A new audio amplifier topology with push-pull transimpedance stage - Part 2: Biasing, stability and AC performance

Another view on figure 6 reveals two last components whose function I've not yet disclosed - Q12 and Q13. These transistors limit the collector current of Q10 and Q11 to a value of about six times the quiescent current. This is necessary because otherwise overload conditions can increase the collector current and power dissipation beyond the maximum rating of typical low power parts. Usually this current limiting scheme will not interfere with the slew rate capability of the amplifier; if necessary, the emitter resistors R11 and R12 may be bypassed with a capacitor to restrict the current limiting to lower frequencies.

7. Further Improvement of Power Supply Rejection
As analysed in some detail in the previous sections, the basic power supply rejection limitation of prior art amplifier architectures is absent in the novel topology presented in this article. However, we also need to consider second-order effects at some point.

If the power supply rejection of a typical implementation, based on figure 6, is simulated it is found that the power supply rejection is indeed independent of frequency (that is, at least up to about 10 kHz), but limited to roughly 60 dB (figure related to the output). At high frequencies this is a definite improvement over the performance of standard amplifier topologies, but the restless urge for perfection asks for more.

It is possible to remove many of the second-order effects by circuit modifications (e.g., by replacing R7 and R8 in figure 6 with current sources, or by substituting the simple Widlar current mirror formed by Q6 and Q7 with a more elaborate implementation) and thereby improve power supply rejection, but a far more powerful and yet simpler approach exists.

Due to the voltage gain of typical power amplifier implementations the output voltage swing is far greater than the voltage swing at the amplifier input. Thanks to the structure of the novel topology, only the folded cascodes in the transimpedance stage will see the full output voltage swing. This means that only these circuit elements, which represent a very minor ripple injection route, need to be connected to the main high voltage power supply.

All preceding stages are easily powered from a supply with lower voltage, which must just accommodate the input voltage swing. Such a low voltage power supply is straightforward to derive from the main power supply by the use of simple voltage regulators; note that this comes without the cost of additional mains transformer secondary windings, rectifiers, large smoothing capacitors and so forth. It is hence a very economical solution, adopted without hesitation even for cost sensitive applications.

In figure 7 a possible basic implementation of such an amplifier with regulated supplies for the amplifier front-end is shown. Even if the used shunt regulators (D1, D2, R7 and R8) are just implemented with resistors and discrete zener diodes as shown, the ripple rejection will approach 40 dB, pushing the overall amplifier power supply rejection to 100 dB.

At low frequencies, it is possible to advance things another order of magnitude by the use of more elaborate regulators (e.g., by replacing R7 and R8 with current sources). At high frequencies, the output buffer also contributes some power supply rejection limitations which are not easily removed, and layout effects will be more difficult to control. Currently these figures are based merely on simulation results; however, I have no indication that these should be misleadingly optimistic.

Besides the power supply rejection improvement, there is another benefit from the use of such low voltage, regulated power supplies for the amplifier front-end: the need for transistors with large breakdown voltage ratings is eliminated, and we can choose from a far greater range of transistors with improved performance, e.g., with lower noise and higher h_FE. Also as power dissipation is reduced, smaller (and standard surface mounted) packages might be used which is beneficial to reduce cost.

I shall not fail to point out that the use of very low power supply voltages for the input stage may pronounce common-mode distortion at some point. If supply rails substantially below ±15 V are used, this should be carefully evaluated. Usually the use of a bootstrapped cascode [1, 2] for the input differential pair will be a complete cure.