The Class i low-distortion audio output stage (Part 3)

-November 13, 2012

In the third article of a four-part series, detailed simulations of an audio power amplifier using the "Class i" output driver introduced in part two shows its flexible control of output device current and its exceptional temperature stability.

This material originally appeared as one long article in Linear Audio Volume 2, published in September 2011. Linear Audio is a book-format audio magazine published half-yearly by Jan Didden.

A complete circuit using 'real' components
In part 2 we looked at the Class i driver stage analytically. Now it's time to put it into a practical audio power amplifier.

A push-pull output stage using the Class i driver circuit is shown in figure 6. The small signal devices used here are the npn FMMT625 and the pnp FZT796A from Diodes, Inc. (formerly Zetex). These are high voltage SOT-23 devices whose SPICE models declare a high Early voltage (VAF), which seemed like a good idea. 'Ordinary' transistors could be used for the current sources and current mirrors; in simulation, changing the models for these devices has scarcely any effect.

Figure 6: A complete Class i push-pull output stage.

The output stage is a triplet Darlington, as recommended by Cordell, whose suggested output and driver devices are also used here [5]. By the way, when using such compound devices, always return the bleed resistors to the top of the emitter resistors, or cross-connect them to the other stage. Don't connect them directly to the output node (homework: why?).

Emitter resistors are set to 0.33 ohms, as in Cordell's circuits. In an amplifier designed to deliver higher current, I suggest reducing these resistors and accepting the higher quiescent current, to reduce the loaded voltage drop. My later practical designs usually used 0.047 ohms. The mirror degeneration resistors are set to drop 0.5 V, and mirror imbalance parameter is set to m=1.2. Tail current of each long-tailed pair is nominally 5 mA (set approximately through resistor parameters).

The circuit also shows a couple of additions: unequal degeneration resistors and resistors in the doublet base connections. In the circuit analyzed, the parameter rx=0, and rt=8 ohms; the simulator calculates the expressions in braces. This enables us to correct for the temperature variation of the quiescent current.

The tail current sources have a roughly CTAT (complementary to absolute temperature) behaviour, and so a low-value resistor in series with the emitter of the feedback transistor (or difference between emitter resistances) can provide an extra voltage that falls with temperature just as the long-tailed pair's offset is rising. We can therefore temperature-stabilize the offset voltage in somewhat the same way as a bandgap reference is stabilized.

This additional resistor does not affect the ideal linearity in any way; it just increases the effective value of the constant K in the defining equation. In practice, the reduction in loop gain around the driver does eventually increase the sensitivity of the circuit to non-idealities in the output devices, but not until the local loop gain has been significantly reduced.

The increase in the limiting and quiescent currents can be easily calculated because the change in offset voltage is just the device current flowing in the effective resistance difference between the degeneration resistors. Surely you must have been expecting some homework!

As with all power amplifiers using slow BJT output devices, the response varies both with the current demanded and the output voltage; figure 7 shows the AC response with the static output voltage stepped from -30 V to 30 V onto a 4 ohm load (a highly recommendable test for all amplifiers!). The change in output device characteristics with current causes the amplitude and phase response to vary, but not in any way that's more difficult to contend with than with conventional output stages.

The poor high-frequency response of large BJTs can cause even the very tight local feedback loop to exhibit some stability issues without the use of compensation. In figure 6, a simple pole-zero network has been wrapped around each feedback transistor as an example of a compensation scheme. The halves can be analyzed independently by simply breaking the connection between the emitters and loading each half separately.

You should of course do detailed bench and simulation studies with your particular transistor choices. In a subsequent practical implementation, stability was improved by removing the zero-defining resistors [11]. Output Zobel networks should also be used to ensure that load impedance is predictable at these high frequencies.

Figure 7: AC response of figure 6 as static output voltage varies from -30 V to +30 V.

Loading comments...

Write a Comment

To comment please Log In