Interviews

May: I've got to say that it's pretty well broken up, distributed over the whole amplifier. It's a pretty stable power supply—not that any unregulated supply doesn't have its limitations—but it's a good transformer, a good power supply. I think the real breakthrough on Polaris was the linearity correction circuit for the MOSFET output stage. Basically what we do is compare an idealized output signal with the actual output signal of the amplifier. And introduce local error-correction networks within the output stage proper. It's a small feedback loop existing right around the output stage.

And if you read the manufacturer's data—I can show you the phrase out of the Toshiba applications guide—they flat-out say you've got ten times the distortion from MOSFETs than you do from bipolars.

Atkinson: Are you saying that the highly touted power MOSFETs are basically more nonlinear than a good bipolar? That a MOSFET has a crooked-looking transfer function?

May: Yes. Consider that theoretically the maximum error you can have with a good bipolar transistor used as an emitter-follower is a change in the base-emitter voltage. Is delta Vbe. And that's a small percentage of the total power-supply voltage. MOSFETs, however, for want of a better term, are "glitchy" in their threshold region, very non-linear in the region that corresponds to zero crossing. Take a look at MOSFET amplifiers: most have their output devices turned on pretty hard. Their bias currents are set pretty high.

Atkinson: That would be appear to be supported by the fact that about five years ago, there was much discussion in the audiophile press about improving the sound of what was then a very common MOSFET amp, the Hafler DH200. If you up the quiescent current of the output stage of each channel to 450mA, it'll sound like a whole different amplifier.

May: It will. We run bias currents of 100mA per part on Polaris, which on 60V is quiescently dissipating 6W per part. Yeah, the amplifier is sonically improved when you drive it a little bit harder. It then takes a pretty demanding source to reveal the differences in the amplifier.

Atkinson: Your next power amplifier is a Mark II version of the class-AB Andromeda. What will be the major differences?

May: We'll be providing more output-current capability, repackaging the unit, taking advantage of some of the later-generation output devices and small-signal devices, and incorporating the output-stage linearity correction that we use in Polaris. I've run an Andromeda prototype, 400W into 4 ohms, at less than 0.05% distortion without loop feedback. And I judged the damping factor on that amplifier open-loop to be well in excess of 100, carried out to a frequency—5, 10kHz—where it's gonna fall out of the question.

Atkinson: So when you close the loop, you're only going to introduce a small amount of overall negative feedback.

May: You really need a minimum of 20dB feedback to get the amplifier to be under the loop control. 20dB is a realistic target figure. Feedback is there to stabilize; not to Band-Aid. And 20dB of feedback with a wideband amplifier is going to pull that distortion to 0.005%. Now how does it correlate with what we hear? I think that'll be a good-sounding amplifier.

Atkinson: I understand that one thing you felt was very important for a preamplifier, something you have incorporated in the Sumo Athena, is the ability to drive relatively high currents. You've got high voltage swing and you've got high current swing. Why is that necessary in a line-level preamplifier?

May: You've got a large amount of cable capacitance. Some of the cables around are 40pF per foot or higher. And you just want the amplifier to be able to totally control what the cable is doing and not get it into a situation where you're slew-rate limited by its limited current capability.

Atkinson: It would appear that some preamplifiers are intrinsically unstable into highly capacitive loads; their designers have to buffer the output with a highish-value series resistor.

May: Yes, they are. The output impedance of our Athena preamplifier is very small, and we deliberately put in a relatively low-value resistor, 75 ohms, so the output resistance of the preamplifier typically looks like 75 ohms. The output stage itself runs in class-A. Of course, it will revert to class-AB if pushed really hard, if you swing about 20V into 2k. But you've got to push it hard. For any practical load, or the IHF standard load of 10k and 1000pF, it's class-A.

Atkinson: Preamplifiers in the Athena's price range tend to have at least some integrated-circuit operational amplifiers. The Athena, however, is totally discrete. Why is that?

May: We don't like the sound of ICs. One, they require tremendous amounts of feedback. Two, their open-loop frequency response is horrible. Even the fastest devices open-loop are down 3dB at 100kHz. They have like 120dB of open-loop gain at low frequencies; they are intrinsically high-negative-feedback devices. I don't know to couch this because it's a difficult thing. I'm not totally against negative feedback—a little bit of feedback is good—but I repeat that a lot of it tends to be a Band-Aid. In addition, by going totally discrete, first we could get high voltage capability. We've got the bipolars running on 35V rails in the Athena. Second, we can get the output capability which you can't get from a conventional op-amp. And even if you do, you've got a heating problem. A thermal problem. What you do in the output stage goes right back to the input.