Web: http: // www. pearl - hifi . com 86008, 2106 33 Ave.

SW, Calgary, AB; CAN T2T 1Z6

E-mail: custserv @ pearl - hifi . com Ph: + .1.403.244.4434 Fx: + .1.403.245.4456
Inc.
Perkins Electro-Acoustic Research Lab, Inc. ❦ Engineering and Intuition Serving the Soul of Music

Please note that the links in the PEARL logotype above are “live”
and can be used to direct your web browser to our site or to
open an e-mail message window addressed to ourselves.

To view our item listings on eBay, click here.

To see the feedback we have left for our customers, click here.

This document has been prepared as a public service . Any and all trademarks
and logotypes used herein are the property of their owners.

It is our intent to provide this document in accordance with the stipulations with
respect to “fair use” as delineated in Copyrights - Chapter 1: Subject Matter and
Scope of Copyright; Sec. 107. Limitations on exclusive rights: Fair Use.

Public access to copy of this document is provided on the website of Cornell Law School
at http://www4.law.cornell.edu/uscode/17/107.html and is here reproduced below:

Sec. 107. - Limitations on exclusive rights: Fair Use

Notwithstanding the provisions of sections 106 and 106A, the fair use of a copyrighted work, includ-
ing such use by reproduction in copies or phono records or by any other means specified by that section,
for purposes such as criticism, comment, news reporting, teaching (including multiple copies for class-
room use), scholarship, or research, is not an infringement of copyright. In determining whether the use
made of a work in any particular case is a fair use the factors to be considered shall include:

1 - the purpose and character of the use, including whether such use is of a
commercial nature or is for nonprofit educational purposes;

2 - the nature of the copyrighted work;

3 - the amount and substantiality of the portion used in relation to the copy-
righted work as a whole; and

4 - the effect of the use upon the potential market for or value of the copy-
righted work.

The fact that a work is unpublished shall not itself bar a finding of fair use if such finding is made
upon consideration of all the above factors

♦ PDF Cover Page ♦

♦ Verso Filler Page ♦
 Part IV — This circuit,

the “long-tailed–pair”,

is considered by the

PUSH - PULL DRIVERS author to be the best

of the phase splitters.

BY GEORGE FLETCHER COOPER

A lightly edited, redrawn and re-typeset version of an
article that appeared in Radio-Electronics in March, 1953.

N THIS SERIES OF FOUR ARTICLES about the average DC potential, because any unbalanced dif-
I phase-splitting stage that provides the link
between a single-ended voltage amplifier and a push-
ference will pull the undeflected spot away from the
center of the tube.
pull [output] stage, we have examined several poor
circuits and two classes of acceptable circuits. We T H E L O N G - TA I L E D – PA I R
have seen that a single tube with a split load is good One phase-splitting circuit is particularly good
if you do not need too much drive while the anode- for cathode-ray oscilloscope work. It has other
follower or see-saw circuit gives more output, but uses too, but we shall come back to those after we
less gain. For conventional audio work, the tubes and have examined its characteristics. The circuit
supply voltages you plan to use will determine itself is known as the “long-tailed–pair” or more
whether you take a single-tube or two-tube phase- prosaically, as the Schmitt cathode-coupled phase
splitting approach. inverter. If you look at the basic circuit in Fig. 1,
There are some special jobs, though, for which you will see why the name “long-tailed–pair” was
these push-pull driver types are not suitable. The adopted. In a rather interesting way, this circuit is
most important of these is when you want to go related to the anode-follower discussed last
down to extremely low frequencies, or even all the month. That circuit, you may remember, could be
way to zero frequency. The anode follower (see Part described simply by saying that the second tube is
III, in the February issue) includes one coupling driven by the difference in the plate-voltage swings
capacitor, so that it will not stay balanced once the of the two tubes.
capacitor starts to take control; the split-load cir- In the long-tailed–pair, the second tube has its
cuit (Part II, January, 1953) has the disadvantage grid grounded and the effective drive to this tube is
that the two output terminals are at different DC applied to its cathode. The driving voltage is equal to
potentials. A symmetrical direct-coupled deflection the cathode resistance R k multiplied by the differ-
amplifier for a cathode-ray oscilloscope calls for a ence in plate currents of the two tubes. Referring
phase splitter that provides two outputs at the same again to Fig. 1, suppose we raise the potential of grid
Y
B+ (a) B+
RL1 RL2 A I2 RK RL1
C1 0
Output
g
C2
Input Output I1 RK Y Input
1 2 1
RIn
V1 V2 X Y V1
P1 A 0 V2
G
P2
G 2
Power Supply & RL2
Signal Common
RK (b) RK
X
B– +
Fig. 2. (a) Current–see-saw diagram of
Power Supply &
the “long-tailed–pair”. (b) See-saw diagram Signal Common RIn
o f b a l a n c e d - vo l t a g e p h a s e i nve r t e r
Fig. 1. Circuit of the “long-tailed–pair” reprinted from last month’s article for com- Fig. 3. The “long-tailed–pair” redrawn to
cathode-coupled phase inverter. parison with the balanced-current type. show the grounded-grid operation of V2.

♦ AN 23.2 - Page 17 ♦
1 by 1 volt, thereby causing an extra current I1, to RL2 I 1
flow through V1-R L1. The current in V2-R L2 will drop = 1 = 1+
RL1 I 2 g m + Rk
by an amount I2, and the cathode potential will
change by (I1 – I2)Rk. This change is the input to V2, Typical values
and if Rk is very large, we can have (I1 – I2) very for this circuit are:
small and still get the neccessary drive at V2’s cath-
ode. In fact, we can draw Fig. 2a for comparison Rk = 5KΩ and
with Fig. 2b, which is the see-saw–voltage diagram gm = 5ma/v (5,000 µmhos),
from last month’s article. Looking at Fig. 2a, you will
notice that in order to get any drive at all to V2, it is giving a ratio: RL2 / RL1=1.04.
necessary to have unequal values of I1 and I2 (C1-Rk This means that, even without unequal load
and C2-Rk)and the drive for V2 is represented by the resistors, we get a balance correct to 4%, which is
value O-g. quite good.
The only way to get equal swings at the two
plates is to use slightly different plate-load resis- THE RIGOROUS APPROACH
tances RL1 and RL2. We can do some very simple cal- This result, obtained by very simple reasoning,
culations to see the sort of differences we can expect applies quite well if V1 and V2 are pentodes, though
in the two values. it leaves a rather tricky gap if you wonder how to
At plate 1, we have a voltage swing of I1RL1. The bypass the screen grids right down to zero fre-
input to V2’s cathode is (I1 – I2)RK, which gives a quency. It shows that the circuit looks good, anyway.
plate current of gm(I1 – I2)RK RL2, which is, of course, And now we come up against the main problem of
also equal to I1RL2. Therefore, gm I1RK RL1 = I2(1 + gm the writer of technical articles: do you want it easy,
RK )R1,2 and: or do you want it right? Sometimes the writer can
I1 1 put in all the mathematics—if he is French he seems
= 1+ to put in nothing but the mathematics—sometimes
I2 g m Rk
he collects a dense mass of small-print equations at
Since we want: the end. Usually he must tack between the Scylla of
editorial condemnation and the Charybdis of long-
I1 RL1 = I2 RL2 haired readers who write and point out the smallest
deviation from rigour.
we have: My preference is to start off with the exact math-
ematical solution and then simplify. That way, any
reader who wants to get more detail can build up on

B+ B+ B+
Spot-
RL1 RL2 RL1 RL2 centering RL1 RL2
pot.

Input Output Output CG1 Output

1 2 1 2 1 2

Input
V1 V2 V1 V2 V1 V2

RG1 RG2
Power Supply & RB
Signal Common
RK

Pentode
used as RK V3 B– RK CG2

B– Power Supply &

Signal Common
Fig. 5. When using the “long-tailed–pair”
to feed the balanced deflecting plates of a
Fig. 4. A high-impedance pentode as a cathode-ray oscilloscope, the grid of V2 Fig. 6. One method of eliminating the
common–cathode-resistor swamps out may be returned to an adjustable, DC negative voltage supply for the cathode
variations in tubes 1 and 2 that might unbal- divider to restore the beam to the center of return. RB is tapped off the total cathode
ance the phase-inverter currents. the ‘scope screen. resistance at the desired bias point.

♦ AN 23.2 - Page 18 ♦
the sound foundations I have provided. If I simplify 5KΩ. We can choose RL2 = 50KΩ and we obtain:
first, there is nothing to build on later. RL1 / RL2 = 51 x 5,000 / 50,000 + 10,000 (50 x 5,000)
It will be much easier to understand the long- = 255,000 / 310,000 = .823 whereas with the
tailed–pair if we redraw the circuit in the form of Fig. simple formula we had: RL1 / RL2 = 1/1.04 = 0.96.
3. This is not as elegant and symmetrical as Fig. 1 but There is quite a difference between a 17.7% and
it is a good deal more useful. V2, you now see, is a a 4% unbalance; we are obviously justified in
grounded-grid stage, a subject we discussed recently in adopting the more rigorous approach. Energetic
RADIO-ELECTRONICS (October, 1952 issue). I shall save readers may care to consider what happens if V2
myself a lot of trouble by using some of the results I ages and Rp2 gets bigger: as Rp2 represents only
worked out in that article. In particular, the input 10,000 in the total 310,000 of the numerator it can-
impedance (RIn ), looking in at the cathode, is RL2 + not be a very serious factor. If µ 2 gets smaller,
Rp2 / 1 + µ2 , and the gain is (1 + µ2 ) RL2 / RL2 + Rp2. things are rather more complicated, because µ2,
First, let us look at the stage V1. The tube has a appears in both denominator and numerator; any-
plate load RL1, and there is some negative feedback way, µ2 is more nearly constant during the life of
because of the unbypassed cathode resistor. If we the tube.
represent the cathode circuit (RK in parallel with the
input resistance of V 2 ) as R x the gain is just O S C I L L O S C O P E A P P L I C AT I O N S
µ1RL1 [ RL1 + Rp2 + (1 + µ1) Rx ] and the voltage across For applications in which a good, stable balance
the cathode circuit is Rx / RL1 , times the voltage at is needed, a much higher value of RK would be used.
the plate. So if we have 1 volt at the grid of V1 we As an example, we might take RK = 50KΩ, when:
have µ1RL1 /[ RL1 + Rp1 + (1 + µ1) Rx ] volts at plate l,
µ1Rx /[ RL1 + Rp + (1 + µ1) Rx ] volts at the common RL1 /RL2=2,550,000/2,560,000 = 0.996
cathode, which we can write as Vx, and [ (1 + µ2)RL2 /
+ Rp2 ] Vx at plate 2. and the unbalance is only about .5%. Changes in
We want the voltage at plate 1 to be the same as tube characteristics can only unbalance the circuit
the voltage at plate 2, so we put: to the same limited extent assuming that we start
with correctly proportioned load resistors, so that
µRL1
=
(
 1+ µ R 
2 ) L2
Vx
we have a very satisfactory circuit here but 50KΩ is
a very long tail.
( ) R
RL1 + R p1 + µ 1 + 1 Rx  L 2 + R P2 
 The reader who has been watching these num-
bers carefully may be getting a little worried. If each
and: tube draws 5mA, there will be a total of 10 mA
through the cathode resistor. all this talk of 50KΩ
Vx = µ1Rx /RL1 + Rp1 + (1 + µ1)Rx implies a drop of 500V across RK. Even if we drop RK
to 20KΩ, we still need a –200V supply if we are to
There is one more equation, because Rx is made work with the grids around ground potential. This is
up of Rk in parallel with Rin—[ (RL2 + Rp2 ) /(1 + µ2) ]— not a serious matter in oscilloscope circuits, because
so Rx = RK (RL2 + Rp2) / (RL2 + Rp2 + µRk). negative high-voltage supplies are generally used. In
That last paragraph is difficult but true. If you some other circuits, where the grid of V1 is con-
assume the tubes are similar and struggle with it for nected directly to the plate of a preceding tube it is
a while, you’ll come out with the equation for true an advantage to have the whole tube circuit lifted up
push-pull balance: above ground. But 200–500V is rather high and
when an extra-large cathode impedance is needed
RL1 / RL2 = (1 + µ) RK /RL2 + Rp2 + µRk for special high-balance jobs, special circuit tricks
are usually adopted.
This doesn’t look too bad, and if we assume fur- One very important circuit uses a pentode in
ther that we have pentodes, so that µ and Rp2 are place of RK. For example, a 6AQ5 with 200V on the
very large 1 + µ ≈ µ and gm = µ/Rp, we arrive at: screen grid will pass 10mA with only 20 volts on the
plate, but the impedance so far as constancy of cur-
RL1 / RL2 = gmRK /(1 + gm) rent is concerned will be very high, certainly above
100KΩ. This means that we can make RL1 = RL2, and
which is equivalent to the answer given by our very still have a virtually perfect balance. The form the
simple treatment. circuit takes is shown in Fig. 4, which also shows a
It’s interesting to compare the exact solution to voltage regulator tube as a screen bypass, to ensure
the simple one for a typical case: consider a 12AT7 operation down to zero frequency.
with Rp = 10KΩ, µ = 50, gm = 5,000 µmhos and RK = Another way of providing the necessary high

♦ AN 23.2 - Page 19 ♦
impedance in the cathode circuit is to use a satu- interested in the variations of a voltage which is
rated diode for RK. The tungsten-filament diodes always well away from zero, this push-pull input is
used as noise sources in receiver testing give full not very satisfactory, because the DC component of
emission (saturate) at a relatively low plate voltage, the input will deflect the spot away from the center
and increases in plate voltage give almost no change of the screen. The arrangement of Fig. 6 is then more
in current. The only disadvantage in using them is useful. The grid of V2 is connected to point A, which
that the diode current depends on the filament tem- provides a positioning voltage to bring the spot near
perature, so the filament current must be stabilized. the center of the screen. The grid of V1 takes the
Our purpose in seeking such highly accurate input signal, which is converted to push-pull to
balance is not to get 10.01 watts from an amplifier avoid defocusing and trapezium distortion [key-
instead of 10 watts, but to meet the requirements of stone-ing]. A capacitor (shown in dotted lines) can
some special measuring instruments. If you look be added to keep supply-hum off the grid of V2,
back at Fig. 1, you will see that apart from the where it would be amplified as an ordinary signal A
ground connection of the grid of V2, the circuit is resistor (also shown dotted) is sometimes added to
absolutely symmetrical. In fact, this ground connec- improve the smoothing. (In some oscilloscopes the
tion is there only because we assumed a grounded time constant of this RC filter is so long that the spot
input: we are really using the voltage G1 – G2 as the goes on drifting long after you have taken your hand
input, and deriving two, equal, antiphase outputs off the positioning control.)
from this. Suppose, however, that we connect both Stages of this kind can be connected in cascade
grids together, and then apply a signal. There will be if you have generous power supplies. The subject is
an in-phase [common-mode–error] signal at the slightly outside our present field, but you can see
plates, but it will be relatively small, since the large that if the first pair is to operate at about zero grid-
cathode resistance RK provides a great deal of feed- volts, the plates will be up at about +l00V, while the
back. We can easily calculate what will happen, cathode resistor is returned to, say, –150V. The sec-
because we can assume that each tube has 2RK in its ond stage grids are then at +100, so we have a 250V
cathode and then treat one tube alone. From grid to drop in the second cathode resistor (assuming this
plate, the gain is: also goes back to –150V. The second stage plates will
be at +200–250 volts, so that the supplies needed will
µRL / [RL + RP + (µ + 1) 2RK ]. be +400V, 0 and –150V. You just apply Ohm’s law to
find the resistance values.
Connecting a 12AT7 (which has a µ = 50 and an Although this circuit is not used much for
Rp = 10KΩ), in circuit with an RL = 50KΩ and an RK = audio amplifiers, we should examine how it can be
50KΩ gives a gain of about 0.5. So for push/push used, and, in particular, how we can get away from
[common-mode] input the gain from grid to plate is this negative-voltage line. A simple and apparently
less than unity. For push-pull [differential-mode] symmetrical form of circuit is shown in Fig. 6. Bias
input, applied between the grids instead of to grids in is provided by RB, which has its usual value for the
parallel, the gain is 30, so the balance [common- tubes (about 100–500Ω) according to the operating
mode–rejection] ratio is 0.5:30, 1 to 60 or –35.6dB. If conditions, and the grids return to the bottom of RB
a pentode is used for the cathode resistor this figure through RG1, RG2. The coupling resistor RK lifts the
can be increased still more. whole group of resistors up to perhaps +100V, so
This circuit is used in electroencephalography: that blocking capacitors C G1, C G2 are needed to
the two grids are connected to electrodes applied to connect the input and to provide the AC ground on
the head of a patient and the tiny brain-currents pro- grid 2. This circuit is not as symmetrical as it looks,
duce a push-pull [differential-mode] voltage between as you will see if you consider it as redrawn in Fig.
the electrodes. Stray 60Hz. fields produce a relatively 7. At high frequencies, the grid of V2 is grounded
large push-push [common-mode] voltage that must but at low frequencies, when CG2 is no longer a low
be eliminated,because it would mask the brain sig- impedance, the grid is returned to somewhere
nals even if it did not overload the final stages of the between ground and the top of R K. At zero fre-
recording amplifier. quency, the coupling resistance is down to R B ,
For use with oscilloscopes there are two possi- which is too small to provide any satisfactory sort
bilities. In the first, the two grids can be regarded as of balance. In practice this means that we must
the two input terminals, and we have the feature make 2π f CG2RG2 >>1 at the lowest frequency we
that push-push voltages are discriminated against, intend to use. We also need 2π f CG1RG1>1 if we are
while push-pull voltages are applied to the deflecting to get the signal into the circuit at all. At high fre-
plates of the oscilloscope tube. This is excellent if quencies the only sources of trouble are tube
you wish to work around zero voltage. But if you are capacitance, in particular the grid-cathode capaci-

♦ AN 23.2 - Page 20 ♦
tance which is effectively in parallel with RK, since B+

grid 2 is grounded. RL1 RL2

If you now look back at Fig. 1 you will, no doubt,
admire the elegant simplicity of the circuit: just CG1 Output
three resistors, a pair of tubes, and that little –Ve
sign. Every few years I come back to this point of Input
V1 V2
indecision. It’s a good simple circuit but where will I
get that negative supply? For special jobs, with dou-
RG1
ble input, it is possible to elaborate the long- RB
tailed–pair to give a really well-balanced system, RG2
though the balance is usually not as good as you can
get with a transformer. But the long-tailed–pair stays
balanced down to zero frequency. RK
I hope that in these four articles I have suc- CG2
ceeded in making it clear that it doesn’t cost any Power Supply &
Signal Common
more to make your push-pull circuit really balanced.
If you want to use a reasonable amount of negative Fig. 7. One method of eliminating the negative-voltage supply for
feedback, the shoddy circuits described in Part I will the cathode return. RB is tapped off the total cathode resistance (RB
add to your troubles and, as far as I can see they, + RK ) at the desired bias point.
amount to nothing more than a public avowal that
you “couldn’t care less.’’ If an amplifier is designed
on that basis, I would expect it to be pretty badly
constructed as well—I’d stay well away from it.

♦ AN 23.2 - Page 21 ♦