Towards the end of the 1950s, transistors ceased to be merely an interesting laboratory curiosity, and emerged into the component catalogues as valid and usable devices, for a number of low power and low voltage applications. However, few people seriously considered them as possible competitors for tubes in higher power applications, or possible options for use in high quality audio systems. There were a number of good reasons for this, of which the major one was that transistors, at that time, mostly meant germanium PNP diffused junction devices, and these were very temperature sensitive, both in respect of the output (collector) current for a given forward bias, and in respect of the electrical characteristics of the device itself. This latter problem arose because, if the diffused junction regions within the transistor got sufficiently hot, say > 180degr. the diffusion processes, by which they were manufactured, would continue, and the internal junction boundaries would shift.
At this time, a germanium transistor would be made by taking a small, thin wafer of single crystal germanium containing trace quantities of arsenic or antimony, as an N-type impurity, and then heating the wafer so that two small pellets of indium- previously spot welded facing each other on opposite sides of the wafer- would cause P-type impurity diffusion zones to move in towards each other from opposite sides of the wafer, as shown in FIG. 1, to form a PNP device. Since the high frequency response of the device, its current gain and its breakdown voltage would depend on the thickness of the zone (the base junction) separating these two regions, if this changed during use it would cause the circuit performance to change also. (In principle, an NPN transistor could also be made by the same technique, and would, presumably, suffer from the same snags. However, neither arsenic nor antimony- the only practicable N-type electron donor materials - were found to behave very satisfactorily as alternatives to indium.)
FIG. 1 Diffused junction transistor
Because of their small size and low supply voltage requirements, germanium PNP devices offered, within the constraints imposed by their sensitivity to heat, a number of advantages for use in things like heating aids and portable radios. For reasons of economy in power consumption, and the need to avoid much heat dissipation in the output transistors, a typical circuit for a low power audio amplifier- having, say, an output power of 500mW - would use a circuit of the kind shown in FIG. 2, in which the output devices were operated in class AB, with a very low quiescent collector current. Inevitably, this arrangement led to a fairly high level of crossover distortion, and because the circuit used both driver and output transformers, it would be difficult to employ a useful amount of NFB to reduce its distortion, but perhaps, as the output stage for a portable radio, its performance would be thought to be adequate.
FIG. 2 Transformer coupled amp
However, the circuit of FIG. 2 is a very simple one, and much better designs are possible, as, for example, in the 15 watt power amplifier shown in FIG. 3, due to Mullard Ltd (Reference Manual of Transistor Circuits, 2nd edition, 1961, pp. 178-180). In this, by rearranging the circuit somewhat, it has proved possible to dispense with the output LS coupling transformer entirely, so that if the driver transformer is reasonably well designed, it will be possible to employ a small measure of overall NFB, from LS output to signal input, to reduce the distortion and make the frequency response of the circuit more uniform. The quoted performance for this design was: THD <4%, bandwidth 150Hz-7kHz, +1.5dB. The way that NFB is applied to the power amplifier, as shown, makes the overall gain dependent on the ratio of R7 to the external circuit impedance seen at Q1 base.
FIG. 3 15 watt audio amplifier
The Lin Circuit
As has been seen in the case of the tube audio amplifiers described above, the output transformer is a bulky and expensive component whose performance is likely to have a crucial effect on the overall performance of the amplifier design. Since the output impedance of a transistor, used as an emitter-follower, could well be less than an ohm, it should be possible to implement the basic audio amplifier structure shown in FIG. 1 (a variable gain voltage amplifier stage driving the loudspeaker via some means of impedance conversion) by the use of a pair of push-pull emitter-followers, rather than a transformer, as the output impedance-matching mechanism.
However, there was at this time an additional requirement- that these output emitter- followers should be based on output transistors of the same polarity. This need arose because although a few NPN transistors were available these were all small-signal types, and were quite unsuitable for use as one half of an output emitter-follower. All of these problems appeared to have been solved at a stroke by the ingenious circuit layout proposed by Lin (Lin, H.C., Electronics, pp. 173-175, September 1956), and is shown in FIG. 4. In this, the output emitter-followers were what Lin termed a quasi-complementary pair, in which the upper half (Q2,Q4) was a conventional Darlington pair and the lower half (Q3,Q5) was a compound emitter-follower. This allowed the output voltage, at C4, to follow the amplified signal voltage at the collector of Q1, but at a low enough impedance to drive a 16 ohm LS directly.
FIG. 4 Audio amplifier due to Lin
The performance of Lin's design (THD <1% at 400Hz at 6W output, bandwidth of 30Hz-15kHz, + 1.5dB), while not yet as good as could be obtained from a run of the mill tube audio amplifier, nevertheless offered a workable design option. As would be expected, the stability of the voltages and currents in this design was not very good, though Lin had employed DC NFB, via R2, in addition to the AC NFB through R9 and C5, to hold the collector voltage of Q1 to a suitable value. The forward bias applied to the output quasi-complementary emitter-followers, which would need to be about 0.2V, at 25ohm was provided by the voltage drop across R5, caused by the collector current of Q1, and an attempt was made to compensate for changes in the junction temperature of the output transistors by connecting a negative temperature coefficient thermistor (TH1) across R5.
As can be seen from FIG. 5, germanium junction transistors have a less abrupt turn-on characteristic than silicon ones, and since the output devices will probably run warm, their actual Vb/Ic graph is more likely to be that of the dashed line curve than the solid line, 20 degr. C one -- and this will act to lessen the magnitude of the potential crossover-type discontinuities which lurk within any push-pull system.
However, the somewhat unpredictable performance of germanium transistors of that time, together with their proneness to thermal runaway, discouraged audio amplifier manufacturers from making commercial designs of this type. This had to wait for another five years for the introduction of silicon transistors made by variants of the Fairchild planar process. These became available initially in small signal versions, and, because the manufacturing techniques favored this, in NPN (+ve rail) constructions. These allowed the design of high quality, low noise, low distortion, small-signal gain stages which needed no setting-up adjustments, and which, in my view at the time, were a great improvement, in terms of freedom from mains hum and microphony, on their thermionic tube predecessors.
FIG. 5 Turn-on characteristics
The implementation of audio power amplifier designs with predictable and stable performance characteristics demanded equally reliable and robust output transistors, which meant, in practice, those made using silicon planar construction, and when these became available, they were offered principally as NPN types, so it was using these in Lin-type quasi-complementary circuit arrangements that the first high fidelity solid- state audio amplifiers were made. Unfortunately, this approach led to a type of malfunction which was overlooked by the designers at the time, but which fairly soon became the subject of hostile comment from the users of this new Hi-Fi equipment, and this was the problem of output stage asymmetry.
Quasi-complementary Output Stage Asymmetry
This problem is illustrated in FIG. 6. The input base voltage vs. collector current relationship of a simple NPN/NPN Darlington pair based on silicon junction transistors is shown in the upper fight-hand quadrant of the drawing, and the equivalent characteristics of a silicon transistor compound emitter-follower are shown in the lower left-hand quadrant. Not only are these curves different in slope - which makes the push-pull transfer characteristics asymmetrical, even when operated at the optimum quiescent bias, as shown in the diagram 'a' - but, since the output impedance of an emitter-follower is approximately 1/gm, and the slope of the curve in the lower LH quadrant is markedly steeper than that of the upper RH, the output impedance will be different as well; a factor which will become apparent on low output impedance loads, such as, for example, LS driver units at some parts of their frequency response curve.
FIG. 6 Output pair asymmetry
FIG. 7 Comparative THD curves for transistor and tube amplifiers
Inevitably, this asymmetry in the transfer curves of the two halves of the output stage leads to a degree of residual crossover distortion, which is worsened if the chosen quiescent current setting (the choice of which will always be a matter of some compromise, because what would be the best setting for one half of the output pair would not be the best for the other) is not the optimum value. Since the circuit layout of the amplifier is likely to be fairly simple, with a limited number of phase-shifting elements, it is possible to use a large amount of NFB to reduce the measured full output power total harmonic distortion (THD) level. Since it is inconvenient, on a production assembly line, to have to adjust the output quiescent currents of each half of a stereo amplifier, some designers - having noted that, in an amplifier using a high degree of NFB, the actual quiescent current setting which was chosen made relatively little difference to the measured value of the residual full output power THD - opted either to use some fixed value which was rather less than the optimum, or to use no forward bias at all. This could lead to the situation shown in FIG. 7, in which the THD could be quite low at the rated power output, but would worsen as the output power level was reduced.
I have shown a typical solid state audio amplifier design of this period, the middle 1960s, in FIG. 8. The circuit I have chosen is that of the Leak Stereo/Delta 70. This is a typical example of design thinking of that time, and most of the contemporary Hi- Fi shops would have similar designs from a wide range of audio amplifier manufacturers on their shelves.
FIG. 8 Leak Delta 70 amplifier
Listener Fatigue
It was fairly generally accepted that there was a difference in the sound quality given by the new transistor amplifiers and that of the tube amplifiers that they sought to supplant, and this quickly led to the emergence of two camps in the Hi-Fi field, those who liked the new sound, and those who rejected it, and described the tonal quality as hard, or thin or clinical. On the other side were those who argued that, since the output power available was now greater, and the full output power bandwidth and distortion figures were both better than those of earlier systems, what the listener was now heating was the actuality of the music, and not some rounded-off version, all of whose rough edges had been removed by the inadequacies of the amplifier output transformer. The anti solid state protagonists retaliated by describing the new technology as giving 'transistor sound' and complaining that it caused listener fatigue.
In reality, the new solid state amplifiers suffered from a number of shortcomings, which were largely overlooked by the design engineers because they occurred in areas which had not previously been regions of concern. The first of these, already mentioned, was that of the dependence of the distortion level on the output power. In tube designs operating in class A (that condition in which the amplifying devices are conducting for the whole of the signal voltage excursion) it could be taken for granted that the worst output THD figure would occur just below the onset of clipping, and this would decrease, ultimately disappearing into the noise background, as the output power decreased. However, as had been seen in FIG. 7, a solid state amplifier operating under zero forward bias (class B) conditions would have a distortion figure which would worsen as the output power was reduced.
An additional factor is also shown in FIG. 7, in respect of the output power available before the onset of signal peak clipping. A tube amplifier which used only a modest amount of NFB, say 15dB or less, would have a distortion figure which would worsen only gradually as it was driven into overload, and if the listener was prepared to accept a moderate level of peak clipping, the tube amplifier could actually sound louder than the apparently higher powered transistor version. The relatively soft clipping of the traditional tube amplifier, when driven into overload, is one of the more highly valued characteristics of this system, in the view of some contemporary users. This is mainly a feature of the low level of NFB which is used in some such designs, coupled with their freedom from latch up, a frequent feature of badly designed solid state amplifiers.
The use of large amounts of NFB to reduce the apparent distortion level of the amplifier, especially under class B operation, leads to two further problems, of which the first is that when the output transistors are cut off, the system gain is zero and consequently the amount of NFB applied through the feedback loop is also zero. In practice, this means that if the signal voltage swing takes it through the zero voltage axis it will pass into a dead zone, beyond which the full amplifier gain will operate to urge the voltage swing across to the opposite conduction region. However, in the dead zone the amplifier is switched off, and any low level signals which are present in this region will be lost - thus justifying the allegations of the thinness of tone of the amplifier, a characteristic feature of zero-biased or under-biased output stage operation.
In the circuit design of FIG. 8, the maker' s recommended quiescent current setting is 30mA, although, in practice, experience would suggest that the optimum current for the upper Darlington pair will be of the order of 80-100mA, and the lower compound emitter-follower stage would be optimally biased at an Iq of some 40-50mA. Using a higher bias current setting than the optimum would, as shown in diagram of FIG. 9, lead to a worsening of the THD above some low power level, determined by the actual Iq setting, in exchange for a substantial improvement in the distortion at very low output power levels as the output stages effectively returned to class A operation. Such an amplifier performance might look worse on the specification sheet, but could be more pleasant to listen to.
Fig. 9 Effect on THD of increased value of quiescent current. over-biased
transistor pair; Power output (watts)
The second incipient problem in solid state amplifier designs of this period was that of inadequate stability margins in the feedback loop. Like most of the other problems, this was worsened by the lack of symmetry of the output stage, in that not only were the dynamic (amplitude and rate of change related) electrical characteristics of the transistor itself frequency, temperature and current dependent, but they would also vary depending on which of the two output emitter-follower groups was conducting.
To add to this complex mix of difficulties, the LS load which was coupled to the amplifier had a reactance which was continuously variable, dependent on the frequency and amplitude levels of the input signal. All of this presented a substantially greater challenge to the loop stability of the amplifier than that offered by the conventional resistive dummy load- so that amplifiers which behaved quite stably on the test bench might well pass through regions of instability under live conditions with an LS load, which would lead to the occurrence of brief bursts of HF oscillation buried, but not hidden from the ears of the listener, within the signal.
Design engineers working in this field in the mid-1960s were acutely aware of the need for some improvement on the type of performance given by the standard quasi-complementary (Q/C) output pair, and a number of options were explored with this aim in mind.
Alternative Circuit Choices
In the absence of PNP power transistors, or, when such devices had become available, obtainable only in relatively low voltage and low power versions, there were two possible options open to the designer - to improve the symmetry of the quasi- complementary pair circuit, so that NPN output power transistors could be used exclusively without audible performance penalties, or to bias the output emitter- followers so that they operated in class A, in which condition the crossover distortion would be greatly reduced. The first of these routes was chosen by the Acoustical Manufacturing Company in their Quad 303 power amplifier, in which they elaborated the two-transistor quasi-complementary pair into a triplet, as shown in FIG. 10a (report, Wireless World, April 1968, p. 67). In this layout, an almost exact symmetry of the Vin/Iout curves was obtained, though there was still some small difference between the two triples in respect of the optimum quiescent current, of which the mean value was only, in any case, about 4mA. This low value of optimum Iq led to the minor drawback that it did not allow any significant margin of operation m class A, which would act as a cushion if there were unexpected variations in the operating conditions or device characteristics.
FIG. 10 Improved Q/C output stages -- (a) Quad output triples; (b) JLH output
triples
FIG. 11 Improved Q/C output stages -- (a) Shaw's circuit (b) Baxandall's version
(c) JLH 75W amp output stage
At this time the design of a high power, high quality audio amplifier presented an interesting technical challenge, in the absence of any high voltage, high power PNP transistors which could be used in conjunction with NPN power transistors in an output stage having complementary symmetry. The general philosophy used by Quad seemed to offer an answer to this problem, and I have shown in FIG. 10b a layout for a Q/C triplet which I tested for use in a high power amplifier. Under DC or LF conditions, the two halves of this triple were virtually identical, and the optimum quiescent current (--100mA) was also the same for both emitter-follower groups. Used in the output stage of an experimental amplifier design, this output configuration gave a distortion figure at low output power levels which was less than my then ability to measure it- at the time my test bench THD meter had a lower measurement limit of about 0.05% over the range 100Hz-10kHz - and the amplifier did not appear to exceed this threshold value over the output power range from 10mW to 30 watts.
Another method for improving the symmetry of the output stage was suggested by Shaw (Shaw, I.M., Wireless World, June 1969, pp. 265-266), and shown in FIG. 1 l a. Baxandall did an analysis of this layout (Baxandall, P.J., Wireless World, September 1969, pp. 416--417) and proposed a rather more straightforward way of achieving the same end, using the circuit layout shown in FIG. 11 b. Unfortunately he did not extend his analysis to show a fully worked out amplifier design based on his analysis. I was attracted by the simplicity of this approach, in which the diode, D 1, simulated the effect on the driver transistor of the base-emitter junction of the lower output transistor. The effect of this diode in imitating the missing output transistor junction could be improved, especially at higher frequencies, by adding a capacitor, C 1, across this diode to simulate the output transistor forward junction capacitance, as shown in FIG. 11c. I adopted this circuit for the output layout of a 75 watt amplifier design, eventually published in Hi-Fi News (Hi-Fi News and Record Review, November 1972, pp. 2120-2123) in what became a very popular constructional project. It is easy for an author to think favorably of his own designs, but my personal feeling, then and now, was that with this design, and others of similar quality which were then being offered, junction transistor audio amplifiers had come of age, and that their users need not feel that something had been lost for ever with the passing of thermionic tube operated designs. I have, for the record, shown the circuit of this 75 watt amplifier in FIG. 12.
Although there are one or two innovations in this circuit, the design is fairly straightforward, and consists of an input long-tailed pair stage, with a junction FET used to provide a very high dynamic impedance constant current source tail to improve the emitter signal transfer between Q1 and Q4. PNP transistors are used in this stage so that Q5, the main voltage amplifying stage, could be a high voltage NPN device with good HF response. The stage gain was increased by the use of a DC bootstrap circuit (Q3,R3,R6) as the load for Q1. This also gives a low drive impedance for Q5, which also helps to maintain the stage gain. The voltage drop which is developed across an amplified diode, Q6, due to Q5 collector current, is used to provide the forward bias (about 3V) needed to make the output transistor groups operate at the best point on their combined push-pull transfer characteristics. Q6 is mounted on the output transistor heat sink to provide a measure of thermal compensation for the quiescent operating current, and helps to maintain this at the desired level ( -100mA). A bootstrapped load resistor (C8, R13) is used to increase the dynamic impedance of R13, the collector load of Q5. The operation of the Shaw/Baxandall technique used to increase the symmetry of the output Q/C transistor layout has already been described and illustrated in FIG. 11. HF stability for all likely combinations of reactive loudspeaker loads is ensured by the main, dominant lag capacitor, C9, connected between Q5 collector and Q4 base- in which position it does not contribute to slew rate limiting or slewing induced distortion, an immunity which is assisted by the input low pass network R2/C2. Since a large amount of NFB (approximately 46dB) is employed to maintain a very low level of distortion over the whole available output power range, the feedback loop characteristics are tailored by the HF step networks R9/C6, R3/C3, R4/C4 and the output Zobel network C14/R31 so that the loop phase characteristics are satisfactory. Typical performance figures for the design shown in FIG. 12 are: output power 75 watts into an 8 ohm load, bandwidth 15Hz-20kHz, (upper end set by R2/C2), THD <0.01% at all power levels below the onset of clipping, unconditionally stable into all combinations of load impedance or reactance.
FIG. 12 75 watt amp
Class A Operation
The other option which was open to the circuit designer, even in the absence of satisfactory PNP power transistors, was to operate the amplifier in class A, a possibility which, as shown in FIG. 13, could be realised using only NPN polarity output devices. Since this is not a push-pull layout, crossover distortion cannot occur, but, since it is not a push-pull system, the output power available is limited, as in any other single ended layout, by the choice of the output stage operating current, and this, in turn, is limited by the permissible thermal dissipation of the output transistors. With reasonably efficient loudspeaker units, the bulk of normal listening would take place at output power levels which did not exceed a watt or two and the possible output power from such a class A system would be entirely adequate.
I had designed and built this amplifier for my own use, like all of my audio circuit designs up to that time, and I only offered it for publication because the use of output transistors in class A had become, at that time, a matter of topical interest, principally because a commercial amplifier using this principle, made by J.E. Sugden Ltd, had attracted very favorable reviews in the Hi-Fi press, who applauded its freedom from transistor sound.
The structure of the circuit shown in FIG. 13 is very simple, with Q1 acting as a grounded emitter amplifier stage, with Q2 as an active collector load, driven in phase opposition to Q1 by Q3. The loop gain of the amplifier is increased by bootstrapping the load resistor for Q3 by C1. Because the transition frequency of the output transistors is of the order of 4MHz, whereas those of Q3 and Q4 are in the 400MHz range, the circuit has an in-built dominant lag in its loop NFB characteristics. This ensures that the loop gain has fallen below unity before the loop phase angle reaches 180 degr. No additional HF compensation networks are therefore necessary to ensure complete loop stability, even with reactive loads.
FIG. 13 JLH 10 watt class 'A' WW 4/69
Fully Complementary Designs
With the continuing development of epitaxial base and similar structures for silicon transistors, PNP power transistors became more readily obtainable, although initially in relatively limited voltage ratings, at prices which approached those of existing NPN power devices. This provided an incentive to the circuit designers to provide amplifier systems which took advantage of this new technology, and offered the possibility of reducing low signal level crossover distortion to a level where it would no longer be audibly detectable. Two of the circuit designs which made use of this new-found freedom were due to Locanthi (Locanthi, B. N., dr. Audio Eng. Soc., July 1967, pp. 290--294) and Bailey (Bailey, A.R., Wireless World, May 1968, pp. 94-98). Of these, the Bailey circuit offered a somewhat lower level of THD and I have shown the circuit used in FIG. 14.
Although NPN and PNP power transistors were nominally exact equivalents of one another, in reality there were significant differences between these structures which lessened the symmetry of the amplifier circuits built around them. Of these differences, the most obvious was that the current carrying majority carriers were electrons, in the case of the NPN devices, and holes in the case of the PNP ones, and since electrons have greater mobility, performance differences show up at higher frequencies. The second difference, due to the nature of the emitter/base diffusion interface is that, although the makers quote identical safe operating area (SOAR) curves, PNP power transistors are nevertheless more prone to failure in use than NPN ones.
There are a number of interesting design features in the circuit of FIG. 14, of which the most important, in terms of its influence on subsequent designs, was the output overload protection circuitry arranged around transistors Q5 and Q8. These are arranged to monitor both the output current from Q7 and Q10 (by means of the voltage drop across R25 and R26) and also the voltage present across Q7 and Q10. If the combination of voltage and output current is such to approach the secondary breakdown region of the maximum working limits of the device (see, for example, FIG. 12), Q5 and Q8 will conduct, and limit the drive voltage applied to the bases of Q6 and Q9.
Bailey had also configured the circuit so that it operated between a symmetrical pair of voltage rails- anticipating the circuit configuration used in the so-called direct coupled layouts which subsequently became very popular (see, for example, FIG. 12). However, he had chosen to retain a simple grounded base input amplifying stage, of which the inevitable Vb_ e DC offset of Q1 was lessened by a bias current derived from the amplified diode circuit built around Q2, rather than the more straightforward input long-tailed pair layout of FIG. 8a. Since the circuit of FIG. 14 gave a DC output voltage offset under no-signal conditions, a reversible (AC working) electrolytic capacitor was required to isolate the loudspeaker from the amplifier output. Such capacitors are not commonly used, and are therefore expensive.
FIG. 14 Bailey 30W amplifier
In the article in Wireless World in which he described this design, Bailey demonstrated, by oscilloscope traces of the relative Vin/Iout transfer characteristics, the fundamental lack of symmetry of the existing and widely used simple quasi- complementary layout, and showed the superiority of the fully complementary design in the way the harmonic distortion of the amplifier progressively decreased towards zero as the output power was reduced- behavior which was typical of tube amplifiers operating in class A, but not found in most of the early class B (low or zero operating quiescent current) transistor designs. Bailey also showed the performance of his design when driven with a square-wave input signal and coupled to resistive or reactive loads. A lack of overshoot in, or significant distortion of, a square wave or similar type of signal, when the amplifier is caused to drive a reactive load, makes, I believe, an important contribution to good overall sound quality in an audio amplifier.
Gain Stage Designs
The gain stages between the signal input point and the output devices are normally operated in class A and are arranged to give as wide a bandwidth, as high a gain and as low a phase shift between input and output as is possible. To reduce the difficulty in keeping the final amplifier stable, when overall NFB is applied, the gain block is normally restricted to two amplifying stages, and to get as high a gain from these stages as is practicable, the collector load for the second stage will be arranged to have a high dynamic impedance. In the amplifier designs shown so far ( FIGs 4, 8, 12-14), this increase in the AC impedance for a given DC resistance has been achieved by bootstrapping the load resistor (by coupling its supply-line end by a capacitor connected to the output of the amplifier). In addition to increasing the AC impedance of the load resistor this also has the practical effect of increasing the possible output voltage swing which that stage can deliver. However, this technique is essentially that of applying positive feedback around the output stage. When this is an emitter-follower stage, or some similar arrangement, the gain will be less than unity and the amplifier will not be unstable. On the other hand, positive feedback has the effect of increasing both the stage gain and the distortion of the stage across which it is applied. In the present context, if that stage suffers from crossover distortion, the ill effects of this will be magnified by a drop in the dynamic impedance of the load resistor and a reduction in the driver stage gain at the crossover point. Modem design practice therefore tends to favor a high dynamic impedance load, such as a constant current source or the output from a current mirror, as the means of optimizing driver stage gain. Typical arrangements of this kind are shown in FIGs 15a-15c.
The layout of FIG. 15a is a fairly typical long-tailed pair input stage, in which a constant current source, of one of the types shown in FIGs 17 or 24, has been used as the tail in order to assure the integrity of the signal transfer between the emitters of Q 1 and Q2. This approach is favored in IC manufacture where it is easier, and less expensive of chip area, to manufacture an active device than a resistor- especially one of high value. Similarly, the gain of Q3 could be increased by replacing R2 with a further constant current source.
FIG. 15 Typical gain stages
In the modification shown in FIG. 15b, a current mirror, such as one or other of the types shown in FIG. 25, has been used to combine the outputs of the two transistors of the long-tailed pair, which will substantially increase the input stage gain. Once again a constant current source could be used in place of R1, with a further increase in stage gain. As modified in this way layouts of the kind shown in FIG. 15b form the basic structure of the bulk of both operational amplifier circuits (because it does not need any resistors) and of a large proportion of Hi-Fi audio amplifier circuitry.
In all of these layouts the polarity of the devices could be reversed (i.e. by substituting NPN for PNP devices, and vice versa) and other types of transistor, such as JFETs or MOSFETs, could be used, at the choice of the designer. Similarly, the gain can be further increased, especially at the higher end of the frequency band, by connecting a cascode transistor, of one of the forms such as are shown in the layouts of FIGs 7 or 15, between its collector and the collector load.
An interesting further development of this idea is shown in FIG. 15c, in which the current mirror, used to combine the outputs of the two antiphase signal streams, is transferred to form the collector load of the second gain stage transistor, Q3. This idea appears to have been originated by National Semiconductors, and has been used in several of their IC designs, such as the LH0061. It has also been adopted by Hitachi as the basis of an audio amplifier design (Hitachi Ltd, Power Mosfet Application Manual (1981), pp. 110-115), having the desirable qualities of symmetry and a very high gain from just two stages.
HF Compensation Techniques
This is the somewhat misleading term which is given to the adjustment of the amplifier gain and phase characteristics, as a function of frequency, so that when overall loop feedback is applied the amplifier remains stable - ideally with a wide margin in terms of the gain or the phase angle which exists between the working condition of the amplifier and the onset of instability. While a two gain-stage amplifier of the kind shown in FIG. 15 would most probably be stable if an NFB signal was returned from the output to Q2 base, by way of some suitable network, if a push-pull output emitter-follower pair, of the kind shown in FIGs 12 or 14, were to be interposed in the feedback path, the loop phase shift would approach 180 degr. at some upper or lower frequency at which the loop gain was equal to, or exceeded, unity, and the amplifier would oscillate.
With direct-coupled circuits the LF phase shift will not exceed a safe level, so the problems of loop instability are confined to the HF end of the signal pass-band, and it was (and is) customary to achieve the necessary HF loop stability by imposing a single-pole, dominant lag characteristic on the gain/phase relationships of the system by connecting a small capacitor (Cfb) between the collector and base of the second amplifying transistor (Q3) in FIG. 16a, since this arrangement gives the best performance, in terms of THD, at the high frequency end of the pass-band. However, this approach leads to the problem that it imposes a finite speed of response to any rapidly changing input signal while Cfb charges or discharges through its associated base or collector circuits - mainly being limited by the collector current of Q1. This effect is illustrated in FIG. 27.
If a composite input signal which includes some rapid rate of change of input voltage level is applied to the input of the amplifier it is possible that the input device (Q 1) will be driven into cut-off or saturation because no compensating feedback signal has yet arrived at the base of Q2. If this happens, there will be a complete loss of signal during this period because Q3 will be paralyzed while the charge on Cfb is returning towards its normal level. This problem was described by Otala in a published paper (Otala, M.J., J. Audio Eng. Soc., 1972, No. 6, pp. 396-399), and he coined the term Transient Intermodulation Distortion to describe the audible effects of this type of malfunction. A simpler description, suggested by Jung (Jung, W.C., Hi-Fi News and Record Review, November 1977, pp. 115-123), is slewing-induced distortion (or slew rate limiting), and this effect can be seen clearly on an oscilloscope connected to the output of an amplifier when a suitable input signal is applied.
FIG. 16 HF compensation methods
This type of problem is not an inevitable consequence of dominant-lag type HF compensation since there are ways of avoiding it ( Hi-Fi News and Record Review, January 1978, pp. 81-83). Of these, the simplest is just to introduce an RC low pass network at the input of the amplifier to limit the possible rate of change of the input voltage - as R1/C 1 in FIG. 16. A better approach is to include the whole of the amplifier gain stages within the bandwidth limiting system, as used, for example, by Bailey (C3 in FIG. 14), and illustrated in FIG. 16b. Then, provided that the possible rate of change of the collector voltage of Q3 (which is determined by its collector current and the circuit capacitances associated with its collector) is faster than the rate of change permitted by R1/C1 (which is within the control of the circuit designer) slew rate limiting will not occur.
Although the conventional scheme shown in FIG. 16a is better from the point of view of THD in the 10kHz-20kHz part of the pass-band, it is much inferior in respect of the normal square wave into reactive load type of test, and it has always seemed unwise to me to choose a design approach in which an inaudible improvement--from 0.1% to 0.02% THD at 20kHz- has been bought at the cost of worsening the (almost certainly audible) transient error from, say, 2% to 40%, as measured, for example, at 10kHz in the manner shown in FIG. 17. In this, the amplifier, operating with a simulated reactive load, is fed with a good quality square wave, and an amplitude and time-delay corrected square waveform, derived from this, is subtracted from the amplifier output. Ideally the residue should be zero, and the closer the amplifier output approximates to its input waveform under reactive load conditions, the better it will probably sound.
Amplifier under test
FIG. 17 Square wave generator; Comparator; Output Load; Amplitude and delay
adjust
Symmetry in Circuit Layout and Slewing Rates
A residual problem with any system in which there is stray, or other circuit capacitances, is that the maximum possible slewing rates will not necessarily be the same for a negative-going or a positive-going signal excursion. This is because the outputs of circuits do not necessarily have the same ability to source or sink current, so there must inevitably be differences in the rate in which any associated capacitances can be charged or discharged. To take the case of the very simple amplifier layout shown in FIG. 15a, if there was some capacitance between its output and the 0V line this capacitance could be discharged very rapidly if Q3 were turned fully on, but would, perhaps, only charge up again, towards the positive line, at a slower rate, which would depend on the value of R2. This problem would be worsened if a constant current source were used instead of a resistor, as in the, apparently much preferable, circuit of FIG. 16a. This prompted some designers, such as Bonjiomo (Audio, February 1974, pp. 47-51) and Borbely (Audio Amateur, February 1984, pp. 13-24), to propose circuits of the form shown in FIG. 18, in which two amplifier blocks of the kind shown in FIG. 16a are coupled together as a mirror-image pair. The only drawback with this type of layout is that there is normally rather greater difficulty in achieving a stable quiescent current in the output transistors- a thing which is very desirable in any class AB output stage for optimally low levels of crossover distortion.
Stability of Output Current
By the early 1970s audio amplifiers operating in class AB -- by which I mean those in which it is intended that a small residual current will pass through the output devices under zero signal conditions - had mainly achieved very high standards of performance, although the one remaining disadvantage which they all shared was that the preferred setting of the quiescent current could be quite critical, and would need to be set up on the test bench for each amplifier as the final step of its assembly.
Moreover, there was no guarantee that this Iq setting, when correctly adjusted (ideally while monitoring the output with an oscilloscope and a distortion meter), would remain at the chosen value, or even that this chosen value would still be the correct one, during the aging of the circuit components, or as the ambient temperature of the amplifier changed.
FIG. 18 Fully symmetrical gain stage
A number of design proposals were offered as a means for ensuring the stability of the quiescent current, but, in general, these all suffered from disabling flaws in their design, so that, in practical terms, the designers were left with the options of trying to ensure quiescent current stability in the face of operating changes or to choose a design approach in which the actual quiescent current value was not particularly critical. An example of the latter approach was my 15-20 watt class AB design of 1970 ( Wireless World, July 1970, pp. 321-324). Various circuit arrangements had been adopted to minimize the effect of changes in the temperature of the output devices, but one bold approach, due to Blomley (Blomley, P., Wireless World, February 1971, pp. 57-61 and March 1971, pp. 127-131), is shown, in a slightly simplified form, in FIG. 19. In this the output devices, a complementary pair of silicon planar transistors, are permanently biased into conduction, and the input signal, after amplification, is chopped into two halves by a pair of switching transistors (Q1 and Q2), and these halves are then passed to the output transistor triples for reassembly into an enlarged and power-augmented version of the original signal. The snag, of course, is that a correct forward bias must now be chosen for the small-signal switching transistors, which merely moves the problem of choosing and maintaining the operating current away from the output transistors and back to the earlier switching stage.
FIG. 19 The Blomley amplifier
A typical example of bipolar transistor operated audio amplifier design of the mid- 1970s, incorporating many of the contemporary design features, is shown in FIG. 20. This has an excellent performance both in terms of its THD (better than 0.01% at 50 watts output over the frequency range 30Hz-15kHz) and square wave into reactive load (no significant overshoot when coupled to an 8 ohm load in parallel with capacitor values from 1nF to 41nF). Q2 and Q3 form a constant current source for the input long-tailed pair (Q1/Q4) with a preset potentiometer (RV1) connected between its emitters to allow the DC offset at the IS output terminals to be reduced to a very low level. Q5 and Q6 form a current mirror load for the input stage and Q10 acts to protect Q7 from an excessive input drive signal.
Q8 and Q9 form a constant current load for the main gain stage transistor (Q7) which drives the output devices on either side of the forward bias generating network (RV2/C7) through hang-up prevention resistors (R1 l/R12). The output transistors are connected as compound emitter-followers because this arrangement gives an output which more nearly equals the input (because of its higher loop gain) and because it offers a lower output impedance. This arrangement also has the advantage that the base-emitter junctions (which will become hot in use) are not directly involved in determining the best forward bias setting - this depends on Q11 and Q13 which have a much lower output current level. Q9 senses the ambient temperature, and adjusts Q7 collector current and the voltage drop across RV2 as required.
Output overload protection could be by means of an output fuse, as shown, or an output cut-out relay, or a current limited power supply. The use of Bailey-type output protection circuitry had fallen from favor at that period because it was thought to operate prematurely during high level signals, especially at higher audio frequencies where the impedance of many commercial LS systems may fall to a low level, and be seen as an apparent output short-circuit.
FIG. 20 1970s 60 watt amplifier
Although there were still areas in which improvements could be made, a comparison between the performance given by the late 1950s design of FIG. 3 and that of the mid-1970s design shown in FIG. 20 indicates the extent of the progress made.