The improved class-B amplifier design used an op-amp driving a discrete component output stage. The purpose of this was to simplify the circuit. I was only concerned to demonstrate the output stage and did not want to complicate the design further when there are plenty of perfectly good low distortion op-amps available at low cost. Even higher power designs could use op-amps, but high voltage versions are harder to obtain and inevitably more expensive, so discrete designs are more justifiable. Using the original version, if more power is needed this can be obtained using a bridge circuit. Another design note deals with some of the bridge amplifier options.
One question is why not use op-amps in the output stage, as shown in some of the first diagrams. The use of devices with low offset voltages could avoid the problems of both setting up and thermal stability of quiescent current without the need for the stabilisation circuit suggested. The use of bootstrapped supply lines for the op-amps would avoid the need for high voltage devices, and op-amps with FET input devices would ensure there was no problem of non-linear input impedance in the output stage. What then is the problem? What lead me to abandon this option is that unpleasant things can happen when such a design is overdriven, and also in the crossover region.
When the output stage clips the heavy local feedback becomes ineffective and the op-amps will also clip. The outputs of the op-amps will swing up (or down as the case may be) to their supply line voltage. This is unimportant until the amplifier wants to come out of clipping. This cannot happen until the outputs of the op-amps swing back down to the normal operating level, and this happens at or below the maximum slew rate of the op-amp. (To be more exact, at the small input signal slew rate, which is not generally anywhere near the usual specification for max. slew rate.) There is therefore a short delay while this happens during which the power amplifier is latched up.
Even if recovery time from overload is not considered an important specification, something similar happens in the crossover region. When one half of the output stage cuts off the op-amp driving that half will again swing its output voltage towards one of its supply lines, and have to slew back towards the normal operating point when the signal needs to cross back to that half of the output stage. There are ways to reduce how far the op-amp output swings when cut off, e.g. using a diode in the feedback network, but calculations suggest that even then the delay before slewing back to linear operation will cause crossover distortion unless the quiescent current is set very high and the op-amps have a very high slew rate.
Looking at specifications for op-amps it seems that devices are available with astonishingly high slew rates, 200 V/usec or more, often with only modest unity gain bandwidths. This is misleading, because such figures are invariably only the large input signal slew rates. Some years ago I tried a design using op-amps driving power mosfets, and gave up after trying every trick in the book to make it stable. This has perhaps unduly influenced my attitude. If anyone wants to try the op-amp approach I wish them luck, but confidently predict that they will regret it.
Another option I thought of but rapidly discarded is to use a small signal version of the output stage simply to split the signal into two halves. The two signals can then be applied to two output sub-amplifiers both of which are biased into their linear region with no signal applied, so neither ever cuts off. The problems described above are far less severe in this case, but another problem arises, which is the difference in gain of the two halves of the circuit. The most convenient circuit I could think of had the problem that the gain of each half was dependant on four or five resistors, and if each was 1% tolerance this gave a theoretical variation of 5% in the gain for each half of the signal. If they both varied in opposite directions this gives a 10% difference in gain between positive and negative half-cycles. Hardly a low distortion figure! There is an added problem that something similar was published some years ago in Wireless World, 'New Approach to Class B Amplifier Design'. by Peter Blomley, Feb 1971 P.57-61, and although the method of splitting the signal was completely different to my design there was some mention of patents, (application no. 53916/69) so even though I never heard of that design being used commercially there may be some danger of being accused of patent infringement. For anyone who is still interested in this approach I include a simplified circuit diagram for one of several variations I thought of using this approach:
Resistors are added to bias the output sub-amplifiers into conduction even with no signal from the splitter stage. One half of the splitter stage will still switch off on one half-cycle with the problems mentioned earlier, and a suitable op-amp design needs to be used, e.g. using discrete components, or a transistor array such as the CA3046, so that recovery from overload can be optimised. The splitter and input stage can be operated from a low voltage regulated supply, and the variable control could be replaced by a fixed resistor if the op-amp offset voltages are low enough to accurately calculate the necessary value. The power transistors would need to be either darlington pairs or power mosfets to enable them to be driven by the op-amps. My own experience of trying to drive power mosfets with op-amps was not encouraging.
