36 Wireless World, April 1976

'Transistor driver for valve amplifiers

Design for Williamson and other output stages

by Seth Berglund
Lunds University, Sweden

There are certainly a lot of valve audio driver. On the contrary, by starting with
amplifiers still in use, and many of them a bandwidth of lMHz, the high-fre
have an inherent quality of perfor quency response can be exactly formed
mance that makes it reasonable to give up to this frequency, using simple
them a thorough repair, with or without operational amplifier techniques, and so
an accompanying modernization. The it can easily be changed to suit diffennt
work needed for repair may tend to output transformers: The output vol
grow, however, since it may not be tage of the driver is sufficient even for
sufficient to replace valves and a few large output tubes such as the KT88;
electrolytic capacitors. A general they are assumed to work in class A or
degradation o f components may have i1
AB in t e design that follows.
taken place, and in nearly all instances
of modernization it should be advanta Output valve biasing
geous to replace the rectifying valve by
Fig. 1 . A constant voltage Vgo is in this
When direct coupling to the output
circuit added to a normal cathode bias. valves is used, the grids can still be kept
silicon or maybe selenium rectifiers. So
there may be some doubt as to what is at zero potential for the quiescent pOint,
really needed and what is worthwhile. with a normal cathode bias for class A
For those who are interested in giving or AB operation. But this is not neces
their valve amplifier a positive moder sary and in my opinion not at all the best
nization that will result in obvious Cg way. Let us therefore look at other ways
improvements, a description is here of biasing. For the sake of simplicity,
0-1 .......
---- -+
given of a transistor driving amplifier single valve biasing is discussed first,
that can replace the voltage amplifying Rg and the valves shown as triodes with
stages of many existing power ampli the usual assumption of zero grid
fiers. The Williamson amplifierl has current, i.e. anode and cathode currents
been chosen as a typical example for the are identical. If thus la is the anode
discussion that follows, because it is a current of a triode and Rh the outer
well-known design. Other amplifier cathode resistance, the negative grid
designs that have been used for compa voltage with a normal cathode bias is
rison are those designs by MullarQ.2 and Vgk = laRk '
by GEC3 Fig. 2. Further development of the It is possible, although not often used
'the original idea was to design an circuit in Fig. 1 by means of a gain . in practise, to modify the influence of the
amplifier with a bandwidth sufficiently anode current on this bias voltage by
function A.
in excess of the output transformer the addition of a constant voltage to the
bandwidth, so that the only phase shift circuit, either in series with the cathode
to take account of should be that of the allowable without instability. T his was or, normally with less effort, in the grid
transformer. A d.c. amplifier with a found to be the case for several output circuit, shown as the voltage V.o in Fig.
bandwidth of about IMHz was thought transformers, at least with a resistive 1. The grid bias voltage is now Vgh
to be sufficient. Direct coupling from load. With the Williamson transformer = Vgo - laRk '
the input stage to the signal grids of the and output stage according to the It is important to note that Vgo may be
output valves leaves the output trans original design, an essentially flat positive as indicated in the figure, or
former as the only cause of phase shift amplitude-frequency response was negative. In the first case a larger
at the low frequency end, and the shift obtained up t o IMHz, and there was resistance Rk is required than for simple
tends to only 90. So there are no some stabilitv margain. cathode bias, which makes the gria
problems of instability from negative If a loudspeaker or some load of a voltage more dependent on the anode
feedback' at the low frequency end, complex nature is inserted, the phase current, i.e. there is a better stabiliza
provided that the usual precautions as shift tends to become too large, and the tion of the quiescent point. In the case
to supply line filtering are taken. only way to stability is then to reduce of negative polarity for VgO' the grid
At the high frequency end of the the closed-loop gain. So extended voltage becomes less dependent on the
transformer passband there is usually bandwidth is no radical solution for anode current, as Rk must be dimin
one main resonant frequency, often at stability at the high frequency end in ished. For growing negative values of
about 100kHz, around which the phase the same way as direct coupling is for Vgo' it becomes in the limit equal to the
shift passes 90 by a considerable low frequencies. And after all, the aim deSired grid voltage. Then Rh must go to
amount but does not reach 180. It was should not be amplification up to radio zero and the result is a constant grid
thought therefore that with a band frequencies, but an l.f. amplifier with a Voltage.
width of at least IMHz for the driver, defined upper frequency limit. T his does A grid bias that has.exactly the same
the normal amount of 20dB frequency not mean that it is a wasted effort to dependence on the combination of a
independent feedb a c k should be start with a large bandwidth for the constant voltage and the anode current
Wireless World, April 1976 37

as that of Fig. I, but with improved Fig. 3. Circuit principles for a complete
means for the choice, can be obtained push-pull amplifier, but with output
by a circuit as shown in Fig. 2. W ith the transformer omitted. The inputs are for
notations according to this figure, and signal input and feedback.
provided that the operational amplifier
of voltage gain A has zero offset, the,
constant part of the grid bias is VgO
A VI' and the total grid bias becomes

Vgk =
A VI - (1 + A)IaR k

In this circuit Rk can be a small

resistance, which is an advantage for
large output tubes where the power
dissipated in Rk for a normal bias may
be considerable. Most important is,
however, the ease of adjustment to a
desired bias.
The bypass capacitor Ck has retained
its function, and the time constant RkCk
is chosen as for normal cathode bias. oV
However, if Rk is small, so that it causes
only negligable feedback by itself, the
bias time-constant may be introduced
by a separate RC-link, either before or
after the amplifier.
In the foregoing figures the bias and
signal voltages have been mixed in the
usual way by a grid resistor R and a
coupling capacitor C . If a ful signal
g
f
feedback from Rk is wanted, corre
sponding to an unbypassed cathode
resistance in Fig. I, some other type of
mixing circuit is needed. This also holds,
if direct coupling of the signal to the
valve grid is used.

Arrangement of amplifier Resistance R2 is selectable for choice of differential stages reduce hum. Only the
To explain the main features of the voltage gain. negative high tension voltage needs a
complete push-pull amplifier, its layout As the amplified part of the bias is a certain stabilization.
is first shown by the simplified circuit of common-mode one, it corresponds to a The input common-emitter l o n g
Fig. 3. The circuit comprises three common cathode resistance with the tailed pair o f Fig. 3 i s a dual n-p-n
differential stages, namely a signal value RkAl2, and the time constant of transistor Trl' and it is completed by an
input stage, a biasing stage for the the RC-link is RCI2. The constant part n-channel d ual f.e.t. Tr2' the two
sensing of the currents in the output of the grid bias is simply an offset transistors o f which are used a s input
valves, and between these a mixing and voltage, effected by an adjustment of source followers. This makes the ampli
amplifying stage that drives the valves. the potentiometer R2, which is therefore fier compatible with valve amplifiers
It is a symmetrical circuit throughout found to have the double function of with regard to input impedance as well
for the input signals, and the necessary determining the gain by its resistance as to independency of the characteris
d.c. balance is obtained at the emitter value and the constant voltage by its tics of the driving source. So all that is
side of the input stage, in the figure by adjustment. required of the preamplifier is that it
means of the potentiometer RJ Another The resistances of Rk may be so small shall give sufficient voltage.
important feature is that the differential that their direct influence on the valve The d.c. balancing potentiometer R4J,
stages are all supplied by a current bias becomes negligible. They cause a a I5-turn trimming potentiometer, has
source at the emitter side, instead of small lowering of the effective valve been moved away from the main signal
just by a common emItter resistor. A transconductances. path into the f.e.t. source circuit, where
high common-mode rejection ratio is Because the collector resistances of it gives a smooth adjustment of the
thereby obtained, which means that the the mixing stage become fairly large, differential balance. By this change the
input signal and the negative feedback there is ample signal amplification two resistors R7 and R8 also become
around the amplifier can be fed differ available in this stage for local feedback more freely selectable for their function
entially to the input stage without to be applied. This is used in the 'to determine the local feedback of the
danger of adverse secondary effects. amplifier for determining the response stage and the gain of the amplifier. They
The current source for the mixing, by means of the impedances Z. should be matched, so as not to cause
stage, a single transistor in Fig.3, acts additional asymmetry to be balanced
with the differential pair as a com Amplifier design out. It is the combination of f.e.t. and
mon-mode amplifier for the Signals The complete amplifier is shown in the bipolar transi?tor pairs that gives the
from the preceding biasing stage, so circuit diagram of Fig. 4. Although the good input property, together with an
that the two stages together give a number of components has grown, the easily variable amplification and a large
common-mode voltage gain from cath fundamental simpliCity as evidenced by bandwidth. Dual transistors must be
odes to grids that corresponds to the Fig.3 is retained, and there are not any used to reduce temperature drift, see
gain function denoted by A in Fig. 2. The hidden difficulties such as the need for later.
gain to a sufficiently good approxima tricky adjustments or special demands The mixlI1g stage has been developed
tion is on the power supply voltages, wh::h to a cascode configuration, which is
may vary within large limits, The very important with regard to harmonic
demands on filtering are not very large distortion because the output voltage
either, since the current sources for the swing is large. It isalso important that
38 Wireless World, April 1976

the Miller feedback capacitance is kept drift is that the current SQurce fQr the class A with a high load impedance is the
very lQW sO' that the lQading Qn the mixing stage, Tr9 and T rio, is a Qnly case where the capacitance may be
preceding stage can be cQntrQlled as cQmplementary pair amplifier. Qmitted to' some advantage.
desired, and the amplifier as a whQle be The gain as defined by Fig.2 is nearly
given sufficient bandwidth. The main 70, which means that the bias circuit Response and distortion
lQcal feedback is by means Qf the cQrresPQnds to' a cQmmQn-cathQde The amplitude-frequency resPQnse Qf
emitter resistQrs RI3 and R14, but they resistQr Qf 350 Qhms. A cQmmQn-mQde the cQmplete amplifier is shQwn in Fig.
need nQt be matched as their CQunter CQnstant grid vQltage Qf abQut + 5V is 5: withQut feedback by curve B, and
partsR7 and Rs, as the balancing actiQn added by adjustment Qf R42. The quies with 20dB Qverall negative feedback by
Qf I is amplified by the input stage. cent grid-cathQde vQltage is abQut -45V curve C. The lQw-frequency resPQnse
LQcal feedback by the twO' impe and the valves wQrk in class A. fQr small signals is flat dQwn to' 10Hz
dances Z starts at a value Qf abQut I2dB A negative feedback that senses the bQth with and withQut feedback. Exact
fQr lQW frequencies, but increases differential direct vQltages acrQSS the curves shQwing the fall belQw 10Hz are
within the frequency range 20 to' cathQde resistQrs has alsO' been added to' nQt interesting, but it is PQssible to'
200kHz to' abQut 26 dB. It fQrms the the circuit. It cQnsists Qf th matched select a value fQr C2 that gives an
amplitude resPQnse as shQwn in Fig.5, resistQr pairs R31, R32 and R5, R6 tQgether optimum resPQnse to' square waves at
curve A. The impedances Z dO' nQt cause with the capacitQrs C3 , C4. This feed lQW frequencies.
any cQmmQn-mQde feedback but act back is cQupled to' the amplifier inputs There is a dip in transfQrmer resPQnse
tQgether fQr the differential feedback, SO' and has an upper frequency limit Qf at abQut 50kHz, which cannQt be
they dO' nQt need matching fQr their abQut 1Hz. It has the same stabilizing eliminated by simple feedback circuits.
actiQn. HQwever, matching is needed effect Qn the balance between the tube It causes SQme ringing in square-wave
fQr the cQllectQr IQads Qf transistQrs Tr5 currents as twO' separate cathQde resis tests, which Qf CQurse has nQthing to' dO'
and Tr6 fQr symmetry in driving the tQrs Qf 200 Qhms, cQnnected tQgether in with instability. The capacitance of C6 in
Qutput valves. The twO' cQllectQr resis a lQng-tailed pair cQnfiguratiQn but the feedback lQQP has, hQwever, been
tQrs RI5 and RI6 shQuld be matched, and withQut influence Qn the CQmmon chQsen SO' large that it has a damping
alsO' the impedances as they alsO' lQad mQde bias. influence Qn the ringing. The series
the cQllectQrs. All the abQve values are easily resistance Qf R36 has been chQsen as a
As to' the valve CQmmQn-mQde bias changed fQr desired bias cQnditiQns, but cQmprQmise to' give abQut the same
ing, there are Qnly twO' alteratiQns frQm a general discussiQn Qf valve biasing is frequency resPQnse when lQaded by a
the simplified circuit Qf Fig.3. One is that Qutside the sCQpe Qf thi article. _0 __
certain brQadband lQadspeaker as with
the PQtentiQmeter fQr adjustment Qf the A capacitance Qf 22 f1F was Qriginally a resistive lQad. A capacitance inserted
constant vQltage part has been split up used fQr C5, but is Qmitted in the circuit as C6 in the feedback lQQP withQut a
in twO' fixed resistQrs, RI9 and R20' and a Qf Fig. 4. HQwever, Qutput triQdes in series resistance Qften gives a gOQd
I5-turn trimming PQtentiQmeter, R42 frequency resPQnse with a resistive
This makes the selectiQn Qf resistances load, but QscillatiQns when a lQad
fQr a desired value Qf the amplificatiQn speaker is cQnnected. Its influence Qn the
fairly easy, and prQvides fQr a smQQth Fig. 4. CQmplete circuit diagram of the feedback must therefQre always be
adjustment Qf the CQnstant vQltage. The power amplifier. Valves work in class A carefully checked.
Qther change, mainly fQr temperature as C5 is made zero (see text). The branch R22 and C7 between the

+ 2 5V -200V

R 17 R 15
Rll R9 22k
22k
4k7
6k8 out

Tr5
-2 5V
M P SA92

R23
lk
Trl + 2 5V
Cl BCY89
R18 R3 5
lOOn
S 12k 4k7
in o---t
,
RI R3 IN414 8
R32 C6
470k 18k R7 R13 220p
4M7
150
220 Tr8 R36
R5
BC107 2k415IL
470k R 41 load
0--+ 5k
R6 Tr7
470k R8 BC107
R4 2 20 R 14
R2
R31 18k 150 R
470k 42
4 M7 2k

R24
Tr4 lk
BCl77

RIO Tr6
4 k7
M P SA92

+25V

R I6
22k
Wireless World, April 1976 39

output terminals has been found valua

ble with several output transformers, 20
and is therefore recommended. It has no
effect on the response within the
' co
audible band, but represents a resistive 10
load at high frequencies. Values are not z
:;{
critical. . (!)
It has been an aim to choose about w 0

the same high frequency limit for the

response without feedback as in the -'
w
improved version design by Williamson er
-10
to make a comparison of the final result
fairly easy. It could be an advantage,
however, to choose a lower high -20
frequency limit by a change of the
impedances Z. 5 10 50 100 500 1k 5k 10k 50k lOOk 500k 1p00k
Total harmonic distortion of the FREQUENCY (Hz)
driver is quite low. For 30V r.m.s. output
on each side it is only about 0.05% at low
frequencies and rises to about 0.1% at
.20kHz. This leads to a low distortion for Fig. 5. Amplitude frequency response The main'cause of differential drift is
the whole amplifier even without curves for the driver (A) and for the i
the input dual f.e.t. Although ts thermal
overaH feedback: at 1kHz this distortion complete amplifier without (B) and drift of gate-source voltage difference
is only 0.08% for IOW and 0.2 % for 15W with feedback (C). for specified working conditions is less
output power. than 40.uV/deg C, its drift in the circuit
The overall feedback works fully may be larger, on account of shifts of
within the audible band, but the maxi quiescent points. There is also up to
mum output power. falls at the low and mentioned d.c. input voltage, being IO.uV/deg C drift in the dual bipolar
high frequency ends. At a distortion of about 700mV. However, the two tran transistor, and some additional drift
less than a quarter of a percent the sistors must be mounted close together, from the transistor pair Tr3' Tr4 As a
available output power with resistive so that they experience the same summation a temperature drift of up to
load is ?,OW at middle frequencies and ambient temperature change. Prefera 100 tJ.V/deg C referred to the input of.
15W at 20Hz and 15kHz. bly they should be plastics transistors the amplifier will be assumed.
The total harmonic distortion, mea and clamped together, but a dual To find what the above drift means as
sured at 20Hz, 1kHz, and 15kHz and transistor is not necessary. a drift in quiescent current for the
with an output power of 10 and 15W is There is also a temperature drift from valves, the d.c. feedback from the
summarized in the table below. The differences in the internal heating of the cathodes to the input circuit will first be
figures are given in percentage distor transistors, for instance at power supply assumed inoperative. The differential
tion, but include what there may be of variations. This is kept low by means of voltage amplification to the grids is 450
hum and 'noise in the prototype ampli low collector currents. For the same and the transconductance is IOmA/V,
fier. reason the design assures a small which gives 0.45mA/deg C differential
collector current for Tr9 in the current drift for the anode currents, or 9mA
Power Total harmonic distortion (%) source, and the transistor drift is partly for a change in temperature of 20deg
output balanced out by D3. The balance is not C. This is at the limit of what should be
as good as for a couple of equal allowable, but, on the other hand, fairly
(W) 20Hz 1kHz 15kHz
transistors, but here the drift is inside wide limits as to the causes are
10. 0.05 0.01 0.1 the feedback loop and has less influence assumed.
15 0.1 0.02 0.25 on the valve currents, about one third of The picture of drift changes radically,
that of the preceding transistor pair. however, if the d.c. differential feedback

Circuit working conditions

In all d.c. amplifiers there is a tempera
ture drift that must be taken account of.
In this case there are really two, namely
a common-mode drift in the biasing
circuit and a differential drift for the
signal path. Drift in the output valves is
not considered.
An obvious cause of common-mode
bias drift is the difference in change of
base-emitter voltage with temperature
for the transistor pair Tr7 and Trs. The
two transistors should be of the same
current amplification class, BC 107A in
the prototypes, in which case the
difference may be assumed to' be
O.lmV /deg C at the most. The drift
voltage is equal in its effect to a false
reading of the direct input voltage on
the base of Tr8, and results in a
corresponding shift of the anode cur
rents of the valves.
If an ambient temperature change as
large as 20deg C is assumed, the
false reading is not more than 2m V,
which is less than 0.3% of the above-
40 Wireless World, April 1976

o +25V
o

SF

o
-25V o
is inserted. The feedback is 14dB from Fig. 6. Components in feedback tension voltage for the cascode stage
d.c. to about 1Hz, and the above anode circuits, Rs> C3, Rs. C4 and C2 are not i;lnd the maximum grid peak-to-peak
drift becomes less than 2mA for a included on board; neither are R15 and voltage is about two. When smaller
2Odeg C temperature change. The R16. Mono printed boards are available output tubes are used, such as EL 34,
feedback also reduces d.c. drift from for 2 inclusive from M. R. Sagin, 11 . EL506 or EL84, the negative voltage
other causes, such as changes of Villiers Road, London NW9. should be lowered, but the above ratio
component values with time. Its equi not made smaller - a value between
valence to a pair of separate cathode two and three is preferred. The collector
resistances has already been shown. currents for the cascode stage should be
difference is acceptable, although a
The above feedback may, on the maintained, and the collector resistors
closer tolerance may be required for the
whole, be regarded as a possibility chosen accordingly.
resistor pairs in the d.c. feedback, or the
rather than a necessity, and 14dB is value of RI may prove not to be
certainly more than necessary. The time sufficient
constant in the feedback circuit is so In a first construction, the d.c. References
large that temporary deviations from feedback should be omitted, and put I. Williamson, D.T.N. High quality ampli
symmetry in the signal (musical) vol into effect only as a finishing touch. fier: new version, Wireless World, vo!. 55,

tage should not cause appreciable d.c. 1949, p. 282.

For the positive and negative supply
Williamson, D.T.N. The WiIliamson Am
shifts. voltages of 25V in Fig. 4 the recom
plifier, A Wireless World publication, Iliffe &
Stabilization is needed for the nega mended. values are 25 to 30V, but there Sons, Ltd.
tive high tension voltage, because a is no need for symmetry. The value of 2. Ferguson, W.A. Design for a 20-watt
10% variation of this voltage would collector currents for the cascode stage high quality amplifier, Wireless World, vo!.
cause too large variations in the valve is 6 to 7mA. The currents of the other 61, 1955, p.223.
bias. A simple stabilization, for instance stages are evident from the values of the High Quality Sound Reproduction, Mul
by means of a series resistance from resistors R38, R39 and R4o, since the lard Ltd.
a -300V supply feeding a chain of six voltage across these is about 4.3V. 3. How to Build the Osram 912-PLUS,
O.4W, 33V zener diodes is sufficient. The General Electric Co, Ltd.
Any transistor in the Phili p s
Heath, W. ran and Woodville, G. R.
voltage is of course not critical. BFQIO-16 family may b e used a t the
"Design for a 50-watt amplifier," Wireless
input, and there are of course also other
World, vo!. 63, 1957, p.158.
replacement types, instance the
for
Constructional details Siliconix E401. There are also a number
The layout of the circuit on a printed of replacements for the M otorola
circuit board or otherwise is not critical. MPS-A92, for instance M PS-U60, BFTl9
It has already been mentioned that the (RCA), and BFW43 and BFW44 (SGS
two transistors of the pairs Tr3' Tr4 and Ates). There are numerous replace
Tr7' Tr8 should be mounted for close ments for BC107 and BC I77, and also for
thermal connexion, and so should Trg be the dual transistor BCY89, which is the
with D3. To avoid heating effects from least expensive of the BCY87-89 family.
the collector resistors RI5 'and RI6,
mount them with the valves, and not on Concluding remarks
a p.c. board. The circuit should be One reason for the choice of KT88
mounted away from the mains trans valves connected as triodes was that
Logic design course
former and filtering choke to avoid they put high demands on the driver, Digital System Desi gn is the name of a course
induced hum from stray magnetic and so are suitable for presentation of to be held at Chelsea College, Pulton Place,
fields. It should also be kept away from driver qualities. The same valves con London SW6, from May 17 to 21. This course

any hot air stream or heat radiation nected as pentodes or with a distributed is designed to give practising engineers and

from the valves. These precautions do load are more easily driven because the scientists a formal approach to the logical
;
design of digital .s stems and should prove
not cause any problem, as the circuit Miller capactiance is lower. An obvious
useful to those engineers and scientists
may be given fairly small dimensions. conclusion is therefore that the driver
working in the field of digital electronics who
Simple metal shields have been used in should suit most power amp lifiers
have had no previous training in methods of
the prototype amplifiers. except for very large ones that require logic design. Enquiries should be addressed
Five-percent resistors have been several output valves in parallel. to Professor J. E. Houldin at the above
use, and for matched pairs a 2% The ratio between the negative high address.