• WARNING: Tube/Valve amplifiers use potentially LETHAL HIGH VOLTAGES.
    Building, troubleshooting and testing of these amplifiers should only be
    performed by someone who is thoroughly familiar with
    the safety precautions around high voltages.

M. Koster’s "Schadeode" Push-Pull Version?

I was looking at Michael J Koster's 'Schadeode' circuit, as found here:
RE: hybrid - Michael Koster - Tube DIY Asylum

and here:
Tube DIY Asylum

Looks like a fantastic way to use 6L6s.

So I was thinking.... Why not push-pull? I threw it all into spice, and as long as you can match up a couple of DN2540 or 01N100D MOSFETs, the results are nothing short of spectacular. BUT...

How likely is it that matching these MOSFETs would result in a well-balanced push-pull circuit? Would using the two MOSFETs in a LTP with a good stiff CCS in the tail balance them well enough?

I'm wondering why nobody else has posted about this, since it seems an obvious thing to try, and looks so good in simulations.

What do you think?

(Sorry I can't post a schematic. That will have to wait until I get home.)
--
 
I breadboarded similar thingy, with 6J52P driving GU-50 with cathode bias. MOSFET has higher transconductance, and more linear V to I conversion. The drawback is, linear current swing drives non-linear resistance of parallel feedback. That's why I went by a different path, when a pentode drives more linear impedance of a feedback voltage divider.
 
Here's a schematic of what I was thinking of.
 

Attachments

  • _PP-Schadeode_EXPERIMENTAL_00.png
    _PP-Schadeode_EXPERIMENTAL_00.png
    31.7 KB · Views: 1,006
The drawback is, linear current swing drives non-linear resistance of parallel feedback. That's why I went by a different path, when a pentode drives more linear impedance of a feedback voltage divider.

I'm trying to understand this.

Are you saying that a pentode input tube with its own separate plate resistor and screen resistor, with a 'plate-to-plate' feedback resistor from the output tube plate to the plate of the input pentode, results in the input tube working into a more linear impedance than if the feedback resistor is also the plate (or in this case, drain) load resistor for the input stage?

In other words, are you criticizing the 'Schadeode' idea in general, whether for SE or PP, or is this only an issue if trying to use the 'Schadeode' idea in a push-pull configuration?

--
 
Isn't it just the way feedback should behave ? Less amplification of the output stage gives less load and more amplification of the driver to compensate. After all every amp is non linear and the feedback is there to correct things (allmost).

Mona
 
I'd figure that the higher the transconductance and gain of the input device, the better it would be able to correct non-linearities in the output device.

Is the problem in this M. Koster-inspired push-pull circuit that the drain resistor and feedback resistor have been combined into one? (R4 and R5 in the schematic.)

691037d1531190612-koster-schadeode-push-pull-version-_pp-schadeode_experimental_00-png


Is it a better idea to take the more traditional approach of RC-coupling the input stage to the output stage, and adding a separate feedback resistor from the output pentode plate to the drain (or plate) of the input device? (Actually from output pentode plate to output pentode control grid, but it amounts to the same thing in this circuit.)

In other words, is it better for the input device to work against a voltage divider composed of passive resistors than it is to use its internal resistance as one leg of the feedback voltage divider?
 
Isn't it just the way feedback should behave ? Less amplification of the output stage gives less load and more amplification of the driver to compensate. After all every amp is non linear and the feedback is there to correct things (allmost).

Mona

Yes, if feedback divider's ratio is stable. But if is non-linear, the result is non-linear. Thst's why I always use the best resistors in feedback voltage divider.
 
Concerning the original Schadeode schematic, why is the cathode bypass cap connected to B+ rather than gnd? Makes PSR worse and requires a high voltage cap ... Cannot see why it should be beneficial .

That's the 'ultrapath' connection of cathode bypass. It's controversial.

In this particular amp, since the MOSFET drain is DC-coupled to the 6L6 grid, the 6L6 cathode has to sit at >70V, so a high voltage capacitor would be required for either connection.

Ultrapath connection would need 500V - 70V = 430V or higher rated. Normal cathode bypass to ground would need 70V or higher rated.

In my proposed push-pull variation, the B+ would be at about 450V, so a 450V rated cap would work (450V - 70V = 380V). Those are easy to find.

No comments on whether a push-pull version would be do-able? I'm wondering how difficult it would be to select a good pair of DN2540 close enough in both current draw and gain/transconductance to work as a push-pull pair.

Yes, if feedback divider's ratio is stable. But if is non-linear, the result is non-linear. Thst's why I always use the best resistors in feedback voltage divider.

So, when one of the resistors is replaced by the plate resistance of the driver tube (or drain-source resistance of a MOSFET), does that mean you now have a lower quality resistor in the feedback divider?

How variable is the Rd-s of a DN2540 MOSFET with only about 4mA Id-s?
 
So, when one of the resistors is replaced by the plate resistance of the driver tube (or drain-source resistance of a MOSFET), does that mean you now have a lower quality resistor in the feedback divider?
Mosfet output impedance in schadeode is assumed to be high and not a concern in Wavebourne's comment. Non linear effective feedback resistance (R1 / gain of V3) due to non linear gain of V3 is. I think he prefers using a different topology, a low output impedance stage driving a voltage divider or something similar.
How variable is the Rd-s of a DN2540 MOSFET with only about 4mA Id-s?
I would use Id > 10ma if I want to use mosfet as input, member TheGimp discuss FET Source Follower Distortion thread at low Id.
 
I think the usual complaint about leaving out the driver load resistor is that it is hard to get enough operating current thru a driver pentode that way (large voltage drop thru Rfdbk). The Mosfet should be fine on that account. A driver triode would certainly call for some load resistor to control the Rfdbk ratio.

Pentode or Mosfet driver output Z should be high enough to ignore in the Rfdbk divider. However, both pentode and Mosfet have an Iout = k Vin ^2 transfer (square law if current is up enough) so a driver cathode or source resistor is called for to linearize the driver V to I in the operating range.

While the "Schade" shunt Fdbk works well to turn the output device into a linear triode, one issue to consider is the OT leakage L. The plate is not the best actual Vout sampling point. Typical OTs with UL taps should have better coupling to the secondary at the UL taps (for UL use-age), so could make a better place to take the "Schade" Fdbk from. -If- it remains stable with the OT leakage L now (more so) in the N Fdbk loop. And the two P-P sides usually have wildly different leakage L for typical single bobbin OTs. But since these are near constants, they could be compensated for in the driver loops.
 
Last edited:
Exactly, that's confirming how I usually see it - the plate of the driver shunts the high impedance feedback signal coming from the plate of the output tube. If the impedance of the driver plate is too low, it shunts the feedback signal too heavily. Since a Pentode has a high impedance at it's plate, it is far better suited to this style of feedback. If you have to use a triode, increase it's effective plate resistance by using a high-mu triode with an un-bypassed cathode resistor.
 
Pentode or Mosfet driver output Z should be high enough to ignore in the Rfdbk divider. However, both pentode and Mosfet have an Iout = k Vin ^2 transfer (square law if current is up enough) so a driver cathode or source resistor is called for to linearize the driver V to I in the operating range.

Yes, I found that to be necessary in the MOSFET LTP. In my schematic, both DN2540s have 220 ohm source load resistors, 'above' the ccs in the tail. In spice simulation, without those source load resistors gain goes wildly high and I'm sure in real life balance would go right out the window.

While the "Schade" shunt Fdbk works well to turn the output device into a linear triode, one issue to consider is the OT leakage L. The plate is not the best actual Vout sampling point. Typical OTs with UL taps should have better coupling to the secondary at the UL taps (for UL use-age), so could make a better place to take the "Schade" Fdbk from. -If- it remains stable with the OT leakage L now (more so) in the N Fdbk loop. And the two P-P sides usually have wildly different leakage L for typical single bobbin OTs. But since these are near constants, they could be compensated for in the driver loops.

Is that the argument for 'E-Linear' connection? OK, I'll try a simulation using the screen taps to feed the drain load/Rnfb resistors to the MOSFET LTP. The resulting amp would have less gain that way, certainly.

I have a beater Dynaco ST70 chassis that's been stripped down to its sockets and transformers several times over its lifetime. I was thinking of building this up in that chassis. It will be tight, though. The 1.5k cathode resistors for the 6L6s will need breathing room, as they will be dissipating over 6W each. I'll need to use 15W rated resistors minimum. 25 watters if I can find them.
--
 
Yes, if feedback divider's ratio is stable. But if is non-linear, the result is non-linear. Thst's why I always use the best resistors in feedback voltage divider.
If the driver is a triode you are right,the Ri changes giving a non-linear devider.
But here the driver is a mosfet, allmost a constant current source, no real divider there.It has to produce a current that gives Va-Vg1 on 100k resistance.
With no feedback (100k to +B) the driver current has to give only Vg1 on the resistance.Less distortion at the cost of more input, as usual with fb.
Mona
 
If the driver is a triode you are right,the Ri changes giving a non-linear devider.
But here the driver is a mosfet, allmost a constant current source, no real divider there.It has to produce a current that gives Va-Vg1 on 100k resistance.
With no feedback (100k to +B) the driver current has to give only Vg1 on the resistance.Less distortion at the cost of more input, as usual with fb.
Mona

The same way we can assume that an output tube has similarly almost linear transfer curve (well, straight line), so why feedback?

But if the tube is not linear, driver's current would drive non-linear impedance of non-linear device that is the feedback resistor's resistance divided by real output stage's momentary gain. What would we get then?
 
I combined a bunch of stuff in this idea.

1) M. Koster 'Schadeode' but with push-pull and ultralinear OPT.
2) Doug "Bandersnatch" 'E-Linear' screen grid to control grid feedback.
3) RCA-style negative feedback from output tube plate to driver source.

Combining them all, I get what looks like a well-performing circuit. LTspice predicts THD of around 0.05% at 1W out into 4 ohms, about 1% THD at 20W out. H2 and H3 equal (about -75dB) at 1W out.

However, the straight push-pull 'Schadeode' with plate-to-grid feedback and ultralinear OPT connection looks 'too good to be true' in LTspice. Would you believe 0.005% THD at 1W out into 4 ohms? Nah, I don't either.

Anyway, here's the mish-mash version, attached. Please feel free to criticize it for all it's worth.
--
 

Attachments

  • _PP-UL_Schadeode_ELinear_plate-to-cathode-FB_idea.png
    _PP-UL_Schadeode_ELinear_plate-to-cathode-FB_idea.png
    43.2 KB · Views: 328
Last edited:
RE: Wavebourn

The output tube impedance variation and near square law Vgrid to Iplate are distortion causing factors for sure. But the voltage on the OT/plate is the real deal output, what we are measuring, so getting the N Fdbk to linearize that is what we want.

----------------------------------------------------

Gee, I was just about to mention the RCA 50 Watt'r schematic with "Schade" N Fdbk -and- N Fdbk to the driver cathodes, driver CFB (Mosfet sources here). (schematic below) The N Fdbk to the driver cathode/source linearizes the V to I conversion in the driver stage nicely. The low Z plate load for the driver, however, mostly winds down the overall driver CFB gain.

0.005% dist. Huh! Probably some coincidence from the device models, but still, very interesting, can't be much wrong with it. Would leave Citation II and Mac 275 way behind. Schematic? (just delete C3 and C4?)

You might try C3 and C4 connected to the SG1 and SG2 OT taps instead of the plates.

And then maybe try changing the R4,R13 and R5,R14 Schade Fdbks into just loads to B+, so making it totally driver CFB based. Some optimizing R values may be needed for 6L6 grid bias level. Then we have some comparison between driver CFB and shunt "Schade" effectiveness.
 

Attachments

  • RCA_50Watt_Amp.JPG
    RCA_50Watt_Amp.JPG
    534 KB · Views: 310
Last edited:
0.005% dist. Huh! Probably some coincidence from the device models, but still, very interesting, can't be much wrong with it. Would leave Citation II and Mac 275 way behind. Schematic? (just delete C3 and C4?)

The problem with removing C3 and C4, allowing DC current to flow from output tube plates to driver MOSFET sources, is that it seriously messes with the driver stage biasing. I could RC-couple the MOSFET drains to the output tubes' grids, but then I'll be increasing the parts count even more. I was going to attempt to cram all of this into a Dyna ST70 chassis, and it's tight in there.

That 0.005% THD at 1W into 4 ohms figure was attained by really goosing the current through the 6L6s, beyond what the ST70 power transformer can do. But even with it cooled down to within the ST70 power transformer's current budget, LTspice is saying 0.01% THD at 1W. The schematic is attached to this post.

To get it to that 0.005% THD @ 1W/4R figure, change R5 and R20 to 39k, leaveing R4 and R19 at 47k. That will raise the voltage at the DN2540 drains, reducing the grid bias on the 6L6s so that they draw 68mA each instead of 55mA each. That will draw a total of 150mA from the B+ supply. The extra current reduces 3rd harmonic distortion a bit.

In simulation, THD is lower if you take the feedback from the output tubes' plates to the driver sources. I think there's just more gain there, so more feedback. There's less gain from the screen grids, so there's less gain for feedback available from there.

smoking-amp said:
You might try C3 and C4 connected to the SG1 and SG2 OT taps instead of the plates.

Do you mean don't tap feedback off the plates at all? Tap two feedback loops from the screens? (Output tube screens to MOSFET drains, and also output tube screens to MOSFET sources?)
 

Attachments

  • _PP-Schadeode_EXPERIMENTAL_01.png
    _PP-Schadeode_EXPERIMENTAL_01.png
    44.9 KB · Views: 315
Last edited: