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

[Part 1 looks at advantages and disadvantages of different power supply technologies as well as the design considerations involved in choosing and evaluating a mains transformer. Part 2 considers the pros and cons of external supplies, inrush current control, RF emissions from bridge rectifiers, and relay supplies.]

Power-Supply Rail Rejection in Amplifiers
The literature on power amplifiers frequently discusses the importance of power-supply rejection in audio amplifiers, particularly in reference to its possible effects on distortion ^[4]!

I have (I hope) shown in earlier chapters that regulated power supplies are just not necessary for an exemplary THD performance. I want to confirm this by examining just how supply-rail disturbances insinuate themselves into an amplifier output, and the ways in which this rail injection can be effectively eliminated. My aim is not just the production of hum-free amplifiers, but also to show that there is nothing inherently mysterious in power-supply effects, no matter what subjectivists may say on the subject.

The effects of inadequate power-supply rejection ratio (PSRR) in a typical Class-B power amplifier with a simple unregulated supply may be twofold:

1. A proportion of the 100 Hz ripple on the rails will appear at the output, degrading the noise/ hum performance. Most people find this much more disturbing than the equivalent amount of distortion.

2. The rails also carry a signal-related component, due to their finite impedance. In a Class-B amplifier this will be in the form of half-wave pulses, as the output current is drawn from the two supply rails alternately; if this enters the signal path it will degrade the THD seriously.

This must not be confused with distortion caused by inductive coupling of half-wave supply currents into the signal path – this effect is wholly unrelated and is completely determined by the care put into physical layout; I labeled this Distortion 6 (induction distortion).

Assuming the rail bypass capacitors are connected correctly, with a separate ground return, ripple and distortion can only enter the amplifier directly through the circuitry. It is my experience that if the amplifier is made ripple-proof under load, then it is proof against distortion components from the rails as well. This bold statement does, however, require a couple of qualifications.

Firstly, the output must be ripple-free under load, i.e. with a substantial ripple amplitude on the rails. If a Class-B amplifier is measured for ripple output when quiescent, there will be a very low amplitude on the supply rails and the measurement may be very good, but this gives no assurance that hum will not be added to the signal when the amplifier is operating and drawing significant current from the reservoir capacitors.

Spectrum analysis could be used to sort the ripple from the signal under drive, but it is simpler to leave the amplifier undriven and artificially provoke ripple on the HT rails by loading them with a sizeable power resistor; in my work I have standardized on drawing 1 A. Thus one rail at a time can be loaded; since the rail rejection mechanisms are quite different for V+ and V-, this is a great advantage. Drawing 1 A from the V- rail of the typical power amplifier in Figure 9.7 degraded the measured ripple output from -88 dBu (mostly noise) to -80 dBu.

Secondly, I assume that any rail filtering arrangements will work with constant or increasing effectiveness as frequency increases; this is clearly true for resistor-capacitor (RC) filtering, but is by no means certain for electronic decoupling such as the NFB current-source biasing used in the design in Chapter 7. (These will show declining effectiveness with frequency as internal loop gains fall.) Thus, if 100 Hz components are below the noise in the THD residual, it can usually be assumed that disturbances at higher frequencies will also be invisible, and not contributing to the total distortion.

To start with some hard experimental facts, I took a power amplifier – similar to Figure 9.7 – powered by an unregulated supply on the same PCB (the significance of this proximity will become clear in a moment) driving 140 W rms into 8 Ω at 1 kHz. The PSU was a conventional bridge rectifier feeding 10,000 µF reservoir capacity per rail.

The 100 Hz rail ripple under these conditions was 1 V peak to peak. Superimposed on this were the expected half-wave pulses at signal frequency; measured at the PCB track just before the HT fuse, their amplitude was about 100 mV peak to peak. This doubled to 200 mV on the downstream side of the fuse – the small resistance of a 6.3 A slow-blow fuse is sufficient to double this aspect of the PSRR problem, and so the fine details of PCB layout and PSU wiring could well have a major effect. (The 100 Hz ripple amplitude is of course unchanged by the fuse resistance.)

It is thus clear that improving the transmitting end of the problem is likely to be difficult and expensive, requiring extra-heavy wire, etc., to minimize the resistance between the reservoirs and the amplifier. It is much cheaper and easier to attack the receiving end, by improving the poweramp's PSRR. The same applies to 100 Hz ripple; the only way to reduce its amplitude is to increase reservoir capacity, and this is expensive.