Class A amplifiers always have had a good reputation among audiophiles. When talking about low power designs these are in most cases designed as class A. Talking about audio power amplifiers, the picture is quite different. Here class B and A/B amplifiers play a dominant role, while class A is more seldom. In this article we will have a look at the different types.
To explain the difference between class A, B and A/B, it is easiest to look at the current amplifier itself. In figure 1 it is shown an example of a typical version with complementary transistors.
The transistors Q1 and Q2 make the drivers, while Q3 and Q4 make the power transistors. The resistors R1 and R2 determines the quiescent point for the drivers, while the resistors R3 and R4 prevent thermal ’runaway’ for the quiescent point IQ for the power transistors. The bias voltage VBIAS thus makes the quiescent current IQ. For class A operation nor Q3 or Q4 should be switched off at maximum voltage swing to the load RL.
Normally the quiescent current is calculated with the maximum continuous output power P in mind. Minimum quiescent current therefore will be found as IQ = sqrt(P/2RL).
For P = 25 W RMS and RL = 8 ohm, IQ = 1.25 A is needed. This is half of the maximum current delivered to the load. Similarly, for 100 W RMS the demand is a quiescent current of minimum IQ = 2.5 A. The voltage swing at the load is a minimum of 20 V peak and 40 V peak, respectively. The power supply voltage should be chosen somewhat higher than this to allow voltage drop over transistors and resistors.
When the loss in the drivers and the resistors not are taken into account, the power efficiency is 50 %. For 25 W RMS continuous output power, Q3 and Q4 accordingly dissipate 50 W, while these transistors for 100 W RMS continuous output power dissipate 200 W. Notice that the power efficiency for this push pull output stage is the double of the power efficiency of a 'single ended' (constant current) output stage.
Now a number of 50 % for the power efficiency does not sound that bad, but it is calculated at full continuous output power. A music signal would be very boring if it consisted of one single tone and was played at maximum output level. In fact the power efficiency is much lower. It will depend on the load, the content of the music signal and the applied volume. The power loss however will be the same.
Consider the example with the 100 W amplifier, where the output transistors alone dissipate more than 200 W. This requires large heat sinks. Using a conventional heat sink as an example, having an effective thermal resistance of 0.25 W/K, the temperature will rise with more than 50 degrees above ambient temperature. Adding the fact that heat sinks not exactly are cheap, we already have two reasons to avoid class A drive.
Additionally the high quiescent current makes large demands on the power supply. In order to keep the ripple on a reasonably low level, both the transformer and the electrolytic capacitors must be in the ‘Rolls Royce’ class. Since the heat dissipation is so high, it is not simply sufficient to choose a high value for the capacitors, they should not exsiccate the first few years. From the foregoing, it should be obvious that the choice of class A is not taken to save money.
By reducing the biasing (VBIAS) to a level where IQ is less than half of the peak current to the load, the amplifier will work in class A/B. If VBIAS is further reduced, we will reach a point that may be defined as class B. In the literature you may see that IQ = 0 is defined as class B. This however tastes more off class C, since there is no opinion of the level of the signal that is needed to get Q3 and Q4 into conduction.
Normally one will set IQ to a few mA. It is an optimum setting of VBIAS here. The simplest way to find this is to force a sine wave signal to the input and monitor the output spectrum with a spectrum analyzer. The optimum setting of VBIAS is when this spectrum has a minimum of harmonic components. The optimum setting of VBIAS may also be done with an oscilloscope. The output signal should not show any crossover effects, but this method is not that useful, as it relies on pure visual inspection.
Both if the quiescent current is too low or too high around this optimum class B point, the harmonic components will increase. Thus it is good reasons to keep this optimum setting of the quiescent point. In practice it is difficult to retain this value, this is due to among other factors thermal fluctuation. If IQ is too high, this corresponds to class A/B. Thus it is not necessarily true that class A/B is better than class B, but more on his later on.
Now, for non-class A amplifiers there are cross over distortion; the transition between the power transistors is a non-linear process. In contrast to what sometimes is claimed, the current to the load is not sinusoid (for a sinusoid input signal) even if the power transistors are exact complementary. In addition we will have switching distortion stemming from the fact that the transistors have to be switched on and off.
For power amplifiers this kind of distortion is severe, due to the fact that large charges shall be moved in a limited time. The effect of cross over and switching distortion is that a large number of unwanted harmonic components are produced, among others of odd order. Since the latter are perceived by the ear as signal clipping, it should be unnecessary to say that the result is not quite kind to your listening experience.
The simplest way to reduce the distortion is to make use of negative feedback in one or another way. The unwanted harmonic components may be attenuated by a typical factor of 20-30 dB when in the audio band. The method has its limitations, since the relative level of these harmonic components will increase with the frequency as the feedback factor normally is reduced with frequency. In addition even harmonic components are attenuated in a similar manner, since the feedback itself do not alter the distribution between more well sounding even harmonic components and odd harmonics.
There are a few other methods to reduce the distortion. Here we will look only at a couple of them. Firstly, the level of unwanted harmonic components stemming from the switching distortion may be reduced by a simple modification to figure 1, see figure 2.
Compared to figure 1, it is observed that Q1 and Q2 not automatically are switched off if Q3 and Q4 are. Now there is a way to discharge the power transistors. Since the switching distortion stems from the transistors being switched off, it is worth trying to avoid this. This takes us to so-called ’Non Switching’ power amplifiers. The description is considerable more sober than ’New Class A’, ’Optical Class A’ or whatsoever.
An often-used method is to measure the current through the power transistors, directly or indirectly. If the current becomes too low, the bias voltage VBIAS is increased. The method is applicable if the current delivered to the load is not too high. It is, however, a possibility that this method has audible adverse effect, since the bias voltage is sliding as a function of the music signal itself, even with a small time delay.
Other methods rest on the use of other biasing methods than shown in figure 1 and 2. However, these are considerable more rare in audio power amplifiers.
When we now have looked at methods to reduce the switching distortion for non-class A amplifiers, is it not anything to do with the cross-over distortion? For class B amplifiers one may avoid the cross over distortion by not using the complementary pair. Instead one has to fight ’only’ with the switching distortion. For the biasing method in figure 1 and 2 it is not possible to eliminate the cross over distortion unless operating in class A.
By increasing the quiescent current considerably from class B, however, the cross over distortion can be moved to arise at higher power levels. A class A/B amplifier at 100 W RMS can be set to operating at a quiescent current of e.g. 1.25 A, corresponding to a class A output power at 25 W RMS. Now most of the time the amplifier would operate in class A. The cross over distortion now would occur at a higher power level.
The fact that the distortion occurs at higher levels is interpreted by our hearing as more natural. Secondly there would be no cross over or switching distortion at lower power levels, where our hearing is most sensitive. A distortion increasing at lower levels is not favored by our hearing. In this way, a class B amplifier (with switching and cross over distortion) reminds of a CD player. This could be a good explanation to why a CD player with an output stage made up of valves, despite their high content of harmonic components, has a good reputation for its ’musicality’
It should appear from the previous section that class B and A/B amplifiers with their amendment of ill sounding harmonics, stemming from the use of the transistors used as switches, not can be made to give the same sonic quality as class A. Additionally the harmonic distortion would be higher even when the transistors are working in the normal active range. This is due to the fact that the excess from the quiescent point would be large even for small output levels. To cope with this problem, some designers are using a current amplifier as shown in figure 3a.
Optimal quiescent point is easier to maintain with this compound emitter follower, since there is only to base-emitter junctions instead of four. The distortion also is lower, since it in principle is two common emitter stages with feedback. The switching distortion is however high, since the drivers are switched off when the power transistors are. Another draw back with this circuit is that parasitic oscillations may be a problem, especially with complex loads. With fast transistors this coupling has a tendency to oscillate with ease. After trying it, it is not my first choice even for class A amplifiers, where there is little to gain by using fast transistors.
The modified version shown in figure 3b, retain roughly speaking the good properties to the previous circuit. The distortion is somewhat higher, but the stability is better owing to local feedback of the drivers. Besides the drivers are not switched off automatically when the power transistors are. Due to good thermal stability, relatively low distortion and relatively simple construction, this current amplifier is a good choice for use in a class A power amplifier.
It is a widespread misunderstanding that class A amplifiers need oversized power supplies compared to class B and A/B amplifiers. It is however easy to undersize the power supply of especially class B amplifiers. These consume very little current without signal, giving little ripple. When applying signal to the amplifier, the case is dramatically changed, in oppose to class A amplifiers. Operating in class B, the current draw is nearly half rectified sinusoid at sinusoid input signal. In class A amplifiers, the current draw is a sinusoid when the input signal is sinusoid (i.e. a constant mean current is drawn from the power supply).
If the two channels are supplied from a common supply, the affairs are worsened by cross talk between the two channels. Global feedback (the output signal must be compared to a signal that is not linked to the power supply voltages) may attenuate some of the resulting harmonics and preventing them from reaching the loudspeaker. A further improvement may be obtained by making the supply represent low impedance seen from the amplifier, e.g. by using large capacitance values or by using a heavy regulated supply.
But the situation is worsened if the voltage amplifier shares the power supply with the current amplifier. A remedy is to construct the voltage amplifier to have a very high PSRR. Additionally the power supply to the voltage amplifier at least should be RC-filtered. The best result, of course, is obtained by using a separate supply for the voltage amplifier. Some designers also use regulated power supply to reduce the ripple and the impedance. Even if this supply is critical (since the stage has voltage amplification), hopefully this part of the amplifier is operated in class A. It is in other words ’only’ ripple (and noise) to consider; so passive RC-filtering should be sufficient when separate power supply is used.
Since the input impedance to the current amplifier of a class B amplifier vary strongly around the cross over between the two power transistors, this represent a very nonlinear load for the voltage amplifier. This means that the voltage amplifier distorts more than it would do when using class A, where the input impedance is nearly constant. The output impedance of the amplifier operated in class B also varies heavily around the cross over between the two power transistors. In class A the variation is small, and additionally the value is smaller, thanks to the high current in the power transistors.
Negative feedback may reduce the output impedance, but not the relative variation for class B (and A/B). As long as the output impedance not increases with the frequency in the audible spectrum worth mentioning, this represents a minor problem. Some designers use a relative low impedance load for the voltage amplifier. This is reducing the influence of the nonlinear load that the current amplifier is representing and at the same time is reducing the effect of the Early effect. This implies that the current amplifier is nearly current driven, which is an advantage with complex loads. The output impedance also is reduced.
However, the method is not without disadvantage, as the voltage amplifier has to deliver higher current to this low resistive load, implying higher distortion. Other designers trust negative feedback of astronomical size at low frequencies and a compensation capacitor, giving voltage drive at higher frequencies. Such amplifiers have a typical OPAMP- characteristic with low open loop cut-off frequency and relatively low Slew Rate value (caused by a large compensation capacitor).
From the above it should be obvious that the choice of class A not is taken just to be eccentric. In contrary it is not difficult to find good arguments to let the amplifier work in class A.
The advantages can be summarized as:
The last point is referring to both mechanical and electrical design. In addition to e.g. the items regarding methods to reduce cross over distortion and switching distortion, it is worth mentioning printed circuit board layout, wiring etc, which are critical for class B amplifiers.
Negative feedback is several times mentioned as a plausible method to improve, but not eliminate, corruption produced by class B (and A/B) amplifiers (still it is made such amplifiers without global feedback known for theirs sonic qualities).
As it appear from the above, designing a current amplifier with appurtenant power supply, without any possibility for error correction and operated in class B is not a promising job. A class A amplifier designed without any form of global feedback, is a more grateful task. Negative feedback, however, has its advantages:
As the feedback also tries to correct offset from the quiescent point because of the temperature, a DC servo or blocking capacitor may be avoided. Also disturbances at the amplifier output are corrected by aid of the feedback.
It is a relevant question to ask whether use of negative feedback implies some kind of information loss. It is also a fact that amplifiers both with and without global feedback that have good sonic qualities are available. It is considered that the linearization is the most important advantage of negative feedback. It is more difficult to design linear amplifiers without feedback. Here the distortion is high at high levels, and more complicated design is needed to reduce the distortion. It is very difficult to compare amplifiers with and without global feedback. One may argue that amplifiers with feedback have a tendency to ’compress’ the sound and be less transparent, but the differences are in that case marginal. The challenge often is two-sided: to design an amplifier with negative feedback preserving the good properties from a non-feedback amplifier, or to design a non-feedback amplifier with sufficient linearity.