[Ian Iveson[_2_]]

Following the 10c thread, I checked my slightly musty copy of the
quintessentially English Morgan Jones, "Valve Amplifiers", Newnes
1999.

I was curious because I thought his circuit uses a zener to restrict
the upswings of the signal from the cathode-current sensing resistor.
Using a fixed reference isn't ideal if the bias current is adustable,
because then the zener clipping point won't necessarily be correct. As
it happens, he doesn't use a zener. This is what he does instead.

First, he uses a 1R current sense resistor at the cathode of the
valve, followed by an opamp set to a gain of 100. This minimises the
effect of this part of the circuit on the operation of the valve. Each
10mA of cathode current thus results in 1V output from the opamp.

He generates two reference voltages, one twice the other, from a
convenient +ve voltage using a pot and two fixed resistors. The upper
reference is used together with a transistor/diode clamp to limit the
upswings of the signal from the opamp. A low-pass filter then smooths
this to a steady DC before it is compared to the lower of the two
reference voltages, using a second opamp.

The ensuing difference signal is then used to alter the grid voltage
via a second transistor.

Because the signal to the filter is always limited to twice the set
bias level, its AC component remains more or less symmetrical when the
valve cuts off during AB operation, so it smooths to a steady DC
level. With a triode, the servo will tend to very slightly raise bias
current during extended periods of AB, which is better than reducing
it as others tend to do.

That's two opamps, two transistors, a diode, and a bunch of resistors
and caps to maintain the fixed bias of one output valve. Morgan
doesn't use the circuit in any of his amps AFAICS, so there's no
example of how the scheme is extended to the one or more other output
valves.

It wouldn't be necessary to replicate the whole circuit for the other
valves, because they need to be set at the same level, to balance the
transformer current. The set reference voltages could be made common,
and the rest of the circuit replicated. Alternatively the unclamped
signal could be smoothed and used as the reference for the other
servos, which then wouldn't need clamps.

It's possible that reading the above will not result in a working
diagram in your head. The actual circuit does work. Buy the book. It
is British and we are short of money.

Some thought, in the details, must be given to stability. Each servo
is a feedback circuit, and all are related in various ways via various
delays to themselves and to each other.

Ian