The problem with class-B amplifiers is that the crossover non-linearity occurs at small signal levels. The high order harmonics liable to be created by the discontinuities when each half of the output stage switches on or off will therefore not fall to insignificant levels at low signal levels in the same way that other sources of high harmonics do. Adjusting quiescent current in the output stage for maximum linearity can reduce the distortion level, but maintaining the correct adjustment is difficult, and careful thermal compensation is needed to keep errors at a low level. Avoiding the switching by operating the output stage in class-A, i.e. at such a high quiescent current that both halves remain conducting at all signal levels, can generate an inconvenient amount of heat if high power output is needed. A compromise sometimes chosen is known as class-AB, and here the quiescent current is higher than for class-B, giving good linearity for small and moderate output power. Unfortunately as the signal level is increased further each half will switch off at a certain level, one for positive and the other for negative polarity input. The variation in output stage gain as this happens is far higher than the gain variations for an accurately adjusted class-B stage, but because it only happens at relatively high signal levels it will hopefully be masked by the high level signal.
There is clearly some advantage in using the maximum possible quiescent current to delay the onset of distortion to as high a signal level as possible. For a conventional emitter follower bipolar junction transistor (bjt) output stage using emitter resistors, with quiescent current 500mA, the maximum linear output current is 1A. For a mosfet output stage at the same quiescent current, with no source resistors, linearity is maintained up to 2A (provided the n and p-channel devices are well matched; not so easy in practice). The greater output before distortion sets in is because of the square-law characteristic of mosfets, at least in theory. At 500mA the gate voltage change needed to switch one mosfet off will be enough to increase the current in the other to 2A, which is then available to drive the load. There is another advantage for mosfets that when one device switches off there is no sharp jump in stage gain, only a jump in rate of change of gain. (Again this is in theory for ideal square-law devices).
There is an alternative approach, which I first encountered in the letters pages of Wireless World as a suggestion to improve the performance of a 741 op-amp by adding a current source at the output to ensure that one half of the output stage would remain conducting at all times. (M.L.G.Oldfield, Wireless World, March 1973, p.119) The addition of a current source can be illustrated by a slightly rearranged version of my original feedforward amplifier, here used as a conventional class-B amplifier with one simple addition.
The addition is at the bottom right-hand corner, and is just a npn transistor plus emitter resistor, designed in this example to add 100mA to the amplifier output. The amplifier should be adjusted for minimum distortion without this current source, operating in conventional class-B mode, and the quiescent current should then be a little over 10mA. Adding the 100mA source the upper half of the output stage will, with no signal, also operate at 100mA while the lower half BD912 will remain switched off. The bottom half will only start to switch on when an output current in excess of 100mA is required, and in the same region the top half will also be starting to switch off. The resulting crossover should be about the same as in the correctly biased class-B mode, but now only occurs at a higher output signal level.
To return to the earlier example of using 500mA quiescent current it was pointed out that a standard bjt class-AB stage then remains fairly linear up to 1A peak output, while using mosfets extends this to 2A. The 'improved' method unfortunately would only remain linear up to 500mA at this quiescent current, so the improved linearity when crossover does occur must be balanced against the reduction in signal level at which it occurs. (e.g. a tenth of the crossover distortion, but starting at half the signal level compared to normal class-AB). My own choice would be to use the mosfet version, giving potentially the highest linear output level, and allowing such high levels of overall feedback that any crossover effects can anyway be reduced down to the noise level.
The circuit above is just an example, and could certainly be improved. However, without the added current source the addition of just two resistors is sufficient to far more effectively eliminate crossover distortion, as shown on this site as the 'Improved Class-B Amplifier'.HOME.