Home | Audio Magazine | Stereo Review magazine | Good Sound | Troubleshooting

There was only a limited range of circuit options open to the audio amplifier designer before the widespread availability and adoption of solid state semiconductor components, and for this reason the variety of commercially successful designs evolved for tube operated audio power amplifiers was also fairly limited. In this Section I have mainly concerned myself with those circuits produced by UK manufacturers, or those designs offered both by the tube manufacturers' laboratories and by independent designers for home construction.

By the early 1950s the expected standard of performance of a high quality audio amplifier was that it should have an output power of at least 10 watts, that it should have a full power bandwidth of at least 20Hz-20kHz, and that its harmonic distortion, at 1 kHz and full output power, would be better than 0.2%. In order to achieve this sort of performance, it was necessary, in practice, that overall (output to input) NFB should be used. It was not normally feasible to do this with the simple transformer coupled layout shown in FIG. 4 because the inevitable phase shifts, introduced within the feedback loop by the two coupling transformers, would make the amplifier oscillate continuously, except where the designer had chosen to use very small- and probably unhelpful- feedback levels.

In the case of the designs using only one transformer (that for coupling the output tubes to the loudspeaker) it was possible to design the circuit to allow useful amounts of NFB to be employed-- a value for 13A ~ of 20-26dB was a typical figure--provided that the output transformer was well designed, and this meant that, for the designer, the extent of the choices available to him were in the type of input amplifier stage (triode, pentode, cascode or long-tailed pair), the type of phase splitter (split load, floating paraphase, long-tailed pair or simple inverter), the type of output tubes,(pentode or beam-tetrode, straight or triode connected), whether an additional amplifier stage (usually a push-pull pair of triodes) between the phase splitter and the output stage was judged to be necessary and, finally, in the case of pentode or beam- tetrode output tubes, whether the output transformer had primary tapping points to which the screened grids could be connected- the so-called ultra-linear or distributed load type of connection.

In spite of the seemingly large number of different design layouts these choices allowed, in practice, when one looked at the designs offered at the time, they did not appear to differ greatly from one another, and I will examine the more notable examples of these later in this Section. Three designs, by McIntosh of the USA, Quad of the UK, and Sonab of Sweden, did, however, strike off on directions of their own, and are worthy of particular consideration.

The McIntosh Amplifier

In the years preceding and following the 1939-1945 war, the McIntosh company of the USA became notable, among the radio and radio-gramophone equipment which filled the British radio shops, for producing massive, immaculately engineered equipment offering a superb performance at a high price, for those who wanted and could afford the best. (The company is still in business, and its philosophy appears to be unchanged). At a time when it was thought that amplifier output powers in the range 7-15 watts, and THD figures in the range 0.5-1%, would be entirely adequate for domestic audio purposes, McIntosh offered the circuit shown in FIG. 1, which had an output power of 50 watts, and a THD figure of 0.2% from 50Hz to 10kHz (McIntosh, F.H., and Gow, G.J., Audio Engineering, December 1949).

FIG. 1 The McIntosh system

There were three basic design problems to be solved in this application. The first of these was that obtaining such high output power levels, even with fixed bias, entails driving the output tubes in class AB 1. This is a condition where the level of negative grid bias is chosen so that one or other of the output tubes will be driven into cut-off at some part of the operating cycle. In normal circumstances this will lead to cross-over distortion, due, in part, to the imperfect coupling between the two halves of the output transformer primary, and to shitts in the input bias level due to the high output impedance of the driver stage.

McIntosh solved both of these problems by using a trifilar wound output transformer, in which there was very tight inductive coupling between the windings, and the cathodes as well as the anodes of the output tubes were coupled to the load. The final problem, that of achieving satisfactory loop stability with two coupling transformers within the loop, was solved in part by taking great care in the design of the coupling transformers to achieve a high winding inductance at the same time as preserving low levels of leakage reactance - the driver transformer is also trifilar wound (i.e. the windings are made by laying the three wires of the separate windings alongside each other and at the same time). The other method used to preserve NFB loop stability was to derive the NFB lines from the cathode of each output tube to that of each input tube rather than from the output to the loudspeaker. The latter arrangement is preferable on technical grounds.

An additional contribution to the close coupling of the two halves of the output stage is provided by connecting the output tube screening grids (G2) to the anodes of the tubes on the opposite side of the transformer.

The Williamson Amplifier

The period which saw the introduction of the McIntosh amplifier was one of very great activity on the part of audio amplifier designers, and a number of novel circuit designs were published, of which some were offered specifically for the use of the home constructor. Among these, by far the most celebrated, was the Williamson amplifier, of which the circuit design is shown in FIG. 2. This was designed by an enthusiastic audiophile who was working for the GEC receiving tube development laboratories at Hammersmith, London. His circuit, first described in a GEC internal memo in 1944, was subsequently published in April-May 1947 (Williamson, D.T.N., Wireless World, April 1947, pp. 118-120; idem. May 1947, pp. 161-163). The springboard for Williamson's design was the prospect of the renewed availability of the GEC KT66 output beam-tetrode for general use. (This tube design had been introduced in 1937, but, in the UK, its use in domestic applications during the 1939-1945 period of hostilities had been restricted due to the overriding priority given to military requirements.) The KT66 beam-tetrode tube was particularly suited to audio output use, in that its output waveform distortion, unlike that of an output pentode in which the distortion was mainly third harmonic, was predominantly second harmonic and, if the output tubes were matched, this type of distortion could be substantially reduced by the use of the output tubes in push-pull. Moreover, by connecting its G2 to the anode, the KT66 could be made to behave in a manner very closely analogous to that of an output power triode, without any of the problems associated with high power triode construction.

FIG. 2 The Williamson amplifier (voltages in brackets are AC signal levels)

In his introduction to the Wireless World articles, Williamson examined the various circuit options available to the audio amplifier designer, and concluded that for the requirements of low harmonic distortion, negligible phase shift within the chosen frequency band (10Hz-20kHz), good transient response and low output resistance to be met, it was essential that the loudspeaker output transformer should be included within the negative feedback loop. This, in turn, required that the transformer primary incremental inductance should be greater than 100H, and the primary:primary and primary:secondary leakage reactances should be less than 33mH. These figures were calculated from the premise that in order to ensure loop stability the phase shift introduced by this component should not exceed 90 degr. at 3.3Hz and 60kHz: a requirement which was easier to meet if the anode impedance of the output tubes was not too high, so that the required primary:secondary turns ratio for the output transformer would also not be too large. For a triode connected KT66, Ra would be 2500-ohm and the optimum total primary to secondary turns ratio for a 15 ohm LS load would be 25:1.

A triode tube, used as a low power voltage amplifier, has a very low level of non- linearity distortion, especially if the output signal voltage is small in relation to the total possible output voltage swing. In Williamson's circuit the output voltage swing at the anode of the input tube, V1, was only 2.7V peak, which is very small in relation to the possible 100V swing at this point.

In order to minimize the LF phase shift within the amplifier, the anode of V 1 was directly connected to the split-load phase splitter, V2, and since the DC voltage on V 1 anode was about + 100V, the voltage on V2 cathode would be about 105V and on its anode would be about +200V. Since V2 is effectively a cathode follower, it will have a signal gain very close to unity and negligible waveform distortion. Moreover, since the same current flows through both R7 and R8, it must follow that if the load impedances are identical, the signal voltages across both of these load resistors would be identical.

Because of the relatively low gain from the input of V1 to the output of the phase splitter, it was necessary to have a further gain stage between this and the output tubes, and here, again, a push-pull pair of triodes was used. Since the cancellation of second harmonic distortion requires that the output tubes are matched - both in characteristics and in operating conditions- Williamson included preset controls (RV1 and RV2 in FIG. 2) to allow matching of both grid bias and signal drive levels.

Since the final circuit structure now gave an adequate gain without the use of (electrolytic) cathode bypass capacitors, these were omitted and this provided a small amount of local negative feedback in each gain stage. This helped to improve the gain/bandwidth product and the gain and phase linearity of these stages, and also eliminated a component whose performance was sometimes less than desirable, as well as being prone to long-term deterioration.

Provided that the specification of the output transformer, as given by Williamson, was met, the performance of this amplifier was excellent, and was far ahead of any of its competitors. The articles describing its circuit were reprinted on a world-wide scale, and its performance served as a model to which other designers could aspire. Since it was not easy to improve on this design in terms of performance, most of the competitive designs sought to offer a higher output power than the 15 watts which the Williamson gave, though, in the UK, many of the commercial designers judged that 12 watts would be adequate for all normal domestic purposes.

Distributed Load Systems

The dilemma which faced audio amplifier designers in their choice of output tubes was that of balancing output stage efficiency (and available output power) against harmonic distortion. It was a matter of general observation among the audio enthusiasts of this period that the type of sound quality given by a triode output stage, which mainly generated second harmonic distortion, was preferable to the rather hard or shrill reproduction which was typical of the output pentode, which mainly produced third and fifth harmonics. Output beam-tetrodes, of the KT type, which tended more to a triode distortion characteristic, sounded better than pentodes, but triodes were still the enthusiasts' preference. (With modem beam-tetrodes or pentodes it was possible to get something fairly closely approaching triode characteristics by connecting G2 to the anode.) The problem was in output efficiency, which was only 27% for a push-pull pair of triodes, as compared with nearly 40% for a pair of pentodes using the same HT voltage and anode current.

Understandably, circuit designers began to explore ways in which an output beam- tetrode (or pentode) tube could be made to give a higher output power while still retaining the low distortion characteristic of the triode stage, and it was shown by Hailer and Keroes (Audio Engineering, pp. 15-17, November 1951) that this could be achieved by connecting G2 to a tap on the output transformer primary winding, as shown in FIG. 3. The effect of this method of transformer connection, in terms of relative output power, output impedance and harmonic distortion, is shown graphically in FIG. 4. Hailer and Keroes termed this the ultralinear (U-L) connection, and this raised the hackles of the linguistic purists, one of whom -- I forget who -- likened it to the thirteenth chime of a crazy clock, which served only to cast doubt on all that had gone before. Nevertheless the name stuck.

FIG. 3 The ultralinear connection

FIG. 4 Distributed load effects

The practical effect of the U-L connection was to offer the designer a graded choice between the virtues and drawbacks of triode or pentode outputs. In terms of distortion and output (anode) impedance- low values of which made the design of high quality output transformers easier- there was a continual improvement as one moved from pentode to 100% triode, but for maximum output power for a given level of distortion, the 25% tapping was the best. The final point of concern to the designer was that the stage gain in the pentode connected mode was substantially higher than the triode, and this would facilitate the design of the preceding stages of the amplifier. The best compromise value for this was the 40% tap, and this was the ratio favored for the high quality end of the market.

Much attention was drawn to this technique by the publication by Hailer and Keroes of a circuit which offered ultralinear operation as a means of improving the performance of the Williamson amplifier (Audio Engineering, pp. 26-27, June 1952). This prompted an indignant response from Williamson and Walker (the latter being the head of the Acoustical Manufacturing Company, the design of whose Quad audio amplifier used a similar, but allegedly superior, philosophy and had been on sale since 1945). Their response (Amplifiers and Superlatives, Williamson, D.T.N., and Walker, P.J., Audio Engineering, pp. 75-81, April 1954) gives one of the more comprehensive analyses of this topic, and compares U-L operation with cathode coupled systems of the kind used by both Quad and McIntosh.

The Quad System

This method of operation of the output tube is shown schematically in FIG. 5. This arrangement offered a number of practical advantages, among which was the fact that the local negative feedback introduced into the cathode circuit, by a winding which was about 10% of that in the anode, reduced the output stage distortion from 2-3%, a typical value for a triode connected output tube, to some 0.7%- a figure which would be reduced still further by overall NFB. It also offered an output stage efficiency which was of the same order as that of an output pentode. Moreover, the local NFB reduced the phase shift due to the output transformer and made the specification for this component less stringent. The full circuit of the Quad 12 watt power amplifier is shown in FIG. 6. Since this design was offered commercially in 1945, it has the distinction of being one of the first true Hi-Fi power amplifiers, with the NFB loop including the output transformer as well as the whole signal amplifying chain.

FIG. 5 The Quad circuit

There are a number of interesting (and surprising) features in the Quad II design, which I have redrawn in FIG. 6 so that these features are easier to see.

Superficially the circuit uses a single pentode amplifying stage followed by a simple phase inverter of the kind shown in FIG. 5, in which the ratio of R7 to R8 (R4:RVI in FIG. 5) is chosen to give the correct output drive voltage at V2 anode. However, there are complications in that the NFB signal, from a tap on the output transformer secondary, is applied to both of the input tube cathode circuits, and while this will appear as negative feedback at V1, it will be positive feedback when applied to V2.

PFB has the effect of increasing both the stage gain and the distortion of the stage to which it is applied, and its use, within the NFB loop, to increase the loop gain and thereby to reduce the overall distortion level has been employed many times.

However, in this instance, the increase in the gain of V2 makes it necessary to reduce the input signal to V2 grid to a much lower level than would normally have been used.

The actual values of R7 and R8 would imply a stage gain of 253 for V2, which is rather more than twice that which would have been expected.

Another unusual factor is the cross-coupling, by C1, of the G2 grids of V 1 and V2, instead of the more normal separate decoupling of both grids to the ground (0V) or 2-3%, a typical value for a triode connected output tube, to some 0.7%--a figure which would be reduced still further by overall NFB. It also offered an output stage efficiency which was of the same order as that of an output pentode. Moreover, the local NFB reduced the phase shift due to the output transformer and made the specification for this component less stringent. The full circuit of the Quad 12 watt power amplifier is shown in FIG. 6. Since this design was offered commercially in 1945, it has the distinction of being one of the first true Hi-Fi power amplifiers, with the NFB loop including the output transformer as well as the whole signal amplifying chain.

There are a number of interesting (and surprising) features in the Quad II design, which I have redrawn in FIG. 6 so that these features are easier to see.

Superficially the circuit uses a single pentode amplifying stage followed by a simple phase inverter of the kind shown in FIG. 5, in which the ratio of R7 to R8 (R4:RVI in FIG. 5) is chosen to give the correct output drive voltage at V2 anode. However, there are complications in that the NFB signal, from a tap on the output transformer secondary, is applied to both of the input tube cathode circuits, and while this will appear as negative feedback at V1, it will be positive feedback when applied to V2.

PFB has the effect of increasing both the stage gain and the distortion of the stage to which it is applied, and its use, within the NFB loop, to increase the loop gain and thereby to reduce the overall distortion level has been employed many times.

However, in this instance, the increase in the gain of V2 makes it necessary to reduce the input signal to V2 grid to a much lower level than would normally have been used.

The actual values of R7 and R8 would imply a stage gain of 253 for V2, which is rather more than twice that which would have been expected.

Another unusual factor is the cross-coupling, by C 1, of the G2 grids of V 1 and V2, instead of the more normal separate decoupling of both grids to the ground (0V) or cathode line. This, and the shared cathode bias resistor (R4), helps to restore the equality of the two drive signals applied to the output tubes.

FIG. 6 The Quad amplifier

The Baxandall Amplifier

There was clearly a degree of rivalry between amplifier designers at this time, since, shortly after Williamson had described his 15 watt design, in 1947, Baxandall proposed in 1948 a circuit design for an amplifier offering a comparable standard of quality at a lower cost- but also over a more limited frequency range (30Hz-15kHz) and with a lower (10 watt) power output - (Baxandall, P.J., Wireless World, pp. 2-6, January 1948). There were two innovative features shown in Baxandall's design, of which the first was the decision to use pentode (or tetrode) tubes throughout, including a push-pull beam-tetrode output stage. Baxandall argued that the extra gain obtained this way, for the same number of gain stages, would allow more NFB to be applied, without instability, and this would reduce the hum and noise in the system and would also allow a substantial reduction in the overall THD. The second innovation proposed in this circuit was to derive the overall NFB signal from a separate secondary winding on the output transformer. The absence of any significant output load on this winding greatly reduced the internal phase shifts within this part of the transformer and facilitated the use of quite a high level of NFB -- up to 36dB, adjustable by RV3 -- without the need for the elaborate and costly design described by Williamson. Unfortunately this idea had been patented by C.G. Mayo, of the BBC Engineering Research Dept., and any commercial manufacturer who sold transformers to Baxandall's design would have to pay royalties to the BBC. I have shown the circuit of the Baxandall 10W amplifier in FIG. 7. In terms of its circuit structure, it uses a pair of SP61 high slope pentode tubes (V1 and V2) as a floating paraphase phase splitter directly driving an output pair of 6L6s, V3 and V4.

These tubes were the RCA (USA) equivalent of the Marconi-Osram KT66s and had a somewhat lower permissible anode voltage rating and anode dissipation.

FIG. 7 The Baxandall amplifier

Interestingly, Baxandall proposed the use of a square wave test signal as a means of uncovering wide-band performance defects, and used this method, with a 50Hz input signal, to show the very considerable improvement which NFB made to the overall LF performance- a possible weak feature in any tube system. It is a pity, in retrospect, that the performance of this amplifier was not tested also with a high frequency square wave- say 10kHz- while driving a reactive load, in that this would have allowed a more direct comparison to be made between this amplifier and contemporary solid state audio designs, which are often tested in this way.*


* Measurements, in 1966, on two tube amplifiers, one of commercial origin and one of my own design, showed that the handling of a 10kHz square wave was relatively poor, even on a resistive load. Unfortunately, although the memory remains, the oscilloscope photographs from these tests are no longer available.


Baxandall had taken particular care in tailoring the loop phase shifts within this design, and had chosen the values of the coupling capacitors (C2,C5) and those used for screen decoupling (C3,C6) to constrain the amplifier LF phase lead to an adequately low level. A similar function, at HF, was performed by the step network (C1, R3) across the anode load resistor (R4) of V1, and the Zobel networks (R21,C9 and R22,C10) across the two transformer half primaries. The Zobel networks also protected the output transformer (and output tubes) against damaging high voltage swings if the amplifier was over-driven, or operated into a very high impedance (or open-circuit) load.

Unfortunately, although the more attractive features of this design were copied, usually in a simplified form, by many amateur constructors, Baxandall's design did not succeed in displacing the Williamson circuit as the leader in public esteem, or in overall sound quality. The weak link, both in terms of the cost and complexity of the output transformer, and in the amplifier performance as a whole, was, of course, the fact that the NFB was drawn from a different winding from that which fed the loudspeaker. This led to the possibility that if the loudspeaker load current was distorted due to the non-uniformities of the moving coil magnetic system- which would usually be the case - the output signal from the amplifier would also be distorted- because of the effective high output impedance of the power amplifier.

Baxandall noted this fact in two places in his article describing the design, but blamed the fault on imperfections in the loudspeaker- as if there would ever be a perfect LS unit.

FIG. 8 20-25W amp (JLH 1951)

However, as I noted, this didn't prevent the general layout of his circuit being copied by various amateur designers. An example of this is given in a circuit of my own, shown in FIG. 8. This was designed around a relatively inexpensive 20W. output transformer made by the Wharfedale Wireless Works, which had rather poorer figures for primary winding inductance and leakage reactance than Williamson's, but which would nevertheless allow the circuit shown to work quite stably with about 15dB of NFB if local feedback, around the output tubes, (R16,R19) restricted their stage gain to about 12". I made several of these amplifiers at the time (1951-1952), one of which was made for a local music loving cabinet maker in exchange for a most magnificent oak radiogram cabinet which he then built to my design.

Baxandall's 5 Watt Design

In March 1957, Baxandall returned to the field of audio amplifier design with a simplified, relatively low power (5W) design, because he felt that there was a niche for an inexpensive design intended for use with relatively efficient loudspeaker systems (Baxandall, P.J., Wireless World, pp. 108-113, March 1957). In this design, of which I have shown the circuit in FIG. 9, Baxandall retained both the use of output pentodes and a fairly high level of NFB to keep the residual hum level low, and the total harmonic distortion (THD) below the 0.1% target level at 400Hz and 5 watts output, though he now used the split load phase-splitter system (V2) proposed by Williamson.

FIG. 9 Baxandall's simple amp

The use of high levels of NFB with an output transformer of relatively simple construction brought with it the probability of HF instability - which, if it occurred while handling a speech or music signal, would lead to a greater degree of impairment in the audible quality than any possible improvement due to reduced steady-state THD. To provide the very necessary correction to the poor HF loop stability, Baxandall added a further output Zobel network (R22,C10) across the whole secondary of the transformer, in addition to the networks (C8, R20, C10, R21) retained across the output transformer half primaries. Further circuit refinements for the purposes of HF NFB loop compensation were the lag-lead networks (C2, R8 + C3, R9) interposed in the signal line between V1 and V2.

The Radford Tube Amplifiers

These designs were marketed by Radford* Audio Ltd, of Long Ashton, Bristol, over a period of nearly twenty years, and enjoyed a reputation among music lovers which was second to none, mainly because of the very high quality of the output transformer and the general attention to detail in components and design. They were normally available in output power ratings from 7 to 30 watts, though units up to 100 watts could be provided to special order. The circuit I have shown in FIG. 10 is that of their MA15 15 watt mono version, but the other designs were very similar in form.

In terms of its circuit structure, the output stage uses a push-pull pair of output power pentodes, operating in class A, and coupled to an output transformer which is used in a 40% U-L mode. The output tubes are individually cathode biased and are driven by an ECC83 double-triode long-tailed pair phase splitter. Negative feedback is taken from the whole of the output transformer secondary winding to the cathode circuit of V1 via a phase correcting network (R28,C17), of which the component values are switched to suit the output load impedance chosen. (This is also switch selected, but I have omitted the details of this in the interests of clarity.) Local NFB is applied between anode and grid of V1, and additional loop phase correcting networks are included across V1 anode load (R7,C3) and in the grid drive to the output tubes (R16,C8 and R17,C9). Apart from the primary inductance of the transformer the only LF phase error inducing components are C 10, R 18 and C 11 ,R 19, and this ensures loop LF stability.


*Arthur Radford, the owner and founder of the firm, was, in his own view, a transformer designer who 'dabbled a bit in electronics '. However, he had not come to terms with 'solid state' electronics, and although his firm manufactured and sold transistor operated audio power amplifiers, they were not of his design, and the amplifier which he preferred, and used in his own home, was the STA25 -- a 25 watt stereophonic amplifier which was one of the last of his own creations. I had the great privilege of acquaintanceship with Mr Radford, and was invited to his home on one occasion in the early 1970s so that we could do a comparative listening trial between his own domestic STA25 and my recently designed 75 watt (Hi-Fi News) transistor operated unit. I had expected that the two power amplifiers, which, for the purposes of the trial, were both driven from Radford's own tube preamp, would sound the same, and, so far as I could tell, this was indeed the case. However, I was flattered to find that Mr Radford thought that my design was superior in sound quality to his, and he then invited me, as a consultant, to do some electronic circuit design work for him, which suggested that his judgment was not just a compliment intended to please a guest: not that I thought that such an act would have been in his nature.)


FIG. 10 The Radford design

The Leak TL/12

This circuit, which I have shown in FIG. 11, was conservatively rated at 15 watts, and used a circuit which was very similar to that of the Radford, the main differences being in the capacitative--rather than direct--coupling between V1 anode and V2 grid, and that the output tubes were triode (rather than U-L) connected. As in the Radford an HF loop phase correction network (R23/C12) was included in the NFB line, and, once again, the values of these components were switched to suit the output transformer secondary ratio chosen. The only other loop compensation component was the 1nF capacitor across R6 in the cathode circuit of V1.

The Mullard 5/10 and 5/20 Designs

The Mullard 5/10, so described because it had an output power of 10 watts and used five tubes, including the rectifier, and its companion design with 20 watts output, formed the third of this closely similar group of designs, and, because Mullard Ltd was a wholly owned subsidiary of the Dutch Philips Electronics NV organization, these circuits could be taken as representing a broad consensus of European audio amplifier design. Unlike the situation in the UK and the USA, beam-tetrode tubes were not used, output stages being based solely on pentode tube types.

The design I have shown in FIG. 12 is the Mullard 5/10, a design which was enormously popular among amateur constructors, and for which output transformers and mains transformers were available from a wide range of suppliers. The main differences between the 10 and 20 watt units were that the higher powered version used a higher HT line voltage, used EL34s rather than the smaller, pentode connected, EL84s as the output tubes, and that the output was U-L connected. Mullard noted that the best performance -- presumably in respect of output power - was with the screen tapping points being taken at 20% from the center, though they did give technical details for the use of the 43% tapping connections, which was an operating mode favored by the average constructor and transformer supplier.

FIG. 11 The Leak TL12

The quoted harmonic distortion for both these designs, at their rated output powers, was in the range 0.3-0.4%, which, though entirely adequate for most amateur use, did not compete, in the minds of the audio enthusiasts, with the less than 0.1% THD at full output power which was claimed for the Williamson, the Radford or the Leak designs.

Full details of these and other Mullard audio designs were reprinted by Mullard in 1959 in a booklet Mullard Circuits for Audio Amplifiers and this was reprinted in 1993 by Audio Amateur Publications, of Peterborough, NH, USA. The GEC Audio Amplifiers.

A major milestone in audio amplifier design was the Williamson amplifier, of 1947, and the British General Electric Company, as the owners of the Marconi/Osram tube manufacturing plant were D.T.N. Williamson's employers. The celebrity which his work had generated, even after Williamson had left the GEC and had joined Ferranti, provided an incentive to their tube development laboratories to develop a range of audio amplifier designs, of which the most significant, from my point of view, was a 30 watt U-L design shown in FIG. 13, which was basically a somewhat modified and simplified version of Williamson's 15 watt circuit.

FIG. 12 The Mullard 5/10 amp

FIG. 13 GEC 30W Amplifier

This amplifier could be built with KT66s, but gave a better performance, <0.2% THD at 30 watts, if the (then) newly introduced KT88s were used as the output tubes. With either KT88s of the recently introduced KT77s (a redesign of the KT66) the power of this amplifier could be increased to 60 watts by employing fixed bias rather than cathode bias. This is done by simply removing the cathode bias resistors of V5 and V6 and returning their cathodes to the 0V rail, and then applying a negative grid voltage of some 48 volts in value to the earthy ends of the grid resistors R14 and R15. In the revised 60 watt circuit the negative grid bias for each output tube was obtained separately from small preset potentiometers connected across a simple 55V negative DC supply.

FIG. 14 The GEC '912 +'

The GEC 912+ Amplifier

The period from 1950 to 1965 was one of great activity on the part of amateur audio enthusiasts who tried out a wide variety of experimental ideas. The GEC company reflected the mood of the moment and offered a metal cone loudspeaker driver unit, and an economical and easy to build 12-14 watt audio power amplifier- the GEC 912+. The + indicated that it included a simple preamplifier with a gramophone record replay frequency response equalization stage and a passive tone control circuit. The actual output power depended on the design and efficiency of the output transformer.

In the circuit which I have shown in FIG. 14, I have deleted the input, single tube, preamplifier and tone control circuitry, which would provide the normal input to V1, in the interests of clarity. I have also deleted the presence switch, in the NFB line, with which an RC series network could be switched across either R11 or R12, depending on the effect sought.

In the view of the amateur constructor this amplifier was in direct competition with the highly successful Mullard 5/10 design, with the GEC unit offering a somewhat greater power, at a rather worse full output power (typically 0.8% THD) distortion figure. In terms of circuit design, the GEC amplifier used a single triode input amplifier (V1), followed by a split load phase splitter (V2) which provided the drive for the output (U-L connected) pentodes. The total loop feedback was not high, because of the limited gain of the amplifying stages. This was the reason for the relatively poorer THD figure for this design, but the greater loop stability then allowed 'presence control' type tinkering with the frequency response.

Other GEC Audio Designs

It is noteworthy that GEC had produced a number of application reports during the 1950s which covered a wide range of tube amplifier designs in this field, from 3W to 1100W. Most of these designs were gathered together in a booklet. An approach to audio frequency amplifier design, originally published in 1957, and reprinted in 1994 by Audio Amateur Publications, of Peterborough, NH, USA. In addition to these design booklets GEC's tube development laboratories continued to produce application reports relating to tube audio amplifier design up to 1979 -- by which time almost all the audio design effort had followed the engineering developments in semiconductor technology into the field of the solid state.

It is somewhat to be regretted that amongst all this effort little novel circuitry emerged from the GEC's audio engineers, except in the very high power designs, and these were aimed at public address use, rather than high quality domestic audio applications, and mainly were not of a type which would allow the use of overall loop NFB to attain low levels of harmonic and intermodulation distortion, and mainly didn't employ overall NFB.

FIG. 15 The Brimar 25P1 amp

The Brimar 25P1 Design

This is, in many ways, similar to the Leak TL 12 or the Radford STA 25 designs, but has some interesting design features, and achieves a very good performance with a relatively simple circuit layout. The quoted performance is 25 watts output at less than 0.1% THD, at 1kHz, a bandwidth of 25Hz-20kHz, +0.2dB, an input sensitivity of 480mV for 25W output, and a hum and noise figure of >85dB referred to 20 watts output. Other figures which are very good for a tube amplifier design are: Rise time 6uS, Phase shift, <20 ~ at 25kHz, and Zout <0.4f2 from 50Hz to 10kHz.

Structurally, the circuit consists of a high gain pentode input stage, with an HF phase compensation network, C2 R24, across its anode load resistor. This drives a pair of triodes, V2,V3, which act as a floating paraphase phase splitter, and which in turn drive an output push-pull pair of beam-tetrodes, V4,V5. (The output tubes specified were STC type 5B255Ms, but these were closely similar to the contemporary KT66s, which I have shown.) An innovative feature is the use of a resistor, R12, which is common to both V2 and V3 anode circuits to derive the paraphase input signal to V3.

This leads to rather less loss of gain than the rather more common grid-derived paraphase input signal. A further subsidiary NFB loop, which contributes a further -10dB to the total, and assists in achieving a good loop stability, is drawn from V2 anode circuit by way of R9. The circuit shown is for a 15 ohm output load, and the values of R23 and C12 are those appropriate for this load figure.

I consider this circuit to be an excellent example of tube audio design practice, along with the Brimar SP55 preamplifier quoted in Section 5, which could well serve as models to those designers wishing to return to tube operated audio systems.

The Sonab Design

Sadly, from the point of view of the historian, there have been relatively few genuinely original tube circuit designs in the audio amplifier field, but the OA6 Mk 1 design offered by Sonab, of Sweden, during the period 1966-1970, is a remarkable example of circuit innovation. I have shown the basic layout of this design in FIG. 16, simplified by omitting the various NFB loops as well as the power supplies to the second grids of the output pentodes. A somewhat unusual feature of the design is that it requires a conducting path between its LS output terminals, and it will not work, electrically, without such a low resistance output load. I have assumed that this will always be the case for the purposes of the circuit analysis.

FIG. 16 The Sonab system

The output stage is connected as a White type cathode follower (see Jones, M., Tube Amplifiers, Newnes, 1995, p. 91). In this layout, the output cathode follower, V4, feeds an active load, V5, which is driven in phase opposition to V4. This phase inversion of the signal is provided by an additional triode amplifier stage, V3, inserted between V2 and V4. The cathode potential of V4 is arranged to sit at about half the input HT line voltage by taking its grid to a voltage divider network, R18/R19, which is connected between V+ and the 0V line via the LS load. Because of the low impedance of this output stage- approximately half that of V4 acting as a cathode follower on its own- it is possible, with this layout, to drive a high impedance LS unit (say, 50-100 ohms), or a pair of headphones, directly, without the need for an impedance matching output transformer.

In the complete circuit of the OA6/1 design, shown in FIG. 16, several further innovative features are added. Firstly, the input tube, V1, is operated in a virtual earth configuration, with the input and NFB signals being summed at its grid. This shunt type NFB is theoretically superior to the more common series type NFB, where the feedback signal is returned to the input tube cathode, because it is free from any possible common-mode errors. Secondly, both voltage mode NFB (taken across the LS load, via NFB 1) and current mode NFB (derived, via NFB2, from the current flowing through the LS load by a current transformer, TR1) signals are employed. In addition, a third NFB path is taken directly from the amplifier output to the cathode of V2.

Since Sonab felt that the availability of suitable high impedance LS units might be limited, they provided, as the other half of the OA6 design, a mirror image circuit in which only voltage feedback was used, and in order to provide a low Z output the amplifier output was coupled to the load via a conventional output transformer. Since the voltage NFB signal in the low Z half of this amplifier was taken from the primary of the output transformer, this would leave the design open to the inherent fault Baxandall found, in his design of FIG. 7, that the inevitable loudspeaker load non- linearities would distort the waveform at the amplifier output- regardless of the absence of distortion in the circuit of the amplifier itself.

By modern standards, the resistance values employed in this design, such as in the input and NFB paths to V1, are very high, and would lead to a relatively high background thermal noise level, as well as some DC level instability, due to tube aging processes, as a consequence of high value grid leak resistances.

Increasing Available Output Power

As has been seen, the use of the U-L output transformer connection provides a simple means for increasing the output power obtainable from a given design, but, with suitable protection arrangements to prevent inadvertent operation of the system without output grid bias, fixed bias operation can also substantially increase the available output. For example, in the case of U-L connected KT77, at 430V HT, an output power of 34 watts is quoted at 2.5% THD, while a similar fixed bias operation will allow an output of 48 watts at 1% THD. For the KT88, the comparable figures, at 460V HT, are 50 watts and 80 watts respectively.

FIG. 17 The Sonab 0Q6/1 amp


With hindsight it is easy to see the pitfalls into which some of the pioneering tube amplifier designers fell. The need, for example, for the amplifier LS output to be taken from the winding, and the whole of that winding, from which the loop NFB signal was drawn, was not generally appreciated until the late 1950s, this allowed the situation in which the amplifier could be virtually distortion flee, until the LS load was connected.

By the time of the Radford and Mullard designs it was accepted that the NFB signal must be taken from the LS output connections, and if the transformer primary:secondary ratio was changed to suit different LS loads then the attenuator network (cf. R28 and C17 in FIG. 10) in the feedback path must also be changed to suit this new condition.

In a tube amplifier, the output transformer is a source of waveform distortion, so, for an output THD of better than 0.1% over the bulk of the audible frequency spectrum, overall loop NFB, including the output transformer, must be employed. The amount of this which it was needful to employ would depend mainly on the quality of the output transformer, and on the type of output tube used. For a well made LS transformer, and triode connected output tubes, this degree of linearity could be met with some 20dB of NFB. For U-L connected beam-tetrodes, this performance would require some 26dB of NFB, while pentode or beam-tetrode output tubes, an NFB ratio of nearer 30dB was needed. Beam tetrodes were, in general, slightly more linear than pentodes.

In order to have an adequate gain margin to allow sensible amounts of NFB to be employed- without the need for very high input signal voltages - either a driver stage must be used between the phase splitter and the output tubes, or a phase splitter with both a high gain and a large permissible output voltage swing, such as a floating paraphase or a long-tailed pair, is required.

Unless the required gain can be obtained with relatively few gain stages, the use of NFB will lead to problems with HF and LF instability. Poor loop stability at the LF end of the spectrum will probably cause motor-boating which will be audible. HF instability may cause continuous oscillation at frequencies which are outside the heating range, which will spoil the sound quality and may damage the amplifier or the LS units. An insidious problem with poor loop stability margins is that the instability may be sporadic, triggered by certain combinations of signal frequency, output signal level and load reactance. This is most easily checked by the use of an oscilloscope, a variable frequency square wave input and a range of non-resistive loads. A number of HF/LF phase compensation networks have been shown in FIGs 7, 9 and 10.

Prev. | Next

Top of Page   All Related Articles    Home

Updated: Saturday, 2015-05-23 8:26 PST