Negative Supply-Rail Rejection
The V- rail is the major route for injection, and a tough nut to analyze. The well-tried wolf-fence approach is to divide the problem in half, and in this case the fence is erected by applying RC filtering to the small-signal section (i.e. input current-mirror and VAS emitter), leaving the unity-gain output stage fully exposed to rail ripple. The output ripple promptly disappears, indicating that our wolf is getting in via the VAS or the bottom of the input pair, or both, and the output stage is effectively immune.
We can do no more fencing of this kind, for the mirror has to be at the same DC potential as the VAS. SPICE simulation of the amplifier with a 1 V (0 dBV) AC signal on V- gives the PSRR curves in Figure 9.10, with Cdom stepped in value.
Figure 9.10: Negative-rail rejection varies with Cdom in a complex fashion; 100pF is the optimal value. This implies some sort of cancellation effect
As before there are two regimes, one flat at -50 dB and one rising at 6 dB/octave, implying at least two separate injection mechanisms. This suspicion is powerfully reinforced because as Cdom is increased, the HF PSRR around 100 kHz improves to a maximum and then degrades again, i.e. there is an optimum value for Cdom at about 100 pF, indicating some sort of cancelation effect. (In the V+ case, the value of Cdom made very little difference.)
A primary LF ripple injection mechanism is Early effect in the input-pair transistors, which determines the -50 dB LF floor of curves in Figure 9.10, for the standard input circuit (as per Figure 9.10 with Cdom = 100 pF).
To remove this effect, a cascode structure can be added to the input stage, as in Figure 9.11.
Figure 9.11: A cascoded input stage; Q21, Q22 prevent AC on V- from reaching TR2, TR3 collectors, and improve LF PSRR. B is the alternative Cdom connection point for cascode compensation
This holds the Vce of the input pair at a constant 5 V, and gives curve 2 in Figure 9.12.
Figure 9.12: Curve 1 is negative-rail PSRR for the standard input. Curve 2 shows how cascoding the input stage improves rail rejection. Curve 3 shows further improvement by also decoupling the TR12 collector to V-
The LF floor is now 30 dB lower, although HF PSRR is slightly worse. The response to the Cdom value is now monotonic, simply a matter of more Cdom, less PSRR. This is a good indication that one of two partly canceling injection mechanisms has been deactivated.
There is a deep subtlety hidden here. It is natural to assume that Early effect in the input pair is changing the signal current fed from the input stage to the VAS, but it is not so; this current is in fact completely unaltered.
What is changed is the integrity of the feedback subtraction performed by the input pair; modulating the Vce of TR1, TR2 causes the output to alter at LF by global feedback action. Varying the amount of Early effect in TR1, TR2 by modifying VAF (Early intercept voltage) in the PSPICE transistor model alters the floor height for curve 1; the worst injection is with the lowest VAF (i.e. Vce has maximum effect on Ic), which makes sense.
We still have an LF floor, though it is now at -80 rather than -50 dB. Extensive experimentation showed that this is getting in via the collector supply of TR12, the VAS beta-enhancer, modulating Vce and adding a signal to the inner VAS loop by Early effect once more. This is easily squished by decoupling the TR12 collector to V-, and the LF floor drops to about -95 dB, where I think we can leave it for the time being (curve 3 in Figure 9.12).