36 World, April 1976 ' driver for valve

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 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 of components may have AB in ti1e design that follows. taken place, and in nearly all instances of modernization it should be advanta­ Output valve biasing Fig. constant voltage is in this geous to replace the rectifying valve by 1. A Vgo When direct coupling to the output circuit added to a normal bias. silicon or maybe selenium . So valves is used, the grids can still be kept 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 andin 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 and the valves shown as with Rg 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 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 It is possible, although not often used GEC3• Fig. 2. Further development of the 'the original idea was to design an . in practise, to modify the influence of the circuit in Fig. 1 by means of a 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. This 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 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 to 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 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. This 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 feedback should be start with a large bandwidth for the constant voltage and the anode current WirelessWorld, 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. With the transformer omitted. The inputs are for notations according to this figure, and signal input and feedback. provided that the 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)IaRk

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 theamplifier. 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 fulf signal g 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 long­ Fig. 3. The circuit comprises three common cathode resistance with the tailed pair of Fig. 3 is 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 dual f.e.t. Tr2' the two sensing of the currents in the output of the grid bias is simply an offset of which are used as 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 , 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 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 . 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 because the output voltage either, since the current sources for the swing is large. It is·also 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 Trio, 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 parts R7 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 t curve A. The impedances Z dO' nQt cause wi h 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 R17 R15 R R9 22k 22k ll 4k7 6k8 out

Tr5 MPSA92 -2 5V R23 lk Trl + 2 5V Cl BCY89 lOOn R18 R35 4k7 in o---t S 12k , RI IN414 8 R32 R3 C 470k 6 4M7 18k R 7 R13 220p 150 220 Tr8 R3 R 6 5 BC107 2k415IL 470k load R41 0--+ 5k R6 Tr7 470k R8 BC107 R 220 R 4 R14 R 2 18k 150 R 31 470k 42 4M7 2k

R24 lk Tr4 BCl77

RIO T k7 r6 4 MPSA92

+25V

RI6 22k WirelessWorld, 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. . (!) w It has been an aim to choose about 0 � the same high frequency limit for the �-' response without feedback as in the w er improved version design by Williamson -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 the input dual f.e.t. Although its 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 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 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 (W) 20Hz 1kHz 15kHz allowable, but, on the other hand, fairly 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 EL34,

feedback also reduces d.c. drift from for £2inclusive 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, 1949, p. 282. tage should not cause appreciable d.c. 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 is sufficient. The Any transistor in the Philips General Electric Co, Ltd. Heath, W. ran and Woodville, G. R. voltage is of course not critical. BFQIO-16 family may be used at the "Design for a 50-watt amplifier," Wireless input, and there are of course also other World, vo!. 63, 1957, p.158. replacement types, for instance the Constructional details Siliconix E401. There are also a number The layout of the circuit on a printed of replacements for the Motorola circuit board or otherwise is not critical. MPS-A92, for instance MPS-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 BCI77, 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 Logic design course mounted away from the mains trans­ valves connected as triodes was that former and filtering choke to avoid they put high demands on the driver, Digital System Design 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 load are more easily driven because the scientists a formal approach to the logical from the valves. These precautions do ; 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 should suit most power amplifiers Simple metal shields have been used in 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.