Amplifiers & feedback

[capslock]

Jon, you said over at Madisound that you had a paper by Baxendall that proves that NFB simply shifts distortion to higher harmonics. Do you have a reference?

The only way I can see this happen is when amps are optimized for maximum LF open loop gain. I don't see how this can happen if the amp is designed for reasonable open loop linearity, and if its open loop gain is moderate and only begins to roll of above the audio band.

Most current designs falls somewhere between those two categories. I can see that they will not be effective in surpressing higher order harmonics, so they shift the relative balance of harmonics (which may be a bad thing), but they do not actually amplify the higher harmoncis.

I have a hunch the whole paper may be an apples and pears kind of argument, and I'd like to see what kinds of circuits he considered comparable.

[JonMarsh]

I'll have to go rummaging around to find that- it's in my "off site" storage (a climate controlled 10' X 20' unit). The basic premise that for a circuit with non linearity "x" at a certain level, with the typical low order THD of "benign" circuit issues (in other words, not inherently high order issues like crossover notch distortion which have plagued many amplifiers) if you applied 40 dB of feedback around that, how much is the total lienarity reduced, and whats the spectral distribution? The low order harmonics decrease of course, the higher for all the obvious reasons don't decrease, but actually usually increase in level.

I've measured quite a few designs open loop, including some units like Hafler MOSFET output amps, and frankly, most of the them stink.

Really, there are two main issues as I see it- inherent open loop linearity of the power buffer stage (and stability of that stage into a variety of load angles), and the linearity of the voltage again stage.

For example, the most recent Ayre design has some further tweaks to the supersonic compensation of the output stage (uses RF techniques in a way- something I first came up with for a wideband MOSFET output stage) (think "grid neutralization" and you know the whole story), plus the bias track output devices and some tweaks to the driver design. Then, the voltage gain stage uses a new cascode circuit compared with the V5/V6 configuration, and the net result is pretty nice- about 0.02% at 250 watts out, monotonically decreasing with level. I haven't played myself with the bias track transistors, but we'd been going the same direction on the other stuff. (Chas is nice since I'm an old friend and have made some contributions, and we share info on the amp schematics).

It's likely you could make a nice sounding amp if you had something like the Ayre or Theta Digital or BAT circuits and only applied a gentle 10 dB or so global feedback, but some other designs with variable feedback have had anecdotal evidence that they sound better without it. Tube designs, too.

Unless you start with the premise of each stage having to stand on their own two feet without global feedback, rarely are the individual stages implemented that well, IMO. I've taken that stance before I switched to working with NFB designs, and everyone would say, "Why do you have so many parts and all those cascode stages, and that really weird looking current mirror, and local feedback loops? You don't need that to get good numbers. " But I wasn't looking for good numbers.

[capslock]

I agree that you can get gain and medium load impedance buffer stages to be distortion free up to - 90 to 100 dB. But even making an output stage to operate at -70 dB by itself is a major challenge.

On a side note: in the app note for the thermal trak transistors, in the no load 1 and 10 kHz THD curves, the conventional bias always has an edge for low and medium voltage swing - a case of overbias and cross-conduction?

Also, isn't it strange how they use 2x 3 diodes to compensate for 2x 4 (maybe only 2x 1) junctions that see thermal stress?

[Victor]

I must agree that making a push-pull output driver to operate at low distortion is difficult if not impossible. I would say that -70 dB is not doable unless you make a compound circuit with for instance high-side PNP devices. This will have voltage gain and you can use local feedback to control the distortion down to better then -70 dB. Such circuit tends to oscillate and the layout will need a lot of attention.

In my own efforts to power amp designs I also subscribe to a view that you must start with a very liner open loop approach and that means cascoding. Then you do not need all that much overall NFB. However, I also noticed that if you keep the conventional output push-pull buffer outside the global feedback loop, then the THD is rather high. Naturally even with a global NFB the high order harmonics are not affected because the excess gain at high frequency is close to unity.

Victor