Partial feedback DHP SE amp

[Fuling]

Good evening!

I have a few of these babies that I want to put to some use:
http://frank.pocnet.net/sheets/066/a/AL1.pdf

The datasheet recommends 7k primary impedance, but since such transformers are rare I´ve ordered a pair of Hammond 125CSE that will be wired for a 5k plate load. I´ll loose a little power but that´s not much of an issue.

Input/driver stage will be PC86
( http://frank.pocnet.net/sheets/010/p/PC86.pdf ).

First I thought about cathode feedback by putting the OPT secondary between the AL1´s cathode and GND, but after reading through a couple of threads here on the forum I got curious about "Partial feedback" or "Plate to Plate feedback", ie a resistor from the drivers plate to the power tubes plate like the RH84 and RH807 amps.

Since I don´t know anything about the maths behind that feedback resistor I will just start at, say, 100k and adjust until it works as intended
(Unless someone can give me a hint about a proper resistor value)

I should mention that I´ve peeked at John Broskie´s article about this topology but I didn´t get much wiser...

[audiousername]

The feedback should be applied to a high impedance node. It will work better if you leave the cathode of the PC86 unbypassed.

[Fuling]

Ah, you´re right! I read about that just a couple of days ago.
Thanks!

[Eli Duttman]

Regulate the screen grid B+ for lower distortion. Got any VR150s?

[Fuling]

Nope, no VR150s. Would an IRF820 and a zener string be considered blasphemy?
(There will be plenty of sand in the filament supplys, so the amp is already contaminated)

[Eli Duttman]

Originally posted by Fuling
Nope, no VR150s. Would an IRF820 and a zener string be considered blasphemy?
(There will be plenty of sand in the filament supplys, so the amp is already contaminated)

Heck, no. "Sand" can be useful. Just that it's easier to kill the noise of VR tubes than that of Zener diodes.

[Fuling]

kill the noise

I guess film caps at strategic locations can work wonders here?

I´ll have to see if there´s space for this inside the chassis, I´m not so sure about that

[PRR]

A 1935 AL1 driven by a 1969 PC86? 14mA/V into 2.8mA/V? That's really spanning history!

> I don´t know anything about the maths behind that feedback resistor I will just start at, say, 100k and adjust until it works as intended

For small-signal analysis:

The PC86 is a voltage-to-current converter. The AL1 with a 100K resistor around it is a current-to-voltage converter. Assuming (wrongly) that the AL1 has infinite gain, the total gain (to the AL1 plate) is just Gm*100K.

If you add an unbypassed cathode resistor to the PC86, you are replacing the intrinsic Gm with a new, more-linear, but lower, Gm. "The feedback should be applied to a high impedance node." You are robbing Peter to pay Paul. Total gain available to cancel errors is still Gm. The cathode resistor will linearize the input stage; may even be necessary depending on signal level.

Since the AL1 does not have infinite voltage gain, and the PC86 has an internal plate resistance plus a DC supply resistor, gain is less.

Assuming the PC86's Gm at this (rich!) operating point is around 13mA/V, and 100 ohm unbypassed cathode resistor, and AL1 voltage gain is 12, with no feedback resistor the gain is about 270; with 100K feedback the gain is about 178. A bare 3.6dB feedback: some but not a lot. 50K feedback gives gain of 132, 6dB feedback, and so on. For "normal" hi-fi power amp sensitivity, total gain (to plate) of about 50 to 100 is called for, try 27K.

OK, for huge feedback, remove the 12K plate-feed resistor and use it for the feedback resistor. The DC conditions are (nearly) unchanged. But it will distort at very low power. When the PC86 plate tries to go high, the AL1 plate goes low, below the PC86's standing plate bias. The PC86 will overload long before the AL1. There must be a simple equality for when the feedback resistor becomes "too low", but I can't figure it out.

On the bench, reduce the 100K and watch the PC86 cathode current. (Put a 10 ohm unbypassed resistor under it.) As Rfb is reduced, the PC86 has to work harder, and will clip before the AL1 is maxed-out. When you get there, go back up in Rfb value, 1.5X to 2X higher.

> I thought about cathode feedback by putting the OPT secondary between the AL1´s cathode and GND

AL1 is only 3 watts. Assuming you take an 8 ohm tap, you have 5V RMS or 7V peak at that tap, while the AL1 needs 14V peak drive. Very little feedback. For 6dB feedback you would want a 33 ohm tap.

[mach1]

I am getting rather confused with this local / partial feedback thang.

Broskie used a worked example of the effect of local feedback where all current to the driver tube is applied via a resistor connected to the output tube's (300B) plate. He states that the minimum voltage drop across this resistor should be equal to the current flowing through the output tube x opt primary impedance (Ip x Rpri). The value of this resistor on the circuit's upper frequency rolloff is also discussed, but no mention is made of the impact of the value of the output tube's grid resistor (Rg) on the circuit dynamics.

However, a chapter of an old text posted by Pete Millett on the asylum (I will the post link shortly when I find it) states that the amount of feedback applied in this type of circuit is directly proportional to the ratio of Rp to Rg (where feedback is greatest when Rg >>> Rfb).

One could therefore assume that the required voltage drop across Rfb (as defined in the Broskie article) drops proportionally with the reduction in feedback due to the impact of Rg.

Neither of these texts explores the partial feedback scenario (as per the circuit diagram in this post), where plate current to the driver tube is applied via a combination of conventional plate resistor from the Ht line + Rfb from the output tube plate.

Could somebody please explain how to analyse the partial feedback scenario in the terms I have described, and in particular how to calculate the required voltage drop(s) accross Rp and Rfb to get full power output ?

Some mathematics would be great.

[mach1]

Correction to previous post

Sorry i meant to say

the amount of feedback applied in this type of circuit is directly proportional to the ratio of Rfb to Rg (where feedback is greatest when Rg >>> Rfb).

paulm