The Bell System Technical Journal

January, 1934

Stabilized Feedback Amplifiers*

By H. S. BLACK

This paper describes and explains the theory of the feedback principle
and then demonstrates ho\\ stability of amplification and reduction of
modulation products, as \veil as certain other advantages, follow when
stabilized feedback is applied to an amplifier. The underlying principle
of design by means of which singing is avoided is next set forth. The paper
concludes \Vith some examples of results obtained on amplifiers which have
been built employing this new principle.
The carrier-in-cable system dealt with in a companion paper 1 involves
many amplifiers in tandem with many telephone channels passing through
each amplifier and constitutes, thereforel an ideal field for application of
this feedback principle. A field trial of this system was made at Morris
town1 Ne'v Jersey, in which seventy of these amplifiers were operated in
tandem. The results of this trial were highly satisfactory and demon
strated conclusively the correctness of the theory and the practicability
of its commercial application.

INTRODUCTION
UE TO advances in vacuum tube development and amplifier
D technique, it is now possible to secure any desired amplification
of the electrical waves used in the communication field. When many
amplifiers are worked in tandem, ho\vever, it becomes difficult to keep
the overall circuit efficiency constant, variations in battery potentials
and currents, small when considered individually, adding up to produce
serious transmission changes for the overall circuit.. Furthermoret
although it has remarkably linear properties, when the modern vacuum
tube amplifier is used to handle a number of carrier telephone channels,
extraneous frequencies are generated which cause interference between
the channels. To keep this interference within proper bounds involves
serious sacrifice of effective amplifier capacity or the use of a push-pull
arrangement which, while giving some increase in capacity, adds to
maintenance difficulty.
However, by building an amplifier whose gain is deliberately made,
say 40 decibels higher than necessary' (10,000 fold excess on energy
basis), and then feeding the output back on the input in such a way
Presented at \\Tinter Convention of A. I. E. E., New York City, Jan. 23-26,
1934. Published in Electrical E1tgineering, January 1934.
1

1 11 Carrier in Cable,. by A. B. Clark and B. \V. Kendall1 presented at.the A. I. E. E.

Sum mer Con ven ti on Chicago, I 11., J unc, 19 3 3; pub I ished in Eleclr ical Engineerin.g
July. 1933, and in Bell S)'S. Terh. Jour., Ju 1)1 1933.
t 1

1
 2 BELL SYSTEf TECHNICAL JOURNAL

as to throw away the excess gain, it has been found possible to effect
extraordinary improvement in constancy of amplification and freedom
from non-linearity. By employing this feedback principle, amplifiers
have been built and used whose gain varied less than 0.01 db with a
change in plate voltage from 240 to 260 volts and whose modulation
products were 75 db below the signal output at full load . For an
amplifier of conventional design and comparable size this change in
plate voltage would have produced about 0. 7 db variation while the
modulation products would have been only 35 db down; in other
'vords, 40 db reduction in modulation products was effected. (On an
energy basis the reduction was 10,000 fold . )
Stabilized fee dback possesses other advantages including reduced
delay and delay distortion, reduced noise disturbance from the power
supply circuits and various other features best appreciated by practical
designers of amplifiers.
It is far from a simple JJroposition to employ feedback in this way
because of the very special control required of phase shifts in the
an1plifier and feedback circuits, not only throughout the useft1l fre
quency b and but also for a \vide range of frequencies above and below
this band. Unless these relations are maintained, singing will occurt
usually at frequencies o u tside the useful range. Once having achieved
a design, however, in \vhich proper phase relations are secured, expe
rience has demonstrated that the performance obtained is perfectly
reliable.
CIRCUIT ARRANGEMENT
In the amplifier of Fig. 1, a p o rtion of the output is returned to the
input to produce feedback action. The upper branch, called the
-circuit, is represented as containing active elements such as an
amplifier while the lower branch, called the /j-circuit, is shown as a
passive network. The way a voltage is modified after once traversing
each circuit is denoted and f3 respectively and the product, {3, repre
sents how a vo l t age is modified after making a single journey around
amplifier and feedback circuits. Both and f3 are complex quantities,
functions of frequency, and in the generalized concept either or both
may be greater or less in absolute value than unity.2
Figure 2 sho\vs an arrangement convenient for some purposes where,
by using balanced bridges in in put and output circuits, interaction
between input and output is avoided and feedback action and amplifier
impedances are made independent of the properties of circuits con
nected to the amplifier.
2 . is not used in the sense that it is someti1nes usedt namely, to denote the

a111p)ification constant of a particular tube but as the con1plex ratio of the output
to the jnput voltage of the anlplifier circuit.
 ST..:1 BILIZED FEl!.IJ B.1 CK ..4 ill P LI FIERS 3

e ---..

>--....-...
.. E + N + 0
f3(E +N+D)

rE EDBACK aRCUIT
"

Fig. l-A111plifier system \vith feedback.

e-Signalinput voltage.
-Propagation of an1plifier circuit.
e-Signal output voltage without feedback.
1i-Noise OUtJlUt \1oltage without feedback.
d(E)-Distortion output \oltage without feedback.
{J-Propagation of feedback circuit.
E-Signal otttput voltage with feedback.
N-N oise output voltage \Vi th feedback.
D-Distortion output ''oltage with feedbackA
rhe output ''oltage with feedback is E + N + D and is the sun1 of e + n + d(E),
the value \vithout feedback plus ,B[E + N +DJ due to feedback.

E +N+D = e + 1t + d(E) + .B[E + N + D]

[E + N + D](l - fj) = e + n + d(E)
e + n + d(E)
E + J.V + D
1 - /3 1 - fJ 1 - /3
=

If [ fJ I >> lt E .:... - ;. Under this condition the a111plification is independent of

.but does depend upon {3. Consequentl) the O'\"er-all characteristic \vill be con
trolled by the feedback circuit \vhich nla; include equalizers or other corrective
net\vorks.

GENERAL EQUATION
In Fig. 1, f3 is zero \Vithout feedback and a signal voltage, e0, applied
to the input of the -circuit produces an output voltage. This is
made up of \vhat is wanted, the amplified signal, E0, and components
that are not \Vanted, namely, noise and distortion designated N0 and
Do and assumed to be generated within the amplifier. It is further
assumed that the noise is independent of the signal and the distortion
generator or modulation a function only of tl1e sig11al 011tp11t. Using the
11otation of Fig. 11 the output \vithout feedback may be written as:

Eo +No+ Do = eo + n + d(Eo), (1)

\V11ere zero sulJscripts refer to conditions \\rithout feedback.
 +t:--

..- b:l
R tl1
"' 0
,
t""'4
..
I
t""4
I

,,.,,.
,

..,,/)
c -
L
,J


-

t-;

Cs
-.24V -70V +130V

t--i
Fig. 2-Circuit of a negative feedback amplifier.
 .\' 1A BILIZRD FEED R.:1 CK A AfP LIFIER.<; 5

With feedba<:k, /3 is not zero and the input to t11e -circuit becomes
eo + {3(E + N + D). The dutput is E + N + D ar1d is equal to
[e0 + {j(E + N + D)] + n + d(E) or:
en + n + d(E)
E + N+ D (2)
.

1 1 1 p
=

- - -

In the output. signal, noise and modulation are divided by (1 - {3),

and a s sumi ng I 1-{3 j > 1, al 1 are reduced.

CHANGE IN G.i\IN DUE TO FEEDBACK

From equation (2), the amplification with feedback equals the
amplification without feedback divided by (1 {3). The e ffect of -

adding feedback, therefore, usually is to cha n ge the gain of the amplifier

and this change wil I be expressed as:

1
GcF = 20 ?og10 1 _ {3 (3)

where GcF is db cliange in gain due to feedback. 1/(1 - {3) will be used
as a quantitative measu re of the effect of feedback and the feedback
referred to as positive feedback or negative feedback according as the
absolute value of 1/(1 fl) is greater or less than unity.
- Positive
feedback increases the gain of the amplifier; negative feedback r ed uc e s
it. The term feedback is not limited merely to those cases where the
absolute value of 1/(1 ) is other than unity.
-

From fj = j ,u I and (3), it may be shown that:

l_!
(4)
which is the equation for a family of concentric circles of radii
10-GcFl10 about tl1e point 1, 0. Figure 3 is a polar diagram of the
vector field of {j =J f3 I Using rectangular instead of polar
l_!.
coordinates, Fig. 4 corresponds to Fig. 3 and may be regarded as a
d i agr am of the field of {3 where th e parameter is db change in gain
due to feedback. From these diagrams all of the essential properties
of feedback action can be obtained such as change in amplification,
effect on Jinearity change in stability due to variations in various
t

parts of the system, reduction of noise, etc. Certain significant

boundaries have been designated similarly on both figures.
For example, boundary A is the locus of zero change in gain due to
feedback. Along this parametric contour line where the absolute
magnitude of amplification is not changed by feedback action, values
of I /3 I range from zero to 2 and the phase shiftt <P, around the amplifier
6 BELL S1'.91'Eli1 TECilNJC.-1 L JO URN.1 JJ

and feedback circuits eq tlt ls cos-1 I /JI /2 a11cl, the refore, lies between
- 90 and + 90. For all conditio11s inside or above this boundary,
the gain \vi th feedback is increased; outside or below, the gain is
decreased.
STABILITY
From equation (2), eo/(1 - fl) is the amplified signal with feedback
and, therefore, /(1 - {3) is an index of the amplification. It is of
course a complex ratio. It \vill be designated AF and referred to as
the amplification with feedback.
To consider the effect of feedback upon stability of amplification,
the stability will be viewed as the ratio of a change, SAF, to AF where
oAF is due to a change in either or {3 and the effects may be derived
by assuming the variations are small.

(5)

(6)

[ oA F ...:_ {3 [ .
o/3
AF J 1 - {3 {3J (7)

If {3 >> 1 , it is seen that or the -circuit is stabilized by an amount

corresponding to the reduction in amplification and the effect of intro
duci n g a gain or loss in the -circuit is to produce no material change
in the overall amplification of the system; the stability of amplification
as affected by {1 or the j3-circuit is neither appreciably i mprove d nor
degraded since increasing the loss in the JS-circuit raises the gain of
the amplifier by an amoun t almost corresponding to the loss intro
duced and vice-versa. If and {3 are both varied and the variations
sufficiently small, the effect is the same as if each were changed sepa ...

rately and the two results then combined.

In certain practical applications of amplifiers it is the change in
gain or ammeter or voltmeter reading at the output that is a measure
of the stability rather than the complex ratio previously treated. The
conditions surrounding gain stability may be examined by considering
the absolute value of AF This is shown as follows Let (db) represent
the gain in decibels corresponding to AF Then

(db) = 20 log10 I AF I ,

O(db) -'- 8.686 [ OI ;, ' ] . (8)

 ST 4BILIZED
.. fa'lE/JB11 CK ..4'!PL! FIERS 7

To get the absolute \ralue of the amplification: Let

{3 = I {j I ' I (9)
rAF) j[ I
(10)
,. 1 2 cp +
=

- i ,B l cos I f3 J 2
The stability of amplification which is proportional to the gain
stability is given by:

(11)

(12)

(13)

BOUNDARY J BOUNDARY C

lft! =cos t =COS

2
l;/St
BOUNO>.RY E
o::COS cp
-6DB
BOUNMRY l ____,,.,,_

-3 B
ODS
-BCXJ::" NOARY H
....

- lO D B

BOUNDARY G
COS= -I I Ba.JN QARY r
/ COS<p=r
/ I
_....

BOUNDARY A
l...u..81 - cos "''t'
2
ZERO CHANGE
HF

1: ig. 3-'rhe ,crt or field of ./j. See capt ion for 1-ig. 4.
8 BELL .5'1r::iTEM TECIINICAL JOURNAL

t--++--\--t--Jir--+-=:t"#-L-L.)('
0

..340 g
tf)
0
3 0 .....+--1
... ---l--\+-Ac-....; ..,,c.....,,_Lf--i,_.-----,-A--+- 330
-

0
0
o
o 0
o

-+:.,_---111""'----..y;--"t---:i,__. 320
4

c
w
w
l.i..
80 2B0
0 0
2 z
:::i
0 90 --+--....
. ti---++--""''---W'-W""".....,...,p"'-_,.+-_.,.._.,._--llf 270 6
a: a:
< <
100 .__1--.....,,.---..t..._-HL--+--+-+---7-+----f+---trt--\:-t--f---t-----t 260
r

11 OD 1--r-#o--+---#--+-__,.-+-r....t----,R---+---.ll---,J'---+-:--+-7'---i 2 50
r20D ...--li-----fi,__--M---11'.--4-.,__+--f--J.J--1f.-+--J-+-f--j-+---+-I 240
<
z
Q_ 130 230

ti') U)
ttl 140

220
a::
! "
Q c,, ,.,
'.SOO 210 0
-
I ,

G
160 180
0 .2 .4 6 .8 1.0 1.2 1,4 1.6 1.8 2.0 2.2 2.4 2.6 2.8 3.0

I _,p I
Fig. 4-Phase shift around the feedback path plotted as a function of ! /3 I ,
the absolute value of /3.
fj is complex quantity which represents the ratio by which the amplifier and
a

feedback (or 1nore generally IL and (3) ntodify a voltage in a single trip around the
closed path.
First, there is a set of boundary curves indicated as A, B, C, D, E, F, G, H, I, and J
which gives either limiting or significent values of f3 and <I>.
Secondly, there is a family of curves in which db change in gain due to feedback is
I I
the parameter.
Boundaries
A. Conditions in which gain and 111odulation are unaffected by feedback.
B. Constant a111plification ratio against small variations in [ ,8 j.

Constant change in gain,

I
l pj , against variations in !JJI and /Pl.
Stable phase shift through the amplifier against variations in 'lls.
The boundary on which the stability of amplification is unaffected by feedback.
C. Constant amplification ratio against small variations in I. I.
Constant phase shift through the amplifier against variations in fl>,u.

The absolute magnitude of the \'oltage fed back 11 llPI is constant against

\7ariations in [ I and I fJ l.
 STABILIZED FEEDB4CK AJ.fPLIFIERS 9

A curious fact to be noted from (11) is that it is possible to choose

a value of ,{1 (namely, ! J = sec <I>) so that the numerator of the

right hand side vanishes. This means that the gain stability is
perfect, assuming differentia: variations in f I - Referring to Figs. 3
and 4, contour C is the locus of j /3 ! sec <Pandit includes all ampli
=

fiers whose gain is unaffected by small variations in [I. In this way

it is even possible to stabilize an amplifier who se feedback is positive,
i.e., feedback may be utilized to raise the gain of an amplifier and, at
the same timet the gain stability \Vith feedback need not be degraded
but on the contrary improved. If a similar procedure is followed
with an amplifier whose feedback is negative, the gain stability will
be theoretically perfect and independent of the reductions in ga in due
to feedback. Over too wide a frequency band practical difficulties
will limit the improvements possible by th e se methods.
With negative feedback, gain stability is al,vays improved by an
amount at least as great as corresponds to the reduction in gain and
generally more; with positive feedback, gai n stability is never degraded
by more than would correspond to the increase in gain and under
appropriate conditions, assuming the variations are not too great,
is as good as or much better than \Vithout feedback. With positive
feedback, the variations in . or {3 must not be permitted to become
sufficiently great to cause the amplifier to sin g or give rise to instabil
ity as defined in a fol lowing section on ''Avoiding Singing.''

l\11 OD ULATI ON
To determine the effect of feedback action upon modulation pro
duced in the amplifier circuitt it is convenient to assume that the
o utp ut of undistorted signal is made the same with and without feed

back and that a comparison is then made of the difference in modula

tion with and without fe e db ack. Therefore, with feedb ac k, the i npu t
is changed to e = e0(1 - {3) and, referring to equation (2), the out
put voltage is e0, and the generated modulation, d(E), assumes its
value without feedback, d(Eo). and d (E) / (1 - fJ) becomes d(Eo) / ( 1 - {J)
which is D0/(1 - {3). This relationship is approximate because the

D. I /j=I
E. ct>
l.=
90. I n1proven1ent in gain stabilit)' corresponds to ty.ice db reduction
1n gain.

f' and G. Constant an1plification ratio against ''ariations in <I>.

Constant phase shift through the an1plifier against \rariations in l,u I and
I j
II. Same properties as B.
I

I. Sa111e 11roperties as E.

.. w h'1c h ! 1
J . C on d.1t1ons 1n
' 1/3 1 t h e overa ll gain JS
. .
. t h e exact negative
I ! fJ I
_

in,'erse of the t ransn1ission through the -circuit.

10 BELL SYSTEM TECHNICAL JOURNAL

voltage at the input without feedback is free from distortion and with
feedback it is not and, hence, the assumption that the generated
modulation is a function only o.f the signal output used in deriving
equation (2) is not necessarily justified.
From the relationship D = Do/( 1 {3), it is to be concluded that
-

modulation with feedback will be reduced db for db as the effect of

feedback action causes an arbitrary db reduction in the gain of the
amplifier, i.e., when the feedback is negative. With positive feedback
the opposite is true, the modulation being increased by an amount
corresponding to the increase in amplification.
If modulation in the .8-circuit is a factor, it can be shown that
usually in its effect on the output, the modulation level at the output
due to non-linearity of the ,8-circuit is approximately {3/(1 -{1) multi
plied by the modulation generated in the .a-circuit acting alone and
without feedback.
ADDITIONAL EFFECTS
Noise
A criterion of the lOrth of a reduction in noise is the reduction in
signal-to-noise ratio at the output of an amplifier. Assuming that
the amount of noise introduced is the same in two systems, for example
\Vith and "rithout feedback respectively, and that the signal outputs
are the samet a comparison of the signal-to-noise ratios will be affected
by the amplification between the place at which the noise enters and
the output. Denoting this amplification by a and ao respectively, it
can be shown that the relation between the two noise ratios is

(a0/a) (1 {3). This is called the noise index.

-

If noise is introduced in the power supply circuits of the last tube,

a0/a = 1 and the noise index is (1 fJ). As a result of this relation
-

less expensive power supply filters are possible in the last stage.

Pliase Shift. Envelope Delay, Delay Distortion

In the expression AF = [/(1 .B)] [!, O is the overall phase shift
-

with feedback, and it can be shown that the phase shift through the
amplifier with feedback may be made to approach the phase shift through
the {3-circuit plus 180 degrees. The effect of phase shift in the {j-circuit
is not correspondingly reduced. It will be recalled that in reducing
the change in phase shift with frequency, envelope delay, which
is the slope of the phase shift with respect to the angular velocity,
w = 2Trf, also is reduced. The delay distortion likewise is reduced
because a measure of delay distortion at a particular frequency is the
difference between the envelope delay at that frequency and the least
envelope delay i11 the band.
 .S"lll 111 II ZlIJ JlEED B.4 L-.K 11 J.lfPLI FIEl.\' 1l

f:J-( 'irc:ttit Eq11al-izaJ01i

Referri11g to equatio11 (2), t11e output voltage, E, a1)1)roacl1es - eu/13
as 1 {J {J and equals it in absolute value if cos

where {J
-

[ /3 j =
-

<I> 21 I
Under these circumstances increasing the loss
l_!.
= :/3
in the -circuit one db raises the gain of the amplifier one db and vice
versa, thus givi11g any gain-frequency characteristic for which a like
loss-frequency cl1aracteristic can be i nserted in the /j-circuit. This
procedure has been termed jl-circuit equalization. It possesses other
advantages which cannot be dwelt upon here.

A VOID SINGING
Having considered the theory up to this point, experimental evidence
\\"as readily acquired to demonstrate that {J might assume large values,

JO i-. C

I 00 11.C

\1 10
..... 1
I 8 KC

10 "'c

2KC

5TABLl. U ... STABLC

Fig. 5-1\feasurecl fJ characteristics of t\vO a n1plifiers.

for example 10 or 10,000, provided <I> \\ras not at the same time zero.
However one noticeable feature about the field of (3 (Figs. 3 and 4) is
t

that it implies that even though the phase shift is zero and the absolute
value of fj exceeds unity, self-oscillations or singing \\rill not result.
This may or ma),. not be true. \Vhen the author first thought about
this matter he suspected that owing to practical non-linearity. singi ng
Vtlould result \\"he never the gain around the closed loop equalle d or
exceeded the loss and simultaneously the phase shift was zero, i.e. t

{J I {3 J + jo >- 1. Results of experiments, ho,vever, seemed to

=

indicate something more \\ras invol\1ed and these matters \Vere de

scribed to l\I r. H. N)qt1ist, ,,ho de\-eloped a m ore general criterion
12 BELL SY )TRM TECHNICAL J(JURNAL
..

for freedotn from i11stability 3 applicable to a11 atnplifier having li11ear

positive constants.
To use this criterio11, plot /3 (the modulus and argument vary with
frequency) and its complex conjugate in polar coordinates for all
values of frequency from 0 to + co. If the resulting loop or loops
do not enclose the point ( 1, 0) the system wil I be stable, otherwise
not. 3 The envelope of the transient response of a stable amplifier

-. I
I 1
,
-

I
/
'
I
I
j
I
! I
90 I

FEEDBACK]l
.

NO

80
I
I
rt .
'
I I i
I \
I
I I \.
70

I
PE.RATlNG RANG;__.i '
I

60
II
/
;
I
4-40 KC
i
I

I
I
'
I
II
z
< 50
-

Cl

jjii"
I
I

'
'
I

I
'
\
FEEDBACK I
I

\'

I
I
a:=:
r
40 """"
La.. t '
I '
I
I -
...J I
a. I I

30
! I
-
.

I
I
I
FEEDBACK i: I
I
I
-- - - .. i---------- '- ----i--
.....
!
-- -

I
'
- --

20
.,,.
_.-
"'

I
,,
'
""" ,,_
I
I
I

'
I \
I I
I I \
10 l
I
I I
I
I r
I I
0
I

J,Ooo 1qooo
'

100 roo.ooo
FREQUENCY- CYCLES

Fig. 6-Gain frequency characteristics with and \Vithout feedback of amplifier of

Fig. 2.

always dies away exponentially with time; that of an unstable amplifier

in all physically realizable cases increases with time. Characteristics
A and B in Fig. S are results of measurements on two different
amplifiers; the amplifier having ,B-characteristic denoted A was stable;
t he other unstable.
The number of stages of amplification that can be used in a single
amplifier is not significant except insofar as it affects the question of
avoiding singing. Amplifiers with considerable negative feedback
For a complete description of the criterion for stability and instability and
3

exactly what is meant by enclosing the point (1, 0), reference should be made to
"Regeneration Theory "-H. Nyquist, Bell S)1stem Technical Jo11,rnal, Vol. XI,
pp. 126-147, July, 1932.
 .' TA BILIZED FEEDB.4 CK A "AfPL I FIERS 13

have been tested where the number of stages ranged from one to five
inclusive. In every case the feedback path w as from the output of
the last t u b e to t he input of the first tube.

90 i

80
\ '

' J.
,,

, '

' ....
...
--.... ...

- ....
/
.. ... ... 3F'
,.
...

WI T H FEED BAC K I
....
. ... _

.. ,,
I

\
\
\
\
\
GR ' D \
'
Pa.SITIVE

m 60
I
'
\
\
l
'
aI I
'

'
\
z '
0 '
I \
[I: \ '

50
l
\
\
..

l
0 I
a: , '
l: '
-

.... '...... l
'
\
'
c
.II\
'

\
z
'
'...

'
... ...

<
.... .

... l
...
....

40 2F

0 l
....

z
0
... ... ,
/ '

u
tAJ
VJ
Y. '
'
'

NO f'EtDBACK
,
' I
,.
\
\
'
.
I

'
30
'"
'
'

'
'

HARMO N I C MEASURED AT 15 KC '

'
'
'

20
0 10 20 30 40 50

O UTPUT O F' F1.JNOAM E NTA L.. - M I LLfAMPERE.S I NTO 600 OH MS

Fig. 7-Mod ulation characterist ics wit h a n d \Vithout feedback for the ampli fier of
l<ig. 2 .

EXPERIMENTAL RE SULTS
Figures 6 a nd 7 sho,,r ho\\r the gai n-frequency and modulation char
acterist ics of the three-st age impedance cou pl ed amplifier of Fig. 2
are improved by negative feedback. In Fig. 7 , the improvement in
harmonics is n ot exactly equal to the db reduction in gain. Figure 8
14 BELL S 1r.) TEA1 TE CHNICA L JO UR 1VA L

shows measuremen t s o n a different ampli fier in \vhich harmonics are

reduced as negative feedback is increased , db for db over a 65 db range.
That the gai n with frequency is practical l y independe11 t of small vari
ations in I I is shown by Fig. 9. This is a characteristic of the Morris
town am plifier described in the paper by Messrs. Clark an d Kendall 1
which meet s the se vere requirements imposed upon a repeater amplifier
for use in cable carrier syst ems. Designed to amplify frequencies from 4 kc

IOO -
80

,I
/
95 15
FUNDA M E NTAL O U T PUT H E L D CON S TA N T
AT 2 0 MI L L IAMPERES INTO 6 00
/

90 /
70
m
/
v / 65 0
85
...J

z
/
/' / // 60 IJ...

60
,
w
'

0
c( v /" / i-
c 75 -
... ,,
55
:z " w
:::> v ,. ::::;)
I..... 70 o / "
1 "
.so
Q
"
LL) / z
r
65 -?-"",, _/ ,,, /
45 Q
I-
S) v , , I
/
<
""*'
. " o ..;. "'"
60 40 5
55
JI
v
'

"' f<,.
'/
&/'
35
Q

al
c v o:.!', g
,'
z 50 c.,G 30 z
-
z
v r Q,. t-
45 .J
'
,_/
25 z
-I-
0
v

." LaJ
/
<(
_J 40 ,, /
" "
/
20 w
>
::i
c v / /
,"
0
0

35
v " 15 g:
/
-
30 ,
/4 _ .;
fO
/
" al

25 /
/ '
s Q
"
,

20
- "
0
0 5 10 15 20 25 30 35 40 45 50 55 60 6.5

DB R E DUCT ION IN GAI N DUE TO FEED BACK

Fig. 8-I mprove me n t of harmonics wit h feedback. One exan1ple of another

a n1pl ifier in which with 60 db feedback, harn1onic currents in t he out put a re onl y
one-thousandth and t he i r e nergy one- million t h of the val ues \Vithout feedback.

to 40 kc the maximum ch ange i n gain d ue to variations in plate voltage

does not exceed 7/1 0000 db per vol t and at 20 kc the change is o nly
1 /2 0000 db per volt. This illustrates t h at for smal l changes in I I ,
the ratio of the sta b ility without feedback to the stability with feed
backJ called the stability index, approaches I 1 - /3 I 2/ ( 1 - I fl I cos rll)
and gain stabi lity is im proved at least as much as the gai n is reduced
and usually more and is theoretical I y perfect if cos cf> = 1 / r /3 J .
.
1 I.... oc cit .
 S TA BILIZED f/EEDBA CK A AfPLIFIEJ(S 15

1000

500

J
)(
w
c
z
- 100
\.
>
.... "-
'

- '--""'
-
...J
50
\
Ill

ui
II

\
j \
I \
I i\
/ '.._
10
0.1 as I s too
F R[QUENCY IN K I LOCVC L ES

DB CHAN G E IN GAIN W I TH O U T F' E E D B AC K

-4 -.2 0 +. 2 +4
+ .0 10 ---- -.....-----r- 50.0tO

NO R MA L O P E R AT I NG VO LTAGE
2 50 2 VOLT S
40 f(
z
-
+ .oos --c 50.005 z
< <
CJ) u
- ....

<!
- .....

z<
-m a:
....1- - - - ....
WW
0
50,000 IL
w
z I.&. ...J
a...
<
I l: :E
U 1- A v.,C
- D05 49.9 9 5 <
C:C 3:
0

- .0 1 0 "------.i....-----...a.-- 4 9.990
240 24 5 250 255 2 60
P L AT E B AT T E RY SUPPLY VOLTA G E

Fig. 9-Representative gai n sta bility of a si ngle an1pl i fie r as detern1ined by

n1easu ring 69 feedback amplifiers in tanden1 at l\-I orristo\vn , N. J .
The u pper figu re sho\vs t he absol u te \'a l u e of t h e stabi l it' i ndex. I t ca n be seen
t hat be t ween 20 a n d 25 kc t he i m1lrove n1e n t i n st ability is n1ore t h a n 1000 to 1 yet
the red uct ion i n gai n wa s less t ha11 35 d b .
The lower figure sho\VS change i n ga in of t h e feedback a mpl i fi er wit h cha nges in
the plate bat t ery voltage and t he correspond ing change s in ga i n \Vit hou t feedback.
At so111e freq u e n c i es t he c h a nge i n ga i n i s of the san1e sign a s wit ho u t feedback a nd
at ot hers i t j s of 011posite sign and i t ca n be seen t hat near 23 kc t he st abi l ity must
be pe rfect .
16 BELL S YS TEM TE Cil1VICA L JO URNA L

Figure 10 indicates the effectiveness with which the gain of a feed

back amplifier can be made independent of vari ations i n input ampli
tude up to practically the overload point of the amplifier. These
measurements were made on a three-stage amplifier designed to work
from 3 . 3 kc to 50 kc.
Figure 1 1 shows that negative feedback may be used to improve
ph ase shift and red uce delay and del ay distortion . These measurements

84

80
"

---
76
WITHOUT FEEDBACK
-......
"""

72
"
'\
68
\

64
\

'

'
60

56 ..

m
c
z
52
z

$
48

40
r
-
-

W I TH FCCOBACK
36

32

28

24

2 4 8 10 12 16 18 20 22 24 .
M1 L LrAMPERES INTO 600 i..,:,

Fig. 1 0-Gain-load characterist ic \\'it h and wit h out feedback for a low le vel a111pl ifier
design ed t o amplif ) freq uencies fron1 3 . 5 to 50 kc .
'
 .S T4 BILIZED FEED B.-l L"K 11 .J.lfPLI }'JEJ<}) 17

280
5
26 0
\.
4
'- J.S NI TO MOO N ----I
i 2,a
0 \-WITHOUT FEEDBACK o
t=-o
z
I\
fll

z z 3 a: u
240 - o s ....
'\ )- 0
< lo.I 2 \,. VI !!!
Oj
' ...J If)
l!' \
220 I' ....J
Wl )... -
w
w
' :.'I THOUT fEEDB.ACK O .J
:2
I n;'" 1 K < :::E
...I

5 200
....
01
lo.lz
0-

w 0 100 1000 10000

0
z 180
... .. -- - "" '
fREQUENCY CYCLES PER SECOND IN
-
... -
.. ""' ''
'"" "'

t ' lo.
...

160
--
...

,
!E' ' W I TH FEED BA C K
.... .

j"...,
...
en
14 0
... .

w
-

""""l"-o ""11.'

"'
-..

""'
1 20
...

Q..
t'----._
1 60 ......
r--...
!'-...to..
80 I""'

60
20 '00 1000 10000 20000
rFlEQUENC V I N CYC LES PER SECO ND

F ig. 1 1 -Phase shift , delay, a n d delay d is t ort i o n with a n d wit hout feedback fur a

single t ube voice fre q u e ncy a m pl i fier.

were made on one-tube amplifier, 35-8500 cycles,

an experime nt al
feeding b ack around the low side \V i ndings of the input and output
tr a n sformers.
Figure 1 2 gi ves the gai n -freq uen cy ch aracteristic of an amplifier
with and without feedbac k when i n the {3-circ ui t there was a n equal izer

""""
,
/
BO
i,..o.... f

,,
,,
\ "
,,,,,. WI T HOUT FEED8"CK
"""
v

m 60
j /
Q
z /

,. '

loi'
z
< 40
"
/ i,..o'
....-

v v
"" " l..--"'...
v
,,
l.--' ........
i,... lrH FEE DBACK
zo

#"

v
./ v
.JI

v
0
v /
200 IOOO 5000 IQ,000
FREQUENCY I N CYCLE.5 PER SECON D

F ig. 1 2-Crai n-freq ue ncy characterist ic of a n a mpl i fier ,, i t h a n equal izer in t he

P-circ uit. T h i s \Va s designed to ha,e a gai n frequency characterist ic '\\.' tth feed back
of t he sa m e shape a s t he loss freq uenC)' c h aracterist ic of a n on -loaded t elephone cable.
18 13ELL .)' l,..'5 TEM TE l'l-INI CA L JO URNA L

desig11ed t o make the gai n-frequency characteristic of the amplifier

with fee db ac k _of the same shape as the loss-frequency characteristic
of a non -loaded telephone cable.

CONCLUSION
The feedback amplifie r dealt with in this paper was developed
11rimarily with requi rements in mind for a cable carrier telephone
system , i nvolving many ampli fiers i n tandem with many telephone
c ha n n el s passing through each amplifier. l\1ost of the examples of
feedback ampl i fier performance have nat ural ly been drawn fro m
ampli fiers designed for this field of operation . In this fiel d , vacuum
t ube amplifiers normal ly possessi ng good ch aracteristics with respect
to stability and freedom from distortion are made t o possess super
l atively good characteri stics by application of the feedback pri nci ple.
However, certain types of ampli fiers in which economy has been
secured by sacri ficing perform ance characteristics , particularly as
regards distortion , can be made to possess im proved characteristics
by the application of feedback. Discussion of these amplifiers is
beyond the scope of t h i s paper.