|Home | Audio Magazine | Stereo Review magazine | Good Sound | Troubleshooting|
The evolution of power amplifier design
The design of electronic equipment for use in audio applications presents some unique problems to the circuit engineer, not least because the final result is intended to appeal to the human ear. For this reason, the relevance of specifications based purely on engineering considerations will always be a matter for some debate, and although some technical standards have been adopted because they were supported by the results of deliberate listening trials, held with audience panels composed of engineers or musicians, they exist mainly either because they are generally in accord with experience, or just because they seem to be sensible.
The contemporary preoccupation with the pursuit of perfection in both the electronics circuitry and the other hardware used in audio systems is a relatively recent phenomenon, and has come about because of the development of very high quality sound storage and reproduction systems, which are themselves mainly of fairly recent origin. In earlier times other, not usually electronic, considerations were thought to be of greater importance, if only because the possible quality of the results given by the other items in the recording and reproduction chain was not very high anyway.
Obviously, since the domestic audio equipment market has always been a very competitive one, the expectations of the user, and the apparent value for money offered by the products, have been very import ant considerations in the eyes of the manufacturers.
However, the dominant design factor in audio amplifiers, is, and has always been, the nature of the hard ware which was available at the time. For this reason, the main flow of audio design has nearly always followed the evolution of electronic circuit components and the electromechanical transducers, such as loudspeakers or gramophone pick-ups, which are used in association with it. Even the nature of the circuitry employed has usually been dictated by the advantages and limitations of the components, and the needs, in respect of input signal voltage or output power demands, of the input/output transducers.
Output power and bandwidth:
The levels of output power which will be needed, from any audio amplifier system, depend so heavily on the use to which it’s to be put that it’s difficult to suggest more than tentative figures for this part of the specification. For equipment which is to be used only in a domestic environment, with reasonably efficient loud speaker units -- with, say, input/output efficiencies above 90dB, s.p.l. (sound pressure level, per watt, at one meter distance, axially, from the speaker) -- and with typical 'easy listening' or light classical music, it’s unlikely that more power than 3 watts per stereo channel will ever be required, but if LS units having similar efficiencies are to be used, on the same type of music, for lecture purposes, even in only a medium sized hall, then up to 30 watts of power will probably be needed.
For the 'hard rock' or 'heavy metal' music enthusiast, with less efficient loudspeakers -- say, less than 85 dB, s.p.l. per watt/meter -- and very tolerant neighbors, then something over 100 watts per channel will certainly be necessary. Still larger powers would be needed to reproduce, in a large room, a realistic sound level equivalent to that of a grand piano played with panache, even with reasonably efficient (85-90dB s.p.l.) LS units.
It’s useful, also, to remember that, under lecture conditions, because of the sound deadening effect of furnishings, or simply because of the presence of a number of people in an audience, the effective sound level reaching the listener will be significantly reduced, by comparison with that, in the same room, when empty. So, taking these factors into account, one might divide these various uses into groups, such as 'average domestic' -- 3-10 watts, 'general purpose' -10-30 watts, 'pop music' -- 60-120 watts, and 'studio/lecture' -- 80-500+ watts.
It’s generally assumed, as a basis for the necessary bandwidth for audio systems, that the human ear can respond to sound frequencies over the range 20Hz 20kHz. However, in reality, very few adult men, even when young, can hear much above 16kHz, and this upper limit of frequency response will decrease gradually with increasing age. Also, very few women can hear much below about 35^0Hz. Children may hear sounds at frequencies well above 20kHz, but not usually those much below 100-150Hz. Again, few LS units, and very few listening rooms, will allow sound pressure waves to be reproduced at frequencies below 35Hz.
As for the program sources, FM radio is restricted to an upper limit of 15kHz, though frequencies down to 30Hz can be transmitted. With a few exceptions, AM radio transmitters do not, nowadays, offer signals above 5kHz, because of international frequency allocation agreements. Vinyl, and some cassette tape machines, can reproduce frequencies up to 20kHz, though only at low levels, and with substantial (10% or more) waveform distortion. Similar constraints exist below 35Hz, with both tape and vinyl LPs. Compact discs do, however, offer a true 20Hz-20kHz low distortion program source.
A point which must be kept in mind, in any consideration of the bandwidth specification for an audio amplifier, is that too wide an extension of the HF or LF bandwidth can allow the intrusion of acoustically unpleasant and unwanted noises into the final sound.
An unexpectedly high background hiss level on an FM signal could easily be due to this cause. This problem can arise because the inevitable HF or LF nonlinearities, and other mechanical imperfections, in the LS units or headphones used to reproduce the signal, may allow audible cross- modulation effects between signals which are themselves outside the audio band, and would normally have been quite inaudible. So, for this reason, it’s sensible, in amplifier design, not to seek to preserve, purely because it looks good on the specification sheet, a high frequency response which ex tends too far beyond the possible frequency range of the ears of any listener. Having said that, care should also be exercised in the way in which the bandwidth is curtailed, since too steep a low-pass filter characteristic can itself impair the sound quality of the system, by modifying its transient response.
Most sounds produce air pressure waveforms which are irregular, non-sinusoidal and asymmetrical in form, especially those which are caused by percussive actions. At first glance, it would seem obvious that realism, in the reproduction of these sounds, depends on the ability of the system to generate, at the ear of the listener, air pressure patterns which are closely similar to those of the original sound. However, in a normal, medium sized, listening room, having various kinds of 'cavity resonances', filled with sound-absorbing furniture, and with reflecting walls, floors and ceilings, this is an impossible task. This can easily be demonstrated, in any listening room, by connecting a microphone to an oscilloscope and then watching the large changes in the reproduced waveform which occur as a result of relatively small movements in the microphone position, even when it’s exposed to exactly the same audio signal.
So, the fact that the ear can recognize, and regard as realistic, such a wide range of possible acoustic wave forms, which can change, literally from inch to inch, with movement of the microphone or the ear, is as much a tribute to the sound recognition ability of the brain as it’s to the precision of the reproducing equipment in use.
However, though the ear may be very tolerant of errors in the flatness of frequency response of the system, or the time of arrival of the individual frequency components which go to make up a complex signal, there are various types of waveform distortion which can be shown to degrade the sound quality, and one of the tasks of the audio engineer is to make all of these distortions imperceptibly small.
Under steady-state conditions, the waveform or harmonic distortion of an amplifier, using a constant amplitude single frequency (sinusoidal) input signal, can be measured quite easily, and this gives a convenient and widely used method of assessing one particular aspect of amplifier quality. Unfortunately, mainly due to the claims of the sales departments of the amplifier manufacturers and their advertising agencies, who are aware that a large number of 'O's behind the decimal point in the total harmonic distortion figure looks good on paper, this part of the overall specification has assumed an excessive degree of importance in the minds of many potential purchasers of audio equipment.
It’s certainly true that there are some very objectionable forms of harmonic distortion, such as the dissonant high-order odd harmonics, such as those at the 7th, 9th, 11th and upwards, which can be generated by 'crossover' effects in class B push-pull amplifiers; and it’s desirable that these should be kept below 0.05%, at all signal levels, if their presence is not to be detectable. However, most musical signals are rich in naturally occurring harmonics, mostly of the lower harmonic orders, such as the 2nd, 3rd, and 4th, and the presence of these harmonics adds richness to the quality of the sound. Also, in the sounds produced by many of the stringed instruments, there are a whole range of overtones, of which the higher ones are not strictly in harmonic relationship to the fundamental tone. Musicians refer to these overtones as 'partials', and the size, and frequency distribution, of these, for any given note, determines the sound quality of the instrument, and in violins, for example, would help to distinguish a Stradivarius from a Joe Bloggs.
It’s not, therefore, the presence of low order harmonic distortion which is undesirable in an audio amplifier -- indeed, at levels below, say, 0.5% such harmonics are probably undetectable -- but the inter modulation (IM) distortion which such nonlinearities in the transfer characteristics will cause. This kind of distortion leads to the generation of spurious compo site signals, as a result of the mixing of the various components of any input signal, as illustrated in FIG1, and causes confusion and a lack of clarity or 'transparency' in the final sound.
FIG1 Effect of nonlinear transfer characteristics in generating intermodulation distortion.
There is not a direct and simple relationship between harmonic and intermodulation distortions, but, as a general rule, the lower the harmonic distortion, the lower the IM distortion will be, and the more transparent the sound quality will appear. Once again, where the waveform distortion (THD) figure is below 0.05%, IM distortion will usually be negligible, and no further improvement in THD is likely to give audibly detect able benefits.
No international standard has yet been adopted for the measurement, or the specification of the accuracy, of the transient response of an amplifier system, apart from the 'settling time' measurement which is some-times quoted in high quality operational amplifiers, which is illustrated in FIG2. Because no specification exists, it often seems that little attention is paid in commercial audio amplifier design to the nature of the response of the circuit to an input step-function or square-wave signal, except, perhaps, with a purely resistive output load which is quite unrepresentative of real-life operating conditions. This is a matter for regret since, now that circuit technology allows the design of amplifier circuits having a very high degree of steady-state linearity, it’s probable that the residual errors in the transient response of the equipment --which can be quite large -- arc responsible for most of the audible differences between one otherwise high quality unit and another.
FIG2 Illustration of the settling time specification in operational amplifiers.
There are several mechanisms by which transient response errors can modify the sound of an audio system. Of these, the most common is the effect due to the presence of an 'overshoot' following the leading edge of a step-function, as shown in FIG3a. This will lead to a frequency response of the type shown in FIG3b, when the system is fed with a swept frequency square-wave input signal, or any other signal source of wideband characteristics. One effect of this is to redistribute the spectral energy of such a wideband signal, to give a higher output at the point of the upward kink in the frequency response, which appears to over-emphasize high frequency components in the signal and lead to a 'hard' or 'bright' quality to the sound.
A further cause of poor sound quality, as a result of transient malfunction, is that where the poor transient response is due to inadequate HF stability margins in the amplifier feedback loop, especially under highly reactive LS load conditions. This can also modify the behavior and output voltage on wideband signal frequency distribution of a wideband signal, to cause energy peaks at those frequencies associated with poor loop stability.
Many of these problems in transient response have arisen because designers have endeavored to produce electronic power amplifier circuits with unnecessarily low levels of harmonic distortion, in order to satisfy a presumed customer demand, promoted by advertising hyperbole, and have used over complex, or inadequately stable, feedback loops in the pursuit of this end.
FIG3 Relationship between transient
Most audio amplifier systems will be required to operate into a loudspeaker load, and this can often present problems in matching. This difficulty arises partly because, in order to achieve the flattest practicable frequency response, the loudspeaker designers will usually employ multiple driver units, with complex inter-unit crossover networks -- which can offer very low impedances at certain parts of the audio frequency pass-band -- and partly because the LS unit has dynamic characteristics of its own. Because of this, in addition to the inductive or capacitative nature of the load which it presents, the LS load can provoke unexpected responses in the amplifier driving it. Unless the designer of the amplifier can be very certain of the nature of the load with which his design will be used, it’s prudent for him to assume worst-case conditions, since an unsatisfactory combination of amplifier output and load characteristics may significantly worsen the performance that the amplifier might otherwise be expected to offer.
In spite of these residual problems, it’s undoubtedly true that, with a few exceptions, most of the amplifier circuits currently offered as Hi-Fi units, in the mid- to upper price range, are capable of a vastly better performance, both in respect of technical specification and in terms of audible results, than all but the very best of their predecessors of some decades ago, if only because these earlier designs were aimed to satisfy much more modest specifications, were based on a more limited range of circuit components, and were evolved at a time of a more limited understanding of the field. Since there is a tendency for nostalgic feelings towards early audio amplifying equipment, which is supposed to have been capable of a degree of sonic perfection lost in the progress of circuit design, it may be instructive to look at the actual types of circuit used in the earlier days of 'audio' and at the technical standards which were attained. It should be remembered that when electronic devices were first employed to amplify audio signals it was almost sufficient that the equipment worked at all, and relatively little attention was directed towards the actual quality of its performance, so long as it was not so bad as to be actually objectionable to the user.
Early amplifier designs
Simple valve designs:
In the early years of 'radio' most amplifier systems were operated from primary cells (dry batteries) as the principal source of power, at least so far as their high tension (HT) DC supplies were concerned, and since these were an expensive source of energy, economy in the use of power was a major consideration. Indeed, Briggs, in his classic book on loudspeakers (Briggs, G. A. and Cooke, R. E., Loudspeakers, 5th Edition (1958). pp. 11-13. (Wharfedale Wireless Works, Bradford, Yorks)), says that the success of one of his early Wharfedale LS driver designs was almost entirely due to its high sensitivity.
Similarly, the most common input transducer element, the gramophone record 'pick up', was almost invariably of piezo-electric type, in which a sandwich of foil electrodes and a piezo-electric crystal (normally Rochelle salt) was arranged so that the lateral displacement of the stylus, as it followed the undulations of the gramophone record groove, would cause a twisting action of the crystal sandwich, which would produce an audio voltage output. Because of the high intrinsic capacitance of the crystal transducer element, and the high load resistance which would normally be used with it, the high frequency response of this type of pick-up was very poor, and because of its stiffness its waveform distortion was very high.
However, in the same way that the principal quality sought, at that time, in the LS unit was that it should produce a high sound level for a modest electrical input, with wide bandwidth or flatness of frequency response being minor considerations, so, in the case of the gramophone pick-up, a high electrical output was one of the major requirements, in order to reduce the number of stages of amplification which would be required to achieve an adequate output signal level. By comparison, the relatively poor high frequency response or linearity of such crystal pick-ups was of secondary importance.
A typical battery operated audio amplifier:
A typical battery operated audio amplifier for gramophone record reproduction, of this period, would be of the form shown in FIG4. The anticipated output of the gramophone pick-up would be of the order of 0.5-1V RMS at 1kHz, with a 78 RPM shellac record of average' loudness'. This would be large enough not to require a high degree of subsequent amplification, though somewhat too small to drive the output stage (V2) directly. An input stage, V1, would therefore be added to provide some additional amplification of the signal.
It’s necessary, in almost all thermionic valves, to apply some negative bias voltage to the control grid to set the anode current to the required level, but since battery operated valves use a directly heated, barium oxide coated, filament as the electron source, and all of these filaments are connected in parallel, it’s not convenient to develop the required negative grid bias voltage by just putting a resistor in the cathode circuit, as is done with valves having indirectly heated cathodes.
For the first stage, which had a high value of input resistor (R1), the residual grid current flow through this resistor would usually provide an adequate negative bias voltage to ensure that V1 would operate in a reasonably linear part of its I/Vg characteristic. This technique is known as grid current biasing. Ideally, the grid bias voltage should be chosen to ensure that the anode potential was about half the available supply voltage, which is the preferred operating condition.
However, since the available HT voltage will be between 90V and 120V, and the required output voltage swing from this stage would only be some 5V RMS, the precise anode voltage level would be far from critical. Although this type of biasing technique is simple to use, it will cause the larger positive-going peaks of the input signal to clip, through the diode action of the grid-cathode circuit, and this will lead to some worsening of the circuit distortion characteristics.
For the output stage, a pentode valve would normally be used in the interests of efficiency, and the anode current of this would normally be controlled by an externally applied negative bias voltage, typically of the order of 3V-4.5V, derived from a separate grid bias battery. These batteries would offer a range of output voltages between 1.5 and 9V so that the user could make his own choice between economy and harmonic distortion.
Although the use of a pentode output valve was preferred because of its higher efficiency, such valves also introduced a large measure of third-harmonic distortion, which gave a somewhat shrill quality to the sound of the amplifier. In order to make the sound somewhat more bland it was customary to connect a small capacitor (C5) across the primary winding of the LS transformer.
Sometimes a simple 'tone control' circuit, consisting of a variable resistor and a rather larger capacitor (CJRV^) would also be provided to allow the user, by adjusting the resistor value, to opt for a progressively more 'mellow' sound characteristic. Such an amplifier, for a consumption of some 10-15mA from the HT source, would give an output of 200-300mW, with an overall THD figure of about 10-20%, mainly com posed of third harmonics, and with a bandwidth, at -3dB points, from gramophone record input to LS output, of some 150-2500Hz.
FIG4 Battery operated two-valve gramophone amplifier.
Mains operated circuitry:
The desire for greater amplifier output powers, and for freedom from the cost and performance limitations imposed by battery sourced power supplies, led to the growing adoption of valves with indirectly heated cathodes. These were of a type in which the heated filament was electrically isolated from the cathode, as described in Section 3, and which could therefore be heated by a low voltage AC supply, derived from a low voltage winding on the main power supply transformer, without problems due to the introduction of mains frequency hum. The HT supplies could also be derived from a rectified AC voltage provided by the same transformer, fed directly from the household electric mains supply, and it became practicable to design amplifier systems in which economy in the use of input power was of minor importance, and more attention could be paid to the linearity and bandwidth of the circuit.
Two fairly typical circuits of the period are shown in FIGs 5 and 9.6. Both of these used push-pull output stages, because these were more efficient, and the symmetrical push-pull layout of the output valves reduced the extent of even harmonic distortion for a given output power level.
FIG5 Simple push-pull mains powered valve amplifier
FIG6 RC-coupled push-pull audio amplifier
In the circuit of FIG5, an inter-stage coupling transformer, TR2, is used to provide the symmetrical anti-phase drive to the control grids of the output valves, whereas in the more elaborate circuit of FIG6, a 'floating para-phase' inverter stage (V2, with R7/R9 and C5), is used to generate the required phase inverted signal input for V4. Both of these amplifiers would give a very acceptable quality of the output sound, and although the layout of FIG6 is theoretically superior, there were some very high quality inter-stage coupling transformers, made by companies such as Varley or Ferranti, which offered an excellent transmission bandwidth, with very low distortion in the output signal, so, from the point of view of the user, both designs would be comparably good.
A point to note in all of these designs is the extensive use of supply line decoupling circuits, such as R2/C2 in FIGs 4-9.6, as well as #8/C3, and C8 in FIG6. These were essential, since, with the typical power supply circuitry of the period, it was inevitable that the fluctuating anode current demand of the output stages would cause voltage ripple on the HT line, and this could intrude into the signal path via the inter-stage coupling capacitors, to cause a characteristic form of low frequency instability, known as motor boating from the sound it would produce in the loudspeakers.
Mains powered circuits of the type of FIGs 5 and 9.6, would offer 6-8 watts of audio signal into a typical 3 ohm moving coil LS load, at perhaps 5% harmonic distortion -- ignoring the contribution of the pick-up element -- with an amplifier bandwidth of ?????-10kHz. These, together with the 3-4 watt out put mains valve operated versions of the circuit of FIG4; commonly referred to as a 'single ended' design, to distinguish it from circuits with push-pull output stages; remained typical of the type of output stages used in the bulk of audio equipment up to the late 1940s, and in many medium price radios and 'radiograms' for at least a decade beyond this period.
High quality valve operated amplifiers:
The demands of radar and airborne navigation equipment had led to an enormous development in electronic circuit technology during the period 1940-45, and this led circuit designers to realize that the use of negative feedback (NFB) would allow a consider able improvement in audio system performance, both in respect of bandwidth and in respect of harmonic and intermodulation distortions.
Although many comparable audio amplifier designs were offered at this time, based on the use of various combinations of negative feedback within the circuit, the one which had the greatest impact on performance expectations in the UK and other English language countries was that published shortly after the war by D. T. N. Williamson (Wireless World, Aug. 1949, pp. 282-284, Oct. 1949, p. 365, Nov. 1949, p. 423). The author of this circuit worked, at the time, for the Marconi-Osram valve company, and his design exploited the qualities of the recently introduced 'KT66' output beam-tetrodes, which he used in combination with a carefully designed loudspeaker output transformer.
The Williamson amplifier:
The principal stimulus to this design, and to other similar designs of the same period, was the realization that greater benefits would arise from the use of a NFB loop which enclosed the whole amplifier, including the output transformer, rather than the use of a number of separate feedback loops around individual areas of the circuit, as had been used in the past, although the adoption of such a design approach would require much greater care in the choice of the circuit and component characteristics. In particular, this use of overall NFB required that the total phase shift within the circuit must be maintained at less than 180° within the frequency band in which the loop gain was greater than unity, otherwise the circuit would be unstable, and the amplifier would break into oscillation. Moreover, even when this initial condition was satisfied, it was also essential that there should be an adequate margin of stability within the loop, otherwise bursts of oscillation could be provoked when input signals, especially those of a transient nature, occurred at parts of the frequency spectrum where the stability margin was low. This type of malfunction would greatly impair the quality of the reproduced sound.
This requirement for strictly controlled phase characteristics within the loop discouraged the use of inter stage coupling transformers, since these would inevitably suffer from substantial phase shifts at the extremes of their frequency pass-bands. These single loop feedback designs therefore employed other circuit arrangements for generating the required symmetrical pair of drive waveforms for the output valves.
FIG7 Phase-splitting circuit used in Williamson amplifier
In the Williamson design, this was done by the use of a split-load amplifier stage, of the design shown in FIG7. Since this type of stage would provide only a relatively low undistorted output voltage swing, Williamson followed it by a pair of further triode valve amplifiers to drive the push-pull output valves, giving the complete design shown in FIG8.
Because the anode circuit impedance of both triode and triode-connected beam-tetrode valves is high, typically in the range 500-10,000 ohms, it’s necessary, in order to get the best efficiency, to use an output transformer to match the relatively low load impedance of the loudspeaker to that of the output valves, and it’s easier to minimize phase errors due to this transformer if the transformation ratio is not too high.
Although loudspeaker manufacturers responded to the evolution of these higher quality audio amplifiers by making loudspeakers with 15 ohm, rather than the previously standard 3 ohm voice coil impedance, to facilitate output transformer design, nevertheless an output coupling transformer remained essential.
From time to time, circuits did appear in which the layout was arranged to allow the loudspeaker to be directly coupled to the output valves, to eliminate the need for an output coupling transformer altogether, but the inconvenience, and low overall efficiency, of these designs prevented their becoming popular.
The performance of any amplifier using overall negative feedback derived from the output transformer secondary winding is crucially dependent on the characteristics of this transformer, for which Williamson's specification is given below. This stipulated a much higher quality component for this position than had been used previously, and this, in turn, stimulated a number of transformer manufacturers to provide improved quality components of their own designs.
The Williamson output transformer specification
Primary load impedance Primary inductance Series leakage inductance (whole primary to whole secondary)
Primary resistance 10,000 ohms; 100 henrys; min 30 millihenries; max 250 ohms, max
FIG8 The Williamson amplifier
Triode valves have better distortion characteristics -- they generate some second harmonic, but very little third or other odd-order harmonics -- than either output pentodes or beam-tetrodes, when they are used in the output stages of power amplifiers, but they are relatively inefficient in use. Also, because of the difficulty of getting adequate electron emission from an indirectly heated cathode, all of the contemporary output triode valves, such as the PX4 or the PX25, had directly heated cathodes, which led to difficulties with cathode bias arrangements, and a proneness to mains hum.
Beam tetrode output valves, of the KT66 type used by Williamson, generate less distortion than the output pentodes they were designed to replace, but they are still less linear than triodes. Williamson therefore used triode-connected beam tetrodes in the output stage, which avoided both the problems of pentode-type distortion characteristics and the hum problems associated with directly heated triodes, while the push-pull output arrangement tended to cancel out the remaining second-order distortion. This circuit design offered an overall performance, quoted below, which was substantially better than almost all of the contemporary audio amplifier designs.
Williamson amplifier. Performance specification
Output power, into 15 ohm load --- Total Harmonic Distortion --- Bandwidth 15 W --- less than 0.1% at 15W --- 2Hz-100kHz
However, the overall efficiency of this design, at maximum output, was only of the order of 23%, and this encouraged other circuit designers to explore various alternative 'distributed load' type output stage configurations, some of which gave better efficiencies with beam-tetrodes or pentodes, without significantly worsening the output stage distortion characteristics.
Alternative output stage connections
FIG9 The ultra-linear output stage connection
FIG10 Quad 22 power amplifier
FIG11 The McIntosh cathode and anode coupled output stage.
Of the alternatives to the simple push-pull triode-connected beam-tetrode output stage, the most popular was the so-called 'ultra-linear' or partial triode connection. In this, the screen grids of the beam-tetrodes were taken to taps on the primary of the output transformer, as shown in FIG9. This allowed an increased output power for the same supply voltage and anode current.
Another arrangement, exploited in Britain by the Acoustical Manufacturing Co. (Quad), employed an output layout in which the cathodes of the output valves were taken to a second primary winding on the output transformer, as shown in FIG10. A similar layout was employed in the USA by McIntosh (McIntosh, F. H., and Gow, J. G., Audio Engineering, Dec. 1949, p. 9), in a 50 watt design, shown in FIG11.
In general, such layouts did not offer a better performance than that given by the original Williamson design.
Output stage biasing:
Although in audio amplifiers based on thermionic valves, the total thermal dissipation of the output stages is a matter of less consequence than in the case of transistors, which are devices having a much lower thermal inertia and generally a much lower ability to dissipate heat, nevertheless, even with valves, there is an effective upper limit on the permissible anode current and consequent heat dissipation. Beyond this level the life expectancy of the valve will be much reduced.
If the anode current under no-signal conditions could be reduced somewhat, a higher peak output power could be obtained, without increasing the aver age current rating. The current which flows in the valve is largely controlled by the bias voltage applied to the control grid, and the possible operating conditions are normally described as being in one or other of the various 'classes', such as 'class A', in which the bias level is chosen so that the anode current remains substantially the same, within the whole of the output power range, so long as this is below the 'clipping level'. If the control grid negative bias is increased to the level at which little or no anode current flows under quiescent conditions, then the stage is said to operate in class B. Similarly, if the negative grid bias is increased still further, so that anode current only flows during the peak positive swings of the grid drive voltage, this is referred to as class C. This latter condition is not used in audio amplifiers, which all operate either in class A or at some intermediate quiescent current level, some way towards the class B condition. This is loosely described as 'class AB'. All of the valve output stages considered so far operated in class A, defined, above, as a condition in which the mean anode current is the same in the quiescent state as it’s at full power. This has the great benefit that the harmonic distortion due to the output stage will generally decrease as the output power is reduced, whereas in class B, or other conditions in which the zero signal anode current is less than that at full power, not only will this reduction in distortion not occur, but the actual output circuit contribution to the THD may actually increase as the signal level is reduced. The failure of circuit designers to appreciate this point led to the unsatisfactory performance of almost all of the early transistor operated audio amplifiers -- all of which operated either in class B or in some class AB operating condition which was close to this -- and led, in due course, to the emergence of a band of Hi-Fi devotees who claimed that the only true high quality audio reproduction was that obtained from valve operated amplifiers.
Modern audio amplifier circuit design
Early transistor operated audio power amplifiers:
Although transistors had been available since the early 1950s, it was not until the end of that decade that circuit engineers felt adequately confident of their design skills to offer audio amplifiers, based on these devices, which they felt could compare in performance or output power with the very highly developed thermionic valve operated audio power amplifiers which were then commonplace. This was partly because, during the 1950s, the only mass produced transistors were germanium junction types, which were generally only available in PNP forms. Although a few NPN germanium and silicon transistor types were also marketed, these generally had a low performance and were only offered in very low power versions.
A truly complementary output stage layout, using symmetrical PNP and NPN output transistors, of the type now typical of almost all contemporary audio power amplifiers, was therefore impracticable and most of the transistor designs used inter-stage transformer coupling of the types shown in FIGs 12 and 9.13. The latter design is the output stage of a 15 watt audio power amplifier, using germanium transistors, due to Mullard (Mullard Ltd., Reference Manual of Transistor Circuits 2nd Edition (1961), pp. 178- 180).
As with the previously described valve amplifier designs, the use of overall loop negative feedback (NFB) was desirable in order to improve the linearity and flatness of the frequency response of the system.
The need for inter-stage and output coupling transformers greatly limited the extent to which NFB could be employed without sacrifice of loop stability. More over, as seen in Section 7, unless an adequate amount of NFB could be employed it might impair rather than improve the performance of the system.
FIG12 Typical transformer-coupled germanium transistor power amplifier
FIG13 Mullard 15 watt power amplifier (c.1960)
FIG14 Impedance conversion stage
FIG15 The UN quasi-complementary output stage configuration
The Lin quasi-complementary circuit layout:
The major design breakthrough in this field came with the introduction of the so-called quasi-complementary output stage, due to Lin (Lin, H. C, Electronics, Sept. 1956, pp. 173-175), which largely set the type of all subsequent transistor audio amplifier designs. In this, a small-signal voltage amplifier stage, operating in class A, was used to drive a combination of Darlington and compound emitter follower layouts, of the forms shown in FIGs 14a and 14b, which acted as an impedance converter between the output of the small signal amplifier and the loudspeaker, and allowed the construction of a push-pull output stage, without the need for either input or output transformers, as shown in FIG15. This layout appeared to offer the design solution for which the circuit engineers had been waiting, in that it allowed the construction of a high quality transformerless audio amplifier, which could be built with identical types of power output transistor. How ever, herein lay a hidden source of trouble.
At the time (1956) that Lin proposed his output stage layout, the only commercially available transistors were of germanium type. In these, at room temperatures and above, the relationship between forward base voltage and collector current, shown in FIG16a, had a much more gradual turn-on characteristic at the origin of the graph than is the case with silicon devices, whose base-voltage/collector-current relationship is as shown in FIG16b. This meant that a push-pull arrangement using germanium transistors had a much less abrupt discontinuity at the crossover point -- from one output device conducting to the other- than would be the case if silicon transistors were used. Unfortunately, by the time this circuit came into popular use among the more adventurous of the Hi-Fi amplifier manufacturers, germanium transistors had largely been replaced by silicon types, which led to the problem of inadequately specified distortion characteristics noted above.
In any push-pull output system, in which the two halves of the output stage are caused to handle the positive- and negative-going segments of the output waveform, sequentially, there is likely to be some discontinuity at the crossover point unless both halves operate at a linear region of their transfer characteristics, as would invariably be the case with the previous, class A biased, valve operated power amplifiers. However, with transistors, it was customary to choose a level of bias at which they would operate in class AB or class B, with little or no quiescent collector current. This was done in order to lessen the thermal dissipation of the output stages, which would give a longer life expectancy, and allow a larger output power for any given heat-sink size. This was necessary, at the time, because the manufacturing techniques employed, both in respect of the integrity of the solder joint between the silicon chip and the metal 'header' -- the mounting plate of the device -- and the stability of the thickness of the diffused junction regions were inadequate to prevent deterioration of the devices during use. This was a major problem with germanium power transistors, but was still a likely failure mechanism with silicon transistors, even though these had markedly better thermal characteristics than was the case with germanium devices. Since these problems were worsened at higher junction temperatures, it was very desirable to keep the output devices as cool as possible.
Operation in class B or AB inevitably leads to the generation of high order types of harmonic distortion (7th, 9th, 11th, etc), due to the presence of a kink in the transfer characteristics of the push-pull stage at the point of the crossover from one half of the output pair to the other. This kind of distortion would be inevitable, even if the two halves of the output pair are identical, but in the case of a quasi-complementary layout using silicon transistors this problem is made much worse because the input/output transfer slopes of the two quasi-complementary halves are markedly different, as shown in FIGs 17a and 17b.
FIG17 Asymmetry in upper and lower halves of quasi-complementary pair.
Accurate choice of the quiescent forward bias applied to the output transistors could lessen the extent of the distortion due to this cause, but this would be difficult to guarantee in a commercial product in tended for use in a domestic setting because the precise bias necessary would depend on the junction tempera ture of the output transistors and would also change with time. If too high a quiescent current setting was chosen, thermal runaway might occur, where an in crease in junction temperature would lead to an in crease in quiescent current which would, in turn, lead to a further increase in junction temperature. This would cause an uncontrolled increase in operating current and could lead to the failure of the output transistors. Such an event would reflect badly on the manufacturers of the equipment.
For this reason, commercial designs tended to favor the use of little or no forward bias on the output transistors, and rely on the use of high amounts of negative feedback to linearize the output characteristics. This would readily allow the achievement of THD figures, at full output power, which were at least as good as those given by the traditional valve amplifier designs -- in which the possible amount of NFB was limited by the phase errors introduced by the loudspeaker coupling transformer.
Amplifier sound and listener fatigue:
After the initial enthusiasm generated by the novelty of this new breed of amplifier design, which offered higher powers in a much more compact housing, and which dissipated so little heat that it could be housed almost anywhere, without a specific need to allow for ventilation, some dissatisfaction began to be expressed by a growing number of users in respect of the sound quality these amplifiers gave. This dissatisfaction gave rise to the descriptions of 'amplifier sound' or 'listener fatigue', and related mainly to the general thinness and lack of body or 'warmth' in the sound quality of these new types of amplifier, coupled with various other, difficult to define, aspects of their performance, which made the user reluctant to spend much time listening to them.
Although there was a fairly lively debate at the time, in the technical press, about just what it was that was responsible for this reaction on the part of the user, several points soon emerged. Of these, the first was that- although the use of NFB could lessen the extent of the measured nonlinearity of the amplifier at high output powers -- because the gain of the system fell to near zero at the crossover point, even apparently very high amounts of NFB could not eliminate the high order harmonic distortion due to the kink in the cross over characteristic, and this kind of distortion would become progressively more conspicuous at low out put-power levels. This meant that a graph of total harmonic distortion (THD), versus output power would show the characteristic illustrated in FIG18a, with an increasing distortion level as the output power was reduced, rather than that of FIG18b, which was typical of most earlier valve operated amplifiers. Not only was this high level of small-signal distortion unsatisfactory for the majority of listeners, who would do the bulk of their listening at low power levels, but this type of defect would lead to the kind of signal input vs. signal output transfer characteristic shown in FIG19, where the relative gain of the amplifier was greatly reduced for input signal levels around the zero signal datum line. This had the predictable result that all small signals, such as the dying away sounds of instrument tones, would be sup pressed, along with a lot of other low signal level sounds, and amply explained the complaints of 'thinness' of tone.
FIG18 Comparison between distortion characteristics of valve vs. quasi-complementary transistor output stages
FIG19 Input/output transfer characteristics of incorrectly biased push-pull stage
Listener fatigue was probably a combination of the ear's objection to exposure to alien high-order harmonic distortions, due to crossover effects, and of the bad transient performance of the amplifiers, especially on 'awkward' types of LS load, where the injudicious use of too high a level of NFB, in the interests of a good apparent full output power THD figure, had eroded the stability margins of the amplifier and might have caused sporadic instability following sudden changes in signal level.
Improved transistor amplifier designs
Output circuit alternatives:
With the realization that much greater care must be exercised in the design of transistor amplifier output stages, if a performance was to be obtained which would compare with that given by earlier valve operated designs -- particularly in respect of low signal level distortion characteristics -- came a great amount of design activity aimed at remedying crossover errors.
Fully complementary output stages:
This took two routes; of these the first was that of improving the symmetry of the output stage by the use of more truly complementary output devices, which were beginning to become available from the semi conductor manufacturers in the mid- to late 1960s, of which a typical example is that due to Bailey (Bailey, A. R., Wireless World, May 1968, pp. 94-98), whose circuit is shown in FIG20. This approach was popular among designers in the USA where fully complementary power output transistors had first become readily available.
The Shaw improved quasi-complementary layout Alternatively, steps could be taken to improve the symmetry of the Lin quasi-complementary layout, and a circuit for this purpose was disclosed by Shaw (Shaw, I. M., Wireless World, June 1969, pp. 148 153), and is shown in FIG21. Two later improvements to Shaw's circuit, of which the first was proposed by Baxendall (Baxandall, P. J., Wireless World, Sept. 1969, pp. 416-417), and the second by the present author (Linsley Hood, J. L., Hi-Fi News and Record Review, Nov. 1972, pp. 2120-2123), are shown in FIGs 22a and 22b. Versions of these latter layouts are still in commercial use, in well thought-of amplifier designs.
FIG21 Shaw's improved quasi-complementary output stage
FIG20 Fully complementary output stage of Bailey 30W amplifier