter's reset function is placed The binary counter is provided with CARRY TO and
at the fourth output line of the second decoder. This BORROW FROM pins, permitting counters to be cascaded
hookup resets the binary counter to six each time the for generating more than 15 outputs. With two binary
twelfth ring output is enabled, causing the ring counter counters, up to 256 outputs can be produced, as long as
to step sequentially from six to 12 and then repeat, as an appropriate number of decoders is used.

Sure -clocking ring counter. Input clock pulses sequentially step data through binary counter to latches (flip-flops) to NAND gates to BCD -
to -decimal decoders. Switch S1 permits clocking up or down, while switch S2 selects number of stages in ring. S2 also carries reset signal that
sets binaiy counter to its preset input number. Counting proceeds only for clock transition from logic 0 to logic 1.

INPUT CLOCK PULSE ALL ICs: NATIONAL SEMICONDUCTOR

CLOCK CLOCK DM7475 DM7400 RING

DOWN UP OUTPUTS

A 0

BINARY DECODER 1

2
COUNTER
B B 3
4
PRESET DM7442
INPUTS DM7563
C
O 6
7
8

LOAD
CLEAR (PRESET)
A 0
120ns
MIN
DECODER
CLOCK
PULSE B 3 s,

6
DM7442

7 -1-1 6
7

8
9

10

11

12

59
 ration (b) makes it possible to keep voltage gain inde-
pendent of output load conditions and to be determined
Complementary output stage by a resistance ratio:
improves op -amp response eo/e, = R2/R1
where R1 is the load resistance for the op amp, and R2 is
by Robert Gagnon and Richard Karwoski the input resistor to the output stage.
Raytheon Co., Equipment division, Sudbury, Mass. If R2/R1 = 10, the op amp output voltage swing
need only be ± 1 v to realize a circuit output voltage
swing of ±-10 v. In practice, a resistance ratio of 6.7, with
The performance of a conventional 741 -type oper- R1 = 270 ohms, yields the best over-all results because
ational amplifier can be considerably enhanced if it's of how the op amp responds to this load resistance
given a complementary -transistor output stage. The op value.
amp's gain -bandwidth product is extended from its For a load resistance of around 300 ohms, the op
normal 1 megahertz to 7.5 MHz for a 250 -ohm load re- amp's maximum output voltage swing is reduced by
sistor, while slew rate is increased from 0.5 to 5 volts/ about 30%, but its slew rate remains approximately 0.5
microsecond. Similarly. the full -power bandwidth yips for small voltage excursions. By fixing the gain of
reaches 50 kilohertz, as opposed to 15 kHz, and the the output stage at about 10, the slew rate of the over-
bandwidth at a voltage gain of -2 becomes 2.5 MHz, all circuit is increased by a factor of 10, and the op
rather than 330 kHz. amp's frequency response is extended.
The output stage (a), which contains transistors Qi Undesirable high -order effects of the 741 -type op
and Q2, acts as a current buffer, providing extra load amp are far enough beyond its nominal 1 -MHz cross-
drive capability. It is basically a bootstrap configuration over frequency that instability due to the gain added by
using degenerative feedback. Transistor Q2 is the princi- the output stage is not a problem. A 100-kilohm cermet
pal source of load current, and the stage's dynamic in- potentiometer is included in the circuit to permit cancel-
put impedance is the product of the load resistance and lation of crossover effects. It should be adjusted at the
the current amplification factors of both Qi and Q2. full -power bandwidth limit -50 kHz and a maximum
The two outboarded transistors form two comple- output voltage of 20 v peak -to -peak.
mentary pairs with the output transistors inside the op
amp-outboarded transistor Qi complements internal
transistor Q3, while Q2 complements Q4. This configu-

Boosting op amp performance. Output stage (a) multiplies op amp slew rate by 10 and extends crossover frequency to 7.5 megahertz. As
shown in (b), external Oi and Q2 complement internal transistors 03 and 04, respectively. Resistor R1 acts as op amp load resist-
ance so that resistance ratio R2/R1 fixes overall circuit gain. Additional gain lets op amp operate at low output voltage.

+v

2.2 NF
18052

07
2N2905

330 sz

500 S-2 eo
ei-AAAi-=

1 al
01
2N2219

COMPENSATE0
18052 z6
2.2 NF NORMAL

01 1 10
FREQUENCY (MHz)

lal (b)

60
 and Qi's emitter current becomes:
IE1 = V2/12.3
Generator independently Both V1 and V2 can range between 0 and 2 volts.
varies pulse rate and width Transistor Qi has a minimum hFE of 250, making its
base current much smaller than its collector current so
by Mahendra Shah that constant current Tel is maintained nearly equal to
Univ. of Wisconsin, Space Science and Engineering Center, Madison, Wis. constant current IE1. Because transistor Q2 is also a
large-hFE device, it draws very little base current and al-
most all of Ici passes through resistor R4 and diode D1.
Capacitor charging current can be used to change the Furthermore, the base -emitter voltage of Q2 and the
output frequency of a voltage -controlled pulse gener- forward voltage drop of D1 are about the same, permit-
ator that also offers an independent control over output ting voltage V3 to be written as:
pulse width. The generator's output -pulse frequency V3 = R4IC1
can range from 1 to 75 kilohertz and output pulse width And the constant charging current, 1c2, for timing ca-
from 100 nanoseconds to 18 seconds. A dual retrig- pacitor C1 can be expressed in terms of voltage
gerable one-shot performs the actual pulse generation, 1c2 = R4V1/R3R5
while an operational amplifier and a couple of transis- One-shot OS1 can be retriggered during its on state
tors convert the control voltage into a proportional within 100 nanoseconds of the end of the timing period,
charging current for the frequency timing capacitor. fixed by its timing components. The positive trigger in-
The voltage divider created by resistors R1 and R2 de-put of OS1 is connected to its Q terminal, and a small ca-
termines the voltage level at the noninverting input of pacitive load (C2) is added at the Q terminal to increase
the op amp. Since the op amp and emitter -follower Qi the retriggering delay by about 20 ns.
form a unity -gain buffer having a high input imped- At the end of OSi's timing cycle, its Q output makes a
ance, the large open -loop gain of the op amp keeps the high -to -low transition, and its Q output makes a slightly
noninverting input voltage (Vi) equal to the inverting delayed low -to -high transition, retriggering OS1 back
input voltage (V2): into its on state for another timing period. This cycle re-
V1 = V2 = [R2/(R1 + ROW; peats at frequency f:

Voltage -variable rep rate. Input control voltage determines output frequency of pulse generator. Collector currents developed by transistors
Ql and 02 are directly proportional to control voltage Vi. Constant current IC2 charges timing capacitor C1, controlling pulse frequency of
one-shot OS1, which is retriggered by its own 0 output. OS,'s 0 output triggers one-shot 0S2, which controls output pulse width.

+5 V PULSE -WIDTH
CONTROL

0<C3< 1,000 pF R5
R4
O<R6< 60 1(12 820 12 820 12
(±5%) 1±5%) R6
100 kS2
Dl
IN4148 D2
2N3251
F17

1C2
5.1 k.12

C3
1900 pF 0.022 µF
R1 (±10%) (±10%) TEXAS INSTRUMENTS
10 kit SN74123N
(±1%)

a, 15 14. 16 6
In
+ 2N2484 1 13 10
3
CD

-J
FAIRCHILD
pA741
6 OS, 0S2
+4.5 V
0 SL11...
2-=" 2
2 12
_.1 1.4_1 ps

R2
CC
OUTPUT
1k&2
(±1%) L
-
C.1
R3 UTPUT
820 .12 C2
I 330 pF

-5 V

61
 f = 1/(T1 + 60 ns) V1. The graph shows that the pulse generator's fre-
where T1 is OS1's on -time. One-shot 0S2 is triggered by quency is practically linear over the control voltage
the low -to -high transitions of OS1's Q output; these range of 5 to 22 v (0.45 to 2 V for VI).
transitions generate positive -going pulses at OS2's Q Generator frequency range can be extended to higher
output. The width of OS2's output pulses is controlled or lower frequencies by scaling the value of capacitor
independently by resistors R6 and R7 and capacitor C3. C1, but this capacitor's charging current, 1C2, should not
Timing period T1, which is usually much greater than be made greater than a few milliamperes. Voltage con-
60 ns, is inversely proportional to voltage V1. The cir- trol of output pulse width can be obtained by replacing
cuit's output pulse frequency is almost linearly pro- resistors 11'6 and R7 with a voltage -controlled constant -
portional to voltage V1 and, therefore, to control voltage current source.

voltage V2 makes its low -to -high transition.

Besides protecting the input of inverter I1, diode D1
Photodetector senses motion isolates point A, allowing the potential at this point to
in noisy surroundings rise to 12 v. Without the diode, point A would be
clamped at Ifs threshold voltage of 8 v and voltage Vi
by Richard T. Laubach would never reach 12 V.
National Cash Register Co., Cambridge, Ohio Once the circuit is in its high state, transistor voltage
Vi must drop below 4 v to pull point A below Ifs thres-
hold and to switch V2 back to ground. When V1 be-
A digital phototransistor amplifier, consisting of a few comes less than 4 v, the phototransistor's pull-up resist-
readily available inexpensive components, can detect ance increases and speeds up the rate of decline of V1.
slowly moving objects in an electrically noisy environ- Feedback resistance Rf causes Vi's switching voltage
ment. The circuit works reliably, even if long cables link levels to be different-first 11 v, then 4v-thereby pro-
its optical sensing section to its amplifying section. viding 7 v of hysteresis and noise immunity.
The circuit detects objects that break the light beam Inverter 13 and diode D2 buffer voltage V2, converting
between the light -emitting diode and the phototransis- it to a 5-v transistor -transistor -logic signal. If desired,
tor. When the light beam is interrupted, transistor -out- the 12-v output of V2 can be used directly as the logic
put voltage V1 increases, causing the potential at point input voltage.
A to rise until the threshold voltage (about 8 volts) of The circuit is operating successfully in an electrome-
inverter II is chanical printer where the cable from the phototransis-
Both inverters I1 and 12 then switch, and voltage V2 tor to the amplifier input is a twisted pair of wires 10
jumps from 0 to 12 v. This voltage jump supplies addi- feet long that run near solenoid driver cables. Moving
tional current at point A through feedback resistor Rf, printed matter is detected by an existing circuit without
reducing the pull-up resistance seen by the phototran- any output -signal bounce, even though the edge of the
sistor so that voltage Vi jumps from 8.5 to 11 v when matter momentarily stops as it breaks the light beam.
Optical detector. When object breaks LED light beam, phototransistor output voltage V1 increases, switching on inverters I, and 12. Voltage
V1 then jumps up because additional current through resistor Fif to point A decreases phototransistor pullup resistance. Voltage V2 stays high
until V1 drops to 4 V, increasing pullup resistance and causing V2 to return to ground. Inverter 13 converts V2 to TTL-compatible output.

+12 V
+5 V

43 IcS2

PHOTOTRANSISTOR: TEXAS INSTRUMENTS LS600

LED: TEXAS INSTRUMENTS L23 12 V

11 12, 13: % FAIRCHILD 9109 V2 0

62
 Transistor Q3 functions as a unity -gain stage, trans-
ferring input voltage V; to produce voltages Vi and V2,
Quasi -matched MOSFETs form which are equal in magnitude to V, but 180° out of
filterless squaring circuit phase with each other:
V1 = -V2 =
by W.V. Subbarao Effectively, the input to the main gate (GO of transistor
North Dakota State University, Fargo, N D Qi is V1, and the input to the main gate of transistor Q2
IS -V1. Input voltage V, can then be considered as the
VGs voltage for both Qi and Q2.
By compensating inexpensive dual -gate MOSFETs so The MOSFET transfer characteristic can now be used
that they appear matched, a squaring and frequency - to solve for the drain current that flows in resistor Rn:
doubling circuit can be made to function reliably over a ID1,2 = ID1 1D2 = 2IDSS 2IDSS(Vi/VP)2
broad frequency range without the aid of a filter. Con- The output signal voltage contribution can be separated
ventional filterless squaring circuits require costly high - from the output dc offset voltage of 2RDIDss:
quality matched components. V. = 2RDIDss(Vi/Vp)2
Biasing MOSFETs Qi and Q2 to operate in their deple- Letting K = 2RDIDss/Vp2 allows the output voltage to
tion region causes MOSFET behavior to resemble that of be written as:
the junction field-effect transistor. The drain current Vo = KVJ2,
(ID) of either Qi or Q2 is given by: which is a squaring function.
ID = IDss(1 - V6s/Vp)2 The value of resistor RD is held to 1 kilohm to prevent
where IDss is the drain current with both gates shorted the dc offset current of 2IDss from saturating the MOS-
to the source terminal, VGs is the voltage between FETs when the circuit is operating. Also, peak output
shorted gates and the source terminal, and Vp is the voltage swing is restricted to about 1 volt to keep from
pinchoff voltage (the VGS value when ID = 0). driving the MOSFETs away from their optimum mid-
Since Qi and Q2 are not matched, they may, however point bias condition.
have about the same Vp value (approximately -1.5 For a sinusoidal input, V. is also a sinusoid at double
volts) but exhibit different IDss values. If Q2's IDss cur- the input frequency and with a voltage gain of approxi-
rent is lower than that of Q1, say 3 milliamperes as com- mately 3.5. Resistors R1 and R2 are adjustable so that
pared to 4 mA, the control gate (G2) of Q2 can be driven the Qi-Q2 match can be preserved for changing input
positive, making Q2 more conductive so that its IDss signal conditions. This permits the circuit to operate
current is compensated to equal that of Qi. In this way, from 200 hertz to 1 megahertz without any distortion. n
Qi and Q2 can be made to look matched with the same
IDss and Vp values. The transfer curves show how Q2's
characteristic tracks Qi's characteristic.

Compensated MOSFETs double frequency. Filterless squaring circuit works from 200 hertz to 1 megahertz. MOSFETs 01 and 02 are oper-
ated in their depletion mode, causing them to square voltages applied to their main gates (GO. Adjusting control gate (G2) voltage forces
MOSFETs to simulate a matched pair. Unity -gain transistor 03 drives Oi and 02 with equal voltages of opposite polarity (V1 = -V2 - V1).

63
 the bridge is variable from this resistance null:
Feedback linearizes = R + AR
where AR represents the change in bridge resistance.
resistance bridge For the standard bridge:
V0' = V1(0R)/(2R +AR)
by Robert D. Guyton This is not a linear relationship, since AR appears in the
Mississippi State University, State College, Miss. denominator. The standard bridge, therefore, is usually
limited to those applications that involve only small val-
ues of AR.
With the addition of a feedback circuit, the output volt- The feedback circuit shown in the figure alters the
age of a standard resistance bridge can be made to vary bridge excitation voltage so that the relationship be-
linearly for a change in bridge resistance. The feedback tween the output voltage and OR becomes linear:
circuit provides good sensitivity for a wide range of Vo = (1 +2Rf/R)V(AR)/ 2R
bridge resistance variations, while maintaining the full- Feedback resistor Rf determines the circuit's sensitivity
scale linearity of the bridge output voltage to within to the change in bridge resistance. Bridge excitation
0.1%. Furthermore, either an ac or a dc voltage may be voltage can be written as:
used to excite the bridge. V1 = V + RVo/(R + 2Rf)
The output voltage of the standard bridge remains when:
zero as long as its four resistance arms are equal to each R1 = R2(R+2Rf)/R
other, with a resistance value of R ohms. One arm of Resistance R can vary from 0 to 2R.
Linearized resistance bridge. Adding feedback circuit to standard resistance bridge allows output voltage V. to be linearly related to vari-
ations in bridge resistor Rx. Changes in bridge resistance produce output voltage that is linear to within 0.1%. Circuit sensitivity is set by value
of feedback resistor Pf. Amplifiers control bridge excitation voltage VI, which can be due to either ac or dc source.

Rf
2.2ks-z
R2 5k fl

lOkE2
301(f.-2
R,
5k1-2

R R2
2.2kI2
tokn Z 4.31(12

+6V

ALL RESISTORS: ±0.1%

OP AMPS: BURR -BROWN 3024/15

64
 soldering iron, allowing tip temperature to be set from
near room ambient to 125°C. A potentiometer permits
Soldering iron converts to the temperature setting to be varied continuously. The
tip -temperature -sensing element is a thermistor, which
constant -temperature probe is installed by drilling a small hole in the tip and then
by Mahendra J. Shah epoxy -mounting the thermistor in place.
Univ. of Wisconsin, Space Science an: Engineering Center, Madison, Wis. For best results, an isolation transformer should be
placed between the source of line power to the tip and
the triac at the output of the control circuit. The heat
Designing a circuit that has good temperature stability transfer of the probe can be improved by using silicone
requires pinpointing those components that are the ma- vacuum grease between the tip and the component un-
jor drift contributors. These components can then be der test.
properly specified and compensated for temperature Sometimes, testing component drift with a constant -
drift, if necessary. Unfortunately, the designer must temperature probe can reduce parts cost. For instance,
frequently run temperature stability tests on the entire suppose a regulated 600-v power supply that contains a
circuit because he cannot heat individual components high -stability zener reference costing about $24 is tested.
selectively. And employing a soldering iron as a selec- When the temperature of the whole supply is increased
tive heat source does not produce precise temperature by approximately 50°C, the supply output drops around
test results. 5 v.
However, a conventional line -operated soldering iron A transistor junction is found, with the probe, to be
can be easily converted into a constant -temperature the principal cause of the drift. Substituting a general-
probe. Since most low -power circuits operate very near purpose zener selling for about $1 for the high -stability
ambient room temperature and most circuit compo- device will decrease over-all supply drift. The tempera-
nents have a low thermal resistance for efficient heat ture -drift errors of the transistor junction and the low-
transfer, the device being heated by the probe will be cost zener almost cancel each other, since they have
within a few degrees of the probe temperature. The drift about the same magnitude but are of opposite polarity.
contribution of the component can then be measured by It is also possible to build a probe that lowers compo-
noting the change in the circuit parameter of interest. nent temperature below room ambient by using a
The temperature -control circuit in the diagram regu- thermoelectric cooling element in a temperature -control
lates the voltage applied to the tip of a 115 -volt, 27 -watt loop. CI

Pinpointing component drift. Constant -temperature probe can be built by controlling power to tip of soldering iron. Control circuit maintains
tip temperature within a few degrees of desired setting, from room ambient to 125°C. Thermistor located inside probe tip acts as the tempera-
ture -sensing element. An isolation transformer (not shown) should be placed between the line input and circuit's output triac.

PROBE TIP
2k0 (115-V
27-W
IRON)

101(C2 TEMPERATURE
1052
ADJUST
3

14 RCA 4
115/30 CA3059 RCA
Vrms 30kE1 40724
WI RAMP 7
100k12 SLOPE
9 TRIAC
2A CONTROL

LINE 320 I.LF 10052

+ (20V) 10k12. 1.2k1 470k12

loon
10052
2N4852 410--vVV-411
2N3646

IN649

THERMISTOR.
F EN EWAL GA45J 1,
"1--- 1.2 µ F
50 ka @ 23°C. 125 pF
(25V) (25V1

2N3646

PULSE TRANSFORMER
*INSERT ISOLATION TRANSFORMER

65
 Two inverters can also be used as the delay element,
as shown in (b). Now the gate inputs have the same po-
Exclusive -OR gate makes larity, causing the gate output to be low for either a
bidirectional one-shot high or low input signal. The output will be a train of
positive pulses with a width equal to a single inverter's
by Tim O'Toole delay time. These output pulses may be easily extended
Tektronix Inc., Beaverton, Ore. or stretched by inserting an RC timing network between
the two inverters.
Applying the operating principles of circuits (a) and
An exclusive -OR gate is an ideal device for building a (b) permits the realization of a bidirectional one-shot (c)
bidirectional one-shot by running the same signal into that has an addressable output pulse polarity. Gates G1
both gate inputs, but putting a time delay on one of the and G2 simply duplicate the one-shot of (a) and feed
inputs. The gate will then produce a pulse for every ris- one of the inputs of the exclusive -OR output gate, G3. A
ing or falling edge of the input signal, and the width of flip-flop, comprised of gates G4 and G5, drives G3's
the pulse is determined by the time delay. other input. The polarity of the output pulses is now se-
In (a), the gate inputs are inverted with respect to lectable, since the state of the flip-flop controls output
each other so that the gate output is in the high state for pulse polarity.
either a high or low input signal. When the input signal If longer or more accurate output pulse widths are re-
switches from high to low, both gate inputs are low until quired, the delay element in any of the three one -shots
the inverter has a chance to change states. During this can be changed to a delay line, such as the ones now
time, the gate output goes low until the inverter available in 14- and 16 -pin dual -in -line packages. The
switches and causes the gate output to return to the high delay device could even be a standard off -the -shelf one-
state. This process repeats when the input signal shot, for example, Texas Instruments' SN74121 or Fair-
switches from low to high. Therefore, for every transi- child's 9601.
tion of the input signal, the gate puts out a negative -go-
ing pulse having a width equal to the delay time of the
inverter.

Selectable one-shot polarity. Either positive or negative output pulses can be generated by controlling high/low states of inputs to exclu-
sive -OR gate. One-shot (a) uses time delay of single inverter to produce negative pulse train. One-shot (b) has additional inverter for positive
output pulse train. Bidirectional one-shot (c) has addressable output polarity, which is determined by state of G4 -G5 flip-flop.

1/6 7404

-LITT
1/4 7486

1/6 7404 1 t6 7404

lhl

7400
1/4 7486

J L
G
1 G,
1/4 7486
Jt_f
XL
7400

G4 )D
1-7
SELECT

7400 1 05 )0

66
 number being decoded. An operational amplifier, con-
Gated MOSFET acts as nected as a comparator, is used to switch each MOSFET
on and off. Each comparator's inverting input is set to a
multiplexing switch dc reference voltage of approximately 1.6 volts.
When a logic 0 appears at the selected decoder out-
by Glen Coers put, the associated comparator's output goes to -14 v.
Texas Instruments, Dallas, Texas The MOSFET that is tied to the comparator output then
conducts, gating the analog input voltage to the volt-
meter or chart recorder. When the decoder output re-
A four -terminal MOSFET makes a handy series analog turns to logic 1, the comparator output switches to
gate for time -multiplexing either ac or dc signals. An ar-14 v, turning off the MOSFET.
ray of these MOSFETS can, for instance, be used to gate a Since MOSFETs are bilateral devices, they can handle
number of voltages that are being monitored period- both positive and negative voltages. For this circuit, the
ically by a chart recorder or a voltmeter, as shown in the maximum input voltage swing is ±-10 v, but can be ex-
figure. Here, the BCD -to -decimal decoder switches one tended by raising the level of the positive and negative
MOSFET analog gate at a time for a given interval before supply voltages that bias the MOSFET's gate terminal.
switching another. The multiplexer circuit has good isolation and leakage
The decoder outputs are high except for the output properties.
MOSFET gate. Analog signals can be gated selectively by using a BCD -to -decimal decoder to switch array of MOSFETs. All decoder outputs
are high, except for the one selected to be the output number. When selected decoder output goes low, output of associated comparator
also goes low, turning on MOSFET and permitting analog input to pass to circuit's output. Both positive and negative inputs can be gated.

E,

CHART
271,1! RECORDER
3N16O OR 3N161

3.3 kl2 VOLT-

SN72741 METER

15 V

E,

INPUTS

SN 1442
3N160 OR 3N161

3.3k22
SN72741 BCD -TO -DECIMAL
DECODER

15 V

TO OTHER
ALL ICs TEXAS INSTRUMENTS ANALOG GATES

67
 The motor's advance sequence is dictated by a pulse
stream that increments the motor's winding -sequence
Stepper drive circuit counter and also triggers the monostable multivibrator
at the input of the drive circuit. When the Q output of
boosts motor torque this one-shot goes low, its Q output goes high, turning
by E. Wolf
off transistor Qi and turning on transistors Q2 and Q3.
Redactron Corp., Hauppauge, N.Y. Since diode D1 is reverse -biased, the potential at Q2's
collector rises to about 24 v, and the capacitor is re-
stricted to a discharge path through the motor windings.
The output power of a stepping motor can be boosted Therefore, the voltage available for the coil common
when it's stepping, while the dissipated power during its lead is nearly two times the supply voltage-or approxi-
dwell (holding) intervals is minimized, by effectively mately 48 v. With this boosted voltage, motor -winding
doubling the supply voltage of the motor -drive circuit current rises rapidly to enhance available torque.
with capacitor charge. The boosted drive circuit shown Once the current buildup time is over, the one-shot
is intended for four -phase 28 -volt motors. completes its timing cycle, turning off transistors Q2 and
During the dwell intervals, the drive circuit supplies Q3, while turning on transistor Qi, which is current -lim-
the pair of motor coils that were energized during the ited by resistor R1. Capacitor C1 recharges to the sup-
previous clockwise step. Transistor Qi is on, but con- ply -voltage level. This cycle repeats for every stepping
ducts only leakage current because transistors Q2 and pulse.
Q3 are off. This permits capacitor Ci to charge to the The timing of the one-shot is not critical. A reason-
supply voltage of 24 v. (Transistors Qi and Q2 form a able timing period would be half of the shortest period
complementary Darlington transistor pair.) between advance pulses.

Stepping up torque. Drive circuit for stepping motor boosts available stepping power without increasing supply voltage. During motor dwell
time, capacitor C1 charges to 24 -volt supply level. When advance pulse triggers one-shot, transistor 01 turns off while transistors 02 and 03
turn on. Coil common voltage then builds to twice the supply level because 02's collector voltage rises by 24 V.

+24 V

1 kS2 MOTOROLA
D1 I IN4003
02
MJE 1091
+5 V
CI
4.7 kS2 COIL COMMON
470 SZ
Q3 1001.LF
2N3566 MOTOR MOTOR MOTOR MOTOF
(35 VI
+5 v 1 kn COIL COIL COIL COIL
A B C D
470 52 0V MOTOROLA
Qi
1 kS2
MJE 1101

2.2 S2

STANDARD
DRIVE CIRCUIT 0V
ADVANCE ONE-
PULSES SHOT
74121

ADVANCE PULSES
CLOCKWISE
ENERGIZATION
SEQUENCE
AB
BC
CD
DA
MOTOR COMMON VOLTAGE
+48 V
+24 V fTh 1i

68
 age over a center -frequency range of 4.5:1.
Voltage -tuned filter varies Resistor R1 and supply voltage VGG bias the FET at
-3 volts, thereby setting the filter's first breakpoint for
center frequency linearly Vc = 0 at. 1,460 hertz. For negative values of control
voltage, only diode D1 conducts, and the gain of the
by Vassilios J. Georgiou amplifier is determined by resistors R2 and R3.
University of Massachusetts, Amherst, Mass. For positive control voltages, diode D2 conducts, and
amplifier gain is about half the value it is for negative
control voltages. As a result, the lower portion (right-
Although a voltage -tunable multiple -feedback active hand side) of the filter's tuning curve tilts upward and
filter generally offers constant gain and constant band- aligns with the upper portion (left-hand side) to form a
width throughout its tuning range, its tuning curve for linear characteristic.
center frequency versus control voltage is usually highly Diodes D3 and D4 define the second filter break-
nonlinear for extended frequencies. This is due to the point-at Vc = 2.4 v-by further reducing amplifier
nonlinearity of the field-effect transistor, which is used gain and extending linear operation to 570 Hz. Prefera-
as a variable resistor, and because the center frequency bly, the germanium diodes, D1 and D2, should be gold -
varies inversely with the square root of the FET's drain - doped so that they have a low forward -voltage drop. 0
source resistance. Employing feedback, therefore, to
BIBLIOGRAPHY
linearize the FET'S behavior is not a solution. G. Deboo and R. Hedlund, -Automatically Tuned Filter Uses IC Op Amps," EDN/EEE,
Instead, a modified version of the diode function gen- Feb. 1, 1972, p. 38.

erator can be used to drive the FET'S gate terminal. For

the voltage -tunable bandpass filter shown, center fre-
quency remains nearly linear for changing control volt-

Straight-line tuning. Modified diode function generator (in color) drives FET variable resistor for voltage -tunable bandpass filter, causing
center frequency to vary linearly with control voltage over 4.5:1 frequency range. Gain of function -generator amplifier is reduced for poditive
control voltages to raise lower section of (uncompensated) tuning curve. Compensated filter remains nearly linear.

-2.5

2.0 UNCOMPENSATED

COMPENSATED
-1.5

17;
x - 1.0

- 0.5
CONTROL VOLTAGE IV)
-4 -3 -2 -1 0 1 2 3
1 I t 1 1 1 i

-1 -2 -3 -4
FET GATE VOLTAGE IV)
6 kS2 1
SIGNAL
VOLTAGE Nil -/%111---.

SILICON
IN914 OUTPUT
VGG VOLTAGE
6.2 kS1
GERMANIUM -7.4 V Vo

11.4 Ic12
R,
13.4 kS2
CONTROL
VO1 TAGE V`

10 ki-2
DI 6.98 k12

-6 V

:',1(),,),;(), A

69
 temperature -induced change in the base -emitter voltage
of transistor Q2 to be cancelled by an equal -in -magni-
Temperature -stable decoder tude, but opposite -in -polarity, change in the base -emit-
for modulated pulse widths ter voltage of transistor Q3.
The circuit's output voltage is referenced to the lower
by H.R. Beurrier point of capacitor Ci's charge/discharge cycle. When
Bell Telephone Laboratories, Murray Hill, N.J. the input is negative, diode D1 reverse -biases transistor
Q2's collector so that capacitor C1 always discharges to
the same level. Diode D2 and resistors R5 and R6 clamp
Besides offering exceptional temperature stability, a the capacitively coupled input control signal so that
pulse -width -modulation decoder for remote propor- transistor Q2's base is driven slightly negative. Diode D3
tional radio control produces a presettable fail-safe simply fixes the base voltage of transistor Qi.
analog output when the input control signal is inter- With the component values shown, the circuit will
rupted. The circuit converts a time -modulated pulse in- convert a pulse width of 1.25 ±0.63 milliseconds to an
put to an analog output. analog output of 2.4 ±0.6 volts dc. The 1-kilohm input
Transistors Qi and Q2 form a sawtooth generator resistor simply reduces the loading on the driving
with a ramp output that starts when the input control source.
signal goes positive and that resets when the input goes
negative. The control signal switches Q2 alternately on +4.8 V
and off.
When the base of Q2 is driven positive by the input,
this transistor turns off and capacitor Ci charges with
the constant current supplied by transistor Qi. (The
longer the input pulse duration, the higher Ci's ramp
voltage and the resulting output voltage.) When Q2 is
driven negative, it conducts, pulling its emitter voltage
in the negative direction and partially discharging ca- ANALOG
pacitor C1. 1 R2 OUTPUT

Emitter -follower Q3 acts as a peak -voltage detector PULSE

47µF
(+2.4 V)
INPUT
that charges capacitor C2 positively when its base is
driven positive by the ramp voltage of capacitor C1. If 3

the voltage across C1 is too negative to forward -bias the 4.7 /IF
base -emitter junction of Q3, capacitor C2 discharges
through resistor R1. This portion of the circuit, there-
fore, acts as a diode peak detector with power gain. Re- SILICON
sistor R2 and capacitor C3 are connected as a simple
first -order output filter.
Once the input signal is terminated, the base voltage PWM decoder. Circuit detects pulse -width -modulated signals, sup-
of transistor Q2 settles to a level determined by the volt- plying dc analog output over wide operating temperature range. If in-
age divider of resistors R3 and R4. This voltage level is put control signal is interrupted, decoder output goes to preset fail-
coupled to the output through transistors Q2 and Q3, safe level. Transistors ()land 02 and capacitor C1 make up sawtooth
which are connected as cascaded emitter -followers. The generator that drives. transistor 03, which acts as peak -voltage de-
complementary arrangement of Q2 and Q3 causes any tector. 02 and 03 are cascaded complementary emitter-tollowers.

circuits and linear diode gates are used. The organ,

which requires only one gate for each frequency, adds
C-MOS sums up tones tones to derive the harmonic content of each voice.
for electronic organ Sound reproduction varies smoothly, too, so that there
are no unpleasing key clicks.
by Robert Woody The circuit shown generates the eight frequencies of
Hercules Inc., Radford, Va. note A. Adjacent frequencies of the same note have 2:1
ratios and are separated by an octave. Similar circuit ar-
rangements generate frequencies for notes B through G
An electronic organ that produces a wide variety of and five sharps, for a total of 12 notes. These 12 notes,
voices, either singly or in combination, can be built with each having eight frequencies, comprise the 96 fre-
relatively simple circuitry if complementary-mos logic quencies in the organ. Individual frequencies are spaced

70
at 6% intervals, from 32.7 to 7,902 hertz. goes to the supply voltage level, D1 is back -biased so
The eight square -wave outputs of the Hartley oscilla- that current through R1 charges C1, causing the gate
tor and the binary divider have precise frequencies, but output to have a rising slope. The capacitor discharges
are not musical because square waves contain only the quickly through the diode when the square wave re-
fundamental frequency and odd (not even) harmonics. turns to ground, and the cycle repeats.
Furthermore, if these square -wave frequencies are Resistor R2 and capacitor C2 slow the application of
turned on and off directly by key switches, the sounds current to resistor R1, permitting the gate to turn on and
begin and end too abruptly to be musically pleasing, off gradually without clicks. Since C1 charges for half
and key clicks will be heard. the cycle of square -wave frequency f, time constant
A diode gating circuit, like the one consisting of diode R1C1 is half the period off, or:
D1, resistor R1, and capacitor C1, is used to convert each C1 =
square wave to a sawtooth. This supplies the even har- Values for capacitor C2 are selected to approximate the
monics and helps to turn the tones on and off gradually switching times of organ pipes. Since the output of each
without clicks. gate is linearly proportional to its dc biasing voltage, the
The square wave alternates between the supply volt- tone amplitude needed can be metered out precisely.
age and ground. While the square wave is at ground, D1 To sum tones to create the desired organ voices,
is forward -biased by current flowing through R1, and which comprise several harmonically related tones, each
the gate output is also at ground. When the square wave gate output must be filtered to reduce its harmonics
C-MOS makes organ music. Complementary-MOS ICs add tones to produce eight frequencies that make up note A. Each of the remaining
six notes, B through G, plus five sharps, requires its own Hartley oscillator and binary divider network to generate all 96 organ frequencies.
Diode gate at each divider output converts that square wave to a sawtooth to get the necessary even harmonics.

f C1 C2
Note (pH
HARTLEY (Hz) (PH
OSCILLATOR
RCA Y,CD4001AE Al 55 0.15 0.47
A2 110 0.082 0.33
A3 220 0.047 0.22

=JO (A8) A4
A5
440
880
0.022
0.01
0.22
0.22
A6 1,760 0.0047 0.22
10 k12 (A7)
0.0047 pF
1

12
A7 3,520 0.0022 0.22
A8 7,040 0.001 0.22

60-130 mH BINARY
DIVIDER

MILLER OUTPUT
#6324 (Volpk-pk
0.1 µF (A2)
4 AAAAII

0.082µF
D1 OFF
(All OUTPUT ON
NW_ CI
_AAA_
0 15 )IF 131

56 IcS2
RCA CD4004AE 0.33 µF
56 kE2
C2
0.47 µF
R2
56 kS2

dc BIAS
ALL DIODES: 1N3064
Vim SUPPLY = 10 V
30 V
164.8 Hz dc BIAS
(E3)

0-30 V KEY BUSES

>): TO OTHER KEY SWITCHES

V8 ---0 t,

71
 HARMONIC CONTENT OF SOME TYPICAL ORGAN VOICES
down to the simplest voice produced in the organ,
HARMONIC FREQUENCIES
VOICE
2 3 4 5 6
namely the flute. Since the eight frequencies of a note
1 7 8 9 10 11 12
are multiples or submultiples of each other, they can be
OPEN FLUTE I I m
handled by one low-pass filter. This means that only 12
low-pass filters are needed for all 96 frequencies.
OPEN FLUTE,
OCTAVE HIGHER i The diodes between the gates and the key switches-
diode D2, for instance-allow a given tone to be
ABOVE FLUTES
sounded by a number of different key switches. All of

!III'.
SOUNDED TOGETHER I a I a

the key switches indicated are operated by the key for

TRUMPET note Al. The voltages (VI, V2, V3, . . .) on the key buses
determine the amplitude at which the tones are
FRENCH HORN I 1 I sounded (V5 controls the fundamental frequency tone,
VIOLIN III I II III a I a
V2 the second harmonic, V3 the third, .). The key bus
.

voltages set the harmonic content and, therefore, the

voice of all the keys on the keyboard.
.

STOPPED FLUTE I I I To conserve circuitry, the third harmonic, note A3.

borrows a tone from the gate (not shown) used for note
CLARINET I I .
I I . I . E3. The resulting 0.2 -Hz frequency error cannot be
heard.

has excellent saturation characteristics, the VcE drop

Transistor gating circuit can be held to approximately 10 millivolts by using a
forced beta of 0.1.) Driving several high -impedance in-
cuts signal delay to 100 ps puts or perhaps one low -impedance input will cause
some loss of noise immunity. Although driving a termi-
by Arthur J. Metz nated line is not recommended, the gate may be driven
Tektronix Inc., Beaverton, Ore. from a terminated line.
As with any ECL design, care must be taken in laying
out the circuit to realize maximum performance. Since
Frequently in emitter -coupled -logic design, a high- the fun logic -voltage swing appears at the transistor's
speed data signal must be gated by a dc control signal base terminal, the base biasing resistor should be lo-
that is generated by a contact closure or a transistor - cated close to that terminal to eliminate the -
transistor -logic gate. When the propagation delay in the mission line effects of an interconnecting lead.
high-speed signal path must be held to a minimum, a
simple transistor circuit can probably provide the fastest
way to perform the gating function.
Propagation delay for the transistor gate is as low as
100 picoseconds, compared to the I -nanosecond delays
of the fastest integrated -circuit gates presently avail-
able. Moreover, the transistor gate consumes less power
and costs less than the lc gate, especially if an addi-
tional Ic gate package is needed to perform the gating
function.
A typical application for the transistor gate is shown
in the figure. Here, signal delay between the flip-flop's
(5- output and its D input must be minimized to realize
the fastest possible toggle rate.
With switch Si open, the transistor's base -emitter
junction is reverse -biased. An internal pull -down re-,
sistor of about 50 kilohms at the flip-flop's D input
holds all data inputs near ground potential, reverse -
biasing the flip-flop's input stage (by more than 3 volts)
and assuring the rejection of any signal that passes
through the internal capacitance of the transistor. When
the switch is closed, the transistor saturates and pro-
vides a low -impedance signal path. For rrL appli- Gating ECL signals. Bipolar transistor gate can transfer logic sig-
cations, the function of switch Si is implemented by the nals with propagation delay of as little as 100 picoseconds. With
appropriate TTL device. switch closed, transistor saturates and gates signals from flip-flop's
The noise immunity of the gating circuit is main- 10 output to its D input. When. switch is open, transistor is cut off,
tained by keeping the transistor's collector -emitter volt- and strong reverse bias at D input rejects all unwanted stray signals.
age drop low. (With the type 2N2907A transistor, which The switch can be a pair of contacts or a TTL device.

72
 C1 is charged to produce an output voltage step:
Staircase generator AV° = VREFT/RIC1
Between one-shot output pulses, transistor Qi is off, and
resists output drift the integrator becomes a hold circuit and maintains the
output constant.
by Maxwell Strange Capacitors C1 and C2 are the two components with
NASA, Goddard Space Flight Center, Greenbelt, Md. the greatest effect on step height stability. If the same
type of capacitor is used in both the one-shot and the
integrator sections of the circuit, the temperature coeffi-
Tracking capacitors that mutually cancel temperature cients of C1 and C2 will cancel. Staircase risetime is pro-
drift make a simple analog staircase generator, which is portional to capacitor C2, while integrator slope is pro-
as accurate and stable as expensive circuits that employ portional to capacitor C1, so that step height is
precision digital -to -analog converters. Additionally, the unaffected by a similar percentage change in both ca-
strictly analog circuitis easier to adjust for any number pacitors. The period of the one -shot's output pulse is
of steps and to any step amplitude. directly proportional to the ratio of C2/C1.
The generator essentially consists of two sections, a As for the output voltage droop that occurs during
one-shot and an integrate -and -hold circuit. The one- the integrator's hold mode, the value of C1 must bz
shot, which drives the integrate -and -hold circuit, is trig- large enough to keep it negligible over the staircase
gered by an oscillator or system clock that determines cycle. For the components shown, output droop is only
the generator's stepping rate. During the high period about 1 millivolt in 10 seconds, and step amplitude is
(T) of the one -shot's output pulse, integrating capacitor stable within ±0.2% from 0°C to 50°C.

Stepping up. Staircase generator employs one-shot to drive integrate -and -hold circuit. During one-shot period, capacitor C1 charges and
steps up output voltage. When one-shot is off, integrator section holds step height constant. Output voltage droop is kept to 1 millivolt in 10
seconds. Step amplitude drift is held to ±0.2% because temperature coefficients of same -type capacitors Ci and C2 cancel.

h-T--2 32ms START/RESET

CLOCK
INPUT OV
OV
STAIRCASE
OUTPUT

0.5V MIN NATIONAL

LM201
v
Av.= 0.500 V/STEP
NATIONAL
LM208

0.001 NF
(FREQUENCY
COMPENSATION/

'1% METAL FILM

"5% COMPOSITION
C, and C2: POLYCARBONATE OR MYLAR

73
 the capacitor to about 39 v through the diode bridge.
The transformer, which has a no-load secondary voltage
Four -ampere power supply rating of 28 v ac, can deliver the 4-A operating current
but will not deliver this voltage under loading because
costs just S13 to build of its core and copper losses. Although capacitor voltage
by Joseph Ennis
may drop to around 24 v during peak loading, the series
Automation Industries, Inc., Vitro Laboratories Division, Silver Spring, Md. regulator will continue to provide a smooth 20-v output.
A current of more than 4 A could be controlled by the
regulator transistor with the appropriate heat sinking,
The cost of building a regulated power supply can be but the heat sink would cost more than the transistor.
lowered to around $13 if a large capacitor is used to The heat -sink area is designed to handle only normal
store energy at a higher voltage than is necessary. Un- worst -case operating conditions and does not allow for
der normal operating conditions, the supply, which is any current foldback dissipation.
primarily intended for powering a stereo amplifier, can Instead, this dissipation is provided by two replace-
deliver an output of 4 amperes at 20 volts with load ment -type automobile lamp bulbs. Besides acting as
fluctuations down to 18 hertz and with regulation to fuses when there is a short circuit at the output, they re-
better than 5%. duce the voltage drop across the regulator transistor,
A high -value capacitor, one measuring tens of thou- thereby decreasing the power it has to dissipate during
sands of microfarads, stores cliarge so that only a small normal supply operation.
amount of transformer iron is needed to produce the To change the supply's output voltage to 15 v, the
4-A operating current. The resulting higher -than -re- 23-v zener is replaced by a 17-v one. If a 5-v supply is
quired capacitor voltage is then dropped to the desired needed, a transformer with a secondary voltage lower
20-v level with a transistorized series regulator. More- than 28 v should be selected to reduce the voltage drop
over, two inexpensive incandescent lamp bulbs are used that the regulator must handle. A negative -voltage sup-
for short-circuit protection, rather than a more costly ply can be constructed by substituting complementary
current foldback technique. transistor types TIP30 and TIP32 for the type TIP29
With no load at the output, the transformer charges and TIP31 transistors.

Economical regulated supply. Parts cost for power supply is pared to absolute minimum by storing energy in 33,000-microfarad capacitor
at higher -than -required voltage level. This allows a fairly lossy, and therefore inexpensive, transformer to be used. Incandescent lamp bulbs
serve as fuses in case of a short circuit and reduce voltage seen by series -regulator transistor. Output is 20 volts at 4 amperes.

BULBS: GENERAL ELECTRIC

1206 (201 EACH)
TRIAD DIODES: 100-V PI V, 6A
MZ5 (391 EACH)
$4.35) HEAT SINK TEXAS INSTRUMENTS
($1.09) TIP31 ($1.35)

115Vac 28 V ac

20 V
0-4A
33,000µF GENERAL TEXAS
HEAT SINK ($2.64) ELECTRIC 23V INSTRUMENTS
(241) ZD23 TIP29 1291)
TOTAL PARTS COST: $13.12 IS1 20)

train. The harmonic frequency generator, which consists

of a clock pulse generator, a decade counter, and a few
Binary division produces NOR gates and flip-flops, produces square -wave outputs
at frequencies f. through 10f0.
harmonic frequencies The clock frequency must be 2n times faster than the
by Donald DeKold frequency of the highest harmonic of interest (n is the
Santa Fe Junior College, Gainesville, Fla. number of flip-flops used for the binary division).
Therefore, to produce the highest harmonic, 10f0 in this
case, the clock output is simply divided down by 2n. For
Harmonically related frequencies-more specifically, a all harmonics but the fifth, however, the clock signal
fundamental frequency and its first nine overtones-can must be properly gated before it can be divided.
be generated with binary division of a blanked pulse To understand why this is so, consider what happens

74
with the ninth harmonic. The uppermost NOR gate the reduction of phase distortion is still evident.
passes and inverts the first nine clock pulses to reach it. The proper locations of this waveform's transitions-
But the arrival of the tenth pulse coincides with the arri- those that a true square wave of frequency 9(2n-4)fo
val of a high from the decade counter. Since the would produce if divided in the same way-are denoted
counter's output stays high for the full duration of the by the colored dashed lines. The leftmost switching
tenth clock pulse, the gate's output remains low, pre- edge, which exhibits the largest error, is advanced by
venting the tenth pulse from propagating. This tenth - 18° from the true edge. At any stage of division, the
pulse rejection occurs every 2nfo times per second. maximum uncertainty in a transition will always be less
Clearly, the gate's output pulse frequency is nine - than the period of the clock frequency.
tenths that of the clock frequency, because one pulse is Harmonic frequencies 6f0, 7f0, and 8f0 are developed
blanked for every ten delivered. The gate's output in much the same way as 9fo. Two clock pulses must be
waveform may be regarded as the 2nth overtone of 9f0, blanked for 8fo, three for 7fo, and four for 6f0. To gener-
but one that is badly distorted with respect to phase. ate harmonics 5f0, 4f0, 3f0, 2f0, and f0, binary divisions of
The nine pulses making up a full period of this wave- harmonics 10fo, 8f0, and 6f0 are performed as indicated.
form are cumulatively advanced in time from their Complementary-Mos integrated circuits can be used
proper locations as they progress through one complete to build this harmonic generator. If RCA's type
period of the binaryffiovertone of theffifundamental. CD4017A decade counter is chosen, the maximum
This phase distortion can be almost eliminated by clock frequency is limited to 5 megahertz. (An unused
successively dividing the gate output by two with flip- dual -input NOR gate in RCA'S type CD4001A package
flops, as shown in the timing diagram. Waveforms A can be employed as the clock inverter.)
through F illustrate how the blanked space can be made
smaller by flip-flop divisions of 2, 4, 8, and 16. Although
only 21/2 cycles of divided -by -16 waveform F are shown,

Harmonic generator. Single clock signal can be used to create fundamental frequency f° and its first nine harmonics. Clock frequency can
be divided directly for tenth and fifth harmonics, but other harmonics must be gated to produce appropriately blanked pulse train. (For in-
stance, one clock pulse out of ten is blanked by top NOR gate for harmonic 9f..) Flip-flops then divide gated outputs.

fr1012")f
(A)
CLOCK
DIVIDE -8Y-2" 10fo
PULSE
n F LIP -F LCPS
GENERATOR RCA CD4000A

101 ICI

DIVIDE -BY -2" 9f0

n FLIP FLOPS

8 8f°
DIVIDE -BY -2"
7 n FLIP-FLOPS
5
7f0
4 -RCA IhCD4002A
3 DIVIDE -BY 2'
n FLIP-FLOPS
2
RCA MD4002A, DIVIDE BY 2 6fo
CD4001A, CD4000A FLIP= -LDP
INVERTER DIVIDE -8Y-2"
n FLIP-FLOPS
r 5fo

DIVIDE -BY 2
FLIP-FLOP
1 2 3 4 5 6 7 8 910 4f0
r
(CLOCK) MUIR If 1 J. 1
DIVIDE -BY -2
FLIP-FLOP
(GATE) Mil11E11.11f if 1. - .1 1. J - 3fo
(÷2)
DIVIDE BY -2

L FLIP-FLOP 2f.

(-81
DIVIDE -BY -2
FLIPFLOP
(÷ 16)
I 1E' r 15.7° 13.5° 11 2

75
 (in 0.5 to 0.7 millisecond) and triggers the monostable.
Precision auto tachometer The current through R1 keeps Qi on and prevents Ci
from charging until the points close. If the points
squelches point bounce bounce upon closure, they will not be open long enough
to allow C1 to charge and fire Qi again.
by James B. Young For every point opening, the monostable produces a
Canadian General Electric Ltd., Peterborough, Ont., Canada pulse having a fixed width and amplitude. Normally,
the output stage of the operational amplifier produces a
negative saturation voltage. But when a positive trigger
A tachometer circuit for automobiles with capacitive - from the relaxation oscillator is applied through diode
discharge ignition systems suppresses point bounce D1, the op -amp's output switches to a positive satura-
while measuring motor rpm accurately to within 1%. tion voltage, causing capacitor C2 to charge positively
The circuit, which has an operating temperature range through resistor R2. Capacitor C2 stores the charge until
of -20°F to 150°F, can also be used as a temperature - e- is greater than e+, and the op amp switches back to
compensated ratemeter or to eliminate relay -contact its stable state.
bounce. Diode D2 clamps the voltage across C2 to about
Many automobile tachometers do not work properly -0.6 v, while diode D3 provides temperature compensa-
with a capacitive -discharge ignition because this type of tion for changes in D2's junction -voltage drop. Both of
system employs the breaker points only for triggering these diodes should be kept in thermal contact with
an SCR. The voltage waveform across the breaker each other. Since the op amp is left floating so that it
points, therefore, consists of a series of 14 -volt pulses, can be operated from a car's single supply voltage, it
rather than the 200-v spikes that exist in the usual ket- has a small positive output voltage when in its untrig-
tering ignition system. gered state, making capacitor C3 necessary to decouple
The tachometer circuit shown is composed of three the meter.
sections: a relaxation oscillator at the input for point Zener diode D4 and resistor R3 regulate the output
bounce suppression, a monostable multivibrator for against supply voltage variations, and the thermistor
pulse generation, and a buffer for driving a meter. compensates for temperature variations in the base -
Unijunction transistor Qi is operated with an emitter emitter voltage of transitor Q2. If a meter with a full-
current that is larger than its valley current so that it scale current rating of less than 5 milliamperes is used,
will not turn off after triggering. When the points open, the thermistor, as well as transistor Q2 and resistor R.4,
capacitor C1 charges through resistor R1 until Qi fires can be omitted.
0
Measuring rpm. Intended primarily for automobiles with capacitive -discharge ignitions, tachometer circuit accurate within 1% is immune to
breaker -point contact bounce. When points open, capacitor C1 charges until unijunction transistor 01 fires and triggers one-shot formed by
op amp. Point bounce is suppressed because C1 takes 0.7 millisecond to charge before 01 can fire. Circuit can operate from -20°F to 150°F.

8 CYLINDERS: 0.47 MS2

6 CYLINDERS: 0.68 mil
4 CYLINDERS: 0.90 MS2
C2
R2
+12 V
+12 V
0.01 /IF FAIRCHILD
+12 V
µA 741
SUPPLY: 10-17 V, ±1%
1.8 kS2 C3

8 µF

D,
R,
D4 METER
200 E-2 (100 mA)
(6.6 V)

BREAA-KER
POINTS 1.4 4(12 1.4 kS2
R3
168 Si N)
1.1°
200 S2@ 25°C
(THERMONETICS
1810)
50 12
L

76
 being desensitized at high signal levels and permits low-
level targets to be detected even in the proximity of
Agc rf threshold detector heavy clutter.
provides fast slewing With the threshold power level (typically 0 to 5 dBm)
applied at the detector's rf input, the input voltages of
by Roland J. Turner the operational amplifier are equalized so that there is
American Electronics Communications Corp., Lansdale, Pa. no output voltage. For signal levels below the threshold,
the op amp's inverting input swings positive, relative to
the noninverting input, and the op -amp output is
In both radar and communication systems, an auto- clamped by diode D1 near the quiescent bias voltage set
matic -gain -control loop is commonly employed to keep by potentiometer R1. This clamp prevents the output of
signal level constant for enhanced signal detection. As the op amp from swinging to the negative supply volt-
systems become more adaptive, the agc loop must be age and enhances the recovery time of the detector at its
more sophisticated, frequently forcing simple circuits to threshold level.
perform multiple functions. As soon as the rf input develops 3 millivolts of recti-
Meeting this demand is an rf threshold detector for fied bias voltage across capacitor C1, the detector's out-
the agc loop in an rf receiver. The detector can process put goes positive to 1 v, thereby activating the agc loop
signals of 1 megahertz to 1 gigahertz with a slew rate of in which it is installed. Only the differential offset volt-
several volts a microsecond. It also permits delayed agc age between the dual Schottky diodes, D2 and D3, is im-
operation and can perform over a wide range of tem- pressed across the op amp's differential input. The op
perature and supply -voltage variations. amp provides 50 decibels of gain, while Schottky diode
Furthermore, below the detection threshold, the sig- D3 provides automatic temperature stabilization for the
nal can be processed linearly in the rf section of the re- detector. Diodes D4 and D5 are temperature -stabilized
ceiver. Above the detection threshold, the output of the reference diodes that desensitize the detector to power
rf section is rapidly leveled so that such detrimental ef- supply variations.
fects as limiting and hangup cannot occur in subsequent The rf section of the receiver is gated by switching the
i-f stages. This fast action prevents the receiver from voltage at the detector's gate input from its normal
Closing the loop. Threshold detector for agc loop in rf receiver can handle broadband signals with frequencies of 1 megahertz to
gigahertz at slew rates as fast as 1 volt/microsecond. Dual Schottky diodes and high -slew -rate op amp account for circuit's speed. Tempera-
ture compensation, from 0°C to 60°C, is provided by diodes D2 and D3, while diodes D4 and D5 compensate for supply variations of ±6%.

+12 Vdc
A

a 4 46.4 kt2
100 f2 100 E2
BIAS
w3
ADJUST
5 kE2
2

1 MP. 750 P.
0 1

0
-29 -27 -25 -23 -21 -19 -17 10 kE2
POWER INPUT IdBm1

10 kt2
GATE I

.01
INPUT BUFFER
IN4148 IN4376 0.1 pF 0.1 pF AMPLIFIER
3 pF OUTPUT TO
33 pF ATTENUATOR
38.3 kS2
10-8 Vdc,
ri INPUT
261 100 mA)
(+5 dBm) c,
IN4376 31.6 E2
NATIONAL
0.15µH 5PF 6
" NH0002
4
03

F
6 7
3 33 pF 330 pF

1000 SZ 100 .E2 100 E2

-9.7 V
0.1 pF
HARRIS
HA2620
05
HP5082 SERIES
IN758A

-12 Vdc

77
-10-v bias level to 10 v. This overrides any signal condi- sumed by the static offset voltage of the transfer func-
tion, forcing the detector's output to go to 8 vdc and tion over the same temperature range.
forcing the associated attenuator in the agc loop to its Moreover, the transfer function remains constant and
maximum attentuation state. unaffected by ±6% variations in the supply voltages.
For power levels above the detection threshold, the Less than 0.3 dB of loop dynamic range is consumed by
detector produces 4.2 v at its output per decibel of input static offsets in the transfer function for the same supply
power over the operating temperature range of 0°C to variations. For signals greater than 2 dB above thres-
60°C. Less than 0.5 dB of loop dynamic range is con- hold, the detector slew rate exceeds 1 v/ps.

come logic 0, and an input pulse at frequency f2 clocks

Phase comparator FF2 and produces a logic 0 at its Q output, the preset
delay containing three NAND gates resets the flip-flops.
for servo loops The propagation delay time of these NAND gates assures
that both flip-flops are preset.
by Francis E. Adams The time lapse between FF1 being clocked and FF2
San Bernadino Microwave Society, Corona, Calif. being clocked determines the duty cycle of the pulse
train seen by the algebraic adder. As input pulse fre-
quencies f1 and f2 approach each other, this duty cycle
An easy -to -build phase comparator offers a number of becomes smaller. When the two inputs are in phase
advantages for use with dc servo motors and in voltage- = f2), the duty cycle is zero, except for a narrow con-
controlled -oscillator loops. The output of the compara- tribution due to the preset time of the NAND gates,
tor is zero for inputs that are in phase, and, unlike the which is usually less than 1 microsecond.
output of discriminators, it remains real, even for ex- If f1 is greater than f2, there is a positive output cur-
treme differences in the two input -pulse frequencies. rent that is proportional to the phase error. If f2 is
Moreover, the circuit, which is composed of low-cost, greater than f1, then there is a negative output current,
readily available ics, can operate at a frequency that is which is also proportional to the phase error. For
limited only by the type of devices employed. G = f2, the output current is zero.
Both J -K flip-flops, FF1 and FF2, are preset so that
their Q outputs are at logic 1. After an input pulse at
frequency f1 clocks FF1 and causes its Q output

PRESET DELAY

00-00-cCI SN7401
ALL ICs: TEXAS INSTRUMENTS

a
ft f2
FF1 FF2

C C K
f2 INPUT f2 > ft
0
ALGEBRAIC ADDER
SN7473 ft > f2
INPUT
TO LEAD -LAG
NETWORK

IN914
f2 > ft
+15 V
FF2,171 0 39 k11
SN7401

GROUND TO
FFI,Q 0 INHIBIT
ft > f2
FF2, Q 0 Comparing pulse frequencies. Phase comparator produces output
that is proportional to difference between input -pulse frequencies.
Output is positive when 11 is greater than f2, zero when fl equals f2,
FF1, Q 0
and negative when f1 is less than f2. Duty cycle of signal driving alge-
braic adder depends on when flip-flops FF1 and FF2 are clocked. In -
phase inputs result in zero output because duty cycle is zero.

78
Diode switching matrices
make a comeback
Poor noise immunity of standard logic gates has restricted diode use,
but cost advantages of matrices now can be realized by constructing
these networks with high -noise -immunity integrated circuits

by Dave Guzeman, Teledyne Semiconductor. Palo Alto, Calif.

O Since the advent of high -noise -immunity logic, diode that of conventional TTL becomes important for code
switching matrices are being used more extensively in conversion applications; for example, decimal -to -ex-
logic systems, particularly those for code conversion, be- cess -three encoders that use keyboard switch inputs.
cause diode gates cost less than standard logic gates. With high -noise -immunity logic, the keyboard switches
Once a very popular logic building block, the diode can be located remotely from the logic cards, and worst -
switching matrix had lost ground because of the poor case noise immunity can be as high as 4.3 v.
noise immunity of conventional logic gates. The excess -three encoder of Fig. 2(a) consists of 10
The two most common 5 -volt logic families, transis- keyboard switches, four high -noise -immunity dual -in-
tor -transistor logic (TTL) and diode -transistor logic put NAND gates and several diodes. Essentially, the en-
(DTL), usually experience noise immunity degradation coder uses the same principle as the simple diode AND
because of the voltage drop, about 0.7 v, across a for- gate. As individual keyboard switches are closed, the
ward -conducting diode. Improvement in circuit noise diodes tied to the supply voltage through pull-up resis-
immunity becomes important for code conversion ap- tors become forward -biased. In this case, the diodes are
plications, such as decimal -to -excess -three encoders. wired to make the input to the NAND gates the comple-
And substituting a simple diode gate for a standard gate ment of the desired excess -three code.
can reduce parts costs-often by a factor of four. The NAND gates restore the input signal levels for full
A noise immunity problem arises whenever simple noise immunity in the system. An inhibit line is also
diode gates are driven by conventional TTL or DTL. provided to block or transfer data from the keyboard
Since noise immunity voltage for any gate is the differ- switches into the system logic.
ence between the guaranteed input threshold voltage This encoder, however, may present a problem since
and the gate output voltage, the already narrow 400 - its output becomes 1111, which can be mistaken for
millivolt noise immunity offered by most TTL is easily some other number, causing an unwanted output when
exceeded by the additional diode voltage drop. the inhibit line drops low. A more usable circuit is one
whose output goes to 0011, the excess -three code for
Noise immunity: problem and solution
Suppose a simple diode AND gate is driven by a stan- +5 V
dard era, active output gate whose maximum output al
voltage in the logic 0 state is 0.4 v. As shown in 0.4 V
hi- 0.7 V /1.1V
Fig. 1(a), the forward -biased diode at input A, which AI
drops around 0.7 v, makes the logic 0 output voltage
(VOL) of the driving gate equal to 1.1 v. Since the guar-
D- B
OUTPUT

14
anteed input logic 0 threshold of conventional TTL de-
vices is 0.8 v, the noise immunity becomes 0.8 - 0.4 =
ALL DIODES IN914 +12 V
0.4 v, exceeding TTL's 400-mv noise immunity.
Consider the same circuit when high -noise -immunity
logic is used, as indicated in Fig. 1(b). Now, the guaran-
teed driving gate Vol, increases to 1.5 v and the guaran-
teed input threshold increases to as high as 5 v. Adding
the 0.7-v diode drop to the 1.5-v VoL of the driving gate
brings the output voltage of the diode AND gate to 2.2 v.
The worst -case logic 0 noise immunity becomes 5 -
2.2 = 2.8 v. Many devices in high -noise -immunity fam- 1. Noise Immunity. When conventional TTL gate with active outpu
ilies have open -collector or passive pull-up outputs drives diode AND gate (a), noise immunity of TTL driving gate is ex-
whose guaranteed VOL is 0.5 v. Then noise immunity of ceeded because of the diode's forward voltage drop. Using high -
the diode AND gate is better yet: logic 0 output is 0.5 + noise -immunity logic driving gate (b) solves the problem since its
0.7 = 1.2 v, and noise immunity is 5 - 1.2 = 3.8 v. guaranteed input threshold is 5 V, rather than the 0.8 V of standard
This significant improvement in noise immunity over TTL. Noise immunity, which is 0.4 V for (a), increases to 2.8 V for (b).

79
 +12V +12V +12V +12 V
+12V +12V +12V +12V
b
ALL 8.2 kS2 ALL 8.2 kS2

ALL DIODES 1N914

_L-
5

)0( TELEDYNE
321 w
w p
1 Cr 1- LaJ
x=
2 I-0. L.Li2
4 t335
8 to
STROBE TELEDYNE
302 HD,
INHIBIT

2. Excess -three encoding. Diode matrix (a) converts keyboard

switch inputs to complementary excess -three code. NAND gates in-
vert the signal and restore signal level. Another arrangement (b) Ic
+12V +12V +12V +12V
codes the 1111 output, which could be mistaken for another num- ALL 8.2 kS2
ber, to 0011 (excess -three code for zero) whenever strobe line goes

/
low. Third encoder (c) employs quad latch for interfacing keyboard °
_L-
outputs with holding register. In this switching matrix, the diodes pro- 1

vide a true, rather than complementary, excess -three output code. 2

3 )1/
(al 4 ALL DIODES IN914
7400
5
)11.

J 0 0 OUTPUT _L- \ 6 Jkle

\91 D,
/471 7471 7471 74/1
K K T it
7404
K
_L-
T 8
-L- .., GI
CLOCK __. GI
TELEDYNE
2 02 (12
370 (12
4 03 QUAD a3
b
8.2 kS2 8 04 LATCH - - (13
IN914
-. --+12 V Q4
04
0 -0 0 OUTPUT
ENABLE
TELEDYNE TELEDYNE TELEDYNE IN914 TELEDYNE
312 _ 312 _ 312 _ 312 _
K 0 K T 0 K T a
the keyboard outputs must be fed into a holding regis-
CLOCK -
ter, as indicated in Fig. 2(c). Here, the diodes are con-
nected within the matrix to provide true excess -three
3. Saving dollars with diodes. Substituting a simple diode gate for outputs, rather than the complementary form. These
standard logic gates can considerably reduce parts cost-often by a outputs then drive a quad latch, which consists of four
factor of four. For example, serial NAND gate and inverter (a) of Type D flip-flops. The latch is provided with an enable
BCD decoder/counter can be replaced by diode AND gate (b). input that prevents data from being entered into the
flip-flops, except when the enable line is low.
zero, rather than 1111. To implement such an encoder, Besides encoder circuitry, diode gates can be used ex-
shown in Fig. 2(b), several NAND gates are used to dis- tensively in combinational logic. For example, many
connect all keyboard switches from their common circuits in digital systems use NAND gates followed by
ground and to ground output lines 4 and 8. This corre- inverters. In most instances, this common configuration
sponds to the complement of 0011. Since the NAND can be replaced by a simple diode AND gate, at cost sav-
gates are open -collector devices, the matrix resistors ings as high as four to one. The BCD decoder/counter of
serve as pull-ups for the NAND gates, while protecting Fig. 3(a) provides a typical circuit for diode gate substi-
their outputs from being grounded. tution. In Fig. 3(b), two diodes and a resistor replace the
Another encoder variation should be implemented if NAND gate and inverter.

80
Approximating true log output
at high frequencies
New way to build high -frequency log amplifiers eliminates interstage
phase shift by using twin -gain amplifier blocks that consist of a
unity -gain non -limiting amplifier and a high -gain limiting amplifier

by Douglas Clifford, Hewlett-Packard Co., Loveland. Colo

0 Many circuit analysis applications, such as those in- The second technique, which can achieve equal accu-
volving tuned receivers and spectrum analyzers, require racy with less compensating circuitry, is also based on
displaying signals with a very broad dynamic range. an exponential characteristic-that of transistor collector
Signal levels can typically vary from -120 to + 10 deci- current. A single, simple transistor amplifier can usually
bels referred to 1 milliwatt. But to display even a por- provide exponential gain over a range of more than 20
tion of such a signal requires some form of compression. dB. For larger dynamic ranges, transistors must be cas-
Usually the signal is processed by forming a logarithmic caded with intervening gain blocks to allow each stage
function of signal level. to operate within its range.
A recently developed technique eliminates the objec- The third method, successive limiting, has been the
tions of poor frequency response above 5 megahertz most popular at high frequencies because of its simplic-
that is characteristic of the previously least complicated ity. Successive limiting approximates the logarithmic
method, successive limiting. The new design is similar curve by summing either the rf signal outputs or the de-
to successive limiting, but it eliminates interstage phase tected video signal outputs of a series of linear ampli-
shift by using twin -gain amplifier blocks to approximate fiers that limit at a predetermined level. Figure I illus-
the log curve as a series of straight lines. trates the circuit and its output characteristic.
Poor frequency response is the main limitation of suc-
Comparing approximation methods
cessive limiting. At frequencies above 5 MHz, phase shift
Successive limiting, or linear approximation, is the between stages causes the rf signal components to be
most sophisticated of three traditional nonlinear meth- summed out of phase, and the logarithmic response
ods, which use progressively more complex compensat- deteriorates unless delay networks are inserted.
ing circuitry to improve accuracy. The new twin -gain In applications where the log of the rf signal must be
amplifier technique achieves accuracy with no compen- detected before further processing, detectors can be
sating circuitry. placed at the output of each amplifier. This solves the
The method that requires heaviest compensation em- phase shift problem, but complicates circuitry and in-
ploys feedback with a nonlinear element. Generally, a creases over-all thermal sensitivity because the detectors
transistor's base -emitter junction is used, since its char- do not track each other as temperature changes.
acteristic is exponential. A recently developed technique eliminates both the

RF
INPUT V.

USED WITH
N CASCADED STAGES
SUCCESSIVE DETECTOR
DETECTOR DETECTOR DETECTOR
DETECTION
LOG AMP

INDIVIDUAL
DELAY
0 DELAY 410 DELAY STAGES

OUTPUT
DETECTOR

LOG OUTPUT

1. Successive limiting. Summing the outputs of cascaded limiting amplifiers yields logarithm of input signal. Limited output of each stage is
added to that of previous stage to approximate log curve. Delay networks become necessary when operating above 5 megahertz to compen-
sate for phase shift between stages. Optional detectors allow H signal to be detected before additional processing.

81
 N= 8 7 5 0
M= 0 3 7 8

Vo
8 V, -

6V, N)

4V, -
2 V, - (Vo)n,

--80 -70 -60 -50 -40 -30 -20 -10 0

(a) (b) Vi (am)

2. Double -amplifier log stage. Twin -gain amplifier block (a) can be cascaded to form log amplifier that does not need phase -shift correc
tion. Low -gain (GL) amplifier does not limit input signals, but high-gai n (GH) amplifier does, at limit level VL. For logarithmic output to result
GI, must be unity. Plot (b) shows how eight cascaded stages of twin-gain amplifiers approximate log curve. Final output voltage, (Vo)(m+H),
reflects addition of outputs from both limiting (M) and non -limiting (N) stages for total signal range of -80 to 0 dBm.

need for multiple detectors and rf phase shift problems. and K2 = T - Kllog(VL)
The method is similar to successive limiting, because a Since V,(GH + 1)N is greater than or equal to VL:
number of identical blocks are used with output charac- (Vo)(M+N) = VLM + Vx, for Vx between 0 and VL.
teristics that approximate the log curve with a series of This last equation indicates that Vo is proportional to
straight lines. M, which is in turn related to log(V,), making Vo pro-
As shown in Fig. 2(a), the signal is applied simulta- portional to the log of the input. It should also be noted
neously to two amplifiers that form a twin -gain circuit. that GL must equal unity for the log approximation to
The low -gain (GL) amplifier does not limit an input sig- hold.
nal, but the high -gain (GH) amplifier does, limiting it at A plot of the two parts of (Vo)(m+N) illustrates the
a voltage level of VL. The relationship between output logarithmic relationship between the output and the in-
voltage (Vo) and input voltage (V,) can be written as: put. As an example, the graph of Fig. 2(b) results for
V. = V,(GH + GL) cascaded twin -gain blocks when M + N = 8, GH + 1 =
when V, is less than VL/GH, and: 10 dB, and VL = 0 dB into 50 ohms. Assuming GL = 1
Vo = V,GL + VL and N=0 yields the output of M limiting blocks:
when V, is greater than VL/GH. (Vo)m = VLM
If V, is less than VL/(GH + GL)N, the limit level for N And setting M = 0 and keeping GL = 1 gives the out-
cascaded amplifier blocks, the overall amplifier will be put voltage for N non -limiting blocks:
linear. However, when the magnitude of V, causes the (Vo)N = V,(GH + 1)N
high -gain amplifiers in each block to limit, the system The graph shows the contributions of both (Vo)m and
becomes nonlinear. (VO)N for all combinations of M and N. The two outputs
When all the stages are limiting, the output of one is are simply added together for a total output voltage.
still the input drive to the next. For M stages driven into Both output voltage level and input signal range be-
limiting: come larger with increasing M and decreasing N.
(Vo)m = VL( 1 + GL2 + GL3 + . . + GLm-1) + V,GLm
Combining N stages of non -limiting blocks and M From theory to hardware
stages of limiting blocks gives: A log amplifier with almost any given dynamic range
(Vo)(m+/.) = VL( 1 + GL + GL2 + . + m-1) + and log fidelity can be constructed with twin -gain am-
V,GLm(GH + GL)N plifier blocks. The fewer the stages used, the more gain
If GL is set to unity, then: each stage must provide for a given range, and the
(Vo)(m+N) = VLM + V,(GH + 1 )N greater the deviation from the actual log curve.
The maximum non -limiting gain for N §tages is found The computer printout of Fig. 3 compares an eight -
by setting M = 0 and letting Vo = VL: stage log amplifier to a five -stage one. For both compu-
(GH +1)N = VL/V, tations, the maximum and minimum errors (differ-
or: ences) between the calculated values of (Vo)(m+N) and
N = log(VL/V,)/log (GH +1) the actual log of V, are printed. Also shown are the
Substituting T = M +N, where T is the total number of computed output voltages for inputs from -80 to -10
gain blocks: dBm. Output voltage figures are normalized to reflect a
M = T+ log(NYVL)/log(GH + 1) 1-v output change for a 10 -dB input step.
If GH and VL are the same for each block, this equation The calculation for the eight -stage log amplifier, in
reduces to: Fig. 3(a), indicates a maximum error of 0.03 v and a
M = K2 Kilog(V,) minimum error of 0 v when GH = 6.68 dB and VI, = -1
where K1 = 1 /log(GH +1) dam into 50 ohms. (Here, maximum error represents

82
 (a) GAIN DB,LIMIT 76. 68.- 1

MAX ERROR= 3.12457E-02 MIN ERROR= 0

DB= -80 VO UT= 1.00247
DB= -70 VO UT= 2.00211
DB= -60 VO UT= 3.00176
DB= -50 VO UT= 4.0014
DB= - 40 VO UT= 5.00105
DB= - 30 VO UT= 6.0007
DB= -20 VO UT= 7. 00035
DB= -10 VO UT= 8

(b) GAIN DB,LIMIT 714,-5

VMAX= .699427
MAX ERROR= 8. 13584E- 02 MIN EkkOk= 8.99811E-02
DB= -80 VOUT= .979577
DB= -70 VOU I= 2. 07 618
DB= - 60 VO U'1'= 3.00364
DB= - 50 VO UT= 3.92999
DB= - 40 V0111= 5.01443
UB= -30 VO UT= 5.98255
08= 20 VO U'1 = 6.91574
08= -10 VU UT= 8

3. Computer comparison. Printouts for eight -stage (a) and five -

stage (b) twin -gain log amplifiers list normalized (1 V = 10 dB) out-
put voltage for inputs from -80 to -10 dBm. Maximum and minimum
errors across this range are computed as difference between calcu-
lated (Vo)(m+H) and actual log of V1. Eight -stage amplifier, whose
GH = 6.68 dB and VL = -1 dBm, gives maximum error of 0.03 V
and minimum error of 0 V. Five -stage amplifier, with GH = 14 dB
and VL = -5 dBm, provides errors of 0.08 V max and 0.09 V min.

the largest positive deviation from the desired output,

and minimum error the largest negative.) Error levels 4. Building a log amp. Two differential pairs (a) drive common load
are higher for the five -stage amplifier, as indicated in to realize twin -gain log amplifier. Limit level VL of high -gain (GH) pair
Fig. 3(b), with GH = 14 dB and VL = -5 dBm into 50 is determined by allowable peak -to -peak current through RL. Single
ohms. Maximum error becomes 0.08 v, while minimum transistor (b) can also be used. When diodes are forward -biased,
error is about 0.09 v, bringing the peak -to -peak error to amplifier gain is GH, and limit level is fixed by current flowing in diode
0.17 v or 1.7 dB. limiter. For both circuits, low gain GL is unity.
As can be seen from the circuit equations, the critical
design parameters are limit level VL and high gain GH. pacitances are small to obtain broadband frequency
These quantities are influenced by the desired log range performance.
and number of stages. A computer program is generally Another design method, shown in Fig. 4(b), uses the
used to determine them for the dynamic range and log same amplifier to provide both GL and GH. Diodes Di
fidelity desired. and D2 are forward -biased for small signal levels so
One way to implement a twin -gain log amplifier is to that:
use two differential transistor pairs driven in parallel GH = RIARB 2rb) = RL/(RB+2rb)
and feeding a common load, as illustrated in Fig. 4(a). where R1 = RL, and rb is diode resistance. Low gain GL
Transistors Qi and Q2 form the unity -gain amplifier, is obtained through the emitter resistance path:
while transistors Q3 and Q4 provide a high gain of GH GL = RL/RI = 1
and limit at VL. This design, it should be noted, is com- The limit level for the high -gain portion of the ampli-
pletely analogous to the theoretical model. Differential fier is determined by how much current is flowing in the
pairs are used because of the excellent limiting they diode limiter:
provide when they are overdriven. VL = 11R2 + 12R3
The gain of the low -gain amplifier is: where VL is the peak -to -peak voltage of the limit level.
GL = RL/(2R1 + 2re,) = RL/(2R2 + 2re) = 1 When the signal begins to reverse -bias the diodes, turn-
where R1 = R2, and re° is transistor dynamic emitter re- ing them off, the effective transistor emitter resistance
sistance. High gain GH is determined by: increases, and gain drops. The final amplifier gain ap-
GH = RO2R3 +20 = RL/(2R4+ 2re) proaches unity.
where R3 = RI. Limit level VL is fixed by the peak -to - Test results for this log amplifier configuration verify
peak current swing allowed in RL due to the high -gain its predicted performance. An eight -stage amplifier, for
amplifier: example, with a nominal operating frequency of 20
VL = IRL wiz, provides an 80 -dB input range for signal levels
Usually, GH and VL are found first from a computer from -80 to 0 dBm. Moreover, its maximum output de-
model of the circuit. Actual resistor values can then be viation from a true log curve is only 0.3 dB. For this
determined from the equations. Amplifier bandwidth is tight peak -to -peak ripple performance, the emitter re-
established by components RL, L and C. Transistors sistors must be adjusted to assure that GL = 1 when a
should be chosen so that amplifier input and output ca - stage is limited.

83
 Doubling op amp
summing power
Design equations make calculations for non -inverting amplifier input
as simple as those for inverting input so that amplifier efficiency
is maximized when several signals are summed at the same time

by Raymond G. Kostanty, Bendix Corp., Navigation and Control division, Los Angeles, Calif.

111 Summation of multiple signals is a useful circuit terminal. For the inverting amplifier input:
technique whenever more than one signal must be am- E./En = KI1, E0/Ei2 = KI2, and E./EIN = KIN
plified or attenuated with a single amplifier-for ex- where K1 represents the desired gain of the signal ap-
ample, in instruments or avionics equipment. And the plied to the inverting input, and subscripts 1 through N
versatile operational amplifier makes it possible to sum designate signal number. For the non -inverting input:
simultaneously on both its inverting and non -inverting Eo/EN1 = KN1, Eo/EN2 = KN2, and E./ENN = KNN
inputs. where KN is the desired signal gain through the non -in-
Unfortunately, designers frequently utilize only the verting input. Once a convenient value for feedback re-
inverting input because calculations for the non -invert- sistor RF is selected, the input resistors can be found:
ing input are usually tedious. A common practice is to R11 = R12 = RF/KI2, and RIN = RF/KIN (1)
use the inverting input of another amplifier, rather than RN1 = RF/KN1, RN2 = RF/KN2, and RNN =
going through the mathematics for the non -inverting in- RF/KNN (2)
put of a single stage. The values of resistors Rs and RA depend on the rela-
A new set of equations simplifies summation circuit tionship of the sum of the inverting gains to the sum of
design by making the resistors for one input as easy to the non -inverting gains.
calculate as the resistors for the other. Thus, a single op Let: SKIN = K11 + K12 + + KIN
amp may perform as well as two that use only the in- and: 1KNN = KN1 KN2 KNN
verting input. If: 1+ SKIN is greater than or equal to MKNN (3)
Relationships between signal gain, input and feed- then: Rs = 00
back resistors, and input and output voltages are un-
complicated for the inverting input, and each can be
handled separately. But previously, the non -inverting Rs
input required the solution of simultaneous equations- 1-"AN-0---"VA RF

one equation for each applied input. Although these RIN

KIN
equations were only first -order functions, their solution EIN 1/V \
could become cumbersome for three or more inputs.
The new equations use up to two grounded resistors- Kit
Ri2
V\/.,
one for each side of the amplifier-to control available E12
R
amplifier gain. This technique permits all signal gains E,,
Kip
v\A,
and all input resistors to be calculated without simulta-
neous equations and with only simple arithmetic. KN, RN1
EN1 A/V
Circuit functions are elementary
KN2
EN2
A few basic assumptions are necessary to minimize
equation complexity. The amplifier, for example, is as-
sumed to have infinite gain, infinite input resistance, ENN
and zero output resistance. Moreover, each of the am-
plifier's inputs is considered to be directly coupled from
a low -impedance source resistance.
The equations set the dc resistance from the ampli-
fier's inverting input to ground equal to the dc resist- 1. Double summation. Finding resistor values when signals are
ance from its non -inverting input to ground. This min- summed at both inverting (I) and non -inverting (N) inputs of op amp
imizes output null shift with changing temperature. can be.reduced to elementary calculations. Up to N inputs to either
Output/input relationships for the fundamental am- terminal can be accommodated. Simple equations relate input and
plifier configuration of Fig. 1 are easily defined. Let E. output voltages to signal gain. Input resistor values depend on RF
be output voltage, E1 an input voltage at the inverting and signal gain: RIN = RF/KIN, and RNN = RF/KNN. Resistors Rs
terminal, and EN an input voltage at the non -inverting and RA are determined after comparing gain summations.

84
and: RA = RF/(1 4 -SKIN EKNN) (4) Figure 2(b) shows this completed design.
If: 1 + SKIN is less than EKNN (5) If a non -inverting, rather than inverting, input is
then: Rs = RF/(K' - SKIN) (6) added to the first amplifier example, resistor values,
and: RA = RF/(1+K' - EKNN) (7) when RF is 50 kilohms, are unchanged for R11, R12, RNI,
where: K' = RF/Rs + SKIN (8) and RN2. For the new input, KN3 = 3:
If Eq. 5 rather than Eq. 3 is satisfied, a minimum value RN3 = 50k/3 = 16.7 kilohms
is arbitrarily assigned to K', which represents the sum of Equation 3 is no longer satisfied, since SKIN = 3 and
the inverting gains: EKNN = 5.5, but Eq. 5 is:
K' must be greater than or equal to EKNN - (9) 1 + 3 is less than 5.5
Now, the minimum value of K' is found from Eq. 9:
Utilizing the results K' is greater than or equal to 5.5 - 1 = 4.5
Using this value for K', Rs and RA can be computed
Using these equations is not difficult and can be illus-
trated by a few examples. Suppose that an amplifier with Eqs. 6 and 7, respectively:
must be designed for K11 = 1, K12 = 2, KN1 = 0.5, and Rs = 33.3 kilohms
KN2 = 2; all other inputs are zero. First, a convenient RA = 00
value must be chosen for RF. (It should be noted that The finished circuit is shown in Fig 2(c). If some of
the overall resistance level of the final circuit is pro- the non -inverting gains are expected to increase or some
portional to RF, and that amplifier offset current limits of the inverting gains to decrease, a larger value for K',
the maximum value of RF.) Selecting an arbitrary RF of such as 8, would be more appropriate. Resistors Rs and
50 kilohms allows values of the input resistors to be RA then become Rs = 10 kilohms and RA =
found with Eqs. 1 and 2: 14.3 kilohms, as indicated in Fig. 2(d).
R11 = 50k/1 = 50 kilohms Modified designs
R12 = 50k/2 = 25 kilohms
RN1 = 50k/0.5 = 100 kilohms Given an existing circuit, it is possible to modify it
RN2 = 50k/2 = 25 kilohms with the design equations. Suppose non -inverting gain
Since SKIN = 3 and EKNN = 2.5, Eq. 3 is satisfied: must be increased by 20% for the amplifier of Fig. 3(a),
+ 3 is greater than or equal to 2.5
1 in which RF = 50 kilohms, R11 = 25 kilohms, R12 = 50
so that: kilohms, RNI = 10 kilohms, Rs = 16.7 kilohms, and
Rs = 00 RA = 25 kilohms. Existing signal gains must be deter-
and from Eq. 4: mined first, using Eqs. 1 and 2:
RA = 33.3 kilohms K11 = 50k/25k = 2
The final design is illustrated in Fig. 2(a). If signal con- K12 = 50k/50k = 1
ditions remain the same, but another input, K13 = 3, is KNI = 50k/ 10k = 5
added, input resistor values do not change, and: Next, find K' for SKIN = 3. From Eq. 8:
R13 = 50k/3 = 16.7 kilohms K' = 50k/16.7k + 3 = 6
Because Eq. 3 is still satisfied, Rs = 0q but: For the modified circuit, KNI must be 20% higher, or:
RA = 11.1 kilohms KNI = 5(1.2) = 6

Rs
2 V/V 33.3 kS2

1 V/V
":" 2 V/V
0.5 V/V 1 V/V
2 V/V 0.5 V/V
2 V/V
3 V/V

16.7 kt2 RF
Rs
3 V/V 10 kS2 RF

2 V/V
'VVV-- 50 kt2
25 kft kE2
2 V/V
1 V/V 50 kit.
1 V/V
100 kit E0
0.5 V/V 0.5 V/V
2 V/V 2 V/V
3 V/V
Id)
A 14.3 kS2

2. Amplifier examples. When another inverting input is added to amplifier (a), as in (b), values of existing input resistors are not affected,
but RA changes value If non -inverting input is added instead, as in (c), Rs becomes necessary and RA is eliminated. Amplifier (d) is final de-
sign for modified version of (c) that can handle increased non -inverting gains or decreased inverting gains.

85
 EDCIN

EDCI2

EDCI1

EACIN

EACI2

EACI1

EACN1

EACN2
EACNN

EDCN1

EDCN2

E DCNN

4. Generalized amplifier. For complete design analysis, both ac

(AC) and dc (DC) elements and signals must be taken into account.
3. Making changes. New design equations are also useful for mod Dc relationships are simply those developed for amplifier of Fig. 1.
ifying existing circuits. When non -inverting gain of amplifier (a) is in- And ac solutions for signal gain and input resistors are identical, but
creased by 20%, from 5 to 6, resistors to inverting input are unaf- involve ac quantities. Since computing RsAc and Rpm! requires both
fected. As shown in (b), only RN1 and RA must be changed. ac and dc parameters, dc solution should always be found fir$t.

Checking the minimum value for K' with Eq. 9: E./EACIN = KACIN and Ea/EACNN = KACNN
K' = 6 is greater than or equal to 6 - 1 And input resistor computations are also equivalent:
Since the inequality holds, K' = 6 is satisfactory. RACIN = RF/KACIN and RACNN = RF/KACNN
The value of Rs does not change because K' and the in- The values for resistors RSAC and RAAC, however, de-
verting signal gains remain the same, but: pend on both ac and dc signal conditions.
RN1 = 50k/6 = 8.33 kilohms If: + EKDCIN + EKAciN is greater than or
1

and from Eq. 7: equal to EKDCNN + EICACNN

RA = 50k/(1+ 6 - 6) = 50 kilohms then: RsAc = 00
The modified amplifier circuit is drawn in Fig. 3(b). and: (RADcRAAc)/(11.Apc + R AAc) = RF/
Instead of just a 20% increase, suppose KNI is doubled (1 + EKDCIN +EKACIN -
in value to 10. Since the minimum K' now changes: 21(DcNN - EKACNN)
K' is greater than or equal to 10 - 1 = 9 If: 1 + EKDCIN + XKACIN is less than
Select K' = 10, then: EKDcNN + EKACNN
RN1 = 50k/10 = 5 kilohms then: (RSACRSDC)/(RSAC + R spc) = RF/
Rs = 50k/(10 - 3) = 7.13 kilohms (K'Ac - /KDCIN IKACIN)
RA = 50k/(11 - 10) = 50 kilohms and: (RAACRADC)/(RAAC +RADC) = RF/
All the design equations developed assume that the 11 K'AC - IEKDC NN+EKAcNNA
impedance of a source resistance is low and can effec- where: K'AC = RF/RSDC EKDCIN +
tively be neglected. However, this is not always true. If RF/RsAc EICACIN
source resistances are some constant known values, the And the minimum value for K'Ac becomes:
equations are easily modified. Simply subtract a given K'Ac is greater than or equal to
source resistance from the appropriate calculated input EKDcNN +EKACNN - I
resistor. For example, if RF is 50 kilohms, K11 is 2, and When using these equations to analyze a circuit, the dc
the source resistance driving E11 is 2 kilohms, then: properties must be calculated before tackling any ac
R11 = 25k - 2k = 23 kilohms problems. Of course, the dc input resistors and RSDC
Ac considerations and RADC are computed with Eqs. 1 through 9.
A few practical circuit considerations should also be
Amplifier inputs can also be capacitively coupled. remembered. Signal frequencies applied to capacitive
The generalized amplifier configuration of Fig. 4 shows inputs should be at least 100 times greater than the
both ac and dc circuit elements and parameters. (Of maximum frequency on the direct inputs so that capaci-
course, subscript AC denotes an ac element or signal, tors C1 and C2 can properly isolate frequency bands.
while subscript DC is for a dc element or signal.) And time constants CiRsAc and C2RAAC, as well as the
Although the equations used thus far involve only dc time constant formed by each input capacitor with its
quantities, ac signal analysis parallels the dc. Ac out- input resistor, should be about 0.l/6.28fac seconds,
put/input relationships, for instance, are the same as where fa,. is the lowest frequency on any of the capaci-
those for a dc signal: tive inputs. El

86
One-shot timing performance:
don't take it for granted
Study of industry standard monostable multivibrator-the 9601 -
type of one-shot-reveals that vendor -to -vendor variations in
timing accuracy are wide enough to prevent device interchangeability

by David E. Green, Honeywell Information Systems Inc. Billerica, Mass.

All too often, designers presume that a component's Timing errors become especially significant for small
data sheet is precise. Inaccurate performance specifica- values of timing capacitance. For instance, the one -
tions are not intentional misrepresentations on the part shot's minimum true output pulse is specified to have a
of the manufacturer. Instead, they result from problems maximum value of 65 nanoseconds, but approximate
inherent in processing techniques, device geometries, or extremes of 30 and 92 ns were observed.
packaging effects. But, no matter what its cause, exces- Examining the problem
sive performance variations-particularly those of so-
called building blocks that can be used in a number of Pulse width variation for the monostable is specified
applications and that are readily available off -the -shelf at a minimum of 3.08 microseconds and a maximum of
from several manufacturers-must be brought to the de- 3.76 ps about a nominal output of 3.42 µs. Or, in -specifi-
signer's attention. cation variations of timing accuracy can be thought of
Consider, for example, the industry standard re- as ± 10%. Therefore, after accounting for timing resistor
triggerable monostable multivibrator, typified by the and timing capacitor tolerances, as well as supply volt-
9601. A recent study of a dozen sample lots of these age variations, the designer must deal with an overall
one -shots from several different companies shows that timing limit tolerance of ± 15% to ±20%, making any ad-
the variations in output pulse width are so severe- ditional one-shot timing inaccuracies difficult to handle.
sometimes approaching 60%-that some manufacturers Adding a trimmer potentiometer is the simplest way
are re-evaluating their current designs, whereas others to compensate for this wide tolerance spread, or even
are already in the process of redesign. poor one-shot timing-but an acceptable trimmer is

vcc vcc

4 ki2
1N4524
cx

c, 1 02 1 5 pF 2N709
10 )FT, .110 pF
1.5 V
15 pF 24 13 pF 50 ns MAX
13
8 1N3064
2 0 1.5 V
INPUT 9601
vcc
6
OUTPUT PULSE -01
4 IcS2
1N4524 WIDTH

35 V
- 5 pF 2N709
--- V

14-OUTPUT PULSE -01

15 pF Z 241,1.2 13 pF >1N3064 WIDTH

1. Testing the one-shot. Test circuit for retriggerable monostable multivibrator employs unit TTL load on each output.Timing network formed
by Rx and Cx determines output pulse width at Q and Q. Capacitors C1 and C2 are fixed at 10 picofarads (including stray capacitance and
fixture capacitance), and minimum Cx is held to 5 pF. Output pulse width is measured at device's low-level logic threshold.

87
 A
/1
11
/ 1
/
1

SAMPLE ft
I
LOT eft
I Se.
1
I
1

I
1
2 I

3 \

44Zes.).7,...1ftr.
7

-E

Ja"

- 1C

- 15

ns 378 ns 29 Ais .20 las 3 vs 23.97

MEAN PULSE WIDTH

Rx 4k 50 kS2 4 k1,1 34 k!...! 50 kSZ 34 1(2 50 ksz

C. 5 pF 5 pF 1,600 pF 500 pF 500 pF 1,600 pF 1,600 pF

2.PlodIng the results.Pulse width variations of over 500 9601 -type one -shots are shown as percentage deviation frommean pulse width (77)
for entire group. Test points for 12 sample lots are plotted individually. Values noted for Rx and Cx are those used to obtain indicated mean
pulse widths from 68 ns to 24 ps. Scattered distribution of results indicates significant discrepancy between actual and specified timing per-
formance Also, output pulse variations from one vendor to next are so great that parts cannot be used interchangably.

88
generally more expensive than the multivibrator itself.
And for most printed circuit boards, no provision is Published timing data
made for adding a trimmer at some later date if a prob- The expected output pulse width of the widely used
lem arises. 9601 type of monostable multivibrator is easily deter-
Specified one-shot timing accuracy implies that the mined from its data sheet. Generally, how the unit's
part holds pulse -width variations to -±-1 0% for outputs timing is computed depends on the value of timing ca-.
above and below the nominal 3.4 -us value. However, pacitance Cx. When Cx is greater than 1,000 pico-
extensive testing of over 500 of these multivibrators farads, output pulse width can be computed:
shows that this is not true. The study notes the variation T = 0.32RxCx(1 + 0.7/Rx)
in output pulse width for 12 sample lots of 9601 -type where T is in nanoseconds, timing resistance Rx in
kilohms, and Cx in picofarads. For Cx values of less
one -shots from a number of different vendors. than 1,000 pF, manufacturer's timing curves, like
The test circuit used is illustrated in Fig. 1. Stray ca- those shown, must be used.
pacitance and capacitance from the test fixture are rep-
resented by capacitors C1 and C2. The total value for
Sach of these capacitors is fixed at 10 picofarads to per-
mit the use of a variety of test fixtures. Timing capaci- lo'
tance Cx is held to a minimum value of 5 pF since the Rx = 50 kit
minimum data sheet value of 0 pF is not really achie- Rx = 30 ki-Z
vable in practice. Rx = 20 k S."2
Rx = 10 k2
Both Q and Q outputs are terminated with a unit Rx = 5 kg2
transistor -transistor logic load and a 15-pF load capaci- 1103
tor that includes probe capacitance. (In this case, a unit H

load is the equivalent of a standard TTL gate that re- A

quires the one-shot to supply 60 microamperes for its
high output and to sink 1.4 milliamperes for its low out- a_

put.) H
All pulse widths for the study were measured at H
102
3
the low-level logic threshold of the Q output.
The graph of Fig. 2 summarizes test results by scom-
paring the mean pulse width for each sample lot (x) to
the mean pulse width for all the units tested 01). Final
values of 31 are indicated along the horizontal axis, 10 1 1 I 1

10
I

102 10
showing the mean pulse width obtained for the timing TIMING CAPACITANCE, Cx (pE)
components used. For every combination of timing
component values, there is a set of 12 data points, each
point representing the percentage deviation of mean
pulse width z for a given vendor from the group mean A singular advantage to testing pulse width is the
width of TE, which ranges from 68 ns to 24 its. ability to select those vendors whose products fall
within the desired tolerance of ±-10% over the full tim-
Analyzing test results ing range. One-shot timing should be monitored for all
Several observations can be made from the graph. pulse widths of interest by using permissible combina-
For instance, the data point at 2.29 us, near the stan- tions of Rx, between 5 and 50 kilohms, and Cx, up to
dard test point of 3.4 us, suggests that the number of 1,000 pF.
units meeting specification is not very high since devia- As this study decidedly points out, in -spec perform-
tions from z are significant. Also the ±10% tolerance ance at one pulse width does not imply across-the-board
band is drastically exceeded for several pulse widths, in- in -spec performance. One vendor's products, for in-
dicating a serious discrepancy from predicted perform- stance, exhibited unorthodox characteristics even
ance. though the one -shots passed the standard 3.4 -us ± 10%
For some sample lots, output pulse width diverges pulse width test for Rx = 10 kilohms and Cx = 1,000
from its specified value as pulse width increases, even pF. Some devices displayed a variation of ±40% around
though published timing data states that the output a 68-ns mean pulse width, while others could not gener-
width should be linearly related to timing component ate pulses wider than 3.5 us under any conditions.
values when Cx is greater than 1,000 pF. Once the vendors whose products track together are
In addition, for narrow output pulses, performance selected, additional testing should still be done for abso-
tracking from vendor to vendor is particularly poor, lute certainty of timing performance. To virtually elimi-
with a highly undesirable broad distribution of data nate any excessively inaccurate one -shots, the tests
points. A user changing vendors would encounter sig- should fix the allowable ± 10% timing tolerance as the 3c
nificant timing performance differences, especially since limit of a Gaussian distribution of test results. Based on
the plot deals only with deviation from a mean, and the graph of Fig. 2, the recommended pulse widths and
does not show worst -case errors. For example, when timing networks are: 85 ns ±10%, 5 kilohms and 5 pF;
Rx = 50 kilohms and Cx = 5 pF, supposedly identical 380 ns ±-10%, 50 kilohms and 5 pF; 3.4 us ± 10%, 10
devices from vendor 6 and vendor 12 can generate kilohms and 1,000 pF; and 17 us ± 10%, 50 kilohms and
pulses ranging from 237 to 732 ns in width. 1,000 pF. 0

89
Active resonators save steps
in designing active filters
Resonator model allows active filter sections to be treated as
common components, side-stepping design details that bog down analysis;
the model employs a simulated inductance that remains inherently lossless
by Randy Brandt, Integrated Electronics Inc., Los Gatos, Calif.

Active filters make possible the realization of inex- it = (Vo)Lp/Rf = (Vo)Bp/sR'RfC'

pensive high -Q networks at low frequencies. Passive fil- Solving for the equivalent impedance of the loop with
ters, on the other hand, tend to become costly, lossy, these two equations gives:
and unwieldy in physical size when large inductances ZL = (Vo)Bp/if =
are required. Since active filters simulate inductance, showing ZL to be inductive in nature. Therefore, ampli-
extremely large values can be realized while holding fiers A2 and A3, and feedback resistor 111 can be re-
down filter cost, losses, and size. Furthermore, the ever- plAced with the equivalent inductance:
decreasing price of today's monolithic operational am- L = R'RfC' (I)
plifier, the heart of any active filter, is constantly im- connected in a negative feedback loop around amplifer
proving the attractiveness of building an active, rather Al. Now the general form of the active resonator can be
than passive, filter. drawn as depicted in Fig. 1(b).
However, because of the proliferation of active filter Identifying equivalent inductance L is the key to de-
analysis techniques, only a core of full-time design spe- riving the active resonator model. Once L is defined in
cialists can differentiate between the merits of various terms of existing components in an active filter section,
approaches and select the optimum solution. The occa-
sional filter designer finds it easier to build a passive cir-
cuit, despite the advantages of the active approach.
But a basic building block, called the active resona-
tor, allows even the occasional filter designer to handle
an active filter as an ordinary circuit component. This
active resonator model is common to all active filter sec-
tions. It consists of a tuned RLC circuit for resonance
and an op amp for the necessary gain and isolation.
Modeling the active resonator
An active filter section, regardless of how many am-
plifiers it contains, is the fundamental repetitive portion
of a complete active filter. Generally, the active filter
section can perform as a filter itself or be cascaded to re-
alize higher -order filter functions. The term "active
resonator" is simply another label for active filter sec-
tion. With varying degrees of difficulty, any active filter
can be reduced to active resonator form.
Because of its utility and simplicity, a sound choice
for demonstrating how to form an active resonator
model is the popular biquad active filter. This filter is
particularly noted for its insensitivity to variations in
component values. When it has three amplifiers, as
shown in Fig. 1(a), the biquad filter provides both low-
pass and bandpass outputs. By adding a fourth sum-
ming amplifier, the circuit can supply a high-pass, all -
pass, or notch output.' 1. Resonator model. Biquad filter section (a) can oe reduced to ac-
The network's transfer function between its band- tive resonator model (b) by replacing amplifiers A2 and A3 with
pass output, (V.)Bp, and its low-pass output, (VOLp, is equivalent inductance L. This simulates L with non -inverting integra-
that of a non -inverting integrator: tor in negative feedback loop, but other techniques can be used.
(Vo)Lp/(Vo)Bp = 1 /sR'C' Very high but lossless inductances can be realized. Single -amplifier
where s is the Laplace variable. Feedback current is: resonator can be treated as complete filter section.

90
the active resonator model can be used to represent that
particular active filter section. IW
Since L is known for the three -amplifier biquad net-
work, it can be handled as a single -amplifier circuit. Om
This reduces analysis complexity and allows the de- splane
signer to concentrate on his over-all filter requirements
without being unduly concerned with the details of each
filter section.
For the biquad filter, inductance L may be simulated
in several ways by using different non -inverting integra- a
tors in a negative feedback loop around an operational
amplifier.2 Of course, any other technique for simu-
lating a stable inductance is also suitable for the biquad
network or whatever filter section is being modeled as
an active resonator. la)

Because L is simulated, its inductance can be very

large and yet remain absolutely lossless if ideal ampli-
fiers are used. Although ideal operational amplifiers are RESONANCE
fictitious, practical op amps can approach the ideal so 40
that inductance L stays virtually pure (lossless). This is I
Q = 0 - 100
true whenever resistance into the summing junction is SE, 20 Q = 0 - 10
SLOPE = 6dB/OCTAVE 0= 0 - 1
an order of magnitude less than the amplifier open-loop
input impedance, and when the gain of the feedback CD

loop is high enough to prevent open -loop rolloff from

-20-
affecting desired Q.
100 fo
Therefore, assuming that loop gain and amplifier in- 0-01
FREQUENCY (Hz)
put impedance are sufficiently large, the only lossy ele- (b 1

ment in the network is the parallel resistor, RQ. It can be

regarded as the Q -setting resistor, even though the 2. Resonator response. Characteristic behavior of active resonator
product of RQC really determines the 3 -decibel band- is identical to that of tuned RLC circuit. Pole -zero plot (a) illustrates
width, and both the RQC and the LC products establish bandpass nature of resonator transfer function-there are two com-
the Q of the active resonator. plex -conjugate poles and two zeros. Magnitude of pole vector is
fixed by radian center frequency wo, while vector direction is function
Matching the transfer function of 0. Slope of resonator magnitude response (b) depends on Q near
The transfer function of the resonator's tuned circuit resonance and then becomes a constant value.
must be identical to that of its passive counterpart. For
the entire resonator of Fig. 1(b): zw coefficient of s in the denominator. At w = coo, re-
Vo/V, = s/R,C(s2 + s/RQC + 1 /LC) sponse magnitude becomes:
Analyzing this function in the complex s -plane yields A(coo) = h/Aco
the graph of Fig. 2(a). The location of the complex -con- For the active resonator, the response magnitude at coo
jugate pair of poles is determined by the roots of the is found by substituting h = 1 /RiC and Au) = 1 /RQC in
denominator, while the positions of two transmission this last equation. Then:
A(w.) = RQ/R, (4)
zeros (roots of the numerator) are fixed at frequencies of
zero and infinity. Since the impedance, at resonance, of a parallel
This means that the logarithmic magnitude response tuned LC network with lossless inductors is infinite, the
has a bandpass characteristic with a shape as illustrated closed -loop gain of the active resonator at resonance is
in Fig. 2(b). The slope of each curve in the vicinity of simply equal to a resistance ratio. The active resonator,
resonant frequency fc, is a function of Q. The slope away then, has all the properties of a parallel RLC network
from resonance is eventually 6 dB per octave because of and can be described in familiar terms once inductance
the zeros that are located at zero and infinity. L is identified.
The transfer function of a passive parallel RLC net- Designing with the resonator
work can be written as:
G(s) = hs/(s2 + /as + co02) The placement of poles and zeros in the complex -
where h is the attenuation factor of the network, and frequency plane fully defines the shape of a filter's re-
Ow = wo/Q. Comparing the coefficients of G(s) to the sponse. Furthermore, the locations of these poles and
coefficients of Vo/V, for the active resonator yields: zeros are determined strictly by the Q and coo of the
coo = 1 /(LC)1/2 (2) transfer function. And since Q and coo are known for the
and: active resonator, complete active filter networks can
Aw = 1 /RQC or Q = cooRgC (3) now be designed.
and: In general, filter design with either active or passive
h = 1 /R,C sections is a matter of obtaining the best curve fit over a
The magnitude response of G(s) reaches a maximum band of frequencies for a given set of specifications. The
at w = coo, and the 3 -dB bandwidth is determined by the designer usually begins by searching through tables or

91
 Bandpass transformations
The appropriate low-pass approximation' function for character. Further, for each pair of complex -conjugate
transforming to a second -order bandpass function has a poles and each pair of zeros, there is a filter section with
single pair of complex -conjugate poles in the left-hand and f, given by the Q and fo of the transformed pole
s -plane. To transform fourth -order functions, the trans- pair.
formation equations for the second -order functions are The bandpass 0 transformation for a low-pass com-
used twice. plex -conjugate pole pair can be written as:
Transforming each low-pass complex -conjugate pole
(a) into its bandpass equivalent requires six parameters: ry (r2y2+ 8ry - 16y 16) 1/21 1/2
the over-all Q of the bandpass filter at its center fre- 8y
quency (0.), the bandpass center frequency (f0), and the Here, y is defined as:
real and imaginary parts of the two low-pass poles. y = (Re/0.)2
Since the vectors drawn from the origin to the poles in where Re is the real part of the low-pass pole pair, and 0,
(a) are defined in terms of 0 and f0, it is convenient to is the cutoff -frequency Q. And r is defined as:
write the transforms with Q and fo as dependent vari- r = 1 + (1m/Fle)2
ables. As shown in (b), the low-pass complex -conjugate where lm is the imaginary part of the low-pass pole pair.
pole pair transforms into the bandpass plane as two pairs The f, transformation is:
of complex -conjugate poles and two pairs of zeros. 00% = [Opy1/2 (Qp2y 1 )1/2]f0
The bandpass pole Qs are identical, and the center fre- 00% = foi [Oy1/2 (Qp2y 1)1/2]
quencies are geometrically symmetrical about the center where (f0')1 and (f0')2 are the resonant frequencies of the
frequency of the over-all filter. The zero pairs are located bandpass filter sections, and fo is the resonant frequency
at co = 0 and (.,) = 00, thereby establishing the bandpass of the low-pass filter.

BANDPASS

DOUBLE
ZERO

0= secla

( b)

by using a computer program to find the mathematical Hz, a maximum passband ripple (Amax) of 0.1 dB, and
function that most closely approximates his require- geometrical symmetry.
ments. In doing so, he may select any one or a combina- First, the filter type and function order that best fit
tion of such well-known functions as Butterworth, the specifications must be found. To do this, the band-
Chebyshev, or Bessel. width at -60 dB (Mw) is computed from the upper (fu)
Once the function or functions are chosen, the de- and lower (fL) frequencies about geometric mean fn:
signer decides on the order of the filter, based on char- fu = f02/fL = (2,125)2/1,270 = 3,560 HZ
acteristics such as shape, minimum stopband attenua- Then, for this case:
tion, maximum passband ripple, group delay, and Af60 = fu -fL = 3,560 - 1,270 = 2,290 Hz
phase response. The order of a filter function fixes the And the center -frequency Q (Q) of the entire filter is:
number of sections needed for a given response. And Qc = fo/Af3 = 2,125/400 = 5.32
the coefficients of the filter's characteristic equation de- Shape factor On is determined next:
termine how each section is to be tuned for Q and f0. = Af60/Af3 = 2,290/400 = 5.13
As an example, a typical fourth -order bandpass fil- The design step that follows often involves the trans-
ter-the type frequently used in low -speed telephone formation of a normalized filter function into the proper
data communications systems-illustrates designing filtering plane. After transforming and denormalizing
with the active resonator. Suppose the specifications this function, the designer can tune each resonator to a
are: a center frequency (f0) of 2,125 hertz, an over-all particular Q and cutoff frequency. Tables of transfer
passband gain [A(con)] of 200 (46 dB), a 3 -dB bandwidth functions are usually written for low-pass functions so
(M3) of 400 Hz, a minimum stopband attenuation that realizing other filter types requires transforming
(Amin) of 60 dB at a lower cutoff frequency (fL) of 1,270 the low-pass function into the desired function.

92
 Using the computed value of 2s, and specified values
of Amin and Amax, and published filter data3 shows that Off -the -shelf resonators
a fourth -order Chebyshev filter will satisfy the ex-
ample's skirt, bandwidth, and ripple requirements. The Packaged active resonators, available as standard
normalized coefficients of the low-pass Chebyshev func- product lines from more than a dozen companies, can
tion are found in a table of. low-pass Chebyshev poly- almost reduce active filter design to a matter of re-
nominals.4 For a ripple of 0.1 dB and an order of four, sistor selection. Some manufacturers refer to their
lines of active resonators as universal active filters, but
the filter's characteristic equation is given as: the circuit function is the same.
D(s) = (s2 +0.458s+ 1.153)(s2 + 1.616s + 0.789) The larger suppliers include Beckman Instruments
Solving for the roots of D(s) yields: Inc., Fullerton, Calif.; Datel Systems Inc., Canton,
SA1,A2 = -0.229 ±1.05j, Sm.,B2 = -0.808 ±0.369j Mass.; Kinetic Technology Inc., Santa Clara, Calif.;
These low-pass roots contain two complex - Optical Electronics Inc., Tucson, Ariz.; and TRW Semi-
conjugate pole pairs, which transform (see panel, conductor division, Lawndale, Calif.
"Bandpass transformations") into bandpass equivalents A standard family of hybrid microcircuit active
of four complex -conjugate pole pairs and eight zeros. resonators also is offered by Integrated Electronics
Since there are four pairs of poles, four active resona- Inc., principally for use in audio and subaudio appli-
tors are needed to attain the requirements. cations. In large quantities, they range in price from
about $5 to $8.
Finding resonator components
From the low-pass-to-bandpass transform equations
in the panel, the values of Q and fp can be computed for Figure 3(a) shows the bandpass filter configuration
each resonator in the bandpass filter. Low-pass roots sAi using the active resonator representation. It should be
and sA2 characterize the first two resonators: remembered that the resonator model employs an
Qi = 23.3, (f0)1 = 2,350 Hz equivalent inductance requiring practical simulation
Q2 = 23.3, (f0)2 = 1,925 Hz and that each resonator section actually consists of three
while sBi and sB2 characterize the last two resonators: amplifiers and associated circuitry, as noted in Fig. 3(b)
= 6.6, (f0)3 = 2,135 Hz for the input resonator. Once the components for the
Q4 = 6.6, (f0)4 = 2,120 Hz resonators are known, those required to complete each

0 4.58 H L
0 5.56 H L
1,000 pF 5.63HO L

C , 1,000 pF L
8.26HO
Ho
11,000 pF
Ro
1.58 MS -2
C 111,000 pF
Ro
490 kSZ

495 kS2 Ro
158 kil
Ri
250 1.93 MS2
Ri
250 kSt

190 kft
v

RECOMMENDED OP AMPS:
TYPE 741
FAIRCHILD gA776 1 LOW
SOLITRON UC4250 POWER

Ri

bI

3. From model to design. Fourth -order bandpass filter (a) is represented by four cascaded resonators. Each resonator, with its simulated
inductance, actually requires three amplifiers (b), making 12 op amps necessary for entire filter. Component values can be found, once
and fo are known for each resonator. Capacitor values are assigned for computation convenience. R9 principally determines 0 of resonator.

93
 Rf Rf Rf Rf
-1VVV -"OA. -A/VV
45.8 k12 55.6 kit 56.3 kS2 82.6
1+ V
1+"
158 k11 8 12 8 12 12

Vi -AAN-- 3
7
3
7
3 7 3
7

R RESONATOR RESONATOR RESONATOR RESONATOR

R Ri
2
6
2
6
d6 2 6 2

4 4 4 5 4
250 kS2 5
250 kit 190 kS2
-4---

RQ R0 -V RQ -v
Wv
1.58 Mf2 490 k.S2 495 kS2 1.93 MS2

0
RESONATORS:
INTEGRATED
- 10 ELECTRONICS
µAR 1700

- 20

- 30

- 40

- 50

- 60

70
2 3 4 5 FREQUENCY (kHz)

4. Ready-made resonators. Active filter design is considerably simplified by using packaged resonators that are available for a number of
filtering functions. Those in this diagram are biquad networks connected to realize bandpass filter of Fig. 3. Only three components must be
added to each package-R, for gain adjustment, RQ to fix resonator 0, and Rf to set resonant frequency.

filter section can be found. panel, "Off -the -shelf resonators." These packaged cir-
To keep the analysis simple (neglecting the input im- cuits often reduce the final design step to choosing re-
pedance of practical operational amplifiers), a conve- sistor values, once Q and fo are known.
nient value is chosen for capacitor C. Generally, C For instance, the µAR 1700 resonator from Integrated
should vary between 800 and 1,000 picofarads for re- Electronics is a biquad network that can be used for the
sistor values to remain reasonable. By letting C = 1,000 resonator sections in the fourth -order bandpass filter
pF and computing the radian center frequency (co. = design example. Components C, L, RQ, and Ri are com-
27Tfo) for each resonator, L can be found from Eq. 2: puted as before; of these, only RQ are R1 are required
L= 1/coo2C for the actual filter implementation, as indicated in
The value of resistor RQ is determined with Eq. 3: Fig. 4. None of these component values differ from
RQ = Q/w0C those already calculated.
Since there are four resonators and the over-all required Next, the "internal" resonator parts are found. In ad-
gain is 200, then each resonator must provide a gain of dition to setting C' = 1,000 pF, R' is assigned a value of
50. Eq. 4 becomes: 100 kilohms for convenience, and then Rf is computed
A(coo) = RQ/124 = 50 from Eq. 1:
which can be solved for input resistor R1: Rf = L/R'C'
R1 = RQ/50 The resulting values of Rf are noted in the figure. When
After the component values for the four resonators the µAR 1700 resonator is used to build an active filter,
are computed, the other components needed to simulate RQ effectively sets resonator Q, and Rf effectively sets
inductance L can be found. Again, the value of 1,000 pF resonator fo. The magnitude response of the fourth -
is chosen for capacitor C' to simplify the calculations. order bandpass filter is also illustrated.
Feedback resistor Rf is set equal to resistor R':
Rf = R' = R REFERENCES:
so that Eq. 1 can be used to find R: 1. J. Tow, "Step -by -Step Active Filter Design," IEEE Spectrum, December 1969, pp. 6 4-68.
2. L. C. Thomas, "The Biquad: Part I-Some Practical Design Considerations," IEEE Trans-
R = (L/C)" actions on Circuit Theory, Vol. CT -18, No.. 3, May 1971, pp. 350-357.
Active resonators, however, are available as standard 3. A. Zverev, Handbook of Filter Synthesis, John Wiley & Sons, 1961.
4. R. P. Salton, E. L. Key, "A Practical Method of Designing RC Active Filters,"
products from several manufacturers, as noted in the M. I. T. Lincoln Laboratory, Tech. Report No. 50, May 6, 1954.

94
Active filters get
more of the action
Built around the new, better, cheaper IC op amps, active filters
are beginning to move in on passive filters; their advantages include
flat passbands with sharp corners, inherent isolation, and easy tuning

by Lucinda Mattera, Circuit Design Editor

For all practical purposes, an active filter is just an change, while the passive filter's steadily drops, making
RC network tacked on to an operational amplifier-or the performance the deciding factor in this region.
amplifiers-and its price and performance depend es- Over the audio range and below, the active filter has
sentially on those of the op amp. Recently IC op amps a flat constant -amplitude passband and sharp corner
have improved so much on both counts that active fil- frequencies. Moreover, even when operating at fre-
ters have started challenging passive filters at the lower quencies below 10 Hz, it provides reasonably high val-
end of the frequency spectrum. In fact, the very lowest ues of Q, allowing good selectivity to be obtained. In
end is by now all theirs. contrast, the vlf passive filter requires very high induc-
A great many are being built in-house. They are rela- tance values so that the inductors are physically large
tively straightforward to design, because they can be and cannot provide the Q needed for good selectivity.
easily handled in mathematical terms and lend them- Above 20 kHz, the op amp is the limiting factor. Its
selves to computer -aided design. But with the manufac- gain -bandwidth product determines the highest fre-
turers of active filters busily cutting prices and introduc- quencies at which the filter can work. The output power
ing standard products that can be customized by the level of most low-cost general-purpose lc op amps be-
buyer, the make -or -buy decision is becoming more dif- gins to roll off at about 20 kHz, attenuating the filter
ficult than it was a year ago. output voltage. But op amps with gain -bandwidth prod-
Presently, communications applications in telephone ucts large enough to sustain filter operation above 20
equipment and various military programs are making kHz are expensive, costing $5-$10 instead of $1.
heaviest use of active filters. They are also being ex- Comparing active and passive
ploited in some areas of data acquisition, notably tele-
metry, medical research, oceanography and the moni- In general, it's the component parts of the active filter
toring of geophysical phenomena. But much larger and that contribute most to its cost-the IC op amps cost
practically untapped markets exist in process control, nearly a dollar each, whereas resistors and capacitors
data acquisition, computer terminals, and modems. get only a few cents apiece. With the passive filter, on
Other potential markets include pager systems, petro- the other hand, the design time needed to produce de-
chemistry, and instrumentation. sired performance is the expensive element.
From this listing, it's clear that active filters perform This can be inferred from a comparison of active ver-
the same functions as passive filters, providing either sus passive performance:
low-pass, high-pass, bandpass or band -reject (notch) The behavior of practical inductors is very nonlinear
outputs. Their response characteristics are also identi- at low frequencies, especially with respect to tempera-
cal, since they can realize a Butterworth, Chebyshev, ture coefficient. But because even low-cost general-pur-
Bessel, Gaussian or Cauer (elliptic) response. As a re- pose op amps offer dependable temperature perform-
sult, deciding which type of filter will do the better job ance, the active filter can achieve a corner frequency as
boils down to the traditional design tradeoff between tight as 100 parts per million per °C without too much
cost and performance. difficulty.
Two frequency breakpoints -100 hertz and 20 kilo- Obtaining a frequency accuracy within only 5% can
hertz-can be used as soft guidelines for the decision. be troublesome with the passive filter, but is easy with
Active filters s,hould be considered only if they are to the active.
operate at very low frequencies or in the audio region- Getting sharp corner frequencies, which are inherent
say, from 0.001 Hz to 20 kHz. Above 20 kHz, the passive in the active filter, requires many inductors and capaci-
filter costs less than the active and provides comparable tors in the passive -filter. This makes the passive bigger
or better performance. Below 100 Hz, passive perform- than the active for the same skirt selectivity.
ance is quite sloppy, so that the active filter is the more The active filter can supply a controlled gain over its
practical choice almost regardless of cost. From 100 Hz entire operating frequency range. The passive filter al-
to 20 kHz, however, the active filter's cost does not really ways attenuates a signal.

95
 Additionally, the active filter is easily tuned to the de-
sired operating frequency with a resistor, while the pas- Data acquisition: promising
sive must be tuned with an inductor and/or capacitor.
Impedance matching at either the active filter's input Although computer -controlled data acquisition sys-
or its output is unnecessary because its op amp provides tems are popular, only a few contain analog -to -digital
it with inherent isolation in the form of a high input im- converters that use filters to precondition incoming
pedance and a low output impedance. For the passive analog signals for improved data analysis accuracy.
filter, impedance matching can prolong design time The filters that are used, however, are generally ac-
inordinately, and proper termination may even involve tive, since much of the data being collected occurs at
frequencies of 20 hertz or below.
active devices (an emitter -follower or an op amp) at
each end. Only rather specialized data gathering systems-for
example, systems for special automobile testing, off-
Realizing higher -order filter functions is a simple shore drilling operations, and checking bridge con-
matter of cascading active -filter sections, each of which struction-presently employ filters. For the most part,
provides at least a second -order filter function. This is a process control, where a good deal of data acquisition
marked improvement on implementing higher -order is done in the noisy environment of industrial machin-
functions with the passive filter, which requires the ad- ery, remains a holdout against filtering.
dition of inductors and capacitors that make the design Manufacturers selling data acquisition systems with
more complex and interaction problems more acute. prefiltering include: Digital Equipment Corp., Maynard,
Mathematically, in fact, the active -filter section can Mass.; Data General Corp., Southboro, Mass.; and
be treated as a second -order transfer function that is Analogic Corp., Wakefield, Mass. Another company,
Datel Systems Inc., Canton, Mass., will be introducing
synthesized as a pole pair rather than a low-pass or a prefiltered multiplexing system at the Wescon show.
high-pass filter section. This makes the active filter easy
to manipulate on paper or to design with the aid of a
computer without special complicated programs. The least used-the positive -feedback filter (Fig. 1)-is
Other cost considerations also the only one to employ a noninverting op amp.
This filter, which is also known as the voltage -controlled
Admittedly, the passive filter operates without being voltage source, usually requires the fewest resistors and
tied to a power supply, while the active cannot. Op -amp capacitors and is easily tunable over a wide frequency
power supplies are common to so many applications, range. It is the only practical active filter circuit that of-
however, that the addition they make to the price is fers zero phase shift without needing a second amplifier
usually very reasonable. for signal inversion.
And even so, there are situations in which active fil- The other three circuits incorporate negative feed-
ters are more cost-effective than are passive. Several ac- back, using the op amp in its inverting mode. For these,
tive filters, for instance, can fit on one printed -circuit the output signal is out of phase with the input by 180°.
board, while at least one board is usually needed for Common names for the three are multiple -feedback,
each passive filter. The cost of associated hardware like state -variable, and biquad active filters.
pc boards, connectors, and cables could be substantial
enough to tip the balance toward active filters. Negative -feedback filters
Moreover, active filters often help lower production The multiple -feedback active filter (Fig. 2), which is
costs. When housed in small tc-compatible packages the most widely used, consists of a single op amp and an
they can save assembly time, and are adaptable to exist- associated RC network. As a bandpass filter, it offers
ing automatic production machinery, like insertion and constant -gain -and -bandwidth performance that does
wire -wrapping equipment. not change as the filter's center frequency is shifted.
Today's practical active filter is one that uses an ac- Two variations of this filter-the bridged -T and the
tive device to synthesize inductance. There are four ba- twin -T active filters- have only one feedback path. But
sic types. they are less popular than the multiple -feedback ver-
1. Non -inverting. Positive -feedback active fitter (low-pass, in this 2. Most popular. Multiple -feedback filter nas two or more negative-
case) maintains zero phase shift between input and output. This feedback paths. Here a bandpass function is realized. The shining of
least -used filter circuit minimizes component count.
center frequency does not affect filter's gain and bandwidth.

r-vv\-->-* e0

96
 Rf

sion, because they need more passive components and

have higher output impedances. e,
Both the state -variable (Fig. 3) and biquad (Fig. 4)
active filters realize a transfer function through the BANOPASS LOW-PASS,
analog computing technique of integrating and sum- OUTPUT 0 UTPI IT

ming. The former is based on general state -variable

theory, while the latter is a modified state -variable filter 4. Modification. Biquad active filter offers most advantages of state -
that involves quadratic equations. variable circuit, but cannot produce a high-pass function without ad-
Simultaneous low-pass, high-pass, and bandpass out- ditional amplifier. sets gain, RQ sets Q, and Rf sets frequency.
puts are available from the state -variable filter. The bi-
quad filter provides simultaneous low-pass and band- a high-pass function. But the values of the resistors and
pass functions, but requires an additional amplifier to capacitors in a biquad may be so high that their tem-
realize a high-pass output. This choice of output func- perature coefficients become inadequate. In that case,
tion is why these circuits are often referred to as univer- the state -variable filter is to be preferred, since its per-
sal active filters or active resonators [Electronics, "Ac- formance parameters are set by component ratios and it
tive resonators save steps in designing active filters," can make use of low resistor and capacitor values, with
April 24, p.106]. their better temperature coefficients. For example, the
Both these types of filter require two or three op ampt, Q of the state -variable is determined by a resistance ra-
each, in addition to the necessary resistors and capaci- tio, not a single resistor value as with the biquad.
tors, making them rather more expensive than the mul- Deciding to make or buy
tiple -feedback filter. In exchange for this higher cost,
however, they do offer a choice of output functions and As a rule, active -filter users build their own circuits if
simpler design procedures. Also, they are relatively in- they have the necessary design experience, and if they
sensitive to passive -component tolerances. The Q, gain, use filters in large quantities. Active filter suppliers find
and operating frequency of the two can be indepen- their main customers are among those users who lack
dently adjusted with resistors. design know-how, or who have particularly stringent
Better temperature stability can be obtained from the performance requirements.
state -variable and biquad filters than from the multiple - In contrast to the ones built in-house on pc boards,
feedback filter, but op -amp quality must be better; the purchased active filters are always plug-in modules that
multiple -feedback filter is not as sensitive to amplifier are housed in ic-compatible packages. These ready-
performance. On the other hand, since the gain -band- made units are either modular encapsulated assemblies
width product of cascaded amplifiers is smaller than the (Fig. 5) of packaged discrete -type components or preci-
gain -bandwidth product of a single amplifier, the upper sion hybrid circuits (Fig. 6) that contain chip amplifiers,
frequency limit of the state -variable and biquad filters chip and/or discrete capacitors, and thick -film and/oi
tends to be lower than that of the multiple -feedback fil- thin-film resistors. They are usually negative -feedback
ter circuit. filter circuits of the three types discussed above, and can
As for the relative merits of the biquad and the state - range in price from about $26 for a single standard unit
variable filters, the former is usually the better for a to a few hundred dollars for a single special unit.
low-pass or bandpass function, and the latter better for Only about four years ago, the market for active fil-
ters seemed to have a bright future. But many poten-
3. Versatile. State -variable filter requires three op amps to supply tially large markets have remained relatively unpene-
high-pass, low-pass, and bandpass outputs at same time. Operating trated because of widespread in-house design, the firm
frequency, gain, and Q are inaependently adjustable. Also, circuit is establishment of passive filters, and the lack of filtering
not sensitive to passive -component tolerances. in many data gathering applications. Nevertheless, ac-
tive filter sales are expected to be close to $10 million
for 1972, climbing to $19 million by 1975.
While the current market is growing at the rate of
10% to 15% per year, usage of active filters is probably
growing much faster since many more filters seem to be
being built in-house than are being sold by filter manu-
facturers. To attract more sales, some filter suppliers are
hoping to launch the same kind of customer -education
campaign that was extremely successful for the op amp.
That goal should be helped along by the declining
prices of recent and upcoming new filter products.
Because most filters are custom items, their manufac-
turer is primarily in the business of selling engineering

97
 5. Ott -the -shelf. Modular -type filters manufactured by Frequency Devices are assembled from discrete amplifiers, resistors, and capacitors.
Several venoors supply similar filters in both standard and custom models, as either factory -tuned or customer -tuned circuits.

services. Filter makers who do supply standard filters self. If we can't give him what he wants for his price,
attribute only about 30% of their sales to them. he'll build it himself." Burr -Brown expects to expand its
Several companies are offering the universal active standard line with a state -variable filter before the end
filter as a standard product, but its reception has been of this year, and to add a voltage -tunable filter shortly
mixed. Some users say that its principal feature, its vari- afterwards.
ability, is useless since filter needs are usually special- Another manufacturer of modular -type filters is Fre-
ized. The filter's data sheet is said to be difficult to un- quency Devices Inc. of Haverhill, Mass. This small com-
derstand, and tuning it to desired specifications may not pany has an interesting way of keeping parts cost low.
be easy. Tuning can be further complicated when sev- Alan Schutz, director of engineering, explains, "We buy
eral of the devices are cascaded, because their tuned capacitors with tolerances of ±20%, but know their ac-
component values generally differ from resonator to tual values to within ±0.5%. Resistor values are then
resonator. Finally, the universal filter is expensive be- easy to determine with a simple computer program."
cause it contains more than one op amp. A recently introduced family of tunable active filters
However, as one supplier points out, this filter can be
used as a basic building block by the manufacturer to 6. Chips and films. Hybrid active filters tend to be customized cir-
realize a whole line of filters, without harming the fil- cuits of thick- and thin-film resistors, chip capacitors, and chip ampli-
ter's tuning versatility. The manufacturer can choose fiers. Pictured is digitally tunable bandpass filter made by Sprague
the filter function he wants and then encapsulate. Even- Electric for the military. Molded tantalum capacitors (on left and right
tually, therefore, the universal filter will cost the manu- sides) are sometimes used to furnish required capacitance.
facturer less to make-and his customer less to buy.
Filter makers stress lower pricing
One of the oldest suppliers of modular -type active fil-
ters, Burr -Brown Research Corp., Tucson, Ariz., is
keenly aware of the need for a low-cost ready-made
unit. This month, the company is introducing its series
ATF76 filters that are about half as expensive as its ear-
lier units and are intended to be priced about 10% be-
low competing lines. The new filters are fixed -tuned
units that employ lc op amps instead of discrete transis-
tors. Single -unit prices begin at $29 for a two -pole low-
pass Butterworth filter.
Ron Gadway, product marketing engineer, data con-
version products, sums up the filter maker's biggest
problem, "Our competition is really the customer him-

98
 8. Hybrid maker. Active filter suppliers must also service other mar-
kets to survive, asserts Eugene Donovan, Sprague Electric.

11.00011w

7. Module maker. Alan Schutz of Frequency Devices thinks that 9. Instrument maker. Variable active filters can be used as design
suppliers must teach customers how to select and use active filters. aid to find filter needed, observes Ernest Lutfy, Krohn -Hite Corp.

from Frequency Devices frees the customer from the predominantly custom market, Sprague intends to in-
difficult and time-consuming job of capacitor trimming. troduce a second line of standard filters in the next
The capacitors are inside the package, and tuning is ac- month or so. The filters will be tunable, using external
complished with outboarded resistors. Frequency De- resistors to control Q, frequency and gain. Both band-
vices is one of the few companies with a standard series pass and notch functions will be available for fre-
of voltage -tunable active filters. Tuning ranges for these quencies of 30 Hz to 10 kHz. Pricing is not yet firm.
extend down to 0.01 Hz and up to 50 kHz. Single -unit Motorola Semiconductor, Phoenix, Ariz., is aiming
pricing starts at $210. particularly for penetration of the modem area with ac-
Manufacturers that make hybrid active filters do so, tive hybrid filters. But, observes Don Kessner, manager
for the most part, as a byproduct of their other hybrid of systems, industrial applications, "because of cost, the
business. A well-known supplier of hybrid circuits for hybrid active filter will probably never replace the pc-
several years is the Semiconductor division of Sprague board active filter used in those modems where space is
Electric Co., in Worcester, Mass. Eugene Donovan, op- not critical." Only when the "so-called all-mos data
erations manager, hybrid circuits, points out, "Anybody modem" becomes practical does he feel that a high -vol-
who depends on filters totally is going to have a difficult ume special market may develop for the inexpensive
time. At Sprague, we build hybrids and can serve sev- hybrid filter. Intended for low -speed operation, this
eral markets with the same technology, because the modem would comprise two packaged circuits-a digital
components used in our converter products can also be MOs chip and an analog front-end conditioner for the
used in our filters." chip. The analog portion could be a hybrid filter.
In an attempt to break hybrid filters out of the Kinetic Technology of Santa Clara, Calif., which re -

99
covered from bankruptcy officially this last December,
is selling both modular and thick -film hybrid active fil-
ters. The company's standard hybrid filters are state-
variable circuits that provide a variety of response char-
acteristics, from Butterworth to elliptic. The least expen-
sive filter in this family sells for $26 singly, dropping to
less than $5 in quantities over 10,000. It includes an ex-
tra op amp.
Fred Glynn, Western regional sales manager, offers a
bit of customer advice. "If frequency accuracy within
1% is acceptable," he says, "the customer can use inex- 10. Variable active filter. When filter frequency must be contin-
pensive 1% resistors to tune our filters without difficult uously varied, an instrument -type filter can be used. Corner fre-
trimming. If his accuracy requirement is tighter, we sug- quencies and filter function can be selected from front panel.
gest that he use a less expensive filter [of ours] and tune
it with a fixed resistor and a trimmer potentiometer."
For completeness' sake, the instrument type of active
filter (Fig. 10) should also be mentioned. It can provide
a number of filtering functions over a wide frequency
range. The user selects the ftInctions he wants (low-pass,
high-pass, or a combination of these) and then dials the
frequency he wants.
Krohn -Hite Corp., Cambridge, Mass., manufactures
a series of these variable active filters. The company's
instruments can perform low-pass, high-pass, band-pass
and band -reject filtering from 0.001 Hz to 3 MHz. Prices
range from $450 to $2,075 for frequency accuracies of
±2% to -±10%.
"Because our instruments are tunable and can be op-
erated over a wide frequency range, they are sometimes
used as a design aid to help pinpoint the fixed -fre-
quency filter that will be needed," notes Ernest Lutfy,
sales manager. Medical researchers are also using the
variable filter, for example, to test the frequency re-
sponse of the ear, or to listen to the heartbeat of a fetus
while filtering out unwanted signals from its expectant
mother.
Users justify building in-house
11. Tight fit. Intertel Inc. employs narrow -band active filters to elimi-
In-house active filter design is done by companies nate noisy sidebands in receiver portion of pc -board 201 -type
who can spread engineering costs over a number of modem. Complete modem measures only 9.5 by 12.9 inches.
products. "We seldom buy outside because it's just too
expensive," says Farouk Al -Nasser, a principal scientist Inc. in Burlington, Mass., uses active filters exclusively
at the test instrument division of Honeywell Inc., in to keep its modems as small as possible. "We need very
Denver, Colo., who is involved in designing data stor- special filters and prefer designing our own," says
age systems. But he cautions that learning how to de- George Harlem, marketing coordinator. "Nobody else
sign active filters may take as long as two to three years, can build them as cheaply as we can, and nothing that
because existing literature does not cover practical con- we could buy is equivalent in terms of performance."
siderations for the op amp. Most filter theory still treats Many designers actively prefer active filtering to pas-
the op amp as an ideal element having infinite gain and sive. Jerry Holsinger, Intertel president, explains why:
perfect isolation. "In a sense, passive filter design is a black art. Active fil-
Many modem manufacturers have a staff of active fil- ter design is really a totally new concept in which the
ter design specialists. One of these is Paradyne Corp. of general-purpose building block is not a physical compo-
Clearwater, Fla. "The cost of active filters depends on nent, but simply a second -order transfer function." He's
who designs them," says Tam Saliga, senior engineer. referring to the fact that an active filter can be com-
"Using only one op amp, we can build a [multiple -feed- pletely described by its pole -zero plot, which always
back] filter with four to five poles for around $1 to contains a pole pair, permitting the filter to be easily
$1.50." manipulated mathematically.
Paradyne even standardizes its active filter designs to One designer, John DeFalco, who is a lead engineer
some extent to realize additional cost savings. Most of at Honeywell Information Systems in Billerica, Mass.,
the filters the company incorporates in its modems use personalizes his preference for active filters. "Passive fil-
the same two values of capacitance, enabling Paradyne ter design is just a matter of looking up data in tables
to buy capacitors in very large volume. and charts-it's a lot more fun designing an active fil-
Specializing in pc -board modems (Fig. 11), Intertel ter," he says.

100
The integrated Schmitt trigger:
a versatile design component
Availability of the Schmitt trigger in an IC version is expanding
its use in common circuits, such as multivibrators and pulse
stretchers; here are a few design tips on how best to apply it

by John A. DeFalco, Honeywell Information Systems, Billerica, Mass.

As basic digital circuits go, the Schmitt trigger re- trigger. Logically, the integrated Schmitt functions as a
quires careful and sometimes time-consuming design. four -input NAND gate; tying all four of the inputs to-
That fact has deterred many circuit and systems design- gether allows the circuit to be driven by signals with
ers from working with it, despite its versatility. larger amplitudes.
Recently, however, several manufacturers have intro- Analyzing the IC Schmitt
duced an integrated dual Schmitt trigger, the type 7413,
which is completely compatible with the popular type Multiple -emitter transistor Qi operates as a diode (its
7400 TR circuits. collector -base junction is shorted). Transistors Q2 and
To use the type 7413 to its best advantage, the de- Q3 are the actual Schmitt trigger, while transistors Q4
signer should first take an inside look at the chip and through Q9 make up a NAND gate with conventional
see how it operates. That means understanding how to TTL-type output characteristics. Two of these transistors,
derive the values of the major parameters from the cir- Q5 and Q9, function as emitter -base diodes (collector
cuit diagram. open).
The circuit's most important parameters are its posi- When the input (assuming all four inputs are tied to-
tive- and negative -going threshold voltages, VT+ and gether) is low, Q2 is off and Q3 is saturated. To calculate
VT-. VT+ must be exceeded to switch the output from a the positive -going threshold voltage, the circuit of
high to a low state, while at VT_ the output switches Fig. 2a (left) can be used to obtain the Thevenin
from low to high. equivalent (right) with transistor Q3 in saturation. Sum-
These voltages, together with the effects of tempera- ming voltage drops around the input loop,
ture and power supply variations, can be easily calcu- VT+ = -VBE1 Vr2 VE
lated for the chip from the circuit diagram of Fig 1. where VBE1 is the base -emitter voltage of transistor Qi
Each half of the type 7413 Schmitt trigger is actually a and Vy2 is the emitter -base voltage required to turn on
combination of 7400 -type logic and the basic Schmitt transistor Q2. Voltage VE from the emitters of Q2 and

vcc

4 kS2 2 kS2 1.2 kS2 1.5 ks2 1.6 kS2 2 130 12

'/s,' 0,

INPUTS OUTPUT
0,,
7
460 S2

AAAA 1 5 kil
1 kS2

1. Integrated Schmitt circuit. Each Schmitt trigger contained in the type 7413 package functions logically as a four -input NAND gate. Tying
all four inputs together maximizes the size of signal that can be accepted. Transistors 02 and 03 are basic Schmitt; remaining transistors form
TTL-type NAND gate. Multiple -emitter transistor Oi and transistors 05 and 09 operate as diodes. Output of Schmitt is TTL-compatible.

101
Q3 to ground can be found from the Thevenin equiva- The equation for VT-- becomes:
lent circuit: VT- = -VBE1 VBE2 - V73 + VC2 = -VBE1 V1
YE = (46012)(4.6 v)/(750 S2 + 450 S2) = 1.74 v Replacing V1 with Eq. 3:
Assuming values of VBE1 = 0.7 v and Vy2 = 0.55 v: VT- = -VBE1 VBE2 (VCC -V73)/(1+RC2/RE)
VT+ = -0.7 v + 0.55 v + 1.74 v = 1.59 v Since VBE1 approximately equals VBE2, this equation
Negative -going threOiold VT- can be determined by reduces to:
assuming transistor Q2 is on and saturated. As the input VT- = (Vcc - V73)/(1 Rc2/RE)
to the Schmitt trigger drops, transistor Q3 begins to turn With the circuit component values shown and with Vy3 =
on and the circuit conditions of Fig. 2b apply. When Q3 0.55 v, the negative -going threshold voltage can be eas-
turns on, transistor Q2 is in the active state and its col- ily evaluated:
lector voltage is: VT-- = (5 v - 0.55 v)/[1 + (21d2)/(460 2)1 = 0.83 v
V02 = Vcc 1C2RC2 (1) Both threshold voltages vary considerably with sup-
Computing Q2's collector current: ply voltage Vcc. For example, a 1-v change in Vcc will
Ic2 = [a2(Vi - VBE2) /RE (2) cause a variation of 380 millivolts in positive -going
where a2 is the common -base current gain of transistor threshold voltage
Q2. Substituting Eq. 2 into Eq. 1 and solving for Vi with dVT+/dVcc = 0.38
a2 = 1 yields: Similarly, for VT_:
V1 = VBE2 -V73 VCC RC2(V1 VBE2)/RE dVT_/dVcc = 141+ Rc2/RE) = 0.185
or: The type 7413 Schmitt trigger can operate over a
V1 = VBE2 (Vcc - Vy3)/(1 +RC2/RE) (3) wide temperature range. Positive -going threshold VT+ is
almost insensitive to temperature changes because its
thermal variations either cancel or are negligible. And
negative -going threshold VT_ is only slightly tempera-
ture -dependent:
VT_ = Vcc -V7/(1 +RC2/RE)
Differentiating this equation with respect to tempera-
ture (T) yields the negative -going threshold tempera-
ture dependence:
dVT_/dT = dVy/(1 +Rc2/RE)dT
Since dVy / dT = -2 my/°C, then:
dVT_/dT = 0.37 my /°C.
Widely used Schmitt circuits
The integrated Schmitt trigger is useful in a variety of
applications. For instance, Fig. 3a shows a simple RC
multivibrator that can be implemented with only one-
half a type 7413 Schmitt trigger, plus a single resistor
and a single capacitor.
Circuit operation is simple. Capacitor C is initially
uncharged and has a voltage across it that is less than
VT+; the circuit's output level is high. The capacitor
charges through resistor R and through multiple -emit-
ter transistor Q. When the voltage across C reaches
VT+ (about 1.6 v), the Schmitt changes state so that its
Vcc output goes low, discharging C through R and saturated
1 02 R c2
1.2 kS2
+5 V output transistor Q7. When capacitor voltage drops to
2 kS2
Vc2 VT_ (about 0.8 v), the Schmitt switches to its high state
and C begins to charge again.
Several dc considerations influence the selection of
resistor R. It must not draw excessive current from the
VBE2
output, but its value must be high enough to prevent
VI
RE
the loading of subsequent stages. Its maximum value is
460 E2 restricted by the current -sinking capability of output
transistor Q7. A resistance of 390 ohms is recommended
by one manufacturer of the type 7413 Schmitt for a
fanout of 2.
With R = 390 ohms, the equations for output pulse
2. Calculating threshold voltages. Positive and negative switching width are:
thresholds are the most important Schmitt parameters. Simple way tH = 0.150
of finding positive -going voltage (VT+) needed to switch from high to tL = 0.34C
low state is to use Thevinin equivalent (right) of circuit (left) in (a), where tH is the period of time (in nanoseconds) during
with 02 in cutoff and 03 in saturation. To compute negative -going which the multivibrator is high, and tL the duration (in
threshold (VT_), 02 is saturated while 03 is cut off, as in (b). nanoseconds) of the low output period. (The units of ca -

102
 140 S2 ,-SILICON DIODE
The Schmitt trigger reviewed
tH
A variation of the basic Eccles -Jordan bistable multi -
1390 St)
vibrator, the Schmitt trigger is also called an emitter -
INPUT OUTPUT coupled binary. Unlike the conventional binary, it ex-
hibits hysteresis, switching from one state to another
1/27413
at different threshold voltages, depending on the di-
rection of the triggering input. Additionally, the
Schmitt's input terminal is not involved in regenerative
he- t switching and so keeps the same potential during and
after transition, creating an excellent triggering circuit.
OUTPUT
Normally, the Schmitt operates with a loop gain of
greater than unity. Circuit hysteresis can be elimi-
nated by making the loop gain equal to unity through
resistance adjustment. The positive- and negative -go-
ing threshold voltages can also be shifted.
Vcc
For a more -than -unity loop gain, input voltage v1
must reach positive -going threshold V1 before the
Schmitt can switch to its high output state (v0 = VH).
When vi decreases, it must pass the V1 threshold level
OUTPUT before the transition to the low output state occurs
(vo = VL). The circuit is unstable in the region be-
'/27413 tween VL and VH. For the low output state, 01 is off
and 02 on; for the high state, 01 is on and Q2 off.
(c)

Vcc

3. Using the Schmitt. Simple multivibrator (a) requires only external

resistor and capacitor; adding diode and second resistor achieves
50% duty cycle. For pulse stretcher (b), extra resistor compensates
for unit -to -unit processing variations so that output pulse width is
more predictable. Biasing Schmitt at its mid -threshold point enables
it to convert sine waves to square waves (c).

pacitance are picofarads.) Multivibrator frequency in

hertz, therefore, will be:
f = (2.02 x 10-3)/C

j
where C is in farads. Vo
As the timing equations indicate, the multivibrator
VH
output is low approximately 70% of the time. Adding a '4 LOOP GAIN 1

resistor and a diode, as shown by the dashed lines, re- LOOP GAIN = 1
duces the discharge cycle time, yielding a multivibrator LOOP GAIN - 1

with a duty cycle of 50%. With the values shown, output VL

frequency becomes:
f = (3.3 x 10-3)/C V2 v,

Again, C is expressed in farads and fin hertz.

A relatively simple one-shot (or pulse stretcher) can
be built with the circuit configuration of Fig. 3b. When square wave, as shown in Fig. 3c. Resistors R1 and R2
an input pulse is applied to NAND gate G1, its output bias the Schmitt trigger at the mean value of its thres-
goes low, but the Schmitt trigger's output remains high. hold voltages to obtain the necessary 50% output duty
Before the Schmitt can switch to its low output state, ca- cycle. To avoid differentiating the input signal, the im-
pacitor C must charge to VT+. The charging resistor is pedance of capacitor C at the operating frequency of in-
the base resistor of transistor Ql. Output pulse width for terest should be much less than the parallel resistance of
this circuit is nominally: R1 and R2:
t = 1.95C C approximately equals (4.3 x 10-3)/f
where t is in nanoseconds and C in picofarads. where C is in farads and fin hertz.
Variations in the charging resistor cause output pulse The converter can be operated at frequencies up to
width to vary from unit to unit. Performance can be im- about 8 megahertz. If all four inputs of the Schmitt trig-
proved by inserting a resistor between the supply and ger are tied together, peak input voltage can be as high
the capacitor. Pulse width then becomes: as 6.5 V. If only one input is used, maximum input volt-
t = 0.26C age becomes 5.5 V peak. The minimum input voltage
for tin nanoseconds and C in picofarads. Pulse width is should be about 1 V peek.
reduced with the additional resistor. BIBLIOGRAPHY
Another useful circuit converts a sine wave to a "TTL Schmitt Trigger," Texas Instruments Application Report, Bulletin CA -152

103
Electronic fuel injection
reduces automotive pollution
An MOS read-only memory is at the heart of a British system
that minimizes pollutants by measuring exact quantities
of fuel and timing their insertion into combustion chambers
by Malcolm Williams, Joseph Lucas (Electrical) Ltd., Shirley near Solihull, England

0 In a massive effort to limit automotive exhaust pollu- consequence, gives rise to fewer partially burnt pollu-
tion, one of the devices being evaluated carefully is an tants. It is also important to obtain an accurate mea-
electronically controlled fuel -injection system. surement of engine conditions and fast, accurate mea-
Besides the advantage over standard carburetion of surement of air -mass flow. The system must be able to
increased power offered by fuel injection, the accurate supply the optimum amount of fuel when the air in the
cylinder -to -cylinder fuel distribution and optimum fuel inlet and exhaust manifolds is resonating.
control under turbulent manifold air conditions enable The engine parameters that best meet these require-
fuel injection to minimize exhaust pollution. ments are throttle angle and engine speed. But the cost
It is unlikely, however, that a fuel -injection system of a complex control function is prohibitive if standard
alone will meet the latest legal emission requirements, analog electronic techniques are applied. However, by
and some form of exhaust treatment also will be neces- using a digital memory to store the fuel -demand charac-
sary. Nonetheless, the cost of exhaust treatment in a ve- teristics, a system becomes economically attractive.
hicle fitted with fuel injection would be lower than it A prototype fuel injection system has been developed
would be without fuel injection. for a standard 150 -cubic -inch, six -cylinder Triumph se-
To achieve low -pollution exhaust emissions, a high dan. Engine speed and throttle angle information are
air -to -fuel ratio is required over the engine's complete stored in a digital memory containing 256 seven -bit
working range. This minimizes wasted fuel, and, as a words. Each word in the mos read-only memory repre-

FUEL TANK

11

FUEL
PUMP PRESSURE
FILTER
REGULATOR
SOLENOID (ONE FOR
INJECTOR EACH CYLINDER)

C:=1
ATMOSPHERIC
PRESSURE SENSOR
( ILY
THROTTLE
ENGINE
TEMPERATURE
AIR INLET
TEMPERATURE ANGLE SENSOR SENSOR
SENSOR

SPEED
SENSOR
POWER -4- TIMING ANALOG TO
AMPLIFIERS DIGITAL
CIRCUIT DIGITAL CRANK
MEMORY
CONVERTER ANGLE
SENSOR

ANALOG TO -4- INTER-

DIGITAL POLATION
CONTROL CONVERTER FUNCTION
UNIT

1. FIll 'er up. The electronic fuel -injection system shown as part of the auto power plant is designed to provide accuiate amounts of fuel to
the engine cylinders based on demand (throttle angle), engine speed, engine temperature, and ambient conditions.

104
 VOLTAGE
CLOCK
PROPORTIONAL TO
THROTTLE POSITION
UP
THROTTLE 4 BIT POWER SOLENOID
SHAPING DUAL
ANGLE UP/DOWN STAGE --100 INJECTOR
SENSOR
CIRCUIT COMPARATOR -- COUNTER GROUP 1 BANK
SUMMING DOWN GROUP 1
AMPLIFIER
TRIANGULAR .00
SUMMING
WAVE
NETWORK
GENERATOR
VOLTAGE PROPORTIONAL
TO FUEL REQUIRED
90° PHASE D GITAL 00 - VOLTAGE -
LAGGED MEMORY -so - SUMMING STEERING
TO -TIME -
TRIANGULAR 256 -WORD -ow - NETWORK r CIRCUIT
WAVE 7 -BIT n: CONVERTER

GENERATOR SUMMING A
AMPLIFIER
.00
SUMMING -IP- SYNCRONIZATION
NETWORK CIRCUIT

SUMMING AMPLIFIER up r --
AUXILIARY -11410 TIME -TO - 4BT ENGINE BAROMETRIC POWER
DUAL STAGE
DISTRIBUTOR VOLTAGE UP/DOWN TEMPERATURE PRESSURE
COMPARATOR GROUP 2
CONTACTS CONVERTER COUNTER SENSOR SENSOR

VOLTAGE DOWN
PROPORTIONAL SOLENOID
TO ENGINE SPEED INJECTOR
AIR
TEMPERATURE BANK
SENSOR GROUP 2

2. Electronic controller. The complete fuel -injection system comprises TTL, MOS, and linear devices to process analog information from
engine and throttle to produce signals that help determine the exact fuel quantity to be fed into each cylinder.

engine's
V REFERENCE
Both digital and linear ICs were used. Their functions
VOLTAGE include counting, comparison, and amplification to con-
FOLLOWER vert analog data into a form that will address a digital
OUTPUT
read-only memory and for interpolation between the
VOLTAGE exact digital steps. Interpolation is necessary because
PROPORTIONAL the memory covers the range of possible engine fuel re-
TO ENGINE quirements and must progress smoothly from one state
SPEED
to another.
When the driver switches on the ignition, the electric
fuel pump mounted in the trunk sends fuel to the sole-
noid injectors in the engine. Injection pressure is con-
stant, and the controller determines the quantity of fuel
injected by controlling the time the injector is open.
Hence, each word stored in the memory corresponds to
a time interval.
AUXILIARY SYNCHRONIZED
On turning over the engine, auxiliary contacts in the
DISTRIBUTOR TRIGGER distributor cause the solenoid injectors to commence
CONTACTS CIRCUIT fueling at the correct point in the engine cycle and also
provide engine -speed information to the control circuit,
3. Speeding ticket. To get engine -speed information, auxiliary con- selecting the appropriate memory word (Fig. 2). With a
tacts are mounted on the distributor. The circuit ;shown above con- cold engine, the fuel supply is increased over the
amount a hot engine would need by multiplying the
verts the time measured between contact pulses into analog voltaae.
This voltage is converted to digital signals. quantity stored in the digital memory by a factor de-
pendent on the engine's temperature sensor. This sen-
sents a different quantity of fuel. By reading out the sor, which is fitted on the engine block, is tailored to
right word at the right time, the correct quantity of fuel provide easy starting with minimum emissions and
is injected into the induction port of the engine. The se- thereby function as an automatic choke.
lection of the word is controlled with the information The driver controls the power delivered by the engine
supplied by strategically placed sensors that monitor the by depressing the accelerator pedal, which controls the

105
throttle aperture in the air intake. The angle of the re- the comparator threshold level is reached.
stricting butterfly valve, in conjunction with the engine The counter is then allowed to count up the clock
speed, measures air flow. Both the driver's demand and pulses, thus changing the digital word until equality of
the engine's air intake can be measured by coupling a both signals again is achieved. The digital signal re-
potentiometer to the throttle shaft. As the throttle is de- mains constant until an error greater than half the least -
pressed, extra air flows into the engine, and simulta- significant bit occurs, at which point the digital value
neously a new word is addressed in the memory, thus changes to reduce the error. This particular a -d con-
optimizing the amount of fuel injected into the cylin- verter tracks the throttle signal rapidly.
ders. The auxiliary contacts mounted in the distributor
To account for changes in air density due to climatic provide engine -speed information, which is turned into
conditions and radiant heating by the engine of the in- an analoe voltage from the time between contact pulses
take air, an atmospheric pressure sensor and air -inlet
temperature sensor are also installed. EXHAUST POLLUTION EMISSIONS IN EXPERIMENTAL CAR 10.
(grams/mile)
Data converted
Achieved
The throttle angle and engine -speed information Pollutant U. S. Federal California
on Test Engine*
both have to be converted into digital form to address
the memory. Whenever the combination of throttle 1971 LEGISLATIONI 2

angle and engine speed hit a precise point in the Carbon Monoxide 23 23 3.5
memory, the engine's fuel demand is obtained by direct Hydrocarbons 2.2 2.2 0.8
reference. However, when the values of throttle angle Nitrogen Oxides 4.0 0.8
and engine speed are between these exact memory sites,
the engine fuel demand is calculated by interpolation
from the adjacent stored data. 1975 LEG ISLATION3' 4

For the throttle signal, a potentiometer output pro- Carbon Monoxide 3.4 12 3.5
portional to throttle angle is first shaped to define more Hydrocarbons 0.41 0.5 0.8
closely the changes at the initial low throttle opening. Nitrogen Oxides 3.0 1.0 0.8
The signal is then converted into digital form by a "7-mode California cycle
simple analog -to -digital converter, which uses a feed- REFERENCES
back resistor network. For signals of equal amplitude, 1. Federal Register, Vol. 33, No. 108, June 4, 1968
the output from the amplifier is zero, and the digital 2. State of California Air Resources Board, Nov. 20, 1968
word -signal feedback from the counter is equal to the 3. Federal Register, Vol. 36, No. 128, Part II, July 2, 1971
input analog signal. When this analog signal increases, 4. State of California Air Resources Board, Resolution 20-4, Jan. 21,1970
the output from the amplifier decreases from zero until

12 1000110 1000100 1000000 0111101

3.6 1000001 0111101 0110111 0110010

7.5 0110110 0110000 0101010 0100011

5.7 0100111 0100010 0011011 0010101

I 1183 1364 1580 1800

ENGINE SPEED (RPM)

4. Memory lane. Digital words representing throttle angle and en 5. Fuel program. Each box in the memory program has a digital
gine speed are fed into a decoder circuit, which then selects a pro code number representing location and proper amount of fuel re-
gramed line in the memory matrix to calculate fuel quantity. quired for that particular operating condition.

106
(Fig. 3). On closure of the first contacts, capacitor C 1 The fuel surface represented by the stored informa-
rapidly charges toward a reference voltage. After about tion in the memory would consist of a number of flat
5 microseconds, S 1 opens, and the capacitor decays planes (Fig. 6), the height being proportional to the fuel
through a fixed resistor R for the remainder of the en- quantity to be delivered per engine stroke. However,
gine revolution. When the second distributor contact the required surface cannot have these discrete steps.
closes, this voltage is sampled for about 10 ps through The control unit achieves a smoothly varying surface by
switch S2 and is stored on capacitor C2. The voltage fol- linearly interpolating stored data whenever the combi-
lower acts as a buffer to prevent discharge of capacitor nation of the throttle angle and engine speed do not hit
C2. This cycle is repeated for every engine revolution. an exact memory site (Fig. 7).
Another a -d converter changes the analog voltage Circuits interpolate
representing the engine speed to a digital word. The two
four -bit digital words-one representing throttle angle, There are two interpolation circuits-one for engine
the other representing engine speed-are fed to decod- speed and one for throttle angle. Each generates a posi-
ing circuits; each decoder circuit then selects a particu- tive and negative triangular wave of fixed amplitude
lar line in the memory matrix (Fig. 4). and frequency. The effect, for each parameter, is to
make the memory sites each side of an intermediate
Engine -speed conversion point read out alternately for periods proportionate to
For example, if line A on the throttle input is selected the distance of the sites from the intermediate point. To
and a positive voltage is impressed on the line, then do this, the triangulations superimpose on the sensor
diode D will conduct. If line B is sensed by the engine - voltage from the summing amplifier. To distinguish be-
speed decoder, the positive voltage through diode D tween the two parameters, one wave generator is placed
will be sensed on line B if the particular line sensed has 90° out of phase behind the other.
no diode on line . If the particular line sensed has no When a sensor output voltage corresponds exactly to
diode, then a negative voltage would be sensed. So by one of the 16 matrix input levels tied to that sensor, the
the presence or absence of a diode, a binary 1 or 0 can positive and negative triangular oscillators are equal in
be programed for any particular input line. amplitude, and the running voltage level in the com-
This example shows a simple memory matrix of 16 parator is within the threshold levels identifying that
one -bit words. To increase the number of bits, separate particular matrix line. When the sensor voltage moves
matrixes should be addressed simultaneously. The dig- slightly, one triangle peak pushes the comparator volt-
ital memory used in the controller consists of 256 words, age within the levels identifying the next matrix line for
each of seven bits. mos transistors are used to form the a fraction of the triangle cycle time.
matrix instead of diodes. Half -way between the lines, half the triangle extends
The digital information (Fig. 5) stored in the square into the next line level, and so on. To connect voltage
corresponding to the intersection of throttle angle 9.6° levels and matrix lines, each line in each parameter is
and engine speed 1,580 rpm is 0110111. The data in the identified by a four -bit digital word, and an a -d con-
adjacent sites in the memory bear no relation to each verter translates voltage into word. Hence, two four -bit
other, and whatever the optimum fuel required by the words act on each of the seven memory planes, identi-
engine at the appropriate site, it can be provided by the fying in each plane a particular memory site. All seven
control system. sites are read out simultaneously, making a seven -bit

6. Injection plot. Without interpolation, a three-dimensional plot of 7. Bumpy road. With interpolation to facilitate continuous engine
speed, throttle angle, and fuel amount would look like this series o operation, the surface of an X -Y plot becomes a series of undula-
discrete and flat planes produced by an X -Y plotter. tions, assuring smoother engineer operation.

107
 A B

01

401
1

q r'G

)
s St

I
J K. L

M N 0 P

9. It's a gas. Prototype unit contains both ICs and discrete semi-
conductors, but follow-on versions will replace analog with digital
THROTTLE ANGLE
devices. Protrusions on the top of the unit are power resistors used
to set the time -constant of the solenoid fuel injectors.

8. Gas station. When fuel demand coincides with an exact memory which, by means of the steering circuit, turns on an ap-
site, the interpolation signal causes only that site to be addressed propriate group of injectors. At the same time, the volt-
(point F); but when fuel needs don't correspond to an exact site- age -to -time converter commences its timing period.
which is normally the case-an interpolation of as many as four adja- This period is a direct function of the voltage sup-
cent word sites may be necessary (point Y). plied from the memory, but it's modified, depending on
engine temperature, air temperature, and barometric
word identifying an exact injector opening time. Be- pressure. At the end of the timing period, the steering
cause most throttle and engine -speed selections will be circuit is reset, and the appropriate group is turned off,
somewhere between exact sites, most of the time the cutting the fuel supply to those particular cylinders.
memory output, under the influence of the interpolation
circuits, is cycling around four adjacent sites. Prototype has standard parts
Fig. 8 shows the effect of the triangular waves on the The prototype unit specifically developed to control
memory where the memory sites, alphabetically la- the fuel system for the test vehicle (Fig. 9) is 2.25 inches
belled, represent the engine conditions when the fuel by 6.75 in. by 10.25 in. All the components, as well as
data have been measured. The solid lines represent the some discretes, are readily available ics; for instance, a
decision points where the a -d converter changes from standard MOS ROM has been programed with the infor-
one word to the next, thus selecting a different memory mation pertaining to the Triumph that had been ob-
site. When the driver's demand, i.e., throttle angle and tained from engine tests. However, in production units,
the resulting engine speed, are both sent to the same custom -designed integrated circuits would be used ex-
memory site, the interpolation signal causes only that clusively.
memory site to be addressed. For example, if F is the Although the system is in the early development
selected point, the resultant path traveled through the phase, it has already shown considerable improvement
matrix is the square bounded by p, q, r, and s. When the over currently available fuel -control devices. Develop-
driver demands more power, point x would be selected, ment is being continued to assess the full potential of
and the output from the memory would be alternatively this type of control system. The pollution figures from
the words stored at memory sites F and G. this unit have fallen short of meeting the new Federal
As the matrix is swept at a constant velocity, the time and California requirements for 1975 (see Table 1), al-
for which each word appears at the output of the though it did perform within the limits of previous legis-
memory is proportional to the throttle -angle position lation. This unit has shown that accurate fueling under
between F and G. A lag in the summing circuit aver- both steady-state and transient conditions can reduce
ages the memory output to give linear interpolation be- pollutants and work in conjection with a total antipollu-
tween F and G. With the throttle angle and engine tion system.
speed set to select the point Y-the typical operating sit- Current work is aimed at digitizing the analog speeu-
uation-the output from the memory consists of the four measurement stage and the analog fuel -measurement
adjacent word sites G, H, K, and L in cyclical sequence. stage after the memory. A digital interpolation tech-
These words are averaged to produce a two-dimension- nique is being devised so that the only analog stage in
al interpolated surface (Fig. 7). the system will be throttle -angle sensing. This should
To control the fuel delivered to the engine, auxiliary mean that all the electronics can be packed on one
distributor contacts activate the synchronization circuit, bipolar lc and one MOS IC. El

108
Take a bit of advice:
use 16 -bit converters carefully
Their wide dynamic range and extremely fine resolution allow design
simplification, while improving performance and cutting costs.
But take care in applying them, or their performance will be wasted.

by Wayne Marshall and Cyril Brown, Analog Devices Inc., Norwood, Mass.

A 16 -bit analog -to -digital or digital -to -analog con- millivolt common -mode signals. In short, factors of only
verter that actually works is something of a modern secondary concern in the design and use of a 12 -bit con-
technological marvel (See Part I of this two-part article verter can have an overwhelmingly adverse impact on
in the last issue of Electronics). True 16 -bit devices are the operation of a 16 -bit unit.
now available off -the -shelf, but at prices on the order of To understand how and where these precision con-
$1,000. The extremely fine resolution, monotonicity, verters can be most effectively applied, it is helpful to
high accuracy, linearity, and broad dynamic range that begin by studying their key characteristics, then exam-
characterize these converters make them ideal for a ining applications that these characteristics suggest, and
wide variety of applications-from nuclear research to finally discussing methods to ensure that the devices
precision function generation. will operate properly in a real -world environment.
However, the unsophisticated user may waste the ad- Would you believe 65,536:1?
vantages of the converter's tight performance specifica-
tions unless he exercises extreme care in circuit design. The most outstanding characteristic of a 16 -bit con-
Potential perils that must be considered include thermo- verter is its extremely wide dynamic range-that is, the
couple effects, voltage drops in short lengths of wire, ratio of full-scale value to the value of the least -signifi-
low-level noise, radio -frequency interference, and even cant bit (LsB) it can handle. For a 16 -bit converter, the

GAIN -SWITCHING
-6 V 0 LOCIC
+8 V 0 1
+1 V

-0.5 V
+0.09 V 0 10-BIT
}DIGITAL
A-D OUTPUT
CONVERTER

V
DIGITALLY ADJUSTED
-4-
OFFSET CURRENT PROGRAMMABLE
CURRENT
SOURCE

-6 V 0

+8 V 0
16 -BIT }16 -BIT
+1 V A- D DIGITAL

-0.5 V °04° CONVERTER OUTPUT

+0.09 V

1. Simplicity. Complexity of conventional gain -changing data -acquisition circuitry (top) is reduced by exploiting wide dynamic range of 16 -
bit converter (bottom). Bottom circuit is not only simpler, it's also more accurate, faster, and less expensive.

109
dynamic range is 216:1, or 65,536:1, or slightly more
than 96 dB. Thus a d -a converter with a full-scale out-
put voltage of 10 v can change its output in steps as 0 cc

small as 150 µV, approximately. 16-BIT TIME TIME

D -A VFC
This wide dynamic range means that many naturally CONVERTER
occurring variables-like light and sound levels-can be
handled by these converters without range -switching or
variable -gain amplifiers. The high resolution implied by
a wide dynamic range means that accurate ratio mea- DIGITAL
surements can be made between two large -amplitude COUNTER COMPAR- FREQUENCY
COUNTER
signals that differ by only a small amount. ATOR
A true 16 -bit converter must have a linearity specifi-
cation commensurate with its resolution. For 16 bits,
this means that the maximum deviation from perfect
linearity should never exceed 0.0015% of full scale. CLOCK
Some applications that only need, say, 12 bits of resolu-
tion are better served by a 16 -bit converter because of CHART RECORDER,
its superior linearity. PRINTER, ETC.
Scintillation counters, ion chambers, and other nu-
clear instruments, for example, are frequently used to
separate energy levels, charge levels, or particle mo-
menta into well defined categories. The number of cate- =.-
gories that can be obtained is determined by the resolu-
tion of the converter, but the uniformity of the category
"widths" is determined by its linearity. 2. Function generator. D -a converter is used as ramp generator in
Monotonicity is the tough one setup for testing linearity of voltage -to -frequency converter.

Monotonicity is the toughest single specification that solute ones, are required. Nonetheless, of course, the
a converter must meet, because it imposes a maximum relative magnitudes must be measured with considera-
limit of ±1/2 LSB on the converter's step -to -step non - bly certainty, but this requires high linearity if the con-
linearity. What is most difficult is preserving the con- verter uses a reference common to all sources, rather
verter's monotonicity over a useful temperature range. than absolute accuracy.
Monotonicity is particularly important in digital servo Closely related to a converter's accuracy is its stability
systems. If such a system is trying to find a null, and the over both time and temperature. This stability must be
converter is non -monotonic around the balance point, commensurate with the need for repeating measure-
the null balance will never be attained. ments day after day without introducing discrepancies
Speed is a converter parameter that can be all-impor-
between one day's results and the next. This is often a
tant in one application and insignificant in another. relative accuracy factor, since the absolute values of
Automatic testing systems and CRT displays obviously readings may be unimportant so long as they can be re-
demand fast converters. However, many slower systems peated with precision. An example occurs in computer -
can benefit from using high-speed a -d converters to output microfilming, where it may be necessary to up-
fight noise. The trick: take several readings of the same date data on a piece of film that was originally recorded
input signal in quick succession and average them. days, or even months, earlier.
Noise will be reduced proportionally to the square root
of the number of readings. Eliminating range switching
Accuracy's true meaning What applications do the preceding characteristics
suggest? One obvious one is the elimination of range -
The definition of accuracy is a semantic problem that switching and variable -gain amplifiers. An example of
tends to separate the purists-who think in terms of Na- such an application is in a multichannel data acquisi-
tional Bureau of Standards certification-from the engi- tion system that must digitize analog signals developed
neers and scientists with data to acquire or distribute. by thermocouples, strain gauges, and other transducers
Since the best zener reference sources drift at some 5 (Fig. 1). The upper diagram of Fig. 1 shows the rather
ppm/ °C, it is difficult to see how a converter can claim elaborate arrangement required to switch amplifier gain
any reference to NBS certified standards. The reality is automatically from channel to channel to bring the low-
that 16 -bit converters don't claim absolute accuracy to level signals up to the converter's full-scale input range.
0.0015%, but offer 15 ppm resolution, linearity to 15 The bottom illustration shows the dramatic reduction in
ppm, and excellent stability, which can be exploited for complexity provided by a 16 -bit converter.
valuable results. But the 16 -bit machine does more than simply elimi-
Absolute accuracy is certainly the critical specifica- nate a few extra components. It can improve perform-
tion for a digital voltmeter, which must make error -free ance and cut costs at the same time. To understand
readings from a variety of unrelated sources. By con- how, suppose that the system of Fig. 1 requires the
trast, an a -d converter usually takes readings within a nominally 8-v input to be resolved within 1 millivolt
given system, where relative voltage magnitudes, not ab- over the range of 8 to 9 v. Since a 10 -bit converter with

110
a 10-v reference provides only slightly more than 100 wide variety of output functions with a quantizing noise
increments between 8 v and 9 v, preamplification by 10 level approximately 96 dB below full scale.
is needed to obtain the desired resolution. The usefulness of such a function generator is illus-
How can a voltage between 8 v and 9 v be multiplied trated in Fig. 2. A setup is shown for testing the line-
by 10 and still fit into the converter's 10-v amplitude arity of precision voltage -to -frequency converters
range? The answer is to use an op amp and apply to its (vFCs). The automatic testing setup makes use of the
summing junction a stable bias current having a magni- d -a converter's excellent linearity as a ramp generator.
tude that is exactly sufficient to make the amplifier out- Clock pulses accumulated in the up -down counter con-
put zero when the 8-v input is applied. This will allow tinually update the d -a converter, feeding an analog
the 1-v difference signal between 8 v and 9 v to be ex- voltage ramp to the VFC undergoing the linearity test.
panded to fill the converter's full 10-v span. In this way, By holding the quantizing noise of the converter's ramp
the 10 -bit converter can resolve the amplified signal waveform down to the -96-dB level, while simulta-
variation to 1 part in 1,024, or slightly better than 1 my. neously ensuring linearity at least tenfold better than
However, in addition to a programable-gain amplifier, the voltage -to -frequency converter's own performance
the auxiliary equipment now requires a bias source out- specification, highly accurate calibration curves can be
put to be fed into the amplifier's summing junction. plotted.
If the system involves many suppressed -zero readings The technique can also be applied to digital fre-
like the foregoing, the circuitry for generating the pre- quency synthesizers to produce output waveforms in
cise biasing currents adds significantly to the complexity which harmonically related noise is some 96 dB below
of the over-all conversion equipment; it also adds error the full-scale signal level. In such an application the
sources at the 10 -bit level, even though 1-mv resolution converter's output would be arranged for bipolar signal
of the expanded 1-v signal swing requires performance swings.
at roughly the 13 -bit level. An undistorted view
Although a 16 -bit a -d converter costs appreciably
more than any 10 -bit counterpart, over-all cost of Although CRT displays are widely used for presenting
equipment is considerably reduced, and performance is data to human operators -in aircraft flight controls, in-
improved: the 1-v interval between 8 v and 9 v is re- teractive computer terminals, newsprint preparation,
solved into more than 6,000 150-,uv steps, with 16 -bit, and so on -there are many additional applications for
rather than 10 -bit, operation. Further, because it takes the basic idea of using a deflected electron beam as a
an automatically switched amplifier roughly 500 micro- tool for generating and displaying data that have not
seconds for each range change, the digitizing rate is yet been fully developed.
greatly enhanced by use of the 16 -bit converter, since For example, "photographs" of Mars and other plan-
range switching is eliminated. ets could be transmitted to earth -borne cam-
A 16 -bit converter can be surprisingly useful in preci- eras, but they would have to undergo a considerable
sion function generation. Because the LSE is such a tiny amount of digital massaging before emerging in con-
fraction of the full-scale output of a 16 -bit d -a con- ventional photographic format. Similarly, in geological
verter, the converter can be programed to produce a and mineral explorations, where sonar and explosive re-
verberations are echoed from inner -earth strata and re-
3. Precise positioning. To approach the quality of a real photo- corded for later transformation into pictures of the
graph while using a TV -type raster -scanning approach, this CBS
Laboratories electron -beam recording system uses a 16 -bit d -a con- BRIGHTNESS
verter to set the beam's vertical position. Computer -enhanced photo CONTROL
8 -BIT
of earth was taken from an orbiting satellite. DAC

ELECTRON
GUN
16 BIT
DAC
VERTICAL
POSITION
CONTROL

SWEEP

/I
COMPUTER GENERATOR

HORIZONTAL

/ -
SWEEPS

STORED DEFLECTION
DATA SYSTEM

PHOTOGRAPHIC
PLATE

111
earth's geological cross-section, the raw data is pro-
cessed by a computer before being converted into an
image. E
In both these instances, the final picture is produced FEED BACK CURRENT
by exposing a piece of photographic film inside a vac- INDEPENDENT OF LOAD
uum system to a digitally positioned electron beam.
And it is often wise to use a 16 -bit d -a converter to posi-
tion the beam, even if 16 bits of positioning are not
needed. The reason is a need for the 16 -bit converter's
linearity. LOAD LOAD
CURRENT
To see why this is so, consider the electron -beam
recording system marketed by CBs Laboratories (Fig. 3).
This system uses a raster -scanning process not unlike
that of conventional television, but with eight bits of
bright -to -dark -gray shading, and digital control of the NEGLIGIBLE
AMPLIFIER
vertical position. To avoid visible distortion in the pic; BIAS CURRENT
tures it produces, the CBS system imposes tight toler-
ances on the spacing between its horizontal scan lines.
Because a d -a converter's specifications allow an er-
ror of up to ±1/2 LSB on its output amplitude, it would be
possible for successive horizontal scan lines to vary by 4. Error reduction. Four -wire output connection eliminates errors
as much as half a line width in their vertical positioning caused by voltage drop between converter output and load.
if the d -a converter's resolution were the same as the
vertical resolution of the system. By using a 16 -bit con- data from a 16 -bit d -a converter to a remote load. Be-
verter in a system that requires only 12 bits of vertical cause the dc power for the a -d converter must travel a
resolution, CBS has cut the sloppiness with which its finite distance from the power supply, and perhaps pass
4,096 scan lines are positioned from ±1/2 line width to through printed -circuit board interconnections with rel-
±-1 /32 line width. The quality of the picture that results atively high resistances, there is likely to be a voltage
is evident in Fig. 3. drop across the power cable and connectors that will
Preserving the performance raise the entire converter to some finite common -mode
voltage. This voltage may easily exceed a few 15011v
For the user to get all of the performance of which a least -significant bits. In the absence of some subtle pre-
16 -bit converter is capable, he must avoid the inadver- cautions in feeding analog data into the measuring cir-
tent introduction of errors on the order of 1 LSB. This is cuitry, the digital conversion is therefore likely to be in
not as easy as it sounds. For example, just 15 mA-the error by many least -significant bits, because of this com-
full-scale output of some d -a converters-flowing mon -mode effect alone. The solution is to use a high-
through 2 feet of 18 -gauge wire will drop approximately performance differential input circuit, or buffer, at the
165 luv, the equivalent of slightly more than 1 LSB. a -d converter's front end. The common -mode rejection
To combat this problem, converter manufacturers of such a circuit then scales the common -mode voltage
have introduced four -terminal output -connection tech- down to negligible proportions.
niques (Fig. 4) that allow the user to include the imped- Shot noise, 1/f noise, and thermal noise can only be
ance of the connecting circuit inside the amplifier's held to an irreducible minimum by careful circuit de-
feedback loop. In short, feedback signals are derived sign and component selection. To keep the noise down,
from the load, ensuring that the converter's output am- extreme care must be taken with the wiring associated
plifier places the desired voltage at the actual load ter- with the converter. In particular, analog and digital
minals, and not merely at the converter's output pins. grounds should be kept separate to prevent the leakage
In some applications-CRT displays and certain auto- of large digital pulses onto the analog lines.
matic test systems, for example-the output -voltage Because of high source impedances and other con-
transients produced by the d -a converter cannot be tol- straints imposed by various signal sources, it is much
erated. Deflection amplifiers tend to integrate these more difficult to minimize noise effects when measuring,
glitch spikes, creating dc offsets that may persist for instead of generating, precise analog values. In fact, the
many microseconds, creating a noticeable distortion in 16 -bit a -d converter is vulnerable to radio -frequency in-
any smooth CRT trace. terference, 60 -Hz pickup, feedthrough from digital
Similarly, semiconductor testers that use d -a con- power -supply transients, and even magnetic coupling
verters to control applied voltages and currents cannot from nearby motors and transformers. Therefore, even
tolerate glitches either. When these converters try to if all possible care has been taken to minimize circuit
creep up slowly on critical threshold voltages, large noise, it still may be necessary to use the data -averaging
glitch spikes can burst through the critical region, trig- technique mentioned earlier. This technique exploits
ger erroneous responses, and even damage sensitive de- the converter's high speed by taking several readings of
vices. In such applications, the use of a sample -and - the same input data. In this way, averaging techniques
hold deglitch module becomes mandatory. will yield a reading that approximates more closely the
The problem of getting data into a 16 -bit a -d con- true analog input. (Noise is reduced in proportion to the
verter is analogous to the difficulties in transmitting square root of the number of readings.)

112
Tables shorten design time
for active filters
Both low-pass and high-pass Butterworth -Thomson filters, which use either
unity -gain or infinite -gain amplifiers, can be designed from tables of
normalized component values; filter orders as high as 10 are covered

by Farouk Al -Nasser, Honeywell Test Instruments division, Denver, Colo.

El As the lc operational amplifier grows steadily better subscript T the Thomson filter, the equation that de-
and cheaper, active filters are becoming more popular fines the location of a Butterworth -Thomson pole can
than passive. Their performance is superior and more be written as:
predictable, they have no need for impedance match- PBT = RTmeXPH(OB - MOB - 9T))]
ing, nor are nonlinear inductor properties a problem, where m is a parameter that varies between 0 and 1.
since the active filter uses an op amp and an associated When m = 0, the Butterworth -Thomson poles are the
RC network to simulate inductance. same as those of the Butterworth filter:
But perhaps the greatest attraction that active filters PET = eXP(JOB) = PB
hold for the designer is that they lend themselves to a When m = 1, the Butterworth -Thomson and the
building-block approach. Being inherently isolated, Thomson poles are the same:
they can be cascaded to realize higher -order filter func- PBT = RTeXP(JOT) = PT
tions, and the passive -component values needed to im- From these equations, the pole locations, the magnitude
plement each filter block or section can be computed in and angle of the pole vectors, and the filter Q factor can
a normalized form and tabulated. This also means that be computed for various values of parameter m. Tables
active -filter design can be reduced to a matter of look- 1 through 6 list this data for m = 0, 0.2, 0.4, 0.6, 0.8,
ing up tables, once the engineer has determined the and 1 for both even and odd filter orders from 2 to 10.
type of filter required and chosen a filter circuit. Data for intermediate values of parameter m can be ob-
Selecting the right type of active filter depends on the tained by interpolation.
signal being processed, as well as the time- and fre- From the general transfer function of the But-
quency -domain properties that the filtered signal must terworth -Thomson filter, which can be found in refer-
have. The Butterworth type of filter, for instance, exhib- ence literature, both the frequency- and time -domain
its flat passband amplitude characteristics up to the cut- characteristics can be obtained. Graphs 1 and 2 relate
off frequency, a rolloff of -6n decibels per octave be- filter gain to normalized frequency for m = 0.4 and
yond cutoff (where n is the order of the filter), and m = 0.8, respectively. The group delay characteristics,
nonlinear phase characteristics that lead to a nonlinear again for m = 0.4 and m = 0.8, are drawn in Graphs 3
group delay. On the other hand, the Thomson filter and 4. For any order filter, in either the frequency or
(also referred to as the normalized Bessel filter) has a time domain, the Butterworth -Thomson filter changes
linear phase response in the passband, and an ampli- smoothly from the Butterworth to the Thomson charac-
tude response that is not as flat as that of the But- teristics, as parameter m varies from 0 to 1.
terworth filter in the passband and stopbands. Group Active -filter circuits synthesize a second -order trans-
delay for the Thomson filter is constant up to cutoff. fer function with one or more op amps and an RC net -
In some cases, a compromise characteristic that trades
off the best properties of the Butterworth and the
Thomson filter is needed. The result is the class of But-
terworth -Thomson filter. Its characteristics vary
smoothly between the maximally flat amplitude of the
Butterworth filter and the maximally flat envelope de-
lay of the Thomson filter.
Butterworth -Thomson filter properties
In the complex frequency plane, the poles of the But-
terworth -Thomson filter lie between those of the But-
terworth and Thomson types, as shown in Fig. 1. The
actual pole location is denoted by P, the length of the 1. Pole locations. Butterworth -Thomson fitter compromises the flat
vector to the pole by R, and the angle of this pole vector amplitude of the Butterworth filter with the flat envelope delay of the
by O. Letting subscript BT represent the Butterworth - Thomson filter Its !Soles (PBT) are located between those (PB) of the
Thomson filter, subscript B the Butterworth filter, and Butterworth filter and those (PT) of the Thomson filter.

113
 TABLE 1 TABLE 2

POLE LOCATIONS FOR m = 0 POLE LOCATIONS FOR m = 0.2

Angle from Angle from

Order Real Imaginary Magnitude neg. real 0 Order Real Imaginary Magnitude neg. real G
factor factor
axis (deg.) axis (deg.)

2 -0.70711 0.70711 1.00000 45.00000 0.70711 2 -0.74314 0.66913 1.00000 42.00000 0.67282
3 -1.00000 0.00000 1.00000 0.00000 3 -0.98804 0.00000 0.98804 0.00000
-0.50000 0.86603 1.00000 60.00000 1.00000 -0.55189 0.84115 1.00604 56.73050 0.91145
4 -0.92388 0.38268 1.00000 22.50000 0.54120 4 -0.92089 0.35967 0.98863 21.33390 0.53678
-0.38268 0.92388 1.00000 67.50000 1.30656 -0.43822 0.91164 1.01150 64.32650 1.15409
5 -1.00000 0.00000 1.00000 0.00000 5 -0.98484 0.00000 0.98484 0.00000
-0.80902 0.58779 1.00000 36.00000 0.61803 -0.81940 0.55883 0.99182 34.29390 0.60521
-0.30902 0.95106 1.00000 72.00000 1.61803 -0.36431 0.94842 1.01598 68.98710 1.39440
6 -0.96593 0.25882 1.00000 15.00000 0.51764 6 -0.95464 0.24348 0.98520 14.30820 0.51601
-0.70711 0.70711 1.00000 45.00000 0.70711 -0.72778 0.67917 0.99546 43.02160 0.68390
-0.25882 0.96593 1.00000 75.00000 1.93185 -0.31255 0.97057 1.01966 72.15020 1.63120
7 -1.00000 0.00000 1.00000 0.00000 7 -0.98335 0.00000 0.98335 0.00000
-0.90097 0.43388 1.00000 25.71430 0.55496 -0.89756 0.41069 0.98706 24.58720 0.54986
-0.62271 0.78245 1.00000 51.48570 0.80294 -0.65070 0.75798 0.99897 49.35490 0.76761
-0.22252 0.97493 1.00000 77.14290 2.24698 -0.27429 0.98524 1.02271 74.44310 1.86431
8 -0.98079 0.19509 1.00000 11.25000 0.50980 8 -0.96628 0.18372 0.98359 10.76510 0.50896
-0.83147 0.55557 1.00000 33.75000 0.60134 --0.83600 0.52926 0.98945 32.33720 0.59178
-0.55557 0.83147 1.00000 56.25000 0.89998 -0.58822 0.81141 1.00220 54.06020 0.85188
-0.19509 0.98079 1.00000 78.75000 2.56292 -0.24483 0.99562 1.02528 76.18490 2.09390
9 -1.00000 0.00000 1.00000 0.00000 9 -0.98250 0.00000 0.98250 0.00000
-0.93969 0.34202 1.00000 20.00000 0.53209 -0.93021 0.32331 0.98480 19.16590 0.52934
-0.76604 0.64279 1.00000 40.00000 0.65270 -0.77746 0.61608 0.99197 38.39430 0.63795
-0.50000 0.86603 1.00000 60.00000 1.00000 -0.53590 0.85034 1.00512 57.78010 0.93779
-0.17365 0.98481 1.00000 80.00000 2.87939 -0.22142 1.00334 1.02748 77.55510 2.32017
10 -0.98769 0.15643 1.00000 9.00000 0.50623 10 -0.97154 0.14743 0.98266 8.62893 0.50572
-0.89101 0.45399 1.00000 27.00000 0.56116 -0.88731 0.43102 0.98646 25.90860 0.55587
-0.70711 0.70711 1.00000 45.00000 0.70711 -0.72420 0.68154 0.99446 43.26170 0.68660
-0.45399 0.89101 1.00000 63.00000 1.10135 -0.49204 0.87947 1.00775 60.77410 1.02406
-0.15643 0.98769 1.00000 81.00000 3.19623 -0.20237 1.00930 1.02938 78.66240 2.54337

TABLE 3 TABLE 4
POLE LOCATIONS FOR m = 0.4 POLE LOCATIONS FOR m = 0.6
Angle from Angle from
Order Real Imaginary Magnitude neg. real Ct
Order CI
factor Real Imaginary Magnitude neg. real
axis (deg.) axis (deg.)
factor
2 -0.77715 0.62932 1.00000 39.00000 0.64338 2 -0.80902 0.58779 1.00000 36.00000 0.61803
3 -0.97622 0.00000 0.97622 0.00000 3 -0.96454 0.00000 0.96454
-0.60258 0.81318 1.01211
0.00000
53.46100 0.83981 -0.65189 0.78218 1.01822 50.19150 0.78098
4 -0.91747 0.33698 0.97740 20.16790 0.53266 4 -0.91363 0.31462 0.96629
-0.49363 0.89617 19.00180 0.52882
1.02312 61.15300 1.03633 -0.54872 0.87744 1.03489 57.97950 0.94300
5 -0.96990 0.00000 0.96990 0.00000 5 -0.95519 0.00000 0.95519 0.00000
-0.82884 0.52982 0.98371 32.58790 0.59342 -0.83734 0.50078 0.97566 30.88180
-0.42026 0.94279 1.03222 0.58260
65.97430 1.22806 -0.47673 0.93409 1.04871 62.96140 1.09989
6 -0.94334 0.22850 0.97062 13.61640 0.51446 6 -0.93203 0.21388 0.95625 12.92460
-0.74738 0.65067 0.51300
0.99093 41.04320 0.66294 ,0.76590 0.62165 0.98643 39.06480 0.64397
-0.36750 0.97258 1.03970 69.30030 1.41455 -0.42357 0.97184 1.06013 66.45050 1.25144
7 -0.96698 0.00000 0.96698 0.00000 7 -0.95088 0.00000 0.95088 0.00000
-0.89375 0.38787 0.97429 23.46010 0.54506 -0.88954 0.36543 0.96168 22.33300 0.54055
-0.67773 0.73250 0.99794 47.22410 0.73623 -0.70377 0.70606 0.99690 45.09330 0.70826
-0.32766 0.99328 1.04593 71.74330 1.59604 -0.38258 0.99892 1.06968 69.04350 1.39798
8 -0.95191 0.17265 0.96744 10.28030 0.50816 8 -0.93769 0.16189 0.95156 9.79540
-0.83984 0.50312 0.50740
0.97901 30.92440 0.58286 -0.84300 0.47717 0.96868 29.51160 0.57454
-0.62016 0.79008 1.00440 51.87040 0.80979 -0.65132 0.76748 1.00660 49.68070
-0.29645 1.00853 0.77274
1.05120 73.61980 1.77299 -0.34991 1.01938 1.07777 71.05470 1.54005
9 -0.96530 0.00000 0.96530 0.00000 9 -0.94841 0.00000 0.94841 0.00000
-0.92061 0.30503 0.96982 18.33180 0.52673 -0.91089 0.28716 0.95508 17.49770
-0.78804 0.58929 0.52426
0.98401 36.78860 0.62434 -0.79779 0.56242 0.97611 35.18290 0.61176
-0.57134 0.83319 1.01026 55.56030 0.88411 -0.60627 0.81458 1.01543
-0.27128 1.02026 53.34040 0.83744
1.05571 75.11010 1.94581 -0.32320 1.03545 1.08472 72.66520 1.67811
10 -0.95561 0.13869 0.96562 8.25786 0.50524 10 -0.93990 0.13020 0.94887
-0.88324 0.40843 7.88679 0.50477
0.97310 24.81720 0.55087 -0.87879 0.38623 0.95992 23.72580
-0.74041 0.65560 0.98895 0.54616
41.52340 0.66784 -0.75575 0.62934 0.98348 39.78510 0.65066
-0.52990 0.86636 1.01557 58.54810 0.95826 -0.56752 0.85167 1.02344
-0.25052 1.02959 1.05963 56.32220 0.90168
76.32480 2.11490 -0.30089 1.04844 1.09076 73.98720 1.81256

work. First -order transfer functions, on the other hand, and -order sections are cascaded. Odd -order filters are
are realized with a simple passive resistor -capacitor implemented by cascading a third -order section with
combination, so that an active third -order filter is ac- one or more second -order sections. For a seventh -order
tually an active second -order circuit cascaded with a filter, therefore, one third -order section is cascaded with
passive first -order RC network. two second -order sections.
To realize high -order filters, second -order and/or Different circuits can be used to realize the basic sec-
third -order filter sections are cascaded. If the filter order ond -order or third -order active filter section. Two popu-
is even, only second -order sections are. used. For ex- lar configurations, which require only one op amp per
ample, a sixth -order filter is obtained when three sec - section, are the positive -feedback and the multiple -

114
 TABLE 5 TABLE 6

POLE LOCATIONS FOR m = 0.8 POLE LOCATIONS FOR m = 1

Angle from 0
Angle from neg. real
Imaginary Magnitude neg. real
Q Order Real Imaginary Magnitude factor
Order Real factor axis (deg.)
axis (deg.)

33.00000 0.59618 2 -0.86603 0.50000 1.00000 30.00000 0.57735

2 -0.83867 0.54464 1.00000

0.00000 0.95300 0.00000 3 -0.94160 0.00000 0.94160 0.00000

3 -0.95300 0.71137 1.03054 43.65250 0.69105
-0.69963 0.74822 1.02436 46.92200 0.73207 -0.74564
4 -0.90476 0.27092 0.94445 16.66970 0.52193
4 -0.90939 0.29260 0.95531 17.83570 0.52524 0.80554
0.86753 -0.65721 0.83016 1.05882 51.63250
-0.60331 0.85544 1.04678 54.80600
5 -0.92644 0.00000 0.92644 0.00000
5 -0.94071 0.00000 0.94071 0.00000 0.56354
0.57265 -0.85155 0.44272 0.95976 27.46960
-0.84491 0.47173 0.96768 29.17570 1.08250 56.93570 0.91646
-0.53356 0.92225 1.06547 59.94860 0.99845 -0.59059 0.90720

6 -0.90939 0.18570 0.92816 11.54110 0.51032

6 -0.92071 0.19962 0.94210 12.23290 0.51162 0.61119
37.08630 0.62678 -0.79965 0.56217 0.97749 35.10790
-0.78333 0.59213 0.98195
-0.53855 0.96169 1.10222 60.75080 1.02331
-0.48063 0.96824 1.08097 63.60070 1.12454

0.00000 / -0.91949 0.00000 0.91949 0.00000

7 -0.93505 0.00000 0.93505
0.32167 0.93695 20.07870 0.53236
-0.88496 0.34336 0.94923 21.20590 0.53632 -0.88000
0.68325 -0.75274 0.65047 0.99485 40.83160 0.66082
-0.72878 0.67871 0.99588 42.96240 1.12626
-0.43895 1.00204 1.09397 66.34370 1.24611 -0.49669 1.00251 1.11881 63.64390

0.50668 8 -0.90968 0.14124 0.92058 8.82567 0.50599

8 -0.92361 0.15142 0.93595 9.31054
0.55961
-0.84549 0.45143 0.95846 28.09880 0.56681 -0.84733 0.42590 0.94834 26.68610
-0.68166 0.74367 1.00881 47.49090 0.73997 -0.71114 0.71865 1.01103 45.30110 0.71085
-0.40517 1.02805 1.10501 68.48960 1.36362 -0.46217 1.03439 1.13294 65.92450 1.22567

9 -0.93181 0.00000 0.93181 0.00000

g -0.91550 0.00000 0.91550 0.00000
0.52192 -0.89113 0.25266 0.92626 15.82950 0.51971
-0.90106 0.26971 0.94056 16.66360
0.58941
-0.80671 0.53551 0.96827 33.57720 0.60014 -0.81480 0.50858 0.96050 31.97150
0.79658 -0.67436 0.77305 1.02586 48.90070 0.76061
-0.64063 0.79453 1.02063 51.12050
1.32191
-0.37716 1.04877 1.11453 70.22020 1.47752 -0.43314 1.06007 1.14515 67.77530

7.51572 0.50433 -0.90913 0.11396 0.91625 7.14465 0.50391

10 -0.92441 0.12196 0.93242 11)
0.53755
-0.87399 0.36442 0.94692 22.63440 0.54172 -0.86885 0.34300 0.93410 21.54300
-0.78377 0.57591 0.97261 36.30850 0.62047
-0.77020 0.60276 0.97803 38.04680 0.63491
51.87030 0.80979
-0.60482 0.83542 1.03137 54.09630 0.85262 -0.64175 0.81758 1.03937
-0.40832 1.08127 1.15580 69.31190 1.41531
-0.35349 1.06572 1.12281 71.64960 1.58817

feedback active filters. The first uses the op amp in its Tables 7 and 8 give the normalized capacitor values
noninverting unity -gain mode, while the second uses it that satisfy these transfer functions when all the resist-
as an infinite -gain device. ors have the same value. Table 7 lists the capacitances
The cutoff frequency and the Q factor of the positive - for the positive -feedback Butterworth -Thomson filter
feedback filter are more sensitive to passive component for m = 0.2, 0.4, 0.6, and 0.8 for both even and odd fil-
tolerances and amplifier gain variations than the mul- ter orders from 2 to 10.
tiple -feedback filter. However, for the same cutoff fre- for the multiple -feedback Butterworth -Thomson filter.
quency, the positive -feedback version does not demand Interpolation yields the normalized capacitor values for
as much bandwidth from the amplifier. Additionally, intermediate values of parameter m.
both the second- and third -order positive -feedback fil- How to use the tables and graphs
ters require one fewer resistor than the same -order mul-
tiple -feedback filters. Now that all the data needed to put an active But-
terworth -Thomson filter together is at hand, a general
Low-pass Butterworth -Thomson filters design procedure can be outlined. Of course, the filter
A second -order low-pass filter section using a unity - characteristics that will satisfy the specifications must
gain amplifier and another using an infinite -gain ampli- first be selected. This is done by choosing the best value
fier are shown in Fig. 2a and 2b, respectively. The cir- of parameter m, as well as the necessary filter order,
cuits in Fig. 2c and 2d are the comparable third -order from Tables 1 through 6 and Graphs 1 through 4. (The
low-pass filter sections. desired filter operating frequency is set equal to the nor-
The transfer function of the second -order low-pass malized frequency of 1 radian/second.)
positive -feedback filter of Fig. 2a can be written as: If a low-pass function is required, the first steps are to
G(s) = 1 /[s2C1C212.112.2 + sC2(Ri + R2) + 1] .find the proper normalized capacitor values for each fil-
where s is the Laplace transform variable. For the sec- ter section from Table 7 or 8, depending on the circuit
ond -order low-pass multiple -feedback filter of Fig. 2b: preferred, and then choose a convenient value of resist-
G(s) = (-R3/R1)/[s2R2R3C1C2 ance. (A different resistance value may be assigned for
+ sC2(R2 + R3+ R2R3/Ri) + 1] each section.) Actual capacitor values (Ca) are found by
For the third -order positive -feedback filter of Fig. 2c: a simple division:
G(s) = 1 /[s3C1C2C3R1R2R3 Ca = Ca/coaR
+ s2(C2C3R3(Ri + R2) + CiC3Ri(R2 + R3)) where Ca is the normalized capacitor value, coo is the ac-
+ s(C3(Ri + R2 + R3) + CIRO + 1] tual filter operating frequency, and R is the resistance
For the third -order multiple -feedback filter of Fig. 2d: value chosen for a given section.
G(s) = 1 /[s3R1R2R3C1C2C3 High-pass filter functions synthesized with the posi-
+ s2((R2R3 + R3R4 + R2R4) tive -feedback circuit can also be designed from the nor-
(RiCiC3/R4)+ (RI +R2)R3C1C3) malized capacitor values given for low-pass filters in
+ s((R3R4 + R2R3 + R2R4 + RiR3 + R IRO Table 7. Since the positions of resistors and capacitors
(C3/R4)+ (RIR2Ci/R4)) + (RI R 2)/R4] cannot be interchanged for the multiple -feedback cir-

115
 ( ,I(APii )
likAI'll :,
GAIN FOR m= 0.4
GAIN FOR m = ILO
NORMALIZED FREQUENCY (RADIANS/s) -4. NORMALIZED FREQUENCY (RADIANS s) ---4.
0.1
10 0.1
10 0

- ----`..s,
0 0
Nk, L
ti
20
_
2 20
0.2

a 4
n= 2 \ -40 03
-1
Z -6 n 8 dillikial 1NIngillMI 60 z
.
;7
n=6
-
23.

ft
L.,
n=10 Iiii.- z
cn 6 n - 8 %
-60 co -8 lit
n. 10
,...s.
11

8
- -80
*1
10 .1 s -100
I -n -
10 , .. -100 -12 120

GRAPH 3
GRAPH 4
GROUP DELAY1FOR m = 0.4
GROUP DELAY FOR m =0.8
10
10

Ad\

5
5

4
4

^=2

02 04 06 0.8 1.2 14 16 18 0.2 04 06 08 1 12 14 16 18

NORMALIZED FREQUENCY (RADIANS/s)
NORMALIZED FREQUENCY (RADIANS/s)

cuit, Table 8 cannot be used for this high-pass filter. Fig. where ca. is the actual filter operating frequency.
3 shows the two active filter circuits that correspond to A few specific examples will clarify the general proce-
the low-pass filters of Fig. 2a and 2c. The second -order dures just outlined.
high-pass positive feedback filter is illustrated in Fig. A fourth -order multiple -feedback low-pass But-
3a, and the third -order version of this filter is drawn in terworth -Thomson filter is needed with m = 0.4 and
Fig. 3b. wo = 10,000 radians/second. From table 8, the norma-
Designing the filter lized capacitor values can be obtained. For the first fil-
ter section:
The design procedure for an active high-pass But- C1. = 2.8812 and C2. = 0.3633
terworth -Thomson filter is similar to the one for a low- For the second section:
pass filter. After the correct value for parameter m and C1. = 1.4147 and C2n = 0.6753
the desired filter order have been selected, the norma- Setting the resistors for both sections equal to 1,000
lized capacitor values are picked from Table 7. Norma- ohms, the denormalizing conversion factor becomes:
lized resistor values are then computed: wait = 104(103) = 107
R. = 1/C. The actual capacitor values can then be computed. For
A convenient value is next chosen for capacitor C; a dif- the first section:
ferent value of C may be chosen for each filter section. CI = 2.8812/107 = 0.28812 microfarad
The actual resistor values (Ra) can then be found from C2 = 0.3633/107 = 0.3633p.F
the normalized resistor values: And for the second section:
Ra = Rn/cooC C1 = 1.4147/107 = 0.14147 µF

116
 C2 = 0.6753/107 = 0.06753 itif comes high and filter Q large, the finite gain -bandwidth
The component values needed to build the filter are product of an op amp, as well as its sensitivity parame-
now known. ters, can no longer be neglected.
A fifth -order positive -feedback high-pass But- A common assumption is to consider the actual op
terworth -Thomson filter must be designed for m = 0.2 amp gain -bandwidth product, fa, to be at least an order
and coo = 1,000 radians/second. The normalized ca- of magnitude larger than the desired filter cutoff fre-
pacitor values are found in Table 7. For section 1: quency, fo. However, for a second -order positive -feed-
C1, = 1.3369, C2,, = 1.7467, and C3,, = 0.4276 back filter:
For section 2: fa/fo = - foa/fo) (1)
Clo = 2.7449 and C2,, = 0.3529 where Q is the desired filter Q, and foa is the actual filter
The normalized resistor values for section 1 become: cutoff frequency. This frequency ratio can also be ex-
R1,, = 1/1.3669 = 0.732 pressed solely in terms of filter Q:
R2,, = 1/ 1.7467 = 0.573 fa/fo = Q/(Qa/Q - 1) (2)
R3n = 1/0.4276 = 2.34 where Qa is the actual Q. The same relationships can be
And for section 2: developed for the second -order multiple feedback filter,
Rlo = 1/2.7449 = 0.365 which uses an infinite -gain as opposed to a unity -gain
R2 = 1/0.3529 = 2.84 amplifier:
Choosing a convenient value for capacitor C -let C = fa/fo = 1.5Q/(1 - foa/fo) = 1.5Q/(Qa/Q - 1)
0.1µF = 10-7 farads -allows the denormalizing conver- Suppose that the actual cutoff frequency, foa, must
sion factor for the resistances to be calculated: deviate less than 5% from the desired cutoff frequency,
woC = 103(10-7) = 10-4 fo, and that actual filter Q, Qa, must be within 2% of the
The actual resistor values for section 1 can then be read- desired filter Q. Using Eq. 1 yields:
ily determined: fa/fo must be greater than Q/0.05 = 20Q
R1 = 0.732/10-4 = 7.32 kilohms or, from Eq. 2:
R2 = 0.573/10-4 = 5.73 kilohms fa/fo must be greater than Q/0.02 = 50Q
R3 = 2.34/10-4 = 23.4 kilohms Since the latter condition is the more stringent, it is the
And for section 2: one that must be obeyed.
R1 = 0.365/10-4 = 3.65 kilohms If a fourth -order filter, with m = 0.2 and fo = 1,000
R2 = 2.84/10-4 = 28.4 kilohms hertz, is being designed, Table 2 indicates that Q =
The filter can now be assembled from the computed 0.53678 for section 1 and Q = 1.15409 for section 2.
component values. The gain -bandwidth product of the amplifier for the
second section must satisfy the condition:
Practical considerations fa must be greater than 50Qfo = 5(1.15409)(103)
For the most part, literature on active filter design fa must be greater than 57.7045 kilohertz
supposes the op amp to be ideal -that is, it is assumed to achieve less than 2% accuracy in the Q of the filter. A
to have infinite gain, infinite input impedance, and zero multiple -feedback filter would require even more am-
output impedance. But when operating frequency be- plifier bandwidth, because:

2. Low-pass filters. Second -order low-pass filter can be implemented with unity -gain amplifier in positive -feedback circuit (a), or with infinite -
gain amplifier in multiple -feedback circuit (b). Third -order filters (c) and (d) are second -order filters that have been cascaded with first -order
RC network at input end. Tables 7 and 8 give normalized capacitor values for these low-pass filters having orders up to 10.

117
 TABLE 7

NORMALIZED CAPACITOR VALUES USING UNITY -GAIN AMPLIFIERS

m = 0.2 I m = 0.4 I m = 0.6 m = 0.8
Order Section 1
I

C(1) C(2) C(3) I C(1) C(2) C(3) I C(1) C(2) C(3) I C(1) C(2) C(3)
2 1 1.3456 0.7431 I 1.2868 0.7771 I 1.2361 0.8090 I 1.1924 0.8387
3 1 1.3950 3.2775 0.2190 I 1.3963 3.2851 0.2185 I 1.3963 3.2853 0.2185 I 1.3951 3.2779 0.2189
4 1 1.0859 0.9422 1.0900 0.9604 1.0945 0.9785 1.0996 0.9965
2 2.2819 0.4283 2.0258 0.4716 1.8224 0.5123 1.6575 0.5506
5 1.3669 1.7467 0.4276 1.3796 1.7651 0.4234 I 1.3922 1.7841 0.4192 1.4048 1.8036 0.4150
2 7449 0 35291 2 37% 0.3944 2.0976 0.4335 1.8742 0.4700
6 1 1.0475 0.9835 1.0601 1.0013 1.0729 1.0193 1.0861 1.0374
2 1.3741 0.7344 1.3380 0.7611 1.3057 0.7871 1.2766 0.8124
3 3.1995 0.3006 2.7211 0.3400 2.3609 0.3769 2.0806 0.4113
7 1 1.3546 1.5423 0.4920 1.3726 1.5601 0.4860 1.3909 1.5791 0.4799 1.4094 1.5992 0.4738
2 1.5368 0.6520 1.4755 0.6805 1.4209 0.7081 1.3722 0.7348
3 3.6458 11_2622 3 0519 0.2995 2.6138 0.3344 2.2782 0.3668
8 1 1.0349 0.9988 I 1.0505 1.0171 1.0664 1.0356 1.0827 1.0544
2 1.1962 0.8539 1.1907 0.8762 1.1862 0.8984 1.1827 0.9204
3 1.7000 0.5856 1.6125 0.6147 1.5353 0.6428 1.4670 0.6698
4 4.0845 0.2329 3.3733 0.2683 2.8578 0.3012 2.4681 0.3318
9 1 1.3487 1.4715 0.5197 1.3696 1.4889 0.5127 1.3908 1.5077 0.5056 1.4124 1.5279 0.4984
2 1.2862 0.7901 1.2690 0.8139 1.2535 0.8373 1.2396 0.8604
3 1.8660 0.5305 1.7503 0.5598 1.6494 0.5880 1.5610 0.6150
4 4.5162 0.2097 3.6863 0.2434 3.0941 0.2747 2.6514 0.3036
10 1 1.0293 1.0061 1.0465 1.0249 1.0639 1.0439 1.0818 1.0633
2 1.1270 0.9118 1.1322 0.9327 1.1379 0.9537 1.1442 0.9747
3 1.3808 0.7323 1.3506 0.7570 1.3232 0.7814 1.2984 0.8052
4 2.0324 0.4845 1.8871 0.5138 1.7621 0.5418 1.6534 0.5686
5 4.9415 0.1910 3.9918 0.2231 3.3235 0.2529 2.8289 0.2804

TABLE 8

NORMALIZED CAPACITOR VALUES USING INFINITE -GAIN AMPLIFIERS

m = 0.2 I m = 0.4 I m = 0.6 m = 0.8
Order Section I

C(1) C(2) C(3) I C(1) C(2) C(3) I C(1) C(2) C(3) I C(1) C(2) C(3)
2 1 2 2294 0 4485 I 2 3314 0.4289 2.4271 0.4120 I 2.5160 0.3975
3 1 1_3495 5.2591 0.1405 I 1.3423 5.1960 0.1419 1.3337 5.1220 0.1436 I 1.3236 5.0383 0.1457
4 1 2.8266 0.3620 2.8812 0.3633 2.9355 0.3648 I 2.9894 0.3665
2 1.2850 0.7606 1.4147 0.6753 1.5370 0.6075 1.6518 0.5525
5 1 1.2640 2.9081 0.2771 1.2713 2.9226 0.2757 1.2785 2.9372 0.2742 1.2857 2.9520 0.2728
2 1.0588 0.9150 1.1833 0.7932 1.3004 0.6992 1.4100 0.6247
6 1 2.9506 0.3492 3.0039 0.3534 3.0578 0.3576 3.1121 0.3620
2 2.2033 0.4580 2.2833 0.4460 2.3613 0.4352 2.4372 0.4255
'1 0 90111 1.0665 1.0199 0.9070 1.1306 0.7870 1.2340 0.6935
7 1 1.2297 2.7070 0.3202 1.2436 2.6252 0.3174 1.2578 2.6444 0.3146 1.2722 2.6647 0.3117
2 1.9561 0.5123 2.0416 0.4918 2.1244 0.4736 2.2045 0.4574
3 0.7867 1.2153 0.8986 1.0173 1.0031 0.8713 1.1004 0.7594
8 1 2.9964 0.3450 3.0512 0.3502 3.1067 0.3555 3.1631 0.3609
2 2.5618 0.3987 2.6287 0.3969 2.6952 0.3954 2.7611 0.3942
3 1.7569 0.5667 1.8442 0.5375 1.9284 0.5118 2.0094 0.4890
d fl AQ87 1 1E1 c n 90411 1 1944 0 9037 0.9526 0.9955 0.8227
9 1 1.2138 2.5053 0.3388 1.2311 2.5240 0.3353 1.2487 2.5440 0.3318 1.2667 2.5654 0.3282
2 2.3703 0.4287 , 2.4416 0.4230 2.5120 0.4178 2.5813 0.4132
3 1.5914 0.6220 1.6794 0.5834 1.7640 0.5498 1.8450 0.5203
4 0 6792 1 5054 0.7302 1.2288 0.8241 1.0314 0.9109 0.8838
10 1 3.0184 0.3431 3.0746 0.3488 3.1317 0.3546 3.1898 0.3606
2 2.7355 0.3757 2.7982 0.3774 2.8611 0.3793 2.9241 0.3814
3 2.1969 0.4603 2.2711 0.4502 2.3441 0.4411 2.4156 0.4328
4 1.4535 0.6775 1.5414 0.6290 1.6255 0.5874 1.7058 0.5511
5 0.5729 1.6472 0.6693 1.3306 0.7587 1.1078 0.8412 0.9430

must be greater than 1.5(50)Qfo

fa
must be greater than 1.5(50)(1.15409)(103)
fa
must be greater than 86.5568 kHz
fa
The infinite -gain amplifier, however, has lower sensi-
tivity parameters than the unity -gain amplifier. 0
BIBLIOGRAPHY
Farouk Al -Nasser, Tables Speed Design of Low -Pass Active Filters," EDN, March 15,
1971.
Aram Budak, Active Network Synthesis," class notes, Colorado State University, 1971.
A. Budak and D. Petrela, Frequency Limitations of Active Fitters Using Operational Ampli-
fiers, IEEE Transactions on Circuit Theory, Vol. CT -19, No. 4, July, 1972.
R. Golden and J. Kaiser, Root and Delay Parameters for Normalized Bessel and But-
terworth Low -Pass Transfer Functions, IEEE Transactions on Audio and Electrocoustics,
Vol. AU -19, No. 1, March, 1971.
Analysis and Synthesis of Transitional Butterworth -Thomson Fitters and Bandpass Ampli-
fiers," RCA Review, Vol. 18, March, 1957. pp. 60-94.

3. High-pass filters. Circuits (a) and (b) correspond to low-pass

configurations on Fig. 2a and 2c, but provide high-pass function. Re-
ciprocals of normalized capacitor values in Table 7 are used to de-
termine resistor values for these high-pass filters.

118
Instrumentation amplifier
conditions computer inputs
Two -stage amplifier detects more than 100 multiplexed signals
at computer -speed scanning rates with minimum noise over wide
frequency range; signal levels can range from microvolts to volts

by Thomas C. Lyerly, IBM General Systems Division, Boca Raton, Fla.

Since the job of data gathering has become so com- computer analog input subsystem, such as the typical
plex that computers are needed to process measurement one shown in Fig. 1 for the IBM System/7 computer. As
information, increasingly stringent performance re- many as 128 analog input signals may be connected
quirements are being placed on the instrumentation through single -pole low-pass filters to the multiplexer,
amplifier. A single instrumentation amplifier must which is made up of p -channel MOSFET switches. The
frequently accept more than 100 multiplexed signals multiplexer can handle up to eight parallel groups of
with levels ranging from a few microvolts to several signals, with each group consisting of up to 16 parallel
volts. The amplifier must also reject unwanted noise sig- channels.
nals from dc to hundreds of megahertz, as well as pro- The differential output of the multiplexer is con-
vide a noise -free, properly scaled, single -ended output nected to an instrumentation amplifier through a balun
signal that can be sampled quickly for processing. transformer that serves as a high -frequency noise filter
Precision low-level signal conditioning can be for the amplifier. Multiplexed analog input signals must
achieved with a two -stage amplifier that employs cas- be accurately processed by the amplifer at scanning
code transistor pairs, degenerative feedback, and bal- rates as fast as 20 kilohertz without any adverse effects
anced frequency compensation to provide exceptional from noise sources that are both external and internal to
gain -bandwidth performance without gain -adjustment the analog input subsystem.
potentiometers. Even in noisy environments, this ampli- External noise may be separated into two types-
fier can detect microvolt signals at high scanning rates. common-mode, and normal- or differential -mode sig-
It can also be adapted for programable gain selection. nals. Differential -mode noise is the easier to reject. It
The environment for the precision amplifier is the can be filtered by passive low-pass networks located at

LOW-PASS
NORMAL -MODE INSTRUMENTATION
FILTER MULTIPLEXER AMPLIFIER
BALUN

ANALOG
INPUT -AA/V-4 it
OUTPUT

COMMON -
MODE
INPUT
,s,
CHANNEL
DRIVER
15 UP TO HIGH -FREQUENCY
DIGITAL PARALLEL 7 GROUPS OF COMMON -MODE
INPUTS
CHANNELS 16 CHANNELS FILTER
GROUP
DRIVER

1. The environment. Analog input computer subsystem challenges the instrumentation amplifier. Data is filtered before being multiplexed to
eliminate normal- or differential -mode noise. Balun transformer acts as high -frequency common -mode noise filter for the amplifier.

119
 characteristics and can be adapted for programable
INVERTING
INPUT I>A1
R1 OUTPUT gain selection.
Amplifiers Al and A2 provide the desired input char-
acteristics and signal ranging. A third amplifier, A3, acts
Rf as an inverting unity -gain device for differential inputs
and operates as a precision broad -bandwidth subtractor
\2 for unwanted common -mode signals. To reject com-
mon -mode signals over a large bandwidth. Al and A2
must perform as broadband unity -gain amplifiers, gen-
L__
erating balanced common -mode signals for subtractor
amplifier A3.
GAIN The common -mode rejection ratio of the over-all am-
SELECTION
SWITCHES
Poi plifier can be expressed as:
3 CMRR = (Itr/R1)(1-1-2Rm/R0/[(1/CMRR4-1/CMRR2)
(1-1-2Rmiltn)(Rr/Ri) + (Rf/CMRR3(111+R1)
+ Rf/(Ri+Rf) - Re/(Ri'+111'))/
R1'011' + Rol
where CMRR1, CMRR2, and cmRR3 are the individual
common -mode rejection ratios of amplifiers A1, A2, and
A3, respectively.
Other common -mode input errors, not included in
this equation, are caused by the voltage -divider effect
NON -
INVERTING
"s between the source impedance of each analog input line
R', Rf _L
INPUT -r and the common -mode input impedance of its ampli-
fier. To prevent common -mode signals from being con-
2. Basic amplifier design. Double -amplifier input eases demands verted into differential signals from source imbalance at
on individual amplifiers Al and A2. Amplifier A3 rejects unwanted the input of the over-all amplifier, amplifiers Al and A2
common -mode noise signals by subtracting balanced common - must maintain a high common -mode input impedance
mode signals from A1 and A2. over a large bandwidth.
The high impedance of these two amplifiers also
the multiplexer inputs, as shown in Fig. 1, since a given helps the over-all amplifier to reject the transients gen-
channel is generally sampled at a rate that is much erated when the multiplexer channels are switched.
slower than the speed at which the amplifier scans the These transients are caused by common -mode voltages,
channels. sometimes as high as 10 volts, that are switched into the
Common -mode noise is more difficult to reject be- instrumentation amplifier inputs at rates that can ap-
cause low -frequency common -mode filtering reduces proach several hundred volts per microsecond.
the multiplexer's ac common -mode rejection ratio,
while increasing its common -mode settling time. For The first stage
the analog input subsystem to operate at fast multi- A first -stage design that satisfies the tough input per-
plexer -scanning rates, the instrumentation amplifier formance requirements is illustrated in Fig. 3. Basically,
must be able to reject common -mode noise over a large the circuit attempts to eliminate amplifier input errors
bandwidth. Therefore, a balun transformer is placed at caused by offset voltage, bias current, offset -voltage
the amplifier input as a high -frequency common -mode drift, offset -current drift, and input impedance.
filter to reject the common -mode noise beyond the Input offset voltage can be reduced to less than 2 mi-
bandwidth of a practical amplifier. crovolts with the nonlinear potentiometer network of re-
(This noise is filtered to prevent the amplifier from sistors R1, R2, and R3 and the voltage divider set up by
rectifying rf noise on the input lines and effectively pro- resistors Ri and R. The input -offset voltage drift with
ducing dc offset errors. The balun transformer of Fig. 1 temperature is also low because the difference between
is a low-pass common -mode filter with a bandwidth of the base -emitter voltage drops of matched input transis-
approximately 5 megahertz.) tors Qi and Q2 is nearly zero.
Noise within the analog input subsystem, which takes Additionally, input bias current can be adjusted to
the form of capacitively coupled noise signals and less than 1 nanoampere by parallel potentiometers R6
switching transients, is primarily caused by the multi- and R7, which are driven by the high -impedance con-
plexer because of the device capacitance of the MOSFET stant -current source, consisting of cascode transistors Q3
switches. This internal noise source appears to the am- and Q4 and resistor Rs. The potentiometers are also
plifier to be primarily common -mode noise as a result of driven by the common -mode input voltage through a
the balanced design of the multiplexer and the low-pass feedback network in the emitter circuit of the Q1 -Q2 in-
input filter. put transistor pair. The network, made up of transistors
Configuration outline Q6, Q6, and Q7 and resistor Rs, supplies a constant volt-
age for biasing resistors R10 and R11 that is independent
An instrumentation amplifier that provides excellent of the common -mode input voltage.
rejection of unwanted common -mode noise is sketched The input -impedance component of the amplifier
in Fig. 2. The circuit also offers good gain -bandwidth from the feedback bias network is several thousand

120
megohms. Therefore, this feedback circuit provides a pedance node A and the value of resistor RE.
stable bias current without reducing the input imped- When the first- and second -stage designs are cas-
ance of the amplifier. A heated -substrate matched -input caded, they form a differential amplifier with a stable
transistor pair can further improve the bias current, the and well-defined open -loop dc voltage gain that can ex-
offset -current drift, and the offset -voltage drift of the ceed 140 decibels. Since the open -loop gain is very
amplifier. large, the errors in the closed -loop gain caused by a fi-
To increase the amplifier's common -mode input im- nite open -loop gain are reduced to a negligible level.
pedance, FETS Q8 and Q9 are connected in a cascode Therefore, gain -adjustment potentiometers are not re-
configuration to feed the common -mode emitter voltage quired, arid the closed -loop gain of the instrumentation
of transistors Qi and Q2 back to their collectors. The amplifier can be established by resistive networks alone.
feedback reduces the effective input capacitance, but in- To satisfy the fast sampling -rate requirements of the
creases the effective input resistance between the base analog -input subsystem, the settling time of the ampli-
and collector of both input transistors. The impedance fier must be minimized. Settling time is determined by
in the emitter loop of the Q1 -Q2 input pair is also in- both bandwidth and slew rate, which is a measure of the
creased by a cascode pair, Qio and Qii, along with re- rate at which the amplifier output can change.
sistor R12, in the emitter -current source. Slew -rate limiting
The first -stage design can provide a common -mode
input resistance of more than 2,000 megohms and an in- Slew rate is limited in feedback amplifiers when the
put capacitance of no more than a few tenths of a pi- maximum output rate of change that is fed back to the
cofarad. It also offers a good common -mode rejection inverting input is less than the rate of change at the
ratio, as well as power -supply rejection ratio because of noninverting input. Under these conditions, a large dif-
the high drain output impedance of the FET cascode cir- ferential error is produced across the amplifier inputs.
cuits. (Voltages V1 and V2 are zener-stabilized levels.) The differential input impedance drops while the ampli-
fier is slew -rate limiting and may be further reduced if
The second stage the input step voltage exceeds the amplifier differential
Performance requirements for the second stage are breakdown voltage.
determined primarily by the programable signal rang- The low input impedance that occurs during slew rate
ing and gain -bandwidth requirements of the over-all in- limiting can produce large errors if low-pass filters with
strumentation amplifier. An ideal amplifier has an infi- long recovery time constants are associated with the
nite open -loop gain so that its closed -loop voltage gain multiplexer, because charge is transferred from the fil-
is determined only by external resistive networks. How- ter capacitors to the amplifier inputs. To minimize this
ever, a practical amplifier's gain accuracy decreases for error, the amplifier must have a fast slew rate, as well as
increasing values of closed -loop gain because it has a fi-
nite open -loop gain. The closed -loop gain for the prac-
tical amplifier of Fig. 2 can be written as:
CLG = (1 +2R,./Rri)(Rf /R1)[1 /(1+ 1/A)
(1 + 2Rm/Ra][1/(1 + (Ri + Rf)/ ARO]
where A represents open -loop gain:
A=Al=A2=A3
Instrumentation amplifiers with large closed -loop
gains usually require gain -adjustment potentiometers if
a high degree of gain accuracy must be maintained.
However, a second -stage design with sufficient gain can
eliminate the need for potentiometers without degrad-
ing gain accuracy. A schematic of a high -gain second
stage is shown in Fig. 4.
The dc gain of the stage is directly proportional to the
impedance at the drain terminal of FET Qi. A high -im-
pedance node is developed at point A because the Qi-
Q2 cascode pair drives the Q3 -Q4 cascode pair and FET
Q5 is connected as a source -follower. The dc resistance
from node A to ground may be as high as 500 meg-
ohms, depending on the quiescent current of the cas-
code pairs and the resistance in the emitter circuits of
transistors Q2 and Q.
Cascading the stages
The circuit permits voltage gains in excess of several 3. First stage. Differential input amplifier stage almost zeroes out er-
million to be achieved with a single stage. To make the rors caused by offset voltage, offset -voltage drift, bias current, off-
gain of the stage less sensitive to changes in transistor set -current drift, and input impedance. The circuit, which has
parameters, emitter degeneration is provided by the re- matched input transistors, Q1 and 02, offers a high common -mode
sistors labeled RE so that the voltage gain is approxi- input resistance, but it holds input capacitance to a fraction of a. pi-
mately equal to the ratio of the resistance of high -im- cofarad. Cascode transistor pairs act as constant -current sources.

121
 140
-\ \ -6 dB/OCTAVE

120
\ -12 dB/OCTAVE

100
\ OPEN -LOOP RESPONSE
\ V FOR GAIN OF 512
\
80
OPEN -LOOP RESPONSE
AMPLIFIER CLOSED -LOOP
RESPONSE FOR GAIN
\ FOR UNITY GAIN

60 OF 512
-6 dB/OCTAVE

40

20 AMPLIFIER CLOSED -LOOP

RESPONSE FOR UNITY GAIN

10 102 10' 104 106 106

FREQUENCY (Hz)

4. Second stage. Since this stage can attain voltage gains of several 5. Response curves. Cascaded first and second stages result in
million without losing gain accuracy, gain -adjustment potentiometers amplifier that maintains its unity -gain bandwidth of almost 10 mega-
are not needed. Circuit gain approximately equals the resistance ra- hertz over wide range of closed -loop gains. For instance, the upper
tio of high -impedance node A and resistor RE. frequency breakpoint is 125 kilohertz for a closed -loop gain of 512.

a relatively high open -loop input impedance. nent of high -impedance node A of the second stage es-
In a practical amplifier, one that must operate over a tablish the first break point of the amplifier. Since the
wide range of closed -loop resistive component
sponse time is achieved for large closed -loop gains by occurs at a relatively low frequency with only a small
extending the natural break frequencies of the open - value of capacitance, thereby permitting slew rates of
loop amplifier to obtain the largest possible gain -band- several hundred volts per microsecond to be realized.
width product. Amplifier capacitance, which determines The second break frequency, at around 4 kHz, is de-
bandwidth, can then be kept small, and the slew rate, termined by time constants R13C1 and R14C2 in the first
which is also fixed by amplifier capacitance, can be held stage. The resulting two -pole response reduces the
at a large value. open -loop gain at a rate of 12 dB/octave, which in-
A high open -loop gain presents a difficult frequency - creases slew rate and the high -frequency loop gain more
compensation problem when a fast unity -gain closed- than the 6-dB/octave rate.
loop response is needed. For example, suppose the The unity -gain bandwidth can be determined by a
same bandwidth must be available from a single -pole 6- lag -lead network that is digitally selected at the high im-
dB/octave amplifier for unity -gain operation and for a pedance node of the second stage, as shown in Fig. 4.
closed -loop gain of 512. To do this, the amplifier's open - This type of unity -gain compensation has the advantage
loop gain must be reduced by a factor of 512 at the de- of a 12-dB/octave rolloff. Unlike the compensation for a
sired bandwidth by increasing the internal compensa- gain of 512, the unity -gain compensation increases
tion capacitance by a factor of 512. The increase in ca- bandwidth while decreasing slew rate so that the settl-
pacitance decreases the slew rate by a factor of 512 and ing times for the two gain settings are approximately
degrades the unity -gain time response. One solution to the same. Similarly, opposing compensation networks
the problem is to select the appropriate frequency -com- can be traded off for gain values between unity and 512.
pensation components digitally so that the settling time Two additional lead -compensation networks in the
for each gain setting is optimized. first stage (time constants R15C1 and R16C2) and one in
The simplified Bode plots of Fig. 5 show the individ- the second stage (time constant 2REC4) complete the
ual frequency responses of the over-all amplifier, along over-all amplifier's internal frequency compensation.
with the closed -loop responses for unity gain and a gain These three lead networks cancel poles in those cascode
of 512. The bandwidth for a closed -loop gain of 512 is transistor pairs that have two natural poles.
determined primarily for first -stage capacitors C1 and The gain degeneration employed throughout the am-
C2 and second -stage capacitor C1, all of which assume plifier design stabilizes the lag and lead compensation
small values. The 125 -kilohertz upper frequency limit is networks, and provides an over-all gain -bandwidth per-
determined by the RC networks of Fig. 2. formance that is determined by passive components, in-
The capacitive component and the resistive compo- stead of sensitive semiconductor parameters. 0

122
Dynamic zero -correction method
suppresses offset error in op amps
In data -acquisition systems. the offset voltage and offset voltage
temperature drift of a FET-input op amp can be held to only
a few microvolts by a sample -and -hold correction technique

by Richard C. Jaeger and George A. Hellwarth, IBM General Systems Division, Boca Raton, Fla

0 Although today's monolithic FET-input operational nously between the cycles of a multiplexer or an analog -
amplifier offers the advantages of high input imped- to -digital converter so that the switches and demodula-
ance, large open -loop gain, fast slew rate, and low in- tion filter become a measure -and -hold circuit. After
put -bias current, it often has a high initial offset voltage measuring the magnitude of the input offset voltage
that drifts with time and with changing temperature. (with the amplifier input, shorted by a switch), the cir-
However, certain dynamic zero -correction techniques cuit holds the correction voltage inserted at the ampli-
can drive the initial offset voltage to zero and reduce the fier input. This technique is sometimes hampered by er-
effect of initial offset current when the amplifier is used rors in the sample -and -hold circuit and by feedback
for signal conditioning in a multiplexed or sampled instability during the correction cycle.
data -acquisition system. One such scheme keeps the
Zeroing out offset error
amplifier's input voltage to only a few microvolts and its
offset voltage temperature drift to merely a few hun- In the improved zero -correction scheme (Fig. 1), the
dredths of a microvolt per degree Celsius. amplifier is also driven by a multiplexer, and the offset
This method overcomes the shortcomings of previous voltage is dynamically eliminated between multiplexer
dynamic zero -correction techniques. One method, for cycles. A set of switches, which are synchronously
example, controls drift by inserting a dc correction sig- linked to the multiplexer timing control, change the am-
nal into the amplifier's input with a periodically oper- plifier's operating mode to eliminate offset error. mos-
ated switch or modulator, an auxiliary ac -coupled am- FETS are usually used as the switches because of their
plifier, and a demodulator switch and filter. This operating speed and predictable switching action.
scheme is not only expensive, but it produces carrier - Zero correction is implemented by disconnecting and
frequency noise from the switches and recovers slowly grounding both inputs of amplifier A1, allowing the for-
from an overload condition. ward gain of the over-all amplifier to generate a large
Recent methods require switches operating synchro- output voltage that is fed back to the input of interstage

1. Dynamic zero -correction. Offset voltage of over-all amplifier is held to a few microvolts by inserting correction voltage at input of
amplifier A2. By applying correction voltage at interstage between Aland A2, errors due to sample -and -hold inaccuracies can be minimized.
Sample -and -hold circuit stores correction voltage while over-all amplifier has its feedback loop open and is disconnected from source.

123
 +15 V

740 -TYPE
OUTPUT

TO GATE
DRIVE A

10 20 30 40

TEMPERATURE (DC)

3. Almost -zero drift. Offset voltage of zero -corrected 740 -type op amp
TO GATE DRIVE A is initially 3µV and has temperature drift of under 0.05 µV7°C.
- TO GATE -
DRIVE B
+15 V
GATE expressed in terms of Ai's offset and A2's gain:
GATE
DRIVE A Vos = Vosi/(1 + GA2)
DRIVE B
where Vos1 is the equivalent amplifier input offset volt-
10 Id/
age without correction, and GA2 is the open -loop gain of
A2. The correction voltage stored on capacitor Ci is:
GATE Vc1 GAiGA2Vos1/(1 + GA2) + V053
DRIVER
where V083 is the offset voltage of unity -gain amplifier
A3, and GA1 is the open -loop gain of amplifier Al.
-15 V Because the correction voltage is applied after the
p -CHANNEL MOSFETs, FLOATING SUBSTRATES
first stage of the over-all amplifier, the effect of errors
caused by sample -and -hold inaccuracies is significantly
reduced. Resistor Reg is made equal to the parallel com-
bination of resistors R1 and R2 to correct for the error
0 caused by bias current Liu, which flows through these
resistors during normal amplifier operation. Similarly,
the error due to bias current 1B2 is compensated for by
lam.11N11
resistor R3, which has the same value as resistor Rs.
Getting practical
0
4, Figure 2 shows a practical zero -correction scheme for
ERROR -STORAGE
(2 m /DIV)
the popular FET-input 740 -type op amp. The compo-
MODE AMPLIFICATION
MODE
nents are numbered to correspond with the labels of
Fig. 1. Amplifiers Al and A2 of Fig. 1 become the op -
2. Practical circuit. Zero -correction scheme for popular 740 -type amp's input and output stages, respectively.
op amp employs MOSFET switches to cycle op amp between error The sample -and -hold circuit, which is formed by
storage and amplification modes. Sample -and -hold circuit, which is switch S5, transistors Q1 and Q2, and capacitor C1, in-
composed of switch S5, transistors 01 and 02, and capacitor C1, in- jects a correction -signal current, instead of a correction -
serts correction current, rather than correction voltage, into an inter- signal voltage as shown in Fig. 1, into the interstage off-
mediate stage of op amp. Scope trace shows op -amp output. set adjustment port of the op amp. Resistor Re is in-
cluded to stabilize the negative feedback of the correc-
amplifier A2. The feedback voltage is sampled and held tion loop during error storage.
while amplifier Ai is returned to normal operation by When switch S3 opens, the correction voltage on ca-
reconnecting its input to the signal source. pacitor C1 is disturbed because the circuit's gate -drive
The over-all amplifier is cycled between two modes of signal is coupled through the gate -source capacitance of
operation-amplification and error storage. The error the MOSFET that is being used as the switch. Capacitor
storage cycle begins when switches S1 and S2 are C2, which is returned to the gate -drive signal having op-
opened and switches S3, S4, and S5 are closed. This posite polarity, neutralizes this error. The scope trace
causes Ai to amplify its own offset voltage and the depicts the circuit's output as the op amp cycles be-
sampling circuit to derive the correction voltage. tween the error -storage mode and the amplification
Opening switches S3, S4, and S5 and closing switches mode, where the source voltage is zero.
Si and S2 start the amplification cycle by driving Ai A plot (Fig. 3) of offset voltage (referred to the op
with the input -signal source and closing the over-all amp's input) as a function of temperature illustrates the
amplifier's normal feedback loop. The correction volt- remarkably low drift this zero -correction scheme can at-
age remains applied to amplifier A2 so that Ai's offset is tain. The circuit brings the initial offset voltage down to
reduced by an amount determined by Az's gain. only a few microvolts and holds offset voltage tempera-
The zero -corrected amplifier offset, therefore, can be ture drift to less than 0.05 µv/ °C. 0

124
Programs in Getel speak
test engineer's language
English -based language and translator can be applied to automatic
test equipment, making it easy for engineers and technicians
to write in new routines and check out program parameters

by Gene Kierce, General Electric Co., Schenectady, N.Y.

BecaUse each computer -controlled test system must named Mark III), it is not actually in use. Rather, a
be addressed in its own language, only the operators smaller translator, the Mark I, is in regular use, since it
have had the opportunity to become fluent in those lan- can fit into a minicomputer with an 8,192 -word
guages. As a result, many engineers with product qual- memory.
ity responsibilities have not been able to easily check The Mark I can translate from Getel into any lan-
what parameters actually were being measured. Such gauge that the user desires. In the case of the GE -PAC
difficulties have even spread to the equipment trouble- 30 -GE -MAC test system, the Getel compiler translates
shooters, who had to figure out what the test equipment test statements directly into GE -PAC 30 assembly code.
was testing before they could even begin the task of A medium -size translator, the Mark II, has also been
finding out what was wrong with the equipment under developed. It can fit into a 16,384 -word machine, and is
test. mounted in a computer, but has not yet been used at GE
To solve such problems, test equipment engineers because the Mark I has been able to serve all the users'
from 15 different departments at General Electric Co. needs to this time.
joined forces to develop a single language and a trans- Filling the blanks. The Mark I's translation from Ge-
lator that applies it to any piece of automatic test gear. tel into a tester's language is performed with a tem-
language, Getel (for general plate -type approach. The Getel statement is considered
test engineer language, pronounced jee-tell), is based on to have a fixed portion that corresponds directly with a
English, so that an operator needs little formal training statement in the translated language, and a variable
and almost anyone else can understand what's actually portion, which contains the specific numbers relating to
being tested. the test at hand. For example in the statement
Experience with the language has shown that tech- 100 SET 11' TO 50 OHMS
nicians and engineers can learn to read and understand the value for the load, 50, is a variable. The remainder
a Getel program in anywhere from a few minutes to a of the statement is fixed and thus, if the variable part is
couple of hours and can be programing proficiently in deleted, the statement appears with a hole in this place
less than a week. The use of English also allows even a (denoted by an asterisk) and thus can be considered to
typist to correct many spelling and programing errors, be a template.
and troubleshooters can use the source listing as a 100 SET 'LP TO * OHMS
checkout routine. Such templates are part of the automatic test equip-
Getel is now in wide use in many General Electric ment (ATE) conversion table, which is composed by a
plants, furnishing inputs to a wide range of systems test engineer for each particular automatic test system.
such as the GE -MAC tester controlled by, a GE -PAC 30 The ATE table thus relates Getel to a particular test sys-
controller, the Hewlett-Packard rf tester, the tape - tem, requiring only the specific numbers to fill the holes
driven Systomation Fixit tester for checking wiring, and and thus complete an instruction to the automatic test
the tape -driven Texas Instruments T1553 integrated -cir- system in its own language. For example, the corre-
cuit tester. sponding statement in GE -PAC 30 assembly code for the
Mini -translator. Although English is the most conve- above term, as might be required for the GE -MAC test
nient language for an operator, it is difficult to translate systems would be
into a useful format for controlling machines. Origi- LHI RE,2
nally, the plan was to develop a large central software BAL RF, MONITR
translator to process Getel. The translator was to pro- B* +8
duce an intermediate language that could be processed DC A(Z99999)
*
by a smaller machine, which in turn would control a *
*
specific instrument. However, the language and the
translator became larger and larger as the development Z99999 DC 50
went on, so that only the largest computers could be The statement in Getel is called a source statement
used to process the translation program. Although this and the translated form is called the object statement.
software translator has actually been designed (and Once the templates have been created for the ATE con -

125
version table, any engineer can write programs in Getel After getting together all the necessary prints, sche-
without concerning himself with the details of the par- matics, and specifications the test operator must estab-
ticular automatic test equipment that will perform the lish parametric requirements for the unit to be tested.
tests. However, note that not all Getel statements will This is best done by a man at present, since he can eval-
have holes-many will be control -type statements that uate the circuit performance and thus set limits on the
have direct counterparts in the object language that various parameters to be tested. However, some design
controls the tester. automation programs are closing the gap since they can
Getel is based on a logical assignment of tasks to man be designed with a capability to analyze the circuit and
and the computer, taking advantage of the capabilities relate certain internal parameters to input-output re-
of each. The early steps of carrying out a test program quirements.
either are creative or require decision -making ability Next, the operator must determine the best test se-
that is best performed by a man, while the later steps quence, taking into account the likelihood of failure for
are best assigned to the computer, since it can store and each test and the relative importance of each test in
retrieve data without error. To demonstrate the assign- meeting the specifications. The next step is to determine
ment of tasks, here are the typical steps in the process of what actions must be taken based on the results of the
performing a test. test.

1. Easy reading. Test routine for simple inverter circuit is written as instructions for manual test (left column), and also in programing lan-
guages for four different testers. Getel program for same test is at right. Note close correspondence between Getel terminology and instruc-
tions for manual test that would be performed by a technician at the test bench.

8 16
CIRCUIT INPUT REQUIREMENTS : CIRCUIT OUTPUT REQUIREMENTS :

+12 volts dc on pin 8 Unloaded output, pin 16

R2 -12 volts dc on pin 9 Logical 0, or 'down level'
6 --1114---
Ground on pin 7 High limit: +0.5 volts dc 1
12 ma.
Logical 1: +6.1 volts dc 1 Low limit: +0.0 volts dc
Logical 0: + 0.0 volts dc J pin 6 Logical 1, or 'up level'
High limit: +12.0 volts dc
9 7 0 ma.
Low limit: +11.3 volts dc

MANUAL TEST
INSTRUCTION A B C D GETEL

1.0 Test 1 01 @TEST NO. 01@ TID001 980000040001 04 100 Test 1

10
Apply 12 volts dc
02 FB +12.00, TB08
BV11200 01 01 20850800 20 110 Apply 12.0VDC
to pin 8 reference 03 FC -12.00, TB09 BC1PL2 02 11 20850800 31 to 'pin 8'
to pin 7. 04 TGO7 W1001 08
0341 20807 800 120 Apply -12.0 VDC
Apply -12 volts dc 05 FE +012.0, TE16 04061 0704800 20 to 'pin 9'
BV21 200
to pin 9 reference 88
to pin 7. 06 FD +06.10, TD06 BC2NL2 1 00000047 1 20 39 130 Apply 12.0 MADC
07 MV16/H +0.5v/L +0.0v W1102 to 'pin 16'
Apply 12 ma dc 11 0000030000 20
to pin 16 reference 08 PFC W0911 89
22 1 600000000 140 Measure 'DC
to pin 7. 39
BIXO3M1200 21 400007 0000 voltage',10±.1%
Apply 6.1 volts dc 20
to pin 6 reference BVX03M1 200 202050000000 87 150 VDC at 'pin 16'
to pin 7. 39
BV30610 160 Record 'result'
Measure output dc 24
BC3PL2
voltage on pin 16 81 170 Compare 'result'
reference to pin 7. W0703 34 to '0.5, '0.0'
21 goto 'HIGHM',
Compare output dc W201403 96 'LOWM'
voltage against high 31
limit -0.5 volts dc MTRSVPOR
24 180 Perform Test 2
-and low limit -0
MAXP0050P0 85
volts dc. 190 HIGH; MESSAGE
33
Record the value MI NPOOOOPO 20 TESTI FAILED
if it fails. 96 HI
RI U4
33
200 LOWM; MESSAGE
TESTI FAILED
LO

A: AAI Model 1000 IC tester; B: TI 553 IC tester; C: Fairchild Model 5000; D: GE -PAC 200 controller

126
2. Test run. At GE, test engineers use the Mark II time-sharing system to prepare Getel programs. After accessing the system, Getel is com-
piled with the Mark I translator program (Getel MI). In this example, Getel is translated to Fortran for use on the Hewlett-Packard rf tester. The
output of the system is a tape to be inserted in the tester to control the actual test.

1. BETEL
PROGRAM TEST ENGINEER 2: ACCESS MARK II TIML SOARING
WRITTEN DESIGNS TEST
FUR
GETEL LANGUAGE
PIECE OF
FARDVVARE REF. MANUAL DIAL MARK II GIVE USER NUMBER,
TIMESHARE SYSTEM AND GETEL
TELEPHONE NO. PROGRAM NAME
GETEL MARK I
USERS MANUAL

TEST ENGINEER PUT GETEL TAPE IN

PROGRAM
WRITES TEST TELETYPEWRITER CREATE FILE
PUNCHED ON
PROCEDURE IN READER AND PRESS FOR OUTPUT
PAPER TAPE
GETEL RUN

ACCESS GETEL ATE TABLE WAS WRITTEN AND

GIVE ATE
MARK II COMPILER PUT IN TIME-SHARE SYSTEM
TABLE NAME
AND RUN AT AN EARLIER TIME
COMPILE
USING
GL TEL
MARK I
TELL WHETHER LINE TURN TELETYPE.
GIVE OUTPUT GIVE GETEL TEST COMPILE
NUMBERS WANTED WRITER PUNCH
FILE NAME PROGRAM NAME COMPLET
ON OUTPUT ON

TAPE READY
TO BE RUN IN OUR CASE THIS (011 AIN
OUTPUT TAPE IS IN AIIS01.
FORTRAN 11INARY 1API
BY 1 Olt IRAN
COMPIt A (ION
IS
TAPE MUST
LIST OUTPUT FILE ATE
COMPUTER BE COMPILED
AND GET OBJECT
CONTROL ON COMPUTER
TAPE
IN ATE
LOAD HEWLETT-
PACKARD FORTRAN
SIMULTANEOUSLY COMPILER FROM
GET OBJECT 4. LIST GETEL MARK I OUTPUT AND MAGNETIC TAPE
LISTING CONTINUE IF NECESSARY

OPTIONAL
ASSEMBLY
LISTING
LOAD HEWLETT-
PACKARD BASIC
SELECT PUT FORTRAN LOAD ADDRESS OF CONTROL SYSTEM
RELOCATABLE FROM MAGNETIC
OUTPUT PROGRAM IN FORTRAN START
BINARY TAPE TAPE
OPTIONS PHOTOREADER AND PRESS RUN

LIBRARY ROUTINES 6. I OAD

FROM MAGNETIC TAPE AND II !: I HARDWAltl HARDWARE
GETS TESTED

LOAD ADDRESS LOAD ABSOLUTE LOAD ADDRESS

SELECT ABSOLUTE HOOK UP
OF BASIC BINARY TAPE IN OF ABSOLUTE
OUTPUT BINARY HARDWARE TO
CONTROL SYSTEM HEWLETT-PACKARD BINARY TAPE
OPTIONS TAPE BE TESTED
AND PRESS RUN COMPILER AND RUN

127
 Likewise, the operator is best able to decide how In many cases, single instructions for the manual test
much data is enough and how best to control the oper- must be expanded into several subinstructions for the
ator displays. He also can explore other possible ways of automatic tester.
performing the test, if he decides that the test station In Getel, there is little difference between the instruc-
cannot do it. tions for manual test and the instructions for automatic
However, after this, the computer can take over. The testing. Only line numbers have been added and all la-
computer can be instructed about what equipment is bels have been apostrophized.
available and what its capabilities are. From this instru- Most users at General Electric access Getel through
ment pool, the computer can choose the instruments, GE's Mark II time-sharing system (the Mark II in this
since it can even keep track of which instruments are in case bears no relationship to the presently unused Mark
use on other tests and which instruments are available II translator for Getel; the redundancy occurred be-
for this test. cause of the assignment of the name Mark III to the
The computer can then translate the instructions into large translator, and this term propagated through later
machine code to operate the instruments, using look -up versions of Getel translators).
tables. This is one step that a man performs poorly-it is Step by step. Before the test engineer signs onto the
an unthinking task and errors often creep in. Also, the Mark II system, he familiarizes himself with the unit to
translation of these machine codes back into English to be tested and designs the test. He then converts the test
see if any errors have been introduced is unnecessary instructions to a Getel program. The Getel program
with the computer. The machine can also optimize test then is punched on paper tape in ASCII (American Stan-
time by checking the machine code for duplications. dard Code for Information Interchange), and then the
Thus, with Getel, the user can describe the parameter to engineer dials the GE Mark II system on a telephone
be tested and the computer will select the instrument hooked up to a teletypewriter, as shown in the diagram
for the job. in Fig. 2.
Test instrumentation falls into three categories: Upon request from the time-sharing system, he gives
sources (power supplies, signal generators); sensors his user number to identify himself as a qualified user,
(digital voltmeters, electronic counters); and loads (re- and the system then queries him on which of the two
sistors, capacitors, inductors). To control such instru- available languages-Basic or Fortran-he wants to use.
ments, the operator uses as statements simple impera- Since the Mark I translator is written in Fortran, he re-
tive sentences containing verbs that are commonly used sponds with that selection. He then gives a name to his
by test engineers. Prefixes, suffixes, and units abbrevia- Getel program and enters the paper tape into the tele-
tions for Getel are taken from IEEE standards and Gov- typewriter reader for insertion in the time-sharing
ernment documents. memory. He then creates a file in the time-sharing sys-
The user also can define synonyms and insert stan- tem in which the translated program (the object pro-
dard routines into the body of the program to compose gram) will be stored.
new complex test functions based on groups of simple After telling the system that he wishes to use the Ge-
test functions. tel Mark I program, he goes through some bookkeeping
As an example of how Getel differs from languages steps with the teletypewriter, loads in the proper ATE
used in the automatic testers, consider a typical inverter conversion table, and the system then delivers a
circuit that is common enough to be tested practically punched tape of the object program for use on the au-
on any of several types of automatic testers. The test tomatic test system that he intends to use. This tape can
will be programed in each of several languages and also then either be processed further on the test system's
in Getel. (The actual meaning of each command in the computer or the test can be run directly from the tape,
languages is not really important for this discussion-the depending on the tester in use.
example is presented only to illustrate complexities.)
Testing. To test the inverter, appropriate supply volt-
ages must be applied along with a certain signal level to
the input. Then the output must be measured and com- Anyone can use Getel
pared with the design value, and the output must be re- Getel is available to users outside General Electric. It
corded if the inverter fails the test. has been put on GE'S Network Service, a time-sharing
The circuit is shown in Fig. 1, along with a set of test service centered in Cleveland. Remote concentrators in
instructions written for interpretation by a technician Los Angeles, Kansas City, Atlanta, Teaneck, N.J.,
actually doing manual testing. The instructions in the Schenectady, N.Y., and Washington, D.C., link the cen-
column for tester A-AAI Corp.'s model 1000, a popular tral computer complex-which comprises a GE -635 cen-
integrated circuit tester-are almost self-evident. This tral processor, a GE -4020 communications processor,
tester uses a pseudo -variable -field technical programing and a disk file complex-to remote buffer units in major
language, in which the alphabetic terms are used; FB metropolitan cities. Customers can gain access to the
means fix bias supply B, and TB means tie (connect) bias system with a local phone call. Overseas, the network
can be used in such cities as Amersfoort, Brussels, Lon-
supply B, but still an engineer might find it difficult to don, Manchester, Birmingham, and Paris. In the west-
interpret the instructions with certainty unless he ac- ern hemisphere, outside the continental United States,
tually knew the language. the network can be dialed from remote units in such
Tester B, 553, also uses some alphabetic charac- cities as Mexico City, San Juan, Montreal, Toronto,
ters, but Tester C, Fairchild's model 5000, uses numeric Vancouver, and Anchorage.
code only, as does Tester D, the GE -PAC 200 controller.

128
GETEL VOCABULARY

OPERATIONS SQUARE WAVE TRIANGULAR WAVE AM SIGNAL

APPLY BW FREQ MODUL-AMPTD

CALCULATE IMP BW CAR R-AMPTD

CLEAR PHASE-JIT FER IMP CARR-FREQ

COMPARE RT PHASE -JITTER HARMONICS

COMPLETED FT NON LIM NONHARMONICS

CONNECT TIME ASYM TIME ASYM NOISE

CONTINUE OVERSHOOT PEAK- DEGEN PHASE -JITTER

DEFINE UNDERSHOOT DC -OFFSET

DELAY PRESHOOT
DISCONNECT DROOP DC SIGNAL
END RINGING UNITS
EXECUTE ROUNDING POWER

EXTERNAL DC -OFFSET AAC CUR-LIM

GOTO AMP AC VOLT-LIM
AMPS AC IMP
IF GOTO
ADC RIPPLE
IF -PERFORM MULTIPLIERS
AMP DC
INSERT RIPPLE -F RE()
AMPS DC
LET NOISE
P (10-12) VAC
MEASURE VOLT AC DISTORTION
N (10-9)
NOTE VOLTS AC CUR MONITOR
U (10-6)
PAUSE VDC VOLT MONITOR
M (10-3)
VOLT DC
PERFORM K (103) VOLTS DC
PROCEDURE MEG (106) WATT
READ FM SIGNAL
G (102) WATTS
RECORD A/SEC
FM -INDEX
RESERVE AMP/SEC
AMPS/SEC CAR R-FREQ
RESET I !MI IN I IIVAI
I

ROUTINE DB MODUL-FREQ
ROUTINE END DEG HARMONIC DIST
TIME
SELECT DEG -C NONHARMONIC DIST
SEC
SENSE MIN DEG -F AMPTD-MODUL
SHARE HR DEG -K MODUL-DIST
TEST DEG -R NOISE
START
TITLE STOP FD
TURNOFF TRIG HY
TURNON FIRST POS SLOPE HZ SUFFIXES
USE FIRST NEG SLOPE HERTZ

CHANNEL A HR -TRMS (true rms)

OHM -AU (average)
I t .I(' SlliNAI PCT -PP (peak -to -peak)
PPS -P (peak)
ZERO MAG RAD
ONE MAG REV
RT SEC
FT S -PARAMETERS
NOISE V/HZ
FREQ
SKEW TIME

129
Sceptre rules benevolently
over computer -aided design
Some practical tips on solving nonlinear problems with Sceptre reduce
the complexity and cost of this computer program; tradeoffs must
be made between circuit models, solution accuracy, and running time

by George C. Kenney, North American Philips Corp., Briarcliff Manor, N.Y.

Solving linear and nonlinear circuit problems is the Sceptre accepts input system data in two forms: as
province of a powerful computer program called nonlinear, first -order, differential equations or as engi-
Sceptre. Unfortunately, its reign has been limited by its neering descriptions of a circuit (similar to ECAP). Initial
complexity, high cost, and lengthy computer time. dc conditions are computed or supplied by the user if
Sceptre, however, can be a practical and very valuable known, then a transient analysis of the system's re-
engineering instrument if used properly. sponse to forcing functions-such as time -varying volt-
Sceptre, which stands for system for circuit evaluation ages or currents-is carried out. It is also possible to run
and prediction of transient radiation effects, can deter- initial conditions only and then, after inspection, run
mine initial dc conditions and transient responses to transient conditions, or compute only the transient
general forcing functions for linear and nonlinear sys- mode by supplying the initial conditions.
tems. Although it is primarily intended for circuits, it Sceptre recognizes standard passive elements such as
can find solutions for systems in any discipline. resistors, capacitors, inductors, transformers, and the
Until recently, Sceptre was available only by batch ideal diode. Linear dependent sources-like voltage -
run at a computer facility, making it costly, time-con- controlled current or voltage sources or current -con-
suming, and difficult to use. As a result, many nonlinear trolled voltage or current sources-can also be used.
problems were solved by hand calculations, circuit ex- These circuit elements enable the user to synthesize
periments, or by piecewise linearization with ECAP Ebers-Moll models for transistors as well as diodes that
(electronic circuit analysis program). are operational in the large- or small -signal region.
But now, input/output data for a Sceptre program Models stored on tape can be retrieved for use at any
can be handled by a remote teletypewriter terminal point in a circuit. The models can be interconnected
(through software provided by Computer Sciences groups of passive or active, linear and nonlinear ele-
Corp., Infonet div., El Segundo, Calif.). This allows the ments. Nonlinear "black boxes" are implemented by ta-
user to solve nonlinear problems with an ease ap- bles or a Fortran expression. A special section allows
proaching that of time-shared ECAP, and with reason- special parameters to be defined that may be unrelated
able expenditure of time and cost. to the primary problem. Moreover, the user may write
his own Fortran program and insert it into a normal
1. Circuit problem. Hysteresis of this flip-flop can be computed and Sceptre run as a subroutine.
plotted with Sceptre. Circuit elements are labeled for easy computer Using Sceptre. Printout is in tabular and/or selected
callout, with letters L and R differentiating between left and right side graphical form. Sceptre will print any or all sources, in
of flip-flop. Current sources are denoted by J. addition to circuit element currents or voltages, which
are defined as a function of time or of any other
parameter. The plots are easily requested, automatically
scaled, and labeled.
An automatic termination feature stops the program
upon satisfaction of any Fortran expression involving
any of the circuit elements. For example, the statement
TERMINATE IF( -1R7 GT.10.I.AND.R9 LT.I2.2)
will terminate the run when the reverse current through
R, is greater than 10.1 milliamperes, and the voltage
across R.. is less than 12.1 volts.
Sceptre uses the state -variable principle as its mathe-
matical basis. Capacitor branch voltages and inductor
link currents are the state variables chosen to form a
network of first -order, nonlinear, differential equations.
This method is very efficient since computation time re-
lates only to the number of state variables and not the
number of elements.

130
 12.1 V

RST
2052 EST
RLL RLR
0.316k 1-2 0.3161(2

IC3L R3L R3R C3R

0.5 pF 3.16k52 3.161(2 0.5 pF

+5V
RC 102 RC 102
ti
-5V CCSL CCSR
OpF 10pF
FREQUENCY = 1kHz
R6L R6R
102 JCSL JCSR 10 2

R1L R1R
0.73
CESL JESL JESR -CESR 0.73
10pF 10pF
1(S2 TC1L
0.5 JFL aNIE JFR aNIE TC1R
0.5
pF pF
REL RER
ES 1052 0 02 1052

-T VR
VR
R2L RJC
42k 2
JC
5mA 1M52
R2R _L
42k52

12.1V ESB
RSB
(a) HIGH FREQUENCY MODEL 2012

11112.1V
RST
RLL 20 2 EST
0.316k S2

R3L
3.161(2 R3R
3.16k52

+2V

CCSL CCSR
-2V 10 pF 10pF
FREQUENCY = 100kHz
11-*
JCSL JCSR

CESL JESL CESR

10pF 10pF R1R
R1L
0.73k2 JFL aNIE JFR aN IE JESR 0.73k2

01 Q2 - VR
ES T 0,2V
R2L R2R
VR 421(2 JC
0,2V 5mA 42k52

1_
12.1V ESB
RSB 1

(b) LOW -FREQUENCY MODEL 2012

2. Sceptre models. High -frequency model (a) on flip-flop includes transistor speed-up capacitors and lead resistance. These circuit ele-
ments, however, severely slow up solutions at low frequencies because Sceptre must take thousands of steps to arrive at answer. Changing
to model shown in (b) speeds solution. Increasing frequency of forcing function and decreasing its amplitude also helps.

131
 Ebers-Moll pnp transistor model
There are usually two circuit equations associated with the Another version of the equation is:
Ebers-Moll transistor model: one for the collector -current,
If., and the other for the emitter current, IF. Ii: = IE0(e6N vivE - 1)/(1 - aiaN)
The collector current is: where IH is the base -emitter saturation current with the
lc = Ics(e8i valet - 1) base -collector open. The program listing for IF becomes:
IF = DIODE EQUATION (ON, VB.E)
where L., is the base -collector saturation current with the The model's circuit elements are defined as: rH, the base -
base -emitter shorted, and 0, the slope of the natural log of spreading resistance; re, the series collector resistance; C,,
le versus the collector diffusion capacitance; C1 e, the collector junc-
This equation can also be written as: tion transition capacitance; CdE, the emitter diffusion ca-
ic = Ico(eeivB'c' - 1)/(1 - aiaN) pacitance; and C,H, the emitter junction transition capaci-
tance.
where Ie is the base -collector saturation current with the
base -emitter open, a, the inverted common -base current
gain, and a, the normal common -base current gain. For
computer program listing, the collector current becomes:
r MODE EQI ATION (OD Vli'C)
Similarly, the expression for the emitter current can be
written as:
IE = IEs(eNvivE - 1)
where I is the base -emitter saturation current with the
base -collector shorted, and 0, the slope of the natural log
of 1H versus VH,H.

The first step in preparing a problem for analysis with transition collector capacitances become CCSL (or
Sceptre is to draw an equivalent circuit using resistors, ccsR). Also, small speed-up capacitors (5 pF) and lead
inductors, capacitors, transformers, sources, and any resistances (10 ohms) are included for better accuracy.
necessary stored models. Circuit elements may be linear Because the flip-flop's hysteresis is reflected in the dif-
or nonlinear, and defined by a numerical constant, ference between the collector voltage of Q, and Q2, re-
tabular list, or Fortran expression. sistor RO is added. Circuit hysteresis, which is a function
Next, name or number all nodes and name each cir- of R1L (and R IR) and R3L (and R3R), can be pre-
cuit element. Then choose current flow direction in each dicted by computing the voltage across RO.
passive element and source. Assign circuit values in a Operating frequency range for the flip-flop is ki- 1

constant set of parameter units. (For high speed, a good lohertz to 10 megahertz. Therefore, a forcing function
set is kilohms, picofarads, microhenries, milliamperes, with a frequency of 1 kHz is a reasonable choice for the
volts, and nanoseconds.) excitation voltage. To guarantee that the flip-flop will
Circuit elements can now be easily combined to form trigger, the amplitude of the forcing function will be
an Ebers-Moll large -signal transistor equivalent circuit 5 v. Let the forcing function, then, be E, = -5cos27
(see "Ebers-Moll pnp transistor model," p. 74). The 1000t, where t is time.
emitter current, 4, and the collector current, 1, are en- Fail-safe. For the model of Fig. 2a, Sceptre will cor-
tered as DIODE EQUATION (0,V) or DIODE TABLE rectly solve the dc case but will fail to give a transient
XY to achieve numerical convergence for the initial solution. The reason for the failure will be indicated as
condition solution. SMALLER MINIMUM STEP SIZE REQUIRED.
An example will illustrate how to use Sceptre and This type of failure, which is sometimes called the Ei-
how to implement model changes in order to achieve a genvalue problem, is caused by a conflict in circuit time
successful run. The circuit hysteresis of the flip-flop in constants that will incapacitate any nonlinear analysis,
Fig 1 is to be predicted by the computer. The diagram computer or otherwise. The flip-flop has natural fre-
uses a labeling scheme that is convenient for calling out quency modes caused by short time constants that re-
circuit elements in a program. Besides the usual nu- quire very small time steps for accurate analysis. Be-
merical differentiation between components, the letters cause of the long running time to complete the solution,
L and R distinguish between elements on the left side Sceptre will decline to solve the problem.
and right side of the circuit. There are several elements aggravating the flip-flop's
The Sceptre model of the flip-flop in Fig. 2(a) uses a time -constant problem. Speed-up capacitors, header ca-
slightly modified version of the Ebers-Moll transistor pacitances, and transistor lead resistance do not affect a
equivalent shown on p. 74. Since the flip-flop's transis- low -frequency solution, but they do complicate solution
tors never saturate, current source ail, can be omitted. processes and increase program time.
Diode JCSL (or icsR) replaces current source lc, and Step size is determined by the specified problem du-
diode JESL (or JESR) replaces source IE. Both the diffu- ration (stop time) and by the smallest time constant of
sion and transition emitter capacitances are combined the circuit. The initial step size chosen by Sceptre is 10-3
in capacitor CESL (or CESR), while the diffusion and times the stop time. This step size is tested by the inte-

132
gration routine for accuracy, and if that size is found to Figure 3 reproduces the successful run of the pro-
be erroneous, a new step size that is half the previous gram input data and the hysteresis plot. The forcing
value is attempted. function, E,, is entered as a table; all diodes are repre-
The test process is repeated until a small enough step sented by the exponential diode equation. The program
size is found or until the step size is smaller than 10-5 requests a plotted output of VRO, the voltage across re-
times the stop time. If the step size required is less than sist& RO, versus E8. The plot correlates fact with
this value, the run is terminated because a smaller min- theory -computed hysteresis is 0.4 v; the measured is
imum step size is required. 0.5 v.
A successful run can be achieved without changing A brief look at cost considerations is also essential. Be
the low -frequency hysteresis by removing all speed-up cause Sceptre is a large program that requires expensive
capacitors and all lead resistances as shown in Fig. 2(b). Fortran compilation, a good deal of its cost tends to be
It is also helpful to decrease the magnitude and period a fixed overhead. The flip-flop problem, for example,
of the forcing voltage, E5, to speed up the integration. can be run with identical numerical results on an IBM
Actually measuring circuit hysteresis helps to deter- 360-65 or a Univac 1108, using a conversational, re-
mine how to modify E. A measurement will show that mote, job entry teletypewriter. The price of the IBM run
the hysteresis is about 0.5 v at 1 kHz as well as at 100 will be approximately $90, 80% of which is overhead,
kHz; therefore, changing the frequency of E. from 1 kHz while the Univac solution costs $30 with 20% overhead.
to 100 kHz will speed the solution without affecting ac- Defined parameters. Another example will serve to il-
curacy. Reducing the amplitude of E. from 5 to 2 v also lustrate the flexibility of Sceptre through the use of its
aids the solution without impairing the triggering of the special defined parameters section. Suppose the prob-
flip-flop. Switching from a cosine wave to a triangle lem is to solve a set of first -order, simultaneous, differ-
wave is another simplification. To see the change in hys- ential equations that may be entirely independent of
teresis at high frequencies, another run could be made any electrical network:
at 10 megahertz with the original model and original E.. X'(t) = -6X(t) + 5Y(t) + 10

RrPPTRE SYSTEM INPUT PLOT OF VMS VS ES

VRO

2.500E000 1
CIRCUIT DESCRIPTION 1 1 1 1 1
FLIP FLOP
ELEMENTS
ES.11-1ETARIE 11TIME)
2.000E00 1--°-
ERBL.3-2.0
R2L.2-8'42
JESL.0-3.010DE EQUATION(1.2E-11.35.)
CESL.0-3.10
EREL.6-7.i 1.500E000 1----
-3.
R3L.15-323.16
JC11.4-3.DIODE EQURTION(1.2E-11.35.)
CCsL.4-3.10 .I
01..3-4.,97.JESL 1.000E...00 1----
ERcL.3.!_4.11
RIL.i0-5..316 . I

RLR.10-15..316
RST.10-17Y.02
EST.17-0.12.1 5.000E-01 1--
RSB.11-9..n2
ELMO-9.12.1 1.

ER.O..11.0
R114,11-12,73 1-FVS-TE-RES4S --0.
R211.12-8.42 0.
rT
EREIR.12-13.0
JESR.10-13EDIODE EOURTION(1.2E-11.35.)
coR.16-13.10
R5R.5-I3.3.16
JCSM.14-13xDIODE EOURTIONI1.2E-11.35.) -5.000E-01 1--
CCSR.14-13.10 1
JI.R.13-14,97.4ESR ,
.FRCR.15-14.0
RJC.7-8.1c00
R0415-5.1801 -1.0000.00 1----
ERER,16..7.0 li
FUNCTIONS
TOLE I
0.-2 1
At1J2 -1.500..00
E2 1----
10E3.-2
OUTPUTS
YRA,RLOTLE5)
01.4f010(RER010/21 .RN4c,J!LOT
RUN CONTROLS
U74114tteCgnTIONS
STOP Timp.12p3
MINIMUM STEP SIZE ..001 J
RERUN DEScRIPTIoRI1)
.4.000E00 2.000000 0.
-5.000E.00 -1.000E*00 1.000E.D0

(a) PROGRAM INPUT DATA (b) HYSTERESIS PLOT

3. Computer printout. Program listing (a) shows set of Fortran instructions needed to predict flip-flop hysteresis. A graph of the hysteresis
(b) is obtained by requesting a plot of the collector difference voltage (between 0, and 02), represented by VRO, versus the forcing function,
E,. Distance between colored lines is predicted hysteresis voltage, which is 0.4 volt compared with measured value of 0.5 V.

133
 Modified pnp Ebers-Moll model
A few minor modifications of the conventional Ebers-Moll where Ts is storage time
transistor model improves its speed without adversely af- CdE TE IE
fecting its accuracy. Eliminating extraneous resistances where TE is the minority carrier transit time
and carefully combining diffusion and transition capaci- TE = I /(0.11,ha = I/1.2wT
tances implements the change. where cos, is the alpha cutoff frequency, and co, is the
The equation for transistor collector current, Ie, can be beta cutoff frequency
written as: CU` = Cob° CCE
IC = - 1) where Co, is the output capacitance with the emitter open
CIE = Cibu CCE
while the expression for transistor emitter current, 1E, be- where C,,, is the input capacitance with the collector open.
comes:
IE = IE(o.euN valE - II
CBC
All modified circuit elements are defined in terms of the
original model:
C, = 3C./4
CH'e = 4+ CdC
CE = CdE
Cc = Car + Cu` = CAC + CH.(' CE CCE
CcE = collector -emitter header capacitance
rs, = base -spreading resistance
Cd,. = 01 Ts lc

Y'(t) = 5X(t) - 7Y(t) equations must be reduced to a set of first -order ones.
The initial conditions are X(0) = 6 and Y(0) = 5. Limitations. There are several important factors to
Each of the derivatives may be entered under the keep in mind when using Sceptre. The first one is to
program listing DEFINED PARAMETERS in explicit know the application. Sceptre can solve very compli-
form. A proper sequence would be: cated nonlinear problems; if the problem is linear, use
CIRCUIT DESCRIPTION ECAP. It is often worthwhile to simplify the problem by
DIFFERENTIAL EQUATION EXAMPLE using different models at different frequencies.
(Enter data, for instance, as in Fig. 3(a).) Another consideration is computer models. For most
DEFINE PARAMETERS nonlinear applications, transistors can satisfactorily be
DPX = EQUATION 1 (PX,PY) handled by Sceptre with the Ebers-Moll model shown
DPY = EQUATION 2 (PX,PY) in "Modified pnp Ebers-Moll model," above. It offers a
PX = 6 good compromise between speed and accuracy. Always
PY = 5 remember that the simplest satisfactory circuit model
OUTPUTS has the best.chance of running.
PX(X), PY(Y), XSTPSZ, PLOT Sceptre is limited to two types of active dependent
FUNCTIONS circuit elements for accurate dc (initial condition) analy-
(Define equations X'(t) -6X(t) + 5Y(t) + 10 and sis. These are a current source that depends on a current
Y'(t) = 5X(t) - 7Y(t) by using dummy variables A and a nonlinear element, like the diode. The diode must
and B for X and Y, respectively.) be entered as a diode equation or a diode table. Any
EQUATION 1 (A,B,) = (-6.*A + 5.*B + 10.) other active dependent circuit elements (such as volt-
EQUATION 2 (A,B) = (5.*A-7.*B + 2.) age -dependent current sources, passive elements that
RUN CONTROLS depend on either voltages or currents, and tables for ze-
INTEGRATION ROUTINE = TRAP ner diodes, tunnel diodes, or any device whose charac-
STOP TIME = 100 teristic curve does not monotonically increase) may
END cause convergence difficulties in the dc solution.
Since the derivatives of PX and PY (DPX and DPY, re- However, these elements present no difficulty for the
spectively) are entered in the program, the functions of transient solution portion of Sceptre. They even offer an
PX and PY must be updated at each integration step.
alternate method of obtaining a circuit's initial condi-
It should be noted that the initial values of the vari- tion when the dc method fails.
ables X and Y are entered as PX = 6 and PY = 5; the To approximate a dc steady-state solution, some ini-
differential equations themselves are entered under DE- tial conditions can be assumed and a transient solution
FINED PARAMETERS. Since the quantities X and Y run (with the forcing function set to zero and all capaci-
are treated like the state variables of a general transient tors given equal values) for a duration of five constants.
problem, they are subject to the same step -size limita- Using the steady-state values obtained in this way as
tions in whatever Sceptre integration routine is used. initial conditions, a transient run with a proper forcing
When the defined parameters feature is used, it function can be made. This technique should only be
should be remembered that Sceptre can only solve non- used after unsuccessful attempts to solve the dc case.
linear, first -order, differential equations. Higher -order Finding initial conditions by transient response will al -

134
ways require two runs to achieve a final solution.
Another difficulty often encountered is the Eigen- Nonlinear computer programs
value or small -time -constant problem. First, it should
be determined whether the elements causing the trouble At present, there are four computer -aided design pro-
grams for solving nonlinear circuit problems: Sceptre,
are necessary to get the desired solution. About 90% of NET -1, Circus and TRAC. Each one has its own special
the time, they are superfluous and only serve to bog application and, when properly matched to a problem,
down the program. will minimize both time and cost. Generally, Sceptre is
One method of solution is to use a simple model at considered the most powerful, flexible, and expensive
low frequencies and a complex one at high frequencies, of the four; it can solve many problems beyond the
where the forcing -function frequency and the Eigen- reach of the others.
values are compatible. A good low -frequency model is NET- I (network analysis program) employs predeter-
one that has no capacitances except large diode shunt mined stored Ebers-Moll models of junction diodes
capacitors and that uses E, at the highest possible fre- and bipolar transistors for a good compromise between
speed and accuracy. But the program's flexibility is lim-
quency without appreciably changing the dc solution. ited. NET -1 does not allow the user to account for such
An appropriate, but not overly complex, model things as junction breakdown, base narrowing or con-
should be used for a second run at higher frequencies. ductive modulation.
This two -model technique will probably be more satis- Circus (circuit simulator) employs a somewhat
fying and less expensive than using the same model for awkward charge -control model that is mathematically
high and low frequencies, as can be done with ECAP. equivalent to the Ebers-Moll model. Program restric-
Even if the problem involves conflicting Eigenvalues tions include acceptance of only pulse or sine inputs,
and all attempts at separate solution are unsatisfactory, limited element library, and no tables or analytical ex-
there is yet another approach. The user can allow pressions.
Sceptre to take a smaller step size than the normal 10 - TRAC (transient radiation analysis by computer) cannot
handle stored elements, tables, or analytical expres-
times the stop time. A specific minimum step size (for sions. However, a problem that can be solved with
example, MINIMUM STEP SIZE = 0.00001) is called TRAC will be done more economically and faster than
out as a RUN CONTROLS instruction. Sceptre follows with Sceptre.
the same process to determine step size-starting with
10-3 times the stop time and dividing by two each inte-
gration pass until the specified limit is reached. is plotted as a function of time.
It is also useful to call out a maximum step size (like Sceptre has three integration routines: TRAP, XPO,
MAXIMUM STEP SIZE = 100) that is about 1% of the and RUK, which (in this order) are increasingly accu-
stop time. The program can then take large steps in rate, time consuming, and costly. For most applications,
time toward the solution when the derivative of the XPO, which gives a 2.5 -term approximation of the Tay-
state variables are changing slowly. For example, a cir- lor series, offers optimum speed and accuracy.
cuit problem may require very small step sizes in the Although Sceptre can handle up to 300 circuit
beginning, but it could correctly take large steps later, branches, it is advisable to limit the number of branches
near the steady-state condition. to 70. The Fortran compiler cannot handle more than
Excessive computer time is not a concern when a very 70 branches at once, and large programs become diffi-
small step size is specified. A built-in counter logs inte- cult to run.
gration passes and automatically terminates the run There are several reasonable topological restrictions
when the number of passes exceeds 20,000. When step imposed by Sceptre for dc solutions. Two are shown in
size is varied, it is a good idea to call for XSTPSZ, Fig. 4-a voltage -source inductor loop and a current -
PLOT under the OUTPUTS instruction so that step size source capacitor cut set (dual of a loop, connecting one
node to another node). The dc case for these problems
is solved by adding a small resistor, Rx, to the loop and
a large resistor, R), to the cut set. The resistors should
be eliminated for the transient run since cut sets and
El loops are acceptable for the integration routine. It
should be noted that a diode, instead of a resistor, can
O be used to prevent the formation of a cut set, as it does
in the Ebers-Moll model.
la) VOLTAGESOURCE (b) CURRENT -SOURCE
INDUCTOR LOOP CAPACITOR CUT SET Very small and very large passive element values
should be avoided in Sceptre because of the time con-
stant problem. Also, zero values of resistance, induc-
tance, or capacitance will cause a run to terminate in
Fortran. Because Sceptre can plot the current through
voltage sources and the voltage across current sources, a
Ic) DC SOLUTION FOR LOOP Id) DC SOLUTION FOR CUT SET zero -impedance element can be represented by a zero -
value voltage source, and a zero -value current generator
4. T r.polog Ica I restrictions. To solve the dc case (initial conditions) makes an ideal infinite -impedance element. Moreover,
for a voltage -source inductor loop (a) or a current -source capacitor the diode capacitances in the Ebers-Moll model should
cut set (b), Sceptre requires the addition of a resistance-in series not be neglected since their exclusion would cause solu-
for the loop (c) and shunting for the cut set (d). tion errors.

135
Optinet guides electronic products
from design through marketing
Software system goes beyond usual optimization and anaysis functions
of CAD programs-helping users develop a prototype,
control production, and dramatically expedite marketing procedures

by Robert Hall, Dean Hall Associates Inc Los Altos, Calif.

The various people involved in designing, building, Basically, Optinet requires three types of files-net-
and selling electronic equipment have various require- work, frequency, and performance-to solve an analysis
ments-the engineers want to optimize design, the pro- problem. They are numbered to correspond to each
duction staff aims for maximum efficiency, and the mar- other; for example, files N I, F 1, and P 1 form a set, as
keting men seek to reach the right customer. Now they do files N2, F2, and P2. Up to 15 complete file sets are
have a software package that can satisfy all of their available. The network file gives the circuit description,
goals. Called Optinet, the program can be used to opti- the frequency file lists the frequencies of interest within
mize circuit designs, keep track of production controls, a design band, and the performance file defines circuit
and expedite marketing procedures. requirements.
Optinet includes a full set of computational facilities Data files (30 in all) are independent of the other files
and uses them to perform modeling with variable para- and contain values of frequency -dependent circuit ele-
meters. Also provided is storage for models, solutions, ments or circuit specifications, which are defined in ei-
and intermediate results, as well as commercial product ther a network or performance file. These data files can
data. Another feature is a dial -up access capability that be used anywhere in the analysis.
brings the system to manufacturers and customers. A variable file is required for design, because the user
The software package is available through National must specify the circuit elements he is willing to vary to
css Inc., Stamford, Conn., and Sunnyvale, Calif. By meet desired performance specifications. This file allows
next March, other time-sharing services are expected to either a circuit element as called out in a network file or
offer the program. a circuit parameter as specified by a performance file to
Optinet generally requires a large computer facility become a variable. Limits and nominal values for the
on the order of the IBM 360/67. The program is written variables are established by the file, which can contain
in Fortran 4. Though Optinet is one of the fastest and up to 15 lines.
least -expensive analysis programs available, design A criterion file is also required for Optinet's design
analysis itself is not always the highest -priority consid- capability and can have up to 15 lines. The criterion file
eration when introducing a new product. With its sensi- defines a function that represents best circuit perform-
tivity analysis, worst -case diagnosis, production mod- ance. The function usually has one form for perform-
eling, and market data capabilities, Optinet can cut ance optimization and another for worst -case analysis.
costs and time involved in the critical process of trans- For example, an optimization criterion could be a func-
ferring the product to the production and marketing tion whose minimum value represents best perform-
stages. ance, while a worst -case criterion could be a function
Optinet is designed for devices, components, and sub- that goes to zero if desired specifications are not met.
systems that can be described in terms of piecewise- Optinet provides a sensitivity analysis capability, too.
linear ac models. Additional modeling capability is It can vary individual parameters by any desired per-
added regularly so that the system will be able to centage and print the resulting variations of the per-
handle dc, nonlinear and transient calculations some- formance requirements. A trace capability lets the user
time next year. plot frequency -dependent performance as a function of
A closer look. Modeling with Optinet is straight- a variable parameter over its entire range.
forward. Networks are broken down into subnetworks, Ample storage facilities are another feature: up to 15
sections, and elements. A catalog of more than 75 ele- complete problem groups can be saved in a private disk
mentary sections provides easy access to lumped, dis- area. When a problem group is retrieved, the contents
tributed, active, passive, analytical, numerical, and of all files are reset exactly as they were when the prob-
hypothetical models. Up to 15 networks can be defined lem group was saved. The problem then may be contin-
simultaneously, each with its own frequency range and ued from the last result.
performance requirements, permitting modeling of Information for use by multiple subscribers may be
piecewise-linear and multistate networks. A total of 15 stored in read-only files for general access. The contents
elements in these networks may be identified as the are retrieved by code names supplied along with the in-
variable parameters of the model. formation at the time it is stored. These names also can

136
 (a) AMPLIFIER BLOCK DIAGRAM

INPUT GAIN - OUTPUT

MATCHING LEVELING MATCHING
CIRCUIT CIRCUIT CIRCUIT

HEWLETT PACKARD 35821E

IbI TRANSISTOR CHARACTERIZATION

FORWARD NETWORK 2 REVERSE NETWORK 5

AVERAGE VALUES AVERAGE VALUES

GM DB 6.49855 GM DR = -19.0961
R OHM = 19.0001 R OHM = 39.0471
OHM = 10.2284 OHM . -46.3705
RHM .470666 RHM = .482

MAXIMUM VALUES MAXIMUM VALUES

DB 9.39645 AT F 1.6 GHZ GM DB = -17.4579 AT F = 3 GHZ

GM .
OHM . 20.2859 AT F = 3 GHZ R OHM 52.4157 AT F = 1.6 CH7
R
OHM 16.3528 AT F 3 GHZ OHM . -40.3196 AT F = 3 GH7
RHM .48 AT F = 2.39999 GHZ. RHM .520001 AT F . 3 GHZ

MINIMUM VALUES MINIMUM VALUES

GM DR = 4.02796 AT F GHZ
3 GM DB = -21.2096 AT F = 1.6 GHZ
OHM = 18.17 AT F = 1.6 GHZ R OHM = 28.2906 AT F = 3 GHZ
OHM = 4.14764 AT F = 1.6 GHZ OHM = -52.9882 AT F 1.6 GHZ
RHM .469999 AT F = 2.1 GHZ RHM = .46 AT F = 1.7 GHZ

1. Desired amplifier. Transistorized microwave amplifier (a) must supply gain of 12 decibels 3- 1 dB and have 50 -ohm input and output im-
pedances from 1.6 to 3 GHz. Due to transistor transmission characteristics, gain -leveling and matching circuits can be separated. Important
transistor properties are tabulated in (b) for forward and reverse performance.

be stored in a directory that can be printed out during center frequency band of 2 to 2.5 gigahertz; a total
any Optinet session. bandwidth of 1.6 to 3 GHz; a minimal number of stages;
The program uses a time-shared computer system and 50 -ohm input and output impedances, with max-
primarily because it is the best choice of communi- imum reflection coefficients of 0.2.
cations. Through time-sharing, Optinet can rapidly Data on a suitable transistor is prestored in the Opti-
service scattered users by providing mutual access to net public data file by the device's manufacturer. Using
models and performance criteria. Hewlett-Packard's type 35821E transistor for this ex-
Furthermore, Optinet offers a number of convenience ample, it is characterized in the file by S parameters at
facilities. For example, it controls the cost of compu- 15 volts and 15 milliamperes over a frequency range of
tation. Total dollar expenditure is reported on demand, 100 to 4,500 megahertz.
and estimates of the cost of computation are available The major transistor design problems are gain vari-
before the computation is run. ation and mismatched input and output impedances.
Information and diagnostics keyed to the current Since this particular transistor has low reverse trans-
state of the problem may be obtained by typing "?". mission, gain -leveling and matching circuits can be sep-
Successively deeper explanations are given when "?" is arated as shown in Fig. 1(a). Transistor properties rele-
typed again. Also included is an adaptive editor to vant to the problem are computed from S parameters
speed the flow of information, manage files, and adjust stored in the public file, as shown in Fig. 1(b).
to the user's style. Optinet describes the transistor for both forward (in-
Design example. How Optinet works is best illus- put) and reverse (output) characteristics. Average, max-
trated by following a hypothetical project from design imum, and minimum values are tabulated for gain
through marketing, and restricting the example to four magnitude (GM), resistance (R), reactance (x), and re-
basic areas: design, interpretation of measured data, flection coefficient (RHM). Also noted is the frequency at
production control, and marketing. which the value occurs as well as its units callout. For
Suppose the desired product is a broadband amplifier example, maximum input resistance is 20.2859 ohms at
that uses microwave transistors. Its specifications are: a 3 GHz, while minimum output resistance is 28.2906
minimum gain of 12 decibels with a flatness of ± 1 dB; a ohms at 3 GHz.

137
 (a) PARTIAL AMPLIFIER NETWORK 16.6 nH

RS 1.23 pF 4.3 nH

10 -'MHO RL

SOURCE H -P 35621E H -P 35821E LOAD

SECTION 1 SECTION 2 SECTION 3 SECTION 4 SECTION 5 SECTION 6
CATALOG S1 65 4 2 65 L1
ELEMENT 1/5 RS S11M S21M L L S11M S21M RL
ELEMENT 2/6 XS S11P S21P C S11P S21P XL
ELEMENT 3/7 ZO S12M S22M G S12M S22M
ELEMENT 4/8 S12P S22P S12P S22P

lb) NETWORK FILE

N2
SECTION CATA Lon FLF.MFNT1/5 El EMENT2/6 FLEMENT3/7 E I FMENT4/E
1 S1 50 0 5n
2 65 1C1D9 1C21)111 1C3D11 1041112
105D13 106014 icinn irvir,
3 16.6C12 1.23C9 1E-5
4 2 4.3C10
5 G5 109 1D10 1D11 1D12
1013 11114 1D]5 1D1C
Li 50 0

(dl VARIABLE FILE

V
(c) FREQUENCY FILE DATA FILE LINE notimori NED-BOUND POS-BOUND VA LUE
1 9 1 2 1.23
F2 D13 2 10 2 6 4.3
LINE FREQUENCY LINE DATA 3 11 11 13 12
1 1.6 1 2.95 12 10 2C 16.6
2 1.7 2 2.3
3 1.8 3 2.62
1.9 4 2.54 (e) PERFORMANCE FILE
5 2 5 2.37 P2
6 2.1 6 2.3 LINE MEASURE FORM WEIORT REFERENCE cnvnnaF
7 2.2 7 2.15 1 nm PLOT 10 12C11 SO
8 2.3 8 2.68 2 D1
q 2.39999 9 1.98 3 X 02
10 2.49999 10 1.91
11 2.59999 11 1.86
12 2.69999 12 1.77 (f) CRITERION FILE
13 2.79991 13 1.73 C
14 2.89999 14 1.63 LINE OPERATOR SOURCES
15 3 15 1.59 1 SUM P2

2. Gain -shaper. Partial amplifier (a) is described by its network file (b). Frequency file (c) sets test frequencies in operating band, while data
file shows corresponding values of transistor gain. Variable file (d) establishes specification limits, performance file (e) defines amplifier gain
requirements, and criterion file (f) optimizes gain flatness.

Of the several procedures that could be used to de- maximum input resistance and minimum output resist-
sign the amplifier with Optinet, a very direct one is to ance nearly match at the high end of the operating fre-
first ignore input and output matching (since about the quency band if a series inductance is used to balance
same amount of gain is lost across the entire band due out reactance. It can also be seen that the gain rolloff,
to mismatching), and find approximate element values from a maximum of 9.4 dB at 1.6 GHz to a minimum of
for an interstage circuit that provides adequate control 4 dB at 3 GHz, is about 6 dB per octave.
over gain slope. Next, amplifier input and output im- Since there are two transistors, approximately 12 dB
pedances are computed and approximate element val- of gain rolloff must be introduced at the low end to flat-
ues for 50 -ohm matching networks are found. Finally, ten amplifier gain response. Thus, at least three reactive
the complete circuit is assembled, overall performance elements are needed to increase the gain control cir-
criteria are defined, and optimization is used to adjust cuit's reactance slope. Values will be chosen to provide
the various element values for best performance. the flattest and highest gain within the operating band.
The data in Fig. 1(b) indicates that the transistor's Figure (2a) illustrates a partial amplifier design and

138
Fig. 2(b) is its description. The network is partitioned tion multiplied by the appropriate data file entry.
into sections that correspond to lines in the network file The variable file of Fig. 2(d) specifies a set of variable
and elementary sections in Optinet's catalog of subcir- parameters whose values are to be adjusted for best per-
cuits. Elementary sections are represented by mnemon- formance: up to 15 numbers of a maximum of 25 stored
ics (like Si and L1) or numbers in the catalog. in common locations Cl to C25 can become variables.
The mnemonics are listed in the CATALOG column of These are listed (by number) in the COMMON column. A
network file N2 to indicate which sections are needed. range for each variable must be defined by listing its
Certain element values are required to complete the lower limit (NEG-BOUND) and upper limit (PosBouND).
section description. For example, Fig. 2(a) shows that Nominal variable values are called out in the VALUE
section 1 contains a voltage source (S1), described by a column.
source resistance (Rs), a reactance (xs), and a character- Variable file V presents the data for amplifier gain
istic impedance (zo). These elements are assigned ap- and the passive gain -leveling circuit elements shown in
propriate numerical values each time they are used, as Fig. 2(a). Line 1 calls for the 1.23-picofarad capacitor
done on the first line of the network file. stored in common location C9, line 2 the 4.3-nanohenry
Eight S parameters are required to completely de- inductor in C10, and line 4 the 16.6-nH inductor in C12.
scribe each transistor's performance at any frequency. Line 3 stores amplifier gain in common location C11
The S parameters are complex numbers forming a 2 -by - and confines its value to 12 ± 1 dB. Storage locations for
2 matrix, and are represented by mnemonics S 11M gain -control elements are also noted in the network file.
through S22P (M is for magnitude, P for phase). The Values of variables may be assigned or changed by
double -number designation fixes the S parameter's row making an appropriate entry in the variable file, as well
and column location in the matrix. For instance, S11M as in any file where the common location is referenced.
is the magnitude of the S parameter in the first row and Variables are required for all of Optinet's higher com-
first column of the matrix; S21P is the phase of the S puting functions, such as sensitivity analysis and optimi-
parameter in the second row and first column of the zation. During optimization, any changes in variable
matrix. Because each transistor is described by identical parameter values are readily examined by printing the
S parameters, only one set of eight must be stored; these variable file.
values go into data files D9 through D16. The performance file of Fig. 2(e) defines the proposed
Also included is a master numbering system for the network's performance requirements. Line 1, for in-
circuit elements. Every network file has four ELEMENT stance, requests the gain magnitude (GM) to be
columns, differentiated by numbers 1/5, 2/6, 3/7, and presented in the form of a plot versus frequency when-
4/8. A number is assigned to each element in a network ever the network is analyzed. The REFERENCE column
section. For any single section, Optinet can accept up to cites the nominal value of the variable of interest (gain
eight elements, the maximum required to completely magnitude in this case) and its storage location. A
describe any network section. From the network file of callout of 12C11, for example, refers to the variable
Fig. 2(b), it can be seen that section 2 comprises ele- listed on line 3 of the variable file; its nominal value is
ments 1 and 5 (1C1D9 and 105D13) through elements 12 and it's stored in common location C 11.
4 and 8 (1C4D12 and 1C8D16). The gain affects the criterion function because its
Section 3 of the partial amplifier calls for catalog sub - weight is positive (not zero). Weight value is assigned
circuit number 4, which is a parallel inductance (L), ca- by the user and reflects the relative importance of the
pacitance (C), and conductance (G). Because the con- quantity being computed.
ductance merely accounts for minor parasitic losses, it is Two other factors influence the weight value. One is
specified as 10v5 (noted as 1E-5 in Fortran) mho. the magnitude of the quantity, and the other is how of-
Program files. Network elements may be made vari- ten it is computed during a run. Gain, for example, is
able and/or frequency -dependent. Figure 2(c) shows a computed 15 times (because of the 15 frequencies in the
frequency file; it gives the frequency points to be eval- band) and has a high value of 10 to 15 dB. Reflection
uated and the data file for S21M. The latter represents coefficient, however, is found only once in the band and
the variation of element S21M over the frequency range has a low value of 0.1 to 0.2. To make these two quan-
called for in frequency file F2. Whenever network per- tities approximately equal in importance, the user could
formance is computed, the appropriate gain for S21M is assign a weight of 1 to the gain and a weight of 1,500 to
chosen from data file D13. If the frequency of interest is the reflection coefficient.
2 GHz on line 5 of the frequency file, corresponding gain Since gain is specified as a point -by -point measure
is 2.37 on line 5 of the data file. (Purr) in performance file P2, it contributes a value to
If an element value is stored in one of 25 special loca- the criterion function at each point in the operating fre-
tions, called common storage locations, it can be used quency band. This value is the WEIGHT (10) multiplied
for several elements in a network and/or it can be made by the square (sQ) of the difference between the gain
a variable parameter. To be stored in one of these loca- magnitude and the reference. The COMPARE column in
tions, a flag designated by a mnemonic Cl through C25 the performance file indicates what mathematical oper-
must immediately follow the element's numerical value. ation should be carried out. if

Element values stored in this way act as multipliers. Minimizing the criterion value, WEIGHT (GM-REF)2,
Data files can be treated similarly. If data files are will bring the gain closer to the reference at each point
called out as shown for the eight S parameters (1C1 in the band. Since the reference is a variable parameter,
through 1C8) in line 2 of network file N2, the element an optimization search can adjust it to a value that gives
value is the number stored in the selected common loca- the flattest gain curve.

139
 (a) FINAL AMPLIFIER DESIGN 17.4 nH

2.9 nH 2.36 nH 3.14 pF 1.4 pF 4 nH

-____mm_______4,..._____mers__ _II_ .
6.1 nH 3.7 nH

1 8 pF 3 nH 10 ' MHO 1.26 pF

35821E 35821E

i
(c) PERFORMANCE CHECK

(b) NETWORK FILE FORWARD NETWORK

AVERAGE VALUES
2

GM Ds 14.4017
RHM .137713

NO MAXIMUM VALUES

SECTION CATALOG ELEMENT1/5 ELEMENT2/6 ELEMENTS/7 ELEMENT4/8 GM 09 14.9977 AT F 2.79999 DMZ

RHM .214116 AT F 2.59999 0142

2 2 2.89173016 MINIMUM VALUES

3 11 1. 80025011
4 2 2.361160I8 GM DB 13.997 AT F 2.2 GHZ
RUM .015975 AT F 2 GHZ
5 12 3.0268IC19
6 1 3.13776C20 PROCEED WITH PLOTS? (Y OR (42( Y
7 05 1C1D9 1C2D10 1C3D11 1C4D12 GM DB 13.95 14.25 14.55 14.65 15.15 15.4'
105D13 1061114 1C7015 1C8D16 $ OHM 0 .0500001 .1 .15 .1
8 4 17.412C12 1.39919C9 1E-5
9 2 3.99244C10 I I I 1 1 I

1.6
10 65 109 1010 1011 1012 1. 7 ;;
1013 1D14 1015 1016 1.4 a :::::
11 2 6.0872C13 1.9
12 11 1.25798014 2 ...-- REFLECTION
13 2 3.71329C15
2.1 COEFFICIENT
FRE()
11Hz
2.2
2.3
2.39999
-i
K.:::- -

2.49999 14.
2.59999 .. 1
2.69999 GAIN ...
2.79999
2.69999
3 4r. ___

3. Optimum amplifier. Final optimization results in amplifier shown in (a) and described by network file (b). Optinet can also predict per-
formance of finished amplifier. Performance check (c) notes that minimum gain is 14 dB at 2.2 GHz, and maximum input reflection coefficient
is 0.22 at 2.6 GHz. Gain and reflection coefficients are plotted.

For optimization, only the first line of performance 2.2 GHz. To improve the reflection coefficient, another
file P2 is used, because the other lines have no weights. computer run would be necessary and some degrada-
When an analysis is run, the network gain magnitude is tion permitted in gain performance.
plotted, and network input resistance (R) and reactance With the initial part of the product design finished,
(x) are automatically computed and stored in files D1 the time and expense incurred thus far can be exam-
and D2 for use in solving the input matching problem. ined. Slightly more than 24 engineering hours in addi-
The criterion file of Fig. 2(f) gives the final require- tion to seven hours on a time-shared terminal are re-
ments for a function whose minimum represents best quired. A reasonable time -and -cost estimate for the
performance. Contributions from one or more perform- entire design task is about a week of engineering time
ance files can be combined to create the criterion func- and around $300 to $700 for computer time, including
tion in the way specified by the OPERATOR column. Pos- the investigation of alternative circuits.
sible operators are sum (sum), maximum (mAx), and The design is not complete until sensitivity and worst -
product (PRO). Sum and maximum are generally used case performance are checked. Optinet's sensitivity
for design purposes, while product is usually listed for analysis indicates the change in network performance as
worst -case analysis. Performance files to be combined to each variable circuit element is individually increased
form the criterion function are listed in the SOURCES by some small amount. Worst -case analysis predicts cir-
column. cuit performance if all the variables are changed simul-
First -order design. Now that the Optinet files needed taneously. For this amplifier, the key performance spec-
for a solution are established, the amplifier design prob- ifications are minimum gain and maximum input
lem can be run. Figure 2 contains all the files needed to reflection coefficient.
establish an interstage gain control circuit. For sensitivity analysis, these two performance mea-
Next, an output matching network can be found by sures are called for by altering performance file P2 as
computing input and output impedances across the op- shown in Fig. 4(a). The variables to be changed are the
erating band. The complete first -order design is then gain and circuit elements whose optimized values are
optimized to produce the best performance. indicated for the amplifier of Fig. 3(a). Nominal value
The final amplifier design, its network file, and a final of each variable will be increased by 1%.
performance check are shown in Fig. 3. As indicated in Figure 4(b) illustrates the results of the analysis (po-
Fig. 3(c), the amplifier's maximum reflection coefficient tentially troublesome areas are surrounded by boxes).
does not quite meet desired specifications; its value is Nominal values of gain and reflection coefficient are
0.216 at 2.6 GHz. Minimum gain, however, exceeds the printed, along with the incremental change that occurs
original design objective with a value of about 14 dB at in these quantities when a given variable is increased.

140
Numbers 1 through 11 correspond to the variables
listed on the same -number lines in the variable file. (a) PERFORMANCE FILE FOR SENSITIVITY ANALYSIS
A 1% increase in variable 1, the 1.4-pF capacitor in
Fig. 3(a), causes a 0.003 increase in the reflection P2
LINE MEASURE FORM WEIGHT REFERENCE COMPARE

coefficient and a 0.07 -dB increase in the gain. For this 1

2
CM
RHM
MIN
MAX

design example, an increase in any variable causing

more than a 1% change in a performance measure is (b) SENSITIVITY ANALYSIS RESULTS
considered potentially troublesome. Therefore, any in- FORWARD NETWORK 2

crement greater than 0.002 in the reflection coefficient PERFORMANCE

PARAMETER VALUE
INCREMENT OUE TO 12 CHANGE IN VARIABLE
1 2 3

and greater than 0.14 in gain is surrounded by a box in MAX RHM .216116 En= 5.02765E-4

Fig. 4(b). (Note that gain performance never goes out of MIN GM 13.997 .070467 .046E922 .0131348

specification; taken as a whole, the amplifier design is PERFORMANCE

PARAMETER VALUE
INCREMENT DUE TO 1% CHANGE IN VARIABLE
7

satisfactory.) MAX RHM .216116 .00141792 I.R0125E-4 -7.512E-4 MEMO

The cost of sensitivity analysis is usually nominal (un- MIN GM 13.997 .00505447 .00344276 -.004911772 -.

der $10 in this case). The analysis not only pinpoints the PERFORMANCE INCREMENT DUE TO 11 CHANGE IN VARIABLE
VALUE 9 10 11
PARAMETER
elements that may need to be adjusted in the final de- MAX RHM .216116 .00196241 8.47638E-4 .00172377
.0O244141 -.0E295639
sign but also helps to define acceptable tolerances on MIN CM 13.997 -.00534058

other elements. And it could be useful to overdesign a

circuit by using more than the minimum number of cir-
4. Sensitivity analysis. Performance of final amplifier is checked
cuit elements, thereby reducing all of their sensitivities.
for sensitivity to changes in individual variables. A 1% increase in
Worst -case analysis is the next step. Optinet can de-
termine what specifications will not be met, at what fre- each variable is introduced separately. Performance file (a) indicates
that gain and reflection coefficient should be tested. Results in (b)
quencies in the band, as well as for what combinations
have possible trouble spots noted with a box.
of component values and value tolerances. The neces-
sary files are given in Fig. 5(a); these must be used
along with the network file of Fig. 3(b). The worst -case performance study indicates the S
Worst -case analysis. For the amplifier example, the parameters that should be avoided at specific fre-
worst -case criterion function must go to zero if the de- quencies for acceptable amplifier performance. Vari-
sired specifications are not met at any point in the band. ations in transistor parameters generally are compen-
The variables in the variable file are normalized values sated for by adjusting circuit elements, and Optinet can
of the eight S parameters that characterize each transis- determine appropriate ranges and parameters.
tor. The boundary values shown reflect a ± 10% toler- Another application for the Optinet system is inter-
pretation of measured can deter-
ance as specified by the manufacturer.
Performance file P2 places an upper bound (UB) on mine the values of the variable model parameters that
minimum gain to assure that the first term in the crite- provide the best fit between computer and measured
rion function will be 0 if the gain falls below 13 dB. Line data. Using this feature requires that a bench model of
2 of the file sets a lower bound (LB) for the maximum the Optinet-designed amplifier be built and tested.
reflection coefficient to make certain that the second Because of tolerance ranges in purchased compo-
term of the criterion function is 0 if the maximum re- nents, it's highly unlikely that the test data for the pro-
flection coefficient in the band exceeds 0.33. The prod- totype will accurately match predicted performance.
uct operator (FRO) in the criterion file sets the criterion But Optinet accepts the measured performance data
function equal to the product of the terms in file P2. and finds the new variable values needed to make mea-
When the worst -case performance objectives defined sured performance match the computed. The program
by the performance file are exceeded, Optinet termi- indicates those parts of the circuit responsible for any
nates the search and prints the results as shown in Fig. discrepancy and notes corrective action required.
Implementing this Optinet feature involves using the
5(b). The normalized S parameters that cause worst -case
performance are listed in the VALUE column of a new same files that are employed for design. However, the
variable file. (Note that none of the worst -case S para- measured data is stored in data files to establish ampli-
meter values exceed their specified tolerance.) . fier performance for each frequency in the band. These
According to the analysis, the reflection coefficient data files are then listed in the REFERENCE column of a
will be 0.346 at 3 GHz and, therefore, not within specifi- performance file. Optinet will automatically adjust cir-
cation. Minimum gain, however, stays within design ob- cuit element values to achieve the best match for the
jectives -its value is 13.9 dB at 1.8 GHz. Worst -case anal- reference (measured) data. The user then can determine
ysis always identifies the frequency with the poorest those elements that require adjustment and by what
performance. amount.
Production modeling. Transferring a product from
When the limit for the maximum reflection
coefficient is changed to 0.5, gain variation is empha- development to production usually causes many prob-
sized. Running through worst -case analysis again yields lems. Some can be reduced or even eliminated if both
the results shown in Fig. 7(c) and a different set of S - an Optinet model and prototype are available to pro-
parameter values. Gain will be just over 13 dB at 2.2 duction personnel.
GHz and still within specification. Reflection coefficient A study of the prototype quickly shows what toler-
also meets the desired limit with a value of 0.346 at 3 ance must be held and what performance changes oc-
GI -1z. Cost for this worst -case analysis is about $55. cur, with certain circuit element variations. As pilot

141
 models are built and checked. Optinet can interpret test
data for the adjustment of the production process to
(a) FILES DEFINING WORST -CASE ANALYSIS provide target performance. Optinet also can even con-
V
tinuously monitor production so that drifts can be cor-
LINE COMMON NEC. -ROUND Pm -Boutin
rected before the product goes out of specification.
VALUF
1 1 .9 1.1 1
Optinet also can compensate for variations in state-
2
3 3
.9
.9
1.1 1 of-the-art components that may be the only devices
1.1
4 .9 1.1 1
available for a given production run. With an Optinet
5
6
5
6
.9 1.1 production model of the whole product, data on the
.9
7 7
1.1
1.1
doubtful parts can be entered into the program. Optinet
8 8 .9 1.1 then will investigate the corrective action possible
through controlled changes in other circuit parameters.
P2 Another consideration in production modeling is an-
LINE
1
MEASURE. FORM WEIsHT RFFFI'EMCF
Oft MIN
COMPARE swering requests for quotations that involve some de-
1 13 UR
2 RHM MAX 100 .33 LP
viation from standard performance. It may be difficult
to determine quickly whether these special specifica-
C tions can be met on the production line. With Optinet,
LINE OPERATOR SOURCES the special specifications can be used in the optimiza-
1 PRO P2
tion routine. Knowing the best values for the elements
and the best performance obtainable can greatly im-
prove bidding accuracy and production planning.
(b) WORST -CASE ANALYSIS RESULTS Marketing. The final step in developing a new prod-
V
uct is marketing. Through its public data file, Optinet
LINE. COMMON NES-ROUND
can make this task easier, as well as reduce time and ex-
POS-ROUND VALUE
1 1 .9 1.1 .999343 pense for both seller and buyer. Since the program is
2
3
2
3
.9
.9
1.1 .945808 available through a time-shared network, any sub-
1.1
4 .9 1.1
1.0627
.949303
scriber has easy dial -up access to the public file and,
5 5 .9 1.1 .990329 therefore, to the Optinet product model. A potential
6
7 7
.9
.9
1.1
1.1
.990838 customer can explore the properties of the product or
1.02744
8 8 .9 1.1 1.04C263
even run through a complete design, as was done here
with the Hewlett-Packard transistor.
FORWARD NETWORK 2 Instead of spending months exchanging samples,
MAXIMUM VALUES
building prototypes, and trying to agree on mutually ac-
RHM = .346239 AT F = 3 nPz ceptable product performance, the manufacturer and
MINIMUM VALUES his customer can trim this exploratory period to a mat-
ter of weeks. Costs, too, are dramatically reduced be-
nm DR = 13.9353 AT F = 1.8 cause less engineering time is needed.
nm7
Suppose a conventional sample -and -build period is
about three months. Engineering costs, including sup-
(c) MODIFIED WORST -CASE ANALYSIS port and overhead expenses, normally would be about
V
$10,500. Optinet could reduce the time factor to two
LIME rommor! tiEn-snumn PrIs-ROUND VALUE
weeks, dropping engineering costs down to $1,750. Add-
1
2
1 .9 1.1 1.04318 ing about $700 for computer time brings the total ex-
2 .9
3 3 .9
1.1 .915040 penditure to $2,450, a significant saving.
1.1 1.07918
4 4 .9 1.1 1.03349 Since Optinet's public data file has the same capacity
5
6
5
6
.9 1.1 .912443 as its working file, there is ample space to store product
.9 1.1 .904025
7 7 .9 1.1 1.00344
data, frequency ranges, performance properties, and
8 8 .9 1.1 .921671 component tolerances. The amplifier example could be
FORWARD NETWORK 2
represented explicitly by its network file description. Or
MAXIMUM VALUES the circuit's two -port performance may be given in
RHM = .324711 PT F = 3 sul
terms of S parameters or any other equivalent nu-
merical description.
mINIMUM VALUES If an Optinet subscriber already has a model in his
GM PR = 13.0834 AT F = 2.2 C.M7
private files, it can be transferred to the public disk
without charge and stored for $10 to $20 per month.
The public file can be updated from subscribers' private
files at any time.
As the Optinet system grows in scope, the catalog of
5. Worst -case analysis. If all circuit variables are changed at the products in its public data file also will grow. It should
same time, amplifier performance can be checked for worst -case be remembered, however, that the program's most up-
conditions. Files (a) define worst -case gain (GM) and reflection to-date source of product information is its directory,
coefficient (RHM) when transistor S parameters are varied by which can be listed simply by giving a command during
± 10%. Printouts (b) and (c) show results. an Optinet terminal session.

142
Program calculates load effects
of high-speed digital circuits
Digital interconnections are analyzed by program called Line as if
they were transmission lines; from a minimum of,input data it
accurately computes line voltage, plots ringing characteristics

by Laurence P. Flora, Burroughs Corp., Pasadena, Calif.

High-speed digital circuitry often lands its designer where:

with the job of analyzing the loading effects, like circuit K, = (R2R3 - ZoR, -.ZoR3)/(R2R3 + ZoR, + ZoR3)
ringing and delay, that result from line discontinuities. and:
This analysis is highly complex, and requires a com- K2 = (R,R, + R2R3 - ZoR, - ZoR, - ZoR3)/
puter's speed and accuracy. (R,R3 + R2R3 + ZoR, + Zolt, + ZoR3)
However, it has to be performed in terms of trans- and:
mission line equations, and ECAP (electronic circuit r = Ci(R,R,R, + ZoR5R2 + ZoR5R3)/(R,R3
analysis program) and its equivalents are really only + R2R3 + ZoR, + ZoR, + ZoR3)
satisfactory for standard lumped -constant circuitry. A and Zo is the line output impedance.
lumped -element model doesn't fit transmission lines too Now the solution can be generalized, and compu-
accurately. Besides, the programs take an extremely tation time minimized. Since the reflection voltage may
long time to solve even moderately complex problems assume an infinite number of waveforms, the final solu-
of this nature. tion must be some approximation that produces accu-
To help the engineer plot an accurate analysis of a rate results from a minimum of information. With such
high-speed transmission line, a program called Line was an approximation, comparatively little need be known
recently developed. It's based on a general load config- about the line's immediate past history, and the com-
uration that represents all circuits to be used in the sys- plexity due to multiple reflections is minimized.
tems under study. It also is based on what the general Of course, any waveform can be approximated as a
load effects are when a unit step or unit ramp function series of small steps. But a better approach in this con-
is transmitted down the line, and closely approximates text is to use a series of small ramps, since this is nearly
any reflection without reference to the line's history an exact approximation of a pulse with a finite risetime
from time zero. The program, which is written in For- and falltime - the usual shape of waveforms trans-
tran, is conversational, and can be run on a time-shared mitted over lines in high-speed digital systems.
terminal. In contrast to existing computer programs, Assume that the input waveform is a ramp with a
Line can precisely analyze a transmission line problem, slope of A volts per second up to time T1. Then, at T
but it cannot handle a general circuit consisting of vari- the slope increases by an amount B, which gives a new
ous lumped components. slope of A + B. By superposition, reflection voltage is:
Line makes use of the accurate, yet general, load con- Vr = A[Va].,U(t) + B[Vr(t - TilrampU(t - T1)
figuration of Fig. 1. Besides being an adequate repre- where U(t) is a unit step function, and U(t - T5) a de-
sentation of the circuits used in most systems, this RC layed unit step.
network may also model a purely resistive discontinuity The forward voltage at time T which is represented
(i.e., when R, = 00) or purely capacitive discontinuity
(R5 = 00, R2 = 0, and R3 = co).
Computing the reflection
To understand how Line solves for general load ef-
fects, consider the response of a line to a unit step or a
unit ramp function. These two forcing functions are
chosen since either provides a good approximation for
most input waveforms.
For the load of Fig. 1, the reflection voltage due to a
unit step function is:
[Vr(t)]step = K2 + (K, - K2)e-tir (1)
while a reflection voltage due to a unit ramp function
can be expressed as: 1. General load. RC network accurately represents most loaded
[Vr(t)]ramp = K2t + r(K,) - 1(2)(1 - e-t/r) (2) transmission lines in high-speed digital systems. When R2 - cc, it is
Now on educational leave at Massachusetts Institute of Technology, Cambridge, Mass. resistive; when 13, 00, R2 - 0 and R, oo, it is capacitive.

143
 ENTER

REQUEST COMMAND
READ AL STRING

WHAT IS NEXT
WORD ?

NO MORE ADD CHANGE LOAD RUN EXECUTE PLOT

WORDS AND LOADS COMPUTE GRAPH

PRINT ERROR
MESSAGE

READ LINE IMPEDANCE, ARE DISCONTINLI I YES READ DISCONTINUITIES,

PROPAGATION VELOCITY CONSTANTS STORE, MAX AND MIN VOLTAGES

NO

READ CALCULATION TIME READ CONSTANTS PLACE PLOTTING

INTERVAL, LINE LENGTH CHARACTER IN OUTPUT
ARRAY FOR EACH VOLT-
AGE TO BE PLOTTED
YES READ TIME LIMIT,
IS LINE TOO LONG
INITIAL LINE VOLTAGE, PRINT ARRAY
GENERATOR VOLTAGE
NO
REPEAT FOR
PRINT FP OR EACH
STORE CONSTANTS VOLTAGE
V

DOES SEGMENT CONTAIN NO

DISCONTINUITY?

[READ GENERATOR YES

IMPEDANCE
FIND NEW INCIDENT NEW INCIDENT
VOLTAGE FROM
VOLTAGE = INCIDENT
REFLECTION CONSTANTS
AND DATA FOR VOLTAGE OF
IND GENERATOR ADJACENT SEGMENT ADJACENT SEGMENT
REFLECTION CONSTANTS

STORE INCIDENT VOLTAGES

READ DISCONTINUITY AT THIS DISCONTINUITY
DATA

FIND REFLECTION ADD LEFT AND RIGHT

CONSTANTS FOR EACH INCIDENT VOLTAGES
DISCONTINUITY

STORE TOT AL VOLTAGE

REPEAT FOR
EACH
SEGMENT

2. Flow chart. Once initial data iS entered, Line computes incident voltage and reflection constants for each line segment, storing only inci-
dent and transmitted voltages from previous adjacent segment. There are three basic instructions: LOAD (describing physical line proper-
ties), RUN (establishing initial electrical conditions), and PLOT (graphing reflection voltage at several load points).

144
 any waveform, including a step with a fast risetime.
The instruction sequence for Line is shown in the
flow diagram of Fig. 2. To start the program, the com-
mand RUN LINE (or equivalent instruction, depending
on the time-sharing system used) is used. After this,
there are three operating instructions that can be re-
peated as many times as desired.
Operating instructions
The first of these is LOAD or LOADS, which allows the
user to describe the line's physical properties. For
proper operation, Line requires the line impedance, the
3. Input voltage. Line accepts input waveforms as piecewise linear propagation velocity on the line, how often (at what
functions with up to 20 breakpoints. For example, the voltage wave- time intervals) the voltages should be computed, the to-
form shown is described by an initial value (0,0) and four break- tal length of the line, the impedance of the driving gen-
points with coordinate locations of (1,3), (4,3), (5,0), and (10,0). erator, and the positions and values of loads on the line.
To describe a load for Line, load position (distance
by V, (T,), can be expressed as: from the generator), and the values of R R2, R, and C,
must be entered. Up to 19 loads can be accommodated.
Vf(T2) = A [K2T, r (Ki - K2) (1 - e
Constants representing the loads are saved in a disk file.
Any consistent set of units may be used. For instance,
the system of ohms, feet, seconds, and farads is accept-
And the final equation for the reflected voltage is: able, and velocity would then be entered in feet per sec-
V. = V,(Ti)e-th'r + (A + B)[Vr(c10],amp ond. But this system may be cumbersome, and a more
+ (ATIVr(c101.,ep - (AT,)(K, + 1)e -"'r (3) convenient set of units is kilohms, inches, nanoseconds,
where: and picofarads. Velocity is then expressed in inches per
V,(T,) = forward voltage at the end of the last time nanosecond. (It should be noted that the unit of capaci-
interval tance multiplied by the unit of resistance must be con-
Vr(dt)ramp = the unit ramp reflection of Eq. 2 sistent with the unit of time.)
A + B = the new ramp slope The second operating instruction describes the inci-
Vr(dt)step = the unit step reflection of Eq. 1 dent voltage waveform, and uses the command RUN,
AT, = the incident voltage at the end of the last time EXECUTE, R, E, or COMPUTE. For this computation, Line
interval requests the maximum time for which the voltages are
dt = the time interval. to be computed, the initial line voltage, and the break-
Note that only two pieces of information about the points of the input voltage waveform. For instance, an
line's history need be saved for the next calculation-the input voltage like the one illustrated in Fig. 3 would be
incident voltage, AT and the transmitted voltage, described by initial voltage (0, 0) and breakpoints (1,3),
V,(T,), at the end of the last time interval. (4,3), (5,0), and (10,0).
The time at which the last breakpoint occurs must be
Using the results equal to or greater than the maximum time for which
To program Eq. 3 into a computer, the values of cir- the voltages are computed. When this last breakpoint is
cuit variables R R2, R2, and C or mathematical vari- entered, Line calculates the voltage waveforms at every
ables K1, K.2, and r must be stored. Also, the input load and writes them in a disk file.
waveform must be represented in memory, for example, If one of the commands describing the incident volt-
by a table of voltages or by an equation. age is given before the physical properties of the line
The reflected and transmitted voltages are calculated have been established (through a LOAD command), an
at each discontinuity, but need to be stored for only one error message is typed. When this occurs, the command
time interval. However, it is convenient to save the RUN LINE must be entered again.
transmitted voltage for each interval since the instan- The third and last operating instruction, PLOT or
taneous line voltage is the sum of two transmitted volt- GRAPH, causes Line to graph the voltage at one or more
ages (one for each direction). Some provision should of the load points. If several separate plots are desired,
also be made for the distance between loads so that the the command must be given several times. Line prints
transmitted voltage (plus the reflected voltage due to each load's number and position, and asks for the num-
the voltage incident in the opposite direction) reaches ber of waveforms the user wants plotted, plus the num-
the next load after a suitable time delay. bers of those loads for which plots are desired.
The program Line reads the values of R,, R2, R2, andAlong with the load numbers, Line lists load posi-
C and calculates and stores K K2, and r. The input tions, and the typewriter character with which each load
waveform is described by a piecewise linear function will be plotted. Finally, the program asks for the desired
that has up to 20 breakpoints. Total line voltage at each minimum and maximum voltage scale for each plot. All
discontinuity is also stored and can be plotted. The plots are superimposed. If either the PLOT or GRAPH
computation interval is determined from the resolution command is given before the voltage waveforms have
requested by the user; it is the output time interval di- been computed, an error message is printed, and the
vided by 10. This allows a very close approximation of command RUN LINE must be given again.

145
 RUN LINE
RUNNING

WHAT IS YOUR COMMAND?ADD LOADS .50002 01 .7500S

0.
I I I
WHAT IS THE LINE IMPEDANCE?.100
.1000B 01:
WHAT IS THE PROPAGATION VELOCITY?8 .2000E 01:
.30002 01:
AT WHAT TIME INTERVALS DO YOU WANT THE VOLTAGES?1 .4000E 01: &
.5000E 01:
HOW LONG IS THE LINE?60 .60002 01:

:gggE gl:: &

THE LINE WILL BE BROKEN INTO SEGMENTS .800E 00 LONG. &
.9000E 01 *
WHAT IS THE IMPEDANCE OF THE GENERATOR?.020 &
.10002 02
.1100E 02 &
DISCONTINUITIES ARE OF THE FORM R1 IN PARALLEL WITH &
.1200E 02 :
C, ALL IN SERIES WITH R2, ALL IN PARALLEL WITH R3. &
.13002 02
INPUT LOAD POSITION, C, R1, R2, AND R3.
,1400E 02
INPUT 5 ZEROS IF NO MORE DISCONTINUITIES.
.1500E 02
?1,0,1000,1000,1000
.1600E 02
?12,10,1000,.100,1.0
?24,5,1000,.200,2.0 .1700E 02
.18002 02
?36,15,1000,.067,.667
.1900E 02
?48,5,1000,.2,2 02
760,10,1000,.1,1 .20002
.21002 02
?0,0,0,0,0
.22002 02
WHAT IS YOUR COMMAND?RUN ,2300E 02
.24002 02
.25002 02
VOLTAGES WILL BE COMPUTED FOR EVERY *
.1002 01 INCREMENT IN TIME. .2600E 02
.27002 02 * &
WHAT IS THE MAXIMUM TIME?100
.28002 02
WHAT IS THE INITIAL LINE VOLTAGE?0 .2900E 02 &
.30002 02 * & :

INPUT WAVEFORM BREAKPOINTS BY ENTERING .31002 02 *A :

TIME,VOLTAGE .32002 02 *:
.33002 02'
75,5
?100,5 .3400E $4*
o2 :

THE VOLTAGES AT THE DISCONTINUITIES HAVE .36002 02 : & *

.3700E 02 : & *
BEEN WRITTEN IN THE OUTPUT FILE LINEV. .3800E 02 : & *
.39002 02 : & *
WHAT IS YOUR COMMAND?PLOT AND STOP.
.4000E 02 : &
NUMBER 1
.41002 02 : & *
2 3 4 5 6
POSITION 0.80 12.00 24.00 36.00 48.00 60.00 .4200E 02 : &
.4300E 02 & *
HOW MANY VOLTAGES DO YOU WANT PLOTTED?3 02 &
.4400E :

.4500E 02 : & *
INPUT THE NUMBERS OF THE DISCONTINUITIES 02
.4600E : &
AT WHICH THE VOLTAGES ARE TO BE PLOTTED. .47002 02 : & *
?1,3,6
.48002 02 :&*
DISCONTINUITY PLOTTING .4900E 02 :&*
NUMBER POSITION CHARACTER .50002 02 :&
1 .800E 00 .5100E 02
3 .240E 02 .52002 02
6 .600E 02 .5300E 02
WHAT ARE THE MINIMUM AND MAXIMUM VOLTAGES?0,10 02 &:
.55002 02 'A.
.5600. 02 '4:
'4:
:Nga g
4. Line example. Sample program and plot shows total line voltage 02.5900E 02 'A:

versus time for three points of discontinuity along a line. Different .6100E 02 *&:
.6200E 02 *4:
typewriter characters represent the three ringing waveforms, which .63002 02 *4:
.6400E 02 *&:
are plotted at 1 -nanosecond intervals. .6500E 02 *4:
.66002 02 *4:
.67002 02 *:
A RUN command may be given several times, as can .6800E 02
.69002 02
the PLOT command. RUN causes Line to find the re- .7000E 02 *2
*:
.71002'02
sponse of the loaded line to different input waveforms. .7200E 02
.7300E 02
Either STOP or END terminates the program. .7400E.20 02
It is also possible to obtain the voltage response at .7600E
02
02
some point on the Line where there is no load, without .77002
.78002
02
02 :1
affecting voltage at that point. The user simply inserts a .79002
.80002
02
02
:&
;&
load where desired by letting R2 and R3 be several or- .81002 02 :&
.8200E 02 :&
ders of magnitude larger than the line impedance. .83002 02 :&
To make Line's commands more readable, punc- .8400E
.8500E
02
02
:&
:&
tuation and the words ADD, CHANGE and AND may be .86002
.87002
02
02
used. Then a simple command to completely describe a .8800E
.89002
02
02
problem, compute the waveforms, obtain a plot, and .9000E
.9100E
02
02
stop would be: ADD LOADS, RUN, PLOT AND STOP. .92002 02
.9300E 02
Figure 4 is a sample run of the program to find the .9400E 02
ringing characteristics of a line at three points of dis- .95002
.96002
02
02
continuity for 100 nanoseconds. Total line voltage is .97002
.98002
02
02
plotted at 1-ns intervals along the line. The program re- .99002
.1000E
02
03
stricts minimum and maximum voltage excursions to b
and 10 volts, respectively, if the line voltage exceeds END LINE 1 MIN, 33.4 SEC
these limits. 0
Printed copies of Line are available at $0.75 each from the author, Laurence P. Flora, 40
Highland Ave., Cambridge, Mass. 02139.

146
Computer helps design of
complementary MOS logic
Accurate characterization of MOSFET current -voltage curve permits computer
modeling of silicon -gate C/MOS logic circuits; switching performance of both
combinational and sequential logic circuitry can be predicted or analyzed

by James Foltz and Fuad Musa, Motorola Semiconductor Products Inc., Phoenix, Ariz.

Complementary metal -oxide -semiconductor logic latter capacitances is constant, their individual values
circuitry is being more widely used as the demand in- vary with biasing conditions and also depend on the di-
creases for digital systems that operate at low power lev- rection of current flow between drain and source termi-
els and from low supply voltages. ctmos satisfies these nals. Detailed descriptions of how to evaluate mos ca-
requirements with its extremely low dynamic power dis- pacitances are available.1,2,3,4
sipation and its essentially zero quiescent power con- This model does not account for any source or drain
sumption. Moreover, silicon -gate cimos can operate series resistances because the effect of these resistances
from very low supply voltages-as little as 1 volt, can be neglected, due to the extremely low current de-
thereby further reducing power requirements. veloped by the MOSFET at silicon -gate cimos operating
Complementary logic structures, however, can be- voltage levels. (Peak current is about 100 microamperes,
come quite complex, making the use of computer -aided while drain -source voltage approximately equals the
analysis essential, since the basic cimos logic element, 1.5-v gate -source voltage).
the inverter, requires two devices-an n -channel and a Characterizing the model
p -channel mos field-effect transistor. To use the com-
puter, a practical model of the silicon -gate MOSFET is In the linear region, the current generator, IDS, can be
needed that satisfies cimOS requirements. written as: 5,6,7
The model must be suitable for both transient analy- IDS = Bo[(VG - Vo - VD/2)VD -(VG - Vo - VS/2)Vs
sis and determining the effect of mos device pa- -(2/3)KAIVD 20F13/2-IVS 2ciF13/2)] ( I)
rameters, especially threshold voltage, on circuit per- where all voltages are referenced to substrate potential:
formance. As threshold voltage becomes smaller, oper- B. = Zp.C./L
ating speed can be increased, but possibly at the price of Vo = OMs - Qss/Co + 24 F
higher power consumption.
The MOSFET model
Because of its simplicity and accuracy, the low -fre-
quency large -signal equivalent circuit of the MOSFET
(Fig. 1) provides a good basis for developing the appro-
priate equations for cimos computer analysis. Through
Sceptre, a general-purpose computer program, and
MOSFET current and voltage relationships, this equiva-
lent circuit can be used to predict cimos switching
speed, power dissipation, and the effect of threshold
voltage on transient response and power -speed product.
The MOSFET model is a four -terminal device consist-
ing of drain -source current generator IDS and five ca-
pacitors that represent the device's gate -source CGS,
gate -drain COD, gate -substrate COB, drain -substrate
CDB, and source -substrate CSB capacitances. For an re-
channel MOSFET, current flows from the drain terminal
to the source terminal; for a p -channel transistor, cur-
rent flows from source to drain.
Model capacitors CDB and CSB represent junction ca-
pacitances and can be evaluated readily by conven- 1. MOSFET model. Low -frequency large -signal equivalent circuit of
tional methods. However, since CGS, COD, and CGB are MOSFET serves as model for complementary MOS logic. Equation
mos gate -oxide capacitances, they are functions of pro- for current generator, 'Ds, represents model's current -voltage char-
cessing parameters and device geometry, as well as acteristic, which can be applied to C/MOS design or analysis. Cur-
MOSFET terminal voltages. Although the sum of these rent flows from drain to source for n -channel device.

147
 n -CHANNEL (n11
160 VT ", 0.5 V Vc, 1.8 V x
-160

140 -140

120 VG = 1 6 V
-120

100 -100
<c
VG = 1.4 V
3 80 -80
O
60 -60
VG = 1.2 V
X X
40 - 40
VG = 1 v
20 -20
1 l I

0 0.5 1.0 1.5 2.0 0 -1.0 -1.5

Vos (V) Vos (VI

OPTIMIZED PARAMETERS

Device Bo Vo ND X0 C1SS /CO L VT

type (A/V2) (V) (cm -3) (microns) (V) (microns) IV)

n1 2.85 X 10-4 -0.57 4.3 X 1015 0.117 0.27 3.9 0.48

p1 -1.4 X 10 4 0.24 1.1 X 1015 0 115 0.30 5.2 -0.25

n2 8.9 X 10 4 -0.68 1.12 X 1016 0.122 0.24 20.0 1.20

p2 -3.49 X 10 4 0.14 4.72 X 1015 0.122 0.33 25.0 -1.0

2. Mathematical versus practical. Device parameters needed to compute model current IDs, are noted in table for small (nl and pl) and
large (n2 and p2) transistors. Parameters are mathematically optimized to approximate the current -voltage characteristics of real devices. I -V
curves for nl and p1 MOSFETs embody the comparison of predicted results (solid lines) with measured data (test points).

VDD
2.25 - 2.2501 00

2.0 - 2.0000 00

IN
1.75 - 1.75o1 00 MEASURED

1.5 - I.500E 00

1.25 - 12501.0

0
1.0 - 1.000E 00

VDUT
0.75 - 7.500E-02 (0.5 V/cm, 2 µs/cm)

0.5 - 5.0000-01

0.25 - 2.5000-02

0 - 0.0

-0.25 -
2.0001 02 6.0000 02 1.0000 03 1.404E 03 1.0000 03
0.0 4.000E 02 0.0000 02 1.200E 03 1.0000 03 2.0002 03

I I I I I I I I I I I

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

TIME (µs)
(a) (b) PREDICTED

148
 Ko = (2Eoes6N01/2/C0
VG is the gate voltage, VD the drain voltage, Vs the
source voltage, Z the channel width, L the channel
length, it the inversion layer mobility, Co the gate -oxide
capacitance per unit area,. Eo the permittivity of free
space, Es the relative dielectric constant of silicon, ND
the substrate impurity concentration, Ords = he meta
semiconductor work function difference, Qss the fixed
positive charge at the silicon/silicon-oxide interface, OF
the Fermi substrate potential, and e the electron charge.
In the saturation region, VDSAT replaces VD in Eq. 1
and is defined by:5
VDSAT = Va - Vo 410/2[1 -(1 ± 4(Va
- Vo + 24)/K02)1/211/2 (2)
(In Eq. 2, where a choice is indicated, the sign is + for
n -channel and - for p -channel.) To simplify compu-
tations, the channel conductance in the linear region,
gds, can be regarded as a linear function of VG, provided
that Vs = 0 and VD) is much less than 241F:
dins = Bo(V - VT)
gds =
UV D
V D = constant
where threshold voltage VT is:
VT = Vo K0(2001/2
When channel conductance gds is plotted against gate
voltage VG for an actual mos device, the value of vari-
able Bo (slope) becomes smaller with increasing gate
voltages. This is due to reduced surface mobility and to
the series source resistance, which lowers the gate -
source voltage below the applied gate voltage. For all
silicon -gate MOSFETs, variable Bo can be approximated
by a constant when VG is less than or equal to 2 v.
Modeling C/MOS logic
A computer -aided technique can be developed to op-
timize the values of Qss, Vo, oxide thickness Xo, channel
length L, and impurity concentration ND so that the
current -voltage relationships calculated with Eqs. 1 and
2 closely approximate those measured for practical sili-
con -gate cimos devices.
The table in Fig. 2 lists the most important optimized
parameters for two pairs of n -channel and p-channel
MOSFETS. The upper two sets of values, n1 and p1, are
for small transistors, while the lower two, n2 and p2, are
for large transistors. Both the predicted (with the
model's equations) and the actual (from existing transis-
tors) current -voltage characteristics of devices n1 and pl
also appear in Fig. 2. The differences between the pre-
dicted and the measured curves are small.
The basic element of the ctmos logic family is the in-
verter, which is shown in Fig. 3a along with its transfer
characteristic. Figure 3b shows the inverter's predicted
and measured switching performance. The printout re-
flects the computed switching speed, based on the opti-
mized parameters of transistors n1 and p 1. And the os-
cilloscope trace displays the switching action of a circuit

3. C/MOS logic Inverter. Basic C/MOS logic element is inverter (a)

comprising p -enhancement MOSFET (top) and n -enhancement
MOSFET (bottom). Voltage transfer function shows transition from 4. Minimizing device count. Flip-flop (a) requires 20 MOSFETs, as
logic to logic 0. Computer plot (b) of output voltage versus time
1
shown in (b). But three transistors are redundant and can be re-
projects inverter switching performance using n1 and p1 param- moved-one between either 01 or 02, another between 04 or Q5,
eters. Scope display is for inverter built with n1 and pl MOSFETs. and the third between Qg or 07. Final flip-flop (c) uses 17 MOSFETs.

149
 1.75 --- 1.750+00
1 1

1.5 --- 1.500+00

1
1
1
1
1.25 --- 1.250+00 1-- MEASURED
1
1
1
1
TRIGGER
1.0 --- 1.000+00 1--

1
1
1
0.75 --- 7.500-01 1-- Q OUTPUT
1
1
1
1
0.5 --- 5.000-01 1-- (0.5 V/cm, 10 ps/cm)
1
1
1
0.25 - 2.500-01 1--
1
1
1
1
0 --- 0.000 1--
1
1
1
1
-0.25 --- -2.500-01 1--
1
1
1
1
--- -5.000-01 1--
1

-0.75 - -7.500-01
1
1
1

1.000+04
1 1

3.000+04 5.000+04 7.000+04 9.000+04

0.000 2.000+04 4.000+04 6.000+04 8.000+04 1.000+05

0 10 20 30 40 50 60 70 80 90 100

TIME (As)

PREDICTED

5a. Flip-flop Q output. Predicted switching performance is plotted by computer for O output of toggle flip-flop that is simulated with n2 and
p2 parameters of Fig. 2. Oscilloscope trace displays same 0 output for actual flip-flop built with practical n2 and p2 MOSFETs.

built with actual nl and pl transistors. Each switching speed product, which is defined as VDDIDD/frequency,
waveform is the voltage across load capacitor CL (2 pi- may not necessarily be constant with frequency if the
cofarads) when the inverter is operated from a 1.5-v dc input signal is sinusoidal or has varying rise and fall
supply and is driven by a 1.5-v input having 100 - times. The MOSFET model can be used to predict the
nanosecond rise and fall times. power -speed product by employing a function subrou-
Power -speed product tine in conjunction with Sceptres to calculate the aver-
age inverter current drain, IDD
One of the most important performance parameters For example, if an inverter's sum of threshold volt-
of any switching circuit is its power -speed product; ages is about half a 1.5-v supply, its power -speed prod-
power consumption and operating speed are directly uct goes up by a factor of four when input rise and fall
proportional to each other. Here, the c/mos inverter times are increased from about 0.1 microsecond to
has the advantage of dissipating power only during its around 5 /is. The power -speed product for an inverter
transition, when charging or discharging its load and in- having a sum of threshold voltages that is greater than
ternal capacitances. However, I2R power may be dis- the supply does not change with slower inputs.
sipated if both MOSFETs conduct simultaneously during
transition, when the supply voltage and the input signal Sequential logic circuits
amplitude exceed the sum of the n- and p -channel In addition to combinational logic, sequential ctmos
threshold voltages. Dissipated power then becomes a logic circuit performance can also be predicted with the
function of input rise and fall times and increases with MOSFET equivalent circuit. Sequential circuits can be
slower signals. Therefore, the ctmos inverter's power - considered as interconnected combinational logic

150
 2.25 - 2.250+00
1
1
1 1 1 1 1
1

2.0 - 2.000+00
1
1

1
1 1
1 1
1
1.75 -- 1.750+00 1--
1
1 1

1,5 - 1.500+00
1
1
1--
1
1 frfr
.1,11
1

1
1
1.25 --- 1.250+00 1--
1
1

.1
1
1
1.0 - 1.000+00
1
1

-
1
1 1
1 1
0.75 7.500-01 1-- MEASURED
1
1
1 TRIGGER *It
1
0.5 5.000-01 .01 --1
1

,
1

0.25 - 2.500-01
1
1
I__
1
1
Q OUTPUT
.
...
--I
1
1
**
1 (0.5 V/cm, 10 ps/cm) 1
1 1
0 - 0.000 ..* ******** .., ....1
1 1
1 1

-0.25 - -2.500-01
1
1
1
1

1.000+04
1

3.000+04
1

5.000+04
1 1

7.000+04
1 1

9.000+04
1

0.000 2.000+04 4.000+04 6.000+04 8.000+04 1.000+05

0 10 20 30 40 50 60 70 80 90 100

TIME ( us)

PREDICTED

5b. Flip-flop a output. Computer predicts switching action of flip-flop a output by plotting output voltage versus time. Again, device para-
meters of n2 and p2 MOSFETs are used to simulate flip-flop. Switching performance of practical flip-flop is shown by scope trace.

blocks, like gates and inverters, with appropriate feed- The MOSFET model can be employed to predict the
back. A toggle (type T) flip-flop, for example, can be switching performance of the toggle flip-flop, as was
implemented with the blocks noted in Fig. 4a. done for the inverter. The parameters of MOSFET5 n2
In this form, the flip-flop requires 20 transistors, as in- and p2 can be used for computation, and actual n2 and
dicated in Fig. 4b. A closer inspection, however, reveals p2 transistors can be used to breadboard the flip-flop.
that a few transistors serve the same functional purpose. Both the predicted and measured Q and Q output
P -enhancement transistors Qi and Qz, for instance, forms are noted in Fig. 5, when each output drives a 7-
have identical bias conditions; their gates and sources pF load. The longest part of the switching cycle, about
are common. One of these transistors can be eliminated 20 is, is the fall time of the 0:5 output, Fig. 5b. 0
when the drains of the two have been tied together. This REFERENCES
is possible, since their common drains will be buffered 1. C.T. Sah, "Characteristics of the Metal -Oxide -Semiconductor Transistor," IEEE Transac-
from the logic output node by transistor Q3. tions on Electron Devices, ED -11, July, 1964, pp. 324-345.
2. C.T. Sah, N.C. Pao, "The Effects of Fixed Bulk Charge onhe Characteristics of Metal -
The same reasoning applies to transistors Q4 and Q5, Oxide -Semiconductor Transistors," IEEE Transactions on Electron Devices, ED -13, April,
1966, pp. 393-409.
since their drains are also common, once Qi and Q2 are 3. Paul Nygaard, "An investigation of Voltage -variable Gate -Oxide Capacitance," Motorola
tied together. Moreover, transistors Q6 and Q7 have an Inc., Internal Communications, 1971.
4. William Lenin, "MOS voltage -variable Capacitance Study," Motorola Inc., Internal Com-
identical biasing arrangement, allowing one of them to munications, 1972.
be eliminated after connection of their drains, which 5. A.S. Grove, "Physics and Technology of Semiconductor Devices," Wiley, 1967.
6. D. Frohman-Bentchkowsky, A.S. Grove, "Conductance of MOS transistors in Satura-
will be buffered from the output by transistor Q8. In all, tion," IEEE Transactions on Electron Devices, ED -16, January, 1969, pp. 108-113.
three devices are redundant and may be removed. The 7. Robert H. Crawford, "MOSFEr in Circuit Design,'' McGraw-Hill, 1967.
8. Paul Nygaard, "Sceptre Subprogram Library," Motorola Inc., Internal Communications,
final flip-flop, Fig. 4c, requires 17 transistors. 1970.

151
Aedcap: the circuit designer's
computer -assisted slide rule
This powerful conversational software system solves circuit problems
easily; designs can be analyzed quickly for dc, ac, and transient analysis,
and their sensitivity to component tolerances can also be determined

by Ronald A. Rohrer and Jorge E. Rodriguez, Sof Tech Inc., Waltham, Mass.

Although the virtues of computer -aided circuit de- The program can simulate circuits composed of linear
sign are well known by now, most engineers still avoid resistors, capacitors, and inductors; independent voltage
using the computer because they don't know how to in- and current sources; voltage -controlled current sources;
terface with the machine. A year -old fully conversa- junction diodes; bipolar junction transistors; and mos
tional computer program called Aedcap, however, al- and junction field-effect transistors. Theoretically, there
lows the designer to communicate with the computer in is no limit to the size of the circuit Aedcap can accom-
an instruction -reply format that is written in engineer- modate, but, from a practical point of view, circuit size
ing terms. is restricted by the amount of computer memory avail-
Available on a time-shared basis, Aedcap (Auto- able. Presently, hundreds of circuit nodes can be
mated Engineering Design Circuit Analysis Program) analyzed in a single computer run.
provides as much analytical power as batch -processed Models are built-in
software packages while making efficient use of com-
puter time to hold down design costs. Circuits can be Diode and transistor models are predefined; the user
treated as prototypes being probed with an instrument. merely specifies the electrical parameters that character-
Aedcap can perform linear and nonlinear dc analysis, ize the active device. If the user chooses to leave a para-
small -signal ac analysis, and large -signal transient anal- meter value unspecified, the program inserts a default
ysis. All three analysis modes can be easily inter - value, which usually simplifies the model.
coupled. Moreover, the roles of signals and parameters Once characterized, a model can be stored so that it
can be interchanged so that a single computation can can be called out by name when needed to describe a
assess accurately the effects of every circuit parameter circuit. Models for physical devices are used to reduce
perturbation on a selected signal. the circuit being described to N nodes (not including

MODELS MODELS INITIAL FILE SYSTEM

MODELS
CONDITIONS USER AEDCAP MODELS
STANDARD ELEMENTS
MODELS MODELS MAN-MACHINc NEW ELEMENTS
INTERFACE
SMALL -SIGNAL LARGE SIGNAL NETWORKS
NONLINEAR
AC TRANSIENT FUNCTIONS
DC
ANALYSIS ANALYSIS ANALYSIS
INITIAL
TEXT EDITOR
CONDITIONS
QUIESCENT TIME
FREQUENCY STATIC
OPERATING
RESPONSE RESPONSE
POINT TRANSFER
CHARACTERISTIC SUPERVISORY
MIP AND DEFINED ANALYSIS
COMMANDS INTERFACE
SPARSE MATRIX MANIPULATOR

A
SENSITIVITY, SENSITIVITY, SENSITIVITY,
NOISE STATISTICAL, STATISTICAL, STATISTICAL, DATA
ANALYSIS AND WORST -CASE
ANALYSIS
V
AND WORST -CASE
ANALYSIS
AND WORST -CASE
ANALYSIS
STRUCTURE
GENERATOR ANALYSIS
ROUTINES
,0 0- -0
ADJOINT CONCEPT PRINT OUTPUT
OR POST
PLOT PROCESSOR
(al (b)

1. Framework. Flowcharts map Aedcap's analysis (a) and control (b) structures The program can perform dc, ac, or transient analysis, as
well as sensitivity, statistical, or worst -case analysis. Sparse matrix (one having many zero -valued entries) conserves computer time; adjoint
network technique permits reversing roles of signals and parameters. Aedcap's interactive framework eases interface between user and
computer. Conversational language and built-in active device models keep program instructions simple.

152
the datum or ground node) and B branches. Each equivalents, and the resulting linear circuit is solved at
branch is defined by a relationship between the branch sequential frequency points.
current and the branch voltage. The unknown circuit Large -signal transient analysis is used to determine
variables are the N node voltages. the time -domain response of a circuit to various input
Aedcap employs nodal analysis to solve circuit prob- waveforms, starting with the initial conditions found
lems. The program applies Kirchoff's current law at with a dc analysis.
each node to produce a set of N independent equations Full sensitivity analysis capability
in matrix form. A large circuit often has a nodal admit-
tance matrix that is typically 75% to 95% sparse. (Spar- Aedcap allows the user to perform dc or ac sensitivity
sity denotes the percentage of zero -value entries in the analysis, finding the change in a circuit output para-
matrix). The sparse matrix approach saves computer meter caused by variations in one or more circuit ele-
time. Flow charts for Aedcap's analysis and control ments. Since dc or ac sensitivity analysis is an inherent
structures are presented in Fig. 1. part of any dc or ac solution, its presentation does not
To perform dc analysis, Aedcap replaces capacitors require a new solution of the circuit. Worst -case analy-
and inductors by open and short circuits, respectively, sis and statistical analysis are also available as part of a
and then solves the circuit for fixed source values. A set dc or ac solution.
of static transfer characteristics can be obtained by solv- Worst -case analysis computes the output degradation
ing for sequential values of a given source. that will occur if all the element parameter tolerances
Of course, dc analysis is used to determine a circuit's vary collectively in the most pessimistic possible man-
quiescent operating point so that linearized model para- ner. A design that passes this test may be used with the
meters can be computed for small -signal ac analysis. utmost assurance that it will function properly.
Nonlinear elements are then replaced by linearized Statistical analysis with Aedcap predicts the percen-

Bipolar transistor model

Aedcap offers a built-in bipolar junction transistor model
that is an extension of the Ebers-Moll charge -control VI3E
model used by most programs. The model includes an QBEI = CjE0 f0 dV/(1 - Vicl)BEY/2
emission coefficient in each exponential term that char-
acterizes a junction. Additionally, three series parasitic QBE2 --- aFTFIBE
resistances and an output resistance augment the dc Vac
transistor characterization, while nonlinear capacitors QBCI = CjC0 dV/(1 - V/Oscy/2
model the base and depletion layer charge storage. A I.
linear collector -substrate capacitance is also added to
complete the modeling of transistor charge -storage ef- QBC2 = aRTRIBC
fects. In all, 18 parameter values are needed. Default values are
The circuit shown is an npn transistor model. It is char- put in by Aedcap when parameters are not specified.
acterized by six equations-two defining junction cur- For a pnp transistor, polarity is reversed for voltages
rents IBE and IBC, and the other four defining capaci- VBE and Vgc, currents IBE and IBC, and dependent gener-
tances OBE', OBE2, QBC1, and QBC2: ators aftlBc and aFIBE. Voltages VBE and VBC coincide
IBE = IEB[eXP(C1VBERIBEkT) - 1] with terminal potential differences only when the voltages
IBC = les[exp(qVite /nBekT) - 1] across resistors rB', re', and rE' are negligible.

TRANSISTOR PARAMETERS 0
ro

Parameter Aedcap Parameter Default

Symbol Name Name Value

Type Transistor type, npn or pnp

OF BF Forward common -emitter current gain 100
OR BR Reverse common -emitter current gain 1

re' RB Base ohmic resistance 0

rc' RC Collector ohmic resistance 0
rE' RE Emitter ohmic resistance 0
ccs CCS Collector -substrate capacitance 0
TF TF Forward transit time 0
TR TR Reverse transit time 0
C jE CI CJEO Zero -bias base -emitter capacitance 0
Cico CJCO Zero -bias base -collector capacitance 0
I Es I ES Base -emitter saturation current 1 X 10-14
I cs ICS Base -collector saturation current 2 X 10 14
rIBE NBE Base -emitter emission coefficient 1

nBc NBC Base -collector emission coefficient 1

(DBE PHIBE Base -emitter junction potential 1

PHIBC Base -collector junction potential 1

(PBC
ro RO Output resistance .

153
 LOGIN PROCEDURE

ATTACH AEDCAP TIME-SHARING FACILITY LOGOUT

PRINTF LISTF ERASE COST QUIT

CKT MANUAL BATCH

MODEL
FILE
FNCTN CKTNAME
{EDIT
DESIGN NAME ANALYZE
QUIT
FNCTNL
CONTROL
COM-
SET UP DATA
STRUCTURE FOR CKT
NEW FILE FILE ALREADY
EXISTS
REPEAT
CONTROLS

QUIT SHOW ACAN

INPUT EDIT OCAN
IRAN
SPECIFY OUTPUT LISTS USE SHOW
AGAIN AND DEVICE NAME PARM VALUE
BLANK SWEEP parm
OUTPUT S
BOTTOM 1 FROM value
CHANGE TO value
CONTINUE FROM
DELETE PRINT BY value
TO
FIND QUIT PLOT TIMES value
VS
GOTO OCT integer
INPUT QUIT DEC integer
.4111
INSERT PROBE
LOCATE SENS
SAVES
NEXT CONTINUE WORSTL
PRINT PERFORMANCE OF WORSTH
TO
REPLACE ANALYSIS PROGRAMS
BY
TOP { TIMES AND/OR PRINTING OR
PLOTTING OF RESULTS SAVES- t 1
UP OCT
VERIFY QUIT DEC SPECIFY LIST OF VARIABLES
DOUBLE CARRIAGE TO BE SAVED BY ANALYSIS
RETURN FILE UPON COMPLETION
ROUTINE
OF PROGRAMS QUIT
1

2. Road map. Using Aedcap does not require special skills. Road map charts basic chain of command and shows how they are related to
each other. The three primary command areas are COM-, controls, and input/edit. Instructions typed on terminal by user are underlined.

tage of circuits that will be out of tolerance in a produc- by typing ATTACH AEDCAP. The system responds with
tion run. This analysis is a variation of sensitivity analy- the word COM-, signifying that it is ready to accept any
sis that yields a simulated standard deviation of the command listed in the COM- area on the road map.
desired circuit output parameter. The standard devia- (User -supplied commands are underlined).
tion figure gives a tolerance or voltage deviation within The designer then calls for a circuit already filed in
which roughly two-thirds of the circuits in a production the system, or he describes a new circuit. He may do this
run will fall. Before his design is firm and the produc- by typing DESIGN AMPLIFIER CKT, if "amplifier" is the
tion line set up, the designer can learn what his yield is formal name by which he wishes to file his circuit. The
likely to be. system then searches for any circuit with this name. If it
In practice, Aedcap's statistical analysis has only lim- does not find one, the system notifies the user, who may
ited accuracy. But, taken in conjunction with other de- then proceed to describe his circuit.
sign uncertainties, the predictions Aedcap makes can be After executing any edit commands given by the user,
useful to engineers engaging in statistical design. the system returns to the COM- area, from which the
Instructions are conversational user can exit with the command ANALYZE. ANALYZE
precedes requests for the various analysis routines listed
Three types of commands are available to the Aedcap under controls. When the analysis is complete, the sys-
user. The road map of Fig. 2 outlines the instructions tem returns to COM-. If he wishes, the user may now ter-
included in each major command area. Control com- minate the session and log out to study his results. The
mands, for instance, are the various analyses that may circuit he has just analyzed may be filed in the Aedcap
be performed, while input/edit commands are the in- system for future reference or for subsequent additional
structions needed to describe or modify circuits. The analysis.
central command area, designated as COM-, provides ac-
cess to the controls and input/edit sections, and also Pointing out some limitations
contains several auxiliary commands. Parameter determination and modeling are still two
Sitting at his computer terminal, the user logs into the of the most significant limitations to the effective appli-
time-sharing network and then gains access to Aedcap cation of circuit -simulation systems. Aedcap attempts to

154
minimize these problems with a large model library. should suggest to him that this may occur. He can avoid
Device models can be stored permanently and made the problem by biasing the simulated circuit so as to
available for future designs and other engineering users. eliminate the superfluous solution.
Although Aedcap's built-in bipolar junction transis- The bistable example illustrates a general property of
tor model (see panel) is adequate for most circuit appli- simulation circuits. Practical circuits will always present
cations, it does neglect a number of higher -order effects more possibilities than could be envisioned by any set
that may occasionally be important. For example, the of program equations. The designer, therefore, must
common -emitter forward current gain, is not a con- have some idea of what he wants his circuit to do, and
stant, as assumed in the Ebers-Moll and Aedcap mod- how it is likely to perform before he can make intelli-
els. However, a reasonable representation of the beta gent use of a simulation system. He cannot expect the
variation can be obtained by augmenting Aedcap's computer to describe all the possibilities inherent in a
built-in model with other standard elements. particular design.
Other shortcomings of the modeling equations can Using Aedcap is easy
often be handled in a similar manner. Aedcap's mod-
eling equations do not currently extend to microwave A typical design session will illustrate how to work
frequencies, but microwave devices can be simulated with Aedcap. Figure 3a shows an every -day circuit-a
through suitable manipulations. differential transistor pair containing transistors Qi and
Another common problem is getting the series expan- Q2, resistors R1 and R2, voltage sources E1 and E2, and
sions for the solutions to converge. Aedcap minimizes ideal current source J1. The circuit has five nodes, which
this problem, with the help of the designer. For ex- are numbered. (The ground node is always made 0 by
ample, a bistable circuit can be particularly trouble- Aedcap convention; otherwise, any node numbering
some, since the computer does not know which of the scheme will do.)
two stable modes represents the desired solution. The The Aedcap description for the differential pair also
designer's knowledge of the circuit he is analyzing appears in Fig. 3a. It is a listing, by node connection, of
the circuit elements. Only a few simple rules must be
followed to describe a circuit. Each line in the table de-
(a)
scribes an element; the node that lies at the higher po-
12 V tential is listed first. Therefore, supply E1 has its node
connections written as 5 and then 0, since node 5 is at 12
volts and node 0 is ground.
For transistors, the collector node is listed first, fol-
5 0 el dc 12. lowed by the base and emitter nodes, respectively.
1 0 e2 dc xfer ac 1. Transistors, it should be noted, must be separately de-
2 3 ql q2n3563
4
1

0 3 q2 q2n3563 scribed as models so that the computer can refer to

5 2 rl 1k them when they are used in a circuit. Failure to specify
5 4 r2 1k all 18 parameter values of Aedcap's built-in transistor
3 0 j1 dc jvary
model will cause default values to be used.
The differential pair of Fig. 3a is to be a preamplifer
in a larger circuit configuration. Its load resistance will
be 1 kilohm, and a single -ended voltage gain of
50 ± 10% is desired. The power supplies are + 12 v and
-12 v. (The negative supply is to be added later in place
of the ideal current source.) Transistor model param-
eters are known and stored in the Aedcap file.
From circuit design experience, it is known that the
gain of this circuit depends on the value of transistor
5 0 gl dc 12. emitter current, IE. The design procedure, then, consists
10 e2 dc xfer act. of establishing an ideal current source (JO to determine
2 3 ql q2n3563
4
1

0 3 q2 q2n3563 the correct value of IE. After this value is found, the
5 2 rl 1k ideal current source can be replaced by an emitter resis-
5 r2 1k
4
tance and a negative supply.
3 10 r3 1.2k
10 0 e3 dc -12. A computer run
Once the circuit is described, the current for J1 can be
estimated and tested. Since the frequency response of
3. Designing with Aedcap. Differential transistor pair (a) must sup this preamp should be essentially flat, ac voltage gain
ply voltage gain of 50 ±10%. After approximating value of ideal cur can be quickly checked by computing the voltages at
rent source that provides desired gain, real supply (E3) and series re nodes 2 and 4 for a signal frequency of 1 hertz (for sim-
sistors (R3) can be substituted, as in (b). Tables are Aedcap circuit pler calculations). Knowing the voltage at either node 2
descriptions. Once all circuit nodes are numbered, elements are de or node 4 would be sufficient, but finding the potential
scribed on a line by listing the higher -potential node first. Transistor at both nodes will help confirm expected performance.
terminal nodes must be ordered-collector, base, then emitter. Also To test a 2 -milliampere value of J1, the user simply
a transistor must be identified as one of Aedcap's built-in models. types a request for an ac analysis, specifying the 2-mA

155
 AEDCAP CIRCUIT PREAMP
F2 GAIN(VM(4),VM(1))
F2 0.00E+00 1.00E+01 2.00E+01 3.00E+01 4.00E+01
103 1.00E+13 3.44E+01
1.26E+03 3.44E+01
1.58E+03 3.44E+01
2.00E+03 3.44E+01
2.51E+03 3.44E+01
3.16E+03 3.44E+01
3.98E+03 3.44E+01
5.01E+03 3.44E+01
6.31E+03 3.44E+01
7.94E+03 3.44E+01
104 1.001.04 3.44E011
1.26E+04 3.44E+11
1.58E+04 3.44E+01
2.00E.04 3.44E+01
2.51E+04 3.44E+01
3.16E+04 3.44E+01
3.98E+14 3.44E+01
5.01E+04 3.44E+01
6.31E+04 3.44E+11
7.94E+04 3.44E+01
105 1.00E+05 3.44E+01
1.26E+05 3.44E+01
1.58E+05 3.44E+01
2.00E+05 3.44E+01
2.51E+05 3.44E+01
3.16E+05 3.44E+01
3.98E+05 3.44E+01
5.011.05 3.44E+01
6.31E+05 3.44E+01
7.94E+05 3.43E+01
106 1.00E+06 3.43E+01
1.26E+06 3.41E+01
1.58E+36 3.40E+01
2.00E+06 3.37E+01
2.51E+06 3.34E+01
3.16E+06 3.28E+01
3.98E+06 3.21E+11
5.01E+06 3.12E+01
6.311+06 3.00E+01
7.94E+06 2.86E+01
10' -- 1.00E+07 2.70E+01
1.26E+07 2.53E+01
1.58E+07 2.35E+01
2.00E+07 2.16E+01
2.51E+07 1.96E+01
3.16E+07 1.76E+01
3.98E+07 1.55E+01
5.01E+07 1.34E+01
6.31E+07 1.12E+01
7.94E+07 9.03E+00
108 - 1.00E+08 6.80E+00
1.26E+08 4.57E+00

0 10 20 30 40
GAIN (dB)

4. Performance check. Computer plot of frequency response for differential pair in Fig. 3b shows that gain is about 35 dB out to 1 MHz.

estimate in place of NARY in the table of Fig. 3a. At the to Q2's base and then looking at the output voltage
same time, he requests a printout of the voltages at transfer curve at node 4.
nodes 2 and 4. The computer will find that both volt- The circuit's pulse transient response can also be
ages are approximately 7.3 v. checked out by using standard Aedcap pulse functions
Since the estimate of 2 mA yields a voltage gain of or any user -desired pulse function. And the sensitivity
only 7.3, the analysis is repeated for a current source of of the circuit can be tested to see how well the design is
10 mA. This gives a satisfactory gain of about 55.3. likely to fare under production conditions.
Ideal current source Ji can now be replaced by a real Since the circuit's frequency response was initially as-
resistor and a real supply. By calling in'the Aedcap edi- sumed to be flat, a sensible computation is checking the
tor, source J1 can be removed, and a 1.2-Idlohm resistor gain of the differential pair over a broad range of fre-
(R3) and a -12-v supply (E3) can be inserted. The prac- quencies, taking advantage of Aedcap's ability to do full
tical differential pair is illustrated in Fig. 3b, along with frequency sweeps. The computer printout of Fig. 4
its Aedcap description. shows the results of sweeping the circuit from 1
Again, circuit gain is checked by computing the volt- kilohertz to 100 megahertz and obtaining a plot of gain
ages at nodes 2 and 4 for a 1 -Hz input. If the gain is not (in decibels) versus frequency. The graph verifies that
within the desired specification of 50 ±-10%, the value of the frequency response will be essentially flat to 1 MHz.
resistor R3 must be changed. A computer run shows
that the gain is within tolerance, indicating that the cir- Program availability
cuit is essentially designed. Aedcap requires a large in-house computer, or it can
At this point, there are a number of possible analyses be used interactively through the time-sharing facility
that can be performed. For example, the circuit's dc of National css in Stamford, Conn. The computer ter-
transfer characteristics can be determined by plotting minals that the program can now accommodate include
the output voltages at nodes 2 and 4 as functions of dc the IBM model 2741, Teletype models 33 and 35, and
input E2. Or the common -mode gain can be computed the Tektronix types T4002, T4002A, and T4010 graphic
by first calling in the Aedcap editor to connect Qi's base computer terminals.

156
Easy -to -use PRINT TITLE, IDENTIFY TRANSISTOR

Hypi program READ DATA

makes possible WHAT ARE VALUES OF VCB, JN

transition from FOR I = 1 TO M NEXT I 1

nonlinear to linear IS INPUT VALUE OF JN

GREATER THAN
STORED VALUE OF JN
YES
STORE STATEMENT
DATA FOR NEXT LOWER
VALUE OF JN
AS J5, B5, D5, U5, E5

transistor model NO

IS INPUT VALUE OF JN STORE STATEMENT

YES DATA FOR NEXT LARGER
Acting as an interface between linear LESS THAN
STORED VALUE OF JN VALUE OF JN
AS J6, B6, D6, U6, E6
and nonlinear analysis programs,
NO
a new computer program called Hypi
GO SUB TO GET
converts parameters of nonlinear STORE TABLE VALUES
AS JI, 61, D1, Ul, El LINEAR APPROX
AS J1, Bl, DI, Ul, El
charge -control transistor model
to those of linear hybrid -pi model EXIT

by John R. Greenhaum, General Electric Co., Syracuse, N.Y. CALCULATE VBE

CALCULATE SMALL SIGNAL PARAMETERS

PRINT RESULTS

YES
NEW RUN

NO

END

Most nonlinear computer programs can analyze cir-

cuits only in the time domain, while linear computer
programs perform analysis in the frequency domain.
However, practically all available transistor perform-
ance data is in nonlinear form, making it inappropriate
for the linear programs. Consequently, an intermediate
program is needed to convert nonlinear transistor mod-
els to linear models, so that the linear programs can be
used for frequency analysis.
A new short program, which is called Hypi (pro-
nounced high -pie), enables the user to describe his cir-
cuit with a nonlinear transistor model, but perform his
analysis with a linear computer program, like Cornap
(Cornell network analysis program), ECAP (electronic
circuit analysis program), ACnet (ac network analysis

157
1. Hypi computer program. Flowchart outlines how Hypi converts nonlinear charge -control transistor model to linear hybrid -pi transistor
model. The user simply supplies data for transistor collector current and base -collector voltage. Previously stored table contains data for
charge -control model parameter values. (Linear interpolation is used to bridge table values.) After reading user input data, Hypi solves equa-
tions relating charge -control parameters to hybrid -pi parameters. Sample program develops model for type 2N1711 transistor.

NYPII. NYPIP. CONTINUED

100 PRINT"PRC.ChAM TO DERIVE SMALL SIGNAL HYSPID PI FROM LANGE SIGNAL" 00d GO SUS 1270
11D PRINT DATA" 810 LOT 81:13
12V PRINT " 622 LET y1.05
ISO kiR bl.b8,k2.1.C.N3.hb.Al.A1,A2.42,11.1C.12.118,13.1C9 830 LET 12:26
140 hEr vi.PNII.P2.PNIL,TI:TNETA W.T2.TWETA 1,41.01.02.k2 840 GO SUS 1270
150 1.01 21.VCE, V2:VAC. 113.18i, JI.JN, J02.11. J3.JC, JAI:a 65d LAT 01:13
14O CI.OJE,C2.CDC,E:3.ETC-C4.CTE,C5.CE,C6.CC,81.GME,R4.ROE 660 LE7 11:85
170 PIM pi.hCC 872 LET 72.U6
114 Ncm 6(11:01.t D(1).31, 11(1),ICN. r(1):7C1 891/ GO SUS 1272
ISO PnIAT :DOLL 011711. 908 LET U1.13
240 PNINT" 910 LET X1.05
210 DAIA 23.,7.081,27.. -12,i1.1 -10,1.62E-13,.420-13 926 LEI 72:06
22.4 dtAJ N1,142.03.01,A2,12,13 938 GO SUS 1270
232 JA1A 9144 LAT E1.Y3
240 NEKJ VI.P2,71,T2,61.02 356 AU TO 1006
050 LeT A5.1.E7 560 NEXT I
262 DIG C(50>,b(516),2(5b).11(50),V(92) 1600 LET V3.LOGIJI/127/TI
276 LOT M.22 1610 LEI J2.13.0(EXP(TE4V2)-1.1
dwd VAIST 1026 LET J3aJb/21-J1
294 00A DATA TABLE 14 CE4uTONIC ILENEASIMG CULFENTS 1030 LEI J4.J1/01-J2
360 JAL% .8225E-3.e...56,5.E-103.0-6 1642 LOT CI.Tibute(J1.12)
310 J974 .OBIE-3,2.,.56,5.2-10,3.E-6 1.56 LOT C0.70..E1b(22.13)
322 JAI0 1060 LAT C3.42/(P2,,2)'1112
333 .;ATA .010 -3,28.,.56,5.E -10,3.E-6 1070 LET C4.A1/(rl-W3)'111
340 JA14 .02E -3,25,06,58-10,3E-6 1000 LEI CS:CI...CA
350 DATA .63E-3,28,.58 5E-12 3E-6 11.90 LET C6:C22C3
36.1 2ATA 1109 LAT 61.11.21
370 JAIN .072-3,30,.67,50-12,3E-6 1:18 LET 114.81/01
380 dA7A 1128 PRINT
SSJ 321A .0E -3,55..72,5E -10.3E-6 1139 PRINT
402 2A14 .34-3,05.03.50-10,3E-6 1110 PRINT -FOR VC15."1V21.00."1.11
410 Gbf. 1159 PRINT"
422 ;22a .56-3,70.,.67,5.,-10,3.1-6 1100 PRINT'RBeiR1.'RCr'1112.'8E.'SR3,'ffE.'IN4
430 JATA 1.0-3,96.,.12.5.0-18,3.E-6 1176 PRINT 'ROC.-11,5,"CC:":06,"CEd"gC5,'GME.'101
440 BAIA 2o -3,105,57,50-16,3E-6 1160 PRINT'VOE.'1113
450 GAT. SE -3,117,53,5E -14,3E-6 IMO PRINT
1220 PRINT" IF YOU WANT A NEW RUN TYPE I. 008 NO Nil RUN TYPE 9'
470 JAI.: 7i -3,140,47,5E -10,3e-6 1230 LET 1:2
.80 JATA 10E -3,145,012,5E -10,3E-6 1240 INPUT X
496 DATA 200-3,156,38,52-10,3E..6 1250 IF X.1 TNEN550
530 DATA 300 -3,160,36,5E -10,3E-6 1253 STOP
510 GAIN 500-3,160,33,50-10,33-6 1266 REM SUROUTINE FOR LINEAR APPROXIMATION
520 Fyn 1:1 TO II 1270 LET A . (Y2 -Y1)/(.16-0)
531 Niid C(1).11(1),D(I)d(1).V(1) 1280 LET 0.Y1-04.15
542 NEXT 1 1290 LET 13 .A4J14.0
556 OWINT YOUN WALKS FON VC6,J11 MC; 1390 RETURN
560 Rif VC6.I/2 lb VOLTS J8.21 IN AMPS 1319 END
570 INPUT V2,JI
580 PRINT
590F011 1.1 10 M
COO IF .11.4(1) THEN 6711
618 IF J1dO(1) THIN 730
620 LET 01:0(1)
630 LET D10(1)
640 LET UMW
65d LET 01:6(1)
660 SO TO 1000
670 LET 25:C(1)
60C LET 95.6(1)
690 LET 55.2(1>
700 LOT U5.11(1)
711 LET 05.8(I)
7211 60 TO 960
730 LET .16.C(1)
746 LET 86.0(I)
750 LET 06.2(1)
760 LET 116.11(1)
778 LET 06.0(1)
700 LAT YI415
790 LET Y2.06

program), or any one of several other programs. parameters for the charge -control model, current den-
Developed at General Electric, Hypi can be run on a sity .41 and base -collector voltage Vgc, have been com-
time-shared basis and is written in Basic language. It puted manually. Or, again, the nonlinear program, Cir-
converts the parameters of the nonlinear large -signal cus (circuit simulator), can be used to find the JN and
Beaufoy-Sparkes charge -control transistor model to the VBc values, and then Hypi used to finish the conversion.
parameters of the familiar linear small -signal hybrid -pi
transistor model. The panel, "From charge -control to Examining Hypi
hybrid -pi," shows the two models and the conversion Hypi's flowchart (Fig. 1) is a generalized description
equations. of how the hybrid -pi model data is obtained. The pro-
There are several ways to solve these equations. For gram identifies the transistor type being examined,
instance, they can be solved directly, if not very conve- reads the data describing the charge -control transistor
niently, with the nonlinear program, Sceptre (system for characteristics, and then asks the user for the operating
circuit evaluation and prediction of transient radiation conditions specific to his circuit's performance.
effects) [Electronics, Aug. 16, 1971, p. 721. Alternatively, Hypi represents the charge -control model parameters
Hypi can be used alone, after the values for two of the of forward gain $N., inverse gain PI, normal time con -

158
stant TN, and inverse time constant Ti, as points on the in statements 210 and 230. Statement 210 provides data
curves normally used to describe these functions. (This for base bulk resistance RBB, collector bulk resistance
data can be obtained from published transistor liter- Rcc, emitter bulk resistance REE, base -emitter capaci-
ature.) The Hypi program automatically provides linear tance A1, base -collector capacitance A2, saturation emit-
interpolation between adjacent parameter values when ter current 'Es, and saturation collector current Ics.
necessary. Therefore, a proper result is obtained when Statement 230 enters data for intrinsic base -emitter po-
an evaluation is requested for a value of current density tential 01, intrinsic base -collector potential 02, and con-
JN that does not exactly correspond with one of the stants 01, 02, NI, and N2.
listed entries. Statement 270 indicates the number (m) of different
User input data is examined to determine whether it collector current values to be stored in the program.
agrees with the charge -control model data previously Statements 300 through 510 list these various values of
stored or whether an interpolation is required. The current JNm, in addition to the values for forward gain
charge -control model equations are then solved, and the fiNin, inverse gain thm, normal time constant TNm, and
results inserted in the conversion equations for the hy- inverse time constant Tim. This creates a data table, with
brid -pi model. The outcomes of these computations are the current values given in increasing order.
labelled with the hybrid -pi parameter descriptions. User input data for current JN and voltage VCB is re-
A sample Hypi program is also included in Fig. 1. quested by program statements 550, 560, and 570. With
Statements 100 through 200 identify the program, the this input information, Hypi searches its data table to
transistor that is being evaluated (in this case, type determine if the input current value agrees with a stored
2N1711), and the variables in the program. Statements current value. If no agreement is found, a linear inter-
210 through 540 cause the data describing the transistor polation is performed between the two stored current
to be read into the program. By changing this data, dif- values between which the input current value falls.
ferent transistors can be modeled. This procedure is described in statements 590
The charge -control model parameters are called out through 960. The lower data point values for RN, fil, TN,

R1 L
+47 V ---41,---"VVV-

0
2N1711
03
2N1490

02
2N1711
+24 Vdc

RL
LOAD

R2L

ALL INDUCTORS IL) AND RESISTORS R1 AND R2 ARE DUE TO LEAD LENGTHS

NONLINEAR TRANSISTOR OPERATING CONDITIONS

03

VE3c Jr, (A) VEic (V) JN (A) VBc (V) JN (A)

25 -5.54 0.0132 -13.90 0.0027 6.16 0.996

50 -12.40 0.00536 -14.00 0.00626 -13.00 0.519

125 -16.70 0.00228 -14.10 0.00848 -17.30 0.220

250 -18.20 0.00151 -14.10 0.00924 -18.80 0.119

2. Sample analysis. Voltage regulator can be examined for potential instabilities. Nonlinear Circus program is used first to determine collec-
tor currents and base -collector voltages of all three transistors for four different load conditions. Table shows results of Circus analysis of the
transistors for load resistances of 25 to 250 ohms, which cause load current for the 24 -volt regulator to vary from 1 to 10 amperes.

159
 From charge -control to hybrid -pi
When a transistor operates under varying conditions, a
nonlinear model is needed to describe device behavior
properly. One such model, a simplified version of the
Beaufoy-Sparkes charge -control transistor model, is Para- Sample
Definition
shown along with a tabulation of its parameters and some meter value
typical values.
The equations to determine the current generators for
the model are: R BB Base bulk resistance 55 S2
11 = (1 /fiN + 1)JN - Ji
12 = -JN + (1 /p, + 1)JI RCC Collector bulk resistance 5S2
where:
REE Emitter bulk resistance 1 mil
JN = lEs[exp(ONVBE) - 1]
J1 = Ics[exP(B1VBc) - 1]
Rc Collector reverse -bias leakage resistance 10 M11
The depletion and diffusion capacitances for the model
can be expressed as: Emitter reverse -bias leakage resistance 30 MS1,
RE
CTE = Al /(01 VBE)N1
CTC = A2/(c52 - VB14N2 CTC Collector depletion capacitance
CDE = O\TN(JN IES)
CDC = ICS) CTE Emitter depletion capacitance
The accuracy of nonlinear transistor modeling and the
convenience of linear problem -solving can be combined CDC Collector diffusion capacitance
by converting the nonlinear charge -control model param-
eters to the parameters of the linear hybrid -pi model CDE Emitter diffusion capacitance
shown. Four equations must be solved:
ICS Saturation collector current when VBE = 0 0.485 pA
gmE = ONJN
RBE = PN/ONJN = flN/gmE Saturation emitter current when VBC = 0 3.5 fA
I ES
Cc = A2 / (02 - VCC)N2 OiTI(J1 + ICS)
CE = A1/(4)1 - VCE)Nl 8NTN(JN 1ES) Forward current generator
JN
Generally, base -collector resistance RBc is assumed to
be so large that it can be neglected. Inverted current generator
NONLINEAR CHARGE -CONTROL MODEL
ON Normal beta with Vetc = 0, VBE = 0 72
CTE CTC

II 11

Iii
01 Inverse beta 0.62
CDE CDC
Base -emitter capacitance 3.7 pF
REE Rc RCC
A2 Base -collector capacitance 3.3 pF

01 Intrinsic base -emitter junction potential 1.1 V

RBB 02 Intrinsic base -collector junction potential 1.1 V

B
ON q/mkT, 1 < m < 2, T = 25°C 40.1

LINEAR HYBRID -PI MODEL 0, q/mkT, 1 < m < 2, T = 25°C 29.4

Cc
N1 Constant, 0.33 (graded junction)
to 0.5 (step junction) 0.34

N2 Same as Ni, usually N2 < Ni 0.10

R BE grnE VBE

TN Normal storage time constant 120 ps

Inverted storage time constant 35 ns

and TI are stored as J5, B5, D5, U5, and E5, respectively. ating conditions to be examined and, therefore, other
Upper data point values for these same parameters are parameter values to be generated at the user's option. If
stored respectively as J6, B6, D6, U6, and E6. no further modeling is required, the program stops.
The values for the hybrid -pi model parameters are Using Hypi
calculated with the equations listed in statements 1000
through 1100. Printout instructions for these computed A design example will illustrate how the Hypi pro-
values are given in statements 1140 through 1180. State- gram can simplify circuit analysis. The voltage regulator
ments 1220 through 1250 allow additional circuit oper- of Fig. 2 is to be examined to identify any potential cir-

160
la)
_112.909 ELAPSED TIME r. A SECONDS
259 RES RUN 1
___fL, __,..i.6.412.1.0.:-.01. -4 8.8111UW.-ILL- -.E.A.8/1311E _9.1._ Z.SZ_V12!E 91 3.470963E 91
2.1600000 02 4.3200006 02 6.486004 n, 8.6900000. 82 0.05000 E 9t
0.94aii1:489081i-O: -5.9.26600005 2.9680120-05 1.187201E- 0 2.0776040-01 1.959005E-04
103- .
.- - ,

I I
1 - I I I

I I I 1 I 1
I

i I .1
----H.._ ____1--- ' _ II I

1
1 1

1 .

I I I
.1

I 1 I I

!
...
I ____ __
V.Iii.ig 63 . - ' '
-5--
_JL__ ) I
1 1 _J .

I 1 1 ; 1 -+ t

-I _
t
__I
I
.2 _ t.I
; 1
I.

t I I
I

I 1 I

I L 1_

I I

1 1___ 1 L 1_
3.1622E 66 -TS
.1. ...I

I I I I t

I _1 1- L I I

I 1 I I
I
1 II --
-1.-
I

1
I

__I
I I I 1 I I I
I_ .1 1 ___ 1
1

L-.____..1.
1 I

_1 L. _L
105 - 9.99980 04
L I -I 1 1
I 1 i
i i

L_.._ I I t I

; I I I i
__L 1 4.- I _I 1
I I I I I I
.__L_ _.-L L
I I I I I

1._ L . i 1 . _ L .
3.1622E OS - I

.1___ L- I
I

I
I
L
I
1 I

I
1

I
I
L
I
L __

1__
I
I
I I
I GAIN -

I. _l_ 1 1

106- 9.9997E 05
____1
1

--I-
I I
--I---- -I.--I-_ I I

1. _P 1 it _IL
I 1 I I 1 1

J
I I I I I

.1 __L__. -.1- 1_1L.-

I.1622e 06 !
1
I I

.1 _
I

1.-
I1I I

I-_'
I

I
1

,L L

107 - 9.9997E 06

3.1622E 07
__L - .
1_
L. .

I I

108- 9.99960 07 !

GAIN: -78 -50 -21 +7.9 +37 +65

PHASE: 0 220 432 648 864 1,080

DELAY: -1.5 x 10-4 -0.6 x 10-4 +0.3 x 10-4 +1.2 x 10-4 +2.1 x 10-4 +3.0 x 10-4

161
 (b)

_17.94 112469 ELAPSED _TIME SECONDS

25V REG RUN I.

TRANSFER FUNCTION_ CRLTICAL FREQUENCIES - Hi

OUTPUT VARIABLE - I RL1

SOURCE VARIABLE - 14

GAIN CONSTANT IS 5.9664833D-03

P Q_LE P_O_S LT IONS 25R0 EDSITIONS

REAL PART IMAGINARY PART ORDER REAL PART IMAGINARY PART ORDER
-2.74806490 08 O. 1 -2,7480622D 08 O. _ .

-5,38803550 U6 O. 1 1.3145697D 06 O. 1
-5.695079910_16_ -2,3229i_9/0_07_ 1_ -5_._6927_901D_ _06 -2,_11999710 07 L
-7,41325480 05 4,07107 22D 06 1 -6.6625225D 05 -4.2086433D 06 1
-7.0510_9_64 a_ 15__ a. E1.12. !4 89_1_16 1_ -1,80701 9D 16 4.5857293Q Q5 .1
-2.2069888D 06 O. 1 -6.0152818D 05 O. 1
25_9n2n7n n4 n_ 1
7.1,50/532D_94 O. L.
02082720 03 1.38692 070 04 1

TRANSFER FUNCTION POLYNOMIALS - RADIANS

SCALED RY 1.000D 08 IN FRFOUENCY AND 1.0000 03 IN IMPEDAU_E_
DENOMINATOR DEGREE NUMERATOR DEGREE
COEFFECIENTS IN S. COEFFECIENTS IN S
1,000_001101(.1(110_04D_SLQ_ 1_Z_ .C110110ANDOCIfICIQ011. CIO 10
1.86421780292370D 01 11 1.82507986367618D 01 9
2.6744959R2R19170 11 10 1.-952.21164.522AD.86D_11 8
5.344559084114750 01 9 4.441617336461980 01 7
3.104.44159315380D 01 1-19_3A 6163 216.Ak ID 01
1.169884435379860 01 7 3.59471232049698D 00 5
3414516494686251110 4,6247967M/9830-01 4
6.148752089695850-01 5 -5.280801379067990-03 3
8.910538394723560-02 4 -1.88926194533646D-03 2
6.59065593013516D-03 3 -1.279931354482660-04 1
A, LA17.321.8137/61D- 06_ 2 2.63368.06764232.6a-07
5.683390326587980-09
113/1411111 7_12:11 2 0

3. Frequency response. Nonlinear transistor model data computed by Circus is entered into Hypi so that hybrid -pi model and, therefore,
linear program can be used to analyze regulator. Subsequent frequency analysis by linear Cornap program provides plot (a) of regulator
characteristics for 1 -ampere load, as well as tabulation (b) of pole -zero locations. Potential instability is outlined in color.

cuit instabilities that could cause unwanted oscillation. regulator's gain, phase and delay characteristics when
First, the collector currents and base -collector voltage load current is about 1 A. Amplitude values and peak
drops of all the transistors are found with the nonlinear amplitude locations vary with transistor collector cur-
Circus program for four different load conditions. In rent. The peak (outlined in color) in the vicinity of 14
this case, load current ranges from approximately 1 am- kilohertz for all characteristics indicates the presence of
pere to 10 milliamperes, as load resistance is varied a potentially critical pole.
from 25 to 250 ohms. The table in Fig. 2 gives the re- Cornap's tabulated output format (Fig. 3b) for the
sults of this analysis. regulator's frequency -domain transfer function and
The current and voltage values are then entered into pole -zero locations lists the potentially troublesome
the Hypi program; a separate run is needed for each pole (outlined in color) as having its real part located at
transistor. The hybrid -pi model data supplied by Hypi 1,021 hertz and its imaginary part at 13,869 Hz. An
can then be entered into any one of several linear analy- analysis of this pole location reveals that the pole's
sis programs. For this example, the widely used linear phase angle is approximately 86°, which implies ex-
program, Cornap, is chosen for convenience. With a treme sensitivity to ringing and probable oscillation if
command of only a single instruction, Cornap can pro- parameter values change even slightly. 1:1
vide transfer functions, pole -zero locations, and plots.
Figure 3a is a Cornap frequency -domain plot of the The author wishes to admovAedge J.E. Hooper's help in developing the Hypi program

162
Other McGraw-Hill Books for Electronics Engineers
(See inside front cover for more titles)

ELECTRONIC FLASH, STROBE CIRCUIT DESIGN FOR

By HAROLD E. EDGERTON, Institute Professor Emeritus, AUDIO, AM/FM, AND TV
Massachusetts Institute of Technology Prepared By THE ENGINEERING STAFF OF
384 pages, 209 illustrations TEXAS INSTRUMENTS INCORPORATED
Everything you need to know about today's electronic flash 352 pages, 71/4 x 91/4, 145 illustrations
- whether for your work in professional photography, in re-
search, or as a hobby - is included in this authoritative book Stressing time- and cost-cutting approaches, this practical
by a leading world expert on the subject. You are given volume is packed with proven new procedures for solving
theoretical and practical information that guarantees you typical problems in audio, AM/FM, and TV circuit de-
better photographs through numerous techniques developed sign. With examples and procedures illustrating the latest
by Dr. Edgerton and now widely used in commerce and in- available transistor devices, it covers such important topics
dustry. He describes typical systems employed today, plus as design of IF strips, neutralized and unneutralized am
some of the more unusual and interesting uses in science, plifiers, IF amplifier designs for AM/FM and FM IF
assuring you a thorough understanding of what the tech- amplifier circuit applications, specific design examples for
niques can mean to you. To anyone who must learn quickly
and expettly about electronic flash, the book is a clear guide each major TV receiver system, UHF and VHF tuners,
to all the fundamentals by the man who pioneered in the sync separators, vertical oscillators, video amplifier sys-
creation of the modem stroboscope. tems, TV automatic gain control, and more.

INTEGRATED ELECTRONICS HANDBOOK OF MATERIALS AND

By RODNEY SORKIN, Electronic Engineer, U. S. PROCESSES FOR ELECTRONICS
Department of Defense
224 pages, 147 illustrations, Edited by CHARLES A. HARPER, Aerospace Division.
Westinghouse Electric Corporation
Here is the first book to assemble all the essential elements 1,344 pages, 803 illustrations
of integrated electronics into one compact reference. It not
only covers the whole technology-components, construction An indispensable working tool for all engineering and tech-
techniques, circuit characteristics, and design principles - nical personnel in the electronics and electrical industries,
but gives equal weight to both fabrication and applications. this handbook offers the greatest wealth of basic data and
In this way, you get a clear working knowledge of the tech- applications guidance ever assembled in a single volurhe. 21
nology not available in any other book, and most particularly top specialists give you the kind of guidelines you need for
of the various factors and limitations that affect your own all major materials and processes you use on the job, and
specialty. Whether you use it to brush up on the field, get a whether you are engaged in design, development, assembly,
better grasp of the theory, or as a design aid - you will find or related activities in the electronics or electrical industries,
it the most useful work of its kind in print. Thoroughly up to you will turn to this massively complete handbook again and
date, and putting emphasis where you want it, the book gives again for practical, on-the-job guidance. It will not only save
you a revealing new perspective on a rapidly expanding you time and money, it will help you produce a better, more
technology. marketable product. Hundreds of tables, drawings and
photographs.

RADAR HANDBOOK TRANSISTOR CIRCUIT DESIGN

Prepared By THE ENGINEERING STAFF OF
Edited by MERRILL I. SKOLNIK, Superintendent, Radar TEXAS INSTRUMENTS INCORPORATED
Division, U.S. Naval Research Laboratory, Washington, D.C.
1,200 pages, 1,140 illustrations 523 pages, 71/4 x 91/4, 526 illustrations
A clear, complete picture of all modern radar design tech- You can solve your circuit design problems quickly and
niques, applications, and concepts is offered in this world's accurately with the help of this authoritative guide which
first one -volume handbook on radar engineering. 51 top translates the most frequently used elements of transistor
radar experts have contributed their know-how and experi- theory into practical solutions and gives you a large num-
ence to this superb reference, which provides full, accurate ber of circuit examples with clear, easy -to -follow design
information on radar subsystems, radar techniques, analysis procedures. It classifies all commercially available tran-
of performance, signal management, environmental effects, sistors, showing their relationship to each other and to
and system applications. In view of its enormous range of
subjects covered, no matter whether you are involved with the five basic fabricating techniques. It shows how to in-
radar in air traffic control, police speedometers, or astronomy, terpret data sheets and device numbers, illustrates equiva-
here is a single book that will be of invaluable aid in solving lent circuits, details techniques for measuring transistor
any technical problem you may meet, whatever your special- characteristics, and demonstrates the application of tran-
ization. A perfect on-the-job reference for men who need to sistorized circuitry to air navigation, radar, remote con-
know radar applications and concepts from other specialties. trol, and other areas.

PLASTIC COATINGS FOR ELECTRONICS ELECTRONIC DIGITAL TECHNIQUES

By JAMES J. LICARI, Supervisor, Research Division of the By PAUL M. KINTNER, Manager, Digital Systems and
Advanced Chemistry Laboratories of Autonetics, Materials Products Development, Industrial Systems Division,
and Processes Laboratory Cutler -Hammer.
400 pages, 130 illustrations 315 pages, 6 x 9, 215 illustrations
Available at last - a practical, one -volume guide to the selec- Here is a down-to-earth approach to digital systems geared
tion and specification of plastic coatings. This new book to the needs of the practicing engineer. Non -mathematical
combines theoretical and practical data on plastic coatings in presentation, this practical guide is concerned only with
for both commercial and military electronic applications. It actual design problems and the techniques for their solu-
proceeds from a discussion of basic chemistry and properties tion. All of the most up-to-date methods are used. Written
of coatings, to functions, manufacturing technology, and
specific electrical applications such as printed circuit boards, in an easy -to -understand style, the book is ideal for self -
resistors, capacitors, and similar devices, and concludes with study. It uses algebra or other mathematical formulations
a detailed survey of types of specifications used in obtaining only when they specifically contribute to a design tech-
and applying coatings. Numerous tables and lists provide nique. All topics are carefully developed - with numerous
instant access to data on coating properties, specifications, illustrations and examples - making the volume as com-
and tests. Clear cut graphs, diagrams, and examples offer prehensive as possible. A large number of self -testing
added facts and figures at a glance. problems have also been included for this purpose.

Use the attached order card or write to:

Dept. 23-A191-2000-1/McGraw-Hill Book Company/P.O. Box 400/Hightstown, N.J. 08520
f

..!,j
3kre4