Reference

Conventional wisdom has it that the series of harmonics we lump together as measured "THD" are created by diverse nonlinearities, occurring principally in the power devices of well-engineered units. The causes include: 1) fundamentally non-linear gain or transconductance, true of all devices; 2) opposing devices passing conduction among themselves, alias "crossover distortion," true of class-AB operation; 3) mutual mismatch between paralleled devices; 4) non-complementarity: pnp and npn devices are rarely truly identical opposites in every respect; 5) thermal feedback—where temperature changes follow the signal and in turn alter the device characteristics; 6) all the other stuff nobody can be bothered to categorize, commonly called "noise"—but is it really?

Items 1–6 can be reduced to very low levels by good design, which involves choosing (or stumbling across) one of the right combinations of types of parts, topologies, and design philosophies, and extensive testing against the bounds of realistic operating conditions. In designing a new power amplifier over the past two years, I felt I had reduced these defects to below the threshold of concern. Still, I had some distortion.

Many problems are hidden by sloppy word usage. Creative thought requires terms taken for granted to be continually redefined and "error-checked" (footnote 9). Noise is truly a random, stochastic thing. Noise is hiss. Hum and buzz are not noise, but periodic signals related to the AC line's generator frequency. If you carelessly call hum "noise," you may be upset when use of a low-noise IC fails to reduce that "noise."

A great deal of so-called "noise" is really signal feedthrough, the brothers Crosstalk, Backtalk, and Sidetalk. Crosstalk is familiar enough, Backtalk obvious enough. Sidetalk introduces to your music a number of signals inside electronic equipment which are quite different from the audio signal at the input and output. In power amplifiers in particular, this signal may include strong elements of mains-related signals, and unilateral (half-wave) audio signals in the output stage. A great advantage of amplifiers using MOSFET outputs is that these hostile signals can be more readily limited to the two output supply-rail nodes alone, rather than occurring throughout a number of nodes in preceding driver stages (footnote 10).

When such "X-talk" contributions have been engineered below audibility, you will still be left with distortion, unless the amplifier's output stage is biased into class-A operation. If you concede that class-A is gross on ecological grounds for powers above a few watts, you're left with class-AB. Then, however linear your circuit, however much feedback you use, and however big your power supply (a traditional type is assumed), a distortion product will be present or will eventually obtrude with decreasing load impedance and increasing frequency.

"Surely with a butch enough supply?"
Well, even using a 100,000µF reservoir capacitor, and superconducting, resistance-less wiring, every amplifier with a conventional power supply has a significant impedance. It increases toward either frequency extreme (as seen in fig.4), and always drops a voltage in proportion to the current dragged through it. If the current varies, the voltage on the rails (as seen by the output stage) varies too. In a class-A amplifier, current draw should be constant. (Any change indicates a shift from class-A into class-AB operation.)

Turning to actual figures, the impedances in fig.4 are very much best-case. A 1 milliohm impedance at 1kHz (0.001 ohms), while seemingly small, will cause a voltage perturbation of 10mV if 10 amperes of peak current are drawn from the amplifier's output stage by the speaker. This perturbation is 60dB below the typical output signal of 10V, but only at 1kHz. At low and high frequencies, it can range up to just 10 or 20dB below the signal, particularly with earthly power supplies. (The model used to plot fig.4 assumes superconductor construction, in supercooled orbit around Pluto, maybe.)

To make matters even worse, rectifier conduction modulates this impedance at 120 or 100Hz, momentarily reducing it at low frequencies to the transformer's source impedance, with a variable, load-dependent pulse repetition frequency. Ugh.

Thereafter, any amplifier relies solely on Power Supply Rejection (PSR, or local feedback, footnote 11) to remove this large, high–garbage-content signal from the output. PSR is broadly related to the amount of feedback you use. It invariably decreases with increasing frequency—and decreases sooner the more feedback you use—but nested and current-mode feedback can stave off the inevitability.

The upshot is that, particularly if you want to use low-feedback, class-AB circuits, power-rail garbage will leak into your music. To overcome this once and for all, my amplifier project needed an audio-grade power regulator. Most designers abandoned regulators in audio amplifiers 15 or more years ago because they didn't seem necessary, or were too complex and expensive for the nominal gains (footnote 12). A fresh approach was needed.

DC voltage stability, the usual obsession of regulator designers, is irrelevant. Instead, we need to reduce supply impedance and feedthrough to vanishing levels and have noise as low as a line stage, yet be able to source peak currents of over, say, 50 amperes. With the help of the latest US IC technology, Japanese/UK power devices and electrolytic capacitors, and MicroCAP III circuit simulation, I found it surprisingly easy to construct a unit with exceptional specifications. At the node from which the power devices draw their current, measured and computed impedance was reduced to micro-ohms at low frequencies, and no more than 2 milliohms at 200kHz, a decade above the audio band. As a rough measure, supply-induced residue is now below –60dB at all psychoacoustically significant frequencies, typically below –110dB below 1kHz, and rising only slightly at high currents above 15 amperes.

What does all this mean? Suddenly, the power rails are as noise-free as the line input you are listening through. If one can afford a big enough transformer, the regulator circuit can supply 40 amperes continuously. "One-shot" current into the most greedy passive-crossover/speaker combination is practically unlimited, as theoretical instantaneous current capability exceeds 10,000 amperes—although to get this, the sinking path would need to have a net resistance under some 20 micro-ohms!

The Outlook after Re-regulation
A regulator with these attributes dramatically affects the amplifier's measured performance. The noise spectra become clear, while the THD+N can drop an order of magnitude or more, depending on how much negative feedback is in use and how good the conventional power supply is. Need I say that the regulator affects sonics? Pinpoint images, tonal clarification, inter-transient silence, all the stuff of class-A, perhaps?

The ramifications are interesting. First, clear evidence has emerged supporting the phenomenon of power-supply modulation. This was recently highlighted by Greg Ball (footnote 13), an Australian audiophile engineer who shares my thoughts about the superiority of class-A amplifiers: it is their absence of power-supply modulation (which amplifiers with output stages biased into class-A cannot, by definition, have) rather than the simplistic absence of crossover distortion (which class-B amplifiers do, by definition, possess, however minutely) which is responsible for the improvement in sound quality. It is now much harder for objectivists to laugh at listeners' reactions to amplifier output-stage class. At the same time, the search for engineering ecology has squared the circle: Clean up the power inside an amplifier, and you can use fewer parts and less juice to sound as good as class-A, and probably sell off your mains conditioner. Suddenly, class-A's supremacy is no longer certain.

---

Footnote 9: "Power amplifier" is itself a misnomer, as power is rate of energy usage, not a thing in itself (like volts or amps); the concept of "amplifying" power is thus highly dubious. "Loudspeaker driver" would be much more apposite.

Footnote 10: MOSFET amplifiers do not require high continuous drive currents and the drivers do not need to be "wrapped" around the output devices, so their driver supplies are readily separated from the output stage and individually regulated, like a line stage.

Footnote 11: Power Supply Rejection is the ability of a circuit to cancel or ignore signals on the DC supply rails, usually by using negative feedback.

Footnote 12: Commercial hi-fi amplifiers currently available that use regulated power supplies for their output rails are the Mark Levinson No.20.6 (and its predecessors, the No.20 and 20.5), the Krell Audio Standard, and models from Exposure, Linn, and Naim.—John Atkinson

Footnote 13: Greg Ball, "Distorted Truth," Electronics+Wireless World, May 1992.