Class A Output
We will start with the simplest audio output stage circuit - a single-transistor class A arrangement.
With class A operation, the transistor is biased so that it is passing collector current continuously, very much like the common emitter configurations we discussed previously. The circuit draws the same average amount of current from the supply regardless of output volume. This is not good news for battery sets, where we want to keep the current consumption as low as possible to increase the battery life.
The circuit does have some advantages however. It is fairly easy to set up correctly, there is minimal distortion, and it only uses one transistor. Many early transistor car radios used class A output stages similar to this (often with additional refinements to improve stability), because they could deliver a few watts with only one expensive power transistor. The high current consumption was not a problem with a car battery to supply it, and the heat produced could be easily dissipated by the metal radio case.
This circuit is the transistor equivalent to the normal pentode output stage in a valve radio, and operates in the same way. The biasing is usually arranged to that the collector voltage is about 10% less than the supply voltage. Re is included for bias stabilisation as detailed previously. The speaker is connected via a transformer for impedance matching and to remove the DC bias. The signal to the base passes through a transformer to provide isolation and impedance matching.
This circuit shows the output and driver stage of a typical car transistor radio. R29 sets the bias current through the output transistor. The collector current is shown on the circuit diagram as 550mA. R30 helps to improve stability by reducing the base current if the collector current increases (collector voltage drops). The driver circuit is similar to the output stage, but works at much lower current (3.6mA).
The output power would probably be about 3W. Since the output stage consumes 550mA at 14V (typical charging voltage), the power consumed from the supply is 7.7W (power = current x voltage). The efficiency of the stage is therefore just under 40%. The maximum theoretical efficiency of a class-A output stage is 50% at maximum output power, rapidly decreasing as the output level drops.
A few early pocket transistor radios, such as the Perdio PR4 shown here, also used this arrangement. In order to keep the battery consumption acceptable, the audio output from such sets is typically a few tens of milliwatts, so they don't make a lot of noise! This was done solely to save money - transistors were expensive, so if you could get rid of one or two you could make the set cheaper. This Perdio uses a reflex IF/AF stage (discussed later) and achieves loudspeaker output with just four transistors.
The main disadvantage of class A operation is
the continuous high current consumption. This is because the transistor has
to be biased so that it is passing a significant amount of collector current
all the time. The current is the same whether the audio output is silent or
very loud. A class A output stage is therefore not very efficient. At maximum
volume it is no more than 50% efficient (meaning that 50% of the power the circuit
consumes drives the speaker, the rest is wasted as heat). Because the current
consumption is constant, the circuit becomes considerably less efficient at
Class B and AB Output (Two Transformers)
With a class B output stage, instead of biasing the transistor so that it is passing collector current continuously, we bias it right on the threshold where it is just not passing current. We then use the base current increases caused by the audio input to take it over the threshold and increase the collector current. Thus the average current the circuit draws depends on the volume level.
However an audio signal goes positive and negative. The arrangement I have just described would only amplify the positive half cycles. The negative half cycles would reduce the base current still further, and since its bias level is already at a point when no collector current is flowing, these negative half cycles would just be lost and not amplified at all.
We therefore need two transistors, one to amplify the positive half cycles and one to amplify the negative half cycles. To make this work we need circuits to split the input into two signals to drive the two bases, and to combine the two amplified halves into a whole signal to drive the speaker.
The simplest way of doing this, and the method used in many transistor radios in the late 1950s and much of the 1960s, is to use two transformers.
The basic circuit is shown above. The input transformer produces two signals that are out of phase (relative to the centre tap). These connect to the bases of the two transistors. Resistors R1 and R2 set the DC bias conditions. The output transformer is connected as the collector load of the two transistors. The emitter resistor Re stabilises the bias conditions as described before.
In theory this would work well. The circuit would consume no current (other than the small current through R1 and R2), with no signal. Any input signal would take the transistors into conduction, one on the positive half cycles and the other on the negative half cycles. The output transformer would combine the two halves, and match the impedance of the speaker.
The circuit is sometimes called a push-pull output stage (on the basis that one transistor is pushing and the other is pulling). The maximum theoretical efficiency of this circuit (if we disregard the current through R1 and R2) is 75%, and this applies at any volume, not just at maximum volume. This is clearly much better for battery operated sets.
However, it isn't quite that simple in practice. The main problem is that this nice threshold we spoke about, where the transistors are at the brink of starting to pass collector current, is not as nice as we would like. There is actually a vague curved bit between the point where we have no collector current and the point where the collector current starts to follow the base current in a relatively linear manner.
The circuit with therefore not give a nice transition between transistors as the signal passes through zero. The result will be audio distortion, known as "crossover distortion", as shown here (bottom waveform).
To get around this we have to increase the bias a bit, so that the transistors are passing a small amount of collector current all the time. As the signal passes through zero there is now a small area where both transistors are contributing to the amplifying.
Because the transistors are passing a small collector
current all the time, the circuit is not true class B. It is mostly class B
but with a bit of class A thrown in. It is therefore known as class AB, although
it will often be referred to as class B since this is still fundamentally what
This is not the end of our problems however. Since we have two transistors each amplifying half the signal, we clearly need them to amplify by the same amount otherwise the result will again be distortion as shown here.
When the circuit is assembled, transistors would need to be tested to see how much they amplify, and a pair having similar amplification (known as "gain") would be used. Set manufacturers obviously did not want to be messing around doing this, so the transistor manufacturers did it for them. Rather than actually pairing transistors, they tested the gain of every transistor as part of the final test, and decided which band or range of gains it belonged in. The transistor was then marked to show its gain band - generally by means of a coloured dot on the top on the case. The set manufacturer simply had to fit two transistors having the same coloured dots, and would know that the gains were close enough. Because of the banding, the gains could still differ by maybe 10 or 15%, but this is good enough for your average portable transistor radio.
If you ever need to replace a transistor in any sort of class AB output stage, you should always make sure it is in the same gain band as the other one in the circuit. Generally if one has failed the other one will have been stressed, and it is easier to replace them both, probably with two from a scrap set. If you dismantle scrap sets for the parts, try to keep the output transistors in their pairs.
So we have dealt with the cross-over distortion, and we have matched the transistor gains to avoid this distortion. We aren't there yet though!
Previously I mentioned that the forward drop voltage of any germanium PN junction, such as a transistors base-emitter junction, decreases as the temperature increases. Assume we have carefully chosen the values of R1 and R2 to give a collector current of 5mA through each transistor at 20°C (roughly room temperature). As the temperature increases the forward drop of the base-emitter junction decreases. Since we are holding the bases at a constant voltage by R1 and R2, the base current will increase. As a result, the collector current through the transistors will also increase. In this silent state, a few mA increase in collector current isn't really a problem.
However at higher volumes the current increases considerably - this is the whole point of the design. This higher current (peaks of over 100mA being typical) causes the transistors to become warmer. As they become warmer, their gain increases, and so the collector current increases, and they become even warmer. We have a nasty vicious circle that is appropriately known as "thermal runaway". Left unchecked it would soon result in the demise of the transistors.
I mentioned Re in passing earlier and said that it stabilises the bias. This is actually the first line of defence against thermal runaway. As the collector current increases, the voltage drop also increases. Since the voltage on the transistor bases is held constant by R1 and R2, the base-emitter bias is reduced, thus reducing the collector current. Re reduces the efficiency of the circuit a bit (because some power is wasted there rather than being delivered to the output transformer), but this is something we are generally happy to live with. For small sets with outputs of perhaps a couple of hundred milliwatts, adding Re (typical value being somewhere between one and ten ohms) is sufficient to prevent thermal runaway, and is a cheap and simple solution.
Bias Stability Circuits
We can do more however, and it doesn't significantly increase the complexity. Rather than relying on the increasing current to control the biasing, we can monitor the temperature of the transistors and use that to reduce the bias directly.
Three possible ways of doing this are shown here. In (a) a thermistor "T" is connected in parallel with the bottom bias resistor. A thermistor is similar to a resistor, except that its resistance decreases as its temperature rises (thermistors that increase in resistance as the temperature increases are also available). The thermistor is mounted close to the transistors. If the transistors start to warm up, the thermistor also warms up and its resistance starts to decrease. This decreases the base bias to the transistors, thus reducing the current, which should stop the transistor getting any warmer.
In fig (b) we have replaced the thermistor with a diode. Because the diode is also a PN junction made from the same material as the transistor, its forward drop should vary with temperature in a similar manner to the base-emitter junction of the transistors. This works reasonably well, however because a diode junction is constructed differently to a transistor base-emitter junction, the voltage to temperature ratio does not follow quite as closely as we would like.
Fig (c) shows a further refinement. A transistor of a similar type to the output transistors is connected with its base and collector linked. In this state it is turned fully on. The collector voltage is the same as the base voltage and the majority of the current goes through the collector. This device will follow the temperature to voltage variations of the output transistors very accurately, and is probably the best solution to the problem.
However, in the era we are interested in, transistors were not cheap, so the idea of adding an extra one just to stabilise the bias did not go down well with financial departments who wanted the sets as cheap as possible. There was of course the opportunity for the marketing department to count this transistor when proclaiming the number of transistors fitted in the set, but this was questionable morally because the transistor was just in a supporting role and was not actively amplifying or processing the signals.
The circuits in figs (b) and (c) have an additional advantage. Because the voltage across a PN junction remains relatively constant with varying current, the bias to the output stage is held more constant as the battery voltage drops, compared to a resistor only circuit or a resistor-thermistor circuit. The set will therefore perform well as the battery voltage drops, increasing the useful life of the battery.
A set designed for a 9V battery and using resistor biasing will start to sound distorted when the battery voltage drops to about 7V. A similar set using a transistor biasing circuit will continue to work (at reduced volume) down to about 5V, by which point the 9V battery is pretty well dead. Some manufacturers made this a selling point, quite rightly. For example, Pye/Ekco/Invicta coined the term "Peak Performance Control" or "PPC" to describe it.
You may have noticed that Re is not included in these circuits. If the biasing correction is well designed, Re is not needed because the bias correction will take care of everything. By eliminating Re, the efficiency and output power of the circuit is increased. In practice however, a single thermistor or diode/transistor junction will not provide complete compensation, so many sets will still have a small Re as well as other bias correction measures.
Why Bother with Class AB?
So there is a lot more messing around needed to get a good reliable class AB output stage, compared to a class A stage. Why not just use class A?
As mentioned before, the theoretical efficiency of a class B output stage is 75%. Since we need to use class AB, and have losses due to Re (if fitted) and the current through the biasing circuit, we aren't going to achieve this sort of figure. Realistically we could probably expect around 60 to 70%. This compares favourably to the 50% maximum of class A, particularly bearing in mind that class AB remains reasonably efficient at lower volumes whereas the efficiency of class A quickly drops as the volume is reduced.
Taking some real figures, the combined collector current of a class AB output stage with no signal is around 5 to 10mA. At full volume the average supply current would probably exceed 100mA and at a normal listening level it would be around 25mA. A class A output stage capable of delivering the same output power would need to consume around 200mA continuously. At normal listening level the class AB stage consuming 25mA is eight times more efficient than the class A stage consuming 200mA.
With class A, it is impossible to achieve a useful
output power level from a battery supply, if we want the batteries to last for
reasonable time, whereas with class AB this is possible. Without class AB output
stages, portable battery powered transistor radios would probably not exist.