[Radio Wrangler]

Hi Andy,

You've just run into two classical problems:

Valve amplifiers with quite low amounts of negative feedback because several factors in their design tend to cause more phase shift than can be easily handled. The output transformer is best known and probably the least tractable cause, but there are others. Leak's Point One design got down to about 1% distortion open-loop and feedback reduced this to the 0.1% of its name. This implies he achieved a stable and manufacturable design with about 20dB of feedback. For a valve amplifier, that's a lot. Arthur Radford was a designer of transformers, and he designed some special ones for his amplifiers, but Arthur Bailey contribured a phase splitter that was effectively cascoded to both outputs. This greatly reduced another contributor of phase lag. All these things combine.

You can put an open-loop amplifier together quite easily. Nick a circuit section from here, borrow one from there, play about with a few standard sections for the rest. It's how a lot of electronic circuits, and all sorts of other things are designed though 'inspired by' usually replaces the 'nicked' word.

What you can't do is nick someone else's feedback design. You have to evaluate the gain and phase characteristics across the audio band and up to silly-looking frequencies and then work out how much feedback you can safely employ. It won't be as much as you'd like, so you see if your feedback network can compensate for it a bit. It can, but only a bit... a bit smaller than you'd like. So you revisit the main amplifier design. Where can you reduce phase lag? Every little helps. Anode load resistors fall in value, bias currents go up. Cathode followers may sprout. Fancy interleaved transformer winding schemes appear. And with all of this you can eventually dare risk 10-20dB. Of course, this reduces the gain of the finished amplifier, so you needed to start with this built into the open loop design as extra gain to be chucked away when the loop is closed.... but the amplifier still needs to do the low lag thing even with this elevated gain.

So the open loop amplifier needs to be designed with feedback in mind.

I once did a silly valve amplifier. Lots of feedback, trivial distortion numbers, ridiculous bandwidth. Output transformers were right out. It used banks of PL519s in a totem-pole configuration and the earlier stages looked more like a fast oscilloscope Y amplifier. It worked. It worked very well. What you heard was what was put into it. All 'Valve sound' was engineered out. The power consumption was terrible, the heat from the thing was too much. It served as a comparison with some other designs I was trying, then it came apart and the bits recycled into other things.

To understand feedback and stability, the easiest way is to learn the maths. Like those junior school sums about 'If it takes seven men three days to dig a trench 3ft wide and 5 ft deep 300yards long.....' Would you rather have to do the sum or have to dig the trench? Treat maths as a labour saving device which actually works. If you see it as unpleasant, the alternatives are worse.

You said you'd heard that the phase shift had to be less than 180 degrees. That's sort of right. It's just incomplete, though. It's less than 180 degrees over the entire frequency range where the gain around the loop is greater than unity. So you don't test at a spot frequency, you have to go to much higher frequencies than audio. Measure the gain and phase until you get so high the gain falls below 1.

THEN you run into the nasty that the things you could add to run the gain down before the phase shift builds up will themselves create extra phase shift. You get sucked into a nasy race. You need a plan to have any chance of winning.

David