May,

I949

\Virele!is World

I8g

WHEN NEGATIVE FEEDBACK

ISN'T NEGATIVE
The Cause and Prevention of Oscillation and Distortion
By "CATHODE RAY"

OST
M

experimenters
have played about

who

feedback can go wrong, let us

much
with negative feedback
must have had some results that

make sure we see the thing itself

quite clearly.
We start off w ith any ordinary

were not according to plan.

Unless, that is, they confined

amplifier, represented
dotted box in Fig. I.

their researches to a single resist

ance-coupled stage.
For it is
of an amplifier is fed back over
several stages there is a great risk

call these voltages 'Vi and

respectively .
Then v0/vi is

of oscillation, usually at some

frequency outside the working

voltage

range of the amplifier. An audio

amplifier,
for
example,
would
most probably oscillate at an

amplification ;
bv
A.
,

denote it
Negative
applied; in

feedback

this
example, by con-

Being in

audible, it might go unnoticed as

oscillation, and the inexperienced

necting part of the

output voltage in

thing
done entirely outside the

low output and quality of thcc

amplification.
It may not be so generally

i_

the load imped

ance is kept the

oscillation there may be peaks in

the
frequency
characteristic

quite contrary to what one usually

intends!

same, it has no

dealt

there is still A,

effect on what goes

on inside the box.

of negative
pretty fully

what

:;;!,_
V ;;;:

dotted box ; and

so, provided that

realized that even if there is no

with in

series with the

input. It is some- 11

experimenter would be at a loss

to account for the disappointingly

This seamy side

feedback has been

the

that when we apply a certain

signal voltage to its input we get
a certain signal voltage at the
output.
For convenience let us

well known that when the output

inaudible frequency.

by

We know

The amplification

highbrow

let

is
t
Vit

v0

the
us

next

and the gain of the whole outfit

as
everybody

is A/(1 + AB),
knows.
What I w ant

to emphasize is
that in any inquiry into negative
feedback it is fatal to use the

external input, Vi, as the starting

point . Always begin with 'Vi or v0
Then the other (v0 or vi) follows

ro1n ordinary amplifier know

ledge, without any feedback com
plications.

And when once you

have decided how much of v0 to

feed back, Vi is arrived at merely

by adding this fed-back voltage

to

'V;.

say

Though

ri:bi_!_E1

<>-'------<>

Au;

0-------0

"merely

by

adding," I don't
mean that it is

always
just
as
simple as adding

and 3. If it
were, then the

v0 2

Fig. 1.
In an amplifier using
negative feedback, the amplifier
proper (within the dotted lines)
works quite normally, and the
results
of
the
feedback
are
accounted for by what is done

outside it.

complications we
are about to con
sider would not
arise. The source
of all the trouble
is what we have

denoted by the de

ceptively
simple
letter A.3
It is
what is known as
a

vector quantity.

writers refer to as " the litera

ture," but mostly in a mathe
matical
style
that
only
the

and an input signal of vi volts

still delivers an output of v0
volts. In practice, of course, the

That is to say , it doesn't merely

denote the number of times the
output voltage is greater than the

Actually it is not a really difficult

subject, and requires only an ele

unit, represented by the outer box.

The dotted line in Fig. 1 is an

input. One could say, for example,

lag.
that A was 150, with 30

amplifier and the feedback circuit

a re both made up into a single

brighter students are likely to

regard as other than forbidding.

mentary knowledge of vectors or

"j" notation, or preferably

of the
both.

Readers

invisible boundary, and its input

Another way of presenting this

sary to apply to the external i nput

-and generally the most con

venient-is by the "j " notation,

, terminals" may be merely un

marked points on the wiring.
So the voltage which it is neces

who have been

able to follow my dissertations on

these1 should find it an excellent

example to test their proficiency.

Probably the real stumbling block

terminals (call it Vi) in order to

get v0 at the output is equal to vi

itself are a trifle hazy.

fraction B, then the fed-back

voltage is Bv 0, which is equal to
ABvi.2 So the total external input
must be vi + A Bvi = vi(I + AB),

plus the fed-back voltage, which

is a fraction of v0 If we call this

in most cases is that the basic

principles of negative feedback

So before

we start considering how negative

"j " .

- Feb., 1948, p. 68.

"A.C Bridges," Apri l, 1948, p. 139.
H Phase," June, 1948,
p. : '..

input, but it must also specify the

phase of the output relative to the

If the whole of v0 is fed

back,

B - 1.

information is by means of a
vector in a diagram. Still another

which

in

130 + j75.4
3

this

case

would

be

Strictly speaking the same complications

apply to B too, but they are usually less

important there, provided that one sticks to

simple resistance feedback circuits and does not

try any tricks with inductance or capacitance,
to make the feedback depend on frequency.

4 130 and 75 are d erived from 150 and 30

in the usual way by multiplying 150 by cos 0
and sin 30 respectively.

Wireless Worltl

190
When Negative Feedback Isn't
NegativeThe simplest case is the one

with zero phase difference between

output and input voltages, because
then the fed-back voltage adds
directly to 11 to give the input
This can be shown
voltage, V;.
in

simple vector diagram, Fig. 2.

vt

b Id

Fig. 2. Vector diagram applying

to Fig. I when the output voltage
is in ph ase with the input.

The vector oa is drawn to repre

sent vi, and oc is drawn A times
as long, to represent v0 Being in
the same direction, it represents
an output exactly in phase with
the internal input (vi), such as
would be the case with an ideally
simple cathode follower or 2-stage
resistance-coupled amplifier.
ob
is then marked off along oc to
rpresent the'' fed-back fraction,
Bv0; the external input, ad, is
oa + ob. (In a cathode follower,
the whole of the output is fed
back, so ob coincides with oc, and
the input voltage is greater than
the output by the amount 'V;.)
Since in
this case all
the
quantities are in phase, it is much
easier to add them by simple
arithmetic than to draw a vector
diagram.
The only purpose of
Fig. 2 was to show the principle
of the thing, for comparison with
other cases. And of course the j
method is quite unnecessary in
this case, because j indicates the
out-of-phase component, which is
non-existent.
Next, consider a simple resist
ance-coupled
audio
amplifier,
Fig. 3 (in which provision for grid
bias and other details have been
omitted for clearness). The only

visible components whose be

haviour depends on frequency are
the coupling capacitors C1 and c.,
and they are normally chosen so
that their reactance at all working
frequencies is negligible, in which
case the output voltage is in
phase with the input and Fig . 2
app lies.
At very low frequencies, how
ever, the reactance of C1 is

appreciable in comparison with

R2, and these two components
form a sort of potential-divider.
Only part of the output of Vr
reaches the input ofV2. What is
more, the< current through a
capacitor leads the voltage across
'0
it by 90 ; and, since the voltage
across Ri must be in phase with
the current through H.2 and C1,
the voltages across R2 a nd C1
out of phase with one
are 90
So when the frequency
another.
is low enough for the reactance of
cl to be appreciable, not only
does the amplification begin to
drop, but also the phase of the
output starts to lead the input.
As a matter of fact, it is the
phase that is the first to start

changing noticeably. This doesn't

matter in a " straight " amplifier
used for listening purposes only,
because the ear cannot detect even

the maximum phase shift. But if

negative feedback is used it does
matter. To see how, we must go
into the matter more closely.
Assume that the signal input to
VI can be varied in frequency but
is constant in amplitude, yielding
a certain outp ut (v al ) at the
anode.
If the resistance R2 is
very large compared with R1 and
r1a (the anode resistance of Vr),
then the additional impedance of
cl at very low frequencies will
not affect vat appreciably.
So
we shall assume that
va l
is
constant too, and therefore can be
represented - by a vector line of
fixed length (oe in Fig. 4).
The voltages across C1 and R2,
which we can call v cl and v g2
respectively, can also be repre
sented by vectors , which will have

May,

1949

they differ in phae by 90 , their

vectors, fe and of in Fig. 4, must
always be at right angles to one
another.
You can make a working mod.el
of this vector diagram under these
conditions by sticking pins in the
points o and e and pushing the

right-angled corner of a card

between them, ignoring the part
of the card below oe.
One ege
of the car d will form the vector

of

and the Other fe.

Except at low frequencies, t4e
reactance of cl is so small com
pared with R 2 that the voltage
across it (v c1) is negligible ; this
condition is represented by holding
the card so that its edge of
coincides with oe,. and fe dis

appears. But as the frequency is

reduced, v cl correspondingly in
creases, as can be shown by
bringing fe into view, still keeping
the card pressed against the pins.
/

/
/
I
I
I

---

'

"

'f

Fig. 4. Vector diagram applying

to the C1R2 portion of Fig. 3 (and
also to C3R!), showing how v 02
is. related to Va1

To do this you must turn the card

anti-clockwise, so that its edge of
indicates a phase-shift, <f>. B ut at
first its length is hardly affected.
grows, howi>:ver,
As
vc1
v02
.
dwindles at an mcreasmg rate ;
until finally, when vcl becomes

relatively large, v u2 rapidly dis

appears while the angle of phase
difference
ap

proaches 90 quite
slowly.
The cor
ner of the card
Ci
(as is proved in
geometry) traces
out the circum
ference of a semicircle, dotted in

Fig. 4.
To make the
chap_ges in v 0 2 and
<P clearer in rela
tion to frequency,
Fig. 3.
The effects of the coupling capacitances,
they
can
be
C1 and C3, and the stray capacitances (not shown)
plotted
on
a
in this amplifier cir.cuit are considered in detail.
frequency b a s e
to fulfil two conditions. The first
as in Fig. 5.
The frequen cy
is that they must of course always
scale shown holds good for all
add up (vectorially) to equal Vai
combinations of C1 (in F) and
And since, as we have just seen,
R2 (in MO), which when multiplied

"gz

R4

OUTPUT V0

LO----'---+----- t ---0-1
-'--

May, i949

\Vireless \Vorld

If an amplliier could be made

strictly according to Fig. 3 there
would be no limit ; but un
fortunately there are the " inv i sib l e
com
p o n e n t s ''-stray
capacitances. One
I o
J...
lot of these, in
09
,__ __
-"'
/ Ug1
eluding the input
80
oa
'
v.,
J
capacitance of V2
70
01

d
/
and the output
06
S- 60
'
capacitance of Vr,
05
'
'
comes across R2,
0:4
.: 40
:::;
',
so we shall call it
03
v
30
'
C2
By using
2J
01
v
Thevenin's theo
OI
----rem5 we can boil
os
10"
DO 01 002
01
oos
02
down the parts of
1FREQUENCY (c/s)
the c irc u i t con
cerned to Fig. 6,
Fig. 5.
Both .P and v gz in Fig. 4 depend on
in which R is
frequency ; here they are plotted on a frequency
base to bring out this relationship.
equal to I{al R1
and H.2 in parallel,
fed by a generator giving a
must be divided by the number of
constant voltage equal to v av
megohm-microfarads.
If there is another coupling, when C2 is removed. This voltage
has been marked 1! aio
C3R4 in Fig. 3, it behaves simi
Now the only difference between
larly ; and the combined effect
this problem and the one already
of the two is calculated by adding
solved for C1R2 is that the desired
their individual phase shifts and
v u2 comes across the capacitance
multiplying amplitude ratios. So
instead of the resistance ; so of
the total phase shift due to the
course one wants this capacitive
two couplings approaches r8o0
reactance to be as large as possible
lead at the lowest frequencies.
relative to R.
Tht> appropriate
In practical amplifiers this very
low-frequency behaviour is gener
ally a good deal more complicated.
Capacitors used for smoothing the
0
main power supply, decoupling
individual valve feeds, and by
passing bias resistors, tend to
r:p 0
become ineffective ; with the re
I+-- BANDWIDTH
- 30
<.:>
sult that the impedances they are
\
-45
:5
supposed to short-circuit cause
-60
various positive or negative feed
-90
FREQUENCY
backs that may do almost any
thing to the frequency characteri
Fig. 7.
Typical frequency char
stic of the amplifier. The cunning
acteristic of an amplifier of the
designer can so,metimes turn t!).ese
Fig. 3 type, without feedback.
The cut-off frequencies, which
effects to his advantage, as for
limit the effe,:tive bandwidth, are
example in bass-boost circuits;
detemiirted r:by the RC (time.::
or he may make the capacitances
constii.nt)''vitlues of the circuit.
as large as can be afforded, to
push the trouble below the lowest
vector diagram is like Fig. 4 in
working frequency.
But, as we
reverse.
The. frequency curves
shall see, that may not dispose of it.
have the s?:me s lfape as those in
So much in the meantime for the
; l}at they top are
lpw frequencies; what about the Fig. 5 exC
reversed; the amplitude ratio is
high?
until some ,::'fairly
practically
high frequeti'C'y, when it begins to
Fig. 6. The effect
fall off, artd at the same time' the
of stray capaci
phase shift begins to grow-but
tance in Fig. 3 is
this time it i.? a lag.
As with
made clear with
_
C1R2, at the frequency which
the help of this

together are equal to r megohm

microfarad.
For other combina
tions the shapes of the curves are
the same but the frequency figures

?!.>'

g so

E'1J
;)"

"equ iva l e n t
circuit."

3 " Th6venin's,
p.109

Theorem,"

:\larch-,

1949,

Wireles!li World
When Negative Feedback Isn't
Negative-:-

makes the reactance equal to the

resistance, the phase shift is 45

and the amplitude ratio o.707

(i.e. 1/y2). Obviously the lower

the combined valve and coupling
resistance
(R) the higher the
frequency before the phase begins
to shift and amplification falls off.
Putting all this together, then,
the frequency characteristic of a
re sista nce-c oupled a mpli fier with

one series-C coupling and one

shunt-C stray ea pacitance, and

leaving out of account any other

influences :mch as power-supply

impedance, is as shown in Fig. 7.

The two curves together specify A;
I Al being the symbol for its
numerical magnitude alone. The
two sloping ends are copied from
Fig. 5 and its mirror image, and
can be made to apply to any
amplifier by placing them so that
the points where \A\ has dropped
to 0.707 times maximum come

where the appropriate resistances

and
capacitive reactances
are

ob is of c our s e unch ang ed, but oc

has been made 1.41 times longer
and given the corresponding phase
lag of 45. The required input,
given by vectorial ly a dding ob
and oc, is ad, which is much less
than 1.41 times longer than ob,
and also its angle of lag is much
smaller than rf>. The more nega
tive feedback is used, the less is
t he phase shift and drop in
amplification due to whatever
oc does.

So the effects of negative feed

back on the frequency character
istic, Fig. 7, are : (1) The fiat
top is lowered (from A to A/
(r + AB), as we saw at the
beginning) ;
(2) the fall-off at

each

less

pronounced

= 450

o-.:t"/"':::
:::' ::
::: ::
:: : --+a

-------

(a)

can be seen from Fig. 2. Let oc

represent the wanted output. Then
oa represents the inp ut required to

give it, with no feedback. Over the

fiat-top par t of Fig. 7 the length

voltage

must

be

increased

by

'a
(c)
d

..

(d)
== 180
Fig . 8. Vector diagrams showing

various conditions in amplifiers

The vectors
with feedback.
represent the same voltages as in

the factor 1/0.707

1.41.
With feedback, a greater input,
oa + ob
say, is needed, so \Al
is low even over the fiat region.
But ob, which can be made by
far the larger part of oa + ob,
is a. constant proportion of oc,
so a falling off in the internal
gain of the amplifier, which
affects oa only, has relatively little

at the same time it sw ings round

nearly at right angles to ob,
Fig. S(b). So finally the ampli
fication and phase suffer almost
as badly as th e y did (at some
lower frequency) without feed
back.
Similarly, at the low

phase is affected too, so at the high

frequency end the vector diagram
becomes something l ike Fig . Sa.

that includes
within the feedback loop two
RC circuits of the same tendencv,
the vecto r oc grows twice as faS't,

effect on the overall gain. '

It must be remembered that the

Fig.

2.

frequency end.
, With an amplifier

r8o0

phase shift approaches

This is where things begin

its

to get interesting.
Fig. S (c) ,
for example, shows the condition

where each of two similar RC

circuits is giving a lag of 60
and from Fig. 5 it can be ascer
tained that the relative ampli
fication is 0.5 X 0.5, so oc must
be four times as long as in Fig. 2.
In spite of thi s, ad is act ua ll y
shorter th an in F ig. 2, so the
overall gain is higher.
(This
assumes, of course, that the
amplifier can handle the internal

input without being overloaded ;

if not, distortion may be violent).

So instead of the overall ampli

fication falling off,

as it

would

with no negati ve fee dback, it

rises. This can't go on, though ;
as </> approaches 90 per RC

circuit

the internal amplification

drops off so rapidly that oc

becomes immense, and ad like
wise.
But now consider what may

happen

with

three

similar

RC

circuits. At the frequency where

each introduces a lag of 60,
the total lag is 180.
And i f
oa in Fig. 2 wa s one-eighth of
ob, it is now equal to it, so we
get the result shown in Fig. 8(d),
where ad has shrunk to nothing.
In other words, the amplifier

( b)

How it does this

of oa will be constant, correspond

ing to constant a mplifr::a tion .
But at low or high frequenc ie s,
wher e the ampl i ficati on falls, the
length of oa has to be
increased to keep the
c
output constant.
For
example, at the marked
frequencies, where \Al drops to
o. 707 of its maximum, the input

(3) the phase shift at each end

is less. But the benefits (2) and
(3), can't last for ever as the
frequency is raised.
In the end
the internal input, represented
by the vector oc, must become
large-even larger than ob-and

equal. The amplifier frequency

band is commonly regarded as
e xtendin g from one of these fre4 uencies to the other.
One of the objects of negative
feedback is the widening of this
frequency band.

is

end

and

May, 1949

will give output at this frequency

without any input at all.
In
still other words, it is self
osci llat ing .
The same thing is liable to
happen at a frequency lower
than the working range, if there
are three RC cir cuits of the
series-C type.
At first it might seem a very
unlikely coincidence that oc would
be exactly equal to ob when
q, was exactly 180, and so the
risk of oscillation would be small.
But this is not so.
Make oc
in Fig. S(d) any length you like,
less than ob. Then the external
input, ad, must be in phase with
ob. So if ad is reduced, say to
zero by shorting the external
input terminals oc must in
crease correspondingly to preserve
But that makes
the balance.
oa and consequently ad increase,
so oc must increase more. And

so on, until the amplifier is

overloaded and its amplification

reduced to the point at which

ob
oc and oscillation is main
tained at a steady amplitude.
=

Vve have just found that if

an

May, 1949

Wireless Worlll

circuit e b rac ed b:
negative feedback loop contains
three similar RC circui ts there
will be oscillation unless AB is

amplifier
a

back over
cluding no

only one stage, in

transfornwr. But one
stag e with heayy feedback gives
hardly any amplification.
Two
stages, again with no transformer,
offer more useful possibilities,
without risk of oscillation, but

less than 8. (By " similar " I

mean having the s3.me RC; values

and tending to cut frequencies at

the same end.)
With four such

can develop peaks.

circuits the critical phase shift

in each is only 45 and the ratio

in each (see Fig. 5 again) is
o. 707, so the oscillation value of
AB is only r/0.7074
4.
But
we can easily see
from the diagrams

-;;

that even if feedback is kept well

below these fatal :::
figures it may still
be enough to raise

peaks, as in Fig.
9; and these may

cause things like

gra moph o n c

j j'-\
FREQUENCY

Fig. 9.
Typical effect of applying
negative feedback to an amplifier
having more than one circuit
cutting off at about the same

frequency.

what

As is explained in the books,

inductance and the stray capaci

tance. A feature of such a circuit

in

circuit can easily cause high

The truth of this can be shown

a more professional manner

by the "j" method; and any

and the input (across the whole)

frequency oscillation.
It can be shown that at the
low-frequency
end the trans

be No. But if you

body who wants to go into the

matter more deeply is advised to
consult an article by C. F. Brockel

is that the phase angle between

the output (across the capacitance)

above resonance.
So feedback
across one transformer and one

to cover the same

frequency band
the answer would

circuit, composed of the leakage

swings from a small lag below

the resonant frequency, to 90
at resonance, and approaches 180

design each stage

try combining the

effects of circuits having different
cut-off frequencies you will find
that more feedback can be used
before peaks appear. In particular,
if three shunt-C circuits arc in
cluded, as there usuallv will be in
three stages, it is best to make one
of them cover a narrower frequenc,
band than the other two.

at high frequencies a transformer

usually becomes approximately
equivalent to a series resonant

RC

adopts

would normally be
icL-------------- a sound econom
al principle - to

scratch and motor

rumbles to be brought into undes irable prominence.
If a transformer is included in
the system, the danger is greater,
at least at the high-frequ ency
end.

ls it possible

to include more than two phase

shifting circuits (counting one
transformer as two circuits), to
combine high amplification with
a full measure of the benefits of
negative fccdback ?
If one

:z:

193

sby in the March 1949 Wireless

He shows how one

Engineer.
can

design

for

"maximal

flat

ness," which means "staggering"

the cut-off frequencies of the
circuits so that feedback can be
used to extend the frequency
coverage as far as possible, just
short of allowing peaks to appear.
The tendency to peak, controlled

former is roughly equivalent to

one RC circuit.

in this way, is useful for squaring

to use a lot of feedback.

Yet,
paradoxicaily , in using it one

frequency curve, without going

so far as the curve of Fig . 9.
If your amplifier gives trouble

r isk of causing oscillation and

peakiness. The advie one usually
gets about this is to see to it

coupling resistance for the middle

stage and higher values for the two

To make an extremely stable

and level amplifier it is necessary

seems certain to run a serious

that the amplification has fallen

well below the danger point at

the frequencies where the phase

shift is 180. But, as we have

seen in arriving at Fig. 5, the drop
and the shift are bound together
by the nature of the circuit.
One line of policy is to feed

the shoulders of the amplification/

when you feed back over three

stages, then try using a low anode
outer ones.

Or, if a transformer

is included, make sure that the

other circuits cut off at a higher
frequency.
Of course, it is best

to work out the design fully and

check by tests ; but the foregoing
trial-and-error

than nothing.

hints

are

better