Interviews

Atkinson: Do you think that the correlation between good sound and low loop feedback is due to the basic linearity of the circuit? Or is it due to other factors? For instance, if you have very high amounts of loop feedback, you can then have time-domain problems.

May: Well, the basic amplifier is more linear. But there are a lot of aspects of it. You have to look at what the open-loop frequency response is with phase compensation. There you want to keep the energy storage in the loop as small as possible. In other words, get the widest bandwidth open-loop. But the phase-compensation should have adequate safety margins when you close the loop.

And consider feedback itself: a paper that came out of the Bell Labs Tech Journal in about 1957 clearly points out that if you throw feedback around an amplifier with second-harmonic distortion, you've then got an amplifier with second and third. You have a feedback multiplication factor. RCA and some of the individuals in the UK did some work in the '40s related to vacuum tubes—I can't find any reason why it shouldn't be equally valid today—they weighted the structure of the distortion, putting extremely heavy emphasis on high orders of distortion and multiplying it by the square. I'll take the fifth as an example—it would be weighted 25 times whatever it measured. The sixth would be multiplied 36 times.

My designs have very little even-order distortion. And from a listening standpoint, time and time again when we've taken amplifiers that have second-harmonic distortion and removed it, their sound cleans up. Stop and think about it. Say we've got a high-level, high-frequency burst of energy, from brass or something of that nature, that lasts for a hundredth of a second. If the amplifier introduces second-harmonic distortion, this is equivalent to a DC level shift in the amplifier, which translates into hearing a low-frequency tone along with the original sound.

Atkinson: If you have to have an amplifier which is nonlinear, then you must arrange its distortion spectrum to be both odd-order and low-order.

May: That's correct. To me, the only distortion—I'm talking about when an amplifier is approaching overload—that's at all liveable with is third. Because it's symmetrical, it cancels out. Have fun with the very high-order, odd-order harmonics. But stay away from things that show DC level shifts.

Atkinson: What you're saying is that you can't have second harmonic distortion in an amplifier because it represents the DC level shift, and you can't have harmonics higher than fourth because their audibility is very severe . . .

May: Yes, but then you have an amplifier which is very symmetrical, an amplifier which is very well behaved up to and in heavy overload. To me a good amplifier is one which you can overdrive by 6 or 8dB and only just begin to detect compression.

Atkinson: Yet very often you get amplifiers that go DC-unstable if you drive them into clipping. You start to see your woofer cones slowly pumping.

May: That's a no-no. That's a problem that just absolutely shouldn't exist.

As a teenager, I got my hands on an old Concertone 15ips machine which could be set up pretty good. An interesting thing about it was that with the power amplifier I was using, I preferred the sound of program material that had been recorded on tape and played back to the sound of program material which passed directly through the electronics chain bypassing the tape. When you stop and think about it, the tape's a good low-pass filter.

Atkinson: It's conditioning the signal so that it doesn't stress the following electronics.

May: With a lot of program sources, you end up with out-of-band components that'll create havoc with the power amplifier. We don't routinely think that there will be a problem with a power amplifier with what's going to happen at 100kHz. But if you get a couple of tones up there and the amplifier is non-linear, you'll get intermodulation products. The difference tone will come right back down in the audio band. If the linearity isn't the right type, you get a thing called cross-modulation where amplitude and modulation on one tone transfer to the other tone. And these side-band products come crashing down into the audio band and you can hear it.

Atkinson: But not with pure audio-frequency tones when you're doing your traditional distortion measurements. You have a music-related distortion which is not detectable on a pure tone.

May: Absolutely. A dynamic distortion. A lot of factors we routinely look at concerning the performance of the amplifiers are out in the multi-MHz region—even driving capacitive loads—because we want to find out if the amplifiers are stable; if they do, in fact, behave well out there. We have found a correlation between the listenability of the amplifier, the smoothness of its top end, and its performance in the 100kHz to 1MHz region. As I said initially, there are a lot of things that I would like to be able to measure that we don't routinely measure; I'd like to bring them back to something that we can correlate to.

Atkinson: The most recent Sumo amplifier is the Polaris. I understand that you inherited the brief for the design—a 100W, single-ended class-AB amplifier using MOSFETs—when Califone took over Sumo. How have the resources been allocated in that amplifier? Is the power supply most important, or is it the use, say, of special output transistors?