• 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.

Folded/shunt cascode differential question

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
I recently started reading a bit about the folded/shunt cascode (thanks, Rod Coleman for your interesting posts). It's a cool circuit and I have a potential project for it as the differential front end on a push-pull amp (Allen Wright style).

shunt-cascode.png


I understand the basic workings and am comfortable enough to be able to calculate bias/dissipation/etc as a single-ended cap-coupled amplifier. However if simply mirroring the layout in a LTP arrangement, one runs into CCSs on top and bottom of the circuit.

Notwithstanding Morgan Jones, this is usually frowned on. But in the case of the folded cascode, I wonder if the multiple CCSs are an issue. My thinking is that surplus current is driven through the shunt anyways, and so CCS current differences at anode or cathode (with respect to the tube) are ok. Thus the two CCSs would not simultaneously be trying to set a different current. The cathode CCS sets current through the triodes and the anode CCS sets current through the triodes and the shunt.

differential-shunt-cascode.png


Has anyone tried this kind of circuit as differential? Maybe a current mirror instead of a CCS would make a good anode load, too?
 
There would be plenty in the CCS, I'm sure.

Back to my original post though, I guess I could substitute the anode loads in a cascode LTP for resistors too. The tail CCS would still set the balance/splitting and the resistors would create some feedback for the transistor (I think).
 

PRR

Member
Joined 2003
Paid Member
....My thinking is that surplus current is driven through the shunt anyways, and so CCS current differences at anode or cathode (with respect to the tube) are ok.....

Your thinking is correct. You need to be more confident.

But why stop there? Why have CCSes in the top? All you really need is much-much more than transistor emitter impedance hOE. Assume 1mA. hOE is 27 Ohms. Much-much-more is 2,700r for 1% accuracy. Pencil 5K for easy math. This leg has 2mA so 10V of drop, which is surely acceptable.

You have fixed bias on bases. With the resistors this wants to be referenced to the resistor supply. This actually does come out to be a "CCS", so we arrive at the same place.

The un-implemented tail CCS needs detailing. We can wrap NFB around the whole shebang by combining the transistor legs.
 

Attachments

  • Sodacose-4.gif
    Sodacose-4.gif
    7.8 KB · Views: 1,271
I recently started reading a bit about the folded/shunt cascode (thanks, Rod Coleman for your interesting posts). It's a cool circuit and I have a potential project for it as the differential front end on a push-pull amp (Allen Wright style).

I understand the basic workings and am comfortable enough to be able to calculate bias/dissipation/etc as a single-ended cap-coupled amplifier. However if simply mirroring the layout in a LTP arrangement, one runs into CCSs on top and bottom of the circuit.

Notwithstanding Morgan Jones, this is usually frowned on. But in the case of the folded cascode, I wonder if the multiple CCSs are an issue. My thinking is that surplus current is driven through the shunt anyways, and so CCS current differences at anode or cathode (with respect to the tube) are ok. Thus the two CCSs would not simultaneously be trying to set a different current. The cathode CCS sets current through the triodes and the anode CCS sets current through the triodes and the shunt.


Has anyone tried this kind of circuit as differential? Maybe a current mirror instead of a CCS would make a good anode load, too?


Glad you're enjoying the shunt cascode!

I haven't implemented a differential, but it should work very well.

The shunt cascode will offer high-to-very-high gain and low distortion, if designed carefully.

My thoughts:

- The input and output are both referred the ground, so if possible, keep the cathode at a low-impedance to ground, and the anodes at a high impedance. I understand the motive for PRR's anode resistors, but depletion FETs (DN2540s are perfect here) improve the measured performance enough to justify the small extra effort - they are very easy to assemble. PNP or PFET CCS can be made to work, but the performance is degraded.

High impedance in the anode circuit, plus the fact that the instantaneous (and average) supply current is perfectly static, even at large output swings, mean that the gain-stage is quiet, and needs no supply regulation.

With a differential stage, some consideration must be given to the circuit balance, to account for the high gain. I would suggest that individual resistors are used in the cathodes of each triode, before joining at the long tail. At least 50Ω, as a starting estimate, depending on the gm of the triode. The gain is reduced, but the feedback also has a useful linearising effect. A dual triode would seem like a good idea for closer matching, and many pleasant TV bottles have high gm, low-voltage operation and come in duals.

I might prefer to use a resistor as the long tail, because common-mode rejection should not be the most important design choice, but of course, it depends on what feeds the inputs.

As usual with high-gain stages, care must be taken to avoid running out of phase margin. Watch out for loop-area of the grid, anode, CCS, and load-resistor circuits; stopper the anode as well as the grid.
 
In case anyone wonders about how good the shunt cascode is: here's some spectrum.

This is a 6Э5П in triode running at 202V and 32mA through my shunt-cascode explorer board. The load resistor returns to negative voltage in this case. The performance and sound is very pleasing, but I need to add a low-frequency servo at some point, to avoid the need to occasionally adjust the current-source.

The stage is driven directly by an ES9018 DAC, and the measurement is taken at an output of ~50V rms - almost 150V peak to peak.
 

Attachments

  • 49V8rms-145Vpp-109V-idle.png
    49V8rms-145Vpp-109V-idle.png
    49.1 KB · Views: 1,296
  • 20170120_085131.jpg
    20170120_085131.jpg
    320.3 KB · Views: 618
But why stop there? Why have CCSes in the top?

We can wrap NFB around the whole shebang by combining the transistor legs.

Thanks, PRR! I was part of the way there after thinking about it a bit more yesterday. The emitter sets the anode voltage and drop across the load, so using a resistor makes calculating the PNP current easy (given the cathode CCS). The PNP is creating a 'vertical' loadline for the tube (just like a regular cascode), so the load in parallel to it isn't so critical from the tube's perspective. I like the feedback idea too (kind of like Baby Huey).

Rod, I really appreciate your thoughts. The point about PSRR and CSS over resistor in the anode makes sense. This choice (for me) would probably be determined by tube choices and how easy a quiet supply would be to create.

It's the potential for a single gain stage (if you can call cascode a single stage) that has really drawn me to this circuit. I'd like to do a "simple" two-stage PP design with trioded EL34/KT88/etc outputs. If it works out, the next step would be an AB2 amp along the same lines (grid drivers add that extra layer of complexity).
 
For my 300B driver, the load resistor is 27K, and the gm of the 6Э5П starts at 30mS, degenerated to about 7mS by the action of a 110Ω cathode resistor, unbypassed.

So the gain in this case is 7e-3 * 27e3 = about 190 (45.5dB)

Meanwhile, a D3A with 24KΩ load, and 53.3Ω unbypassed will yield about 12.9mS and x309 gain (50dB), but I have not needed to use that yet.

In other practical terms, I have gone from DL103 Moving Coil to 300B grid in 3 stages of shunt cascode: 2SK369 JFET shunt cascode MC stage, 6H23 shunt cascode EQ stage, EF184 shunt cascode driver stage. All of these stages run open-loop, and degenerated by cathode resistors.
 
shunt-cascode-pp.png


Some quick rough guesses as to values. The 150V base reference would probably be a 0D3. Around 100x gain.

This puts 5mA through the PNP (MJE350 just because I have some). Less current would allow for a larger anode resistor, less dissipation across the PNP, and higher collector quiescent voltage. OTOH, more transistor current increases gm. But is that advantage swamped by the relatively low gm of the tube?
 

PRR

Member
Joined 2003
Paid Member
...more transistor current increases gm. But is that advantage swamped by the relatively low gm of the tube?

The transistor is working common-base. Its Gm hardly matters as long as it is high. The stage gain is Gm(tube) * RL(bjt) for practical purpose. (Never forgetting "RL" includes load and stray-capacitance suckage.)

...It's the potential for a single gain stage (if you can call cascode a single stage) that has really drawn me...

Don't get over-drawn yet.

The "folding" doesn't, by itself, change the action. It is still a cascode. With some simplifying assumptions they both work the same.

The folding does un-stack the supply voltage (at the cost of those plate loads). If you need large swing from small supply, this may be compelling.

The "simplifying" is assuming all devices flow the same current. This is "only natural" in cascode, though a diversion is possible. Un-equal current is "easy" in the folded form. Why would we do that? The Gm which matters is the tube(s), and good Gm means high current. However it *may* be that the BJT does not need such a high current. While current "usually" increases as you go from input to output, sometimes we do otherwise. The folded form can do this by choice of top and bottom current sources. You "can" do this in the cascode but the diversion adds unnecessary parts, may not save supply power, or may want an added supply.
 

Attachments

  • Sodacose-5.gif
    Sodacose-5.gif
    18.5 KB · Views: 621

PRR

Member
Joined 2003
Paid Member
I have often wondered if you could find a vacuum diode with complementary characteristics to the lower tube in a shunt cascode, to get the distortion down...

This wonderment leads down the rabbet hole. Diode compensation has been used in the past, notably to improve DC drift without an extra triode, but there's more to explore.

Somebody (you?) recently posted a hollow-diode distortion for musicians. (Fender did it 20+ years back: diode-strapped 12AX7 worked just like a pair of Ge or Si diodes.) Your site has a clever curve tracer which would find possible "complements".

I'm sure you've seen the pre-distortion diode plan in the LM13700 (below). For the same THD you can run 5X the level; or for the same level you can reduce THD an astonishing 30:1. (Yet musicians seem to prefer the soft-onset to the sudden clipping.)

I'm simming a thingie which uses complementarly driven tube grid-cathode impedances to good effect and some advantages over the way it has been done since the 1930s; but I'm not sure of my thoughts yet, and it has no application to what Sodacose is doing.
 

Attachments

  • LinearizingDiodes.gif
    LinearizingDiodes.gif
    21.2 KB · Views: 618
Last edited:
The Gm which matters is the tube(s), and good Gm means high current. However it *may* be that the BJT does not need such a high current. While current "usually" increases as you go from input to output, sometimes we do otherwise. The folded form can do this by choice of top and bottom current sources.

That's the meat of what I was wondering. Thanks!

I misspoke about collector voltage in the earlier post though. Obviously decreasing collector current lowers collector voltage (everything else constant).

Seems like I'm looking to strike a balance between transistor dissipation and a high enough collector voltage to avoid clipping the output. Not that I need extreme swing with the current plan (shooting for around 100x), but still an important aspect of this circuit.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.