The Class i low-distortion audio output stage (Part 2)
Kendall Castor-Perry - October 16, 2012
In this, the second part of four articles, Kendall Castor-Perry introduces the driver circuit for what he calls the "Class i" output stage. He shows through detailed analysis that it can deliver perfectly resistive output impedance, through complete cancellation of transistor non-linearity.
Introducing the Class i driver
We saw in part 1 that using a long-tailed pair form of power stage driver could give us a nice way of creating the small bias voltages needed to set the quiescent current. We also saw, though, that the linearity performance of such a stage is exquisitely dependent on the value of this bias voltage, and is never actually perfect. In this part, we'll look at a method for fixing this robustly.
If we don't want the current in the inactive half of the output stage to fall to zero, we need that half to transform topologically into a constant current source/sink with a predictable value. And the offset-equipped differential input amplifier discussed in part 1 immediately suggests a route to doing this. Consider the circuit of figure 4 compared to part 1's figure 1. The base of the input transistor has been connected instead to the output terminal, driving a current out to a load.
The feedback transistor runs at m times the current in the input transistor. Because of the resulting built-in offset voltage of the imbalanced long-tail pair, the effect is to produce a circuit that is unwilling to deliver lower than a certain value of current, set by this offset voltage across Re. And that sounds like what we want out of half of our output stage; lots of current when we want current, and non-zero current (soaked up by the other half, of course) when we don't. There's just one problem, though – we lost our input terminal, in order to make this happen.
So, here's the clever part: just put another transistor back in, to act as an input path. We get the circuit shown in figure 5. The resultant input transistor ‘doublet' is the characteristic signifier of the Class i driver. Depending on the relationship between the voltages at the two bases, the current-splitting in the long-tailed pair is controlled either by the input signal or by the connection back to the output node.
We'll presently see that the resulting circuit neatly hands off control from one half to the other as the load current direction changes. But what makes it special is that it actually does work exactly, and can deliver solid design equations that can be used for quantitative work. It is possible to dimension a circuit that not only essentially eliminates both the switching and transconductance modulation components of crossover distortion, but also works out of the box in production, over temperature, with absolutely no need for any trimming components or complicated control loops.
The output impedance is indeed purely resistive; under these admittedly rather ideal conditions, the output stage cannot introduce any current-dependent distortion. Only two parameters need to be chosen; the emitter resistor and a parameter K (equal to the ratio of the mirrors in the collector feeds). In a practical discrete circuit, this ratio is set by using unequal degeneration resistors in the collector mirrors. Early effect in the mirror transistor can be neglected as long as several hundred mV is dropped across those resistors.
Patent searches in the late 1990s indicated that of the many investigators of what amounts to an ‘analogue OR gate' operation required for this kind of current control , Nakayama  got closest to the Class i configuration. However, his patent (filed 1983, granted 1985) shows elaborate circuits of unnecessary complexity, and I feel that he missed the atomic elegance and ideal operation of the Class i doublet.