Audio amplifier power supply design - Part 3: Power-supply rail rejection

Having peeled two layers from the LF PSRR onion, something needs to be done about the rising injection with frequency above 100 Hz. Looking again at the amplifier schematic in Figure 9.7, the VAS immediately attracts attention as an entry route. It is often glibly stated that such stages suffer from ripple fed in directly through C_dom, which certainly looks a prime suspect, connected as it is from V- to the VAS collector. However, this bald statement is untrue.

In simulation it is possible to insert an ideal unity-gain buffer between the VAS collector and C_dom, without stability problems (A1 in Figure 9.13) and this absolutely prevents direct signal flow from V- to VAS collector through C_dom; the PSRR is completely unchanged.

C_dom has been eliminated as a direct conduit for ripple injection, but the PSRR remains very sensitive to its value. In fact the NFB factor available is the determining factor in suppressing V- ripple injection, and the two quantities are often numerically equal across the audio band.

The conventional amplifier architecture we are examining inevitably has the VAS sitting on one supply rail; full voltage swing would otherwise be impossible. Therefore the VAS input must be referenced to V-, and it is very likely that this change of reference from ground to V- is the basic source of injection. At first sight, it is hard to work out just what the VAS collector signal is referenced to, since this circuit node consists of two transistor collectors facing each other, with nothing to determine where it sits; the answer is that the global NFB references it to ground.

Consider an amplifier reduced to the conceptual model in Figure 9.14, with a real VAS combined with a perfect transconductance stage G and unity-gain buffer A1. The VAS beta-enhancer TR12 must be included, as it proves to have a powerful effect on LF PSRR.

To start with, the global NFB is temporarily removed, and a DC input voltage is critically set to keep the amplifier in the active region (an easy trick in simulation). As frequency increases, the local NFB through C_dom becomes steadily more effective and the impedance at the VAS collector falls. Therefore the VAS collector becomes more and more closely bound to the AC on V-, until at a sufficiently high frequency (typically 10 kHz) the PSRR converges on 0 dB, and everything on the V- rail couples straight through at unity gain, as shown in Figure 9.15.

There is an extra complication here; the TR12/TR4 combination actually shows gain from V- to the output at low frequencies; this is due to Early effect, mostly in TR12. If TR12 was omitted the LF open-loop gain drops to about -6 dB.

Reconnecting the global NFB, Figure 9.16 shows a good emulation of the PSRR for the complete amplifier of Figure 9.14. The 10–15 dB open-loop gain is flattened out by the global NFB, and no trace of it can be seen in Figure 9.16.

Now the NFB attempts to determine the amplifier output via the VAS collector, and if this control was perfect the PSRR would be infinite. It is not, because the NFB factor is finite, and falls with rising frequency, so PSRR deteriorates at exactly the same rate as the open-loop gain falls. This can be seen on many op-amp spec sheets, where the V- PSRR falls off from the dominant-pole frequency, assuming conventional op-amp design with a VAS on the V- rail.

Clearly a high global NFB factor at LF is vital to keep out V- disturbances. In Chapter 5, I rather tendentiously suggested that apparent open-loop bandwidth could be extended quite remarkably (without changing the amount of NFB at HF where it matters) by reducing LF loop gain; a high-value resistor R_nfb in parallel with C_dom works the trick. What I did not say was that a high global NFB factor at LF is also invaluable for keeping the hum out, a point overlooked by those advocating low NFB factors as a matter of faith rather than reason.

Table 9.2 shows how reducing global NFB by decreasing the value of R_nfb degraded ripple rejection in a real amplifier.