coppatone GM70 /4

Yeah, I have spent some time thinking + doodling, and only now do I have something to show that looks good to me and works out.

But first, thanks to jc morrison for chipping in with a comment to the previous instalment. Looking forward to more of that. Yes, the ECC81/12AT7 comparison was a bit of hyperbole to get the message across. jc’s manic laughter, in reply, balances that out—enjoyed that. My point however is not gain, it is drive and power transfer. I always imagine a tube coupled out by a series resistor as trying to control the next stage by means of shaking a rope. The larger the resistance, the looser the rope. Happy I could avoid the resistor altogether.

Last time I ended with pondering current sinks, cascoding, ring-of-two and teddyReg, and I did get around to putting that all together:

The tube on the left is a simple CCS, protected by the resistor on top of it (should drop half of the 500V at design current), the latter also forms a filter with the cap. At the bottom we see the ring-of-two, enhanced with the led, setting the sink current. The tube on the right cascodes this and its plate is the current sink output, dealing with the large DC voltage, large voltage swing and heat dissipation. Looks nifty and appealing to set up with a dual-triode. But too complicated for my taste to include in the coppatone amp.

But now let us look at some of the elements that make up my current schematic. First up, the bypassed resistor on the kathode of the second stage:

My efforts to remove this resistor resulted in getting quite a bit of respect for it. With its large value (3k9), it not only provides for a lot of self-biasing compensation for the Ed, but also intervenes when things go whacky with either the 437A or the Ed.

Similarly, jc points out—while talking about the series resistor between first and second stage—‘the 10k resistor does many jobs!’ Ah, I was too obsessed getting rid of it to spot that. I guess he means that when the Ed starts to draw a mA of grid current, there is 10V of intervention in its grid voltage.

Back to the kathode of the second stage. Looking at the resistor as an auto-B- mechanism for that stage (below, left), I came up with a simple VR tube stage (right):

Two VR tubes fix the B- for the second stage to 300V. And if things do ‘go whacky’ and the VR tubes cannot handle it anymore, then the resistor does the same job of intervening.

Why the VR tubes? Because then there are no caps left in this choke loaded stage. In the design on the left, all signal current through the tube must go through the bypass cap shown and the decoupling cap of the B-. This combines with the inductance of the plate choke to a third-order high-pass that is very hard to tame. If there is one thing that is very worthwhile, in my experience, it is removing capacitors around a choke-loaded stage, resulting in a single-order response without capacitor mush.

Second element: driving the grid of the GM70 into A2. This is tough. Payback time, really. From the class A2 operating point we got 3dB of extra power (headroom); a cleaner first (ten) Watts; using a reasonable 7K output transformer and operation below 1000V. But now we have to work for it. Looking at the datasheet graph again:

I have marked the slope of the final 20V of grid drive (+40 to +60V). Grid current rises by ~20mA (to a maximum of 30mA) during that final stretch. Grid impedance is ~1kOhm, there. A 2A3 or 300B is not going to push that around, even if they are helped by a low-DCR choke taking care of about half the current.

Remember, the stage driving the GM70 needs to be able to swing 260Vpeak for headroom reasons. This means that a step-down (auto-)transformer stage is not in the cards (1040Vpeak into a 4:1 step-down? you design it). So we are looking at inserting a kathode follower before the GM70. Direct-coupling a follower to the second stage turned out to be relatively easy (left):

Just as before, the grid of the GM70 is connected via a choke to ground; a safety feature just in case a stage before goes whacky or a tube is pulled. With the two chokes in series around ground, this asks for for having a look at the variant on the right, forced symmetry by auto-former action and all. There are two reasons that make me say ‘better not’:

1. the load of the GM70 grid is exposed to the plate of the second stage. The point of the follower is to buffer that;
2. vaguely it looks like that fad of the 80s (with a long shade into the 90s): SRPP (but then again, the 80s are back, aren’t they?).

What are the requirements for the kathode follower tube?

• To have a fighting chance of driving the GM70 grid all the way, it should have an output impedance ≤ 100 Ohm. This means we are looking for a tube with a transconductance (gm) ≥ 10mA/V.
• Biasing: there is some negative voltage coming from the direct coupling to the second stage (say, -9V in case of a 2A3), but most is provided by the self-bias of the kathode choke DCR. This self-bias voltage eats away the headroom of the stage, so it better not rise much beyond +40 Volts.
• Operating point: in principle it could run on 1mA, because it needs to sink only tens of µA and source the 30mA. But no tube is going to be on fire at 1mA and have a gm of 10mA/V. So some decent current has to be allowed for, like 40mA. Too much current, however, gets us in trouble with the previous point and the next.
• Plate dissipation: 260V × 40mA = 10.4W. 12 Watts or so would not hurt.
• Max plate voltage ≥ 275V.
• On top of that I want to take this opportunity to inject some of that punk attitude—that jc was exhorting to do in his comment—in the form of cheap, unfashionable TV tubes. No collectable DHTs. Also, no skinny noval tubes and a top cap would be cool.

After a lot of rummaging in TDSL, then going through whole series of TV PL tube types, I came up with EL/PL508 (triode), half a 6528 and the 6BX7 (paralleled). After that I picked the collective brain of the Joelist. Among the suggestions, Jeffrey Jackson proffered the N339 (groovy, baby!) and jc insisted I check out the 6AV5, so I did. All of these tubes meet the requirements, have been checked to work using their (triode) curves and are cheap (the 6528 excepted).

Please note that this search also turned up lots of candidate tubes that simply do not work out: either not meet the ratings; do not bias in ‘the zone’ or simply do not have the gm at the part of the voltage swing where it counts. Some of the more popular types that do not work out include: 6550, EL34, EL86, 6BL7, 6L6, any other EL/PL5xx type.

Another result of this search is a community pointer: cheap, punk tubes for the first and second stage. For the input stage, the 6BX7 has the voltage ratings and the µ (10) to make it work. For those who really need that bit more gain, there is the related 6BL7. For the second stage, there is the mighty fine 6CB5A that Thomas Mayer has been championing for a while now, building his SE amplifier concept around it. I really wished this one (fat, linear, TV + top cap) would work out for the kathode follower, but it doesn’t.

The third element is the GM70 bias supply. It does not need to handle the GM70 DC or AC current, but it does need to shunt the 30mA peak grid current. While working on it, I had at one point this:

(no, the ‘0’s of the 500V and 840V supplies are not connected to each other)

Above, the separate 840V supply of the GM70 is referenced to the top of a VR tube, comfortably higher in voltage than the nominal 70V bias needed for the GM70 (105V from a 0C3 will do). Then the variable resistance under the kathode follower should have biased the kathode voltage anywhere between +25 and +45V, setting up the GM70. I say ‘should have’ because this does not work. That sizeable transconductance we require from the follower tube also means that to move the its kathode one volt up or down, we must change the current through it by 10mA. That is really not practical. So it was back to the drawing board.

And thus it became inevitable to use two buffers to control both the grid and the kathode of the GM70. Direct coupling makes that the grid of a GM70-grid-driving-follower cannot be placed independently, which makes it impossible to also use this follower for biasing the GM70. So here we go:

(again, the ‘0’s of the 500V and 840V supplies are not connected to each other)

I picked a MOSFet (say, a DN2540) for the bias follower because its transconductance is an order of magnitude greater than can be expected from a tube (100 vs. 10mA/V). Having another tube follower in the bias position would mean doubling the disruption the grid current causes (the followers are in series for this current). It is very satisfying that the MOSFet shunts the grid current coming off the GM70 kathode directly back to where it belongs: the plate of the grid-driving-follower.

So if the MOSFet is so great, why not use it also on the grid of the GM70? (I always ask myself about a regulator circuit: ‘would you listen to this if configured as an amplifier?’) Well, you listen all the time to the grid-driving-follower, from the first (milli-)Watt to the last. I cannot bring myself to place a MOSFet in that position, capacitances and MOS-mist and all. In the bias position it just sits there, most of the time, coming into action when trouble starts. Btw, I could use some advice where it comes to protection and deployment of MOSFets under these circumstances, so comments are welcome if you got some experience to share.

The final element is the one that ended the previous instalment and started this one: biasing the second stage via the first one. Using a MOSFet set-up like above is one of the options, but one useful result of the bias-circuit-fiasco was that this type of circuit can be used for the first stage:

A choke in series with some variable resistance instead of the current source. Here, adjusting bias is a simple voltage divider between the dynamic rp of the input tube and the variable resistance under it. Cool, but there is one snag: the first stage, being now choke-loaded, should not interact with the B- decoupling capacitor (on the right, above). No caps around choke-loaded stages:

The extra resistor to the junction of the VR tubes and the second stage kathode is necessary to supply the current this stage is suppose to draw (25mA).
Putting it all together, we get this schematic [v1.1, corrected input biasing]:
Which looks very satisfying. Between the input and output transformers, we see four stages direct-coupled by nothing but wire. The first three stages are all choke-loaded and are capacitor-free. The output stage, however, is wholly conventional: the OPT and the capacitor at the end of the 840V power supply close the loop. But this loop is self-contained, no AC or DC current of the output stage can find its way through the earlier stage circuits or ground. Thus no cock-up at low frequencies, like you get with ‘ultra-path’ type of circuits. A great relief is that the number of power supplies required has been kept in check: one +/-500V supply and the 840V output supply.

Although not unsafe—knock on wood—the coppatone design does ask users to check and adjust currents every once in a while. This in contrast to the safe-as-houses silbatone RI-25. jc had to design an amp with no user adjustments, for billionaires who might engage in some tube rolling. That certainly is a tougher assignment than mine.

That's it for today. Coming up: values for everything on the schematic and then, power supplies.